Plectis Atlas
Search

↑ and ↓ move, Enter opens, Esc closes.

Optimal logarithmic mixing of the Thorp shuffle

OpenAI groups related papers into a “family”. OpenAI’s catalogue describes this one:

Proves that the Thorp shuffle randomizes \(N=2^d\) labeled cards in \(\Theta(\log N)\) physical shuffles, settling its optimal mixing order for power-of-two deck sizes. Convergence is in total variation from the worst initial ordering and concerns the entire permutation, not just individual card positions. (OpenAI’s Lean note)

Lean is a language for mathematical statements and proofs that a computer can verify. The assessments below ask whether those statements express the result OpenAI describes here.

Both independent assessments find that the Lean statements express or imply the whole result described above. Read the assessments

Four problems have been reported by others; the atlas has not judged them.

What the papers state

Corollary 1.3 (Singular ranks and full-deck mixing)

In Compatibility entropy and the spectrum of a Thorp sweep

Possible match to OpenAI’s description, selected from shared wording. No review confirming the selection is recorded.

For every nontrivial \lambda\vdash n and 1\le r\le D_\lambda,

s_r(T_{n,\lambda})\le(16D_\lambda r)^{-1/(2p_*)}.

If s_j(T_n) denotes the full regular singular list, including all multiplicities and zeros, then

s_j(T_n)\le j^{-1/(2p_*)}\qquad(1\le j\le n!).

The constant singular value is the first value. The physical Thorp chain has worst-start 1/4 mixing time \Theta(d); more precisely its lower bound is 2d-O(1). For every fixed integer M>p_*, the worst-start total-variation distance after Md physical shuffles tends to zero as d\to\infty.

Theorem 1.1 (Conditional permutation and full-deck mixing)

In Conditional permutations in a revealed switching environment

Possible match to OpenAI’s description, selected from shared wording. No review confirming the selection is recorded.

Let n=2^d, and start the shuffle from any deterministic deck. After J(s_0+1) sweeps, where s_0=400 and J=40000, the conditional free-slot law defined above satisfies

\sup_{\text{initial layouts}}\E_{\mathcal E} \|\nu_{\mathcal E}-U_{n/2}\|_{\TV}\longrightarrow0 \qquad(d\longrightarrow\infty).

Consequently the full-deck law approaches uniform, uniformly in the starting deck, after 16\,040\,400d physical shuffles. For every integer t\ge0, after t physical shuffles the full-deck distance from uniform is at least 1-2^{tn/2}/n!. Thus the full-deck mixing time is of order d.

No Lean statement in the release is paired with this theorem.

The release does not say which paper theorem its catalogue describes. The statements above share the most wording with the description; the atlas has not reviewed that choice. Identifying it is listed under Questions awaiting review.

16 more possible matches

  • Theorem 1.1 in Random coordinate frames and partial permutation laws.
  • Theorem 1.1 (Regular-trace smoothing) in Random-subspace tests and trace smoothing for coordinate sweeps.
  • Theorem 1.2 (A fixed moment with a degree saving) in A strict four-row permanent inequality and permutation moments.
  • Theorem 1.2 (A bounded regular moment) in Compatibility entropy and the spectrum of a Thorp sweep.
  • Theorem 1.1 (Binary sweep contraction) in Conditional coordinate sweeps and analytic transfer.
  • Theorem 1.1 (Fixed lists and a full-deck consequence) in Conditional information under deterministic coordinate sweeps.
  • Theorem 1.1 (Routing density and representation contraction) in Routing densities and representation contraction for Thorp sweeps.
  • Corollary 1.2 (Optimal order of mixing) in Row–column symmetry and contraction of coordinate sweeps.
  • Corollary 1.2 (Fixed-sweep convergence of the full deck) in Signed tensor densities and diagram budgets for the Thorp shuffle.
  • Theorem 1.1 (Complementary cosets control Fourier matrices) in From partial permutation information to Fourier bounds.
  • Theorem 1.1 (Optimal-order mixing) in Optimal-order mixing of the Thorp shuffle.
  • Theorem 1.1 (A robust four-row inequality) in A strict four-row permanent inequality and permutation moments.
  • Theorem 1.1 (Weighted compatibility) in Compatibility entropy and the spectrum of a Thorp sweep.
  • Proposition 1.2 (Support obstruction) in Conditional coordinate sweeps and analytic transfer.
  • Theorem 1.1 (Weighted sweep moments) in Row–column symmetry and contraction of coordinate sweeps.
  • Theorem 1.1 (A moment bound for every signed occurrence) in Signed tensor densities and diagram budgets for the Thorp shuffle.
Reported proof verification
13 Lean statements in the release; independent groups report that Comparator accepted runs for all these files at the version shown here.

