{"dictionary":[{"anchor":"model","at":"b3853-3914","end_byte":3914,"end_line":12,"heading_id":"the-model-and-the-result","line":12,"markdown":"Support statements refer to the relative topology of $[0,1)$.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Time","paper":"short","section":"1. The model and the result","start_byte":3853,"tex":"[0,1)","words":"The overlap interval from zero up to, and excluding, one, used as the time domain."},{"end_line":15,"line":14,"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.extend","tex":null,"words":"Extends a function on [0,1) to the whole real line by zero, so ordinary integrals can be written."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":22,"heading_id":"the-model-and-the-result","line":17,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":"\\mathcal U","words":"The admissible class: nonnegative, nondecreasing, right-continuous, integrable functions on [0,1), bundled with these proofs."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":18,"heading_id":"the-model-and-the-result","line":18,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter.val","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":"\\gamma","words":"The function itself, from [0,1) to the reals."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":19,"heading_id":"the-model-and-the-result","line":19,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter.nonneg","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The function takes nonnegative values."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":20,"heading_id":"the-model-and-the-result","line":20,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter.monotone","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The function is nondecreasing."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":21,"heading_id":"the-model-and-the-result","line":21,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter.right_continuous","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The function is right-continuous at every point of [0,1)."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":22,"heading_id":"the-model-and-the-result","line":22,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter.integrable","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The function is Lebesgue integrable on [0,1); it may still grow near one."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":29,"heading_id":"the-model-and-the-result","line":24,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":null,"words":"A probability space with a standard Brownian motion on it, which every Lean definition below takes explicitly as W."},{"end_line":25,"line":25,"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem.law","tex":null,"words":"The probability measure of the space; the papers' expectation is integration against it."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":27,"heading_id":"the-model-and-the-result","line":27,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem.B","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":"B","words":"The Brownian path, indexed by nonnegative real time."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":29,"heading_id":"the-model-and-the-result","line":29,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem.brownian","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":null,"words":"States that the process is standard Brownian motion under the measure."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":36,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":33,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.incrementFiltration","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":null,"words":"The filtration generated by the Brownian increments after time t, namely the increments of B from t onward."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":41,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":38,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":"\\alpha","words":"A process bounded by one and progressive for the increments after t; Lean indexes it by time elapsed since t."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":39,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":39,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control.val","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":"\\alpha","words":"The control process; Lean reads the paper's value at time s at elapsed time s minus t."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":40,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":40,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control.progressive","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":null,"words":"The control is progressively measurable for the Brownian increments after t."},{"anchor":"model","at":"b4270-4475","end_byte":4475,"end_line":41,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":41,"markdown":"The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control.bound","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4270,"tex":"|\\alpha|\\le1","words":"The control is bounded by one at every time and every sample point."},{"anchor":"model","at":"b4270-4475","end_byte":4475,"end_line":47,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":43,"markdown":"The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4270,"tex":"\\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|-\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds","words":"The random reward of one control, sample by sample, whose expectation the value function maximises."},{"anchor":"model","at":"b4270-4475","end_byte":4475,"end_line":51,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":49,"markdown":"The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4270,"tex":"\\Phi_\\gamma(t,x)","words":"Lean defines this as a supremum of expected payoffs over controls; the short paper presents it through the backward equation."},{"anchor":"model","at":"b5019-5106","end_byte":5106,"end_line":54,"heading_id":"the-model-and-the-result","line":53,"markdown":"Write $u=\\Phi_x$ for the spatial gradient and\n$a=\\Phi_{xx}$ for its spatial derivative.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.gradient","paper":"short","section":"1. The model and the result","start_byte":5019,"tex":"u(t,x)","words":"The spatial derivative of the value function, with u equal to the first x-derivative of Phi."},{"anchor":"model","at":"b5019-5106","end_byte":5106,"end_line":57,"heading_id":"the-model-and-the-result","line":56,"markdown":"Write $u=\\Phi_x$ for the spatial gradient and\n$a=\\Phi_{xx}$ for its spatial derivative.","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.curvature","paper":"short","section":"1. The model and the result","start_byte":5019,"tex":"a(t,x)","words":"The second spatial derivative of the value function, written as the derivative of the gradient."},{"anchor":"model","at":"b4044-4544","end_byte":4544,"end_line":60,"heading_id":"the-model-and-the-result","line":59,"markdown":"The Parisi functional $\\mathcal P(\\gamma)$ is defined through an\nauxiliary value function $\\Phi_\\gamma(t,x)$, in which $x\\in\\mathbb R$\nis a spatial variable and the overlap parameter $t$ plays the role of\ntime. Write $\\Phi=\\Phi_\\gamma$ and use subscripts for partial\nderivatives. The backward equation, terminal condition and functional are\n$$\n\\Phi_t=-\\tfrac12\\bigl(\\Phi_{xx}+\\gamma(t)\\Phi_x^2\\bigr),\n\\qquad \\Phi(1,x)=|x|,\n\\qquad\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)-\\tfrac12\\int_0^1t\\gamma(t)\\,dt.\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","paper":"short","section":"1. The model and the result","start_byte":4044,"tex":"\\mathcal P(\\gamma)","words":"The Parisi functional, which Lean computes from the value function on the chosen Brownian system W."},{"anchor":"main-result","at":"b6367-6420","end_byte":6420,"end_line":63,"heading_id":"the-model-and-the-result","line":62,"markdown":"If $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","paper":"short","section":"1. The model and the result","start_byte":6367,"tex":null,"words":"The coefficient minimises the Parisi functional over all order parameters, with W held fixed throughout."},{"anchor":"model","at":"b3506-3608","end_byte":3608,"end_line":66,"heading_id":"the-model-and-the-result","line":65,"markdown":"and define the locally finite\nStieltjes measure by\n$$\n\\mu_\\gamma([0,t])=\\gamma(t),\\qquad 0\\leq t<1.\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","paper":"short","section":"1. The model and the result","start_byte":3506,"tex":"\\mu_\\gamma","words":"A measure on [0,1) giving each interval [0,t] mass gamma(t); Lean states this as a property of a measure."},{"anchor":"model","at":"b5107-5300","end_byte":5300,"end_line":75,"heading_id":"the-model-and-the-result","line":68,"markdown":"Let $B$ be standard Brownian\nmotion, and define the diffusion $X$ and its self-consistency function\n$q$ by\n$$\ndX_t=\\gamma(t)u(t,X_t)\\,dt+dB_t,\\qquad X_0=0,\n\\qquad q(t)=\\mathbb E[u(t,X_t)^2].\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":"X_t","words":"A progressive, almost surely continuous solution of the integrated equation on [0,1], starting at zero, for the given W."},{"anchor":"model","at":"b5107-5300","end_byte":5300,"end_line":79,"heading_id":"the-model-and-the-result","line":77,"markdown":"Let $B$ be standard Brownian\nmotion, and define the diffusion $X$ and its self-consistency function\n$q$ by\n$$\ndX_t=\\gamma(t)u(t,X_t)\\,dt+dB_t,\\qquad X_0=0,\n\\qquad q(t)=\\mathbb E[u(t,X_t)^2].\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.squareMoment","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":"q(t)","words":"The second moment of the gradient along a given process X, supplied as an argument."},{"anchor":"main-result","at":"b6273-6622","end_byte":6622,"end_line":88,"heading_id":"the-model-and-the-result","line":81,"markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$\nFor the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","paper":"short","section":"1. The model and the result","start_byte":6273,"tex":null,"words":"The whole conclusion of Theorem 1, plus existence of the measure and of a diffusion."},{"anchor":"contact","at":"b10475-10624","end_byte":10624,"end_line":113,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","line":112,"markdown":"We fix a minimiser. Itô's formula cancels the drift of $u(t,X_t)$:\n$$\ndu(t,X_t)=a(t,X_t)\\,dB_t,\n\\qquad q(0)=0,\\qquad q'(t)=\\mathbb E[a(t,X_t)^2].\n$$","module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.curvatureMoment","paper":"short","section":"2. Contact conditions and the exclusion of gaps","start_byte":10475,"tex":"\\mathbb E[a(t,X_t)^2]","words":"The second moment of the curvature along the constructed diffusion."},{"anchor":"model","at":"b5437-5474","end_byte":5474,"end_line":276,"heading_id":"the-model-and-the-result","line":275,"markdown":"The SDE has a unique strong solution.","module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion","paper":"short","section":"1. The model and the result","start_byte":5437,"tex":"X_t","words":"The diffusion Lean constructs on W, which supplies the existence clause of the theorem."},{"anchor":"model","at":"b3506-3608","end_byte":3608,"end_line":110,"heading_id":"the-model-and-the-result","line":109,"markdown":"and define the locally finite\nStieltjes measure by\n$$\n\\mu_\\gamma([0,t])=\\gamma(t),\\qquad 0\\leq t<1.\n$$","module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","paper":"short","section":"1. The model and the result","start_byte":3506,"tex":"\\mu_\\gamma","words":"The Stieltjes measure Lean constructs from the coefficient, with an atom of mass gamma(0) at zero."},{"anchor":"main-result","at":"b6273-6622","end_byte":6622,"end_line":33,"heading_id":"the-model-and-the-result","line":19,"markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$\nFor the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","paper":"short","section":"1. The model and the result","start_byte":6273,"tex":null,"words":"Theorem 1 for a given Stieltjes measure and the constructed diffusion, stated before transfer to every diffusion."},{"anchor":"main-result","at":"b6273-6622","end_byte":6622,"end_line":59,"heading_id":"the-model-and-the-result","line":35,"markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$\nFor the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","paper":"short","section":"1. The model and the result","start_byte":6273,"tex":null,"words":"The main theorem: a minimiser on W satisfies the full-support conclusion."},{"anchor":"qaoa-boundary","at":"b42353-42413","end_byte":42413,"end_line":10,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":10,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.standardGaussian","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":null,"words":"The standard normal law on the real line."},{"anchor":"qaoa-boundary","at":"b42353-42413","end_byte":42413,"end_line":15,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":14,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gaussianProduct","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":null,"words":"The product of independent standard normal laws over a finite index set."},{"anchor":"qaoa-boundary","at":"b42353-42413","end_byte":42413,"end_line":22,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":21,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coordinate","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":"Z_{j+1}","words":"The coordinate of a Gaussian vector, index reduced modulo N+1; Lean counts from zero, so index j is the paper's Z with index j+1."},{"anchor":"qaoa-boundary","at":"b42230-42291","end_byte":42291,"end_line":24,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":24,"markdown":"a cutoff $0<T<1$, and a mesh $t_j=j\\Delta$ with $\\Delta=T/N$.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.stepSize","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42230,"tex":"\\Delta","words":"The mesh width T divided by N."},{"anchor":"qaoa-boundary","at":"b42230-42291","end_byte":42291,"end_line":25,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":25,"markdown":"a cutoff $0<T<1$, and a mesh $t_j=j\\Delta$ with $\\Delta=T/N$.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.meshTime","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42230,"tex":"t_j","words":"The j-th mesh point, equal to j times the mesh width."},{"anchor":"qaoa-boundary","at":"b42353-42652","end_byte":42652,"end_line":31,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":27,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":"Y_j","words":"The Euler recursion for the discretised diffusion, written for arbitrary coefficient and drift function."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":38,"heading_id":"the-model-and-the-result","line":33,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":"\\mathcal U","words":"The admissible class, using a function on all reals whose constraints apply only on [0,1)."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":34,"heading_id":"the-model-and-the-result","line":34,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter.coeff","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":"\\gamma","words":"The coefficient, a function on all reals; its values outside [0,1) are unconstrained."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":35,"heading_id":"the-model-and-the-result","line":35,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter.nonneg","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The coefficient is nonnegative on [0,1)."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":36,"heading_id":"the-model-and-the-result","line":36,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter.monotone","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The coefficient is nondecreasing on [0,1)."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":37,"heading_id":"the-model-and-the-result","line":37,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter.rightContinuous","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The coefficient is right-continuous at every point of [0,1)."},{"anchor":"model","at":"b3348-3505","end_byte":3505,"end_line":38,"heading_id":"the-model-and-the-result","line":38,"markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter.integrable","paper":"short","section":"1. The model and the result","start_byte":3348,"tex":null,"words":"The coefficient is integrable on the interval from zero to one."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":47,"heading_id":"the-model-and-the-result","line":40,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":null,"words":"A probability space with a standard Brownian motion, bundled together with its sample space, unlike the support model."},{"end_line":43,"line":43,"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace.μ","tex":null,"words":"The probability measure; the papers' expectation is integration against it. This symbol differs from the Stieltjes measure."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":45,"heading_id":"the-model-and-the-result","line":45,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace.B","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":"B","words":"The Brownian path, indexed by nonnegative real time."},{"anchor":"model","at":"b5107-5143","end_byte":5143,"end_line":47,"heading_id":"the-model-and-the-result","line":47,"markdown":"Let $B$ be standard Brownian\nmotion,","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace.brownian","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":null,"words":"States that the process is standard Brownian motion under the measure."},{"end_line":52,"line":51,"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace.filtration","tex":null,"words":"The natural filtration of the Brownian motion."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":57,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":54,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedFiltration","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":null,"words":"The filtration generated by the Brownian increments after time t."},{"anchor":"model","at":"b4550-4622","end_byte":4622,"end_line":60,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":59,"markdown":"Controls are progressively measurable for Brownian increments after $t$.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.Admissible","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4550,"tex":"|\\alpha|\\le1","words":"The control condition: progressive for increments after t and bounded by one; here a property, indexed by elapsed time."},{"anchor":"model","at":"b4270-4475","end_byte":4475,"end_line":66,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":62,"markdown":"The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.controlPayoff","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4270,"tex":"\\mathbb E\\left[\\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|-\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right]","words":"The expected reward of one control; unlike the support model it already includes the expectation and uses elapsed time."},{"anchor":"model","at":"b4270-4475","end_byte":4475,"end_line":69,"heading_id":"the-sk-model-and-the-parisi-variational-problem","line":68,"markdown":"The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.phi","paper":"long","section":"1. The SK model and the Parisi variational problem","start_byte":4270,"tex":"\\Phi_\\gamma(t,x)","words":"Lean defines this as a supremum of expected payoffs over admissible controls; the short paper presents it through the backward equation."},{"anchor":"model","at":"b5019-5106","end_byte":5106,"end_line":72,"heading_id":"the-model-and-the-result","line":71,"markdown":"Write $u=\\Phi_x$ for the spatial gradient and\n$a=\\Phi_{xx}$ for its spatial derivative.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gradient","paper":"short","section":"1. The model and the result","start_byte":5019,"tex":"u(t,x)","words":"The spatial derivative of the value function."},{"anchor":"model","at":"b5019-5106","end_byte":5106,"end_line":75,"heading_id":"the-model-and-the-result","line":74,"markdown":"Write $u=\\Phi_x$ for the spatial gradient and\n$a=\\Phi_{xx}$ for its spatial derivative.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.curvature","paper":"short","section":"1. The model and the result","start_byte":5019,"tex":"a(t,x)","words":"The second spatial derivative of the value function, written as the derivative of the gradient."},{"anchor":"model","at":"b4044-4544","end_byte":4544,"end_line":78,"heading_id":"the-model-and-the-result","line":77,"markdown":"The Parisi functional $\\mathcal P(\\gamma)$ is defined through an\nauxiliary value function $\\Phi_\\gamma(t,x)$, in which $x\\in\\mathbb R$\nis a spatial variable and the overlap parameter $t$ plays the role of\ntime. Write $\\Phi=\\Phi_\\gamma$ and use subscripts for partial\nderivatives. The backward equation, terminal condition and functional are\n$$\n\\Phi_t=-\\tfrac12\\bigl(\\Phi_{xx}+\\gamma(t)\\Phi_x^2\\bigr),\n\\qquad \\Phi(1,x)=|x|,\n\\qquad\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)-\\tfrac12\\int_0^1t\\gamma(t)\\,dt.\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.parisi","paper":"short","section":"1. The model and the result","start_byte":4044,"tex":"\\mathcal P(\\gamma)","words":"The Parisi functional in this separate model, computed on the chosen Brownian space W."},{"anchor":"main-result","at":"b6367-6420","end_byte":6420,"end_line":81,"heading_id":"the-model-and-the-result","line":80,"markdown":"If $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","paper":"short","section":"1. The model and the result","start_byte":6367,"tex":null,"words":"The coefficient minimises the Parisi functional over all order parameters, with W held fixed throughout."},{"anchor":"model","at":"b5107-5300","end_byte":5300,"end_line":88,"heading_id":"the-model-and-the-result","line":83,"markdown":"Let $B$ be standard Brownian\nmotion, and define the diffusion $X$ and its self-consistency function\n$q$ by\n$$\ndX_t=\\gamma(t)u(t,X_t)\\,dt+dB_t,\\qquad X_0=0,\n\\qquad q(t)=\\mathbb E[u(t,X_t)^2].\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","paper":"short","section":"1. The model and the result","start_byte":5107,"tex":"X_t","words":"An adapted, almost surely continuous solution of the integrated equation on [0,1]; here X runs on real time."},{"anchor":"model","at":"b1969-2183","end_byte":2183,"end_line":90,"heading_id":"the-model-and-the-result","line":90,"markdown":"For independent standard Gaussian couplings $J_{ij}$, the pure,\nzero-field [Sherrington–Kirkpatrick (SK) Hamiltonian][SK] is\n$$\nH_n(\\sigma)=n^{-1/2}\\sum_{i<j}J_{ij}\\sigma_i\\sigma_j,\n\\qquad \\sigma\\in\\{-1,1\\}^n.\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.spin","paper":"short","section":"1. The model and the result","start_byte":1969,"tex":"\\sigma_i","words":"Encodes a Boolean as the spin value plus one or minus one."},{"anchor":"model","at":"b1969-2183","end_byte":2183,"end_line":95,"heading_id":"the-model-and-the-result","line":92,"markdown":"For independent standard Gaussian couplings $J_{ij}$, the pure,\nzero-field [Sherrington–Kirkpatrick (SK) Hamiltonian][SK] is\n$$\nH_n(\\sigma)=n^{-1/2}\\sum_{i<j}J_{ij}\\sigma_i\\sigma_j,\n\\qquad \\sigma\\in\\{-1,1\\}^n.\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.hamiltonian","paper":"short","section":"1. The model and the result","start_byte":1969,"tex":"H_n(\\sigma)","words":"The SK Hamiltonian on n spins, summing over pairs i<j, with an n-by-n coupling array of which those pairs are used."},{"anchor":"model","at":"b2184-2330","end_byte":2330,"end_line":98,"heading_id":"the-model-and-the-result","line":97,"markdown":"With the maximisation convention, its limiting ground-state energy per\nspin is\n$$\nP_*:=\\lim_{n\\to\\infty}n^{-1}\\mathbb E\\max_\\sigma H_n(\\sigma).\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateSequence","paper":"short","section":"1. The model and the result","start_byte":2184,"tex":"n^{-1}\\mathbb E\\max_\\sigma H_n(\\sigma)","words":"The expected maximum energy per spin at size n."},{"anchor":"model","at":"b2184-2330","end_byte":2330,"end_line":100,"heading_id":"the-model-and-the-result","line":100,"markdown":"With the maximisation convention, its limiting ground-state energy per\nspin is\n$$\nP_*:=\\lim_{n\\to\\infty}n^{-1}\\mathbb E\\max_\\sigma H_n(\\sigma).\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateValue","paper":"short","section":"1. The model and the result","start_byte":2184,"tex":"P_*","words":"The limit of the finite-size sequence; the Lean limit is meaningful once convergence is proved, which the scalar conclusion does."},{"anchor":"scalar-value","at":"b36389-36476","end_byte":36476,"end_line":104,"heading_id":"a-martingale-formula-for-the-ground-state-value","line":102,"markdown":"$M_t=u(t,X_t)$, the bounded gradient martingale, which satisfies\n$dM_t=a(t,X_t)\\,dB_t$.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gradientStopped","paper":"short","section":"5.1 A martingale formula for the ground-state value","start_byte":36389,"tex":"M_t","words":"The gradient martingale before time one, set equal to its terminal limit U from time one onward."},{"anchor":"scalar-value","at":"b36855-37016","end_byte":37016,"end_line":118,"heading_id":"a-martingale-formula-for-the-ground-state-value","line":106,"markdown":"Let $U$ be the almost-sure and\n$L^2$ limit of $M_t$. The value identity is\n$$\nP_*=\\mathbb E[UB_1]\n   =\\int_0^1\\mathbb E[a(t,X_t)]\\,dt.                \\tag{10}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","paper":"short","section":"5.1 A martingale formula for the ground-state value","start_byte":36855,"tex":null,"words":"Existence of the ground-state limit, the martingale with terminal spin U, and the value identities for the given minimiser and diffusion."},{"anchor":"qaoa-boundary","at":"b42353-42652","end_byte":42652,"end_line":122,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":120,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.eulerPath","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":"Y_j","words":"The Euler path built from this model's coefficient and gradient."},{"anchor":"qaoa-boundary","at":"b42353-42652","end_byte":42652,"end_line":126,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":124,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.rawCoefficient","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":"a(t_j,Y_j)","words":"The curvature along the Euler path at mesh point j, before normalisation."},{"anchor":"gaussian","at":"b93513-93593","end_byte":93593,"end_line":130,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","line":128,"markdown":"c_j=(\\mathbb E a(t_{j-1},Y_{j-1})^2)^{-1/2},\\qquad\ng_j=c_ja(t_{j-1},Y_{j-1})Z_j.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.normalizer","paper":"long","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93513,"tex":"c_{j+1}","words":"The inverse root second moment of the raw coefficient; Lean counts from zero, so the paper's index is j+1."},{"anchor":"qaoa-boundary","at":"b42353-42652","end_byte":42652,"end_line":134,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":132,"markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gaussianIncrement","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353,"tex":"g_{j+1}","words":"The normalised coefficient times the next Gaussian coordinate; Lean index j is the paper's index j+1."},{"anchor":"qaoa-boundary","at":"b42912-43123","end_byte":43123,"end_line":139,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":136,"markdown":"With $\\Phi_{\\rm G}$ the standard normal distribution function, set\n$$\n\\begin{aligned}\nF_N&=\\operatorname{sgn}\\{u(T,Y_N)+2\\Phi_{\\rm G}(Z_{N+1})-1\\},\\\\\n\\mathbb E[F_N\\mid Z_1,\\ldots,Z_N]&=u(T,Y_N).\n\\end{aligned}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42912,"tex":"F_N","words":"The finite spin: the sign of the terminal gradient plus uniform rounding noise from the last Gaussian coordinate."},{"anchor":"gaussian","at":"b94403-94465","end_byte":94465,"end_line":145,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","line":141,"markdown":"A_j=\\mathbb E[F_Ng_j],\\quad B_i=\\mathbb E[F_NZ_i],\\quad A_0=0.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientA","paper":"long","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94403,"tex":"A_j","words":"The correlation of the finite spin with the normalised increment, set to zero at j=0."},{"anchor":"gaussian","at":"b94403-94465","end_byte":94465,"end_line":149,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","line":147,"markdown":"A_j=\\mathbb E[F_Ng_j],\\quad B_i=\\mathbb E[F_NZ_i],\\quad A_0=0.","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientB","paper":"long","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94403,"tex":"B_i","words":"The correlation of the finite spin with the Gaussian coordinate of paper index i."},{"anchor":"qaoa-boundary","at":"b43303-43590","end_byte":43590,"end_line":153,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":151,"markdown":"The circuit construction\nneeds one finite quantity, the adjacent pairing\n$$\n\\begin{aligned}\nS_{T,N}&:=\\sum_{j=1}^N\n \\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}],\\\\\n\\lim_{N\\to\\infty}S_{T,N}\n&=I_T:=\\int_0^T\\mathbb E[a(t,X_t)]\\,dt.\n\\end{aligned}                                      \\tag{11}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43303,"tex":"S_{T,N}","words":"The adjacent pairing, summing A at j times B at j+1 for j from zero to N."},{"anchor":"qaoa-boundary","at":"b43303-43590","end_byte":43590,"end_line":161,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","line":155,"markdown":"The circuit construction\nneeds one finite quantity, the adjacent pairing\n$$\n\\begin{aligned}\nS_{T,N}&:=\\sum_{j=1}^N\n \\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}],\\\\\n\\lim_{N\\to\\infty}S_{T,N}\n&=I_T:=\\int_0^T\\mathbb E[a(t,X_t)]\\,dt.\n\\end{aligned}                                      \\tag{11}\n$$","module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","paper":"short","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43303,"tex":null,"words":"Normalisers are eventually positive, the pairing tends to the cutoff integral, and that integral tends to the ground-state value."},{"anchor":"evidence-scope","at":"b53118-53437","end_byte":53437,"end_line":171,"heading_id":"formal-verification","line":163,"markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses, and its third\nclause identifies the ground-state and Parisi values from a minimiser,\nconstructing the diffusion internally.","module":"ComparatorChallenges.SKValue","name":"OAI.SKValue.value_consequences","paper":"short","section":"Formal verification","start_byte":53118,"tex":null,"words":"The scalar statement: three clauses giving value identities, finite Gaussian convergence, and equality of ground-state and Parisi values."},{"anchor":"formal-scalar","at":"b243180-243450","end_byte":243450,"end_line":15,"heading_id":"the-separate-scalar-statement","line":13,"markdown":"The second formal statement has three clauses. Given a minimizer and an\nassociated diffusion, it establishes convergence of the finite-spin\nground-state values, a bounded martingale with a spin-valued terminal\nlimit, convergence of the gradient and the value identities.","module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","paper":"long","section":"12.2 The separate scalar statement","start_byte":243180,"tex":null,"words":"For a minimiser and any diffusion, the scalar value conclusion holds."},{"anchor":"formal-scalar","at":"b243451-243599","end_byte":243599,"end_line":20,"heading_id":"the-separate-scalar-statement","line":17,"markdown":"Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.","module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","paper":"long","section":"12.2 The separate scalar statement","start_byte":243451,"tex":null,"words":"For a minimiser, any diffusion and a cutoff T strictly between zero and one, the finite Gaussian conclusion holds."},{"anchor":"formal-parisi-clause","at":"b245775-245893","end_byte":245893,"end_line":27,"heading_id":"the-separate-scalar-statement","line":25,"markdown":"The third declaration, `ground_state_parisi_target`, assumes a minimizer\nalone and concludes $P_*=\\mathcal P(\\gamma)$.","module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","paper":"long","section":"12.2 The separate scalar statement","start_byte":245775,"tex":"P_*=\\mathcal P(\\gamma)","words":"The ground-state value equals the Parisi functional at a minimiser; no diffusion hypothesis is needed."}],"names":{"OAI.SKValue.Admissible":"α is an admissible control","OAI.SKValue.BrownianSpace":"Bundled space with Brownian motion","OAI.SKValue.BrownianSpace.B":"The Brownian path","OAI.SKValue.BrownianSpace.brownian":"B is standard Brownian motion","OAI.SKValue.BrownianSpace.filtration":"The natural Brownian filtration","OAI.SKValue.BrownianSpace.μ":"The probability measure","OAI.SKValue.Certificate.integrated_coercivity":"Integrated coercivity in the value proof","OAI.SKValue.FiniteGaussianConclusion":"The finite-Gaussian conclusion","OAI.SKValue.GradientStrip":"Gradient bounds on a fixed strip","OAI.SKValue.IsDiffusion":"X is a diffusion for γ","OAI.SKValue.IsDiffusion.L1_terminal":"Xₜ → X₁ in L¹","OAI.SKValue.IsDiffusion.contactPotential_hasDerivAt":"G′(s) = s − q(s)","OAI.SKValue.IsDiffusion.curvatureExpr_hasDerivAt":"Time derivative of the curvature moment","OAI.SKValue.IsDiffusion.curvatureSlope_hasDerivAt":"Time derivative of the curvature slope","OAI.SKValue.IsDiffusion.curvature_convexOn":"Curvature moment convex on plateaus","OAI.SKValue.IsDiffusion.curvature_generator_nonneg":"Coercive moment is nonnegative","OAI.SKValue.IsDiffusion.curvature_mean_integrable":"Mean curvature is integrable in time","OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact":"Contact gives unit curvature moment","OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip":"Finite-Gaussian limit on a fixed strip","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact":"Finite-Gaussian limit from contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact":"Ground state equals Parisi, given contact","OAI.SKValue.IsDiffusion.jetMoment_continuous":"Jet moments are continuous","OAI.SKValue.IsDiffusion.jetMoment_hasDerivAt":"Time derivative of a jet moment","OAI.SKValue.IsDiffusion.literal_coercivity":"Coercivity along the value-proof diffusion","OAI.SKValue.IsDiffusion.moment_convexOn_zero":"Second moment convex on zero plateaus","OAI.SKValue.IsDiffusion.parisi_first_value":"Parisi value from E|X₁|","OAI.SKValue.IsDiffusion.parisi_value_assembly":"Terminal formulas at the Parisi value","OAI.SKValue.IsDiffusion.polyJet_generator":"Itô formula for polynomial jets","OAI.SKValue.IsDiffusion.scalar_value_of_contact":"Scalar value from contact","OAI.SKValue.IsDiffusion.strip_value_le_groundState":"Strip integral below ground state","OAI.SKValue.IsDiffusion.terminal_brownian_covariance":"Covariance of U with Brownian motion","OAI.SKValue.IsDiffusion.terminal_compact_product":"Expected U X at time T","OAI.SKValue.IsDiffusion.terminal_gradient_product":"U X₁ = |X₁| almost surely","OAI.SKValue.IsDiffusion.terminal_product_value":"Expected product of U and X₁","OAI.SKValue.IsDiffusion.unitMomentMartingale":"The gradient is a unit-moment martingale","OAI.SKValue.IsDiffusion.value_expectation":"Mean of Φ along the diffusion","OAI.SKValue.IsMinimizer":"γ minimises the Parisi functional","OAI.SKValue.IsMinimizer.coeff_constant_of_contactPositive":"γ constant where G is positive","OAI.SKValue.IsMinimizer.coeff_eq_of_contactPositive":"γ equal where G is positive","OAI.SKValue.IsMinimizer.contactPotential_nonneg":"G is nonnegative","OAI.SKValue.IsMinimizer.contactPotential_zero":"G vanishes on (0,1)","OAI.SKValue.IsMinimizer.contact_origin_of_positive":"G(0) = 0 if γ(0)>0","OAI.SKValue.IsMinimizer.gradient_moment_contact":"Contact: q(t) = t on [0,1)","OAI.SKValue.IsMinimizer.ground_state_parisi_of_diffusion":"Ground state equals Parisi, given diffusion","OAI.SKValue.IsMinimizer.moment_eq_at_contact":"q(s) = s at interior contact","OAI.SKValue.JetExpr.curvatureExpr_sourceP":"Source of a², γ part","OAI.SKValue.JetExpr.curvatureSlope_sourceA":"Slope source, part without γ","OAI.SKValue.JetExpr.curvatureSlope_sourceP":"Slope source, γ part","OAI.SKValue.JetExpr.sourceA":"Generator source, part without γ","OAI.SKValue.JetExpr.sourceP":"Generator source, part multiplying γ","OAI.SKValue.OrderParameter":"Admissible order parameter on ℝ","OAI.SKValue.OrderParameter.coeff":"The coefficient γ on ℝ","OAI.SKValue.OrderParameter.integrable":"γ is integrable on (0,1]","OAI.SKValue.OrderParameter.monotone":"γ is nondecreasing on [0,1)","OAI.SKValue.OrderParameter.nonneg":"γ is nonnegative on [0,1)","OAI.SKValue.OrderParameter.origin_eq_of_constant":"γ(0) = c on initial plateaus","OAI.SKValue.OrderParameter.rightContinuous":"γ is right-continuous on [0,1)","OAI.SKValue.OrderParameter.zero":"The zero order parameter","OAI.SKValue.ScalarValueConclusion":"The scalar-value conclusion","OAI.SKValue.SourceStripRegularity":"The strip regularity bounds","OAI.SKValue.UnitMomentMartingale":"Martingale with second moment t","OAI.SKValue.UnitMomentMartingale.exists_continuous_terminal":"Terminal limit of a continuous martingale","OAI.SKValue.UnitMomentMartingale.exists_sequence_terminal":"Terminal limit along a sequence","OAI.SKValue.UnitMomentMartingale.integral_past_mul":"E[fₛ fₜ] = s","OAI.SKValue.UnitMomentMartingale.terminal_condExp":"Conditional expectation of the terminal U","OAI.SKValue.UnitMomentMartingale.terminal_mean_square":"L² convergence to U","OAI.SKValue.UnitMomentMartingale.terminal_square":"U² = 1 almost surely","OAI.SKValue.UnitMomentMartingale.weighted_terminal_expectation":"Weighted terminal expectation","OAI.SKValue.coefficientA":"Correlation A with normalised increments","OAI.SKValue.coefficientB":"Correlation B with Gaussian coordinates","OAI.SKValue.contactPotential":"The potential G","OAI.SKValue.controlPayoff":"Expected control payoff","OAI.SKValue.coordinate":"Coordinate of a Gaussian vector","OAI.SKValue.coupled_euler_L2_convergence":"L² convergence, coupled Euler scheme","OAI.SKValue.curvature":"The curvature a = ∂ₓₓΦ","OAI.SKValue.diffusionProcess_isDiffusion":"The constructed process solves the equation","OAI.SKValue.diffusion_gradient_brownian_covariance":"Gradient covariance with Brownian motion","OAI.SKValue.diffusion_gradient_isometry":"Itô isometry for the gradient","OAI.SKValue.diffusion_gradient_stopped_martingale":"Stopped gradient is a martingale","OAI.SKValue.driven_euler_L2_convergence":"L² convergence, driven Euler scheme","OAI.SKValue.endpoint_condition":"Endpoint condition on a hump","OAI.SKValue.euler":"Euler recursion on a uniform mesh","OAI.SKValue.eulerPath":"Euler path of the diffusion","OAI.SKValue.exists_diffusion":"A diffusion exists","OAI.SKValue.expectedMaximum_le_parisi":"Finite-spin maximum below n times Parisi","OAI.SKValue.finiteGaussian_from_diffusion":"Finite-Gaussian limit from a diffusion","OAI.SKValue.finiteGaussian_from_strip_and_moments":"Finite-Gaussian limit from strip and moments","OAI.SKValue.finiteRawMartingale":"Raw martingale of the finite scheme","OAI.SKValue.finiteRawMartingale_coordinate_coefficient":"Coefficient on a coordinate","OAI.SKValue.finiteRawMartingale_normalized_coefficient":"Coefficient on a normalised increment","OAI.SKValue.finite_coefficient_L2_errors":"L² error of the rounded coefficients","OAI.SKValue.finite_gaussian_target":"Finite-Gaussian limit at a minimiser","OAI.SKValue.finite_rounding_coefficient":"Rounding preserves correlations","OAI.SKValue.finite_rounding_signed_spin":"Rounded sign is a Boolean spin","OAI.SKValue.gaussianIncrement":"Normalised Gaussian increment","OAI.SKValue.gaussianProduct":"Product of standard Gaussians","OAI.SKValue.gaussian_rounding_mean":"Gaussian rounding keeps the mean","OAI.SKValue.gradient":"The gradient u = ∂ₓΦ","OAI.SKValue.gradientMeshError":"Gradient error on the mesh","OAI.SKValue.gradientMeshError_bound":"Explicit gradient mesh error bound","OAI.SKValue.gradientMeshError_bound_linear":"Mesh error linear in the step","OAI.SKValue.gradientMeshError_tendsto_zero":"Gradient mesh error tends to zero","OAI.SKValue.gradientStopped":"Gradient martingale with terminal limit U","OAI.SKValue.gradient_bound":"The gradient is bounded by one","OAI.SKValue.gradient_terminal_tendsto":"Gradient tends to sign x","OAI.SKValue.groundStateSequence":"Expected maximum energy per spin","OAI.SKValue.groundStateValue":"The SK ground state value","OAI.SKValue.groundStateValue_le_parisi":"Ground state below Parisi","OAI.SKValue.ground_state_parisi_target":"Ground state equals Parisi at minimisers","OAI.SKValue.hamiltonian":"SK Hamiltonian on n spins","OAI.SKValue.jetMoment":"Moment of a polynomial in jets","OAI.SKValue.meshMaxError":"Maximum error on the mesh","OAI.SKValue.meshTime":"Mesh time jT/N","OAI.SKValue.mesh_coefficient_errors":"L² errors of the mesh coefficients","OAI.SKValue.mesh_coefficients_norms":"Norm bounds on the mesh coefficients","OAI.SKValue.mesh_leading_norm":"Norm bound on the leading coefficients","OAI.SKValue.mesh_raw_depends":"Raw coefficient depends on past only","OAI.SKValue.monotone_grid_sum_tendsto":"Monotone Riemann sums converge","OAI.SKValue.normalizedIncrement_mean":"Normalised increment has mean zero","OAI.SKValue.normalizedIncrements_orthogonal":"Normalised increments are orthogonal","OAI.SKValue.normalized_family_orthonormal":"Normalised increments are orthonormal","OAI.SKValue.normalized_shifted_sum_le_groundState":"Normalised shifted sum below ground state","OAI.SKValue.normalizer":"Normaliser of the raw coefficient","OAI.SKValue.not_three_slope_pattern":"No three-slope pattern for convex f′","OAI.SKValue.parisi":"The Parisi functional","OAI.SKValue.perturbed_shifted_grid_tendsto":"Perturbed Riemann sums converge","OAI.SKValue.phi":"The value function Φ","OAI.SKValue.phi_terminal":"Terminal value Φ(1,x) = |x|","OAI.SKValue.polyJet":"Polynomial in jets of Φ","OAI.SKValue.positive_gap_endpoints":"Endpoints of G's positive gap","OAI.SKValue.positive_hump_impossible":"No positive hump of G","OAI.SKValue.positive_right_hump_impossible":"No positive hump on the right","OAI.SKValue.rawCoefficient":"Curvature along the Euler path","OAI.SKValue.roundedSign":"The rounded finite spin","OAI.SKValue.scalar_value_target":"Scalar value at a minimiser","OAI.SKValue.shiftedCoefficientSum":"Adjacent pairing of A and B","OAI.SKValue.shiftedCoefficientSum_le_groundState":"Shifted coefficient sum below ground state","OAI.SKValue.shiftedFiltration":"Filtration of increments after t","OAI.SKValue.shiftedPairing_tendsto_error_zero":"Shifted pairing error tends to zero","OAI.SKValue.sourceStripRegularity":"Strip bounds hold for every γ","OAI.SKValue.spatialJet":"Scaled spatial jet c ∂ₓⁿψ′","OAI.SKValue.spin":"Boolean spin ±1","OAI.SKValue.standardGaussian":"The standard Gaussian law","OAI.SKValue.standardGaussian_cdf_hasLaw_uniform":"Gaussian CDF transform is uniform","OAI.SKValue.stepSize":"Mesh width T/N","OAI.SKValue.uniform_rounding_mean":"Uniform rounding keeps the mean","OAI.SKValue.value_consequences":"The scalar-value theorem","OAI.SKValueG.adaptiveOffset":"Offsets after adaptive columns","OAI.SKValueG.adaptiveResidual":"Residual vectors after adaptive columns","OAI.SKValueG.adaptive_spin_lower":"Adaptive spin lower bound","OAI.SKValueG.adaptive_value_identity":"Adaptive column identity","OAI.SKValueG.columnLinear_bound":"Bound on the column's linear term","OAI.SKValueG.columnResidual":"Residual after one Gaussian column","OAI.SKValueG.columnResidual_norm_le":"Column residual has smaller norm","OAI.SKValueG.columnSplitCoeff_distance":"Column split preserves distances","OAI.SKValueG.columnSplitCoeff_inner":"Column split preserves inner products","OAI.SKValueG.columnVector":"Column vector in the Gaussian split","OAI.SKValueG.column_revealed_energy":"Column vector recursion for revealed energy","OAI.SKValueG.constant_limit_integral_le":"Fatou bound for a constant limit","OAI.SKValueG.empiricalColumnEnergy":"Empirical column energy","OAI.SKValueG.empiricalColumnEnergy_ae_tendsto":"Empirical energy limit","OAI.SKValueG.empiricalVector":"Empirical vector of samples","OAI.SKValueG.empirical_finite_spin_lower":"Empirical energy below the finite-spin value","OAI.SKValueG.empirical_inner_tendsto":"Empirical inner products converge","OAI.SKValueG.finite_adaptive_SK_lower":"Finite adaptive lower bound","OAI.SKValueG.gaussianProduct":"Product of standard Gaussians","OAI.SKValueG.gramSchmidtNormed_approx":"Gram–Schmidt of nearly orthonormal vectors","OAI.SKValueG.groundStateSequence":"Expected maximum energy per spin","OAI.SKValueG.groundStateSequence_tendsto":"Finite-spin value converges","OAI.SKValueG.groundStateValue":"The SK ground state value","OAI.SKValueG.hamiltonian":"SK Hamiltonian on n spins","OAI.SKValueG.normalizedOffset_rows":"Normalised offset as column energies","OAI.SKValueG.normalized_column_energy_tendsto":"Limit of the normalised column energy","OAI.SKValueG.orthogonalRemainder":"Remainder after orthogonal projections","OAI.SKValueG.predictableDirection":"Predictable direction from the history","OAI.SKValueG.predictable_coefficients_le_groundState":"Predictable correlations below ground state","OAI.SKValueG.predictable_coordinate_mean_zero":"Predictable times later coordinate: mean 0","OAI.SKValueG.quadraticCoeff":"Quadratic coefficient wᵢ wⱼ/√2","OAI.SKValueG.quadraticValue":"Expected maximum of the quadratic process","OAI.SKValueG.quadraticValue_column":"One Gaussian column split","OAI.SKValueG.quadraticValue_ge_offset":"Quadratic value dominates each offset","OAI.SKValueG.quadraticValue_spin":"Quadratic value of the SK spins","OAI.SKValueG.random_column_identity":"Gaussian column split with random direction","OAI.SKValueG.spin":"Boolean spin ±1","OAI.SKValueG.standardGaussian":"The standard Gaussian law","OAI.SKValueG.transposeGaussian_preserving":"Gaussian matrix transpose preserves law","OAI.ZeroTemperatureSK.BrownianSystem":"Probability space with Brownian motion","OAI.ZeroTemperatureSK.BrownianSystem.B":"The Brownian path","OAI.ZeroTemperatureSK.BrownianSystem.abs_increment":"Mean absolute Brownian increment","OAI.ZeroTemperatureSK.BrownianSystem.brownian":"B is standard Brownian motion","OAI.ZeroTemperatureSK.BrownianSystem.law":"The probability measure","OAI.ZeroTemperatureSK.Certificate.A":"Certificate constant A = 1/13","OAI.ZeroTemperatureSK.Certificate.DF₂":"Derivative of F₂ along the jets","OAI.ZeroTemperatureSK.Certificate.F₂":"Certificate polynomial F₂","OAI.ZeroTemperatureSK.Certificate.F₂b":"Partial derivative of F₂ in b","OAI.ZeroTemperatureSK.Certificate.F₂j":"Partial derivative of F₂ in j","OAI.ZeroTemperatureSK.Certificate.P₀":"Certificate term P₀","OAI.ZeroTemperatureSK.Certificate.Q":"Certificate polynomial Q","OAI.ZeroTemperatureSK.Certificate.T":"Certificate polynomial T","OAI.ZeroTemperatureSK.Certificate.WF₂":"Weighted divergence of F₂","OAI.ZeroTemperatureSK.Certificate.coercivity_constant":"The constant exceeds 1/1000","OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral":"From sharp constant to 1/1000","OAI.ZeroTemperatureSK.Certificate.hasDerivAt_flux":"Derivative of the certificate flux","OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux":"Integrated coercivity from vanishing flux","OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity":"Pointwise coercivity inequality","OAI.ZeroTemperatureSK.Certificate.rational_identity":"The rational certificate identity","OAI.ZeroTemperatureSK.Certificate.remainder":"Certificate remainder","OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound":"Certificate remainder is at least c·v⁴","OAI.ZeroTemperatureSK.Certificate.weightedSquares":"Weighted squares in the certificate","OAI.ZeroTemperatureSK.Control":"Admissible control bounded by one","OAI.ZeroTemperatureSK.Control.bound":"The control is bounded by one","OAI.ZeroTemperatureSK.Control.progressive":"The control is progressive","OAI.ZeroTemperatureSK.Control.val":"The control process","OAI.ZeroTemperatureSK.FiniteVerification.cascade_eq_control_sup":"Cascade equals the control supremum","OAI.ZeroTemperatureSK.FullSupportConclusion":"The full-support conclusion","OAI.ZeroTemperatureSK.Gaussian.gaussian_abs_moment":"First absolute Gaussian moment","OAI.ZeroTemperatureSK.Heat.BackwardShape":"The backward shape conditions","OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale":"Rescaling in (0,1] keeps shape","OAI.ZeroTemperatureSK.Heat.ForwardDatumShape.jump":"Forward datum shape survives jumps","OAI.ZeroTemperatureSK.Heat.ReverseBurgers":"Reverse-time Burgers solution on [0,H]","OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg":"Ratio sign on the closed interval","OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg_interior":"Ratio sign before the endpoint","OAI.ZeroTemperatureSK.Heat.SmoothBurgers.pos_halfLine":"Odd increasing r positive for x>0","OAI.ZeroTemperatureSK.Heat.SmoothBurgers.rate_nonpos":"Burgers flow preserves rate sign","OAI.ZeroTemperatureSK.Heat.SmoothBurgers.second_nonpos":"Burgers flow preserves half-line concavity","OAI.ZeroTemperatureSK.Heat.SmoothBurgers.wronskian_equation":"Parabolic equation for the backward Wronskian","OAI.ZeroTemperatureSK.Heat.SmoothBurgers.wronskian_nonneg":"Burgers flow preserves the Wronskian sign","OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse":"Forward score slope at least 1/t","OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos":"Forward flow preserves half-line concavity","OAI.ZeroTemperatureSK.Heat.backwardBoundary_shape":"Backward shape at grid times","OAI.ZeroTemperatureSK.Heat.backwardWronskian":"The backward Wronskian","OAI.ZeroTemperatureSK.Heat.boundaryRatio_nonneg":"Ratio sign at grid times","OAI.ZeroTemperatureSK.Heat.bounded_linear_nonpositive":"Half-line maximum principle","OAI.ZeroTemperatureSK.Heat.burgersRate":"Burgers rate ½∂ₓ²r + r∂ₓr","OAI.ZeroTemperatureSK.Heat.cascade":"Finite Parisi cascade","OAI.ZeroTemperatureSK.Heat.cascade_terminal_stability":"Cascade stable in terminal data","OAI.ZeroTemperatureSK.Heat.certB":"Certificate variable b = −w/r","OAI.ZeroTemperatureSK.Heat.certJ":"Certificate j = ∂ₓs − sv/r","OAI.ZeroTemperatureSK.Heat.certL":"Certificate variable L = −r ∂ₓ²s","OAI.ZeroTemperatureSK.Heat.certR":"Certificate variable R = r²","OAI.ZeroTemperatureSK.Heat.certV":"Certificate variable v = ∂ₓr","OAI.ZeroTemperatureSK.Heat.cert_flux_continuous":"The certificate flux is continuous","OAI.ZeroTemperatureSK.Heat.cert_flux_derivative":"Divergence identity for the flux","OAI.ZeroTemperatureSK.Heat.cert_halfLine_coercive":"Pointwise coercivity on the half-line","OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs":"Seven signs on the half-line","OAI.ZeroTemperatureSK.Heat.certificate_growth":"Exponential bounds on F₂ and divergence","OAI.ZeroTemperatureSK.Heat.datumScore":"Score of the forward datum","OAI.ZeroTemperatureSK.Heat.datumScore_jump":"Forward score after a height jump","OAI.ZeroTemperatureSK.Heat.density_tilted_pairing":"Tilted pairing against a Gaussian density","OAI.ZeroTemperatureSK.Heat.deriv_logSemigroup_tilted":"Heat step derivative is tilted mean","OAI.ZeroTemperatureSK.Heat.derivative_interpolation":"Landau derivative interpolation","OAI.ZeroTemperatureSK.Heat.finiteBurgersSource":"Burgers source of the finite scheme","OAI.ZeroTemperatureSK.Heat.finiteDensity":"Finite-step density","OAI.ZeroTemperatureSK.Heat.finiteDensity_first_formula":"Density at the first grid time","OAI.ZeroTemperatureSK.Heat.finiteDensity_step_pairing":"Density advances one grid step","OAI.ZeroTemperatureSK.Heat.finiteGradient_mild":"Mild form of the finite-step gradient","OAI.ZeroTemperatureSK.Heat.finiteGradient_uniform_smoothing":"Uniform smoothing of finite-step gradients","OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape":"Backward shape within each finite step","OAI.ZeroTemperatureSK.Heat.finite_density_coercivity":"Finite-step coercivity, density form","OAI.ZeroTemperatureSK.Heat.finite_forward_ratio":"r∂ₓs − s∂ₓr ≥ 0 stepwise","OAI.ZeroTemperatureSK.Heat.forwardBoundary":"Forward factor at grid times","OAI.ZeroTemperatureSK.Heat.forwardScore":"The forward score s","OAI.ZeroTemperatureSK.Heat.forwardScore_odd":"The forward score is odd","OAI.ZeroTemperatureSK.Heat.forwardSegment_ratio":"Ratio sign propagates across one step","OAI.ZeroTemperatureSK.Heat.gaussianLogDensity_derivative":"Derivative of the Gaussian-weighted density","OAI.ZeroTemperatureSK.Heat.gaussian_expBound_tendsto_zero":"Gaussian decay beats exponential growth","OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity":"Coercivity for Gaussian-weighted densities","OAI.ZeroTemperatureSK.Heat.iteratedDeriv_finiteBurgersSource_bound":"Derivative bounds on the Burgers source","OAI.ZeroTemperatureSK.Heat.logSemigroup":"Log-exponential heat step","OAI.ZeroTemperatureSK.Heat.logSemigroup_curvature_pos":"Heat step preserves strict convexity","OAI.ZeroTemperatureSK.Heat.logSemigroup_even":"Heat step preserves evenness","OAI.ZeroTemperatureSK.Heat.logSemigroup_lipschitz":"Heat step preserves Lipschitz bounds","OAI.ZeroTemperatureSK.Heat.logarithmic_heat_equation":"Hopf–Cole equation for the log-heat flow","OAI.ZeroTemperatureSK.Heat.ratioWronskian":"The ratio Wronskian","OAI.ZeroTemperatureSK.Heat.ratioWronskian_equation":"Parabolic equation for the ratio Wronskian","OAI.ZeroTemperatureSK.Heat.ratioWronskian_first_jump":"Ratio Wronskian sign, first jump","OAI.ZeroTemperatureSK.Heat.ratioWronskian_jump":"Ratio Wronskian across a height jump","OAI.ZeroTemperatureSK.Heat.scaled_rate":"Burgers rate of a rescaled field","OAI.ZeroTemperatureSK.Heat.scaled_wronskian":"Wronskian of a rescaled field","OAI.ZeroTemperatureSK.Heat.smooth_limit_of_uniform_derivative_bounds":"Smooth limit under uniform derivative bounds","OAI.ZeroTemperatureSK.Heat.softAbs":"Soft absolute value log(2 cosh Mx)/M","OAI.ZeroTemperatureSK.Heat.softAbs_backward_shape":"Soft |x| has the backward shape","OAI.ZeroTemperatureSK.Heat.softAbs_bounds":"Soft absolute value near |x|","OAI.ZeroTemperatureSK.Heat.tangent_bound_of_concave_odd":"Tangent bound for concave odd r","OAI.ZeroTemperatureSK.Heat.tendstoUniformly_deriv_of_bound":"Uniform convergence of first derivatives","OAI.ZeroTemperatureSK.Heat.tendstoUniformly_deriv_zero":"Uniform null limits have null derivatives","OAI.ZeroTemperatureSK.Heat.uniform_gradient_bound_of_mild":"Uniform gradient bound, mild equation","OAI.ZeroTemperatureSK.Heat.varianceHeat":"Heat semigroup of variance t","OAI.ZeroTemperatureSK.Heat.varianceHeat_exp_eq_exp_log":"Hopf–Cole identity for the heat flow","OAI.ZeroTemperatureSK.Heat.varianceHeat_gradient_bound":"Heat gradient bound of order 1/√h","OAI.ZeroTemperatureSK.Heat.varianceTilted_lower_window":"Tilted kernel lower bound, window","OAI.ZeroTemperatureSK.IsDiffusion":"X is a diffusion for γ","OAI.ZeroTemperatureSK.IsMinimizer":"γ minimises the Parisi functional","OAI.ZeroTemperatureSK.IsStieltjesMeasure":"μ is γ's Stieltjes measure","OAI.ZeroTemperatureSK.Nonuniform.euler":"Euler scheme on a variable mesh","OAI.ZeroTemperatureSK.OrderParameter":"Admissible order parameter γ","OAI.ZeroTemperatureSK.OrderParameter.addCDF":"Order parameter plus a CDF","OAI.ZeroTemperatureSK.OrderParameter.addCDF_difference":"Difference after adding a CDF","OAI.ZeroTemperatureSK.OrderParameter.chord":"Chord between two order parameters","OAI.ZeroTemperatureSK.OrderParameter.integrable":"γ is integrable on [0,1)","OAI.ZeroTemperatureSK.OrderParameter.local_bound":"γ bounded by its integral","OAI.ZeroTemperatureSK.OrderParameter.monotone":"γ is nondecreasing","OAI.ZeroTemperatureSK.OrderParameter.nonneg":"γ is nonnegative","OAI.ZeroTemperatureSK.OrderParameter.removeCDF":"Order parameter minus a dominated CDF","OAI.ZeroTemperatureSK.OrderParameter.removeCDF_difference":"Difference after removing a CDF","OAI.ZeroTemperatureSK.OrderParameter.right_continuous":"γ is right-continuous","OAI.ZeroTemperatureSK.OrderParameter.val":"The function γ","OAI.ZeroTemperatureSK.Parabolic.exponential_lower_barrier":"Exponential lower barrier","OAI.ZeroTemperatureSK.Parabolic.nonnegative_fullLine_bounded":"Minimum principle on the whole line","OAI.ZeroTemperatureSK.Parabolic.nonnegative_halfLine_bounded":"Minimum principle on the half-line","OAI.ZeroTemperatureSK.Parabolic.nonnegative_of_parabolic_inequality":"Parabolic minimum principle on an interval","OAI.ZeroTemperatureSK.Smoothing.weighted_absorption":"Absorption in a √(b−t)-weighted norm","OAI.ZeroTemperatureSK.SupportDeterministic.InSupport":"t lies in supp μ","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure":"μ is γ's Stieltjes measure","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.cdf_add_Ioc":"γ(b) = γ(a) + μ((a,b])","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.eq_cdf_of_measure_Ioc_zero":"γ constant where μ is null","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.eq_cdf_of_support_free":"γ is constant off the support","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.finiteOnCompacts":"μ is finite on compact sets","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support":"Full support gives γ(a) < γ(b)","OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.strictMonoOn_of_full_support":"Full support gives strict increase","OAI.ZeroTemperatureSK.SupportDeterministic.Time":"The overlap interval [0,1)","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.contact_conditions":"Contact conditions on the support","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos":"Zero on support from nonpositive integrals","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min":"G's minimum gives slope ≤ 1","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_eq_one_of_full_contact":"Full contact gives q′ = 1","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature":"Full support from contact and curvature","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one":"No gap with slope below one","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative":"No gap with strictly convex q′","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap":"No bounded gap at positive height","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap":"No bounded gap at height zero","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.value_contact_of_potential_min":"G's interior minimum gives contact","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal":"G vanishes on support, abstract form","OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap":"γ constant on a support gap","OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap":"Missing point gives a support gap","OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau":"Support points beyond every time","OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap":"γ vanishes on initial support gaps","OAI.ZeroTemperatureSK.Time":"The overlap interval [0,1)","OAI.ZeroTemperatureSK.WeakIto.completion_square_feedback":"Completing the square at the feedback","OAI.ZeroTemperatureSK.WeakIto.constant_interval_verification_upper":"Verification upper bound for constant γ","OAI.ZeroTemperatureSK.WeakIto.exists_feedback_verification":"Near-optimal control from feedback","OAI.ZeroTemperatureSK.WeakIto.expected_verification_step":"One expected verification step","OAI.ZeroTemperatureSK.WeakIto.integrated_completion_square":"Integrated completion of the square","OAI.ZeroTemperatureSK.abs_le_value":"The value dominates |x|","OAI.ZeroTemperatureSK.actualJet":"Spatial derivatives along the diffusion","OAI.ZeroTemperatureSK.actualQuartic_pos":"Fourth curvature moment is positive","OAI.ZeroTemperatureSK.actualThirdSquare_pos":"Second moment of ∂ₓ³Φ is positive","OAI.ZeroTemperatureSK.actual_coercivity":"Spatial coercivity at fixed time","OAI.ZeroTemperatureSK.actual_jet_even_moment_pos":"Nonzero jets have positive even moments","OAI.ZeroTemperatureSK.approxError_limit":"Approximation error tends to zero","OAI.ZeroTemperatureSK.approxValue":"Approximate value of step n","OAI.ZeroTemperatureSK.approxValue_derivative_uniform":"Uniform derivative of the approximate value","OAI.ZeroTemperatureSK.approxValue_error":"Error bound for the approximate value","OAI.ZeroTemperatureSK.approxValue_uniform":"Approximate value converges uniformly","OAI.ZeroTemperatureSK.approxValue_uniform_derivative_bounds":"Uniform derivative bounds, approximate value","OAI.ZeroTemperatureSK.coercivePolynomial_scaled":"Coercive integrand in terms of jets","OAI.ZeroTemperatureSK.compactDrift_lipschitz":"Strip drift is Lipschitz in x","OAI.ZeroTemperatureSK.compactGradient_continuous":"Strip gradient is continuous","OAI.ZeroTemperatureSK.compactGradient_derivative_continuous":"Strip gradient derivatives are continuous","OAI.ZeroTemperatureSK.compactGradient_family":"Strip gradient is a smooth family","OAI.ZeroTemperatureSK.compactGradient_lipschitz":"Strip gradient is Lipschitz in x","OAI.ZeroTemperatureSK.compactGradient_stability":"Gradient is stable in the coefficient","OAI.ZeroTemperatureSK.compactJetField":"x-derivatives of u as time fields","OAI.ZeroTemperatureSK.compactJetField_generator_one":"Itô generator of the curvature a","OAI.ZeroTemperatureSK.compactJetField_generator_two":"Itô generator of ∂ₓa","OAI.ZeroTemperatureSK.compactMixedDrift":"β coefficient times the averaged gradient","OAI.ZeroTemperatureSK.compactMixedDrift_bound":"The mixed drift is bounded","OAI.ZeroTemperatureSK.compactMixedGradient":"Average of the two gradients","OAI.ZeroTemperatureSK.compactSquare_generator":"Itô generator of u²","OAI.ZeroTemperatureSK.compactValueRate_eq":"Formula for the value's time derivative","OAI.ZeroTemperatureSK.contactIntegrand_integrable":"q(t) − t is integrable","OAI.ZeroTemperatureSK.contactPotential":"The potential G","OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal":"∫G dμ ≤ 0 on compacts","OAI.ZeroTemperatureSK.contactPotential_continuousOn":"G is continuous on [0,1]","OAI.ZeroTemperatureSK.contactPotential_hasDerivAt":"G′(s) = s − q(s)","OAI.ZeroTemperatureSK.contactPotential_nonneg":"G is nonnegative","OAI.ZeroTemperatureSK.contactPotential_remove":"Removable mass has ∫G ≤ 0","OAI.ZeroTemperatureSK.controlPayoff":"Control payoff, sample by sample","OAI.ZeroTemperatureSK.controlPayoff_coeff_le":"One-sided payoff bound in the coefficient","OAI.ZeroTemperatureSK.controlPayoff_convex":"Samplewise payoff is convex in x","OAI.ZeroTemperatureSK.controlPayoff_le":"Upper bound on the payoff","OAI.ZeroTemperatureSK.controlPayoff_space_le":"One-sided spatial bound on the payoff","OAI.ZeroTemperatureSK.control_values_bddAbove":"Expected payoffs are bounded above","OAI.ZeroTemperatureSK.curvature":"The curvature a = ∂ₓₓΦ","OAI.ZeroTemperatureSK.curvatureField_generator":"Itô generator of a²","OAI.ZeroTemperatureSK.curvatureMoment":"The curvature moment E[a(t, Xₜ)²]","OAI.ZeroTemperatureSK.curvatureMoment_continuousOn":"Curvature moment continuous on [0,T]","OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit":"Curvature moment continuous on [0,1)","OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant":"Formula for q‴ on a plateau","OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant":"Formula for q″ on a plateau","OAI.ZeroTemperatureSK.curvature_nonneg":"The curvature is nonnegative","OAI.ZeroTemperatureSK.curvature_not_identically_zero":"The curvature is not identically zero","OAI.ZeroTemperatureSK.curvature_terminal_bound":"Curvature bound of order 1/√(1−t)","OAI.ZeroTemperatureSK.deriv_compactValueRate":"x-derivative of the value's time derivative","OAI.ZeroTemperatureSK.differenceField":"Value difference as a time field","OAI.ZeroTemperatureSK.differenceField_generator":"Itô generator of the value difference","OAI.ZeroTemperatureSK.diffusion":"The constructed diffusion","OAI.ZeroTemperatureSK.diffusion_interval_pos":"Xₜ charges every interval","OAI.ZeroTemperatureSK.diffusion_isDiffusion":"The constructed process solves the equation","OAI.ZeroTemperatureSK.diffusion_uniqueness":"Pathwise uniqueness of the diffusion","OAI.ZeroTemperatureSK.eulerState":"Euler state on a Gaussian sample","OAI.ZeroTemperatureSK.expectedSKMaximum":"Expected maximum energy per spin","OAI.ZeroTemperatureSK.expected_controlPayoff_convex":"Expected payoff is convex in x","OAI.ZeroTemperatureSK.expected_hasDerivWithinAt_right":"Right generator formula along a diffusion","OAI.ZeroTemperatureSK.expected_transition_tiltedMean":"Transition law as a tilted mean","OAI.ZeroTemperatureSK.extend":"Extension by zero to ℝ","OAI.ZeroTemperatureSK.finiteCDF":"CDF of a finite measure","OAI.ZeroTemperatureSK.finiteCDF_fubini":"Fubini for a finite measure's CDF","OAI.ZeroTemperatureSK.finiteCDF_integral":"CDF as an integral of indicators","OAI.ZeroTemperatureSK.finiteValue_control_error":"Error of the finite-step value","OAI.ZeroTemperatureSK.finite_density_identity":"Finite-step density is the grid law","OAI.ZeroTemperatureSK.finite_feedback_coercivity":"Finite-step coercivity along the diffusion","OAI.ZeroTemperatureSK.full_support":"Full support of SK minimisers","OAI.ZeroTemperatureSK.gradient":"The gradient u = ∂ₓΦ","OAI.ZeroTemperatureSK.gradient_abs_le_one":"The gradient is bounded by one","OAI.ZeroTemperatureSK.gradient_odd":"The gradient is odd in x","OAI.ZeroTemperatureSK.gradient_terminal_small":"Gradient small near zero, late times","OAI.ZeroTemperatureSK.gradient_zero":"Gradient vanishes at x = 0","OAI.ZeroTemperatureSK.groundStateValue":"The SK ground state value","OAI.ZeroTemperatureSK.hasDerivAt_of_right_on_Ioo":"From right derivatives to a derivative","OAI.ZeroTemperatureSK.incrementFiltration":"Filtration of increments after t","OAI.ZeroTemperatureSK.integral_pos_of_interval_support":"Positive mean when intervals are charged","OAI.ZeroTemperatureSK.meshTime":"Mesh time jT/N","OAI.ZeroTemperatureSK.minimizer_contact":"Contact and slope bounds","OAI.ZeroTemperatureSK.minimizer_full_support_actual":"Full support for the constructed diffusion","OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau":"No terminal plateau at a minimiser","OAI.ZeroTemperatureSK.minimizer_variational_inequality":"First-order inequality at a minimiser","OAI.ZeroTemperatureSK.mixedPath_error":"Coupling bound for the mixed path","OAI.ZeroTemperatureSK.mixedQ":"Second moment along the mixed diffusion","OAI.ZeroTemperatureSK.mixedQ_perturbation_limit":"Mixed second moment converges along chords","OAI.ZeroTemperatureSK.mixedQ_weighted_limit":"Weighted mixed moment converges","OAI.ZeroTemperatureSK.mixedStripDrift":"Mixed drift on a strip","OAI.ZeroTemperatureSK.mixed_difference_integral":"Value difference as a mixed integral","OAI.ZeroTemperatureSK.mixed_difference_right":"Right derivative of the mixed difference","OAI.ZeroTemperatureSK.mixed_difference_terminal_bound":"Terminal bound on the mixed difference","OAI.ZeroTemperatureSK.normalizedParisiDifference_eq":"Normalised Parisi difference formula","OAI.ZeroTemperatureSK.normalizedValueDifference_limit":"Limit of the normalised value difference","OAI.ZeroTemperatureSK.normalizedValueDifference_truncation":"Truncating the normalised value difference","OAI.ZeroTemperatureSK.parameterMeasure":"The Stieltjes measure of γ","OAI.ZeroTemperatureSK.parameterMeasure_isStieltjes":"The Stieltjes measure exists","OAI.ZeroTemperatureSK.parisi":"The Parisi functional","OAI.ZeroTemperatureSK.parisi_chord_derivative":"One-sided derivative along a chord","OAI.ZeroTemperatureSK.penalty_difference":"Penalty term along a chord","OAI.ZeroTemperatureSK.perturbation":"Chord perturbation of step n","OAI.ZeroTemperatureSK.perturbation_distance_limit":"Perturbations converge to γ","OAI.ZeroTemperatureSK.positiveCoeff":"Positive finite-step coefficient","OAI.ZeroTemperatureSK.positiveCoeff_sample_bound":"Sampled coefficient at most γ(t)+1","OAI.ZeroTemperatureSK.positiveCoeff_sample_eq":"Positive coefficient at the sampled index","OAI.ZeroTemperatureSK.positiveCoeff_sample_limit":"Sampled coefficient converges to γ(t)","OAI.ZeroTemperatureSK.positiveGamma_L1":"Positive step functions converge in L¹","OAI.ZeroTemperatureSK.positivePath_error":"Coupling bound for an approximating path","OAI.ZeroTemperatureSK.positivePath_sample_limit":"Approximating paths converge at sample times","OAI.ZeroTemperatureSK.positivePath_uniform_error":"Approximating paths converge uniformly","OAI.ZeroTemperatureSK.positiveValue_derivative_uniform":"Uniform derivative of the positive scheme","OAI.ZeroTemperatureSK.positiveValue_uniform_derivative_bounds":"Uniform derivative bounds, positive scheme","OAI.ZeroTemperatureSK.positive_gap_curvature":"q‴ positive on positive plateaus","OAI.ZeroTemperatureSK.quarticMoment_scaled":"Quartic integrand in terms of jets","OAI.ZeroTemperatureSK.removeCDF_monotone":"γ minus dominated CDF is monotone","OAI.ZeroTemperatureSK.sampleCoercive_integral_limit":"Limit of the sampled coercive moment","OAI.ZeroTemperatureSK.sampleJet_bound":"Uniform bound on sampled jets","OAI.ZeroTemperatureSK.sampleJet_limit":"Sampled jets converge almost surely","OAI.ZeroTemperatureSK.sampleQuartic_integral_limit":"Limit of the sampled quartic moment","OAI.ZeroTemperatureSK.sampleTime":"Grid time sampled from t","OAI.ZeroTemperatureSK.sample_coercivity":"Coercivity at sampled grid times","OAI.ZeroTemperatureSK.secondCurvatureField_generator":"Itô generator of (∂ₓa)² − 2ca³","OAI.ZeroTemperatureSK.spinDisorder":"Gaussian disorder on n spins","OAI.ZeroTemperatureSK.spinHamiltonian":"SK Hamiltonian on n spins","OAI.ZeroTemperatureSK.squareMoment":"Second moment of u along X","OAI.ZeroTemperatureSK.squareMoment_bounds":"The second moment lies in [0,1]","OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment":"q(t) integrates the curvature moment","OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt":"Derivative of the gradient moment","OAI.ZeroTemperatureSK.squareMoment_integrable":"The second moment is integrable","OAI.ZeroTemperatureSK.squareMoment_weighted_integrable":"Weighted second moment is integrable","OAI.ZeroTemperatureSK.squareMoment_zero":"Second moment vanishes at zero","OAI.ZeroTemperatureSK.stripA_continuous":"Strip curvature moment is continuous","OAI.ZeroTemperatureSK.stripA_hasDerivAt_on_constant":"Curvature moment derivative on a plateau","OAI.ZeroTemperatureSK.stripA_hasDerivWithinAt_right":"Right derivative of the curvature moment","OAI.ZeroTemperatureSK.stripDrift":"Bounded Lipschitz drift on a strip","OAI.ZeroTemperatureSK.stripQ":"Second moment on a strip","OAI.ZeroTemperatureSK.stripQ_eq_integral":"Strip moment as curvature integral","OAI.ZeroTemperatureSK.stripQ_hasDerivWithinAt_right":"Right derivative of strip moment","OAI.ZeroTemperatureSK.stripQ_zero":"Strip moment vanishes at zero","OAI.ZeroTemperatureSK.stripSecond_eq_actual":"The q″ expression as an expectation","OAI.ZeroTemperatureSK.stripSecond_hasDerivAt_on_constant":"q″ expression derivative on plateaus","OAI.ZeroTemperatureSK.stripSecond_zero_pos":"q″ expression positive at c=0","OAI.ZeroTemperatureSK.stripThird":"Expression for q‴ on a strip","OAI.ZeroTemperatureSK.stripThird_coercive":"Coercive lower bound, q‴ expression","OAI.ZeroTemperatureSK.stripThird_eq_actual":"The q‴ expression as an expectation","OAI.ZeroTemperatureSK.stripThird_pos":"The q‴ expression is positive","OAI.ZeroTemperatureSK.terminal_plateau_denominator":"Denominator bound on a terminal plateau","OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one":"Terminal plateau gives q(t)<t near one","OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect":"Square-root terminal deficit","OAI.ZeroTemperatureSK.terminal_plateau_transition":"Diffusion transition on a terminal plateau","OAI.ZeroTemperatureSK.thirdCurvatureField_eq":"Polynomial form of the q‴ field","OAI.ZeroTemperatureSK.third_not_identically_zero":"∂ₓa is not identically zero","OAI.ZeroTemperatureSK.value":"The value function Φ","OAI.ZeroTemperatureSK.valueRate":"Time derivative of the value","OAI.ZeroTemperatureSK.value_coefficient_stability":"Value is stable in the coefficient","OAI.ZeroTemperatureSK.value_contDiff":"The value is smooth in x","OAI.ZeroTemperatureSK.value_convex":"The value is convex in x","OAI.ZeroTemperatureSK.value_derivative_continuousOn":"x-derivatives of the value are continuous","OAI.ZeroTemperatureSK.value_derivative_hasDerivWithinAt_right":"Right time derivative of x-derivatives","OAI.ZeroTemperatureSK.value_derivative_time_bound":"x-derivatives of Φ Lipschitz in time","OAI.ZeroTemperatureSK.value_even":"The value is even in x","OAI.ZeroTemperatureSK.value_integral_rate":"The value integrates its rate","OAI.ZeroTemperatureSK.value_lipschitz":"The value is 1-Lipschitz in x","OAI.ZeroTemperatureSK.value_on_terminal_plateau":"Explicit value on a terminal plateau","OAI.ZeroTemperatureSK.value_regular":"Value is regular before time one","OAI.ZeroTemperatureSK.value_space_le":"One-sided spatial bound on the value","OAI.ZeroTemperatureSK.value_terminal_uniform":"Uniform bound on Φ minus |x|","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds":"Bounded x-derivatives of Φ on strips","OAI.ZeroTemperatureSK.zero_gap_curvature":"q″ positive on zero plateaus"},"records":{"assumptions.full_support_minimizer":{"lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: minimal among order parameters, with the Parisi value computed on the given Brownian system.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":29,"line":24,"lines":[24,29],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem","note":"A probability law and a real Brownian process on the sample space.","plain":"Probability space with Brownian motion","relation":"defines"}],"short":"Admissible minimiser premise","statements":[{"anchor":"model","at":"b3841-4231","end_byte":4231,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The convention: mass of [0,t] equals γ(t), so an atom of size γ(0) is at zero.","plain":"μ is γ's Stieltjes measure","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\mathcal U$ consist of nonnegative, nondecreasing, right-continuous\nfunctions $\\gamma:[0,1)\\to[0,\\infty)$ with\n$\\int_0^1\\gamma(t)\\,dt<\\infty$. Their Stieltjes measures satisfy\n$\\mu_\\gamma([0,t])=\\gamma(t)$, including\n$\\mu_\\gamma(\\{0\\})=\\gamma(0)$. They are locally finite. They need not be\nprobability measures, and their total mass may be infinite. Support is taken\nrelative to [0,1).","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":3841},{"anchor":"formal","at":"b240278-240424","end_byte":240424,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Quantifies over W and γ, then takes hmin.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[82,83],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The two witnesses, on the given Brownian system.","plain":"The full-support conclusion","relation":"part"}],"lean_source":"statement","markdown":"Its quantifiers are $\\forall W\\,\\forall\\gamma$, followed by the minimizer\nhypothesis, and it supplies both witnesses on the given Brownian system.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":240278},{"anchor":"model","at":"b3348-3505","end_byte":3505,"heading_id":"the-model-and-the-result","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","paper":"short","role":"definition","section":"1. The model and the result","start_byte":3348},{"anchor":"evidence-scope","at":"b52898-53043","end_byte":53043,"heading_id":"formal-verification","lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: minimal among order parameters, with the Parisi value computed on the given Brownian system.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"asserts the\nsupport and moment conclusions of Theorem 1, with an admissible\nminimiser as its hypothesis, and constructs the associated diffusion.","paper":"short","role":"mention","section":"Formal verification","start_byte":52898},{"anchor":"order-parameter-example","at":"b7752-7904","end_byte":7904,"heading_id":"what-a-gap-means","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model. Nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"In general, an admissible **order parameter** is a function\n$\\gamma:[0,1)\\to[0,\\infty)$ that is nondecreasing, right-continuous\nand has finite integral.","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":7752},{"anchor":"source-record","at":"b62874-62970","end_byte":62970,"heading_id":"formal-verification-and-sources","lean":[{"clause":"(hmin : IsMinimizer W γ)","end_line":59,"line":35,"lines":[36,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The theorem takes a minimiser as a hypothesis. Attainment and the external Parisi formula appear nowhere in it.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Minimal among order parameters, with the Parisi value computed on the given Brownian system.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"statement takes the minimiser as a\nhypothesis, so the external Parisi formula and its attainment","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62874},{"anchor":"value-and-diffusion","at":"b14717-14911","end_byte":14911,"heading_id":"the-zero-coefficient","lean":[{"clause":"brownian : IsBrownianReal B law","end_line":29,"line":24,"lines":[29,29],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem","note":"Support model. Rate-one noise is taken as given. Lean has no variance-rate parameter.","plain":"Probability space with Brownian motion","relation":"defines"}],"lean_source":"statement","markdown":"Our\nnormalisation $\\xi(t)=t^2/2$ gives rate one, so the noise\nis **Brownian motion** $B$, a continuous random path starting at zero\nwhose increments over disjoint time intervals are independent.","paper":"expository","role":"definition","section":"3.1 The zero coefficient","start_byte":14717},{"anchor":"value-and-diffusion","at":"b14849-15005","end_byte":15005,"heading_id":"the-zero-coefficient","lean":[{"clause":"brownian : IsBrownianReal B law","end_line":29,"line":24,"lines":[29,29],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem","note":"Support model. Both facts are packaged in IsBrownianReal, which Lean assumes here.","plain":"Probability space with Brownian motion","relation":"defines"},{"end_line":25,"line":14,"lines":[18,19],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.BrownianSystem.abs_increment","note":"hasLaw_sub is the increment law the Lean uses, from the Brownian structure in W.brownian.","plain":"Mean absolute Brownian increment","proof_lines":[18,25],"relation":"proof"}],"lean_source":"statement","markdown":"whose increments over disjoint time intervals are independent. Over an\ninterval of length $h$, an increment has the law of the centred\nGaussian $\\sqrt h Z$,","paper":"expository","role":"explanation","section":"3.1 The zero coefficient","start_byte":14849},{"anchor":"dependencies","at":"b12866-13009","end_byte":13009,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":67,"line":50,"lines":[50,51],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.OrderParameter.local_bound","note":"The bound before time one.","plain":"γ bounded by its integral","proof_lines":[52,67],"relation":"part"},{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"A gap gives a constant interval.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"}],"lean_source":"statement","markdown":"Monotonicity gives bounded\ncoefficients before time one and nonnegative coefficient jumps, and it\nturns a support gap into a constant interval.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12866},{"anchor":"dependencies","at":"b13911-13959","end_byte":13959,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The minimiser is the hypothesis hmin.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"}],"lean_source":"statement","markdown":"The support proof is conditional on a minimizer.","paper":"long","role":"explanation","section":"2.3 Prior results and their roles","start_byte":13911},{"anchor":"dependencies","at":"b17220-17348","end_byte":17348,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"hmin is assumed.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":24,"line":16,"lines":[22,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"The value theorem assumes IsMinimizer too.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"The full-support theorem below remains conditional on a minimizer, and neither of the selected formal interfaces constructs one.","paper":"long","role":"explanation","section":"2.3 Prior results and their roles","start_byte":17220},{"anchor":"contact","at":"b31857-31991","end_byte":31991,"heading_id":"from-minimality-to-contact","lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Lean minimises over order parameters; the measure is read off afterwards, and no mass constraint appears.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The minimisation takes place over nonnegative Stieltjes measures with an\nintegrable cumulative function, and their total mass is free.","paper":"long","role":"explanation","section":"3. From minimality to contact","start_byte":31857},{"anchor":"main-result","at":"b6834-6881","end_byte":6881,"heading_id":"the-model-and-the-result","lean":[{"clause":"integrable : Integrable (extend val)","end_line":22,"line":17,"lines":[22,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: nonnegative, monotone, right-continuous and integrable on [0,1).","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"Integrability allows $\\gamma$ to grow near one.","paper":"short","role":"mention","section":"1. The model and the result","start_byte":6834}]},"assumptions.qaoa_model_and_limits":{"lean":[],"short":"QAOA model and limit order","statements":[{"anchor":"qaoa","at":"b104974-105063","end_byte":105063,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"The cost and mixer are\n$$\nC_{n,J}=n^{-1/2}\\sum_{i<j}J_{ij}Z_iZ_j,\\qquad B_n=\\sum_iX_i.\n$$","paper":"long","role":"definition","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":104974},{"anchor":"qaoa","at":"b105755-106526","end_byte":106526,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"A depth-$p$ state, with $p$ counting cost–mixer pairs, is\n$$\n|\\psi_{n,p,J}(\\boldsymbol\\gamma,\\boldsymbol\\beta)\\rangle\n=e^{-i\\beta_pB_n}e^{-i\\gamma_pC_{n,J}}\\cdots\n e^{-i\\beta_1B_n}e^{-i\\gamma_1C_{n,J}}|+\\rangle^{\\otimes n}.\n$$\nHere the real angle vectors $\\boldsymbol\\gamma,\\boldsymbol\\beta$ are\ndeterministic, and $\\boldsymbol\\gamma$ is separate from the Parisi\ncoefficient $\\gamma(t)$. A *word* means a finite ordered product of cost\nand mixer exponentials. Its pattern and angles are fixed independently of\n$n,J$, while its cost operator depends on $n,J$. Define\n$$\nv_p(\\boldsymbol\\gamma,\\boldsymbol\\beta)\n=\\lim_{n\\to\\infty}\\frac1n\\mathbb E_J\n \\langle\\psi_{n,p,J}|C_{n,J}|\\psi_{n,p,J}\\rangle,\n\\qquad Q_p=\\sup_{\\boldsymbol\\gamma,\\boldsymbol\\beta\\in\\mathbb R^p}v_p.\n$$","paper":"long","role":"definition","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":105755},{"anchor":"qaoa-boundary","at":"b40764-41453","end_byte":41453,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"Let $\\widehat Z_i,\\widehat X_i$ be Pauli operators, and define the cost\nand mixer by\n$$\nC_{n,J}=\\frac1{\\sqrt n}\\sum_{i<j}J_{ij}\\widehat Z_i\\widehat Z_j,\n\\qquad B_n=\\sum_i\\widehat X_i.\n$$\nThe eigenvalues of the cost are $H_n(\\sigma)$. QAOA alternates cost\nevolution and mixing, starting from the uniform state\n$|+\\rangle^{\\otimes n}$ ([Farhi, Goldstone and Gutmann (2014)][FGG]).\nFor a fixed sequence\n$W=(\\theta_1,\\beta_1,\\ldots,\\theta_p,\\beta_p)$, write\n$$\n|\\psi_{n,J}(W)\\rangle\n=e^{-i\\beta_p B_n}e^{-i\\theta_p C_{n,J}}\\cdots\n e^{-i\\beta_1 B_n}e^{-i\\theta_1 C_{n,J}}|+\\rangle^{\\otimes n},\n$$\n$$\nv(W)=\\lim_{n\\to\\infty}\\frac1n\\mathbb E_J\n\\langle\\psi_{n,J}(W)|C_{n,J}|\\psi_{n,J}(W)\\rangle.\n$$","paper":"short","role":"definition","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":40764},{"anchor":"spins-and-overlap","at":"b2451-2896","end_byte":2896,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":93,"line":91,"lines":[91,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. Sum over i<j, divided by the square root of n. The couplings are indexed by all ordered pairs; only i<j enter.","plain":"SK Hamiltonian on n spins","relation":"defines"},{"end_line":96,"line":95,"lines":[95,96],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinDisorder","note":"Support model. Product of standard Gaussians over all ordered pairs, so independence is built into the measure.","plain":"Gaussian disorder on n spins","relation":"defines"}],"lean_source":"statement","markdown":"For each unordered pair we draw an independent standard Gaussian\n$J_{ij}$, a normal random variable with mean zero and variance one.\nIndependence means that learning the other couplings leaves the\nprobability distribution of this one unchanged.\nThese coefficients, or **couplings**, are drawn once and then held\nfixed while we maximise the energy, or **Hamiltonian**,\n\n$$\nH_n(\\sigma)=\\frac1{\\sqrt n}\\sum_{1\\le i<j\\le n}J_{ij}\\sigma_i\\sigma_j.\n$$","paper":"expository","role":"definition","section":"1. The finite optimisation problem","start_byte":2451},{"anchor":"qaoa-boundary","at":"b61859-61917","end_byte":61917,"heading_id":"further-reading","lean":[],"lean_source":"record","markdown":"develops its circuit construction and\nthe order of limits.","paper":"expository","role":"mention","section":"Further reading","start_byte":61859},{"anchor":"spins-and-overlap","at":"b2281-2450","end_byte":2450,"heading_id":"the-finite-optimisation-problem","lean":[{"clause":"(σ : Fin n → Bool)","end_line":93,"line":91,"lines":[91,91],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. Lean encodes each sign as a Boolean, read as +1 or -1 inside the sum.","plain":"SK Hamiltonian on n spins","relation":"defines"}],"lean_source":"statement","markdown":"At each of $n$ positions, or **sites**, we choose a sign $\\sigma_i\\in\\{-1,1\\}$,\ncalled a **spin**, and a configuration is the vector\n$\\sigma=(\\sigma_1,\\ldots,\\sigma_n)$.","paper":"expository","role":"definition","section":"1. The finite optimisation problem","start_byte":2281},{"anchor":"spins-and-overlap","at":"b2898-2980","end_byte":2980,"heading_id":"the-finite-optimisation-problem","lean":[{"clause":"J (i,j) * (if σ i then 1 else -1) * (if σ j then 1 else -1)","end_line":93,"line":91,"lines":[93,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. Each term is the coupling times the product of two signs.","plain":"SK Hamiltonian on n spins","relation":"defines"}],"lean_source":"statement","markdown":"A positive coupling rewards equal signs and a negative one rewards\nopposite signs.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":2898},{"anchor":"spins-and-overlap","at":"b2981-3215","end_byte":3215,"heading_id":"the-finite-optimisation-problem","lean":[],"lean_source":"record","markdown":"For example, positive couplings on pairs $12$ and $23$ favour\n$\\sigma_1=\\sigma_2=\\sigma_3$, which conflicts with a negative coupling\non pair $13$. This inability to maximise every pair's contribution at\nonce is called **frustration**.","note":"Illustration only: frustration has no Lean counterpart.","paper":"expository","role":"example","section":"1. The finite optimisation problem","start_byte":2981},{"anchor":"spins-and-overlap","at":"b3216-3266","end_byte":3266,"heading_id":"the-finite-optimisation-problem","lean":[{"clause":"/ Real.sqrt n","end_line":93,"line":91,"lines":[93,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. The same factor, as a division.","plain":"SK Hamiltonian on n spins","relation":"defines"}],"lean_source":"statement","markdown":"The factor $n^{-1/2}$ sets the scale of\nthe model.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":3216},{"anchor":"spins-and-overlap","at":"b3268-3447","end_byte":3447,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":93,"line":91,"lines":[91,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. The formula is this definition at n=2, a direct instance. The maximum |J12|/√2 is not in the Lean.","plain":"SK Hamiltonian on n spins","relation":"part"}],"lean_source":"statement","markdown":"For two spins, $H_2(\\sigma)=J_{12}\\sigma_1\\sigma_2/\\sqrt2$, and\nchoosing equal or opposite signs according to the sign of $J_{12}$ gives\n$\\max_\\sigma H_2(\\sigma)=|J_{12}|/\\sqrt2$.","paper":"expository","role":"example","section":"1. The finite optimisation problem","start_byte":3268},{"anchor":"spins-and-overlap","at":"b3543-3813","end_byte":3813,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":16,"line":14,"lines":[14,15],"module":"OAI.Probability.SKSupport.Density.DensityHeat","name":"OAI.ZeroTemperatureSK.Heat.heatKernel_formula","note":"Support model. The density for variance t; t=1 is this φ. Positivity, total integral and the interval formula come from mathlib's gaussianReal.","proof_lines":[16,16],"relation":"part"}],"lean_source":"statement","markdown":"A **probability density** assigns probability by area. For a standard\nGaussian random variable $Z$, the density is\n$\\varphi(z)=(2\\pi)^{-1/2}e^{-z^2/2}$, which is nonnegative, has total\nintegral one and, for $a\\le b$, gives\n\n$$\n\\Pr(a\\le Z\\le b)=\\int_a^b\\varphi(z)\\,dz.\n$$","note":"Textbook background. The couplings use mathlib's gaussianReal 0 1, and the Lean writes the density formula for its heat kernel.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":3543},{"anchor":"spins-and-overlap","at":"b4357-4484","end_byte":4484,"heading_id":"the-finite-optimisation-problem","lean":[],"lean_source":"record","markdown":"The expected two-spin maximum is therefore $1/\\sqrt\\pi$. Signs chosen\nbefore the coupling is seen have expected energy zero, so","note":"No Lean counterpart: the Lean has no two-spin case.","paper":"expository","role":"example","section":"1. The finite optimisation problem","start_byte":4357},{"anchor":"spins-and-overlap","at":"b4414-4585","end_byte":4585,"heading_id":"the-finite-optimisation-problem","lean":[{"clause":"∫ J, sSup (Set.range (spinHamiltonian n J)) ∂spinDisorder n","end_line":99,"line":98,"lines":[99,99],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.expectedSKMaximum","note":"Support model. The maximum is inside the integral: signs are chosen after each coupling draw.","plain":"Expected maximum energy per spin","relation":"defines"}],"lean_source":"statement","markdown":"Signs chosen\nbefore the coupling is seen have expected energy zero, so the order\nmatters: we choose signs for each realised coupling and then average\nthe achieved maximum.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":4414},{"anchor":"spins-and-overlap","at":"b4893-5048","end_byte":5048,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":93,"line":91,"lines":[92,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"Support model. Only two-spin products appear, and there is no one-spin term.","plain":"SK Hamiltonian on n spins","relation":"defines"}],"lean_source":"statement","markdown":"The model is **pure** because every term contains\nexactly two spins, and has **zero external field** because there is\nno additional term $h\\sum_i\\sigma_i$.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":4893},{"anchor":"spins-and-overlap","at":"b4216-4258","end_byte":4258,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":191,"line":180,"lines":[180,181],"module":"OAI.Probability.SKSupport.Foundations.Const","name":"OAI.ZeroTemperatureSK.Heat.semigroup_eq_scaled","note":"The same fact as an identity of averages: the proof shows √h times a standard Gaussian has law gaussianReal 0 h.","proof_lines":[182,191],"relation":"part"}],"lean_source":"statement","markdown":"so\n$\\sqrt h Z$ has variance $h$ for $h>0$.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":4216}]},"assumptions.value_diffusion":{"lean":[{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Scalar model: stopped process adapted, almost surely continuous on [0,1], and solving the integral equation there.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Uses this premise as a hypothesis of its first two clauses, with 0<T and T<1 in the second. The third omits it.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"part"}],"short":"Scalar diffusion premise","statements":[{"anchor":"targets","at":"b5337-5417","end_byte":5417,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Scalar model: stopped process adapted, almost surely continuous on [0,1], solving the integral equation there.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model: progressive, continuous on [0,1], X_0=0, and the same integral equation.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":72,"line":71,"lines":[71,72],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gradient","note":"Scalar model: u is the x-derivative of the control value phi.","plain":"The gradient u = ∂ₓΦ","relation":"defines"},{"end_line":75,"line":74,"lines":[74,75],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.curvature","note":"Scalar model: a is the x-derivative of u.","plain":"The curvature a = ∂ₓₓΦ","relation":"defines"}],"lean_source":"statement","markdown":"Write $u=\\Phi_x$, $a=\\Phi_{xx}$, and\n$dX_t=\\gamma(t)u(t,X_t)\\,dt+dB_t$, $X_0=0$.","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":5337},{"anchor":"formal-scalar","at":"b243180-243599","end_byte":243599,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Three clauses. The first two take a minimiser and a diffusion, the second also 0<T<1; the third takes only the minimiser.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Scalar model: stopped process adapted, almost surely continuous on [0,1], and solving the integral equation there.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"The second formal statement has three clauses. Given a minimizer and an\nassociated diffusion, it establishes convergence of the finite-spin\nground-state values, a bounded martingale with a spin-valued terminal\nlimit, convergence of the gradient and the value identities. Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":243180},{"anchor":"formal-parisi-clause","at":"b245894-246060","end_byte":246060,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":27,"line":25,"lines":[25,27],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"The proof applies the equality with a supplied diffusion to diffusionProcess.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"proof"},{"end_line":77,"line":74,"lines":[74,74],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.diffusionProcess_isDiffusion","note":"The diffusion predicate for that process.","plain":"The constructed process solves the equation","proof_lines":[75,77],"relation":"proof"}],"lean_source":"statement","markdown":"Its proof chooses\n$X=\\texttt{diffusionProcess}\\ W\\ \\gamma$ and proves the diffusion\npredicate for that process before applying the equality with a supplied\ndiffusion.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":245894},{"anchor":"model","at":"b5127-5300","end_byte":5300,"heading_id":"the-model-and-the-result","lean":[{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"The SDE in integral form, with X adapted and continuous.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":79,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.squareMoment","note":"The moment q, for any process X.","plain":"Second moment of u along X","relation":"defines"}],"lean_source":"statement","markdown":"Brownian\nmotion, and define the diffusion $X$ and its self-consistency function\n$q$ by\n$$\ndX_t=\\gamma(t)u(t,X_t)\\,dt+dB_t,\\qquad X_0=0,\n\\qquad q(t)=\\mathbb E[u(t,X_t)^2].\n$$","paper":"short","role":"definition","section":"1. The model and the result","start_byte":5127},{"anchor":"evidence-scope","at":"b53314-53437","end_byte":53437,"heading_id":"formal-verification","lean":[{"end_line":27,"line":25,"lines":[25,27],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"The third clause: only the minimiser is assumed, and its proof supplies the diffusion.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"proof"}],"lean_source":"statement","markdown":"and its third\nclause identifies the ground-state and Parisi values from a minimiser,\nconstructing the diffusion internally.","paper":"short","role":"explanation","section":"Formal verification","start_byte":53314},{"anchor":"value-and-diffusion","at":"b21408-21639","end_byte":21639,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model: the integral equation, path continuity and start at zero.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Scalar model: stopped process adapted, almost surely continuous on [0,1], solving the integral equation. No separate X_0=0 clause.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"Following it generates the state\n\n$$\nX_t=B_t+\\int_0^t\\gamma(s)u(s,X_s)\\,ds,\\qquad X_0=0.\n$$\n\nThis is the **associated diffusion**, whose existence and uniqueness for\na fixed Brownian motion follow from the cited regularity theorem.","paper":"expository","role":"definition","section":"3.3 The response, the state and the averaged square","start_byte":21408},{"anchor":"value-and-diffusion","at":"b21098-21407","end_byte":21407,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":54,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.gradient","note":"Support model. A deterministic function, built from value before any path is run.","plain":"The gradient u = ∂ₓΦ","relation":"defines"},{"clause":"gradient W γ s (X (Real.toNNReal s) ω))","end_line":75,"line":68,"lines":[75,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model. The drift at time s depends only on s and the state at s.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"We first fix $\\gamma$ and obtain the deterministic value and slope by\noptimising the expected future payoff. We then use that slope as a rule\nalong a random path. Computing an expected future payoff reveals nothing\nabout the future noise, and at time $t$ the rule needs only $t$ and the\nstate already reached.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":21098},{"anchor":"value-and-diffusion","at":"b21640-21729","end_byte":21729,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"gradient W γ s (X (Real.toNNReal s) ω))","end_line":75,"line":68,"lines":[75,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model. The slope evaluated at the state already reached.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"The\nrule is called **feedback** because it uses the slope at the currently\nreached state.","paper":"expository","role":"definition","section":"3.3 The response, the state and the averaged square","start_byte":21640},{"anchor":"value-and-diffusion","at":"b22096-22182","end_byte":22182,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"ContinuousOn (fun t : ℝ => X (Real.toNNReal t) ω) (Set.Icc 0 1)","end_line":75,"line":68,"lines":[71,71],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model. The definition requires path continuity. A jump of γ only changes the integrand.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"Its drift\naccumulates through time, so $X$ stays continuous even where $\\gamma$\njumps.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":22096},{"anchor":"value-and-diffusion","at":"b21974-22095","end_byte":22095,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"X (Real.toNNReal t) ω = W.B (Real.toNNReal t) ω +","end_line":75,"line":68,"lines":[73,73],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model. The integral equation, which is the Lean definition. Euler steps are separate.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"This discrete approximation explains the order of operations,\nwhile the integral equation defines the continuous process.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":21974},{"anchor":"changed-case-answer","at":"b57018-57129","end_byte":57129,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"Changing the coefficient after $1/2$ leaves $X_{1/4}=B_{1/4}$, because\nits drift has been zero up to that time.","note":"Not in the Lean: it has no lemma for this example. IsDiffusion's drift contains γ on [0,t], so zero γ there would give X_t=B_t.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":57018},{"anchor":"model","at":"b5019-5106","end_byte":5106,"heading_id":"the-model-and-the-result","lean":[{"end_line":54,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.gradient","note":"","plain":"The gradient u = ∂ₓΦ","relation":"defines"},{"end_line":57,"line":56,"lines":[56,57],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.curvature","note":"","plain":"The curvature a = ∂ₓₓΦ","relation":"defines"}],"lean_source":"statement","markdown":"Write $u=\\Phi_x$ for the spatial gradient and\n$a=\\Phi_{xx}$ for its spatial derivative.","paper":"short","role":"definition","section":"1. The model and the result","start_byte":5019}]},"assumptions.value_minimizer":{"lean":[{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Same form as the support model, with this model's own functional.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Coefficient on the real line; the conditions apply on [0,1) and integrability on (0,1].","plain":"Admissible order parameter on ℝ","relation":"defines"},{"end_line":47,"line":40,"lines":[40,47],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace","note":"Bundles the sample space with the law, unlike the support model's BrownianSystem.","plain":"Bundled space with Brownian motion","relation":"defines"}],"short":"Scalar minimiser premise","statements":[{"anchor":"targets","at":"b6457-6561","end_byte":6561,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Scalar model: the first two clauses take IsMinimizer and IsDiffusion; the third takes only IsMinimizer.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Same form as the support model, with this model's own functional.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":47,"line":40,"lines":[40,47],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace","note":"Bundles the sample space with the law, unlike the support model's BrownianSystem.","plain":"Bundled space with Brownian motion","relation":"defines"}],"lean_source":"statement","markdown":"Its formal statement is separate\nand keeps the minimizer and diffusion hypotheses where they are needed.","paper":"long","role":"mention","section":"1. The SK model and the Parisi variational problem","start_byte":6457},{"anchor":"formal-parisi-clause","at":"b246061-246152","end_byte":246152,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Hypothesis hmin only, with no diffusion argument. No clause produces a minimiser.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Same form as the support model, with this model's own functional.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"No diffusion is required from the caller, and none of these\nclauses constructs a minimizer.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":246061},{"anchor":"formal-transport","at":"b247853-248166","end_byte":248166,"heading_id":"relating-the-two-statements","lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model. No Lean theorem transports minimality to the scalar model.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Scalar model. No Lean theorem transports minimality to the support model.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"**Value and minimizer.** Equality of the payoff sets gives equality of their\nsuprema for every $x$ and $0\\le t<1$. Only then identify the first and second\nspatial derivatives and the Parisi penalties. Transport minimality against\nevery admissible competitor, including restriction or extension of that\ncompetitor.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":247853},{"anchor":"evidence-scope","at":"b53129-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Same form as the support model, with this model's own functional.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"The diffusion premise: X is adapted, continuous, and solves the SDE with this model's gradient.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Coefficient on the real line; the conditions apply on [0,1) and integrability on (0,1].","plain":"Admissible order parameter on ℝ","relation":"defines"}],"lean_source":"statement","markdown":"[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"mention","section":"Formal verification","start_byte":53129},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"The scalar model's statement: value identities and the finite Gaussian limit, for a minimiser.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Same form as the support model, with this model's own functional.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62987},{"anchor":"dependencies","at":"b10412-10617","end_byte":10617,"heading_id":"outline-of-the-proof","lean":[{"end_line":24,"line":16,"lines":[17,21],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Both clauses take one minimiser and one diffusion together.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"**Value and circuits.** We keep the same minimizer and diffusion through\n[the value identity in §6](#scalar-value-identities) and\n[the finite Gaussian construction in §7](#finite-gaussian-approximation).","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":10412},{"anchor":"approximation-choices","at":"b29646-29751","end_byte":29751,"heading_id":"the-order-of-the-approximation-limits","lean":[{"end_line":24,"line":16,"lines":[17,21],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"W, γ and X are bound once in each clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"A single admissible minimiser\nand its associated diffusion stay fixed throughout the scalar construction.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":29646}]},"circuit.additive_responses":{"lean":[],"short":"Additive cut responses","statements":[{"anchor":"qaoa-synthesis","at":"b140426-141309","end_byte":141309,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"There remains a response inside the lower formula $f$. A cut can pass\nthrough a descendant used to compute $f$, rather than through one of\nthe displayed three-edge paths. The source first writes the exact\ndifference of the affected Hamiltonian terms before applying the\npreceding subtree circuit. That circuit is fixed with respect to the\nnew cut input. Exclusions ensure that the downward coefficients are\nindependent of that active boundary argument. The resulting additive\ncut-response rule retains both first derivatives and mixed second\nderivatives:\n$$\n\\sum_\\alpha a_\\alpha q(H_\\alpha\\mathrm{vac})\n +\\frac12\\sum_{\\alpha,\\lambda}a_\\alpha a_\\lambda\n       \\langle H_{\\alpha\\lambda}\\rangle_0.\n$$\nHere $H_\\alpha$ is centred by parity, so no limit of the extensive\nzero-input Hamiltonian is required. The self-adjoint generator is\nobtained from the limiting time group of cut ratios.","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":140426},{"anchor":"qaoa-synthesis","at":"b141677-142134","end_byte":142134,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"The mixed coefficients $c_{\\alpha\\lambda}$ must remain even when two\nchannels share descendants. A bounded scalar example makes the issue\nvisible: for a shared standard normal $Q$, take\n$H(y_1,y_2)=\\cos(Q+y_1+y_2)-\\cos Q$. At zero both first derivatives\nare $-\\sin Q$ and are centred, but\n$\\mathbb E H_{12}=-e^{-1/2}$. The two cross terms in the half-sum\ntherefore contribute $-e^{-1/2}a_1a_2$. Centred first responses do not\nremove a mixed second response.","paper":"long","role":"example","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":141677}]},"circuit.bounded_trigonometric":{"lean":[],"short":"Bounded trigonometric approximation","statements":[{"anchor":"qaoa-regularization","at":"b124183-125031","end_byte":125031,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"If the bounded output is $m^*$, shrink it to\n$(1-\\kappa)m^*$, with fixed $0<\\kappa<1$, before taking\n$\\theta_\\kappa=\\tfrac12\\arcsin((1-\\kappa)m^*)$. This margin keeps\nthe derivative of inverse sine bounded on its used range, and the output\nloss is at most $\\kappa$ in $L^2$.\n\nWe then truncate the transmitted functions and approximate the resulting bounded maps by trigonometric\npolynomials. The source makes these choices from output\ntoward leaves: each outer Lipschitz function is fixed before its inputs\nare approximated. The Gaussian isometry above converts an error in a\ntransmitted child vector into the same $L^2$ error in its receiving\nposition. Energy continuity then bounds the accumulated loss. The\nbounded output has the form $m=\\sin(2\\theta)$. The transmitted columns are\nbounded, while the Gaussian fields themselves remain unbounded.","paper":"long","role":"explanation","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":124183},{"anchor":"qaoa-boundary","at":"b48379-48572","end_byte":48572,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"[OpenAI (2026b), §§3–5][Q] smooths the spin and approximates every\ntransmitted function by a bounded smooth map that preserves oddness.\nSmoothing the final sign alone would be insufficient.","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":48379}]},"circuit.classical_fixed_minimizer":{"lean":[],"short":"Bounded input for given minimizer","statements":[{"anchor":"qaoa-regularization","at":"b127916-128066","end_byte":128066,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"The resulting classical\nnear-optimality proposition supplies a finite bounded odd formula,\nzero on helpers, with energy as close to $P_*$ as required.","paper":"long","role":"mention","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":127916},{"anchor":"qaoa-assembly","at":"b161440-161796","end_byte":161796,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"The bounded formula is chosen before the colors and helper law. The\nexcluded receiver and ancestor colors prevent a dependency from returning\nthrough a traversed branch. Reserved helper colors are absent from every\ndependency, and the output is zero on helpers. Color refinement and small\npositive helper probabilities jointly control $\\delta_c$ and $p_H$.","paper":"long","role":"explanation","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":161440}]},"circuit.classical_positive":{"lean":[],"short":"Alternative bounded synthesis input","statements":[{"anchor":"alternative-assembly","at":"b219040-219225","end_byte":219225,"heading_id":"where-the-temperature-routes-meet","lean":[],"lean_source":"record","markdown":"Both routes produce a bounded finite classical calculation whose rooted\nenergy is close to $P_*$. They reach that common input by different\ntheorems and different finite approximations.","paper":"long","role":"explanation","section":"11.1.1 Where the temperature routes meet","start_byte":219040},{"anchor":"alternative-assembly","at":"b220789-221019","end_byte":221019,"heading_id":"where-the-temperature-routes-meet","lean":[],"lean_source":"record","markdown":"One can choose $\\beta_0$ and $T$ finitely for the required\nerror and then stop. The later circuit argument needs that finite\nchoice and never asks for a joint limit of the temperature, message\nmesh, graph degree and circuit depth.","paper":"long","role":"explanation","section":"11.1.1 Where the temperature routes meet","start_byte":220789}]},"circuit.classical_primary":{"lean":[],"short":"Near-optimal bounded synthesis input","statements":[{"anchor":"qaoa-assembly","at":"b162401-162935","end_byte":162935,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"We allocate the errors after choosing the desired $\\varepsilon>0$, and\napply the\nclassical near-optimality proposition with the two tolerances\n$\\varepsilon/4$ and $\\eta_H=(\\varepsilon/8)^2$. It supplies one\nformula and its helper law together, with classical loss below\n$\\varepsilon/4$ and $0<p_H<\\eta_H$. In its proof the uncoloured\nformula is fixed first, then the small positive helper probabilities\nand the excluded-colour approximation are selected. Thus changing a\nhelper law does not silently preserve an already fixed formula.","paper":"long","role":"statement","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":162401}]},"circuit.classical_strong_limit":{"lean":[],"short":"Strong limit of local formulas","statements":[{"anchor":"alternative","at":"b208327-208537","end_byte":208537,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"A separate norm-convergence argument with a uniform first occupation\nmoment identifies this edge limit with the rooted energy. A\nfinite-dimensional central limit theorem alone does not identify that\nobservable.","paper":"long","role":"mention","section":"11.1 The separate positive-temperature route","start_byte":208327}]},"circuit.color_exclusion":{"lean":[],"short":"Seed reencoding and ancestor exclusion","statements":[{"anchor":"qaoa-regularization","at":"b126883-127356","end_byte":127356,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"The types can be encoded together\nwith a Gaussian seed in one raw Gaussian variable: its sign supplies\nthe odd symmetry, while disjoint binary digits of a uniform transform\nof its absolute value supply independent coordinates. The resulting\nfinite type label is even and the Gaussian seed is odd under reversal.\nThe associated pullback preserves vacuum, parity and $\\mathsf L$.\nThus adding types preserves the energy identification and needs no\nadditional circuit resource.","paper":"long","role":"statement","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":126883},{"anchor":"qaoa-regularization","at":"b127424-127915","end_byte":127915,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"For the already fixed formula, the exclusion estimate is\n$$\n\\|m^{\\rm col}-m\\|^2\\le C(\\alpha+p_H),\n$$\nwhere $\\alpha$ is the largest nonhelper type mass and $p_H$ is the total\nhelper mass. Its constant is independent of the number of types. The\nproof uses orthogonality of the disjoint masks before applying the\nLipschitz estimate in the original, fixed list of formula coordinates.\nTreating the refined types as an arbitrary growing list of independent\narguments would not provide this bound.","paper":"long","role":"statement","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":127424}]},"circuit.energy_transfer":{"lean":[],"short":"Edge energy needs occupation control","statements":[{"anchor":"qaoa-rooted","at":"b119233-120325","end_byte":120325,"heading_id":"rooted-realisation-and-the-observable-it-preserves","lean":[],"lean_source":"record","markdown":"Let $f_D$ be\ncentred, invariant under rooted permutations of complete child\nsubtrees, and supported at a fixed finite radius. Suppose its occupation\nembedding converges strongly to $v$, with a uniformly bounded expected\ntop-generation occupation $\\langle f_D,\\mathsf Nf_D\\rangle$. Put\n$M=D+1$ and $R_M=\\sqrt{1-\\mathsf N/M}$.\n\nTo obtain the decomposition, we delete the distinguished edge and project\nonto the vacuum on either resulting half-tree. There are four pairings.\nThe vacuum–vacuum pairing is zero by centring. If only one side is\nexcited, the other side forces the root and all remaining branches into\ntheir vacuum, and root symmetry gives the term\n$M^{-1/2}\\langle\\mathsf L^*f_D,R_Mf_D\\rangle$ or its conjugate.\nIn the pairing where both sides are excited, each projected vector has\nsquared norm $M^{-1}\\langle f_D,\\mathsf Nf_D\\rangle$ by symmetry\namong the $M$ branches. Cauchy–Schwarz bounds their pairing by that same\nquantity. Finally, deleting one vacuum child multiplies the $m$-particle\nsector by $\\sqrt{\\binom{M-1}{m}/\\binom{M}{m}}=\\sqrt{1-m/M}$,\nwhich identifies $R_M$.","paper":"long","role":"statement","section":"8.1 Rooted realisation and the observable it preserves","start_byte":119233},{"anchor":"qaoa-rooted","at":"b120326-120884","end_byte":120884,"heading_id":"rooted-realisation-and-the-observable-it-preserves","lean":[],"lean_source":"record","markdown":"Thus the exact edge decomposition is\n$$\n\\langle f_{D,i},f_{D,j}\\rangle\n =\\frac2{\\sqrt M}\\operatorname{Re}\\langle\\mathsf L^*f_D,R_Mf_D\\rangle\n   +\\varepsilon_D,\n\\qquad\n|\\varepsilon_D|\\le\\frac{\\langle f_D,\\mathsf Nf_D\\rangle}{M}.\n$$\nVacuum deletion also gives\n$\\|R_Mf_D-f_D\\|^2\\le M^{-1}\\langle f_D,\\mathsf Nf_D\\rangle$.\nConsequently,\n$$\n\\lim_{D\\to\\infty}\\frac{\\sqrt D}{2}\n \\langle f_{D,i},f_{D,j}\\rangle=\\mathcal E(v).\n$$\nStrong convergence alone does not control the scaled remainder, and the\noccupation estimate for the relevant finite circuits comes later.","paper":"long","role":"statement","section":"8.1 Rooted realisation and the observable it preserves","start_byte":120326}]},"circuit.even_p_spin_obstruction":{"lean":[],"short":"Even higher-spin ceiling","statements":[{"anchor":"qaoa-depth","at":"b232317-232833","end_byte":232833,"heading_id":"depth-angle-choice-and-the-order-of-quantifiers","lean":[],"lean_source":"record","markdown":"The pure even higher-spin model has a different thermodynamic-first\nconclusion. For every even interaction order $q\\ge4$, [Basso,\nGamarnik, Mei and Zhou, arXiv:2204.10306v2, Theorem 4, (3.15),\nand its proof in §4.2](https://arxiv.org/abs/2204.10306v2) show\n$$\nv_p^{(q)}(\\boldsymbol\\theta,\\boldsymbol\\beta)\n\\le\\eta_{\\mathrm{OGP}}(G_q)<\\eta_{\\mathrm{OPT}}(G_q)\n$$\nfor every finite $p$ and every fixed angle vector, in that source's\nnormalization. The same ceiling bounds their supremum over all finite\n$p$ and angles.","paper":"long","role":"statement","section":"11.3 Depth, angle choice and the order of quantifiers","start_byte":232317}]},"circuit.external_mode_energy":{"lean":[],"short":"Persistent Gaussian mode and energy","statements":[{"anchor":"qaoa-seed","at":"b154281-154974","end_byte":154974,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"In the limit, $\\ell_r$ and $w_r$ become two orthogonal external real\nGaussian modes at each root. The initialization uses $q(\\ell)$ and\nevery simulated seed gate uses the same $q(w)$: choosing\n$t=\\sigma/2$ implements $e^{-i\\sigma Zq(w)}$. Different vertices\nhave independent local copies. This is the required joint law of the\nseeded word. A finite induction uses the Weyl identity\n$$\n\\langle W(f)\\Omega,W(g)\\Omega\\rangle\n =\\exp\\!\\left(-\\tfrac12\\|g-f\\|^2\n                  +i\\operatorname{Im}\\langle f,g\\rangle\\right)\n$$\nto match the relevant finite Gram matrices. Escaping high directions\nsupply precisely the vanishing pairings between the new external\nmodes and endogenous child directions.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":154281},{"anchor":"qaoa-seed","at":"b155067-155941","end_byte":155941,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"**The energy pairing after seed removal.** Matching the seed law and its\nfinite Gram matrices is not yet enough. For a centred low insertion\n$h=\\sum_j s_j\\otimes W(f_j)\\Omega$, the one-particle coefficient of\n$W(f_j)\\Omega$ is $i e^{-\\|f_j\\|^2/2}f_j$. Hence\n$$\n\\langle h,\\mathsf Lh\\rangle\n =-i\\sum_j\\langle s_j,+\\rangle e^{-\\|f_j\\|^2/2}\n                      \\langle f_j,h\\rangle.\n$$\nIn the seeded representation, $\\mathsf L$ places the root-seed factor\nin its vacuum and creates $h$ only in the endogenous child space.\nExternal components therefore contribute zero to the last pairing.\nThe high-direction estimate yields the same zeros in the actual\nlimit. The actual insertion differs from its low insertion by $O(r)$,\nso energy continuity completes the comparison. Matching Gram\nmatrices without this one-particle calculation would not identify\n$\\mathsf L$ or its energy.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":155067}]},"circuit.finite_formula_word_approximation":{"lean":[],"short":"Gaussian data by ordinary word","statements":[{"anchor":"qaoa-assembly","at":"b160428-161439","end_byte":161439,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"**Approximating a specified finite observable.** First fix $T,N$ and the\npositive finite normalizers in §7. The rooted construction in §8.1 has\nenergy exactly $S_{T,N}$. Let $z$ be a smoothed version of its rounded\nspin $F_N$, let $m=\\sin(2\\theta)$ be the bounded formula's spin output,\nand let $m^{\\mathrm{col}}$ be its color-excluded approximation. Write\n$$\n\\delta_s=\\|z-F_N\\|_2,\\qquad\n\\delta_b=\\|m-z\\|_2,\\qquad\n\\delta_c=\\|m^{\\mathrm{col}}-m\\|_2.\n$$\nThese measure errors in the spin outputs themselves. The rooted\nshift is a contraction, so\n$$\n|\\mathcal E(f)-\\mathcal E(g)|\n\\le(\\|f\\|_2+\\|g\\|_2)\\|f-g\\|_2.\n$$\nAll these outputs have norm at most one. Apply synthesis with tolerance\n$\\eta_s$ and total helper probability $p_H$, then remove the seeds with\nenergy tolerance $\\eta_r$. The source's absolute synthesis bound and\nseed-removal theorem, followed by fixed-word transfer, give one ordinary\nword $W$ with\n$$\n|v(W)-S_{T,N}|\n\\le2(\\delta_s+\\delta_b+\\delta_c)\n       +2\\eta_s+2\\sqrt{p_H}+\\eta_r.\n\\tag{8.2}\n$$","paper":"long","role":"statement","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":160428},{"anchor":"qaoa-assembly","at":"b162129-162344","end_byte":162344,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"This gives an absolute approximation of each specified finite\n$S_{T,N}$ in the construction, whether or not it is already near $P_*$.\nIt preserves the energy and makes no claim about the entire classical\noutput law.","paper":"long","role":"statement","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":162129},{"anchor":"qaoa-boundary","at":"b44717-45046","end_byte":45046,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"**Proposition 5 (circuit approximation; a consequence of\n[OpenAI (2026b), §§2–5][Q]).** For each fixed $T,N$ in the preceding\nconstruction with finite positive normalisers, and every $\\eta>0$,\nthere is a finite ordinary QAOA word $W$, with deterministic angles\nindependent of $n$ and $J$, such that\n$$\n|v(W)-S_{T,N}|<\\eta.\n$$","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44717},{"anchor":"qaoa-boundary","at":"b48336-49269","end_byte":49269,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"**The construction behind Proposition 5.**\n[OpenAI (2026b), §§3–5][Q] smooths the spin and approximates every\ntransmitted function by a bounded smooth map that preserves oddness.\nSmoothing the final sign alone would be insufficient. The class\nexclusions of the construction prevent return through a traversed\nbranch and reserve helper classes on which the output is zero, with\npositive probabilities fixed during synthesis. This finite formula is\nthen approximated in energy using temporary Gaussian controls,\nwith the same sitewise seeds throughout the word. Removing those\ncontrols produces an ordinary word, whose limiting energy is $v(W)$ by\nthe fixed-word transfer. The approximation errors are absolute and can\nbe chosen successively for each fixed $S_{T,N}$, before it is chosen\nclose to $P_*$. Energy is the preserved quantity, and the conclusion\nabout a measured configuration below uses a further concentration\ntheorem.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":48336}]},"circuit.fixed_word_concentration":{"lean":[],"short":"Fixed-word measurement moments","statements":[{"anchor":"qaoa-quantifiers","at":"b224083-224482","end_byte":224482,"heading_id":"what-is-averaged-and-a-fixed-word-probability-consequence","lean":[],"lean_source":"record","markdown":"Farhi, Goldstone, Gutmann and Zhou supply the additional input. They prove, with\nthe same Hamiltonian and gate-sign convention, that for every fixed word\n$$\n\\mathbb E_{J,\\Sigma}E_n\\longrightarrow v_W,\n\\qquad\n\\mathbb E_{J,\\Sigma}E_n^2\\longrightarrow v_W^2.\n$$\nTheir second expression is the quantum expectation of $(C_{n,J}/n)^2$\naveraged over disorder, not merely the square of the conditional mean.","paper":"long","role":"statement","section":"11.2 What is averaged, and a fixed-word probability consequence","start_byte":224083}]},"circuit.fixed_word_transfer":{"lean":[],"short":"Fixed tree word transfers","statements":[{"anchor":"qaoa-transfer","at":"b157909-158327","end_byte":158327,"heading_id":"transfer-of-the-resulting-ordinary-word","lean":[],"lean_source":"record","markdown":"After all these choices there is one finite ordinary word $W$, fixed\nindependently of degree, size and disorder. On the $(D+1)$-regular\ntree use\n$C_D=D^{-1/2}\\sum_{\\{i,j\\}\\in E}Z_iZ_j$, counting each edge once.\nThe edge expectation is defined on the finite light cone described in\n§8.1. The source applies the fixed-word transfer theorem\n$$\nv_W=\\lim_{D\\to\\infty}\\frac{\\sqrt D}{2}\n \\langle+|W_D^*Z_iZ_jW_D|+\\rangle.\n$$","paper":"long","role":"statement","section":"8.5 Transfer of the resulting ordinary word","start_byte":157909},{"anchor":"qaoa-transfer","at":"b158796-159270","end_byte":159270,"heading_id":"transfer-of-the-resulting-ordinary-word","lean":[],"lean_source":"record","markdown":"Basso et al. use a negative tree cost in (2.3). Let\n$V=\\prod_{i\\in A}X_i$ on one bipartition of the finite light cone.\nEvery edge has exactly one endpoint in $A$, hence\n$$\nVZ_iZ_jV=-Z_iZ_j,\\qquad VB_DV=B_D,\\qquad\nV|+\\rangle=|+\\rangle.\n$$\nConsequently, if $C_D^-=-C_D$, then $VW(C_D^-,B_D)V=W(C_D,B_D)$\nfor the same fixed word and angles, while its measured edge expectation\nchanges sign. This converts the negative cut convention in their (2.6)\nto the positive energy above.","paper":"long","role":"derivation","section":"8.5 Transfer of the resulting ordinary word","start_byte":158796},{"anchor":"qaoa-boundary","at":"b48951-49010","end_byte":49010,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"whose limiting energy is $v(W)$ by\nthe fixed-word transfer.","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":48951}]},"circuit.gate_calculus":{"lean":[],"short":"Smooth exclusion-gate calculus","statements":[{"anchor":"qaoa","at":"b108896-109446","end_byte":109446,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"**Smooth replacement.** Fix the height, a finite expression and channel\nlist, and the required derivatives. Uniform recursive weighted bounds,\nconvergence of operators and adjoints, and strongly commuting signals with\nthe stated joint exponential bounds permit substitution inside a fixed\nsurrounding sequence in the exclusion-gate class defined in §8.2. Constants depend on that fixed family. This is the input\nused in §8.3 to replace classical aggregates by finite gates; see\n[the foundations, smooth topology and gate calculus][q-gate-calculus].","paper":"long","role":"statement","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":108896},{"anchor":"qaoa-regularization","at":"b132864-133563","end_byte":133563,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"For the formula fixed here, its finite expression tree supplies the finite\nchannel list. Exclusion supplies the commuting physical control algebra,\nand normalized-sum limits and the preceding product estimates supply the signal\nlimits and weighted bounds. Conjugation by the already fixed preceding\ncircuit preserves strong commutation, including when signals share\ndescendants. The gate-calculus theorem applies these facts inductively to\nthe formula and its adjoints, at each cut and inside the fixed surrounding\nsequence. It is this theorem, with these constructed inputs, that licenses\nthe later replacements. Its conditional root-seed bounds also give the\nmasked-column estimates used in §8.3.","paper":"long","role":"statement","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":132864}]},"circuit.gaussian_root_calculus":{"lean":[],"short":"Root shift and Gaussian chaos","statements":[{"anchor":"qaoa-rooted","at":"b116625-117413","end_byte":117413,"heading_id":"rooted-realisation-and-the-observable-it-preserves","lean":[],"lean_source":"record","markdown":"For any finite list of fixed centred real child vectors, their parent\npositions $\\mathsf G(f)=q(f)$ are jointly centred Gaussian, and for $f,g$,\n$$\n\\mathbb E[\\mathsf G(f)\\mathsf G(g)]=\\langle f,g\\rangle,\n\\qquad\n\\|[\\mathsf G(f)-\\mathsf G(g)]\\mathrm{vac}\\|=\\|f-g\\|.\n$$\nThe directions are fixed elements of the child space, even when their\nvalues depend on the child's Gaussian coordinates.\nThe receiving root's own seed is a separate, independent tensor factor.\n\nThe root shift\n$$\n\\mathsf Lf=\\omega\\otimes a^*(f)\\Omega\n$$\nis an isometry on centred vectors, and is extended by zero on the\nvacuum. Its adjoint extracts the root-vacuum, one-child component,\nincluding a constant receiving-root seed. For a classical function of\northonormal Gaussian positions, this is its first Gaussian chaos.","paper":"long","role":"statement","section":"8.1 Rooted realisation and the observable it preserves","start_byte":116625},{"anchor":"qaoa-rooted","at":"b121132-121224","end_byte":121224,"heading_id":"rooted-realisation-and-the-observable-it-preserves","lean":[],"lean_source":"record","markdown":"The elementary continuity bound\n$$\n|\\mathcal E(v)-\\mathcal E(w)|\n \\le(\\|v\\|+\\|w\\|)\\|v-w\\|\n$$","paper":"long","role":"statement","section":"8.1 Rooted realisation and the observable it preserves","start_byte":121132}]},"circuit.helper_contraction":{"lean":[],"short":"Pinned helper contraction","statements":[{"anchor":"qaoa-synthesis","at":"b136911-138205","end_byte":138205,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"The second operation implements an aggregation. Let $I$ be the\nreceiver region, $J$ the sender region, and $K$ an unused helper region\nof fixed positive probability $p_K$. They are disjoint. The lower\nseed-only formula $f$ excludes these regions along its cross-edge\npaths. With $A_{ij}=D^{-1/2}1_{i\\sim j}$, the desired propagation has\ngenerator\n$$\nH=\\sum_{i\\in I,j\\in J}P_iA_{ij}f_j.\n$$\nPartition $J$ into parts $J_l$ of probabilities $p_l$. Two levels of\nspin commutators generate the commuting Hamiltonian\n$$\n\\widetilde H_{\\mathcal P}\n =\\frac1{p_K}\\sum_l\\sum_{i\\in I,\\,j,j'\\in J_l,\\,k\\in K}\n P_iA_{ij}X_jA_{jk}X_kA_{kj'}X_{j'}f_{j'}.\n$$\nFor a fixed receiver–sender edge $i\\sim j$, the terms with $j=j'$\ncontain $X_j^2=1$, and $f_j$ is seed-only and commutes with these spin factors.\nOn helper-pinned states $X_k=1$, so their sum is exactly\n$$\n\\frac1{p_K}P_iA_{ij}f_j\n \\sum_{k\\in K}A_{jk}A_{kj}\n=P_iA_{ij}f_j\\frac{N_K(j)}{Dp_K},\n$$\nwhere $N_K(j)$ counts helper neighbours of $j$. Conditional on the receiver\nedge, its endpoint $i\\in I$ is not a helper and the other $D$ neighbours\nhave mean helper count $Dp_K$. Its binomial variance is $Dp_K(1-p_K)$,\nso the ratio tends to one in $L^2$ at fixed $p_K>0$.\nTerms with $j\\ne j'$ are simple paths and must disappear under\npartition refinement.","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":136911},{"anchor":"qaoa-synthesis","at":"b138366-139135","end_byte":139135,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"Helper pinning is a statement about every helper spin in the light\ncone. Its finite-degree projection is\n$$\n\\Pi_{K,V}=\\prod_{v\\in V}\n \\left[1-1_K(s_v)\\frac{1-X_v}{2}\\right].\n$$\nIt retains the seed variables and helper-labelled descendant spaces\nwhile fixing the helper spins in $|+\\rangle$. The ideal propagation,\nthe exact Hamiltonian $\\widetilde H_{\\mathcal P}$, and the surrounding\ngates that avoid $K$ preserve this compressed space. Multiplying an\ninsertion by $1_{K^c}(s_o)$ only masks the root label and does not impose\nthis descendant condition. Individual finite commutator factors may\nleave the compressed space. Accordingly, the proof first establishes\nthe exact contraction there, then chooses finite commutator products\non the full space to approximate it.","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":138366}]},"circuit.high_girth_maxcut":{"lean":[],"short":"Fixed angles on high-girth graphs","statements":[{"anchor":"qaoa-depth","at":"b234134-234823","end_byte":234823,"heading_id":"limits-of-the-regular-maxcut-consequence","lean":[],"lean_source":"record","markdown":"The original manuscript also gives a consequence for MaxCut. For each\n$\\eta>0$ it selects one finite depth and normalized angles, and then\na degree threshold, such that every finite simple unweighted $(D+1)$-regular graph of\ngirth greater than $2p+1$ has expected cut fraction at least\n$$\n\\frac12+\\frac{P_*-\\eta}{\\sqrt D}\n$$\nfor all sufficiently large $D$. The usual MaxCut cost is\n$H_{\\rm MC}=\\frac12\\sum_{\\{i,j\\}}(I-Z_iZ_j)$, so its angles are\n$2\\boldsymbol\\gamma/\\sqrt D$: the normalized angles are fixed,\nwhile these physical cost angles depend on degree. The sign conversion\nuses a bipartition of the relevant tree neighbourhood of each edge, so\nthe whole graph need not be bipartite.","paper":"long","role":"statement","section":"11.4 Limits of the regular-MaxCut consequence","start_byte":234134}]},"circuit.later_finite_size_comparison":{"lean":[],"short":"Later bounded-angle finite-size estimate","statements":[{"anchor":"qaoa-finite-size","at":"b236554-237549","end_byte":237549,"heading_id":"a-later-finite-size-estimate-and-its-limited-depth-consequence","lean":[],"lean_source":"record","markdown":"Boulebnane, Khan, Subrahmanya, Herman, Farhi, Villalonga and Shaydulin\nlater gave a quantitative comparison with the fixed-angle limit\n[arXiv:2609.40309v1, Corollary S3.1, pp. 46–47][decoupling]. They use the\nordered-tuple Gaussian mixed-spin cost with covariance polynomial\n$\\xi(t)=\\sum_q c_q^2t^q$ of finite degree.\nLet $\\Gamma>0$ bound every cost angle in a depth-$p$ word. Choose\n$x>0$ bounding $\\xi(2),\\xi'(1),\\xi''(33)$, with\n$\\Gamma^2p^2x\\ge1$. The mixer angles need no bound. Set\n$$\nB_p=256p^4(2e\\Gamma^2p^2x)^p,\\qquad\nE_p=70\\Gamma^2p^2x+p\\log2,\\qquad\nR_p=(2E_p/\\log2)^{2p}.\n$$\nFor\n$$\n0<\\varepsilon\\le4\\Gamma px\n \\min\\{e^{-1/2},2^{-(8p^2+3)/100}\\},\n$$\nthe source asserts $|\\nu_n-\\nu_\\infty|\\le\\varepsilon$ whenever\n$$\nn\\ge T_p:=\\max\\left\\{\n2^{8p^2+3}B_p\n \\left(\\frac{2^{\\deg\\xi+4}\\Gamma^2p^2x^2}{\\varepsilon^2}\\right)^{R_p}\n \\left[2\\log_2\\frac{4\\Gamma px}{\\varepsilon}+\\deg\\xi\\right]^2,\n\\frac{16\\Gamma px}{\\varepsilon}\\right\\}.\n$$\nHere $\\nu_n$ is the disorder-averaged energy per spin.","paper":"long","role":"statement","section":"11.5 A later finite-size estimate and its limited depth consequence","start_byte":236554},{"anchor":"qaoa-finite-size","at":"b237551-238057","end_byte":238057,"heading_id":"a-later-finite-size-estimate-and-its-limited-depth-consequence","lean":[],"lean_source":"record","markdown":"The estimate applies to each near-optimal word from §8.6 separately.\nGiven $\\zeta>0$,\nchoose a finite word $W_\\zeta$ with\n$\\nu_\\infty(W_\\zeta)\\ge P_*-\\zeta/2$. Its depth $p_\\zeta$ and cost\nangles are finite. We pad by a zero layer if needed and take\n$\\Gamma_\\zeta=\\max\\{1,\\max_j|\\theta_j(W_\\zeta)|\\}$. Choose a positive\n$\\varepsilon_\\zeta\\le\\zeta/2$ satisfying the displayed small-error\ncondition. With these fixed parameters, the threshold gives\n$$\n\\nu_n(W_\\zeta)\\ge P_*-\\zeta\\qquad(n\\ge T_{p_\\zeta}).\n$$","paper":"long","role":"example","section":"11.5 A later finite-size estimate and its limited depth consequence","start_byte":237551}]},"circuit.masked_controls":{"lean":[],"short":"Finite masked local controls","statements":[{"anchor":"qaoa-synthesis","at":"b135837-136819","end_byte":136819,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"There are two different approximation operations. First, seed\nconjugations of $X$ generate $\\cos(2ts)X$ and $\\sin(2ts)Y$.\nTrigonometric approximation and matrix commutators give the required\nbounded measurable masked spin controls. Sign averaging of masked\n$X$ flips isolates an interaction between chosen types. Since the cost\nterms commute, this isolation is exact. This sign averaging is the\nrefocusing method of [Jones and Knill (1999), §II, p. 2](https://arxiv.org/abs/quant-ph/9905008v1):\northogonal sign rows cancel unwanted Ising couplings, while identical\nrows retain a selected coupling. The source applies it to the finite\npartition of seed masks.