Papers

Each paper opens as a reading page in the browser, with the PDF beside it. The statements each paper opens with are on its reading page.

Optimal-order mixing of the Thorp shuffle

Undated, current edition.

Read it herePDF

We prove that the Thorp shuffle on \(2^d\) cards mixes in \(\Theta(d)\) complete shuffles. The full permutation law after \(1600d\) shuffles converges to uniform in total variation as \(d\to\infty\), uniformly over the initial deck, while a support count gives a lower bound of \(2d-O(1)\).

Random coordinate frames and partial permutation laws

Undated, current edition.

Read it herePDF

For the Thorp shuffle on \(2^d\) cards, we prove that the worst-start total-variation distance after \(32800d\) physical shuffles tends to zero as \(d\to\infty\). Combining our fixed-list estimate with the companion Fourier transfer improves this bound to \(512d\) shuffles. These results follow from bounds on partially observed permutation laws in random coordinate frames.

From partial permutation information to Fourier bounds

Undated, current edition.

Read it herePDF

We show how information about partial permutations controls full permutation laws. Let n tend to infinity through multiples of eight. If the images of a uniformly chosen \(7n/8\) labels approach the uniform injection law in average total variation, and the sign mean tends to zero, then the product of two independent permutations with the given law converges to uniform on Sn.

Conditional information under deterministic coordinate sweeps

Undated, current edition.

Read it herePDF

We bound the information remaining after paths of specified cards in the Thorp shuffle on \(n=2^d\) cards have been observed. After a fixed number of deterministic coordinate sweeps, the joint endpoint law of further cards is close to uniform on the available positions, on average over the observed paths, provided a fixed positive fraction of labels lies outside both lists. Combined with a Fourier transfer, these bounds give full-deck mixing after \(2048d\) physical shuffles.

Conditional permutations in a revealed switching environment

Undated, current edition.

Read it herePDF

Fix half the labels in a Thorp shuffle on \(2^d\) positions and reveal their complete trajectories. We prove that, after an explicit absolute number of coordinate sweeps, the conditional permutation of the remaining labels approaches uniform in expected total variation as \(d\to\infty\), uniformly in the initial layout. The resulting constructions give full-deck mixing in a constant multiple of d physical shuffles.

Random-subspace tests and trace smoothing for coordinate sweeps

Undated, current edition.

Read it herePDF

We prove that an absolute number of coordinate sweeps of the Thorp shuffle on \(n=2^d\) positions brings the full permutation to total-variation distance at most \(\frac12n^{-5}\) from uniform, uniformly over the initial deck. Thus \(O(d)\) physical shuffles suffice, which is optimal in order.

Conditional coordinate sweeps and analytic transfer

Undated, current edition.

Read it herePDF

For coordinate sweeps with near-uniform line laws and line sizes among powers of two in a suitable fixed interval, we prove a Schatten-moment bound after conditioning on any feasible collection of prescribed card trajectories. An analytic transfer to binary sweeps then shows that a fixed number of sweeps mixes the Thorp shuffle on \(2^d\) cards in \(O(d)\) physical steps, matching the order of the support lower bound.

A strict four-row permanent inequality and permutation moments

Undated, current edition.