\n\nSecond, finite commutator products\nthen approximate the desired pointwise matrices with smooth bounds.\nThey can be substituted after the surrounding circuit has been fixed.\nThe local masks are measurable functions of the seed. Smoothness is\nrequired only in the continuous aggregate inputs, and the mask labels\nneed none.","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":135837}]},"circuit.one_measurement_near_optimum":{"lean":[],"short":"One measurement near optimum","statements":[{"anchor":"qaoa-quantifiers","at":"b225028-225440","end_byte":225440,"heading_id":"what-is-averaged-and-a-fixed-word-probability-consequence","lean":[],"lean_source":"record","markdown":"Now run the circuit construction with accuracy $\\varepsilon/2$ and\nfreeze its resulting word. Since $v_W\\ge P_*-\\varepsilon/2$,\n$$\n\\Pr_{J,\\Sigma}\\{E_n<P_*-\\varepsilon\\}\n\\le \\frac{4}{\\varepsilon^2}\n       \\mathbb E_{J,\\Sigma}(E_n-v_W)^2\\longrightarrow0.\n$$\nThis follows from the expected-value construction **together\nwith** the external second-moment theorem, and it lies outside both\nselected formal interfaces.","paper":"long","role":"statement","section":"11.2 What is averaged, and a fixed-word probability consequence","start_byte":225028}]},"circuit.ordinary_updates":{"lean":[],"short":"Ordinary cost and mixer updates","statements":[{"anchor":"qaoa-transfer","at":"b159591-160025","end_byte":160025,"heading_id":"transfer-of-the-resulting-ordinary-word","lean":[],"lean_source":"record","markdown":"For an ordinary cost or mixer the root updates are, respectively,\n$$\nu\\longmapsto e^{-itZq(v_Z)}u,\n\\qquad\nu\\longmapsto e^{-i\\beta X}u,\n\\qquad v_Z=u^*Zu\\,\\mathrm{vac}.\n$$\nThe source proves smooth convergence for these updates, their\nadjoints and fixed angle derivatives. The resulting occupation\nbounds supply the hypothesis of the root-shift energy proposition\nin §8.1. Thus its energy is the scaled edge limit required for\ntransfer.","paper":"long","role":"statement","section":"8.5 Transfer of the resulting ordinary word","start_byte":159591}]},"circuit.ordinary_word_positive":{"lean":[],"short":"Alternative-route ordinary word","statements":[{"anchor":"alternative-assembly","at":"b220421-220634","end_byte":220634,"heading_id":"where-the-temperature-routes-meet","lean":[],"lean_source":"record","markdown":"Both routes then use the same bounded approximation, exclusions, seeded\nsynthesis, seed removal and fixed-word transfer. The quantum construction\nis shared, and its classical input has been supplied independently.","paper":"long","role":"explanation","section":"11.1.1 Where the temperature routes meet","start_byte":220421}]},"circuit.ordinary_word_primary":{"lean":[],"short":"Finite near-optimal ordinary word","statements":[{"anchor":"qaoa-assembly","at":"b162401-163292","end_byte":163292,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"We allocate the errors after choosing the desired $\\varepsilon>0$, and\napply the\nclassical near-optimality proposition with the two tolerances\n$\\varepsilon/4$ and $\\eta_H=(\\varepsilon/8)^2$. It supplies one\nformula and its helper law together, with classical loss below\n$\\varepsilon/4$ and $0<p_H<\\eta_H$. In its proof the uncoloured\nformula is fixed first, then the small positive helper probabilities\nand the excluded-colour approximation are selected. Thus changing a\nhelper law does not silently preserve an already fixed formula. With\n$\\eta=\\varepsilon/8$, synthesis gives\n$$\n\\mathcal E(v)\\ge\\mathcal E(m)-2\\eta-2\\sqrt{p_H}\n >P_*-3\\varepsilon/4.\n$$\nSeed removal uses less than $\\varepsilon/4$. The resulting fixed\nordinary word therefore satisfies $v_W>P_*-\\varepsilon$ after the\nfixed-word transfer. Its thermodynamic limit is taken only after\nall approximation choices have been made.","paper":"long","role":"statement","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":162401},{"anchor":"qaoa-depth","at":"b228339-228584","end_byte":228584,"heading_id":"depth-angle-choice-and-the-order-of-quantifiers","lean":[],"lean_source":"record","markdown":"The source theorem says that for every $\\varepsilon>0$, there is one\nfinite word $W_\\varepsilon$ such that\n$$\n\\lim_{n\\to\\infty}\\mathbb E E_n(W_\\varepsilon)\\ge P_*-\\varepsilon.\n$$\nThe cost operator uses $J,n$, but this word and its angles do not.","paper":"long","role":"statement","section":"11.3 Depth, angle choice and the order of quantifiers","start_byte":228339}]},"circuit.orthogonal_quadratics":{"lean":[],"short":"Orthogonal mask quadratics","statements":[{"anchor":"qaoa-synthesis","at":"b139219-139935","end_byte":139935,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"The estimate controlling a growing partition is\n$$\n\\rho^2=\\sum_l p_l^2\\le p_J\\max_l p_l\\longrightarrow0.\n$$\nAt a helper root, each part supplies two commuting fields with columns\n$d_l,e_l$ on disjoint sender masks, of weighted norm at most\n$C\\sqrt{p_l}$. The remaining correction is a sum of centred quadratics\n$$\n\\sum_l\\bigl[q(d_l)q(e_l)-\\langle d_l,e_l\\rangle\\bigr].\n$$\nNormal ordering and disjoint-mask orthogonality bound its quadratic\nkernels by $C\\rho$ in Hilbert–Schmidt norm. The source separately\nbounds the exponentials in the weighted topology, uniformly in the\nnumber of parts. Smallness of the generator on smooth vectors alone\nwould not justify inserting its exponential into the surrounding\ncircuit.","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":139219}]},"circuit.positive_finite_step":{"lean":[],"short":"EAMS finite-step edge identity","statements":[{"anchor":"alternative","at":"b206686-207608","end_byte":207608,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"The finite-degree output is a message calculation on the $(D+1)$-regular\ntree, where $\\partial i$ is the set of neighbours of vertex $i$.\nWrite $t_r=r\\delta$ and $b(t,x)=\\gamma(t)m(t,x)$.\nInitialise independent standard normal labels $g_i$, with\n$u^0_{i\\to j}=u_i^0=g_i$, $A^{-1}_{i\\to j}=1$ and\n$x^0_{i\\to j}=x_i^0=\\sqrt\\delta\\,g_i$. For $0\\le r<L$ set\n$$\nu^{r+1}_{i\\to j}=D^{-1/2}\\!\\sum_{\\substack{v\\in\\partial i\\\\v\\ne j}}\n A^{r-1}_{v\\to i}u^r_{v\\to i},\\qquad\nu^{r+1}_i=(D+1)^{-1/2}\\!\\sum_{v\\in\\partial i}\n A^{r-1}_{v\\to i}u^r_{v\\to i}.\n$$\nEach $x$ then takes the Euler step\n$x^{r+1}=x^r+\\delta b(t_r,x^r)+\\sqrt\\delta\\,u^{r+1}$ with its corresponding\ndirected-edge or vertex message. The directed-edge law fixes the common\nnormaliser:\n$$\nc_{D,r}=(\\mathbb E a(t_r,x^r_{i\\to j})^2)^{-1/2},\\quad\nA^r_{i\\to j}=c_{D,r}a(t_r,x^r_{i\\to j}),\\quad\nB_i^r=c_{D,r}a(t_r,x_i^r),\\quad\nz_i^L=\\sqrt\\delta\\sum_{r=1}^{L}B_i^{r-1}u_i^r.\n$$","paper":"long","role":"definition","section":"11.1 The separate positive-temperature route","start_byte":206686},{"anchor":"alternative","at":"b207609-208149","end_byte":208149,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"The one-step lag in $A^{r-1}u^r$ gives conditional centring from the new\nouter Gaussian layer, although successive finite-degree messages need not\nbe independent. At fixed $\\delta,L$, the joint vertex messages converge to\nthe independent $U_r$ above, and $z_i^L$ converges to\n$\\sqrt\\delta\\sum_{r=1}^Lc_{\\delta,r-1}a(t_{r-1},X^\\delta_{r-1})U_r$.\nThe source's separate edge calculation gives\n$$\n\\lim_{D\\to\\infty}\\frac{D+1}{2\\sqrt D}\\mathbb E[z_o^Lz_v^L]\n  =\\delta\\sum_{r=1}^{L-1}c_{\\delta,r}\\mathbb E a(r\\delta,X_r^\\delta),\n\\qquad o\\sim v.\n$$","paper":"long","role":"statement","section":"11.1 The separate positive-temperature route","start_byte":207609}]},"circuit.positive_input":{"lean":[],"short":"Finite-temperature input family","statements":[{"anchor":"alternative","at":"b201450-202215","end_byte":202215,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"The positive-temperature appendix supplies a separate classical input to\nthe same synthesis and seed-removal theorems. Fix $\\beta_0>1$, let\n$\\mu_{\\beta_0}$ be its Parisi measure, and write\n$q=q_{\\beta_0}=\\max\\operatorname{supp}\\mu_{\\beta_0}$.\nHere $\\gamma(t)=\\beta_0\\mu_{\\beta_0}([0,t])$,\n$m=\\phi_{\\beta_0,x}$, $a=\\phi_{\\beta_0,xx}$ and $X$ belong to the\npositive-temperature PDE, whose terminal value is\n$\\beta_0^{-1}\\log(2\\cosh(\\beta_0x))$. They are distinct from the\nzero-temperature objects in §§1–8. The input credited to\n[Lopatto, arXiv:2607.11756v3, Theorem 1.1](https://arxiv.org/pdf/2607.11756v3)\ngives $\\operatorname{supp}\\mu_{\\beta_0}=[0,q]$ for $\\beta_0>1$, and\nits smooth density below $q$ gives the regularity used on a fixed\n$[0,T]$ with $0<T<q$.","paper":"long","role":"statement","section":"11.1 The separate positive-temperature route","start_byte":201450},{"anchor":"alternative","at":"b202578-202831","end_byte":202831,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"His diffusion $\\widehat X$ has noise coefficient $\\beta_0$, hence\n$X=\\widehat X/\\beta_0$ has unit noise and drift\n$\\beta_0\\mu_{\\beta_0}([0,t])m$. His identity\n$\\beta_0^2\\mathbb E\\widehat\\Phi_{yy}(t,\\widehat X_t)^2=1$\nis exactly $\\mathbb E a(t,X_t)^2=1$.","paper":"long","role":"explanation","section":"11.1 The separate positive-temperature route","start_byte":202578}]},"circuit.positive_mesh":{"lean":[],"short":"Normalized mesh convergence","statements":[{"anchor":"alternative","at":"b208150-208326","end_byte":208326,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"Only afterwards take $\\delta\\downarrow0$ with $L=\\lfloor T/\\delta\\rfloor$.\nThe source bounds the discrepancy from\n$\\int_0^T\\mathbb E a(t,X_t)\\,dt$ by $C(\\beta_0,T)\\sqrt\\delta$.","paper":"long","role":"statement","section":"11.1 The separate positive-temperature route","start_byte":208150}]},"circuit.positive_rooted_clipping":{"lean":[],"short":"Positive-temperature clipping","statements":[{"anchor":"alternative-assembly","at":"b218640-218770","end_byte":218770,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"After clipping and bounded approximation, the result enters the same\nsynthesis and seed-removal route as the primary construction.","paper":"long","role":"mention","section":"11.1 The separate positive-temperature route","start_byte":218640}]},"circuit.positive_value":{"lean":[],"short":"Finite-temperature scalar approximation","statements":[{"anchor":"alternative","at":"b203059-203239","end_byte":203239,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"The source derives $\\mathbb E a(t,X_t)^2=1$ on $[0,q]$ and\n$1-q\\le\\beta_0^{-1}$. The value estimate is\n$$\n\\int_0^T\\mathbb E a(t,X_t)\\,dt\n\\ge P_*-\\frac{1/4+\\log2}{\\beta_0}-(q-T).\n$$","paper":"long","role":"statement","section":"11.1 The separate positive-temperature route","start_byte":203059},{"anchor":"alternative","at":"b204932-205116","end_byte":205116,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"The cutoff removes at most $q-T$, because $0\\le\\mathbb E a\\le1$.\nThis proves the displayed estimate. Choose first a sufficiently large\nfinite $\\beta_0$, then $T<q$ close enough to $q$.","paper":"long","role":"explanation","section":"11.1 The separate positive-temperature route","start_byte":204932}]},"circuit.rooted_realization":{"lean":[],"short":"Rooted realization of scalar sum","statements":[{"anchor":"qaoa-rooted","at":"b117655-118864","end_byte":118864,"heading_id":"rooted-realisation-and-the-observable-it-preserves","lean":[],"lean_source":"record","markdown":"We start with the standard root seed $g_0=S$ and recursively set\n$Z_{j+1}=\\mathsf G(g_j)$. Suppose $g_0,\\ldots,g_{j-1}$ are centred\northonormal vectors. The joint Gaussian law of their lifts and the covariance\nidentity give independent standard normals $Z_1,\\ldots,Z_j$. In the scalar\nconstruction of §7, the multiplier\nof $Z_j$ in $g_j$ depends only on its predecessors. Conditional centring\ntherefore makes $g_j$ orthogonal to $g_1,\\ldots,g_{j-1}$, and the chosen\nnormaliser $c_j$ gives unit norm. Independence of the receiving-root\nseed gives orthogonality to $g_0$. This closes the induction. Compatible\nheight embeddings ensure that the earlier variables are the same\nvariables when the formula is embedded at greater height.\n\nOddness of $u$ and evenness of $a$ make $Y_j$ and $g_j$ odd under\nsimultaneous seed reversal. The extra centred rounding variable\n$2F_{\\rm G}(Z_{N+1})-1$ is odd as well, so $F_N$ is centred. It is\nindependent of the receiving-root seed, hence\n$\\langle F_N,g_0\\rangle=0=A_0$. With $A_j,B_{j+1}$ as in §7,\nfirst-chaos projection gives\n$$\n\\mathsf L^*F_N=\\sum_{j=0}^NB_{j+1}g_j,\\qquad\n\\mathcal E(F_N):=\\operatorname{Re}\\langle F_N,\\mathsf LF_N\\rangle\n =\\sum_{j=0}^NA_jB_{j+1}.\n$$","paper":"long","role":"statement","section":"8.1 Rooted realisation and the observable it preserves","start_byte":117655},{"anchor":"qaoa-boundary","at":"b46513-47100","end_byte":47100,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"We start with a standard normal receiving-root seed $g_0$ and put\n$Z_{j+1}=\\mathsf G(g_j)$ recursively. The orthonormal child directions\ngive independent standard Gaussian lifts, which reproduce the scalar\nlaw. Since the finite spin $F_N$ depends only on those lifts, projection\ngives the exact rooted-energy identity\n$$\n\\begin{aligned}\n\\mathsf L^*F_N&=\\sum_{j=0}^N\\mathbb E[F_NZ_{j+1}]g_j,\\\\\n\\mathcal E(F_N):=\\operatorname{Re}\\langle F_N,\\mathsf L F_N\\rangle\n&=\\sum_{j=0}^N\\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}]\n =S_{T,N}.\n\\end{aligned}                                     \\tag{12}\n$$","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":46513},{"anchor":"qaoa-boundary","at":"b46831-47100","end_byte":47100,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"statement","markdown":"$$\n\\begin{aligned}\n\\mathsf L^*F_N&=\\sum_{j=0}^N\\mathbb E[F_NZ_{j+1}]g_j,\\\\\n\\mathcal E(F_N):=\\operatorname{Re}\\langle F_N,\\mathsf L F_N\\rangle\n&=\\sum_{j=0}^N\\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}]\n =S_{T,N}.\n\\end{aligned}                                     \\tag{12}\n$$","note":"Not in the Lean: it has no rooted operators or energy. Only the sum S_{T,N} exists there, as shiftedCoefficientSum.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":46831},{"anchor":"qaoa-boundary","at":"b47101-47164","end_byte":47164,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"statement","markdown":"Independence of the receiving-root seed removes the $j=0$ term.","note":"Not in the Lean: no seed appears. coefficientA sets the j=0 term to zero by definition.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":47101}]},"circuit.sampled_finite_instance_optimum":{"lean":[],"short":"Measured configuration near instance maximum","statements":[{"anchor":"sampled-finite-instance","at":"b226418-226798","end_byte":226798,"heading_id":"what-is-averaged-and-a-fixed-word-probability-consequence","lean":[],"lean_source":"record","markdown":"**Comparison with the same finite instance.** The concentration of the\nfinite maximum in §11.1.2 strengthens this to a near-optimal sampled\nconfiguration. For every $\\varepsilon>0$, there is one finite\ndeterministic word $W$, independent of $n$ and $J$, such that\n$$\n\\Pr_{J,\\Sigma}\\!\\left\\{\nH_{n,J}(\\Sigma)\\ge\\max_\\sigma H_{n,J}(\\sigma)-\\varepsilon n\n\\right\\}\\longrightarrow1.\n$$","paper":"long","role":"statement","section":"11.2 What is averaged, and a fixed-word probability consequence","start_byte":226418},{"anchor":"sampled-finite-instance","at":"b226799-227629","end_byte":227629,"heading_id":"what-is-averaged-and-a-fixed-word-probability-consequence","lean":[],"lean_source":"record","markdown":"To see this, choose the word with $v_W\\ge P_*-\\varepsilon/4$ and\nwrite $r_n=\\mathbb E(E_n-v_W)^2\\to0$ and\n$M_n=\\max_\\sigma H_{n,J}(\\sigma)/n$. The convergence\n$\\mathbb E M_n\\to P_*$ makes its bias at most $\\varepsilon/4$\nfor all sufficiently large $n$. The Gaussian coefficient-vector bound\nfrom §11.1.2 and Chebyshev's inequality for $E_n$ then give, for $n\\ge2$,\n$$\n\\Pr_J\\{M_n>P_*+\\varepsilon/2\\}\n \\le2\\exp\\!\\left\\{-\\frac{n^2\\varepsilon^2}{16(n-1)}\\right\\},\n\\qquad\n\\Pr_{J,\\Sigma}\\{E_n<v_W-\\varepsilon/4\\}\n \\le\\frac{16r_n}{\\varepsilon^2}.\n$$\nOutside these two events, $M_n-E_n\\le\\varepsilon$. Their union\ntherefore has probability tending to zero. The maximum and the measured\nenergy use the same disorder, and the union bound needs no independence.\n\nOne fresh preparation and measurement suffices for this asymptotic\nstatement.","paper":"long","role":"derivation","section":"11.2 What is averaged, and a fixed-word probability consequence","start_byte":226799},{"anchor":"qaoa-boundary","at":"b50885-51306","end_byte":51306,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"**Corollary 6 (a measured configuration on the same instance).**\nFor every $\\varepsilon>0$ there is a finite ordinary QAOA word\n$W_\\varepsilon$, chosen independently of $n$ and $J$, such that one\ncomputational-basis measurement $\\Sigma$ satisfies\n$$\n\\Pr_{J,\\Sigma}\\!\\left\\{\n\\max_\\sigma H_{n,J}(\\sigma)-H_{n,J}(\\Sigma)\\leq\\varepsilon n\n\\right\\}\\longrightarrow1\n\\qquad(n\\to\\infty).                               \\tag{14}\n$$","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":50885},{"anchor":"qaoa-boundary","at":"b51308-52073","end_byte":52073,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"**Proof.** For a fixed word, put\n$Y_n=H_{n,J}(\\Sigma)/n$ and\n$K_n=\\max_\\sigma H_{n,J}(\\sigma)/n$.\n[Farhi, Goldstone, Gutmann and Zhou, §4, equations (13)–(22)][FGGZ]\nprove\n$$\n\\mathbb E_{J,\\Sigma}Y_n\\to v(W),\\qquad\n\\mathbb E_{J,\\Sigma}Y_n^2\\to v(W)^2.\n$$\nThe second moment is $\\mathbb E_J\\langle(C_{n,J}/n)^2\\rangle$,\nso it includes measurement variance as well as disorder fluctuations.\nThus $Y_n\\to v(W)$ in joint $L^2$. Gaussian concentration of the\nmaximum and its defining mean limit give $K_n\\to P_*$ in probability.\nChoose $W$ with $P_*-v(W)<\\varepsilon/3$. With probability tending\nto one, both $K_n-P_*<\\varepsilon/3$ and\n$v(W)-Y_n<\\varepsilon/3$, which proves (14). The two energies use\nthe same $J$, and the argument needs no independence between them.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":51308},{"anchor":"qaoa-boundary","at":"b51733-51832","end_byte":51832,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"The mean limit only. Concentration of the maximum is outside the Lean.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"part"}],"lean_source":"statement","markdown":"Gaussian concentration of the\nmaximum and its defining mean limit give $K_n\\to P_*$ in probability.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":51733}]},"circuit.seed_packets":{"lean":[],"short":"Small packet and tail estimates","statements":[{"anchor":"qaoa-seed","at":"b145024-146090","end_byte":146090,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"Fix $b>0$ and $0<r<1$, and put\n$$\nR_*=e^{-i\\pi X/4},\\qquad \\delta_r=\\arcsin r,\\qquad\n\\gamma_r=b/r,\\qquad e=Z\\mathrm{vac},\\qquad\n\\ell_r=Y e^{-2i\\gamma_r Zq(e)}\\mathrm{vac}.\n$$\nApply, in chronological order, cost $\\gamma_r$, mixer $\\delta_r$,\ncost $-\\gamma_r$ and mixer $\\pi/4$. Just before the inverse cost the\ninsertion is $v_{2,r}=\\cos(2\\delta_r)e+\\sin(2\\delta_r)\\ell_r$. Define\n$$\nd_{0,r}=\\gamma_r(v_{2,r}-e)\n       =-2br\\,e+2b\\sqrt{1-r^2}\\,\\ell_r,\n\\qquad W_Z(f)=e^{iZq(f)}.\n$$\nExpanding the first mixer and using the Weyl identity gives the exact split\n$$\nu_r=a_{0,r}U_{O,r}+a_{1,r}U_{S,r},\\qquad\nU_{O,r}=R_*W_Z(d_{0,r}),\\quad\nU_{S,r}=R_*XW_Z(-2\\gamma_re-d_{0,r}),\n$$\nwhere\n$$\na_{0,r}=\\cos\\delta_r\\,e^{i\\vartheta_r},\\quad\n a_{1,r}=-ir e^{-i\\vartheta_r},\\quad\n\\vartheta_r=\\gamma_r^2\\operatorname{Im}\\langle v_{2,r},e\\rangle.\n$$\nBoth component operators are unitary. The special component\n$\\psi_{S,r}=a_{1,r}U_{S,r}\\mathrm{vac}$ therefore has norm exactly $r$.\nIts normalized contribution to the insertion is\n$$\nw_r=u_r^*Z\\psi_{S,r}/r,\\qquad \\|w_r\\|=1.\n\\tag{8.1}\n$$","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":145024},{"anchor":"qaoa-seed","at":"b146531-147343","end_byte":147343,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"The source's high-direction lemma gives that $\\ell_r,w_r$ are centred\nunit vectors, become mutually orthogonal, and escape every fixed\noccupation sector. Splitting a uniformly bounded family of occupation\nmoments at a fixed sector cutoff shows that its pairings with these\ndirections vanish: first let $r\\downarrow0$, then send the cutoff to\ninfinity. The special component must nevertheless be retained because\nthe later echo magnifies its insertion contribution by $1/r$.\n[OpenAI (2026b), §5.2, exact initialization and high directions][q-seed-packets].\n\nA packet has the form $W(d_r/r)\\eta_r$, where the scaled centre $d_r$\nis bounded and the tail $\\eta_r$ has uniformly bounded occupation\nmoments. Packets with centres separated by a fixed positive amount\nhave overlap smaller than every fixed power of $r$.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":146531}]},"circuit.seed_removal":{"lean":[],"short":"Finite ordinary approximation of seeds","statements":[{"anchor":"qaoa","at":"b109933-110368","end_byte":110368,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"**Seed removal.** Fix that finite seeded word and an energy tolerance.\nReuse the same Gaussian at every occurrence on a vertex. Packet control and\nthe one-particle comparison then produce one finite ordinary word with the\nrequired energy error. The [seed-removal theorem][q-seed-statement] supplies\nan energy conclusion, which is stronger than convergence of seed marginals,\nand so makes the ordinary-word transfer in §8.5 applicable.","paper":"long","role":"statement","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":109933},{"anchor":"qaoa-seed","at":"b157278-157698","end_byte":157698,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"The choices are therefore nested. We make $b$ small to control\ninitialization, perturb the fixed target angles to ensure visibility and\nchoose all finite probe lists. Finally we choose $r$ small. A value of\n$r$ fixes every large angle and every finite product. The theorem\nallows those angles to be large or negative. It makes no claim of\nuniform bounds as the target word length grows or as the tolerance\ntends to zero.","paper":"long","role":"explanation","section":"8.4 Removing the seed controls without changing their joint law","start_byte":157278},{"anchor":"qaoa-boundary","at":"b48900-49010","end_byte":49010,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"Removing those\ncontrols produces an ordinary word, whose limiting energy is $v(W)$ by\nthe fixed-word transfer.","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":48900}]},"circuit.seeded_synthesis":{"lean":[],"short":"Seeded synthesis with helper loss","statements":[{"anchor":"qaoa","at":"b109448-109765","end_byte":109765,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"**Helper contraction and synthesis.** Fix a bounded, finite, type-excluded\nodd formula, zero on helpers, and positive helper masses. The contraction\nconverges smoothly at every cut and for the specified responses. For each\n$\\eta>0$, finite seeded synthesis gives\n$$\n\\mathcal E(v)\\ge\\mathcal E(m)-2\\eta-2\\sqrt{p_H}.\n$$","paper":"long","role":"statement","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":109448},{"anchor":"qaoa-synthesis","at":"b143449-143940","end_byte":143940,"heading_id":"seeded-synthesis-helper-contraction-and-nested-responses","lean":[],"lean_source":"record","markdown":"If the resulting insertion $v$ obeys\n$\\|1_{H^c}(v-v_{\\rm cl})\\|<\\eta$, seed commutation and unitarity give\n$\\|1_Hv\\|=\\sqrt{p_H}$. First omit the helper part of $v$, at energy cost at most\n$2\\sqrt{p_H}$. Compare the two insertions projected off helpers.\nTheir norm difference is below $\\eta$, so their energies differ by\nat most $2\\eta$. The projected classical insertion still has energy\n$\\mathcal E(m)$ because $m=0$ on helpers. Thus\n$$\n|\\mathcal E(v)-\\mathcal E(m)|\\le2\\eta+2\\sqrt{p_H}.\n$$","paper":"long","role":"statement","section":"8.3 Seeded synthesis: helper contraction and nested responses","start_byte":143449},{"anchor":"qaoa-boundary","at":"b48764-48899","end_byte":48899,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"This finite formula is\nthen approximated in energy using temporary Gaussian controls,\nwith the same sitewise seeds throughout the word.","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":48764}]},"circuit.selective_accuracy":{"lean":[],"short":"Selective pulse accuracy","statements":[{"anchor":"qaoa-seed","at":"b149337-149718","end_byte":149718,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"The separate accuracy\nlemma then follows the actual insertion drift\nthrough every probe prefix. Its ordinary-component error is $o(r)$, and\nthe error on the normalised special component is $o(1)$ and hence\nalso contributes $o(r)$ to the state. These bounds are uniform over\nthe bounded-tail families needed for insertions, and their constants\nrefer to the chosen finite pulse list.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":149337},{"anchor":"qaoa-seed","at":"b150183-151231","end_byte":151231,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"The error scale needed below follows from the source's quantitative\nprobe estimate. For a fixed finite pulse list, separated packet centres\nand bounded designated tail moments, write $\\varepsilon_r=o(r)$ for\nthe discarded vacuum remainder. At each probe prefix $j$, with current\ninsertion $v_{P,r}^{(j)}$ and huge-cost direction $v_r$, the proof gives\n$$\n\\max_j|\\operatorname{Im}\\langle v_{P,r}^{(j)},v_r\\rangle|\n \\le C_M(r^2+\\varepsilon_r+r^M).\n$$\nChoose a fixed $M>1$. The right side is $o(r)$: within each ordinary\npacket the relevant Hermitian expectation is real, distinct packet\ncentres have overlap $O(r^M)$, and terms containing only special\npackets have size $O(r^2)$. Unitarity carries the norm remainder, which\nreceives no unproved occupation bound. The exact zero clocks then\ncancel the first-order displacement on ordinary packets, leaving\n$o(r)$ error there. The prescribed action on a normalized special\npacket has $o(1)$ error, multiplied by its $O(r)$ amplitude.\nThese are the unequal accuracy requirements that the next echo uses.","paper":"long","role":"derivation","section":"8.4 Removing the seed controls without changing their joint law","start_byte":150183}]},"circuit.selective_control":{"lean":[],"short":"Zero-clock selective rotations","statements":[{"anchor":"qaoa-seed","at":"b147746-148819","end_byte":148819,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"For the finite list of nonzero special centres $h_j\\in\\mathbb R^2$,\nthe source constructs the same rotation $e^{-i\\theta X}$ at every\ncentre, with *zero clocks*: the signed pulse durations on each of two\nperpendicular axes in the $ZY$ plane sum exactly to zero. Choose the\naxes so that $c_{j,a}=h_j\\cdot P_a\\ne0$, and the square pairs\n$(c_{j,1}^2,c_{j,2}^2)$ are distinct except for equal or opposite centres.\nBrackets have zero clock components and generate $iX$ coefficients\n$c_{j,1}c_{j,2}F(c_{j,1}^2,c_{j,2}^2)$. Polynomial interpolation makes\nall these coefficients one. Equal or opposite centres have the same\nproduct, so they cause no obstruction to this common rotation.\n\nThis does not give arbitrary independent rotations: opposite centres\nobey $U_{-h}=XU_hX$. Exact finite reachability of the required common\nrotation follows by taking finitely many reachable conjugates spanning\nthe generated Lie algebra. Their product map contains an identity\nneighbourhood, and subdivision reaches the whole one-parameter path.\nClosure alone would not give exact cancellation.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":147746}]},"circuit.sign_smoothing":{"lean":[],"short":"Finite sign smoothing","statements":[{"anchor":"qaoa-regularization","at":"b123963-124182","end_byte":124182,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"We fix $T,N$ first. The independent continuous rounding coordinate makes\nthe event at the sign discontinuity null. Bounded continuous odd\napproximations to that sign therefore converge in $L^2$ by dominated\nconvergence.","paper":"long","role":"explanation","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":123963}]},"circuit.singular_echo":{"lean":[],"short":"Cost-flip echo keeps finite signal","statements":[{"anchor":"qaoa-seed","at":"b153199-154131","end_byte":154131,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"Consider a forward cost $t/r$, a selective flip of the special\ncomponent, the reverse cost $-t/r$, and a selective undoing of that\nflip. The packet invariant transports the special component so that\nits contribution to the insertion changes sign. For the actual\ninsertion directions before and after the first flip, it gives\n$$\nv'_r-v_r=-2r\\,w_r+o(r).\n$$\nThe Weyl product of the two costs is formed with these actual\ndirections before making any approximation. Their large centres\ncancel, leaving a bounded displacement\n$$\n\\frac tr(v'_r-v_r)=-2t\\,w_r+o(1).\n$$\nApproximating each direction by a low component with merely $O(r)$\nerror before this subtraction would leave an uncontrolled order-one\nerror. The remaining scalar Weyl phase is independent of the root\nspin and cancels in insertions, so it need not itself converge. The\nsecond flip restores the transported special component so that the\nsame construction can be used again.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":153199}]},"circuit.upper_bound_padding":{"lean":[],"short":"Upper bound and identity padding","statements":[{"anchor":"qaoa","at":"b107195-107378","end_byte":107378,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"A cost expectation is bounded above by its\nmaximum eigenvalue, and zero-angle padding makes $Q_p$ nondecreasing.\nIt therefore suffices to construct one finite word for every accuracy.","paper":"long","role":"statement","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":107195},{"anchor":"qaoa-assembly","at":"b163293-163444","end_byte":163444,"heading_id":"fixing-the-word-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"Since $Q_p\\le P_*$ and\nzero padding makes $Q_p$ nondecreasing, this proves the source's\nthermodynamic-first conclusion from its stated analytic inputs.","paper":"long","role":"statement","section":"8.6 Fixing the word before the thermodynamic limit","start_byte":163293}]},"circuit.visibility":{"lean":[],"short":"Finite visibility choices","statements":[{"anchor":"qaoa-seed","at":"b156042-157165","end_byte":157165,"heading_id":"removing-the-seed-controls-without-changing-their-joint-law","lean":[],"lean_source":"record","markdown":"**Why the selective rotations are available at every required stage.**\nThe limiting projection of each special centre must be nonzero. The\nsource imposes finitely many visibility conditions on the angles of\nthe formal seeded word. At each relevant standard stage they have\nform\n$$\n-2\\operatorname{Im}\\langle v_Z^{\\rm form},\n        -2bz\\,e+yD^{\\rm form}\\rangle\\ne0,\n\\qquad z,y\\in\\{\\pm1\\},\n$$\nwhere $e=Z\\mathrm{vac}$ and $D^{\\rm form}$ accumulates echoes whose\nreverse costs have already occurred. These functions are real\nanalytic by the finite Weyl formulas. Setting all target angles to\nzero, while retaining initialization, gives\n$4bz e^{-8b^2}\\ne0$, so none is identically zero. An arbitrarily small\nangle perturbation avoids their finite union of zero sets and the\nzero echo-time hyperplanes. The huge-cost stages need a separate\ntransverse-projection estimate, after which their limiting projection pairs\nalso have a nonzero coordinate.\n\nFinite pulse lists can now be chosen\nsuccessively from limits already established at earlier stages.\nNo list is chosen from a limit that assumes its own successful\nimplementation.","paper":"long","role":"statement","section":"8.4 Removing the seed controls without changing their joint law","start_byte":156042}]},"circuit.weighted_finite_horizon":{"lean":[],"short":"Corrected weighted finite-tree estimate","statements":[{"anchor":"weighted-tree","at":"b213549-214244","end_byte":214244,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"**A finite estimate with the required hypotheses.** The source argument\ndoes provide a usable replacement. Fix $\\delta>0$, $L\\ge1$ and\n$L\\delta<1$. Allow the coefficients $a_k,b_k$ to depend on degree, but\nrequire bounds independent of $k$ at the finitely many mesh times:\n$|a_k|\\le K$ with $K\\ge1$, and spatial Lipschitz constants $L_a,L_b$\nfor $a_k,b_k$. Use the original initialization and the same normalized\nfunction for node and directed-edge coefficients, with\n$\\mathbb E(A_{i\\to j}^r)^2=1$. Then\n$$\n\\left|k\\mathbb E[z_o^Lz_v^L]\n -2\\sqrt{k-1}\\,\\delta\\sum_{r=1}^{L-1}\\mathbb E B_o^r\\right|\n \\le C(K,L_a,L_b,\\delta,L).\n$$\nThe constant is independent of $k$, and the sum is empty when $L=1$.","paper":"long","role":"statement","section":"11.1 The separate positive-temperature route","start_byte":213549},{"anchor":"weighted-tree","at":"b215989-217482","end_byte":217482,"heading_id":"the-separate-positive-temperature-route","lean":[],"lean_source":"record","markdown":"For $s=\\ell-1$, the adjacent-time summand of Proposition 2.1 is\n$$\nQ_s=\\mathbb E[A_{v\\to o}^{s-1}(u_{v\\to o}^s)^2B_v^{s-1}B_o^s].\n$$\nLet $\\mathcal G_{i\\to j}^{r,+}$ be the labels within distance $r$\nof $i$ in its cavity after removing edge $ij$. Condition on\n$\\mathcal G_{v\\to o}^{s-1,+}\\vee\\mathcal G_{o\\to v}^{s,+}$.\nThe three coefficient factors are measurable there: their vertex balls\nsplit into these two disjoint cavity balls. The newest layer driving\n$u_{v\\to o}^s$ is outside this field. Adding the initial label at $v$\nto the source's minus field does not change the conditional law, since\nthat message and its variance exclude that label. Conditioning therefore replaces $(u_{v\\to o}^s)^2-1$ by $(\\tau_{v\\to o}^s)^2-1$.\nIts contribution is bounded by $K^3\\sqrt{V_s/D}$. Replacing the two\nnode coefficients by directed-edge coefficients then costs at most\n$K^2L_a(\\sqrt{d_{s-1}}+\\sqrt{d_s})/\\sqrt k$.\nThe two directed-edge trees are independent, and exact normalization\nreduces their product expectation to $\\mathbb E A_{o\\to v}^s$.\nOne final comparison with $\\mathbb E B_o^s$ proves\n$$\n|Q_s-\\mathbb E B_o^s|\\le\\frac{C_s}{\\sqrt D},\\quad\nC_s=K^3\\sqrt{V_s}+K^2L_a\\sqrt{d_{s-1}}\n                         +(K^2+1)L_a\\sqrt{d_s}.\n$$\nFor the same-time remainder in that proposition,\n$|R_\\ell|\\le K^{2\\ell+2}$ by boundedness and Cauchy–Schwarz. Hence\nthe asserted finite bound holds, for example, with\n$$\nC=2\\delta\\sum_{s=1}^{L-1}C_s\n                  +\\delta\\sum_{\\ell=1}^LK^{2\\ell+2}.\n$$","paper":"long","role":"derivation","section":"11.1 The separate positive-temperature route","start_byte":215989}]},"circuit.weighted_unitary_calculus":{"lean":[],"short":"Unitary products and commuting substitution","statements":[{"anchor":"qaoa-regularization","at":"b130362-131030","end_byte":131030,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"Let $K_D(x)$ be a smooth unitary family with $K_D(0)=1$, uniformly\ncontrolled derivatives and adjoint derivatives that converge smoothly\nin the stated weighted topology to those of a limiting family $K$,\nand centred first vacuum derivatives. Write\n$K_\\alpha=\\partial_\\alpha K(0)=iT_\\alpha$ and\n$K_{\\alpha\\lambda}=\\partial_\\alpha\\partial_\\lambda K(0)$.\nFor a fixed finite parameter list $a$, the source proves\n$$\nK_D(a/\\sqrt D)^{\\otimes D}\\longrightarrow\n\\exp\\!\\left\\{\n i\\sum_\\alpha a_\\alpha q(T_\\alpha\\mathrm{vac})\n +\\frac12\\sum_{\\alpha,\\lambda}a_\\alpha a_\\lambda\n \\left\\langle K_{\\alpha\\lambda}\n             +\\tfrac12\\{T_\\alpha,T_\\lambda\\}\\right\\rangle_0\n\\right\\}.\n$$","paper":"long","role":"statement","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":130362},{"anchor":"qaoa-regularization","at":"b131740-132639","end_byte":132639,"heading_id":"finite-regularisation-and-the-calculus-needed-for-compilation","lean":[],"lean_source":"record","markdown":"For a fixed finite list of strongly commuting self-adjoint signals\n$S_D=(S_{D,1},\\ldots,S_{D,b})$, their polynomials must converge smoothly,\nand their joint exponentials must converge strongly on smooth inputs and\nsatisfy, for every output weight $m$,\n$$\n\\|\\mathsf W_D^m e^{it\\cdot S_D}\\xi\\|\n \\le C_m(1+|t|)^{k_m}\\|\\mathsf W_D^{r_m}\\xi\\|,\n$$\nuniformly in $D$ and $t\\in\\mathbb R^b$. The coefficient family $F_D(x)$\nand its adjoint obey the smooth bounds above and converge with all their\njets on compact parameter sets. Each coefficient must commute with the\nsignals' spectral projections. For unbounded coefficients this means\ncommutation in the closed-operator sense. Seed-dependent substitution also\nrequires strong commutation with the receiving seed multiplication algebra.\nThese hypotheses permit the joint spectral and Fourier calculation of\n$F_D(S_D)$, including its derivatives and adjoints.","paper":"long","role":"statement","section":"8.2 Finite regularisation and the calculus needed for compilation","start_byte":131740}]},"full_support":{"lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, conditional on a minimiser: every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The five clauses: measure, diffusion, full support, strict increase, both moments for every associated diffusion.","plain":"The full-support conclusion","relation":"defines"}],"short":"Full support of SK minimisers","statements":[{"anchor":"gaps","at":"b71060-71391","end_byte":71391,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, conditional on a minimiser: every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The five clauses: measure, diffusion, full support, strict increase, both moments for every associated diffusion.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Theorem 5.1 (full support).** In the pure zero-field SK model with\n$\\xi(t)=t^2/2$, let $\\gamma\\in\\mathcal U$ minimize $\\mathcal P$. Then\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and $\\gamma$ is strictly increasing.\nFor every associated diffusion started at zero,\n$\\mathbb E u(t,X_t)^2=t$ and $\\mathbb E a(t,X_t)^2=1$ for\n$0\\le t<1$.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71060},{"anchor":"main-result","at":"b6273-6622","end_byte":6622,"heading_id":"the-model-and-the-result","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, conditional on a minimiser: every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The five clauses: measure, diffusion, full support, strict increase, both moments for every associated diffusion.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$\nFor the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6273},{"anchor":"full-support-statement","at":"b25262-25665","end_byte":25665,"heading_id":"the-full-support-theorem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, conditional on a minimiser: every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The five clauses: measure, diffusion, full support, strict increase, both moments for every associated diffusion.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Source theorem ([OpenAI (2026a), Theorem 1.1][support-manuscript]).**\nIn the pure, zero-field model with $\\xi(t)=t^2/2$, let $\\gamma$ be an\nadmissible integrable minimiser of $\\mathcal P$. Its Stieltjes measure\nhas full relative support on $[0,1)$, and $\\gamma$ is strictly increasing.\nFor every associated diffusion and every $0\\le t<1$,\n\n$$\n\\mathbb E[u(t,X_t)^2]=t,\\qquad\n\\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"expository","role":"statement","section":"4. The full-support theorem","start_byte":25262},{"anchor":"full-support-statement","at":"b25782-25856","end_byte":25856,"heading_id":"the-full-support-theorem","lean":[{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0. The role of ξ''=1 has no single statement: Lean fixes unit noise throughout.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"clause":"value W γ 0 0","end_line":60,"line":59,"lines":[60,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Zero field is built in: state zero at time zero, unit noise.","plain":"The Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"Zero field supplies\n$q(0)=0$, and $\\xi''=1$ enters the curvature estimate.","paper":"expository","role":"explanation","section":"4. The full-support theorem","start_byte":25782},{"anchor":"full-support-statement","at":"b25857-25941","end_byte":25941,"heading_id":"the-full-support-theorem","lean":[],"lean_source":"statement","markdown":"Nonzero field\nand general mixed-spin covariance functions lie outside the statement.","note":"Not in the Lean: a remark on scope. Lean's model has no field and no covariance parameter to vary.","paper":"expository","role":"explanation","section":"4. The full-support theorem","start_byte":25857},{"anchor":"targets","at":"b5504-5618","end_byte":5618,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model. hmin is the assumption; the first two clauses of the conclusion supply the measure and a diffusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[82,83],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Existence of a Stieltjes measure and of a diffusion.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The first result assumes an admissible minimizer. It constructs its\nStieltjes measure and an associated diffusion,","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":5504},{"anchor":"targets","at":"b5619-5696","end_byte":5696,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, taking hmin. Its conclusion has these clauses, for every Stieltjes measure.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[84,85],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Full support for every Stieltjes measure, then strict increase.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"proves\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and strict increase of $\\gamma$,","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":5619},{"anchor":"targets","at":"b5697-5794","end_byte":5794,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, taking hmin. The last clause of its conclusion gives both moments for every diffusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[86,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Both moments, for every diffusion of the coefficient.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"and obtains $\\mathbb E u(t,X_t)^2=t$ and\n$\\mathbb E a(t,X_t)^2=1$ for every associated diffusion.","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":5697},{"anchor":"gaps","at":"b71060-71273","end_byte":71273,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support and strict increase are two clauses of FullSupportConclusion, for every minimiser on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[84,85],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Support and strict increase.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Theorem 5.1 (full support).** In the pure zero-field SK model with\n$\\xi(t)=t^2/2$, let $\\gamma\\in\\mathcal U$ minimize $\\mathcal P$. Then\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and $\\gamma$ is strictly increasing.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71060},{"anchor":"gaps","at":"b71274-71391","end_byte":71391,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The last clause of FullSupportConclusion, for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[86,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Both moments, for each diffusion.","plain":"The full-support conclusion","relation":"defines"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Every diffusion is the constructed one, almost surely.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"For every associated diffusion started at zero,\n$\\mathbb E u(t,X_t)^2=t$ and $\\mathbb E a(t,X_t)^2=1$ for\n$0\\le t<1$.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71274},{"anchor":"gaps","at":"b82209-82484","end_byte":82484,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Gives the first sentence. The theorem stops there: no density and no statement that γ(0)=0.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"}],"lean_source":"statement","markdown":"The proof establishes full support, strict increase and the two\nmoment identities. Smooth absolute continuity, including the later\nconclusion $\\gamma(0)=0$, uses the source's further argument from\nsaturated consistency, and neither contact nor the exclusion of an\ninitial gap","paper":"long","role":"explanation","section":"5. Excluding terminal, interior and initial gaps","start_byte":82209},{"anchor":"density","at":"b165655-165796","end_byte":165796,"heading_id":"smoothness-of-the-minimizing-coefficient","lean":[],"lean_source":"statement","markdown":"We reconstruct this argument because it gives a further conclusion of the original paper, outside the selected full-support formal statement.","note":"Not in the Lean: a remark. FullSupportConclusion has no density clause.","paper":"long","role":"explanation","section":"9.1 Smoothness of the minimizing coefficient","start_byte":165655},{"anchor":"overview","at":"b488-726","end_byte":726,"heading_id":"a-long-reasoning-record","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model. Takes the minimiser as hypothesis hmin; the conclusion also gives strict increase and both moments.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[84,84],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Full support for each Stieltjes measure; the minimiser is the hypothesis of full_support.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"OpenAI's full-support manuscript proves that an admissible integrable\nminimizer of the zero-temperature Parisi functional for the pure zero-field\nSherrington–Kirkpatrick model has Stieltjes support equal to every overlap\nscale in [0,1).","paper":"long","role":"statement","section":"A long reasoning record","start_byte":488},{"anchor":"abstract","at":"b328-510","end_byte":510,"heading_id":"full-support-and-positive-density-in-the-zero-temperature-sk-model","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Takes a minimiser as the hypothesis hmin. Full support is one clause of its conclusion, beside strict increase and both moments.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ)","end_line":88,"line":81,"lines":[84,84],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The five clauses: measure, diffusion, full support, strict increase, both moments for every associated diffusion.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"For the pure, zero-field Sherrington–Kirkpatrick model at\nzero temperature, the Stieltjes measure of every integrable minimiser of\nthe Parisi functional has full support on $[0,1)$","paper":"short","role":"mention","section":"Full support and positive density in the zero-temperature SK model","start_byte":328},{"anchor":"model","at":"b2409-2505","end_byte":2505,"heading_id":"the-model-and-the-result","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Takes a minimiser as the hypothesis hmin. Its conclusion gives full support and strict increase. The positive density stated next is outside the Lean.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":88,"line":81,"lines":[84,85],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The clauses for full support and strict increase.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"We study the minimiser itself. Its increments charge every\ninterval of overlap scales below one,","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":2409},{"anchor":"gap-proof","at":"b16825-16879","end_byte":16879,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"In Lean those inputs are proved, so the theorem carries only hmin.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"}],"lean_source":"statement","markdown":"This proves Theorem 1 from the stated analytic inputs.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":16825}]},"lean.full_support":{"lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"This is the implementing declaration itself.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The proposition the theorem concludes.","plain":"The full-support conclusion","relation":"defines"}],"short":"Lean full-support theorem","statements":[{"anchor":"formal","at":"b239740-240834","end_byte":240834,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"This is the implementing declaration itself.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The proposition the theorem concludes.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The first formal statement assumes a Brownian system and an admissible\nminimizer. It supplies a Stieltjes measure, an associated diffusion, full\nrelative support and strict increase, and the two moment identities for\nevery associated diffusion. Its proof assembles contact, exclusion of a\nterminal plateau, curvature for positive and zero constant intervals,\nand uniqueness to transfer the moments between associated diffusions.\nThis describes the inspected proof's mathematical route, and we claim no\nsuccessful research-target replay.\n\nIts quantifiers are $\\forall W\\,\\forall\\gamma$, followed by the minimizer\nhypothesis, and it supplies both witnesses on the given Brownian system. The moment\nconclusion is universal over every diffusion satisfying the specified\nequation, and covers exactly $0\\le t<1$. The diffusion equation itself\nholds through time one, simultaneously for all those times outside\none null set. Controls are progressive for the natural filtration of\nBrownian increments and bounded by one pointwise. These are not\nadditional regularity or minimizer-existence assumptions.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":239740},{"anchor":"evidence-scope","at":"b52840-53043","end_byte":53043,"heading_id":"formal-verification","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"This is the implementing declaration itself.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The proposition the theorem concludes.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The\n[formal statement of full support][support-challenge] asserts the\nsupport and moment conclusions of Theorem 1, with an admissible\nminimiser as its hypothesis, and constructs the associated diffusion.","paper":"short","role":"explanation","section":"Formal verification","start_byte":52840},{"anchor":"source-record","at":"b62676-62869","end_byte":62869,"heading_id":"formal-verification-and-sources","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"This is the implementing declaration itself.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The proposition the theorem concludes.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"contains a [formal statement][support-interface] of the support and\nmoment identities of Section 4 for a minimiser, and Lean checks OpenAI's\n[proof of the corresponding theorem][support-proof].","paper":"expository","role":"explanation","section":"Formal verification and sources","start_byte":62676},{"anchor":"source-record","at":"b63152-63282","end_byte":63282,"heading_id":"formal-verification-and-sources","lean":[],"lean_source":"statement","markdown":"Neither statement covers\npositive density, the near-ground overlap theorem used in Section 6 or\nthe complete QAOA circuit theorem.","note":"Not in the Lean, as the sentence says: FullSupportConclusion and ScalarValueConclusion have no clause for any of these.","paper":"expository","role":"explanation","section":"Formal verification and sources","start_byte":63152},{"anchor":"targets","at":"b5795-5890","end_byte":5890,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Hypothesis IsMinimizer, conclusion FullSupportConclusion; no density appears.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"}],"lean_source":"statement","markdown":"The selected\nformal statement has exactly this conditional scope and asserts no smooth\ndensity.","paper":"long","role":"mention","section":"1. The SK model and the Parisi variational problem","start_byte":5795},{"anchor":"targets","at":"b5892-6114","end_byte":6114,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"This is the implementing declaration itself.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The proposition the theorem concludes.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Its [Lean implementation at the fixed source revision][support-main] implements the\n[separately stated formal challenge][support-type], and the\n[Comparator configuration][support-config] identifies the intended\ncomparison.","paper":"long","role":"mention","section":"1. The SK model and the Parisi variational problem","start_byte":5892},{"anchor":"formal","at":"b240835-240882","end_byte":240882,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The exact support statement: full_support.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"}],"lean_source":"statement","markdown":"See the\n[exact support statement][support-type]","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":240835},{"anchor":"verification","at":"b254119-254229","end_byte":254229,"heading_id":"source-checks-and-finite-calculation","lean":[{"end_line":92,"line":90,"lines":[90,92],"module":"ComparatorChallenges.SKFullSupport","name":"OAI.ZeroTemperatureSK.full_support","note":"The challenge statement ends in sorry.","plain":"Full support of SK minimisers","proof_lines":[92,92],"relation":"states"}],"lean_source":"statement","markdown":"The challenge is an\nadmitted specification, so its successful build and export are not a\nproof of the theorem.","paper":"long","role":"mention","section":"13. Source checks and finite calculation","start_byte":254119},{"anchor":"verification","at":"b255457-255865","end_byte":255865,"heading_id":"source-checks-and-finite-calculation","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The single root of that extraction. The build and traversal facts are in the receipt, outside the Lean.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"}],"lean_source":"statement","markdown":"A subsequent [source-bound observation of 26 October 2026](https://wcook04.github.io/lean-exposition-site/verification/full-support-build-extraction-observation-20261026.json)\nrecords a successful local Lake build of the Support implementation and\nv2 extraction of the single conditional full-support root. Its selected\ntraversal contains 1,674 declarations. The root reports no sorry, unsafe\nor partial flag","paper":"long","role":"mention","section":"13. Source checks and finite calculation","start_byte":255457},{"anchor":"verification","at":"b256095-256196","end_byte":256196,"heading_id":"source-checks-and-finite-calculation","lean":[],"lean_source":"statement","markdown":"This does not establish\nminimiser existence, positive density, the scalar target or QAOA convergence.","note":"Not in the Lean: none of these. FullSupportConclusion has no minimiser existence, density, scalar value or QAOA clause.","paper":"long","role":"explanation","section":"13. Source checks and finite calculation","start_byte":256095},{"anchor":"overview","at":"b1756-1882","end_byte":1882,"heading_id":"a-long-reasoning-record","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The first of the two formal statements.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"The second of the two formal statements.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"part"}],"lean_source":"statement","markdown":"The two\nformal statements concern full support and the scalar value consequences,\nand neither states QAOA circuit convergence.","paper":"long","role":"mention","section":"A long reasoning record","start_byte":1756},{"anchor":"evidence-scope","at":"b52720-52983","end_byte":52983,"heading_id":"formal-verification","lean":[{"end_line":92,"line":90,"lines":[90,92],"module":"ComparatorChallenges.SKFullSupport","name":"OAI.ZeroTemperatureSK.full_support","note":"The retained statement, whose proof is sorry here; Main.lean proves the theorem of the same name.","plain":"Full support of SK minimisers","proof_lines":[92,92],"relation":"states"},{"end_line":88,"line":81,"lines":[81,88],"module":"ComparatorChallenges.SKFullSupport","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"OpenAI's formalisation\nin [Lean][Lean4] with [mathlib][mathlib] contains two formal statements\nthat bear on this paper. The\n[formal statement of full support][support-challenge] asserts the\nsupport and moment conclusions of Theorem 1, with an admissible\nminimiser","paper":"short","role":"explanation","section":"Formal verification","start_byte":52720},{"anchor":"evidence-scope","at":"b52720-52839","end_byte":52839,"heading_id":"formal-verification","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The full-support implementation's main theorem. That it uses Lean and mathlib is a fact about its source, which no theorem states.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"},{"end_line":24,"line":16,"lines":[16,24],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"The scalar implementation's main theorem. Its toolchain is likewise a fact about the source.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"part"}],"lean_source":"statement","markdown":"OpenAI's formalisation\nin [Lean][Lean4] with [mathlib][mathlib] contains two formal statements\nthat bear on this paper.","paper":"short","role":"explanation","section":"Formal verification","start_byte":52720},{"anchor":"evidence-scope","at":"b53658-53712","end_byte":53712,"heading_id":"formal-verification","lean":[{"end_line":88,"line":81,"lines":[81,88],"module":"ComparatorChallenges.SKFullSupport","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The support statement: five clauses, none on density or circuits.","plain":"The full-support conclusion","relation":"defines"},{"end_line":171,"line":163,"lines":[163,170],"module":"ComparatorChallenges.SKValue","name":"OAI.SKValue.value_consequences","note":"The scalar statement: three clauses, none on density or circuits.","plain":"The scalar-value theorem","proof_lines":[171,171],"relation":"states"}],"lean_source":"statement","markdown":"Neither covers\nCorollary 4 or the convergence of QAOA.","paper":"short","role":"explanation","section":"Formal verification","start_byte":53658}]},"lean.value_consequences":{"lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"This is the implementing declaration itself. Its middle clause is finite_gaussian_target.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"clause":"ScalarValueConclusion W γ X","end_line":15,"line":13,"lines":[14,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First clause, for a minimiser and a supplied diffusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Third clause. It assumes only the minimiser and constructs the diffusion internally.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"short":"Lean scalar-value theorem","statements":[{"anchor":"formal-scalar","at":"b243180-244072","end_byte":244072,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"This is the implementing declaration itself. Its middle clause is finite_gaussian_target.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"clause":"ScalarValueConclusion W γ X","end_line":15,"line":13,"lines":[14,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First clause, for a minimiser and a supplied diffusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Third clause. It assumes only the minimiser and constructs the diffusion internally.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"lean_source":"statement","markdown":"The second formal statement has three clauses. Given a minimizer and an\nassociated diffusion, it establishes convergence of the finite-spin\nground-state values, a bounded martingale with a spin-valued terminal\nlimit, convergence of the gradient and the value identities. Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.\n\nThese are the declarations `scalar_value_target` and\n`finite_gaussian_target`. Their caller supplies the same minimizer and\nassociated diffusion. The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper. Those internal\ninputs are not additional hypotheses of the exported clauses. The\nconvergence clause establishes existence of the finite-spin limit.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":243180},{"anchor":"formal-parisi-clause","at":"b245775-246152","end_byte":246152,"heading_id":"the-separate-scalar-statement","lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"end_line":24,"line":16,"lines":[22,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its third clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"The third declaration, `ground_state_parisi_target`, assumes a minimizer\nalone and concludes $P_*=\\mathcal P(\\gamma)$. Its proof chooses\n$X=\\texttt{diffusionProcess}\\ W\\ \\gamma$ and proves the diffusion\npredicate for that process before applying the equality with a supplied\ndiffusion. No diffusion is required from the caller, and none of these\nclauses constructs a minimizer.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":245775},{"anchor":"evidence-scope","at":"b53129-53437","end_byte":53437,"heading_id":"formal-verification","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"This is the implementing declaration itself. Its middle clause is finite_gaussian_target.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"clause":"ScalarValueConclusion W γ X","end_line":15,"line":13,"lines":[14,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First clause, for a minimiser and a supplied diffusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Third clause. It assumes only the minimiser and constructs the diffusion internally.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"lean_source":"statement","markdown":"[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses, and its third\nclause identifies the ground-state and Parisi values from a minimiser,\nconstructing the diffusion internally.","paper":"short","role":"explanation","section":"Formal verification","start_byte":53129},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"This is the implementing declaration itself. Its middle clause is finite_gaussian_target.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"clause":"ScalarValueConclusion W γ X","end_line":15,"line":13,"lines":[14,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First clause, for a minimiser and a supplied diffusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Third clause. It assumes only the minimiser and constructs the diffusion internally.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"explanation","section":"Formal verification and sources","start_byte":62987},{"anchor":"targets","at":"b6808-6971","end_byte":6971,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"This is the implementing declaration itself. Its middle clause is finite_gaussian_target.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"states"},{"clause":"ScalarValueConclusion W γ X","end_line":15,"line":13,"lines":[14,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First clause, for a minimiser and a supplied diffusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Third clause. It assumes only the minimiser and constructs the diffusion internally.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"lean_source":"statement","markdown":"The scalar clauses are in a [separate Lean implementation][value-main], with their\nown [formal challenge][value-type] and [Comparator configuration][value-config].","paper":"long","role":"mention","section":"1. The SK model and the Parisi variational problem","start_byte":6808},{"anchor":"evidence-scope","at":"b53118-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":171,"line":163,"lines":[163,171],"module":"ComparatorChallenges.SKValue","name":"OAI.SKValue.value_consequences","note":"Three clauses; the retained statement ends in sorry and Main.lean proves it.","plain":"The scalar-value theorem","proof_lines":[171,171],"relation":"states"}],"lean_source":"statement","markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"explanation","section":"Formal verification","start_byte":53118}]},"qaoa_thermodynamic_first":{"lean":[],"short":"QAOA approaches the SK optimum","statements":[{"anchor":"qaoa","at":"b106762-107075","end_byte":107075,"heading_id":"from-the-scalar-formula-to-one-ordinary-qaoa-word","lean":[],"lean_source":"record","markdown":"The source target is $\\lim_{p\\to\\infty}Q_p=P_*$. Its needed lower\nbound has the literal quantifiers\n$$\n\\forall\\varepsilon>0\\quad\\exists p\\in\\mathbb N,\n\\boldsymbol\\gamma,\\boldsymbol\\beta\\in\\mathbb R^p:\n\\quad\\lim_{n\\to\\infty}\\frac1n\\mathbb E_J\n\\langle\\psi_{n,p,J}|C_{n,J}|\\psi_{n,p,J}\\rangle\n\\ge P_*-\\varepsilon.\n$$","paper":"long","role":"statement","section":"8. From the scalar formula to one ordinary QAOA word","start_byte":106762},{"anchor":"qaoa-boundary","at":"b41540-41778","end_byte":41778,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"[OpenAI (2026b), Theorem 1.1][Q] proves that for\nevery $\\varepsilon>0$ there is one finite $W$ with\n$v(W)\\geq P_*-\\varepsilon$, whose depth and angles depend on\n$\\varepsilon$ but not on $n$ or $J$. The instance still determines\n$C_{n,J}$.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":41540},{"anchor":"qaoa-boundary","at":"b49775-50138","end_byte":50138,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"Near-optimality follows from the decomposition\n$$\nP_*-v(W)\\leq\n (P_*-I_T)+|I_T-S_{T,N}|+|S_{T,N}-v(W)|.            \\tag{13}\n$$\nChoose $T$ so that $1-T<\\varepsilon/3$, then $N$ so that the\nsecond term is below $\\varepsilon/3$, and finally apply Proposition 5\nwith $\\eta<\\varepsilon/3$. These finite choices give\n$v(W)>P_*-\\varepsilon$ before $n$ tends to infinity.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":49775},{"anchor":"qaoa-boundary","at":"b61412-61971","end_byte":61971,"heading_id":"further-reading","lean":[],"lean_source":"record","markdown":"The separate circuit theorem in [OpenAI (2026b)][qaoa-manuscript]\nresolves the eventual-optimality conjecture for the Quantum Approximate\nOptimization Algorithm (QAOA) in\n[Basso–Farhi–Marwaha–Villalonga–Zhou, §7, equation (7.2)][bfmvz].\nThe *Long Reasoning Record*, <span class=\"pdf-companion\" data-pdf-target=\"../long/reasoning-record.pdf\" data-pdf-anchor=\"qaoa-construction\">[§8](../long/reasoning-record.md#qaoa-construction)</span>, develops its circuit construction and\nthe order of limits. It requires additional arguments beyond full support.","paper":"expository","role":"mention","section":"Further reading","start_byte":61412},{"anchor":"dependencies","at":"b8939-9059","end_byte":9059,"heading_id":"outline-of-the-proof","lean":[],"lean_source":"record","markdown":"These branches show\nwhy the two selected formal statements fall short of a formal proof of the\ncomplete circuit theorem.","note":"No Lean declaration here mentions a circuit.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":8939},{"anchor":"qaoa-boundary","at":"b49775-49901","end_byte":49901,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The first two terms tend to zero by these clauses. The third is outside the Lean.","plain":"The finite-Gaussian conclusion","relation":"part"}],"lean_source":"statement","markdown":"Near-optimality follows from the decomposition\n$$\nP_*-v(W)\\leq\n (P_*-I_T)+|I_T-S_{T,N}|+|S_{T,N}-v(W)|.            \\tag{13}\n$$","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":49775},{"anchor":"qaoa-boundary","at":"b49902-50000","end_byte":50000,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Lean gives both limits, without the rate 1−T.","plain":"The finite-Gaussian conclusion","relation":"part"}],"lean_source":"statement","markdown":"Choose $T$ so that $1-T<\\varepsilon/3$, then $N$ so that the\nsecond term is below $\\varepsilon/3$,","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":49902}]},"scalar.conditional_fixed_strip":{"lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Exactly this lemma, with analytic bounds and the unit moment as explicit hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"states"}],"short":"Conditional fixed-strip implication","statements":[{"anchor":"formal-scalar","at":"b243747-244001","end_byte":244001,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":72,"line":61,"lines":[61,72],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Derives the strip bounds from sourceStripRegularity and the unit moment from contact, then applies the helper.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"proof"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The helper, with the analytic bounds and the unit moment as explicit hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper. Those internal\ninputs are not additional hypotheses of the exported clauses.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243747}]},"scalar.conditional_rounding":{"lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Integrated form: correlation with any integrable function of the first N coordinates, for measurable U bounded by one.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"states"},{"clause":"∫ z, Real.sign (a+2*cdf (gaussianReal 0 1) z-1) ∂gaussianReal 0 1 = a","end_line":186,"line":181,"lines":[182,182],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gaussian_rounding_mean","note":"The one-variable fact: for |a| at most 1, the rounded sign has mean a.","plain":"Gaussian rounding keeps the mean","proof_lines":[183,186],"relation":"part"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"The sign F_N, with Real.sign, which is zero at zero.","plain":"The rounded finite spin","relation":"defines"}],"short":"Rounding preserves conditional mean","statements":[{"anchor":"gaussian","at":"b94269-94674","end_byte":94674,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Integrated form: correlation with any integrable function of the first N coordinates, for measurable U bounded by one.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"states"},{"clause":"∫ z, Real.sign (a+2*cdf (gaussianReal 0 1) z-1) ∂gaussianReal 0 1 = a","end_line":186,"line":181,"lines":[182,182],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gaussian_rounding_mean","note":"The one-variable fact: for |a| at most 1, the rounded sign has mean a.","plain":"Gaussian rounding keeps the mean","proof_lines":[183,186],"relation":"part"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"The sign F_N, with Real.sign, which is zero at zero.","plain":"The rounded finite spin","relation":"defines"}],"lean_source":"statement","markdown":"With $F_{\\rm G}$ the standard Gaussian distribution function, define\n$$\nF_N=\\operatorname{sgn}(u(T,Y_N)+2F_{\\rm G}(Z_{N+1})-1).\n$$\n$$\nA_j=\\mathbb E[F_Ng_j],\\quad B_i=\\mathbb E[F_NZ_i],\\quad A_0=0.\n$$\nThe extra coordinate is independent and its distribution function is uniform.\nFor any $v\\in[-1,1]$, the random sign of $v+2U-1$ has mean $v$.\nTherefore\n$\\mathbb E[F_N\\mid Z_1,\\ldots,Z_N]=u(T,Y_N)$ exactly.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94269},{"anchor":"gaussian","at":"b94675-95236","end_byte":95236,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Gives only that rounding preserves correlations. The L² gap 1−T and the deterministic-sign remark are not in the Lean.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"part"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"The sign F_N, with Real.sign, which is zero at zero.","plain":"The rounded finite spin","relation":"defines"}],"lean_source":"statement","markdown":"A deterministic sign of the gradient would not preserve this conditional\nmean. Rounding preserves correlations with functions of the first $N$\ncoordinates, which is exactly what the two coefficient vectors need.\nIt does not make the whole rounded spin close in $L^2$ to that conditional\nmean. Indeed, with $m_N=u(T,Y_N)$,\n$$\n \\mathbb E[(F_N-m_N)^2]=1-\\mathbb E m_N^2\\longrightarrow1-T.\n$$\nHere $m_N\\to u(T,X_T)$ in $L^2$ at fixed $T$ and full support gives\n$\\mathbb E u(T,X_T)^2=T$. The independent rounding variance disappears\nonly in the later endpoint limit.","paper":"long","role":"explanation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94675},{"anchor":"qaoa-boundary","at":"b42912-43302","end_byte":43302,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Integrated form: correlation with any integrable function of the first N coordinates, for measurable U bounded by one.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"states"},{"clause":"∫ z, Real.sign (a+2*cdf (gaussianReal 0 1) z-1) ∂gaussianReal 0 1 = a","end_line":186,"line":181,"lines":[182,182],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gaussian_rounding_mean","note":"The one-variable fact: for |a| at most 1, the rounded sign has mean a.","plain":"Gaussian rounding keeps the mean","proof_lines":[183,186],"relation":"part"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"The sign F_N, with Real.sign, which is zero at zero.","plain":"The rounded finite spin","relation":"defines"}],"lean_source":"statement","markdown":"With $\\Phi_{\\rm G}$ the standard normal distribution function, set\n$$\n\\begin{aligned}\nF_N&=\\operatorname{sgn}\\{u(T,Y_N)+2\\Phi_{\\rm G}(Z_{N+1})-1\\},\\\\\n\\mathbb E[F_N\\mid Z_1,\\ldots,Z_N]&=u(T,Y_N).\n\\end{aligned}\n$$\nThe second equality holds because $\\Phi_{\\rm G}(Z_{N+1})$ is\nuniform on $[0,1]$, so this random rounding preserves every correlation\nwith a function of the preceding coordinates.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42912},{"anchor":"gaussian","at":"b94269-94468","end_byte":94468,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"F_N.","plain":"The rounded finite spin","relation":"defines"},{"end_line":145,"line":141,"lines":[141,145],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientA","note":"A_j, with the index shifted and A₀=0.","plain":"Correlation A with normalised increments","relation":"defines"},{"end_line":149,"line":147,"lines":[147,149],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientB","note":"B_i.","plain":"Correlation B with Gaussian coordinates","relation":"defines"}],"lean_source":"statement","markdown":"With $F_{\\rm G}$ the standard Gaussian distribution function, define\n$$\nF_N=\\operatorname{sgn}(u(T,Y_N)+2F_{\\rm G}(Z_{N+1})-1).\n$$\n$$\nA_j=\\mathbb E[F_Ng_j],\\quad B_i=\\mathbb E[F_NZ_i],\\quad A_0=0.\n$$","paper":"long","role":"definition","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94269},{"anchor":"gaussian","at":"b94469-94674","end_byte":94674,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":186,"line":181,"lines":[181,182],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gaussian_rounding_mean","note":"The mean of the random sign.","plain":"Gaussian rounding keeps the mean","proof_lines":[183,186],"relation":"part"},{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Lean states the consequence: correlations with functions of the earlier coordinates agree.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"part"}],"lean_source":"statement","markdown":"The extra coordinate is independent and its distribution function is uniform.\nFor any $v\\in[-1,1]$, the random sign of $v+2U-1$ has mean $v$.\nTherefore\n$\\mathbb E[F_N\\mid Z_1,\\ldots,Z_N]=u(T,Y_N)$ exactly.","note":"Lean does not state the conditional expectation, only its effect on these correlations.","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94469},{"anchor":"gaussian","at":"b94754-94886","end_byte":94886,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"K ranges over functions depending only on earlier coordinates.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"part"}],"lean_source":"statement","markdown":"Rounding preserves correlations with functions of the first $N$\ncoordinates, which is exactly what the two coefficient vectors need.","paper":"long","role":"explanation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94754},{"anchor":"qaoa-boundary","at":"b42912-43123","end_byte":43123,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"F_N.","plain":"The rounded finite spin","relation":"defines"},{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"The conditional mean appears as an identity against every test function K of the earlier coordinates.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"states"}],"lean_source":"statement","markdown":"With $\\Phi_{\\rm G}$ the standard normal distribution function, set\n$$\n\\begin{aligned}\nF_N&=\\operatorname{sgn}\\{u(T,Y_N)+2\\Phi_{\\rm G}(Z_{N+1})-1\\},\\\\\n\\mathbb E[F_N\\mid Z_1,\\ldots,Z_N]&=u(T,Y_N).\n\\end{aligned}\n$$","paper":"short","role":"definition","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42912},{"anchor":"qaoa-boundary","at":"b43124-43204","end_byte":43204,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":143,"line":140,"lines":[140,141],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.standardGaussian_cdf_hasLaw_uniform","note":"The law of the Gaussian distribution function, under the standard normal, is uniform on (0,1].","plain":"Gaussian CDF transform is uniform","proof_lines":[142,143],"relation":"states"},{"end_line":179,"line":153,"lines":[153,154],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.uniform_rounding_mean","note":"","plain":"Uniform rounding keeps the mean","proof_lines":[155,179],"relation":"part"}],"lean_source":"statement","markdown":"The second equality holds because $\\Phi_{\\rm G}(Z_{N+1})$ is\nuniform on $[0,1]$,","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43124},{"anchor":"qaoa-boundary","at":"b43208-43302","end_byte":43302,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"Integrated form: correlation with any integrable function of the first N coordinates, for measurable U bounded by one.","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"states"}],"lean_source":"statement","markdown":"this random rounding preserves every correlation\nwith a function of the preceding coordinates.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43208}]},"scalar.constant_curvature":{"lean":[{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Support model, case c>0: curvatureMoment is q', so two derivatives give q'''. Strict positivity only. The scalar model proves its own curvature_convexOn.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Case c=0: the derivative of curvatureMoment is q'', positive on the interval. The identity for q'' is elsewhere.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The quantitative 1/1000 bound, multiplied through by c^2, for the strip expression at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"proof"}],"short":"Curvature on constant intervals","statements":[{"anchor":"curvature","at":"b43953-44349","end_byte":44349,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"Case c>0: the actual inequality, multiplied through by c², at a continuity point. stripThird is q''' on the interval.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Support model, case c>0: curvatureMoment is q', so two derivatives give q'''. Strict positivity only. The scalar model proves its own curvature_convexOn.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Case c=0: the derivative of curvatureMoment is q'', positive on the interval. The identity for q'' is elsewhere.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"Let $\\gamma\\in\\mathcal U$ be\nconstant with value $c$ on an open interval $I\\subset(0,1)$, and let $X$\nbe its diffusion started at zero. Then\n\n$$\n\\begin{cases}\nq'''(t)\\ge \\dfrac1{1000c^2}\n  \\mathbb E\\bigl[(c\\Phi_{xx}(t,X_t))^4\\bigr]>0,&c>0,\\\\[4pt]\nq''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0,&c=0,\n\\end{cases}\\qquad t\\in I.\n$$\n\nThus $q'$ is strictly convex for $c>0$ and strictly increasing for\n$c=0$.","paper":"long","role":"statement","section":"4. Strict curvature on a constant interval","start_byte":43953},{"anchor":"fixed-time-coercivity","at":"b69835-70384","end_byte":70384,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Case c=0: the derivative of curvatureMoment is q'', positive on the interval. The identity for q'' is elsewhere.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"Expands the q″ expression as an expectation; at c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":63,"line":58,"lines":[58,59],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualThirdSquare_pos","note":"Positivity of the expectation.","plain":"Second moment of ∂ₓ³Φ is positive","proof_lines":[60,63],"relation":"proof"},{"end_line":73,"line":46,"lines":[46,47],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.third_not_identically_zero","note":"The third derivative is nonzero somewhere.","plain":"∂ₓa is not identically zero","proof_lines":[48,73],"relation":"proof"}],"lean_source":"statement","markdown":"When $c=0$, the heat equation gives\n$q''(t)=\\mathbb E\\Phi_{xxx}(t,X_t)^2$ without division by $c$.\nStart the drift-free evolution at an earlier point of $I$. Its\nGaussian transition kernel gives the law of $X_t$ a positive\ndensity everywhere. If the expectation were zero, continuity would\nforce $\\Phi_{xxx}(t,\\cdot)\\equiv0$. Then $\\Phi(t,\\cdot)$ would\nbe an even quadratic polynomial. Its one-Lipschitz bound forces that\npolynomial to be constant, contradicting its bounded distance from\n$|x|$. Therefore $q''(t)>0$, giving strict increase of $q'$.","paper":"long","role":"derivation","section":"4.6 The zero interval as a separate case","start_byte":69835},{"anchor":"gap-proof","at":"b12934-13348","end_byte":13348,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The 1/1000 bound multiplied through by c², at a continuity point. The next tab identifies stripThird with q‴ on a plateau.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":46,"line":39,"lines":[39,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"On a plateau, q‴ is the strip moment stripThird.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Support model, case c>0: curvatureMoment is q', so two derivatives give q'''. Strict positivity only. The scalar model proves its own curvature_convexOn.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Case c=0: the derivative of curvatureMoment is q'', positive on the interval. The identity for q'' is elsewhere.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"**Lemma 2 (constant-interval curvature; [OpenAI (2026a),\nProposition 5.1][FS]).** Let $\\gamma\\in\\mathcal U$, with its\nassociated diffusion and moment $q$. If $\\gamma=c$ on an open\ninterval $I\\subset(0,1)$, called a plateau, then for $t\\in I$\n$$\nq'''(t)\\geq\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\quad\\text{if }c>0.                               \\tag{3}\n$$\nIf $c=0$, then $q''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0$.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":12934},{"anchor":"contact-and-gap","at":"b33656-33936","end_byte":33936,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The quantitative 1/1000 bound, multiplied through by c^2, for the strip expression at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":46,"line":39,"lines":[39,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Identifies stripThird with q''' on the interval.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Case c>0. curvatureMoment is q', so two derivatives give q'''. This declaration states strict positivity only.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"**Strict-curvature input ([OpenAI (2026a), Proposition 5.1][support-curvature]).**\nFor any admissible coefficient equal to $c>0$ on an open interval,\nits moment curve satisfies, strictly inside that interval,\n\n$$\nq'''(t)\\ge\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\qquad(r<t<s).\n$$","paper":"expository","role":"statement","section":"5.3 Strict curvature creates an impossible integral","start_byte":33656},{"anchor":"curvature","at":"b45512-45866","end_byte":45866,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[{"end_line":109,"line":101,"lines":[101,104],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.compactJetField_generator_two","note":"The drift of the third derivative is −3c a b, the last display unscaled.","plain":"Itô generator of ∂ₓa","proof_lines":[105,109],"relation":"proof"}],"lean_source":"statement","markdown":"On a positive constant interval put\n$r=c\\Phi_x$, $v=r_x$, $w=r_{xx}$ and $z=r_{xxx}$.\nIn forward time,\n$r_t=-\\tfrac12r_{xx}-rr_x$ and $dX_t=r(t,X_t)\\,dt+dB_t$.\nDifferentiating the PDE and applying Itô's formula gives\n\n$$\ndr(t,X_t)=v(t,X_t)\\,dB_t,\n$$\n$$\ndv(t,X_t)=-v(t,X_t)^2\\,dt+w(t,X_t)\\,dB_t,\n\\qquad\ndw(t,X_t)=-3v(t,X_t)w(t,X_t)\\,dt+z(t,X_t)\\,dB_t.\n$$","note":"Lean works with the unscaled derivatives and the coefficient c, and writes generators, never stochastic differentials.","paper":"long","role":"derivation","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":45512},{"anchor":"curvature","at":"b45868-45955","end_byte":45955,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[{"end_line":37,"line":30,"lines":[30,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"The time derivative of the curvature moment.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"proof"},{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"It equals E[b²−2ca³], the same identity unscaled.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"}],"lean_source":"statement","markdown":"Since $q'=c^{-2}\\mathbb E v^2$, the drift of $v^2$ yields\n$c^2q''=\\mathbb E(w^2-2v^3)$.","paper":"long","role":"derivation","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":45868},{"anchor":"curvature","at":"b45956-46097","end_byte":46097,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[{"end_line":163,"line":150,"lines":[150,154],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.secondCurvatureField_generator","note":"The generator of b²−2ca³ is d²−12cab²+6c²a⁴.","plain":"Itô generator of (∂ₓa)² − 2ca³","proof_lines":[155,163],"relation":"proof"},{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"The result as an expectation of the actual jets.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"The drifts of $w^2$ and $v^3$ are,\nrespectively, $z^2-6vw^2$ and $3vw^2-3v^4$. Therefore\n\n$$\nc^2q'''=\\mathbb E\\bigl[z^2-12vw^2+6v^4\\bigr].\n$$","paper":"long","role":"derivation","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":45956}]},"scalar.correlation_vector_control":{"lean":[{"clause":"(∑ j : Fin N, (meshA T N γ u j-Real.sqrt (stepSize T N)/\n      predictableNormalizer (meshRaw T N γ u j) (gaussianProduct (Fin (N+1))))^2)≤\n      gradientMeshError T N γ u","end_line":100,"line":84,"lines":[89,91],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both squared ell-2 errors are bounded by gradientMeshError, on a fixed strip with positive active moments.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"},{"clause":"Tendsto (fun N ↦ gradientMeshError T N γ u) atTop (𝓝 0)","end_line":54,"line":44,"lines":[46,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"That error tends to zero as the mesh refines, which supplies the o(1).","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"part"},{"clause":"(∑ j : Fin N, (meshB T N γ u j)^2)≤1 ∧ |meshB T N γ u N|≤1","end_line":105,"line":102,"lines":[104,104],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"The B vector has ell-2 norm at most 1, and the last coordinate B_{N+1} is at most 1 in absolute value.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"}],"short":"Control of coefficient vectors","statements":[{"anchor":"gaussian","at":"b95238-96114","end_byte":96114,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"clause":"(∑ j : Fin N, (meshA T N γ u j-Real.sqrt (stepSize T N)/\n      predictableNormalizer (meshRaw T N γ u j) (gaussianProduct (Fin (N+1))))^2)≤\n      gradientMeshError T N γ u","end_line":100,"line":84,"lines":[89,91],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both squared ell-2 errors are bounded by gradientMeshError, on a fixed strip with positive active moments.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"},{"clause":"Tendsto (fun N ↦ gradientMeshError T N γ u) atTop (𝓝 0)","end_line":54,"line":44,"lines":[46,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"That error tends to zero as the mesh refines, which supplies the o(1).","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"part"},{"clause":"(∑ j : Fin N, (meshB T N γ u j)^2)≤1 ∧ |meshB T N γ u N|≤1","end_line":105,"line":102,"lines":[104,104],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"The B vector has ell-2 norm at most 1, and the last coordinate B_{N+1} is at most 1 in absolute value.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"}],"lean_source":"statement","markdown":"The approximation\n$$\nM_N=\\sum_{j=1}^N\\sqrt\\Delta\\,a(t_{j-1},Y_{j-1})Z_j\n =u(T,Y_N)+o_{L^2}(1)\n$$\ngives a common residual for both coefficient families. In\n$\\mathbb E[M_NZ_i]$, a summand with $j<i$ is independent of centered\n$Z_i$, and one with $j>i$ vanishes on conditioning on the first $j-1$\ncoordinates. The $j=i$ term is\n$\\sqrt\\Delta\\,\\mathbb E a(t_{i-1},Y_{i-1})$.\nAlso $\\mathbb E[M_Ng_j]=\\sqrt\\Delta/c_j$ by orthonormality.\n\nSince the union of the two families need not be orthogonal, we apply\nBessel's inequality separately to the orthonormal family $(g_j)$ and to\nthe family $(Z_i)$. This gives the vector estimates\n$$\n(A_j)_{j=1}^N=(\\sqrt\\Delta/c_j)_{j=1}^N+o_{\\ell^2}(1),\n$$\n$$\n(B_i)_{i=1}^N=\n(\\sqrt\\Delta\\,\\mathbb E a(t_{i-1},Y_{i-1}))_{i=1}^N+o_{\\ell^2}(1).\n$$\nA sum of growing length needs these vector estimates, since coordinatewise\nerrors would not control it.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":95238},{"anchor":"gaussian","at":"b96255-96928","end_byte":96928,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"clause":"(∑ j : Fin N, (meshA T N γ u j-Real.sqrt (stepSize T N)/\n      predictableNormalizer (meshRaw T N γ u j) (gaussianProduct (Fin (N+1))))^2)≤\n      gradientMeshError T N γ u","end_line":100,"line":84,"lines":[89,91],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both squared ell-2 errors are bounded by gradientMeshError, on a fixed strip with positive active moments.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"},{"clause":"Tendsto (fun N ↦ gradientMeshError T N γ u) atTop (𝓝 0)","end_line":54,"line":44,"lines":[46,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"That error tends to zero as the mesh refines, which supplies the o(1).","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"part"},{"clause":"(∑ j : Fin N, (meshB T N γ u j)^2)≤1 ∧ |meshB T N γ u N|≤1","end_line":105,"line":102,"lines":[104,104],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"The B vector has ell-2 norm at most 1, and the last coordinate B_{N+1} is at most 1 in absolute value.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"}],"lean_source":"statement","markdown":"The fixed-$T$ Euler argument gives $e_N\\to0$. Bessel's inequality gives\n$\\|A_{1:N}-\\alpha\\|_2,\\|B_{1:N}-\\beta\\|_2\\le e_N$.\nAlso $\\|B_{1:N}\\|_2\\le1$ and $\\|\\alpha\\|_2\\le C_T$ for sufficiently\nlarge $N$, with $C_T$ independent of the mesh. Cauchy–Schwarz gives\n$$\n\\begin{aligned}\n\\left|\\sum_{j=1}^{N-1}(A_jB_{j+1}-\\alpha_j\\beta_{j+1})\\right|\n&\\le\\|A-\\alpha\\|_2\\|B_{2:N}\\|_2\n +\\|\\alpha_{1:N-1}\\|_2\\|B-\\beta\\|_2\\\\\n&\\le(1+C_T)e_N.\n\\end{aligned}\n$$\nThe unindexed differences in this inequality use coordinates $1,\\ldots,N$,\nand restricting to the indicated subvector only decreases their norms.\nThe rounding coordinate contributes at most\n$|A_NB_{N+1}|\\le e_N+\\sqrt\\Delta/c_N$.","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":96255},{"anchor":"qaoa-boundary","at":"b44036-44333","end_byte":44333,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"clause":"(∑ j : Fin N, (meshA T N γ u j-Real.sqrt (stepSize T N)/\n      predictableNormalizer (meshRaw T N γ u j) (gaussianProduct (Fin (N+1))))^2)≤\n      gradientMeshError T N γ u","end_line":100,"line":84,"lines":[89,91],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both squared ell-2 errors are bounded by gradientMeshError, on a fixed strip with positive active moments.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"},{"clause":"Tendsto (fun N ↦ gradientMeshError T N γ u) atTop (𝓝 0)","end_line":54,"line":44,"lines":[46,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"That error tends to zero as the mesh refines, which supplies the o(1).","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"part"},{"clause":"(∑ j : Fin N, (meshB T N γ u j)^2)≤1 ∧ |meshB T N γ u N|≤1","end_line":105,"line":102,"lines":[104,104],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"The B vector has ell-2 norm at most 1, and the last coordinate B_{N+1} is at most 1 in absolute value.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"}],"lean_source":"statement","markdown":"Bessel's inequality, applied\nseparately to $(g_j)$ and $(Z_j)$, controls the errors of both\ncorrelation vectors in $\\ell^2$, and their bounded norms control the\ngrowing pairing by Cauchy–Schwarz. The terms with $j<N$ give a Riemann\nsum, and the last term, from the rounding coordinate, vanishes.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44036},{"anchor":"gaussian","at":"b95238-95334","end_byte":95334,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"clause":"Tendsto (fun N ↦ gradientMeshError T N γ u) atTop (𝓝 0)","end_line":54,"line":44,"lines":[46,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"The squared L² residual tends to zero as N grows, for any γ with a gradient strip.","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"part"},{"end_line":26,"line":24,"lines":[24,26],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finiteRawMartingale","note":"M_N.","plain":"Raw martingale of the finite scheme","relation":"defines"},{"end_line":54,"line":51,"lines":[51,54],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.gradientMeshError","note":"The squared L² norm of the residual.","plain":"Gradient error on the mesh","relation":"defines"}],"lean_source":"statement","markdown":"The approximation\n$$\nM_N=\\sum_{j=1}^N\\sqrt\\Delta\\,a(t_{j-1},Y_{j-1})Z_j\n =u(T,Y_N)+o_{L^2}(1)\n$$","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":95238},{"anchor":"gaussian","at":"b95335-95667","end_byte":95667,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":63,"line":51,"lines":[51,59],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finiteRawMartingale_normalized_coefficient","note":"The g_j correlations.","plain":"Coefficient on a normalised increment","proof_lines":[60,63],"relation":"proof"},{"end_line":80,"line":65,"lines":[65,71],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finiteRawMartingale_coordinate_coefficient","note":"The Z_i correlations.","plain":"Coefficient on a coordinate","proof_lines":[72,80],"relation":"proof"}],"lean_source":"statement","markdown":"gives a common residual for both coefficient families. In\n$\\mathbb E[M_NZ_i]$, a summand with $j<i$ is independent of centered\n$Z_i$, and one with $j>i$ vanishes on conditioning on the first $j-1$\ncoordinates. The $j=i$ term is\n$\\sqrt\\Delta\\,\\mathbb E a(t_{i-1},Y_{i-1})$.\nAlso $\\mathbb E[M_Ng_j]=\\sqrt\\Delta/c_j$ by orthonormality.","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":95335},{"anchor":"gaussian","at":"b95738-95861","end_byte":95861,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":123,"line":82,"lines":[82,95],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finite_coefficient_L2_errors","note":"Both vector estimates, one Bessel inequality each.","plain":"L² error of the rounded coefficients","proof_lines":[96,123],"relation":"proof"}],"lean_source":"statement","markdown":"Bessel's inequality separately to the orthonormal family $(g_j)$ and to\nthe family $(Z_i)$. This gives the vector estimates","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":95738},{"anchor":"gaussian","at":"b95862-96010","end_byte":96010,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":100,"line":84,"lines":[84,94],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both expansions, with error the squared residual.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"}],"lean_source":"statement","markdown":"$$\n(A_j)_{j=1}^N=(\\sqrt\\Delta/c_j)_{j=1}^N+o_{\\ell^2}(1),\n$$\n$$\n(B_i)_{i=1}^N=\n(\\sqrt\\Delta\\,\\mathbb E a(t_{i-1},Y_{i-1}))_{i=1}^N+o_{\\ell^2}(1).\n$$","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":95862},{"anchor":"gaussian","at":"b96378-96492","end_byte":96492,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":105,"line":102,"lines":[102,105],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"B has norm at most one.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"},{"end_line":129,"line":107,"lines":[107,113],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_leading_norm","note":"α has norm at most √T D, so Lean has this for every N.","plain":"Norm bound on the leading coefficients","proof_lines":[114,129],"relation":"part"}],"lean_source":"statement","markdown":"Also $\\|B_{1:N}\\|_2\\le1$ and $\\|\\alpha\\|_2\\le C_T$ for sufficiently\nlarge $N$, with $C_T$ independent of the mesh.","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":96378},{"anchor":"obligations","at":"b251213-251283","end_byte":251283,"heading_id":"logical-limits-of-the-argument","lean":[],"lean_source":"statement","markdown":"Coordinatewise convergence does not control a growing coefficient sum.","note":"Not in the Lean: a remark. The Lean's own estimate is the squared ℓ² bound in mesh_coefficient_errors.","paper":"long","role":"explanation","section":"12.4 Logical limits of the argument","start_byte":251213},{"anchor":"qaoa-boundary","at":"b43816-43914","end_byte":43914,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":123,"line":82,"lines":[82,95],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finite_coefficient_L2_errors","note":"The rounded spin replaces u(T,Y_N) in both coefficient vectors.","plain":"L² error of the rounded coefficients","proof_lines":[96,123],"relation":"part"},{"end_line":33,"line":25,"lines":[25,30],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.finite_rounding_coefficient","note":"","plain":"Rounding preserves correlations","proof_lines":[31,33],"relation":"proof"}],"lean_source":"statement","markdown":"Conditional-mean preservation transfers these correlations to $F_N$\nfor the first $N$ coordinates.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43816},{"anchor":"qaoa-boundary","at":"b44036-44233","end_byte":44233,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":100,"line":84,"lines":[84,94],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficient_errors","note":"Both errors are bounded by gradientMeshError, the squared residual above.","plain":"L² errors of the mesh coefficients","proof_lines":[95,100],"relation":"part"},{"end_line":123,"line":82,"lines":[82,95],"module":"OAI.Probability.SKValue.Equations.FiniteRounded","name":"OAI.SKValue.finite_coefficient_L2_errors","note":"","plain":"L² error of the rounded coefficients","proof_lines":[96,123],"relation":"proof"}],"lean_source":"statement","markdown":"Bessel's inequality, applied\nseparately to $(g_j)$ and $(Z_j)$, controls the errors of both\ncorrelation vectors in $\\ell^2$, and their bounded norms control the\ngrowing pairing by Cauchy–Schwarz.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44036},{"anchor":"qaoa-boundary","at":"b44161-44233","end_byte":44233,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":110,"line":93,"lines":[93,104],"module":"OAI.Probability.SKValue.Equations.ShiftedPairing","name":"OAI.SKValue.shiftedPairing_tendsto_error_zero","note":"","plain":"Shifted pairing error tends to zero","proof_lines":[105,110],"relation":"proof"},{"end_line":105,"line":102,"lines":[102,104],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_coefficients_norms","note":"The B vector has ell-2 norm at most 1, and the last coordinate B_{N+1} is at most 1 in absolute value.","plain":"Norm bounds on the mesh coefficients","proof_lines":[105,105],"relation":"part"},{"end_line":129,"line":107,"lines":[107,113],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_leading_norm","note":"","plain":"Norm bound on the leading coefficients","proof_lines":[114,129],"relation":"part"}],"lean_source":"statement","markdown":"and their bounded norms control the\ngrowing pairing by Cauchy–Schwarz.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44161}]},"scalar.euler_normalization":{"lean":[{"clause":"Tendsto (fun N ↦ ∫ ω, (meshMaxError T X (coupledEuler T γ u B) N ω)^2 ∂μ)\n      atTop (𝓝 (0 : ℝ))","end_line":95,"line":80,"lines":[91,92],"module":"OAI.Probability.SKValue.Gaussian.CoupledCoordinates","name":"OAI.SKValue.coupled_euler_L2_convergence","note":"The convergence of expected squared maximum error, for Euler driven by Brownian increments under strip hypotheses.","plain":"L² convergence, coupled Euler scheme","proof_lines":[93,95],"relation":"part"},{"clause":"∀ i j : Fin N,\n      (∫ z, normalizedIncrement (H i) (coordinate N i) (gaussianProduct (Fin (N+1))) z *\n        normalizedIncrement (H j) (coordinate N j) (gaussianProduct (Fin (N+1))) z\n        ∂gaussianProduct (Fin (N+1))) = if i=j then 1 else 0","end_line":57,"line":35,"lines":[39,42],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.normalized_family_orthonormal","note":"Orthonormality of the normalised increments, for any predictable coefficients with positive second moments.","plain":"Normalised increments are orthonormal","proof_lines":[43,57],"relation":"part"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,157],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Its first clause gives eventual positivity of active normalisers. Uniform convergence of c_j to 1 is not stated here.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"short":"Euler convergence, normalised directions","statements":[{"anchor":"gaussian","at":"b93291-94267","end_byte":94267,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"clause":"Tendsto (fun N ↦ ∫ ω, (meshMaxError T X (coupledEuler T γ u B) N ω)^2 ∂μ)\n      atTop (𝓝 (0 : ℝ))","end_line":95,"line":80,"lines":[91,92],"module":"OAI.Probability.SKValue.Gaussian.CoupledCoordinates","name":"OAI.SKValue.coupled_euler_L2_convergence","note":"The convergence of expected squared maximum error, for Euler driven by Brownian increments under strip hypotheses.","plain":"L² convergence, coupled Euler scheme","proof_lines":[93,95],"relation":"part"},{"clause":"∀ i j : Fin N,\n      (∫ z, normalizedIncrement (H i) (coordinate N i) (gaussianProduct (Fin (N+1))) z *\n        normalizedIncrement (H j) (coordinate N j) (gaussianProduct (Fin (N+1))) z\n        ∂gaussianProduct (Fin (N+1))) = if i=j then 1 else 0","end_line":57,"line":35,"lines":[39,42],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.normalized_family_orthonormal","note":"Orthonormality of the normalised increments, for any predictable coefficients with positive second moments.","plain":"Normalised increments are orthonormal","proof_lines":[43,57],"relation":"part"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,157],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Its first clause gives eventual positivity of active normalisers. Uniform convergence of c_j to 1 is not stated here.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"We first fix $0<T<1$, then a mesh $N\\ge1$,\n$\\Delta=T/N$, $t_j=j\\Delta$.\nOn independent standard normals $Z_1,\\ldots,Z_{N+1}$, set\n$$\nY_0=0,\\quad Y_j=Y_{j-1}+\\sqrt\\Delta Z_j+\n \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\n$$\n$$\nc_j=(\\mathbb E a(t_{j-1},Y_{j-1})^2)^{-1/2},\\qquad\ng_j=c_ja(t_{j-1},Y_{j-1})Z_j.\n$$\nFor sufficiently large $N$, all $c_j$, $1\\le j\\le N$, are finite\nand positive. Couple the increments to Brownian motion. On the fixed\nstrip, the gradient and curvature are bounded and Lipschitz. Monotonicity\ngives, for the left-grid projection $\\pi_N$,\n$$\n\\int_0^T|\\gamma(s)-\\gamma(\\pi_N(s))|\\,ds\n \\le\\Delta(\\gamma(T)-\\gamma(0)).\n$$\nDrift error estimates and discrete Gronwall yield\n$\\mathbb E\\max_{j\\le N}|Y_j-X_{t_j}|^2\\to0$, hence\n$\\max_j|c_j-1|\\to0$. The $g_j$ are orthonormal in $L^2$:\nthey have unit variance and conditional mean zero given their past.\nTheir predictable multipliers may depend on earlier coordinates, so\northonormality is weaker than independence.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93291},{"anchor":"qaoa-boundary","at":"b42162-42911","end_byte":42911,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"clause":"∀ i j : Fin N,\n      (∫ z, normalizedIncrement (H i) (coordinate N i) (gaussianProduct (Fin (N+1))) z *\n        normalizedIncrement (H j) (coordinate N j) (gaussianProduct (Fin (N+1))) z\n        ∂gaussianProduct (Fin (N+1))) = if i=j then 1 else 0","end_line":57,"line":35,"lines":[39,42],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.normalized_family_orthonormal","note":"Orthonormality of the normalised increments, for any predictable coefficients with positive second moments.","plain":"Normalised increments are orthonormal","proof_lines":[43,57],"relation":"part"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,157],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Its first clause gives eventual positivity of active normalisers. Uniform convergence of c_j to 1 is not stated here.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"**The finite Gaussian input.** Fix the minimiser and its diffusion,\na cutoff $0<T<1$, and a mesh $t_j=j\\Delta$ with $\\Delta=T/N$.\nThe mesh count $N$ is distinct from the number of spins $n$.\nFrom independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$\n[OpenAI (2026a), Proposition 7.2][FS] proves that the denominators\nare positive for all sufficiently large $N$. Since $h_j$ depends\nonly on $Z_1,\\ldots,Z_{j-1}$, the $g_j$ are centred and\northonormal in $L^2$, although they need not be independent Gaussians.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42162},{"anchor":"value-and-diffusion","at":"b21730-21894","end_byte":21894,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":43,"line":38,"lines":[41,43],"module":"OAI.Probability.SKSupport.Regularity.ClosedTime","name":"OAI.ZeroTemperatureSK.eulerState","note":"The same step on a mesh: previous state, scaled Gaussian, then step times γ times gradient, both evaluated at the left end.","plain":"Euler state on a Gaussian sample","relation":"defines"}],"lean_source":"statement","markdown":"With $\\gamma$ and $u$ supplied, a short forward step of\nlength $h$ from state $x$ at time $t$, with $t+h<1$, is approximated by\n\n$$\nx+h\\gamma(t)u(t,x)+\\sqrt h Z.\n$$","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":21730},{"anchor":"value-and-diffusion","at":"b21896-21973","end_byte":21973,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"gradient W γ (meshTime T N j) (eulerState W γ T N j z)","end_line":43,"line":38,"lines":[43,43],"module":"OAI.Probability.SKSupport.Regularity.ClosedTime","name":"OAI.ZeroTemperatureSK.eulerState","note":"The gradient depends on state j, which does not depend on coordinate j of the noise.","plain":"Euler state on a Gaussian sample","relation":"defines"}],"lean_source":"statement","markdown":"The response is evaluated before the fresh independent noise $Z$ is\nobserved.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":21896},{"anchor":"gaussian","at":"b93676-93929","end_byte":93929,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":80,"line":74,"lines":[74,77],"module":"OAI.Probability.SKValue.Processes.EulerL1","name":"OAI.SKValue.monotone_grid_sum_tendsto","note":"Riemann sums of a monotone coefficient converge; the Lean limit replaces this displayed rate.","plain":"Monotone Riemann sums converge","proof_lines":[78,80],"relation":"proof"},{"end_line":95,"line":80,"lines":[80,92],"module":"OAI.Probability.SKValue.Gaussian.CoupledCoordinates","name":"OAI.SKValue.coupled_euler_L2_convergence","note":"The coupling to the path.","plain":"L² convergence, coupled Euler scheme","proof_lines":[93,95],"relation":"proof"}],"lean_source":"statement","markdown":"Couple the increments to Brownian motion. On the fixed\nstrip, the gradient and curvature are bounded and Lipschitz. Monotonicity\ngives, for the left-grid projection $\\pi_N$,\n$$\n\\int_0^T|\\gamma(s)-\\gamma(\\pi_N(s))|\\,ds\n \\le\\Delta(\\gamma(T)-\\gamma(0)).\n$$","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93676},{"anchor":"gaussian","at":"b93930-94051","end_byte":94051,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":81,"line":34,"lines":[34,53],"module":"OAI.Probability.SKValue.Processes.MeshMaxError","name":"OAI.SKValue.driven_euler_L2_convergence","note":"The maximal L² Euler bound.","plain":"L² convergence, driven Euler scheme","proof_lines":[54,81],"relation":"proof"},{"end_line":27,"line":25,"lines":[25,27],"module":"OAI.Probability.SKValue.Processes.MeshMaxError","name":"OAI.SKValue.meshMaxError","note":"The maximum over the grid.","plain":"Maximum error on the mesh","relation":"defines"}],"lean_source":"statement","markdown":"Drift error estimates and discrete Gronwall yield\n$\\mathbb E\\max_{j\\le N}|Y_j-X_{t_j}|^2\\to0$, hence\n$\\max_j|c_j-1|\\to0$.","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93930},{"anchor":"gaussian","at":"b94052-94267","end_byte":94267,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":57,"line":35,"lines":[35,42],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.normalized_family_orthonormal","note":"Orthonormal for predictable multipliers, for any such family.","plain":"Normalised increments are orthonormal","proof_lines":[43,57],"relation":"part"},{"end_line":267,"line":256,"lines":[256,261],"module":"OAI.Probability.SKValue.GroundState.PredictableNormalizer","name":"OAI.SKValue.normalizedIncrements_orthogonal","note":"Orthogonality off the diagonal.","plain":"Normalised increments are orthogonal","proof_lines":[262,267],"relation":"proof"}],"lean_source":"statement","markdown":"The $g_j$ are orthonormal in $L^2$:\nthey have unit variance and conditional mean zero given their past.\nTheir predictable multipliers may depend on earlier coordinates, so\northonormality is weaker than independence.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":94052},{"anchor":"qaoa-boundary","at":"b42162-42291","end_byte":42291,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":24,"line":24,"lines":[24,24],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.stepSize","note":"","plain":"Mesh width T/N","relation":"defines"},{"end_line":25,"line":25,"lines":[25,25],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.meshTime","note":"","plain":"Mesh time jT/N","relation":"defines"},{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"0<T<1, a minimiser and any solution X of the SDE.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"**The finite Gaussian input.** Fix the minimiser and its diffusion,\na cutoff $0<T<1$, and a mesh $t_j=j\\Delta$ with $\\Delta=T/N$.","paper":"short","role":"definition","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42162},{"anchor":"qaoa-boundary","at":"b42353-42652","end_byte":42652,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":15,"line":14,"lines":[14,15],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gaussianProduct","note":"Lean indexes coordinates from 0, so paper Z_j is coordinate j−1 and paper g_j is gaussianIncrement (j−1).","plain":"Product of standard Gaussians","relation":"defines"},{"end_line":31,"line":27,"lines":[27,31],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","note":"The recursion for Y_j.","plain":"Euler recursion on a uniform mesh","relation":"defines"},{"end_line":130,"line":128,"lines":[128,130],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.normalizer","note":"The inverse square root of the second moment.","plain":"Normaliser of the raw coefficient","relation":"defines"},{"end_line":134,"line":132,"lines":[132,134],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gaussianIncrement","note":"","plain":"Normalised Gaussian increment","relation":"defines"}],"lean_source":"statement","markdown":"From independent standard normals $Z_1,\\ldots,Z_{N+1}$, form\n$$\n\\begin{aligned}\nY_0&=0,\\\\\nY_j&=Y_{j-1}+\\sqrt\\Delta Z_j+\n       \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\\\\\nh_j&=\\frac{a(t_{j-1},Y_{j-1})}\n {\\sqrt{\\mathbb E a(t_{j-1},Y_{j-1})^2}},\\qquad g_j=h_jZ_j\n \\quad(1\\leq j\\leq N).\n\\end{aligned}\n$$","paper":"short","role":"definition","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42353},{"anchor":"qaoa-boundary","at":"b42653-42764","end_byte":42764,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":49,"line":29,"lines":[39,39],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"On one strip, given the unit curvature moment. At a minimiser, finite_gaussian_target gives it outright.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"states"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,157],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Its first clause gives eventual positivity of active normalisers. Uniform convergence of c_j to 1 is not stated here.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":136,"line":47,"lines":[60,65],"module":"OAI.Probability.SKValue.Processes.MeshMeanBound","name":"OAI.SKValue.finiteGaussian_from_strip_and_moments","note":"Each second moment is within r N of one, by uniform_moment_positive.","plain":"Finite-Gaussian limit from strip and moments","proof_lines":[60,136],"relation":"proof"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Proposition 7.2][FS] proves that the denominators\nare positive for all sufficiently large $N$.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42653},{"anchor":"qaoa-boundary","at":"b42765-42911","end_byte":42911,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":57,"line":35,"lines":[35,42],"module":"OAI.Probability.SKValue.Gaussian.CoordinateDependsBefore","name":"OAI.SKValue.normalized_family_orthonormal","note":"Orthonormal.","plain":"Normalised increments are orthonormal","proof_lines":[43,57],"relation":"part"},{"end_line":133,"line":125,"lines":[125,128],"module":"OAI.Probability.SKValue.GroundState.PredictableNormalizer","name":"OAI.SKValue.normalizedIncrement_mean","note":"Centred.","plain":"Normalised increment has mean zero","proof_lines":[129,133],"relation":"part"},{"end_line":68,"line":66,"lines":[66,67],"module":"OAI.Probability.SKValue.Processes.MeshRaw","name":"OAI.SKValue.mesh_raw_depends","note":"The predictability hypothesis hp, for the raw coefficient.","plain":"Raw coefficient depends on past only","proof_lines":[68,68],"relation":"part"}],"lean_source":"statement","markdown":"Since $h_j$ depends\nonly on $Z_1,\\ldots,Z_{j-1}$, the $g_j$ are centred and\northonormal in $L^2$, although they need not be independent Gaussians.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":42765},{"anchor":"qaoa-boundary","at":"b40540-40573","end_byte":40573,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The Lean stops at the limit of the finite pairing. The circuit theorem and the rooted energy are outside it.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"We give its finite Gaussian input","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":40540},{"anchor":"scalar-value","at":"b39964-40073","end_byte":40073,"heading_id":"the-classical-algorithm","lean":[{"end_line":63,"line":41,"lines":[41,48],"module":"OAI.Probability.SKValue.Equations.RealFiltration","name":"OAI.SKValue.IsDiffusion.unitMomentMartingale","note":"Used in Lean for the value identity. The algorithm of this subsection is outside it.","plain":"The gradient is a unit-moment martingale","proof_lines":[49,63],"relation":"part"}],"lean_source":"statement","markdown":"In our notation, the bounded martingale $M_t$ and $\\mathbb E a^2=1$ give\nthe required variance normalisation.","paper":"short","role":"explanation","section":"5.2 The classical algorithm","start_byte":39964}]},"scalar.finite_gibbs_comparison":{"lean":[],"short":"Finite Gibbs comparison","statements":[{"anchor":"gibbs-comparison","at":"b221287-222177","end_byte":222177,"heading_id":"finite-temperature-values-and-exact-gibbs-samples","lean":[],"lean_source":"record","markdown":"This comparison samples the exact Gibbs distribution. The QAOA Born law\nin §11.2 is a separate distribution, and none of the following bounds\nidentifies the two. For one fixed instance, let $L_n=\\max_\\sigma H_{n,J}(\\sigma)/n$,\n$\\mathcal F_{n,\\beta}=(n\\beta)^{-1}\\log\\sum_\\sigma e^{\\beta H_{n,J}(\\sigma)}$,\nand let $\\mathcal U_{n,\\beta}$ be the mean energy per spin under the\nGibbs law proportional to $e^{\\beta H_{n,J}(\\sigma)}$. Then\n$$\n0\\le\\mathcal F_{n,\\beta}-L_n\\le\\frac{\\log2}{\\beta},\\qquad\n0\\le L_n-\\mathcal U_{n,\\beta}\\le\\frac{\\log2}{\\beta}.\n$$\nThe first is the log-sum bound. The second uses\n$\\mathcal F=\\mathcal U+S/(n\\beta)$, where the Gibbs entropy satisfies\n$0\\le S\\le n\\log2$, together with $\\mathcal U\\le L_n\\le\\mathcal F$.\nThus both disorder averages approach $P_*$ along every joint\n$n\\to\\infty$, $\\beta\\to\\infty$, by the already established\nconvergence of $\\mathbb E L_n$.","paper":"long","role":"statement","section":"11.1.2 Finite-temperature values and exact Gibbs samples","start_byte":221287},{"anchor":"gibbs-comparison","at":"b222179-222982","end_byte":222982,"heading_id":"finite-temperature-values-and-exact-gibbs-samples","lean":[],"lean_source":"record","markdown":"There is also a finite probability estimate. Every configuration has\nGaussian coefficient-vector norm $\\sqrt{(n-1)/(2n^2)}$ after division\nby $n$, and their maximum has the same Lipschitz bound. Gaussian\nconcentration gives, for $n\\ge2$,\n$$\n\\Pr_J(|L_n-\\mathbb E L_n|>s)\n\\le2\\exp\\{-n^2s^2/(n-1)\\}.\n$$\nFor an exact Gibbs draw $\\Sigma^{\\rm G}$, count the at most $2^n$\nconfigurations with deficit at least $s$ and bound the partition sum\nbelow by its maximal term. This gives, for every instance,\n$$\n\\Pr\\{H_{n,J}(\\Sigma^{\\rm G})/n\\le L_n-s\\mid J\\}\n\\le\\min\\{1,e^{n(\\log2-\\beta s)}\\}.\n$$\nThe lower tail below $P_*-s$ is bounded by\n$\\Pr_J(L_n<P_*-s/2)+\\min\\{1,e^{n(\\log2-\\beta s/2)}\\}$;\nthe upper tail is at most the upper tail of $L_n$. Hence an exact\nGibbs sample also approaches $P_*$ in joint probability.","paper":"long","role":"statement","section":"11.1.2 Finite-temperature values and exact Gibbs samples","start_byte":222179}]},"scalar.finite_maximum_concentration":{"lean":[],"short":"Finite maximum concentrates","statements":[{"anchor":"gibbs-comparison","at":"b222179-222478","end_byte":222478,"heading_id":"finite-temperature-values-and-exact-gibbs-samples","lean":[],"lean_source":"record","markdown":"There is also a finite probability estimate. Every configuration has\nGaussian coefficient-vector norm $\\sqrt{(n-1)/(2n^2)}$ after division\nby $n$, and their maximum has the same Lipschitz bound. Gaussian\nconcentration gives, for $n\\ge2$,\n$$\n\\Pr_J(|L_n-\\mathbb E L_n|>s)\n\\le2\\exp\\{-n^2s^2/(n-1)\\}.\n$$","paper":"long","role":"statement","section":"11.1.2 Finite-temperature values and exact Gibbs samples","start_byte":222179},{"anchor":"dependencies","at":"b16786-17219","end_byte":17219,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"The cited theorem is stated for the almost-sure normalized maximum. Its expected-value formulation used here follows from Gaussian concentration. For\n$L_n=n^{-1}\\max_\\sigma H_n(\\sigma)$, the Lipschitz constant in the Gaussian couplings is\n$\\sqrt{(n-1)/(2n^2)}$. Hence $L_n-\\mathbb E L_n\\to0$ in probability. Since the almost-sure limit is the deterministic Parisi value, the deterministic sequence $\\mathbb E L_n$ has the same limit.","paper":"long","role":"derivation","section":"2.3 Prior results and their roles","start_byte":16786},{"anchor":"dependencies","at":"b17048-17219","end_byte":17219,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"Hence $L_n-\\mathbb E L_n\\to0$ in probability. Since the almost-sure limit is the deterministic Parisi value, the deterministic sequence $\\mathbb E L_n$ has the same limit.","note":"Lean uses no concentration: its P_* is the limit of expectations, which it proves exists by subadditivity.","paper":"long","role":"derivation","section":"2.3 Prior results and their roles","start_byte":17048}]},"scalar.finite_maximum_unique_modulo_flip":{"lean":[],"short":"Finite maximiser unique up to flip","statements":[{"anchor":"overlap-pairs","at":"b174831-175341","end_byte":175341,"heading_id":"near-ground-configurations-at-a-prescribed-overlap","lean":[],"lean_source":"record","markdown":"At a fixed finite size the picture is different. If $\\sigma,\\tau$\ndiffer at $k$ sites, precisely the $k(n-k)$ edges crossing that set change\ntheir spin product. Their independent Gaussian contributions give\n$$\n\\operatorname{Var}(H_n(\\sigma)-H_n(\\tau))=\\frac{4k(n-k)}n.\n$$\nFor $0<k<n$ this is positive, so the two energies tie with probability\nzero. A finite union over pairs, together with\n$H_n(-\\sigma)=H_n(\\sigma)$, shows that almost surely the exact maximum\nhas one maximizing pair $\\{\\sigma_*,-\\sigma_*\\}$.","paper":"long","role":"statement","section":"9.3 Near-ground configurations at a prescribed overlap","start_byte":174831},{"anchor":"overlap-payoff","at":"b43330-43676","end_byte":43676,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"At a fixed finite size, almost surely the only exact maximisers are\na configuration and its global reversal. Indeed, two configurations\nthat are neither equal nor global reversals differ in some pair product.\nTheir energy difference is then a Gaussian with positive variance, so a\ntie has probability zero, and there are only finitely many pairs.","paper":"expository","role":"statement","section":"6. Return to the spins","start_byte":43330},{"anchor":"spins-and-overlap","at":"b6414-6530","end_byte":6530,"heading_id":"the-finite-optimisation-problem","lean":[],"lean_source":"record","markdown":"Reversing all the signs of one configuration\npreserves its energy and reverses the sign of its overlap with another.","paper":"expository","role":"mention","section":"1. The finite optimisation problem","start_byte":6414}]},"scalar.fixed_strip_error_budget":{"lean":[],"short":"Fixed-strip error budget","statements":[{"anchor":"gaussian-error-budget","at":"b102870-103111","end_byte":103111,"heading_id":"error-estimates-at-a-fixed-cutoff","lean":[],"lean_source":"record","markdown":"In particular, for each fixed minimizer and $T<1$ there are finite\n$C_{\\gamma,T}$ and $N_0(\\gamma,T)$ such that\n$$\n|S_{T,N}-P_*|\\le(1-T)m(T)+C_{\\gamma,T}\\sqrt{T/N}\n\\qquad(N\\ge N_0(\\gamma,T)).\n$$\nThe constants are not uniform as $T\\uparrow1$.","paper":"long","role":"statement","section":"7.1 Error estimates at a fixed cutoff","start_byte":102870},{"anchor":"gaussian-error-budget","at":"b102635-102869","end_byte":102869,"heading_id":"error-estimates-at-a-fixed-cutoff","lean":[],"lean_source":"record","markdown":"Combining this with the vector and rounding-endpoint estimate gives\n$$\n \\begin{aligned}\n |S_{T,N}-P_*|\n \\le{}&(1-T)m(T)\n   +\\bigl(2+\\sqrt{T(1+d_N)}\\bigr)e_N\n   +\\sqrt{\\Delta(1+d_N)}\\\\\n &+TK_Td_N+TL_T\\delta_N+\\Delta .\n \\end{aligned}\n$$","paper":"long","role":"derivation","section":"7.1 Error estimates at a fixed cutoff","start_byte":102635},{"anchor":"gaussian-error-budget","at":"b99332-99812","end_byte":99812,"heading_id":"error-estimates-at-a-fixed-cutoff","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Qualitative limits only. The N^(-1/2) rate and its constants are not in this statement.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"### 7.1 Error estimates at a fixed cutoff\n\nWe fix the minimizer first and then $T<1$. The constants in the bounds\nbelow depend on that minimizer and strip. They give an $N^{-1/2}$ rate\nthere, with no uniformity as $T\\uparrow1$. The derivative bounds are\nfinite, but no numerical values for them are available as inputs to an\nalgorithm that chooses a mesh.\n\nThe calculation is a deduction from the source's strip estimates and goes\nbeyond the qualitative selected formal statement.","paper":"long","role":"explanation","section":"7.1 Error estimates at a fixed cutoff","start_byte":99332},{"anchor":"gaussian-error-budget","at":"b101313-101525","end_byte":101525,"heading_id":"error-estimates-at-a-fixed-cutoff","lean":[{"end_line":133,"line":59,"lines":[59,65],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.gradientMeshError_bound","note":"Lean has an explicit bound on e_N squared, on its own strip constants.","plain":"Explicit gradient mesh error bound","proof_lines":[66,133],"relation":"proof"}],"lean_source":"statement","markdown":"For the\nmartingale residual, Itô isometry and the same increment estimate give\n$$\ne_N\\le K_T\\delta_N+\n\\sqrt T\\{L_T\\delta_N+L_T\\sqrt\\Delta+(H+\\Gamma L_T)\\Delta\\}.\n$$\nThus $\\delta_N,e_N=O_{\\gamma,T}(\\sqrt\\Delta)$.","note":"The Lean constants differ from these; the rate √Δ is not stated as a conclusion.","paper":"long","role":"derivation","section":"7.1 Error estimates at a fixed cutoff","start_byte":101313},{"anchor":"gaussian-error-budget","at":"b101526-101671","end_byte":101671,"heading_id":"error-estimates-at-a-fixed-cutoff","lean":[],"lean_source":"statement","markdown":"These are deductions from the compact-strip bounds and stochastic\nidentities. The selected formal statement contains no such quantitative\nclause.","note":"Not in the Lean: a remark. finite_gaussian_target and its conclusion state no rate.","paper":"long","role":"explanation","section":"7.1 Error estimates at a fixed cutoff","start_byte":101526}]},"scalar.fixed_strip_input_bounds":{"lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"Holds for every Brownian space and coefficient, with no minimiser or diffusion premise. The top-level proofs take the strip bounds from here.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"states"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"The same bounds, for every cutoff, in the form the top-level proofs supply.","plain":"The strip regularity bounds","relation":"defines"},{"clause":"(hg : GradientStrip T γ.coeff (gradient W γ) K L)\n    (hD : 0≤D) (hLu : 0≤Lu) (hLa : 0≤La)\n    (ha : ∀ t∈Icc (0 : ℝ) T, ∀ x, |curvature W γ t x|≤D)\n    (huLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |gradient W γ s x-gradient W γ t y|≤Lu*(|s-t|+|x-y|))\n    (haLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |curvature W γ s x-curvature W γ t y|≤La*(|s-t|+|x-y|))","end_line":49,"line":29,"lines":[31,37],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Takes these bounds as premises hg, hD, hLu, hLa, ha, huLip and haLip. It does not derive them; sourceStripRegularity does.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"part"},{"end_line":36,"line":18,"lines":[18,36],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.GradientStrip","note":"The structure behind premise hg, with its bounds and integrated equation.","plain":"Gradient bounds on a fixed strip","relation":"defines"}],"short":"Fixed-strip analytic bounds","statements":[{"anchor":"formal-scalar","at":"b243747-244001","end_byte":244001,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"The bounds, for every γ with no minimiser or diffusion premise. They fill the helper's premises.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"part"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"The same bounds, for every cutoff, in the form the top-level proofs supply.","plain":"The strip regularity bounds","relation":"defines"},{"end_line":36,"line":18,"lines":[18,36],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.GradientStrip","note":"The structure behind premise hg, with its bounds and integrated equation.","plain":"Gradient bounds on a fixed strip","relation":"defines"},{"clause":"(hg : GradientStrip T γ.coeff (gradient W γ) K L)\n    (hD : 0≤D) (hLu : 0≤Lu) (hLa : 0≤La)\n    (ha : ∀ t∈Icc (0 : ℝ) T, ∀ x, |curvature W γ t x|≤D)\n    (huLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |gradient W γ s x-gradient W γ t y|≤Lu*(|s-t|+|x-y|))\n    (haLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |curvature W γ s x-curvature W γ t y|≤La*(|s-t|+|x-y|))","end_line":49,"line":29,"lines":[31,37],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Takes these bounds as premises hg, hD, hLu, hLa, ha, huLip and haLip.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper. Those internal\ninputs are not additional hypotheses of the exported clauses.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243747},{"anchor":"gaussian","at":"b93676-93791","end_byte":93791,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"Proves these bounds for every order parameter and cutoff T<1. The coupling itself is in coupled_euler_L2_convergence.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"part"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"The same bounds, for every cutoff, in the form the top-level proofs supply.","plain":"The strip regularity bounds","relation":"defines"},{"end_line":36,"line":18,"lines":[18,36],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.GradientStrip","note":"The structure behind premise hg, with its bounds and integrated equation.","plain":"Gradient bounds on a fixed strip","relation":"defines"},{"clause":"(hg : GradientStrip T γ.coeff (gradient W γ) K L)\n    (hD : 0≤D) (hLu : 0≤Lu) (hLa : 0≤La)\n    (ha : ∀ t∈Icc (0 : ℝ) T, ∀ x, |curvature W γ t x|≤D)\n    (huLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |gradient W γ s x-gradient W γ t y|≤Lu*(|s-t|+|x-y|))\n    (haLip : ∀ s∈Icc (0 : ℝ) T, ∀ t∈Icc (0 : ℝ) T, ∀ x y,\n      |curvature W γ s x-curvature W γ t y|≤La*(|s-t|+|x-y|))","end_line":49,"line":29,"lines":[31,37],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Takes these bounds as premises hg, hD, hLu, hLa, ha, huLip and haLip.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"Couple the increments to Brownian motion. On the fixed\nstrip, the gradient and curvature are bounded and Lipschitz.","paper":"long","role":"explanation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93676}]},"scalar.fixed_strip_input_diffusion":{"lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The caller: it takes the minimiser, the diffusion and 0<T<1 as hypotheses and derives the other premises of the helper itself.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"clause":"(hX : IsDiffusion W γ X) {T K L D Lu La : ℝ}\n    (hT : 0<T) (hT1 : T<1)","end_line":49,"line":29,"lines":[30,31],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Premises hX, hT and hT1. No minimiser appears among its hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"part"},{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"The diffusion predicate used in premise hX.","plain":"X is a diffusion for γ","relation":"defines"}],"short":"Fixed-strip diffusion, cutoff premises","statements":[{"anchor":"formal-scalar","at":"b243680-243924","end_byte":243924,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Its hypotheses hmin and hX are what the caller supplies.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":72,"line":61,"lines":[61,72],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Derives the strip bounds from sourceStripRegularity and the unit moment from contact, then applies the helper.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"proof"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The helper. No minimiser appears among its hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"Their caller supplies the same minimizer and\nassociated diffusion. The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243680}]},"scalar.fixed_strip_input_unit_moment":{"lean":[{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"Derives this moment from contact and strip regularity, as the top-level proofs do.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"clause":"(hmom : ∀ t∈Icc (0 : ℝ) T, (∫ ω, (curvature W γ t (X t ω))^2 ∂W.μ)=1)","end_line":49,"line":29,"lines":[38,38],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Takes the moment as premise hmom on [0,T]. It does not derive it; the contact lemma does.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"part"}],"short":"Unit curvature-moment premise","statements":[{"anchor":"formal-scalar","at":"b243747-244001","end_byte":244001,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"Derives this moment from contact and strip regularity, as the top-level proofs do.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"clause":"(hmom : ∀ t∈Icc (0 : ℝ) T, (∫ ω, (curvature W γ t (X t ω))^2 ∂W.μ)=1)","end_line":49,"line":29,"lines":[38,38],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Takes this moment as premise hmom.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper. Those internal\ninputs are not additional hypotheses of the exported clauses.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243747}]},"scalar.formal_contact_input":{"lean":[{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Exactly this identity, on [0,1), inside the scalar model.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"states"}],"short":"Scalar contact identity","statements":[{"anchor":"formal-contact","at":"b180140-180421","end_byte":180421,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Exactly this identity, on [0,1), inside the scalar model.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"states"}],"lean_source":"statement","markdown":"A second route obtains $q=t$ from the variational inequalities. It uses\nconvexity on intervals where the tail potential is positive, without first\nproving full Stieltjes support. The scalar implementation uses this\nargument, which is independent of the formal full-support theorem.","paper":"long","role":"explanation","section":"10.1 Contact from convexity on constant intervals","start_byte":180140},{"anchor":"formal-contact","at":"b182729-182873","end_byte":182873,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Exactly this identity, on [0,1), inside the scalar model.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"states"}],"lean_source":"statement","markdown":"Thus $G$ cannot have a positive component, and $G\\equiv0$ on\n$(0,1)$. Differentiation gives $q(t)=t$ there, with the origin supplied\nseparately.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":182729},{"anchor":"formal-contact","at":"b180423-180585","end_byte":180585,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":74,"line":72,"lines":[72,74],"module":"OAI.Probability.SKValue.Variation.ContactVariations","name":"OAI.SKValue.contactPotential","note":"G, for a given diffusion.","plain":"The potential G","relation":"defines"},{"end_line":110,"line":105,"lines":[105,107],"module":"OAI.Probability.SKValue.Variation.ContactVariations","name":"OAI.SKValue.IsMinimizer.contactPotential_nonneg","note":"G is nonnegative.","plain":"G is nonnegative","proof_lines":[108,110],"relation":"part"},{"end_line":139,"line":124,"lines":[124,126],"module":"OAI.Probability.SKValue.Variation.ContactVariations","name":"OAI.SKValue.IsDiffusion.contactPotential_hasDerivAt","note":"The derivative of G.","plain":"G′(s) = s − q(s)","proof_lines":[127,139],"relation":"part"}],"lean_source":"statement","markdown":"In the scalar model put $q(t)=\\mathbb E[u(t,X_t)^2]$,\n$f(t)=q(t)-t$ and $G(t)=\\int_t^1f(s)\\,ds$. The source first obtains\n$G\\ge0$, $G(1)=0$, $q(0)=0$ and $G'=-f$.","paper":"long","role":"explanation","section":"10.1 Contact from convexity on constant intervals","start_byte":180423},{"anchor":"formal-contact","at":"b180586-180871","end_byte":180871,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":125,"line":116,"lines":[116,120],"module":"OAI.Probability.SKValue.Variation.ContactIntegral","name":"OAI.SKValue.IsMinimizer.coeff_constant_of_contactPositive","note":"Constant on an interval where G is positive.","plain":"γ constant where G is positive","proof_lines":[121,125],"relation":"part"},{"end_line":114,"line":105,"lines":[105,108],"module":"OAI.Probability.SKValue.Variation.ContactIntegral","name":"OAI.SKValue.IsMinimizer.coeff_eq_of_contactPositive","note":"The mass argument for two points.","plain":"γ equal where G is positive","proof_lines":[109,114],"relation":"proof"}],"lean_source":"statement","markdown":"Compact-mass variations\nalso show that on any interval where $G>0$ the coefficient is constant:\nif its cumulative mass increased between two points there, integration\nof the strictly positive $G$ against that positive increment would\ncontradict the mass-removal variational inequality.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":180586},{"anchor":"formal-contact","at":"b180961-181153","end_byte":181153,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":75,"line":38,"lines":[38,41],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.positive_gap_endpoints","note":"The component of positivity, with its endpoints.","plain":"Endpoints of G's positive gap","proof_lines":[42,75],"relation":"proof"}],"lean_source":"statement","markdown":"Suppose now that $G(x)>0$ for some $0<x<1$. Choose the component\n$(a,b)$ of this positivity set containing $x$, allowing $a=0$ or\n$b=1$. Its coefficient is constant, say $c\\ge0$, and $G(b)=0$.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":180961},{"anchor":"formal-contact","at":"b181255-181425","end_byte":181425,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":18,"line":10,"lines":[10,11],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.OrderParameter.origin_eq_of_constant","note":"Right continuity gives γ(0)=c.","plain":"γ(0) = c on initial plateaus","proof_lines":[12,18],"relation":"proof"},{"end_line":17,"line":10,"lines":[10,11],"module":"OAI.Probability.SKValue.Variation.ContactOrigin","name":"OAI.SKValue.IsMinimizer.contact_origin_of_positive","note":"So G(0)=0.","plain":"G(0) = 0 if γ(0)>0","proof_lines":[12,17],"relation":"proof"}],"lean_source":"statement","markdown":"If $c>0$, one also has $G(a)=0$. The only extra case is $a=0$:\nright continuity gives $\\gamma(0)=c>0$, so the mass-removal condition\nat that initial atom forces $G(0)=0$.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":181255},{"anchor":"formal-contact","at":"b181426-181722","end_byte":181722,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":149,"line":137,"lines":[137,139],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.endpoint_condition","note":"The later point z, in both cases.","plain":"Endpoint condition on a hump","proof_lines":[140,149],"relation":"proof"},{"end_line":136,"line":110,"lines":[110,116],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.positive_hump_impossible","note":"The hump lemma that uses these points.","plain":"No positive hump of G","proof_lines":[117,136],"relation":"proof"}],"lean_source":"statement","markdown":"The mean-value theorem on\n$[a,x]$ and $[x,b]$ now provides points $r<s$ with\n$f(r)<0<f(s)$. There is a later point $z\\le b$, with $z<1$, for\nwhich $f(z)\\le f(s)$. For $b<1$ take $z=b$, since contact gives\n$f(b)=0$. For $b=1$, choose $z$ sufficiently close to one and use\n$f(z)=q(z)-z\\le1-z<f(s)$.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":181426},{"anchor":"formal-contact","at":"b181724-182165","end_byte":182165,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":102,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.not_three_slope_pattern","note":"The impossible pattern of three secant slopes.","plain":"No three-slope pattern for convex f′","proof_lines":[88,102],"relation":"proof"},{"end_line":92,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.curvature_convexOn","note":"The convexity on a constant interval.","plain":"Curvature moment convex on plateaus","proof_lines":[78,92],"relation":"part"}],"lean_source":"statement","markdown":"The three intervals $[a,r]$, $[r,s]$ and $[s,z]$ therefore have\nsecant slopes for $f$ of signs respectively negative, positive and\nnonpositive. Applying the mean-value theorem once more gives three\nordered points where $f'$ has those signs. This is impossible when\n$f'=q'-1$ is convex: its middle value cannot exceed the larger of\nits two exterior values. The scalar source supplies precisely this\nnonstrict convexity on a constant interval.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":181724},{"anchor":"formal-contact","at":"b182302-182670","end_byte":182670,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":27,"line":8,"lines":[8,13],"module":"OAI.Probability.SKValue.Coercivity.ZeroGapHump","name":"OAI.SKValue.positive_right_hump_impossible","note":"The zero-coefficient hump; there is no hypothesis on G a.","plain":"No positive hump on the right","proof_lines":[14,27],"relation":"proof"},{"end_line":110,"line":93,"lines":[93,95],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.moment_convexOn_zero","note":"Convexity of q on a zero interval.","plain":"Second moment convex on zero plateaus","proof_lines":[96,110],"relation":"part"}],"lean_source":"statement","markdown":"If $c=0$, the source instead gives convexity of $f=q-t$. A mean-value\nargument between $x$ and $b$ gives $f(s)>0$, and the same right-endpoint\nargument produces $z>s$ with $f(z)\\le f(s)$. Since $f(a)=0$,\n$$\n\\frac{f(s)-f(a)}{s-a}>0\\ge\n\\frac{f(z)-f(s)}{z-s},\n$$\ncontradicting monotonicity of adjacent secant slopes of a convex\nfunction. This case does not need $G(a)=0$.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":182302},{"anchor":"formal-contact","at":"b182729-182798","end_byte":182798,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"G vanishes on (0,1).","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"part"}],"lean_source":"statement","markdown":"Thus $G$ cannot have a positive component, and $G\\equiv0$ on\n$(0,1)$.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":182729},{"anchor":"formal-contact","at":"b182799-182873","end_byte":182873,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":82,"line":75,"lines":[75,82],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"On all of [0,1): differentiation at interior points, with the origin handled separately.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"states"},{"end_line":150,"line":141,"lines":[141,144],"module":"OAI.Probability.SKValue.Variation.ContactVariations","name":"OAI.SKValue.IsMinimizer.moment_eq_at_contact","note":"q(s)=s where G(s)=0.","plain":"q(s) = s at interior contact","proof_lines":[145,150],"relation":"proof"}],"lean_source":"statement","markdown":"Differentiation gives $q(t)=t$ there, with the origin supplied\nseparately.","paper":"long","role":"derivation","section":"10.1 Contact from convexity on constant intervals","start_byte":182799}]},"scalar.formal_finite_gaussian_clause":{"lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Exactly this implication, with the conclusion unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"},{"clause":"0 < T → T < 1 → IsMinimizer W γ → IsDiffusion W γ X →\n      FiniteGaussianConclusion W γ X T","end_line":24,"line":16,"lines":[20,21],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its second clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"short":"Finite-Gaussian branch","statements":[{"anchor":"formal-scalar","at":"b243451-243679","end_byte":243679,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Also gives the limit to the ground-state value as S rises to one, a third clause.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"},{"clause":"0 < T → T < 1 → IsMinimizer W γ → IsDiffusion W γ X →\n      FiniteGaussianConclusion W γ X T","end_line":24,"line":16,"lines":[20,21],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its second clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.\n\nThese are the declarations `scalar_value_target` and\n`finite_gaussian_target`.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":243451},{"anchor":"formal-scalar","at":"b244074-244586","end_byte":244586,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":31,"line":27,"lines":[27,31],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","note":"Uses coordinates 0 to N−1.","plain":"Euler recursion on a uniform mesh","relation":"defines"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The range up to N, with coefficientA zero at j=0.","plain":"Adjacent pairing of A and B","relation":"defines"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"Uses coordinate N.","plain":"The rounded finite spin","relation":"defines"},{"end_line":161,"line":155,"lines":[157,157],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Eventual positivity, for j<N.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The index match in the Gaussian statement is exact. For\n$\\Delta=T/N$ and $t_j=j\\Delta$, the Euler path uses\n$Z_1,\\ldots,Z_N$ and produces $Y_0,\\ldots,Y_N$.\nThe source normalizer indexed from zero corresponds to manuscript\n$c_1,\\ldots,c_N$, and its eventual positivity covers precisely those\nnormalizers. A separate $Z_{N+1}$ performs mean-preserving rounding.\nIt enters $B_{N+1}$ and is no Euler increment. Thus the finite sum is\n$\\sum_{j=0}^{N}A_jB_{j+1}$ with $A_0=0$, using exactly the displayed\nindex ranges.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244074},{"anchor":"evidence-scope","at":"b53118-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Finite-Gaussian clause only: the implication from minimiser and diffusion, unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"},{"clause":"0 < T → T < 1 → IsMinimizer W γ → IsDiffusion W γ X →\n      FiniteGaussianConclusion W γ X T","end_line":24,"line":16,"lines":[20,21],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its second clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"mention","section":"Formal verification","start_byte":53118},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The finite Gaussian half of this sentence, for a minimiser and a supplied diffusion. Conclusion unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62987}]},"scalar.formal_model_transport_open":{"lean":[{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Scalar side of the comparison: coefficient on real time. No Lean theorem maps it to the support model.","plain":"Admissible order parameter on ℝ","relation":"defines"},{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support side of the comparison: coefficient on the subtype Time. No Lean theorem maps it to the scalar model.","plain":"Admissible order parameter γ","relation":"defines"}],"short":"Model transport unproved","statements":[{"anchor":"formal-transport","at":"b249840-250143","end_byte":250143,"heading_id":"relating-the-two-statements","lean":[],"lean_source":"statement","markdown":"Once those obligations are proved, an application must use the same\ntransported minimizer and associated diffusion in all premises. No such\ncross-model transport theorem is supplied here. The manuscript proof in\n§§3–7 and the scalar implementation's independent argument remain\navailable without it.","note":"Not in the Lean: no theorem transports a minimiser or diffusion between the support and scalar models.","paper":"long","role":"statement","section":"12.3 Relating the two statements","start_byte":249840},{"anchor":"formal-transport","at":"b246632-246882","end_byte":246882,"heading_id":"relating-the-two-statements","lean":[],"lean_source":"statement","markdown":"The support and scalar models encode closely related control problems.\nA direct formal application still needs a specified map of their objects.\nThe following contract lists the obligations, and it asserts no\nequivalence of their complete interfaces.","note":"Not in the Lean: no theorem connects the support and scalar models. The obligations below are a contract, proved nowhere.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":246632},{"anchor":"evidence-scope","at":"b53545-53657","end_byte":53657,"heading_id":"formal-verification","lean":[],"lean_source":"statement","markdown":"The two formal statements use separate models of the order\nparameter, which have not been related to each other.","note":"Not in the Lean: the support and scalar models define their objects separately, and no Lean theorem connects them.","paper":"short","role":"statement","section":"Formal verification","start_byte":53545},{"anchor":"formal-transport","at":"b246884-247258","end_byte":247258,"heading_id":"relating-the-two-statements","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support: a function on Time.","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Scalar: a function on ℝ, with conditions only on [0,1).","plain":"Admissible order parameter on ℝ","relation":"defines"}],"lean_source":"statement","markdown":"**Coefficients.** Restrict the scalar real-time coefficient to $[0,1)$.\nConversely, choose an extension compatible with the scalar right-continuity\npredicate. Prove equality almost everywhere of the coefficients used by each\npayoff and penalty. The scalar endpoint and exterior values are unconstrained\nby the shape conditions, so two extensions need not be literally equal.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":246884},{"anchor":"formal-transport","at":"b247260-247525","end_byte":247525,"heading_id":"relating-the-two-statements","lean":[{"end_line":29,"line":24,"lines":[24,29],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.BrownianSystem","note":"Support: Ω is a parameter.","plain":"Probability space with Brownian motion","relation":"defines"},{"end_line":47,"line":40,"lines":[40,47],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.BrownianSpace","note":"Scalar: Ω is a field.","plain":"Bundled space with Brownian motion","relation":"defines"}],"lean_source":"statement","markdown":"**Brownian space.** Bundle or unbundle the same sample space, sigma algebra,\nprobability law and Brownian process. Identify the natural filtration and\nevery elapsed-increment filtration. Equality in law alone does not identify\nadmissible controls on a chosen space.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":247260},{"anchor":"formal-transport","at":"b247527-247851","end_byte":247851,"heading_id":"relating-the-two-statements","lean":[{"end_line":47,"line":43,"lines":[43,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support: integral over [t,1], using α(s−t).","plain":"Control payoff, sample by sample","relation":"defines"},{"end_line":66,"line":62,"lines":[62,66],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.controlPayoff","note":"Scalar: integral over [0,1−t], using α(s).","plain":"Expected control payoff","relation":"defines"}],"lean_source":"statement","markdown":"**Controls and payoffs.** Bundle an admissible scalar control as a support\ncontrol and conversely. With $s=t+r$, prove equality between the absolute\nintegral over $[t,1]$ using $\\alpha(s-t)$ and the elapsed integral over\n$[0,1-t]$ using $\\alpha(r)$. Preserve progressive measurability and the\npointwise bound $|\\alpha|\\le1$.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":247527},{"anchor":"formal-transport","at":"b248168-248431","end_byte":248431,"heading_id":"relating-the-two-statements","lean":[{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support: indexed by ℝ≥0.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":88,"line":83,"lines":[83,88],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Scalar: indexed by ℝ.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"**Diffusion and moments.** Reindex the same process between nonnegative and\nreal time, and stop it at one where required. Transport the integral equation\non one full-measure event for all $t\\in[0,1]$, path continuity, initial value,\ngradient and moment integrals.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":248168},{"anchor":"formal-transport","at":"b248433-248651","end_byte":248651,"heading_id":"relating-the-two-statements","lean":[{"end_line":75,"line":68,"lines":[70,70],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"IsProgressive, for the natural filtration.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":88,"line":83,"lines":[84,85],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsDiffusion","note":"Adapted after stopping, with continuity almost surely.","plain":"X is a diffusion for γ","relation":"defines"}],"lean_source":"statement","markdown":"The diffusion comparison has a version issue. The support predicate asks\nfor a globally progressive process, while the scalar predicate asks for\nadaptedness of the stopped process and almost-sure continuity on $[0,1]$.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":248433},{"anchor":"obligations","at":"b250393-250710","end_byte":250710,"heading_id":"logical-limits-of-the-argument","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model, namespace ZeroTemperatureSK. No theorem relates it to the scalar model.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"},{"end_line":24,"line":16,"lines":[17,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Scalar model, on its own structures. No theorem relates it to the support model.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"part"}],"lean_source":"statement","markdown":"The full-support theorem and scalar-value theorem use related but\ndifferent formal representations. Their mathematical conclusions can be\ndiscussed together, but a direct formal implication would require the\ncoefficient, Brownian law, control value and diffusion translations\ndescribed in §12, with common witnesses.","paper":"long","role":"explanation","section":"12.4 Logical limits of the argument","start_byte":250393},{"anchor":"formal-transport","at":"b249489-249712","end_byte":249712,"heading_id":"relating-the-two-statements","lean":[],"lean_source":"statement","markdown":"This settles the version construction at the ordinary mathematical level.\nIts implementation in the two formal predicates, and the coefficient,\nfiltration and value identifications listed above, remain separate\nobligations.","note":"Not in the Lean: neither the version construction nor the identifications. No theorem connects the two models.","paper":"long","role":"explanation","section":"12.3 Relating the two statements","start_byte":249489}]},"scalar.formal_parisi_clause":{"lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"clause":"IsMinimizer W γ → groundStateValue = parisi W γ","end_line":24,"line":16,"lines":[23,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its third clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"short":"Parisi identity branch","statements":[{"anchor":"formal-parisi-clause","at":"b245775-246152","end_byte":246152,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"clause":"IsMinimizer W γ → groundStateValue = parisi W γ","end_line":24,"line":16,"lines":[23,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its third clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"The third declaration, `ground_state_parisi_target`, assumes a minimizer\nalone and concludes $P_*=\\mathcal P(\\gamma)$. Its proof chooses\n$X=\\texttt{diffusionProcess}\\ W\\ \\gamma$ and proves the diffusion\npredicate for that process before applying the equality with a supplied\ndiffusion. No diffusion is required from the caller, and none of these\nclauses constructs a minimizer.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":245775},{"anchor":"evidence-scope","at":"b53314-53437","end_byte":53437,"heading_id":"formal-verification","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"clause":"IsMinimizer W γ → groundStateValue = parisi W γ","end_line":24,"line":16,"lines":[23,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its third clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"and its third\nclause identifies the ground-state and Parisi values from a minimiser,\nconstructing the diffusion internally.","paper":"short","role":"mention","section":"Formal verification","start_byte":53314},{"anchor":"gaussian-expectation-limit","at":"b193702-193788","end_byte":193788,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":27,"line":25,"lines":[25,27],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"No diffusion in the hypotheses.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"end_line":77,"line":74,"lines":[74,77],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.diffusionProcess_isDiffusion","note":"The constructed diffusion.","plain":"The constructed process solves the equation","proof_lines":[75,77],"relation":"proof"}],"lean_source":"statement","markdown":"The third scalar endpoint constructs its own diffusion before making\nthis application.","paper":"long","role":"explanation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193702},{"anchor":"formal-parisi-clause","at":"b246153-246331","end_byte":246331,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":15,"line":13,"lines":[13,15],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"First of the three, line 13.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":27,"line":25,"lines":[25,27],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Last of the three, ending line 27.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"}],"lean_source":"statement","markdown":"The exact declarations are\n[MainConclusions, lines 13–27][scalar-main-conclusions]; their analytic\nsuppliers and applications are in [ContactAssembly][scalar-contact-assembly].","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":246153}]},"scalar.formal_scalar_clause":{"lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Exactly this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"},{"clause":"IsMinimizer W γ → IsDiffusion W γ X → ScalarValueConclusion W γ X","end_line":24,"line":16,"lines":[18,18],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its first clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"short":"Scalar-value branch","statements":[{"anchor":"formal-scalar","at":"b243180-243450","end_byte":243450,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Exactly this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"},{"clause":"IsMinimizer W γ → IsDiffusion W γ X → ScalarValueConclusion W γ X","end_line":24,"line":16,"lines":[18,18],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its first clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"The second formal statement has three clauses. Given a minimizer and an\nassociated diffusion, it establishes convergence of the finite-spin\nground-state values, a bounded martingale with a spin-valued terminal\nlimit, convergence of the gradient and the value identities.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":243180},{"anchor":"formal-scalar","at":"b243601-243924","end_byte":243924,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"The first named declaration: this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second named declaration, for the same minimiser and supplied diffusion, with 0<T<1.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"The compact-strip bounds the implementation derives, for any order parameter.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"proof"},{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"The unit curvature moment, from contact.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"},{"clause":"IsMinimizer W γ → IsDiffusion W γ X → ScalarValueConclusion W γ X","end_line":24,"line":16,"lines":[18,18],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its first clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"These are the declarations `scalar_value_target` and\n`finite_gaussian_target`. Their caller supplies the same minimizer and\nassociated diffusion. The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243601},{"anchor":"evidence-scope","at":"b53118-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Value clause only: the implication from minimiser and diffusion, unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"},{"clause":"IsMinimizer W γ → IsDiffusion W γ X → ScalarValueConclusion W γ X","end_line":24,"line":16,"lines":[18,18],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its first clause.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"mention","section":"Formal verification","start_byte":53118},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"The value-identity half of this sentence, for a minimiser and a supplied diffusion. Conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62987},{"anchor":"scalar-identities","at":"b60955-61068","end_byte":61068,"heading_id":"further-reading","lean":[{"clause":"groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ","end_line":118,"line":106,"lines":[117,117],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Scalar model. A clause of the conclusion of scalar_value_target: the first moment of the curvature, integrated in time, gives the value.","plain":"The scalar-value conclusion","relation":"part"}],"lean_source":"statement","markdown":"They use the first curvature moment\n$\\mathbb E[a]$, distinct from the second moment $\\mathbb E[a^2]=1$\nused here.","paper":"expository","role":"explanation","section":"Further reading","start_byte":60955},{"anchor":"formal-scalar","at":"b244074-244586","end_byte":244586,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":31,"line":27,"lines":[27,31],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","note":"Uses coordinates 0 to N−1.","plain":"Euler recursion on a uniform mesh","relation":"defines"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The range up to N.","plain":"Adjacent pairing of A and B","relation":"defines"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"Uses coordinate N.","plain":"The rounded finite spin","relation":"defines"}],"lean_source":"statement","markdown":"The index match in the Gaussian statement is exact. For\n$\\Delta=T/N$ and $t_j=j\\Delta$, the Euler path uses\n$Z_1,\\ldots,Z_N$ and produces $Y_0,\\ldots,Y_N$.\nThe source normalizer indexed from zero corresponds to manuscript\n$c_1,\\ldots,c_N$, and its eventual positivity covers precisely those\nnormalizers. A separate $Z_{N+1}$ performs mean-preserving rounding.\nIt enters $B_{N+1}$ and is no Euler increment. Thus the finite sum is\n$\\sum_{j=0}^{N}A_jB_{j+1}$ with $A_0=0$, using exactly the displayed\nindex ranges.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244074},{"anchor":"formal-scalar","at":"b244587-244865","end_byte":244865,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":22,"line":21,"lines":[21,22],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coordinate","note":"The zero-based coordinate.","plain":"Coordinate of a Gaussian vector","relation":"defines"},{"end_line":126,"line":124,"lines":[124,126],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.rawCoefficient","note":"a(t_j,Y_j).","plain":"Curvature along the Euler path","relation":"defines"},{"end_line":130,"line":128,"lines":[128,130],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.normalizer","note":"c_{j+1}.","plain":"Normaliser of the raw coefficient","relation":"defines"}],"lean_source":"statement","markdown":"Total definitions at $N=0$ make no extra mathematical\nclaim about the construction for $N\\ge1$.\n\nMore explicitly, the\nzero-based formal quantities `coordinate N j`, `rawCoefficient ... j`\nand `normalizer ... j` correspond to $Z_{j+1}$,\n$a(t_j,Y_j)$ and $c_{j+1}$ for $0\\le j<N$.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244587},{"anchor":"formal-scalar","at":"b244866-245008","end_byte":245008,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":161,"line":155,"lines":[157,157],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Eventually in N, for all j<N, with T fixed outside.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The positivity quantifiers\nare $\\exists N_0\\,\\forall N\\ge N_0\\,\\forall j<N$, simultaneously over\nthe growing mesh, for each already fixed $T$.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244866},{"anchor":"formal-scalar","at":"b245067-245256","end_byte":245256,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":104,"line":102,"lines":[102,104],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gradientStopped","note":"Stopped at U from time one.","plain":"Gradient martingale with terminal limit U","relation":"defines"},{"end_line":118,"line":106,"lines":[109,114],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"U²=1, and convergence almost surely and in L².","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The formal martingale stops at its terminal random variable $U$ for all\nlater times. Then $U^2=1$ almost surely, and the gradient converges to $U$ almost surely and in $L^2$ from below\none.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":245067},{"anchor":"formal-scalar","at":"b245257-245452","end_byte":245452,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":111,"line":97,"lines":[97,101],"module":"OAI.Probability.SKValue.Equations.IntegralAbsSqrtSecond","name":"OAI.SKValue.IsDiffusion.terminal_gradient_product","note":"The internal lemma.","plain":"U X₁ = |X₁| almost surely","proof_lines":[102,111],"relation":"proof"}],"lean_source":"statement","markdown":"The internal terminal-gradient lemma proves $UX_1=|X_1|$\nwithout requiring $X_1$ to have no atom at zero. That identity is used\ninside the implementation and is no extra conjunct of the endpoint.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":245257}]},"scalar.integrated_coercivity":{"lean":[{"clause":"(1/1000:ℝ)*(∫ x,v x^4*f x)≤∫ x,(z x^2-12*v x*w x^2+6*v x^4)*f x","end_line":58,"line":12,"lines":[26,26],"module":"OAI.Probability.SKValue.Coercivity.IntegratedCoercivity","name":"OAI.SKValue.Certificate.integrated_coercivity","note":"Scalar model, as the scalar proof uses it: the 1/1000 bound, assuming the correction integrates to zero, with sign and growth hypotheses.","plain":"Integrated coercivity in the value proof","proof_lines":[27,58],"relation":"states"},{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Support model, no caller in either proof. Sharper constant 1661507/1521000000, vanishing flux as hypotheses, written with R=r^2 and b=-w/r. The next row reaches 1/1000.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"clause":"(1 / 1000 : ℝ) * (∫ t, ρ t * (v t)^4)","end_line":202,"line":196,"lines":[197,197],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral","note":"The final step from 1661507/1521000000 down to 1/1000.","plain":"From sharp constant to 1/1000","proof_lines":[199,202],"relation":"part"}],"short":"Conditional integrated coercivity","statements":[{"anchor":"curvature","at":"b62768-62931","end_byte":62931,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Support model, no caller in either proof. Sharper constant 1661507/1521000000, vanishing flux as hypotheses, written with R=r^2 and b=-w/r. The next row reaches 1/1000.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"clause":"(1 / 1000 : ℝ) * (∫ t, ρ t * (v t)^4)","end_line":202,"line":196,"lines":[197,197],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral","note":"The final step from 1661507/1521000000 down to 1/1000.","plain":"From sharp constant to 1/1000","proof_lines":[199,202],"relation":"part"},{"clause":"(1/1000:ℝ)*(∫ x,v x^4*f x)≤∫ x,(z x^2-12*v x*w x^2+6*v x^4)*f x","end_line":58,"line":12,"lines":[26,26],"module":"OAI.Probability.SKValue.Coercivity.IntegratedCoercivity","name":"OAI.SKValue.Certificate.integrated_coercivity","note":"Scalar model, as the scalar proof uses it: the 1/1000 bound, assuming the correction integrates to zero, with sign and growth hypotheses.","plain":"Integrated coercivity in the value proof","proof_lines":[27,58],"relation":"states"}],"lean_source":"statement","markdown":"Integration\nremoves only $\\mathcal W(F_2)$, giving\n\n$$\n\\mathbb E_f[z^2-12vw^2+6v^4]\n \\ge\\frac{1661507}{1521000000}\\mathbb E_fv^4\n \\ge\\frac1{1000}\\mathbb E_fv^4.\n$$","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":62768},{"anchor":"curvature","at":"b24108-24337","end_byte":24337,"heading_id":"the-integrated-estimate","lean":[{"end_line":46,"line":35,"lines":[35,44],"module":"OAI.Probability.SKValue.Coercivity.Coercivity","name":"OAI.SKValue.Certificate.rational_identity","note":"Scalar model. The polynomial identity, with B=b−2v and b=−w/r; remainder holds the κv⁴ term. R≥0 is the next row.","proof_lines":[45,46],"relation":"part"},{"end_line":76,"line":48,"lines":[48,52],"module":"OAI.Probability.SKValue.Coercivity.Coercivity","name":"OAI.SKValue.Certificate.pointwise_coercivity","note":"Scalar model. Under the sign hypotheses, the terms of R are nonnegative. This is the decomposition as an inequality.","proof_lines":[53,76],"relation":"part"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Lemma 4.1][FS] supplies an explicit polynomial $F_2$\nand a decomposition\n$$\nz^2-12vw^2+6v^4\n=\\mathcal W(F_2)+\\kappa v^4+\\mathcal R,\n\\qquad\n\\mathcal R\\geq0,\\qquad\n\\kappa=\\frac{1661507}{1521000000}>\\frac1{1000}.\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":24108},{"anchor":"curvature","at":"b25022-25207","end_byte":25207,"heading_id":"the-integrated-estimate","lean":[{"end_line":168,"line":122,"lines":[122,128],"module":"OAI.Probability.SKValue.Coercivity.IntegratedCoercivity","name":"OAI.SKValue.BackwardShape.integrated_certificate","note":"Scalar model: (5) for the density exp(cA) times a forward weight, from the shapes and the joint sign hJ.","proof_lines":[129,168],"relation":"states"},{"clause":"(1/1000:ℝ)*(∫ x,v x^4*f x)≤∫ x,(z x^2-12*v x*w x^2+6*v x^4)*f x","end_line":58,"line":12,"lines":[26,26],"module":"OAI.Probability.SKValue.Coercivity.IntegratedCoercivity","name":"OAI.SKValue.Certificate.integrated_coercivity","note":"Scalar model, as the scalar proof uses it: the 1/1000 bound, assuming the correction integrates to zero, with sign and growth hypotheses.","plain":"Integrated coercivity in the value proof","proof_lines":[27,58],"relation":"states"},{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Full-support model: sharper constant, with signs and vanishing flux assumed. No other Lean proof calls it.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"}],"lean_source":"statement","markdown":"Integration removes $\\mathcal W(F_2)$ and keeps the fourth-moment\nmargin, giving\n$$\n\\int_{\\mathbb R}(z^2-12vw^2+6v^4)f\\,dx\n\\geq\\frac1{1000}\\int_{\\mathbb R}v^4f\\,dx.           \\tag{5}\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":25022},{"anchor":"curvature-transfer","at":"b53645-54166","end_byte":54166,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":188,"line":173,"lines":[179,180],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_feedback_coercivity","note":"The final inequality at a finite-step grid time, along the feedback path, multiplied through by c². The correction C stays inside the proof.","plain":"Finite-step coercivity along the diffusion","proof_lines":[181,188],"relation":"part"},{"end_line":162,"line":106,"lines":[141,153],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The correction in Lean: add certDivergence, compare pointwise, integrate it away on the half-line, then reflect.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Whole-line form with the sharper constant, flux and signs as hypotheses. Nothing in this tree calls it.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"clause":"(1/1000:ℝ)*(∫ x,v x^4*f x)≤∫ x,(z x^2-12*v x*w x^2+6*v x^4)*f x","end_line":58,"line":12,"lines":[26,26],"module":"OAI.Probability.SKValue.Coercivity.IntegratedCoercivity","name":"OAI.SKValue.Certificate.integrated_coercivity","note":"Scalar model: the 1/1000 bound only, assuming directly that the correction integrates to zero.","plain":"Integrated coercivity in the value proof","proof_lines":[27,58],"relation":"part"}],"lean_source":"statement","markdown":"At this stage write\n$R_c=d^2-12cab^2+6c^2a^4$. The source constructs a correction\n$C$ from the spatial derivatives and the density, with\n\n$$\n\\mathbb E[C(t,X_t)]=0,\\qquad\nR_c(t,x)+C(t,x)\\ge\\frac{c^2}{1000}a(t,x)^4.\n$$\n\nA rational polynomial identity gives the second inequality by writing\nthe remainder as a sum of nonnegative terms. The density bounds and vanishing boundary\nproducts justify the first. Consequently\n\n$$\nq'''(t)=\\mathbb E[R_c(t,X_t)]\n=\\mathbb E[(R_c+C)(t,X_t)]\n\\ge\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4].\n$$","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53645}]},"scalar.lean_gaussian_target":{"lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Exactly this implication, with the conclusion unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"short":"finite_gaussian_target","statements":[{"anchor":"formal-scalar","at":"b243451-243679","end_byte":243679,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Also gives the limit to the ground-state value as S rises to one, a third clause.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.\n\nThese are the declarations `scalar_value_target` and\n`finite_gaussian_target`.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":243451},{"anchor":"formal-scalar","at":"b244074-244586","end_byte":244586,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":31,"line":27,"lines":[27,31],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","note":"Uses coordinates 0 to N−1.","plain":"Euler recursion on a uniform mesh","relation":"defines"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The range up to N, with coefficientA zero at j=0.","plain":"Adjacent pairing of A and B","relation":"defines"},{"end_line":139,"line":136,"lines":[136,139],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.roundedSign","note":"Uses coordinate N.","plain":"The rounded finite spin","relation":"defines"},{"end_line":161,"line":155,"lines":[157,157],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Eventual positivity, for j<N.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The index match in the Gaussian statement is exact. For\n$\\Delta=T/N$ and $t_j=j\\Delta$, the Euler path uses\n$Z_1,\\ldots,Z_N$ and produces $Y_0,\\ldots,Y_N$.\nThe source normalizer indexed from zero corresponds to manuscript\n$c_1,\\ldots,c_N$, and its eventual positivity covers precisely those\nnormalizers. A separate $Z_{N+1}$ performs mean-preserving rounding.\nIt enters $B_{N+1}$ and is no Euler increment. Thus the finite sum is\n$\\sum_{j=0}^{N}A_jB_{j+1}$ with $A_0=0$, using exactly the displayed\nindex ranges.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244074},{"anchor":"evidence-scope","at":"b53118-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Finite-Gaussian clause only: the implication from minimiser and diffusion, unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"mention","section":"Formal verification","start_byte":53118},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The finite Gaussian half of this sentence, for a minimiser and a supplied diffusion. Conclusion unfolded in FiniteGaussianConclusion.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The three clauses: normaliser positivity, mesh limit and endpoint limit.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62987}]},"scalar.lean_parisi_target":{"lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"}],"short":"ground_state_parisi_target","statements":[{"anchor":"formal-parisi-clause","at":"b245775-246152","end_byte":246152,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"}],"lean_source":"statement","markdown":"The third declaration, `ground_state_parisi_target`, assumes a minimizer\nalone and concludes $P_*=\\mathcal P(\\gamma)$. Its proof chooses\n$X=\\texttt{diffusionProcess}\\ W\\ \\gamma$ and proves the diffusion\npredicate for that process before applying the equality with a supplied\ndiffusion. No diffusion is required from the caller, and none of these\nclauses constructs a minimizer.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":245775},{"anchor":"evidence-scope","at":"b53314-53437","end_byte":53437,"heading_id":"formal-verification","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"}],"lean_source":"statement","markdown":"and its third\nclause identifies the ground-state and Parisi values from a minimiser,\nconstructing the diffusion internally.","paper":"short","role":"mention","section":"Formal verification","start_byte":53314}]},"scalar.lean_scalar_target":{"lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Exactly this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"short":"scalar_value_target","statements":[{"anchor":"formal-scalar","at":"b243180-243450","end_byte":243450,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Exactly this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The second formal statement has three clauses. Given a minimizer and an\nassociated diffusion, it establishes convergence of the finite-spin\nground-state values, a bounded martingale with a spin-valued terminal\nlimit, convergence of the gradient and the value identities.","paper":"long","role":"statement","section":"12.2 The separate scalar statement","start_byte":243180},{"anchor":"formal-scalar","at":"b243601-243924","end_byte":243924,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"The first named declaration: this implication, with the conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"states"},{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second named declaration, for the same minimiser and supplied diffusion, with 0<T<1.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"The compact-strip bounds the implementation derives, for any order parameter.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"proof"},{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"The unit curvature moment, from contact.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"These are the declarations `scalar_value_target` and\n`finite_gaussian_target`. Their caller supplies the same minimizer and\nassociated diffusion. The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243601},{"anchor":"evidence-scope","at":"b53118-53313","end_byte":53313,"heading_id":"formal-verification","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Value clause only: the implication from minimiser and diffusion, unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate\n[formal statement of the scalar consequences][value-challenge] covers\nthe value identity and the finite Gaussian approximation, with\nminimiser and diffusion premises for those clauses,","paper":"short","role":"mention","section":"Formal verification","start_byte":53118},{"anchor":"source-record","at":"b62987-63151","end_byte":63151,"heading_id":"formal-verification-and-sources","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"The value-identity half of this sentence, for a minimiser and a supplied diffusion. Conclusion unfolded in ScalarValueConclusion.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction concluded by the target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"A separate [formal statement][value-interface] concerns the\nvalue identity and the finite Gaussian approximation, and its\n[proof][value-proof] has not been checked.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62987}]},"scalar.manuscript_terminal_value":{"lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model: U and the value identities. Quadratic variations appear integrated in the last rows. The unit moment comes from contact without full support.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"},{"end_line":63,"line":41,"lines":[41,48],"module":"OAI.Probability.SKValue.Equations.RealFiltration","name":"OAI.SKValue.IsDiffusion.unitMomentMartingale","note":"Scalar model: the gradient process is a unit-moment martingale, given strip bounds and contact. No differential form of dM or its quadratic variation is stated.","plain":"The gradient is a unit-moment martingale","proof_lines":[49,63],"relation":"part"},{"end_line":104,"line":87,"lines":[87,95],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.IsDiffusion.terminal_brownian_covariance","note":"Scalar model: E[U B_T] equals the integral of E a up to T, the integrated form of the covariation with the noise.","plain":"Covariance of U with Brownian motion","proof_lines":[96,104],"relation":"part"}],"short":"Terminal martingale value formulas","statements":[{"anchor":"value","at":"b83463-83841","end_byte":83841,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model, diffusion supplied by the caller. Gives U, the martingale and the identities; the tail bound by 1−T is not in the Lean.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"}],"lean_source":"statement","markdown":"**Proposition 6.1 (terminal value).** For the minimizer and diffusion of\nTheorem 5.1 there is a spin-valued terminal variable $U_1$ such that\n$u(t,X_t)=\\mathbb E[U_1\\mid\\mathcal F_t]$ and\n$$\nP_*=\\mathbb E[U_1B_1]=\\int_0^1\\mathbb E a(t,X_t)\\,dt.\n\\tag{6.1}\n$$\nIts finite-cutoff tail lies in $[0,1-T]$. Besides the support theorem,\nthe proposition uses the attained Parisi formula.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":83463},{"anchor":"value","at":"b83843-84043","end_byte":84043,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":333,"line":288,"lines":[288,293],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.exists_continuous_terminal","note":"Almost sure convergence.","plain":"Terminal limit of a continuous martingale","proof_lines":[294,333],"relation":"proof"},{"end_line":391,"line":370,"lines":[370,375],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_mean_square","note":"L² convergence.","plain":"L² convergence to U","proof_lines":[376,391],"relation":"proof"},{"end_line":368,"line":335,"lines":[335,340],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_square","note":"U²=1 almost surely.","plain":"U² = 1 almost surely","proof_lines":[341,368],"relation":"proof"}],"lean_source":"statement","markdown":"The bounded martingale $u(t,X_t)$ converges almost surely and in $L^2$\nto $U_1$. Its second moment tends to one because $q(t)=t$.\nSince $|U_1|\\le1$, $\\mathbb E U_1^2=1$ forces $U_1^2=1$ almost surely.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":83843},{"anchor":"value","at":"b85078-85249","end_byte":85249,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":79,"line":37,"lines":[73,79],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"hPB combines the identities; hCov and he give the curvature integral.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"},{"end_line":337,"line":277,"lines":[277,294],"module":"OAI.Probability.SKValue.Processes.MeshBrownian","name":"OAI.SKValue.diffusion_gradient_brownian_covariance","note":"Covariance of the gradient with Brownian motion, by Euler approximation.","plain":"Gradient covariance with Brownian motion","proof_lines":[295,337],"relation":"proof"}],"lean_source":"statement","markdown":"Combining the identities and using stochastic integral covariance gives\n$$\nP_*=\\mathbb E|X_1|-\\int_0^1t\\gamma(t)\\,dt\n =\\mathbb E[U_1B_1]=\\int_0^1\\mathbb E a(t,X_t)\\,dt.\n$$","note":"Lean obtains the covariance by Euler approximation, with no stochastic integral.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":85078},{"anchor":"scalar-value","at":"b36323-37016","end_byte":37016,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"(10), with U and its limits, at a minimiser, in the scalar model. The differential of M and its brackets are not written.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"}],"lean_source":"statement","markdown":"The support identities give a scalar representation of $P_*$. Set\n$M_t=u(t,X_t)$, the bounded gradient martingale, which satisfies\n$dM_t=a(t,X_t)\\,dB_t$. Its quadratic variations distinguish the two\nHessian moments:\n$$\nd\\langle M\\rangle_t=a(t,X_t)^2\\,dt,\n\\qquad d\\langle M,B\\rangle_t=a(t,X_t)\\,dt.\n$$\nFull support gives $\\mathbb E a(t,X_t)^2=1$, so each increment\n$M_t-M_s$ has unconditional variance $t-s$. This normalises the\ndiscrete increments below. The first moment of $a$ gives instead the\ncovariance with the driving noise. Let $U$ be the almost-sure and\n$L^2$ limit of $M_t$. The value identity is\n$$\nP_*=\\mathbb E[UB_1]\n   =\\int_0^1\\mathbb E[a(t,X_t)]\\,dt.                \\tag{10}\n$$","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":36323},{"anchor":"scalar-value","at":"b37017-37513","end_byte":37513,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model: U and the value identities. Quadratic variations appear integrated in the last rows. The unit moment comes from contact without full support.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Boundedness gives $|U|\\leq1$, while $\\mathbb E M_t^2\\to1$ gives\n$\\mathbb E U^2=1$, so $U^2=1$ almost surely.\n\nThe cancellation behind (10) runs as follows. Terminal convergence of\nthe convex value functions gives $UX_1=|X_1|$ (the equality is\nautomatic on $X_1=0$), and Itô's formula gives\n$\\mathcal P(\\gamma)=\\mathbb E|X_1|-\\int_0^1t\\gamma(t)\\,dt$.\nMultiplying $X_1=B_1+\\int_0^1\\gamma(t)M_t\\,dt$ by $U$ cancels\nthat integral, because $\\mathbb E[UM_t]=t$, and the Parisi formula\nthen gives (10).","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37017},{"anchor":"scalar-identities","at":"b60861-61068","end_byte":61068,"heading_id":"further-reading","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Conclusion also gives convergence of the finite-spin values and the martingale property. Scalar model, with diffusion supplied by the caller.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"}],"lean_source":"statement","markdown":"Further martingale arguments recover the ground-state value from the\nterminal slope response. They use the first curvature moment\n$\\mathbb E[a]$, distinct from the second moment $\\mathbb E[a^2]=1$\nused here.","paper":"expository","role":"mention","section":"Further reading","start_byte":60861},{"anchor":"dependencies","at":"b7776-7961","end_byte":7961,"heading_id":"outline-of-the-proof","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model: the same formulas, with its own Guerra bound and lower bound in place of the attained Parisi formula.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":118,"line":106,"lines":[115,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The three formulas for the ground-state value.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The value argument then uses bounded-martingale convergence and the\nattained Parisi formula to express $P_*$ as a terminal Brownian correlation\nand as an integral of expected curvature.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":7776},{"anchor":"approximation-choices","at":"b30520-30707","end_byte":30707,"heading_id":"the-order-of-the-approximation-limits","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model: U and the value identities. Quadratic variations appear integrated in the last rows. The unit moment comes from contact without full support.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"part"},{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The full conjunction: finite-spin limit, terminal variable U with its limits, and the three value identities.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Terminal value.** Keep the chosen minimiser and diffusion fixed. Bounded\nmartingale convergence and integrability of $\\gamma$ justify\n$t\\uparrow1$ without any terminal derivative bound.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":30520},{"anchor":"value","at":"b83548-83646","end_byte":83646,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model. The stopped gradient is a martingale ending at U, with |U|≤1 and U²=1 almost surely.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"end_line":118,"line":106,"lines":[109,112],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"U is bounded with U²=1, and the gradient stopped at U is a martingale.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":55,"line":27,"lines":[27,32],"module":"OAI.Probability.SKValue.Processes.TerminalCondExp","name":"OAI.SKValue.UnitMomentMartingale.terminal_condExp","note":"The conditional-expectation form, for s<1.","plain":"Conditional expectation of the terminal U","proof_lines":[33,55],"relation":"proof"}],"lean_source":"statement","markdown":"there is a spin-valued terminal variable $U_1$ such that\n$u(t,X_t)=\\mathbb E[U_1\\mid\\mathcal F_t]$","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":83548},{"anchor":"value","at":"b83651-83720","end_byte":83720,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"clause":"∃ U : W.Ω → ℝ,\n    Martingale (gradientStopped W γ X U) W.filtration W.μ ∧\n    (∀ᵐ ω ∂W.μ, |U ω| ≤ 1 ∧ (U ω)^2 = 1 ∧\n      Tendsto (fun t ↦ gradient W γ t (X t ω)) (𝓝[<] (1 : ℝ)) (𝓝 (U ω))) ∧\n    Tendsto (fun t ↦ ∫ ω, (gradient W γ t (X t ω) - U ω)^2 ∂W.μ)\n      (𝓝[<] (1 : ℝ)) (𝓝 (0 : ℝ)) ∧\n    groundStateValue = (∫ ω, |X 1 ω| ∂W.μ) - ∫ t in (0 : ℝ)..1, t * γ.coeff t ∧\n    groundStateValue = ∫ ω, U ω * W.B 1 ω ∂W.μ ∧\n    groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ ∧\n    IntervalIntegrable (fun t ↦ ∫ ω, curvature W γ t (X t ω) ∂W.μ) volume 0 1","clause_in":"OAI.SKValue.ScalarValueConclusion","clause_lines":[109,118],"clause_module":"OAI.Probability.SKValue.Model","end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model. The two identities hold for groundStateValue, the limit that defines P_*.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"end_line":118,"line":106,"lines":[116,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The two formulas, with integrability of the curvature mean.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"$$\nP_*=\\mathbb E[U_1B_1]=\\int_0^1\\mathbb E a(t,X_t)\\,dt.\n\\tag{6.1}\n$$","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":83651},{"anchor":"value","at":"b83721-83762","end_byte":83762,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":34,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"The lower end of the interval only.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"}],"lean_source":"statement","markdown":"Its finite-cutoff tail lies in $[0,1-T]$.","note":"The Lean conclusion has no bound by 1−T.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":83721},{"anchor":"value","at":"b83843-83923","end_byte":83923,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":333,"line":288,"lines":[288,293],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.exists_continuous_terminal","note":"Almost sure convergence.","plain":"Terminal limit of a continuous martingale","proof_lines":[294,333],"relation":"proof"},{"end_line":391,"line":370,"lines":[370,375],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_mean_square","note":"L² convergence.","plain":"L² convergence to U","proof_lines":[376,391],"relation":"proof"},{"end_line":63,"line":41,"lines":[41,48],"module":"OAI.Probability.SKValue.Equations.RealFiltration","name":"OAI.SKValue.IsDiffusion.unitMomentMartingale","note":"The gradient along X has the structure these lemmas use.","plain":"The gradient is a unit-moment martingale","proof_lines":[49,63],"relation":"proof"}],"lean_source":"statement","markdown":"The bounded martingale $u(t,X_t)$ converges almost surely and in $L^2$\nto $U_1$.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":83843},{"anchor":"value","at":"b83924-84043","end_byte":84043,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":368,"line":335,"lines":[335,340],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_square","note":"The spin conclusion.","plain":"U² = 1 almost surely","proof_lines":[341,368],"relation":"proof"}],"lean_source":"statement","markdown":"Its second moment tends to one because $q(t)=t$.\nSince $|U_1|\\le1$, $\\mathbb E U_1^2=1$ forces $U_1^2=1$ almost surely.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":83924},{"anchor":"value","at":"b84044-84177","end_byte":84177,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":55,"line":27,"lines":[27,32],"module":"OAI.Probability.SKValue.Processes.TerminalCondExp","name":"OAI.SKValue.UnitMomentMartingale.terminal_condExp","note":"The identification.","plain":"Conditional expectation of the terminal U","proof_lines":[33,55],"relation":"proof"}],"lean_source":"statement","markdown":"Uniform integrability identifies $u(t,X_t)=\\mathbb E[U_1\\mid\\mathcal F_t]$,\nwhich is the conditional-expectation identity used below.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84044},{"anchor":"value","at":"b84179-84260","end_byte":84260,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":44,"line":37,"lines":[37,39],"module":"OAI.Probability.SKValue.Equations.IntegralAbsSqrtSecond","name":"OAI.SKValue.IsDiffusion.L1_terminal","note":"L1 convergence, which is what Lean uses.","plain":"Xₜ → X₁ in L¹","proof_lines":[40,44],"relation":"proof"}],"lean_source":"statement","markdown":"The drift bound $|\\gamma u|\\le\\gamma\\in L^1$ gives convergence of $X_t$\nto $X_1$,","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84179},{"anchor":"value","at":"b84179-84493","end_byte":84493,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":141,"line":117,"lines":[117,122],"module":"OAI.Probability.SKValue.Evolution.TerminalError","name":"OAI.SKValue.gradient_terminal_tendsto","note":"The gradient tends to the sign away from zero.","plain":"Gradient tends to sign x","proof_lines":[123,141],"relation":"proof"},{"end_line":111,"line":97,"lines":[97,101],"module":"OAI.Probability.SKValue.Equations.IntegralAbsSqrtSecond","name":"OAI.SKValue.IsDiffusion.terminal_gradient_product","note":"U X₁ = |X₁| almost surely, with no claim about X₁=0.","plain":"U X₁ = |X₁| almost surely","proof_lines":[102,111],"relation":"proof"}],"lean_source":"statement","markdown":"The drift bound $|\\gamma u|\\le\\gamma\\in L^1$ gives convergence of $X_t$\nto $X_1$, and uniform terminal convergence of $\\Phi$ to $|x|$, together\nwith convexity, gives $u(t,x_t)\\to\\operatorname{sgn}(x)$ when\n$x_t\\to x\\ne0$. Thus $U_1X_1=|X_1|$, including when $X_1=0$, so we\nnever need the probability of that event.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84179},{"anchor":"value","at":"b84495-84583","end_byte":84583,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":72,"line":57,"lines":[57,66],"module":"OAI.Probability.SKValue.Processes.StripPaths","name":"OAI.SKValue.IsDiffusion.value_expectation","note":"The identity with E u² in the integrand, before contact is used.","plain":"Mean of Φ along the diffusion","proof_lines":[67,72],"relation":"proof"}],"lean_source":"statement","markdown":"Itô's formula gives\n$$\n\\mathbb E\\Phi(t,X_t)=\\Phi(0,0)+\\frac12\\int_0^t\\gamma(s)s\\,ds.\n$$","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84495},{"anchor":"value","at":"b84584-84804","end_byte":84804,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":69,"line":42,"lines":[42,50],"module":"OAI.Probability.SKValue.Evolution.TimeWeightedIntervalIntegrable","name":"OAI.SKValue.IsDiffusion.parisi_first_value","note":"The first value formula, given the moment identity.","plain":"Parisi value from E|X₁|","proof_lines":[51,69],"relation":"proof"}],"lean_source":"statement","markdown":"The terminal limit first gives\n$\\Phi(0,0)=\\mathbb E|X_1|-\\tfrac12\\int_0^1s\\gamma(s)ds$.\nSubtracting the second half from the definition of the functional yields\n$\\mathcal P(\\gamma)=\\mathbb E|X_1|-\\int_0^1s\\gamma(s)\\,ds$.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84584},{"anchor":"value","at":"b84865-85077","end_byte":85077,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":117,"line":94,"lines":[94,102],"module":"OAI.Probability.SKValue.Evolution.L1Terminal","name":"OAI.SKValue.IsDiffusion.terminal_product_value","note":"E[U X₁] = E[U B₁] + ∫ tγ, with contact already used.","plain":"Expected product of U and X₁","proof_lines":[103,117],"relation":"proof"},{"end_line":92,"line":61,"lines":[61,70],"module":"OAI.Probability.SKValue.Evolution.L1Terminal","name":"OAI.SKValue.IsDiffusion.terminal_compact_product","note":"The same at a cutoff T, then the L1 limit.","plain":"Expected U X at time T","proof_lines":[71,92],"relation":"proof"}],"lean_source":"statement","markdown":"The diffusion equation, conditional expectation and Fubini give\n$$\n\\mathbb E[U_1X_1]=\\mathbb E[U_1B_1]\n +\\int_0^1\\gamma(s)\\mathbb E[u(s,X_s)^2]\\,ds.\n$$\nBounded $u,U_1$ and integrable $\\gamma$ justify these terms.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84865},{"anchor":"value","at":"b85250-85346","end_byte":85346,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":85,"line":73,"lines":[73,76],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.IsDiffusion.curvature_mean_integrable","note":"Integrability of the mean curvature on (0,1), which uses the unit second moment.","plain":"Mean curvature is integrable in time","proof_lines":[77,85],"relation":"proof"}],"lean_source":"statement","markdown":"The last integral is absolutely bounded by one because\n$\\mathbb E|a|\\le(\\mathbb E a^2)^{1/2}=1$.","paper":"long","role":"explanation","section":"6. Recovering the value from a martingale","start_byte":85250},{"anchor":"value","at":"b85549-85643","end_byte":85643,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":391,"line":370,"lines":[370,375],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_mean_square","note":"Lean proves only that this error tends to zero; the exact value 1−T is not stated.","plain":"L² convergence to U","proof_lines":[376,391],"relation":"part"}],"lean_source":"statement","markdown":"The prediction error and the energy tail have different sizes:\n$$\n\\|U_1-u(T,X_T)\\|_2^2=1-T,\n$$","note":"The exact value 1−T is not stated in Lean.","paper":"long","role":"explanation","section":"6. Recovering the value from a martingale","start_byte":85549},{"anchor":"scalar-value","at":"b36323-36384","end_byte":36384,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":59,"line":49,"lines":[49,52],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.scalar_value_of_contact","note":"Its only hypotheses: X solves the SDE and q(t)=t on [0,1).","plain":"Scalar value from contact","proof_lines":[53,59],"relation":"contains"}],"lean_source":"statement","markdown":"The support identities give a scalar representation of $P_*$.","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":36323},{"anchor":"scalar-value","at":"b36385-36476","end_byte":36476,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":63,"line":41,"lines":[41,48],"module":"OAI.Probability.SKValue.Equations.RealFiltration","name":"OAI.SKValue.IsDiffusion.unitMomentMartingale","note":"M packaged as a bounded martingale with E M_t²=t. The stochastic integral is not written.","plain":"The gradient is a unit-moment martingale","proof_lines":[49,63],"relation":"part"},{"end_line":31,"line":26,"lines":[26,31],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale","note":"","plain":"Martingale with second moment t","relation":"defines"},{"end_line":30,"line":27,"lines":[27,28],"module":"OAI.Probability.SKValue.Evolution.TerminalError","name":"OAI.SKValue.gradient_bound","note":"","plain":"The gradient is bounded by one","proof_lines":[29,30],"relation":"part"}],"lean_source":"statement","markdown":"Set\n$M_t=u(t,X_t)$, the bounded gradient martingale, which satisfies\n$dM_t=a(t,X_t)\\,dB_t$.","paper":"short","role":"definition","section":"5.1 A martingale formula for the ground-state value","start_byte":36385},{"anchor":"scalar-value","at":"b36477-36623","end_byte":36623,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":130,"line":71,"lines":[87,88],"module":"OAI.Probability.SKValue.Processes.MeshBrownian","name":"OAI.SKValue.diffusion_gradient_isometry","note":"Lean states the integrated moment, by Euler discretisation and not by bracket calculus.","plain":"Itô isometry for the gradient","proof_lines":[89,130],"relation":"proof"},{"end_line":337,"line":277,"lines":[293,294],"module":"OAI.Probability.SKValue.Processes.MeshBrownian","name":"OAI.SKValue.diffusion_gradient_brownian_covariance","note":"The same for the covariance with B.","plain":"Gradient covariance with Brownian motion","proof_lines":[295,337],"relation":"proof"}],"lean_source":"statement","markdown":"Its quadratic variations distinguish the two\nHessian moments:\n$$\nd\\langle M\\rangle_t=a(t,X_t)^2\\,dt,\n\\qquad d\\langle M,B\\rangle_t=a(t,X_t)\\,dt.\n$$","paper":"short","role":"explanation","section":"5.1 A martingale formula for the ground-state value","start_byte":36477},{"anchor":"scalar-value","at":"b36624-36730","end_byte":36730,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":60,"line":49,"lines":[49,51],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.integral_past_mul","note":"E M_s M_t=s, which with E M_t²=t gives the increment variance.","plain":"E[fₛ fₜ] = s","proof_lines":[52,60],"relation":"part"},{"clause":"secondMoment : ∀ t ∈ Ico (0 : ℝ) 1, ∫ ω, (f t ω)^2 ∂μ = t","end_line":31,"line":26,"lines":[31,31],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale","note":"","plain":"Martingale with second moment t","relation":"part"}],"lean_source":"statement","markdown":"Full support gives $\\mathbb E a(t,X_t)^2=1$, so each increment\n$M_t-M_s$ has unconditional variance $t-s$.","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":36624},{"anchor":"scalar-value","at":"b36778-36854","end_byte":36854,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":337,"line":277,"lines":[293,294],"module":"OAI.Probability.SKValue.Processes.MeshBrownian","name":"OAI.SKValue.diffusion_gradient_brownian_covariance","note":"E[M_T B_T] is the integral of E a up to T, for a gradient on a strip.","plain":"Gradient covariance with Brownian motion","proof_lines":[295,337],"relation":"states"},{"end_line":104,"line":87,"lines":[87,95],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.IsDiffusion.terminal_brownian_covariance","note":"The same with the terminal value U in place of M_T, for T<1.","plain":"Covariance of U with Brownian motion","proof_lines":[96,104],"relation":"part"}],"lean_source":"statement","markdown":"The first moment of $a$ gives instead the\ncovariance with the driving noise.","paper":"short","role":"explanation","section":"5.1 A martingale formula for the ground-state value","start_byte":36778},{"anchor":"scalar-value","at":"b37017-37125","end_byte":37125,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":368,"line":335,"lines":[335,340],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_square","note":"","plain":"U² = 1 almost surely","proof_lines":[341,368],"relation":"states"},{"end_line":368,"line":335,"lines":[354,358],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_square","note":"E U²=1, by dominated convergence along terminalTime.","plain":"U² = 1 almost surely","proof_lines":[341,368],"relation":"proof"}],"lean_source":"statement","markdown":"Boundedness gives $|U|\\leq1$, while $\\mathbb E M_t^2\\to1$ gives\n$\\mathbb E U^2=1$, so $U^2=1$ almost surely.","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37017},{"anchor":"scalar-value","at":"b37127-37282","end_byte":37282,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":111,"line":97,"lines":[97,101],"module":"OAI.Probability.SKValue.Equations.IntegralAbsSqrtSecond","name":"OAI.SKValue.IsDiffusion.terminal_gradient_product","note":"For U the terminal limit, assuming differentiability of Φ along the path near time one.","plain":"U X₁ = |X₁| almost surely","proof_lines":[102,111],"relation":"states"}],"lean_source":"statement","markdown":"The cancellation behind (10) runs as follows. Terminal convergence of\nthe convex value functions gives $UX_1=|X_1|$ (the equality is\nautomatic on $X_1=0$),","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37127},{"anchor":"scalar-value","at":"b37283-37367","end_byte":37367,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":69,"line":42,"lines":[42,50],"module":"OAI.Probability.SKValue.Evolution.TimeWeightedIntervalIntegrable","name":"OAI.SKValue.IsDiffusion.parisi_first_value","note":"Given strip regularity and q(t)=t: the Itô step gives ½∫γq, here ½∫tγ.","plain":"Parisi value from E|X₁|","proof_lines":[51,69],"relation":"states"},{"end_line":69,"line":42,"lines":[51,61],"module":"OAI.Probability.SKValue.Evolution.TimeWeightedIntervalIntegrable","name":"OAI.SKValue.IsDiffusion.parisi_first_value","note":"","plain":"Parisi value from E|X₁|","proof_lines":[51,69],"relation":"proof"}],"lean_source":"statement","markdown":"and Itô's formula gives\n$\\mathcal P(\\gamma)=\\mathbb E|X_1|-\\int_0^1t\\gamma(t)\\,dt$.","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37283},{"anchor":"scalar-value","at":"b37368-37513","end_byte":37513,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":117,"line":94,"lines":[94,102],"module":"OAI.Probability.SKValue.Evolution.L1Terminal","name":"OAI.SKValue.IsDiffusion.terminal_product_value","note":"E[UX₁]=E[UB₁]+∫tγ, given the martingale and continuous gradient paths.","plain":"Expected product of U and X₁","proof_lines":[103,117],"relation":"states"},{"end_line":92,"line":61,"lines":[84,92],"module":"OAI.Probability.SKValue.Evolution.L1Terminal","name":"OAI.SKValue.IsDiffusion.terminal_compact_product","note":"The identity below time one, then a limit.","plain":"Expected U X at time T","proof_lines":[71,92],"relation":"proof"},{"end_line":45,"line":27,"lines":[27,36],"module":"OAI.Probability.SKValue.Processes.WeightedTerminalExpectation","name":"OAI.SKValue.UnitMomentMartingale.weighted_terminal_expectation","note":"E[U M_t]=t, under the integral.","plain":"Weighted terminal expectation","proof_lines":[37,45],"relation":"proof"}],"lean_source":"statement","markdown":"Multiplying $X_1=B_1+\\int_0^1\\gamma(t)M_t\\,dt$ by $U$ cancels\nthat integral, because $\\mathbb E[UM_t]=t$, and the Parisi formula\nthen gives (10).","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37368},{"anchor":"scalar-value","at":"b37474-37513","end_byte":37513,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":47,"line":36,"lines":[36,47],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"The paper imports the Parisi formula. Lean proves P*=P(γ) here, from weak Guerra above and the finite Gaussian construction below.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"},{"end_line":79,"line":37,"lines":[73,79],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"hPB and hPB.trans he: the two forms of (10) at the Parisi value.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"proof"}],"lean_source":"statement","markdown":"and the Parisi formula\nthen gives (10).","paper":"short","role":"derivation","section":"5.1 A martingale formula for the ground-state value","start_byte":37474},{"anchor":"scalar-value","at":"b37867-37903","end_byte":37903,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":59,"line":49,"lines":[49,52],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.scalar_value_of_contact","note":"hmom, contact on all of [0,1), is the only analytic hypothesis.","plain":"Scalar value from contact","proof_lines":[53,59],"relation":"part"}],"lean_source":"statement","markdown":"The argument uses only full\nsupport.","paper":"short","role":"explanation","section":"5.1 A martingale formula for the ground-state value","start_byte":37867},{"anchor":"scalar-value","at":"b37514-37549","end_byte":37549,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"The same identities with the Parisi value, assuming contact, unit curvature moment and strip regularity.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"states"},{"end_line":104,"line":87,"lines":[87,95],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.IsDiffusion.terminal_brownian_covariance","note":"Scalar model: E[U B_T] equals the integral of E a up to T, the integrated form of the covariation with the noise.","plain":"Covariance of U with Brownian motion","proof_lines":[96,104],"relation":"part"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Corollary 7.1][FS]","paper":"short","role":"mention","section":"5.1 A martingale formula for the ground-state value","start_byte":37514}]},"scalar.minimizer_attainment":{"lean":[],"short":"Infimum is attained","statements":[{"anchor":"model","at":"b4934-5172","end_byte":5172,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[],"lean_source":"record","markdown":"Auffinger and Chen proved the zero-temperature Parisi formula and its\nattainment, which identify $P_*$ with $\\min_{\\gamma\\in\\mathcal U}\\mathcal P(\\gamma)$.\nThe support theorem itself takes a minimizer as given and proves a property\nof it.","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":4934},{"anchor":"main-result","at":"b6710-6833","end_byte":6833,"heading_id":"the-model-and-the-result","lean":[],"lean_source":"statement","markdown":"The existence of a minimiser is a separate input, from\n[Auffinger and Chen (2017), Theorem 1 and the proof of Lemma 3][AC].","note":"Not in the Lean: its theorems take a minimiser as the hypothesis hmin.","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6710},{"anchor":"value-and-diffusion","at":"b17494-17626","end_byte":17626,"heading_id":"allowing-a-response-to-the-noise","lean":[],"lean_source":"record","markdown":"The ground-state Parisi\nformula and the attainment of its minimum give\n\n$$\nP_*=\\min_{\\gamma\\text{ admissible}}\\mathcal P(\\gamma).\n$$","paper":"expository","role":"statement","section":"3.2 Allowing a response to the noise","start_byte":17494},{"anchor":"value-and-diffusion","at":"b17709-17841","end_byte":17841,"heading_id":"allowing-a-response-to-the-noise","lean":[],"lean_source":"record","markdown":"This equality and attainment are external inputs from\n[Auffinger–Chen (2017), Theorem 1 and the proof of Lemma 3][auffinger-chen],","paper":"expository","role":"mention","section":"3.2 Allowing a response to the noise","start_byte":17709},{"anchor":"dependencies","at":"b13502-13778","end_byte":13778,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"The ground-state formula of Auffinger and Chen supplies two facts with different uses. Their Theorem 1 identifies the ground-state limit with an infimum over $\\mathcal U$. To choose a coefficient attaining that infimum, we also need their construction in the proof of Lemma 3.","note":"Lean has no attainment: it proves the formula for a given minimiser and never constructs one.","paper":"long","role":"explanation","section":"2.3 Prior results and their roles","start_byte":13502},{"anchor":"dependencies","at":"b14231-14919","end_byte":14919,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"Here is the part of that construction which supplies the witness. Let $\\alpha_\\beta$ be a positive-temperature Parisi distribution function. The source bounds $\\beta\\alpha_\\beta(t)$ uniformly on each interval ending before one. Helly selection and a diagonal subsequence produce a nonnegative, nondecreasing right-continuous limit $\\gamma_0$. In the SK normalization, the source's weighted integral bound also controls $\\int_0^1\\beta\\alpha_\\beta(t)\\,dt$. For every nonnegative nondecreasing $h$, splitting at $1/2$ gives $\\int_0^1 h(t)\\,dt\\le4\\int_0^1t h(t)\\,dt$: the first half is bounded by $h(1/2)/2$, and on the second half $t\\ge1/2$. Fatou's lemma then gives $\\gamma_0\\in\\mathcal U$.","note":"The Helly and Fatou compactness argument has no Lean counterpart.","paper":"long","role":"derivation","section":"2.3 Prior results and their roles","start_byte":14231},{"anchor":"dependencies","at":"b14921-16184","end_byte":16184,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"Mass can concentrate at time one during this limit, so compact convergence alone does not identify the limiting control problem. Write the limiting measure of the time coefficients as\n$$\n\\nu_0(dt)=\\gamma_0(t)\\,dt+m\\delta_1(dt),\\qquad m\\geq0.\n$$\nFor an arbitrary admissible control, let $S$ denote its terminal state with coefficient $\\gamma_0$. To see why the terminal choice is admissible, let the original control be\n$\\alpha$ and put $S_s=B_s+\\int_0^s\\gamma_0(r)\\alpha_r\\,dr$.\nFor a cutoff $\\tau<1$ use $\\alpha_s$ before $\\tau$ and\n$g_k(S_s)$ afterwards, where $g_k$ is a continuous function bounded by\none and converging pointwise to the sign with $\\operatorname{sgn}(0)=1$.\nThis control uses only information available at time $s$. At fixed\n$\\tau,k$, we first take the temperature limit and then let $\\tau\\uparrow1$\nand $k\\to\\infty$. The continuous terminal evaluation has thereby\napproached $\\operatorname{sgn}(S)$ without anticipating a future state. The endpoint atom then contributes\n$$\n|S+m\\operatorname{sgn}(S)|-|S|=m\n$$\nto the reward and\n$$\n\\tfrac12m\\bigl(\\operatorname{sgn}(S)^2+1\\bigr)=m\n$$\nto the combined control and Parisi penalties. These contributions cancel. Taking the supremum over the remaining controls proves $P_*\\geq\\mathcal P(\\gamma_0)$.","note":"Imported from Auffinger and Chen; no Lean counterpart.","paper":"long","role":"derivation","section":"2.3 Prior results and their roles","start_byte":14921},{"anchor":"dependencies","at":"b16393-16724","end_byte":16724,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"record","markdown":"Thus $\\gamma_0$ attains the infimum. The cancellation is a step in the imported existence proof and adds no assumption about the Stieltjes measure $d\\gamma_0$. In particular, $\\nu_0$ here is the limit of coefficient-times-Lebesgue measures, distinct from the order parameter's Stieltjes measure, and the two must not be identified.","note":"Lean uses only the Stieltjes measure; this limit measure does not occur.","paper":"long","role":"derivation","section":"2.3 Prior results and their roles","start_byte":16393}]},"scalar.minimizer_coefficient_budget":{"lean":[],"short":"Minimiser coefficient budget","statements":[{"anchor":"coefficient-budget","at":"b87384-88459","end_byte":88459,"heading_id":"recovering-the-value-from-a-martingale","lean":[],"lean_source":"record","markdown":"**A coefficient budget from the first curvature moment.** Full support\nalso bounds the minimizing coefficient. Put $m(t)=\\mathbb E a(t,X_t)$.\nThe compact-strip identity\n$$\nda(t,X_t)=-\\gamma(t)a(t,X_t)^2\\,dt+b(t,X_t)\\,dB_t\n$$\nholds before any smoothness conclusion for $\\gamma$, and the bounded strip\nderivatives make its stochastic integral a martingale.\n[OpenAI (2026a), proof of Corollary 6.4][fs-density].\nAt zero, $X_0=0$, $a(0,0)\\ge0$ and $\\mathbb E a(0,X_0)^2=1$\ngive $a(0,0)=1$. Taking expectations and using\n$\\mathbb E a(t,X_t)^2=1$ therefore yields\n$$\nm(t)=1-\\int_0^t\\gamma(s)\\,ds,\\qquad 0\\le t<1.\n$$\nConvexity gives $m(t)\\ge0$. Letting $t\\uparrow1$ and using\nmonotonicity of $\\gamma$ on every tail gives\n$$\n\\int_0^1\\gamma(s)\\,ds\\le1,\\qquad\n0\\le\\gamma(t)\\le\\frac{m(t)}{1-t}\\le\\frac1{1-t}.\n$$\nIndeed, $(s-t)\\gamma(t)\\le\\int_t^s\\gamma\\le m(t)$ for\n$t<s<1$. Integrating the first-moment identity also gives\n$$\nI(T)=T-\\int_0^T(T-s)\\gamma(s)\\,ds,\\qquad\nP_*=1-\\int_0^1(1-s)\\gamma(s)\\,ds.\n$$\nThe second equality uses (6.1) and so depends on its ground-state\nidentification.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":87384}]},"scalar.minimizer_total_coefficient_mass":{"lean":[],"short":"Coefficient total mass one","statements":[{"anchor":"coefficient-budget","at":"b88615-89067","end_byte":89067,"heading_id":"recovering-the-value-from-a-martingale","lean":[],"lean_source":"record","markdown":"**Proposition 6.2 (terminal normalization in zero field).** For the\nminimizer and diffusion of Theorem 5.1,\n$$\n\\lim_{t\\uparrow1}m(t)=0,\\qquad\n\\int_0^1\\gamma(s)\\,ds=1,\\qquad\nm(t)=\\int_t^1\\gamma(s)\\,ds,\n$$\nand consequently\n$$\nP_*=\\int_0^1s\\gamma(s)\\,ds.\n$$\nThus $\\gamma(s)\\,ds$ has total mass one. This says nothing about the\ntotal mass of the Stieltjes measure $d\\gamma$. The proposition is an\nanalytic consequence outside the selected formal endpoints.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":88615},{"anchor":"coefficient-budget","at":"b91269-91565","end_byte":91565,"heading_id":"recovering-the-value-from-a-martingale","lean":[],"lean_source":"record","markdown":"First let $t\\uparrow1$ and then $\\delta\\downarrow0$. This proves\n$m(1-)=0$. The compact-strip identity for $m$ now gives total mass one\nand the tail formula. Tonelli and (6.1) give the stated expression for\n$P_*$. In particular,\n$$\n0\\le P_*-I(T)=\\int_T^1(s-T)\\gamma(s)\\,ds\n\\le(1-T)m(T)=o(1-T).\n$$","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":91269},{"anchor":"coefficient-budget","at":"b89069-89195","end_byte":89195,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":130,"line":103,"lines":[103,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"The sign for each finite-step problem.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"}],"lean_source":"statement","markdown":"**Proof.** The backward comparison in §4.3 gives\n$\\Phi_{xxx}(t,x)\\le0$ for $x>0$ in every positive finite-step\napproximation.","note":"Lean has the finite-step sign but no terminal normalisation: Proposition 6.2 is not formalised.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":89069}]},"scalar.paper_admissible":{"lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: the admissible coefficient on the interval [0,1).","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":60,"line":59,"lines":[59,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model: the Parisi functional, built on the control value `value`.","plain":"The Parisi functional","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Scalar model: the same conditions, on a coefficient defined on all reals.","plain":"Admissible order parameter on ℝ","relation":"defines"},{"end_line":78,"line":77,"lines":[77,78],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.parisi","note":"Scalar model: the same formula, with this model's own value function phi.","plain":"The Parisi functional","relation":"defines"}],"short":"Admissible coefficient and normalisation","statements":[{"anchor":"model","at":"b3841-4548","end_byte":4548,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: the admissible coefficient on the interval [0,1).","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":60,"line":59,"lines":[59,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model: the Parisi functional, built on the control value `value`.","plain":"The Parisi functional","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Scalar model: the same conditions, on a coefficient defined on all reals.","plain":"Admissible order parameter on ℝ","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\mathcal U$ consist of nonnegative, nondecreasing, right-continuous\nfunctions $\\gamma:[0,1)\\to[0,\\infty)$ with\n$\\int_0^1\\gamma(t)\\,dt<\\infty$. Their Stieltjes measures satisfy\n$\\mu_\\gamma([0,t])=\\gamma(t)$, including\n$\\mu_\\gamma(\\{0\\})=\\gamma(0)$. They are locally finite. They need not be\nprobability measures, and their total mass may be infinite. Support is taken\nrelative to [0,1).\n\nLet $B$ be standard Brownian motion. The control value and Parisi\nfunctional are\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_{|\\alpha|\\le1}\\mathbb E\\left[\n \\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n -\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\\right],\n$$\n$$\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)-\\frac12\\int_0^1t\\gamma(t)\\,dt.\n$$","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":3841},{"anchor":"model","at":"b3348-3505","end_byte":3505,"heading_id":"the-model-and-the-result","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"A function on Time; integrability is stated for its extension by zero.","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"The scalar model uses a function on ℝ with the same conditions on [0,1). The two models are not transported.","plain":"Admissible order parameter on ℝ","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$,","paper":"short","role":"definition","section":"1. The model and the result","start_byte":3348},{"anchor":"model","at":"b4044-4544","end_byte":4544,"heading_id":"the-model-and-the-result","lean":[{"end_line":60,"line":59,"lines":[59,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model: Φ(0,0) minus half the integral, with Φ the control value. Lean proves the equation afterwards.","plain":"The Parisi functional","relation":"defines"},{"end_line":51,"line":49,"lines":[49,51],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"Support model: Φ is a control supremum. The equation and terminal condition are theorems about it.","plain":"The value function Φ","relation":"defines"},{"end_line":78,"line":77,"lines":[77,78],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.parisi","note":"Scalar model: the same formula on its own phi. No Lean theorem connects the two models.","plain":"The Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The Parisi functional $\\mathcal P(\\gamma)$ is defined through an\nauxiliary value function $\\Phi_\\gamma(t,x)$, in which $x\\in\\mathbb R$\nis a spatial variable and the overlap parameter $t$ plays the role of\ntime. Write $\\Phi=\\Phi_\\gamma$ and use subscripts for partial\nderivatives. The backward equation, terminal condition and functional are\n$$\n\\Phi_t=-\\tfrac12\\bigl(\\Phi_{xx}+\\gamma(t)\\Phi_x^2\\bigr),\n\\qquad \\Phi(1,x)=|x|,\n\\qquad\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)-\\tfrac12\\int_0^1t\\gamma(t)\\,dt.\n$$","paper":"short","role":"definition","section":"1. The model and the result","start_byte":4044},{"anchor":"order-parameter-example","at":"b7752-7904","end_byte":7904,"heading_id":"what-a-gap-means","lean":[{"end_line":22,"line":17,"lines":[17,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model: the admissible coefficient on the interval [0,1).","plain":"Admissible order parameter γ","relation":"defines"},{"end_line":38,"line":33,"lines":[33,38],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.OrderParameter","note":"Scalar model: the same conditions, on a coefficient defined on all reals.","plain":"Admissible order parameter on ℝ","relation":"defines"}],"lean_source":"statement","markdown":"In general, an admissible **order parameter** is a function\n$\\gamma:[0,1)\\to[0,\\infty)$ that is nondecreasing, right-continuous\nand has finite integral.","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":7752},{"anchor":"value-and-diffusion","at":"b17214-17374","end_byte":17374,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":60,"line":59,"lines":[59,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model: the Parisi functional, built on the control value `value`.","plain":"The Parisi functional","relation":"defines"},{"end_line":78,"line":77,"lines":[77,78],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.parisi","note":"Scalar model: the same formula, built on its own value function phi.","plain":"The Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The **Parisi functional** subtracts an explicit correction from this\noptimised value:\n\n$$\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)\n-\\frac12\\int_0^1t\\gamma(t)\\,dt.\n$$","paper":"expository","role":"definition","section":"3.2 Allowing a response to the noise","start_byte":17214},{"anchor":"spins-and-overlap","at":"b5646-5742","end_byte":5742,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":40,"line":11,"lines":[11,13],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.hamCoeff_distance","note":"Scalar model: E(H(σ)−H(τ))² as a coefficient distance, n(1−R²), which is ξ(t)=t²/2. Lean never names ξ.","proof_lines":[14,40],"relation":"part"},{"clause":"(1 / 2 : ℝ) * ∫ t in (0 : ℝ)..1, t * extend γ.val t","end_line":60,"line":59,"lines":[60,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model. Unit noise and this correction term are written directly, which is the same choice. Lean never names ξ.","plain":"The Parisi functional","relation":"part"}],"lean_source":"statement","markdown":"Our normalisation gives the covariance function $\\xi(t)=t^2/2$ in the\nvariational problem below.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":5646},{"anchor":"order-parameter-example","at":"b8524-8650","end_byte":8650,"heading_id":"what-a-gap-means","lean":[{"clause":"integrable : Integrable (extend val)","end_line":22,"line":17,"lines":[22,22],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model. Lean requires the first integral finite. The equality with the weighted mass integral is not stated.","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"Admissibility\nrequires instead the weighted finiteness\n\n$$\n\\int_0^1\\gamma(t)\\,dt\n=\\int_{[0,1)}(1-s)\\,\\mu_\\gamma(ds)<\\infty.\n$$","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":8524},{"anchor":"order-parameter-example","at":"b7905-8003","end_byte":8003,"heading_id":"what-a-gap-means","lean":[{"clause":"right_continuous : ∀ t, ContinuousWithinAt val (Set.Ici t) t","end_line":22,"line":17,"lines":[21,21],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.OrderParameter","note":"Support model. The field that fixes the value at a jump.","plain":"Admissible order parameter γ","relation":"defines"}],"lean_source":"statement","markdown":"Right-continuity means that at a jump the\nfunction already takes its new value, as in the example.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":7905},{"anchor":"value-and-diffusion","at":"b14328-14525","end_byte":14525,"heading_id":"the-zero-coefficient","lean":[{"end_line":51,"line":49,"lines":[49,51],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"Support model. A supremum over controls of an expected payoff.","plain":"The value function Φ","relation":"defines"},{"clause":"|x + W.B 1 ω - W.B (Real.toNNReal t) ω +","end_line":47,"line":43,"lines":[45,45],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. The final reward is the absolute value of the final state. The one-spin reading is the paper's.","plain":"Control payoff, sample by sample","relation":"defines"}],"lean_source":"statement","markdown":"The variational problem assigns each $\\gamma$ a value by a one-dimensional\nauxiliary optimisation. Its final reward at state $x$ is $|x|$, the\nbest one-spin score $\\max_{\\sigma\\in\\{-1,1\\}}x\\sigma$.","paper":"expository","role":"definition","section":"3.1 The zero coefficient","start_byte":14328},{"anchor":"value-and-diffusion","at":"b15773-15897","end_byte":15897,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"progressive : IsProgressive (incrementFiltration W t) val","end_line":41,"line":38,"lines":[40,40],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Support model. Adapted to the Brownian increments after the start, as in incrementFiltration.","plain":"Admissible control bounded by one","relation":"defines"},{"clause":"bound : ∀ s ω, |val s ω| ≤ 1","end_line":41,"line":38,"lines":[41,41],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Support model. The bound by one.","plain":"Admissible control bounded by one","relation":"defines"}],"lean_source":"statement","markdown":"A **control** $\\alpha_s\\in[-1,1]$ chooses a response at each time\nusing only noise already observed since the starting time.","paper":"expository","role":"definition","section":"3.2 Allowing a response to the noise","start_byte":15773},{"anchor":"value-and-diffusion","at":"b15898-15992","end_byte":15992,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"progressive : IsProgressive (incrementFiltration W t) val","end_line":41,"line":38,"lines":[40,40],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Support model. The filtration restricts noise only. γ enters the payoff as a fixed deterministic function.","plain":"Admissible control bounded by one","relation":"defines"}],"lean_source":"statement","markdown":"The whole\ndeterministic schedule $\\gamma$ is known in advance and the future noise\nis unknown,","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":15898},{"anchor":"value-and-diffusion","at":"b16397-16519","end_byte":16519,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"progressive : IsProgressive (incrementFiltration W t) val","end_line":41,"line":38,"lines":[40,40],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Support model. The restriction itself. The sign example above is not in Lean.","plain":"Admissible control bounded by one","relation":"defines"}],"lean_source":"statement","markdown":"In the continuous control problem,\ndecisions repeat under the same restriction, which the finite example\nonly illustrates.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":16397},{"anchor":"value-and-diffusion","at":"b16549-16743","end_byte":16743,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":51,"line":49,"lines":[49,51],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"Support model. The supremum of expected payoffs.","plain":"The value function Φ","relation":"defines"},{"end_line":47,"line":43,"lines":[43,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. The same payoff. Lean indexes a control by time elapsed since the start, s minus t.","plain":"Control payoff, sample by sample","relation":"defines"}],"lean_source":"statement","markdown":"For a starting position $x$, the value is\n\n$$\n\\Phi_\\gamma(t,x)=\\sup_\\alpha\\mathbb E\\left[\n\\left|x+B_1-B_t+\\int_t^1\\gamma(s)\\alpha_s\\,ds\\right|\n-\\frac12\\int_t^1\\gamma(s)\\alpha_s^2\\,ds\n\\right].\n$$","paper":"expository","role":"definition","section":"3.2 Allowing a response to the noise","start_byte":16549},{"anchor":"value-and-diffusion","at":"b16745-16901","end_byte":16901,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":47,"line":43,"lines":[45,46],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. Noise and accumulated bias, inside the absolute value.","plain":"Control payoff, sample by sample","relation":"defines"},{"end_line":47,"line":43,"lines":[47,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. The running cost.","plain":"Control payoff, sample by sample","relation":"defines"}],"lean_source":"statement","markdown":"Inside the absolute value, noise and the accumulated bias change the\nstate before its final reward is paid. The second integral is the\nrunning control cost.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":16745},{"anchor":"value-and-diffusion","at":"b16902-17029","end_byte":17029,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"sSup (Set.range (fun α : Control W (Real.toNNReal t) =>","end_line":51,"line":49,"lines":[50,50],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"Support model. A real supremum over the Control structure, so every control in the index set is progressively measurable.","plain":"The value function Φ","relation":"defines"},{"end_line":37,"line":30,"lines":[30,33],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.control_values_bddAbove","note":"This bound makes the real supremum a true one.","plain":"Expected payoffs are bounded above","proof_lines":[34,37],"relation":"part"}],"lean_source":"statement","markdown":"The supremum chooses the best permitted rule, and the formal version of\nthe information condition is progressive measurability.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":16902},{"anchor":"value-and-diffusion","at":"b17145-17212","end_byte":17212,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":47,"line":43,"lines":[46,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. Both integrals contain γ, so they vanish at zero. value_on_terminal_plateau with c=0 gives the resulting formula.","plain":"Control payoff, sample by sample","relation":"defines"}],"lean_source":"statement","markdown":"At $\\gamma=0$ both control terms vanish\nand we recover Section 3.1.","paper":"expository","role":"example","section":"3.2 Allowing a response to the noise","start_byte":17145},{"anchor":"value-and-diffusion","at":"b15522-15603","end_byte":15603,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":47,"line":43,"lines":[46,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"Support model. The bias and its quadratic cost.","plain":"Control payoff, sample by sample","relation":"defines"}],"lean_source":"statement","markdown":"For a general $\\gamma$, a response can bias the future state at a\nquadratic cost.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":15522},{"anchor":"value-and-diffusion","at":"b15898-16079","end_byte":16079,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"val : ℝ≥0 → Ω → ℝ","end_line":41,"line":38,"lines":[39,39],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Support model. A control is a function of time and of the path.","plain":"Admissible control bounded by one","relation":"defines"}],"lean_source":"statement","markdown":"The whole\ndeterministic schedule $\\gamma$ is known in advance and the future noise\nis unknown, yet a rule fixed in advance can still respond differently on\ndifferent observed paths.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":15898},{"anchor":"value-and-diffusion","at":"b17030-17144","end_byte":17144,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"∫ ω, controlPayoff W γ t x α ω ∂W.law","end_line":51,"line":49,"lines":[51,51],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"Support model. An integral over the Brownian system's law, with no spin variables.","plain":"The value function Φ","relation":"defines"}],"lean_source":"statement","markdown":"Here the expectation\naverages over Brownian paths for a fixed $\\gamma$, and the spin\ncouplings play no part in it.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":17030},{"anchor":"model","at":"b4550-4622","end_byte":4622,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":41,"line":38,"lines":[38,41],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Control","note":"Progressive for the increments after t, and bounded by one.","plain":"Admissible control bounded by one","relation":"defines"},{"end_line":36,"line":33,"lines":[33,36],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.incrementFiltration","note":"The filtration of Brownian increments after t.","plain":"Filtration of increments after t","relation":"defines"}],"lean_source":"statement","markdown":"Controls are progressively measurable for Brownian increments after $t$.","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":4550},{"anchor":"dependencies","at":"b12807-12865","end_byte":12865,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[],"lean_source":"statement","markdown":"The convention\n$\\xi''=1$ fixes the curvature coefficients.","note":"Lean has no parameter here: the factor one is written into the model.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12807},{"anchor":"model","at":"b4324-4544","end_byte":4544,"heading_id":"the-model-and-the-result","lean":[{"end_line":60,"line":59,"lines":[59,60],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.parisi","note":"Support model: the Parisi functional, built on the control value `value`.","plain":"The Parisi functional","relation":"defines"},{"end_line":55,"line":52,"lines":[52,52],"module":"OAI.Probability.SKValue.Evolution.TerminalError","name":"OAI.SKValue.phi_terminal","note":"Scalar model. The terminal condition.","plain":"Terminal value Φ(1,x) = |x|","proof_lines":[53,55],"relation":"part"},{"end_line":17,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Moments.SquareEquation","name":"OAI.ZeroTemperatureSK.compactValueRate_eq","note":"The right side of the PDE, on a strip.","plain":"Formula for the value's time derivative","relation":"part"},{"end_line":137,"line":121,"lines":[121,125],"module":"OAI.Probability.SKSupport.Regularity.RightRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_hasDerivWithinAt_right","note":"Lean defines Φ by a control supremum and proves this equation as a right time derivative.","plain":"Right time derivative of x-derivatives","proof_lines":[126,137],"relation":"part"}],"lean_source":"statement","markdown":"The backward equation, terminal condition and functional are\n$$\n\\Phi_t=-\\tfrac12\\bigl(\\Phi_{xx}+\\gamma(t)\\Phi_x^2\\bigr),\n\\qquad \\Phi(1,x)=|x|,\n\\qquad\n\\mathcal P(\\gamma)=\\Phi_\\gamma(0,0)-\\tfrac12\\int_0^1t\\gamma(t)\\,dt.\n$$","paper":"short","role":"definition","section":"1. The model and the result","start_byte":4324},{"anchor":"model","at":"b4545-4627","end_byte":4627,"heading_id":"the-model-and-the-result","lean":[{"end_line":55,"line":52,"lines":[52,52],"module":"OAI.Probability.SKValue.Evolution.TerminalError","name":"OAI.SKValue.phi_terminal","note":"Scalar model. Gives Φ(1,x)=|x| for the control value. The one-spin reading is outside the Lean.","plain":"Terminal value Φ(1,x) = |x|","proof_lines":[53,55],"relation":"part"}],"lean_source":"statement","markdown":"The terminal value $|x|$ is the zero-temperature value of one spin\nin a field $x$.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":4545}]},"scalar.paper_minimizer":{"lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: global minimality of the Parisi functional, taken as a hypothesis.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Scalar model: the same inequality over that model's order parameters.","plain":"γ minimises the Parisi functional","relation":"defines"}],"short":"Minimiser hypothesis","statements":[{"anchor":"model","at":"b5090-5172","end_byte":5172,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model: the minimiser is the hypothesis hmin, and no Lean theorem supplies one.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: global minimality of the Parisi functional, taken as a hypothesis.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The support theorem itself takes a minimizer as given and proves a property\nof it.","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":5090},{"anchor":"main-result","at":"b6367-6420","end_byte":6420,"heading_id":"the-model-and-the-result","lean":[{"clause":"(hmin : IsMinimizer W γ)","end_line":59,"line":35,"lines":[36,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"The theorem's hypothesis. The next tabs define minimality in each model.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: global minimality of the Parisi functional, taken as a hypothesis.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Scalar model: the same inequality over that model's order parameters.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"If $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6367},{"anchor":"full-support-statement","at":"b25334-25452","end_byte":25452,"heading_id":"the-full-support-theorem","lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: global minimality of the Parisi functional, taken as a hypothesis.","plain":"γ minimises the Parisi functional","relation":"defines"},{"end_line":81,"line":80,"lines":[80,81],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.IsMinimizer","note":"Scalar model: the same inequality over that model's order parameters.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"In the pure, zero-field model with $\\xi(t)=t^2/2$, let $\\gamma$ be an\nadmissible integrable minimiser of $\\mathcal P$.","paper":"expository","role":"statement","section":"4. The full-support theorem","start_byte":25334},{"anchor":"full-support-statement","at":"b25667-25781","end_byte":25781,"heading_id":"the-full-support-theorem","lean":[{"clause":"(hmin : IsMinimizer W γ)","end_line":59,"line":35,"lines":[36,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Minimality is a hypothesis of the theorem.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model: global minimality of the Parisi functional, taken as a hypothesis.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The hypothesis is minimality, since an arbitrary admissible candidate,\nsuch as the one in Figure 1, can have gaps.","paper":"expository","role":"explanation","section":"4. The full-support theorem","start_byte":25667},{"anchor":"value-and-diffusion","at":"b17376-17493","end_byte":17493,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":63,"line":62,"lines":[62,63],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsMinimizer","note":"Support model. The outer problem. Lean states minimality against every order parameter.","plain":"γ minimises the Parisi functional","relation":"defines"}],"lean_source":"statement","markdown":"The inner supremum chooses a response for fixed $\\gamma$, and the outer\nproblem minimises $\\mathcal P$ over $\\gamma$.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":17376},{"anchor":"value-and-diffusion","at":"b17910-17977","end_byte":17977,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"(hmin : IsMinimizer W γ)","end_line":59,"line":35,"lines":[36,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Minimality is a hypothesis. Lean never shows that a minimiser exists.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"}],"lean_source":"statement","markdown":"The no-gap theorem assumes a minimiser and proves a property of it,","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":17910}]},"scalar.parisi_formula":{"lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Equality at a given minimiser, proved in the scalar model. Taking the infimum and attainment stay outside it.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"},{"clause":"groundStateValue≤parisi W γ","end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Upper bound only, for every order parameter, proved in the scalar model. Equality needs a minimiser, and attainment is not proved.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"short":"Zero-temperature Parisi formula","statements":[{"anchor":"model","at":"b4934-5089","end_byte":5089,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Equality at a given minimiser, proved in the scalar model. Taking the infimum and attainment stay outside it.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"},{"clause":"groundStateValue≤parisi W γ","end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"The upper bound for every order parameter, proved in the scalar model.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"Auffinger and Chen proved the zero-temperature Parisi formula and its\nattainment, which identify $P_*$ with $\\min_{\\gamma\\in\\mathcal U}\\mathcal P(\\gamma)$.","paper":"long","role":"statement","section":"1. The SK model and the Parisi variational problem","start_byte":4934},{"anchor":"model","at":"b3506-3791","end_byte":3791,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":40,"line":11,"lines":[11,13],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.hamCoeff_distance","note":"Scalar proof tree: squared distance of H_n coefficients is n(1-R₁₂²), the convention in increment form. No ξ and no extra Gaussian appear.","proof_lines":[14,40],"relation":"part"}],"lean_source":"statement","markdown":"For overlap $R_{12}=n^{-1}\\sum_i\\sigma_i^1\\sigma_i^2$, the covariance is\n$nR_{12}^2/2-1/2$. Thus the convention is $\\xi(t)=t^2/2$, with\n$\\xi''=1$. A centered Gaussian of variance $1/2$, common to all\nconfigurations, removes the final constant and leaves the expected maximum\nunchanged.","paper":"long","role":"explanation","section":"1. The SK model and the Parisi variational problem","start_byte":3506},{"anchor":"gaussian-expectation-limit","at":"b193363-193903","end_byte":193903,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Equality at a given minimiser, scalar model, constructing its own diffusion. Taking the infimum and attainment stay outside it.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"end_line":47,"line":36,"lines":[36,39],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Same equality with gradient contact as premise; this proof combines the two inequalities.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"},{"clause":"groundStateValue≤parisi W γ","end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Upper bound only, for every order parameter, proved in the scalar model. Equality needs a minimiser, and attainment is not proved.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"At a minimizer, the separate contact and diffusion arguments establish\nthe fixed-strip inputs and eventual positivity. First take the mesh\nlimit to obtain $I(T)\\le P_*$, then let $T\\uparrow1$. The scalar\nterminal assembly identifies that last limit with $\\mathcal P(\\gamma)$.\nTogether with $P_*\\le\\mathcal P(\\gamma)$ this proves equality.\nThe third scalar endpoint constructs its own diffusion before making\nthis application. These sources therefore supply an independent route\nto the equality, alongside the imported Parisi formula in §6.","paper":"long","role":"explanation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193363},{"anchor":"model","at":"b2331-2408","end_byte":2408,"heading_id":"the-model-and-the-result","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Scalar model. P* equals the functional at a minimiser. Lean takes the minimiser as given.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"},{"end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Scalar model. The upper bound, for every coefficient.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"The Parisi formula expresses $P_*$ as a minimum over nondecreasing\nfunctions.","paper":"short","role":"statement","section":"1. The model and the result","start_byte":2331},{"anchor":"model","at":"b2949-3280","end_byte":3280,"heading_id":"the-model-and-the-result","lean":[{"end_line":72,"line":59,"lines":[59,60],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.quadraticValue_spin","note":"Scalar model. Up to the factor √n, the process with covariance nξ(R) has H_n's expected maximum, by equal increments. Convention and citation are outside it.","plain":"Quadratic value of the SK spins","proof_lines":[61,72],"relation":"part"},{"end_line":40,"line":11,"lines":[11,13],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.hamCoeff_distance","note":"Scalar model. The energy increment variance n−nR², which is the stated covariance in distance form.","proof_lines":[14,40],"relation":"part"}],"lean_source":"statement","markdown":"We use the covariance convention $\\xi(t)=t^2/2$, so $\\xi''=1$.\nAdding one independent centred Gaussian of variance $1/2$ to every\nconfiguration's energy gives covariance $n\\xi(R)$ and leaves the\nexpected maximum unchanged, which is the normalisation of the\nground-state Parisi formula of [Auffinger and Chen (2017), Theorem 1][AC].","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":2949},{"anchor":"value-and-diffusion","at":"b17494-17626","end_byte":17626,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Equality at a given minimiser, proved in the scalar model. Taking the infimum and attainment stay outside it.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"},{"clause":"groundStateValue≤parisi W γ","end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Upper bound only, for every order parameter, proved in the scalar model. Equality needs a minimiser, and attainment is not proved.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"The ground-state Parisi\nformula and the attainment of its minimum give\n\n$$\nP_*=\\min_{\\gamma\\text{ admissible}}\\mathcal P(\\gamma).\n$$","paper":"expository","role":"statement","section":"3.2 Allowing a response to the noise","start_byte":17494},{"anchor":"value-and-diffusion","at":"b17628-17909","end_byte":17909,"heading_id":"allowing-a-response-to-the-noise","lean":[{"clause":"groundStateValue≤parisi W γ","end_line":141,"line":129,"lines":[130,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Upper bound only, for every order parameter, proved in the scalar model. Equality needs a minimiser, and attainment is not proved.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"},{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Equality at a given minimiser, proved in the scalar model. Taking the infimum and attainment stay outside it.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"}],"lean_source":"statement","markdown":"Each candidate supplies an upper bound, and minimisation finds the\nsharpest one. This equality and attainment are external inputs from\n[Auffinger–Chen (2017), Theorem 1 and the proof of Lemma 3][auffinger-chen],\nand Gaussian concentration gives the expected-value form used here.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":17628},{"anchor":"spins-and-overlap","at":"b6620-6815","end_byte":6815,"heading_id":"the-finite-optimisation-problem","lean":[{"clause":"groundStateValue=parisi W γ","end_line":27,"line":25,"lines":[26,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Scalar model: equality at any minimiser, proved in Lean though the paper cites it. Existence of a minimiser stays a hypothesis.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"part"},{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Scalar model: the upper bound for every order parameter.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"The Parisi theorem of Section 3 identifies their\nminimum with the limiting spin optimum. That identification is an\nexternal theorem and involves no change of variables in the finite\nmaximisation.","paper":"expository","role":"explanation","section":"1. The finite optimisation problem","start_byte":6620},{"anchor":"model","at":"b1969-2183","end_byte":2183,"heading_id":"the-model-and-the-result","lean":[{"end_line":95,"line":92,"lines":[92,95],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.hamiltonian","note":"Scalar model. Spins are Bool read through spin. Couplings are indexed by all ordered pairs; only i<j enter.","plain":"SK Hamiltonian on n spins","relation":"defines"},{"end_line":93,"line":91,"lines":[91,93],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.spinHamiltonian","note":"The same Hamiltonian, as written in the support model.","plain":"SK Hamiltonian on n spins","relation":"defines"}],"lean_source":"statement","markdown":"For independent standard Gaussian couplings $J_{ij}$, the pure,\nzero-field [Sherrington–Kirkpatrick (SK) Hamiltonian][SK] is\n$$\nH_n(\\sigma)=n^{-1/2}\\sum_{i<j}J_{ij}\\sigma_i\\sigma_j,\n\\qquad \\sigma\\in\\{-1,1\\}^n.\n$$","paper":"short","role":"definition","section":"1. The model and the result","start_byte":1969},{"anchor":"model","at":"b2184-2330","end_byte":2330,"heading_id":"the-model-and-the-result","lean":[{"end_line":100,"line":100,"lines":[100,100],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateValue","note":"Scalar model. Written as limUnder, so the definition alone asserts no limit.","plain":"The SK ground state value","relation":"defines"},{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateSequence","note":"Scalar model. The expected maximum per spin, before the limit.","plain":"Expected maximum energy per spin","relation":"defines"},{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"Scalar model. This declaration proves that the limit exists.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"part"}],"lean_source":"statement","markdown":"With the maximisation convention, its limiting ground-state energy per\nspin is\n$$\nP_*:=\\lim_{n\\to\\infty}n^{-1}\\mathbb E\\max_\\sigma H_n(\\sigma).\n$$","paper":"short","role":"definition","section":"1. The model and the result","start_byte":2184},{"anchor":"model","at":"b3281-3346","end_byte":3346,"heading_id":"the-model-and-the-result","lean":[{"end_line":95,"line":92,"lines":[92,95],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.hamiltonian","note":"Scalar model. The sum runs over pairs only, with no linear term.","plain":"SK Hamiltonian on n spins","relation":"states"},{"clause":"phi W γ 0 0","end_line":78,"line":77,"lines":[78,78],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.parisi","note":"Scalar model. Zero field is the evaluation of the value at x=0.","plain":"The Parisi functional","relation":"part"}],"lean_source":"statement","markdown":"The model has no higher-order interactions and no external field.","paper":"short","role":"mention","section":"1. The model and the result","start_byte":3281}]},"scalar.separate_classical_algorithm":{"lean":[],"short":"Separate classical algorithm","statements":[{"anchor":"algorithm","at":"b176247-177059","end_byte":177059,"heading_id":"a-classical-optimization-algorithm","lean":[],"lean_source":"record","markdown":"The scalar Gaussian construction is not itself an algorithm on the matrix of SK couplings. We obtain a finite-disorder algorithm from the theorem of El Alaoui, Montanari and Sellke. In their cited preprint, Assumption 2 asks for a strictly increasing minimizer of the zero-temperature Parisi functional on $\\mathcal U$. The support theorem gives this property, and the imported attainment argument supplies the minimizer.\n\nTheir Corollary 2.2 then returns, for each fixed $\\varepsilon>0$, a randomized spin configuration within $\\varepsilon n$ of the finite-instance maximum with probability tending to one. Its operation count is the one in the displayed Theorem 3 on p. 7, which the preprint's running references call “Theorem 3.2.” [El Alaoui, Montanari and Sellke, arXiv:2001.00904v1, pp. 7–8][eams20].","paper":"long","role":"statement","section":"9.4 A classical optimization algorithm","start_byte":176247},{"anchor":"algorithm","at":"b177852-178527","end_byte":178527,"heading_id":"a-classical-optimization-algorithm","lean":[],"lean_source":"record","markdown":"The imported operation count is\n$C(\\varepsilon)(\\chi+n)+n\\chi_1$, where $\\chi$ is the cost of evaluating the full gradient and $\\chi_1$ is the cost of one coordinate of the distinct-index gradient. Here $\\nabla\\widehat H_n(x)=Ax$ costs $O(n^2)$, and one coordinate of $\\nabla H_n(x)$ costs $O(n)$. The resulting count is $C(\\varepsilon)n^2$, with $C(\\varepsilon)$ independent of $n$.\n\nReal sums and multiplications have unit cost in the cited model, and the scalar functions and normalizers prescribed for the chosen accuracy are part of that construction. This statement supplies neither a bound for their numerical preprocessing nor a finite-precision bit-complexity bound.","paper":"long","role":"derivation","section":"9.4 A classical optimization algorithm","start_byte":177852},{"anchor":"scalar-value","at":"b38608-39252","end_byte":39252,"heading_id":"the-classical-algorithm","lean":[],"lean_source":"record","markdown":"Full support and existence give an attained, strictly increasing\nminimiser in the monotone zero-temperature class, which is Assumption 2\nof [El Alaoui, Montanari and Sellke, preprint v1][EAMS]. Their\nCorollary 2.2 then gives, for every fixed $\\varepsilon>0$, a randomised\nalgorithm that returns a configuration within $\\varepsilon n$ of the\nfinite SK maximum with probability tending to one over couplings and\nalgorithmic randomness. It costs $C(\\varepsilon)n^2$ real-arithmetic\noperations when the prescribed scalar functions and normalisers are\navailable. This count includes no separate bound for numerical\npreprocessing or finite precision.","paper":"short","role":"statement","section":"5.2 The classical algorithm","start_byte":38608},{"anchor":"algorithm","at":"b179651-179778","end_byte":179778,"heading_id":"a-classical-optimization-algorithm","lean":[],"lean_source":"record","markdown":"It supplies no efficient construction of QAOA angles, no quantum depth bound and no clause of either selected formal interface.","paper":"long","role":"mention","section":"9.4 A classical optimization algorithm","start_byte":179651}]},"scalar.shifted_gaussian_limit":{"lean":[{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second clause is the mesh limit. The third clause is the endpoint limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The fixed-T mesh limit with strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"short":"Shifted Gaussian coefficient limit","statements":[{"anchor":"gaussian","at":"b92520-92887","end_byte":92887,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model. All three clauses: positive normalisers, the mesh limit at fixed T, then the limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The fixed-T mesh limit with strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"**Proposition 7.1 (finite Gaussian approximation).** For a fixed minimizer\nand its diffusion, the construction below has positive active normalizers\nfor all sufficiently fine meshes at every fixed $0<T<1$. Its adjacent\ncorrelation sum converges to $\\int_0^T\\mathbb E a(t,X_t)\\,dt$, and this\nintegral tends to $P_*$ as $T\\uparrow1$. The limits are taken in that order.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":92520},{"anchor":"gaussian","at":"b97343-97487","end_byte":97487,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":136,"line":47,"lines":[47,59],"module":"OAI.Probability.SKValue.Processes.MeshMeanBound","name":"OAI.SKValue.finiteGaussian_from_strip_and_moments","note":"The limit, given the second and first moment inputs.","plain":"Finite-Gaussian limit from strip and moments","proof_lines":[60,136],"relation":"proof"},{"end_line":49,"line":29,"lines":[39,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The same for the diffusion, on one strip.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"}],"lean_source":"statement","markdown":"Hence\n$$\n\\sum_{j=0}^NA_jB_{j+1}\n =\\Delta\\sum_{j=1}^{N-1}\\frac{\\mathbb E a(t_j,Y_j)}{c_j}+o(1)\n \\longrightarrow\\int_0^T\\mathbb E a(t,X_t)\\,dt.\n$$","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":97343},{"anchor":"gaussian","at":"b98195-98305","end_byte":98305,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: its third clause is this limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"Only afterward let $T\\uparrow1$, obtaining $P_*$.\nThe mesh count $N$ is unrelated to the spin-system size $n$.","paper":"long","role":"mention","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":98195},{"anchor":"qaoa-boundary","at":"b43303-43590","end_byte":43590,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"A j B (j+1) with j from 0; A 0=0. Lean j is paper j, and coordinate j is paper Z_{j+1}.","plain":"Adjacent pairing of A and B","relation":"defines"},{"end_line":145,"line":141,"lines":[141,145],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientA","note":"","plain":"Correlation A with normalised increments","relation":"defines"},{"end_line":149,"line":147,"lines":[147,149],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.coefficientB","note":"","plain":"Correlation B with Gaussian coordinates","relation":"defines"},{"end_line":161,"line":155,"lines":[158,159],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The limit.","plain":"The finite-Gaussian conclusion","relation":"part"}],"lean_source":"statement","markdown":"The circuit construction\nneeds one finite quantity, the adjacent pairing\n$$\n\\begin{aligned}\nS_{T,N}&:=\\sum_{j=1}^N\n \\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}],\\\\\n\\lim_{N\\to\\infty}S_{T,N}\n&=I_T:=\\int_0^T\\mathbb E[a(t,X_t)]\\,dt.\n\\end{aligned}                                      \\tag{11}\n$$","paper":"short","role":"definition","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43303},{"anchor":"qaoa-boundary","at":"b43592-44412","end_byte":44412,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The fixed-T mesh limit with strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second clause is the mesh limit. The third clause is the endpoint limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"The same proposition proves this limit. Its proof starts from the\nEuler approximation and the gradient martingale, which give\n$$\n\\left\\|u(T,Y_N)-\\sum_{j=1}^N\\sqrt\\Delta\\,\n a(t_{j-1},Y_{j-1})Z_j\\right\\|_2\\longrightarrow0.\n$$\nConditional-mean preservation transfers these correlations to $F_N$\nfor the first $N$ coordinates. The rounding error itself stays of order\none, since its squared $L^2$ error satisfies\n$\\mathbb E(F_N-u(T,Y_N))^2\\to1-T$. Bessel's inequality, applied\nseparately to $(g_j)$ and $(Z_j)$, controls the errors of both\ncorrelation vectors in $\\ell^2$, and their bounded norms control the\ngrowing pairing by Cauchy–Schwarz. The terms with $j<N$ give a Riemann\nsum, and the last term, from the rounding coordinate, vanishes. All\nthese limits keep $T$ fixed. Only afterwards does $T\\uparrow1$ give\n$P_*$.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43592},{"anchor":"dependencies","at":"b8042-8180","end_byte":8180,"heading_id":"outline-of-the-proof","lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The fixed-T mesh limit with strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second clause is the mesh limit. The third clause is the endpoint limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"The fixed-$T$ Gaussian\napproximation recovers this integral through Euler coupling, orthogonal\nnormalization and mean-preserving rounding.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":8042},{"anchor":"dependencies","at":"b24559-24666","end_byte":24666,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"The fixed-T mesh limit with strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"The second clause is the mesh limit. The third clause is the endpoint limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The shifted sum of A_j B_{j+1}, with A_0=0, over j from 0 to N.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"The particular shifted Gaussian\ncoefficient identity used here is proved in [OpenAI (2026a), §7][fs-cons].","paper":"long","role":"mention","section":"2.3 Prior results and their roles","start_byte":24559},{"anchor":"approximation-choices","at":"b30709-30857","end_byte":30857,"heading_id":"the-order-of-the-approximation-limits","lean":[{"clause":"Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ)) ∧\n  Tendsto (fun S ↦ ∫ t in (0 : ℝ)..S, ∫ ω, curvature W γ t (X t ω) ∂W.μ)\n    (𝓝[<] (1 : ℝ)) (𝓝 groundStateValue)","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[158,161],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: N to infinity at fixed T, then a separate limit in S. No tail bound in 1-T.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Refining N at fixed T, with the strip bounds and the unit curvature moment as hypotheses.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"N tends to infinity at fixed T, then the separate limit in S.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Finite Gaussian coefficients.** Fix $\\gamma,X$ and $T<1$. Refine $N$\nusing the fixed-strip bounds, then control the remaining value tail by $1-T$.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":30709},{"anchor":"gaussian","at":"b92595-92725","end_byte":92725,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: the first clause of FiniteGaussianConclusion, eventual positivity of the normalisers at fixed T.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[157,157],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Eventually in N, all normalisers are positive.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"and its diffusion, the construction below has positive active normalizers\nfor all sufficiently fine meshes at every fixed $0<T<1$.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":92595},{"anchor":"gaussian","at":"b92726-92851","end_byte":92851,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: the second and third clauses of FiniteGaussianConclusion, first N at fixed T, then S rising to one.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"First N at fixed T, then S rising to one.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Its adjacent\ncorrelation sum converges to $\\int_0^T\\mathbb E a(t,X_t)\\,dt$, and this\nintegral tends to $P_*$ as $T\\uparrow1$.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":92726},{"anchor":"gaussian","at":"b93035-93119","end_byte":93119,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The same adjacent pairing, over range (N+1).","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"$$\nS_{T,N}=\\sum_{j=0}^N A_jB_{j+1}\n\\longrightarrow\\int_0^T\\mathbb E a(t,X_t)\\,dt.\n$$","paper":"long","role":"definition","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93035},{"anchor":"gaussian","at":"b93363-93596","end_byte":93596,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":31,"line":27,"lines":[27,31],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.euler","note":"Y, with coordinates numbered from zero.","plain":"Euler recursion on a uniform mesh","relation":"defines"},{"end_line":130,"line":128,"lines":[128,130],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.normalizer","note":"c_j.","plain":"Normaliser of the raw coefficient","relation":"defines"},{"end_line":134,"line":132,"lines":[132,134],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.gaussianIncrement","note":"g_j.","plain":"Normalised Gaussian increment","relation":"defines"}],"lean_source":"statement","markdown":"On independent standard normals $Z_1,\\ldots,Z_{N+1}$, set\n$$\nY_0=0,\\quad Y_j=Y_{j-1}+\\sqrt\\Delta Z_j+\n \\Delta\\gamma(t_{j-1})u(t_{j-1},Y_{j-1}),\n$$\n$$\nc_j=(\\mathbb E a(t_{j-1},Y_{j-1})^2)^{-1/2},\\qquad\ng_j=c_ja(t_{j-1},Y_{j-1})Z_j.\n$$","paper":"long","role":"definition","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":93363},{"anchor":"gaussian","at":"b98195-98244","end_byte":98244,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: the third clause of FiniteGaussianConclusion, with S<1.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[160,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The second limit, with S<1.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Only afterward let $T\\uparrow1$, obtaining $P_*$.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":98195},{"anchor":"gaussian","at":"b98245-98305","end_byte":98305,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"No spin count appears in the definition.","plain":"Adjacent pairing of A and B","relation":"defines"}],"lean_source":"statement","markdown":"The mesh count $N$ is unrelated to the spin-system size $n$.","paper":"long","role":"mention","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":98245},{"anchor":"gaussian","at":"b98854-99037","end_byte":99037,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model: the two limits this choice uses. The choice of T and then N is not itself stated.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"Thus, for every accuracy, the same chosen minimizer supplies a cutoff $T$\nand then one finite function $F_N$ with $|S_{T,N}-P_*|$ below that accuracy.\nThis is the finite scalar input.","paper":"long","role":"explanation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":98854},{"anchor":"qaoa-boundary","at":"b43592-43631","end_byte":43631,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"On one strip, given the unit moment.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"states"},{"end_line":97,"line":58,"lines":[79,97],"module":"OAI.Probability.SKValue.Evolution.MeanDerivativeContinuous","name":"OAI.SKValue.finiteGaussian_from_diffusion","note":"","plain":"Finite-Gaussian limit from a diffusion","proof_lines":[79,97],"relation":"proof"}],"lean_source":"statement","markdown":"The same proposition proves this limit.","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43592},{"anchor":"qaoa-boundary","at":"b43632-43815","end_byte":43815,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":54,"line":44,"lines":[44,46],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_tendsto_zero","note":"gradientMeshError is the square of this norm. Stated for a gradient on a strip.","plain":"Gradient mesh error tends to zero","proof_lines":[47,54],"relation":"states"},{"end_line":54,"line":51,"lines":[51,54],"module":"OAI.Probability.SKValue.Equations.GradientStrip","name":"OAI.SKValue.gradientMeshError","note":"The square of the norm in the display.","plain":"Gradient error on the mesh","relation":"defines"},{"end_line":42,"line":34,"lines":[34,36],"module":"OAI.Probability.SKValue.Equations.GradientErrorConstant","name":"OAI.SKValue.gradientMeshError_bound_linear","note":"With a rate: at most a constant times the step.","plain":"Mesh error linear in the step","proof_lines":[37,42],"relation":"part"}],"lean_source":"statement","markdown":"Its proof starts from the\nEuler approximation and the gradient martingale, which give\n$$\n\\left\\|u(T,Y_N)-\\sum_{j=1}^N\\sqrt\\Delta\\,\n a(t_{j-1},Y_{j-1})Z_j\\right\\|_2\\longrightarrow0.\n$$","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43632},{"anchor":"qaoa-boundary","at":"b44234-44274","end_byte":44274,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":136,"line":47,"lines":[100,113],"module":"OAI.Probability.SKValue.Processes.MeshMeanBound","name":"OAI.SKValue.finiteGaussian_from_strip_and_moments","note":"hvlim, by perturbed_shifted_grid_tendsto: a grid sum of perturbed samples of f.","plain":"Finite-Gaussian limit from strip and moments","proof_lines":[60,136],"relation":"proof"},{"end_line":140,"line":114,"lines":[114,120],"module":"OAI.Probability.SKValue.Evolution.UniformGridRiemannBound","name":"OAI.SKValue.perturbed_shifted_grid_tendsto","note":"","plain":"Perturbed Riemann sums converge","proof_lines":[121,140],"relation":"part"}],"lean_source":"statement","markdown":"The terms with $j<N$ give a Riemann\nsum,","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44234},{"anchor":"qaoa-boundary","at":"b44275-44333","end_byte":44333,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":136,"line":47,"lines":[83,87],"module":"OAI.Probability.SKValue.Processes.MeshMeanBound","name":"OAI.SKValue.finiteGaussian_from_strip_and_moments","note":"hend: the last leading coefficient is at most √δ·D, and |bstar|≤1.","plain":"Finite-Gaussian limit from strip and moments","proof_lines":[60,136],"relation":"proof"}],"lean_source":"statement","markdown":"and the last term, from the rounding coordinate, vanishes.","paper":"short","role":"derivation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44275},{"anchor":"qaoa-boundary","at":"b44334-44412","end_byte":44412,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Two separate clauses: N→∞ at fixed T, then S↑1.","plain":"The finite-Gaussian conclusion","relation":"part"}],"lean_source":"statement","markdown":"All\nthese limits keep $T$ fixed. Only afterwards does $T\\uparrow1$ give\n$P_*$.","paper":"short","role":"explanation","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":44334}]},"scalar.smooth_positive_density":{"lean":[],"short":"Smooth positive density","statements":[{"anchor":"density-positive","at":"b170403-170936","end_byte":170936,"heading_id":"strict-positivity-of-the-density-away-from-zero","lean":[],"lean_source":"record","markdown":"Combining (4.1) with the preceding\nidentity gives\n$$\n2\\gamma'(t)\\mathbb E a^3\n\\geq\\frac{\\gamma(t)^2}{1000}\\mathbb E a^4>0,\n\\qquad 0<t<1.\n$$\nThe final strict inequality uses $\\mathbb E a^4\\geq(\\mathbb E a^2)^2=1$. Cauchy–Schwarz also gives\n$\\mathbb E a^3\\le\\sqrt{\\mathbb E a^2\\mathbb E a^4}\n=\\sqrt{\\mathbb E a^4}\\le\\mathbb E a^4$. Therefore the same estimate proves\nthe quantitative ordinary consequence\n$$\n\\gamma'(t)\\ge\\frac{\\gamma(t)^2}{2000}>0\\quad(0<t<1),\\qquad\n\\mu_\\gamma((r,s])\\ge\\frac{s-r}{2000}\\gamma(r)^2\n\\quad(0<r<s<1).\n$$","paper":"long","role":"statement","section":"9.2 Strict positivity of the density away from zero","start_byte":170403},{"anchor":"density","at":"b166811-167475","end_byte":167475,"heading_id":"smoothness-of-the-minimizing-coefficient","lean":[],"lean_source":"record","markdown":"Convexity gives $a\\geq0$, and the already established second moment gives\n$\\mathbb E a^3\\geq(\\mathbb E a^2)^{3/2}=1$. Thus the quotient\n$$\n\\widetilde\\gamma(t)=\\frac{\\mathbb E b(t,X_t)^2}\n                         {2\\mathbb E a(t,X_t)^3}\n$$\nis well defined and continuous, before any continuity of $\\gamma$ is known. It equals $\\gamma$ almost everywhere.