Read it herePDF

We prove a permanent inequality with exponent strictly below two for laws on S4 sufficiently close to uniform and with exactly uniform coordinate marginals. Applied to a four-row recursion, it shows that an absolute number of coordinate sweeps brings the full permutation of the Thorp shuffle on \(N=2^d\) cards to total-variation distance tending to zero from uniform as \(N\to\infty\). Thus the mixing time has the optimal order \(O(\log N)\) in physical shuffles.

13 Lean statements in this family

Choose one to see it beside the paper theorem it is paired with.

For a configured theorem, the Comparator program tests whether OpenAI’s separate proof proves the exact Lean statement using only the allowed axioms. Some statement files instead configure definitions for comparison by name and type. Acceptance does not establish that a formal statement means the same thing as the paper’s theorem.

BinarySweep

The paper states

Theorem 1.1 (Binary sweep contraction)

In Conditional coordinate sweeps and analytic transfer

There are absolute constants g>0 and d_0 such that, for d\ge d_0 and every \lambda\vdash 2^d,

\|K_\lambda\|_{\op}\le D_\lambda^{-g}.

The sign representation is annihilated. There is an absolute positive integer w such that the law after w independent sweeps converges to uniform in total variation as d\to\infty, uniformly over deterministic initial decks.

The words they share: binary, contraction, initial, sign, representation, decks, independent, deterministic, uniformly, total, variation, law, sweep, integer, uniform, absolute, coordinate.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Binary sweep contraction and mixing

BinarySweep.lean, the release lists it as a main result.

theorem binary_sweep_contraction_and_mixing :    BinaryCoordinateSweeps.BinaryContractionTarget ∧    (∀ d : ℕ, 0 < d → ∑ g : Equiv.Perm (BinaryCoordinateSweeps.Slot d),      BinaryCoordinateSweeps.binaryLaw d g * BinaryCoordinateSweeps.realSign g = 0) ∧    BinaryCoordinateSweeps.UniformSweepMixingTarget := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

two independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.

Read the complete Lean statement

CoordinateSweeps

The paper states

Theorem 3.1 (Conditional sweep moment estimate)

In Conditional coordinate sweeps and analytic transfer

There are a power of two R\ge2, a positive integer q, and z_*>0 such that every grid and every family \mathcal H of prescribed disjoint trajectories defined above satisfy, for every z\in[0,z_*] and every \lambda\vdash s-h,

\log T_\lambda^{\mathcal H}\ \le\ -c(s)F(\lambda)+e(s)h\log s-C(\mathcal H).

Here a zero moment gives -\infty. Use parameters

a=\frac1{100},\quad c_0=e_0=\frac a{100},\qquad c(s)=c_0+\frac1{\sqrt{\log s}},\qquad e(s)=e_0-\frac1{\sqrt{\log s}}.

The empty diagram has dimension one. If h=s, the remaining bijection is the unique empty bijection and its moment is one. The assertion also includes the one-dimensional trivial representation for every h.

The Lean source’s comment reads: /-- Manuscript03 thm:conditional / ac:main, with c₀=e₀=a/100, a=1/100, natural logarithms, unnormalized integer Schatten moments, all feasible disjoint paths (including none and all), and log(0)=−∞. -/

The Lean statement whose source cites this theorem

The Lean source cites this theorem. No review of whether the statements mean the same thing is recorded.

Conditional coordinate-sweep moment estimate

CoordinateSweeps.lean, the release lists it as a main result.

def ConditionalMain : Prop :=  ∃ (r q : ℕ) (zStar : ℝ), 1 ≤ r ∧ 1 ≤ q ∧ 0 < zStar ∧ zStar ≤ 1 / ((2 : ℕ) : ℝ) ∧    ∀ (G : Grid), G.Allowed r → ∀ (h : ℕ) (H : G.Holes h) (hf : H.Feasible),    ∀ z ∈ Set.Icc 0 zStar, ∀ ρ : UnitaryIrrep H.stabilizer,      logMoment (schattenMoment q (H.conditionalAverage hf z ρ)) ≤        ((-c G.size * Real.log ρ.dimension + e G.size * h * Real.log G.size - H.cost : ℝ) : EReal)theorem conditional_main : ConditionalMain := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

The file states one more theorem: all_grid_moment.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

CoordinateTrace

The paper states

Theorem 1.1 (Regular-trace smoothing)

In Random-subspace tests and trace smoothing for coordinate sweeps

There is an absolute P\ge2 such that for every dyadic n\ge2,

\operatorname{Tr}_{\rm reg}|Q_n|^P\le1+n^{-10}.