\n\nApproaching an arbitrary $t<1$ from the right through the full-measure equality set, continuity of $\\widetilde\\gamma$ and right continuity of $\\gamma$ give equality at that $t$ as well. This proves continuity of $\\gamma$. At zero, $X_0=0$, and evenness of $\\Phi(0,\\cdot)$ gives $b(0,0)=0$; hence $\\gamma(0)=0$.","paper":"long","role":"derivation","section":"9.1 Smoothness of the minimizing coefficient","start_byte":166811},{"anchor":"density","at":"b168429-168650","end_byte":168650,"heading_id":"smoothness-of-the-minimizing-coefficient","lean":[],"lean_source":"record","markdown":"Induction proves $\\gamma\\in C^\\infty([0,1))$, with derivatives at zero taken from the right. Since $\\gamma(0)=0$, there is no atom at zero. Its Stieltjes measure is $\\gamma'(t)\\,dt$, and monotonicity gives $\\gamma'\\geq0$.","paper":"long","role":"derivation","section":"9.1 Smoothness of the minimizing coefficient","start_byte":168429},{"anchor":"density-consequence","at":"b30106-30580","end_byte":30580,"heading_id":"strict-positivity-of-the-density","lean":[],"lean_source":"record","markdown":"**Corollary 4 (smooth positive density; [OpenAI (2026a),\nCorollary 6.4][FS]).** Under the hypotheses of Theorem 1,\n$\\gamma(0)=0$ and $\\gamma\\in C^\\infty([0,1))$, with derivatives\nat zero taken from the right. Its Stieltjes measure is\n$\\mu_\\gamma(dt)=\\gamma'(t)\\,dt$, and\n$$\n2\\gamma'(t)\\mathbb E[a(t,X_t)^3]\n\\geq\\frac{\\gamma(t)^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\qquad(0<t<1).                                    \\tag{7}\n$$\nOnly nonnegativity of the density at zero is asserted.","note":"Neither formal statement contains this corollary. The highlights in its proof mark the steps that are in the Lean.","paper":"short","role":"statement","section":"4. Strict positivity of the density","start_byte":30106},{"anchor":"density-consequence","at":"b33409-33622","end_byte":33622,"heading_id":"strict-positivity-of-the-density","lean":[],"lean_source":"record","markdown":"In particular, Cauchy–Schwarz gives\n$$\n\\mathbb E a^3\\leq\\sqrt{\\mathbb E a^2\\,\\mathbb E a^4}\n=\\sqrt{\\mathbb E a^4}\\leq\\mathbb E a^4,\n$$\nso the same inequality implies\n$\\gamma'(t)\\geq\\gamma(t)^2/2000$ for $0<t<1$.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":33409},{"anchor":"density-and-approximation","at":"b60043-60378","end_byte":60378,"heading_id":"further-reading","lean":[],"lean_source":"record","markdown":"Full support alone does not give a density. Additional differentiation\nof the identities $q=t$ and $\\mathbb E[a^2]=1$, followed by\nanother use of the integrated curvature estimate on locally constant\ncomparison coefficients, proves that $\\gamma$\nis smooth, $\\gamma(0)=0$, and\n$\\mu_\\gamma=\\gamma'(t)\\,dt$ with $\\gamma'(t)>0$ for $0<t<1$","paper":"expository","role":"statement","section":"Further reading","start_byte":60043},{"anchor":"order-parameter-example","at":"b13020-13353","end_byte":13353,"heading_id":"what-a-gap-means","lean":[],"lean_source":"statement","markdown":"Its\nmeasure is $c\\delta_0$, supported only at zero, so positivity of the\nfunction falls far short of the conclusion we seek. Full support is also\nweaker than having a density, since a measure with summable positive\nmasses at a dense countable set has full support. The further-reading note returns to the stronger\ndensity conclusion.","note":"Not in the Lean: a remark about measures in general. The Lean conclusion stops at support and strict increase, with no density clause.","paper":"expository","role":"mention","section":"2. What a gap means","start_byte":13020},{"anchor":"density-consequence","at":"b30582-30700","end_byte":30700,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":99,"line":91,"lines":[91,94],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.compactJetField_generator_one","note":"Full-support model. The drift of a, on a strip. The diffusion term is implicit in the generator.","plain":"Itô generator of the curvature a","proof_lines":[95,99],"relation":"part"}],"lean_source":"statement","markdown":"**Proof.** All spatial factors in the expectations below are evaluated\nat $(t,X_t)$. Since $da=-\\gamma a^2\\,dt+b\\,dB$,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":30582},{"anchor":"density-consequence","at":"b30701-30827","end_byte":30827,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":105,"line":94,"lines":[94,96],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.stripA_hasDerivWithinAt_right","note":"Full-support model. The right derivative of E a² is E[b²−2γa³], at every time. The vanishing of the left side is not here.","plain":"Right derivative of the curvature moment","proof_lines":[97,105],"relation":"part"},{"end_line":40,"line":31,"lines":[31,33],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.curvatureExpr_hasDerivAt","note":"Scalar model. The two-sided derivative at continuity points of γ.","plain":"Time derivative of the curvature moment","proof_lines":[34,40],"relation":"part"},{"end_line":74,"line":73,"lines":[73,74],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.JetExpr.curvatureExpr_sourceP","note":"Scalar model. The −2a³ term.","plain":"Source of a², γ part","proof_lines":[74,74],"relation":"part"}],"lean_source":"statement","markdown":"Itô's formula applied\nto $a^2$ gives, for almost every $t$,\n$0=\\frac{d}{dt}\\mathbb E a^2=\\mathbb E b^2-2\\gamma\\mathbb E a^3$,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":30701},{"anchor":"density-consequence","at":"b30828-30902","end_byte":30902,"heading_id":"strict-positivity-of-the-density","lean":[{"clause":"curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Full-support model. The unit moment. The two facts are not combined in the Lean.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"}],"lean_source":"statement","markdown":"where the left side vanishes because full support gives\n$\\mathbb E a^2=1$.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":30828},{"anchor":"density-consequence","at":"b30903-31004","end_byte":31004,"heading_id":"strict-positivity-of-the-density","lean":[],"lean_source":"record","markdown":"Solving for the coefficient yields\n$$\n\\gamma(t)=\\frac{\\mathbb E b^2}{2\\mathbb E a^3}.      \\tag{8}\n$$","note":"Equation (8) is outside the Lean, as is everything built on it to smoothness.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":30903},{"anchor":"density-consequence","at":"b31043-31140","end_byte":31140,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":192,"line":190,"lines":[190,191],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.curvature_nonneg","note":"Full-support model. The sign a≥0 only.","plain":"The curvature is nonnegative","proof_lines":[192,192],"relation":"part"}],"lean_source":"statement","markdown":"Its denominator is at least\ntwo, because $a\\geq0$ and\n$\\mathbb E a^3\\geq(\\mathbb E a^2)^{3/2}=1$.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31043},{"anchor":"density-consequence","at":"b31141-31168","end_byte":31168,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":21,"line":13,"lines":[13,15],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.IsDiffusion.jetMoment_continuous","note":"Scalar model: every polynomial jet moment, on [0,T] for T<1.","plain":"Jet moments are continuous","proof_lines":[16,21],"relation":"states"},{"end_line":32,"line":26,"lines":[26,27],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripA_continuous","note":"Full-support model: E a² on the strip.","plain":"Strip curvature moment is continuous","proof_lines":[28,32],"relation":"part"}],"lean_source":"statement","markdown":"The moments are continuous,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31141},{"anchor":"density-consequence","at":"b31316-31514","end_byte":31514,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":12,"line":11,"lines":[11,12],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.jetMoment","note":"Scalar model. P is a JetExpr. Coordinate k stands for the k-th derivative of Φ_x.","plain":"Moment of a polynomial in jets","relation":"defines"},{"end_line":69,"line":68,"lines":[68,69],"module":"OAI.Probability.SKValue.Equations.PolynomialObservable","name":"OAI.SKValue.polyJet","note":"","plain":"Polynomial in jets of Φ","relation":"defines"}],"lean_source":"statement","markdown":"For smoothness, fix $T<1$ and consider the polynomial moments\n$$\nM_P(t)=\\mathbb E P(\\Phi_x(t,X_t),\\ldots,\\partial_x^m\\Phi(t,X_t)),\n$$\nwhere $P$ is a polynomial in any finite number $m$ of variables.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31316},{"anchor":"density-consequence","at":"b31515-31732","end_byte":31732,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":85,"line":28,"lines":[28,32],"module":"OAI.Probability.SKValue.Control.PolynomialGenerator","name":"OAI.SKValue.IsDiffusion.polyJet_generator","note":"Scalar model. This identity, with Q_P=e.sourceA and R_P=e.sourceP.","plain":"Itô formula for polynomial jets","proof_lines":[33,85],"relation":"states"},{"end_line":14,"line":14,"lines":[14,14],"module":"OAI.Probability.SKValue.Control.PolynomialGenerator","name":"OAI.SKValue.JetExpr.sourceA","note":"","plain":"Generator source, part without γ","relation":"defines"},{"end_line":15,"line":15,"lines":[15,15],"module":"OAI.Probability.SKValue.Control.PolynomialGenerator","name":"OAI.SKValue.JetExpr.sourceP","note":"","plain":"Generator source, part multiplying γ","relation":"defines"}],"lean_source":"statement","markdown":"Spatial differentiation of the PDE and Itô's product rule give\npolynomials $Q_P,R_P$, involving finitely many further spatial\nderivatives, such that\n$$\nM_P(t)-M_P(0)=\\int_0^t\n [M_{Q_P}(s)+\\gamma(s)M_{R_P}(s)]\\,ds.\n$$","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31515},{"anchor":"density-consequence","at":"b31733-31792","end_byte":31792,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":85,"line":28,"lines":[28,32],"module":"OAI.Probability.SKValue.Control.PolynomialGenerator","name":"OAI.SKValue.IsDiffusion.polyJet_generator","note":"Scalar model. γ appears only as γ.coeff t under the integral.","plain":"Itô formula for polynomial jets","proof_lines":[33,85],"relation":"part"}],"lean_source":"statement","markdown":"The identity involves $\\gamma$ and none of its derivatives.","paper":"short","role":"explanation","section":"4. Strict positivity of the density","start_byte":31733},{"anchor":"density-consequence","at":"b31793-31869","end_byte":31869,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":21,"line":13,"lines":[13,21],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.IsDiffusion.jetMoment_continuous","note":"Scalar model. Boundedness comes from polyJet_joint on the strip.","plain":"Jet moments are continuous","proof_lines":[16,21],"relation":"part"}],"lean_source":"statement","markdown":"Bounded\nspatial derivatives on the fixed strip make every moment continuous.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31793},{"anchor":"density-consequence","at":"b31870-31922","end_byte":31922,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":48,"line":22,"lines":[22,25],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.IsDiffusion.jetMoment_hasDerivAt","note":"Scalar model. A derivative at continuity points of γ, from the integral identity. Lean does not go on to prove γ smooth.","plain":"Time derivative of a jet moment","proof_lines":[26,48],"relation":"proof"}],"lean_source":"statement","markdown":"The integral identity then makes every moment $C^1$,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31870},{"anchor":"density-consequence","at":"b32552-32714","end_byte":32714,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":76,"line":75,"lines":[75,76],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.JetExpr.curvatureSlope_sourceA","note":"Scalar model. The source terms of b²−2ca³ without γ.","plain":"Slope source, part without γ","proof_lines":[76,76],"relation":"part"},{"end_line":78,"line":77,"lines":[77,78],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.JetExpr.curvatureSlope_sourceP","note":"Scalar model. The terms multiplied by γ. Lean states the sum, with c frozen, so this display splits it.","plain":"Slope source, γ part","proof_lines":[78,78],"relation":"part"}],"lean_source":"statement","markdown":"The two differentiated moments are now\n$$\n\\frac{d}{dt}\\mathbb E b^2=\\mathbb E[d^2-6\\gamma ab^2],\n\\qquad\n\\frac{d}{dt}\\mathbb E a^3=\\mathbb E[3ab^2-3\\gamma a^4].\n$$","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":32552},{"anchor":"density-consequence","at":"b32715-32914","end_byte":32914,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":52,"line":41,"lines":[41,46],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.curvatureSlope_hasDerivAt","note":"Scalar model. The right side of (9). c is frozen at γ(t), so the 2γ′E a³ term is absent from the Lean.","plain":"Time derivative of the curvature slope","proof_lines":[47,52],"relation":"proof"}],"lean_source":"statement","markdown":"For a smooth time coefficient, differentiating\n$q''=\\mathbb E b^2-2\\gamma\\mathbb E a^3$ therefore gives\n$$\nq^{(3)}+2\\gamma'\\mathbb E a^3\n=\\mathcal C_\\gamma(t).                              \\tag{9}\n$$","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":32715},{"anchor":"density-consequence","at":"b32979-33040","end_byte":33040,"heading_id":"strict-positivity-of-the-density","lean":[{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Full-support model: (4) on a plateau, proved there directly. Lean has no (9) to specialise.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"Full-support model: the right side as an expectation.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"},{"end_line":35,"line":27,"lines":[27,29],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.coercivePolynomial_scaled","note":"Full-support model: the rescaling, as a polynomial identity.","plain":"Coercive integrand in terms of jets","proof_lines":[30,35],"relation":"part"}],"lean_source":"statement","markdown":"On a plateau, $\\gamma'=0$ and $(v,w,z)=c(a,b,d)$ recover (4).","paper":"short","role":"explanation","section":"4. Strict positivity of the density","start_byte":32979},{"anchor":"density-consequence","at":"b33144-33191","end_byte":33191,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Full-support model, scaled by c². Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"},{"end_line":33,"line":12,"lines":[12,17],"module":"OAI.Probability.SKValue.Coercivity.LiteralCoercivity","name":"OAI.SKValue.IsDiffusion.literal_coercivity","note":"Scalar model: the same bound in v, w, z, for every t in (0,1).","plain":"Coercivity along the value-proof diffusion","proof_lines":[18,33],"relation":"states"},{"end_line":38,"line":10,"lines":[10,14],"module":"OAI.Probability.SKValue.Processes.CurvatureNonneg","name":"OAI.SKValue.IsDiffusion.curvature_generator_nonneg","note":"Scalar model. The sign of the moment, from literal_coercivity.","plain":"Coercive moment is nonnegative","proof_lines":[15,38],"relation":"part"}],"lean_source":"statement","markdown":"Proposition 3 bounds the right\nside from below.","paper":"short","role":"explanation","section":"4. Strict positivity of the density","start_byte":33144},{"anchor":"density-consequence","at":"b33245-33308","end_byte":33308,"heading_id":"strict-positivity-of-the-density","lean":[{"clause":"StrictMonoOn (extend γ.val) (Ico (0:ℝ) 1)","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Full-support model. Strict increase. The fact γ(0)=0 is not in the Lean.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"}],"lean_source":"statement","markdown":"Strict increase and\n$\\gamma(0)=0$ give $\\gamma(t)>0$ for $t>0$,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":33245},{"anchor":"density-consequence","at":"b33624-33770","end_byte":33770,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"Full-support model. The Lean limit is of the spatial moment. No time derivative of any moment is taken in actual_coercivity.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"proof"}],"lean_source":"statement","markdown":"Proposition 3 passes spatial moments to the limit, and its proof does\nnot identify $q_\\delta'''$ with the third derivative of the limiting\nmoment.","paper":"short","role":"explanation","section":"4. Strict positivity of the density","start_byte":33624},{"anchor":"density-consequence","at":"b31267-31293","end_byte":31293,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":162,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_even","note":"Full-support model. Evenness of Φ only. The Lean does not go on to γ(0)=0.","plain":"The value is even in x","proof_lines":[157,162],"relation":"proof"}],"lean_source":"statement","markdown":"Evenness\ngives $b(0,0)=0$,","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":31267},{"anchor":"density-consequence","at":"b33327-33408","end_byte":33408,"heading_id":"strict-positivity-of-the-density","lean":[{"end_line":56,"line":51,"lines":[51,52],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"Full-support model. Lean shows E a⁴>0 for any coefficient, by interval positivity, with no use of E a²=1.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"proof"}],"lean_source":"statement","markdown":"Strictness follows from $\\gamma(t)>0$ and\n$\\mathbb E a^4\\geq(\\mathbb E a^2)^2=1$.","paper":"short","role":"derivation","section":"4. Strict positivity of the density","start_byte":33327}]},"scalar.source_diffusion_exists":{"lean":[{"clause":"IsDiffusion W γ (diffusionProcess W γ)","end_line":77,"line":74,"lines":[74,74],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.diffusionProcess_isDiffusion","note":"The concrete process satisfies the diffusion predicate.","plain":"The constructed process solves the equation","proof_lines":[75,77],"relation":"part"},{"clause":"∃X:ℝ → W.Ω → ℝ,IsDiffusion W γ X","end_line":79,"line":78,"lines":[78,78],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.exists_diffusion","note":"The existence consequence.","plain":"A diffusion exists","proof_lines":[79,79],"relation":"part"}],"short":"Diffusion constructed for every coefficient","statements":[{"anchor":"formal-parisi-clause","at":"b245775-246152","end_byte":246152,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":27,"line":25,"lines":[25,26],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.ground_state_parisi_target","note":"Exactly this implication. Its proof supplies diffusionProcess itself.","plain":"Ground state equals Parisi at minimisers","proof_lines":[27,27],"relation":"states"},{"clause":"IsDiffusion W γ (diffusionProcess W γ)","end_line":77,"line":74,"lines":[74,74],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.diffusionProcess_isDiffusion","note":"The concrete process satisfies the diffusion predicate.","plain":"The constructed process solves the equation","proof_lines":[75,77],"relation":"part"},{"clause":"∃X:ℝ → W.Ω → ℝ,IsDiffusion W γ X","end_line":79,"line":78,"lines":[78,78],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.exists_diffusion","note":"The existence consequence.","plain":"A diffusion exists","proof_lines":[79,79],"relation":"part"}],"lean_source":"statement","markdown":"The third declaration, `ground_state_parisi_target`, assumes a minimizer\nalone and concludes $P_*=\\mathcal P(\\gamma)$. Its proof chooses\n$X=\\texttt{diffusionProcess}\\ W\\ \\gamma$ and proves the diffusion\npredicate for that process before applying the equality with a supplied\ndiffusion. No diffusion is required from the caller, and none of these\nclauses constructs a minimizer.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":245775},{"anchor":"dependencies","at":"b9243-9395","end_byte":9395,"heading_id":"outline-of-the-proof","lean":[{"end_line":157,"line":154,"lines":[154,156],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivAt","note":"Support model, constructed diffusion: q'=E a² on (0,1). The strip estimates behind it are separate declarations.","proof_lines":[157,157],"relation":"part"},{"end_line":327,"line":309,"lines":[309,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Support model: the constructed process is a diffusion for every order parameter, with no minimality.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"clause":"IsDiffusion W γ (diffusionProcess W γ)","end_line":77,"line":74,"lines":[74,74],"module":"OAI.Probability.SKValue.Processes.DiffusionExistence","name":"OAI.SKValue.diffusionProcess_isDiffusion","note":"Scalar model: the concrete process satisfies the diffusion predicate.","plain":"The constructed process solves the equation","proof_lines":[75,77],"relation":"part"}],"lean_source":"statement","markdown":"**The diffusion and moment identity.** For any admissible $\\gamma$, the\ncompact-strip estimates justify defining $X,q$ and computing\n$q'=\\mathbb E a^2$;","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":9243}]},"scalar.source_empirical_energy_limit":{"lean":[{"end_line":122,"line":44,"lines":[44,55],"module":"OAI.Probability.SKValue.GroundState.EmpiricalEnergyLimit","name":"OAI.SKValueG.empiricalColumnEnergy_ae_tendsto","note":"Limit for one index j<K; the paper sums over j. The sign is a measurable real function with square one.","plain":"Empirical energy limit","proof_lines":[56,122],"relation":"states"}],"short":"Empirical energy limit","statements":[{"anchor":"gaussian-empirical-limit","at":"b190877-191468","end_byte":191468,"heading_id":"orthogonalization-preserves-the-limiting-correlations","lean":[{"end_line":122,"line":44,"lines":[44,55],"module":"OAI.Probability.SKValue.GroundState.EmpiricalEnergyLimit","name":"OAI.SKValueG.empiricalColumnEnergy_ae_tendsto","note":"Limit for one index j<K; the paper sums over j. The sign is a measurable real function with square one.","plain":"Empirical energy limit","proof_lines":[56,122],"relation":"states"}],"lean_source":"statement","markdown":"Expand the revealed energy before taking its limit:\n$$\n\\begin{aligned}\nE_{n,j}={}&\\langle v_{n,j},\\widehat S_n\\rangle\n \\langle R_{n,j+1}\\widehat S_n,\\widehat z_{j+1,n}\\rangle\\\\\n&+\\frac{\\langle v_{n,j},\\widehat S_n\\rangle^2}{\\sqrt2}\n \\langle v_{n,j},\\widehat z_{j+1,n}\\rangle.\n\\end{aligned}\n$$\nAll terms introduced by projecting away earlier directions vanish,\nas does the second line. The remaining product tends to $a_jb_{j+1}$.\nThus $\\sum_{j<K}E_{n,j}\\to\\sum_{j<K}a_jb_{j+1}$ almost surely.\nThis establishes the random-variable limit, but an expectation bound\nstill requires the next step.","paper":"long","role":"derivation","section":"10.2.3 Orthogonalization preserves the limiting correlations","start_byte":190877},{"anchor":"gaussian-empirical-limit","at":"b190135-190493","end_byte":190493,"heading_id":"orthogonalization-preserves-the-limiting-correlations","lean":[{"end_line":62,"line":56,"lines":[56,61],"module":"OAI.Probability.SKValue.GroundState.EmpiricalVectors","name":"OAI.SKValueG.empirical_inner_tendsto","note":"The strong law for each needed product.","plain":"Empirical inner products converge","proof_lines":[62,62],"relation":"proof"},{"end_line":91,"line":69,"lines":[69,73],"module":"OAI.Probability.SKValue.GroundState.GramSchmidtLimit","name":"OAI.SKValueG.gramSchmidtNormed_approx","note":"Gram–Schmidt changes little when the vectors are nearly orthonormal.","plain":"Gram–Schmidt of nearly orthonormal vectors","proof_lines":[74,91],"relation":"proof"}],"lean_source":"statement","markdown":"We keep $K$ fixed while $n\\to\\infty$. The strong law applies to the\nfinitely many needed products, which are integrable by Cauchy–Schwarz.\nThe empirical Gram matrix tends to the identity. Inductively in\nGram–Schmidt, every projection coefficient tends to zero and every\nremaining norm tends to one; hence\n$\\|v_{n,j}-\\widehat f_{j,n}\\|\\to0$ almost surely.","paper":"long","role":"derivation","section":"10.2.3 Orthogonalization preserves the limiting correlations","start_byte":190135},{"anchor":"gaussian-empirical-limit","at":"b190735-190829","end_byte":190829,"heading_id":"orthogonalization-preserves-the-limiting-correlations","lean":[{"end_line":32,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.GroundState.EmpiricalEnergyLimit","name":"OAI.SKValueG.predictable_coordinate_mean_zero","note":"The cancellation.","plain":"Predictable times later coordinate: mean 0","proof_lines":[26,32],"relation":"proof"}],"lean_source":"statement","markdown":"The last limit uses precisely the fresh-coordinate cancellation stated\nat the start of §10.2.","paper":"long","role":"explanation","section":"10.2.3 Orthogonalization preserves the limiting correlations","start_byte":190735},{"anchor":"gaussian-empirical-limit","at":"b190877-191169","end_byte":191169,"heading_id":"orthogonalization-preserves-the-limiting-correlations","lean":[{"end_line":135,"line":128,"lines":[128,133],"module":"OAI.Probability.SKValue.GroundState.GramSchmidtLimit","name":"OAI.SKValueG.column_revealed_energy","note":"The same two terms.","plain":"Column vector recursion for revealed energy","proof_lines":[134,135],"relation":"part"}],"lean_source":"statement","markdown":"Expand the revealed energy before taking its limit:\n$$\n\\begin{aligned}\nE_{n,j}={}&\\langle v_{n,j},\\widehat S_n\\rangle\n \\langle R_{n,j+1}\\widehat S_n,\\widehat z_{j+1,n}\\rangle\\\\\n&+\\frac{\\langle v_{n,j},\\widehat S_n\\rangle^2}{\\sqrt2}\n \\langle v_{n,j},\\widehat z_{j+1,n}\\rangle.\n\\end{aligned}\n$$","paper":"long","role":"derivation","section":"10.2.3 Orthogonalization preserves the limiting correlations","start_byte":190877},{"anchor":"gaussian-empirical-limit","at":"b191170-191369","end_byte":191369,"heading_id":"orthogonalization-preserves-the-limiting-correlations","lean":[{"end_line":122,"line":44,"lines":[44,55],"module":"OAI.Probability.SKValue.GroundState.EmpiricalEnergyLimit","name":"OAI.SKValueG.empiricalColumnEnergy_ae_tendsto","note":"Almost surely, for the empirical vectors, one index j at a time.","plain":"Empirical energy limit","proof_lines":[56,122],"relation":"states"},{"end_line":176,"line":153,"lines":[153,166],"module":"OAI.Probability.SKValue.GroundState.GramSchmidtLimit","name":"OAI.SKValueG.normalized_column_energy_tendsto","note":"The deterministic limit.","plain":"Limit of the normalised column energy","proof_lines":[167,176],"relation":"proof"}],"lean_source":"statement","markdown":"All terms introduced by projecting away earlier directions vanish,\nas does the second line. The remaining product tends to $a_jb_{j+1}$.\nThus $\\sum_{j<K}E_{n,j}\\to\\sum_{j<K}a_jb_{j+1}$ almost surely.","paper":"long","role":"derivation","section":"10.2.3 Orthogonalization preserves the limiting correlations","start_byte":191170}]},"scalar.source_finite_adaptive_lower":{"lean":[{"end_line":88,"line":74,"lines":[74,80],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.finite_adaptive_SK_lower","note":"Directions and the selected spin's offset are measurable in the revealed column history; bound is by groundStateSequence n.","plain":"Finite adaptive lower bound","proof_lines":[81,88],"relation":"states"}],"short":"Finite adaptive lower bound","statements":[{"anchor":"gaussian-adaptive-lower","at":"b187132-188407","end_byte":188407,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":88,"line":74,"lines":[74,80],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.finite_adaptive_SK_lower","note":"Directions and the selected spin's offset are measurable in the revealed column history; bound is by groundStateSequence n.","plain":"Finite adaptive lower bound","proof_lines":[81,88],"relation":"states"}],"lean_source":"statement","markdown":"We choose each zero-or-unit direction measurably from columns already\nrevealed. Conditional on that history, the current residual candidate\nvectors and offsets are fixed, so the preceding recursion applies.\nMore precisely, conditional on the revealed history, the expected next\ncontinuation value, including its accumulated offsets, equals the current\none. Iteration therefore preserves the initial expected maximum. For any measurable\nchoice of a candidate from the final history, its accumulated offset is\nbounded by the remaining conditional expected maximum. Hence\n$$\n\\mathbb E\\,d_{s(\\mathrm{history})}^{\\rm final}\n\\le\\mathcal V(w,0).\n$$\nContraction of each residual, the finite candidate set and Gaussian\ncolumn moments justify the expectations. Adaptivity does not turn a\nfuture column into part of the conditioning history.\n[Scalar implementation, adaptive recursion][scalar-adaptive].\n\nTake all spins $w_s\\in\\{-1,1\\}^n$ as candidates. The full quadratic\nmatrix has off-diagonal coefficients $(G_{ij}+G_{ji})/\\sqrt2$, which\nare independent standard normals, and a centered diagonal contribution\ncommon to every spin. Thus\n$$\n\\mathcal V(\\mathrm{spins},0)=\\sqrt n\\,\\mathbb E\\max_\\sigma H_n(\\sigma).\n$$\nDividing the offset inequality by $n\\sqrt n$ gives a bound by $e_n$.","paper":"long","role":"statement","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":187132},{"anchor":"gaussian-adaptive-lower","at":"b187132-187548","end_byte":187548,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":114,"line":92,"lines":[92,95],"module":"OAI.Probability.SKValue.GroundState.AdaptiveColumns","name":"OAI.SKValueG.adaptive_value_identity","note":"The value is preserved along the history.","plain":"Adaptive column identity","proof_lines":[96,114],"relation":"proof"},{"end_line":26,"line":20,"lines":[20,26],"module":"OAI.Probability.SKValue.GroundState.AdaptiveColumns","name":"OAI.SKValueG.adaptiveOffset","note":"Accumulated offsets.","plain":"Offsets after adaptive columns","relation":"defines"},{"end_line":18,"line":13,"lines":[13,18],"module":"OAI.Probability.SKValue.GroundState.AdaptiveColumns","name":"OAI.SKValueG.adaptiveResidual","note":"Residual candidates.","plain":"Residual vectors after adaptive columns","relation":"defines"}],"lean_source":"statement","markdown":"We choose each zero-or-unit direction measurably from columns already\nrevealed. Conditional on that history, the current residual candidate\nvectors and offsets are fixed, so the preceding recursion applies.\nMore precisely, conditional on the revealed history, the expected next\ncontinuation value, including its accumulated offsets, equals the current\none. Iteration therefore preserves the initial expected maximum.","paper":"long","role":"derivation","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":187132},{"anchor":"gaussian-adaptive-lower","at":"b187549-187773","end_byte":187773,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":134,"line":118,"lines":[118,122],"module":"OAI.Probability.SKValue.GroundState.AdaptiveColumns","name":"OAI.SKValueG.adaptive_spin_lower","note":"The offset of any measurable final choice.","plain":"Adaptive spin lower bound","proof_lines":[123,134],"relation":"proof"}],"lean_source":"statement","markdown":"For any measurable\nchoice of a candidate from the final history, its accumulated offset is\nbounded by the remaining conditional expected maximum. Hence\n$$\n\\mathbb E\\,d_{s(\\mathrm{history})}^{\\rm final}\n\\le\\mathcal V(w,0).\n$$","paper":"long","role":"derivation","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":187549},{"anchor":"gaussian-adaptive-lower","at":"b188026-188407","end_byte":188407,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":72,"line":59,"lines":[59,60],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.quadraticValue_spin","note":"The normalisation.","plain":"Quadratic value of the SK spins","proof_lines":[61,72],"relation":"proof"},{"end_line":88,"line":74,"lines":[74,80],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.finite_adaptive_SK_lower","note":"The bound, in terms of groundStateSequence.","plain":"Finite adaptive lower bound","proof_lines":[81,88],"relation":"states"}],"lean_source":"statement","markdown":"Take all spins $w_s\\in\\{-1,1\\}^n$ as candidates. The full quadratic\nmatrix has off-diagonal coefficients $(G_{ij}+G_{ji})/\\sqrt2$, which\nare independent standard normals, and a centered diagonal contribution\ncommon to every spin. Thus\n$$\n\\mathcal V(\\mathrm{spins},0)=\\sqrt n\\,\\mathbb E\\max_\\sigma H_n(\\sigma).\n$$\nDividing the offset inequality by $n\\sqrt n$ gives a bound by $e_n$.","note":"Lean proves the normalisation by equal Gaussian increment distances, not through the off-diagonal matrix entries.","paper":"long","role":"derivation","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":188026},{"anchor":"gaussian-adaptive-lower","at":"b188567-188758","end_byte":188758,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":14,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.EmpiricalVectors","name":"OAI.SKValueG.empiricalVector","note":"The normalised empirical vector.","plain":"Empirical vector of samples","relation":"defines"}],"lean_source":"statement","markdown":"To obtain the desired coefficients, take independent rows\n$z^{(1)},z^{(2)},\\ldots$, each with the law of $(z_0,\\ldots,z_K)$,\nand put\n$$\n\\widehat g_n=n^{-1/2}(g(z^{(1)}),\\ldots,g(z^{(n)})).\n$$","paper":"long","role":"definition","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":188567},{"anchor":"gaussian-adaptive-lower","at":"b188759-189085","end_byte":189085,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":134,"line":131,"lines":[131,134],"module":"OAI.Probability.SKValue.GroundState.EmpiricalVectors","name":"OAI.SKValueG.predictableDirection","note":"The direction chosen from the history.","plain":"Predictable direction from the history","relation":"defines"}],"lean_source":"statement","markdown":"Let $v_{n,j}$ be the normalized Gram–Schmidt vector obtained from\n$\\widehat f_{0,n},\\ldots,\\widehat f_{j,n}$, where a zero remainder is\nnormalized to zero. Start the adaptive procedure with a zero direction\nto reveal column $z_0$. After seeing columns through $z_j$,\n$v_{n,j}$ is available, and the next column is $z_{j+1}$.","paper":"long","role":"definition","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":188759},{"anchor":"gaussian-adaptive-lower","at":"b189490-189685","end_byte":189685,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":140,"line":112,"lines":[112,118],"module":"OAI.Probability.SKValue.GroundState.GaussianTranspose","name":"OAI.SKValueG.empirical_finite_spin_lower","note":"The finite bound, for every n>0.","plain":"Empirical energy below the finite-spin value","proof_lines":[119,140],"relation":"proof"},{"end_line":56,"line":51,"lines":[51,53],"module":"OAI.Probability.SKValue.GroundState.GaussianTranspose","name":"OAI.SKValueG.transposeGaussian_preserving","note":"The transposition preserves the law.","plain":"Gaussian matrix transpose preserves law","proof_lines":[54,56],"relation":"proof"}],"lean_source":"statement","markdown":"Transposing the array between\nindependent rows and independent columns preserves its Gaussian law.\nThe finite comparison therefore gives, for every $n>0$,\n$$\n\\mathbb E\\sum_{j<K}E_{n,j}\\le e_n.\n$$","paper":"long","role":"derivation","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":189490},{"anchor":"gaussian-adaptive-lower","at":"b189138-189305","end_byte":189305,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":167,"line":163,"lines":[163,167],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.empiricalColumnEnergy","note":"E_{n,j}.","plain":"Empirical column energy","relation":"defines"},{"end_line":105,"line":104,"lines":[104,105],"module":"OAI.Probability.SKValue.GroundState.GramSchmidtLimit","name":"OAI.SKValueG.orthogonalRemainder","note":"R_{n,j} applied to a vector.","plain":"Remainder after orthogonal projections","relation":"defines"}],"lean_source":"statement","markdown":"Let $R_{n,j}$ project away $v_{n,0},\\ldots,v_{n,j-1}$ and set\n$$\nE_{n,j}=\\langle C_{v_{n,j}}(R_{n,j}\\widehat S_n),\n                       \\widehat z_{j+1,n}\\rangle.\n$$","paper":"long","role":"definition","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":189138},{"anchor":"gaussian-adaptive-lower","at":"b189424-189489","end_byte":189489,"heading_id":"adaptive-directions-and-a-finite-spin-inequality","lean":[{"end_line":193,"line":170,"lines":[170,174],"module":"OAI.Probability.SKValue.GroundState.QuadraticSK","name":"OAI.SKValueG.normalizedOffset_rows","note":"The scaling.","plain":"Normalised offset as column energies","proof_lines":[175,193],"relation":"proof"}],"lean_source":"statement","markdown":"Consequently their revealed\nenergy is exactly $n\\sqrt n E_{n,j}$.","paper":"long","role":"derivation","section":"10.2.2 Adaptive directions and a finite-spin inequality","start_byte":189424}]},"scalar.source_fixed_strip_application":{"lean":[{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Given a minimiser and diffusion. Its conclusion also has the limit to the ground-state value, the proposition's last claim, which this highlight omits.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Helper with exactly these two conclusions, taking every strip bound as a premise.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":72,"line":61,"lines":[61,64],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Same conclusion with gradient contact as premise; this proof derives the strip bounds and the unit moment.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"contains"}],"short":"Fixed-strip mesh limit","statements":[{"anchor":"gaussian","at":"b92520-92887","end_byte":92887,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model. All three clauses: positive normalisers, the mesh limit at fixed T, then the limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Helper with exactly these two conclusions, taking every strip bound as a premise.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":72,"line":61,"lines":[61,64],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Same conclusion with gradient contact as premise; this proof derives the strip bounds and the unit moment.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"contains"}],"lean_source":"statement","markdown":"**Proposition 7.1 (finite Gaussian approximation).** For a fixed minimizer\nand its diffusion, the construction below has positive active normalizers\nfor all sufficiently fine meshes at every fixed $0<T<1$. Its adjacent\ncorrelation sum converges to $\\int_0^T\\mathbb E a(t,X_t)\\,dt$, and this\nintegral tends to $P_*$ as $T\\uparrow1$. The limits are taken in that order.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":92520},{"anchor":"gaussian","at":"b97343-97487","end_byte":97487,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Helper with exactly these two conclusions, taking every strip bound as a premise.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Given a minimiser and diffusion. Its conclusion also has the limit to the ground-state value, the proposition's last claim, which this highlight omits.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":72,"line":61,"lines":[61,64],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Same conclusion with gradient contact as premise; this proof derives the strip bounds and the unit moment.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"contains"}],"lean_source":"statement","markdown":"Hence\n$$\n\\sum_{j=0}^NA_jB_{j+1}\n =\\Delta\\sum_{j=1}^{N-1}\\frac{\\mathbb E a(t_j,Y_j)}{c_j}+o(1)\n \\longrightarrow\\int_0^T\\mathbb E a(t,X_t)\\,dt.\n$$","paper":"long","role":"derivation","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":97343},{"anchor":"formal-scalar","at":"b243451-244001","end_byte":244001,"heading_id":"the-separate-scalar-statement","lean":[{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Given a minimiser and diffusion. Its conclusion also has the limit to the ground-state value, the proposition's last claim, which this highlight omits.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":49,"line":29,"lines":[29,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"Helper with exactly these two conclusions, taking every strip bound as a premise.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"proof"},{"clause":"(∀ᶠ N in atTop, ∀ j < N, 0 < normalizer W γ T N j) ∧\n  Tendsto (shiftedCoefficientSum W γ T) atTop\n    (𝓝 (∫ t in (0 : ℝ)..T, ∫ ω, curvature W γ t (X t ω) ∂W.μ))","clause_in":"OAI.SKValue.FiniteGaussianConclusion","clause_lines":[157,159],"clause_module":"OAI.Probability.SKValue.Model","end_line":72,"line":61,"lines":[61,64],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","note":"Same conclusion with gradient contact as premise; this proof derives the strip bounds and the unit moment.","plain":"Finite-Gaussian limit from contact","proof_lines":[65,72],"relation":"contains"}],"lean_source":"statement","markdown":"Under the\nsame hypotheses and fixed $0<T<1$, it gives eventual positivity of the\nGaussian normalizers and convergence of the finite coefficient sum.\n\nThese are the declarations `scalar_value_target` and\n`finite_gaussian_target`. Their caller supplies the same minimizer and\nassociated diffusion. The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper. Those internal\ninputs are not additional hypotheses of the exported clauses.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":243451}]},"scalar.source_gaussian_column_split":{"lean":[{"end_line":181,"line":158,"lines":[158,163],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue_column","note":"Stated with quadraticValue for the paper's V; the direction v is zero or unit, as in the paper.","plain":"One Gaussian column split","proof_lines":[164,181],"relation":"states"}],"short":"One Gaussian column split","statements":[{"anchor":"gaussian-column-split","at":"b186331-186643","end_byte":186643,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":181,"line":158,"lines":[158,163],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue_column","note":"Stated with quadraticValue for the paper's V; the direction v is zero or unit, as in the paper.","plain":"One Gaussian column split","proof_lines":[164,181],"relation":"states"}],"lean_source":"statement","markdown":"For a finite nonempty family of candidates $w_s$ and offsets $d_s$, define\n$$\n\\mathcal V(w,d)=\\mathbb E_G\\max_s\n  \\{\\langle Q(w_s),G\\rangle+d_s\\}.\n$$\nThe covariance identity and Fubini give the exact recursion\n$$\n\\mathcal V(w,d)=\\mathbb E_Z\\mathcal V\n \\bigl((R_vw_s)_s,(d_s+\\langle C_v(w_s),Z\\rangle)_s\\bigr).\n$$","paper":"long","role":"statement","section":"10.2.1 Revealing one Gaussian column","start_byte":186331},{"anchor":"gaussian-column-split","at":"b185786-186183","end_byte":186183,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":80,"line":70,"lines":[70,73],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.columnSplitCoeff_inner","note":"Equal inner products of coefficients: the same covariance.","plain":"Column split preserves inner products","proof_lines":[74,80],"relation":"proof"},{"end_line":88,"line":82,"lines":[82,85],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.columnSplitCoeff_distance","note":"Equal increment distances, which Lean uses instead.","plain":"Column split preserves distances","proof_lines":[86,88],"relation":"proof"}],"lean_source":"statement","markdown":"Write $\\alpha=\\langle v,w\\rangle$, $\\beta=\\langle v,w'\\rangle$\nand $r=\\langle R_vw,R_vw'\\rangle$. For an independent standard Gaussian\ncolumn $Z$ and matrix $G'$, the covariance of\n$\\langle C_v(w),Z\\rangle+\\langle Q(R_vw),G'\\rangle$ is\n$$\n\\alpha\\beta r+\\frac12\\alpha^2\\beta^2+\\frac12r^2\n=\\frac12(r+\\alpha\\beta)^2\n=\\frac12\\langle w,w'\\rangle^2.\n$$\nThus the new field has the old joint Gaussian law.","note":"Lean compares expected maxima through Gaussian comparison in both directions; it does not state equality of joint laws.","paper":"long","role":"derivation","section":"10.2.1 Revealing one Gaussian column","start_byte":185786},{"anchor":"gaussian-column-split","at":"b185449-185785","end_byte":185785,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":21,"line":20,"lines":[20,21],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticCoeff","note":"Q.","plain":"Quadratic coefficient wᵢ wⱼ/√2","relation":"defines"},{"end_line":15,"line":14,"lines":[14,15],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.columnResidual","note":"R_v.","plain":"Residual after one Gaussian column","relation":"defines"},{"end_line":18,"line":17,"lines":[17,18],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.columnVector","note":"C_v.","plain":"Column vector in the Gaussian split","relation":"defines"}],"lean_source":"statement","markdown":"For $w\\in\\mathbb R^n$, put $Q(w)=w\\otimes w/\\sqrt2$.\nAn independent standard Gaussian matrix $G$ gives a centered field\n$\\langle Q(w),G\\rangle$ with covariance\n$\\langle w,w'\\rangle^2/2$. Given a unit vector $v$, define\n$$\nR_vw=w-\\langle v,w\\rangle v,\\qquad\nC_v(w)=\\langle v,w\\rangle R_vw+\n       \\frac{\\langle v,w\\rangle^2}{\\sqrt2}v.\n$$","paper":"long","role":"definition","section":"10.2.1 Revealing one Gaussian column","start_byte":185449},{"anchor":"gaussian-column-split","at":"b186184-186329","end_byte":186329,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":181,"line":158,"lines":[158,159],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue_column","note":"The hypothesis is v=0 or a unit vector.","plain":"One Gaussian column split","proof_lines":[164,181],"relation":"states"}],"lean_source":"statement","markdown":"The same statement\nholds for $v=0$: then $R_v$ is the identity and $C_v=0$.\nAllowing that case will avoid assuming empirical linear independence.","paper":"long","role":"explanation","section":"10.2.1 Revealing one Gaussian column","start_byte":186184},{"anchor":"gaussian-column-split","at":"b186331-186480","end_byte":186480,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":114,"line":112,"lines":[112,114],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue","note":"V(w,d).","plain":"Expected maximum of the quadratic process","relation":"defines"}],"lean_source":"statement","markdown":"For a finite nonempty family of candidates $w_s$ and offsets $d_s$, define\n$$\n\\mathcal V(w,d)=\\mathbb E_G\\max_s\n  \\{\\langle Q(w_s),G\\rangle+d_s\\}.\n$$","paper":"long","role":"definition","section":"10.2.1 Revealing one Gaussian column","start_byte":186331},{"anchor":"gaussian-column-split","at":"b186481-186643","end_byte":186643,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":181,"line":158,"lines":[158,163],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue_column","note":"The recursion.","plain":"One Gaussian column split","proof_lines":[164,181],"relation":"states"}],"lean_source":"statement","markdown":"The covariance identity and Fubini give the exact recursion\n$$\n\\mathcal V(w,d)=\\mathbb E_Z\\mathcal V\n \\bigl((R_vw_s)_s,(d_s+\\langle C_v(w_s),Z\\rangle)_s\\bigr).\n$$","paper":"long","role":"statement","section":"10.2.1 Revealing one Gaussian column","start_byte":186481},{"anchor":"gaussian-column-split","at":"b186644-186734","end_byte":186734,"heading_id":"revealing-one-gaussian-column","lean":[{"end_line":138,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.GaussianColumn","name":"OAI.SKValueG.quadraticValue_ge_offset","note":"The offset bound.","plain":"Quadratic value dominates each offset","proof_lines":[131,138],"relation":"part"}],"lean_source":"statement","markdown":"Also $d_s\\le\\mathcal V(w,d)$ for each $s$, since the residual Gaussian\nterm has mean zero.","paper":"long","role":"statement","section":"10.2.1 Revealing one Gaussian column","start_byte":186644}]},"scalar.source_ground_state_identity_with_diffusion":{"lean":[{"end_line":23,"line":21,"lines":[21,22],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.IsMinimizer.ground_state_parisi_of_diffusion","note":"Exact premises: a minimiser and a supplied diffusion.","plain":"Ground state equals Parisi, given diffusion","proof_lines":[23,23],"relation":"states"},{"end_line":47,"line":36,"lines":[36,39],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Same equality with gradient contact as premise; this proof combines the two inequalities.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"short":"Ground state equals Parisi","statements":[{"anchor":"gaussian-expectation-limit","at":"b193482-193701","end_byte":193701,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":23,"line":21,"lines":[21,22],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.IsMinimizer.ground_state_parisi_of_diffusion","note":"Exact premises: a minimiser and a supplied diffusion.","plain":"Ground state equals Parisi, given diffusion","proof_lines":[23,23],"relation":"states"},{"end_line":47,"line":36,"lines":[36,39],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Same equality with gradient contact as premise; this proof combines the two inequalities.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"First take the mesh\nlimit to obtain $I(T)\\le P_*$, then let $T\\uparrow1$. The scalar\nterminal assembly identifies that last limit with $\\mathcal P(\\gamma)$.\nTogether with $P_*\\le\\mathcal P(\\gamma)$ this proves equality.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193482},{"anchor":"formal-lower-overview","at":"b183517-183775","end_byte":183775,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":47,"line":36,"lines":[36,39],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Same equality with gradient contact as premise; this proof combines the two inequalities.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"},{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"The upper inequality, for every coefficient, with no minimiser or diffusion premise.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"},{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"The finite-spin limit, stated for the SKValueG sequence.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"part"}],"lean_source":"statement","markdown":"The scalar implementation obtains the value equality by two inequalities.\nIts finite-spin limit follows from superadditivity and a uniform upper\nbound. A finite-profile Gaussian comparison supplies\n$P_*\\le\\mathcal P(\\gamma)$ for every admissible coefficient.","paper":"long","role":"explanation","section":"10.2 A universal bound for predictable correlations","start_byte":183517},{"anchor":"value","at":"b83763-83841","end_byte":83841,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":47,"line":36,"lines":[36,47],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Lean proves the identity from its own upper and lower bounds.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"Besides the support theorem,\nthe proposition uses the attained Parisi formula.","note":"Lean imports no Parisi formula here: its own Guerra upper bound and §10 lower bound replace it.","paper":"long","role":"explanation","section":"6. Recovering the value from a martingale","start_byte":83763},{"anchor":"value","at":"b84805-84864","end_byte":84864,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":47,"line":36,"lines":[43,47],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"le_antisymm of the upper bound and the limit of finite lower bounds.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"The ground-state Parisi formula identifies this with $P_*$.","paper":"long","role":"mention","section":"6. Recovering the value from a martingale","start_byte":84805},{"anchor":"gaussian-expectation-limit","at":"b193556-193701","end_byte":193701,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":47,"line":36,"lines":[41,47],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"parisi_value_assembly identifies the limit; le_antisymm ends the proof.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"The scalar\nterminal assembly identifies that last limit with $\\mathcal P(\\gamma)$.\nTogether with $P_*\\le\\mathcal P(\\gamma)$ this proves equality.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193556}]},"scalar.source_ground_state_upper":{"lean":[{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"For every coefficient, with no minimiser or diffusion premise.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"states"},{"clause":"SKValueG.expectedMaximum n≤(n : ℝ)*parisi W γ","end_line":127,"line":117,"lines":[118,118],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.expectedMaximum_le_parisi","note":"The finite-spin bound for each positive n, before passing to the limit.","plain":"Finite-spin maximum below n times Parisi","proof_lines":[119,127],"relation":"part"}],"short":"Ground state below Parisi","statements":[{"anchor":"formal-lower-overview","at":"b183669-183775","end_byte":183775,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"For every coefficient, with no minimiser or diffusion premise.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"states"},{"clause":"SKValueG.expectedMaximum n≤(n : ℝ)*parisi W γ","end_line":127,"line":117,"lines":[118,118],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.expectedMaximum_le_parisi","note":"The finite-spin bound for each positive n, before passing to the limit.","plain":"Finite-spin maximum below n times Parisi","proof_lines":[119,127],"relation":"part"}],"lean_source":"statement","markdown":"A finite-profile Gaussian comparison supplies\n$P_*\\le\\mathcal P(\\gamma)$ for every admissible coefficient.","paper":"long","role":"statement","section":"10.2 A universal bound for predictable correlations","start_byte":183669},{"anchor":"value-and-diffusion","at":"b17628-17708","end_byte":17708,"heading_id":"allowing-a-response-to-the-noise","lean":[{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Every coefficient, in the scalar model. No Lean theorem maps it to the support model's parisi.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"states"}],"lean_source":"statement","markdown":"Each candidate supplies an upper bound, and minimisation finds the\nsharpest one.","paper":"expository","role":"explanation","section":"3.2 Allowing a response to the noise","start_byte":17628},{"anchor":"dependencies","at":"b16185-16290","end_byte":16290,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"Scalar model. Lean proves this upper bound itself, by its own Guerra interpolation, so it is not an import there.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"states"}],"lean_source":"statement","markdown":"The upper bound proved earlier in that paper gives $P_*\\leq\\inf_{\\gamma\\in\\mathcal U}\\mathcal P(\\gamma)$.","paper":"long","role":"mention","section":"2.3 Prior results and their roles","start_byte":16185}]},"scalar.source_minimizer_curvature_moment":{"lean":[{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"Scalar model, where contact gives the moment without full support. Premises are strip regularity and gradient contact. Theorem 5.1 belongs to the support model.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Scalar model: supplies the gradient contact premise from minimality and the diffusion.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"proof"}],"short":"Unit curvature moment","statements":[{"anchor":"gaps","at":"b71060-71391","end_byte":71391,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Conditional on a minimiser. Every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"Scalar model, a separate proof of the unit curvature moment; premises are strip regularity and gradient contact.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Scalar model: supplies the gradient contact premise from minimality and the diffusion.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"proof"}],"lean_source":"statement","markdown":"**Theorem 5.1 (full support).** In the pure zero-field SK model with\n$\\xi(t)=t^2/2$, let $\\gamma\\in\\mathcal U$ minimize $\\mathcal P$. Then\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and $\\gamma$ is strictly increasing.\nFor every associated diffusion started at zero,\n$\\mathbb E u(t,X_t)^2=t$ and $\\mathbb E a(t,X_t)^2=1$ for\n$0\\le t<1$.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71060},{"anchor":"formal-contact","at":"b182729-183048","end_byte":183048,"heading_id":"contact-from-convexity-on-constant-intervals","lean":[{"end_line":90,"line":84,"lines":[84,87],"module":"OAI.Probability.SKValue.Processes.GradientSquareIntegral","name":"OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","note":"Scalar model, where contact gives the moment without full support. Premises are strip regularity and gradient contact. Theorem 5.1 belongs to the support model.","plain":"Contact gives unit curvature moment","proof_lines":[88,90],"relation":"proof"},{"end_line":82,"line":75,"lines":[75,77],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.gradient_moment_contact","note":"Scalar model: supplies the gradient contact premise from minimality and the diffusion.","plain":"Contact: q(t) = t on [0,1)","proof_lines":[78,82],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $G$ cannot have a positive component, and $G\\equiv0$ on\n$(0,1)$. Differentiation gives $q(t)=t$ there, with the origin supplied\nseparately. The [scalar assembly][scalar-contact-assembly] then obtains\nthe unit curvature moment, finite Gaussian approximation and terminal\nvalue identities from this contact equality.","paper":"long","role":"explanation","section":"10.1 Contact from convexity on constant intervals","start_byte":182729}]},"scalar.source_predictable_lower":{"lean":[{"end_line":134,"line":48,"lines":[48,57],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The sign S appears as spin (F y) for a measurable Boolean F; groundStateValue is the paper's P_*.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"}],"short":"Predictable correlations bound","statements":[{"anchor":"formal-lower-overview","at":"b184536-185211","end_byte":185211,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":134,"line":48,"lines":[48,57],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The sign S appears as spin (F y) for a measurable Boolean F; groundStateValue is the paper's P_*.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"}],"lean_source":"statement","markdown":"**Proposition 10.1 (predictable Gaussian comparison).** Fix $K<\\infty$, independent standard normals $z_0,\\ldots,z_K$,\nand real functions $f_j=f_j(z_0,\\ldots,z_j)$ for $0\\le j<K$.\nAssume measurability, square integrability and\n$\\mathbb E[f_if_j]=\\delta_{ij}$. For any measurable\n$S=S(z_0,\\ldots,z_K)\\in\\{-1,1\\}$, the conclusion is\n$$\n\\sum_{j=0}^{K-1}\\mathbb E[f_jS]\\,\\mathbb E[S z_{j+1}]\\le P_*.\n$$\nThese assumptions allow nonconstant, dependent $f_j$, and even\n$\\mathbb E f_j=0$ is unnecessary. Their order of dependence ensures\n$\\mathbb E[f_i z_{j+1}]=0$ whenever $i\\le j$. The next fresh coordinate,\nrather than centring of $f_i$, is what will remove the projection terms.","paper":"long","role":"statement","section":"10.2 A universal bound for predictable correlations","start_byte":184536},{"anchor":"gaussian-expectation-limit","at":"b191741-192268","end_byte":192268,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":134,"line":48,"lines":[48,57],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The final inequality. Its proof clips at the constant limit, in constant_limit_integral_le, with no direct Fatou step.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"proof"}],"lean_source":"statement","markdown":"Since $S^2=1$, $\\|\\widehat S_n\\|=1$ for $n>0$, and every residual\nprojection contracts. The column formula gives\n$|E_{n,j}|\\le1+\\|\\widehat z_{j+1,n}\\|^2$.\nDefine\n$$\nc_n=\\sum_{j<K}(1+\\|\\widehat z_{j+1,n}\\|^2).\n$$\nThen $\\sum_jE_{n,j}+c_n\\ge0$, $\\mathbb E c_n=2K$ exactly, and\n$c_n\\to2K$ almost surely. Fatou's inequality, the finite bound and\n$e_n\\to P_*$ imply\n$$\n\\sum_{j<K}a_jb_{j+1}+2K\n\\le\\liminf_n\\mathbb E\\bigl[\\sum_{j<K}E_{n,j}+c_n\\bigr]\n\\le P_*+2K.\n$$\nCancelling $2K$ proves the claimed predictable-coefficient inequality.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":191741},{"anchor":"formal-lower-overview","at":"b183776-184127","end_byte":184127,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":134,"line":48,"lines":[48,57],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The general inequality for predictable coefficients.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"},{"end_line":47,"line":36,"lines":[43,47],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"Its use: le_antisymm with the upper bound.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"The reverse\ninequality at a minimizer comes from a more general finite-dimensional\nfact: predictable orthonormal Gaussian functions paired with any spin\nhave a shifted correlation sum at most $P_*$. We reconstruct that fact\nbelow. It supplies the inequality that §7's mesh limit consumes, without\ninvoking the literature's Parisi formula used in §6.","paper":"long","role":"explanation","section":"10.2 A universal bound for predictable correlations","start_byte":183776},{"anchor":"formal-lower-overview","at":"b184536-184795","end_byte":184795,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":134,"line":48,"lines":[48,54],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"ha is the dependence on the first coordinates; hu and ho the orthonormality.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"}],"lean_source":"statement","markdown":"**Proposition 10.1 (predictable Gaussian comparison).** Fix $K<\\infty$, independent standard normals $z_0,\\ldots,z_K$,\nand real functions $f_j=f_j(z_0,\\ldots,z_j)$ for $0\\le j<K$.\nAssume measurability, square integrability and\n$\\mathbb E[f_if_j]=\\delta_{ij}$.","paper":"long","role":"statement","section":"10.2 A universal bound for predictable correlations","start_byte":184536},{"anchor":"formal-lower-overview","at":"b184796-184934","end_byte":184934,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":134,"line":48,"lines":[55,57],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The sum over Fin K, with S the spin of a Bool-valued F.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"}],"lean_source":"statement","markdown":"For any measurable\n$S=S(z_0,\\ldots,z_K)\\in\\{-1,1\\}$, the conclusion is\n$$\n\\sum_{j=0}^{K-1}\\mathbb E[f_jS]\\,\\mathbb E[S z_{j+1}]\\le P_*.\n$$","paper":"long","role":"statement","section":"10.2 A universal bound for predictable correlations","start_byte":184796},{"anchor":"formal-lower-overview","at":"b184935-185211","end_byte":185211,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":32,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.GroundState.EmpiricalEnergyLimit","name":"OAI.SKValueG.predictable_coordinate_mean_zero","note":"The fresh-coordinate cancellation.","plain":"Predictable times later coordinate: mean 0","proof_lines":[26,32],"relation":"proof"}],"lean_source":"statement","markdown":"These assumptions allow nonconstant, dependent $f_j$, and even\n$\\mathbb E f_j=0$ is unnecessary. Their order of dependence ensures\n$\\mathbb E[f_i z_{j+1}]=0$ whenever $i\\le j$. The next fresh coordinate,\nrather than centring of $f_i$, is what will remove the projection terms.","paper":"long","role":"explanation","section":"10.2 A universal bound for predictable correlations","start_byte":184935},{"anchor":"gaussian-expectation-limit","at":"b191741-191895","end_byte":191895,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":49,"line":42,"lines":[42,44],"module":"OAI.Probability.SKValue.GroundState.ColumnBounds","name":"OAI.SKValueG.columnLinear_bound","note":"The column formula, bounded by the norm of z.","plain":"Bound on the column's linear term","proof_lines":[45,49],"relation":"proof"},{"end_line":27,"line":20,"lines":[20,22],"module":"OAI.Probability.SKValue.GroundState.ColumnBounds","name":"OAI.SKValueG.columnResidual_norm_le","note":"Each residual contracts.","plain":"Column residual has smaller norm","proof_lines":[23,27],"relation":"proof"}],"lean_source":"statement","markdown":"Since $S^2=1$, $\\|\\widehat S_n\\|=1$ for $n>0$, and every residual\nprojection contracts. The column formula gives\n$|E_{n,j}|\\le1+\\|\\widehat z_{j+1,n}\\|^2$.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":191741},{"anchor":"gaussian-expectation-limit","at":"b191953-192268","end_byte":192268,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":134,"line":48,"lines":[119,134],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.predictable_coefficients_le_groundState","note":"The inequality, from constant_limit_integral_le and empirical_finite_spin_lower.","plain":"Predictable correlations below ground state","proof_lines":[58,134],"relation":"states"},{"end_line":36,"line":14,"lines":[14,18],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.constant_limit_integral_le","note":"The nonnegative comparison with a constant limit.","plain":"Fatou bound for a constant limit","proof_lines":[19,36],"relation":"proof"}],"lean_source":"statement","markdown":"Then $\\sum_jE_{n,j}+c_n\\ge0$, $\\mathbb E c_n=2K$ exactly, and\n$c_n\\to2K$ almost surely. Fatou's inequality, the finite bound and\n$e_n\\to P_*$ imply\n$$\n\\sum_{j<K}a_jb_{j+1}+2K\n\\le\\liminf_n\\mathbb E\\bigl[\\sum_{j<K}E_{n,j}+c_n\\bigr]\n\\le P_*+2K.\n$$\nCancelling $2K$ proves the claimed predictable-coefficient inequality.","note":"Lean does not use Fatou directly: it clips at the constant limit, as the next sentences say.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":191953},{"anchor":"gaussian-expectation-limit","at":"b192269-192565","end_byte":192565,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":36,"line":14,"lines":[14,36],"module":"OAI.Probability.SKValue.GroundState.ConstantFatou","name":"OAI.SKValueG.constant_limit_integral_le","note":"The helper.","plain":"Fatou bound for a constant limit","proof_lines":[19,36],"relation":"proof"}],"lean_source":"statement","markdown":"The source's constant-limit helper establishes the same nonnegative\ncomparison by clipping at the limiting constant and applying dominated\nconvergence. It shifts to dimensions $n+1$ to avoid dimension zero.\nIt does not apply Fatou directly to the signed energies or presume\nuniform integrability.","paper":"long","role":"mention","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":192269},{"anchor":"gaussian-expectation-limit","at":"b192681-193113","end_byte":193113,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":120,"line":63,"lines":[63,72],"module":"OAI.Probability.SKValue.GroundState.RoundingSpin","name":"OAI.SKValue.normalized_shifted_sum_le_groundState","note":"The specialisation.","plain":"Normalised shifted sum below ground state","proof_lines":[73,120],"relation":"proof"},{"end_line":51,"line":44,"lines":[44,48],"module":"OAI.Probability.SKValue.GroundState.RoundingSpin","name":"OAI.SKValue.finite_rounding_signed_spin","note":"The Bool spin equals the sign almost everywhere.","plain":"Rounded sign is a Boolean spin","proof_lines":[49,51],"relation":"proof"}],"lean_source":"statement","markdown":"For the Euler construction in §7, take $K=N$, relabel its $Z_{j+1}$ as\n$z_j$, and set $f_j=g_{j+1}$. Each normalized increment depends only\non $z_0,\\ldots,z_j$ and the increments are orthonormal when their raw\nsecond moments are positive. The rounded sign is spin-valued almost\nsurely: its tie event is null by the independent last coordinate.\nThe scalar implementation uses a measurable Bool spin equal to that\nsign almost surely.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":192681},{"anchor":"gaussian-expectation-limit","at":"b193130-193306","end_byte":193306,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"Any order parameter, 0≤T<1, N>0, positive second moments.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"states"}],"lean_source":"statement","markdown":"$$\nS_{T,N}=\\sum_{j=1}^N A_jB_{j+1}\\le P_*.\n$$\nThis finite inequality assumes admissibility, $0\\le T<1$, $N>0$ and\npositive active second moments, but no minimizer or diffusion.","paper":"long","role":"statement","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193130},{"anchor":"heat-lower-bound","at":"b200107-200259","end_byte":200259,"heading_id":"the-same-test-at-every-mesh-size","lean":[],"lean_source":"statement","markdown":"**Corollary 10.2 (the heat test).** In this normalization,\n$$\nP_*\\ge\\frac{4\\Gamma(3/4)}{\\sqrt\\pi\\,\\Gamma(1/4)}\n       \\approx0.7627597635.\n\\tag{10.2}\n$$","note":"Not in Lean: the numerical bound is not formalised, only the inequality S_{T,N}≤P_* it starts from.","paper":"long","role":"statement","section":"10.4.1 The same test at every mesh size","start_byte":200107},{"anchor":"heat-lower-bound","at":"b200260-200648","end_byte":200648,"heading_id":"the-same-test-at-every-mesh-size","lean":[{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"Taken at the zero order parameter, this inequality holds in Lean.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"proof"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKValue.Variation.ContactVariations","name":"OAI.SKValue.OrderParameter.zero","note":"The zero order parameter.","plain":"The zero order parameter","relation":"defines"}],"lean_source":"statement","markdown":"For each fixed $T<1$ and $N\\ge1$, the heat normalizers $c_j$ are\npositive. The increments $g_j$ have unit second moments and are\northogonal, since their multipliers depend only on preceding Gaussians.\nThe rounded $F_N$ is a measurable spin after an arbitrary choice on its\nnull tie event. Thus Proposition 10.1 applies with\n$K=N$, $z_j=Z_{j+1}$ and $f_j=g_{j+1}$, giving $S_{T,N}\\le P_*$.","paper":"long","role":"derivation","section":"10.4.1 The same test at every mesh size","start_byte":200260},{"anchor":"heat-lower-bound","at":"b200649-200989","end_byte":200989,"heading_id":"the-same-test-at-every-mesh-size","lean":[],"lean_source":"statement","markdown":"First let $N\\to\\infty$ at fixed $T$, using the displayed exact sum,\nand then let $T\\uparrow1$. The integrand is nonnegative and its endpoint\nsingularity has exponent $1/4<1$. Substitution $s=t^2$ yields\n$$\nP_*\\ge\\frac2\\pi\\int_0^1(1-t^2)^{-1/4}\\,dt\n =\\frac1\\pi B\\!\\left(\\frac12,\\frac34\\right)\n =\\frac{4\\Gamma(3/4)}{\\sqrt\\pi\\,\\Gamma(1/4)}.\n$$","note":"These exact sums and this Beta integral have no Lean counterpart.","paper":"long","role":"explanation","section":"10.4.1 The same test at every mesh size","start_byte":200649},{"anchor":"heat-lower-bound","at":"b200990-201197","end_byte":201197,"heading_id":"the-same-test-at-every-mesh-size","lean":[{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"The theorem has no IsMinimizer hypothesis.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"part"}],"lean_source":"statement","markdown":"The bound follows from the heat calculation and the general Gaussian\ncomparison. Minimality is unnecessary for the inequality. It is what made\nthe different, optimizing construction converge to $P_*$ itself.","paper":"long","role":"explanation","section":"10.4.1 The same test at every mesh size","start_byte":200990}]},"scalar.source_spin_sequence_limit":{"lean":[{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"Stated for the SKValueG sequence, which the scalar proof identifies with the SKValue sequence by rfl.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"states"}],"short":"Finite-spin value converges","statements":[{"anchor":"model","at":"b3346-3501","end_byte":3501,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":95,"line":92,"lines":[92,95],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.hamiltonian","note":"Scalar model: H_n, summed over i<j. The coupling array is indexed by all pairs; only i<j enter.","plain":"SK Hamiltonian on n spins","relation":"defines"},{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateSequence","note":"Scalar model: n⁻¹ E max H_n.","plain":"Expected maximum energy per spin","relation":"defines"},{"end_line":100,"line":100,"lines":[100,100],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateValue","note":"Scalar model: P_* as limUnder of that sequence, which is meaningful once the limit exists.","plain":"The SK ground state value","relation":"defines"},{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"The limit exists. Stated for the SKValueG sequence, which the scalar proof identifies with the SKValue sequence by rfl.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"part"}],"lean_source":"statement","markdown":"H_n(\\sigma)=\\frac1{\\sqrt n}\\sum_{i<j}J_{ij}\\sigma_i\\sigma_j,\\qquad\n\\sigma\\in\\{-1,1\\}^n,\\qquad\nP_*=\\lim_{n\\to\\infty}\\frac1n\\mathbb E\\max_\\sigma H_n(\\sigma).","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":3346},{"anchor":"formal-lower-overview","at":"b183517-183668","end_byte":183668,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"Stated for the SKValueG sequence, which the scalar proof identifies with the SKValue sequence by rfl.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"states"}],"lean_source":"statement","markdown":"The scalar implementation obtains the value equality by two inequalities.\nIts finite-spin limit follows from superadditivity and a uniform upper\nbound.","paper":"long","role":"explanation","section":"10.2 A universal bound for predictable correlations","start_byte":183517},{"anchor":"formal-scalar","at":"b244002-244072","end_byte":244072,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"Stated for the SKValueG sequence, which the scalar proof identifies with the SKValue sequence by rfl.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"states"}],"lean_source":"statement","markdown":"The\nconvergence clause establishes existence of the finite-spin limit.","paper":"long","role":"explanation","section":"12.2 The separate scalar statement","start_byte":244002},{"anchor":"spins-and-overlap","at":"b4586-4801","end_byte":4801,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":101,"line":101,"lines":[101,101],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.groundStateValue","note":"Support model. Written with limUnder, defined even with no limit. Convergence is proved only for the scalar model's own copy.","plain":"The SK ground state value","relation":"defines"},{"end_line":99,"line":98,"lines":[98,99],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.expectedSKMaximum","note":"Support model. Expected maximum divided by n.","plain":"Expected maximum energy per spin","relation":"defines"},{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"Scalar model: convergence of its own sequence, by the same formula. No Lean theorem links it to expectedSKMaximum.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"states"}],"lean_source":"statement","markdown":"For many spins we seek the limiting expected\nmaximum **per spin**,\n\n$$\nP_*:=\\lim_{n\\to\\infty}\\frac1n\\mathbb E\\max_\\sigma H_n(\\sigma),\n$$\n\nwhere the expectation averages over the couplings after the maximum is\ntaken.","paper":"expository","role":"definition","section":"1. The finite optimisation problem","start_byte":4586},{"anchor":"dependencies","at":"b16786-16931","end_byte":16931,"heading_id":"prior-results-and-their-roles","lean":[],"lean_source":"statement","markdown":"The cited theorem is stated for the almost-sure normalized maximum. Its expected-value formulation used here follows from Gaussian concentration.","note":"Not in the Lean: it defines the value through expectations and proves the limit by subadditivity, with no almost-sure theorem and no concentration step.","paper":"long","role":"mention","section":"2.3 Prior results and their roles","start_byte":16786},{"anchor":"formal-lower-overview","at":"b184129-184417","end_byte":184417,"heading_id":"a-universal-bound-for-predictable-correlations","lean":[{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.groundStateSequence","note":"e_n.","plain":"Expected maximum energy per spin","relation":"defines"},{"end_line":263,"line":237,"lines":[237,238],"module":"OAI.Probability.SKValue.GroundState.BlockCoeff","name":"OAI.SKValueG.groundStateSequence_tendsto","note":"The limit.","plain":"Finite-spin value converges","proof_lines":[239,263],"relation":"states"},{"end_line":141,"line":129,"lines":[129,130],"module":"OAI.Probability.SKValue.GroundState.WeakGuerra","name":"OAI.SKValue.groundStateValue_le_parisi","note":"The upper comparison.","plain":"Ground state below Parisi","proof_lines":[131,141],"relation":"part"}],"lean_source":"statement","markdown":"Write $e_n=n^{-1}\\mathbb E\\max_\\sigma H_n(\\sigma)$, with the Hamiltonian\nof §1. The finite-spin convergence and upper comparison are separate\ninputs to this subsection, and their precise suppliers are\n[the superadditive limit][scalar-spin-limit] and\n[the upper comparison][scalar-upper].","paper":"long","role":"mention","section":"10.2 A universal bound for predictable correlations","start_byte":184129}]},"scalar.source_strip_regularity":{"lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"Holds for every Brownian space and coefficient, with no minimiser or diffusion premise.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"states"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"Lists the bounds: value and gradient strips, curvature bound, joint Lipschitz bounds, with constants chosen per cutoff T.","plain":"The strip regularity bounds","relation":"defines"}],"short":"Concrete strip regularity","statements":[{"anchor":"formal-scalar","at":"b243747-243924","end_byte":243924,"heading_id":"the-separate-scalar-statement","lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"Holds for every Brownian space and coefficient, with no minimiser or diffusion premise.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"states"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"Lists the bounds: value and gradient strips, curvature bound, joint Lipschitz bounds, with constants chosen per cutoff T.","plain":"The strip regularity bounds","relation":"defines"}],"lean_source":"statement","markdown":"The implementation derives compact-strip bounds\nfrom `sourceStripRegularity` and the unit curvature moment from contact\nfor those objects before applying its fixed-strip helper.","paper":"long","role":"mention","section":"12.2 The separate scalar statement","start_byte":243747},{"anchor":"dependencies","at":"b10756-11549","end_byte":11549,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":42,"line":22,"lines":[22,22],"module":"OAI.Probability.SKValue.Control.Regularity","name":"OAI.SKValue.sourceStripRegularity","note":"Scalar model, every γ: bounds through the third gradient derivative on each strip. All orders, stability and the terminal estimate are absent.","plain":"Strip bounds hold for every γ","proof_lines":[23,42],"relation":"part"},{"end_line":37,"line":28,"lines":[28,37],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.SourceStripRegularity","note":"Lists the bounds: value and gradient strips, curvature bound, joint Lipschitz bounds, with constants chosen per cutoff T.","plain":"The strip regularity bounds","relation":"defines"}],"lean_source":"statement","markdown":"**Proposition 2.1 (compact-strip regularity).** For $\\gamma\\in\\mathcal U$\nand $0\\le T<1$, the value has bounded, jointly continuous spatial derivatives\nof every positive finite order on $[0,T]\\times\\mathbb R$. If admissible\ncoefficients converge in $L^1$ and their smooth, even, convex, one-Lipschitz\nterminal values converge uniformly to $|x|$, these spatial derivatives\nconverge uniformly on each such strip. The bounds depend on the derivative\norder, the fixed strip and a common coefficient mass bound. Time derivatives need only exist\nalmost everywhere across coefficient jumps. On an open constant interval,\nthe heat formula supplies the additional time smoothness used in §4.\nThe terminal estimate is\n$$\n|x|\\le\\Phi_\\gamma(t,x)\n\\le |x|+\\sqrt{2(1-t)/\\pi}+\\frac12\\int_t^1\\gamma(s)\\,ds.\n$$","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":10756}]},"scalar.source_strip_value_lower":{"lean":[{"end_line":34,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Takes gradient contact as premise; minimality supplies contact at the application.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"states"},{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"The bound on each shifted sum for N>0 and positive raw moments, before the mesh limit.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"proof"}],"short":"Strip integral below ground state","statements":[{"anchor":"gaussian-expectation-limit","at":"b193363-193555","end_byte":193555,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":34,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Takes gradient contact as premise; minimality supplies contact at the application.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"states"},{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"The bound on each shifted sum for N>0 and positive raw moments, before the mesh limit.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"proof"}],"lean_source":"statement","markdown":"At a minimizer, the separate contact and diffusion arguments establish\nthe fixed-strip inputs and eventual positivity. First take the mesh\nlimit to obtain $I(T)\\le P_*$, then let $T\\uparrow1$.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193363},{"anchor":"gaussian-expectation-limit","at":"b193114-193306","end_byte":193306,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":27,"line":14,"lines":[14,17],"module":"OAI.Probability.SKValue.GroundState.LiteralLower","name":"OAI.SKValue.shiftedCoefficientSum_le_groundState","note":"Any order parameter, 0≤T<1, N>0 and positive raw second moments, with no minimiser or diffusion.","plain":"Shifted coefficient sum below ground state","proof_lines":[18,27],"relation":"states"}],"lean_source":"statement","markdown":"It follows that\n$$\nS_{T,N}=\\sum_{j=1}^N A_jB_{j+1}\\le P_*.\n$$\nThis finite inequality assumes admissibility, $0\\le T<1$, $N>0$ and\npositive active second moments, but no minimizer or diffusion.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193114},{"anchor":"gaussian-expectation-limit","at":"b193482-193555","end_byte":193555,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":34,"line":21,"lines":[29,34],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"The mesh limit, via le_of_tendsto.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"states"}],"lean_source":"statement","markdown":"First take the mesh\nlimit to obtain $I(T)\\le P_*$, then let $T\\uparrow1$.","paper":"long","role":"derivation","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193482}]},"scalar.source_terminal_parisi_assembly":{"lean":[{"end_line":79,"line":37,"lines":[37,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"Premises are strip regularity, gradient contact and unit curvature moment. Strip regularity always holds; a minimiser supplies contact, which gives the moment.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"states"}],"short":"Terminal formulas at Parisi value","statements":[{"anchor":"gaussian-expectation-limit","at":"b193556-193638","end_byte":193638,"heading_id":"passing-the-expectation-bound-and-returning-to-the-value","lean":[{"end_line":79,"line":37,"lines":[51,52],"module":"OAI.Probability.SKValue.Evolution.IntegralPrimitiveTerminal","name":"OAI.SKValue.IsDiffusion.parisi_value_assembly","note":"Its last clause: the curvature integral tends to parisi W γ. Premises are strip regularity, gradient contact and unit curvature moment.","plain":"Terminal formulas at the Parisi value","proof_lines":[53,79],"relation":"contains"}],"lean_source":"statement","markdown":"The scalar\nterminal assembly identifies that last limit with $\\mathcal P(\\gamma)$.","paper":"long","role":"mention","section":"10.2.4 Passing the expectation bound and returning to the value","start_byte":193556},{"anchor":"value","at":"b84584-84864","end_byte":84864,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":69,"line":42,"lines":[42,50],"module":"OAI.Probability.SKValue.Evolution.TimeWeightedIntervalIntegrable","name":"OAI.SKValue.IsDiffusion.parisi_first_value","note":"The first value formula, given the moment identity.","plain":"Parisi value from E|X₁|","proof_lines":[51,69],"relation":"proof"},{"end_line":47,"line":36,"lines":[36,39],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","note":"The identification with P_*, proved from the Lean's own upper and lower bounds.","plain":"Ground state equals Parisi, given contact","proof_lines":[40,47],"relation":"proof"}],"lean_source":"statement","markdown":"The terminal limit first gives\n$\\Phi(0,0)=\\mathbb E|X_1|-\\tfrac12\\int_0^1s\\gamma(s)ds$.\nSubtracting the second half from the definition of the functional yields\n$\\mathcal P(\\gamma)=\\mathbb E|X_1|-\\int_0^1s\\gamma(s)\\,ds$.\nThe ground-state Parisi formula identifies this with $P_*$.","paper":"long","role":"derivation","section":"6. Recovering the value from a martingale","start_byte":84584}]},"scalar.terminal_loss":{"lean":[{"clause":"(∫ t in (0 : ℝ)..T,∫ ω,curvature W γ t (X t ω) ∂W.μ)≤groundStateValue","end_line":34,"line":21,"lines":[25,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Only the left inequality, assuming contact. The bound by 1-T has no Lean statement.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"}],"short":"Terminal error bound","statements":[{"anchor":"value","at":"b85549-86004","end_byte":86004,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"clause":"(∫ t in (0 : ℝ)..T,∫ ω,curvature W γ t (X t ω) ∂W.μ)≤groundStateValue","end_line":34,"line":21,"lines":[25,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Only the left inequality, assuming contact. The bound by 1-T has no Lean statement.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"}],"lean_source":"statement","markdown":"The prediction error and the energy tail have different sizes:\n$$\n\\|U_1-u(T,X_T)\\|_2^2=1-T,\n$$\n$$\nP_*-\\int_0^T\\mathbb E a(t,X_t)\\,dt\n=\\mathbb E[(U_1-u(T,X_T))(B_1-B_T)]\\in[0,1-T].\n$$\nThe first identity is orthogonal projection onto $\\mathcal F_T$.\nMartingale orthogonality removes the past Brownian component in the\nsecond. Cauchy–Schwarz bounds the product by $1-T$, and convexity\nmakes it nonnegative, so no uniform terminal curvature bound is needed.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":85549},{"anchor":"gaussian","at":"b98469-98617","end_byte":98617,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":34,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Only the lower inequality; the bound by 1−T is not in Lean.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"}],"lean_source":"statement","markdown":"The QAOA scalar input also uses the tail estimate\n$0\\le P_*-\\int_0^T\\mathbb E a(t,X_t)\\,dt\\le1-T$, from nonnegative\ncurvature and $\\mathbb E a^2=1$.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":98469},{"anchor":"scalar-value","at":"b37904-38033","end_byte":38033,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"clause":"(∫ t in (0 : ℝ)..T,∫ ω,curvature W γ t (X t ω) ∂W.μ)≤groundStateValue","end_line":34,"line":21,"lines":[25,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Only the left inequality, assuming contact. The bound by 1-T has no Lean statement.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"}],"lean_source":"statement","markdown":"Moreover, convexity and\nCauchy–Schwarz give $0\\leq\\mathbb E a\\leq1$, so\n$$\n0\\leq P_*-\\int_0^T\\mathbb E[a(t,X_t)]\\,dt\\leq1-T.\n$$","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":37904},{"anchor":"scalar-value","at":"b37904-37974","end_byte":37974,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":85,"line":73,"lines":[77,80],"module":"OAI.Probability.SKValue.Control.StripRegularity","name":"OAI.SKValue.IsDiffusion.curvature_mean_integrable","note":"The bound |E a|≤1 by Cauchy–Schwarz and E a²=1, kept inside this proof.","plain":"Mean curvature is integrable in time","proof_lines":[77,85],"relation":"proof"}],"lean_source":"statement","markdown":"Moreover, convexity and\nCauchy–Schwarz give $0\\leq\\mathbb E a\\leq1$,","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":37904},{"anchor":"scalar-value","at":"b37975-38033","end_byte":38033,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":34,"line":21,"lines":[21,25],"module":"OAI.Probability.SKValue.Variation.ContactAssembly","name":"OAI.SKValue.IsDiffusion.strip_value_le_groundState","note":"Only the left inequality is stated in the Lean, and it comes from the finite formula.","plain":"Strip integral below ground state","proof_lines":[26,34],"relation":"part"},{"end_line":161,"line":155,"lines":[160,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"The Lean states the limit T↑1, without the rate 1−T.","plain":"The finite-Gaussian conclusion","relation":"part"}],"lean_source":"statement","markdown":"so\n$$\n0\\leq P_*-\\int_0^T\\mathbb E[a(t,X_t)]\\,dt\\leq1-T.\n$$","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":37975},{"anchor":"scalar-value","at":"b38035-38110","end_byte":38110,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"T is fixed in the hypotheses before the limit in N.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"part"}],"lean_source":"statement","markdown":"We can therefore fix the terminal cutoff before discretising the\ndiffusion.","paper":"short","role":"explanation","section":"5.1 A martingale formula for the ground-state value","start_byte":38035}]},"scalar.unshifted_gaussian_limit":{"lean":[],"short":"Unshifted sum has same limit","statements":[{"anchor":"gaussian","at":"b97531-98193","end_byte":98193,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[],"lean_source":"record","markdown":"The same vector estimates also give an unshifted sum with this limit:\n$$\n\\left|\\sum_{j=1}^N A_jB_j-\\sum_{j=1}^N\\alpha_j\\beta_j\\right|\n\\le(1+C_T)e_N\\longrightarrow0,\n$$\n$$\n\\sum_{j=1}^N A_jB_j\n=\\Delta\\sum_{j=1}^N\n  \\frac{\\mathbb E a(t_{j-1},Y_{j-1})}{c_j}+o(1)\n\\longrightarrow\\int_0^T\\mathbb E a(t,X_t)\\,dt.\n$$\nThe latter is a left Riemann sum, since $c_j\\to1$ uniformly and the Euler\ncoupling replaces $Y_{j-1}$ by $X_{t_{j-1}}$. Equal continuum limits\ntherefore leave open which finite expression the circuit needs. The rooted\nenergy calculation in §8.1 selects the adjacent pairing, and the two\nfinite pairings can differ, as [§10.4](#one-step-example) shows.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":97531},{"anchor":"qaoa-boundary","at":"b47294-47378","end_byte":47378,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[],"lean_source":"record","markdown":"An\nunshifted pairing can have the same scalar limit without identifying\nthis energy.","paper":"short","role":"mention","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":47294}]},"scalar_value":{"lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Scalar model. Three conjuncts: scalar value, finite Gaussian convergence, and ground-state value equals Parisi value.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Gives positive normalisers, the coefficient-sum limit at fixed T, and the limit as T tends to one.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"short":"Parisi value identities","statements":[{"anchor":"value","at":"b83463-83841","end_byte":83841,"heading_id":"recovering-the-value-from-a-martingale","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Three conjuncts, for the scalar model: scalar value, finite Gaussian convergence, ground-state value equals Parisi. The tail bound by 1−T is absent.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"part"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"defines"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Gives positive normalisers, the coefficient-sum limit at fixed T, and the limit as T tends to one.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Proposition 6.1 (terminal value).** For the minimizer and diffusion of\nTheorem 5.1 there is a spin-valued terminal variable $U_1$ such that\n$u(t,X_t)=\\mathbb E[U_1\\mid\\mathcal F_t]$ and\n$$\nP_*=\\mathbb E[U_1B_1]=\\int_0^1\\mathbb E a(t,X_t)\\,dt.\n\\tag{6.1}\n$$\nIts finite-cutoff tail lies in $[0,1-T]$. Besides the support theorem,\nthe proposition uses the attained Parisi formula.","paper":"long","role":"statement","section":"6. Recovering the value from a martingale","start_byte":83463},{"anchor":"gaussian","at":"b92520-92887","end_byte":92887,"heading_id":"a-finite-gaussian-spin-and-its-adjacent-correlations","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model. All three clauses: positive normalisers, the mesh limit at fixed T, then the limit to the ground-state value.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"states"},{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its second conjunct is this proposition.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"},{"end_line":161,"line":155,"lines":[155,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Gives positive normalisers, the coefficient-sum limit at fixed T, and the limit as T tends to one.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"**Proposition 7.1 (finite Gaussian approximation).** For a fixed minimizer\nand its diffusion, the construction below has positive active normalizers\nfor all sufficiently fine meshes at every fixed $0<T<1$. Its adjacent\ncorrelation sum converges to $\\int_0^T\\mathbb E a(t,X_t)\\,dt$, and this\nintegral tends to $P_*$ as $T\\uparrow1$. The limits are taken in that order.","paper":"long","role":"statement","section":"7. A finite Gaussian spin and its adjacent correlations","start_byte":92520},{"anchor":"scalar-value","at":"b36855-37016","end_byte":37016,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Its first conjunct gives (10) through ScalarValueConclusion. Its third identifies groundStateValue with the Parisi value.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Let $U$ be the almost-sure and\n$L^2$ limit of $M_t$. The value identity is\n$$\nP_*=\\mathbb E[UB_1]\n   =\\int_0^1\\mathbb E[a(t,X_t)]\\,dt.                \\tag{10}\n$$","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":36855},{"anchor":"qaoa-boundary","at":"b43307-43590","end_byte":43590,"heading_id":"a-fixed-qaoa-circuit-before-the-thermodynamic-limit","lean":[{"end_line":49,"line":29,"lines":[40,41],"module":"OAI.Probability.SKValue.Processes.FiniteGaussianFixedStrip","name":"OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","note":"(11): shiftedCoefficientSum tends to the integral, on one strip, given the unit curvature moment.","plain":"Finite-Gaussian limit on a fixed strip","proof_lines":[42,49],"relation":"states"},{"end_line":153,"line":151,"lines":[151,153],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.shiftedCoefficientSum","note":"The sum over j from 0, with A 0=0.","plain":"Adjacent pairing of A and B","relation":"defines"},{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"At a minimiser its second conjunct gives (11), through FiniteGaussianConclusion.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"}],"lean_source":"statement","markdown":"circuit construction\nneeds one finite quantity, the adjacent pairing\n$$\n\\begin{aligned}\nS_{T,N}&:=\\sum_{j=1}^N\n \\mathbb E[F_Ng_j]\\,\\mathbb E[F_NZ_{j+1}],\\\\\n\\lim_{N\\to\\infty}S_{T,N}\n&=I_T:=\\int_0^T\\mathbb E[a(t,X_t)]\\,dt.\n\\end{aligned}                                      \\tag{11}\n$$","paper":"short","role":"statement","section":"5.3 A fixed QAOA circuit before the thermodynamic limit","start_byte":43307},{"anchor":"scalar-identities","at":"b60861-61068","end_byte":61068,"heading_id":"further-reading","lean":[{"end_line":24,"line":16,"lines":[16,23],"module":"OAI.Probability.SKValue.Main","name":"OAI.SKValue.value_consequences","note":"Scalar model. Three conjuncts: scalar value, finite Gaussian convergence, and ground-state value equals Parisi value.","plain":"The scalar-value theorem","proof_lines":[24,24],"relation":"contains"},{"end_line":118,"line":106,"lines":[106,118],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Further martingale arguments recover the ground-state value from the\nterminal slope response. They use the first curvature moment\n$\\mathbb E[a]$, distinct from the second moment $\\mathbb E[a^2]=1$\nused here.","paper":"expository","role":"mention","section":"Further reading","start_byte":60861},{"anchor":"targets","at":"b6206-6354","end_byte":6354,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model, at a minimiser with a diffusion. Its conclusion gives the martingale, U²=1 and the correlation formula.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"end_line":118,"line":106,"lines":[109,116],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The martingale, its limit U with U^2=1, and the correlation of U with B_1.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"The second result concerns the scalar value. A bounded martingale ends\nat a spin-valued random variable whose Brownian correlation represents\n$P_*$.","paper":"long","role":"explanation","section":"1. The SK model and the Parisi variational problem","start_byte":6206},{"anchor":"targets","at":"b6355-6456","end_byte":6456,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":20,"line":17,"lines":[17,19],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.finite_gaussian_target","note":"Scalar model, at a minimiser with a diffusion: the limit at fixed T, then the limit as S rises to one.","plain":"Finite-Gaussian limit at a minimiser","proof_lines":[20,20],"relation":"contains"},{"end_line":161,"line":155,"lines":[158,161],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.FiniteGaussianConclusion","note":"Convergence at fixed T, then the limit as S rises to one.","plain":"The finite-Gaussian conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Finite Gaussian coefficients approximate that correlation at each\nfixed $T<1$, and then $T\\uparrow1$.","paper":"long","role":"explanation","section":"1. The SK model and the Parisi variational problem","start_byte":6355},{"anchor":"overview","at":"b931-1029","end_byte":1029,"heading_id":"a-long-reasoning-record","lean":[{"end_line":137,"line":98,"lines":[98,112],"module":"OAI.Probability.SKValue.Processes.DiffusionGradientCondExp","name":"OAI.SKValue.diffusion_gradient_stopped_martingale","note":"Scalar proof tree, on a strip ending at T≤1: the gradient along the path, stopped at T, is a martingale. The strip data include |u|≤1.","plain":"Stopped gradient is a martingale","proof_lines":[113,137],"relation":"states"},{"end_line":118,"line":106,"lines":[110,111],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Scalar model: the martingale run to time one with terminal variable U, as concluded by scalar_value_target.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Along the auxiliary\ndiffusion, the spatial gradient of the value function is a bounded\nmartingale.","paper":"long","role":"mention","section":"A long reasoning record","start_byte":931},{"anchor":"overview","at":"b1086-1161","end_byte":1161,"heading_id":"a-long-reasoning-record","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model, at a minimiser with a diffusion. Its conclusion includes the correlation of U with B_1.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"end_line":118,"line":106,"lines":[116,116],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"The correlation formula.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"its\nterminal correlation with Brownian motion gives the ground-state value.","paper":"long","role":"mention","section":"A long reasoning record","start_byte":1086},{"anchor":"abstract","at":"b1327-1376","end_byte":1376,"heading_id":"full-support-and-positive-density-in-the-zero-temperature-sk-model","lean":[{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"Scalar model. Assumes a minimiser and a diffusion. The identity is one clause of the conclusion, shown in the next tab.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"},{"clause":"groundStateValue = ∫ t in (0 : ℝ)..1, ∫ ω, curvature W γ t (X t ω) ∂W.μ","end_line":118,"line":106,"lines":[117,117],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Along the way we derive the scalar value\nidentity","paper":"short","role":"mention","section":"Full support and positive density in the zero-temperature SK model","start_byte":1327},{"anchor":"scalar-value","at":"b36855-36907","end_byte":36907,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":118,"line":106,"lines":[109,114],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"U is the terminal value of the martingale, with both limits.","plain":"The scalar-value conclusion","relation":"part"},{"end_line":286,"line":265,"lines":[281,286],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.exists_sequence_terminal","note":"mathlib martingale convergence along times 1−1/(n+1).","plain":"Terminal limit along a sequence","proof_lines":[270,286],"relation":"proof"},{"end_line":391,"line":370,"lines":[370,375],"module":"OAI.Probability.SKValue.Processes.UnitMomentMartingale","name":"OAI.SKValue.UnitMomentMartingale.terminal_mean_square","note":"","plain":"L² convergence to U","proof_lines":[376,391],"relation":"part"}],"lean_source":"statement","markdown":"Let $U$ be the almost-sure and\n$L^2$ limit of $M_t$.","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":36855},{"anchor":"scalar-value","at":"b36908-37016","end_byte":37016,"heading_id":"a-martingale-formula-for-the-ground-state-value","lean":[{"end_line":118,"line":106,"lines":[116,117],"module":"OAI.Probability.SKValue.Model","name":"OAI.SKValue.ScalarValueConclusion","note":"Gives the limit, the martingale, its spin-valued endpoint U, and the value identities.","plain":"The scalar-value conclusion","relation":"part"},{"end_line":15,"line":13,"lines":[13,14],"module":"OAI.Probability.SKValue.GroundState.MainConclusions","name":"OAI.SKValue.scalar_value_target","note":"The theorem: at a minimiser with its diffusion, this conclusion holds. Scalar model.","plain":"Scalar value at a minimiser","proof_lines":[15,15],"relation":"contains"}],"lean_source":"statement","markdown":"The value identity is\n$$\nP_*=\\mathbb E[UB_1]\n   =\\int_0^1\\mathbb E[a(t,X_t)]\\,dt.                \\tag{10}\n$$","paper":"short","role":"statement","section":"5.1 A martingale formula for the ground-state value","start_byte":36908}]},"support.backward_shapes":{"lean":[{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"Within a step on a uniform mesh. Backward signs only: the forward s claims and the quotient extension at zero are not here.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Odd, v>0, w<=0, rH<=0 and r^2 H_x>=0, the last two as Burgers rate and Wronskian.","plain":"The backward shape conditions","relation":"defines"},{"clause":"BackwardShape (fun x => α*r x)","end_line":86,"line":55,"lines":[56,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"Multiplying by a factor in (0,1] preserves the signs.","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"}],"short":"Backward shape signs","statements":[{"anchor":"curvature","at":"b58126-58603","end_byte":58603,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"All seven signs at x>0, with their sources as hypotheses: BackwardShape, forward concavity and the ratio bound.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Odd, v>0, w<=0, rH<=0 and r^2 H_x>=0, the last two as Burgers rate and Wronskian.","plain":"The backward shape conditions","relation":"defines"},{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"Within a step on a uniform mesh. Backward signs only: the forward s claims and the quotient extension at zero are not here.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"}],"lean_source":"statement","markdown":"previous comparisons give\n\n$$\nR,K,v,B,j,L,S_1\\ge0,\\qquad b=B+2v\\ge0,\\qquad Rb^2=w^2,\n$$\n\nbecause $B=-2H$ and $S_1=2rH_x$. Here $s=-\\partial_x\\log g$\nbelongs to the forward heat factor, while $\\partial_x\\log f=r-s$\nis the logarithmic derivative of the diffusion density. The signs have\nseparate sources: $B\\ge0$ comes from $H\\le0$, $S_1\\ge0$ from\n$H_x\\ge0$ on the positive half-line, $j=s\\rho\\ge0$ from the\nrelative-slope comparison, and $L=-rs_{xx}\\ge0$ from forward concavity.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":58126},{"anchor":"curvature","at":"b52047-52499","end_byte":52499,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":80,"line":63,"lines":[63,65],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.second_nonpos","note":"w≤0 on the half-line for the Burgers flow, given the terminal sign. Lean has no quotient H.","plain":"Burgers flow preserves half-line concavity","proof_lines":[66,80],"relation":"proof"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Odd, v>0, w<=0, rH<=0 and r^2 H_x>=0, the last two as Burgers rate and Wronskian.","plain":"The backward shape conditions","relation":"defines"},{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"Within a step on a uniform mesh. Backward signs only: the forward s claims and the quotient extension at zero are not here.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"},{"clause":"BackwardShape (fun x => α*r x)","end_line":86,"line":55,"lines":[56,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"Multiplying by a factor in (0,1] preserves the signs.","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"}],"lean_source":"statement","markdown":"Twice differentiating gives\n$w_\\tau=\\tfrac12w_{xx}+rw_x+3vw$. At the terminal endpoint\n$w=-2cM^2\\operatorname{sech}^2(Mx)\\tanh(Mx)\\le0$ for $x>0$.\nThe boundary value at zero is zero, and the differentiated tails make\nthe far boundary tend uniformly to zero. Comparison applied to $-w$\ngives $w\\le0$ on the positive half-line.\n\nDefine\n\n$$\nH=\\frac{r_\\tau}{r}=\\frac{w}{2r}+v.\n$$\n\nOddness and $r_x(0)=v(0)>0$ make $w/r$, hence $H$, smooth and even\nat zero.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52047},{"anchor":"curvature","at":"b53106-53654","end_byte":53654,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":166,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.wronskian_nonneg","note":"Burgers flow preserves the Wronskian sign, which is r²H_x≥0. Lean has no quotient H.","plain":"Burgers flow preserves the Wronskian sign","proof_lines":[130,166],"relation":"proof"},{"clause":"BackwardShape (fun x => α*r x)","end_line":86,"line":55,"lines":[56,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"Multiplying by a factor in (0,1] preserves the signs.","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Odd, v>0, w<=0, rH<=0 and r^2 H_x>=0, the last two as Burgers rate and Wronskian.","plain":"The backward shape conditions","relation":"defines"},{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"Within a step on a uniform mesh. Backward signs only: the forward s claims and the quotient extension at zero are not here.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"}],"lean_source":"statement","markdown":"For the second comparison, $wH\\ge0$ and\n$(v/r)_x=w/r-v^2/r^2<0$, so the zeroth-order coefficient has upper\nbound $2\\sup v$. On $[\\varepsilon,A]$, the two lateral errors are\n$H_x(\\varepsilon)=O(\\varepsilon)$ and $H_x(A)=o_A(1)$, uniformly\non the closed step. Taking $\\varepsilon\\downarrow0$ and\n$A\\to\\infty$ gives $H_x\\ge0$. At a backward jump\n$r_{\\rm new}=\\alpha r_{\\rm old}$ with $0<\\alpha\\le1$,\n\n$$\nH_{\\rm new}=H_{\\rm old}-(1-\\alpha)v_{\\rm old},\\qquad\n(H_x)_{\\rm new}=(H_x)_{\\rm old}-(1-\\alpha)w_{\\rm old}.\n$$\n\nThus both signs survive every jump.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":53106},{"anchor":"curvature","at":"b21675-21851","end_byte":21851,"heading_id":"the-integrated-estimate","lean":[{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"Within a step on a uniform mesh. Backward signs only: the forward s claims and the quotient extension at zero are not here.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Odd, v>0, w<=0, rH<=0 and r^2 H_x>=0, the last two as Burgers rate and Wronskian.","plain":"The backward shape conditions","relation":"defines"},{"clause":"BackwardShape (fun x => α*r x)","end_line":86,"line":55,"lines":[56,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"Multiplying by a factor in (0,1] preserves the signs.","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"},{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"The bound s_x≥1/t.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"part"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Concavity of s on the half-line.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"part"}],"lean_source":"statement","markdown":"For $s=-\\partial_x\\log g$, the source proves joint shape restrictions\non $x>0$: $r_x>0$ and $s_x\\geq1/t$, both $r$ and $s$ are concave,\nand $H:=w/(2r)+v\\leq0$ is nondecreasing.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":21675},{"anchor":"curvature","at":"b22403-22501","end_byte":22501,"heading_id":"the-integrated-estimate","lean":[{"end_line":104,"line":102,"lines":[102,103],"module":"OAI.Probability.SKSupport.Density.GaussianDecay","name":"OAI.ZeroTemperatureSK.Heat.quotientExtension_continuous","note":"Continuous extensions only, built from dslope; r odd with r_x(0)>0 keeps the denominator nonzero. Smoothness through zero is not stated.","proof_lines":[104,104],"relation":"part"},{"end_line":115,"line":98,"lines":[98,99],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_continuous","note":"Lean uses less: only continuity of the flux at zero, because it integrates on the half-line.","plain":"The certificate flux is continuous","proof_lines":[100,115],"relation":"proof"}],"lean_source":"statement","markdown":"The quotients $w/r$ and $s/r$ extend smoothly through zero because\n$r,s,w$ are odd and $r_x(0)>0$.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":22403},{"anchor":"curvature","at":"b50850-51118","end_byte":51118,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":292,"line":204,"lines":[204,222],"module":"OAI.Probability.SKSupport.Parabolic.ForwardConstruction","name":"OAI.ZeroTemperatureSK.Parabolic.exponential_lower_barrier","note":"The same barrier, with the same exponent, on a compact cylinder.","plain":"Exponential lower barrier","proof_lines":[223,292],"relation":"states"}],"lean_source":"statement","markdown":"The comparison principle used below is also explicit. If\n$U_\\tau-\\tfrac12U_{xx}-bU_x-dU\\ge0$ on a compact cylinder,\n$d\\le C$, the initial values are nonnegative and the lateral values are\nat least $-\\varepsilon$, then\n$U\\ge-\\varepsilon e^{(\\max(C,0)+1)(\\tau-\\tau_0)}$.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":50850},{"anchor":"curvature","at":"b51119-51204","end_byte":51204,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":202,"line":145,"lines":[145,161],"module":"OAI.Probability.SKSupport.Parabolic.ForwardConstruction","name":"OAI.ZeroTemperatureSK.Parabolic.nonnegative_of_parabolic_inequality","note":"The minimum principle with a strictly positive absorption term.","plain":"Parabolic minimum principle on an interval","proof_lines":[162,202],"relation":"proof"}],"lean_source":"statement","markdown":"Multiply by the corresponding decreasing exponential and apply the\nminimum principle.","paper":"long","role":"derivation","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":51119},{"anchor":"curvature","at":"b51205-51449","end_byte":51449,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":117,"line":12,"lines":[12,28],"module":"OAI.Probability.SKSupport.Parabolic.FullLineComparison","name":"OAI.ZeroTemperatureSK.Parabolic.nonnegative_fullLine_bounded","note":"The whole line, for a drift with βx at most K(1+x²).","plain":"Minimum principle on the whole line","proof_lines":[29,117],"relation":"proof"},{"end_line":231,"line":129,"lines":[129,146],"module":"OAI.Probability.SKSupport.Parabolic.FullLineComparison","name":"OAI.ZeroTemperatureSK.Parabolic.nonnegative_halfLine_bounded","note":"The half-line, with a sign at zero.","plain":"Minimum principle on the half-line","proof_lines":[147,231],"relation":"proof"}],"lean_source":"statement","markdown":"Exhaustion is legitimate when the upper bound for\n$d$ is independent of the cylinder and the lateral errors tend to zero.\nThe first-order coefficient need only be bounded on each cylinder, which\npermits the $1/x$ coefficients that appear below.","paper":"long","role":"explanation","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":51205},{"anchor":"curvature","at":"b51602-51900","end_byte":51900,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":162,"line":148,"lines":[148,151],"module":"OAI.Probability.SKSupport.Parabolic.AffineEvolution","name":"OAI.ZeroTemperatureSK.Heat.logSemigroup_curvature_pos","note":"One log-heat step keeps the second derivative positive.","plain":"Heat step preserves strict convexity","proof_lines":[152,162],"relation":"part"},{"end_line":146,"line":136,"lines":[136,138],"module":"OAI.Probability.SKSupport.Parabolic.AffineEvolution","name":"OAI.ZeroTemperatureSK.Heat.deriv_logSemigroup_tilted","note":"The first derivative is a tilted mean, differentiated again for the mean and variance.","plain":"Heat step derivative is tilted mean","proof_lines":[139,146],"relation":"proof"}],"lean_source":"statement","markdown":"We work in backward time $\\tau$ on a single step, where the equation is\n$r_\\tau=\\tfrac12r_{xx}+rr_x$. The logarithmic heat formula preserves\nstrict positivity of $v$: its second spatial derivative is the tilted\nmean of the old second derivative plus the tilted variance of the old\nfirst derivative.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":51602},{"anchor":"curvature","at":"b51901-52012","end_byte":52012,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":64,"line":49,"lines":[49,50],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.softAbs_backward_shape","note":"The terminal shape, needing c at most M.","plain":"Soft |x| has the backward shape","proof_lines":[51,64],"relation":"part"},{"end_line":86,"line":55,"lines":[55,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"The backward jump, with α in (0,1].","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"}],"lean_source":"statement","markdown":"Initially $v=cM\\operatorname{sech}^2(Mx)>0$, and\npositive rescaling at a backward jump preserves this property.","paper":"long","role":"statement","section":"4.3 Backward and forward comparisons, including jumps","start_byte":51901},{"anchor":"curvature","at":"b52013-52045","end_byte":52045,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":122,"line":112,"lines":[112,114],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.pos_halfLine","note":"From oddness and a positive derivative.","plain":"Odd increasing r positive for x>0","proof_lines":[115,122],"relation":"states"}],"lean_source":"statement","markdown":"Consequently $r(x)>0$ for $x>0$.","paper":"long","role":"statement","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52013},{"anchor":"curvature","at":"b52305-52372","end_byte":52372,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":80,"line":63,"lines":[63,65],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.second_nonpos","note":"The sign of the second derivative of r.","plain":"Burgers flow preserves half-line concavity","proof_lines":[66,80],"relation":"proof"}],"lean_source":"statement","markdown":"Comparison applied to $-w$\ngives $w\\le0$ on the positive half-line.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52305},{"anchor":"curvature","at":"b52374-52422","end_byte":52422,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Parabolic.BurgersConstruction","name":"OAI.ZeroTemperatureSK.Heat.burgersRate","note":"Lean keeps r_τ itself, which is rH. It never defines H.","plain":"Burgers rate ½∂ₓ²r + r∂ₓr","relation":"part"}],"lean_source":"statement","markdown":"Define\n\n$$\nH=\\frac{r_\\tau}{r}=\\frac{w}{2r}+v.\n$$","note":"Lean has no quotient H: it compares r_τ and a Wronskian directly.","paper":"long","role":"definition","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52374},{"anchor":"curvature","at":"b52659-53104","end_byte":53104,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":96,"line":82,"lines":[82,84],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.rate_nonpos","note":"The sign of r_τ, the sign of H since r>0.","plain":"Burgers flow preserves rate sign","proof_lines":[85,96],"relation":"proof"},{"end_line":51,"line":13,"lines":[13,22],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.bounded_linear_nonpositive","note":"A half-line comparison with zero boundary value.","plain":"Half-line maximum principle","proof_lines":[23,51],"relation":"proof"}],"lean_source":"statement","markdown":"The terminal signs are\n$H=(cM-M^2)\\operatorname{sech}^2(Mx)\\le0$ and $H_x\\ge0$ on\n$x>0$, using $M\\ge c$. For the first comparison write\n$v/r+r=1/x+\\eta(x)$, where $\\eta$ is smooth and odd near zero.\nThe radial extension $H(|y|)$ to $\\mathbb R^3$ turns\n$\\tfrac12H_{xx}+x^{-1}H_x$ into $\\tfrac12\\Delta H$.\nIt is smooth at the origin, as is the extra drift, and $H$ tends\nuniformly to zero at infinity. Comparison on expanding balls proves\n$H\\le0$.","note":"Lean proves this differently: r_τ vanishes at zero, so a half-line comparison suffices, with no radial extension to R³.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52659},{"anchor":"curvature","at":"b53106-53429","end_byte":53429,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":166,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Backward.BackwardComparison","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.wronskian_nonneg","note":"r times the derivative of r_τ minus r_x r_τ is nonnegative, which is r²H_x.","plain":"Burgers flow preserves the Wronskian sign","proof_lines":[130,166],"relation":"proof"},{"end_line":109,"line":108,"lines":[108,109],"module":"OAI.Probability.SKSupport.Moments.BurgersJets","name":"OAI.ZeroTemperatureSK.Heat.backwardWronskian","note":"That Wronskian.","plain":"The backward Wronskian","relation":"defines"}],"lean_source":"statement","markdown":"For the second comparison, $wH\\ge0$ and\n$(v/r)_x=w/r-v^2/r^2<0$, so the zeroth-order coefficient has upper\nbound $2\\sup v$. On $[\\varepsilon,A]$, the two lateral errors are\n$H_x(\\varepsilon)=O(\\varepsilon)$ and $H_x(A)=o_A(1)$, uniformly\non the closed step. Taking $\\varepsilon\\downarrow0$ and\n$A\\to\\infty$ gives $H_x\\ge0$.","note":"Lean works with the Wronskian r²H_x; the 1/x drift stays, within the bound βx≤K(1+x²).","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":53106},{"anchor":"curvature","at":"b53430-53654","end_byte":53654,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":29,"line":23,"lines":[23,25],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.scaled_rate","note":"The rate after scaling by α.","plain":"Burgers rate of a rescaled field","proof_lines":[26,29],"relation":"proof"},{"end_line":53,"line":31,"lines":[31,34],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.scaled_wronskian","note":"The Wronskian after scaling by α.","plain":"Wronskian of a rescaled field","proof_lines":[35,53],"relation":"proof"},{"end_line":86,"line":55,"lines":[55,56],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.rescale","note":"All the signs together.","plain":"Rescaling in (0,1] keeps shape","proof_lines":[57,86],"relation":"part"}],"lean_source":"statement","markdown":"At a backward jump\n$r_{\\rm new}=\\alpha r_{\\rm old}$ with $0<\\alpha\\le1$,\n\n$$\nH_{\\rm new}=H_{\\rm old}-(1-\\alpha)v_{\\rm old},\\qquad\n(H_x)_{\\rm new}=(H_x)_{\\rm old}-(1-\\alpha)w_{\\rm old}.\n$$\n\nThus both signs survive every jump.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":53430},{"anchor":"curvature","at":"b52424-52657","end_byte":52657,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":181,"line":173,"lines":[173,177],"module":"OAI.Probability.SKSupport.Moments.BurgersJets","name":"OAI.ZeroTemperatureSK.Heat.SmoothBurgers.wronskian_equation","note":"The equation Lean uses instead, for r²H_x; it keeps a v/r coefficient.","plain":"Parabolic equation for the backward Wronskian","proof_lines":[178,181],"relation":"proof"}],"lean_source":"statement","markdown":"Oddness and $r_x(0)=v(0)>0$ make $w/r$, hence $H$, smooth and even\nat zero. Direct differentiation gives\n\n$$\nH_\\tau=\\tfrac12H_{xx}+(v/r+r)H_x+vH,\n$$\n$$\n(H_x)_\\tau=\\tfrac12(H_x)_{xx}+(v/r+r)(H_x)_x\n             +((v/r)_x+2v)H_x+wH.\n$$","note":"Lean states equations for r_τ and the Wronskian instead of these.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":52424},{"anchor":"curvature","at":"b18464-18616","end_byte":18616,"heading_id":"the-integrated-estimate","lean":[{"end_line":114,"line":113,"lines":[113,114],"module":"OAI.Probability.SKValue.Equations.CoupledComparison","name":"OAI.SKValue.spatialJet","note":"Scalar model. spatialJet c ψ n is c times the n-th derivative of ψ′, so jets 0,1,2,3 are r,v,w,z.","plain":"Scaled spatial jet c ∂ₓⁿψ′","relation":"defines"},{"end_line":35,"line":27,"lines":[27,29],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.coercivePolynomial_scaled","note":"The support model keeps a,b,d and passes to the rescaled polynomial here.","plain":"Coercive integrand in terms of jets","proof_lines":[30,35],"relation":"proof"}],"lean_source":"statement","markdown":"On a plateau with $\\gamma=c>0$, put $r=c\\Phi_x=c u$,\n$v=r_x=c a$, $w=r_{xx}$ and $z=r_{xxx}$. This rescaling makes\nthe derivative identities polynomial.","paper":"short","role":"definition","section":"3. The integrated estimate","start_byte":18464},{"anchor":"curvature","at":"b18617-18660","end_byte":18660,"heading_id":"the-integrated-estimate","lean":[{"end_line":16,"line":12,"lines":[15,16],"module":"OAI.Probability.SKSupport.Parabolic.ReverseBurgers","name":"OAI.ZeroTemperatureSK.Heat.ReverseBurgers","note":"Time derivative of every jet is minus the jet of burgersRate.","plain":"Reverse-time Burgers solution on [0,H]","relation":"part"},{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Parabolic.BurgersConstruction","name":"OAI.ZeroTemperatureSK.Heat.burgersRate","note":"","plain":"Burgers rate ½∂ₓ²r + r∂ₓr","relation":"defines"}],"lean_source":"statement","markdown":"The PDE becomes\n$r_t=-\\tfrac12r_{xx}-rr_x$,","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":18617},{"anchor":"curvature","at":"b19294-19392","end_byte":19392,"heading_id":"the-integrated-estimate","lean":[{"end_line":192,"line":190,"lines":[190,191],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.curvature_nonneg","note":"The sign v≥0, here for a. That it fails to control the mixed term is a remark outside the Lean.","plain":"The curvature is nonnegative","proof_lines":[192,192],"relation":"part"}],"lean_source":"statement","markdown":"The negative mixed term is the obstacle, and convexity, which gives\n$v\\geq0$, does not control it.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":19294},{"anchor":"curvature","at":"b21745-21851","end_byte":21851,"heading_id":"the-integrated-estimate","lean":[{"clause":"BackwardShape (fun x => (c i:ℝ)*deriv (finiteValue c h (softAbs M) (N+1) i t) x)","end_line":130,"line":103,"lines":[106,106],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","note":"r at finite-step data, on each step: BackwardShape bundles r_x>0, concavity, H≤0, H_x≥0. The later rows state the conditions on s.","plain":"Backward shape within each finite step","proof_lines":[107,130],"relation":"part"},{"end_line":17,"line":12,"lines":[12,17],"module":"OAI.Probability.SKSupport.Backward.BackwardRescale","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape","note":"Oddness, r′>0, concavity on x≥0, rate≤0 and Wronskian≥0. With H=rate/r, these last two are H≤0, H_x≥0.","plain":"The backward shape conditions","relation":"defines"},{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"The bound s_x≥1/t.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"part"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Concavity of s on the half-line.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"part"}],"lean_source":"statement","markdown":"on $x>0$: $r_x>0$ and $s_x\\geq1/t$, both $r$ and $s$ are concave,\nand $H:=w/(2r)+v\\leq0$ is nondecreasing.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":21745}]},"support.contact":{"lean":[{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"},{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"}],"short":"Potential vanishes on support","statements":[{"anchor":"contact","at":"b31587-31777","end_byte":31777,"heading_id":"from-minimality-to-contact","lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"}],"lean_source":"statement","markdown":"**Proposition 3.1 (variational contact).** For an admissible minimizer,\nthe potential $G(s)=\\int_s^1(q(t)-t)\\,dt$ is nonnegative on $[0,1)$\nand vanishes on $S=\\operatorname{supp}\\mu_\\gamma$.","paper":"long","role":"statement","section":"3. From minimality to contact","start_byte":31587},{"anchor":"contact","at":"b38737-39198","end_byte":39198,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"},{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"},{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"Since $G\\ge0$, its integral on every compact $K$ is zero. The compact\nintervals $[0,T]$ exhaust $[0,1)$, so $G=0$ $\\mu$-almost everywhere.\nIf a support point $s$ had $G(s)>0$, continuity would give a relative\nneighbourhood on which $G$ stays positive. After restricting to a compact\ninterval containing that neighbourhood, the defining property of support\nwould give it positive $\\mu$-mass, contradicting the zero integral.\nTherefore\n$$\nG(s)=0\\qquad(s\\in S).\n$$","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":38737},{"anchor":"contact","at":"b11477-11607","end_byte":11607,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"G.","plain":"The potential G","relation":"defines"}],"lean_source":"statement","markdown":"first-order cost is half of a tail integral, and minimality makes the\nintegral nonnegative:\n$$\nG(s):=\\int_s^1(q(t)-t)\\,dt\\geq0.\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":11477},{"anchor":"contact","at":"b11680-12285","end_byte":12285,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"},{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"be removed. Let $\\nu=\\mu_\\gamma|_K$ for a compact\n$K\\subset[0,1)$. For $0\\leq\\varepsilon\\leq1$, the measure\n$\\mu_\\gamma-\\varepsilon\\nu=(1-\\varepsilon\\mathbf1_K)\\mu_\\gamma$\nis nonnegative, and its cumulative function is\n$\\gamma-\\varepsilon\\nu([0,\\cdot])\\in\\mathcal U$. Minimality gives\n$$\n0\\leq D_+\\mathcal P(\\gamma)[-\\nu([0,\\cdot])]\n=-\\tfrac12\\int_K G(s)\\,d\\mu_\\gamma(s).\n$$\nThe bounds $|q-t|\\leq1$ and $\\mu_\\gamma(K)<\\infty$ justify the\nchange in the order of integration. Since $G\\geq0$, it vanishes almost\neverywhere for $\\mu_\\gamma$, and by continuity it vanishes on\n$S=\\operatorname{supp}\\mu_\\gamma$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":11680},{"anchor":"contact-and-gap","at":"b28640-28740","end_byte":28740,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":62,"line":57,"lines":[57,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"Stated for every s in Time, from adding a unit atom at s.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"$G(s)/2$ is the marginal cost of placing mass at $s$, and\nat a minimiser $G(s)\\ge0$ for every $s<1$.","paper":"expository","role":"statement","section":"5.1 Allowed changes and their first-order cost","start_byte":28640},{"anchor":"contact-and-gap","at":"b30765-31075","end_byte":31075,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"},{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"}],"lean_source":"statement","markdown":"Since $G\\ge0$, its integral must be zero. If $G$ were positive at a\nsupport point, continuity would bound it below by a positive number\non a neighbourhood carrying positive mass, contradicting the zero\nintegral. Hence $G=0$ on the support. We call this **contact**: the\nnonnegative function touches zero there.","paper":"expository","role":"statement","section":"5.1 Allowed changes and their first-order cost","start_byte":30765},{"anchor":"contact-and-gap","at":"b29119-29485","end_byte":29485,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"Direct instance at s=1/4 and s=3/4, for this candidate, which Lean never writes.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":71,"line":64,"lines":[64,66],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_remove","note":"Direct instance with ν=μ for these finite weights, giving the reduction inequality by removal instead of scaling. Lean writes no such case.","plain":"Removable mass has ∫G ≤ 0","proof_lines":[67,71],"relation":"part"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"}],"lean_source":"statement","markdown":"Suppose temporarily that this two-weight candidate minimised\n$\\mathcal P$. Addition would give two nonnegative costs,\n$G(1/4)\\ge0$ and $G(3/4)\\ge0$, while the allowed reduction would give\n\n$$\n0\\le-\\frac12\\bigl(2G(1/4)+G(3/4)\\bigr).\n$$\n\nBoth costs must therefore be zero, since their positive-weighted sum is\nnonnegative and the reduction forces it to be nonpositive.","paper":"expository","role":"example","section":"5.1 Allowed changes and their first-order cost","start_byte":29119},{"anchor":"contact-and-gap","at":"b27766-27853","end_byte":27853,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":32,"line":28,"lines":[28,29],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF_difference","note":"The change is the cumulative function of the added measure, positive from s onward.","plain":"Difference after adding a CDF","proof_lines":[30,32],"relation":"proof"}],"lean_source":"statement","markdown":"Adding a small atom at scale $s$ raises the cumulative function at\n*every* later scale.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27766},{"anchor":"contact-and-gap","at":"b27854-27955","end_byte":27955,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"Sum of γ and a finite measure's cumulative function, with every field proved. No normalisation appears.","plain":"Order parameter plus a CDF","relation":"defines"}],"lean_source":"statement","markdown":"The change is always allowed, because the total\nmass is free and no compensating removal is required.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27854},{"anchor":"contact-and-gap","at":"b27956-28053","end_byte":28053,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":29,"line":23,"lines":[23,24],"module":"OAI.Probability.SKSupport.Support.FiniteCDF","name":"OAI.ZeroTemperatureSK.finiteCDF_integral","note":"For a unit atom at s this integral is one exactly when s≤t.","plain":"CDF as an integral of indicators","proof_lines":[25,29],"relation":"proof"}],"lean_source":"statement","markdown":"Its direction is\n$\\eta(t)=\\mathbf1_{[s,1)}(t)$, equal to zero before $s$ and one\nfrom $s$ onward.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27956},{"anchor":"contact-and-gap","at":"b28054-28146","end_byte":28146,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"G, without the factor one half.","plain":"The potential G","relation":"defines"},{"end_line":62,"line":57,"lines":[59,62],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"The chord towards γ+1_[s,1), rewritten by Fubini for the Dirac measure at s.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"Thus\n\n$$\nD\\mathcal P(\\gamma)[\\mathbf1_{[s,1)}]\n=\\frac12\\int_s^1(q(t)-t)\\,dt=:\\frac12G(s).\n$$","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":28054},{"anchor":"contact-and-gap","at":"b28148-28316","end_byte":28316,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":26,"line":21,"lines":[21,22],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"Support model. A direct instance with ν=εδ at 1/2, which Lean never writes. Lean takes a unit atom with a chord step.","plain":"Order parameter plus a CDF","relation":"part"}],"lean_source":"statement","markdown":"For the candidate in Figure 1, adding $\\varepsilon$ mass at $s=1/2$\ncreates a jump of height $\\varepsilon$ there and raises the entire\nremaining curve by $\\varepsilon$.","paper":"expository","role":"example","section":"5.1 Allowed changes and their first-order cost","start_byte":28148},{"anchor":"contact-and-gap","at":"b28742-28910","end_byte":28910,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"With ν the two weights and H=q-t. Lean's lemma is general and states no such case.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"contains"}],"lean_source":"statement","markdown":"For the two weights in Section 2, the same accounting gives\n\n$$\n\\int_0^1\\gamma(t)(q(t)-t)\\,dt\n=2G(1/4)+G(3/4).\n$$\n\nEach weight is charged for its entire remaining tail.","paper":"expository","role":"example","section":"5.1 Allowed changes and their first-order cost","start_byte":28742},{"anchor":"contact-and-gap","at":"b29617-29925","end_byte":29925,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":21,"line":12,"lines":[12,21],"module":"OAI.Probability.SKValue.Variation.OrderVariations","name":"OAI.SKValue.OrderParameter.mix","note":"Scalar model. With η=OrderParameter.zero this makes (1-ε)γ admissible. The Lean never takes that η; its contact proof removes mass.","relation":"part"}],"lean_source":"statement","markdown":"Now fix a general minimiser $\\gamma$ and its own $q$, $G$ and\n$\\mu_\\gamma$. The same argument works with the finite sum replaced by an\nintegral. The direction $\\eta=-\\gamma$ is integrable, and\n$(1-\\varepsilon)\\gamma$ remains admissible for $0\\le\\varepsilon\\le1$,\nscaling all existing mass by the same factor.","note":"The support proof does not scale γ. It removes μ restricted to a compact set, in contactPotential_compact_mass_removal.","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":29617},{"anchor":"contact-and-gap","at":"b30094-30376","end_byte":30376,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":158,"line":117,"lines":[120,121],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"The same exchange, written with this indicator, for a finite measure.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"proof"},{"end_line":29,"line":23,"lines":[23,24],"module":"OAI.Probability.SKSupport.Support.FiniteCDF","name":"OAI.ZeroTemperatureSK.finiteCDF_integral","note":"A weight at r counts at t exactly when r≤t.","plain":"CDF as an integral of indicators","proof_lines":[25,29],"relation":"proof"}],"lean_source":"statement","markdown":"To see the order of integration, note that each weight at $s$\ncontributes for exactly the times $t\\ge s$:\n\n$$\n\\begin{aligned}\n\\int_0^1\\gamma(t)(q(t)-t)\\,dt\n&=\\int_0^1\\int_{[0,1)}\\mathbf1_{\\{s\\le t\\}}(q(t)-t)\\,\\mu_\\gamma(ds)\\,dt\\\\\n&=\\int_{[0,1)}G(s)\\,\\mu_\\gamma(ds).\n\\end{aligned}\n$$","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":30094},{"anchor":"contact-and-gap","at":"b30378-30494","end_byte":30494,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":71,"line":64,"lines":[64,71],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_remove","note":"Lean's version: for finite ν≤μ, the integral of G against ν is at most zero.","plain":"Removable mass has ∫G ≤ 0","proof_lines":[67,71],"relation":"proof"}],"lean_source":"statement","markdown":"Thus the allowed reduction gives\n\n$$\n0\\le D\\mathcal P(\\gamma)[-\\gamma]\n=-\\frac12\\int_{[0,1)}G(s)\\,\\mu_\\gamma(ds).\n$$","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":30378},{"anchor":"contact-and-gap","at":"b30496-30763","end_byte":30763,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[],"lean_source":"statement","markdown":"The exchange of integration order is valid even if the total mass is\ninfinite, because the absolute integral is bounded by the area from\nSection 2:\n\n$$\n\\int_{[0,1)}\\int_s^1|q(t)-t|\\,dt\\,\\mu_\\gamma(ds)\n=\\int_0^1\\gamma(t)|q(t)-t|\\,dt\n\\le\\int_0^1\\gamma(t)\\,dt<\\infty.\n$$","note":"Not in the Lean: it never integrates against infinite mass. μ restricted to a compact set is finite, so finiteCDF_fubini suffices.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":30496},{"anchor":"contact-and-gap","at":"b30765-30806","end_byte":30806,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":228,"line":214,"lines":[219,219],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos","note":"The integral is nonpositive by removal and nonnegative by sign.","plain":"Zero on support from nonpositive integrals","proof_lines":[219,228],"relation":"proof"}],"lean_source":"statement","markdown":"Since $G\\ge0$, its integral must be zero.","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":30765},{"anchor":"contact-and-gap","at":"b30807-30976","end_byte":30976,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":228,"line":214,"lines":[220,228],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos","note":"Same idea, without a lower bound: G vanishes almost everywhere, and {G≠0} would be an open set with positive mass.","plain":"Zero on support from nonpositive integrals","proof_lines":[219,228],"relation":"proof"}],"lean_source":"statement","markdown":"If $G$ were positive at a\nsupport point, continuity would bound it below by a positive number\non a neighbourhood carrying positive mass, contradicting the zero\nintegral.","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":30807},{"anchor":"contact-and-gap","at":"b30977-31004","end_byte":31004,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Applied on the compact set below (t+1)/2, since μ need not be finite.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"}],"lean_source":"statement","markdown":"Hence $G=0$ on the support.","paper":"expository","role":"statement","section":"5.1 Allowed changes and their first-order cost","start_byte":30977},{"anchor":"contact-and-gap","at":"b31248-31286","end_byte":31286,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"(hν : ν ≤ μ)","end_line":72,"line":66,"lines":[67,67],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"Lean's removal has this hypothesis: only mass already in μ can be removed.","plain":"Order parameter minus a dominated CDF","relation":"defines"}],"lean_source":"statement","markdown":"Scaling removes only mass that exists.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":31248},{"anchor":"contact-and-gap","at":"b31287-31602","end_byte":31602,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[],"lean_source":"statement","markdown":"In Figure 1 the supported point\n$1/4$ carries no atom, since intervals around it carry mass while the\npoint itself does not. Subtracting $\\varepsilon\\delta_{1/4}$ would\ncreate a downward jump, and for $0<\\varepsilon<1/4$ the cumulative\nvalues would remain nonnegative while the function stopped being\nnondecreasing.","note":"Not in the Lean: this example. removeCDF_monotone needs ν≤μ, which an atom at 1/4 would break, but no failure is proved.","paper":"expository","role":"example","section":"5.1 Allowed changes and their first-order cost","start_byte":31287},{"anchor":"contact-and-gap","at":"b31603-31766","end_byte":31766,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"∀ t, InSupport μ t → G t = 0","end_line":250,"line":230,"lines":[235,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Contact on the support, one way, with no converse. That support alone forbids removing an atom is the paper's remark, with no Lean statement.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"part"}],"lean_source":"statement","markdown":"Support alone therefore permits no atom removal. What we\nhave proved is contact on the support, and the argument allows $G$ to\nvanish at points that carry no mass.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":31603},{"anchor":"endpoints","at":"b39302-39388","end_byte":39388,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":62,"line":57,"lines":[57,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"The addition test at s=0. The negative value of G(0) is not derived in Lean.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"Hence $G(0)=\\int_0^1(q(t)-t)\\,dt<0$, contrary to the allowed\naddition of mass at zero.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":39302},{"anchor":"contact-and-gap","at":"b28549-28634","end_byte":28634,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"∫ r : Time, (∫ t in (r:ℝ)..1, H t) ∂ν","end_line":158,"line":117,"lines":[119,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"The inner integral starts at r: each unit of mass acts on its own tail.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"contains"}],"lean_source":"statement","markdown":"and\nits formula is a tail integral because the tail is where the coefficient\nchanges.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":28549},{"anchor":"contact-and-gap","at":"b31076-31246","end_byte":31246,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[],"lean_source":"statement","markdown":"The scaling argument is the\nvariational argument of [Chen, Proposition 3.2, equations (3.11)–(3.12)][chen-support],\napplied with the integrable-direction formula above.","note":"Not in the Lean: a citation. The support proof replaces scaling by compact removal, in contactPotential_compact_mass_removal.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":31076},{"anchor":"zero-candidate","at":"b47156-47239","end_byte":47239,"heading_id":"a-descent-direction-for-the-zero-candidate","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"Direct instance: ν the unit mass at zero adds the constant one. The Lean does not single out γ=0.","plain":"Order parameter plus a CDF","relation":"part"}],"lean_source":"statement","markdown":"The constant direction\n$\\eta=1$ is admissible and adds an atom at zero, at the cost","paper":"expository","role":"example","section":"7.1 A descent direction for the zero candidate","start_byte":47156},{"anchor":"source-record","at":"b62072-62221","end_byte":62221,"heading_id":"formal-verification-and-sources","lean":[],"lean_source":"statement","markdown":"These notes reconstruct selected parts of OpenAI's analytic manuscript,\nand the global scaling argument for contact follows Chen's variational\nproof.","note":"Not in the Lean: it has no global scaling argument. Contact comes from adding point masses and removing compact pieces of μ.","paper":"expository","role":"explanation","section":"Formal verification and sources","start_byte":62072},{"anchor":"dependencies","at":"b9458-9607","end_byte":9607,"heading_id":"outline-of-the-proof","lean":[{"end_line":92,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"On the support, q(t)=t and q'(t)≤1, the contact the gap argument uses. The rows after it contain G.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"},{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at every time below one, from adding a point mass.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":78,"line":73,"lines":[73,75],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"The integral of G over a compact set is at most zero.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Generic step: a continuous nonnegative G with those compact integrals vanishes on the support.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"}],"lean_source":"statement","markdown":"**Variational contact.** Minimality, admissible mass addition and compact\nmass removal give $G\\ge0$, its vanishing on support and the contact slopes.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":9458},{"anchor":"contact","at":"b31659-31726","end_byte":31726,"heading_id":"from-minimality-to-contact","lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"Nonnegativity at a minimiser.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"states"},{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"G.","plain":"The potential G","relation":"defines"}],"lean_source":"statement","markdown":"the potential $G(s)=\\int_s^1(q(t)-t)\\,dt$ is nonnegative on $[0,1)$","paper":"long","role":"statement","section":"3. From minimality to contact","start_byte":31659},{"anchor":"contact","at":"b36606-36839","end_byte":36839,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"Fubini for a finite measure against a tail integral. The weighted identity for μ itself is not in the Lean.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"part"}],"lean_source":"statement","markdown":"Write $\\mu=\\mu_\\gamma$ and $S=\\operatorname{supp}\\mu$, with support\nrelative to $[0,1)$. Tonelli's theorem gives the useful identity\n$$\n\\int_0^1\\gamma(t)\\,dt\n =\\int_{[0,1)}\\!\\int_s^1dt\\,\\mu(ds)\n =\\int_{[0,1)}(1-s)\\,\\mu(ds)<\\infty.\n$$","note":"Lean uses Fubini only for finite measures, so the weighted identity for μ itself does not appear.","paper":"long","role":"explanation","section":"3.3 Admissible changes of measure","start_byte":36606},{"anchor":"contact","at":"b37031-37105","end_byte":37105,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"G.","plain":"The potential G","relation":"defines"}],"lean_source":"statement","markdown":"Define the tail potential\n$$\nG(s)=\\int_s^1(q(t)-t)\\,dt\\qquad(0\\le s<1).\n$$","paper":"long","role":"definition","section":"3.3 Admissible changes of measure","start_byte":37031},{"anchor":"contact","at":"b37106-37252","end_byte":37252,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":38,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_continuousOn","note":"Continuity.","plain":"G is continuous on [0,1]","proof_lines":[34,38],"relation":"part"},{"end_line":55,"line":47,"lines":[47,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G′, on the open interval.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"lean_source":"statement","markdown":"It is continuous, satisfies $|G(s)|\\le1-s$, and has\n$G'(s)=s-q(s)$ and $G''(s)=1-q'(s)$ on compact strips, with one-sided\ninterpretations at zero.","note":"The support proof uses neither the bound by 1−s nor G′′.","paper":"long","role":"explanation","section":"3.3 Admissible changes of measure","start_byte":37106},{"anchor":"contact","at":"b37443-37549","end_byte":37549,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"γ plus the cumulative function of ν; here ν is the Dirac mass at s.","plain":"Order parameter plus a CDF","relation":"defines"}],"lean_source":"statement","markdown":"| Add $\\varepsilon\\delta_s$, $0\\le s<1$ | $\\eta(t)=\\mathbf1_{[s,1)}(t)$ | $\\varepsilon\\ge0$ | $G(s)\\ge0$ |","paper":"long","role":"definition","section":"3.3 Admissible changes of measure","start_byte":37443},{"anchor":"contact","at":"b37550-37706","end_byte":37706,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":72,"line":66,"lines":[66,72],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"γ minus the cumulative function of ν, for ν ≤ μ.","plain":"Order parameter minus a dominated CDF","relation":"defines"}],"lean_source":"statement","markdown":"| Remove $\\varepsilon\\nu$, where $\\nu=\\mu|_K$ and $K\\subset[0,1)$ is compact | $\\eta_K(t)=-\\nu([0,t])$ | $0\\le\\varepsilon\\le1$ | $\\int_KG(s)\\,\\mu(ds)\\le0$ |","paper":"long","role":"definition","section":"3.3 Admissible changes of measure","start_byte":37550},{"anchor":"contact","at":"b37708-38048","end_byte":38048,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"The competitor; its fields are the conditions listed in this sentence.","plain":"Order parameter plus a CDF","relation":"defines"}],"lean_source":"statement","markdown":"For atom addition, the new cumulative function is nonnegative,\nnondecreasing and right-continuous. Its integral increases by\n$\\varepsilon(1-s)$, so it remains in $\\mathcal U$. This includes $s=0$:\nan atom there adds a constant to $\\gamma$ on all of $[0,1)$. The point\n$1$ is outside the measure's domain and is not an additional competitor.","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":37708},{"anchor":"contact","at":"b38049-38165","end_byte":38165,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":62,"line":57,"lines":[59,62],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"The variational inequality against γ plus a Dirac cumulative function, then Fubini.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"}],"lean_source":"statement","markdown":"The derivative formula gives\n$D_+\\mathcal P(\\gamma)[\\mathbf1_{[s,1)}]=G(s)/2$, proving\n$G\\ge0$ everywhere below one.","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":38049},{"anchor":"contact","at":"b38167-38483","end_byte":38483,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":78,"line":73,"lines":[76,78],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"ν is the restriction of μ to K, finite because K is compact.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"},{"end_line":79,"line":74,"lines":[74,76],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF_difference","note":"The direction is minus the cumulative function of ν.","plain":"Difference after removing a CDF","proof_lines":[77,79],"relation":"part"}],"lean_source":"statement","markdown":"For removal, choose $T<1$ with $K\\subset[0,T]$. Then\n$\\nu(K)\\le\\gamma(T)<\\infty$. The measure\n$\\mu-\\varepsilon\\nu$ is nonnegative for $0\\le\\varepsilon\\le1$,\nso its cumulative function is nonnegative, nondecreasing and\nright-continuous. It lies between zero and $\\gamma$, hence is integrable,\nand $|\\eta_K|\\le\\nu(K)$.","note":"In Lean the competitor is the removal at full strength, γ minus ν; the chord supplies the small steps.","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":38167},{"anchor":"contact","at":"b38608-38736","end_byte":38736,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":71,"line":64,"lines":[67,71],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_remove","note":"The variational inequality, then Fubini.","plain":"Removable mass has ∫G ≤ 0","proof_lines":[67,71],"relation":"proof"}],"lean_source":"statement","markdown":"It follows that\n$$\n0\\le D_+\\mathcal P(\\gamma)[\\eta_K]\n =-\\frac12\\int_0^1\\nu([0,t])(q(t)-t)\\,dt\n =-\\frac12\\int_KG(s)\\,\\mu(ds).\n$$","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":38608},{"anchor":"contact","at":"b38737-39160","end_byte":39160,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":228,"line":214,"lines":[214,218],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos","note":"A continuous nonnegative function with nonpositive integral vanishes on the support.","plain":"Zero on support from nonpositive integrals","proof_lines":[219,228],"relation":"proof"},{"end_line":250,"line":230,"lines":[236,249],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Restricting to the compact Iic b around the point.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"proof"}],"lean_source":"statement","markdown":"Since $G\\ge0$, its integral on every compact $K$ is zero. The compact\nintervals $[0,T]$ exhaust $[0,1)$, so $G=0$ $\\mu$-almost everywhere.\nIf a support point $s$ had $G(s)>0$, continuity would give a relative\nneighbourhood on which $G$ stays positive. After restricting to a compact\ninterval containing that neighbourhood, the defining property of support\nwould give it positive $\\mu$-mass, contradicting the zero integral.","paper":"long","role":"derivation","section":"3.3 Admissible changes of measure","start_byte":38737},{"anchor":"contact","at":"b39199-39325","end_byte":39325,"heading_id":"admissible-changes-of-measure","lean":[{"end_line":71,"line":64,"lines":[64,66],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_remove","note":"Any finite ν below μ is allowed, so the measure's type never enters.","plain":"Removable mass has ∫G ≤ 0","proof_lines":[67,71],"relation":"proof"}],"lean_source":"statement","markdown":"This step applies to atoms, diffuse mass and singular continuous mass.\nRemoving individual atoms alone would not establish it.","paper":"long","role":"mention","section":"3.3 Admissible changes of measure","start_byte":39199},{"anchor":"contact","at":"b40760-41619","end_byte":41619,"heading_id":"the-reach-of-the-first-order-criterion","lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"Support model. Atom addition gives G≥0, as the sentence says. The scaling step after it is not in the Lean.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":228,"line":214,"lines":[214,218],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos","note":"Support model. A continuous nonnegative function with nonpositive integral vanishes on the support, the last step here. The Lean applies it to compact removal.","plain":"Zero on support from nonpositive integrals","proof_lines":[219,228],"relation":"part"}],"lean_source":"statement","markdown":"There is also a global alternative to compact mass removal. Atom addition\nstill gives $G\\ge0$. Now replace $\\mu$ by $(1-\\varepsilon)\\mu$,\nor equivalently $\\gamma$ by $(1-\\varepsilon)\\gamma$, for\n$0\\le\\varepsilon\\le1$. This preserves admissibility, and its direction\n$-\\gamma$ lies in $L^1$, although it may be unbounded. The first\nvariation in §3.2 therefore applies. The signed interchange is justified by\n$$\n\\int_{[0,1)}\\!\\int_s^1|q(t)-t|\\,dt\\,\\mu(ds)\n =\\int_0^1\\gamma(t)|q(t)-t|\\,dt\n \\le\\int_0^1\\gamma(t)\\,dt<\\infty.\n$$\nConsequently\n$$\n0\\le D_+\\mathcal P(\\gamma)[-\\gamma]\n =-\\frac12\\int_{[0,1)}G(s)\\,\\mu(ds).\n$$\nTogether with $G\\ge0$, this gives $\\int G\\,d\\mu=0$. Continuity\nthen gives $G=0$ on relative support, including zero, as in §3.3.\nFinite weighted mass is sufficient throughout, with no assumption of\nfinite total mass or of a density for $\\mu$.","note":"Lean uses only compact removal; this global scaling proof is not formalised. The steps it shares with compact removal are in the Lean.","paper":"long","role":"explanation","section":"3.5 The reach of the first-order criterion","start_byte":40760},{"anchor":"model","at":"b5374-5436","end_byte":5436,"heading_id":"the-model-and-the-result","lean":[{"end_line":92,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"contains"}],"lean_source":"statement","markdown":"minimality will force $q(t)=t$ on the support of $\\mu_\\gamma$.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":5374},{"anchor":"contact","at":"b11351-11607","end_byte":11607,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":62,"line":57,"lines":[59,62],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"The atom is addCDF (dirac s); Fubini turns the variational inequality into the tail integral.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"proof"},{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"G.","plain":"The potential G","relation":"defines"}],"lean_source":"statement","markdown":"Adding a unit atom at $s$ raises the cumulative function at every\n$t\\geq s$, so its direction is $\\eta=\\mathbf1_{[s,1)}$. Its\nfirst-order cost is half of a tail integral, and minimality makes the\nintegral nonnegative:\n$$\nG(s):=\\int_s^1(q(t)-t)\\,dt\\geq0.\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":11351},{"anchor":"contact","at":"b11608-11664","end_byte":11664,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"(1/2:ℝ)","end_line":168,"line":152,"lines":[154,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"The factor one half in the limit; the Lean inequality for G drops it.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"Thus $G(s)/2$ is the first-order cost of the added atom.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":11608},{"anchor":"contact","at":"b11680-11947","end_byte":11947,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":72,"line":66,"lines":[66,72],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"Lean takes the endpoint γ−ν([0,·]) as one order parameter and moves along the chord to it.","plain":"Order parameter minus a dominated CDF","relation":"defines"},{"end_line":78,"line":73,"lines":[76,78],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"ν is μ restricted to K.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"proof"}],"lean_source":"statement","markdown":"be removed. Let $\\nu=\\mu_\\gamma|_K$ for a compact\n$K\\subset[0,1)$. For $0\\leq\\varepsilon\\leq1$, the measure\n$\\mu_\\gamma-\\varepsilon\\nu=(1-\\varepsilon\\mathbf1_K)\\mu_\\gamma$\nis nonnegative, and its cumulative function is\n$\\gamma-\\varepsilon\\nu([0,\\cdot])\\in\\mathcal U$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":11680},{"anchor":"contact","at":"b11948-12054","end_byte":12054,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":71,"line":64,"lines":[64,71],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_remove","note":"The inequality without the factor one half, for any finite ν≤μ.","plain":"Removable mass has ∫G ≤ 0","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"Minimality gives\n$$\n0\\leq D_+\\mathcal P(\\gamma)[-\\nu([0,\\cdot])]\n=-\\tfrac12\\int_K G(s)\\,d\\mu_\\gamma(s).\n$$","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":11948},{"anchor":"contact","at":"b12055-12105","end_byte":12105,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"part"},{"end_line":30,"line":28,"lines":[28,30],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactIntegrand_integrable","note":"From the bound |q|≤1.","plain":"q(t) − t is integrable","proof_lines":[30,30],"relation":"proof"}],"lean_source":"statement","markdown":"The bounds $|q-t|\\leq1$ and $\\mu_\\gamma(K)<\\infty$","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":12055},{"anchor":"contact","at":"b12154-12285","end_byte":12285,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":228,"line":214,"lines":[214,228],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.continuous_zero_on_support_of_integral_nonpos","note":"","plain":"Zero on support from nonpositive integrals","proof_lines":[219,228],"relation":"part"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"K is taken to be Iic b for b past the support point.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"part"},{"end_line":38,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_continuousOn","note":"","plain":"G is continuous on [0,1]","proof_lines":[34,38],"relation":"part"}],"lean_source":"statement","markdown":"Since $G\\geq0$, it vanishes almost\neverywhere for $\\mu_\\gamma$, and by continuity it vanishes on\n$S=\\operatorname{supp}\\mu_\\gamma$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":12154},{"anchor":"contact","at":"b12286-12422","end_byte":12422,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":78,"line":73,"lines":[76,78],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","note":"ν is μ.restrict K and may be any part of μ.","plain":"∫G dμ ≤ 0 on compacts","proof_lines":[76,78],"relation":"part"},{"end_line":72,"line":66,"lines":[66,67],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"","plain":"Order parameter minus a dominated CDF","relation":"part"}],"lean_source":"statement","markdown":"The variation removes a fraction of\nall the mass in $K$, atoms and diffuse mass alike, and assumes no\ndensity or probability constraint.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":12286}]},"support.contact_derivatives":{"lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"},{"clause":"HasDerivAt (contactPotential W γ) (s-squareMoment W γ (diffusion W γ) s) s","end_line":55,"line":47,"lines":[49,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1. The second derivative of G is not stated as a formula here.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"short":"Contact and slope bounds","statements":[{"anchor":"contact","at":"b39437-40043","end_byte":40043,"heading_id":"contact-values-and-endpoint-slopes","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"},{"clause":"HasDerivAt (contactPotential W γ) (s-squareMoment W γ (diffusion W γ) s) s","end_line":55,"line":47,"lines":[49,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1. The second derivative of G is not stated as a formula here.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"lean_source":"statement","markdown":"At $s\\in S\\cap(0,1)$, the function $G$ has an interior minimum of\nvalue zero. Thus $G'(s)=0$ and $G''(s)\\ge0$, giving\n$$\nq(s)=s,\\qquad q'(s)\\le1.\n$$\nIf $0\\in S$, the argument needs the zero-field initial condition in\naddition to the one-sided minimum. We have $G(0)=0$ and $q(0)=0$, so\n$$\nG(h)=G(0)+\\int_0^h(t-q(t))\\,dt\n     =\\frac12(1-q'(0))h^2+o(h^2).\n$$\nNonnegativity for $h>0$ implies $q'(0)\\le1$. This works whether or not\nzero is an atom. The equality $q(0)=0$ comes from symmetry, rather than\nfrom differentiating a one-sided minimum. If $0\\notin S$, no condition\n$G(0)=0$ has been obtained or used.","paper":"long","role":"statement","section":"3.4 Contact values and endpoint slopes","start_byte":39437},{"anchor":"contact","at":"b12424-12780","end_byte":12780,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"},{"clause":"HasDerivAt (contactPotential W γ) (s-squareMoment W γ (diffusion W γ) s) s","end_line":55,"line":47,"lines":[49,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1. The second derivative of G is not stated as a formula here.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"lean_source":"statement","markdown":"At a support point $s>0$, the function $G$ has a minimum. Hence\n$$\nG'(s)=s-q(s)=0,\\qquad G''(s)=1-q'(s)\\geq0,\n\\quad\\text{so}\\quad q(s)=s,\\quad q'(s)\\leq1.       \\tag{2}\n$$\nIf $0\\in S$, contact gives $G(0)=0$, and zero field gives\n$q(0)=0$ independently. Consequently\n$G(h)=\\tfrac12(1-q'(0))h^2+o(h^2)$, and nonnegativity gives the\nsame slope bound at zero.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":12424},{"anchor":"contact-and-gap","at":"b31811-32369","end_byte":32369,"heading_id":"contact-gives-endpoint-conditions","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"},{"clause":"HasDerivAt (contactPotential W γ) (s-squareMoment W γ (diffusion W γ) s) s","end_line":55,"line":47,"lines":[49,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1. The second derivative of G is not stated as a formula here.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"lean_source":"statement","markdown":"We need information about $q$ at the ends of a possible gap. An\ninterior support point $s$ is a minimum of $G$. Since\n$G'(s)=s-q(s)$ and $G''(s)=1-q'(s)$, ordinary calculus gives\n\n$$\nq(s)=s,\\qquad q'(s)\\le1.\n$$\n\nIf zero belongs to the support, contact and symmetry give\n$G(0)=q(0)=0$. Were $q'(0)>1$, continuity would give $q'(t)>1$\nfor small positive $t$, and integration would then give $q(t)>t$, so\n$G'(t)=t-q(t)<0$. Starting from $G(0)=0$, this would force $G(t)<0$,\na contradiction. Hence $q'(0)\\le1$, and this one-sided argument covers\nan atom at zero.","paper":"expository","role":"statement","section":"5.2 Contact gives endpoint conditions","start_byte":31811},{"anchor":"contact-and-gap","at":"b31872-31922","end_byte":31922,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":260,"line":252,"lines":[255,258],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.value_contact_of_potential_min","note":"G is nonnegative and zero at s, so s is a local minimum.","plain":"G's interior minimum gives contact","proof_lines":[255,260],"relation":"proof"}],"lean_source":"statement","markdown":"An\ninterior support point $s$ is a minimum of $G$.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":31872},{"anchor":"contact-and-gap","at":"b31923-32021","end_byte":32021,"heading_id":"contact-gives-endpoint-conditions","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"Both conclusions at every support point, zero included. Lean uses the first derivative and a one-sided argument, never G''.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"contains"},{"end_line":55,"line":47,"lines":[47,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"},{"end_line":260,"line":252,"lines":[259,260],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.value_contact_of_potential_min","note":"The first derivative vanishes at the minimum, so q(s)=s.","plain":"G's interior minimum gives contact","proof_lines":[255,260],"relation":"proof"},{"end_line":293,"line":262,"lines":[262,268],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"Lean never takes G''. Its one-sided argument gives the slope bound at every support point.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"}],"lean_source":"statement","markdown":"Since\n$G'(s)=s-q(s)$ and $G''(s)=1-q'(s)$, ordinary calculus gives\n\n$$\nq(s)=s,\\qquad q'(s)\\le1.\n$$","note":"Lean has one slope argument for all support points; it does not separate interior points from zero.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":31923},{"anchor":"contact-and-gap","at":"b32023-32095","end_byte":32095,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":313,"line":295,"lines":[307,309],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.contact_conditions","note":"At t=0 Lean takes q(0)=0 from the hypothesis hq0; contact gives G(0)=0.","plain":"Contact conditions on the support","proof_lines":[304,313],"relation":"proof"}],"lean_source":"statement","markdown":"If zero belongs to the support, contact and symmetry give\n$G(0)=q(0)=0$.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":32023},{"anchor":"contact-and-gap","at":"b32096-32230","end_byte":32230,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":293,"line":262,"lines":[269,279],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"Lean uses difference quotients near s, not continuity of q': if d>1 then slope q s t>1 close to s.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"}],"lean_source":"statement","markdown":"Were $q'(0)>1$, continuity would give $q'(t)>1$\nfor small positive $t$, and integration would then give $q(t)>t$, so\n$G'(t)=t-q(t)<0$.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":32096},{"anchor":"contact-and-gap","at":"b32168-32230","end_byte":32230,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":293,"line":262,"lines":[280,290],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"The quotient gives q(t)>t, so G'<0 on the interval and G is strictly decreasing.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"}],"lean_source":"statement","markdown":"and integration would then give $q(t)>t$, so\n$G'(t)=t-q(t)<0$.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":32168},{"anchor":"contact-and-gap","at":"b32231-32298","end_byte":32298,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":293,"line":262,"lines":[291,293],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"G falls below zero, contradicting nonnegativity.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"}],"lean_source":"statement","markdown":"Starting from $G(0)=0$, this would force $G(t)<0$,\na contradiction.","paper":"expository","role":"derivation","section":"5.2 Contact gives endpoint conditions","start_byte":32231},{"anchor":"contact-and-gap","at":"b32299-32369","end_byte":32369,"heading_id":"contact-gives-endpoint-conditions","lean":[{"end_line":313,"line":295,"lines":[312,313],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.contact_conditions","note":"The same one-sided lemma at every support point, zero included.","plain":"Contact conditions on the support","proof_lines":[304,313],"relation":"proof"}],"lean_source":"statement","markdown":"Hence $q'(0)\\le1$, and this one-sided argument covers\nan atom at zero.","paper":"expository","role":"explanation","section":"5.2 Contact gives endpoint conditions","start_byte":32299},{"anchor":"contact","at":"b31778-31855","end_byte":31855,"heading_id":"from-minimality-to-contact","lean":[{"end_line":92,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"q(s)=s and the curvature moment at most one, at each support point including zero.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"}],"lean_source":"statement","markdown":"At every $s\\in S$,\n$q(s)=s$ and $q'(s)\\le1$, with a right derivative at zero.","paper":"long","role":"statement","section":"3. From minimality to contact","start_byte":31778},{"anchor":"contact","at":"b39437-39585","end_byte":39585,"heading_id":"contact-values-and-endpoint-slopes","lean":[{"end_line":260,"line":252,"lines":[252,254],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.value_contact_of_potential_min","note":"G′(s)=0 gives q(s)=s.","plain":"G's interior minimum gives contact","proof_lines":[255,260],"relation":"part"},{"end_line":293,"line":262,"lines":[269,293],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"If q′(s)>1 then G falls just after s, contradicting G≥0.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"part"}],"lean_source":"statement","markdown":"At $s\\in S\\cap(0,1)$, the function $G$ has an interior minimum of\nvalue zero. Thus $G'(s)=0$ and $G''(s)\\ge0$, giving\n$$\nq(s)=s,\\qquad q'(s)\\le1.\n$$","note":"Lean never uses G′′: the slope bound is by contradiction, and needs only a right derivative of q.","paper":"long","role":"derivation","section":"3.4 Contact values and endpoint slopes","start_byte":39437},{"anchor":"contact","at":"b39586-39838","end_byte":39838,"heading_id":"contact-values-and-endpoint-slopes","lean":[{"end_line":313,"line":295,"lines":[307,312],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.contact_conditions","note":"At t=0 it uses q(0)=0 directly, then the same slope lemma.","plain":"Contact conditions on the support","proof_lines":[304,313],"relation":"part"}],"lean_source":"statement","markdown":"If $0\\in S$, the argument needs the zero-field initial condition in\naddition to the one-sided minimum. We have $G(0)=0$ and $q(0)=0$, so\n$$\nG(h)=G(0)+\\int_0^h(t-q(t))\\,dt\n     =\\frac12(1-q'(0))h^2+o(h^2).\n$$\nNonnegativity for $h>0$ implies $q'(0)\\le1$.","note":"Lean treats zero by the same slope contradiction, with no second-order Taylor expansion.","paper":"long","role":"derivation","section":"3.4 Contact values and endpoint slopes","start_byte":39586},{"anchor":"contact","at":"b39882-39978","end_byte":39978,"heading_id":"contact-values-and-endpoint-slopes","lean":[{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"The hypothesis hq0 in contact_conditions.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"proof"}],"lean_source":"statement","markdown":"The equality $q(0)=0$ comes from symmetry, rather than\nfrom differentiating a one-sided minimum.","paper":"long","role":"explanation","section":"3.4 Contact values and endpoint slopes","start_byte":39882},{"anchor":"contact","at":"b40082-40237","end_byte":40237,"heading_id":"contact-values-and-endpoint-slopes","lean":[{"end_line":92,"line":80,"lines":[84,92],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"G≥0, compact removal, q(0)=0 and both derivatives are the inputs to contact_conditions, which gives the conclusion.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"}],"lean_source":"statement","markdown":"We have now proved, without any no-gap assumption,\n$$\nG\\ge0\\text{ on }[0,1),\\qquad G=0\\text{ on }S,\\qquad\nq(s)=s\\text{ and }q'(s)\\le1\\text{ for }s\\in S.\n$$","paper":"long","role":"statement","section":"3.4 Contact values and endpoint slopes","start_byte":40082},{"anchor":"gaps","at":"b71467-71713","end_byte":71713,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":92,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"The contact data.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"part"}],"lean_source":"statement","markdown":"Section 3 has established $G\\ge0$, $G=0$ on the support\n$S=\\operatorname{supp}\\mu_\\gamma$, and the contact conditions\n$q(s)=s$, $q'(s)\\le1$ at support points. At zero, the value identity\ncomes from symmetry, and the slope bound requires $0\\in S$.","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":71467},{"anchor":"main-result","at":"b7831-7924","end_byte":7924,"heading_id":"the-model-and-the-result","lean":[{"end_line":92,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"contains"}],"lean_source":"statement","markdown":"Minimality forces $q$ to agree with time at both endpoints,\nwith endpoint slopes at most one.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":7831},{"anchor":"contact","at":"b12424-12595","end_byte":12595,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"contains"},{"end_line":260,"line":252,"lines":[252,260],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.value_contact_of_potential_min","note":"First-order condition at the local minimum.","plain":"G's interior minimum gives contact","proof_lines":[255,260],"relation":"part"},{"end_line":293,"line":262,"lines":[262,293],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"The Lean proof does not use G″. If the right slope of q exceeded one, G would strictly decrease past s.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"},{"end_line":55,"line":47,"lines":[47,49],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_hasDerivAt","note":"G'(s)=s-q(s) for 0<s<1. The second derivative of G is not stated as a formula here.","plain":"G′(s) = s − q(s)","proof_lines":[50,55],"relation":"part"}],"lean_source":"statement","markdown":"At a support point $s>0$, the function $G$ has a minimum. Hence\n$$\nG'(s)=s-q(s)=0,\\qquad G''(s)=1-q'(s)\\geq0,\n\\quad\\text{so}\\quad q(s)=s,\\quad q'(s)\\leq1.       \\tag{2}\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":12424},{"anchor":"contact","at":"b12596-12662","end_byte":12662,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":313,"line":295,"lines":[307,311],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.contact_conditions","note":"At t=0 the proof uses q(0)=0 directly.","plain":"Contact conditions on the support","proof_lines":[304,313],"relation":"part"},{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"}],"lean_source":"statement","markdown":"If $0\\in S$, contact gives $G(0)=0$, and zero field gives\n$q(0)=0$","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":12596},{"anchor":"contact","at":"b12678-12780","end_byte":12780,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":293,"line":262,"lines":[263,268],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_contact_of_potential_min","note":"The same lemma covers s=0 by the right-slope argument. No Taylor expansion of G appears.","plain":"G's minimum gives slope ≤ 1","proof_lines":[269,293],"relation":"proof"}],"lean_source":"statement","markdown":"Consequently\n$G(h)=\\tfrac12(1-q'(0))h^2+o(h^2)$, and nonnegativity gives the\nsame slope bound at zero.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":12678},{"anchor":"contact","at":"b12781-12849","end_byte":12849,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":92,"line":80,"lines":[80,92],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"At every support point, including zero: q(t)=t and slope at most one.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"states"}],"lean_source":"statement","markdown":"These are the contact conditions of\n[OpenAI (2026a), Lemma 6.2][FS].","paper":"short","role":"mention","section":"2. Contact conditions and the exclusion of gaps","start_byte":12781}]},"support.diffusion":{"lean":[{"clause":"IsDiffusion W γ (diffusion W γ)","end_line":327,"line":309,"lines":[310,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Existence: the constructed process satisfies the diffusion predicate on the given Brownian system.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"clause":"∀ᵐ ω ∂W.law, ∀ t : ℝ≥0, t ≤ 1 → X t ω = diffusion W γ t ω","end_line":63,"line":13,"lines":[15,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Every associated diffusion agrees almost surely with the constructed one for all t up to 1.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"},{"end_line":196,"line":163,"lines":[163,168],"module":"OAI.Probability.SKSupport.Diffusion.DiffusionStability","name":"OAI.ZeroTemperatureSK.positivePath_error","note":"A stability bound between a finite-step path and the diffusion, used in the plateau limit. It states neither existence nor uniqueness.","plain":"Coupling bound for an approximating path","proof_lines":[169,196],"relation":"part"}],"short":"Controlled diffusion exists uniquely","statements":[{"anchor":"strip-recovery","at":"b28580-28940","end_byte":28940,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":196,"line":163,"lines":[163,168],"module":"OAI.Probability.SKSupport.Diffusion.DiffusionStability","name":"OAI.ZeroTemperatureSK.positivePath_error","note":"Coupling bound for one approximating path. Lean has a factor 2, and γ(T)T where the paper has the L¹ norm.","plain":"Coupling bound for an approximating path","proof_lines":[169,196],"relation":"part"}],"lean_source":"statement","markdown":"Finally, couple all feedback diffusions with the same Brownian motion.\nOn $[0,T]$, let $L$ be a common spatial Lipschitz bound for their drifts.\nThe difference of the two integral equations gives, for example,\n$$\n\\sup_{s\\le T}|X_s^n-X_s|\n\\le e^{LT}\\left(\\|\\gamma_n-\\gamma\\|_1\n +\\|\\gamma_n\\|_1\\|u_n-u\\|_{\\infty,[0,T]\\times\\mathbb R}\\right)\n \\longrightarrow0.\n$$","paper":"long","role":"statement","section":"2.4 Recovering the compact-strip estimates","start_byte":28580},{"anchor":"formal","at":"b240436-240657","end_byte":240657,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"One ∀ᵐ ω covers every t in [0,1]: the equation holds through time one outside one null set.","plain":"X is a diffusion for γ","relation":"defines"},{"end_line":88,"line":81,"lines":[86,88],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"The moment clause: every diffusion, each t in Time, which is [0,1).","plain":"The full-support conclusion","relation":"defines"},{"clause":"∀ᵐ ω ∂W.law, ∀ t : ℝ≥0, t ≤ 1 → X t ω = diffusion W γ t ω","end_line":63,"line":13,"lines":[15,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Every associated diffusion agrees almost surely with the constructed one for all t up to 1.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"}],"lean_source":"statement","markdown":"conclusion is universal over every diffusion satisfying the specified\nequation, and covers exactly $0\\le t<1$. The diffusion equation itself\nholds through time one, simultaneously for all those times outside\none null set.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":240436},{"anchor":"model","at":"b5437-5614","end_byte":5614,"heading_id":"the-model-and-the-result","lean":[{"clause":"IsDiffusion W γ (diffusion W γ)","end_line":327,"line":309,"lines":[310,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Existence: the constructed process satisfies the diffusion predicate on the given Brownian system.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"clause":"∀ᵐ ω ∂W.law, ∀ t : ℝ≥0, t ≤ 1 → X t ω = diffusion W γ t ω","end_line":63,"line":13,"lines":[15,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Every associated diffusion agrees almost surely with the constructed one for all t up to 1.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"}],"lean_source":"statement","markdown":"The SDE has a unique strong solution. Unless an endpoint limit is\njustified explicitly, all stochastic calculations below take place on\ncompact time intervals ending before one.","paper":"short","role":"statement","section":"1. The model and the result","start_byte":5437},{"anchor":"curvature","at":"b28308-28423","end_byte":28423,"heading_id":"the-integrated-estimate","lean":[{"end_line":69,"line":56,"lines":[56,58],"module":"OAI.Probability.SKSupport.Moments.PositiveGradientError","name":"OAI.ZeroTemperatureSK.positivePath_uniform_error","note":"For grid approximants, with the same Brownian motion.","plain":"Approximating paths converge uniformly","proof_lines":[59,69],"relation":"part"}],"lean_source":"statement","markdown":"By Lemma 2.2, the diffusions\ncoupled through the same Brownian motion satisfy\n$\\sup_{s\\leq T}|X_s^\\delta-X_s|\\to0$.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":28308},{"anchor":"value-and-diffusion","at":"b21442-21639","end_byte":21639,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":75,"line":68,"lines":[68,75],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsDiffusion","note":"Support model. The integral equation, path continuity and start at zero.","plain":"X is a diffusion for γ","relation":"defines"},{"clause":"IsDiffusion W γ (diffusion W γ)","end_line":327,"line":309,"lines":[310,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Existence: the constructed process satisfies the diffusion predicate on the given Brownian system.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"clause":"∀ᵐ ω ∂W.law, ∀ t : ℝ≥0, t ≤ 1 → X t ω = diffusion W γ t ω","end_line":63,"line":13,"lines":[15,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Every associated diffusion agrees almost surely with the constructed one for all t up to 1.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"}],"lean_source":"statement","markdown":"$$\nX_t=B_t+\\int_0^t\\gamma(s)u(s,X_s)\\,ds,\\qquad X_0=0.\n$$\n\nThis is the **associated diffusion**, whose existence and uniqueness for\na fixed Brownian motion follow from the cited regularity theorem.","paper":"expository","role":"statement","section":"3.3 The response, the state and the averaged square","start_byte":21442},{"anchor":"value-and-diffusion","at":"b21501-21639","end_byte":21639,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":327,"line":309,"lines":[309,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Existence for every coefficient, on the given Brownian system.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Almost-sure uniqueness up to time one.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"}],"lean_source":"statement","markdown":"This is the **associated diffusion**, whose existence and uniqueness for\na fixed Brownian motion follow from the cited regularity theorem.","paper":"expository","role":"statement","section":"3.3 The response, the state and the averaged square","start_byte":21501},{"anchor":"model","at":"b5437-5474","end_byte":5474,"heading_id":"the-model-and-the-result","lean":[{"end_line":327,"line":309,"lines":[309,310],"module":"OAI.Probability.SKSupport.Diffusion.CompactDrift","name":"OAI.ZeroTemperatureSK.diffusion_isDiffusion","note":"Existence.","plain":"The constructed process solves the equation","proof_lines":[311,327],"relation":"part"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"Pathwise uniqueness up to time one.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"part"}],"lean_source":"statement","markdown":"The SDE has a unique strong solution.","paper":"short","role":"statement","section":"1. The model and the result","start_byte":5437},{"anchor":"model","at":"b5475-5614","end_byte":5614,"heading_id":"the-model-and-the-result","lean":[{"end_line":16,"line":15,"lines":[15,16],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ","note":"Strip quantities have a cutoff T in Time, so T<1; the moment identities are proved on these.","plain":"Second moment on a strip","relation":"proof"}],"lean_source":"statement","markdown":"Unless an endpoint limit is\njustified explicitly, all stochastic calculations below take place on\ncompact time intervals ending before one.","paper":"short","role":"mention","section":"1. The model and the result","start_byte":5475}]},"support.empty_support_chord_obstruction":{"lean":[],"short":"Empty support, finite chord","statements":[{"anchor":"gaps","at":"b77135-77789","end_byte":77789,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[],"lean_source":"record","markdown":"The same inputs close the empty-support case. Empty Stieltjes support\nforces $\\gamma=0$ everywhere. The\n[zero-interval argument](#zero-plateau) makes $q$ strictly convex on\n$[0,1)$, whereas the positive-plateau argument above makes $q'$ strictly\nconvex. Here $q(0)=0$ and $0\\le q\\le1$. For $0<t<s<1$, convexity gives\n$$\nq(t)\\le\\frac ts q(s)\\le\\frac ts.\n$$\nLetting $s\\uparrow1$ in this bound gives $q(t)\\le t$. Now fix\n$t<u<1$. Strict convexity on the finite interval $[0,u]$ gives\n$$\nq(t)<\\frac tu q(u)\\le t.\n$$\nThus the bounded continuous integrand $q(t)-t$ is strictly negative on\n$(0,1)$, so $G(0)<0$, contrary to the allowed addition of mass at zero.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":77135},{"anchor":"gap-proof","at":"b15181-15507","end_byte":15507,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":146,"line":123,"lines":[123,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Lean has no separate empty case. No terminal plateau forces support past every point, so the support is nonempty.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"proof"},{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"This case uses contact only at\nthe left endpoint.\n\nIf $S$ is empty, then $\\gamma=0$. Lemma 2 makes $q$ strictly\nconvex, and $q(0)=0$ and $q\\leq1$ give $q(t)<(t/b)q(b)\\leq t/b$ for\n$0<t<b<1$. Letting $b\\uparrow1$ gives $q(t)\\leq t$. Strict convexity\napplied once more with $t<b<1$ now gives $q(t)<(t/b)q(b)\\leq t$, so\n$G(0)<0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":15181},{"anchor":"endpoints","at":"b38798-39388","end_byte":39388,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"**Empty support.** Empty support would mean $\\gamma=0$ everywhere. The\npreceding heat argument then gives $q''>0$ throughout $(0,1)$, so $q$\nitself is strictly convex. Recall that $q(0)=0$ and $q\\le1$. For\n$0<t<s<1$, the chord from zero to $s$ gives\n\n$$\nq(t)\\le\\frac ts q(s)\\le\\frac ts.\n$$\n\nLetting $s$ increase to one gives $q(t)\\le t$, where only the numerical\nbound $t/s$ is taken to a limit. For strictness we choose $t<v<1$ and\ncompare with the chord ending at $v$:\n\n$$\nq(t)<\\frac tv q(v)\\le t.\n$$\n\nHence $G(0)=\\int_0^1(q(t)-t)\\,dt<0$, contrary to the allowed\naddition of mass at zero.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":38798},{"anchor":"endpoints","at":"b38966-38999","end_byte":38999,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"end_line":135,"line":125,"lines":[125,126],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"The bound by one.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"part"}],"lean_source":"statement","markdown":"Recall that $q(0)=0$ and $q\\le1$.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":38966}]},"support.endpoint_chord_area":{"lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Same chord bound, then the mean value theorem gives the contradiction from q(a)=a and q(b)=b. The area identity and integral are not stated.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"short":"Integrated endpoint-chord inequality","statements":[{"anchor":"gaps","at":"b79752-80008","end_byte":80008,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[],"lean_source":"statement","markdown":"With $h=q'$ and $C_{a,b}$ its endpoint chord, the contradiction is the exact area identity\n$$\n(b-a)-\\int_a^b h\n=(b-a)\\left(1-\\frac{h(a)+h(b)}2\\right)\n +\\int_a^b(C_{a,b}-h)>0.\n$$\nThe first term is nonnegative, and strict convexity makes the second\npositive.","note":"Not in the Lean: the area identity. no_gap_of_strictConvex_derivative reaches the same contradiction by the mean value theorem, with no integral.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":79752},{"anchor":"gaps","at":"b79255-79530","end_byte":79530,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Same chord bound, then the mean value theorem gives the contradiction from q(a)=a and q(b)=b. The area identity and integral are not stated.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"If the constant coefficient is $c>0$, Section 4 makes $q'$ strictly\nconvex in the gap. Continuity extends the chord comparison to its\nendpoints, giving\n$$\nq'(t)<\\frac{b-t}{b-a}q'(a)+\\frac{t-a}{b-a}q'(b)\\le1,\n\\qquad a<t<b.\n\\tag{5.5}\n$$\nIntegrating contradicts $q(b)-q(a)=b-a$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79255},{"anchor":"gap-proof","at":"b14076-14317","end_byte":14317,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Same chord bound, then the mean value theorem gives the contradiction from q(a)=a and q(b)=b. The area identity and integral are not stated.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Strict convexity of $q'$ and (2) give\n$$\nq'(t)<\\frac{r-t}{r-\\ell}q'(\\ell)\n       +\\frac{t-\\ell}{r-\\ell}q'(r)\\leq1\n\\qquad(\\ell<t<r).\n$$\nIntegrating contradicts $q(r)-q(\\ell)=r-\\ell$, so the geometric\nobstruction is an area deficit under $q'$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14076},{"anchor":"contact-and-gap","at":"b34295-34650","end_byte":34650,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Lean's step is below the larger endpoint slope, the chord's supremum, phrased as a contradiction from q(a)=a and q(b)=b.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Because $q'''=(q')''$, the function $q'$ is strictly convex: between\ntwo points its graph lies strictly below their joining chord.\nThe moment identity makes $q'$ continuous at the endpoints, even if\n$\\gamma$ jumps there, and higher derivatives are needed only inside the\ngap. Hence\n\n$$\nq'(t)<\\frac{s-t}{s-r}q'(r)+\\frac{t-r}{s-r}q'(s)\\le1\n\\qquad(r<t<s).\n$$","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34295},{"anchor":"contact-and-gap","at":"b33141-33516","end_byte":33516,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[],"lean_source":"statement","markdown":"The first two conditions require area $s-r$ under $q'$ across the gap,\nand a calculus model shows how strict convexity can prevent this. On the\nearlier interval $[1/3,2/3]$, put $x=3t-1$ and consider the slope\n$1-x(1-x)$. Its endpoints equal one, but\n\n$$\n\\int_{1/3}^{2/3}[1-(3t-1)(2-3t)]\\,dt\n=\\frac13\\int_0^1[1-x(1-x)]\\,dx=\\frac5{18}.\n$$\n\nThe missing area is $1/3-5/18=1/18$.","note":"Not in the Lean: a worked calculus model. The general step is no_gap_of_strictConvex_derivative, which states neither this slope nor this area.","paper":"expository","role":"example","section":"5.3 Strict curvature creates an impossible integral","start_byte":33141},{"anchor":"contact-and-gap","at":"b33517-33654","end_byte":33654,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":380,"line":370,"lines":[378,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Lean's strict step is below the larger endpoint slope, not the chord itself; then both slopes are at most one.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"The model is a calculus example\nchosen for its shape, and the next input forces the actual $q'$ below\nits endpoint chord in the same way.","paper":"expository","role":"explanation","section":"5.3 Strict curvature creates an impossible integral","start_byte":33517},{"anchor":"contact-and-gap","at":"b34571-34650","end_byte":34650,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":380,"line":370,"lines":[377,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Lean's first bound is the larger endpoint slope, the chord's supremum, not the chord itself. Then max_le gives the bound by one.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Hence\n\n$$\nq'(t)<\\frac{s-t}{s-r}q'(r)+\\frac{t-r}{s-r}q'(s)\\le1\n\\qquad(r<t<s).\n$$","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34571},{"anchor":"changed-cases","at":"b42143-42325","end_byte":42325,"heading_id":"two-checks-on-the-gap-argument","lean":[],"lean_source":"statement","markdown":"**Question 3: tilt the chord.** Suppose a differentiable function $f$ on $[2,5]$ has\n$f(2)=2$, $f(5)=5$, $f'(2)=1/2$ and $f'(5)=3/2$. If $f'$ is\nstrictly convex, can these data hold?","note":"Not in the Lean: a calculus exercise. Its data break hdb, so no_gap_of_strictConvex_derivative does not apply, and the Lean has no tilted-chord version.","paper":"expository","role":"example","section":"5.5 Two checks on the gap argument","start_byte":42143},{"anchor":"changed-cases","at":"b42326-42384","end_byte":42384,"heading_id":"two-checks-on-the-gap-argument","lean":[],"lean_source":"statement","markdown":"What changes if strict convexity\nis weakened to convexity?","note":"Not in the Lean: a question about one calculus example. no_gap_of_strictConvex_derivative assumes strict convexity; the scalar model's positive_hump_impossible assumes plain convexity and different data.","paper":"expository","role":"explanation","section":"5.5 Two checks on the gap argument","start_byte":42326},{"anchor":"changed-cases","at":"b42385-42482","end_byte":42482,"heading_id":"two-checks-on-the-gap-argument","lean":[],"lean_source":"statement","markdown":"The endpoint chord is now tilted, so a\npointwise comparison with the constant one is unavailable.","note":"Not in the Lean: a remark on this example. Its proof compares q' with one, through the larger endpoint slope.","paper":"expository","role":"explanation","section":"5.5 Two checks on the gap argument","start_byte":42385},{"anchor":"changed-cases","at":"b42483-42621","end_byte":42621,"heading_id":"two-checks-on-the-gap-argument","lean":[],"lean_source":"statement","markdown":"Finally,\nkeep the endpoint values and left slope fixed but change $f'(5)$ to\n$2$. Are these data compatible with strict convexity of $f'$?","note":"Not in the Lean: a calculus exercise. Slope 2 also breaks hdb, so the theorem does not apply.","paper":"expository","role":"example","section":"5.5 Two checks on the gap argument","start_byte":42483},{"anchor":"changed-case-answer","at":"b57505-57589","end_byte":57589,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"**[Question 3](#question-3).** A strictly convex $f'$ lies below its endpoint chord.","note":"Not in the Lean as a step: strict convexity is the hypothesis hconvex, and the proof uses only lt_on_openSegment, too weak for these tilted data.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":57505},{"anchor":"changed-case-answer","at":"b57590-57735","end_byte":57735,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"That chord has mean height $(1/2+3/2)/2=1$, hence integral three.\nTherefore $f(5)-f(2)=\\int_2^5f'(t)\\,dt<3$, contradicting the\nprescribed values.","note":"Not in the Lean: a calculus example. f'(5)=3/2 breaks hdb, so no_gap_of_strictConvex_derivative does not apply.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":57590},{"anchor":"changed-case-answer","at":"b57874-57993","end_byte":57993,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"The argument needs a strict loss of area, and the\ncommon endpoint upper bound of one in Figure 2 is only a convenience.","note":"Not in the Lean: a calculus remark. no_gap_of_strictConvex_derivative uses the common bound through max_le hda hdb; the scalar positive_hump_impossible takes no endpoint slope bound.","paper":"expository","role":"explanation","section":"7.4 Answers to the questions","start_byte":57874},{"anchor":"changed-case-answer","at":"b58066-58449","end_byte":58449,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"With the right slope changed to $2$, the chord has mean $5/4$,\nso a strict loss of area can leave the required mean one. Indeed,\n\n$$\nf(t)=2+\\frac{t-2}{2}+\\frac{(t-2)^3}{18}\n$$\n\nhas the required endpoint values, slopes $1/2$ and $2$, and\n$(f')''=1/3>0$. Thus these changed data are compatible.\nStrict convexity produces the contradiction only when the chord\ncannot supply enough area.","note":"Not in the Lean: a calculus example. f'(5)=2 breaks hdb, so no_gap_of_strictConvex_derivative does not apply.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":58066},{"anchor":"changed-case-answer","at":"b57736-57873","end_byte":57873,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"With non-strict convexity the data are possible:\n$f(t)=2+(t-2)/2+(t-2)^2/6$ has affine derivative and satisfies all\nfour endpoint values.","note":"Not in the Lean: a calculus example. no_gap_of_strictConvex_derivative assumes strict convexity, and no declaration mentions this function.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":57736},{"anchor":"gaps","at":"b79752-79929","end_byte":79929,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[],"lean_source":"statement","markdown":"With $h=q'$ and $C_{a,b}$ its endpoint chord, the contradiction is the exact area identity\n$$\n(b-a)-\\int_a^b h\n=(b-a)\\left(1-\\frac{h(a)+h(b)}2\\right)\n +\\int_a^b(C_{a,b}-h)>0.\n$$","note":"Lean does not state this area identity; it uses the mean value form instead.","paper":"long","role":"explanation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79752},{"anchor":"gap-proof","at":"b14076-14210","end_byte":14210,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":380,"line":370,"lines":[376,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"The chord bound, with max of the endpoint slopes in place of the interpolation.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Strict convexity of $q'$ and (2) give\n$$\nq'(t)<\\frac{r-t}{r-\\ell}q'(\\ell)\n       +\\frac{t-\\ell}{r-\\ell}q'(r)\\leq1\n\\qquad(\\ell<t<r).\n$$","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14076},{"anchor":"gap-proof","at":"b14211-14257","end_byte":14257,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":368,"line":361,"lines":[361,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"Lean uses the mean value theorem, not an integral: some interior slope would equal one.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"Integrating contradicts $q(r)-q(\\ell)=r-\\ell$,","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14211},{"anchor":"curvature","at":"b29093-29182","end_byte":29182,"heading_id":"the-integrated-estimate","lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Endpoint bounds q′≤1 and strict convexity give q′<1 inside, then a contradiction with q(a)=a, q(b)=b. No crossing criterion is used.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"},{"end_line":368,"line":361,"lines":[361,365],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"This lemma takes q′<1 inside the gap as its hypothesis, however it was obtained.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"so the endpoint slope bounds give $q'<1$ inside the gap and\nthe same strict area deficit.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":29093}]},"support.first_variation":{"lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter, a weaker form than the real-parameter right derivative.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"},{"clause":"0 ≤ ∫ s in (0:ℝ)..1, (extend β.val s-extend γ.val s)*(squareMoment W γ (diffusion W γ) s-s)","end_line":175,"line":170,"lines":[172,172],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"At a minimiser the chord integral is nonnegative, for directions beta minus gamma.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"short":"One-sided first variation","statements":[{"anchor":"contact","at":"b36025-36473","end_byte":36473,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter, a weaker form than the real-parameter right derivative.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"},{"clause":"0 ≤ ∫ s in (0:ℝ)..1, (extend β.val s-extend γ.val s)*(squareMoment W γ (diffusion W γ) s-s)","end_line":175,"line":170,"lines":[172,172],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"At a minimiser the chord integral is nonnegative, for directions beta minus gamma.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"lean_source":"statement","markdown":"$|\\eta(t)|$. Dominated convergence, followed by subtraction of the linear\nParisi penalty, now yields\n$$\nD_+\\mathcal P(\\gamma)[\\eta]\n :=\\lim_{\\varepsilon\\downarrow0}\n     \\frac{\\mathcal P(\\gamma+\\varepsilon\\eta)-\\mathcal P(\\gamma)}\n          {\\varepsilon}\n =\\frac12\\int_0^1\\eta(t)(q(t)-t)\\,dt.\n$$\nIf $\\gamma$ is a minimiser, this derivative is nonnegative for every\nfeasible direction. It need not vanish: the minimisation has one-sided\nconstraints.","paper":"long","role":"statement","section":"3.2 The feasible right derivative","start_byte":36025},{"anchor":"contact","at":"b34328-34794","end_byte":34794,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":105,"line":86,"lines":[86,89],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_integral","note":"The identity on [0,T], for any pair β, γ, with β-γ in place of εη. Lean adds up Taylor steps, with no Itô formula.","plain":"Value difference as a mixed integral","proof_lines":[90,105],"relation":"states"},{"end_line":86,"line":79,"lines":[79,86],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.mixedStripDrift","note":"The drift of Y on a strip, as a bounded Lipschitz drift.","plain":"Mixed drift on a strip","relation":"defines"},{"end_line":42,"line":36,"lines":[36,38],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.compactMixedDrift_bound","note":"On a strip Lean bounds the drift by the constant β(T).","plain":"The mixed drift is bounded","proof_lines":[39,42],"relation":"part"}],"lean_source":"statement","markdown":"The choice of diffusion\n$$\ndY_t^\\varepsilon\n =\\frac12\\beta(t)(\\widetilde u+u)(t,Y_t^\\varepsilon)\\,dt+dB_t,\n\\qquad Y_0^\\varepsilon=0,\n$$\ncancels the first-order spatial term when Itô's formula is applied to\n$D(t,Y_t^\\varepsilon)$. Its drift has absolute value at most $\\beta(t)$,\nwhich is integrable. On $[0,T]$ the resulting identity is\n$$\nD(0,0)=\\mathbb E D(T,Y_T^\\varepsilon)\n +\\frac{\\varepsilon}{2}\\int_0^T\n       \\eta(t)\\mathbb E[u(t,Y_t^\\varepsilon)^2]\\,dt.\n$$","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":34328},{"anchor":"formal","at":"b241074-242021","end_byte":242021,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter. The bridge to the ordinary right derivative is not in the Lean.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"},{"clause":"0 ≤ ∫ s in (0:ℝ)..1, (extend β.val s-extend γ.val s)*(squareMoment W γ (diffusion W γ) s-s)","end_line":175,"line":170,"lines":[172,172],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"At a minimiser the chord integral is nonnegative, for directions beta minus gamma.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"lean_source":"statement","markdown":"The manuscript's first variation and its formal counterpart differ in one\nuseful respect. The formal result initially gives a derivative\nalong chord lengths $1/(n+1)$. For this particular control functional,\nthat sampled limit determines the ordinary feasible right derivative.\nWith a control fixed, the terminal absolute value is convex in the\ncoefficient while the penalty is linear. The supremum over the same\ncontrol class, and then subtraction of the linear Parisi penalty,\npreserve convexity.\n\nThe secant slopes therefore vary monotonically as the chord shortens, and\neach intervening chord lies between two successive\nsampled chords and has the same limiting slope. If $\\eta$ is an\nintegrable feasible direction, choose $\\varepsilon_0>0$ so that\n$\\gamma+\\varepsilon_0\\eta$ is admissible and rescale the chord. This\nbridge is an ordinary deduction from the definitions and the sampled\nlimit, and no separately named formal theorem states it.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":241074},{"anchor":"contact","at":"b10692-11220","end_byte":11220,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter, a weaker form than the real-parameter right derivative.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"Minimality means that every allowed one-sided change has nonnegative\nfirst-order cost, and the one-sided variation formula computes that\ncost. When $\\gamma$ is replaced by $\\gamma+\\varepsilon\\eta$ and\n$\\varepsilon\\downarrow0$, the functional changes at the rate\n$$\nD_+\\mathcal P(\\gamma)[\\eta]\n=\\tfrac12\\int_0^1\\eta(t)(q(t)-t)\\,dt,                 \\tag{1}\n$$\nfor every integrable direction $\\eta$ for which\n$\\gamma+\\varepsilon\\eta\\in\\mathcal U$ for all sufficiently small\n$\\varepsilon\\geq0$ (see [OpenAI (2026a), Lemma 6.1][FS]).","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10692},{"anchor":"first-variation","at":"b26686-26972","end_byte":26972,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"(𝓝 ((1/2:ℝ)*(∫ s in (0:ℝ)..1, (extend β.val s-extend γ.val s)*(squareMoment W γ (diffusion W γ) s-s))))","end_line":168,"line":152,"lines":[154,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"The same integral with η=β-γ, as a limit along steps 1/(n+1), weaker than a right derivative. No ε₀ is needed.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"Precisely, let $\\eta\\in L^1(0,1)$, meaning $\\int_0^1|\\eta(t)|\\,dt<\\infty$, and suppose\n$\\gamma+\\varepsilon\\eta$ stays admissible for every\n$0\\le\\varepsilon\\le\\varepsilon_0$, for some $\\varepsilon_0>0$.\nThe formula is\n\n$$\nD\\mathcal P(\\gamma)[\\eta]\n=\\frac12\\int_0^1\\eta(t)(q(t)-t)\\,dt.\n$$","paper":"expository","role":"statement","section":"5.1 Allowed changes and their first-order cost","start_byte":26686},{"anchor":"first-variation","at":"b26974-27273","end_byte":27273,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter, a weaker form than the real-parameter right derivative.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"},{"end_line":126,"line":91,"lines":[91,93],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedValueDifference_limit","note":"The term q/2: the limit of the value quotients, along the perturbed coefficients. The fixed-control derivative does not appear.","plain":"Limit of the normalised value difference","proof_lines":[94,126],"relation":"proof"}],"lean_source":"statement","markdown":"Here $q$ belongs to the unperturbed candidate $\\gamma$.\nThe term $-t/2$ comes from the outer correction in $\\mathcal P$, and\n$q(t)/2$ is the sensitivity of the reoptimised control value. That\nsensitivity is the analytic content of the formula, which goes beyond\ndifferentiating with a fixed control.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":26974},{"anchor":"order-parameter-example","at":"b13408-13747","end_byte":13747,"heading_id":"what-a-gap-means","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"Support model. Direct instance with ν the added atom, which Lean never writes. Adding a finite measure keeps every admissibility field.","plain":"Order parameter plus a CDF","relation":"defines"},{"end_line":72,"line":66,"lines":[66,72],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"Support model. Removal requires ν≤μ, the condition in this question. Lean writes no instance here and has no scaling constructor.","plain":"Order parameter minus a dominated CDF","relation":"defines"}],"lean_source":"statement","markdown":"**Question 1: allowed changes.** Return to the two weights\n$2\\delta_{1/4}+\\delta_{3/4}$. For $0<\\varepsilon<1$, which changes\nleave an admissible cumulative function: adding $\\varepsilon$ mass at\n$1/2$, removing it at $1/4$, removing it at $1/2$, or multiplying\nboth weights by $1-\\varepsilon$? Check monotonicity as well as\nnonnegativity.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":13408},{"anchor":"first-variation","at":"b26352-26482","end_byte":26482,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Lean states no derivative in ε: a limit of difference quotients at steps 1/(n+1) along the chord to β.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"The **first variation** in a direction $\\eta$ is the derivative at\nzero of $\\varepsilon\\mapsto\\mathcal P(\\gamma+\\varepsilon\\eta)$.","paper":"expository","role":"definition","section":"5.1 Allowed changes and their first-order cost","start_byte":26352},{"anchor":"first-variation","at":"b26483-26685","end_byte":26685,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":175,"line":170,"lines":[173,175],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"The limit of nonnegative quotients is nonnegative. Lean has only this one-sided inequality.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"lean_source":"statement","markdown":"At a minimum its right derivative is nonnegative whenever small\n$\\varepsilon\\ge0$ leave the candidate admissible. It need not vanish,\nsince the opposite change may violate nonnegativity or monotonicity.","paper":"expository","role":"statement","section":"5.1 Allowed changes and their first-order cost","start_byte":26483},{"anchor":"first-variation","at":"b26974-27029","end_byte":27029,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"clause":"squareMoment W γ (diffusion W γ) s-s","end_line":168,"line":152,"lines":[154,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"The moment uses γ's own diffusion, with no perturbed coefficient inside.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"Here $q$ belongs to the unperturbed candidate $\\gamma$.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":26974},{"anchor":"first-variation","at":"b27030-27094","end_byte":27094,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":150,"line":143,"lines":[143,145],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedParisiDifference_eq","note":"The Parisi quotient is the value quotient minus this correction integral.","plain":"Normalised Parisi difference formula","proof_lines":[146,150],"relation":"proof"},{"end_line":141,"line":133,"lines":[133,135],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.penalty_difference","note":"The correction is linear in the coefficient.","plain":"Penalty term along a chord","proof_lines":[136,141],"relation":"proof"}],"lean_source":"statement","markdown":"The term $-t/2$ comes from the outer correction in $\\mathcal P$,","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27030},{"anchor":"first-variation","at":"b27099-27273","end_byte":27273,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":126,"line":91,"lines":[91,93],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedValueDifference_limit","note":"The limit of the value quotients.","plain":"Limit of the normalised value difference","proof_lines":[94,126],"relation":"proof"}],"lean_source":"statement","markdown":"$q(t)/2$ is the sensitivity of the reoptimised control value. That\nsensitivity is the analytic content of the formula, which goes beyond\ndifferentiating with a fixed control.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27099},{"anchor":"first-variation","at":"b27274-27418","end_byte":27418,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Lean proves the segment formula itself, as a limit along steps 1/(n+1), rather than importing it.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"},{"end_line":23,"line":12,"lines":[12,13],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.OrderParameter.chord","note":"The segment joining γ and β, shown admissible field by field.","plain":"Chord between two order parameters","relation":"defines"}],"lean_source":"statement","markdown":"[Chen–Handschy–Lerman (2018), Proposition 4][chl-optimality]\ngives the derivative along the segment joining any two admissible\ncoefficients.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27274},{"anchor":"first-variation","at":"b27419-27539","end_byte":27539,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[],"lean_source":"statement","markdown":"Taking the second endpoint to be\n$\\gamma+\\varepsilon_0\\eta$ and rescaling the segment parameter gives\nthe formula above.","note":"Not in the Lean: it does no rescaling. Later uses take β=addCDF or removeCDF of a finite measure directly.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":27419},{"anchor":"first-variation","at":"b27540-27696","end_byte":27696,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":105,"line":86,"lines":[86,89],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_integral","note":"The value difference is a terminal term plus the weighted mixed moment.","plain":"Value difference as a mixed integral","proof_lines":[90,105],"relation":"proof"},{"end_line":119,"line":107,"lines":[107,109],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_terminal_bound","note":"The terminal term is bounded by the L1 tail of the coefficient difference.","plain":"Terminal bound on the mixed difference","proof_lines":[110,119],"relation":"proof"},{"end_line":70,"line":50,"lines":[50,52],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedValueDifference_truncation","note":"The two combined for one chord step and one strip.","plain":"Truncating the normalised value difference","proof_lines":[53,70],"relation":"proof"}],"lean_source":"statement","markdown":"The detailed proof in\n[OpenAI (2026a), Lemma 6.1][support-variation] compares value functions\nand bounds the terminal remainder by the $L^1$ tail of $\\eta$.","paper":"expository","role":"derivation","section":"5.1 Allowed changes and their first-order cost","start_byte":27540},{"anchor":"contact-and-gap","at":"b28317-28490","end_byte":28490,"heading_id":"allowed-changes-and-their-first-order-cost","lean":[{"end_line":32,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixedQ","note":"For the changed problem, the Lean uses γ's gradient along a drift mixing both coefficients, whose limit is mixedQ_perturbation_limit.","plain":"Second moment along the mixed diffusion","relation":"proof"}],"lean_source":"statement","markdown":"The coefficient changes only on\nthat tail, yet the optimised value and feedback must be recomputed,\nsince earlier starting states also face a changed future control\nproblem.","paper":"expository","role":"explanation","section":"5.1 Allowed changes and their first-order cost","start_byte":28317},{"anchor":"zero-candidate","at":"b47417-47497","end_byte":47497,"heading_id":"a-descent-direction-for-the-zero-candidate","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"The chord limit holds for any two order parameters, with no minimality.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"The calculation uses the heat and variation formulas and no\nminimiser hypothesis","paper":"expository","role":"explanation","section":"7.1 A descent direction for the zero candidate","start_byte":47417},{"anchor":"changed-case-answer","at":"b55843-56106","end_byte":56106,"heading_id":"answers-to-the-questions","lean":[{"end_line":26,"line":21,"lines":[21,26],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.addCDF","note":"Direct instance: ν is ε times the unit mass at 1/2, and adding is always admissible. The example's heights are not in the Lean.","plain":"Order parameter plus a CDF","relation":"part"},{"end_line":72,"line":66,"lines":[66,72],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.OrderParameter.removeCDF","note":"Direct instance at 1/4: ν≤μ holds since ε is below the weight two. The example's heights are not in the Lean.","plain":"Order parameter minus a dominated CDF","relation":"part"}],"lean_source":"statement","markdown":"**[Question 1](#question-1).** Adding mass at $1/2$ gives cumulative heights\n$0,2,2+\\varepsilon,3+\\varepsilon$, in their successive intervals;\nremoving it at $1/4$ gives heights $0,2-\\varepsilon,3-\\varepsilon$.\nBoth functions remain nonnegative and nondecreasing.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":55843},{"anchor":"changed-case-answer","at":"b56107-56278","end_byte":56278,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"Removing mass at\n$1/2$ instead lowers the height from $2$ to $2-\\varepsilon$ there,\ncreating a downward jump, so it is forbidden even though all values\nremain nonnegative.","note":"Not in the Lean: this example. removeCDF needs ν≤μ, which fails with no mass at 1/2; the Lean proves nothing about this candidate.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":56107},{"anchor":"changed-case-answer","at":"b56279-56339","end_byte":56339,"heading_id":"answers-to-the-questions","lean":[{"end_line":23,"line":12,"lines":[12,23],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.OrderParameter.chord","note":"Direct instance: the chord towards the zero parameter is this scaling, with admissibility built in. The Lean never takes this chord.","plain":"Chord between two order parameters","relation":"part"}],"lean_source":"statement","markdown":"Scaling both weights preserves all admissibility\nconditions.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":56279},{"anchor":"dependencies","at":"b13158-13322","end_byte":13322,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":175,"line":170,"lines":[170,172],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"IsMinimizer first appears here.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"},{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"No IsMinimizer hypothesis.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"contains"}],"lean_source":"statement","markdown":"Minimality first enters through the variational inequality in §3. The\nconstant-interval curvature estimate applies to any admissible coefficient,\nminimiser or not.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":13158},{"anchor":"approximation-choices","at":"b30282-30518","end_byte":30518,"heading_id":"the-order-of-the-approximation-limits","lean":[{"end_line":70,"line":50,"lines":[50,52],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedValueDifference_truncation","note":"The tail bound, uniform in n.","plain":"Truncating the normalised value difference","proof_lines":[53,70],"relation":"proof"}],"lean_source":"statement","markdown":"**First variation.** Fix $\\gamma$, a feasible $\\eta\\in L^1$ and initially\n$\\epsilon>0$. Remove the terminal cutoff before letting\n$\\epsilon\\downarrow0$. The terminal remainder divided by $\\epsilon$ is\nbounded by an $L^1$ tail of $\\eta$.","note":"Lean reverses this order: it fixes T by the tail, then lets n grow on that strip.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":30282},{"anchor":"contact","at":"b33139-33280","end_byte":33280,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":52,"line":52,"lines":[52,52],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.perturbation","note":"Lean perturbs along a chord to another order parameter, at steps 1/(n+1).","plain":"Chord perturbation of step n","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\eta\\in L^1(0,1)$ and suppose that\n$\\gamma+\\varepsilon\\eta\\in\\mathcal U$ for\n$0\\le\\varepsilon\\le\\varepsilon_0$, where $\\varepsilon_0>0$.","note":"Lean has no separate direction: β−γ plays the role of η, and only steps 1/(n+1) occur.","paper":"long","role":"definition","section":"3.2 The feasible right derivative","start_byte":33139},{"anchor":"contact","at":"b33384-33495","end_byte":33495,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":23,"line":12,"lines":[12,23],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.OrderParameter.chord","note":"The chord between two order parameters is one: every field is proved.","plain":"Chord between two order parameters","relation":"part"}],"lean_source":"statement","markdown":"Since $\\mathcal U$ is convex, it is enough\nto check that the endpoint $\\gamma+\\varepsilon_0\\eta$ is admissible.","paper":"long","role":"explanation","section":"3.2 The feasible right derivative","start_byte":33384},{"anchor":"contact","at":"b34064-34327","end_byte":34327,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":29,"line":16,"lines":[16,19],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.differenceField_generator","note":"The generator of the difference under the mixed drift leaves only this source term.","plain":"Itô generator of the value difference","proof_lines":[20,29],"relation":"proof"},{"end_line":99,"line":83,"lines":[83,99],"module":"OAI.Probability.SKSupport.Control.ValueDifference","name":"OAI.ZeroTemperatureSK.differenceField","note":"The difference D as a time field on a strip.","plain":"Value difference as a time field","relation":"defines"}],"lean_source":"statement","markdown":"For $0<\\varepsilon\\le\\varepsilon_0$, put\n$\\beta=\\gamma+\\varepsilon\\eta$,\n$\\widetilde u=\\Phi_{\\beta,x}$ and $D=\\Phi_\\beta-\\Phi_\\gamma$.\nSubtracting the two PDEs gives\n$$\nD_t=-\\frac12D_{xx}\n    -\\frac12\\beta(\\widetilde u+u)D_x\n    -\\frac{\\varepsilon}{2}\\eta u^2.\n$$","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":34064},{"anchor":"contact","at":"b34328-34500","end_byte":34500,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.compactMixedDrift","note":"The drift of Y.","plain":"β coefficient times the averaged gradient","relation":"defines"},{"end_line":16,"line":15,"lines":[15,16],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.compactMixedGradient","note":"The average of the two gradients.","plain":"Average of the two gradients","relation":"defines"},{"end_line":86,"line":79,"lines":[79,86],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.mixedStripDrift","note":"That drift as a bounded Lipschitz drift, so the equation has a solution.","plain":"Mixed drift on a strip","relation":"defines"}],"lean_source":"statement","markdown":"The choice of diffusion\n$$\ndY_t^\\varepsilon\n =\\frac12\\beta(t)(\\widetilde u+u)(t,Y_t^\\varepsilon)\\,dt+dB_t,\n\\qquad Y_0^\\varepsilon=0,\n$$\ncancels the first-order spatial term","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":34328},{"anchor":"contact","at":"b34559-34628","end_byte":34628,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":42,"line":36,"lines":[36,38],"module":"OAI.Probability.SKSupport.Diffusion.MixedDrift","name":"OAI.ZeroTemperatureSK.compactMixedDrift_bound","note":"On a strip Lean bounds it by the constant β(T).","plain":"The mixed drift is bounded","proof_lines":[39,42],"relation":"part"}],"lean_source":"statement","markdown":"Its drift has absolute value at most $\\beta(t)$,\nwhich is integrable.","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":34559},{"anchor":"contact","at":"b34629-34794","end_byte":34794,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":105,"line":86,"lines":[86,89],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_integral","note":"The identity on [0,T], for any pair β, γ.","plain":"Value difference as a mixed integral","proof_lines":[90,105],"relation":"proof"},{"end_line":32,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixedQ","note":"The expectation of u squared along Y.","plain":"Second moment along the mixed diffusion","relation":"defines"}],"lean_source":"statement","markdown":"On $[0,T]$ the resulting identity is\n$$\nD(0,0)=\\mathbb E D(T,Y_T^\\varepsilon)\n +\\frac{\\varepsilon}{2}\\int_0^T\n       \\eta(t)\\mathbb E[u(t,Y_t^\\varepsilon)^2]\\,dt.\n$$","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":34629},{"anchor":"contact","at":"b34795-35068","end_byte":35068,"heading_id":"the-feasible-right-derivative","lean":[],"lean_source":"statement","markdown":"At coefficient jumps the time derivatives need only exist almost\neverywhere. The source obtains this identity by approximating both\ncoefficients by monotone step functions with the same smooth terminal\ndatum, and then passing to the limit using the compact-strip estimates.","note":"Not in the Lean: it does not approximate the coefficients here. Right derivatives on the true strip give the identity directly.","paper":"long","role":"explanation","section":"3.2 The feasible right derivative","start_byte":34795},{"anchor":"contact","at":"b35070-35273","end_byte":35273,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":92,"line":85,"lines":[85,87],"module":"OAI.Probability.SKSupport.Foundations.GradientAbs","name":"OAI.ZeroTemperatureSK.value_coefficient_stability","note":"The time-local bound.","plain":"Value is stable in the coefficient","proof_lines":[88,92],"relation":"proof"},{"end_line":119,"line":107,"lines":[107,109],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_terminal_bound","note":"Applied to the terminal term.","plain":"Terminal bound on the mixed difference","proof_lines":[110,119],"relation":"proof"}],"lean_source":"statement","markdown":"The control representation, applied with the same control on each side,\ngives the time-local difference bound\n$$\n\\sup_x|D(T,x)|\\le\\frac{3\\varepsilon}{2}\n                         \\int_T^1|\\eta(t)|\\,dt.\n$$","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":35070},{"anchor":"contact","at":"b35274-35595","end_byte":35595,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":70,"line":50,"lines":[50,52],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedValueDifference_truncation","note":"The finite-T form with its tail error.","plain":"Truncating the normalised value difference","proof_lines":[53,70],"relation":"proof"}],"lean_source":"statement","markdown":"Hence $T\\uparrow1$, at fixed $\\varepsilon$, removes the terminal term\neven after division by $\\varepsilon$. We obtain the exact finite-change\nidentity\n$$\n\\frac{\\Phi_{\\gamma+\\varepsilon\\eta}(0,0)-\\Phi_\\gamma(0,0)}\n     {\\varepsilon}\n =\\frac12\\int_0^1\\eta(t)\n                       \\mathbb E[u(t,Y_t^\\varepsilon)^2]\\,dt.\n$$","note":"Lean never takes T to one at fixed ε, so this exact identity does not appear.","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":35274},{"anchor":"contact","at":"b35596-35933","end_byte":35933,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":61,"line":58,"lines":[58,59],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.perturbation_distance_limit","note":"L1 convergence of the coefficients.","plain":"Perturbations converge to γ","proof_lines":[60,61],"relation":"part"},{"end_line":63,"line":35,"lines":[35,40],"module":"OAI.Probability.SKSupport.Diffusion.MixedPathStability","name":"OAI.ZeroTemperatureSK.mixedPath_error","note":"The path bound under the shared Brownian motion.","plain":"Coupling bound for the mixed path","proof_lines":[41,63],"relation":"part"}],"lean_source":"statement","markdown":"As $\\varepsilon\\downarrow0$, the coefficients converge in $L^1$ and have\nuniformly bounded $L^1$ norms. On each fixed strip, their gradients converge\nuniformly and have common spatial Lipschitz bounds. Coupling with the same\nBrownian motion and applying Gronwall's inequality therefore gives\n$Y^\\varepsilon\\to X$ uniformly on that strip.","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":35596},{"anchor":"contact","at":"b35934-36037","end_byte":36037,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":112,"line":87,"lines":[87,90],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.mixedQ_perturbation_limit","note":"Convergence of the moment, at each time in the strip.","plain":"Mixed second moment converges along chords","proof_lines":[91,112],"relation":"part"}],"lean_source":"statement","markdown":"The integrand converges to\n$\\eta(t)q(t)$ for each $t<1$, and its absolute value is at most\n$|\\eta(t)|$.","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":35934},{"anchor":"contact","at":"b36038-36125","end_byte":36125,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":48,"line":35,"lines":[35,48],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.mixedQ_weighted_limit","note":"Dominated convergence on the strip.","plain":"Weighted mixed moment converges","proof_lines":[38,48],"relation":"proof"},{"end_line":150,"line":143,"lines":[143,145],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.normalizedParisiDifference_eq","note":"The subtraction of the penalty.","plain":"Normalised Parisi difference formula","proof_lines":[146,150],"relation":"proof"}],"lean_source":"statement","markdown":"Dominated convergence, followed by subtraction of the linear\nParisi penalty, now yields","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":36038},{"anchor":"contact","at":"b36321-36409","end_byte":36409,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":175,"line":170,"lines":[170,175],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"The limit of nonnegative quotients.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"lean_source":"statement","markdown":"If $\\gamma$ is a minimiser, this derivative is nonnegative for every\nfeasible direction.","paper":"long","role":"derivation","section":"3.2 The feasible right derivative","start_byte":36321},{"anchor":"contact","at":"b33661-33827","end_byte":33827,"heading_id":"the-feasible-right-derivative","lean":[{"end_line":23,"line":12,"lines":[12,13],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.OrderParameter.chord","note":"The same chord; ε here is θ.","plain":"Chord between two order parameters","relation":"defines"}],"lean_source":"statement","markdown":"Their chord\n$\\gamma_\\theta=(1-\\theta)\\gamma+\\theta\\gamma_1$ gives this direction\nby taking $\\gamma_1=\\gamma+\\varepsilon_0\\eta$ and\n$\\theta=\\varepsilon/\\varepsilon_0$.","paper":"long","role":"explanation","section":"3.2 The feasible right derivative","start_byte":33661},{"anchor":"model","at":"b5301-5369","end_byte":5369,"heading_id":"the-model-and-the-result","lean":[{"clause":"(squareMoment W γ (diffusion W γ) s-s)","end_line":168,"line":152,"lines":[154,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Limit along sampled chord steps towards another order parameter, a weaker form than the real-parameter right derivative.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"part"}],"lean_source":"statement","markdown":"The discrepancy $q(t)-t$ enters the first variation of $\\mathcal P$,","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":5301},{"anchor":"contact","at":"b10692-10834","end_byte":10834,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":175,"line":170,"lines":[170,172],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.minimizer_variational_inequality","note":"Stated for the direction β−γ towards any order parameter β.","plain":"First-order inequality at a minimiser","proof_lines":[173,175],"relation":"part"}],"lean_source":"statement","markdown":"Minimality means that every allowed one-sided change has nonnegative\nfirst-order cost, and the one-sided variation formula computes that\ncost.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":10692},{"anchor":"contact","at":"b10779-11049","end_byte":11049,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"Lean proves this along chords to another order parameter, as a limit over steps 1/(n+1). That suffices for both variations below.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"proof"}],"lean_source":"statement","markdown":"and the one-sided variation formula computes that\ncost. When $\\gamma$ is replaced by $\\gamma+\\varepsilon\\eta$ and\n$\\varepsilon\\downarrow0$, the functional changes at the rate\n$$\nD_+\\mathcal P(\\gamma)[\\eta]\n=\\tfrac12\\int_0^1\\eta(t)(q(t)-t)\\,dt,                 \\tag{1}\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10779},{"anchor":"contact","at":"b11050-11181","end_byte":11181,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":52,"line":52,"lines":[52,52],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.perturbation","note":"The Lean directions are η=β−γ with β admissible, so the chord stays admissible for a whole sequence of steps.","plain":"Chord perturbation of step n","relation":"proof"}],"lean_source":"statement","markdown":"for every integrable direction $\\eta$ for which\n$\\gamma+\\varepsilon\\eta\\in\\mathcal U$ for all sufficiently small\n$\\varepsilon\\geq0$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":11050},{"anchor":"model","at":"b4856-5018","end_byte":5018,"heading_id":"the-model-and-the-result","lean":[{"end_line":168,"line":152,"lines":[152,154],"module":"OAI.Probability.SKSupport.Support.VariationIntegrals","name":"OAI.ZeroTemperatureSK.parisi_chord_derivative","note":"The first-order cost is an integral of q(s)−s.","plain":"One-sided derivative along a chord","proof_lines":[155,168],"relation":"proof"}],"lean_source":"statement","markdown":"To turn minimality of $\\mathcal P$ into a condition on $\\gamma$, we use\na random process whose gradient moment measures the first-order cost of\nchanging $\\gamma$.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":4856}]},"support.fixed_time_spatial_coercivity":{"lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c squared, so its type needs no c>0. Positivity is needed to divide. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"short":"Spatial coercivity at fixed time","statements":[{"anchor":"fixed-time-coercivity","at":"b63811-64368","end_byte":64368,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c squared, so its type needs no c>0. Positivity is needed to divide. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"**Proposition 4.1 (coercivity at a fixed time).** Let $\\gamma\\in\\mathcal U$\nand let $X$ be its associated diffusion, started at zero. Suppose that\n$0<t<1$, $\\gamma$ is continuous at $t$, and $c=\\gamma(t)>0$. With\n$a=\\Phi_{xx}$, $b=\\Phi_{xxx}$ and $d=\\Phi_{xxxx}$, evaluated at\n$(t,X_t)$, one has\n$$\n\\mathcal C_\\gamma(t):=\\mathbb E[d^2-12cab^2+6c^2a^4]\n \\ge \\frac{c^2}{1000}\\mathbb E a^4.\n\\tag{4.1}\n$$\nThe proposition assumes neither minimality nor a constant neighbourhood\nof $t$, and it is the common estimate behind plateau curvature and\npositive density.","paper":"long","role":"statement","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":63811},{"anchor":"fixed-time-coercivity","at":"b64845-65556","end_byte":65556,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":158,"line":154,"lines":[154,155],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveGamma_L1","note":"Support model. Step coefficients tending to γ in L¹, built as grid sample plus mesh: no plateau, no clipping, no bound (4.3).","plain":"Positive step functions converge in L¹","proof_lines":[156,158],"relation":"proof"},{"end_line":72,"line":44,"lines":[44,45],"module":"OAI.Probability.SKValue.Gaussian.SplitConvergence","name":"OAI.SKValue.OrderParameter.splitError_tendsto","note":"Scalar model. The same L¹ limit for a grid cut at t, with no plateau there either.","proof_lines":[46,72],"relation":"proof"}],"lean_source":"statement","markdown":"**Proof.** Fix $t<T<T'<1$ and a shrinking interval\n$[t-\\delta_n,t+\\delta_n)\\subset(0,T)$. Approximate $\\gamma$ by\n$$\nh_n=\\max\\{\\epsilon_n,\\min\\{K_n,\n                  \\eta_n\\lfloor\\gamma/\\eta_n\\rfloor\\}\\},\n$$\nwhere $0<\\epsilon_n<c<K_n$, $\\epsilon_n,\\eta_n\\downarrow0$ and\n$K_n\\uparrow\\infty$. Use $\\min(c,h_n)$ to the left of the interval,\n$c$ on it, and $\\max(c,h_n)$ to its right. Monotonicity makes these\npositive, nondecreasing, right-continuous finite-step coefficients\n$\\gamma_n$. Clipping toward $c$ outside the interval cannot increase\nthe error, and\n$$\n\\|\\gamma_n-\\gamma\\|_1\n\\le\\epsilon_n+\\eta_n+\\int_0^1(\\gamma-K_n)_+\\,ds\n +2\\delta_n\\sup_{|s-t|\\le\\delta_n}|\\gamma(s)-c|\n \\longrightarrow0.\n\\tag{4.3}\n$$","note":"The Lean's approximant is simpler, so this construction is not there. The L¹ limit is.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64845},{"anchor":"formal","at":"b242709-243043","end_byte":243043,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c squared, so its type needs no c>0. Positivity is needed to divide. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"The [fixed-time supplier `actual_coercivity`][support-fixed-time] assumes\npositive time and continuity of the coefficient at that time. Its scaled\npolynomial inequality is (4.1) multiplied by $c^2$, and division requires\n$c>0$. This gives a direct source destination for Proposition 4.1 without\nadding a minimizing or plateau premise.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":242709},{"anchor":"curvature","at":"b17385-17739","end_byte":17739,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c squared, so its type needs no c>0. Positivity is needed to divide. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"**Proposition 3 (fixed-time spatial-moment estimate).** Let\n$\\gamma\\in\\mathcal U$, let $0<t<1$ be a continuity point of $\\gamma$,\nand suppose that $c=\\gamma(t)>0$. For its associated diffusion,\n$$\n\\mathcal C_\\gamma(t)\\geq\\frac{c^2}{1000}\n                       \\mathbb E[a(t,X_t)^4].\n$$\nNeither minimality nor a constant neighbourhood of $t$ is required.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":17385},{"anchor":"curvature","at":"b26960-27820","end_byte":27820,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Lean inserts no plateau, so γ_δ and (6) do not appear. It proves the fixed-time bound through grid approximants, scaled by c².","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"**Passage to a fixed time.** Now let $\\gamma\\in\\mathcal U$ be continuous\nat $t\\in(0,1)$, with $c=\\gamma(t)>0$. For\n$0<\\delta<\\min(t,1-t)$ define\n$$\n\\gamma_\\delta(s)=\n\\begin{cases}\nc,&t-\\delta\\leq s<t+\\delta,\\\\\n\\gamma(s),&\\text{otherwise}.\n\\end{cases}\n$$\nThe inequalities $\\gamma(t-\\delta)\\leq c\\leq\\gamma(t+\\delta)$\nmake both joins nondecreasing, so $\\gamma_\\delta\\in\\mathcal U$, although\nit need not minimise $\\mathcal P$. Moreover,\n$$\n\\|\\gamma_\\delta-\\gamma\\|_1\n\\leq2\\delta\\,[\\gamma(t+\\delta)-\\gamma(t-\\delta)]\n\\longrightarrow0.\n$$\nWe use a subscript $\\delta$ for its value-function derivatives and\ndiffusion. Lemma 2 and (4), already extended to integrable\ncoefficients, give at the fixed time $t$\n$$\n\\mathcal C_{\\gamma_\\delta}(t)\n=\\mathbb E[d_\\delta^2-12c\\,a_\\delta b_\\delta^2+6c^2a_\\delta^4]\n\\geq\\frac{c^2}{1000}\\mathbb E[a_\\delta^4].           \\tag{6}\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26960},{"anchor":"curvature-transfer","at":"b54606-54654","end_byte":54654,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":143,"line":135,"lines":[135,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Dominated convergence on both sides, then le_of_tendsto_of_tendsto.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"passes the quantitative inequality to the limit.","paper":"expository","role":"derivation","section":"7.3 Where the strict-curvature input comes from","start_byte":54606},{"anchor":"curvature-transfer","at":"b55036-55138","end_byte":55138,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":88,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sample_coercivity","note":"By this point the inequality involves value derivatives only, so only those are passed to the limit.","plain":"Coercivity at sampled grid times","proof_lines":[81,88],"relation":"proof"}],"lean_source":"statement","markdown":"The auxiliary logarithmic derivatives\nneed not converge, because integration has already removed them.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":55036},{"anchor":"fixed-time-coercivity","at":"b64019-64211","end_byte":64211,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Both sides still multiplied by c²; the jets are at the constructed diffusion.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"},{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The same with the time-derivative field.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"}],"lean_source":"statement","markdown":"With\n$a=\\Phi_{xx}$, $b=\\Phi_{xxx}$ and $d=\\Phi_{xxxx}$, evaluated at\n$(t,X_t)$, one has\n$$\n\\mathcal C_\\gamma(t):=\\mathbb E[d^2-12cab^2+6c^2a^4]\n \\ge \\frac{c^2}{1000}\\mathbb E a^4.\n\\tag{4.1}\n$$","paper":"long","role":"statement","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64019},{"anchor":"fixed-time-coercivity","at":"b64212-64368","end_byte":64368,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":143,"line":135,"lines":[135,136],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The hypotheses are 0<t and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"The proposition assumes neither minimality nor a constant neighbourhood\nof $t$, and it is the common estimate behind plateau curvature and\npositive density.","paper":"long","role":"explanation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64212},{"anchor":"fixed-time-coercivity","at":"b64370-64714","end_byte":64714,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"clause":"stripThird W γ T c t = ∫ ξ, (actualJet W γ t 4 ξ)^2-\n      12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2+6*c^2*(actualJet W γ t 2 ξ)^4 ∂W.law","end_line":49,"line":36,"lines":[38,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"The right side of (4.2), for c free.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"},{"end_line":46,"line":39,"lines":[39,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Support model: q‴ on a constant interval, where γ′=0.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":52,"line":41,"lines":[41,46],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.curvatureSlope_hasDerivAt","note":"Scalar model: at a continuity point, with c frozen at γ(t), so no γ′ term appears.","plain":"Time derivative of the curvature slope","proof_lines":[47,52],"relation":"proof"}],"lean_source":"statement","markdown":"When $\\gamma$ is smooth near $t$, Itô's formula gives\n$q''=\\mathbb E b^2-2\\gamma\\mathbb E a^3$ and\n$$\n(\\mathbb E b^2)'=\\mathbb E d^2-6\\gamma\\mathbb E(ab^2),\\qquad\n(\\mathbb E a^3)'=3\\mathbb E(ab^2)-3\\gamma\\mathbb E a^4.\n$$\nDifferentiating the first identity therefore gives\n$$\nq'''(t)+2\\gamma'(t)\\mathbb E a^3=\\mathcal C_\\gamma(t).\n\\tag{4.2}\n$$","note":"Lean has no γ′ term, so identity (4.2) is not stated. Both models reach only the frozen-coefficient case.","paper":"long","role":"explanation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64370},{"anchor":"model","at":"b2572-2711","end_byte":2711,"heading_id":"the-model-and-the-result","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The bound is in the Lean and is used in the gap exclusion. The second use, pointwise positivity, is outside it.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"proof"}],"lean_source":"statement","markdown":"One quantitative bound gives both facts:\nit first excludes constant intervals and then, after an approximation,\ngives pointwise positivity.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":2572},{"anchor":"main-result","at":"b6105-6271","end_byte":6271,"heading_id":"the-model-and-the-result","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The only hypotheses are t>0, t≤T and continuity of γ at t.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"proof"}],"lean_source":"statement","markdown":"The feature used twice is a lower bound\nthat holds throughout every positive constant interval of an admissible\ncoefficient, whether or not the coefficient minimises.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":6105},{"anchor":"curvature","at":"b16991-17248","end_byte":17248,"heading_id":"the-integrated-estimate","lean":[{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"The displayed expectation: jets 2,3,4 are a,b,d, and c is free, taken as γ(t).","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"states"},{"end_line":77,"line":76,"lines":[76,77],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.stripThird","note":"Lean's name for C, taken with c=γ(t). It is an expected field on a strip.","plain":"Expression for q‴ on a strip","relation":"defines"},{"end_line":33,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actualJet","note":"Jet m is the m-th spatial derivative of Φ at the diffusion.","plain":"Spatial derivatives along the diffusion","relation":"defines"}],"lean_source":"statement","markdown":"Both arguments use one spatial moment. Recall $a=\\Phi_{xx}$, write\n$b=\\Phi_{xxx}$ and $d=\\Phi_{xxxx}$, and for $0<t<1$ set\n$$\n\\mathcal C_\\gamma(t)\n=\\mathbb E\\bigl[d^2-12\\gamma(t)ab^2+6\\gamma(t)^2a^4\\bigr],\n$$\nwith the spatial factors evaluated at $(t,X_t)$.","paper":"short","role":"definition","section":"3. The integrated estimate","start_byte":16991},{"anchor":"curvature","at":"b17249-17383","end_byte":17383,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The statement involves γ(t) and spatial jets at time t only.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"The definition requires spatial regularity and no time derivative of\n$\\gamma$, so the estimate we need can be stated at a single time.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":17249},{"anchor":"curvature","at":"b17441-17671","end_byte":17671,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Both sides have an extra factor c²; divide to get this form.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"},{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The same bound written through stripThird.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":33,"line":12,"lines":[12,17],"module":"OAI.Probability.SKValue.Coercivity.LiteralCoercivity","name":"OAI.SKValue.IsDiffusion.literal_coercivity","note":"Scalar model, no transport to the support model: every t in (0,1), no continuity hypothesis, c right-continuous. Jets contain c, so divide by c².","plain":"Coercivity along the value-proof diffusion","proof_lines":[18,33],"relation":"states"}],"lean_source":"statement","markdown":"Let\n$\\gamma\\in\\mathcal U$, let $0<t<1$ be a continuity point of $\\gamma$,\nand suppose that $c=\\gamma(t)>0$. For its associated diffusion,\n$$\n\\mathcal C_\\gamma(t)\\geq\\frac{c^2}{1000}\n                       \\mathbb E[a(t,X_t)^4].\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":17441},{"anchor":"curvature","at":"b17672-17739","end_byte":17739,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,136],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The hypotheses are 0<t and ContinuousAt γ.val t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"Neither minimality nor a constant neighbourhood of $t$ is required.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":17672},{"anchor":"curvature","at":"b18051-18168","end_byte":18168,"heading_id":"the-integrated-estimate","lean":[],"lean_source":"statement","markdown":"We first explain the estimate on a positive plateau and then obtain\nthe fixed-time form by inserting a short plateau.","note":"Not in the Lean: it proves the fixed-time bound first, from finite-step approximants, and the plateau bound follows from it.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":18051},{"anchor":"curvature","at":"b18170-18298","end_byte":18298,"heading_id":"the-integrated-estimate","lean":[{"end_line":52,"line":41,"lines":[41,46],"module":"OAI.Probability.SKValue.Coercivity.CurvatureMoment","name":"OAI.SKValue.IsDiffusion.curvatureSlope_hasDerivAt","note":"Scalar model. Lean freezes the coefficient at c=γ(t), so the moment C appears and the term in γ′ does not.","plain":"Time derivative of the curvature slope","proof_lines":[47,52],"relation":"proof"}],"lean_source":"statement","markdown":"For a smooth coefficient, the two applications are connected by the\nidentity\n$$\nq'''+2\\gamma'\\mathbb E a^3=\\mathcal C_\\gamma.\n$$","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":18170},{"anchor":"curvature","at":"b26960-27383","end_byte":27383,"heading_id":"the-integrated-estimate","lean":[],"lean_source":"statement","markdown":"**Passage to a fixed time.** Now let $\\gamma\\in\\mathcal U$ be continuous\nat $t\\in(0,1)$, with $c=\\gamma(t)>0$. For\n$0<\\delta<\\min(t,1-t)$ define\n$$\n\\gamma_\\delta(s)=\n\\begin{cases}\nc,&t-\\delta\\leq s<t+\\delta,\\\\\n\\gamma(s),&\\text{otherwise}.\n\\end{cases}\n$$\nThe inequalities $\\gamma(t-\\delta)\\leq c\\leq\\gamma(t+\\delta)$\nmake both joins nondecreasing, so $\\gamma_\\delta\\in\\mathcal U$, although\nit need not minimise $\\mathcal P$.","note":"Not in the Lean: it inserts no plateau, so γ_δ does not exist there. actual_coercivity reaches the fixed time through sampled grid approximants.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26960},{"anchor":"curvature","at":"b27572-27820","end_byte":27820,"heading_id":"the-integrated-estimate","lean":[{"end_line":88,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sample_coercivity","note":"The analogue of (6) in Lean: the finite-step inequality at the sampled time and coefficient.","plain":"Coercivity at sampled grid times","proof_lines":[81,88],"relation":"proof"}],"lean_source":"statement","markdown":"Lemma 2 and (4), already extended to integrable\ncoefficients, give at the fixed time $t$\n$$\n\\mathcal C_{\\gamma_\\delta}(t)\n=\\mathbb E[d_\\delta^2-12c\\,a_\\delta b_\\delta^2+6c^2a_\\delta^4]\n\\geq\\frac{c^2}{1000}\\mathbb E[a_\\delta^4].           \\tag{6}\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":27572},{"anchor":"curvature","at":"b27893-28018","end_byte":28018,"heading_id":"the-integrated-estimate","lean":[{"end_line":67,"line":50,"lines":[50,51],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.OrderParameter.local_bound","note":"The bound for any order parameter: γ(t)≤‖γ‖₁/(1−b) when t≤b. Lean forms no γ_δ.","plain":"γ bounded by its integral","proof_lines":[52,67],"relation":"states"},{"end_line":131,"line":125,"lines":[125,126],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.positiveCoeff_sample_bound","note":"Lean bounds the sampled coefficient by γ(t)+1, a direct use of monotonicity.","plain":"Sampled coefficient at most γ(t)+1","proof_lines":[127,131],"relation":"proof"}],"lean_source":"statement","markdown":"Monotonicity gives the\ncommon coefficient bound\n$$\n\\sup_{s\\leq T_1}\\gamma_\\delta(s)\n\\leq\\frac{\\|\\gamma_\\delta\\|_1}{1-T_1}.\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":27893},{"anchor":"curvature","at":"b28019-28307","end_byte":28307,"heading_id":"the-integrated-estimate","lean":[{"end_line":139,"line":115,"lines":[115,118],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_derivative_uniform","note":"For the grid approximants, at every derivative order on [0,T]. Lean forms no γ_δ.","plain":"Uniform derivative of the positive scheme","proof_lines":[119,139],"relation":"part"},{"end_line":83,"line":64,"lines":[64,67],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_uniform_derivative_bounds","note":"One bound for all n, at each order, on [0,T].","plain":"Uniform derivative bounds, positive scheme","proof_lines":[68,83],"relation":"part"},{"end_line":47,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_limit","note":"The two combined, at the sampled path point.","plain":"Sampled jets converge almost surely","proof_lines":[37,47],"relation":"part"}],"lean_source":"statement","markdown":"This is the first uniform estimate, and the bounds on spatial\nderivatives require the smoothing theorem. Lemma 2.1 of\n[OpenAI (2026a)][FS] gives uniform convergence in space and uniform\nbounds for $a_\\delta,b_\\delta,d_\\delta$ on $[0,T]$, with the next\nspatial derivatives bounded as well.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":28019},{"anchor":"curvature","at":"b28424-28610","end_byte":28610,"heading_id":"the-integrated-estimate","lean":[{"end_line":133,"line":108,"lines":[124,133],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"Dominated convergence for the polynomial.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"proof"},{"end_line":106,"line":90,"lines":[99,106],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleQuartic_integral_limit","note":"The same for the fourth moment.","plain":"Limit of the sampled quartic moment","proof_lines":[94,106],"relation":"proof"},{"end_line":57,"line":54,"lines":[54,55],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_bound","note":"","plain":"Uniform bound on sampled jets","proof_lines":[56,57],"relation":"part"}],"lean_source":"statement","markdown":"The derivative bounds control\nevaluation along these paths, so the integrands in (6) converge and\nare uniformly bounded, and dominated convergence passes their\nexpectations to the limit.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":28424},{"anchor":"curvature","at":"b28611-28766","end_byte":28766,"heading_id":"the-integrated-estimate","lean":[{"end_line":83,"line":64,"lines":[64,67],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_uniform_derivative_bounds","note":"One constant for all n, given γ, T and the order. No plateau width occurs in the Lean.","plain":"Uniform derivative bounds, positive scheme","proof_lines":[68,83],"relation":"part"}],"lean_source":"statement","markdown":"These estimates depend only on the common\n$L^1$ bound on the coefficient and the fixed distance from time one,\nwhatever the width of the shrinking plateau.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":28611},{"anchor":"curvature","at":"b28767-28864","end_byte":28864,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c²: both sides have that extra factor. The limit of the sampled inequality, by comparing limits.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"The result is\n$\\mathcal C_\\gamma(t)\\ge\\gamma(t)^2\\mathbb E a^4/1000$,\nwhich proves Proposition 3.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":28767},{"anchor":"curvature","at":"b28767-28947","end_byte":28947,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Scaled by c squared, so its type needs no c>0. Positivity is needed to divide. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"The result is\n$\\mathcal C_\\gamma(t)\\ge\\gamma(t)^2\\mathbb E a^4/1000$,\nwhich proves Proposition 3. Neither a time derivative of the\ncoefficient nor minimality has entered the limit.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":28767},{"anchor":"curvature","at":"b29183-29300","end_byte":29300,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,136],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"No condition on q″ among the hypotheses.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"Proposition 3 supplies more for the next\nargument. It applies at the chosen comparison time without assuming\n$q''=0$,","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":29183},{"anchor":"curvature","at":"b17740-18050","end_byte":18050,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"Lean proves this directly from finite-step data, and derives the plateau form of Proposition 5.1 from it.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"states"}],"lean_source":"statement","markdown":"This is the consequence of [OpenAI (2026a), Proposition 5.1 and\nLemmas 2.1–2.2][FS] stated in the companion's\n<span class=\"pdf-companion\" data-pdf-target=\"../long/reasoning-record.pdf\" data-pdf-anchor=\"fixed-time-coercivity\">[Proposition 4.1, §4.5](../long/reasoning-record.md#fixed-time-coercivity)</span>.","paper":"short","role":"mention","section":"3. The integrated estimate","start_byte":17740},{"anchor":"curvature","at":"b29301-29366","end_byte":29366,"heading_id":"the-integrated-estimate","lean":[{"end_line":106,"line":90,"lines":[90,93],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleQuartic_integral_limit","note":"The quartic side of the inequality passes to the limit, for the Lean approximants.","plain":"Limit of the sampled quartic moment","proof_lines":[94,106],"relation":"proof"}],"lean_source":"statement","markdown":"and its fourth-moment lower bound survives the local\nreplacement,","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":29301}]},"support.fixed_time_spatial_coercivity_at_jumps":{"lean":[],"short":"Spatial coercivity at a jump","statements":[{"anchor":"fixed-time-coercivity","at":"b67730-68650","end_byte":68650,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":33,"line":12,"lines":[12,17],"module":"OAI.Probability.SKValue.Coercivity.LiteralCoercivity","name":"OAI.SKValue.IsDiffusion.literal_coercivity","note":"Scalar model, no transport to the support model: every t in (0,1), no continuity hypothesis, c right-continuous. Jets contain c, so divide by c².","plain":"Coercivity along the value-proof diffusion","proof_lines":[18,33],"relation":"states"},{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The retained support declaration still assumes continuity at t, as this passage says.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"**The spatial inequality at a coefficient jump.** Proposition 4.1 and the\nsampled formal interface assume continuity at $t$, and the spatial-moment\nargument permits a little more. Fix $c=\\gamma(t)>0$ and replace $\\gamma$ by $c$ on\n$[t-\\delta,t+\\delta)$, leaving it unchanged elsewhere. Monotonicity\npreserves both joins and right continuity. For fixed\n$0<\\delta_0<\\min(t,1-t)$ and $0<\\delta\\le\\delta_0$,\n$$\n\\|\\gamma_\\delta-\\gamma\\|_1\n\\le2\\delta\\,\\gamma(t+\\delta_0)\\longrightarrow0.\n$$\nThus local boundedness suffices for the same strip stability and coupled\nexpectation limit. Each replacement has a positive plateau at $t$, so\n(4.1) holds at a jump too, with the stated right-continuous value $c$.\nThis extension concerns spatial moments only. It supplies no time\nderivative $q'''(t)$ or $\\gamma'(t)$ at a jump, and the formal declaration\nkeeps every one of its hypotheses. Neither application below uses the\nextension.","note":"Lean proves this inequality only in the scalar model, by a different route.","paper":"long","role":"statement","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":67730}]},"support.flux_identity":{"lean":[{"end_line":136,"line":112,"lines":[112,122],"module":"OAI.Probability.SKValue.Coercivity.ZeroFlux","name":"OAI.SKValue.Certificate.integral_W_zero","note":"Scalar model only: the whole-line integral of W(F2) against f is zero, under growth hypotheses. The support model has no declaration for this.","proof_lines":[123,136],"relation":"states"},{"clause":"(∫ x in Ioi (0:ℝ), certDivergence r s ρ x)=0","end_line":162,"line":106,"lines":[135,135],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Support model: its proof integrates the divergence over the half-line and obtains zero, which is the step in the Lean here.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"clause":"HasDerivAt (certFlux r s ρ) (certDivergence r s ρ x) x","end_line":179,"line":159,"lines":[161,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"Support model: the divergence identity for F2, where r is nonzero. The zero integral is a proof step of gaussian_shape_coercivity; only scalar integral_W_zero states it.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"},{"clause":"0 ≤ certR r x ∧ 0 ≤ certK r s x ∧ 0 ≤ certV r x ∧ 0 ≤ certJ r s x ∧\n    0 ≤ certB r x-2*certV r x ∧ 0 ≤ certL r s x ∧\n    0 ≤ certZ r x+certB r x*certV r x-2*certR r x*certB r x","end_line":214,"line":181,"lines":[186,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"All seven signs at x>0. Lean works on the half-line and reflects, with no extension through zero.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential bounds on F2 and its divergence, which give decay against the Gaussian density.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"proof"}],"short":"Zero expected divergence","statements":[{"anchor":"curvature","at":"b57789-58247","end_byte":58247,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"clause":"0 ≤ certR r x ∧ 0 ≤ certK r s x ∧ 0 ≤ certV r x ∧ 0 ≤ certJ r s x ∧\n    0 ≤ certB r x-2*certV r x ∧ 0 ≤ certL r s x ∧\n    0 ≤ certZ r x+certB r x*certV r x-2*certR r x*certB r x","end_line":214,"line":181,"lines":[186,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"All seven signs at x>0. Lean works on the half-line and reflects, with no extension through zero.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"end_line":148,"line":148,"lines":[148,148],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.certB","note":"b=-w/r, the quotient.","plain":"Certificate variable b = −w/r","relation":"defines"}],"lean_source":"statement","markdown":"In this certificate only,\n$b=-w/r$ is a quotient. In the unscaled curvature formula, $b$ denotes\ninstead the derivative $\\Phi_{xxx}$. At a fixed positive time\ninside the step define\n\n$$\nR=r^2,\\quad K=rs,\\quad b=-w/r,\\quad B=b-2v,\n\\quad L=-rs_{xx},\\quad S_1=z+bv-2Rb.\n$$\n\nThe subscript on $S_1$ distinguishes it from the support set. The\nprevious comparisons give\n\n$$\nR,K,v,B,j,L,S_1\\ge0,\\qquad b=B+2v\\ge0,\\qquad Rb^2=w^2,\n$$\n\nbecause $B=-2H$ and $S_1=2rH_x$.","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":57789},{"anchor":"curvature","at":"b59062-59926","end_byte":59926,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"clause":"HasDerivAt (certFlux r s ρ) (certDivergence r s ρ x) x","end_line":179,"line":159,"lines":[161,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"Support model: the divergence identity for F2, where r is nonzero. The zero integral is a proof step of gaussian_shape_coercivity; only scalar integral_W_zero states it.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"},{"end_line":162,"line":106,"lines":[131,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Lean integrates on x>0: the flux vanishes at zero because r does, and at infinity by Gaussian tails.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential bounds on F2 and its divergence, which give decay against the Gaussian density.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"proof"}],"lean_source":"statement","markdown":"For $\\mathcal D=r\\partial_x$, direct differentiation gives\n\n$$\n\\mathcal D(R,K,v,b,j)\n =(2Rv,\\ 2vK+Rj,\\ -Rb,\\ -z-bv,\\ -jv+Kb-L).\n$$\n\nAlso $f_x/f=r-s$. Hence\n$\\mathcal W(F)=\\mathcal DF+(v+R-K)F$ satisfies\n\n$$\nf\\mathcal W(F)=\\partial_x(rfF).\n$$\n\nBy itself the identity leaves the boundary terms at infinity to be controlled.\nThe tail estimates give $r=O(1)$, exponential decay of $v,w,z,b$ and their\nderivatives, $s=x/t+O(1)$, $j=1/t+O((1+x)e^{-\\kappa x})$ and $K=O(1+x)$.\nEvery polynomial $F$ in $R,K,v,b,j$ and its differentiated expression\ntherefore has at most polynomial growth. Meanwhile\n$f(t,x)\\le C\\exp(-x^2/(2t)+C|x|)$. Thus every term used below is\nabsolutely integrable and $rfF\\to0$ at both infinities. Smooth\nextension across zero prevents an interior boundary term. Integrating\non $[-A,A]$ and taking $A\\to\\infty$ now gives\n$\\mathbb E_f\\mathcal W(F)=0$.","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":59062},{"anchor":"curvature","at":"b23265-23762","end_byte":23762,"heading_id":"the-integrated-estimate","lean":[{"clause":"HasDerivAt (certFlux r s ρ) (certDivergence r s ρ x) x","end_line":179,"line":159,"lines":[161,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"Support model: the divergence identity for F2, where r is nonzero. The zero integral is a proof step of gaussian_shape_coercivity; only scalar integral_W_zero states it.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"},{"end_line":162,"line":106,"lines":[131,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Lean integrates on x>0: the flux vanishes at zero because r does, and at infinity by Gaussian tails.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential bounds on F2 and its divergence, which give decay against the Gaussian density.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"proof"}],"lean_source":"statement","markdown":"The factorisation of the density gives $f_x/f=r-s$. For a suitable\nexpression $F$ in the shape quantities, define\n$$\n\\mathcal W(F)=r\\partial_xF+(v+r^2-rs)F.\n$$\nThen $f\\mathcal W(F)=\\partial_x(rfF)$. The smooth extensions\navoid an interior boundary term at zero, and the source's differentiated\nGaussian tail estimates give $rfF\\to0$ at both spatial infinities\nfor the polynomial correction used below. Hence\n$\\mathbb E_f\\mathcal W(F)=0$, where $\\mathbb E_f$ denotes integration\nagainst $f(x)\\,dx$.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":23265},{"anchor":"curvature-mechanism","at":"b52378-52718","end_byte":52718,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":162,"line":106,"lines":[135,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The same zero: Lean integrates the flux on x>0, where r(0)=0 removes the boundary term, then reflects.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":179,"line":159,"lines":[159,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"The flux derivative, for this one certificate polynomial F2.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"}],"lean_source":"statement","markdown":"The operation can be written directly for the positive density $f_t$,\nwhere $f_t'$ means differentiation in the state variable $x$ with $t$\nfixed.\nFor a smooth function $F$ with integrable terms and vanishing boundary\nproduct $Ff_t$,\n\n$$\n\\mathbb E\\!\\left[F'(X_t)+F(X_t)\\frac{f_t'(X_t)}{f_t(X_t)}\\right]\n=\\int_{\\mathbb R}(Ff_t)'(x)\\,dx=0.\n$$","note":"The Lean has the integrated form only for the certificate flux, not for general F.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":52378},{"anchor":"correction-example","at":"b52794-53324","end_byte":53324,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":115,"line":90,"lines":[90,93],"module":"OAI.Probability.SKSupport.Foundations.Const","name":"OAI.ZeroTemperatureSK.Heat.standardGaussian_stein","note":"Support model. Gaussian integration by parts for general g; g(y)=2y gives the zero-mean step. The Lean never takes that g or this sign change.","proof_lines":[94,115],"relation":"part"},{"end_line":28,"line":13,"lines":[13,15],"module":"OAI.Probability.SKValue.Coercivity.ZeroFlux","name":"OAI.SKValue.integral_flux_zero","note":"Scalar model. A general zero-flux lemma: an integrable derivative of an integrable function integrates to zero.","proof_lines":[16,28],"relation":"part"}],"lean_source":"statement","markdown":"For a small example of this operation, let $Z$ be standard normal and\n$F(x)=2x$. Its density has logarithmic derivative $-x$, so\n$\\mathbb E[2-2Z^2]=0$. Although $2x^2-1$ is negative near zero,\nadding that zero-mean correction gives\n\n$$\n(2x^2-1)+(2-2x^2)=1.\n$$\n\nHence $\\mathbb E[2Z^2-1]=1>0$. The boundary term vanishes because a\npolynomial times the Gaussian density tends to zero. This illustrates\nhow a correction changes the visible sign without changing the\nexpectation, and it supplies none of the SK-specific sign estimates.","note":"The example is not in the Lean. Gaussian integration by parts and a general zero-flux lemma are.","paper":"expository","role":"example","section":"7.3 Where the strict-curvature input comes from","start_byte":52794},{"anchor":"curvature-mechanism","at":"b51775-51971","end_byte":51971,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":131,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The finite-step comparison: r∂ₓs−s∂ₓr≥0 on x≥0, r from the backward value and s the density's score. Further signs: next tab.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"The seven signs on x>0 that the density and value supply.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"}],"lean_source":"statement","markdown":"The proof needs\nbounds relating this density to the spatial derivatives of the value,\nand [OpenAI (2026a), §3][support-signs] gives the factorisation and\ncomparison arguments that establish them.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":51775},{"anchor":"correction-example","at":"b52794-52945","end_byte":52945,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":115,"line":90,"lines":[90,93],"module":"OAI.Probability.SKSupport.Foundations.Const","name":"OAI.ZeroTemperatureSK.Heat.standardGaussian_stein","note":"Support model. Stein's identity for general g, under integrability. Taking g(y)=2y gives E[2Z²]=E[2], a case the Lean does not write out.","proof_lines":[94,115],"relation":"contains"},{"end_line":88,"line":77,"lines":[77,78],"module":"OAI.Probability.SKSupport.Foundations.Const","name":"OAI.ZeroTemperatureSK.Heat.hasDerivAt_standardGaussianPDF","note":"Support model. The standard density satisfies φ'(x)=-xφ(x), so its logarithmic derivative is -x.","proof_lines":[79,88],"relation":"states"}],"lean_source":"statement","markdown":"For a small example of this operation, let $Z$ be standard normal and\n$F(x)=2x$. Its density has logarithmic derivative $-x$, so\n$\\mathbb E[2-2Z^2]=0$.","note":"The Lean states the general identity. This F is one case of it, which the Lean does not write out.","paper":"expository","role":"example","section":"7.3 Where the strict-curvature input comes from","start_byte":52794},{"anchor":"correction-example","at":"b52946-53085","end_byte":53085,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[],"lean_source":"statement","markdown":"Although $2x^2-1$ is negative near zero,\nadding that zero-mean correction gives\n\n$$\n(2x^2-1)+(2-2x^2)=1.\n$$\n\nHence $\\mathbb E[2Z^2-1]=1>0$.","note":"Not in the Lean: this example. The same step on the proof's polynomial is inside gaussian_shape_coercivity, with certDivergence as the correction.","paper":"expository","role":"example","section":"7.3 Where the strict-curvature input comes from","start_byte":52946},{"anchor":"correction-example","at":"b53086-53175","end_byte":53175,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":63,"line":43,"lines":[43,44],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.gaussian_expBound_tendsto_zero","note":"Support model. exp(-dx²)g(x) tends to 0 at +∞ for exponentially bounded g, which covers polynomials. Only the +∞ side is stated.","plain":"Gaussian decay beats exponential growth","proof_lines":[45,63],"relation":"part"}],"lean_source":"statement","markdown":"The boundary term vanishes because a\npolynomial times the Gaussian density tends to zero.","note":"The general tail limit is in the Lean. This example's boundary term is not named there.","paper":"expository","role":"example","section":"7.3 Where the strict-curvature input comes from","start_byte":53086},{"anchor":"correction-example","at":"b53176-53324","end_byte":53324,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[],"lean_source":"statement","markdown":"This illustrates\nhow a correction changes the visible sign without changing the\nexpectation, and it supplies none of the SK-specific sign estimates.","note":"Not in the Lean: a remark on the example. The SK-specific pointwise bound it does not supply is cert_halfLine_coercive.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53176},{"anchor":"curvature-transfer","at":"b53978-54047","end_byte":54047,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":162,"line":106,"lines":[131,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Flux tends to zero at infinity and vanishes at zero, so the divergence integrates to zero.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"The density bounds and vanishing boundary\nproducts justify the first.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53978},{"anchor":"curvature","at":"b57242-57460","end_byte":57460,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":145,"line":145,"lines":[145,145],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.certV","note":"v, as the derivative of r; certW and certZ follow on the next lines.","plain":"Certificate variable v = ∂ₓr","relation":"defines"}],"lean_source":"statement","markdown":"Here the coefficient is a positive constant $c$ on the step containing\nthe chosen time. The variables are the rescaled spatial derivatives\n$$\nr=c\\Phi_x,\\qquad v=c\\Phi_{xx},\\qquad\nw=c\\Phi_{xxx},\\qquad z=c\\Phi_{xxxx}.\n$$","note":"In Lean the certificate takes r and s as functions; the derivatives of Φ enter only when r is instantiated.","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":57242},{"anchor":"curvature","at":"b59062-59192","end_byte":59192,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":58,"line":55,"lines":[55,58],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.DF₂","note":"Exactly these five components multiply the partial derivatives of F₂.","plain":"Derivative of F₂ along the jets","relation":"defines"}],"lean_source":"statement","markdown":"For $\\mathcal D=r\\partial_x$, direct differentiation gives\n\n$$\n\\mathcal D(R,K,v,b,j)\n =(2Rv,\\ 2vK+Rj,\\ -Rb,\\ -z-bv,\\ -jv+Kb-L).\n$$","paper":"long","role":"derivation","section":"4.4 The polynomial correction with zero expectation","start_byte":59062},{"anchor":"curvature","at":"b59194-59303","end_byte":59303,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":179,"line":159,"lines":[159,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"The hypothesis on ρ is f_x/f=r−s.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"},{"end_line":60,"line":60,"lines":[60,60],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.WF₂","note":"W(F₂).","plain":"Weighted divergence of F₂","relation":"defines"}],"lean_source":"statement","markdown":"Also $f_x/f=r-s$. Hence\n$\\mathcal W(F)=\\mathcal DF+(v+R-K)F$ satisfies\n\n$$\nf\\mathcal W(F)=\\partial_x(rfF).\n$$","note":"Lean proves this only for F₂, not for a general polynomial F.","paper":"long","role":"derivation","section":"4.4 The polynomial correction with zero expectation","start_byte":59194},{"anchor":"curvature","at":"b59384-59926","end_byte":59926,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":162,"line":106,"lines":[131,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The flux tends to zero and the half-line integral of the divergence vanishes.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential bounds on F₂ and W(F₂).","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"part"}],"lean_source":"statement","markdown":"The tail estimates give $r=O(1)$, exponential decay of $v,w,z,b$ and their\nderivatives, $s=x/t+O(1)$, $j=1/t+O((1+x)e^{-\\kappa x})$ and $K=O(1+x)$.