If v is an integer with 2v\ge P, then v independent sweeps have total-variation distance at most \frac12n^{-5} from uniform, from every deterministic initial deck. Here \|\mu-\nu\|_{\rm TV}=\tfrac12\sum_g|\mu(g)-\nu(g)|.

The words they share: smoothing, trace, regular, 1/2, distance, sweeps, integer, absolute, initial, deck, total, variation, uniform.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Regular trace smoothing and full-permutation mixing

CoordinateTrace.lean, the release lists it as a supporting statement.

def MainStatement : Prop :=  ∃ P : ℝ, 2 ≤ P ∧ ∀ d : ℕ, 1 ≤ d →    regularTrace d P ≤ 1 + ((2 : ℝ)^d) ^ (-10 : ℤ) ∧    ∀ v : ℤ, P ≤ 2 * (v : ℝ) → ∀ σ : Equiv.Perm (Card d),      sweepDistance d v.toNat σ ≤ (1/2 : ℝ) * ((2 : ℝ)^d) ^ (-5 : ℤ)theorem regular_trace_smoothing : MainStatement := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

FourRowPermanent

The paper states

Theorem 1.1 (A robust four-row inequality)

In A strict four-row permanent inequality and permutation moments

There are absolute constants p_0\in(4/3,2) and \varepsilon>0 with the following property. Let \nu be a probability measure on S_4 such that

\|\nu-u\|_{\TV}<\varepsilon, \qquad \nu\{\pi:\pi(i)=j\}=\frac14\quad(1\le i,j\le4).

Then every four nonnegative functions on \{1,2,3,4\} satisfy

\E_\nu\prod_{i=1}^4f_i(\pi(i))\le\prod_{i=1}^4\|f_i\|_{p_0}.

The words they share: 4/3, robust, functions, pi, 1/4, probability, permanent, inequality, four, nonnegative, varepsilon, row, constants.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Strict four-row permanent inequality near the uniform law

FourRowPermanent.lean, the release lists it as a supporting statement.

theorem robust_permanent :  ∃ p₀ : ℝ, 4 / 3 < p₀ ∧ p₀ < 2 ∧    ∃ ε : ℝ, 0 < ε ∧ ∀ ν : Law,      IsProbability ν → totalVariation ν < ε → UniformMarginals ν →      ∀ f : Functions, (∀ i j, 0 ≤ f i j) →        permanentExpectation ν f ≤ ∏ i, lpNorm p₀ (f i) := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

OccupiedOverlap

The paper states

Theorem 5.1 (Overlap on an arbitrary occupied board)

In Row–column symmetry and contraction of coordinate sweeps

Take a grid with m rows and n columns, mn=N, and consider any subset of s=N-t cells, with the group G_s on those cells. Let a\vdash s lie in the h,h-hook, 1\le h\le N. Decompose V_a under the within-row permutation subgroup of G_s, and also (separately) the within-column subgroup. For types \sigma,\theta of these two groups, let I_{\sigma,u}: V_\sigma\to V_a, I_{\theta,v}: V_\theta\to V_a be isometric embeddings giving orthogonal copies and spanning the respective isotypic spaces. Then

\sum_{u,v}\|I_{\sigma,u}^* I_{\theta,v}\|_\infty^2 \ \le\ \exp\{C[(m+n+1)h^2\log(N+2)+t]\}\ \min\left\{1,\frac{D_\sigma D_\theta}{D_a}\right\}.

Here \sigma and \theta can each consist of a tuple of partitions, and the infinity norm is the operator norm. For s=0 the empty representation has dimension one.

The Lean source’s comment reads: /-- thm:occupied (weighted.tex:76–84), including the empty occupied subset, arbitrary deletion layout and every orthogonal multiplicity copy. The absolute constant is 128; dimensions, unnormalized trace and operator norm are literal. -/

The Lean statement whose source cites this theorem

The Lean source cites this theorem. No review of whether the statements mean the same thing is recorded.

Row–column overlap on occupied boards

OccupiedOverlap.lean, the release lists it as a supporting statement.

abbrev OccupiedOverlapEndpoint : Prop :=  ∃ C : ℝ, 0 < C ∧ ∀ (m n : ℕ) (Ω : Finset (Board m n))    (a : YoungDiagram) (e : Ω ≃ Cell a) (h : ℕ),    1 ≤ h → h ≤ m * n → InHook a h →    ∀ (r c u v : ℕ)      (ρ : Representation ℂ (rowGroup Ω) (EuclideanSpace ℂ (Fin r)))      (τ : Representation ℂ (columnGroup Ω) (EuclideanSpace ℂ (Fin c))),      Representation.IsIrreducible ρ → Representation.IsIrreducible τ →      IsUnitary ρ → IsUnitary τ →    ∀ (I : Fin u → EuclideanSpace ℂ (Fin r) →ₗᵢ[ℂ] hilbertSpace a)      (J : Fin v → EuclideanSpace ℂ (Fin c) →ₗᵢ[ℂ] hilbertSpace a),      CompleteCopies (V := EuclideanSpace ℂ (Fin r)) (W := hilbertSpace a) ρ ((relabelledUnitary a e).comp (rowGroup Ω).subtype) I →      CompleteCopies (V := EuclideanSpace ℂ (Fin c)) (W := hilbertSpace a) τ ((relabelledUnitary a e).comp (columnGroup Ω).subtype) J →      allCopyOverlap (V := EuclideanSpace ℂ (Fin r)) (W := EuclideanSpace ℂ (Fin c))        (X := hilbertSpace a) I J ≤        Real.exp (C * (((m + n + 1 : ℕ) : ℝ) * (h : ℝ)^2 *

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

The file states one more theorem: occupied_overlap.

Five definitions are compared by name and type only: IsUnitary, hilbertSpace, unitaryRepresentation, relabelledUnitary, OccupiedOverlapEndpoint.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

PartialPermutation

The paper states

No paper theorem is paired with this Lean statement.

no paper statement matches the Lean text clearly enough to pair, so the closest candidates are listed below.

Statements it shares words with:

  • Theorem 13.1 (Entropy clipping and isotypic transfer) in From partial permutation information to Fourier bounds.
  • Theorem 1.1 (Complementary cosets control Fourier matrices) in From partial permutation information to Fourier bounds.
  • Theorem 8.5 (Two-sided Fourier multiplication) in From partial permutation information to Fourier bounds.
  • Theorem 7.3 (Character-projection transfer) in From partial permutation information to Fourier bounds.
  • Corollary 6.2 (From bounded coset densities to Fourier decay) in Random coordinate frames and partial permutation laws.
  • Proposition 4.6 (Branching refinement) in From partial permutation information to Fourier bounds.
  • Corollary 8.6 (Two-factor completion) in From partial permutation information to Fourier bounds.
  • Proposition 2.5 (Fourier amplification) in From partial permutation information to Fourier bounds.

The Lean statement

No paper theorem is paired with it.

Fourier bounds from complementary coset caps

PartialPermutation.lean, the release lists it as a supporting statement.

def ComplementaryCosetsFiniteStatement : Prop :=  ∀ (X : Type*) [Fintype X] [DecidableEq X] (b : ℕ) (_ : 0 < b) (M : Fin b → Set X),    (∀ i, (M i).Nonempty) →    Pairwise (fun i j => Disjoint (M i) (M j)) → (⋃ i, M i) = Set.univ →    ∀ (V : Type) [NormedAddCommGroup V] [InnerProductSpace ℂ V]      [FiniteDimensional ℂ V] (ρ : Representation ℂ (Equiv.Perm X) V),      Representation.IsIrreducible ρ → IsUnitary ρ →      ∀ (f : Equiv.Perm X → ℝ) (B u : ℝ),      (∀ g, 0 ≤ f g) → mass f ≤ 1 → 0 < u →      (∀ i, LeftCosetCap f (blockSubgroup (M i)) B) →      (hsNormSq (complexFourier ρ (fun g => f g)) ≤        b * B * symmetricDegreeConstant u *          (Module.finrank ℂ V : ℝ)^(-1+(u+2)/(b : ℝ))) ∧      (‖LinearMap.toContinuousLinearMap (complexFourier ρ (fun g => f g))‖^2 ≤        b * B * symmetricDegreeConstant u *          (Module.finrank ℂ V : ℝ)^(-1+(u+2)/(b : ℝ)))