\nEvery polynomial $F$ in $R,K,v,b,j$ and its differentiated expression\ntherefore has at most polynomial growth. Meanwhile\n$f(t,x)\\le C\\exp(-x^2/(2t)+C|x|)$. Thus every term used below is\nabsolutely integrable and $rfF\\to0$ at both infinities. Smooth\nextension across zero prevents an interior boundary term. Integrating\non $[-A,A]$ and taking $A\\to\\infty$ now gives\n$\\mathbb E_f\\mathcal W(F)=0$.","note":"Lean uses exponential growth bounds instead of the paper's polynomial bounds.","paper":"long","role":"derivation","section":"4.4 The polynomial correction with zero expectation","start_byte":59384},{"anchor":"verification","at":"b252701-252803","end_byte":252803,"heading_id":"source-checks-and-finite-calculation","lean":[{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"The signs.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Integrability.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"part"},{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"This finite check does not establish the PDE\nsigns, integrability, boundary flux or limiting argument.","note":"Lean states each of these as a separate theorem.","paper":"long","role":"explanation","section":"13. Source checks and finite calculation","start_byte":252701},{"anchor":"curvature","at":"b20318-20476","end_byte":20476,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[135,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"hiZero: the flux derivative integrates to zero on the half-line. This is the route full_support uses.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"The proof therefore subtracts a weighted divergence whose expectation\nunder the law of $X_t$ is zero, and what remains has a positive\nfourth-moment remainder.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20318},{"anchor":"curvature","at":"b23265-23316","end_byte":23316,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[119,121],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The density is written gaussianLogDensity d U, and this have gives its log-derivative.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"The factorisation of the density gives $f_x/f=r-s$.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":23265},{"anchor":"curvature","at":"b23317-23424","end_byte":23424,"heading_id":"the-integrated-estimate","lean":[{"end_line":60,"line":60,"lines":[60,60],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.WF₂","note":"Only for F₂. DF₂ is r∂ₓF₂ written by the chain rule, with R=r², K=rs.","plain":"Weighted divergence of F₂","relation":"defines"}],"lean_source":"statement","markdown":"For a suitable\nexpression $F$ in the shape quantities, define\n$$\n\\mathcal W(F)=r\\partial_xF+(v+r^2-rs)F.\n$$","paper":"short","role":"definition","section":"3. The integrated estimate","start_byte":23317},{"anchor":"curvature","at":"b23425-23463","end_byte":23463,"heading_id":"the-integrated-estimate","lean":[{"end_line":160,"line":127,"lines":[127,141],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.hasDerivAt_flux","note":"For F₂ only, with the chain-rule relations for R, K, v, b, j as hypotheses.","plain":"Derivative of the certificate flux","proof_lines":[142,160],"relation":"part"},{"end_line":179,"line":159,"lines":[159,161],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_derivative","note":"Support model: the divergence identity for F2, where r is nonzero. The zero integral is a proof step of gaussian_shape_coercivity; only scalar integral_W_zero states it.","plain":"Divergence identity for the flux","proof_lines":[162,179],"relation":"part"}],"lean_source":"statement","markdown":"Then $f\\mathcal W(F)=\\partial_x(rfF)$.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":23425},{"anchor":"curvature","at":"b23464-23666","end_byte":23666,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[131,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Lean integrates on (0,∞). The flux vanishes at zero because r does, and at infinity by Gaussian tails.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential growth of the correction.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"part"}],"lean_source":"statement","markdown":"The smooth extensions\navoid an interior boundary term at zero, and the source's differentiated\nGaussian tail estimates give $rfF\\to0$ at both spatial infinities\nfor the polynomial correction used below.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":23464},{"anchor":"curvature","at":"b23667-23702","end_byte":23702,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[135,140],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"hiZero, on the half-line.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":194,"line":165,"lines":[182,183],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"}],"lean_source":"statement","markdown":"Hence\n$\\mathbb E_f\\mathcal W(F)=0$,","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":23667}]},"support.forward_shapes":{"lean":[{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The Wronskian r s_x - s r_x is nonnegative for x>=0. It equals r times j.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"Comparison step for s_x>=1/t on one smooth segment, given the sign at the segment's start.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"proof"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Comparison step for s_xx<=0 on x>=0, given the sign at the segment's start.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"proof"}],"short":"Forward score signs","statements":[{"anchor":"curvature","at":"b58126-58603","end_byte":58603,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"All seven signs on the positive half-line, taking forward concavity and the ratio bound as hypotheses, which the next tabs supply.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The Wronskian r s_x - s r_x is nonnegative for x>=0. It equals r times j.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Comparison step for s_xx<=0 on x>=0, given the sign at the segment's start.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"proof"}],"lean_source":"statement","markdown":"previous comparisons give\n\n$$\nR,K,v,B,j,L,S_1\\ge0,\\qquad b=B+2v\\ge0,\\qquad Rb^2=w^2,\n$$\n\nbecause $B=-2H$ and $S_1=2rH_x$. Here $s=-\\partial_x\\log g$\nbelongs to the forward heat factor, while $\\partial_x\\log f=r-s$\nis the logarithmic derivative of the diffusion density. The signs have\nseparate sources: $B\\ge0$ comes from $H\\le0$, $S_1\\ge0$ from\n$H_x\\ge0$ on the positive half-line, $j=s\\rho\\ge0$ from the\nrelative-slope comparison, and $L=-rs_{xx}\\ge0$ from forward concavity.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":58126},{"anchor":"curvature","at":"b53695-54575","end_byte":54575,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"Comparison step for s_x>=1/t on one smooth segment, given the sign at the segment's start.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"proof"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Comparison step for s_xx<=0 on x>=0, given the sign at the segment's start.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"proof"},{"end_line":143,"line":124,"lines":[124,127],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.ForwardDatumShape.jump","note":"Subtracting δΦ from the datum preserves the signs.","plain":"Forward datum shape survives jumps","proof_lines":[128,143],"relation":"part"}],"lean_source":"statement","markdown":"For the forward score use $a_s=s_x$, temporarily distinguishing it\nfrom the Hessian $a=\\Phi_{xx}$. The forward heat equation gives\n\n$$\ns_t=\\tfrac12s_{xx}-ss_x,\\qquad\n(a_s)_t=\\tfrac12(a_s)_{xx}-s(a_s)_x-a_s^2,\n$$\n$$\n(s_{xx})_t=\\tfrac12(s_{xx})_{xx}-s(s_{xx})_x-3a_s s_{xx}.\n$$\n\nThe first step has $s=x/t$. Thereafter $a_s-1/t$ satisfies the\nlinear equation with drift $-s$ and zeroth-order coefficient\n$-(a_s+1/t)$. For fixed smooth data this coefficient is bounded above\non a closed step, independently of spatial truncation. This uses the\ntail bound and continuity, before its sign is known. Comparison gives\n$a_s\\ge1/t$. The last displayed equation then gives $s_{xx}\\le0$\non $x>0$. At a forward jump,\n\n$$\ns_+=s_-+\\delta\\Phi_x,\\quad\n(a_s)_+=(a_s)_-+\\delta\\Phi_{xx},\\quad\n(s_{xx})_+=(s_{xx})_-+\\delta\\Phi_{xxx}.\n$$\n\nThe backward signs and $\\delta\\ge0$ preserve the forward signs.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":53695},{"anchor":"curvature","at":"b54577-55972","end_byte":55972,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The Wronskian r s_x - s r_x is nonnegative for x>=0. It equals r times j.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":125,"line":102,"lines":[102,110],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg","note":"The half-line comparison on one step.","plain":"Ratio sign on the closed interval","proof_lines":[111,125],"relation":"proof"},{"end_line":187,"line":166,"lines":[166,170],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_jump","note":"Under a jump r s_x − s r_x scales by d/c, as r does, so j is unchanged.","plain":"Ratio Wronskian across a height jump","proof_lines":[171,187],"relation":"proof"}],"lean_source":"statement","markdown":"Concavity alone is insufficient here, so we need one further comparison.\nOn $x>0$ set\n$d_s=s_x/s$, $d_r=r_x/r$ and $\\rho=d_s-d_r$.\nThe heat equation gives $s_t=\\tfrac12s_{xx}-ss_x$, whereas $r_t=-rH$.\nTaking logarithmic derivatives yields\n$$\n(d_s)_t=\\tfrac12(d_s)_{xx}+d_s(d_s)_x-s_{xx},\\qquad\n(d_r)_t=-H_x.\n$$\nSince $\\tfrac12(d_r)_{xx}+d_r(d_r)_x=\\tfrac12(w/r)_x=H_x-w$,\nsubstitution of $d_s=d_r+\\rho$ gives its equation\n\n$$\n\\rho_t=\\tfrac12\\rho_{xx}+d_s\\rho_x+(d_r)_x\\rho\n                      +2H_x-w-s_{xx}.\n$$\n\nAll three terms in the source are nonnegative by the comparisons\nalready proved, and $(d_r)_x<0$. On the first step, concavity gives\n$r(x)=\\int_0^xv(y)\\,dy\\ge xv(x)$, so\n$\\rho=1/x-v/r\\ge0$. Odd Taylor expansions give\n$d_s=1/x+O(x)$, $d_r=1/x+O(x)$ and $\\rho=O(x)$ near zero, and\nthe tail estimates give $\\rho\\to0$ at infinity. Comparison on\n$[\\varepsilon,A]$ therefore proves $\\rho\\ge0$ on each subsequent\nstep provided its initial sign is preserved. Define\n\n$$\nj=s\\rho=s_x-(s/r)v.\n$$\n\nAt a forward jump, with\n$\\alpha=c_+/c_-$ and $\\lambda=(c_+-c_-)/c_-$,\n$r_+=\\alpha r_-$, $v_+=\\alpha v_-$,\n$s_+=s_-+\\lambda r_-$ and $(s_x)_+=(s_x)_-+\\lambda v_-$.\nSubstitution cancels the added terms and gives $j_+=j_-$.\nSince $s_+>0$ on $x>0$, the sign of $\\rho$ is preserved.\nThe quotient $s/r$ extends smoothly and evenly across zero, with\n$j(0)=0$. Reflection then gives $j\\ge0$ everywhere.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":54577},{"anchor":"curvature","at":"b21675-22402","end_byte":22402,"heading_id":"the-integrated-estimate","lean":[{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The Wronskian r s_x - s r_x is nonnegative for x>=0. It equals r times j.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"Comparison step for s_x>=1/t on one smooth segment, given the sign at the segment's start.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"proof"},{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Comparison step for s_xx<=0 on x>=0, given the sign at the segment's start.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"proof"}],"lean_source":"statement","markdown":"For $s=-\\partial_x\\log g$, the source proves joint shape restrictions\non $x>0$: $r_x>0$ and $s_x\\geq1/t$, both $r$ and $s$ are concave,\nand $H:=w/(2r)+v\\leq0$ is nondecreasing. They also give\n$$\nj:=s_x-(s/r)v=r\\,\\partial_x(s/r)\\geq0.\n$$\nThis last inequality couples the backward gradient to the forward\ndensity, and separate monotonicity and concavity do not supply it.\nOn the first step $s=x/t$, so concavity of $r$ gives $j\\geq0$, and a\nparabolic comparison propagates the sign within each step.\nAt a forward jump with $k=c_+/c_-\\geq1$, continuity of $\\Phi$ and $f$\ngives\n$$\nr_+=kr_-,\\qquad s_+=s_-+(k-1)r_-,\\qquad\nj_+=(s_x)_+-\\frac{s_+}{r_+}(r_x)_+=j_-.\n$$\nThe joint sign therefore survives jumps of the coefficient as well.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":21675},{"anchor":"curvature","at":"b53972-54317","end_byte":54317,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":50,"line":15,"lines":[15,17],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.deriv_ge_inverse","note":"The lower bound, with forwardTime for t.","plain":"Forward score slope at least 1/t","proof_lines":[18,50],"relation":"proof"}],"lean_source":"statement","markdown":"The first step has $s=x/t$. Thereafter $a_s-1/t$ satisfies the\nlinear equation with drift $-s$ and zeroth-order coefficient\n$-(a_s+1/t)$. For fixed smooth data this coefficient is bounded above\non a closed step, independently of spatial truncation. This uses the\ntail bound and continuity, before its sign is known. Comparison gives\n$a_s\\ge1/t$.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":53972},{"anchor":"curvature","at":"b54318-54379","end_byte":54379,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":81,"line":52,"lines":[52,54],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.SmoothForward.second_nonpos","note":"Forward concavity on the half-line.","plain":"Forward flow preserves half-line concavity","proof_lines":[55,81],"relation":"proof"}],"lean_source":"statement","markdown":"The last displayed equation then gives $s_{xx}\\le0$\non $x>0$.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":54318},{"anchor":"curvature","at":"b54380-54575","end_byte":54575,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":143,"line":124,"lines":[124,127],"module":"OAI.Probability.SKSupport.Parabolic.ForwardSigns","name":"OAI.ZeroTemperatureSK.Heat.ForwardDatumShape.jump","note":"Subtracting δΦ from the datum preserves the signs.","plain":"Forward datum shape survives jumps","proof_lines":[128,143],"relation":"part"},{"end_line":164,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.datumScore_jump","note":"The score jump itself.","plain":"Forward score after a height jump","proof_lines":[157,164],"relation":"part"}],"lean_source":"statement","markdown":"At a forward jump,\n\n$$\ns_+=s_-+\\delta\\Phi_x,\\quad\n(a_s)_+=(a_s)_-+\\delta\\Phi_{xx},\\quad\n(s_{xx})_+=(s_{xx})_-+\\delta\\Phi_{xxx}.\n$$\n\nThe backward signs and $\\delta\\ge0$ preserve the forward signs.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":54380},{"anchor":"curvature","at":"b54577-54707","end_byte":54707,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":14,"line":13,"lines":[13,14],"module":"OAI.Probability.SKSupport.Density.RatioJets","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian","note":"Lean uses r s_x − s r_x, which is r s ρ. It never defines ρ.","plain":"The ratio Wronskian","relation":"part"}],"lean_source":"statement","markdown":"Concavity alone is insufficient here, so we need one further comparison.\nOn $x>0$ set\n$d_s=s_x/s$, $d_r=r_x/r$ and $\\rho=d_s-d_r$.","paper":"long","role":"definition","section":"4.3 Backward and forward comparisons, including jumps","start_byte":54577},{"anchor":"curvature","at":"b54778-55090","end_byte":55090,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":75,"line":60,"lines":[60,66],"module":"OAI.Probability.SKSupport.Density.RatioJets","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_equation","note":"The equation of r s ρ; its source has the same three signed terms, multiplied by r s.","plain":"Parabolic equation for the ratio Wronskian","proof_lines":[67,75],"relation":"proof"}],"lean_source":"statement","markdown":"Taking logarithmic derivatives yields\n$$\n(d_s)_t=\\tfrac12(d_s)_{xx}+d_s(d_s)_x-s_{xx},\\qquad\n(d_r)_t=-H_x.\n$$\nSince $\\tfrac12(d_r)_{xx}+d_r(d_r)_x=\\tfrac12(w/r)_x=H_x-w$,\nsubstitution of $d_s=d_r+\\rho$ gives its equation\n\n$$\n\\rho_t=\\tfrac12\\rho_{xx}+d_s\\rho_x+(d_r)_x\\rho\n                      +2H_x-w-s_{xx}.\n$$","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":54778},{"anchor":"curvature","at":"b55092-55542","end_byte":55542,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":125,"line":102,"lines":[102,110],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg","note":"The half-line comparison on one step.","plain":"Ratio sign on the closed interval","proof_lines":[111,125],"relation":"proof"},{"end_line":92,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Density.RatioJets","name":"OAI.ZeroTemperatureSK.Heat.tangent_bound_of_concave_odd","note":"r(x) at least x r_x(x), the first-step concavity.","plain":"Tangent bound for concave odd r","proof_lines":[80,92],"relation":"proof"}],"lean_source":"statement","markdown":"All three terms in the source are nonnegative by the comparisons\nalready proved, and $(d_r)_x<0$. On the first step, concavity gives\n$r(x)=\\int_0^xv(y)\\,dy\\ge xv(x)$, so\n$\\rho=1/x-v/r\\ge0$. Odd Taylor expansions give\n$d_s=1/x+O(x)$, $d_r=1/x+O(x)$ and $\\rho=O(x)$ near zero, and\nthe tail estimates give $\\rho\\to0$ at infinity. Comparison on\n$[\\varepsilon,A]$ therefore proves $\\rho\\ge0$ on each subsequent\nstep provided its initial sign is preserved.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":55092},{"anchor":"curvature","at":"b55543-55855","end_byte":55855,"heading_id":"backward-and-forward-comparisons-including-jumps","lean":[{"end_line":187,"line":166,"lines":[166,170],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_jump","note":"Under a jump r s_x − s r_x scales by d/c, as r does, so j is unchanged.","plain":"Ratio Wronskian across a height jump","proof_lines":[171,187],"relation":"proof"},{"end_line":220,"line":189,"lines":[189,193],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_first_jump","note":"The first jump.","plain":"Ratio Wronskian sign, first jump","proof_lines":[194,220],"relation":"proof"}],"lean_source":"statement","markdown":"Define\n\n$$\nj=s\\rho=s_x-(s/r)v.\n$$\n\nAt a forward jump, with\n$\\alpha=c_+/c_-$ and $\\lambda=(c_+-c_-)/c_-$,\n$r_+=\\alpha r_-$, $v_+=\\alpha v_-$,\n$s_+=s_-+\\lambda r_-$ and $(s_x)_+=(s_x)_-+\\lambda v_-$.\nSubstitution cancels the added terms and gives $j_+=j_-$.\nSince $s_+>0$ on $x>0$, the sign of $\\rho$ is preserved.","paper":"long","role":"derivation","section":"4.3 Backward and forward comparisons, including jumps","start_byte":55543},{"anchor":"curvature","at":"b21852-21911","end_byte":21911,"heading_id":"the-integrated-estimate","lean":[{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"The sign of r·j at every time of a step, for finite-step data. With r>0 on x>0 this is j≥0.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"states"},{"end_line":149,"line":149,"lines":[149,149],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.certJ","note":"j.","plain":"Certificate j = ∂ₓs − sv/r","relation":"defines"},{"end_line":14,"line":13,"lines":[13,14],"module":"OAI.Probability.SKSupport.Density.RatioJets","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian","note":"r s_x−s r_x, which is r·j.","plain":"The ratio Wronskian","relation":"defines"}],"lean_source":"statement","markdown":"They also give\n$$\nj:=s_x-(s/r)v=r\\,\\partial_x(s/r)\\geq0.\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":21852},{"anchor":"curvature","at":"b21912-22044","end_byte":22044,"heading_id":"the-integrated-estimate","lean":[{"clause":"0 ≤ ratioWronskian (backwardSegment c h (softAbs M) N i)\n      (forwardScore ((i:ℝ)*h) (forwardBoundary c h (softAbs M) N i)) t x","end_line":131,"line":124,"lines":[128,129],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","note":"Support model. Proves the coupled sign r s'-s r'≥0 on finite steps. That separate shapes cannot supply it is not stated.","plain":"r∂ₓs − s∂ₓr ≥ 0 stepwise","proof_lines":[130,131],"relation":"part"},{"end_line":14,"line":11,"lines":[11,14],"module":"OAI.Probability.SKValue.Equations.CoupledDensity","name":"OAI.SKValue.ForwardShape","note":"Scalar model. The same sign is the field coupled, preserved with slope and concavity through heat steps and jumps.","relation":"part"}],"lean_source":"statement","markdown":"This last inequality couples the backward gradient to the forward\ndensity, and separate monotonicity and concavity do not supply it.","note":"The coupled sign is proved in the Lean and supplied to gaussian_shape_coercivity as hQ. That the separate shapes do not imply it is not shown.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":21912},{"anchor":"curvature","at":"b22045-22172","end_byte":22172,"heading_id":"the-integrated-estimate","lean":[{"end_line":220,"line":189,"lines":[189,220],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_first_jump","note":"The tangent bound x r′≤r gives the sign at the first grid time, after the first jump.","plain":"Ratio Wronskian sign, first jump","proof_lines":[194,220],"relation":"proof"},{"end_line":92,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Density.RatioJets","name":"OAI.ZeroTemperatureSK.Heat.tangent_bound_of_concave_odd","note":"","plain":"Tangent bound for concave odd r","proof_lines":[80,92],"relation":"part"}],"lean_source":"statement","markdown":"On the first step $s=x/t$, so concavity of $r$ gives $j\\geq0$, and a\nparabolic comparison propagates the sign within each step.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":22045},{"anchor":"curvature","at":"b22108-22172","end_byte":22172,"heading_id":"the-integrated-estimate","lean":[{"end_line":125,"line":102,"lines":[102,110],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg","note":"Given the shapes of r and s and the sign at time zero, the sign holds on the closed interval.","plain":"Ratio sign on the closed interval","proof_lines":[111,125],"relation":"states"},{"end_line":87,"line":56,"lines":[56,62],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.forwardSegment_ratio","note":"","plain":"Ratio sign propagates across one step","proof_lines":[63,87],"relation":"proof"}],"lean_source":"statement","markdown":"and a\nparabolic comparison propagates the sign within each step.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":22108},{"anchor":"curvature","at":"b22173-22334","end_byte":22334,"heading_id":"the-integrated-estimate","lean":[{"end_line":187,"line":166,"lines":[166,170],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ratioWronskian_jump","note":"The Wronskian r·j scales by k. As r₊=kr₋, this is j₊=j₋.","plain":"Ratio Wronskian across a height jump","proof_lines":[171,187],"relation":"part"},{"end_line":164,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.datumScore_jump","note":"s₊=s₋+δφ′ with δ=c₊−c₋, which is (k−1)r₋.","plain":"Forward score after a height jump","proof_lines":[157,164],"relation":"part"}],"lean_source":"statement","markdown":"At a forward jump with $k=c_+/c_-\\geq1$, continuity of $\\Phi$ and $f$\ngives\n$$\nr_+=kr_-,\\qquad s_+=s_-+(k-1)r_-,\\qquad\nj_+=(s_x)_+-\\frac{s_+}{r_+}(r_x)_+=j_-.\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":22173},{"anchor":"curvature","at":"b22335-22402","end_byte":22402,"heading_id":"the-integrated-estimate","lean":[{"end_line":122,"line":89,"lines":[116,122],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.boundaryRatio_nonneg","note":"The induction step: comparison, then the jump.","plain":"Ratio sign at grid times","proof_lines":[93,122],"relation":"proof"}],"lean_source":"statement","markdown":"The joint sign therefore survives jumps of the coefficient as well.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":22335},{"anchor":"curvature","at":"b22502-22632","end_byte":22632,"heading_id":"the-integrated-estimate","lean":[{"end_line":100,"line":22,"lines":[22,30],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.ReverseBurgers.ratio_nonneg_interior","note":"The paper cites this. The Lean proves the comparison, with the lower bound hlow and the zero value at x=0 as boundary controls.","plain":"Ratio sign before the endpoint","proof_lines":[31,100],"relation":"proof"},{"end_line":104,"line":102,"lines":[102,103],"module":"OAI.Probability.SKSupport.Density.GaussianDecay","name":"OAI.ZeroTemperatureSK.Heat.quotientExtension_continuous","note":"The paper cites the extensions. The Lean gives them as continuous functions, by dslope.","proof_lines":[104,104],"relation":"part"}],"lean_source":"statement","markdown":"Both the boundary controls of the\ncomparison and these extensions are analytic inputs from\n[OpenAI (2026a), Lemmas 3.2–3.4][FS],","paper":"short","role":"mention","section":"3. The integrated estimate","start_byte":22502},{"anchor":"curvature","at":"b21675-21744","end_byte":21744,"heading_id":"the-integrated-estimate","lean":[{"end_line":137,"line":137,"lines":[137,137],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.datumScore","note":"s at a grid time: x/a minus the derivative of the forward exponent.","plain":"Score of the forward datum","relation":"defines"},{"end_line":21,"line":20,"lines":[20,21],"module":"OAI.Probability.SKSupport.Parabolic.ForwardGeometry","name":"OAI.ZeroTemperatureSK.Heat.forwardScore","note":"The same between grid times.","plain":"The forward score s","relation":"defines"}],"lean_source":"statement","markdown":"For $s=-\\partial_x\\log g$, the source proves joint shape restrictions","paper":"short","role":"definition","section":"3. The integrated estimate","start_byte":21675}]},"support.full_relative_support":{"lean":[{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ) ∧\n  (∀ a b : Time, a < b → γ.val a < γ.val b)","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[84,85],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model. Full support for every Stieltjes measure of the minimiser, and strict increase. Its conclusion also gives both witnesses and the moments.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its first clause. The Lean type Time is [0,1), so support on the whole of Time is the relative statement.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"short":"Support is all of [0,1)","statements":[{"anchor":"gaps","at":"b71060-71273","end_byte":71273,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ) ∧\n  (∀ a b : Time, a < b → γ.val a < γ.val b)","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[84,85],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Conditional on a minimiser. Every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its first clause. The Lean type Time is [0,1), so support on the whole of Time is the relative statement.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"**Theorem 5.1 (full support).** In the pure zero-field SK model with\n$\\xi(t)=t^2/2$, let $\\gamma\\in\\mathcal U$ minimize $\\mathcal P$. Then\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and $\\gamma$ is strictly increasing.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71060},{"anchor":"gaps","at":"b78592-78977","end_byte":78977,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"},{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"γ is constant on each gap.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"},{"end_line":146,"line":123,"lines":[123,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"The terminal exclusion gives support to the right of every point.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"proof"}],"lean_source":"statement","markdown":"The support is relatively closed in $[0,1)$. Its complement is a union\nof relative open intervals, and the Stieltjes measure vanishes on each\nsuch interval. For $x<y$ in one interval,\n$$\n\\gamma(y)-\\gamma(x)=\\mu_\\gamma((x,y])=0,\n$$\nso $\\gamma$ is constant there. The terminal argument excludes a component\nending at one. Every remaining component therefore ends at a point\n$b<1$ of $S$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":78592},{"anchor":"gaps","at":"b81711-81750","end_byte":81750,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its first clause. The Lean type Time is [0,1), so support on the whole of Time is the relative statement.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"The cases are exhaustive, so $S=[0,1)$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81711},{"anchor":"main-result","at":"b6273-6516","end_byte":6516,"heading_id":"the-model-and-the-result","lean":[{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ) ∧\n  (∀ a b : Time, a < b → γ.val a < γ.val b)","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[84,85],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Takes a minimiser as the hypothesis hmin. Its conclusion also includes the moment identities.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ) ∧\n  (∀ a b : Time, a < b → γ.val a < γ.val b)","end_line":88,"line":81,"lines":[84,85],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"These two clauses. Time is [0,1), so support on the whole of Time is the relative statement.","plain":"The full-support conclusion","relation":"defines"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its first clause. The Lean type Time is [0,1), so support on the whole of Time is the relative statement.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6273},{"anchor":"gap-proof","at":"b16561-16652","end_byte":16652,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"(∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ) ∧\n  (∀ a b : Time, a < b → γ.val a < γ.val b)","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[84,85],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Takes the hypothesis hmin. Its conclusion gives full support on Time, which is [0,1), so the statement is the relative one.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its first clause. The Lean type Time is [0,1), so support on the whole of Time is the relative statement.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the relatively closed set $S$.\nConsequently $S=[0,1)$","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":16561},{"anchor":"full-support-statement","at":"b25262-25549","end_byte":25549,"heading_id":"the-full-support-theorem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Full support and strict increase are clauses three and four. Time is [0,1), so support on all of Time is the relative statement.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"μ.support=univ ∧ StrictMonoOn (extend γ.val) (Ico (0:ℝ) 1)","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"The same two clauses written out, for any Stieltjes measure of γ.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"**Source theorem ([OpenAI (2026a), Theorem 1.1][support-manuscript]).**\nIn the pure, zero-field model with $\\xi(t)=t^2/2$, let $\\gamma$ be an\nadmissible integrable minimiser of $\\mathcal P$. Its Stieltjes measure\nhas full relative support on $[0,1)$, and $\\gamma$ is strictly increasing.","paper":"expository","role":"statement","section":"4. The full-support theorem","start_byte":25262},{"anchor":"contact-and-gap","at":"b32428-32763","end_byte":32763,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"clause":"∃ a b : ℝ, 0 ≤ a ∧ a < b ∧ b < 1 ∧ (a = 0 ∨ InSupport μ a) ∧\n      InSupport μ b ∧ ∀ t ∈ Ioo a b, ¬ InSupport μ t","end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The abstract case analysis that exhausts the complement, given contact and curvature inputs.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":146,"line":123,"lines":[123,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"The case reaching one: with no terminal plateau there is support beyond every point, so empty support is excluded too.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"lean_source":"statement","markdown":"The support is relatively closed in $[0,1)$. Starting at a missing\npoint, enlarge its mass-free interval until it reaches support or a\nboundary of the domain. Such a maximal interval has two support\nendpoints, starts at zero outside the support, or reaches one. We treat\nthese three cases in order, and the last includes empty support.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32428},{"anchor":"opening","at":"b638-767","end_byte":767,"heading_id":"a-variational-argument-from-contact-to-strict-curvature","lean":[{"clause":"StrictMonoOn (extend γ.val) (Ico (0:ℝ) 1)","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Strictly increasing on [0,1) is what rules out a constant stretch below one.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"We explain why a minimising order parameter is strictly\nincreasing, that is, why it cannot be constant on any interval below one.","paper":"expository","role":"mention","section":"A variational argument from contact to strict curvature","start_byte":638},{"anchor":"opening","at":"b1104-1154","end_byte":1154,"heading_id":"a-variational-argument-from-contact-to-strict-curvature","lean":[{"end_line":460,"line":411,"lines":[419,424],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The terminal case and the zero-height case are separate hypotheses, hend and hzero.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Initial and\nterminal gaps need separate arguments.","paper":"expository","role":"mention","section":"A variational argument from contact to strict curvature","start_byte":1104},{"anchor":"order-parameter-example","at":"b8995-9160","end_byte":9160,"heading_id":"what-a-gap-means","lean":[{"end_line":72,"line":71,"lines":[71,72],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.InSupport","note":"Uses mathlib's Measure.support on the subtype Time, whose neighbourhoods are the clipped intervals.","plain":"t lies in supp μ","relation":"defines"}],"lean_source":"statement","markdown":"The **support** consists of the points every neighbourhood of which\ncarries positive mass, where a neighbourhood is an interval around the\npoint, clipped to $[0,1)$.","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":8995},{"anchor":"order-parameter-example","at":"b9161-9238","end_byte":9238,"heading_id":"what-a-gap-means","lean":[{"clause":"∀ μ : Measure Time, IsStieltjesMeasure γ μ → μ.support = Set.univ","end_line":88,"line":81,"lines":[84,84],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.FullSupportConclusion","note":"Lean's full support: the support is all of Time. The interval reading is what mathlib's Measure.support means on Time.","plain":"The full-support conclusion","relation":"defines"}],"lean_source":"statement","markdown":"Full support means that every nonempty\nrelatively open interval carries mass.","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":9161},{"anchor":"order-parameter-example","at":"b9483-9706","end_byte":9706,"heading_id":"what-a-gap-means","lean":[],"lean_source":"statement","markdown":"For example, the admissible candidate\n\n$$\n\\gamma(t)=\n\\begin{cases}\nt,&0\\le t\\le1/3,\\\\\n1/3,&1/3\\le t\\le2/3,\\\\\nt-1/3,&2/3\\le t<1\n\\end{cases}\n$$\n\nhas zero increment from $2/5$ to $3/5$, but increment $1/8$ from\n$1/8$ to $1/4$.","note":"Not in the Lean: this candidate appears nowhere. Lean has only general lemmas about Stieltjes measures.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":9483},{"anchor":"order-parameter-example","at":"b9849-9976","end_byte":9976,"heading_id":"what-a-gap-means","lean":[],"lean_source":"statement","markdown":"Both endpoints of the gap remain in the support,\nbecause every neighbourhood of either reaches a region carrying positive\nmass.","note":"Not in the Lean: this candidate appears nowhere. exists_support_gap finds support at the ends of some gap, by compactness.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":9849},{"anchor":"order-parameter-example","at":"b9977-10065","end_byte":10065,"heading_id":"what-a-gap-means","lean":[{"end_line":167,"line":148,"lines":[150,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"Applied to this gap it gives the vanishing increments, a direct instance Lean never writes. The lemma allows height zero.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"}],"lean_source":"statement","markdown":"The upper curve in Figure 1 stays positive across the gap while\nits *increments* vanish.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":9977},{"anchor":"order-parameter-example","at":"b10066-10245","end_byte":10245,"heading_id":"what-a-gap-means","lean":[],"lean_source":"statement","markdown":"This candidate has infinitely many support\npoints, including points arbitrarily close to one, and still has a\nmiddle gap, so neither of those weaker properties gives full support.","note":"Not in the Lean: no counterexample is stated. Its proof uses cofinal support only to find a gap, then excludes that gap.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":10066},{"anchor":"first-variation","at":"b26168-26298","end_byte":26298,"heading_id":"why-a-flat-interval-cannot-minimise","lean":[{"end_line":33,"line":19,"lines":[23,29],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"The Lean proof is this one call: contact, no terminal plateau, and the two gap-curvature cases.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"proof"}],"lean_source":"statement","markdown":"We begin with the first-order cost\nof an allowed change, derive conditions at support points, and then\nexclude every possible gap.","paper":"expository","role":"explanation","section":"5. Why a flat interval cannot minimise","start_byte":26168},{"anchor":"contact-and-gap","at":"b32428-32472","end_byte":32472,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":121,"line":80,"lines":[84,85],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"μ.isClosed_support, transferred from Time to a closed subset of the reals.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"proof"}],"lean_source":"statement","markdown":"The support is relatively closed in $[0,1)$.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32428},{"anchor":"contact-and-gap","at":"b32473-32586","end_byte":32586,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":121,"line":80,"lines":[90,97],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"The largest support point below x, or zero, and the smallest above it, by compactness.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"proof"}],"lean_source":"statement","markdown":"Starting at a missing\npoint, enlarge its mass-free interval until it reaches support or a\nboundary of the domain.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32473},{"anchor":"contact-and-gap","at":"b32587-32689","end_byte":32689,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":121,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"The left end is zero or supported. Reaching one is excluded first, by hright, which supplies support to the right.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"Such a maximal interval has two support\nendpoints, starts at zero outside the support, or reaches one.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32587},{"anchor":"contact-and-gap","at":"b32690-32763","end_byte":32763,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[],"lean_source":"statement","markdown":"We treat\nthese three cases in order, and the last includes empty support.","note":"Not in the Lean: a roadmap. Lean treats the terminal case first, by hend, then one bounded gap split on the height c.","paper":"expository","role":"explanation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32690},{"anchor":"dependencies","at":"b7705-7774","end_byte":7774,"heading_id":"outline-of-the-proof","lean":[{"end_line":33,"line":19,"lines":[19,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Full support and both moments for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"Together these steps give full support and the two moment\nequalities.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":7705},{"anchor":"gaps","at":"b81711-81876","end_byte":81876,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":460,"line":411,"lines":[427,460],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The support by contradiction, then contact at every t.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"The cases are exhaustive, so $S=[0,1)$. Only now can contact be\napplied at every $0<t<1$, giving $q(t)=t$ throughout the domain.\nThe value at zero was already known.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81711},{"anchor":"gap-proof","at":"b16561-16629","end_byte":16629,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[427,432],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The Lean proof is by contradiction: a point outside the support lies in a gap with a support point to its right.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":121,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"Produces, from a missing point with support to its right, a gap of the required kind.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the relatively closed set $S$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":16561}]},"support.full_relative_support_curvature":{"lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Support model: all of Theorem 1. Its proof excludes terminal plateaus through the square-root defect, so the curvature-only route of this step is not taken.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"part"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"},{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"Scalar model: the nearest Lean statement to this route. Curvature and q<=1 rule out positive G, giving contact on (0,1), and support is never mentioned.","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"part"}],"short":"Full support by curvature alone","statements":[{"anchor":"gaps","at":"b78008-78500","end_byte":78500,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"Scalar model, in another form: slope convexity and q≤1 exclude a positive hump of G, even one reaching time one.","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"proof"},{"end_line":149,"line":137,"lines":[137,139],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.endpoint_condition","note":"Scalar model: q≤1 replaces right contact when b=1.","plain":"Endpoint condition on a hump","proof_lines":[140,149],"relation":"proof"},{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Support model: the same exclusion at a minimiser, proved through the square-root defect.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"part"}],"lean_source":"statement","markdown":"A nonempty support with a terminal complementary component has a last\npoint $r<1$. Its plateau\nhas positive height, since a neighbourhood of $r$ has positive measure.\nThe contact values and slope at $r$, including the one-sided case $r=0$,\nand positive-plateau curvature meet the calculus hypotheses. Since\n$0\\le q\\le1$, integration gives $G(r)<0$, contradicting atom addition.\nThis follows from the inputs already stated. We keep the quantitative\nproof above for its stronger estimate (5.1).","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":78008},{"anchor":"main-result","at":"b6273-6622","end_byte":6622,"heading_id":"the-model-and-the-result","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"}],"lean_source":"statement","markdown":"**Theorem 1 (full support; quantitative proof following\n[OpenAI (2026a), Theorem 1.1][FS]).**\nIf $\\gamma\\in\\mathcal U$ minimises $\\mathcal P$, then\n$$\n\\operatorname{supp}\\mu_\\gamma=[0,1),\\qquad\n\\gamma(t_2)>\\gamma(t_1)\\quad(0\\leq t_1<t_2<1).\n$$\nFor the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6273},{"anchor":"gap-proof","at":"b13523-13841","end_byte":13841,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[437,446],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The case analysis: contact at both gap ends, or zero at the left end. Terminal plateaus are hypothesis hend, proved by the square-root defect.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"The scalar model takes the chord route: slope convexity and q≤1 exclude a positive hump of G, even reaching one.","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"proof"}],"lean_source":"statement","markdown":"**Proof of Theorem 1, assuming Lemma 2.** We treat the intervals of the\ncomplement of $S$ by type. An interval with two support endpoints admits\ntwo contact conditions. Initial intervals without support at zero have\ncoefficient zero. For a terminal gap, the bound $q\\leq1$ replaces the\nmissing right contact condition.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13523},{"anchor":"gap-proof","at":"b16561-16879","end_byte":16879,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the relatively closed set $S$.\nConsequently $S=[0,1)$, contact gives $q(t)=t$, and differentiation\ngives $\\mathbb E a^2=1$, including the right-hand value at zero.\nMoreover,\n$\\gamma(t_2)-\\gamma(t_1)=\\mu_\\gamma((t_1,t_2])>0$.\n\nThis proves Theorem 1 from the stated analytic inputs.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":16561},{"anchor":"full-support-statement","at":"b25262-25665","end_byte":25665,"heading_id":"the-full-support-theorem","lean":[{"end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Conditional on a minimiser. The Lean proof excludes terminal plateaus by a second-moment defect, not the source's chord route.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"}],"lean_source":"statement","markdown":"**Source theorem ([OpenAI (2026a), Theorem 1.1][support-manuscript]).**\nIn the pure, zero-field model with $\\xi(t)=t^2/2$, let $\\gamma$ be an\nadmissible integrable minimiser of $\\mathcal P$. Its Stieltjes measure\nhas full relative support on $[0,1)$, and $\\gamma$ is strictly increasing.\nFor every associated diffusion and every $0\\le t<1$,\n\n$$\n\\mathbb E[u(t,X_t)^2]=t,\\qquad\n\\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"expository","role":"statement","section":"4. The full-support theorem","start_byte":25262},{"anchor":"endpoints","at":"b41791-42017","end_byte":42017,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"clause":"μ.support=univ","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Same conclusion. The Lean proof excludes terminal plateaus through the square-root defect, so it does not follow this chord route.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the support. Contact now gives\n$q(t)=t$ for every interior time, and symmetry adds the time zero.\nDifferentiating gives $\\mathbb E[a(t,X_t)^2]=1$, which completes the\nproof of the theorem.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":41791},{"anchor":"gap-proof","at":"b13565-13691","end_byte":13691,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[437,442],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Contact at both ends of the gap; the left end may also be zero.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":460,"line":411,"lines":[447,449],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The left slope bound, where the left end is in the support.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"We treat the intervals of the\ncomplement of $S$ by type. An interval with two support endpoints admits\ntwo contact conditions.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13565},{"anchor":"gap-proof","at":"b13680-13756","end_byte":13756,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":185,"line":169,"lines":[169,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"}],"lean_source":"statement","markdown":"conditions. Initial intervals without support at zero have\ncoefficient zero.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13680},{"anchor":"gap-proof","at":"b13757-13841","end_byte":13841,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":149,"line":137,"lines":[137,149],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.endpoint_condition","note":"This is the scalar model's route: q≤1 replaces right contact when b=1.","plain":"Endpoint condition on a hump","proof_lines":[140,149],"relation":"proof"},{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The support model's proof instead excludes terminal plateaus before any gap is chosen.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"For a terminal gap, the bound $q\\leq1$ replaces the\nmissing right contact condition.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13757}]},"support.integrated_coercivity":{"lean":[{"clause":"(1/1000:ℝ)*(∫ x, finiteDensity c h (softAbs M) N i x*quarticMoment r x) ≤\n      ∫ x, finiteDensity c h (softAbs M) N i x*coercivePolynomial r x","end_line":167,"line":111,"lines":[115,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The inequality with constant 1/1000 for concrete finite-step data. The sharper constant does not appear in its statement.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"clause":"(1661507 / 1521000000 : ℝ) * (∫ t, ρ t * (v t)^4) ≤\n      ∫ t, ρ t * ((z t)^2-12*R t*v t*(b t)^2+6*(v t)^4)","end_line":194,"line":165,"lines":[180,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The sharper constant 1661507/1521000000, with signs and vanishing flux as hypotheses. The finite-data proof does not call this whole-line form.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The half-line integration with reflection that the finite-data theorem calls.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"short":"Integrated lower bound","statements":[{"anchor":"curvature","at":"b62768-62931","end_byte":62931,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"clause":"(1/1000:ℝ)*(∫ x, finiteDensity c h (softAbs M) N i x*quarticMoment r x) ≤\n      ∫ x, finiteDensity c h (softAbs M) N i x*coercivePolynomial r x","end_line":167,"line":111,"lines":[115,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The inequality with constant 1/1000 for concrete finite-step data. The sharper constant does not appear in its statement.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"clause":"(1661507 / 1521000000 : ℝ) * (∫ t, ρ t * (v t)^4) ≤\n      ∫ t, ρ t * ((z t)^2-12*R t*v t*(b t)^2+6*(v t)^4)","end_line":194,"line":165,"lines":[180,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The sharper constant 1661507/1521000000, with signs and vanishing flux as hypotheses. The finite-data proof does not call this whole-line form.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The half-line integration with reflection that the finite-data theorem calls.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"Integration\nremoves only $\\mathcal W(F_2)$, giving\n\n$$\n\\mathbb E_f[z^2-12vw^2+6v^4]\n \\ge\\frac{1661507}{1521000000}\\mathbb E_fv^4\n \\ge\\frac1{1000}\\mathbb E_fv^4.\n$$","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":62768},{"anchor":"curvature","at":"b25022-25207","end_byte":25207,"heading_id":"the-integrated-estimate","lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"This inequality for the finite-step density.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"states"},{"end_line":162,"line":106,"lines":[141,162],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Half-line integration, doubling, then the constant 1/1000.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":202,"line":196,"lines":[196,198],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral","note":"","plain":"From sharp constant to 1/1000","proof_lines":[199,202],"relation":"part"}],"lean_source":"statement","markdown":"Integration removes $\\mathcal W(F_2)$ and keeps the fourth-moment\nmargin, giving\n$$\n\\int_{\\mathbb R}(z^2-12vw^2+6v^4)f\\,dx\n\\geq\\frac1{1000}\\int_{\\mathbb R}v^4f\\,dx.           \\tag{5}\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":25022},{"anchor":"curvature","at":"b24108-24337","end_byte":24337,"heading_id":"the-integrated-estimate","lean":[{"end_line":122,"line":107,"lines":[107,111],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity","note":"Full-support model. The decomposition as an inequality: R≥0 under seven sign hypotheses, with κ explicit.","plain":"Pointwise coercivity inequality","proof_lines":[112,122],"relation":"part"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Lemma 4.1][FS] supplies an explicit polynomial $F_2$\nand a decomposition\n$$\nz^2-12vw^2+6v^4\n=\\mathcal W(F_2)+\\kappa v^4+\\mathcal R,\n\\qquad\n\\mathcal R\\geq0,\\qquad\n\\kappa=\\frac{1661507}{1521000000}>\\frac1{1000}.\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":24108},{"anchor":"curvature-transfer","at":"b53863-54166","end_byte":54166,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"clause":"(1/1000:ℝ)*(∫ x, finiteDensity c h (softAbs M) N i x*quarticMoment r x) ≤\n      ∫ x, finiteDensity c h (softAbs M) N i x*coercivePolynomial r x","end_line":167,"line":111,"lines":[115,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The inequality with constant 1/1000 for concrete finite-step data. The sharper constant does not appear in its statement.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"clause":"(1661507 / 1521000000 : ℝ) * (∫ t, ρ t * (v t)^4) ≤\n      ∫ t, ρ t * ((z t)^2-12*R t*v t*(b t)^2+6*(v t)^4)","end_line":194,"line":165,"lines":[180,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The sharper constant 1661507/1521000000, with signs and vanishing flux as hypotheses. The finite-data proof does not call this whole-line form.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The half-line integration with reflection that the finite-data theorem calls.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"A rational polynomial identity gives the second inequality by writing\nthe remainder as a sum of nonnegative terms. The density bounds and vanishing boundary\nproducts justify the first. Consequently\n\n$$\nq'''(t)=\\mathbb E[R_c(t,X_t)]\n=\\mathbb E[(R_c+C)(t,X_t)]\n\\ge\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4].\n$$","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53863},{"anchor":"curvature-transfer","at":"b54048-54166","end_byte":54166,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":188,"line":173,"lines":[179,180],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_feedback_coercivity","note":"The inequality along the feedback path at a finite-step grid time, multiplied through by c². The equality q'''=E[R_c] is not here.","plain":"Finite-step coercivity along the diffusion","proof_lines":[181,188],"relation":"part"},{"end_line":162,"line":106,"lines":[141,162],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Half-line integration with reflection, for shape data, then the constant weakened to 1/1000.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"Consequently\n\n$$\nq'''(t)=\\mathbb E[R_c(t,X_t)]\n=\\mathbb E[(R_c+C)(t,X_t)]\n\\ge\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4].\n$$","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54048},{"anchor":"curvature-mechanism","at":"b50762-51036","end_byte":51036,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[],"lean_source":"statement","markdown":"The proof in Section 5 used strict curvature on a positive constant\ninterval. We now explain how a correction with expectation zero makes\nthe estimate usable once the required density bounds and algebraic\nidentity are supplied, both of which the sources construct and prove.","note":"Not in the Lean: a roadmap for this section. The estimate it leads to is gaussian_shape_coercivity, with signs and density shape as hypotheses.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":50762},{"anchor":"curvature-transfer","at":"b54168-54300","end_byte":54300,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":162,"line":106,"lines":[143,152],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Pointwise bound with the divergence added, then the divergence integrates away.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"This is how the correction produces a useful lower bound for the\nexpectation even though $R_c$ can be negative at particular states.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54168},{"anchor":"curvature","at":"b20420-20476","end_byte":20476,"heading_id":"the-integrated-estimate","lean":[{"end_line":105,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound","note":"","plain":"Certificate remainder is at least c·v⁴","proof_lines":[95,105],"relation":"part"}],"lean_source":"statement","markdown":"and what remains has a positive\nfourth-moment remainder.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20420},{"anchor":"curvature","at":"b20477-20618","end_byte":20618,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The shape hypotheses appear here: BackwardShape r, then the conditions on s, then hQ for the joint sign.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"For this decomposition to have the required\nsigns, the backward value function and the forward density must satisfy\njoint shape restrictions.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20477},{"anchor":"curvature","at":"b25208-25252","end_byte":25252,"heading_id":"the-integrated-estimate","lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The paper imports (5); the Lean proves it, shape comparisons included.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"states"}],"lean_source":"statement","markdown":"We use (5) as an imported analytic estimate.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":25208},{"anchor":"evidence-scope","at":"b52626-52719","end_byte":52719,"heading_id":"formal-verification","lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"Imported in this paper, proved in the Lean: (5), its shapes and the approximation behind actual_coercivity.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"states"}],"lean_source":"statement","markdown":"integrated estimate (5), the shape comparisons\nbehind it and the compact-strip approximation.","paper":"short","role":"explanation","section":"Formal verification","start_byte":52626},{"anchor":"curvature","at":"b20619-20735","end_byte":20735,"heading_id":"the-integrated-estimate","lean":[{"end_line":122,"line":89,"lines":[89,92],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.boundaryRatio_nonneg","note":"The joint comparison, proved in Lean.","plain":"Ratio sign at grid times","proof_lines":[93,122],"relation":"part"},{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The polynomial identity, proved by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"}],"lean_source":"statement","markdown":"Those comparisons and the exact polynomial\nidentity are the substantial inputs from [OpenAI (2026a), §§3–4][FS].","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20619},{"anchor":"curvature","at":"b20736-20880","end_byte":20880,"heading_id":"the-integrated-estimate","lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"Finite-step inequality.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The approximation, at a fixed time.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"We explain their roles first for finite-step coefficients and then pass\nthe integrated inequality to an integrable coefficient by approximation.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20736}]},"support.moment_identity":{"lean":[{"clause":"HasDerivWithinAt (squareMoment W γ (diffusion W γ)) (curvatureMoment W γ t) (Ici (0:ℝ)) t","end_line":167,"line":155,"lines":[157,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative formula on [0,1). The derivative is taken within t>=0, so it is one-sided only at zero.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"clause":"squareMoment W γ (diffusion W γ) t = ∫ s in (0:ℝ)..t, curvatureMoment W γ s","end_line":144,"line":136,"lines":[138,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"Integrated moment identity below time one. The stochastic-integral form of the gradient is not stated here.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"},{"clause":"0 ≤ squareMoment W γ (diffusion W γ) t ∧ |squareMoment W γ (diffusion W γ) t| ≤ 1","end_line":135,"line":125,"lines":[126,126],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"Nonnegativity and the bound by one. The value at zero is squareMoment_zero.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"part"},{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0 for the constructed diffusion.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"}],"short":"Derivative of the gradient moment","statements":[{"anchor":"contact","at":"b32274-32893","end_byte":32893,"heading_id":"the-moment-function-before-minimisation","lean":[{"clause":"HasDerivWithinAt (squareMoment W γ (diffusion W γ)) (curvatureMoment W γ t) (Ici (0:ℝ)) t","end_line":167,"line":155,"lines":[157,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative formula on [0,1). The derivative is taken within t>=0, so it is one-sided only at zero.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"clause":"squareMoment W γ (diffusion W γ) t = ∫ s in (0:ℝ)..t, curvatureMoment W γ s","end_line":144,"line":136,"lines":[138,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"Integrated moment identity below time one. The stochastic-integral form of the gradient is not stated here.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"},{"clause":"0 ≤ squareMoment W γ (diffusion W γ) t ∧ |squareMoment W γ (diffusion W γ) t| ≤ 1","end_line":135,"line":125,"lines":[126,126],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"Nonnegativity and the bound by one. The value at zero is squareMoment_zero.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"part"}],"lean_source":"statement","markdown":"Fix any $\\gamma\\in\\mathcal U$, without yet assuming that it minimises\n$\\mathcal P$. Compact-strip regularity and Itô's formula give\n$$\nu(t,X_t)=\\int_0^t a(s,X_s)\\,dB_s,\\qquad\nq(t):=\\mathbb E u(t,X_t)^2\n     =\\int_0^t\\mathbb E a(s,X_s)^2\\,ds.\n$$\nThe initial term vanishes because zero field makes $\\Phi$ even and\n$u(0,0)=0$. Thus $q(0)=0$, $0\\le q(t)\\le1$, and\n$$\nq'(t)=\\mathbb E a(t,X_t)^2\\qquad(0\\le t<1),\n$$\nwith a right derivative at zero. The integrand is continuous on each\ncompact time strip, so $q\\in C^1[0,1)$. These facts hold for every\nadmissible coefficient, and the identity $q(t)=t$ is still to be proved.","paper":"long","role":"statement","section":"3.1 The moment function before minimisation","start_byte":32274},{"anchor":"contact","at":"b10475-10691","end_byte":10691,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"HasDerivWithinAt (squareMoment W γ (diffusion W γ)) (curvatureMoment W γ t) (Ici (0:ℝ)) t","end_line":167,"line":155,"lines":[157,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative formula on [0,1). The derivative is taken within t>=0, so it is one-sided only at zero.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0, for the constructed diffusion.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"clause":"squareMoment W γ (diffusion W γ) t = ∫ s in (0:ℝ)..t, curvatureMoment W γ s","end_line":144,"line":136,"lines":[138,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"Integrated moment identity below time one. The stochastic-integral form of the gradient is not stated here.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"},{"end_line":152,"line":142,"lines":[142,143],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit","note":"Continuity of q′ on [0,1), jumps of γ included.","plain":"Curvature moment continuous on [0,1)","proof_lines":[144,152],"relation":"part"}],"lean_source":"statement","markdown":"We fix a minimiser. Itô's formula cancels the drift of $u(t,X_t)$:\n$$\ndu(t,X_t)=a(t,X_t)\\,dB_t,\n\\qquad q(0)=0,\\qquad q'(t)=\\mathbb E[a(t,X_t)^2].\n$$\nIn particular, $q'$ is continuous across jumps of the coefficient.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10475},{"anchor":"value-and-diffusion","at":"b24154-24291","end_byte":24291,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"q'=E[a²] on [0,1), one-sided at zero.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"end_line":135,"line":125,"lines":[125,126],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"0≤q and |q|≤1.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"part"},{"end_line":152,"line":142,"lines":[142,143],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit","note":"q′ is continuous on [0,1), so q is C¹.","plain":"Curvature moment continuous on [0,1)","proof_lines":[144,152],"relation":"part"}],"lean_source":"statement","markdown":"The analytic facts needed in the proof are\n\n$$\nq\\in C^1([0,1)),\\qquad q(0)=0,\\qquad 0\\le q(t)\\le1,\n\\qquad q'(t)=\\mathbb E[a(t,X_t)^2].\n$$","paper":"expository","role":"statement","section":"3.3 The response, the state and the averaged square","start_byte":24154},{"anchor":"control-derivations","at":"b49352-50258","end_byte":50258,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"clause":"HasDerivWithinAt (squareMoment W γ (diffusion W γ)) (curvatureMoment W γ t) (Ici (0:ℝ)) t","end_line":167,"line":155,"lines":[157,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative formula on [0,1). The derivative is taken within t>=0, so it is one-sided only at zero.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"clause":"squareMoment W γ (diffusion W γ) t = ∫ s in (0:ℝ)..t, curvatureMoment W γ s","end_line":144,"line":136,"lines":[138,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"Integrated moment identity below time one. The stochastic-integral form of the gradient is not stated here.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"}],"lean_source":"statement","markdown":"Itô's rule is the chain rule for Brownian paths. Along\n$dX_t=b_t\\,dt+dB_t$, the coefficient of $dt$ (the **drift**)\nin a smooth $f(t,X_t)$ is\n$f_t+b_tf_x+f_{xx}/2$, and its noise term is $f_x\\,dB_t$.\nIf we differentiate the PDE in $x$ and take $b_t=\\gamma u$, the drift\nof the slope cancels:\n\n$$\ndu(t,X_t)\n=\\underbrace{(u_t+\\gamma u u_x+u_{xx}/2)}_{0}\\,dt\n+a(t,X_t)\\,dB_t.\n$$\n\nThus $M_t=u(t,X_t)$ is a bounded **martingale**: its conditional\nexpected later value, given observations up to time $t$, is its present\nvalue. Zero field gives symmetry and $M_0=u(0,0)=0$. Applying Itô's rule once more\nto $M_t^2$ gives\n\n$$\nd(M_t^2)=2M_ta(t,X_t)\\,dB_t+a(t,X_t)^2\\,dt.\n$$\n\nThe noise term has mean zero because its coefficient is nonanticipating\nand bounded on each strip ending before one. Taking expectations and\nintegrating yields $q(t)=\\int_0^t\\mathbb E[a(s,X_s)^2]\\,ds$,\nwhich gives the asserted derivative.","paper":"expository","role":"derivation","section":"7.2 The value equation and the moment identity","start_byte":49352},{"anchor":"value-and-diffusion","at":"b22184-22357","end_byte":22357,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":79,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.squareMoment","note":"Support model. The expected square of the gradient along X.","plain":"Second moment of u along X","relation":"defines"}],"lean_source":"statement","markdown":"Now fix $\\gamma$, follow a path to $X_t$, evaluate the slope there,\nsquare it and average over paths. The result is a deterministic curve\n\n$$\nq(t):=\\mathbb E[u(t,X_t)^2].\n$$","paper":"expository","role":"definition","section":"3.3 The response, the state and the averaged square","start_byte":22184},{"anchor":"value-and-diffusion","at":"b24293-24379","end_byte":24379,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"(Ici (0:ℝ))","end_line":167,"line":155,"lines":[157,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative is taken within t≥0.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"}],"lean_source":"statement","markdown":"Here $C^1$ means continuously differentiable, with derivatives at zero\nfrom the right.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":24293},{"anchor":"value-and-diffusion","at":"b24380-24509","end_byte":24509,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":152,"line":142,"lines":[142,143],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit","note":"Continuity on [0,1), with no condition on jumps of γ.","plain":"Curvature moment continuous on [0,1)","proof_lines":[144,152],"relation":"contains"}],"lean_source":"statement","markdown":"In particular, $q'$ stays continuous across jumps of\n$\\gamma$, and spatial curvature controls the growth of the squared\nresponse.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":24380},{"anchor":"value-and-diffusion","at":"b24443-24509","end_byte":24509,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"q'=E[a²]: the curvature moment is the growth rate of the second moment.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"states"},{"end_line":106,"line":86,"lines":[88,90],"module":"OAI.Probability.SKSupport.Moments.SquareEquation","name":"OAI.ZeroTemperatureSK.compactSquare_generator","note":"The generator of the squared gradient along the drift equals the curvature squared.","plain":"Itô generator of u²","proof_lines":[91,106],"relation":"proof"}],"lean_source":"statement","markdown":"and spatial curvature controls the growth of the squared\nresponse.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":24443},{"anchor":"value-and-diffusion","at":"b24510-24629","end_byte":24629,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"No minimiser hypothesis: γ is any OrderParameter.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"}],"lean_source":"statement","markdown":"These facts hold for every admissible candidate and follow\nfrom [OpenAI (2026a), Lemmas 2.1–2.2][support-regularity].","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":24510},{"anchor":"value-and-diffusion","at":"b24631-24828","end_byte":24828,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":162,"line":155,"lines":[155,162],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_even","note":"Proved differently: evenness passes from finite-step approximations. Lean never reflects controls.","plain":"The value is even in x","proof_lines":[157,162],"relation":"part"}],"lean_source":"statement","markdown":"The initial value has a direct symmetry explanation. Reflecting the\nstarting state, the noise history and the response rule together\npreserves reward and cost, so $\\Phi_\\gamma(t,x)$ is even in $x$.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":24631},{"anchor":"value-and-diffusion","at":"b24829-24904","end_byte":24904,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":181,"line":177,"lines":[177,178],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_zero","note":"The gradient is odd, so it vanishes at zero.","plain":"Gradient vanishes at x = 0","proof_lines":[179,181],"relation":"part"},{"end_line":41,"line":34,"lines":[34,41],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ_zero","note":"X_0=0 and u(t,0)=0 give the zero start; squareMoment_zero then obtains q(0)=0 from the integrated identity.","plain":"Strip moment vanishes at zero","proof_lines":[35,41],"relation":"proof"}],"lean_source":"statement","markdown":"Hence\n$u(t,0)=0$, and the zero-field starting state $X_0=0$ gives $q(0)=0$.","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":24829},{"anchor":"value-and-diffusion","at":"b22359-22553","end_byte":22553,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":79,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.squareMoment","note":"Support model. gradient W γ t is a fixed function; the randomness enters only through X at ω.","plain":"Second moment of u along X","relation":"defines"}],"lean_source":"statement","markdown":"Here $\\Phi_\\gamma$ and $u$ are deterministic functions, while $X_t$ and\nthe response evaluated at it are random. Squaring before averaging keeps\nthe size of both positive and negative responses.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":22359},{"anchor":"value-and-diffusion","at":"b24963-25150","end_byte":25150,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":19,"line":18,"lines":[18,19],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential","note":"The tail integral of q(t)-t.","plain":"The potential G","relation":"defines"}],"lean_source":"statement","markdown":"In Section 5\nthe difference $q(t)-t$ will determine the first-order cost of changing\ncumulative weight, and since an added mass persists at every later time,\nthat cost is a tail integral.","paper":"expository","role":"mention","section":"3.3 The response, the state and the averaged square","start_byte":24963},{"anchor":"control-derivations","at":"b49352-49552","end_byte":49552,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":131,"line":14,"lines":[14,25],"module":"OAI.Probability.SKSupport.Foundations.RightGenerator","name":"OAI.ZeroTemperatureSK.expected_hasDerivWithinAt_right","note":"Lean has no Itô integral here. It differentiates expectations directly, giving the same drift terms and no noise term.","plain":"Right generator formula along a diffusion","proof_lines":[26,131],"relation":"proof"}],"lean_source":"statement","markdown":"Itô's rule is the chain rule for Brownian paths. Along\n$dX_t=b_t\\,dt+dB_t$, the coefficient of $dt$ (the **drift**)\nin a smooth $f(t,X_t)$ is\n$f_t+b_tf_x+f_{xx}/2$, and its noise term is $f_x\\,dB_t$.","note":"The Lean proof here is in weak form: it never writes a stochastic differential.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":49352},{"anchor":"control-derivations","at":"b49553-49728","end_byte":49728,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":45,"line":30,"lines":[30,34],"module":"OAI.Probability.SKSupport.Moments.SquareEquation","name":"OAI.ZeroTemperatureSK.deriv_compactValueRate","note":"The PDE differentiated in x, on a strip. The cancellation appears inside the generator below.","plain":"x-derivative of the value's time derivative","proof_lines":[35,45],"relation":"proof"}],"lean_source":"statement","markdown":"If we differentiate the PDE in $x$ and take $b_t=\\gamma u$, the drift\nof the slope cancels:\n\n$$\ndu(t,X_t)\n=\\underbrace{(u_t+\\gamma u u_x+u_{xx}/2)}_{0}\\,dt\n+a(t,X_t)\\,dB_t.\n$$","paper":"expository","role":"derivation","section":"7.2 The value equation and the moment identity","start_byte":49553},{"anchor":"control-derivations","at":"b49730-49873","end_byte":49873,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":137,"line":98,"lines":[98,112],"module":"OAI.Probability.SKValue.Processes.DiffusionGradientCondExp","name":"OAI.SKValue.diffusion_gradient_stopped_martingale","note":"Proved in the scalar model, stopped at T. The support proof of the moment identity never uses a martingale.","plain":"Stopped gradient is a martingale","proof_lines":[113,137],"relation":"part"}],"lean_source":"statement","markdown":"Thus $M_t=u(t,X_t)$ is a bounded **martingale**: its conditional\nexpected later value, given observations up to time $t$, is its present\nvalue.","paper":"expository","role":"statement","section":"7.2 The value equation and the moment identity","start_byte":49730},{"anchor":"control-derivations","at":"b49874-49919","end_byte":49919,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":181,"line":177,"lines":[177,178],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_zero","note":"u(t,0)=0 for t<1.","plain":"Gradient vanishes at x = 0","proof_lines":[179,181],"relation":"part"}],"lean_source":"statement","markdown":"Zero field gives symmetry and $M_0=u(0,0)=0$.","paper":"expository","role":"derivation","section":"7.2 The value equation and the moment identity","start_byte":49874},{"anchor":"control-derivations","at":"b49920-50018","end_byte":50018,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":106,"line":86,"lines":[86,90],"module":"OAI.Probability.SKSupport.Moments.SquareEquation","name":"OAI.ZeroTemperatureSK.compactSquare_generator","note":"The drift of the square: its generator plus time rate equals the curvature squared.","plain":"Itô generator of u²","proof_lines":[91,106],"relation":"proof"}],"lean_source":"statement","markdown":"Applying Itô's rule once more\nto $M_t^2$ gives\n\n$$\nd(M_t^2)=2M_ta(t,X_t)\\,dB_t+a(t,X_t)^2\\,dt.\n$$","paper":"expository","role":"derivation","section":"7.2 The value equation and the moment identity","start_byte":49920},{"anchor":"control-derivations","at":"b50020-50136","end_byte":50136,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":39,"line":29,"lines":[29,32],"module":"OAI.Probability.SKSupport.Foundations.AdaptedIndepIncrement","name":"OAI.ZeroTemperatureSK.WeakIto.integral_adapted_mul_increment","note":"Support model. Step form: a past-measurable factor times a Brownian increment has mean zero. There is no Itô integral; stripQ_hasDerivWithinAt_right rests on this.","proof_lines":[33,39],"relation":"proof"}],"lean_source":"statement","markdown":"The noise term has mean zero because its coefficient is nonanticipating\nand bounded on each strip ending before one.","note":"The Lean has no Itô integral. The mean-zero step appears one increment at a time, inside stripQ_hasDerivWithinAt_right.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":50020},{"anchor":"control-derivations","at":"b50137-50258","end_byte":50258,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":144,"line":136,"lines":[136,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"q(t) is the integral of q'.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"},{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative formula.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"end_line":85,"line":74,"lines":[74,76],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ_eq_integral","note":"Integrating the right derivative on a strip.","plain":"Strip moment as curvature integral","proof_lines":[77,85],"relation":"proof"}],"lean_source":"statement","markdown":"Taking expectations and\nintegrating yields $q(t)=\\int_0^t\\mathbb E[a(s,X_s)^2]\\,ds$,\nwhich gives the asserted derivative.","paper":"expository","role":"derivation","section":"7.2 The value equation and the moment identity","start_byte":50137},{"anchor":"changed-case-answer","at":"b57130-57480","end_byte":57480,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"The remaining Brownian noise\nstill has variance $3/4$, but the value now includes future control\nterms, and its slope must be recomputed before the square is averaged\nagainst the unchanged state law. The displayed integral using $u_0$\ntherefore no longer applies. Knowing the future coefficient schedule\nreveals nothing about the future random noise.","note":"Not in the Lean: a remark on this example. In squareMoment the gradient is that of the full value, which depends on the whole coefficient.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":57130},{"anchor":"targets","at":"b5418-5502","end_byte":5502,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":79,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.squareMoment","note":"Support model: q for a given process X; the proofs use the constructed diffusion.","plain":"Second moment of u along X","relation":"defines"}],"lean_source":"statement","markdown":"Write $q(t)=\\mathbb E[u(t,X_t)^2]$ for the second moment of this\ngradient along $X$.","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":5418},{"anchor":"contact","at":"b32358-32519","end_byte":32519,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":144,"line":136,"lines":[136,138],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_eq_integral_curvatureMoment","note":"The moment identity only.","plain":"q(t) integrates the curvature moment","proof_lines":[139,144],"relation":"part"},{"end_line":72,"line":43,"lines":[43,45],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ_hasDerivWithinAt_right","note":"The right derivative of the expected square.","plain":"Right derivative of strip moment","proof_lines":[46,72],"relation":"proof"}],"lean_source":"statement","markdown":"Compact-strip regularity and Itô's formula give\n$$\nu(t,X_t)=\\int_0^t a(s,X_s)\\,dB_s,\\qquad\nq(t):=\\mathbb E u(t,X_t)^2\n     =\\int_0^t\\mathbb E a(s,X_s)^2\\,ds.\n$$","note":"Lean never states the stochastic integral: it differentiates the expectation from the right, then integrates.","paper":"long","role":"derivation","section":"3.1 The moment function before minimisation","start_byte":32358},{"anchor":"contact","at":"b32520-32598","end_byte":32598,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":181,"line":177,"lines":[177,178],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_zero","note":"Oddness of the gradient at x=0.","plain":"Gradient vanishes at x = 0","proof_lines":[179,181],"relation":"proof"},{"end_line":41,"line":34,"lines":[38,41],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ_zero","note":"Applied at time zero.","plain":"Strip moment vanishes at zero","proof_lines":[35,41],"relation":"proof"}],"lean_source":"statement","markdown":"The initial term vanishes because zero field makes $\\Phi$ even and\n$u(0,0)=0$.","paper":"long","role":"derivation","section":"3.1 The moment function before minimisation","start_byte":32520},{"anchor":"contact","at":"b32599-32630","end_byte":32630,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"end_line":135,"line":125,"lines":[125,126],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"The two bounds.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"part"}],"lean_source":"statement","markdown":"Thus $q(0)=0$, $0\\le q(t)\\le1$,","paper":"long","role":"statement","section":"3.1 The moment function before minimisation","start_byte":32599},{"anchor":"contact","at":"b32718-32793","end_byte":32793,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":149,"line":146,"lines":[146,147],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn","note":"Continuity of the integrand on each strip.","plain":"Curvature moment continuous on [0,T]","proof_lines":[148,149],"relation":"part"}],"lean_source":"statement","markdown":"The integrand is continuous on each\ncompact time strip, so $q\\in C^1[0,1)$.","paper":"long","role":"derivation","section":"3.1 The moment function before minimisation","start_byte":32718},{"anchor":"contact","at":"b32794-32893","end_byte":32893,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The theorem has no IsMinimizer hypothesis.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"}],"lean_source":"statement","markdown":"These facts hold for every\nadmissible coefficient, and the identity $q(t)=t$ is still to be proved.","paper":"long","role":"explanation","section":"3.1 The moment function before minimisation","start_byte":32794},{"anchor":"contact","at":"b32894-33021","end_byte":33021,"heading_id":"the-moment-function-before-minimisation","lean":[{"end_line":85,"line":74,"lines":[74,76],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.stripQ_eq_integral","note":"The identity on one strip ending at T; the moment is then read off from it.","plain":"Strip moment as curvature integral","proof_lines":[77,85],"relation":"proof"}],"lean_source":"statement","markdown":"The source proves them by smooth approximation on $[0,T]$, $T<1$, and\nnever differentiates the terminal cusp $|x|$ at time one.","paper":"long","role":"explanation","section":"3.1 The moment function before minimisation","start_byte":32894},{"anchor":"contact","at":"b10475-10624","end_byte":10624,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":167,"line":155,"lines":[155,157],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","note":"The derivative of q. Lean applies the generator to u²; no minimiser hypothesis is used.","plain":"Derivative of the gradient moment","proof_lines":[158,167],"relation":"part"},{"end_line":153,"line":151,"lines":[151,152],"module":"OAI.Probability.SKSupport.Moments.StripSquareMoment","name":"OAI.ZeroTemperatureSK.squareMoment_zero","note":"q(0)=0 for the constructed diffusion.","plain":"Second moment vanishes at zero","proof_lines":[153,153],"relation":"part"},{"end_line":106,"line":86,"lines":[86,90],"module":"OAI.Probability.SKSupport.Moments.SquareEquation","name":"OAI.ZeroTemperatureSK.compactSquare_generator","note":"The generator of u² is a². This replaces the differential of u.","plain":"Itô generator of u²","proof_lines":[91,106],"relation":"proof"}],"lean_source":"statement","markdown":"We fix a minimiser. Itô's formula cancels the drift of $u(t,X_t)$:\n$$\ndu(t,X_t)=a(t,X_t)\\,dB_t,\n\\qquad q(0)=0,\\qquad q'(t)=\\mathbb E[a(t,X_t)^2].\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10475},{"anchor":"contact","at":"b10625-10678","end_byte":10678,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":152,"line":142,"lines":[142,143],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit","note":"curvatureMoment is q′. Continuity on [0,1) holds for every order parameter, jumps included.","plain":"Curvature moment continuous on [0,1)","proof_lines":[144,152],"relation":"states"}],"lean_source":"statement","markdown":"In particular, $q'$ is continuous across jumps of the","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10625}]},"support.moment_saturation":{"lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"short":"Both moments from full contact","statements":[{"anchor":"gaps","at":"b71060-71391","end_byte":71391,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Conditional on a minimiser. Every minimising order parameter on the given Brownian system satisfies the conclusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"states"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"**Theorem 5.1 (full support).** In the pure zero-field SK model with\n$\\xi(t)=t^2/2$, let $\\gamma\\in\\mathcal U$ minimize $\\mathcal P$. Then\n$\\operatorname{supp}\\mu_\\gamma=[0,1)$ and $\\gamma$ is strictly increasing.\nFor every associated diffusion started at zero,\n$\\mathbb E u(t,X_t)^2=t$ and $\\mathbb E a(t,X_t)^2=1$ for\n$0\\le t<1$.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":71060},{"anchor":"gaps","at":"b81711-82028","end_byte":82028,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"The cases are exhaustive, so $S=[0,1)$. Only now can contact be\napplied at every $0<t<1$, giving $q(t)=t$ throughout the domain.\nThe value at zero was already known. Differentiating and using the\ngradient-martingale identity gives\n$$\n\\mathbb E a(t,X_t)^2=q'(t)=1,\\qquad 0\\le t<1,\n$$\nwith the right derivative at zero.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81711},{"anchor":"formal","at":"b240425-240546","end_byte":240546,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"The moment\nconclusion is universal over every diffusion satisfying the specified\nequation, and covers exactly $0\\le t<1$.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":240425},{"anchor":"main-result","at":"b6517-6622","end_byte":6622,"heading_id":"the-model-and-the-result","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"For the solution of the displayed SDE and every $0\\leq t<1$,\n$$\nq(t)=t,\\qquad \\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"short","role":"statement","section":"1. The model and the result","start_byte":6517},{"anchor":"gap-proof","at":"b16561-16762","end_byte":16762,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the relatively closed set $S$.\nConsequently $S=[0,1)$, contact gives $q(t)=t$, and differentiation\ngives $\\mathbb E a^2=1$, including the right-hand value at zero.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":16561},{"anchor":"full-support-statement","at":"b25550-25665","end_byte":25665,"heading_id":"the-full-support-theorem","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion on the given Brownian system.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Both identities for the constructed diffusion only.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"part"},{"end_line":63,"line":13,"lines":[13,15],"module":"OAI.Probability.SKSupport.Diffusion.Uniqueness","name":"OAI.ZeroTemperatureSK.diffusion_uniqueness","note":"The pathwise uniqueness that transfers the identities to every associated diffusion.","plain":"Pathwise uniqueness of the diffusion","proof_lines":[16,63],"relation":"proof"}],"lean_source":"statement","markdown":"For every associated diffusion and every $0\\le t<1$,\n\n$$\n\\mathbb E[u(t,X_t)^2]=t,\\qquad\n\\mathbb E[a(t,X_t)^2]=1.\n$$","paper":"expository","role":"statement","section":"4. The full-support theorem","start_byte":25550},{"anchor":"endpoints","at":"b41791-42017","end_byte":42017,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":33,"line":19,"lines":[30,33],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Differentiating full contact gives the unit curvature moment.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":460,"line":411,"lines":[458,460],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Full support and contact give q(t)=t.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"These cases exhaust the complement of the support. Contact now gives\n$q(t)=t$ for every interior time, and symmetry adds the time zero.\nDifferentiating gives $\\mathbb E[a(t,X_t)^2]=1$, which completes the\nproof of the theorem.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":41791},{"anchor":"endpoints","at":"b41842-41926","end_byte":41926,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":460,"line":411,"lines":[459,460],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"One step: contact at every point. Lean's contact already covers zero, using hq0.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Contact now gives\n$q(t)=t$ for every interior time, and symmetry adds the time zero.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":41842},{"anchor":"endpoints","at":"b41927-42017","end_byte":42017,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":482,"line":462,"lines":[462,466],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_eq_one_of_full_contact","note":"A function equal to t has derivative one, one-sided at zero.","plain":"Full contact gives q′ = 1","proof_lines":[467,482],"relation":"proof"}],"lean_source":"statement","markdown":"Differentiating gives $\\mathbb E[a(t,X_t)^2]=1$, which completes the\nproof of the theorem.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":41927},{"anchor":"gaps","at":"b81877-82028","end_byte":82028,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":482,"line":462,"lines":[462,466],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_eq_one_of_full_contact","note":"q=t on the interval forces q′=1, one-sidedly at zero.","plain":"Full contact gives q′ = 1","proof_lines":[467,482],"relation":"proof"},{"end_line":33,"line":19,"lines":[30,33],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"The theorem whose proof applies it.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"Differentiating and using the\ngradient-martingale identity gives\n$$\n\\mathbb E a(t,X_t)^2=q'(t)=1,\\qquad 0\\le t<1,\n$$\nwith the right derivative at zero.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81877},{"anchor":"overview","at":"b1030-1081","end_byte":1081,"heading_id":"a-long-reasoning-record","lean":[{"clause":"squareMoment W γ (diffusion W γ) t=t","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Support model: q(t)=t at a minimiser, for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"Full support makes its second moment equal to time,","paper":"long","role":"mention","section":"A long reasoning record","start_byte":1030},{"anchor":"abstract","at":"b910-953","end_byte":953,"heading_id":"full-support-and-positive-density-in-the-zero-temperature-sk-model","lean":[{"clause":"∀ X : ℝ≥0 → Ω → ℝ, IsDiffusion W γ X → ∀ t : Time,\n    squareMoment W γ X t = (t : ℝ) ∧\n    (∫ ω, (curvature W γ t (X (Real.toNNReal t) ω))^2 ∂W.law) = 1","clause_in":"OAI.ZeroTemperatureSK.FullSupportConclusion","clause_lines":[86,88],"clause_module":"OAI.Probability.SKSupport.Model","end_line":59,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.full_support","note":"Its last clause gives both identities for every associated diffusion.","plain":"Full support of SK minimisers","proof_lines":[37,59],"relation":"contains"},{"clause":"∀ t ∈ Ico (0:ℝ) 1, squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t=1","end_line":33,"line":19,"lines":[22,22],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its last clause gives both identities for the constructed diffusion.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"lean_source":"statement","markdown":"Once the support is full, moment\nidentities","paper":"short","role":"mention","section":"Full support and positive density in the zero-temperature SK model","start_byte":910},{"anchor":"gap-proof","at":"b16654-16762","end_byte":16762,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":33,"line":19,"lines":[30,33],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Differentiating full contact gives the unit curvature moment.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"},{"end_line":482,"line":462,"lines":[462,466],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_eq_one_of_full_contact","note":"Its first case is the one-sided derivative at zero.","plain":"Full contact gives q′ = 1","proof_lines":[467,482],"relation":"proof"},{"end_line":460,"line":411,"lines":[458,460],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"q=t follows by contact at every point.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"contact gives $q(t)=t$, and differentiation\ngives $\\mathbb E a^2=1$, including the right-hand value at zero.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":16654},{"anchor":"main-result","at":"b8119-8166","end_byte":8166,"heading_id":"the-model-and-the-result","lean":[{"end_line":482,"line":462,"lines":[462,466],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.derivative_eq_one_of_full_contact","note":"The identity is in the Lean. Its use to force positive density is outside it.","plain":"Full contact gives q′ = 1","proof_lines":[467,482],"relation":"part"}],"lean_source":"statement","markdown":"Once every gap is excluded, the identity $q'=1$","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":8119}]},"support.near_ground_pairs":{"lean":[],"short":"Near-ground pairs at any overlap","statements":[{"anchor":"overlap-pairs","at":"b172065-172405","end_byte":172405,"heading_id":"near-ground-configurations-at-a-prescribed-overlap","lean":[],"lean_source":"record","markdown":"For every fixed $r\\in[0,1)$ and $\\varepsilon,\\eta>0$,\nthere is a constant $K=K(r,\\varepsilon,\\eta)>0$, independent of $n$, such that\n$$\n\\mathbb P\\left(\\exists\\sigma,\\tau\\in\\{-1,1\\}^n:\n\\left|\\frac1n\\sum_{i=1}^n\\sigma_i\\tau_i-r\\right|<\\varepsilon,\n\\quad \\frac{H_n(\\sigma)}n,\\frac{H_n(\\tau)}n\\ge P_*-\\eta\\right)\n\\ge1-Ke^{-n/K},\\qquad n\\ge1.\n$$","paper":"long","role":"statement","section":"9.3 Near-ground configurations at a prescribed overlap","start_byte":172065},{"anchor":"overlap-pairs","at":"b173030-173850","end_byte":173850,"heading_id":"near-ground-configurations-at-a-prescribed-overlap","lean":[],"lean_source":"record","markdown":"The covariance normalization needs one small adjustment. Writing\n$R=n^{-1}\\sum_i\\sigma_i\\tau_i$, our Hamiltonian has covariance\n$nR^2/2-1/2$. Adjoin a common independent Gaussian $C$ of variance $1/2$.\nThen $\\widetilde H_n(\\sigma)=H_n(\\sigma)+C$ has covariance\n$n\\xi(R)$ with $\\xi(t)=t^2/2$, as in the cited mixed-spin theorem.\nIts expected maximum equals that of $H_n$, so its ground-state limit is\nthe same $P_*$. Apply the theorem with energy tolerance $\\eta/2$.\nOn $|C|\\le\\eta n/2$, the resulting pair also satisfies the displayed\nenergy inequalities for $H_n$. The omitted event has probability at most\n$2e^{-\\eta^2n^2/4}$, which can be absorbed into $Ke^{-n/K}$ by enlarging\n$K$. Since the conclusion for $H_n$ depends only on its couplings,\nthis also removes the auxiliary Gaussian from the probability statement.","paper":"long","role":"derivation","section":"9.3 Near-ground configurations at a prescribed overlap","start_byte":173030},{"anchor":"main-result","at":"b7169-7497","end_byte":7497,"heading_id":"the-model-and-the-result","lean":[],"lean_source":"record","markdown":"For each fixed $\\rho\\in[0,1)$ and $\\delta,\\eta>0$, with probability\ntending to one there are configurations $\\sigma,\\tau$ with\n$|R(\\sigma,\\tau)-\\rho|<\\delta$ whose energies are both at least\n$n(P_*-\\eta)$. This follows from\n[Auffinger, Chen and Zeng (2020), Proposition 1, equation (5)][ACZ]\nin the Gaussian normalisation above.","paper":"short","role":"statement","section":"1. The model and the result","start_byte":7169},{"anchor":"overlap-payoff","at":"b43771-44251","end_byte":44251,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"A separate theorem\nof [Auffinger–Chen–Zeng, Proposition 1, equation (5)][acz] applies at\neach point of the minimiser's support, and combining it with full\nsupport gives the following. For every fixed overlap $r\\in[0,1)$ and tolerances $\\varepsilon,\\eta>0$,\nwith probability over the couplings tending to one exponentially fast\nas $n\\to\\infty$, there exist configurations $\\sigma,\\tau$ with\n\n$$\n|R(\\sigma,\\tau)-r|<\\varepsilon,\n\\qquad H_n(\\sigma)/n,\\ H_n(\\tau)/n\\ge P_*-\\eta.\n$$","paper":"expository","role":"statement","section":"6. Return to the spins","start_byte":43771},{"anchor":"overlap-payoff","at":"b44393-44676","end_byte":44676,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"For example, fix an energy loss $\\eta>0$ per spin and take\n$r=1/2$ and $\\varepsilon=0.02$. With probability tending to one\nexponentially fast, the same instance has two configurations with\nenergies per spin at least $P_*-\\eta$ and a differing-site fraction\nbetween $0.24$ and $0.26$.","paper":"expository","role":"example","section":"6. Return to the spins","start_byte":44393},{"anchor":"opening","at":"b1104-1379","end_byte":1379,"heading_id":"a-variational-argument-from-contact-to-strict-curvature","lean":[],"lean_source":"record","markdown":"Initial and\nterminal gaps need separate arguments. Finally we return to the spins,\nwhere a theorem about their energies gives nearly optimal sign choices\ndiffering on any prescribed positive fraction of positions up to one\nhalf, with arbitrarily small error in that fraction.","paper":"expository","role":"mention","section":"A variational argument from contact to strict curvature","start_byte":1104},{"anchor":"spins-and-overlap","at":"b5137-5326","end_byte":5326,"heading_id":"the-finite-optimisation-problem","lean":[],"lean_source":"record","markdown":"To compare two configurations for the *same couplings*,\nwe define their **overlap**\n$R(\\sigma,\\tau)=n^{-1}\\sum_i\\sigma_i\\tau_i$, their average agreement\nwith a disagreement counted as $-1$.","note":"Overlap has no definition in the Lean: neither selected theorem mentions it.","paper":"expository","role":"definition","section":"1. The finite optimisation problem","start_byte":5137},{"anchor":"spins-and-overlap","at":"b5327-5478","end_byte":5478,"heading_id":"the-finite-optimisation-problem","lean":[],"lean_source":"record","markdown":"If they disagree at $k$ positions,\nthen $R=1-2k/n$. For example, $(+,+,+,+)$ and $(+,+,+,-)$ have\noverlap $1/2$, and two disagreements would give zero.","paper":"expository","role":"example","section":"1. The finite optimisation problem","start_byte":5327},{"anchor":"overlap-payoff","at":"b44253-44391","end_byte":44391,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"If $k$ sites differ, $k/n=(1-R)/2$. The overlap tolerance therefore gives\n\n$$\n\\left|\\frac{k}{n}-\\frac{1-r}{2}\\right|<\\frac\\varepsilon2.\n$$","note":"Elementary arithmetic on overlaps, outside both Lean statements.","paper":"expository","role":"derivation","section":"6. Return to the spins","start_byte":44253},{"anchor":"overlap-payoff","at":"b44803-44991","end_byte":44991,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"The pair may depend on the chosen target and on the couplings. The\ntheorem asserts existence for each fixed target and says nothing about\ntypically sampled pairs or about how to find them.","paper":"expository","role":"explanation","section":"6. Return to the spins","start_byte":44803},{"anchor":"overlap-payoff","at":"b45354-45485","end_byte":45485,"heading_id":"return-to-the-spins","lean":[{"end_line":101,"line":101,"lines":[101,101],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.groundStateValue","note":"The limiting value, in the support model; SKValue defines its own. The near-ground theorem itself is not in Lean.","plain":"The SK ground state value","relation":"defines"}],"lean_source":"statement","markdown":"The threshold $P_*-\\eta$ refers to the limiting value,\nwhereas an exact maximiser attains its particular finite instance's maximum.","paper":"expository","role":"explanation","section":"6. Return to the spins","start_byte":45354},{"anchor":"overlap-payoff","at":"b45512-45816","end_byte":45816,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"**Question 5: use the spin conclusion.** Fix an energy loss $\\eta>0$.\nHow can the theorem give two configurations differing on between 69%\nand 71% of sites, although it is stated for nonnegative overlap targets?\nWhich earlier symmetry is needed? Does the stated theorem also permit\nthe choice $\\eta=1/n$?","paper":"expository","role":"example","section":"6. Return to the spins","start_byte":45512},{"anchor":"overlap-payoff","at":"b44721-44801","end_byte":44801,"heading_id":"return-to-the-spins","lean":[],"lean_source":"record","markdown":"At target zero the differing-site fraction can be arbitrarily close\nto one half.","paper":"expository","role":"example","section":"6. Return to the spins","start_byte":44721},{"anchor":"changed-case-answer","at":"b59403-59939","end_byte":59939,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"record","markdown":"**[Question 5](#question-5).** Choose $r=0.4$ and $\\varepsilon=0.02$.\nThe theorem gives a pair differing on between 29% and 31% of sites.\nReverse every sign of the second configuration. Its energy is unchanged,\nits overlap changes sign, and the differing-site fraction becomes its\ncomplement, between 69% and 71%. This operation uses the same successful\ninstance and causes no further probability loss. The theorem fixes\n$\\eta>0$ before $n$ grows and supplies no rate, so the stated result\nstops short of the stronger choice $\\eta=1/n$.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":59403},{"anchor":"overlap-pairs","at":"b175917-176046","end_byte":176046,"heading_id":"near-ground-configurations-at-a-prescribed-overlap","lean":[],"lean_source":"record","markdown":"It uses the external landscape theorem, full support and attainment, and\nlies outside the two selected formal statements in §12.","note":"No Lean declaration concerns near-ground pairs.","paper":"long","role":"mention","section":"9.3 Near-ground configurations at a prescribed overlap","start_byte":175917}]},"support.no_bounded_gap":{"lean":[{"end_line":391,"line":382,"lines":[382,388],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Abstract calculus contradiction for the positive-height case. It takes contact and curvature as hypotheses.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"Abstract contradiction for the zero-height case, using only the right slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof obtains each gap from support points and applies these two contradictions.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"short":"No bounded support gap","statements":[{"anchor":"gaps","at":"b79006-79716","end_byte":79716,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":391,"line":382,"lines":[382,388],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Abstract calculus contradiction for the positive-height case. It takes contact and curvature as hypotheses.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof obtains each gap from support points and applies these two contradictions.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"First consider a gap $(a,b)$ with both endpoints in $S$, allowing\n$a=0$. Compact-strip regularity gives continuity of $q'$ on $[a,b]$,\neven if $\\gamma$ jumps at an endpoint. Contact gives\n$$\nq(a)=a,\\qquad q(b)=b,\\qquad q'(a),q'(b)\\le1.\n\\tag{5.4}\n$$\nIf the constant coefficient is $c>0$, Section 4 makes $q'$ strictly\nconvex in the gap. Continuity extends the chord comparison to its\nendpoints, giving\n$$\nq'(t)<\\frac{b-t}{b-a}q'(a)+\\frac{t-a}{b-a}q'(b)\\le1,\n\\qquad a<t<b.\n\\tag{5.5}\n$$\nIntegrating contradicts $q(b)-q(a)=b-a$. Equivalently, the mean value\ntheorem applied to (5.4) requires an interior point with $q'=1$,\nwhereas (5.5) excludes every such point. The endpoint derivatives need\nonly be at most one.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79006},{"anchor":"gaps","at":"b80608-80811","end_byte":80811,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":405,"line":393,"lines":[393,405],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"No hypothesis on q′(a) appears.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"}],"lean_source":"statement","markdown":"If $c=0$, the separate heat argument in Section 4 makes $q'$ strictly\nincreasing. Then $q'(t)<q'(b)\\le1$ for $a<t<b$, and the same\ncontradiction follows. The left slope bound is unnecessary in this case.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":80608},{"anchor":"gap-proof","at":"b13843-14317","end_byte":14317,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":391,"line":382,"lines":[382,388],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Abstract calculus contradiction for the positive-height case. It takes contact and curvature as hypotheses.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"Abstract contradiction for the zero-height case, using only the right slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof obtains each gap from support points and applies these two contradictions.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Suppose first that $(\\ell,r)$ is a support gap with endpoints in $S$\nand $r<1$. The coefficient is a positive constant there, because\nsupport at $\\ell$ and the absence of mass immediately to its right\nforce $\\mu_\\gamma([0,\\ell])>0$. Strict convexity of $q'$ and (2) give\n$$\nq'(t)<\\frac{r-t}{r-\\ell}q'(\\ell)\n       +\\frac{t-\\ell}{r-\\ell}q'(r)\\leq1\n\\qquad(\\ell<t<r).\n$$\nIntegrating contradicts $q(r)-q(\\ell)=r-\\ell$, so the geometric\nobstruction is an area deficit under $q'$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13843},{"anchor":"contact-and-gap","at":"b32765-33139","end_byte":33139,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Contact at both ends, from hcontact. Lean does not derive c>0 here: it splits on c, and c=0 takes the zero-curvature route.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"part"},{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"The constant height on a gap, with c≥0 only.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"}],"lean_source":"statement","markdown":"First suppose $(r,s)$ has both endpoints in the support, with\n$0\\le r<s<1$. Its constant height $c$ is necessarily positive:\nfor $t$ inside the gap, a small relative neighbourhood of $r$ lies in\n$[0,t]$ and has positive mass, so that $c=\\mu_\\gamma([0,t])>0$.\nIf $r=0$, this mass is an atom at zero. Contact imposes\n\n$$\nq(r)=r,\\quad q(s)=s,\\quad q'(r)\\le1,\\quad q'(s)\\le1.\n$$","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32765},{"anchor":"contact-and-gap","at":"b34295-34650","end_byte":34650,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":391,"line":382,"lines":[389,391],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"strictConvexOn_of_deriv2_pos, from the second derivative of q' on the open interval.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":380,"line":370,"lines":[377,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Lean's first bound is the larger endpoint slope, the chord's supremum, not the chord itself. Then max_le gives the bound by one.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Because $q'''=(q')''$, the function $q'$ is strictly convex: between\ntwo points its graph lies strictly below their joining chord.\nThe moment identity makes $q'$ continuous at the endpoints, even if\n$\\gamma$ jumps there, and higher derivatives are needed only inside the\ngap. Hence\n\n$$\nq'(t)<\\frac{s-t}{s-r}q'(r)+\\frac{t-r}{s-r}q'(s)\\le1\n\\qquad(r<t<s).\n$$","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34295},{"anchor":"contact-and-gap","at":"b36546-36603","end_byte":36603,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":368,"line":361,"lines":[361,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"Lean's contradiction: the mean value theorem gives slope one somewhere, while q'<1.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"Integration gives $q(s)-q(r)<s-r$, contradicting contact.","note":"Lean uses the mean value theorem where the paper integrates.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":36546},{"anchor":"opening","at":"b827-1103","end_byte":1103,"heading_id":"a-variational-argument-from-contact-to-strict-curvature","lean":[{"end_line":380,"line":370,"lines":[370,375],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"The two endpoint values, two slope bounds and strict convexity. Lean ends with the mean value theorem, not an area.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"Minimality\nfixes the values of an auxiliary curve at the two ends of that interval\nand bounds its slopes there, while an analytic estimate makes the\nderivative of the curve strictly convex in between. The curve is then\nleft with too little area to connect the required values.","paper":"expository","role":"explanation","section":"A variational argument from contact to strict curvature","start_byte":827},{"anchor":"contact-and-gap","at":"b32841-33023","end_byte":33023,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[],"lean_source":"statement","markdown":"Its constant height $c$ is necessarily positive:\nfor $t$ inside the gap, a small relative neighbourhood of $r$ lies in\n$[0,t]$ and has positive mass, so that $c=\\mu_\\gamma([0,t])>0$.","note":"Not in the Lean: it never shows c>0 from a supported left end. It splits on c, and c=0 takes the zero-curvature route.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":32841},{"anchor":"contact-and-gap","at":"b33024-33063","end_byte":33063,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":185,"line":169,"lines":[169,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"Support model, contrapositive: zero outside the support and a mass-free (0,b) force γ=0 there. So positive height puts mass at zero.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"Support model. The Stieltjes measure is built with the atom γ(0)δ at zero.","plain":"The Stieltjes measure of γ","relation":"part"},{"end_line":17,"line":10,"lines":[10,11],"module":"OAI.Probability.SKValue.Variation.ContactOrigin","name":"OAI.SKValue.IsMinimizer.contact_origin_of_positive","note":"Scalar model. When γ(0)>0 the truncated mass has an atom at zero, and contact follows there.","plain":"G(0) = 0 if γ(0)>0","proof_lines":[12,17],"relation":"part"}],"lean_source":"statement","markdown":"If $r=0$, this mass is an atom at zero.","note":"The Lean puts an explicit atom γ(0)δ₀ at zero and uses it when the initial height is positive.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":33024},{"anchor":"contact-and-gap","at":"b33064-33139","end_byte":33139,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"clause":"∀ t, SupportDeterministic.InSupport μ t →\n      squareMoment W γ (diffusion W γ) t=t ∧ curvatureMoment W γ t ≤ 1","end_line":92,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.minimizer_contact","note":"Value and slope at every support point, so at both ends here.","plain":"Contact and slope bounds","proof_lines":[84,92],"relation":"contains"},{"end_line":460,"line":411,"lines":[437,442],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Values at both ends and the right slope, from contact; q(0)=0 when the left end is zero.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":460,"line":411,"lines":[447,449],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The left slope, needed only when c>0.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Contact imposes\n\n$$\nq(r)=r,\\quad q(s)=s,\\quad q'(r)\\le1,\\quad q'(s)\\le1.\n$$","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":33064},{"anchor":"contact-and-gap","at":"b33141-33277","end_byte":33277,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":368,"line":361,"lines":[366,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"Lean states this with the mean value theorem: some point has slope exactly one.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"The first two conditions require area $s-r$ under $q'$ across the gap,\nand a calculus model shows how strict convexity can prevent this.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":33141},{"anchor":"contact-and-gap","at":"b34295-34425","end_byte":34425,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":391,"line":382,"lines":[389,391],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"strictConvexOn_of_deriv2_pos, from the second derivative of q' on the open interval.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"}],"lean_source":"statement","markdown":"Because $q'''=(q')''$, the function $q'$ is strictly convex: between\ntwo points its graph lies strictly below their joining chord.","paper":"expository","role":"derivation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34295},{"anchor":"contact-and-gap","at":"b34426-34570","end_byte":34570,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":152,"line":142,"lines":[142,143],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.curvatureMoment_continuousOn_unit","note":"q' is continuous on [0,1) for every order parameter, jumps of γ included. The next tab has the derivatives only inside the gap.","plain":"Curvature moment continuous on [0,1)","proof_lines":[144,152],"relation":"part"},{"end_line":391,"line":382,"lines":[384,387],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"q' is continuous on the closed interval; curvature is assumed only on the open one.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"}],"lean_source":"statement","markdown":"The moment identity makes $q'$ continuous at the endpoints, even if\n$\\gamma$ jumps there, and higher derivatives are needed only inside the\ngap.","paper":"expository","role":"explanation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34426},{"anchor":"contact-and-gap","at":"b36604-36754","end_byte":36754,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":368,"line":361,"lines":[364,365],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"Direct instance with a=1/3, b=2/3, given q'<1 on the gap. The candidate itself is not in the Lean.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"part"}],"lean_source":"statement","markdown":"For the\ncandidate of Section 2, a minimising plateau on $(1/3,2/3)$ would require the same increment\nto be both equal to and strictly less than $1/3$.","paper":"expository","role":"example","section":"5.3 Strict curvature creates an impossible integral","start_byte":36604},{"anchor":"dependencies","at":"b9967-10198","end_byte":10198,"heading_id":"outline-of-the-proof","lean":[{"end_line":460,"line":411,"lines":[418,424],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The four inputs: contact, no terminal plateau, positive and zero gaps.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"**Exclusion of gaps.** We combine contact with the positive, zero or\nterminal argument, as the case requires. Each case has its own destination:\n[terminal](#terminal-gap), [bounded](#bounded-gaps) and\n[initial](#initial-gaps) gaps.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":9967},{"anchor":"gaps","at":"b78592-78853","end_byte":78853,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"Constancy on a gap.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"},{"end_line":45,"line":42,"lines":[42,45],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.eq_cdf_of_support_free","note":"The increment vanishes when the interval misses the support.","plain":"γ is constant off the support","proof_lines":[45,45],"relation":"proof"}],"lean_source":"statement","markdown":"The support is relatively closed in $[0,1)$. Its complement is a union\nof relative open intervals, and the Stieltjes measure vanishes on each\nsuch interval. For $x<y$ in one interval,\n$$\n\\gamma(y)-\\gamma(x)=\\mu_\\gamma((x,y])=0,\n$$\nso $\\gamma$ is constant there.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":78592},{"anchor":"gaps","at":"b78854-78977","end_byte":78977,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":146,"line":123,"lines":[123,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"The support has a point above each x.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"proof"},{"end_line":121,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"So a missing point lies in a gap ending in the support.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"proof"}],"lean_source":"statement","markdown":"The terminal argument excludes a component\nending at one. Every remaining component therefore ends at a point\n$b<1$ of $S$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":78854},{"anchor":"gaps","at":"b79006-79254","end_byte":79254,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":460,"line":411,"lines":[437,442],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The contact data at both ends.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"First consider a gap $(a,b)$ with both endpoints in $S$, allowing\n$a=0$. Compact-strip regularity gives continuity of $q'$ on $[a,b]$,\neven if $\\gamma$ jumps at an endpoint. Contact gives\n$$\nq(a)=a,\\qquad q(b)=b,\\qquad q'(a),q'(b)\\le1.\n\\tag{5.4}\n$$","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79006},{"anchor":"gaps","at":"b79255-79489","end_byte":79489,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":380,"line":370,"lines":[376,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"Strict convexity bounds q′ by the larger endpoint value.","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"If the constant coefficient is $c>0$, Section 4 makes $q'$ strictly\nconvex in the gap. Continuity extends the chord comparison to its\nendpoints, giving\n$$\nq'(t)<\\frac{b-t}{b-a}q'(a)+\\frac{t-a}{b-a}q'(b)\\le1,\n\\qquad a<t<b.\n\\tag{5.5}\n$$","note":"Lean uses the maximum of the endpoint values rather than the chord itself.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79255},{"anchor":"gaps","at":"b79490-79665","end_byte":79665,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":368,"line":361,"lines":[361,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"The mean value form; Lean does not integrate.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"Integrating contradicts $q(b)-q(a)=b-a$. Equivalently, the mean value\ntheorem applied to (5.4) requires an interior point with $q'=1$,\nwhereas (5.5) excludes every such point.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":79490},{"anchor":"overview","at":"b728-930","end_byte":930,"heading_id":"a-long-reasoning-record","lean":[{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"A gap is a constant interval.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"},{"end_line":391,"line":382,"lines":[382,388],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Contact at the ends contradicts the curvature inside.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"}],"lean_source":"statement","markdown":"The proof turns a hypothetical support gap into a constant interval for the\norder parameter. Variational contact at the ends of that interval then\nconflicts with a strict curvature inequality inside it.","paper":"long","role":"mention","section":"A long reasoning record","start_byte":728},{"anchor":"gaps","at":"b72111-72267","end_byte":72267,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":391,"line":382,"lines":[382,388],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"The positive case.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"}],"lean_source":"statement","markdown":"On a positive plateau with two contact endpoints, strict\nconvexity puts $q'$ below its endpoint chord, although its integral\nmust equal the interval length.","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":72111},{"anchor":"gaps","at":"b72268-72366","end_byte":72366,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"The zero case.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"}],"lean_source":"statement","markdown":"On a zero plateau the heat argument\nmakes $q'$ strictly increasing, below the right contact slope.","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":72268},{"anchor":"abstract","at":"b662-909","end_byte":909,"heading_id":"full-support-and-positive-density-in-the-zero-temperature-sk-model","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The sign that makes q′ strictly convex on a positive plateau, for any order parameter. Minimality enters only through the next tab's endpoint conditions.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"proof"},{"end_line":391,"line":382,"lines":[382,391],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Abstract calculus contradiction for the positive-height case. It takes contact and curvature as hypotheses.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"}],"lean_source":"statement","markdown":"On a positive constant interval, the inequality makes the derivative of\nthe gradient second moment strictly convex, contradicting the endpoint\nconditions imposed by minimality, and together with the bounded\ngradient it also excludes terminal gaps.","paper":"short","role":"mention","section":"Full support and positive density in the zero-temperature SK model","start_byte":662},{"anchor":"main-result","at":"b7925-8118","end_byte":8118,"heading_id":"the-model-and-the-result","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Positive second derivative of q′ on a positive plateau, for any order parameter. The next tabs take this as a hypothesis.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"proof"},{"end_line":391,"line":382,"lines":[382,391],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Positive second derivative of q′ gives strict convexity; the next lemma turns that into the contradiction.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":380,"line":370,"lines":[376,380],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative","note":"","plain":"No gap with strictly convex q′","proof_lines":[376,380],"relation":"proof"}],"lean_source":"statement","markdown":"On a positive constant interval, a\nquantitative estimate makes $q'$ strictly convex, so its integral over\nthe gap is smaller than the gap's length, which contradicts the required\nchange in $q$.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":7925},{"anchor":"gap-proof","at":"b13923-14075","end_byte":14075,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[443,446],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Lean does not show positivity here. It splits on c, and the zero-gap lemma covers a zero plateau whatever the left end is.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"}],"lean_source":"statement","markdown":"The coefficient is a positive constant there, because\nsupport at $\\ell$ and the absence of mass immediately to its right\nforce $\\mu_\\gamma([0,\\ell])>0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13923},{"anchor":"gap-proof","at":"b14258-14317","end_byte":14317,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":368,"line":361,"lines":[366,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"In Lean the deficit appears as a slope contradiction.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"so the geometric\nobstruction is an area deficit under $q'$.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14258},{"anchor":"main-result","at":"b7753-7830","end_byte":7830,"heading_id":"the-model-and-the-result","lean":[{"end_line":460,"line":411,"lines":[437,446],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Contact at both ends, then the constant value on the gap.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"The proof has two stages. A gap with two support endpoints faces two\ndemands.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":7753},{"anchor":"gap-proof","at":"b13843-13922","end_byte":13922,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":121,"line":80,"lines":[80,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"The lemma produces such a gap around any point outside the support; the left end may be zero.","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"Suppose first that $(\\ell,r)$ is a support gap with endpoints in $S$\nand $r<1$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13843}]},"support.no_initial_gap":{"lean":[{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"With a=0 this is the integration contradiction. It needs q(0)=0 and no left slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"clause":"∀ t ∈ Ioo 0 b, γ t = 0","end_line":185,"line":169,"lines":[173,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"The coefficient vanishes on the initial gap when zero lies outside the support.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof covers this case by supplying q(0)=0 at the left end of the gap.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"short":"No initial support gap","statements":[{"anchor":"gaps","at":"b81237-81664","end_byte":81664,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"clause":"∀ t ∈ Ioo 0 b, γ t = 0","end_line":185,"line":169,"lines":[173,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"The coefficient vanishes on the initial gap when zero lies outside the support.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"With a=0 this is the contradiction, by the mean value theorem. It needs q(0)=0 and no left slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof covers this case by supplying q(0)=0 at the left end of the gap.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"If instead the complementary component is $[0,b)$ with $0\\notin S$,\nthere is no mass anywhere in $[0,b)$. Thus $\\gamma=0$ there.\nThe identity $q(0)=0$ comes from zero-field symmetry, and the\nright endpoint gives $q(b)=b$, $q'(b)\\le1$. Here $q''(t)>0$ for\n$0<t<b$, and at zero evenness gives $\\Phi_{xxx}(0,0)=0$ and hence\n$q''(0)=0$. Strict increase on the open interval still forces $q'<1$\nthroughout $(0,b)$, an impossibility.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81237},{"anchor":"gap-proof","at":"b14319-14633","end_byte":14633,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"∀ t ∈ Ioo 0 b, γ t = 0","end_line":185,"line":169,"lines":[173,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"The coefficient vanishes on the initial gap when zero lies outside the support.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"With a=0 this is the integration contradiction. It needs q(0)=0 and no left slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof covers this case by supplying q(0)=0 at the left end of the gap.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"An initial gap $[0,r)$ with $0\\notin S$ and $r\\in S$ carries no\nmass, hence $\\gamma=0$ there. Its right endpoint gives $q(r)=r$\nand $q'(r)\\leq1$, and zero field gives $q(0)=0$. The zero-plateau\nidentity gives $q''>0$, hence $q'(t)<q'(r)\\leq1$, and the same\nintegral contradiction follows without assuming $G(0)=0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14319},{"anchor":"endpoints","at":"b36867-37173","end_byte":37173,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"q''>0 where γ=0: the derivative of q' is positive. The identity with the third derivative is the last tab.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"clause":"∀ t ∈ Ioo 0 b, γ t = 0","end_line":185,"line":169,"lines":[173,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"The coefficient vanishes on the initial gap when zero lies outside the support.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"clause":"stripSecond W γ T c t = ∫ ξ, (actualJet W γ t 3 ξ)^2-2*c*(actualJet W γ t 2 ξ)^3 ∂W.law","end_line":34,"line":22,"lines":[24,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"Expands stripSecond as an expectation; at c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"}],"lean_source":"statement","markdown":"**Initial gap with zero outside the support.** Suppose $[0,s)$ is\nmissing and $s<1$ is the first support point. Then $\\gamma=0$ there,\nso the positive-coefficient bound gives no strictness, and the\nzero-coefficient heat identity takes its place:\n\n$$\nq''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0\\qquad(0<t<s).\n$$","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":36867},{"anchor":"endpoints","at":"b38071-38259","end_byte":38259,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"With a=0 this is the integration contradiction. It needs q(0)=0 and no left slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":460,"line":411,"lines":[411,426],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"Its proof covers this case by supplying q(0)=0 at the left end of the gap.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $q'$ is strictly increasing, and right contact gives\n$q'(t)<q'(s)\\le1$ for $0<t<s$. Symmetry supplies $q(0)=0$ and contact\ngives $q(s)=s$, so integrating again leaves too little area.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":38071},{"anchor":"endpoints","at":"b36914-37001","end_byte":37001,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":185,"line":169,"lines":[169,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"If zero and (0,b) miss the support, γ vanishes on (0,b).","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"}],"lean_source":"statement","markdown":"Suppose $[0,s)$ is\nmissing and $s<1$ is the first support point. Then $\\gamma=0$ there,","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":36914},{"anchor":"endpoints","at":"b38071-38159","end_byte":38159,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":405,"line":393,"lines":[400,405],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"q' strictly increasing, so q'<q'(b)≤1 on the open interval.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $q'$ is strictly increasing, and right contact gives\n$q'(t)<q'(s)\\le1$ for $0<t<s$.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":38071},{"anchor":"endpoints","at":"b38160-38259","end_byte":38259,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":460,"line":411,"lines":[437,441],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"q(a)=a is hq0 when a=0, and contact otherwise; contact gives q(b)=b.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Symmetry supplies $q(0)=0$ and contact\ngives $q(s)=s$, so integrating again leaves too little area.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":38160},{"anchor":"endpoints","at":"b38215-38259","end_byte":38259,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":368,"line":361,"lines":[366,368],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_derivative_lt_one","note":"Again the mean value theorem in Lean.","plain":"No gap with slope below one","proof_lines":[366,368],"relation":"proof"}],"lean_source":"statement","markdown":"so integrating again leaves too little area.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":38215},{"anchor":"endpoints","at":"b38395-38616","end_byte":38616,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"Support model. The value at a comes from the value at b and the rate. The initial gap's Brownian path is not stated.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"part"}],"lean_source":"statement","markdown":"Although the forward path is Brownian here, the backward value receives\nthe data $\\Phi_\\gamma(s,\\cdot)$ from the later coefficient, so the\nformula for the globally zero candidate cannot be reused on this initial\ninterval.","note":"A remark on the initial gap. The general backward relation is in the Lean.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":38395},{"anchor":"endpoints","at":"b38617-38741","end_byte":38741,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":460,"line":411,"lines":[447,449],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"by_cases on whether zero is in the support: if it is, contact gives the left slope.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"The case in which zero itself is supported was already covered\nin Section 5.3, using its atom and the one-sided slope bound.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":38617},{"anchor":"gaps","at":"b80966-81185","end_byte":81185,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":460,"line":411,"lines":[447,449],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The slope bound at a when a is in the support.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"$(0,b)$ is a gap and $0\\in S$, the preceding proof applies using\nthe one-sided contact slope bound at zero. Because $(0,b)$ contains\nno mass, the support condition at zero forces an atom there and hence\n$c=\\gamma(0)>0$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":80966},{"anchor":"gaps","at":"b81237-81365","end_byte":81365,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":185,"line":169,"lines":[169,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"γ vanishes on the gap.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"end_line":460,"line":411,"lines":[450,455],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"So the positive case cannot occur here.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"If instead the complementary component is $[0,b)$ with $0\\notin S$,\nthere is no mass anywhere in $[0,b)$. Thus $\\gamma=0$ there.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81237},{"anchor":"gaps","at":"b81366-81664","end_byte":81664,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":460,"line":411,"lines":[437,446],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The hq0 branch at a=0, then the zero-gap lemma.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"The identity $q(0)=0$ comes from zero-field symmetry, and the\nright endpoint gives $q(b)=b$, $q'(b)\\le1$. Here $q''(t)>0$ for\n$0<t<b$, and at zero evenness gives $\\Phi_{xxx}(0,0)=0$ and hence\n$q''(0)=0$. Strict increase on the open interval still forces $q'<1$\nthroughout $(0,b)$, an impossibility.","note":"Lean does not use q′′(0)=0: strict monotonicity on the closed interval needs only the open interval.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":81366},{"anchor":"gaps","at":"b72367-72499","end_byte":72499,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":460,"line":411,"lines":[447,455],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The case split on whether a is in the support.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"At an initial gap we use the one-sided contact slope if $0\\in S$.\nOtherwise the coefficient is zero, and symmetry supplies $q(0)=0$.","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":72367},{"anchor":"gap-proof","at":"b14319-14412","end_byte":14412,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":185,"line":169,"lines":[169,185],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"The coefficient vanishes on the initial gap when zero lies outside the support.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"states"}],"lean_source":"statement","markdown":"An initial gap $[0,r)$ with $0\\notin S$ and $r\\in S$ carries no\nmass, hence $\\gamma=0$ there.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14319},{"anchor":"gap-proof","at":"b14413-14495","end_byte":14495,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[437,442],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"q a = a comes from q 0 = 0 in the case a=0.