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

two independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.

Read the complete Lean statement

SignedSweepMoment

The paper states

Theorem 1.1 (A moment bound for every signed occurrence)

In Signed tensor densities and diagram budgets for the Thorp shuffle

For each sufficiently small fixed \bar\eta>0, choose a sufficiently small \bar\kappa>0 with \bar\kappa<\bar\eta/256. There is an integer r\ge1, depending only on these fixed constants, such that for every d\ge1, every \lambda\vdash n=2^d, and every occurrence (1),

\log\!\left(D_\lambda\operatorname{Tr}_{V_\lambda}(B_n^*B_n)^r\right) \le\eta_d F(\alpha,\beta)+g_n(l),

where

\eta_d=\bar\eta\left(1-\frac1{2\sqrt d}\right),\qquad \kappa_d=\bar\kappa\left(1-\frac1{2\sqrt d}\right),\qquad g_n(l)=\max\{0,\kappa_d l\log n-l\log(n/l)\}.

Set g_n(0)=0 and interpret the logarithm of a zero trace as -\infty.

The words they share: occurrence, eq, trace, max, moment, signed, sqrt, zero, log, bound, integer, budgets, diagram, tensor, small, logarithm.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Uniform signed-occurrence moment bound for coordinate sweeps

SignedSweepMoment.lean, the release lists it as a supporting statement.

theorem signed_occurrence_moment_bound :    ∃ η₀ : ℝ, 0 < η₀ ∧      ∀ η : ℝ, 0 < η → η < η₀ →        ∃ κ₀ : ℝ, 0 < κ₀ ∧ κ₀ < η / 256 ∧          ∀ κ : ℝ, 0 < κ → κ < κ₀ →            ∃ r : ℕ, 1 ≤ r ∧              ∀ d : ℕ, 1 ≤ d →                ∀ lam : Partition (2 ^ d),                  ∀ (u v l : ℕ) (h : u + v + l = 2 ^ d)                    (α : Partition u) (β : Partition v) (γ : Partition l),                    SignedOccurrence h α β γ lam →                      logMoment (weightedMoment lam r) ≤                        ((coefficient η d * signedEntropy α β +                            remainderBudget κ d l : ℝ) : EReal) := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

two independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.

Read the complete Lean statement

SpinAngle

The paper states

Corollary 1.3 (Singular ranks and full-deck mixing)

In Compatibility entropy and the spectrum of a Thorp sweep

For every nontrivial \lambda\vdash n and 1\le r\le D_\lambda,

s_r(T_{n,\lambda})\le(16D_\lambda r)^{-1/(2p_*)}.

If s_j(T_n) denotes the full regular singular list, including all multiplicities and zeros, then

s_j(T_n)\le j^{-1/(2p_*)}\qquad(1\le j\le n!).

The constant singular value is the first value. The physical Thorp chain has worst-start 1/4 mixing time \Theta(d); more precisely its lower bound is 2d-O(1). For every fixed integer M>p_*, the worst-start total-variation distance after Md physical shuffles tends to zero as d\to\infty.

The words they share: after, bound, constant, deck, distance, fixed, full, including, list, lower, mixing, physical, regular, shuffles, tends, thorp, time, total, variation, zero.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

One-sided angle bound for signed spin types

SpinAngle.lean, the release lists it as a supporting statement.

theorem one_sided_type_angle (W : Finset (R × C))    (D : TypeLabel q) (H : C → TypeLabel q) (K : R → TypeLabel q)    (hD : D.size = W.card)    (hH : ∀ j, (H j).size = Fintype.card (TypeBlocks.Fiber (column W) j))    (hK : ∀ i, (K i).size = Fintype.card (TypeBlocks.Fiber (row W) i)) :    (‖TypeBlocks.projection (column W) H * TypeBlocks.projection (row W) K * D.projection W‖ ^ 2 ≤      min 1 (Real.exp (TypeBlocks.entropy H + TypeBlocks.entropy K - D.entropy) *        (TypeBlocks.dimension H : ℝ) * (TypeBlocks.dimension K : ℝ) * phi W)) ∧    (min 1 (Real.exp (TypeBlocks.entropy H + TypeBlocks.entropy K - D.entropy) *        (TypeBlocks.dimension H : ℝ) * (TypeBlocks.dimension K : ℝ) * phi W) ≤      min 1 (Real.exp (TypeBlocks.entropy H + TypeBlocks.entropy K - D.entropy +        (Fintype.card R * Fintype.card C - W.card : ℕ)) *        (TypeBlocks.dimension H : ℝ) * (TypeBlocks.dimension K : ℝ))) ∧    TypeBlocks.dimension H * TypeBlocks.dimension K ≤      (W.card + 1) ^ ((Fintype.card R + Fintype.card C) * q * (q - 1)) := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

Three definitions are compared by name and type only: row, column, phi.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

ThorpFirstReciprocal

The paper states

No paper theorem is paired with this Lean statement.

no paper statement matches the Lean text clearly enough to pair, so the closest candidates are listed below.

Statements it shares words with:

  • Theorem 1.1 (Complementary cosets control Fourier matrices) in From partial permutation information to Fourier bounds.
  • Theorem 9.2 (Three-group two-sided truncation) in From partial permutation information to Fourier bounds.
  • Proposition 6.1 (Eight-block Fourier bound) in Optimal-order mixing of the Thorp shuffle.
  • Proposition C.1 (Coefficient-space average) in From partial permutation information to Fourier bounds.
  • Corollary 6.2 (From bounded coset densities to Fourier decay) in Random coordinate frames and partial permutation laws.
  • Theorem 5.1 (Overlap on an arbitrary occupied board) in Row–column symmetry and contraction of coordinate sweeps.
  • Proposition 10.2 (Coherent overlap on surviving cells) in Row–column symmetry and contraction of coordinate sweeps.
  • Theorem 13.1 (Entropy clipping and isotypic transfer) in From partial permutation information to Fourier bounds.

The Lean statement

No paper theorem is paired with it.

Reciprocal Specht degrees and eight-block Fourier bound

ThorpFirstReciprocal.lean, the release lists it as a supporting statement.

theorem first_reciprocal_main :    (BddAbove (Set.range (fun n : {n : ℕ // 1 ≤ n} => reciprocalFirst n.val)) ∧    (∀ n : ℕ, 1 ≤ n → reciprocalFirst n ≤ C1) ∧    Tendsto exceptionalFirst atTop (nhds 0)) ∧    (∀ (d : ℕ) (_hd : 3 ≤ d)    (e : Position d ≃ Fin 8 × Fin (2 ^ (d - 3)))    {E : Type uE} [NormedAddCommGroup E] [InnerProductSpace ℂ E]    [FiniteDimensional ℂ E]    (ρ : Representation ℂ (Equiv.Perm (Position d)) E)    [Representation.IsIrreducible ρ]    (_ : ∀ g x y, ⟪ρ g x, ρ g y⟫_ℂ = ⟪x, y⟫_ℂ)    (ν : Equiv.Perm (Position d) → ℝ)    (_ : ∀ g, 0 ≤ ν g) (_ : ∑ g, ν g = 1)    (_ : ∀ (i : Fin 8) (g : Equiv.Perm (Position d)),      ∑ h : Equiv.Perm (Fin (2 ^ (d - 3))),          ν (g * Block.embedding e (Pi.mulSingle i h)) ≤

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

ThorpRemaining

The paper states

Theorem 1.1 (Optimal-order mixing)

In Optimal-order mixing of the Thorp shuffle

As d\to\infty,

\|q_d^{*(1600d)}-U_n\|_{\mathrm{TV}}\longrightarrow0.