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"}],"lean_source":"statement","markdown":"Its right endpoint gives $q(r)=r$\nand $q'(r)\\leq1$, and zero field gives $q(0)=0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14413},{"anchor":"gap-proof","at":"b14496-14633","end_byte":14633,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":405,"line":393,"lines":[400,405],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"With a=0 this is the integration contradiction. It needs q(0)=0 and no left slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"},{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"The zero-plateau\nidentity gives $q''>0$, hence $q'(t)<q'(r)\\leq1$, and the same\nintegral contradiction follows without assuming $G(0)=0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14496},{"anchor":"gap-proof","at":"b14563-14633","end_byte":14633,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":405,"line":393,"lines":[393,399],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"The hypotheses state only the right slope bound.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"}],"lean_source":"statement","markdown":"and the same\nintegral contradiction follows without assuming $G(0)=0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14563},{"anchor":"gap-proof","at":"b14634-14705","end_byte":14705,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":460,"line":411,"lines":[447,449],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature","note":"The gap lemma gives a=0 or a in the support; a support point at zero takes this branch.","plain":"Full support from contact and curvature","proof_lines":[427,460],"relation":"proof"},{"end_line":121,"line":80,"lines":[82,83],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.exists_support_gap","note":"","plain":"Missing point gives a support gap","proof_lines":[84,121],"relation":"part"}],"lean_source":"statement","markdown":"A gap\n$(0,r)$ with $0\\in S$ is already covered by the two-contact case.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14634}]},"support.no_terminal_gap":{"lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Same conclusion for the Lean minimiser, covering c=0.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"},{"clause":"∃ B ∈ μ.support, x < B","end_line":146,"line":123,"lines":[127,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Turns the exclusion into support points beyond every time, which rules out empty support.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"short":"No terminal plateau at minimiser","statements":[{"anchor":"gaps","at":"b75420-75922","end_byte":75922,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Same contradiction, covering c=0, from q<t near one. The explicit (1−s′)^{3/2} bound is not in the Lean.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"},{"clause":"∃ B ∈ μ.support, x < B","end_line":146,"line":123,"lines":[127,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Turns the exclusion into support points beyond every time, which rules out empty support.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"lean_source":"statement","markdown":"We can now read off the contradiction. For $s'$ sufficiently\nclose to one, the estimate holds at every $t\\in[s',1)$, so\n$$\n\\begin{aligned}\nG(s')&=\\int_{s'}^1(q(t)-t)\\,dt\\\\\n&\\le\\frac{(1-s')^2}{2}\n       -\\frac{2C}{3}(1-s')^{3/2}<0.\n\\end{aligned}\n$$\nThe final inequality holds once $\\sqrt{1-s'}<4C/3$. It contradicts\nthe nonnegative cost of adding an atom at $s'$. This proof covers\n$c=0$ as well as $c>0$. In particular, empty support is impossible:\nthen $\\mu_\\gamma=0$ and $\\gamma=0$ on all of $[0,1)$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":75420},{"anchor":"gaps","at":"b71969-72110","end_byte":72110,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"The cusp estimate near one, under a smallness condition on 1−T.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"part"},{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Its proof turns q<t near one into G(b)<0, contradicting contactPotential_nonneg.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"},{"clause":"∃ B ∈ μ.support, x < B","end_line":146,"line":123,"lines":[127,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Turns the exclusion into support points beyond every time, which rules out empty support.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"lean_source":"statement","markdown":"For a terminal interval $(a,1)$, including empty support $[0,1)$,\nthe cusp estimate gives $1-q(t)\\gtrsim\\sqrt{1-t}$ and hence $G<0$\nnear one.","paper":"long","role":"explanation","section":"5. Excluding terminal, interior and initial gaps","start_byte":71969},{"anchor":"gap-proof","at":"b15509-15602","end_byte":15602,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Same conclusion for the Lean minimiser, covering c=0.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"},{"clause":"∃ B ∈ μ.support, x < B","end_line":146,"line":123,"lines":[127,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Turns the exclusion into support points beyond every time, which rules out empty support.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"lean_source":"statement","markdown":"The terminal exclusion was already noted in\n[Auffinger, Chen and Zeng (2020), Remark 7][ACZ].","paper":"short","role":"mention","section":"2. Contact conditions and the exclusion of gaps","start_byte":15509},{"anchor":"endpoints","at":"b41689-41841","end_byte":41841,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Same conclusion for the Lean minimiser, covering c=0.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"}],"lean_source":"statement","markdown":"The absence of terminal plateaus had already been noted in\n[Auffinger–Chen–Zeng, Remark 7][acz].\n\nThese cases exhaust the complement of the support.","paper":"expository","role":"mention","section":"5.4 The remaining initial and terminal gaps","start_byte":41689},{"anchor":"endpoints","at":"b38798-38864","end_byte":38864,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":185,"line":169,"lines":[169,173],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.zero_on_initial_support_gap","note":"With empty support this applies for every b<1, giving γ=0 on (0,1). Direct instance; the Lean does not state it.","plain":"γ vanishes on initial support gaps","proof_lines":[174,185],"relation":"part"},{"clause":"0 ≤ c","end_line":78,"line":53,"lines":[54,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Lean has no chord argument here: γ=0 is the terminal plateau with a=0, c=0.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"},{"end_line":146,"line":123,"lines":[123,127],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.support_cofinal_of_no_terminal_plateau","note":"Excluding plateaus gives support beyond every point, so the support is not empty.","plain":"Support points beyond every time","proof_lines":[128,146],"relation":"part"}],"lean_source":"statement","markdown":"**Empty support.** Empty support would mean $\\gamma=0$ everywhere.","note":"This case and its chord argument are the paper's own; the Lean covers empty support within the terminal plateau case.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":38798},{"anchor":"endpoints","at":"b39554-39742","end_byte":39742,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The case this paragraph rules out. Lean's plateau needs no supported left end and allows c=0, so positive height is not in it.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"part"}],"lean_source":"statement","markdown":"**Terminal gap.** Now suppose a maximal missing interval is $(r,1)$.\nIts left endpoint $r<1$ is supported, and its constant height is\npositive, by the same mass argument as in Section 5.3.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":39554},{"anchor":"endpoints","at":"b40989-41091","end_byte":41091,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":51,"line":15,"lines":[15,18],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one","note":"The Lean proof gives q<t only near one, from the square-root defect, with no convexity.","plain":"Terminal plateau gives q(t)<t near one","proof_lines":[19,51],"relation":"proof"},{"end_line":78,"line":53,"lines":[72,78],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"G(b)<0 at that nearby b, contradicting contactPotential_nonneg. Lean does not reach the left end r.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $q(t)<t$ throughout $(r,1)$, giving\n$G(r)=\\int_r^1(q(t)-t)\\,dt<0$, contrary to the addition test.","note":"The support model proves the same conclusion by the source's terminal estimate rather than this chord barrier.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":40989},{"anchor":"changed-cases","at":"b42648-42763","end_byte":42763,"heading_id":"two-checks-on-the-gap-argument","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The result Question 4 is about. Its proof differs from the chord argument of Section 5.4.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"}],"lean_source":"statement","markdown":"**Question 4: the missing right endpoint.** Put the proof aside and\nexplain why a terminal plateau cannot minimise.","paper":"expository","role":"example","section":"5.5 Two checks on the gap argument","start_byte":42648},{"anchor":"changed-cases","at":"b42764-42868","end_byte":42868,"heading_id":"two-checks-on-the-gap-argument","lean":[{"end_line":51,"line":15,"lines":[15,18],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one","note":"In Lean the replacement is the square-root defect, which puts q below t near one.","plain":"Terminal plateau gives q(t)<t near one","proof_lines":[19,51],"relation":"proof"}],"lean_source":"statement","markdown":"Identify what replaces\nthe missing right contact point and why the contradiction occurs before\ntime one.","paper":"expository","role":"explanation","section":"5.5 Two checks on the gap argument","start_byte":42764},{"anchor":"gaps","at":"b73097-73827","end_byte":73827,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":68,"line":35,"lines":[35,38],"module":"OAI.Probability.SKSupport.Backward.TerminalTailValue","name":"OAI.ZeroTemperatureSK.value_on_terminal_plateau","note":"On the plateau the value is the log-heat evolution of |x|.","plain":"Explicit value on a terminal plateau","proof_lines":[39,68],"relation":"proof"},{"end_line":52,"line":31,"lines":[31,36],"module":"OAI.Probability.SKSupport.Backward.TerminalTransition","name":"OAI.ZeroTemperatureSK.terminal_plateau_transition","note":"The transition, against test functions.","plain":"Diffusion transition on a terminal plateau","proof_lines":[37,52],"relation":"proof"}],"lean_source":"statement","markdown":"Fix $s\\in(a,1)$ and let $\\rho_s$ be the law of $X_s$. Choose\n$R<\\infty$ such that $p_0=\\rho_s([-R,R])>0$. For $c>0$, set\n$$\nh(t,y)=\\mathbb E e^{c|y+\\sqrt{1-t}Z|},\\qquad\n\\Phi(t,y)=c^{-1}\\log h(t,y),\n$$\nwhere $Z$ is standard normal. The drift on this interval is\n$c\\Phi_x=h_x/h$, and the transition density is\n$$\nk_{s,t}(x,y)=p_{t-s}(y-x)\\frac{h(t,y)}{h(s,x)},\\qquad\np_v(z)=(2\\pi v)^{-1/2}e^{-z^2/(2v)}.\n\\tag{5.2}\n$$\nThe heat-semigroup identity normalises this kernel. Equivalently, for\nBrownian motion started at $(s,x)$, $h(t,B_t)/h(s,x)$ is the positive\nmean-one change of measure which adds the drift $h_x/h$. For $c=0$,\nthe same kernel formula applies\nwith $h=1$, and the value function is the ordinary heat evolution of\n$|x|$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":73097},{"anchor":"gaps","at":"b75420-75782","end_byte":75782,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":51,"line":15,"lines":[15,18],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one","note":"q(t)<t on an interval near one.","plain":"Terminal plateau gives q(t)<t near one","proof_lines":[19,51],"relation":"proof"},{"end_line":78,"line":53,"lines":[55,78],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The positive integral of t−q gives G(b)<0, contradicting contactPotential_nonneg.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"We can now read off the contradiction. For $s'$ sufficiently\nclose to one, the estimate holds at every $t\\in[s',1)$, so\n$$\n\\begin{aligned}\nG(s')&=\\int_{s'}^1(q(t)-t)\\,dt\\\\\n&\\le\\frac{(1-s')^2}{2}\n       -\\frac{2C}{3}(1-s')^{3/2}<0.\n\\end{aligned}\n$$\nThe final inequality holds once $\\sqrt{1-s'}<4C/3$. It contradicts\nthe nonnegative cost of adding an atom at $s'$.","note":"Lean uses only the sign of q−t near one, not the explicit (1−s′)^{3/2} estimate.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":75420},{"anchor":"gaps","at":"b75783-75922","end_byte":75922,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The hypothesis allows c=0 and a=0, so empty support is included.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"}],"lean_source":"statement","markdown":"This proof covers\n$c=0$ as well as $c>0$. In particular, empty support is impossible:\nthen $\\mu_\\gamma=0$ and $\\gamma=0$ on all of $[0,1)$.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":75783},{"anchor":"abstract","at":"b827-909","end_byte":909,"heading_id":"full-support-and-positive-density-in-the-zero-temperature-sk-model","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The support model's proof uses a square-root deficit here. The scalar model's proof of contact uses the bounded gradient, closer to this sentence.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"states"}],"lean_source":"statement","markdown":"minimality, and together with the bounded\ngradient it also excludes terminal gaps.","paper":"short","role":"mention","section":"Full support and positive density in the zero-temperature SK model","start_byte":827},{"anchor":"gap-proof","at":"b15603-15666","end_byte":15666,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"Scalar model: slope convexity and q≤1 exclude a positive hump of G, even reaching one. Closest to the sentence's argument.","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"proof"},{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"The support model's proof uses this stronger estimate rather than the curvature alone.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"proof"}],"lean_source":"statement","markdown":"Our argument uses\nonly the curvature needed for the other gaps,","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":15603}]},"support.optimality_criterion":{"lean":[{"clause":"0 ≤ contactPotential W γ s","end_line":62,"line":57,"lines":[58,58],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.contactPotential_nonneg","note":"G is nonnegative at a minimiser. The converse direction of the criterion has no Lean declaration.","plain":"G is nonnegative","proof_lines":[59,62],"relation":"part"},{"end_line":250,"line":230,"lines":[230,235],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.zero_on_support_of_compact_mass_removal","note":"Abstract step giving G=0 on the support, the necessity half. The convexity argument for the converse has no Lean declaration.","plain":"G vanishes on support, abstract form","proof_lines":[236,250],"relation":"part"}],"short":"First-order optimality criterion","statements":[{"anchor":"contact","at":"b41723-42768","end_byte":42768,"heading_id":"the-reach-of-the-first-order-criterion","lean":[],"lean_source":"statement","markdown":"For the actual moment function associated with $\\gamma$, the two\npotential conditions $G\\ge0$ and $G=0$ on $S$ are also sufficient for\nminimality. This ordinary deduction explains exactly how much variational\ninformation has been extracted. For another admissible coefficient\n$\\beta$, let $\\nu=\\mu_\\beta$. The weighted identity above and\n$|G(s)|\\le1-s$ make both measure integrals absolutely convergent, even\nwhen either measure has infinite total mass. Fubini gives\n$$\n2D_+\\mathcal P(\\gamma)[\\beta-\\gamma]\n =\\int_{[0,1)}G(s)\\,\\nu(ds)\n  -\\int_{[0,1)}G(s)\\,\\mu(ds)\\ge0.\n$$\nHere the second integral is zero because $\\mu$ is concentrated on its\nrelative support. The control formula also shows that $\\mathcal P$ is\nconvex in its coefficient: for a fixed control, the terminal absolute\nvalue is convex in the coefficient and both penalties are linear, and taking\nthe supremum over the same control class preserves convexity. Therefore\n$$\n\\mathcal P(\\beta)-\\mathcal P(\\gamma)\n \\ge D_+\\mathcal P(\\gamma)[\\beta-\\gamma]\\ge0.\n$$\nThis proves the converse.","note":"Not in the Lean: only the forward direction exists. The converse and convexity of the Parisi functional in its coefficient have no declaration.","paper":"long","role":"statement","section":"3.5 The reach of the first-order criterion","start_byte":41723},{"anchor":"contact","at":"b43204-43526","end_byte":43526,"heading_id":"the-reach-of-the-first-order-criterion","lean":[],"lean_source":"statement","markdown":"The converse concerns $G$ built from the PDE and diffusion of the same\n$\\gamma$. It does not say that an arbitrary curve with these contact\nproperties comes from a Parisi solution. Nor does it prove full support:\nthe missing input is the strict curvature on an interval where the\ncoefficient is constant, established next.","note":"Not in the Lean: a remark on the scope of the converse, which has no Lean declaration.","paper":"long","role":"explanation","section":"3.5 The reach of the first-order criterion","start_byte":43204},{"anchor":"contact","at":"b42383-42643","end_byte":42643,"heading_id":"the-reach-of-the-first-order-criterion","lean":[],"lean_source":"statement","markdown":"The control formula also shows that $\\mathcal P$ is\nconvex in its coefficient: for a fixed control, the terminal absolute\nvalue is convex in the coefficient and both penalties are linear, and taking\nthe supremum over the same control class preserves convexity.","note":"Lean does not prove convexity of the functional in its coefficient, and the converse optimality criterion has no Lean.","paper":"long","role":"explanation","section":"3.5 The reach of the first-order criterion","start_byte":42383}]},"support.plateau_transfer":{"lean":[{"clause":"stripThird W γ T c t = ∫ ξ, (actualJet W γ t 4 ξ)^2-\n      12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2+6*c^2*(actualJet W γ t 2 ξ)^4 ∂W.law","end_line":49,"line":36,"lines":[38,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"Expands stripThird as the expected polynomial, in unscaled derivatives, so the identity is divided by c squared.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"},{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Third time derivative on a constant interval equals stripThird. The approximation rows follow; the Lean does not keep the value c.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":88,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sample_coercivity","note":"The 1/1000 inequality for one finite-step approximant at a sampled time, before the limit.","plain":"Coercivity at sampled grid times","proof_lines":[81,88],"relation":"part"},{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveCoeff","note":"The approximant: sampled gamma plus the mesh, so on a plateau its height is c plus the mesh and c is not kept.","plain":"Positive finite-step coefficient","relation":"part"},{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"The integrated polynomial passes to the limit at a continuity point of gamma, which is how a plateau's interior is reached.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"part"}],"short":"Plateau-preserving approximation","statements":[{"anchor":"fixed-time-coercivity","at":"b64845-65894","end_byte":65894,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":88,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sample_coercivity","note":"The finite-step inequality at sampled times. Lean builds no plateau at t: the height tends to c by continuity.","plain":"Coercivity at sampled grid times","proof_lines":[81,88],"relation":"proof"},{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveCoeff","note":"Lean's approximant: the grid sample of γ plus the mesh, with no plateau.","plain":"Positive finite-step coefficient","relation":"part"},{"end_line":118,"line":114,"lines":[114,115],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.positiveCoeff_sample_eq","note":"","plain":"Positive coefficient at the sampled index","proof_lines":[116,118],"relation":null}],"lean_source":"statement","markdown":"**Proof.** Fix $t<T<T'<1$ and a shrinking interval\n$[t-\\delta_n,t+\\delta_n)\\subset(0,T)$. Approximate $\\gamma$ by\n$$\nh_n=\\max\\{\\epsilon_n,\\min\\{K_n,\n                  \\eta_n\\lfloor\\gamma/\\eta_n\\rfloor\\}\\},\n$$\nwhere $0<\\epsilon_n<c<K_n$, $\\epsilon_n,\\eta_n\\downarrow0$ and\n$K_n\\uparrow\\infty$. Use $\\min(c,h_n)$ to the left of the interval,\n$c$ on it, and $\\max(c,h_n)$ to its right. Monotonicity makes these\npositive, nondecreasing, right-continuous finite-step coefficients\n$\\gamma_n$. Clipping toward $c$ outside the interval cannot increase\nthe error, and\n$$\n\\|\\gamma_n-\\gamma\\|_1\n\\le\\epsilon_n+\\eta_n+\\int_0^1(\\gamma-K_n)_+\\,ds\n +2\\delta_n\\sup_{|s-t|\\le\\delta_n}|\\gamma(s)-c|\n \\longrightarrow0.\n\\tag{4.3}\n$$\nIntegrability controls the capped tail and continuity at $t$ the last\nterm. Choose smooth terminal parameters $M_n\\ge\\max(n,K_n)$.\nThe terminal values then converge uniformly to $|x|$, and the coefficient\nmasses have a common bound. The finite-step inequality of §4.4 applies\nat $t$, where every $\\gamma_n$ equals the same positive $c$.","note":"Lean builds no plateau at t: it samples at grid times tending to t, and the height tends to c by continuity.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64845},{"anchor":"fixed-time-coercivity","at":"b64593-64764","end_byte":64764,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"q‴ on a constant interval equals stripThird, the case γ′=0. Identity (4.2) with its γ′ term is not stated.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"clause":"stripThird W γ T c t = ∫ ξ, (actualJet W γ t 4 ξ)^2-\n      12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2+6*c^2*(actualJet W γ t 2 ξ)^4 ∂W.law","end_line":49,"line":36,"lines":[38,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"stripThird is the expected polynomial C_γ, for c free.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"Differentiating the first identity therefore gives\n$$\nq'''(t)+2\\gamma'(t)\\mathbb E a^3=\\mathcal C_\\gamma(t).\n\\tag{4.2}\n$$\nOn a plateau the coefficient derivative vanishes.","paper":"long","role":"statement","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":64593},{"anchor":"curvature","at":"b18889-19201","end_byte":19201,"heading_id":"the-integrated-estimate","lean":[{"clause":"stripThird W γ T c t = ∫ ξ, (actualJet W γ t 4 ξ)^2-\n      12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2+6*c^2*(actualJet W γ t 2 ξ)^4 ∂W.law","end_line":49,"line":36,"lines":[38,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"Expands stripThird as the expected polynomial, in unscaled derivatives, so the identity is divided by c squared.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"},{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Third time derivative on a constant interval equals stripThird. The approximation rows follow; the Lean does not keep the value c.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"}],"lean_source":"statement","markdown":"All expectations in this section evaluate spatial functions at\n$(t,X_t)$. Starting from $c^2q'=\\mathbb E v^2$, Itô's formula gives\n$$\n\\begin{aligned}\nc^2q''&=\\mathbb E[w^2-2v^3],\\\\\nc^2q'''&=\\mathbb E[z^2-12vw^2+6v^4]\n       =c^2\\mathcal C_\\gamma(t).\n\\end{aligned}                                      \\tag{4}\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":18889},{"anchor":"curvature","at":"b25733-26315","end_byte":26315,"heading_id":"the-integrated-estimate","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit of (5) at a continuity point, through grid approximants. These keep no plateau near t, so the route differs.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"},{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"The limit involves only the jets; no limit of s is taken.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"part"}],"lean_source":"statement","markdown":"To obtain Lemma 2 for an integrable coefficient, choose positive\nmonotone step coefficients $\\gamma_n$ that keep the plateau near a\nfixed interior time, with $\\|\\gamma_n-\\gamma\\|_1\\to0$, and terminal\nparameters $M_n\\geq\\|\\gamma_n\\|_\\infty$ with $M_n\\to\\infty$.\nSpatial derivatives through order four then converge uniformly on a\nfixed strip below one, and the diffusions driven by the same Brownian\nmotion converge\n(see [OpenAI (2026a), Lemmas 2.1–2.2 and Proposition 5.1][FS]). Only\n$v,w,z$ remain in (5), so the auxiliary logarithmic derivative $s$\nneeds no convergence theorem.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":25733},{"anchor":"curvature-mechanism","at":"b51178-51308","end_byte":51308,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"q'' as the expected polynomial in c, a and b.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"q''' as the expected polynomial in c, a, b and d.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"Differentiating the PDE and applying Itô's rule gives\n\n$$\nq''=\\mathbb E[b^2-2ca^3],\\qquad\nq'''=\\mathbb E[d^2-12cab^2+6c^2a^4].\n$$","paper":"expository","role":"statement","section":"7.3 Where the strict-curvature input comes from","start_byte":51178},{"anchor":"curvature-transfer","at":"b54302-54654","end_byte":54654,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit inequality at a continuity point of γ. Lean samples heights tending to γ(t) and does not keep c fixed.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"},{"end_line":158,"line":154,"lines":[154,155],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveGamma_L1","note":"L1 approximation by positive steps.","plain":"Positive step functions converge in L¹","proof_lines":[156,158],"relation":"proof"},{"end_line":139,"line":115,"lines":[115,118],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_derivative_uniform","note":"Every order m, not only four.","plain":"Uniform derivative of the positive scheme","proof_lines":[119,139],"relation":"proof"},{"end_line":47,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_limit","note":"The two combined along almost every path.","plain":"Sampled jets converge almost surely","proof_lines":[37,47],"relation":"proof"}],"lean_source":"statement","markdown":"To reach an arbitrary admissible coefficient, fix a time inside the\nplateau and approximate in $L^1$ by step coefficients, keeping the same\nvalue $c$ near that time.\nUniform convergence of spatial derivatives through order four on a\nfixed strip below one, together with convergence of the coupled\npaths, passes the quantitative inequality to the limit.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54302},{"anchor":"curvature-mechanism","at":"b51132-51177","end_byte":51177,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":33,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actualJet","note":"The m-th spatial derivative of the value along the diffusion: m=2,3,4 give a, b, d.","plain":"Spatial derivatives along the diffusion","relation":"defines"}],"lean_source":"statement","markdown":"set $b=\\Phi_{xxx}$ and $d=\\Phi_{xxxx}$ there.","paper":"expository","role":"definition","section":"7.3 Where the strict-curvature input comes from","start_byte":51132},{"anchor":"curvature-mechanism","at":"b51310-51377","end_byte":51377,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"clause":"12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2","end_line":49,"line":36,"lines":[39,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"The negative middle term. That it blocks a pointwise argument is a remark, not in the Lean.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"The negative middle term rules out a pointwise positivity argument,","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":51310},{"anchor":"curvature-transfer","at":"b54302-54467","end_byte":54467,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":158,"line":154,"lines":[154,155],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveGamma_L1","note":"L1 approximation by positive steps.","plain":"Positive step functions converge in L¹","proof_lines":[156,158],"relation":"proof"},{"end_line":123,"line":120,"lines":[120,121],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.positiveCoeff_sample_limit","note":"Lean does not keep c fixed: the sampled heights tend to γ(t) at a continuity point.","plain":"Sampled coefficient converges to γ(t)","proof_lines":[122,123],"relation":"proof"}],"lean_source":"statement","markdown":"To reach an arbitrary admissible coefficient, fix a time inside the\nplateau and approximate in $L^1$ by step coefficients, keeping the same\nvalue $c$ near that time.","note":"The Lean proof samples and takes limits rather than freezing the plateau value.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54302},{"anchor":"curvature-transfer","at":"b54468-54605","end_byte":54605,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":139,"line":115,"lines":[115,118],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_derivative_uniform","note":"Every order m, not only four.","plain":"Uniform derivative of the positive scheme","proof_lines":[119,139],"relation":"proof"},{"end_line":27,"line":16,"lines":[16,18],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.positivePath_sample_limit","note":"The coupled paths, on the same Brownian system.","plain":"Approximating paths converge at sample times","proof_lines":[19,27],"relation":"proof"},{"end_line":47,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_limit","note":"The two combined along almost every path.","plain":"Sampled jets converge almost surely","proof_lines":[37,47],"relation":"proof"}],"lean_source":"statement","markdown":"Uniform convergence of spatial derivatives through order four on a\nfixed strip below one, together with convergence of the coupled\npaths,","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54468},{"anchor":"dependencies","at":"b12203-12303","end_byte":12303,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":37,"line":30,"lines":[30,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"First classical time derivative of the curvature moment.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"},{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"The second differentiation.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"}],"lean_source":"statement","markdown":"Inside a constant\ninterval, classical time differentiation permits the strict-curvature\ncalculation.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12203},{"anchor":"dependencies","at":"b11340-11439","end_byte":11439,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"The extra time smoothness, as used.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"}],"lean_source":"statement","markdown":"On an open constant interval,\nthe heat formula supplies the additional time smoothness used in §4.","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11340},{"anchor":"strip-recovery","at":"b28941-29285","end_byte":29285,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":196,"line":163,"lines":[163,168],"module":"OAI.Probability.SKSupport.Diffusion.DiffusionStability","name":"OAI.ZeroTemperatureSK.positivePath_error","note":"The pathwise bound on [0,T].","plain":"Coupling bound for an approximating path","proof_lines":[169,196],"relation":"part"}],"lean_source":"statement","markdown":"This is a deterministic bound under the coupling. The derivative bounds\none order higher ensure uniform continuity when derivatives are evaluated\nat these moving points. Every fixed polynomial in the bounded derivatives\ntherefore converges in expectation. This is precisely the convergence\nneeded for the integrated curvature expression in §4,","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":28941},{"anchor":"strip-recovery","at":"b29290-29358","end_byte":29358,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit involves only spatial jets of the value.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"none of the forward\ndensity scores or their quotients need converge.","paper":"long","role":"explanation","section":"2.4 Recovering the compact-strip estimates","start_byte":29290},{"anchor":"approximation-choices","at":"b29976-30280","end_byte":30280,"heading_id":"the-order-of-the-approximation-limits","lean":[{"end_line":124,"line":119,"lines":[119,119],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveCoeff_bound","note":"Support model. The terminal parameter bounds every step coefficient and grows with n. Steps are grid samples of γ, so none is held at c.","proof_lines":[120,124],"relation":"part"},{"end_line":243,"line":202,"lines":[202,203],"module":"OAI.Probability.SKValue.Equations.CoupledComparison","name":"OAI.SKValue.logCoshTerminal_backwardShape","note":"Scalar model. The log-cosh terminal has the backward shape when c≤M, which is why M must bound the coefficients.","proof_lines":[204,243],"relation":"part"}],"lean_source":"statement","markdown":"**Shape inequalities on a plateau.** Fix the height $c>0$ and a closed\ninterior neighbourhood. Choose step coefficients equal to $c$ there and\nsmooth terminal values with $M_m\\ge\\|\\gamma_m\\|_\\infty$ and\n$M_m\\to\\infty$. The coefficient and terminal approximations may be\nchosen as one compatible sequence.","note":"The Lean keeps no plateau: grid approximants are sampled at the floor of t and continuity passes to the limit. The terminal bound is there.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":29976},{"anchor":"approximation-choices","at":"b30859-31190","end_byte":31190,"heading_id":"the-order-of-the-approximation-limits","lean":[{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"The integrated jet polynomial passes to the limit along sampled grid approximants, dominated by common strip bounds. No density score appears.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"proof"},{"clause":"stripThird W γ T c t = ∫ ξ, (actualJet W γ t 4 ξ)^2-\n      12*c*actualJet W γ t 2 ξ*(actualJet W γ t 3 ξ)^2+6*c^2*(actualJet W γ t 2 ξ)^4 ∂W.law","end_line":49,"line":36,"lines":[38,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"Expands stripThird as the expected polynomial, in unscaled derivatives, so the identity is divided by c squared.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"},{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Third time derivative on a constant interval equals stripThird. The Lean does not keep the value c.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"}],"lean_source":"statement","markdown":"For the plateau transfer, tail estimates for a fixed approximant may depend\non that approximant. The passage to the limit uses only the integrated\npolynomial in the scaled second, third and fourth spatial derivatives of\nthe value function, together with the common strip bounds. The auxiliary\ndensity scores need no uniform bounds.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":30859},{"anchor":"curvature","at":"b44765-45017","end_byte":45017,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"end_line":188,"line":173,"lines":[173,180],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_feedback_coercivity","note":"The finite-step stage, for positive monotone steps and the feedback diffusion. Only the resulting inequality is stated; the correction is in the proof.","plain":"Finite-step coercivity along the diffusion","proof_lines":[181,188],"relation":"part"}],"lean_source":"statement","markdown":"The proof has two stages. In §§4.1–4.4 the coefficient is a positive\nmonotone finite-step function with smooth terminal data. Shape comparisons\nand boundary estimates justify the [polynomial correction](#curvature-algebra)\nand its zero expectation.","note":"Lean uses equal mesh widths: the steps are the cells of length h.","paper":"long","role":"explanation","section":"4. Strict curvature on a constant interval","start_byte":44765},{"anchor":"curvature","at":"b45018-45253","end_byte":45253,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit of the integrated inequality, in spatial jets only.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"In [§4.5](#plateau-limit), approximants\n$\\gamma_n,\\Phi_n,X^n$ approach an arbitrary admissible coefficient.\nOnly the integrated expression in spatial derivatives passes to the limit,\nand the auxiliary forward scores need not converge.","paper":"long","role":"explanation","section":"4. Strict curvature on a constant interval","start_byte":45018},{"anchor":"curvature","at":"b46099-46375","end_byte":46375,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[{"end_line":129,"line":120,"lines":[120,123],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.stripA_hasDerivAt_on_constant","note":"First classical derivative on the interval.","plain":"Curvature moment derivative on a plateau","proof_lines":[124,129],"relation":"proof"},{"end_line":141,"line":131,"lines":[131,134],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.stripSecond_hasDerivAt_on_constant","note":"The second differentiation.","plain":"q″ expression derivative on plateaus","proof_lines":[135,141],"relation":"proof"},{"end_line":59,"line":47,"lines":[47,50],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.hasDerivAt_of_right_on_Ioo","note":"Right derivatives that are continuous give classical ones.","plain":"From right derivatives to a derivative","proof_lines":[51,59],"relation":"proof"}],"lean_source":"statement","markdown":"These are interior-time identities. On a smaller constant interval the\nheat formula gives classical time smoothness and bounded spatial\nderivatives of sufficiently high order. Hence the local martingale terms\nhave zero mean, and the expectation differentiations are justified.","note":"Lean proves classical derivatives from continuous right derivatives, so it uses no zero-mean local martingale.","paper":"long","role":"explanation","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":46099},{"anchor":"fixed-time-coercivity","at":"b65896-66456","end_byte":66456,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":27,"line":16,"lines":[16,18],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.positivePath_sample_limit","note":"Paths converge at the sample times.","plain":"Approximating paths converge at sample times","proof_lines":[19,27],"relation":"part"},{"end_line":47,"line":35,"lines":[35,36],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_limit","note":"The jets converge, for each order m.","plain":"Sampled jets converge almost surely","proof_lines":[37,47],"relation":"part"}],"lean_source":"statement","markdown":"On a fixed strip containing $t$ and ending before one, compact-strip\nregularity gives uniform spatial convergence and bounds through order\nfour. The next derivative bounds give a common spatial Lipschitz\nconstant for those derivatives. Coupling with the same Brownian motion\nand applying Gronwall gives a deterministic bound\n$\\sup_{s\\le T}|X_s^n-X_s|\\to0$. If $D_k^n=\\partial_x^k\\Phi_n$\nand $D_k=\\partial_x^k\\Phi$, then, for $1\\le k\\le4$,\n\n$$\n|D_k^n(t,X_t^n)-D_k(t,X_t)|\n\\le\\|D_k^n-D_k\\|_{\\infty,[0,T]\\times\\mathbb R}\n   +C_{k,T}|X_t^n-X_t|\\longrightarrow0.\n$$","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":65896},{"anchor":"fixed-time-coercivity","at":"b66458-66712","end_byte":66712,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"The curvature integrand.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"part"},{"end_line":106,"line":90,"lines":[90,93],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleQuartic_integral_limit","note":"The quartic integrand.","plain":"Limit of the sampled quartic moment","proof_lines":[94,106],"relation":"part"},{"end_line":57,"line":54,"lines":[54,55],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleJet_bound","note":"The common bound for dominated convergence.","plain":"Uniform bound on sampled jets","proof_lines":[56,57],"relation":"proof"}],"lean_source":"statement","markdown":"After the nonnegative terms are discarded, the two surviving integrands\n$z_n^2-12v_nw_n^2+6v_n^4$ and $v_n^4$ are uniformly bounded,\nwith $v_n=cD_2^n$, $w_n=cD_3^n$ and $z_n=cD_4^n$. Dominated convergence therefore passes their expectations\nto the limit.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":66458},{"anchor":"fixed-time-coercivity","at":"b66714-67357","end_byte":67357,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[],"lean_source":"statement","markdown":"To make this quantitative, let $V,W,Z$ bound both the limiting and approximating\n$v,w,z$, and let $e_v,e_w,e_z$ bound their coupled evaluation errors.\nUniform spatial convergence and the path bound above give these errors\nwith $e_v,e_w,e_z\\to0$. Difference-of-powers identities give\n$$\n\\begin{aligned}\n|\\mathbb E\\mathcal C(v_n,w_n,z_n)-\\mathbb E\\mathcal C(v,w,z)|\n &\\le(12W^2+24V^3)e_v+24VW e_w+2Ze_z,\\\\\n|\\mathbb E v_n^4-\\mathbb E v^4|&\\le4V^3e_v,\n\\end{aligned}\n$$\nwhere $\\mathcal C(v,w,z)=z^2-12vw^2+6v^4$. These constants depend only\non the common strip and the coefficient mass bound, and in particular are\nindependent of the plateau width.","note":"Not in the Lean: it has no difference-of-powers bounds. Both limits pass by dominated convergence, so these constants do not appear.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":66714},{"anchor":"fixed-time-coercivity","at":"b67358-67616","end_byte":67616,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":143,"line":135,"lines":[140,143],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The limit passes only sampleQuartic and sampleCoercive, which are jets of the value.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"After division by $c^2$, this is (4.1).\n\nOnly spatial moments pass to the limit. The auxiliary quantities $s_n$ and\n$j_n$ have already disappeared after integration, so the transfer needs\nneither their convergence nor a uniform bound on their tail constants.","paper":"long","role":"explanation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":67358},{"anchor":"curvature","at":"b18299-18324","end_byte":18324,"heading_id":"the-integrated-estimate","lean":[{"end_line":141,"line":131,"lines":[131,134],"module":"OAI.Probability.SKSupport.Moments.ThirdFields","name":"OAI.ZeroTemperatureSK.stripSecond_hasDerivAt_on_constant","note":"Lean has no γ′. On a constant interval the q″ expression differentiates to the q‴ expression, so the identity's γ′ term never arises.","plain":"q″ expression derivative on plateaus","proof_lines":[135,141],"relation":"part"}],"lean_source":"statement","markdown":"On a plateau $\\gamma'=0$,","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":18299},{"anchor":"curvature","at":"b18661-18888","end_byte":18888,"heading_id":"the-integrated-estimate","lean":[{"end_line":99,"line":91,"lines":[91,94],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.compactJetField_generator_one","note":"The drift of a is −γa², which rescales to −v².","plain":"Itô generator of the curvature a","proof_lines":[95,99],"relation":"part"},{"end_line":109,"line":101,"lines":[101,104],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.compactJetField_generator_two","note":"The drift of b is −3γab, which rescales to −3vw. Lean states generators and leaves the diffusion coefficients implicit.","plain":"Itô generator of ∂ₓa","proof_lines":[105,109],"relation":"part"}],"lean_source":"statement","markdown":"and the diffusion has spatial generator\n$\\tfrac12\\partial_{xx}+r\\partial_x$. Differentiating in space and\napplying Itô's formula gives the drift and diffusion pairs\n$(0,v)$, $(-v^2,w)$ and $(-3vw,z)$ for $r,v,w$, respectively.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":18661},{"anchor":"curvature","at":"b18963-19201","end_byte":19201,"heading_id":"the-integrated-estimate","lean":[{"end_line":148,"line":139,"lines":[139,142],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.curvatureField_generator","note":"The first line of (4), unscaled: the generator of a² is b²−2ca³.","plain":"Itô generator of a²","proof_lines":[143,148],"relation":"proof"},{"end_line":163,"line":150,"lines":[150,154],"module":"OAI.Probability.SKSupport.Moments.JetFields","name":"OAI.ZeroTemperatureSK.secondCurvatureField_generator","note":"The second line of (4), unscaled.","plain":"Itô generator of (∂ₓa)² − 2ca³","proof_lines":[155,163],"relation":"proof"},{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"Expands stripThird as the expected polynomial, in unscaled derivatives, so the identity is divided by c squared.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"Starting from $c^2q'=\\mathbb E v^2$, Itô's formula gives\n$$\n\\begin{aligned}\nc^2q''&=\\mathbb E[w^2-2v^3],\\\\\nc^2q'''&=\\mathbb E[z^2-12vw^2+6v^4]\n       =c^2\\mathcal C_\\gamma(t).\n\\end{aligned}                                      \\tag{4}\n$$","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":18963},{"anchor":"curvature","at":"b19202-19292","end_byte":19292,"heading_id":"the-integrated-estimate","lean":[{"end_line":76,"line":75,"lines":[75,76],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.JetExpr.curvatureSlope_sourceA","note":"Scalar model. The second-order and b² drift terms, unscaled, at slope c.","plain":"Slope source, part without γ","proof_lines":[76,76],"relation":"part"},{"end_line":78,"line":77,"lines":[77,78],"module":"OAI.Probability.SKValue.Equations.PolynomialMoment","name":"OAI.SKValue.JetExpr.curvatureSlope_sourceP","note":"Scalar model. The terms multiplied by γ. Lean states the combination for b²−2ca³.","plain":"Slope source, γ part","proof_lines":[78,78],"relation":"part"}],"lean_source":"statement","markdown":"For example, the drifts of $w^2$ and $v^3$ are\n$z^2-6vw^2$ and $3vw^2-3v^4$, respectively.","paper":"short","role":"example","section":"3. The integrated estimate","start_byte":19202},{"anchor":"curvature","at":"b25733-25993","end_byte":25993,"heading_id":"the-integrated-estimate","lean":[{"end_line":98,"line":97,"lines":[97,98],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveCoeff","note":"Positive monotone grid coefficients: γ at grid points plus the mesh. No plateau is retained; Lean samples at the floor of t instead.","plain":"Positive finite-step coefficient","relation":"part"},{"end_line":158,"line":154,"lines":[154,155],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveGamma_L1","note":"The L¹ convergence of the step functions to γ.","plain":"Positive step functions converge in L¹","proof_lines":[156,158],"relation":"part"},{"end_line":124,"line":119,"lines":[119,119],"module":"OAI.Probability.SKSupport.Regularity.ParameterChords","name":"OAI.ZeroTemperatureSK.positiveCoeff_bound","note":"The terminal parameter positiveTerminal bounds every coefficient.","proof_lines":[120,124],"relation":"part"},{"end_line":88,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sample_coercivity","note":"Lean uses these coefficients here, at the sampled index.","plain":"Coercivity at sampled grid times","proof_lines":[81,88],"relation":"proof"}],"lean_source":"statement","markdown":"To obtain Lemma 2 for an integrable coefficient, choose positive\nmonotone step coefficients $\\gamma_n$ that keep the plateau near a\nfixed interior time, with $\\|\\gamma_n-\\gamma\\|_1\\to0$, and terminal\nparameters $M_n\\geq\\|\\gamma_n\\|_\\infty$ with $M_n\\to\\infty$.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":25733},{"anchor":"curvature","at":"b25994-26147","end_byte":26147,"heading_id":"the-integrated-estimate","lean":[{"end_line":139,"line":115,"lines":[115,118],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_derivative_uniform","note":"All orders, not only four.","plain":"Uniform derivative of the positive scheme","proof_lines":[119,139],"relation":"part"},{"end_line":69,"line":56,"lines":[56,58],"module":"OAI.Probability.SKSupport.Moments.PositiveGradientError","name":"OAI.ZeroTemperatureSK.positivePath_uniform_error","note":"Same Brownian motion, uniformly on the strip.","plain":"Approximating paths converge uniformly","proof_lines":[59,69],"relation":"part"}],"lean_source":"statement","markdown":"Spatial derivatives through order four then converge uniformly on a\nfixed strip below one, and the diffusions driven by the same Brownian\nmotion converge","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":25994},{"anchor":"curvature","at":"b26214-26315","end_byte":26315,"heading_id":"the-integrated-estimate","lean":[{"end_line":133,"line":108,"lines":[108,112],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","note":"The limit involves only the jets; no limit of s is taken.","plain":"Limit of the sampled coercive moment","proof_lines":[113,133],"relation":"part"}],"lean_source":"statement","markdown":"Only\n$v,w,z$ remain in (5), so the auxiliary logarithmic derivative $s$\nneeds no convergence theorem.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":26214}]},"support.positive_curvature":{"lean":[{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Gives positivity of the second derivative of q′ on the interval. The quantitative bound is not in its statement.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The fourth-moment lower bound, multiplied through by c squared, at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"}],"short":"Strict curvature, positive plateau","statements":[{"anchor":"curvature","at":"b43953-44349","end_byte":44349,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"Case c>0: the actual inequality, multiplied through by c², at a continuity point. stripThird is q''' on the interval.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Gives positivity of the second derivative of q′ on the interval. The quantitative bound is not in its statement.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":null}],"lean_source":"statement","markdown":"Let $\\gamma\\in\\mathcal U$ be\nconstant with value $c$ on an open interval $I\\subset(0,1)$, and let $X$\nbe its diffusion started at zero. Then\n\n$$\n\\begin{cases}\nq'''(t)\\ge \\dfrac1{1000c^2}\n  \\mathbb E\\bigl[(c\\Phi_{xx}(t,X_t))^4\\bigr]>0,&c>0,\\\\[4pt]\nq''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0,&c=0,\n\\end{cases}\\qquad t\\in I.\n$$\n\nThus $q'$ is strictly convex for $c>0$ and strictly increasing for\n$c=0$.","paper":"long","role":"statement","section":"4. Strict curvature on a constant interval","start_byte":43953},{"anchor":"fixed-time-coercivity","at":"b68652-69695","end_byte":69695,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"Case c>0: the actual inequality, multiplied through by c², at a continuity point. stripThird is q''' on the interval.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":90,"line":84,"lines":[84,86],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_pos","note":"","plain":"The q‴ expression is positive","proof_lines":[87,90],"relation":null},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Gives positivity of the second derivative of q′ on the interval. The quantitative bound is not in its statement.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"**Corollary 4.2 (positive plateaus).** Suppose now that $\\gamma=c>0$\non an open interval $I\\subset(0,1)$ and $t\\in I$. Then (4.2) gives\n$q'''(t)=\\mathcal C_\\gamma(t)$. To establish strictness for this\nlimiting solution, choose $b\\in I$ later than $t$ and put\n$\\psi=\\Phi_\\gamma(b,\\cdot)$. This is smooth, convex, even and a\nbounded distance from $|x|$, so $\\psi'$ cannot be constant. The\nCole–Hopf formula gives\n\n$$\n\\Phi_{xx}(t,x)\n=\\mathbb E^{\\rm tilt}_{t,x}\\psi''\n  +c\\operatorname{Var}^{\\rm tilt}_{t,x}(\\psi'),\n$$\n$$\n\\nu_{t,x}(dy)=\n\\frac{e^{c\\psi(y)}p_{b-t}(y-x)}\n {\\int_{\\mathbb R}e^{c\\psi(z)}p_{b-t}(z-x)\\,dz}\\,dy.\n$$\n\nThe tilted density is positive everywhere. Hence the tilted variance of\nthe nonconstant continuous function $\\psi'$ is strictly positive, while the\ntilted mean of $\\psi''$ is nonnegative. Thus $v(t,x)>0$ for every $x$, and\n$\\mathbb E v(t,X_t)^4>0$. Combining the limiting inequality with\nthe differentiated identity proves\n$q'''(t)\\ge(1000c^2)^{-1}\\mathbb E v(t,X_t)^4>0$.\nIn particular, $q'$ is strictly convex on $I$.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":68652},{"anchor":"gap-proof","at":"b12934-13348","end_byte":13348,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The fourth-moment lower bound, multiplied through by c squared, at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Gives positivity of the second derivative of q′ on the interval. The quantitative bound is not in its statement.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"The case c=0: positivity of q″ on the interval.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"**Lemma 2 (constant-interval curvature; [OpenAI (2026a),\nProposition 5.1][FS]).** Let $\\gamma\\in\\mathcal U$, with its\nassociated diffusion and moment $q$. If $\\gamma=c$ on an open\ninterval $I\\subset(0,1)$, called a plateau, then for $t\\in I$\n$$\nq'''(t)\\geq\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\quad\\text{if }c>0.                               \\tag{3}\n$$\nIf $c=0$, then $q''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0$.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":12934},{"anchor":"contact-and-gap","at":"b33656-33936","end_byte":33936,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"Multiplied through by c², at a continuity point of γ, which every plateau point is.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":46,"line":39,"lines":[39,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"Identifies stripThird with q''' on the interval.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":56,"line":51,"lines":[51,52],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"Strict positivity of the fourth moment.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"part"}],"lean_source":"statement","markdown":"**Strict-curvature input ([OpenAI (2026a), Proposition 5.1][support-curvature]).**\nFor any admissible coefficient equal to $c>0$ on an open interval,\nits moment curve satisfies, strictly inside that interval,\n\n$$\nq'''(t)\\ge\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\qquad(r<t<s).\n$$","paper":"expository","role":"statement","section":"5.3 Strict curvature creates an impossible integral","start_byte":33656},{"anchor":"contact-and-gap","at":"b34004-34091","end_byte":34091,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":90,"line":84,"lines":[84,90],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_pos","note":"The strict sign, from the quantitative bound. positive_gap_curvature uses only this.","plain":"The q‴ expression is positive","proof_lines":[87,90],"relation":"part"},{"end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The gap argument uses positivity, with no constant.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"The gap argument uses only the strict sign, whatever the constant in\nplace of $1/1000$.","paper":"expository","role":"explanation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34004},{"anchor":"contact-and-gap","at":"b34092-34293","end_byte":34293,"heading_id":"strict-curvature-creates-an-impossible-integral","lean":[{"end_line":56,"line":51,"lines":[51,56],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"Lean uses less: a not identically zero, and every interval charged by X_t, with no pointwise a>0.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"part"}],"lean_source":"statement","markdown":"The final inequality uses $a(t,x)>0$, proved by\nGaussian smoothing on the constant interval\n([OpenAI (2026a), Lemma 2.3][support-regularity]), whereas convexity of\n$\\Phi$ alone would give only $a\\ge0$.","note":"The Lean route to a positive fourth moment differs from the paper's.","paper":"expository","role":"explanation","section":"5.3 Strict curvature creates an impossible integral","start_byte":34092},{"anchor":"curvature-mechanism","at":"b51037-51177","end_byte":51177,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":61,"line":48,"lines":[48,50],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"No IsMinimizer hypothesis: γ is any order parameter constant on (a,b).","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"Take any admissible coefficient, minimising or not, with $\\gamma=c>0$\non an open interval, and set $b=\\Phi_{xxx}$ and $d=\\Phi_{xxxx}$ there.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":51037},{"anchor":"curvature-transfer","at":"b54655-55035","end_byte":55035,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":56,"line":51,"lines":[51,56],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"Also established for the limit, by a different argument: a is not identically zero and X_t charges every interval.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"part"}],"lean_source":"statement","markdown":"Strictness is\nthen established for the limit itself: the constant-coefficient heat\nformula expresses $a$ as a nonnegative mean plus $c$ times a strictly\npositive variance of a nonconstant slope under a positively weighted\nGaussian law. Hence $a(t,x)>0$ and its fourth moment is\npositive. Positivity of the approximants alone would not suffice, since\nit could be lost in the limit.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":54655},{"anchor":"dependencies","at":"b7449-7591","end_byte":7591,"heading_id":"outline-of-the-proof","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < (deriv^[2] (curvatureMoment W γ)) t","end_line":61,"line":48,"lines":[51,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Gives positivity of the second derivative of q′ on the interval. The quantitative bound is not in its statement.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The fourth-moment lower bound, multiplied through by c squared, at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"}],"lean_source":"statement","markdown":"Then, on a constant\ninterval, the shape inequalities and the rational certificate give the\nstrict curvature needed to contradict that contact.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":7449},{"anchor":"dependencies","at":"b9712-9811","end_byte":9811,"heading_id":"outline-of-the-proof","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The hypotheses are an order parameter and c>0; there is no IsMinimizer.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"**Strict curvature.** This step uses only admissibility and a positive\nconstant $c$ on an interval.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":9712},{"anchor":"dependencies","at":"b17853-18133","end_byte":18133,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The unconditional form, with q′ as curvatureMoment: its second derivative is positive throughout the interval.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"Chen's Proposition 4.2 has the conditional\ncrossing form $q''(t)=0\\Rightarrow q'''(t)>0$. The cited rational\ncertificate gives $q'''(t)>0$ throughout a positive constant interval.\nThese are different sufficient mechanisms, and this record makes no\npriority claim for full support.","paper":"long","role":"explanation","section":"2.3 Prior results and their roles","start_byte":17853},{"anchor":"curvature","at":"b44276-44349","end_byte":44349,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"end_line":391,"line":382,"lines":[389,391],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","note":"Strict convexity from the positive second derivative.","plain":"No bounded gap at positive height","proof_lines":[389,391],"relation":"proof"},{"end_line":405,"line":393,"lines":[400,402],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_zero_constant_gap","note":"Strict increase from the positive first derivative.","plain":"No bounded gap at height zero","proof_lines":[400,405],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $q'$ is strictly convex for $c>0$ and strictly increasing for\n$c=0$.","paper":"long","role":"statement","section":"4. Strict curvature on a constant interval","start_byte":44276},{"anchor":"curvature","at":"b44350-44493","end_byte":44493,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Only the interval hypothesis hc constrains γ, and only on Ioo a b; there is no IsMinimizer.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"The coefficient may jump outside $I$ and may be unbounded near\none, provided it is nonnegative, nondecreasing, right-continuous and\nintegrable.","paper":"long","role":"explanation","section":"4. Strict curvature on a constant interval","start_byte":44350},{"anchor":"fixed-time-coercivity","at":"b68771-68819","end_byte":68819,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"clause":"HasDerivAt (deriv (curvatureMoment W γ)) (stripThird W γ T c t) t","end_line":46,"line":39,"lines":[42,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"q‴ on the interval equals stripThird.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":49,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_eq_actual","note":"That is the expectation of the polynomial.","plain":"The q‴ expression as an expectation","proof_lines":[40,49],"relation":"part"}],"lean_source":"statement","markdown":"Then (4.2) gives\n$q'''(t)=\\mathcal C_\\gamma(t)$.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":68771},{"anchor":"fixed-time-coercivity","at":"b68820-69524","end_byte":69524,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"end_line":56,"line":51,"lines":[51,56],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"The quartic moment is positive at each t>0.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"proof"},{"end_line":44,"line":31,"lines":[31,44],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.curvature_not_identically_zero","note":"If curvature vanished the value would be constant, contradicting |x| ≤ value.","plain":"The curvature is not identically zero","proof_lines":[33,44],"relation":"proof"},{"end_line":101,"line":75,"lines":[75,79],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.actual_jet_even_moment_pos","note":"A continuous nonzero function has positive even moment.","plain":"Nonzero jets have positive even moments","proof_lines":[80,101],"relation":"proof"}],"lean_source":"statement","markdown":"To establish strictness for this\nlimiting solution, choose $b\\in I$ later than $t$ and put\n$\\psi=\\Phi_\\gamma(b,\\cdot)$. This is smooth, convex, even and a\nbounded distance from $|x|$, so $\\psi'$ cannot be constant. The\nCole–Hopf formula gives\n\n$$\n\\Phi_{xx}(t,x)\n=\\mathbb E^{\\rm tilt}_{t,x}\\psi''\n  +c\\operatorname{Var}^{\\rm tilt}_{t,x}(\\psi'),\n$$\n$$\n\\nu_{t,x}(dy)=\n\\frac{e^{c\\psi(y)}p_{b-t}(y-x)}\n {\\int_{\\mathbb R}e^{c\\psi(z)}p_{b-t}(z-x)\\,dz}\\,dy.\n$$\n\nThe tilted density is positive everywhere. Hence the tilted variance of\nthe nonconstant continuous function $\\psi'$ is strictly positive, while the\ntilted mean of $\\psi''$ is nonnegative. Thus $v(t,x)>0$ for every $x$, and\n$\\mathbb E v(t,X_t)^4>0$.","note":"Lean proves this differently: curvature need only be nonzero somewhere, and the law charges every interval. There is no tilted variance.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":68820},{"anchor":"fixed-time-coercivity","at":"b69525-69695","end_byte":69695,"heading_id":"coercivity-at-a-fixed-time-and-on-plateaus","lean":[{"clause":"(1/1000:ℝ)*(γ.val t)^4*(∫ ξ, (actualJet W γ t 2 ξ)^4 ∂W.law) ≤\n      (γ.val t)^2 * stripThird W γ T (γ.val t) t","end_line":82,"line":65,"lines":[67,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"Case c>0: the actual inequality, multiplied through by c², at a continuity point. stripThird is q''' on the interval.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":90,"line":84,"lines":[84,86],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_pos","note":"The inequality and positive quartic give positivity.","plain":"The q‴ expression is positive","proof_lines":[87,90],"relation":"proof"},{"end_line":61,"line":48,"lines":[59,61],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"Applied on the interval.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"Combining the limiting inequality with\nthe differentiated identity proves\n$q'''(t)\\ge(1000c^2)^{-1}\\mathbb E v(t,X_t)^4>0$.\nIn particular, $q'$ is strictly convex on $I$.","paper":"long","role":"derivation","section":"4.5 Coercivity at a fixed time and on plateaus","start_byte":69525},{"anchor":"curvature","at":"b43763-43885","end_byte":43885,"heading_id":"strict-curvature-on-a-constant-interval","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The conclusion concerns the time function curvatureMoment.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"The gap argument needs a property of the derivative $q'$, which is distinct\nfrom convexity of the spatial function $\\Phi$.","paper":"long","role":"explanation","section":"4. Strict curvature on a constant interval","start_byte":43763},{"anchor":"fixed-time-coercivity","at":"b70558-70703","end_byte":70703,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The abstract flux lemma; §12.1 states that the proof does not use it.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"end_line":167,"line":111,"lines":[111,114],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The finite-step result.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"proof"},{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The transfer to the original coefficient.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"proof"}],"lean_source":"statement","markdown":"An\nabstract polynomial or flux lemma assumes those analytic hypotheses, and\nthe finite-step comparisons and the transfer argument discharge them.","paper":"long","role":"explanation","section":"4.6 The zero interval as a separate case","start_byte":70558},{"anchor":"gaps","at":"b71714-71863","end_byte":71863,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"No support hypothesis.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"The zero case.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"Section 4 supplies curvature on constant intervals for every admissible\ncoefficient. These premises are available before any support gap is\nexcluded.","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":71714},{"anchor":"gap-proof","at":"b13089-13290","end_byte":13290,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":82,"line":65,"lines":[65,68],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The quantitative bound, multiplied through by c².","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"part"},{"end_line":46,"line":39,"lines":[39,42],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant","note":"On the plateau, q‴ is the strip moment stripThird.","plain":"Formula for q‴ on a plateau","proof_lines":[43,46],"relation":"part"},{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"The strict sign of q‴ on a plateau. Here deriv^[2] of curvatureMoment is q‴.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"}],"lean_source":"statement","markdown":"If $\\gamma=c$ on an open\ninterval $I\\subset(0,1)$, called a plateau, then for $t\\in I$\n$$\nq'''(t)\\geq\\frac{c^2}{1000}\\mathbb E[a(t,X_t)^4]>0\n\\quad\\text{if }c>0.                               \\tag{3}\n$$","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":13089},{"anchor":"gap-proof","at":"b13350-13418","end_byte":13418,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":61,"line":48,"lines":[48,51],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.positive_gap_curvature","note":"No IsMinimizer among the hypotheses.","plain":"q‴ positive on positive plateaus","proof_lines":[52,61],"relation":"part"},{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"}],"lean_source":"statement","markdown":"The lemma holds for every admissible coefficient, minimising or not.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":13350},{"anchor":"curvature","at":"b26316-26344","end_byte":26344,"heading_id":"the-integrated-estimate","lean":[{"end_line":82,"line":65,"lines":[69,82],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_coercive","note":"The fourth-moment lower bound, multiplied through by c squared, at a continuity point.","plain":"Coercive lower bound, q‴ expression","proof_lines":[69,82],"relation":"proof"},{"end_line":90,"line":84,"lines":[84,90],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripThird_pos","note":"Lean only divides for the sign; the bound stays multiplied by c².","plain":"The q‴ expression is positive","proof_lines":[87,90],"relation":"proof"}],"lean_source":"statement","markdown":"Dividing by $c^2$ gives (3),","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26316},{"anchor":"curvature","at":"b26349-26401","end_byte":26401,"heading_id":"the-integrated-estimate","lean":[{"end_line":56,"line":51,"lines":[51,56],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualQuartic_pos","note":"Lean does not use Φ_xx>0. It uses a nonzero value of a somewhere and positive diffusion mass on intervals.","plain":"Fourth curvature moment is positive","proof_lines":[53,56],"relation":"proof"}],"lean_source":"statement","markdown":"strictness follows from $\\Phi_{xx}>0$ on the plateau","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26349}]},"support.rational_certificate":{"lean":[{"clause":"z^2 - 12*R*v*b^2 + 6*v^4 =\n      (z+Q R K v b)^2 + (z+b*v-2*R*b)*T R v j +\n      weightedSquares R K v b (b-2*v) j + WF₂ R K v b j z L +\n      L*(x₀*v^2+h*b*v) + remainder R K v (b-2*v) j","end_line":69,"line":62,"lines":[63,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The polynomial identity for real variables, proved by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"clause":"(1661507 / 1521000000 : ℝ) * v^4 ≤ remainder R K v B j","end_line":105,"line":92,"lines":[94,94],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound","note":"With nonnegative variables, the remainder is at least the v^4 coefficient times v^4.","plain":"Certificate remainder is at least c·v⁴","proof_lines":[95,105],"relation":"part"},{"clause":"(1 / 1000 : ℝ) < 1661507 / 1521000000","end_line":125,"line":124,"lines":[124,124],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"That coefficient exceeds 1/1000.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"}],"short":"Exact rational certificate","statements":[{"anchor":"curvature","at":"b61768-61914","end_byte":61914,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"clause":"z^2 - 12*R*v*b^2 + 6*v^4 =\n      (z+Q R K v b)^2 + (z+b*v-2*R*b)*T R v j +\n      weightedSquares R K v b (b-2*v) j + WF₂ R K v b j z L +\n      L*(x₀*v^2+h*b*v) + remainder R K v (b-2*v) j","end_line":69,"line":62,"lines":[63,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The identity, for arbitrary reals; the proof is ring normalisation.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"states"}],"lean_source":"statement","markdown":"Exact expansion, with $b=B+2v$, gives\n\n$$\nz^2-12Rvb^2+6v^4\n =(z+Q)^2+S_1T_1+\\sum_{i=0}^4p_iP_i\n   +\\mathcal W(F_2)+L(x_0v^2+h_0bv)+P_{\\rm rem}.\n$$","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":61768},{"anchor":"curvature","at":"b60590-60787","end_byte":60787,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":21,"line":21,"lines":[21,21],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.Q","note":"Q.","plain":"Certificate polynomial Q","relation":"defines"},{"end_line":22,"line":22,"lines":[22,22],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.T","note":"T₁.","plain":"Certificate polynomial T","relation":"defines"},{"end_line":25,"line":25,"lines":[25,25],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂","note":"F₂, with its terms in b split off as F₁.","plain":"Certificate polynomial F₂","relation":"defines"}],"lean_source":"statement","markdown":"Define\n\n$$\nQ=Avb+Cv^2+kKv-2kRb+T_0Rv,\\qquad T_1=v(g_0R+h_0j),\n$$\n$$\n\\begin{aligned}\nF_2={}&(Av-2kR)b^2\n +b\\bigl(2Cv^2+2kKv+(g_0+2T_0)Rv+h_0jv\\bigr)\\\\\n &+x_0v^2j+yv^3+mKv^2+n_0Rv^2.\n\\end{aligned}\n$$","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":60590},{"anchor":"curvature","at":"b62648-62931","end_byte":62931,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Keeps the constant 1661507/1521000000, with vanishing flux as hypotheses. Written with R=r^2 and b=-w/r.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"states"},{"clause":"(1661507 / 1521000000 : ℝ) * v^4 ≤ remainder R K v B j","end_line":105,"line":92,"lines":[94,94],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound","note":"With nonnegative variables, the remainder is at least the v^4 coefficient times v^4.","plain":"Certificate remainder is at least c·v⁴","proof_lines":[95,105],"relation":"part"},{"clause":"(1 / 1000 : ℝ) < 1661507 / 1521000000","end_line":125,"line":124,"lines":[124,124],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"That coefficient exceeds 1/1000.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"},{"clause":"z^2 - 12*R*v*b^2 + 6*v^4 =\n      (z+Q R K v b)^2 + (z+b*v-2*R*b)*T R v j +\n      weightedSquares R K v b (b-2*v) j + WF₂ R K v b j z L +\n      L*(x₀*v^2+h*b*v) + remainder R K v (b-2*v) j","end_line":69,"line":62,"lines":[63,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The polynomial identity for real variables, proved by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"}],"lean_source":"statement","markdown":"All square prefactors $Rj,RB,Kv,vB,1$ are nonnegative, as are\n$S_1T_1$, $L(x_0v^2+h_0bv)$ and every remainder monomial. Integration\nremoves only $\\mathcal W(F_2)$, giving\n\n$$\n\\mathbb E_f[z^2-12vw^2+6v^4]\n \\ge\\frac{1661507}{1521000000}\\mathbb E_fv^4\n \\ge\\frac1{1000}\\mathbb E_fv^4.\n$$","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":62648},{"anchor":"curvature","at":"b24108-24337","end_byte":24337,"heading_id":"the-integrated-estimate","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The decomposition. Every term except WF₂ and the v⁴ term of remainder makes up the paper R.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"end_line":25,"line":25,"lines":[25,25],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂","note":"","plain":"Certificate polynomial F₂","relation":"defines"},{"end_line":125,"line":124,"lines":[124,125],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"That coefficient exceeds 1/1000.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Lemma 4.1][FS] supplies an explicit polynomial $F_2$\nand a decomposition\n$$\nz^2-12vw^2+6v^4\n=\\mathcal W(F_2)+\\kappa v^4+\\mathcal R,\n\\qquad\n\\mathcal R\\geq0,\\qquad\n\\kappa=\\frac{1661507}{1521000000}>\\frac1{1000}.\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":24108},{"anchor":"curvature-transfer","at":"b53863-53977","end_byte":53977,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The identity, proved by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"end_line":122,"line":107,"lines":[107,111],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity","note":"The second inequality, after all the terms are shown nonnegative.","plain":"Pointwise coercivity inequality","proof_lines":[112,122],"relation":"part"},{"end_line":105,"line":92,"lines":[92,105],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound","note":"The remainder bound, by positivity on an explicit sum.","plain":"Certificate remainder is at least c·v⁴","proof_lines":[95,105],"relation":"part"},{"end_line":125,"line":124,"lines":[124,125],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"1661507/1521000000 exceeds 1/1000.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"}],"lean_source":"statement","markdown":"A rational polynomial identity gives the second inequality by writing\nthe remainder as a sum of nonnegative terms.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53863},{"anchor":"curvature","at":"b57972-58058","end_byte":58058,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":143,"line":143,"lines":[143,143],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.certR","note":"R, with certK and certB on lines 144 and 148.","plain":"Certificate variable R = r²","relation":"defines"},{"end_line":150,"line":150,"lines":[150,150],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.certL","note":"L.","plain":"Certificate variable L = −r ∂ₓ²s","relation":"defines"},{"clause":"certZ r x+certB r x*certV r x-2*certR r x*certB r x","end_line":214,"line":181,"lines":[188,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"S₁ has no name in Lean; it appears here.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"}],"lean_source":"statement","markdown":"$$\nR=r^2,\\quad K=rs,\\quad b=-w/r,\\quad B=b-2v,\n\\quad L=-rs_{xx},\\quad S_1=z+bv-2Rb.\n$$","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":57972},{"anchor":"curvature","at":"b58122-58213","end_byte":58213,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"All seven signs, on the positive half-line, from BackwardShape, forward concavity and the ratio bound.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"}],"lean_source":"statement","markdown":"The\nprevious comparisons give\n\n$$\nR,K,v,B,j,L,S_1\\ge0,\\qquad b=B+2v\\ge0,\\qquad Rb^2=w^2,\n$$","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":58122},{"anchor":"curvature","at":"b58605-59060","end_byte":59060,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":104,"line":102,"lines":[102,103],"module":"OAI.Probability.SKSupport.Density.GaussianDecay","name":"OAI.ZeroTemperatureSK.Heat.quotientExtension_continuous","note":"Support model. A quotient by r extends continuously across zero, using dslope. Evenness of the extension and j(0)=0, S₁(0)=0 are not stated.","proof_lines":[104,104],"relation":"part"},{"end_line":100,"line":96,"lines":[96,96],"module":"OAI.Probability.SKSupport.Density.GaussianDecay","name":"OAI.ZeroTemperatureSK.Heat.BackwardShape.dslope_nonzero","note":"Support model. The simple zero of r: its slope quotient never vanishes, so the division is sound.","proof_lines":[97,100],"relation":"part"},{"end_line":115,"line":98,"lines":[98,99],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.cert_flux_continuous","note":"Support model. The certificate flux is continuous at zero, so zero is no omitted boundary of the half-line integral.","plain":"The certificate flux is continuous","proof_lines":[100,115],"relation":"part"},{"end_line":102,"line":55,"lines":[55,59],"module":"OAI.Probability.SKValue.Coercivity.ZeroFlux","name":"OAI.SKValue.ExpGrowth.div_odd_monotone","note":"Scalar model. Division by the same simple zero: its proof takes the limit g'(0)/r'(0) at zero and keeps exponential growth.","proof_lines":[60,102],"relation":"part"}],"lean_source":"statement","markdown":"Reflection extends these signs. The quotients were defined for $x>0$,\nwhere $r>0$ and $v>0$ by the tilted mean-and-variance argument in\n[§4.3](#shape-comparisons). At zero, $r$ and $w$ are odd,\n$r_x(0)=v(0)>0$, and division by $r$ is therefore division by a function\nwith a simple zero. This gives the smooth even extension\n$b(0)=-z(0)/v(0)$, and the other comparison identities give $j(0)=0$ and\n$S_1(0)=0$. Thus the integration has no boundary at zero.","note":"The support model extends the quotients across zero with dslope, then integrates on x>0, where the flux vanishes at zero, and doubles.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":58605},{"anchor":"curvature","at":"b60122-60588","end_byte":60588,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":10,"line":10,"lines":[10,10],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.A","note":"The ten constants are lines 10 to 19 of this file, one abbreviation each; h₀ is h.","plain":"Certificate constant A = 1/13","relation":"defines"}],"lean_source":"statement","markdown":"The rational identity below gives a positive remainder,\nand finite arithmetic verifies this third step. Finally,\n[§4.5](#plateau-limit) passes the resulting expectation inequality to the\noriginal coefficient.\n\nSet\n\n$$\n(A,C,k,T_0,g_0,h_0)=\n\\left(\\frac1{13},\\frac{53}{24},\\frac{19}{16},\\frac{73}{48},\n      \\frac{2331}{10000},\\frac{87}{2500}\\right),\n$$\n$$\n(x_0,y,m,n_0)=\n\\left(\\frac{4663}{5000},\\frac{6853}{10000},\n      -\\frac{1051}{400},-\\frac{1073}{500}\\right).\n$$","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":60122},{"anchor":"curvature","at":"b60789-61000","end_byte":61000,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":52,"line":52,"lines":[52,52],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂b","note":"The b-derivative, written as 2Q+T.","plain":"Partial derivative of F₂ in b","relation":"defines"}],"lean_source":"statement","markdown":"The terms containing $z$ motivate this choice. Indeed,\n$$\n\\begin{aligned}\n\\partial_bF_2\n &=2(Av-2kR)b+2Cv^2+2kKv+(g_0+2T_0)Rv+h_0jv\\\\\n &=2Q+T_1,\\\\\n{}[z]\\mathcal W(F_2)&=-\\partial_bF_2=-(2Q+T_1).\n\\end{aligned}\n$$","paper":"long","role":"derivation","section":"4.4 The polynomial correction with zero expectation","start_byte":60789},{"anchor":"curvature","at":"b61292-61422","end_byte":61422,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":53,"line":53,"lines":[53,53],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂j","note":"∂_j F₂.","plain":"Partial derivative of F₂ in j","relation":"defines"}],"lean_source":"statement","markdown":"The other favourable derivative is\n$\\partial_jF_2=x_0v^2+h_0bv\\ge0$, so the $-L$ in $\\mathcal Dj$\nleaves a nonnegative correction.","paper":"long","role":"derivation","section":"4.4 The polynomial correction with zero expectation","start_byte":61292},{"anchor":"curvature","at":"b61424-61766","end_byte":61766,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":27,"line":27,"lines":[27,27],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.P₀","note":"P₀; P₁ to P₄ are lines 28 to 31.","plain":"Certificate term P₀","relation":"defines"},{"end_line":35,"line":33,"lines":[33,35],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.weightedSquares","note":"The weighted sum with these five weights.","plain":"Weighted squares in the certificate","relation":"defines"}],"lean_source":"statement","markdown":"Put\n\n$$\n\\begin{aligned}\nP_0&=Rj(b-7v)^2,&\nP_1&=RB(b-\\tfrac73v)^2,\\\\\nP_2&=Kv(b-\\tfrac92v)^2,&\nP_3&=vB(K+\\tfrac14R-\\tfrac17v)^2,\\\\\nP_4&=(vb-\\tfrac{14}{5}v^2-\\tfrac{13}{4}Kv+\\tfrac43Rv)^2,\n\\end{aligned}\n$$\n$$\n(p_0,p_1,p_2,p_3,p_4)=\n\\left(\\frac{337}{10000},\\frac{183}{2500},\\frac{411}{10000},\n      \\frac{11461}{5000},\\frac{647}{10000}\\right).\n$$","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":61424},{"anchor":"curvature","at":"b61916-62646","end_byte":62646,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":45,"line":37,"lines":[37,45],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder","note":"The same 22 monomials and coefficients.","plain":"Certificate remainder","relation":"defines"}],"lean_source":"statement","markdown":"The table gives the entire remainder, and every unlisted coefficient is\nzero.\n\n| Monomial | Coefficient | Monomial | Coefficient |\n| --- | ---: | --- | ---: |\n| $R^2v^2$ | $223/160000$ | $R^2vB$ | $4007/240000$ |\n| $R^2B^2$ | $1097/120000$ | $RKv^2$ | $83/80000$ |\n| $RKvB$ | $3343/24000$ | $Rv^3$ | $16259/23400000$ |\n| $Rv^2B$ | $30179/5460000$ | $Rv^2j$ | $26/625$ |\n| $RvB^2$ | $107/15000$ | $RvBj$ | $3/2500$ |\n| $RB^3$ | $121/32500$ | $RB^2j$ | $11/10000$ |\n| $K^2v^2$ | $579/20000$ | $K^2vB$ | $207/2500$ |\n| $Kv^3$ | $911/975000$ | $Kv^2B$ | $349/420000$ |\n| $Kv^2j$ | $5011/5000$ | $KvB^2$ | $133/130000$ |\n| $KvBj$ | $87/2500$ | $v^4$ | $1661507/1521000000$ |\n| $v^3B$ | $633799/621075000$ | $v^2B^2$ | $10657/1690000$ |","paper":"long","role":"definition","section":"4.4 The polynomial correction with zero expectation","start_byte":61916},{"anchor":"curvature","at":"b62648-62767","end_byte":62767,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":122,"line":107,"lines":[107,111],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity","note":"The nonnegativity assembled, under the same sign hypotheses.","plain":"Pointwise coercivity inequality","proof_lines":[112,122],"relation":"part"},{"end_line":105,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder_lower_bound","note":"The remainder keeps its v⁴ term.","plain":"Certificate remainder is at least c·v⁴","proof_lines":[95,105],"relation":"part"}],"lean_source":"statement","markdown":"All square prefactors $Rj,RB,Kv,vB,1$ are nonnegative, as are\n$S_1T_1$, $L(x_0v^2+h_0bv)$ and every remainder monomial.","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":62648},{"anchor":"curvature","at":"b62768-62931","end_byte":62931,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The integrated form, for a flux with zero limits.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"},{"end_line":202,"line":196,"lines":[196,198],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral","note":"Replacing the constant by 1/1000.","plain":"From sharp constant to 1/1000","proof_lines":[199,202],"relation":"part"}],"lean_source":"statement","markdown":"Integration\nremoves only $\\mathcal W(F_2)$, giving\n\n$$\n\\mathbb E_f[z^2-12vw^2+6v^4]\n \\ge\\frac{1661507}{1521000000}\\mathbb E_fv^4\n \\ge\\frac1{1000}\\mathbb E_fv^4.\n$$","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":62768},{"anchor":"curvature","at":"b62933-63017","end_byte":63017,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":125,"line":124,"lines":[124,124],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"Lean states only the strict inequality.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"}],"lean_source":"statement","markdown":"The surplus of the first coefficient over $1/1000$ is exactly\n$140507/1521000000>0$.","paper":"long","role":"statement","section":"4.4 The polynomial correction with zero expectation","start_byte":62933},{"anchor":"curvature","at":"b63225-63466","end_byte":63466,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"This estimate with both sides still multiplied by c², at the limit. Assumes positive time and continuity at t.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"},{"end_line":25,"line":19,"lines":[19,20],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.quarticMoment_scaled","note":"","plain":"Quartic integrand in terms of jets","proof_lines":[21,25],"relation":null},{"end_line":35,"line":27,"lines":[27,29],"module":"OAI.Probability.SKSupport.Density.CoerciveMoments","name":"OAI.ZeroTemperatureSK.coercivePolynomial_scaled","note":"","plain":"Coercive integrand in terms of jets","proof_lines":[30,35],"relation":null}],"lean_source":"statement","markdown":"Dividing the integrated inequality by $c^2$ returns\n$$\n\\mathbb E[\\Phi_{xxxx}^2-12c\\Phi_{xx}\\Phi_{xxx}^2\n                  +6c^2\\Phi_{xx}^4]\n\\ge\\frac{c^2}{1000}\\mathbb E\\Phi_{xx}^4.\n$$\nThis is the unscaled estimate needed by the gap argument.","note":"Lean does not divide: it keeps c⁴ and c² on the two sides until stripThird_coercive.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":63225},{"anchor":"verification","at":"b252488-252562","end_byte":252562,"heading_id":"source-checks-and-finite-calculation","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The same identity, as a Lean theorem.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"}],"lean_source":"statement","markdown":"The retained polynomial computation\nin §4 uses exact rational arithmetic.","paper":"long","role":"explanation","section":"13. Source checks and finite calculation","start_byte":252488},{"anchor":"verification","at":"b252563-252624","end_byte":252624,"heading_id":"source-checks-and-finite-calculation","lean":[{"clause":"z^2 - 12*R*v*b^2 + 6*v^4 =\n      (z+Q R K v b)^2 + (z+b*v-2*R*b)*T R v j +\n      weightedSquares R K v b (b-2*v) j + WF₂ R K v b j z L +\n      L*(x₀*v^2+h*b*v) + remainder R K v (b-2*v) j","end_line":69,"line":62,"lines":[63,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The full expansion, closed by ring, so the remainder's coefficients match exactly.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"states"},{"end_line":45,"line":37,"lines":[37,45],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.remainder","note":"The 22 monomials.","plain":"Certificate remainder","relation":"defines"}],"lean_source":"statement","markdown":"Its expansion matches\nall 22 positive remainder\ncoefficients.","paper":"long","role":"statement","section":"13. Source checks and finite calculation","start_byte":252563},{"anchor":"verification","at":"b252625-252700","end_byte":252700,"heading_id":"source-checks-and-finite-calculation","lean":[{"end_line":125,"line":124,"lines":[124,124],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.coercivity_constant","note":"Only the strict inequality.","plain":"The constant exceeds 1/1000","proof_lines":[125,125],"relation":"part"}],"lean_source":"statement","markdown":"The surplus of the $v^4$ coefficient above $1/1000$ is\n$140507/1521000000$.","paper":"long","role":"statement","section":"13. Source checks and finite calculation","start_byte":252625},{"anchor":"dependencies","at":"b18434-18519","end_byte":18519,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"That identity.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"}],"lean_source":"statement","markdown":"The\nrational coercive identity is the step that distinguishes the full-support\nproof.","paper":"long","role":"mention","section":"2.3 Prior results and their roles","start_byte":18434},{"anchor":"curvature","at":"b59967-60331","end_byte":60331,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":214,"line":181,"lines":[181,188],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"The signs.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"part"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Zero integral and positive remainder together.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":143,"line":135,"lines":[135,139],"module":"OAI.Probability.SKSupport.Density.SampleJets","name":"OAI.ZeroTemperatureSK.actual_coercivity","note":"The transfer to the original coefficient.","plain":"Spatial coercivity at fixed time","proof_lines":[140,143],"relation":"part"}],"lean_source":"statement","markdown":"The curvature estimate rests on four separate steps. The comparisons\nsupply the signs above, and the tail estimates make the correction\nintegrate to zero. The rational identity below gives a positive remainder,\nand finite arithmetic verifies this third step. Finally,\n[§4.5](#plateau-limit) passes the resulting expectation inequality to the\noriginal coefficient.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":59967},{"anchor":"curvature","at":"b63018-63224","end_byte":63224,"heading_id":"the-polynomial-correction-with-zero-expectation","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The finite arithmetic, by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The application: its hypotheses are the shapes and the density derivative.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"The rational identity and the positivity of the\ntable reduce to finite arithmetic. Their application to this diffusion\nalso uses the signs, quotient behaviour, integrability and vanishing\nflux proved above.","paper":"long","role":"explanation","section":"4.4 The polynomial correction with zero expectation","start_byte":63018},{"anchor":"obligations","at":"b251134-251193","end_byte":251193,"heading_id":"logical-limits-of-the-argument","lean":[],"lean_source":"statement","markdown":"and an algebraic\ncertificate leaves its analytic hypotheses","note":"Not in the Lean: a remark. integrated_coercivity_of_flux takes the analytic hypotheses as arguments.","paper":"long","role":"explanation","section":"12.4 Logical limits of the argument","start_byte":251134},{"anchor":"curvature","at":"b23764-24107","end_byte":24107,"heading_id":"the-integrated-estimate","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The square (z+Q)² absorbs z², and the terms linear in z cancel through WF₂. Proof: unfold and ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"end_line":25,"line":25,"lines":[25,25],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂","note":"","plain":"Certificate polynomial F₂","relation":"defines"}],"lean_source":"statement","markdown":"What remains is algebra under these analytic restrictions. The\ncorrection polynomial is chosen in two steps: first arrange a square\nand other nonnegative terms containing the highest derivative $z$, then\ncancel their terms linear in $z$ with the weighted divergence. Only\nlower derivatives and the shape quantities then remain in the residual.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":23764},{"anchor":"curvature","at":"b24338-24421","end_byte":24421,"heading_id":"the-integrated-estimate","lean":[{"end_line":25,"line":25,"lines":[25,25],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.F₂","note":"x₀ and y are positive, m and n₀ negative, so the coefficients have both signs.","plain":"Certificate polynomial F₂","relation":"defines"},{"end_line":194,"line":165,"lines":[182,183],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"Only the integral of WF₂ is used.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"proof"}],"lean_source":"statement","markdown":"The coefficients of $F_2$ have both signs. Since the correction\nintegrates to zero,","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":24338},{"anchor":"curvature","at":"b24381-24552","end_byte":24552,"heading_id":"the-integrated-estimate","lean":[{"end_line":122,"line":107,"lines":[113,122],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity","note":"Each term of the identity is shown nonnegative in turn.","plain":"Pointwise coercivity inequality","proof_lines":[112,122],"relation":"proof"}],"lean_source":"statement","markdown":"Since the correction\nintegrates to zero, the signs of $F_2$ and $\\mathcal W(F_2)$ play no\nrole, whereas the shape inequalities make every term of $\\mathcal R$\nnonnegative.","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":24381},{"anchor":"curvature","at":"b24553-24692","end_byte":24692,"heading_id":"the-integrated-estimate","lean":[{"end_line":27,"line":27,"lines":[27,27],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.P₀","note":"R j (b−7v)², with R=r² and b=−w/r, and weight 337/10000 in weightedSquares.","plain":"Certificate term P₀","relation":"defines"},{"end_line":22,"line":22,"lines":[22,22],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.T","note":"g₀Rv+hjv, the second bracket.","plain":"Certificate polynomial T","relation":"defines"},{"clause":"(z+b*v-2*R*b)*T R v j","end_line":69,"line":62,"lines":[64,64],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"z+bv−2Rb is 2rH_x; see cert_halfLine_signs.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"}],"lean_source":"statement","markdown":"Two of its actual terms are\n$$\n\\frac{337}{10000}\\,r^2j\\,(-w/r-7v)^2,\n\\qquad\n(2rH_x)v\\left(\\frac{2331}{10000}r^2+\\frac{87}{2500}j\\right).\n$$","paper":"short","role":"statement","section":"3. The integrated estimate","start_byte":24553},{"anchor":"curvature","at":"b24693-24802","end_byte":24802,"heading_id":"the-integrated-estimate","lean":[{"end_line":214,"line":181,"lines":[195,199],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"j≥0 from the Wronskian.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"proof"},{"end_line":214,"line":181,"lines":[207,214],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"z+bv−2Rb≥0 from the backward Wronskian, which is r²H_x.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"proof"}],"lean_source":"statement","markdown":"The first needs the joint comparison $j\\geq0$, and the second also\nuses $H_x\\geq0$ on the positive half-line.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":24693},{"anchor":"curvature","at":"b24803-24862","end_byte":24862,"heading_id":"the-integrated-estimate","lean":[{"end_line":162,"line":106,"lines":[154,160],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"Lean does not reflect the signs. It proves the half-line inequality and doubles it, both integrands being even.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"lean_source":"statement","markdown":"Reflection preserves\nthese signs on the negative half-line.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":24803},{"anchor":"curvature","at":"b24863-25021","end_byte":25021,"heading_id":"the-integrated-estimate","lean":[{"end_line":214,"line":181,"lines":[200,206],"module":"OAI.Probability.SKSupport.Density.FiniteForwardRatio","name":"OAI.ZeroTemperatureSK.Heat.cert_halfLine_signs","note":"b−2v≥0 from H≤0, and L≥0 from concavity of s.","plain":"Seven signs on the half-line","proof_lines":[189,214],"relation":"proof"}],"lean_source":"statement","markdown":"The remaining terms use\n$H\\leq0$, forward concavity and the other shape bounds, so the\npositivity of the decomposition depends on the shapes the PDE supplies.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":24863},{"anchor":"curvature","at":"b24950-25021","end_byte":25021,"heading_id":"the-integrated-estimate","lean":[{"end_line":122,"line":107,"lines":[107,112],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.pointwise_coercivity","note":"Seven sign hypotheses, which cert_halfLine_signs supplies.","plain":"Pointwise coercivity inequality","proof_lines":[112,122],"relation":"part"}],"lean_source":"statement","markdown":"positivity of the decomposition depends on the shapes the PDE supplies.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":24950},{"anchor":"curvature","at":"b25253-25437","end_byte":25437,"heading_id":"the-integrated-estimate","lean":[{"end_line":69,"line":62,"lines":[62,66],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.rational_identity","note":"The polynomial identity for real variables, proved by ring.","plain":"The rational certificate identity","proof_lines":[67,69],"relation":"part"},{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The integration, with zero flux as hypotheses hbot and htop.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"}],"lean_source":"statement","markdown":"The full polynomial and\nits decomposition are in [Lemma 4.1 and Table 1][certificate], and\nProposition 4.2 integrates them using the signs and zero boundary\nflux supplied by Lemma 3.4.","paper":"short","role":"mention","section":"3. The integrated estimate","start_byte":25253}]},"support.raw_curvature_negative_absolute_terminal":{"lean":[],"short":"Negative raw curvature, absolute-value data","statements":[{"anchor":"curvature","at":"b47452-48438","end_byte":48438,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[],"lean_source":"record","markdown":"The same pointwise obstruction occurs with the actual terminal datum\n$|x|$. For the admissible constant coefficient $\\gamma_c\\equiv c>0$,\nwrite $\\tau=1-t>0$ and use the same heat formula:\n$$\n\\Phi_c(t,x)=c^{-1}\\log\\mathbb E e^{c|x+\\sqrt\\tau Z|}.\n$$\nAt $c=0$ its continuous extension is the heat value, whose derivatives are\n$$\na_0=\\sqrt{\\frac2{\\pi\\tau}}e^{-x^2/(2\\tau)},\\qquad\nb_0=-\\frac{x}{\\tau}a_0,\\qquad\nd_0=\\left(\\frac{x^2}{\\tau^2}-\\frac1\\tau\\right)a_0.\n$$\nAt $x=\\sqrt\\tau$, $d_0=0$ and $b_0^2=a_0^2/\\tau>0$.\nFor fixed $t,x$, Gaussian convolution may be differentiated in $x$\nthrough its kernel and expanded in $c$ near zero, since exponential moments\ndominate every resulting polynomial factor. The apparent singularity\nof $c^{-1}\\log\\mathbb E e^{c|x+\\sqrt\\tau Z|}$ is removable, and its\nspatial derivatives through order four are smooth in $c$. Hence\n$$\nd_c^2-12c a_cb_c^2+6c^2a_c^4\n =-\\frac{12a_0^3}{\\tau}c+O(c^2)<0\n$$\nfor all sufficiently small positive $c$, at that fixed point.","paper":"long","role":"statement","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":47452},{"anchor":"curvature","at":"b19951-20084","end_byte":20084,"heading_id":"the-integrated-estimate","lean":[],"lean_source":"record","markdown":"and that section also finds a negative pointwise value\nwith the actual $|x|$ terminal datum for a small positive constant\ncoefficient","paper":"short","role":"mention","section":"3. The integrated estimate","start_byte":19951},{"anchor":"curvature","at":"b46467-46897","end_byte":46897,"heading_id":"why-the-uncorrected-expression-has-no-pointwise-sign","lean":[],"lean_source":"record","markdown":"A concrete example shows that the last display cannot be read\npointwise. For the smooth finite-step problem $c=M=1$ on $[0,1)$, with terminal\nvalue $\\log(2\\cosh x)$, the exact solution is\n\n$$\n\\Phi(t,x)=\\log(2\\cosh x)+\\frac{1-t}{2},\\qquad r=\\tanh x.\n$$\n\nIndeed, $\\Phi_{xx}+\\Phi_x^2=1$. Writing $R=r^2$ gives\n$v=1-R$, $w=-2rv$ and $z=-2v^2+4Rv$. At $R=1/2$,\n$v=1/2$, $w^2=1/2$ and $z=1/2$, so\n\n$$\nz^2-12vw^2+6v^4=-\\frac{19}{8}<0.\n$$","note":"Not in the Lean: the formal proof uses only the integrated inequality, so this pointwise obstruction is not formalised.","paper":"long","role":"example","section":"4.1 Why the uncorrected expression has no pointwise sign","start_byte":46467}]},"support.regularity":{"lean":[{"clause":"∃ C : ℝ, 0 ≤ C ∧ ∀ t ∈ Icc (0:ℝ) T, ∀ x,\n      |iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[108,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Bounded positive-order spatial derivatives on each strip. Evenness, convexity and Lipschitz bounds are separate declarations.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"clause":"ContinuousOn (fun p : ℝ × ℝ => iteratedDeriv m (value W γ p.1) p.2)\n      (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":68,"line":52,"lines":[54,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity of each spatial derivative on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"},{"clause":"TendstoUniformlyOn (fun n (p : ℝ × ℝ) => iteratedDeriv m (approxValue γ n p.1) p.2)\n      (fun p => iteratedDeriv m (value W γ p.1) p.2) atTop (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":50,"line":26,"lines":[28,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform derivative convergence for one finite-step family with log-cosh terminal value, rather than general L1 approximation.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"short":"Compact-strip regularity","statements":[{"anchor":"dependencies","at":"b10756-11549","end_byte":11549,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"clause":"∃ C : ℝ, 0 ≤ C ∧ ∀ t ∈ Icc (0:ℝ) T, ∀ x,\n      |iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[108,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Bounded positive-order spatial derivatives on each strip. Evenness, convexity and Lipschitz bounds are separate declarations.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"clause":"ContinuousOn (fun p : ℝ × ℝ => iteratedDeriv m (value W γ p.1) p.2)\n      (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":68,"line":52,"lines":[54,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity of each spatial derivative on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"},{"clause":"TendstoUniformlyOn (fun n (p : ℝ × ℝ) => iteratedDeriv m (approxValue γ n p.1) p.2)\n      (fun p => iteratedDeriv m (value W γ p.1) p.2) atTop (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":50,"line":26,"lines":[28,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform derivative convergence for one finite-step family with log-cosh terminal value, rather than general L1 approximation.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"lean_source":"statement","markdown":"**Proposition 2.1 (compact-strip regularity).** For $\\gamma\\in\\mathcal U$\nand $0\\le T<1$, the value has bounded, jointly continuous spatial derivatives\nof every positive finite order on $[0,T]\\times\\mathbb R$. If admissible\ncoefficients converge in $L^1$ and their smooth, even, convex, one-Lipschitz\nterminal values converge uniformly to $|x|$, these spatial derivatives\nconverge uniformly on each such strip. The bounds depend on the derivative\norder, the fixed strip and a common coefficient mass bound. Time derivatives need only exist\nalmost everywhere across coefficient jumps. On an open constant interval,\nthe heat formula supplies the additional time smoothness used in §4.\nThe terminal estimate is\n$$\n|x|\\le\\Phi_\\gamma(t,x)\n\\le |x|+\\sqrt{2(1-t)/\\pi}+\\frac12\\int_t^1\\gamma(s)\\,ds.\n$$","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":10756},{"anchor":"contact","at":"b9822-10213","end_byte":10213,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"clause":"∃ C : ℝ, 0 ≤ C ∧ ∀ t ∈ Icc (0:ℝ) T, ∀ x,\n      |iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[108,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Bounded positive-order spatial derivatives on each strip. Evenness, convexity and Lipschitz bounds are separate declarations.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"clause":"ContinuousOn (fun p : ℝ × ℝ => iteratedDeriv m (value W γ p.1) p.2)\n      (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":68,"line":52,"lines":[54,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity of each spatial derivative on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"},{"clause":"TendstoUniformlyOn (fun n (p : ℝ × ℝ) => iteratedDeriv m (approxValue γ n p.1) p.2)\n      (fun p => iteratedDeriv m (value W γ p.1) p.2) atTop (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":50,"line":26,"lines":[28,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform derivative convergence for one finite-step family with log-cosh terminal value, rather than general L1 approximation.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"lean_source":"statement","markdown":"For an admissible coefficient, $\\Phi$ is even, convex and one-Lipschitz\nin space. Its spatial derivatives of positive order are bounded and\ncontinuous on each strip $[0,T]\\times\\mathbb R$, $T<1$, and the PDE\nholds for almost every time. The compact-strip approximation theorem of\n[OpenAI (2026a), Lemmas 2.1–2.2][FS] justifies the Itô calculations\nbelow even when $\\gamma$ has jumps. This","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":9822},{"anchor":"value-and-diffusion","at":"b18135-19007","end_byte":19007,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":116,"line":105,"lines":[105,106],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.value_lipschitz","note":"The displayed bound, as 1-Lipschitz in the starting state, for every t up to one.","plain":"The value is 1-Lipschitz in x","proof_lines":[107,116],"relation":"states"},{"end_line":48,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.value_convex","note":"Convex in the starting state, by this chord argument, for every t up to one.","plain":"The value is convex in x","proof_lines":[33,48],"relation":"states"}],"lean_source":"statement","markdown":"We first regard the value as a function of the starting state, holding\n$t$ and $\\gamma$ fixed, and write $\\Phi=\\Phi_\\gamma$. Given a permitted\ncontrol, use the same responses to each Brownian history at two starting\nstates $x$ and $y$. The costs are unchanged, and the triangle inequality\nbounds the change in the terminal reward by $|x-y|$, so taking the\nsupremum over controls gives\n\n$$\n|\\Phi_\\gamma(t,x)-\\Phi_\\gamma(t,y)|\\le |x-y|.\n$$\n\nThe value is also **convex**: between two states its graph lies at or\nbelow the chord joining their values. Indeed,\neach fixed-control payoff is an average of translated absolute-value\nfunctions, hence convex. Its chord lies at or below the chord of the\nbest endpoint payoffs, and taking the supremum preserves that upper bound.\n\n[OpenAI (2026a), Lemmas 2.1–2.2][support-regularity] supply spatial\ndifferentiability below time one.","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":18135},{"anchor":"control-derivations","at":"b50335-50632","end_byte":50632,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"clause":"∃ C : ℝ, 0 ≤ C ∧ ∀ t ∈ Icc (0:ℝ) T, ∀ x,\n      |iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[108,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Bounded positive-order spatial derivatives on each strip. Evenness, convexity and Lipschitz bounds are separate declarations.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"clause":"ContinuousOn (fun p : ℝ × ℝ => iteratedDeriv m (value W γ p.1) p.2)\n      (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":68,"line":52,"lines":[54,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity of each spatial derivative on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"},{"clause":"TendstoUniformlyOn (fun n (p : ℝ × ℝ) => iteratedDeriv m (approxValue γ n p.1) p.2)\n      (fun p => iteratedDeriv m (value W γ p.1) p.2) atTop (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":50,"line":26,"lines":[28,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform derivative convergence for one finite-step family with log-cosh terminal value, rather than general L1 approximation.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"},{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"Lean's form of the PDE: integrated in time between a≤b<1, so it imposes no condition at jumps of γ.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"part"}],"lean_source":"statement","markdown":"For each fixed $T<1$, the required spatial derivatives are bounded on\n$[0,T]\\times\\mathbb R$, and step-coefficient and smooth-terminal\napproximations justify the identities. The bounds depend on the\ncoefficient and on the strip, which ends before time one, and the PDE\nholds for almost every time.","paper":"expository","role":"statement","section":"7.2 The value equation and the moment identity","start_byte":50335},{"anchor":"value-and-diffusion","at":"b15303-15352","end_byte":15352,"heading_id":"the-zero-coefficient","lean":[{"end_line":49,"line":27,"lines":[29,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"At t=1 both bounds are zero, so the value is |x|, for every coefficient.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"contains"}],"lean_source":"statement","markdown":"at $t=1$ no noise remains and the\nvalue is $|x|$.","paper":"expository","role":"explanation","section":"3.1 The zero coefficient","start_byte":15303},{"anchor":"value-and-diffusion","at":"b18260-18519","end_byte":18519,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":84,"line":72,"lines":[72,84],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.controlPayoff_space_le","note":"The same control and path at x and y, with the triangle inequality; the cost term cancels.","plain":"One-sided spatial bound on the payoff","proof_lines":[75,84],"relation":"proof"}],"lean_source":"statement","markdown":"Given a permitted\ncontrol, use the same responses to each Brownian history at two starting\nstates $x$ and $y$. The costs are unchanged, and the triangle inequality\nbounds the change in the terminal reward by $|x-y|$, so taking the\nsupremum over controls gives","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":18260},{"anchor":"value-and-diffusion","at":"b18477-18572","end_byte":18572,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":116,"line":105,"lines":[105,106],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.value_lipschitz","note":"The result: the value is 1-Lipschitz in the starting state.","plain":"The value is 1-Lipschitz in x","proof_lines":[107,116],"relation":"states"},{"end_line":103,"line":86,"lines":[90,103],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.value_space_le","note":"Supremum over controls of the pathwise bound.","plain":"One-sided spatial bound on the value","proof_lines":[88,103],"relation":"proof"}],"lean_source":"statement","markdown":"so taking the\nsupremum over controls gives\n\n$$\n|\\Phi_\\gamma(t,x)-\\Phi_\\gamma(t,y)|\\le |x-y|.\n$$","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":18477},{"anchor":"value-and-diffusion","at":"b18574-18783","end_byte":18783,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":48,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.value_convex","note":"The claim itself: the value is convex in the starting state.","plain":"The value is convex in x","proof_lines":[33,48],"relation":"states"},{"end_line":22,"line":14,"lines":[14,22],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.controlPayoff_convex","note":"Each path's payoff is a translated absolute value minus a constant.","plain":"Samplewise payoff is convex in x","proof_lines":[17,22],"relation":"proof"},{"end_line":29,"line":24,"lines":[24,29],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.expected_controlPayoff_convex","note":"The average over paths stays convex.","plain":"Expected payoff is convex in x","proof_lines":[27,29],"relation":"proof"}],"lean_source":"statement","markdown":"The value is also **convex**: between two states its graph lies at or\nbelow the chord joining their values. Indeed,\neach fixed-control payoff is an average of translated absolute-value\nfunctions, hence convex.","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":18574},{"anchor":"value-and-diffusion","at":"b18784-18902","end_byte":18902,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":48,"line":31,"lines":[38,48],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.value_convex","note":"The chord for one control, then the bound by each endpoint's supremum, then the supremum.","plain":"The value is convex in x","proof_lines":[33,48],"relation":"proof"}],"lean_source":"statement","markdown":"Its chord lies at or below the chord of the\nbest endpoint payoffs, and taking the supremum preserves that upper bound.","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":18784},{"anchor":"value-and-diffusion","at":"b18904-19007","end_byte":19007,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":93,"line":86,"lines":[86,87],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_contDiff","note":"Smooth in x, at each time below one.","plain":"The value is smooth in x","proof_lines":[88,93],"relation":"contains"}],"lean_source":"statement","markdown":"[OpenAI (2026a), Lemmas 2.1–2.2][support-regularity] supply spatial\ndifferentiability below time one.","paper":"expository","role":"statement","section":"3.3 The response, the state and the averaged square","start_byte":18904},{"anchor":"value-and-diffusion","at":"b19008-19091","end_byte":19091,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":54,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.gradient","note":"Support model. The spatial derivative of the value.","plain":"The gradient u = ∂ₓΦ","relation":"defines"}],"lean_source":"statement","markdown":"Let $u(t,x)=\\partial_x\\Phi_\\gamma(t,x)$ be the slope\nas the starting state changes,","paper":"expository","role":"definition","section":"3.3 The response, the state and the averaged square","start_byte":19008},{"anchor":"value-and-diffusion","at":"b19189-19279","end_byte":19279,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":57,"line":56,"lines":[56,57],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.curvature","note":"Support model. The spatial derivative of the gradient.","plain":"The curvature a = ∂ₓₓΦ","relation":"defines"}],"lean_source":"statement","markdown":"and write\n$a(t,x)=\\partial_{xx}\\Phi_\\gamma(t,x)$ for the rate at which that\nslope changes.","paper":"expository","role":"definition","section":"3.3 The response, the state and the averaged square","start_byte":19189},{"anchor":"value-and-diffusion","at":"b19280-19345","end_byte":19345,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":17,"line":14,"lines":[14,15],"module":"OAI.Probability.SKSupport.Foundations.GradientAbs","name":"OAI.ZeroTemperatureSK.gradient_abs_le_one","note":"From the 1-Lipschitz bound.","plain":"The gradient is bounded by one","proof_lines":[16,17],"relation":"part"},{"end_line":192,"line":190,"lines":[190,191],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.curvature_nonneg","note":"From monotonicity of the gradient, which comes from convexity.","plain":"The curvature is nonnegative","proof_lines":[192,192],"relation":"part"}],"lean_source":"statement","markdown":"The displayed bound gives $|u|\\le1$, and convexity\ngives $a\\ge0$.","paper":"expository","role":"derivation","section":"3.3 The response, the state and the averaged square","start_byte":19280},{"anchor":"value-and-diffusion","at":"b20023-20112","end_byte":20112,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":57,"line":16,"lines":[30,34],"module":"OAI.Probability.SKSupport.Diffusion.FeedbackVerification","name":"OAI.ZeroTemperatureSK.WeakIto.exists_feedback_verification","note":"On a constant-coefficient segment: a feedback Euler control comes within ε of the value. No exact optimal control is stated.","plain":"Near-optimal control from feedback","proof_lines":[35,57],"relation":"part"}],"lean_source":"statement","markdown":"For general $\\gamma$, the control theorem says that the slope response\nattains the value.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":20023},{"anchor":"value-and-diffusion","at":"b20371-20566","end_byte":20566,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"end_line":30,"line":15,"lines":[20,26],"module":"OAI.Probability.SKSupport.Moments.IntegratedCompletionSquare","name":"OAI.ZeroTemperatureSK.WeakIto.integrated_completion_square","note":"The same completed square, integrated: line 26 uses the nonnegativity of c times (α minus u) squared.","plain":"Integrated completion of the square","proof_lines":[20,30],"relation":"proof"},{"end_line":33,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Moments.IntegratedCompletionSquare","name":"OAI.ZeroTemperatureSK.WeakIto.completion_square_feedback","note":"The value at v=u.","plain":"Completing the square at the feedback","relation":"proof"}],"lean_source":"statement","markdown":"The response-dependent gain per unit time is therefore\n\n$$\n\\gamma vu-\\tfrac12\\gamma v^2\n=\\tfrac12\\gamma u^2-\\tfrac12\\gamma(v-u)^2.\n$$\n\nSince $\\gamma\\ge0$ and $|u|\\le1$, $v=u$ attains its maximum.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":20371},{"anchor":"value-and-diffusion","at":"b20567-20682","end_byte":20682,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[{"clause":"(1/2:ℝ)*extend γ.val t*(gradient W γ t x)^2","end_line":15,"line":14,"lines":[15,15],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.valueRate","note":"The same term, inside the time rate of the value.","plain":"Time derivative of the value","relation":"defines"}],"lean_source":"statement","markdown":"The optimised contribution $\\gamma u^2/2$ also explains why a squared\nslope will enter the sensitivity to $\\gamma$.","paper":"expository","role":"explanation","section":"3.3 The response, the state and the averaged square","start_byte":20567},{"anchor":"control-derivations","at":"b47696-47777","end_byte":47777,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":49,"line":27,"lines":[29,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"At t=1 both bounds are zero, so the value is |x|.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"contains"}],"lean_source":"statement","markdown":"At $t=1$ there is no noise or response left to accumulate, so the value\nis $|x|$.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":47696},{"anchor":"control-derivations","at":"b47836-48054","end_byte":48054,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"Lean's form of the PDE: integrated in time between a≤b<1, so it imposes no condition at jumps of γ. Terminal value: third tab.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"part"},{"end_line":15,"line":14,"lines":[14,15],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.valueRate","note":"The right-hand side.","plain":"Time derivative of the value","relation":"defines"},{"end_line":49,"line":27,"lines":[29,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"At t=1 both bounds are zero, so the value is |x|.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"part"}],"lean_source":"statement","markdown":"Its partial differential equation (PDE) runs backward from that final value:\n\n$$\n\\partial_t\\Phi_\\gamma\n=-\\frac12\\left(\\partial_{xx}\\Phi_\\gamma\n+\\gamma(t)(\\partial_x\\Phi_\\gamma)^2\\right),\n\\qquad \\Phi_\\gamma(1,x)=|x|.\n$$","paper":"expository","role":"statement","section":"7.2 The value equation and the moment identity","start_byte":47836},{"anchor":"control-derivations","at":"b48056-48471","end_byte":48471,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":83,"line":17,"lines":[17,38],"module":"OAI.Probability.SKSupport.Foundations.ExpectedVerificationStep","name":"OAI.ZeroTemperatureSK.WeakIto.expected_verification_step","note":"Support model. One step for a smooth PDE solution: any bounded control gains at most an h^(3/2) error. Here the PDE is the hypothesis.","plain":"One expected verification step","proof_lines":[39,83],"relation":"proof"},{"end_line":87,"line":17,"lines":[17,36],"module":"OAI.Probability.SKSupport.Diffusion.ExpectedFeedbackStep","name":"OAI.ZeroTemperatureSK.WeakIto.expected_feedback_step","note":"Support model. The feedback choice v=u attains the step up to verificationError, on a constant-coefficient step.","proof_lines":[37,87],"relation":"proof"}],"lean_source":"statement","markdown":"With the spatial derivatives $u$ and $a$ defined in Section 3.3,\nthe equation expresses\na recursive choice: choose a response for a short first step, then\nuse the best continuation from the state reached. At a continuity\npoint of $\\gamma$, this gives the formal first-order relation\n\n$$\n\\Phi_\\gamma(t,x)\n=\\sup_{|v|\\le1}\\left\\{\n\\mathbb E\\Phi_\\gamma(t+h,x+\\gamma(t)vh+\\sqrt hZ)\n-\\tfrac12\\gamma(t)v^2h\\right\\}+o(h).\n$$","note":"The paper's recursion is formal. The Lean assumes the PDE and proves the one-step bounds from it.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":48056},{"anchor":"control-derivations","at":"b48685-48804","end_byte":48804,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":30,"line":15,"lines":[20,26],"module":"OAI.Probability.SKSupport.Moments.IntegratedCompletionSquare","name":"OAI.ZeroTemperatureSK.WeakIto.integrated_completion_square","note":"The completed square as an inequality, used in the upper verification.","plain":"Integrated completion of the square","proof_lines":[20,30],"relation":"proof"},{"end_line":33,"line":32,"lines":[32,33],"module":"OAI.Probability.SKSupport.Moments.IntegratedCompletionSquare","name":"OAI.ZeroTemperatureSK.WeakIto.completion_square_feedback","note":"Equality at v=u.","plain":"Completing the square at the feedback","relation":"proof"}],"lean_source":"statement","markdown":"The square completion from Section 3.3 gives a supremum of\n$\\gamma u^2/2$, attained at $v=u$, and hence the stated PDE.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":48685},{"anchor":"control-derivations","at":"b50335-50428","end_byte":50428,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"clause":"|iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[109,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Each positive order, uniformly on the strip.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"}],"lean_source":"statement","markdown":"For each fixed $T<1$, the required spatial derivatives are bounded on\n$[0,T]\\times\\mathbb R$,","paper":"expository","role":"statement","section":"7.2 The value equation and the moment identity","start_byte":50335},{"anchor":"control-derivations","at":"b50429-50508","end_byte":50508,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"clause":"TendstoUniformlyOn (fun n (p : ℝ × ℝ) => iteratedDeriv m (approxValue γ n p.1) p.2)\n      (fun p => iteratedDeriv m (value W γ p.1) p.2) atTop (Icc (0:ℝ) T ×ˢ Set.univ)","end_line":50,"line":26,"lines":[28,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform convergence of derivatives along finite-step, log-cosh terminal approximations.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"lean_source":"statement","markdown":"and step-coefficient and smooth-terminal\napproximations justify the identities.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":50429},{"anchor":"control-derivations","at":"b50509-50632","end_byte":50632,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":114,"line":106,"lines":[106,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"C comes after γ, T and m, so it depends on them.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"}],"lean_source":"statement","markdown":"The bounds depend on the\ncoefficient and on the strip, which ends before time one, and the PDE\nholds for almost every time.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":50509},{"anchor":"control-derivations","at":"b50575-50632","end_byte":50632,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"Stated as an integral identity in time, so no pointwise statement is needed.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"states"}],"lean_source":"statement","markdown":"before time one, and the PDE\nholds for almost every time.","paper":"expository","role":"statement","section":"7.2 The value equation and the moment identity","start_byte":50575},{"anchor":"control-derivations","at":"b50553-50590","end_byte":50590,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"clause":"(hT1 : T < 1)","end_line":114,"line":106,"lines":[107,107],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Every strip ends strictly before one.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"}],"lean_source":"statement","markdown":"the strip, which ends before time one","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":50553},{"anchor":"control-derivations","at":"b48473-48683","end_byte":48683,"heading_id":"the-value-equation-and-the-moment-identity","lean":[{"end_line":80,"line":18,"lines":[18,28],"module":"OAI.Probability.SKSupport.Diffusion.ExpectedBrownianStep","name":"OAI.ZeroTemperatureSK.WeakIto.expected_brownian_step","note":"Support model. Taylor step for a Brownian increment: mean zero, variance h, error h^(3/2). The PDE itself is a hypothesis in the Lean.","proof_lines":[29,80],"relation":"proof"}],"lean_source":"statement","markdown":"Subtracting the present value, dividing by $h$ and expanding by\nTaylor's formula, with $\\mathbb EZ=0$ and $\\mathbb EZ^2=1$, we obtain\n\n$$\n0=\\Phi_t+\\tfrac12\\Phi_{xx}\n  +\\sup_{|v|\\le1}\\gamma(t)(v\\Phi_x-v^2/2).\n$$","note":"The Taylor step is in the Lean. Its PDE has the supremum already evaluated, as c/2 times the squared gradient.","paper":"expository","role":"explanation","section":"7.2 The value equation and the moment identity","start_byte":48473},{"anchor":"model","at":"b4623-4932","end_byte":4932,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"The time equation in integrated form on strips before time one.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"part"},{"end_line":15,"line":14,"lines":[14,15],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.valueRate","note":"The rate is the right side of the paper's PDE.","plain":"Time derivative of the value","relation":"defines"},{"end_line":49,"line":27,"lines":[27,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"At t=1 this squeezes the value to |x|.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"part"}],"lean_source":"statement","markdown":"The control formula fixes the solution, and the PDE is\n$\\Phi_t=-\\tfrac12(\\Phi_{xx}+\\gamma\\Phi_x^2)$, with\n$\\Phi(1,x)=|x|$. For a general admissible coefficient, the solution and its\nspatial derivatives are locally absolutely continuous in time on compact\nstrips, and the time equation holds almost everywhere.","note":"Lean never takes the PDE as a definition: value is the control supremum, as here.","paper":"long","role":"explanation","section":"1. The SK model and the Parisi variational problem","start_byte":4623},{"anchor":"dependencies","at":"b11550-11768","end_byte":11768,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":49,"line":27,"lines":[27,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"The terminal bound; it mentions no derivative.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"part"},{"clause":"∃ C : ℝ, 0 ≤ C ∧ ∀ t ∈ Icc (0:ℝ) T, ∀ x,\n      |iteratedDeriv (m+1) (value W γ t) x| ≤ C","end_line":114,"line":106,"lines":[108,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"The strip bounds, each for a fixed T<1.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"}],"lean_source":"statement","markdown":"It controls the approach to $|x|$ without differentiating the cusp. The\nstrip estimates justify the interior calculations and finite meshes, while\nthe terminal bound and integrability justify the later endpoint limits.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11550},{"anchor":"dependencies","at":"b11850-11996","end_byte":11996,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":190,"line":175,"lines":[175,178],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_derivative_time_bound","note":"Lean states a Lipschitz bound in time, which is stronger than local absolute continuity.","plain":"x-derivatives of Φ Lipschitz in time","proof_lines":[179,190],"relation":"part"}],"lean_source":"statement","markdown":"On a fixed $[0,T]$ with $T<1$, bounded spatial derivatives and locally\nabsolutely continuous time dependence justify the Itô identities and Euler","note":"Lean uses grid and Taylor expansion arguments for these identities, not an Itô formula.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11850},{"anchor":"dependencies","at":"b12007-12202","end_byte":12202,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":137,"line":121,"lines":[121,125],"module":"OAI.Probability.SKSupport.Regularity.RightRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_hasDerivWithinAt_right","note":"A right derivative at every time, stronger than almost everywhere: the coefficient is right-continuous.","plain":"Right time derivative of x-derivatives","proof_lines":[126,137],"relation":"part"}],"lean_source":"statement","markdown":"Across a jump of $\\gamma$, the spatial derivative moments remain\ncontinuous and the time equations hold almost everywhere. This is enough\nfor the contact slopes and the endpoint chord comparison.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12007},{"anchor":"dependencies","at":"b12305-12546","end_byte":12546,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":49,"line":27,"lines":[27,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"The displayed value bound.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"part"},{"end_line":119,"line":107,"lines":[107,109],"module":"OAI.Probability.SKSupport.Regularity.MixedIdentity","name":"OAI.ZeroTemperatureSK.mixed_difference_terminal_bound","note":"The first-variation remainder, bounded by an L1 tail of the coefficient difference.","plain":"Terminal bound on the mixed difference","proof_lines":[110,119],"relation":"proof"}],"lean_source":"statement","markdown":"As $T\\uparrow1$ the controls change. The displayed value bound, integrable\ncoefficient tails and martingale convergence justify the first variation and\nthe terminal value, so the argument uses no differentiated estimate at the\nterminal cusp.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12305},{"anchor":"dependencies","at":"b12728-12806","end_byte":12806,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":162,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_even","note":"Evenness.","plain":"The value is even in x","proof_lines":[157,162],"relation":"part"},{"end_line":181,"line":177,"lines":[177,178],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_zero","note":"u(t,0)=0 at every t<1.","plain":"Gradient vanishes at x = 0","proof_lines":[179,181],"relation":"part"},{"end_line":175,"line":164,"lines":[164,165],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_odd","note":"The reflection.","plain":"The gradient is odd in x","proof_lines":[166,175],"relation":"part"}],"lean_source":"statement","markdown":"Zero field supplies evenness,\n$u(0,0)=0$ and reflection in the shape argument.","note":"In Lean the absence of a field is built into the model: the terminal value is |x|.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":12728},{"anchor":"dependencies","at":"b10756-10965","end_byte":10965,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":114,"line":106,"lines":[106,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Boundedness, for every derivative order.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"end_line":68,"line":52,"lines":[52,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"}],"lean_source":"statement","markdown":"**Proposition 2.1 (compact-strip regularity).** For $\\gamma\\in\\mathcal U$\nand $0\\le T<1$, the value has bounded, jointly continuous spatial derivatives\nof every positive finite order on $[0,T]\\times\\mathbb R$.","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":10756},{"anchor":"dependencies","at":"b10966-11166","end_byte":11166,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":50,"line":26,"lines":[26,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Every order, along the sequence approxValue.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"},{"end_line":139,"line":115,"lines":[115,118],"module":"OAI.Probability.SKSupport.Regularity.PositiveRegularity","name":"OAI.ZeroTemperatureSK.positiveValue_derivative_uniform","note":"The same along the positive sequence.","plain":"Uniform derivative of the positive scheme","proof_lines":[119,139],"relation":"part"},{"end_line":61,"line":45,"lines":[45,48],"module":"OAI.Probability.SKSupport.Parabolic.GradientStability","name":"OAI.ZeroTemperatureSK.compactGradient_stability","note":"For two arbitrary order parameters, only the gradient.","plain":"Gradient is stable in the coefficient","proof_lines":[49,61],"relation":"part"}],"lean_source":"statement","markdown":"If admissible\ncoefficients converge in $L^1$ and their smooth, even, convex, one-Lipschitz\nterminal values converge uniformly to $|x|$, these spatial derivatives\nconverge uniformly on each such strip.","note":"Lean has this at every order only along its own sequences approxValue and positiveValue.","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":10966},{"anchor":"dependencies","at":"b11167-11262","end_byte":11262,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":112,"line":38,"lines":[38,45],"module":"OAI.Probability.SKSupport.Foundations.SmoothingStep","name":"OAI.ZeroTemperatureSK.Heat.finiteGradient_uniform_smoothing","note":"C depends only on the order, the coefficient bound D and the gap δ, uniformly in the terminal datum and grid.","plain":"Uniform smoothing of finite-step gradients","proof_lines":[46,112],"relation":"proof"}],"lean_source":"statement","markdown":"The bounds depend on the derivative\norder, the fixed strip and a common coefficient mass bound.","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11167},{"anchor":"dependencies","at":"b11263-11339","end_byte":11339,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":137,"line":121,"lines":[121,125],"module":"OAI.Probability.SKSupport.Regularity.RightRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_hasDerivWithinAt_right","note":"Lean has right derivatives everywhere instead.","plain":"Right time derivative of x-derivatives","proof_lines":[126,137],"relation":"part"}],"lean_source":"statement","markdown":"Time derivatives need only exist\nalmost everywhere across coefficient jumps.","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11263},{"anchor":"dependencies","at":"b11440-11549","end_byte":11549,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":49,"line":27,"lines":[27,31],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.value_terminal_uniform","note":"The same two bounds, written for value minus |x|.","plain":"Uniform bound on Φ minus |x|","proof_lines":[32,49],"relation":"states"}],"lean_source":"statement","markdown":"The terminal estimate is\n$$\n|x|\\le\\Phi_\\gamma(t,x)\n\\le |x|+\\sqrt{2(1-t)/\\pi}+\\frac12\\int_t^1\\gamma(s)\\,ds.\n$$","paper":"long","role":"statement","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":11440},{"anchor":"strip-recovery","at":"b24999-25337","end_byte":25337,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":22,"line":21,"lines":[21,22],"module":"OAI.Probability.SKSupport.Foundations.Semigroup","name":"OAI.ZeroTemperatureSK.Heat.logSemigroup","note":"Both cases: semigroup at c=0, the log formula otherwise.","plain":"Log-exponential heat step","relation":"defines"},{"end_line":21,"line":19,"lines":[19,21],"module":"OAI.Probability.SKSupport.Foundations.Cascade","name":"OAI.ZeroTemperatureSK.Heat.cascade","note":"Iterating that step over the grid cells.","plain":"Finite Parisi cascade","relation":"defines"}],"lean_source":"statement","markdown":"For finite-step coefficients and a smooth, even, convex, one-Lipschitz\nterminal function $\\psi$, the evolution through a step of height $c$ is\n$$\n\\Phi(t,x)=\n\\begin{cases}\nc^{-1}\\log P_{b-t}(e^{c\\Phi(b,\\cdot)})(x),&c>0,\\\\\nP_{b-t}\\Phi(b,\\cdot)(x),&c=0,\n\\end{cases}\n$$\nwhere $P_h$ denotes convolution with a centred Gaussian of variance $h$.","paper":"long","role":"definition","section":"2.4 Recovering the compact-strip estimates","start_byte":24999},{"anchor":"strip-recovery","at":"b25338-25417","end_byte":25417,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":172,"line":161,"lines":[161,162],"module":"OAI.Probability.SKSupport.Foundations.Semigroup","name":"OAI.ZeroTemperatureSK.Heat.logSemigroup_even","note":"Evenness.","plain":"Heat step preserves evenness","proof_lines":[163,172],"relation":"part"},{"end_line":143,"line":125,"lines":[125,127],"module":"OAI.Probability.SKSupport.Foundations.Semigroup","name":"OAI.ZeroTemperatureSK.Heat.logSemigroup_lipschitz","note":"The Lipschitz bound.","plain":"Heat step preserves Lipschitz bounds","proof_lines":[128,143],"relation":"part"},{"end_line":48,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.value_convex","note":"Convexity comes from the control formula, for the value itself.","plain":"The value is convex in x","proof_lines":[33,48],"relation":"proof"}],"lean_source":"statement","markdown":"The heat formula preserves evenness, convexity and the spatial Lipschitz\nbound.","note":"For convexity the support model uses the control formula, not preservation by the heat step.","paper":"long","role":"explanation","section":"2.4 Recovering the compact-strip estimates","start_byte":25338},{"anchor":"strip-recovery","at":"b25418-25636","end_byte":25636,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":83,"line":17,"lines":[17,38],"module":"OAI.Probability.SKSupport.Foundations.ExpectedVerificationStep","name":"OAI.ZeroTemperatureSK.WeakIto.expected_verification_step","note":"One Euler cell: a Taylor bound on value minus cost, with hPDE and an explicit mesh error.","plain":"One expected verification step","proof_lines":[39,83],"relation":"proof"}],"lean_source":"statement","markdown":"Along a controlled path $dY_s=\\gamma(s)\\alpha_s\\,ds+dB_s$, the drift\nof the value minus the accumulated control cost is\n$-\\gamma(s)(\\alpha_s-\\Phi_x(s,Y_s))^2/2$. It is nonpositive, and is zero\nfor the feedback control.","note":"Lean has no Itô formula here: it adds up one-step Taylor bounds and sends the mesh to zero.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":25418},{"anchor":"strip-recovery","at":"b25637-25755","end_byte":25755,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":297,"line":267,"lines":[267,270],"module":"OAI.Probability.SKSupport.Foundations.StepData","name":"OAI.ZeroTemperatureSK.FiniteVerification.cascade_eq_control_sup","note":"The step value is the supremum over controls.","plain":"Cascade equals the control supremum","proof_lines":[271,297],"relation":"part"},{"end_line":57,"line":16,"lines":[30,34],"module":"OAI.Probability.SKSupport.Diffusion.FeedbackVerification","name":"OAI.ZeroTemperatureSK.WeakIto.exists_feedback_verification","note":"The lower bound, within any ε, by an Euler feedback control.","plain":"Near-optimal control from feedback","proof_lines":[35,57],"relation":"proof"}],"lean_source":"statement","markdown":"This proves the stated control formula for the\nsmooth step problem, including attainability by its feedback diffusion.","note":"Lean does not show attainment: it comes within ε by elementary Euler feedback controls.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":25637},{"anchor":"strip-recovery","at":"b25757-25995","end_byte":25995,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":92,"line":85,"lines":[85,87],"module":"OAI.Probability.SKSupport.Foundations.GradientAbs","name":"OAI.ZeroTemperatureSK.value_coefficient_stability","note":"The coefficient term, with the same constant 3/2.","plain":"Value is stable in the coefficient","proof_lines":[88,92],"relation":"part"},{"end_line":44,"line":36,"lines":[36,39],"module":"OAI.Probability.SKSupport.Foundations.Cascade","name":"OAI.ZeroTemperatureSK.Heat.cascade_terminal_stability","note":"The terminal term, for step values.","plain":"Cascade stable in terminal data","proof_lines":[40,44],"relation":"part"},{"end_line":197,"line":185,"lines":[185,190],"module":"OAI.Probability.SKSupport.Control.GridApprox","name":"OAI.ZeroTemperatureSK.finiteValue_control_error","note":"Both terms together, between a step value and the true value.","plain":"Error of the finite-step value","proof_lines":[191,197],"relation":"part"}],"lean_source":"statement","markdown":"Using the same control in two such problems and then taking suprema gives\n$$\n\\|\\Phi_{\\gamma,\\psi}-\\Phi_{\\widetilde\\gamma,\\widetilde\\psi}\\|_\\infty\n\\le \\|\\psi-\\widetilde\\psi\\|_\\infty\n       +\\frac32\\|\\gamma-\\widetilde\\gamma\\|_{L^1(0,1)}.\n$$","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":25757},{"anchor":"strip-recovery","at":"b25996-26162","end_byte":26162,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":62,"line":35,"lines":[35,38],"module":"OAI.Probability.SKSupport.Foundations.GradientAbs","name":"OAI.ZeroTemperatureSK.controlPayoff_coeff_le","note":"The same control on both sides, pathwise.","plain":"One-sided payoff bound in the coefficient","proof_lines":[39,62],"relation":"proof"}],"lean_source":"statement","markdown":"The drift part costs at most the coefficient difference in $L^1$, by the\nterminal Lipschitz bound, and the quadratic cost contributes another half of\nthat difference.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":25996},{"anchor":"strip-recovery","at":"b26280-26349","end_byte":26349,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":131,"line":112,"lines":[112,113],"module":"OAI.Probability.SKSupport.Foundations.SoftAbs","name":"OAI.ZeroTemperatureSK.Heat.softAbs_bounds","note":"The two-sided terminal bound.","plain":"Soft absolute value near |x|","proof_lines":[114,131],"relation":"states"},{"end_line":69,"line":69,"lines":[69,69],"module":"OAI.Probability.SKSupport.Foundations.SoftAbs","name":"OAI.ZeroTemperatureSK.Heat.softAbs","note":"","plain":"Soft absolute value log(2 cosh Mx)/M","relation":null}],"lean_source":"statement","markdown":"With $\\psi_M(x)=M^{-1}\\log(2\\cosh Mx)$,\n$0\\le\\psi_M-|x|\\le(\\log2)/M$.","paper":"long","role":"definition","section":"2.4 Recovering the compact-strip estimates","start_byte":26280},{"anchor":"strip-recovery","at":"b26350-26602","end_byte":26602,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":41,"line":33,"lines":[33,35],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.approxValue_error","note":"The joint error bound: step and terminal errors against the control-defined value.","plain":"Error bound for the approximate value","proof_lines":[36,41],"relation":"part"},{"end_line":29,"line":21,"lines":[21,21],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.approxError_limit","note":"Both errors vanish together.","plain":"Approximation error tends to zero","proof_lines":[22,29],"relation":"part"},{"end_line":16,"line":15,"lines":[15,16],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.approxValue","note":"Lean takes M=n+1 with the n-th step approximation.","plain":"Approximate value of step n","relation":"defines"}],"lean_source":"statement","markdown":"Choose simultaneously $\\gamma_m\\to\\gamma$ in $L^1$ and $M_m\\to\\infty$.\nThese are two distinct requirements: $M_m\\ge\\|\\gamma_m\\|_\\infty$ alone\ndoes not force terminal convergence. This compatible approximation\ndetermines a unique uniform limiting value.","note":"Lean does not define the limit from this sequence: value is the control supremum, and the sequence converges to it.","paper":"long","role":"explanation","section":"2.4 Recovering the compact-strip estimates","start_byte":26350},{"anchor":"strip-recovery","at":"b26662-26763","end_byte":26763,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":70,"line":54,"lines":[54,55],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.abs_le_value","note":"The lower bound.","plain":"The value dominates |x|","proof_lines":[56,70],"relation":"part"},{"end_line":119,"line":86,"lines":[107,110],"module":"OAI.Probability.SKSupport.Regularity.AbsExtend","name":"OAI.ZeroTemperatureSK.controlPayoff_le","note":"The pointwise inequality the paper quotes, then integrated.","plain":"Upper bound on the payoff","proof_lines":[90,119],"relation":"proof"}],"lean_source":"statement","markdown":"choose the zero control for the lower\nbound, and use $|\\alpha|-\\alpha^2/2\\le1/2$ for the upper bound.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":26662},{"anchor":"strip-recovery","at":"b26765-26925","end_byte":26925,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":67,"line":50,"lines":[50,51],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.OrderParameter.local_bound","note":"The same bound.","plain":"γ bounded by its integral","proof_lines":[52,67],"relation":"states"}],"lean_source":"statement","markdown":"To obtain spatial derivatives, choose $T<b<1$. Monotonicity gives the\nuniform coefficient estimate\n$$\n\\gamma(t)\\le\\frac{\\|\\gamma\\|_1}{1-b}\\quad(0\\le t\\le b).\n$$","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":26765},{"anchor":"strip-recovery","at":"b27070-27170","end_byte":27170,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":142,"line":114,"lines":[114,118],"module":"OAI.Probability.SKSupport.Regularity.BoundedSmoothFamily","name":"OAI.ZeroTemperatureSK.Heat.finiteGradient_mild","note":"For step values; the source is coefficient times gradient times its derivative.","plain":"Mild form of the finite-step gradient","proof_lines":[119,142],"relation":"states"},{"end_line":93,"line":92,"lines":[92,93],"module":"OAI.Probability.SKSupport.Regularity.BoundedSmoothFamily","name":"OAI.ZeroTemperatureSK.Heat.finiteBurgersSource","note":"","plain":"Burgers source of the finite scheme","relation":null}],"lean_source":"statement","markdown":"The differentiated mild equation is\n$$\nU(t)=P_{b-t}U(b)+\\int_t^b\\gamma(s)P_{s-t}[U(s)U_x(s)]\\,ds.\n$$","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":27070},{"anchor":"strip-recovery","at":"b27171-27358","end_byte":27358,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":27,"line":23,"lines":[23,25],"module":"OAI.Probability.SKSupport.Foundations.VarianceHeatGradientBound","name":"OAI.ZeroTemperatureSK.Heat.varianceHeat_gradient_bound","note":"The smoothing bound, with C the first absolute Gaussian moment.","plain":"Heat gradient bound of order 1/√h","proof_lines":[26,27],"relation":"part"},{"end_line":135,"line":56,"lines":[72,130],"module":"OAI.Probability.SKSupport.Foundations.VarianceHeatGradientBound","name":"OAI.ZeroTemperatureSK.Heat.uniform_gradient_bound_of_mild","note":"The pointwise inequality, inside this proof.","plain":"Uniform gradient bound, mild equation","proof_lines":[68,135],"relation":"proof"}],"lean_source":"statement","markdown":"Since $\\|U\\|_\\infty\\le1$ and\n$\\|\\partial_xP_hF\\|_\\infty\\le Ch^{-1/2}\\|F\\|_\\infty$,\n$$\n\\|U_x(t)\\|_\\infty\\le\\frac C{\\sqrt{b-t}}\n +C\\Gamma\\int_t^b\\frac{\\|U_x(s)\\|_\\infty}{\\sqrt{s-t}}\\,ds.\n$$","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":27171},{"anchor":"strip-recovery","at":"b27359-27629","end_byte":27629,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":135,"line":56,"lines":[131,135],"module":"OAI.Probability.SKSupport.Foundations.VarianceHeatGradientBound","name":"OAI.ZeroTemperatureSK.Heat.uniform_gradient_bound_of_mild","note":"Absorption of a weighted supremum on a short step.","plain":"Uniform gradient bound, mild equation","proof_lines":[68,135],"relation":"proof"},{"end_line":167,"line":151,"lines":[151,156],"module":"OAI.Probability.SKSupport.Foundations.InvSqrtNegHalf","name":"OAI.ZeroTemperatureSK.Smoothing.weighted_absorption","note":"The absorption lemma.","plain":"Absorption in a √(b−t)-weighted norm","proof_lines":[157,167],"relation":"proof"}],"lean_source":"statement","markdown":"Substituting this inequality once into its own integral uses\n$\\int_t^s((r-t)(s-r))^{-1/2}\\,dr=\\pi$. It reduces the estimate to an\nordinary backward Gronwall inequality with integrable forcing. This bounds\n$U_x$ uniformly on $[0,T]$ without any terminal derivative bound.","note":"Lean proves this bound differently: it absorbs a √(b−t)-weighted supremum on a short step, with no Gronwall iteration.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":27359},{"anchor":"strip-recovery","at":"b27630-27893","end_byte":27893,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":112,"line":38,"lines":[55,70],"module":"OAI.Probability.SKSupport.Foundations.SmoothingStep","name":"OAI.ZeroTemperatureSK.Heat.finiteGradient_uniform_smoothing","note":"Induction on the order; each step shortens the strip by δ/2.","plain":"Uniform smoothing of finite-step gradients","proof_lines":[46,112],"relation":"proof"},{"end_line":211,"line":191,"lines":[191,197],"module":"OAI.Probability.SKSupport.Foundations.VarianceHeatGradientBound","name":"OAI.ZeroTemperatureSK.Heat.iteratedDeriv_finiteBurgersSource_bound","note":"Only the top derivative is new; the rest are bounded.","plain":"Derivative bounds on the Burgers source","proof_lines":[198,211],"relation":"proof"}],"lean_source":"statement","markdown":"Repeating\non successively smaller strips bounds each higher spatial derivative: put\none derivative on the heat kernel and the rest on $UU_x$, and only the term\n$U\\partial_x^mU$ contains the new highest derivative. The remaining products\nhave already been bounded.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":27630},{"anchor":"strip-recovery","at":"b27895-28203","end_byte":28203,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":55,"line":41,"lines":[41,44],"module":"OAI.Probability.SKSupport.Foundations.Taylor1Bound","name":"OAI.ZeroTemperatureSK.Heat.derivative_interpolation","note":"The same inequality.","plain":"Landau derivative interpolation","proof_lines":[45,55],"relation":"part"},{"end_line":78,"line":57,"lines":[57,61],"module":"OAI.Probability.SKSupport.Foundations.Taylor1Bound","name":"OAI.ZeroTemperatureSK.Heat.tendstoUniformly_deriv_zero","note":"Its use: a uniform limit with bounded second derivative has vanishing derivative.","plain":"Uniform null limits have null derivatives","proof_lines":[62,78],"relation":"part"}],"lean_source":"statement","markdown":"Uniform convergence of the values then yields convergence of every\npositive spatial derivative. For a bounded twice differentiable difference\n$F$ with bounded second derivative, Taylor's formula gives\n$$\n\\|F'\\|_\\infty\\le \\frac{2\\|F\\|_\\infty}{h}\n                         +\\frac h2\\|F''\\|_\\infty\\qquad(h>0).\n$$","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":27895},{"anchor":"strip-recovery","at":"b28204-28335","end_byte":28335,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":191,"line":175,"lines":[175,180],"module":"OAI.Probability.SKSupport.Control.CoefficientApprox","name":"OAI.ZeroTemperatureSK.Heat.tendstoUniformly_deriv_of_bound","note":"One derivative.","plain":"Uniform convergence of first derivatives","proof_lines":[181,191],"relation":"part"},{"end_line":223,"line":201,"lines":[201,206],"module":"OAI.Probability.SKSupport.Control.CoefficientApprox","name":"OAI.ZeroTemperatureSK.Heat.smooth_limit_of_uniform_derivative_bounds","note":"Iterated to every order.","plain":"Smooth limit under uniform derivative bounds","proof_lines":[207,223],"relation":"part"}],"lean_source":"statement","markdown":"First choose $h$ small using the common second-derivative bound, then make\n$\\|F\\|_\\infty$ small. Iterate on derivative differences.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":28204},{"anchor":"strip-recovery","at":"b28336-28468","end_byte":28468,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":102,"line":92,"lines":[92,94],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_integral_rate","note":"The integral form.","plain":"The value integrates its rate","proof_lines":[95,102],"relation":"part"},{"end_line":190,"line":175,"lines":[175,178],"module":"OAI.Probability.SKSupport.Foundations.Equation","name":"OAI.ZeroTemperatureSK.value_derivative_time_bound","note":"Time dependence of every spatial derivative.","plain":"x-derivatives of Φ Lipschitz in time","proof_lines":[179,190],"relation":"part"}],"lean_source":"statement","markdown":"Passing to the\nintegral form of the differentiated PDE proves local absolute continuity\nin time and its almost-everywhere equations.","paper":"long","role":"derivation","section":"2.4 Recovering the compact-strip estimates","start_byte":28336},{"anchor":"approximation-choices","at":"b29753-29974","end_byte":29974,"heading_id":"the-order-of-the-approximation-limits","lean":[{"end_line":50,"line":26,"lines":[26,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"With T and m fixed first, for one sequence.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"lean_source":"statement","markdown":"**Compact-strip regularity.** Fix a derivative order $k\\ge1$, a cutoff\n$T<1$ and a common coefficient $L^1$ bound. Coefficients converge in $L^1$\nwithin that bound, and spatial derivatives converge uniformly on the strip.","paper":"long","role":"explanation","section":"2.5 The order of the approximation limits","start_byte":29753},{"anchor":"strip-recovery","at":"b24809-24892","end_byte":24892,"heading_id":"recovering-the-compact-strip-estimates","lean":[{"end_line":84,"line":66,"lines":[66,69],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.approxValue_uniform_derivative_bounds","note":"Bounds uniform in n, on each strip.","plain":"Uniform derivative bounds, approximate value","proof_lines":[70,84],"relation":"part"}],"lean_source":"statement","markdown":"The curvature argument needs estimates uniform over the approximating\ncoefficients.","paper":"long","role":"explanation","section":"2.4 Recovering the compact-strip estimates","start_byte":24809},{"anchor":"model","at":"b4628-4746","end_byte":4746,"heading_id":"the-model-and-the-result","lean":[{"end_line":50,"line":43,"lines":[43,45],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.approxValue_uniform","note":"In Lean the control formula is the definition for every integrable γ. This approximation is proved for it.","plain":"Approximate value converges uniformly","proof_lines":[46,50],"relation":"proof"}],"lean_source":"statement","markdown":"For an integrable coefficient, the solution of the PDE\nis defined by approximation from bounded monotone coefficients.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":4628},{"anchor":"model","at":"b4747-4854","end_byte":4854,"heading_id":"the-model-and-the-result","lean":[{"end_line":51,"line":49,"lines":[49,51],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.value","note":"","plain":"The value function Φ","relation":"defines"},{"end_line":47,"line":43,"lines":[43,47],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.controlPayoff","note":"","plain":"Control payoff, sample by sample","relation":"defines"},{"end_line":92,"line":85,"lines":[85,87],"module":"OAI.Probability.SKSupport.Foundations.GradientAbs","name":"OAI.ZeroTemperatureSK.value_coefficient_stability","note":"Stability in the coefficient, with constant 3/2 in L1 on [t,1]. The terminal value is fixed at |x| here.","plain":"Value is stable in the coefficient","proof_lines":[88,92],"relation":"part"}],"lean_source":"statement","markdown":"[OpenAI (2026a), §1 and Lemmas 2.1–2.2][FS] give the exact control\ndefinition and the stability theorem.","paper":"short","role":"mention","section":"1. The model and the result","start_byte":4747},{"anchor":"main-result","at":"b6882-6910","end_byte":6910,"heading_id":"the-model-and-the-result","lean":[{"end_line":181,"line":177,"lines":[177,178],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.gradient_zero","note":"From evenness of the value function; it holds for every t<1.","plain":"Gradient vanishes at x = 0","proof_lines":[179,181],"relation":"contains"}],"lean_source":"statement","markdown":"Zero field gives\n$u(0,0)=0$,","paper":"short","role":"mention","section":"1. The model and the result","start_byte":6882},{"anchor":"contact","at":"b9822-9903","end_byte":9903,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":162,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_even","note":"","plain":"The value is even in x","proof_lines":[157,162],"relation":"part"},{"end_line":48,"line":31,"lines":[31,32],"module":"OAI.Probability.SKSupport.Control.ControlPayoffConvex","name":"OAI.ZeroTemperatureSK.value_convex","note":"","plain":"The value is convex in x","proof_lines":[33,48],"relation":"part"},{"end_line":116,"line":105,"lines":[105,106],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.value_lipschitz","note":"","plain":"The value is 1-Lipschitz in x","proof_lines":[107,116],"relation":"part"}],"lean_source":"statement","markdown":"For an admissible coefficient, $\\Phi$ is even, convex and one-Lipschitz\nin space.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":9822},{"anchor":"contact","at":"b9904-10017","end_byte":10017,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":114,"line":106,"lines":[106,109],"module":"OAI.Probability.SKSupport.Regularity.Regularity","name":"OAI.ZeroTemperatureSK.value_uniform_derivative_bounds","note":"Bounded positive-order spatial derivatives on each strip. Evenness, convexity and Lipschitz bounds are separate declarations.","plain":"Bounded x-derivatives of Φ on strips","proof_lines":[110,114],"relation":"part"},{"end_line":68,"line":52,"lines":[52,55],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_continuousOn","note":"Joint continuity of each spatial derivative on the strip.","plain":"x-derivatives of the value are continuous","proof_lines":[56,68],"relation":"part"}],"lean_source":"statement","markdown":"Its spatial derivatives of positive order are bounded and\ncontinuous on each strip $[0,T]\\times\\mathbb R$, $T<1$,","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":9904},{"anchor":"contact","at":"b10022-10058","end_byte":10058,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":137,"line":121,"lines":[121,125],"module":"OAI.Probability.SKSupport.Regularity.RightRegularity","name":"OAI.ZeroTemperatureSK.value_derivative_hasDerivWithinAt_right","note":"Right derivative at every time, for each spatial order. At order zero compactValueRate is the PDE right side.","plain":"Right time derivative of x-derivatives","proof_lines":[126,137],"relation":"proof"}],"lean_source":"statement","markdown":"the PDE\nholds for almost every time.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":10022},{"anchor":"contact","at":"b10059-10208","end_byte":10208,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":50,"line":26,"lines":[26,29],"module":"OAI.Probability.SKSupport.Regularity.JointRegularity","name":"OAI.ZeroTemperatureSK.approxValue_derivative_uniform","note":"Uniform convergence of every spatial derivative on a strip below one.","plain":"Uniform derivative of the approximate value","proof_lines":[30,50],"relation":"part"}],"lean_source":"statement","markdown":"The compact-strip approximation theorem of\n[OpenAI (2026a), Lemmas 2.1–2.2][FS] justifies the Itô calculations\nbelow even when $\\gamma$ has jumps.","paper":"short","role":"explanation","section":"2. Contact conditions and the exclusion of gaps","start_byte":10059}]},"support.source_finite_density_coercivity":{"lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The density-weighted integral form.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"states"},{"end_line":188,"line":173,"lines":[173,180],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_feedback_coercivity","note":"The same inequality as an expectation along the finite-feedback diffusion.","plain":"Finite-step coercivity along the diffusion","proof_lines":[181,188],"relation":"states"},{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The half-line lemma with origin flux and reflection that the first theorem calls.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"}],"short":"Finite-data half-line coercivity","statements":[{"anchor":"formal","at":"b242023-242558","end_byte":242558,"heading_id":"the-support-statement-and-the-derivative-bridge","lean":[{"end_line":162,"line":106,"lines":[106,114],"module":"OAI.Probability.SKSupport.Density.GaussianLogDensity","name":"OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","note":"The half-line lemma: origin flux, then evenness of the integrands for the reflection. finite_density_coercivity calls it.","plain":"Coercivity for Gaussian-weighted densities","proof_lines":[115,162],"relation":"proof"},{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"The resulting integrated inequality, density-weighted.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"end_line":194,"line":165,"lines":[165,181],"module":"OAI.Probability.SKSupport.Density.Certificate","name":"OAI.ZeroTemperatureSK.Certificate.integrated_coercivity_of_flux","note":"The generic flux lemma, with global derivative hypothesis hderiv. Nothing else in the tree calls it.","plain":"Integrated coercivity from vanishing flux","proof_lines":[182,194],"relation":"part"}],"lean_source":"statement","markdown":"At the origin, the manuscript smoothly extends auxiliary quotient\nvariables. The formal coercivity proof for the spatial diffusion density instead works on the positive\nhalf-line, controls the zero flux at the origin and reflects. The two\nmust agree in the resulting integrated inequality, while auxiliary values\nat zero may differ. In particular, the declared generic flux lemma has a\nglobal derivative hypothesis and is not the supplier used here. The\nfinite-density proof uses the Gaussian half-line coercivity lemma and\nreflection.","paper":"long","role":"explanation","section":"12.1 The support statement and the derivative bridge","start_byte":242023},{"anchor":"curvature-transfer","at":"b53400-53644","end_byte":53644,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":188,"line":173,"lines":[173,175],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_feedback_coercivity","note":"Positive, monotone steps, with every height at most M.","plain":"Finite-step coercivity along the diffusion","proof_lines":[181,188],"relation":"states"},{"end_line":69,"line":69,"lines":[69,69],"module":"OAI.Probability.SKSupport.Foundations.SoftAbs","name":"OAI.ZeroTemperatureSK.Heat.softAbs","note":"The log-cosh terminal value, written with a division by M.","plain":"Soft absolute value log(2 cosh Mx)/M","relation":"defines"}],"lean_source":"statement","markdown":"The source first works with positive nondecreasing finite-step\ncoefficients and the terminal approximation $M^{-1}\\log(2\\cosh(Mx))$,\nwith $M$ at least as large as every step height, and its sign\ncomparisons use this bound as well as smoothness.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":53400}]},"support.step_tails":{"lean":[{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"Density at grid times on a uniform mesh, with the Gaussian factor written explicitly.","plain":"Finite-step density","relation":"defines"},{"clause":"(∫ w, g ((finiteFeedback hf hLip c h N 0).solution W.driver ((i:ℝ≥0)*h) w) ∂W.law)=\n      ∫ x, finiteDensity c h f N i x*g x","end_line":72,"line":52,"lines":[56,57],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_density_identity","note":"Identifies finiteDensity as the law of the feedback diffusion at grid times. The differentiated tail expansions are not stated.","plain":"Finite-step density is the grid law","proof_lines":[58,72],"relation":"part"},{"end_line":167,"line":162,"lines":[162,167],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.forwardBoundary","note":"Recursive forward factor, whose step subtracts the height change times the backward value, the jump rule.","plain":"Forward factor at grid times","relation":"defines"},{"end_line":150,"line":117,"lines":[117,121],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.certificate_growth","note":"Exponential bounds on F2 and its divergence, the tail estimates used in the Lean. The paper's tail expansions of l and s are not stated.","plain":"Exponential bounds on F₂ and divergence","proof_lines":[122,150],"relation":"part"},{"end_line":63,"line":43,"lines":[43,44],"module":"OAI.Probability.SKSupport.Density.GaussianTails","name":"OAI.ZeroTemperatureSK.Heat.gaussian_expBound_tendsto_zero","note":"An exponentially bounded function times a Gaussian factor tends to zero at plus infinity, which removes the flux at the far end.","plain":"Gaussian decay beats exponential growth","proof_lines":[45,63],"relation":"part"},{"end_line":69,"line":69,"lines":[69,69],"module":"OAI.Probability.SKSupport.Foundations.SoftAbs","name":"OAI.ZeroTemperatureSK.Heat.softAbs","note":"The smooth terminal value log(2 cosh Mx)/M.","plain":"Soft absolute value log(2 cosh Mx)/M","relation":"defines"}],"short":"Finite-step density and tails","statements":[{"anchor":"curvature","at":"b48864-49584","end_byte":49584,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"Density at grid times on a uniform mesh, with the Gaussian factor written explicitly.","plain":"Finite-step density","relation":"defines"},{"clause":"(∫ w, g ((finiteFeedback hf hLip c h N 0).solution W.driver ((i:ℝ≥0)*h) w) ∂W.law)=\n      ∫ x, finiteDensity c h f N i x*g x","end_line":72,"line":52,"lines":[56,57],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_density_identity","note":"Identifies finiteDensity as the law of the feedback diffusion at grid times. The differentiated tail expansions are not stated.","plain":"Finite-step density is the grid law","proof_lines":[58,72],"relation":"part"},{"end_line":167,"line":162,"lines":[162,167],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.forwardBoundary","note":"Recursive forward factor, whose step subtracts the height change times the backward value, the jump rule.","plain":"Forward factor at grid times","relation":"defines"}],"lean_source":"statement","markdown":"Take $0=t_0<\\cdots<t_N=1$,\n$0<c_1\\le\\cdots\\le c_N\\le M$, and\n$\\gamma(t)=c_i$ on $[t_{i-1},t_i)$. The terminal value is\n$M^{-1}\\log(2\\cosh Mx)$. On a step of height $c$, set $l=c\\Phi$ and\n$h=e^l$. Then $h_t=-\\tfrac12h_{xx}$. For $a<t$ in that step the\ndiffusion transition density is\n\n$$\np_{t-a}(x-y)\\frac{h(t,x)}{h(a,y)},\\qquad\np_t(x)=(2\\pi t)^{-1/2}e^{-x^2/(2t)}.\n$$\n\nIts mass is one by $h(a,\\cdot)=P_{t-a}h(t,\\cdot)$. Factoring the\npositive-time density as $f=e^l g$ gives $g_t=\\tfrac12g_{xx}$ on each\nstep. Put $s=-\\partial_x\\log g$. On the first step,\n$g(t,x)=p_t(x)e^{-l(0,0)}$ and $s=x/t$. At a forward jump\n$\\delta=c_+-c_-\\ge0$, continuity of $f$ gives\n$g_+=g_-e^{-\\delta\\Phi}$. The functions $r$ and $s$ are odd.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":48864},{"anchor":"curvature","at":"b49637-50809","end_byte":50809,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[],"lean_source":"statement","markdown":"The comparison arguments take place on unbounded spatial domains, so\nthe boundary estimates must precede the signs. For these fixed smooth\ndata, Gaussian convolution and the finite jump rules give, as\n$x\\to+\\infty$,\n\n$$\nl(t,x)=cx+C(t)+O(e^{-\\kappa x}),\n$$\n$$\ng(t,x)=\\exp\\!\\left(-\\frac{x^2}{2t}+\\beta(t)x+C_0(t)\\right)\n       (1+O(e^{-\\kappa x})).\n$$\n\nThe errors may be differentiated any fixed finite number of times, and\nthe estimates are uniform on each closed step, with positive time\nbounded away from zero for the forward estimate. Their constants may\ndepend on the step data.\n\nFor example, after completing the square in a\nGaussian convolution, the error is an expectation at\n$Y_x=\\lambda x+b+\\sqrt VZ$ with $\\lambda$ bounded below by a positive\nconstant. Split at $Y_x=\\mathbb EY_x/2$. The first part inherits the\nexponential tail. On the second, the Gaussian probability dominates the\npermitted exponential growth of the error and all its specified\nderivatives. This proves the differentiated estimates by finite induction\nthrough heat evolution and jumps. In particular,\n$s=x/t-\\beta(t)+O(e^{-\\kappa x})$, $s_x=1/t+O(e^{-\\kappa x})$ and\n$s_{xx}=O(e^{-\\kappa x})$.","note":"Not in the Lean: it has no tail expansions. The proof uses cruder exponential bounds against Gaussian decay.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49637},{"anchor":"curvature","at":"b20882-21546","end_byte":21546,"heading_id":"the-integrated-estimate","lean":[{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"Density at grid times on a uniform mesh, with the Gaussian factor written explicitly.","plain":"Finite-step density","relation":"defines"},{"clause":"(∫ w, g ((finiteFeedback hf hLip c h N 0).solution W.driver ((i:ℝ≥0)*h) w) ∂W.law)=\n      ∫ x, finiteDensity c h f N i x*g x","end_line":72,"line":52,"lines":[56,57],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_density_identity","note":"Identifies finiteDensity as the law of the feedback diffusion at grid times. The differentiated tail expansions are not stated.","plain":"Finite-step density is the grid law","proof_lines":[58,72],"relation":"part"},{"end_line":167,"line":162,"lines":[162,167],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.forwardBoundary","note":"Recursive forward factor, whose step subtracts the height change times the backward value, the jump rule.","plain":"Forward factor at grid times","relation":"defines"}],"lean_source":"statement","markdown":"We begin with a positive nondecreasing finite-step coefficient\n$\\gamma_{\\rm step}$, that is, a function constant on each interval of a\nfinite partition, and we replace the terminal value $|x|$ by the smooth\napproximation $M^{-1}\\log(2\\cosh Mx)$, with\n$M\\geq\\|\\gamma_{\\rm step}\\|_\\infty$.\nOn a step of height $c$, set $h=e^{c\\Phi}$. Then\n$h_t=-h_{xx}/2$ and $r=h_x/h$, and the same exponential removes the\ndrift from the forward density equation. Indeed, writing $f=hg$ gives\n$$\n\\begin{aligned}\nf_t&=\\tfrac12(hg)_{xx}-(h_xg)_x\\\\\n   &=-\\tfrac12h_{xx}g+\\tfrac12hg_{xx}\n    =h_tg+\\tfrac12hg_{xx}.\n\\end{aligned}\n$$\nComparison with $f_t=h_tg+hg_t$ yields $g_t=g_{xx}/2$.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":20882},{"anchor":"curvature-mechanism","at":"b51310-51648","end_byte":51648,"heading_id":"where-the-strict-curvature-input-comes-from","lean":[{"end_line":72,"line":52,"lines":[52,57],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_density_identity","note":"Expectations along the finite feedback path as integrals against finiteDensity, at grid times, for finite-step data only.","plain":"Finite-step density is the grid law","proof_lines":[58,72],"relation":"part"},{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"The density, available only for finite-step data at grid times.","plain":"Finite-step density","relation":"defines"}],"lean_source":"statement","markdown":"The negative middle term rules out a pointwise positivity argument, so\nthe source averages instead with the spatial probability density\n$f_t(x)$ of $X_t$ at positive time. As in the half-time example,\ngrouping paths by their reached position gives, for any integrable\nfunction $F$,\n\n$$\n\\mathbb E[F(X_t)]=\\int_{\\mathbb R}F(x)f_t(x)\\,dx.\n$$","note":"The Lean has this density only at the finite-step level; for a general coefficient it passes through limits.","paper":"expository","role":"explanation","section":"7.3 Where the strict-curvature input comes from","start_byte":51310},{"anchor":"curvature","at":"b49008-49087","end_byte":49087,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":175,"line":151,"lines":[151,158],"module":"OAI.Probability.SKSupport.Foundations.ExponentialBoundLipschitz","name":"OAI.ZeroTemperatureSK.Heat.logarithmic_heat_equation","note":"The log of a heat solution, divided by c, solves the step equation.","plain":"Hopf–Cole equation for the log-heat flow","proof_lines":[159,175],"relation":"proof"}],"lean_source":"statement","markdown":"On a step of height $c$, set $l=c\\Phi$ and\n$h=e^l$. Then $h_t=-\\tfrac12h_{xx}$.","paper":"long","role":"definition","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49008},{"anchor":"curvature","at":"b49088-49283","end_byte":49283,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":128,"line":98,"lines":[98,103],"module":"OAI.Probability.SKSupport.Parabolic.TiltedTransition","name":"OAI.ZeroTemperatureSK.expected_transition_tiltedMean","note":"The same transition, as an identity against test functions.","plain":"Transition law as a tilted mean","proof_lines":[104,128],"relation":"proof"},{"end_line":156,"line":134,"lines":[134,138],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity_step_pairing","note":"One step of the density, in paired form.","plain":"Density advances one grid step","proof_lines":[139,156],"relation":"part"}],"lean_source":"statement","markdown":"For $a<t$ in that step the\ndiffusion transition density is\n\n$$\np_{t-a}(x-y)\\frac{h(t,x)}{h(a,y)},\\qquad\np_t(x)=(2\\pi t)^{-1/2}e^{-x^2/(2t)}.\n$$\n\nIts mass is one by $h(a,\\cdot)=P_{t-a}h(t,\\cdot)$.","note":"Lean never writes the transition kernel: it pairs with bounded smooth test functions.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49088},{"anchor":"curvature","at":"b49284-49400","end_byte":49400,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"The explicit density: exp of cΦ, the Gaussian term and forwardBoundary.","plain":"Finite-step density","relation":"defines"},{"end_line":21,"line":20,"lines":[20,21],"module":"OAI.Probability.SKSupport.Parabolic.ForwardGeometry","name":"OAI.ZeroTemperatureSK.Heat.forwardScore","note":"s is x/t minus a smooth correction.","plain":"The forward score s","relation":"defines"}],"lean_source":"statement","markdown":"Factoring the\npositive-time density as $f=e^l g$ gives $g_t=\\tfrac12g_{xx}$ on each\nstep. Put $s=-\\partial_x\\log g$.","paper":"long","role":"definition","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49284},{"anchor":"curvature","at":"b49401-49459","end_byte":49459,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":30,"line":14,"lines":[14,17],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity_first_formula","note":"The first-step density.","plain":"Density at the first grid time","proof_lines":[18,30],"relation":"part"}],"lean_source":"statement","markdown":"On the first step,\n$g(t,x)=p_t(x)e^{-l(0,0)}$ and $s=x/t$.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49401},{"anchor":"curvature","at":"b49460-49549","end_byte":49549,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":164,"line":155,"lines":[155,156],"module":"OAI.Probability.SKSupport.Density.RatioComparison","name":"OAI.ZeroTemperatureSK.Heat.datumScore_jump","note":"The score gains δ times the gradient.","plain":"Forward score after a height jump","proof_lines":[157,164],"relation":"part"},{"end_line":167,"line":162,"lines":[162,167],"module":"OAI.Probability.SKSupport.Backward.GaussianBackwardShapes","name":"OAI.ZeroTemperatureSK.Heat.forwardBoundary","note":"The recursion that includes the jump.","plain":"Forward factor at grid times","relation":"defines"}],"lean_source":"statement","markdown":"At a forward jump\n$\\delta=c_+-c_-\\ge0$, continuity of $f$ gives\n$g_+=g_-e^{-\\delta\\Phi}$.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49460},{"anchor":"curvature","at":"b49550-49584","end_byte":49584,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":85,"line":78,"lines":[78,80],"module":"OAI.Probability.SKSupport.Parabolic.ForwardConstruction","name":"OAI.ZeroTemperatureSK.Heat.forwardScore_odd","note":"s is odd; BackwardShape includes oddness of r.","plain":"The forward score is odd","proof_lines":[81,85],"relation":"part"}],"lean_source":"statement","markdown":"The functions $r$ and $s$ are odd.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49550},{"anchor":"curvature","at":"b49637-49752","end_byte":49752,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[{"end_line":117,"line":12,"lines":[12,28],"module":"OAI.Probability.SKSupport.Parabolic.FullLineComparison","name":"OAI.ZeroTemperatureSK.Parabolic.nonnegative_fullLine_bounded","note":"Support model. Comparison on the whole line. A lower bound and drift growth replace tail estimates, so no such estimate precedes the signs.","plain":"Minimum principle on the whole line","proof_lines":[29,117],"relation":"part"},{"end_line":231,"line":129,"lines":[129,146],"module":"OAI.Probability.SKSupport.Parabolic.FullLineComparison","name":"OAI.ZeroTemperatureSK.Parabolic.nonnegative_halfLine_bounded","note":"Support model. The half-line form, which also needs a sign at x=0.","plain":"Minimum principle on the half-line","proof_lines":[147,231],"relation":"part"}],"lean_source":"statement","markdown":"The comparison arguments take place on unbounded spatial domains, so\nthe boundary estimates must precede the signs.","note":"The comparisons on unbounded domains are in the Lean. Their proofs use no tail estimates.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49637},{"anchor":"curvature","at":"b49988-50809","end_byte":50809,"heading_id":"the-smooth-finite-step-problem-and-its-boundary-control","lean":[],"lean_source":"statement","markdown":"The errors may be differentiated any fixed finite number of times, and\nthe estimates are uniform on each closed step, with positive time\nbounded away from zero for the forward estimate. Their constants may\ndepend on the step data.\n\nFor example, after completing the square in a\nGaussian convolution, the error is an expectation at\n$Y_x=\\lambda x+b+\\sqrt VZ$ with $\\lambda$ bounded below by a positive\nconstant. Split at $Y_x=\\mathbb EY_x/2$. The first part inherits the\nexponential tail. On the second, the Gaussian probability dominates the\npermitted exponential growth of the error and all its specified\nderivatives. This proves the differentiated estimates by finite induction\nthrough heat evolution and jumps. In particular,\n$s=x/t-\\beta(t)+O(e^{-\\kappa x})$, $s_x=1/t+O(e^{-\\kappa x})$ and\n$s_{xx}=O(e^{-\\kappa x})$.","note":"Not in the Lean: it has no differentiated tail estimates. The proof uses exponential bounds against Gaussian decay instead.","paper":"long","role":"statement","section":"4.2 The smooth finite-step problem and its boundary control","start_byte":49988},{"anchor":"curvature","at":"b20882-21169","end_byte":21169,"heading_id":"the-integrated-estimate","lean":[{"end_line":167,"line":111,"lines":[111,116],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.Heat.finite_density_coercivity","note":"hc, hmono and hM are these conditions, with coefficient steps c i of width h.","plain":"Finite-step coercivity, density form","proof_lines":[117,167],"relation":"part"},{"end_line":69,"line":69,"lines":[69,69],"module":"OAI.Probability.SKSupport.Foundations.SoftAbs","name":"OAI.ZeroTemperatureSK.Heat.softAbs","note":"log(2cosh Mx)/M.","plain":"Soft absolute value log(2 cosh Mx)/M","relation":"defines"}],"lean_source":"statement","markdown":"We begin with a positive nondecreasing finite-step coefficient\n$\\gamma_{\\rm step}$, that is, a function constant on each interval of a\nfinite partition, and we replace the terminal value $|x|$ by the smooth\napproximation $M^{-1}\\log(2\\cosh Mx)$, with\n$M\\geq\\|\\gamma_{\\rm step}\\|_\\infty$.","paper":"short","role":"definition","section":"3. The integrated estimate","start_byte":20882},{"anchor":"curvature","at":"b21170-21327","end_byte":21327,"heading_id":"the-integrated-estimate","lean":[{"end_line":144,"line":135,"lines":[135,136],"module":"OAI.Probability.SKSupport.Density.DensityHeat","name":"OAI.ZeroTemperatureSK.Heat.varianceHeat_exp_eq_exp_log","note":"The exponential of the value is a heat flow of the terminal exponential.","plain":"Hopf–Cole identity for the heat flow","proof_lines":[137,144],"relation":"proof"}],"lean_source":"statement","markdown":"On a step of height $c$, set $h=e^{c\\Phi}$. Then\n$h_t=-h_{xx}/2$ and $r=h_x/h$, and the same exponential removes the\ndrift from the forward density equation.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":21170},{"anchor":"curvature","at":"b21214-21546","end_byte":21546,"heading_id":"the-integrated-estimate","lean":[{"end_line":128,"line":104,"lines":[104,108],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.density_tilted_pairing","note":"Lean has no forward PDE. It writes the density in closed form and pairs it with heat kernels by Fubini.","plain":"Tilted pairing against a Gaussian density","proof_lines":[109,128],"relation":"proof"},{"end_line":72,"line":52,"lines":[52,57],"module":"OAI.Probability.SKSupport.Density.FiniteDensity","name":"OAI.ZeroTemperatureSK.finite_density_identity","note":"The closed-form density is the law of the feedback diffusion at grid times.","plain":"Finite-step density is the grid law","proof_lines":[58,72],"relation":"part"}],"lean_source":"statement","markdown":"Then\n$h_t=-h_{xx}/2$ and $r=h_x/h$, and the same exponential removes the\ndrift from the forward density equation. Indeed, writing $f=hg$ gives\n$$\n\\begin{aligned}\nf_t&=\\tfrac12(hg)_{xx}-(h_xg)_x\\\\\n   &=-\\tfrac12h_{xx}g+\\tfrac12hg_{xx}\n    =h_tg+\\tfrac12hg_{xx}.\n\\end{aligned}\n$$\nComparison with $f_t=h_tg+hg_t$ yields $g_t=g_{xx}/2$.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":21214},{"anchor":"curvature","at":"b21547-21673","end_byte":21673,"heading_id":"the-integrated-estimate","lean":[{"end_line":132,"line":130,"lines":[130,132],"module":"OAI.Probability.SKSupport.Density.DensityFubini","name":"OAI.ZeroTemperatureSK.Heat.finiteDensity","note":"The exponent: c φ for the backward factor, a Gaussian term and forwardBoundary for the forward one.","plain":"Finite-step density","relation":"defines"}],"lean_source":"statement","markdown":"One backward\nheat factor and one forward heat factor thus determine the law against\nwhich the moment polynomial is integrated.","paper":"short","role":"explanation","section":"3. The integrated estimate","start_byte":21547}]},"support.stieltjes_measure":{"lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The convention itself: mass of [0,t] equals gamma(t).","plain":"μ is γ's Stieltjes measure","relation":"defines"},{"clause":"IsStieltjesMeasure γ (parameterMeasure γ)","end_line":119,"line":112,"lines":[112,112],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure_isStieltjes","note":"Existence, by the constructed measure. This declaration does not state closed relative support.","plain":"The Stieltjes measure exists","proof_lines":[113,119],"relation":"part"},{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The Stieltjes measure of the coefficient plus an atom of size gamma(0) at zero.","plain":"The Stieltjes measure of γ","relation":"defines"}],"short":"Stieltjes measure with initial mass","statements":[{"anchor":"model","at":"b3989-4231","end_byte":4231,"heading_id":"the-sk-model-and-the-parisi-variational-problem","lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The convention itself: mass of [0,t] equals gamma(t).","plain":"μ is γ's Stieltjes measure","relation":"defines"},{"clause":"IsStieltjesMeasure γ (parameterMeasure γ)","end_line":119,"line":112,"lines":[112,112],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure_isStieltjes","note":"Existence, by the constructed measure. This declaration does not state closed relative support.","plain":"The Stieltjes measure exists","proof_lines":[113,119],"relation":"part"},{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The Stieltjes measure of the coefficient plus an atom of size gamma(0) at zero.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"Their Stieltjes measures satisfy\n$\\mu_\\gamma([0,t])=\\gamma(t)$, including\n$\\mu_\\gamma(\\{0\\})=\\gamma(0)$. They are locally finite. They need not be\nprobability measures, and their total mass may be infinite. Support is taken\nrelative to [0,1).","paper":"long","role":"definition","section":"1. The SK model and the Parisi variational problem","start_byte":3989},{"anchor":"model","at":"b3348-3914","end_byte":3914,"heading_id":"the-model-and-the-result","lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The convention itself: mass of [0,t] equals gamma(t).","plain":"μ is γ's Stieltjes measure","relation":"defines"},{"clause":"IsStieltjesMeasure γ (parameterMeasure γ)","end_line":119,"line":112,"lines":[112,112],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure_isStieltjes","note":"Existence, by the constructed measure. This declaration does not state closed relative support.","plain":"The Stieltjes measure exists","proof_lines":[113,119],"relation":"part"},{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The Stieltjes measure of the coefficient plus an atom of size gamma(0) at zero.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"Let $\\mathcal U$ consist of the nonnegative, nondecreasing,\nright-continuous functions $\\gamma:[0,1)\\to[0,\\infty)$ satisfying\n$\\int_0^1\\gamma(t)\\,dt<\\infty$, and define the locally finite\nStieltjes measure by\n$$\n\\mu_\\gamma([0,t])=\\gamma(t),\\qquad 0\\leq t<1.\n$$\nThus $\\mu_\\gamma((s,t])=\\gamma(t)-\\gamma(s)$, so the measure charges\neach interval with the increment of $\\gamma$ across it. It may have a point mass (an atom) of size\n$\\gamma(0)$ at zero, and its total mass need not be finite or equal to\none. Support statements refer to the relative topology of $[0,1)$.","paper":"short","role":"definition","section":"1. The model and the result","start_byte":3348},{"anchor":"order-parameter-example","at":"b8019-8512","end_byte":8512,"heading_id":"what-a-gap-means","lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The convention itself: mass of [0,t] equals gamma(t).","plain":"μ is γ's Stieltjes measure","relation":"defines"},{"clause":"IsStieltjesMeasure γ (parameterMeasure γ)","end_line":119,"line":112,"lines":[112,112],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure_isStieltjes","note":"Existence, by the constructed measure. This declaration does not state closed relative support.","plain":"The Stieltjes measure exists","proof_lines":[113,119],"relation":"part"},{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The Stieltjes measure of the coefficient plus an atom of size gamma(0) at zero.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"**Stieltjes measure** assigns to each interval the increment of $\\gamma$\nacross it:\n\n$$\n\\mu_\\gamma([0,t])=\\gamma(t),\\qquad\n\\mu_\\gamma((a,b])=\\gamma(b)-\\gamma(a).\n$$\n\nThese cumulative values determine the measure uniquely. The first\nequation allows a mass $\\gamma(0)$ at zero, and mass concentrated at one\npoint is called an **atom**. These weights need not be probabilities of\nfinite-system overlaps, and their total need not be one or even finite,\nalthough the mass of each $[0,T]$ with $T<1$","paper":"expository","role":"definition","section":"2. What a gap means","start_byte":8019},{"anchor":"order-parameter-example","at":"b6900-7216","end_byte":7216,"heading_id":"what-a-gap-means","lean":[{"end_line":12,"line":12,"lines":[12,12],"module":"OAI.Probability.SKSupport.Support.FiniteCDF","name":"OAI.ZeroTemperatureSK.finiteCDF","note":"The cumulative function of a finite measure. This three-step case is one instance; Lean states none.","plain":"CDF of a finite measure","relation":"defines"}],"lean_source":"statement","markdown":"What a gap means\n\nA candidate order parameter accumulates weight. Let us begin with a\nfinite example, placing weight two at scale $1/4$ and weight one at\nscale $3/4$. The cumulative function is zero, then two, then three:\n\n$$\n\\gamma(t)=\n\\begin{cases}\n0,&0\\le t<1/4,\\\\\n2,&1/4\\le t<3/4,\\\\\n3,&3/4\\le t<1.\n\\end{cases}\n$$","paper":"expository","role":"example","section":"2. What a gap means","start_byte":6900},{"anchor":"order-parameter-example","at":"b7218-7423","end_byte":7423,"heading_id":"what-a-gap-means","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"Taking H=1 gives these two counts of one area, for any finite measure. The numbers are not in Lean.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"contains"}],"lean_source":"statement","markdown":"Its area is $2(1/2)+3(1/4)=7/4$. We can count the same area by following\neach weight: the first persists for three quarters of the interval and\nthe second for one quarter, giving $2(3/4)+1(1/4)=7/4$ again.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":7218},{"anchor":"order-parameter-example","at":"b8019-8183","end_byte":8183,"heading_id":"what-a-gap-means","lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The first equation, as the Lean convention.","plain":"μ is γ's Stieltjes measure","relation":"defines"},{"end_line":29,"line":26,"lines":[26,27],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.cdf_add_Ioc","note":"The second equation, in additive form with extended nonnegative values.","plain":"γ(b) = γ(a) + μ((a,b])","proof_lines":[28,29],"relation":"part"}],"lean_source":"statement","markdown":"**Stieltjes measure** assigns to each interval the increment of $\\gamma$\nacross it:\n\n$$\n\\mu_\\gamma([0,t])=\\gamma(t),\\qquad\n\\mu_\\gamma((a,b])=\\gamma(b)-\\gamma(a).\n$$","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8019},{"anchor":"order-parameter-example","at":"b8185-8240","end_byte":8240,"heading_id":"what-a-gap-means","lean":[],"lean_source":"statement","markdown":"These cumulative values determine the measure uniquely.","note":"Not in the Lean: uniqueness of the Stieltjes measure is not stated. full_support instead quantifies over every such measure.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8185},{"anchor":"order-parameter-example","at":"b8241-8352","end_byte":8352,"heading_id":"what-a-gap-means","lean":[{"clause":"ENNReal.ofReal (γ.val ⊥) • Measure.dirac ⊥","end_line":110,"line":109,"lines":[110,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The constructed measure adds this atom of size γ(0) at the bottom of Time.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"The first\nequation allows a mass $\\gamma(0)$ at zero, and mass concentrated at one\npoint is called an **atom**.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8241},{"anchor":"order-parameter-example","at":"b8353-8523","end_byte":8523,"heading_id":"what-a-gap-means","lean":[{"end_line":66,"line":65,"lines":[65,66],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.IsStieltjesMeasure","note":"The only condition is that μ of [0,t] equals γ(t). Nothing limits the total mass.","plain":"μ is γ's Stieltjes measure","relation":"defines"}],"lean_source":"statement","markdown":"These weights need not be probabilities of\nfinite-system overlaps, and their total need not be one or even finite,\nalthough the mass of each $[0,T]$ with $T<1$ is finite.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8353},{"anchor":"order-parameter-example","at":"b8468-8512","end_byte":8512,"heading_id":"what-a-gap-means","lean":[{"end_line":197,"line":187,"lines":[187,188],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.finiteOnCompacts","note":"Finite on compact subsets of Time, such as [0,T].","plain":"μ is finite on compact sets","proof_lines":[189,197],"relation":"states"}],"lean_source":"statement","markdown":"although the mass of each $[0,T]$ with $T<1$","paper":"expository","role":"statement","section":"2. What a gap means","start_byte":8468},{"anchor":"order-parameter-example","at":"b8652-8879","end_byte":8879,"heading_id":"what-a-gap-means","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"Lean's exchange is for finite measures only. It never uses this identity for infinite total mass.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"part"}],"lean_source":"statement","markdown":"This is the double count of area from the example: a unit at $s$\ncontributes at every later time, for a total length $1-s$, and the\ngeneral identity uses the theorem that nonnegative contributions may be\nsummed in either order.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8652},{"anchor":"order-parameter-example","at":"b8880-8993","end_byte":8993,"heading_id":"what-a-gap-means","lean":[{"end_line":12,"line":12,"lines":[12,12],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Time","note":"Time is the half-open interval [0,1), so one is not a point of the domain.","plain":"The overlap interval [0,1)","relation":"defines"}],"lean_source":"statement","markdown":"A mass at one would change none of the\ncumulative values below one, which is why one is excluded from the\ndomain.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":8880},{"anchor":"order-parameter-example","at":"b12957-13144","end_byte":13144,"heading_id":"what-a-gap-means","lean":[{"end_line":110,"line":109,"lines":[109,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"For constant γ the Stieltjes measure vanishes and only the atom remains. Lean does not state this case.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"A constant candidate $\\gamma(t)=c>0$ sharpens the distinction. Its\nmeasure is $c\\delta_0$, supported only at zero, so positivity of the\nfunction falls far short of the conclusion we seek.","paper":"expository","role":"example","section":"2. What a gap means","start_byte":12957},{"anchor":"dependencies","at":"b13010-13156","end_byte":13156,"heading_id":"compact-strip-regularity-and-the-terminal-bound","lean":[{"end_line":158,"line":117,"lines":[117,119],"module":"OAI.Probability.SKSupport.Support.CDFParameters","name":"OAI.ZeroTemperatureSK.finiteCDF_fubini","note":"The Fubini interchange for a finite measure.","plain":"Fubini for a finite measure's CDF","proof_lines":[120,158],"relation":"proof"},{"end_line":197,"line":187,"lines":[187,188],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.finiteOnCompacts","note":"Finite on compact sets.","plain":"μ is finite on compact sets","proof_lines":[189,197],"relation":"part"}],"lean_source":"statement","markdown":"Integrability controls\nterminal tails and Fubini interchanges, while local finiteness of the\nStieltjes measure is enough for compact mass removal.","paper":"long","role":"explanation","section":"2.2 Compact-strip regularity and the terminal bound","start_byte":13010},{"anchor":"model","at":"b3609-3722","end_byte":3722,"heading_id":"the-model-and-the-result","lean":[{"end_line":29,"line":26,"lines":[26,27],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.cdf_add_Ioc","note":"The increment identity, in ENNReal.","plain":"γ(b) = γ(a) + μ((a,b])","proof_lines":[28,29],"relation":"states"}],"lean_source":"statement","markdown":"Thus $\\mu_\\gamma((s,t])=\\gamma(t)-\\gamma(s)$, so the measure charges\neach interval with the increment of $\\gamma$","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":3609},{"anchor":"model","at":"b3734-3852","end_byte":3852,"heading_id":"the-model-and-the-result","lean":[{"clause":"ENNReal.ofReal (γ.val ⊥) • Measure.dirac ⊥","end_line":110,"line":109,"lines":[110,110],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.parameterMeasure","note":"The atom at zero is added by hand to the Stieltjes measure of the coefficient.","plain":"The Stieltjes measure of γ","relation":"defines"}],"lean_source":"statement","markdown":"It may have a point mass (an atom) of size\n$\\gamma(0)$ at zero, and its total mass need not be finite or equal to\none.","paper":"short","role":"explanation","section":"1. The model and the result","start_byte":3734},{"anchor":"model","at":"b3853-3914","end_byte":3914,"heading_id":"the-model-and-the-result","lean":[{"end_line":12,"line":12,"lines":[12,12],"module":"OAI.Probability.SKSupport.Model","name":"OAI.ZeroTemperatureSK.Time","note":"Measures are defined on the subtype Time, so support is taken there.","plain":"The overlap interval [0,1)","relation":"defines"}],"lean_source":"statement","markdown":"Support statements refer to the relative topology of $[0,1)$.","paper":"short","role":"definition","section":"1. The model and the result","start_byte":3853},{"anchor":"model","at":"b3915-4042","end_byte":4042,"heading_id":"the-model-and-the-result","lean":[{"end_line":167,"line":148,"lines":[148,152],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.constant_on_support_gap","note":"A gap here is an open interval (a,b) free of support points.","plain":"γ constant on a support gap","proof_lines":[153,167],"relation":"part"}],"lean_source":"statement","markdown":"A support gap is a relatively open interval of positive length and zero\nmass, and the cumulative coefficient is constant on it.","paper":"short","role":"definition","section":"1. The model and the result","start_byte":3915}]},"support.strict_increase":{"lean":[{"end_line":62,"line":47,"lines":[47,48],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"Stated for points of Time with the cumulative convention of the Stieltjes measure.","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"states"},{"end_line":67,"line":64,"lines":[64,65],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.strictMonoOn_of_full_support","note":"The same implication as strict monotonicity on [0,1).","plain":"Full support gives strict increase","proof_lines":[66,67],"relation":"states"},{"clause":"StrictMonoOn (extend γ.val) (Ico (0:ℝ) 1)","end_line":33,"line":19,"lines":[21,21],"module":"OAI.Probability.SKSupport.Main","name":"OAI.ZeroTemperatureSK.minimizer_full_support_actual","note":"Its second clause gives strict increase for a minimiser.","plain":"Full support for the constructed diffusion","proof_lines":[23,33],"relation":"contains"}],"short":"Full support gives strict increase","statements":[{"anchor":"gaps","at":"b82060-82208","end_byte":82208,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":62,"line":47,"lines":[47,62],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"Positive mass of Ioc a b, from a midpoint in the support.","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"states"}],"lean_source":"statement","markdown":"Finally, every nonempty open subinterval $(a,b)$ receives positive mass. Therefore\n$$\n\\gamma(b)-\\gamma(a)=\\mu_\\gamma((a,b])>0\n\\qquad(0\\le a<b<1).\n$$","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":82060},{"anchor":"gap-proof","at":"b16763-16823","end_byte":16823,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":62,"line":47,"lines":[47,48],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"Stated for points of Time with the cumulative convention of the Stieltjes measure.","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"states"},{"end_line":62,"line":47,"lines":[56,62],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"An open interval around the midpoint has positive mass.","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"proof"}],"lean_source":"statement","markdown":"Moreover,\n$\\gamma(t_2)-\\gamma(t_1)=\\mu_\\gamma((t_1,t_2])>0$.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":16763},{"anchor":"order-parameter-example","at":"b9239-9452","end_byte":9452,"heading_id":"what-a-gap-means","lean":[{"end_line":67,"line":64,"lines":[64,65],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.strictMonoOn_of_full_support","note":"Forward direction only: full support gives strict increase. Lean assembles no converse.","plain":"Full support gives strict increase","proof_lines":[66,67],"relation":"part"},{"end_line":62,"line":47,"lines":[47,48],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"The forward step: full support gives positive mass on (a,b], so γ(a)<γ(b).","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"part"},{"end_line":36,"line":31,"lines":[31,33],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.eq_cdf_of_measure_Ioc_zero","note":"Contrapositive of the converse step: zero mass on (a,b] forces equal values. The full converse is not stated.","plain":"γ constant where μ is null","proof_lines":[34,36],"relation":"part"}],"lean_source":"statement","markdown":"Equivalently,\n$\\gamma$ is strictly increasing: full support gives a positive increment\non each $(a,b]$ with $a<b$, and a positive increment on a smaller interval\ninside any neighbourhood gives positive mass there.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":9239},{"anchor":"order-parameter-example","at":"b9239-9353","end_byte":9353,"heading_id":"what-a-gap-means","lean":[{"end_line":62,"line":47,"lines":[56,62],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.lt_cdf_of_full_support","note":"Exactly this: full support gives positive mass on (a,b], so γ(a)<γ(b).","plain":"Full support gives γ(a) < γ(b)","proof_lines":[49,62],"relation":"states"}],"lean_source":"statement","markdown":"Equivalently,\n$\\gamma$ is strictly increasing: full support gives a positive increment\non each $(a,b]$ with $a<b$,","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":9239},{"anchor":"order-parameter-example","at":"b9354-9452","end_byte":9452,"heading_id":"what-a-gap-means","lean":[{"end_line":36,"line":31,"lines":[31,33],"module":"OAI.Probability.SKSupport.Support.Closure","name":"OAI.ZeroTemperatureSK.SupportDeterministic.IsStieltjesMeasure.eq_cdf_of_measure_Ioc_zero","note":"Contrapositive of this step: zero mass on (a,b] gives equal values. Lean never assembles the converse implication.","plain":"γ constant where μ is null","proof_lines":[34,36],"relation":"part"}],"lean_source":"statement","markdown":"and a positive increment on a smaller interval\ninside any neighbourhood gives positive mass there.","note":"The paper's equivalence is one implication in the Lean; the other direction has no declaration.","paper":"expository","role":"explanation","section":"2. What a gap means","start_byte":9354}]},"support.terminal_chord_barrier":{"lean":[],"short":"Terminal convex-chord obstruction","statements":[{"anchor":"gaps","at":"b76489-77099","end_byte":77099,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[],"lean_source":"record","markdown":"One elementary calculus step suffices. Suppose\n$q\\in C^1([r,1))$, $q(r)=r$, $q'(r)\\le1$, $q\\le1$, and $q'$\nis strictly convex on $(r,1)$. If $q(t)\\ge t$ at some $r<t<1$,\nstrict integration below the finite endpoint chord gives\n$$\nt-r\\le\\int_r^t q'(v)\\,dv\n <\\frac{t-r}{2}\\bigl(q'(r)+q'(t)\\bigr),\n$$\nso $q'(t)>1$. For $t<s<1$, convexity then gives\n$$\nq'(s)\\ge q'(t)+\\frac{s-t}{t-r}\\bigl(q'(t)-q'(r)\\bigr)\n\\ge q'(t)>1.\n$$\nConsequently $q(s)\\ge t+(s-t)q'(t)>1$ for $s$ sufficiently\nclose to one, a contradiction. Thus $q(t)<t$ throughout $(r,1)$.\nAll chord endpoints lie below one, so no derivative at one is used.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":76489},{"anchor":"gap-proof","at":"b14707-15230","end_byte":15230,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[],"lean_source":"record","markdown":"A terminal gap $(r,1)$ with nonempty support has $r\\in S$ and a\npositive plateau height. Suppose that $q(t)\\geq t$ for some $r<t<1$.\nStrict convexity of $q'$ and $q'(r)\\leq1$ give\n$$\nt-r\\leq\\int_r^t q'(s)\\,ds\n<\\frac{t-r}{2}\\bigl(q'(r)+q'(t)\\bigr),\n$$\nso $q'(t)>1$. Convexity then gives $q'(s)>q'(t)>1$ for $s>t$, and\nconsequently $q(s)>t+(s-t)q'(t)>1$ for $s<1$ sufficiently close\nto one, contrary to $q\\leq1$. Thus $q(t)<t$ on $(r,1)$,\nand $G(r)<0$ contradicts minimality. This case uses contact only at\nthe left endpoint.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14707},{"anchor":"endpoints","at":"b40058-41091","end_byte":41091,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"**Reaching the diagonal forces a slope above one.**\nPut $h(t)=q'(t)-1$, so $h(r)\\le0$. If $q(t)\\ge t$ at some\n$r<t<1$, strict integration below the chord on the finite interval\n$[r,t]$ would give\n\n$$\n0\\le q(t)-t=\\int_r^t h(v)\\,dv\n<\\frac{t-r}{2}\\bigl(h(r)+h(t)\\bigr)\n\\le\\frac{t-r}{2}h(t).\n$$\n\nHence $h(t)>0$.\n\n**The excess slope persists and violates the bound.**\nTo control a later slope, we apply the same chord rule on the larger\ninterval $[r,s]$ with $t<s<1$. At its interior point $t$ this gives\n\n$$\nh(t)\\le\\frac{s-t}{s-r}h(r)+\\frac{t-r}{s-r}h(s).\n$$\n\nSolving for the right endpoint gives\n\n$$\nh(s)\\ge h(t)+\\frac{s-t}{t-r}\\bigl(h(t)-h(r)\\bigr)\\ge h(t)>0.\n$$\n\nThe bound holds at every time after $t$, so integration gives\n$q(s)\\ge t+(s-t)(1+h(t))$. Choose any time\n\n$$\nt+\\frac{1-t}{1+h(t)}<s<1.\n$$\n\nSuch a choice exists because $h(t)>0$, and the lower bound now gives\n$q(s)>1$, contradicting $q(s)\\le1$ strictly before time one.\nThus $q(t)<t$ throughout $(r,1)$, giving\n$G(r)=\\int_r^1(q(t)-t)\\,dt<0$, contrary to the addition test.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":40058},{"anchor":"endpoints","at":"b39743-39816","end_byte":39816,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"We have\n$q(r)=r$ and $q'(r)\\le1$, and $q'$ is strictly convex on $(r,1)$.","note":"The support model's terminal proof uses none of these three facts. The scalar contact proof uses slope convexity and q≤1.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":39743},{"anchor":"endpoints","at":"b39817-39938","end_byte":39938,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":135,"line":125,"lines":[129,133],"module":"OAI.Probability.SKSupport.Moments.PerturbationLimits","name":"OAI.ZeroTemperatureSK.squareMoment_bounds","note":"The cap on q, from gradient_abs_le_one. In Lean it is used for integrability, not for this barrier.","plain":"The second moment lies in [0,1]","proof_lines":[127,135],"relation":"proof"}],"lean_source":"statement","markdown":"With no contact condition at one, we use instead the bound $q\\le1$,\nwhich holds because $q=\\mathbb E[u^2]$ and $|u|\\le1$.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":39817},{"anchor":"changed-cases","at":"b42869-43073","end_byte":43073,"heading_id":"two-checks-on-the-gap-argument","lean":[],"lean_source":"statement","markdown":"Then check whether the exact bound $q\\le1$ could be replaced\nby mere boundedness: test the calculus function\n$q(t)=t-3t^2/4+t^3$ on $[0,1)$ against the remaining contact and\ncurvature conditions at $r=0$.","note":"Not in the Lean: a calculus exercise. The test polynomial is not there; the exact cap is squareMoment_bounds.","paper":"expository","role":"example","section":"5.5 Two checks on the gap argument","start_byte":42869},{"anchor":"endpoints","at":"b39939-40056","end_byte":40056,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"We show in two\nsteps that $q$ stays below the diagonal throughout the gap, which makes\nthe tail cost $G(r)$ negative.","note":"The support model also finds q below t, but only near one, by a different estimate.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":39939},{"anchor":"endpoints","at":"b41092-41199","end_byte":41199,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"The argument uses only finite-interval convexity and the bound on $q$,\nand it assumes no value for $q'(1)$.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":41092},{"anchor":"changed-case-answer","at":"b58474-59043","end_byte":59043,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"record","markdown":"**[Question 4](#question-4).** Set $h=q'-1$. Since $q(r)=r$, reaching or passing the diagonal at\n$t$ means $\\int_r^t h\\ge0$. Strict convexity places $h$ below its\nendpoint chord, so that integral is less than\n$(t-r)(h(r)+h(t))/2$. With $h(r)\\le0$, this forces $h(t)>0$.\nThe chord rule on $[r,s]$, solved for $h(s)$, keeps every later value\nat least $h(t)$. Integrating gives\n$q(s)\\ge t+(s-t)(1+h(t))>1$ for a suitable $s<1$, violating the\nbound. If $q$ never reaches the diagonal, then $G(r)<0$ instead.\nThus both possibilities contradict the conditions on a minimiser.","note":"This repeats the chord proof of Section 5.4, which the support model does not use and the scalar contact proof does.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":58474},{"anchor":"changed-case-answer","at":"b59045-59378","end_byte":59378,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"statement","markdown":"For the proposed polynomial, $q(0)=0$, $q'(0)=1$, $q'''=6>0$, and\ndirect integration gives $G(s)=s^3(1-s)/4\\ge0$, with $G(0)=0$.\nIts derivative has minimum $13/16$, so $q$ increases to $5/4$ as\n$t\\uparrow1$. It is bounded, yet $q(9/10)=2043/2000>1$, so the other reduced\nconditions cannot replace the exact cap supplied by $|u|\\le1$.","note":"Not in the Lean: a calculus exercise on a test polynomial. The exact cap comes from gradient_abs_le_one.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":59045},{"anchor":"source-record","at":"b62222-62427","end_byte":62427,"heading_id":"formal-verification-and-sources","lean":[],"lean_source":"record","markdown":"The empty-support and terminal convexity arguments of Section 5.4\nare ordinary deductions from the stated analytic inputs, and the finite\nno-tie argument of Section 6 follows from the Gaussian Hamiltonian.","paper":"expository","role":"mention","section":"Formal verification and sources","start_byte":62222},{"anchor":"gaps","at":"b76347-76488","end_byte":76488,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":73,"line":14,"lines":[14,16],"module":"OAI.Probability.SKValue.Variation.MinimizerContact","name":"OAI.SKValue.IsMinimizer.contactPotential_zero","note":"Scalar model: slope convexity and q≤1 exclude a positive hump of G, even one reaching time one.","plain":"G vanishes on (0,1)","proof_lines":[17,73],"relation":"proof"},{"end_line":149,"line":137,"lines":[137,139],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.endpoint_condition","note":"q≤1 replaces right contact when b=1.","plain":"Endpoint condition on a hump","proof_lines":[140,149],"relation":"proof"}],"lean_source":"statement","markdown":"**A shorter qualitative exclusion.** The curvature input of §4 also\nexcludes a terminal support gap, with no estimate of its boundary layer.","note":"The support model uses the quantitative route; the scalar model formalises this chord exclusion.","paper":"long","role":"explanation","section":"5. Excluding terminal, interior and initial gaps","start_byte":76347},{"anchor":"gap-proof","at":"b14707-14795","end_byte":14795,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Lean excludes every plateau reaching one, height zero included, so this case needs no separate height argument.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"A terminal gap $(r,1)$ with nonempty support has $r\\in S$ and a\npositive plateau height.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14707},{"anchor":"gap-proof","at":"b14796-15265","end_byte":15265,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":136,"line":110,"lines":[110,136],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.positive_hump_impossible","note":"The scalar model's route: convexity of the slope and q≤1 exclude a positive hump of G reaching one.","plain":"No positive hump of G","proof_lines":[117,136],"relation":"proof"},{"end_line":149,"line":137,"lines":[143,149],"module":"OAI.Probability.SKValue.Coercivity.GapBoundary","name":"OAI.SKValue.endpoint_condition","note":"q≤1 produces a later point where q−t has not grown.","plain":"Endpoint condition on a hump","proof_lines":[140,149],"relation":"proof"}],"lean_source":"statement","markdown":"Suppose that $q(t)\\geq t$ for some $r<t<1$.\nStrict convexity of $q'$ and $q'(r)\\leq1$ give\n$$\nt-r\\leq\\int_r^t q'(s)\\,ds\n<\\frac{t-r}{2}\\bigl(q'(r)+q'(t)\\bigr),\n$$\nso $q'(t)>1$. Convexity then gives $q'(s)>q'(t)>1$ for $s>t$, and\nconsequently $q(s)>t+(s-t)q'(t)>1$ for $s<1$ sufficiently close\nto one, contrary to $q\\leq1$. Thus $q(t)<t$ on $(r,1)$,\nand $G(r)<0$ contradicts minimality. This case uses contact only at\nthe left endpoint.\n\nIf $S$ is empty, then $\\gamma=0$.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":14796},{"anchor":"gap-proof","at":"b15118-15180","end_byte":15180,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":51,"line":15,"lines":[15,18],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one","note":"The support model's proof gives q<t only near one, from a square-root deficit in the second moment.","plain":"Terminal plateau gives q(t)<t near one","proof_lines":[19,51],"relation":"proof"},{"end_line":78,"line":53,"lines":[72,78],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"Then G(b)<0 contradicting contactPotential_nonneg.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"proof"}],"lean_source":"statement","markdown":"Thus $q(t)<t$ on $(r,1)$,\nand $G(r)<0$ contradicts minimality.","paper":"short","role":"derivation","section":"2. Contact conditions and the exclusion of gaps","start_byte":15118}]},"support.terminal_defect":{"lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"Fixes a base time s past a; explicit smallness conditions on T play the role of the record's t0.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"short":"Square-root terminal deficit","statements":[{"anchor":"gaps","at":"b72528-72903","end_byte":72903,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"Fixes a base time s past a; explicit smallness conditions on T play the role of the record's t0.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"lean_source":"statement","markdown":"**A terminal plateau violates the variational inequality.** Suppose that $\\gamma=c$ on $(a,1)$, where $0\\le a<1$ and\n$c\\ge0$ is finite. The endpoint $1$ lies outside the support domain and\ncarries no contact condition, so an endpoint-chord argument is\nunavailable. Instead we prove\n$$\n1-q(t)\\ge C\\sqrt{1-t}\n\\tag{5.1}\n$$\nfor some $C>0$ and every $t$ sufficiently close to one.","paper":"long","role":"statement","section":"5. Excluding terminal, interior and initial gaps","start_byte":72528},{"anchor":"gaps","at":"b74385-75342","end_byte":75342,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"Proves (5.1) with its own constant: |u|≤1/2 on a window from a curvature bound, with no tilted sign probabilities.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"proof"}],"lean_source":"statement","markdown":"Now write $\\varepsilon=1-t$ and restrict to\n$|y|\\le\\sqrt\\varepsilon$. Differentiating the terminal heat formula\ngives, with the unweighted interpretation for $c=0$,\n$$\nu(t,y)=\n\\frac{\\mathbb E[\\operatorname{sgn}(y+\\sqrt\\varepsilon Z)\n                     e^{c|y+\\sqrt\\varepsilon Z|}]}\n     {\\mathbb E e^{c|y+\\sqrt\\varepsilon Z|}}.\n$$\nThe denominator is at most\n$D=e^c\\mathbb E e^{c|Z|}<\\infty$. Each sign has unnormalised\nmass at least $p_*:=\\mathbb P(Z\\ge1)>0$, since the threshold\n$|y/\\sqrt\\varepsilon|$ is at most one, and the exponential weight\nis at least one. Hence each sign has tilted probability at least\n$\\theta=p_*/D\\in(0,1/2)$. It follows that\n$$\n|u(t,y)|\\le1-2\\theta,\\qquad\n1-u(t,y)^2\\ge4\\theta(1-\\theta)=:\\kappa>0.\n$$\nMultiplying this uniform deficit by the density lower bound (5.3)\nand the window width $2\\sqrt\\varepsilon$ yields\n$$\n1-q(t)=\\int_{\\mathbb R}(1-u(t,y)^2)f_t(y)\\,dy\n\\ge 2m\\kappa\\sqrt{1-t}.\n$$\nThis proves (5.1) with $C=2m\\kappa$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":74385},{"anchor":"gap-proof","at":"b15936-16081","end_byte":16081,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"Fixes a base time s past a; explicit smallness conditions on T play the role of the record's t0.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"lean_source":"statement","markdown":"A stronger boundary estimate\n$1-q(t)\\geq C\\sqrt{1-t}$, for arbitrary coefficients constant near\none, is proved in [OpenAI (2026a), Lemma 6.3][FS]","paper":"short","role":"mention","section":"2. Contact conditions and the exclusion of gaps","start_byte":15936},{"anchor":"endpoints","at":"b41201-41403","end_byte":41403,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"clause":"C*Real.sqrt (1-(T:ℝ)) ≤ 1-squareMoment W γ (diffusion W γ) T","end_line":208,"line":124,"lines":[129,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"This is the proof the Lean uses. Its hypotheses have only the constant tail: no contact, no minimiser.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"lean_source":"statement","markdown":"By a different terminal argument, the source proves the stronger\nestimate $1-q(t)\\ge C\\sqrt{1-t}$ near one for any finite constant\nterminal coefficient, with no need for the left contact data used here.","paper":"expository","role":"mention","section":"5.4 The remaining initial and terminal gaps","start_byte":41201},{"anchor":"gaps","at":"b73882-74383","end_byte":74383,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":208,"line":124,"lines":[131,146],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"The set |X_s|<1 has positive mass; the constant K contains the kernel lower bound.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"proof"},{"end_line":122,"line":111,"lines":[111,115],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_denominator","note":"The upper bound on the normaliser.","plain":"Denominator bound on a terminal plateau","proof_lines":[116,122],"relation":"proof"},{"end_line":186,"line":146,"lines":[146,152],"module":"OAI.Probability.SKSupport.Density.DensityHeat","name":"OAI.ZeroTemperatureSK.Heat.varianceTilted_lower_window","note":"The lower bound on the window.","plain":"Tilted kernel lower bound, window","proof_lines":[153,186],"relation":"proof"}],"lean_source":"statement","markdown":"Put $L=1-s$ and $\\delta=L/2$. For\n$s+\\delta\\le t<1$, $|x|\\le R$ and $|y|\\le1$,\n$$\np_{t-s}(y-x)\\ge\n(2\\pi L)^{-1/2}e^{-(R+1)^2/(2\\delta)}=:k_*>0,\n$$\nwhile $h(t,y)\\ge1$ and\n$h(s,x)\\le e^{cR}\\mathbb E e^{c\\sqrt L|Z|}=:M<\\infty$.\nIntegrating (5.2) over the part of $\\rho_s$ in $[-R,R]$ therefore\ngives a density $f_t$ of $X_t$ with\n$$\nf_t(y)\\ge p_0k_*/M=:m>0\n\\quad(s+\\delta\\le t<1,\\ |y|\\le1).\n\\tag{5.3}\n$$\nAt time $s$ we used only the positive mass of a compact set, with no\ndensity assumption on $\\rho_s$.","note":"Lean fixes R=1, using positivity of the interval (−1,1), and bounds a tilted expectation rather than a density of X_t.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":73882},{"anchor":"gaps","at":"b74385-75115","end_byte":75115,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":100,"line":77,"lines":[77,81],"module":"OAI.Probability.SKSupport.Support.TerminalCurvature","name":"OAI.ZeroTemperatureSK.gradient_terminal_small","note":"|u| at most 1/2 on a window of radius proportional to the square root of 1−t.","plain":"Gradient small near zero, late times","proof_lines":[82,100],"relation":"proof"},{"end_line":75,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.TerminalCurvature","name":"OAI.ZeroTemperatureSK.curvature_terminal_bound","note":"The bound behind it: curvature at most 2κ over that square root.","plain":"Curvature bound of order 1/√(1−t)","proof_lines":[67,75],"relation":"proof"}],"lean_source":"statement","markdown":"Now write $\\varepsilon=1-t$ and restrict to\n$|y|\\le\\sqrt\\varepsilon$. Differentiating the terminal heat formula\ngives, with the unweighted interpretation for $c=0$,\n$$\nu(t,y)=\n\\frac{\\mathbb E[\\operatorname{sgn}(y+\\sqrt\\varepsilon Z)\n                     e^{c|y+\\sqrt\\varepsilon Z|}]}\n     {\\mathbb E e^{c|y+\\sqrt\\varepsilon Z|}}.\n$$\nThe denominator is at most\n$D=e^c\\mathbb E e^{c|Z|}<\\infty$. Each sign has unnormalised\nmass at least $p_*:=\\mathbb P(Z\\ge1)>0$, since the threshold\n$|y/\\sqrt\\varepsilon|$ is at most one, and the exponential weight\nis at least one. Hence each sign has tilted probability at least\n$\\theta=p_*/D\\in(0,1/2)$. It follows that\n$$\n|u(t,y)|\\le1-2\\theta,\\qquad\n1-u(t,y)^2\\ge4\\theta(1-\\theta)=:\\kappa>0.\n$$","note":"Lean proves this differently: a curvature bound from the mild equation, with no tilted sign probabilities.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":74385},{"anchor":"gaps","at":"b75116-75342","end_byte":75342,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"An existential constant only, with no value 2mκ, under a smallness condition on 1−T. The proof bounds a tilted expectation.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"part"}],"lean_source":"statement","markdown":"Multiplying this uniform deficit by the density lower bound (5.3)\nand the window width $2\\sqrt\\varepsilon$ yields\n$$\n1-q(t)=\\int_{\\mathbb R}(1-u(t,y)^2)f_t(y)\\,dy\n\\ge 2m\\kappa\\sqrt{1-t}.\n$$\nThis proves (5.1) with $C=2m\\kappa$.","paper":"long","role":"derivation","section":"5. Excluding terminal, interior and initial gaps","start_byte":75116},{"anchor":"dependencies","at":"b17532-17685","end_byte":17685,"heading_id":"prior-results-and-their-roles","lean":[{"end_line":208,"line":124,"lines":[124,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"The quantitative estimate.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"lean_source":"statement","markdown":"Their terminal-plateau observation anticipates one endpoint\nobstruction, and the argument below supplies the quantitative\nboundary-layer proof used here.","paper":"long","role":"mention","section":"2.3 Prior results and their roles","start_byte":17532},{"anchor":"gap-proof","at":"b15936-16036","end_byte":16036,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":208,"line":124,"lines":[127,129],"module":"OAI.Probability.SKSupport.Support.ContactPotential","name":"OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","note":"For T near one, with height c≥0 allowed and the constant depending on the plateau.","plain":"Square-root terminal deficit","proof_lines":[130,208],"relation":"states"}],"lean_source":"statement","markdown":"A stronger boundary estimate\n$1-q(t)\\geq C\\sqrt{1-t}$, for arbitrary coefficients constant near\none,","paper":"short","role":"mention","section":"2. Contact conditions and the exclusion of gaps","start_byte":15936}]},"support.zero_candidate_descent":{"lean":[],"short":"Zero coefficient is not minimising","statements":[{"anchor":"gaussian-sanity-check","at":"b194178-195304","end_byte":195304,"heading_id":"the-zero-coefficient","lean":[],"lean_source":"record","markdown":"The admissible coefficient $\\gamma_0=0$ makes the use of minimality in\nthe normalization step explicit. Its diffusion is $X_t=B_t$, and the\n[zero-coefficient heat formula][fs-reg] gives\n\n$$\nu_0(t,x)=2F_{\\mathrm G}\\!\\left(\\frac{x}{\\sqrt{1-t}}\\right)-1,\n\\qquad\na_0(t,x)=\\sqrt{\\frac2\\pi}\\frac{e^{-x^2/(2(1-t))}}{\\sqrt{1-t}}.\n$$\n\nSince $\\mathbb E e^{-\\lambda B_t^2}=(1+2\\lambda t)^{-1/2}$,\n\n$$\n\\mathbb E a_0(t,B_t)=\\sqrt{\\frac2\\pi},\\qquad\n\\mathbb E a_0(t,B_t)^2=\\frac2{\\pi\\sqrt{1-t^2}},\\qquad\nq_0(t)=\\frac2\\pi\\arcsin t.\n$$\n\nThe last identity follows by integrating $q_0'=\\mathbb E a_0^2$\nfrom $q_0(0)=0$. This heat and late-step variation argument is the\nzero-field specialization of [Chen, Handschy and Lerman, Theorem 6\nand its proof, arXiv:1609.04368v4, pp. 13–16](https://arxiv.org/abs/1609.04368v4).\nStrict convexity of $\\arcsin$ gives $q_0(t)<t$\nfor $0<t<1$. Thus $G_0(s)=\\int_s^1(q_0(t)-t)\\,dt<0$ for\nevery $s<1$. Inserting an atom at $s$ is admissible, and its\n[first variation][fs-var] is $G_0(s)/2<0$. In particular, the derivative for inserting\nmass at zero is $1/4-1/\\pi<0$. Hence this coefficient does not\nminimize.","paper":"long","role":"statement","section":"10.3 The zero coefficient","start_byte":194178},{"anchor":"gaps","at":"b77842-77956","end_byte":77956,"heading_id":"excluding-terminal-interior-and-initial-gaps","lean":[],"lean_source":"record","markdown":"The\n[explicit zero-coefficient calculation](#zero-coefficient-example)\nalso gives $D_+\\mathcal P(0)[1]=1/4-1/\\pi$,","paper":"long","role":"mention","section":"5. Excluding terminal, interior and initial gaps","start_byte":77842},{"anchor":null,"at":"b46470-47605","end_byte":47605,"heading_id":"a-descent-direction-for-the-zero-candidate","lean":[],"lean_source":"record","markdown":"We return to the Gaussian value and slope of Section 3. With $\\gamma=0$\nthe state is $B_t$, and differentiating the slope once more gives\n\n$$\nX_t=B_t,\\qquad\nu_0(t,x)=2F_{\\mathrm G}\\!\\left(\\frac{x}{\\sqrt{1-t}}\\right)-1,\n\\qquad\na_0(t,x)=\\sqrt{\\frac2{\\pi(1-t)}}e^{-x^2/(2(1-t))}.\n$$\n\nA Gaussian integral gives\n$\\mathbb E e^{-B_t^2/(1-t)}=\\sqrt{(1-t)/(1+t)}$. Substituting\ninto $q_0'=\\mathbb E[a_0(t,B_t)^2]$ and integrating from zero yields\n\n$$\nq_0'(t)=\\frac2{\\pi\\sqrt{1-t^2}},\\qquad\nq_0(t)=\\frac2\\pi\\arcsin t.\n$$\n\nThis recovers the half-time value $1/3$, and it shows that\n$q_0(t)\\to1$ while $q_0'(t)\\to\\infty$, so convergence of a function\nat one gives no bound on its derivative there. The constant direction\n$\\eta=1$ is admissible and adds an atom at zero, at the cost\n\n$$\nD\\mathcal P(0)[1]\n=\\frac12\\int_0^1\\left(\\frac2\\pi\\arcsin t-t\\right)dt\n=\\frac14-\\frac1\\pi<0.\n$$\n\nA small positive constant coefficient therefore improves the zero\ncandidate. The calculation uses the heat and variation formulas and no\nminimiser hypothesis (see also [Chen–Handschy–Lerman, proof of Theorem 6][chl-optimality]).\n<!-- /passage:zero-candidate -->","paper":"expository","role":"statement","section":"7.1 A descent direction for the zero candidate","start_byte":46470},{"anchor":"endpoints","at":"b39438-39520","end_byte":39520,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[],"lean_source":"record","markdown":"Section 7.1 also computes the exact first variation\n$D\\mathcal P(0)[1]=1/4-1/\\pi$.","paper":"expository","role":"mention","section":"5.4 The remaining initial and terminal gaps","start_byte":39438},{"anchor":"spins-and-overlap","at":"b4259-4356","end_byte":4356,"heading_id":"the-finite-optimisation-problem","lean":[{"end_line":79,"line":54,"lines":[54,55],"module":"OAI.Probability.SKSupport.Foundations.IntegralIoiMulExpSq","name":"OAI.ZeroTemperatureSK.Gaussian.gaussian_abs_moment","note":"Absolute moment of a centred Gaussian. Variance one gives this value.","plain":"First absolute Gaussian moment","proof_lines":[56,79],"relation":"states"},{"end_line":25,"line":14,"lines":[14,17],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.BrownianSystem.abs_increment","note":"At t=0 the increment is B_1, a standard Gaussian, so this gives the square root of 2/π.","plain":"Mean absolute Brownian increment","proof_lines":[18,25],"relation":"contains"}],"lean_source":"statement","markdown":"Symmetry gives the zero mean, whereas\n$\\mathbb E|Z|=2\\int_0^\\infty z\\varphi(z)\\,dz=\\sqrt{2/\\pi}$.","note":"A worked Gaussian integral the Lean also uses, for the terminal-value bound.","paper":"expository","role":"example","section":"1. The finite optimisation problem","start_byte":4259},{"anchor":"value-and-diffusion","at":"b14912-15216","end_byte":15216,"heading_id":"the-zero-coefficient","lean":[{"clause":"value W γ t=varianceLogHeat c (1-t) abs","end_line":68,"line":35,"lines":[38,38],"module":"OAI.Probability.SKSupport.Backward.TerminalTailValue","name":"OAI.ZeroTemperatureSK.value_on_terminal_plateau","note":"With c=0 on a zero tail, varianceLogHeat is the Gaussian average of |x|, which is this formula.","plain":"Explicit value on a terminal plateau","proof_lines":[39,68],"relation":"contains"}],"lean_source":"statement","markdown":"Over an\ninterval of length $h$, an increment has the law of the centred\nGaussian $\\sqrt h Z$, so from time $t$ in state $x$ the remaining\nnoise $B_1-B_t$ has the law of $\\sqrt{1-t}Z$. The expected\nfinal reward is\n\n$$\n\\Phi_0(t,x)=\\mathbb E|x+\\sqrt{1-t}Z|\n=\\int_{\\mathbb R}|x+\\sqrt{1-t}z|\\varphi(z)\\,dz.\n$$","paper":"expository","role":"explanation","section":"3.1 The zero coefficient","start_byte":14912},{"anchor":"value-and-diffusion","at":"b15218-15352","end_byte":15352,"heading_id":"the-zero-coefficient","lean":[{"end_line":25,"line":14,"lines":[14,17],"module":"OAI.Probability.SKSupport.Diffusion.AbsIncrement","name":"OAI.ZeroTemperatureSK.BrownianSystem.abs_increment","note":"At t=0 this is the expected absolute value of B_1. Lean does not name the zero candidate's value.","plain":"Mean absolute Brownian increment","proof_lines":[18,25],"relation":"part"}],"lean_source":"statement","markdown":"At $(t,x)=(0,0)$ we recover the earlier calculation,\n$\\Phi_0(0,0)=\\sqrt{2/\\pi}$, and at $t=1$ no noise remains and the\nvalue is $|x|$.","paper":"expository","role":"example","section":"3.1 The zero coefficient","start_byte":15218},{"anchor":"value-and-diffusion","at":"b17910-18076","end_byte":18076,"heading_id":"allowing-a-response-to-the-noise","lean":[],"lean_source":"record","markdown":"The no-gap theorem assumes a minimiser and proves a property of it, so\nthe value $\\mathcal P(0)=\\sqrt{2/\\pi}$ alone is silent on whether\nthe zero candidate minimises.","paper":"expository","role":"mention","section":"3.2 Allowing a response to the noise","start_byte":17910},{"anchor":"value-and-diffusion","at":"b19405-19739","end_byte":19739,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"For the zero coefficient we can compute the slope directly. Put\n$F_{\\mathrm G}(z)=\\Pr(Z\\le z)=\\int_{-\\infty}^z\\varphi(y)\\,dy$.\nMoving $x$\nslightly raises $|x+\\sqrt{1-t}Z|$ when the final state is positive\nand lowers it when the final state is negative. Consequently\n\n$$\nu_0(t,x)=2F_{\\mathrm G}\\!\\left(\\frac{x}{\\sqrt{1-t}}\\right)-1.\n$$","note":"The Lean never computes this slope.","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":19405},{"anchor":"value-and-diffusion","at":"b19658-20021","end_byte":20021,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"Consequently\n\n$$\nu_0(t,x)=2F_{\\mathrm G}\\!\\left(\\frac{x}{\\sqrt{1-t}}\\right)-1.\n$$\n\nDifferentiating under the integral is legitimate, because the\ndifference quotients of the absolute value are bounded by one and the\nGaussian hits the corner with probability zero. Thus the slope is the probability\nof a positive final state minus the probability of a negative one.","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":19658},{"anchor":"value-and-diffusion","at":"b22584-23070","end_byte":23070,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"**A calculation at half time.** For $\\gamma=0$ we have $X_t=B_t$, and\n$B_{1/2}$ has the same distribution as $Z/\\sqrt2$. We can average\nover paths by grouping them according to the state reached, which at\nhalf time has density $e^{-x^2}/\\sqrt\\pi$. This density describes the\n**law** of the state, assigning probability to positions on the real\nline. The law has total mass one even though the order-parameter\nmeasure $\\mu_0$ is zero, because the two measures describe different\nobjects.","note":"The Lean has no zero-coefficient moment calculation.","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":22584},{"anchor":"value-and-diffusion","at":"b23072-23338","end_byte":23338,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"We average $u_0(1/2,x)^2$ against this density. Substituting\n$x=z/\\sqrt2$ gives\n\n$$\n\\begin{aligned}\nq_0(1/2)\n&=\\int_{\\mathbb R}(2F_{\\mathrm G}(z)-1)^2\\varphi(z)\\,dz\\\\\n&=\\int_0^1(2v-1)^2\\,dv=\\frac13,\n\\qquad v=F_{\\mathrm G}(z),\\quad dv=\\varphi(z)\\,dz.\n\\end{aligned}\n$$","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":23072},{"anchor":"value-and-diffusion","at":"b23340-23682","end_byte":23682,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"The response at zero is zero, yet its mean square at the random state\nis $1/3$, so the identity $q(t)=t$ is far from automatic. Although every\ncontrol is optimal at $\\gamma=0$, $q_0$ uses the value slope $u_0$.\nA discrepancy at one time is too little to lower $\\mathcal P(0)$, and\nSection 7.1 uses the whole curve to find a feasible decrease.","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":23340},{"anchor":"value-and-diffusion","at":"b23709-24129","end_byte":24129,"heading_id":"the-response-the-state-and-the-averaged-square","lean":[],"lean_source":"record","markdown":"**Question 2: change the time and the future.** Keep $\\gamma=0$\neverywhere, but now set $t=1/4$. What variance has the state accumulated, and\nwhat variance remains in its future reward? Write the Gaussian integral\nfor $q_0(1/4)$ without evaluating it.\n\nNow take $\\gamma=0$ before $1/2$ and $\\gamma=1$ afterward.\nDoes that same integral still compute $q(1/4)$? Identify what remains\nunchanged and what must be recomputed.","paper":"expository","role":"example","section":"3.3 The response, the state and the averaged square","start_byte":23709},{"anchor":"changed-case-answer","at":"b56486-57016","end_byte":57016,"heading_id":"answers-to-the-questions","lean":[],"lean_source":"record","markdown":"**[Question 2](#question-2).** For $\\gamma=0$, the state is $X_{1/4}=B_{1/4}$, a\ncentred Gaussian with variance $1/4$, so it has the law of $Z/2$.\nThe remaining independent Gaussian noise $B_1-B_{1/4}$ has variance\n$3/4$ and standard deviation $\\sqrt3/2$.\nThus\n\n$$\nq_0(1/4)=\\int_{\\mathbb R}\n\\left(2F_{\\mathrm G}(z/\\sqrt3)-1\\right)^2\\varphi(z)\\,dz.\n$$\n\nThe two variances agree at half time and differ here. For nonzero $z$\nthe response is now smaller in magnitude, so $q_0(1/4)<1/3$, as the\narcsine formula of Section 7.1 confirms.","note":"A zero-coefficient calculation, not in the Lean.","paper":"expository","role":"example","section":"7.4 Answers to the questions","start_byte":56486}]},"support.zero_curvature":{"lean":[{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Gives positivity of the derivative of q′ on the interval.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"clause":"HasDerivAt (curvatureMoment W γ) (stripSecond W γ T c t) t","end_line":37,"line":30,"lines":[33,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"With stripSecond_eq_actual at c=0, this supplies the identity with the third spatial derivative.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"},{"clause":"stripSecond W γ T c t = ∫ ξ, (actualJet W γ t 3 ξ)^2-2*c*(actualJet W γ t 2 ξ)^3 ∂W.law","end_line":34,"line":22,"lines":[24,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"Expands stripSecond as an expectation; at c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"}],"short":"Strict increase, zero plateau","statements":[{"anchor":"fixed-time-coercivity","at":"b69835-70384","end_byte":70384,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Gives positivity of the derivative of q′ on the interval.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"clause":"HasDerivAt (curvatureMoment W γ) (stripSecond W γ T c t) t","end_line":37,"line":30,"lines":[33,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"With stripSecond_eq_actual at c=0, this supplies the identity with the third spatial derivative.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"},{"clause":"stripSecond W γ T c t = ∫ ξ, (actualJet W γ t 3 ξ)^2-2*c*(actualJet W γ t 2 ξ)^3 ∂W.law","end_line":34,"line":22,"lines":[24,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"Expands stripSecond as an expectation; at c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"}],"lean_source":"statement","markdown":"When $c=0$, the heat equation gives\n$q''(t)=\\mathbb E\\Phi_{xxx}(t,X_t)^2$ without division by $c$.\nStart the drift-free evolution at an earlier point of $I$. Its\nGaussian transition kernel gives the law of $X_t$ a positive\ndensity everywhere. If the expectation were zero, continuity would\nforce $\\Phi_{xxx}(t,\\cdot)\\equiv0$. Then $\\Phi(t,\\cdot)$ would\nbe an even quadratic polynomial. Its one-Lipschitz bound forces that\npolynomial to be constant, contradicting its bounded distance from\n$|x|$. Therefore $q''(t)>0$, giving strict increase of $q'$.","paper":"long","role":"statement","section":"4.6 The zero interval as a separate case","start_byte":69835},{"anchor":"gap-proof","at":"b13291-13348","end_byte":13348,"heading_id":"contact-conditions-and-the-exclusion-of-gaps","lean":[{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"The strict sign of q″.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"At c=0 the moment reduces to the square of jet 3.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":96,"line":92,"lines":[92,96],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_zero_pos","note":"","plain":"q″ expression positive at c=0","proof_lines":[94,96],"relation":"proof"}],"lean_source":"statement","markdown":"If $c=0$, then $q''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0$.","paper":"short","role":"statement","section":"2. Contact conditions and the exclusion of gaps","start_byte":13291},{"anchor":"curvature","at":"b26689-27070","end_byte":27070,"heading_id":"the-integrated-estimate","lean":[{"clause":"∀ t ∈ Ioo a b, 0 < deriv (curvatureMoment W γ) t","end_line":71,"line":63,"lines":[66,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"Gives positivity of the derivative of q′ on the interval.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"clause":"HasDerivAt (curvatureMoment W γ) (stripSecond W γ T c t) t","end_line":37,"line":30,"lines":[33,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"With stripSecond_eq_actual at c=0, this supplies the identity with the third spatial derivative.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"},{"clause":"stripSecond W γ T c t = ∫ ξ, (actualJet W γ t 3 ξ)^2-2*c*(actualJet W γ t 2 ξ)^3 ∂W.law","end_line":34,"line":22,"lines":[24,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"Expands stripSecond as an expectation; at c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"}],"lean_source":"statement","markdown":"The zero plateau has the separate heat identity stated in Lemma 2. Its\nstrictness uses a positive diffusion density and the fact that\n$\\Phi_{xxx}$ cannot vanish identically, since an even quadratic profile\nwould contradict the one-Lipschitz bound and $|x|+O(1)$ growth.\n\n**Passage to a fixed time.** Now let $\\gamma\\in\\mathcal U$ be continuous\nat $t\\in(0,1)$, with $c=\\gamma(t)>0$.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26689},{"anchor":"endpoints","at":"b36979-37173","end_byte":37173,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"The expected polynomial. At c=0 only the squared third derivative remains.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":37,"line":30,"lines":[30,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"Identifies stripSecond with q'' on the interval.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"},{"end_line":96,"line":92,"lines":[92,96],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_zero_pos","note":"The strict sign.","plain":"q″ expression positive at c=0","proof_lines":[94,96],"relation":"part"}],"lean_source":"statement","markdown":"Then $\\gamma=0$ there,\nso the positive-coefficient bound gives no strictness, and the\nzero-coefficient heat identity takes its place:\n\n$$\nq''(t)=\\mathbb E[\\Phi_{xxx}(t,X_t)^2]>0\\qquad(0<t<s).\n$$","paper":"expository","role":"statement","section":"5.4 The remaining initial and terminal gaps","start_byte":36979},{"anchor":"endpoints","at":"b37175-38159","end_byte":38159,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":63,"line":58,"lines":[58,59],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualThirdSquare_pos","note":"The strict sign, for any γ: the third derivative is not identically zero and X_t charges every interval.","plain":"Second moment of ∂ₓ³Φ is positive","proof_lines":[60,63],"relation":"proof"},{"end_line":83,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Diffusion.Diffusion","name":"OAI.ZeroTemperatureSK.diffusion_interval_pos","note":"Lean uses a more general fact: for any γ, at positive time, X_t charges every open interval, with no density and no X=B.","plain":"Xₜ charges every interval","proof_lines":[80,83],"relation":"part"},{"end_line":73,"line":46,"lines":[53,71],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.third_not_identically_zero","note":"Zero third derivative makes a constant, so u=cx since u(0)=0, and the bound at x=2/c forces c=0.","plain":"∂ₓa is not identically zero","proof_lines":[48,73],"relation":"proof"},{"end_line":70,"line":54,"lines":[64,70],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.abs_le_value","note":"The same two inequalities, as a calc block, with the zero control.","plain":"The value dominates |x|","proof_lines":[56,70],"relation":"proof"},{"end_line":44,"line":31,"lines":[37,44],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.curvature_not_identically_zero","note":"The proof ends differently: zero gradient makes the value constant, then x=value+1 contradicts |x|≤value, with no tail slopes.","plain":"The curvature is not identically zero","proof_lines":[33,44],"relation":"proof"}],"lean_source":"statement","markdown":"The inequality is strict for the following reason. On this interval\n$X_t=B_t$, whose density is positive everywhere at positive times. If\nthe expected square were zero, continuity would force\n$\\Phi_{xxx}(t,x)=0$ for every $x$, since a nonzero value would give a\nsmall interval of positive probability on which the square is positive.\nThen $\\Phi(t,\\cdot)$ would be a quadratic polynomial, and its bounded\nslope would make it affine, a linear function plus a constant.\nChoosing zero control, however, gives\n$\\Phi_\\gamma(t,x)\\ge\\mathbb E|x+B_1-B_t|\\ge|x|$, because the average\nabsolute value is at least the absolute value of the average. An affine\nfunction cannot dominate $|x|$ in this way: the positive tail would\nrequire slope at least one and the negative tail slope at most minus\none. The heat identity and the needed regularity are\nin [OpenAI (2026a), Lemmas 2.2–2.3][support-regularity].\n\nThus $q'$ is strictly increasing, and right contact gives\n$q'(t)<q'(s)\\le1$ for $0<t<s$.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37175},{"anchor":"endpoints","at":"b37226-37309","end_byte":37309,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":83,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Diffusion.Diffusion","name":"OAI.ZeroTemperatureSK.diffusion_interval_pos","note":"Lean uses a more general fact: for any γ, at positive time, X_t charges every open interval, with no density and no X=B.","plain":"Xₜ charges every interval","proof_lines":[80,83],"relation":"part"}],"lean_source":"statement","markdown":"On this interval\n$X_t=B_t$, whose density is positive everywhere at positive times.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37226},{"anchor":"endpoints","at":"b37310-37508","end_byte":37508,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":29,"line":14,"lines":[14,29],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.integral_pos_of_interval_support","note":"A continuous nonnegative function, nonzero somewhere, has positive integral against such a law.","plain":"Positive mean when intervals are charged","proof_lines":[18,29],"relation":"proof"}],"lean_source":"statement","markdown":"If\nthe expected square were zero, continuity would force\n$\\Phi_{xxx}(t,x)=0$ for every $x$, since a nonzero value would give a\nsmall interval of positive probability on which the square is positive.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37310},{"anchor":"endpoints","at":"b37509-37641","end_byte":37641,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":73,"line":46,"lines":[53,71],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.third_not_identically_zero","note":"Zero third derivative makes a constant, so u=cx since u(0)=0, and the bound at x=2/c forces c=0.","plain":"∂ₓa is not identically zero","proof_lines":[48,73],"relation":"proof"}],"lean_source":"statement","markdown":"Then $\\Phi(t,\\cdot)$ would be a quadratic polynomial, and its bounded\nslope would make it affine, a linear function plus a constant.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37509},{"anchor":"endpoints","at":"b37642-37810","end_byte":37810,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":70,"line":54,"lines":[64,70],"module":"OAI.Probability.SKSupport.Control.ExpectedControlPayoff","name":"OAI.ZeroTemperatureSK.abs_le_value","note":"The same two inequalities, as a calc block, with the zero control.","plain":"The value dominates |x|","proof_lines":[56,70],"relation":"proof"}],"lean_source":"statement","markdown":"Choosing zero control, however, gives\n$\\Phi_\\gamma(t,x)\\ge\\mathbb E|x+B_1-B_t|\\ge|x|$, because the average\nabsolute value is at least the absolute value of the average.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37642},{"anchor":"endpoints","at":"b37811-37962","end_byte":37962,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"end_line":44,"line":31,"lines":[37,44],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.curvature_not_identically_zero","note":"The proof ends differently: zero gradient makes the value constant, then x=value+1 contradicts |x|≤value, with no tail slopes.","plain":"The curvature is not identically zero","proof_lines":[33,44],"relation":"proof"}],"lean_source":"statement","markdown":"An affine\nfunction cannot dominate $|x|$ in this way: the positive tail would\nrequire slope at least one and the negative tail slope at most minus\none.","paper":"expository","role":"derivation","section":"5.4 The remaining initial and terminal gaps","start_byte":37811},{"anchor":"endpoints","at":"b38260-38393","end_byte":38393,"heading_id":"the-remaining-initial-and-terminal-gaps","lean":[{"clause":"(ht : 0 < (t:ℝ))","end_line":96,"line":92,"lines":[93,93],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_zero_pos","note":"Positive time is a hypothesis, needed by diffusion_interval_pos since X_0 is a point mass. The vanishing at (0,0) is not in the Lean.","plain":"q″ expression positive at c=0","proof_lines":[94,96],"relation":"part"}],"lean_source":"statement","markdown":"Strictness\nwas needed only at positive times, which matters because symmetry can\nmake the third spatial derivative vanish at $(0,0)$.","paper":"expository","role":"explanation","section":"5.4 The remaining initial and terminal gaps","start_byte":38260},{"anchor":"dependencies","at":"b7592-7704","end_byte":7704,"heading_id":"outline-of-the-proof","lean":[{"end_line":71,"line":63,"lines":[63,66],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.zero_gap_curvature","note":"The zero interval.","plain":"q″ positive on zero plateaus","proof_lines":[67,71],"relation":"part"},{"end_line":78,"line":53,"lines":[53,54],"module":"OAI.Probability.SKSupport.Support.NoTerminalPlateau","name":"OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","note":"The terminal plateau, excluded separately.","plain":"No terminal plateau at a minimiser","proof_lines":[55,78],"relation":"part"}],"lean_source":"statement","markdown":"The zero-interval case\nuses a separate heat argument, and the possible terminal plateau needs its\nown exclusion.","paper":"long","role":"mention","section":"2.1 Outline of the proof","start_byte":7592},{"anchor":"fixed-time-coercivity","at":"b69835-69933","end_byte":69933,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"end_line":34,"line":22,"lines":[22,24],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_eq_actual","note":"At c=0 the second term drops out.","plain":"The q″ expression as an expectation","proof_lines":[25,34],"relation":"part"},{"end_line":37,"line":30,"lines":[30,33],"module":"OAI.Probability.SKSupport.Support.GapCurvature","name":"OAI.ZeroTemperatureSK.curvatureMoment_hasDerivAt_on_constant","note":"The time derivative.","plain":"Formula for q″ on a plateau","proof_lines":[34,37],"relation":"part"}],"lean_source":"statement","markdown":"When $c=0$, the heat equation gives\n$q''(t)=\\mathbb E\\Phi_{xxx}(t,X_t)^2$ without division by $c$.","paper":"long","role":"statement","section":"4.6 The zero interval as a separate case","start_byte":69835},{"anchor":"fixed-time-coercivity","at":"b69934-70077","end_byte":70077,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"end_line":83,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Diffusion.Diffusion","name":"OAI.ZeroTemperatureSK.diffusion_interval_pos","note":"Every interval has positive probability, for any coefficient.","plain":"Xₜ charges every interval","proof_lines":[80,83],"relation":"proof"}],"lean_source":"statement","markdown":"Start the drift-free evolution at an earlier point of $I$. Its\nGaussian transition kernel gives the law of $X_t$ a positive\ndensity everywhere.","note":"Lean uses less than a density: positive probability of every interval, proved for all t>0.","paper":"long","role":"derivation","section":"4.6 The zero interval as a separate case","start_byte":69934},{"anchor":"fixed-time-coercivity","at":"b70078-70384","end_byte":70384,"heading_id":"the-zero-interval-as-a-separate-case","lean":[{"end_line":63,"line":58,"lines":[58,63],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualThirdSquare_pos","note":"Positivity of the expectation.","plain":"Second moment of ∂ₓ³Φ is positive","proof_lines":[60,63],"relation":"proof"},{"end_line":73,"line":46,"lines":[46,73],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.third_not_identically_zero","note":"The third derivative is nonzero somewhere.","plain":"∂ₓa is not identically zero","proof_lines":[48,73],"relation":"proof"},{"end_line":96,"line":92,"lines":[92,96],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.stripSecond_zero_pos","note":"The conclusion on the strip.","plain":"q″ expression positive at c=0","proof_lines":[94,96],"relation":"proof"}],"lean_source":"statement","markdown":"If the expectation were zero, continuity would\nforce $\\Phi_{xxx}(t,\\cdot)\\equiv0$. Then $\\Phi(t,\\cdot)$ would\nbe an even quadratic polynomial. Its one-Lipschitz bound forces that\npolynomial to be constant, contradicting its bounded distance from\n$|x|$. Therefore $q''(t)>0$, giving strict increase of $q'$.","paper":"long","role":"derivation","section":"4.6 The zero interval as a separate case","start_byte":70078},{"anchor":"curvature","at":"b26756-27070","end_byte":27070,"heading_id":"the-integrated-estimate","lean":[{"end_line":63,"line":58,"lines":[58,59],"module":"OAI.Probability.SKSupport.Moments.Curvature","name":"OAI.ZeroTemperatureSK.actualThirdSquare_pos","note":"E b²>0 at every positive time. Positive diffusion mass on intervals is used in place of a positive density.","plain":"Second moment of ∂ₓ³Φ is positive","proof_lines":[60,63],"relation":"proof"},{"end_line":29,"line":14,"lines":[14,17],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.integral_pos_of_interval_support","note":"Interval positivity of the diffusion is used in place of a positive density.","plain":"Positive mean when intervals are charged","proof_lines":[18,29],"relation":"proof"},{"end_line":83,"line":77,"lines":[77,79],"module":"OAI.Probability.SKSupport.Diffusion.Diffusion","name":"OAI.ZeroTemperatureSK.diffusion_interval_pos","note":"","plain":"Xₜ charges every interval","proof_lines":[80,83],"relation":"part"},{"end_line":73,"line":46,"lines":[46,73],"module":"OAI.Probability.SKSupport.Moments.StrictMoments","name":"OAI.ZeroTemperatureSK.third_not_identically_zero","note":"If Φ_xxx vanished, u would be linear; boundedness and u(0)=0 force zero slope, and |x|≤Φ then fails.","plain":"∂ₓa is not identically zero","proof_lines":[48,73],"relation":"part"}],"lean_source":"statement","markdown":"Its\nstrictness uses a positive diffusion density and the fact that\n$\\Phi_{xxx}$ cannot vanish identically, since an even quadratic profile\nwould contradict the one-Lipschitz bound and $|x|+O(1)$ growth.\n\n**Passage to a fixed time.** Now let $\\gamma\\in\\mathcal U$ be continuous\nat $t\\in(0,1)$, with $c=\\gamma(t)>0$.","paper":"short","role":"derivation","section":"3. The integrated estimate","start_byte":26756}]}},"steps":{"full_support→support.diffusion→supports":{"route":["OAI.ZeroTemperatureSK.full_support","OAI.ZeroTemperatureSK.diffusion_isDiffusion"]},"full_support→support.full_relative_support→supports":{"route":["OAI.ZeroTemperatureSK.full_support"]},"full_support→support.moment_saturation→supports":{"route":["OAI.ZeroTemperatureSK.full_support"]},"full_support→support.stieltjes_measure→supports":{"route":["OAI.ZeroTemperatureSK.full_support","OAI.ZeroTemperatureSK.parameterMeasure"]},"full_support→support.strict_increase→supports":{"route":["OAI.ZeroTemperatureSK.full_support","OAI.ZeroTemperatureSK.minimizer_full_support_actual"]},"scalar.conditional_rounding→support.regularity→supports":{"route":null},"scalar.constant_curvature→support.fixed_time_spatial_coercivity→supports":{"route":["OAI.ZeroTemperatureSK.stripThird_coercive","OAI.ZeroTemperatureSK.actual_coercivity"]},"scalar.constant_curvature→support.integrated_coercivity→supports":{"route":["OAI.ZeroTemperatureSK.stripThird_coercive","OAI.ZeroTemperatureSK.actual_coercivity","OAI.ZeroTemperatureSK.sample_coercivity","OAI.ZeroTemperatureSK.finite_feedback_coercivity","OAI.ZeroTemperatureSK.Heat.finite_density_coercivity"]},"scalar.constant_curvature→support.moment_identity→supports":{"route":null},"scalar.constant_curvature→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.positive_gap_curvature","OAI.ZeroTemperatureSK.stripThird_pos","OAI.ZeroTemperatureSK.actualQuartic_pos","OAI.ZeroTemperatureSK.actual_jet_even_moment_pos","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds"]},"scalar.correlation_vector_control→scalar.conditional_rounding→supports":{"route":["OAI.SKValue.mesh_coefficient_errors","OAI.SKValue.finite_coefficient_L2_errors","OAI.SKValue.finite_rounding_coefficient"]},"scalar.correlation_vector_control→scalar.euler_normalization→supports":{"route":["OAI.SKValue.mesh_coefficient_errors","OAI.SKValue.finite_coefficient_L2_errors","OAI.SKValue.normalized_family_orthonormal"]},"scalar.euler_normalization→support.moment_saturation→supports":{"route":null},"scalar.euler_normalization→support.regularity→supports":{"route":null},"scalar.formal_finite_gaussian_clause→scalar.formal_contact_input→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsMinimizer.gradient_moment_contact"]},"scalar.formal_finite_gaussian_clause→scalar.source_fixed_strip_application→supports":{"route":["OAI.SKValue.finite_gaussian_target"]},"scalar.formal_finite_gaussian_clause→scalar.source_ground_state_identity_with_diffusion→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact"]},"scalar.formal_finite_gaussian_clause→scalar.source_terminal_parisi_assembly→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.formal_parisi_clause→scalar.source_diffusion_exists→supports":{"route":["OAI.SKValue.ground_state_parisi_target","OAI.SKValue.diffusionProcess_isDiffusion"]},"scalar.formal_parisi_clause→scalar.source_ground_state_identity_with_diffusion→supports":{"route":["OAI.SKValue.ground_state_parisi_target","OAI.SKValue.IsMinimizer.ground_state_parisi_of_diffusion"]},"scalar.formal_scalar_clause→scalar.formal_contact_input→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsMinimizer.gradient_moment_contact"]},"scalar.formal_scalar_clause→scalar.source_ground_state_identity_with_diffusion→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact"]},"scalar.formal_scalar_clause→scalar.source_spin_sequence_limit→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValueG.groundStateSequence_tendsto"]},"scalar.formal_scalar_clause→scalar.source_terminal_parisi_assembly→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.integrated_coercivity→support.rational_certificate→supports":{"route":["OAI.ZeroTemperatureSK.Certificate.coercivity_constant_integral","OAI.ZeroTemperatureSK.Certificate.coercivity_constant"]},"scalar.lean_gaussian_target→scalar.source_fixed_strip_application→supports":{"route":["OAI.SKValue.finite_gaussian_target"]},"scalar.lean_gaussian_target→scalar.source_ground_state_identity_with_diffusion→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact"]},"scalar.lean_gaussian_target→scalar.source_terminal_parisi_assembly→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.lean_parisi_target→scalar.source_diffusion_exists→supports":{"route":["OAI.SKValue.ground_state_parisi_target","OAI.SKValue.diffusionProcess_isDiffusion"]},"scalar.lean_scalar_target→scalar.source_ground_state_identity_with_diffusion→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact"]},"scalar.lean_scalar_target→scalar.source_spin_sequence_limit→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValueG.groundStateSequence_tendsto"]},"scalar.lean_scalar_target→scalar.source_terminal_parisi_assembly→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.manuscript_terminal_value→scalar.parisi_formula→supports":{"route":["OAI.SKValue.scalar_value_target","OAI.SKValue.IsDiffusion.scalar_value_of_contact","OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","OAI.SKValue.groundStateValue_le_parisi"]},"scalar.manuscript_terminal_value→support.moment_identity→supports":{"route":null},"scalar.manuscript_terminal_value→support.moment_saturation→supports":{"route":null},"scalar.manuscript_terminal_value→support.regularity→supports":{"route":null},"scalar.shifted_gaussian_limit→scalar.correlation_vector_control→supports":{"route":["OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip","OAI.SKValue.finiteGaussian_from_diffusion","OAI.SKValue.finiteGaussian_from_strip_and_moments","OAI.SKValue.gradientMeshError_tendsto_zero"]},"scalar.shifted_gaussian_limit→scalar.euler_normalization→supports":{"route":["OAI.SKValue.finite_gaussian_target"]},"scalar.shifted_gaussian_limit→scalar.manuscript_terminal_value→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.source_finite_adaptive_lower→scalar.source_gaussian_column_split→supports":{"route":["OAI.SKValueG.finite_adaptive_SK_lower","OAI.SKValueG.adaptive_spin_lower","OAI.SKValueG.adaptive_value_identity","OAI.SKValueG.random_column_identity","OAI.SKValueG.quadraticValue_column"]},"scalar.source_fixed_strip_application→scalar.source_minimizer_curvature_moment→supports":{"route":["OAI.SKValue.finite_gaussian_target","OAI.SKValue.IsMinimizer.gradient_moment_contact"]},"scalar.source_fixed_strip_application→scalar.source_strip_regularity→supports":{"route":["OAI.SKValue.IsDiffusion.finite_gaussian_of_contact","OAI.SKValue.sourceStripRegularity"]},"scalar.source_ground_state_identity_with_diffusion→scalar.source_ground_state_upper→supports":{"route":["OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","OAI.SKValue.groundStateValue_le_parisi"]},"scalar.source_ground_state_identity_with_diffusion→scalar.source_strip_value_lower→supports":{"route":["OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","OAI.SKValue.IsDiffusion.strip_value_le_groundState"]},"scalar.source_ground_state_identity_with_diffusion→scalar.source_terminal_parisi_assembly→supports":{"route":["OAI.SKValue.IsDiffusion.groundState_eq_parisi_of_contact","OAI.SKValue.IsDiffusion.parisi_value_assembly"]},"scalar.source_ground_state_upper→scalar.source_spin_sequence_limit→supports":{"route":["OAI.SKValue.groundStateValue_le_parisi","OAI.SKValueG.groundStateSequence_tendsto"]},"scalar.source_minimizer_curvature_moment→scalar.formal_contact_input→supports":{"route":["OAI.SKValue.IsMinimizer.gradient_moment_contact"]},"scalar.source_minimizer_curvature_moment→scalar.source_strip_regularity→supports":{"route":["OAI.SKValue.IsDiffusion.curvature_unit_moment_of_contact","OAI.SKValue.SourceStripRegularity"]},"scalar.source_predictable_lower→scalar.source_empirical_energy_limit→supports":{"route":["OAI.SKValueG.predictable_coefficients_le_groundState","OAI.SKValueG.empiricalColumnEnergy_ae_tendsto"]},"scalar.source_predictable_lower→scalar.source_finite_adaptive_lower→supports":{"route":["OAI.SKValueG.predictable_coefficients_le_groundState","OAI.SKValueG.empirical_finite_spin_lower","OAI.SKValueG.finite_adaptive_SK_lower"]},"scalar.source_predictable_lower→scalar.source_spin_sequence_limit→supports":{"route":["OAI.SKValueG.predictable_coefficients_le_groundState","OAI.SKValueG.groundStateSequence_tendsto"]},"scalar.source_strip_value_lower→scalar.source_fixed_strip_application→supports":{"route":["OAI.SKValue.IsDiffusion.strip_value_le_groundState","OAI.SKValue.IsDiffusion.finite_gaussian_fixed_strip"]},"scalar.source_strip_value_lower→scalar.source_predictable_lower→supports":{"route":["OAI.SKValue.shiftedCoefficientSum_le_groundState","OAI.SKValue.normalized_shifted_sum_le_groundState","OAI.SKValueG.predictable_coefficients_le_groundState"]},"scalar.source_terminal_parisi_assembly→scalar.formal_contact_input→supports":{"route":null},"scalar.source_terminal_parisi_assembly→scalar.source_minimizer_curvature_moment→supports":{"route":null},"scalar.source_terminal_parisi_assembly→scalar.source_strip_regularity→supports":{"route":["OAI.SKValue.IsDiffusion.parisi_value_assembly","OAI.SKValue.SourceStripRegularity"]},"scalar.terminal_loss→scalar.manuscript_terminal_value→supports":{"route":null},"scalar.terminal_loss→support.moment_saturation→supports":{"route":null},"scalar.terminal_loss→support.regularity→supports":{"route":null},"scalar_value→scalar.formal_finite_gaussian_clause→supports":{"route":["OAI.SKValue.value_consequences"]},"scalar_value→scalar.formal_parisi_clause→supports":{"route":["OAI.SKValue.value_consequences"]},"scalar_value→scalar.formal_scalar_clause→supports":{"route":["OAI.SKValue.value_consequences"]},"support.backward_shapes→support.step_tails→supports":{"route":["OAI.ZeroTemperatureSK.Heat.finiteHead_backward_shape","OAI.ZeroTemperatureSK.Heat.softAbs"]},"support.contact_derivatives→support.contact→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_contact","OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal"]},"support.contact_derivatives→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_contact","OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt"]},"support.contact→support.first_variation→supports":{"route":["OAI.ZeroTemperatureSK.contactPotential_nonneg","OAI.ZeroTemperatureSK.minimizer_variational_inequality"]},"support.contact→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.contactPotential_nonneg","OAI.ZeroTemperatureSK.contactIntegrand_integrable","OAI.ZeroTemperatureSK.squareMoment_integrable","OAI.ZeroTemperatureSK.squareMoment_bounds"]},"support.contact→support.stieltjes_measure→supports":{"route":["OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal","OAI.ZeroTemperatureSK.IsStieltjesMeasure"]},"support.diffusion→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.diffusion_uniqueness","OAI.ZeroTemperatureSK.stripDrift","OAI.ZeroTemperatureSK.compactDrift_lipschitz","OAI.ZeroTemperatureSK.compactGradient_lipschitz","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds"]},"support.first_variation→support.diffusion→supports":{"route":null},"support.first_variation→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.parisi_chord_derivative","OAI.ZeroTemperatureSK.squareMoment_weighted_integrable","OAI.ZeroTemperatureSK.squareMoment_bounds"]},"support.first_variation→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.parisi_chord_derivative","OAI.ZeroTemperatureSK.normalizedValueDifference_limit","OAI.ZeroTemperatureSK.mixedQ_weighted_limit","OAI.ZeroTemperatureSK.mixedQ_perturbation_limit","OAI.ZeroTemperatureSK.compactGradient_continuous","OAI.ZeroTemperatureSK.value_derivative_continuousOn"]},"support.fixed_time_spatial_coercivity→support.diffusion→supports":{"route":["OAI.ZeroTemperatureSK.actual_coercivity","OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","OAI.ZeroTemperatureSK.sampleJet_limit","OAI.ZeroTemperatureSK.positivePath_sample_limit","OAI.ZeroTemperatureSK.positivePath_uniform_error","OAI.ZeroTemperatureSK.positivePath_error"]},"support.fixed_time_spatial_coercivity→support.integrated_coercivity→supports":{"route":["OAI.ZeroTemperatureSK.actual_coercivity","OAI.ZeroTemperatureSK.sample_coercivity","OAI.ZeroTemperatureSK.finite_feedback_coercivity","OAI.ZeroTemperatureSK.Heat.finite_density_coercivity"]},"support.fixed_time_spatial_coercivity→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.actual_coercivity","OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","OAI.ZeroTemperatureSK.sampleJet_limit","OAI.ZeroTemperatureSK.value_derivative_continuousOn"]},"support.flux_identity→support.backward_shapes→supports":{"route":["OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","OAI.ZeroTemperatureSK.Heat.BackwardShape"]},"support.flux_identity→support.forward_shapes→supports":{"route":null},"support.flux_identity→support.step_tails→supports":{"route":["OAI.ZeroTemperatureSK.Heat.certificate_growth"]},"support.forward_shapes→support.backward_shapes→supports":{"route":["OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","OAI.ZeroTemperatureSK.Heat.boundaryRatio_nonneg","OAI.ZeroTemperatureSK.Heat.backwardBoundary_shape","OAI.ZeroTemperatureSK.Heat.BackwardShape"]},"support.forward_shapes→support.step_tails→supports":{"route":["OAI.ZeroTemperatureSK.Heat.finite_forward_ratio","OAI.ZeroTemperatureSK.Heat.forwardBoundary"]},"support.full_relative_support_curvature→support.contact_derivatives→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.minimizer_contact"]},"support.full_relative_support_curvature→support.contact→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.minimizer_contact","OAI.ZeroTemperatureSK.contactPotential_compact_mass_removal"]},"support.full_relative_support_curvature→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt"]},"support.full_relative_support_curvature→support.no_bounded_gap→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature"]},"support.full_relative_support_curvature→support.no_initial_gap→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature"]},"support.full_relative_support_curvature→support.positive_curvature→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.positive_gap_curvature"]},"support.full_relative_support_curvature→support.stieltjes_measure→supports":{"route":["OAI.ZeroTemperatureSK.full_support","OAI.ZeroTemperatureSK.parameterMeasure"]},"support.full_relative_support_curvature→support.zero_curvature→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.zero_gap_curvature"]},"support.full_relative_support→support.no_bounded_gap→supports":{"route":["OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature"]},"support.full_relative_support→support.no_initial_gap→supports":{"route":["OAI.ZeroTemperatureSK.SupportDeterministic.Variation.full_support_of_contact_and_gap_curvature"]},"support.full_relative_support→support.no_terminal_gap→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau"]},"support.full_relative_support→support.stieltjes_measure→supports":{"route":["OAI.ZeroTemperatureSK.full_support","OAI.ZeroTemperatureSK.parameterMeasure"]},"support.integrated_coercivity→support.flux_identity→supports":{"route":["OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity"]},"support.integrated_coercivity→support.rational_certificate→supports":{"route":["OAI.ZeroTemperatureSK.Heat.gaussian_shape_coercivity","OAI.ZeroTemperatureSK.Certificate.coercivity_constant"]},"support.moment_identity→support.diffusion→supports":{"route":null},"support.moment_identity→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt","OAI.ZeroTemperatureSK.stripA_continuous","OAI.ZeroTemperatureSK.compactGradient_family","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds"]},"support.moment_saturation→support.contact_derivatives→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.minimizer_contact"]},"support.moment_saturation→support.diffusion→supports":{"route":["OAI.ZeroTemperatureSK.diffusion_uniqueness"]},"support.moment_saturation→support.full_relative_support→supports":{"route":["OAI.ZeroTemperatureSK.full_support"]},"support.moment_saturation→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.squareMoment_hasDerivWithinAt"]},"support.no_bounded_gap→support.contact_derivatives→supports":{"route":null},"support.no_bounded_gap→support.endpoint_chord_area→supports":{"route":["OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_positive_constant_gap","OAI.ZeroTemperatureSK.SupportDeterministic.Variation.no_gap_of_strictConvex_derivative"]},"support.no_bounded_gap→support.moment_identity→supports":{"route":null},"support.no_bounded_gap→support.positive_curvature→supports":{"route":null},"support.no_bounded_gap→support.stieltjes_measure→supports":{"route":null},"support.no_bounded_gap→support.zero_curvature→supports":{"route":null},"support.no_initial_gap→support.contact_derivatives→supports":{"route":null},"support.no_initial_gap→support.moment_identity→supports":{"route":null},"support.no_initial_gap→support.stieltjes_measure→supports":{"route":null},"support.no_initial_gap→support.zero_curvature→supports":{"route":null},"support.no_terminal_gap→support.contact→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","OAI.ZeroTemperatureSK.contactPotential_nonneg"]},"support.no_terminal_gap→support.stieltjes_measure→supports":{"route":null},"support.no_terminal_gap→support.terminal_defect→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_no_terminal_plateau","OAI.ZeroTemperatureSK.terminal_plateau_q_lt_near_one","OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect"]},"support.optimality_criterion→support.first_variation→supports":{"route":["OAI.ZeroTemperatureSK.contactPotential_nonneg","OAI.ZeroTemperatureSK.minimizer_variational_inequality"]},"support.optimality_criterion→support.moment_identity→supports":{"route":["OAI.ZeroTemperatureSK.contactPotential_nonneg","OAI.ZeroTemperatureSK.contactIntegrand_integrable","OAI.ZeroTemperatureSK.squareMoment_integrable","OAI.ZeroTemperatureSK.squareMoment_bounds"]},"support.optimality_criterion→support.stieltjes_measure→supports":{"route":null},"support.plateau_transfer→support.diffusion→supports":{"route":["OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","OAI.ZeroTemperatureSK.sampleJet_limit","OAI.ZeroTemperatureSK.positivePath_sample_limit","OAI.ZeroTemperatureSK.positivePath_uniform_error","OAI.ZeroTemperatureSK.positivePath_error"]},"support.plateau_transfer→support.moment_identity→supports":{"route":null},"support.plateau_transfer→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.sampleCoercive_integral_limit","OAI.ZeroTemperatureSK.sampleJet_limit","OAI.ZeroTemperatureSK.value_derivative_continuousOn"]},"support.positive_curvature→support.fixed_time_spatial_coercivity→supports":{"route":["OAI.ZeroTemperatureSK.stripThird_coercive","OAI.ZeroTemperatureSK.actual_coercivity"]},"support.positive_curvature→support.integrated_coercivity→supports":{"route":["OAI.ZeroTemperatureSK.stripThird_coercive","OAI.ZeroTemperatureSK.actual_coercivity","OAI.ZeroTemperatureSK.sample_coercivity","OAI.ZeroTemperatureSK.finite_feedback_coercivity","OAI.ZeroTemperatureSK.Heat.finite_density_coercivity"]},"support.positive_curvature→support.plateau_transfer→supports":{"route":["OAI.ZeroTemperatureSK.positive_gap_curvature","OAI.ZeroTemperatureSK.curvatureMoment_deriv_hasDerivAt_on_constant"]},"support.strict_increase→support.full_relative_support→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual"]},"support.strict_increase→support.stieltjes_measure→supports":{"route":["OAI.ZeroTemperatureSK.minimizer_full_support_actual","OAI.ZeroTemperatureSK.IsStieltjesMeasure"]},"support.terminal_defect→support.diffusion→supports":{"route":null},"support.terminal_defect→support.moment_identity→supports":{"route":null},"support.terminal_defect→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.terminal_plateau_second_moment_defect","OAI.ZeroTemperatureSK.value_regular","OAI.ZeroTemperatureSK.compactGradient_family","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds"]},"support.zero_curvature→support.diffusion→supports":{"route":null},"support.zero_curvature→support.moment_identity→supports":{"route":null},"support.zero_curvature→support.regularity→supports":{"route":["OAI.ZeroTemperatureSK.zero_gap_curvature","OAI.ZeroTemperatureSK.stripSecond_zero_pos","OAI.ZeroTemperatureSK.actualThirdSquare_pos","OAI.ZeroTemperatureSK.actual_jet_even_moment_pos","OAI.ZeroTemperatureSK.value_uniform_derivative_bounds"]}}}