For every integer t\ge0,

\|q_d^{*t}-U_n\|_{\mathrm{TV}} \ge 1-\frac{2^{tn/2}}{n!}.

Consequently the mixing time at distance 1/4 satisfies

t_{\mathrm{mix}}(d)\ge \left\lceil\frac2n\log_2\frac{3n!}{4}\right\rceil =2d-O(1), \qquad t_{\mathrm{mix}}(d)=\Theta(d).

The words they share: optimal, theta, mix, 1/4, time, distance, order, tv, log, integer, thorp.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Frame, spectral, and density estimates for Thorp mixing

ThorpRemaining.lean, the release lists it as a supporting statement.

theorem remaining_main :    FrameMain ∧ InformationMain ∧ SpectrumMain ∧ SignedMain ∧      DegreeSavingMain ∧ FullDensityMain ∧ ForwardMixingMain ∧ OptimalOrderMain := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

two independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.

Read the complete Lean statement

ThorpRouting

The paper states

Proposition 5.1 (High-height cycle tail)

In Routing densities and representation contraction for Thorp sweeps

There are absolute b,\delta>0 and an integer J_0\ge1 such that, uniformly in the tracked input injection of size r and for every integer J\ge J_0,

\begin{aligned}\log_2\E 2^{b\sum_{j\ge J}C_j}&\le O(r2^{-bJ}),\\ \log_2\E 2^{b\sum_jC_j}&\le O(r(r/n)^\delta).\end{aligned}

The first bound also holds with expectation conditional on all input-side switches and all output switches of heights at most s, for every integer 0\le s\le d with 2s<J, uniformly in the fixed switch values and with the same absolute implied constant.

The words they share: high, heights, tail, cycle, switch, delta, log, bound, integer, conditional.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Nine routing-density and representation estimates

ThorpRouting.lean, the release lists it as a main result.

theorem ThorpNine.main :    ThorpNine.Adaptive.Thorp.AdaptiveBounds.MainStatement ∧    ThorpNine.Casimir.Thorp.CasimirBounds.MainStatement ∧    ThorpNine.Dense.Thorp.DenseTruncation.MainStatement ∧    ThorpNine.Contact.Thorp.SparseContact.MainStatement ∧    ThorpNine.Harmonic.Thorp.HarmonicBounds.MainStatement ∧    ThorpNine.Smoothing.Thorp.StrongSmoothing.MainStatement ∧    ThorpNine.Tail.Thorp.StrongTail.MainStatement ∧    ThorpNine.HighTail.Thorp.HighHeight.MainStatement ∧    ThorpNine.SparseSaving.Thorp.SparseRegime.MainStatement := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

One independent group reports that Comparator accepted a run for this Lean statement file: cert-machine. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.

Read the complete Lean statement

ThorpWeightedCompatibility

The paper states

Theorem 1.1 (Weighted sweep moments)

In Row–column symmetry and contraction of coordinate sweeps

There are positive weights W_\lambda, an absolute finite p_*, and real exponents 2\le p_d\le p_* such that, for every d\ge1 and every \lambda\vdash2^d,

D_\lambda^{3/4}\le W_\lambda\le D_\lambda, \qquad W_\lambda\|K_d(\lambda)\|_{p_d}^{p_d}\le1.

Consequently \|K_d(\lambda)\|_{\rm op}\le D_\lambda^{-3/(4p_*)}.

The words they share: weighted, finite, real, row, moments, absolute.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Weighted row–column compatibility bound

ThorpWeightedCompatibility.lean, the release lists it as a main result.

def WeightedCompatibilityTheorem : Prop :=  ∃ L C₀ m₀ : ℝ, 0 < L ∧ 0 < C₀ ∧ 0 < m₀ ∧    ∀ A D : ℕ,    let n : ℝ := A * D    let m := Real.sqrt n    let θ := 1 - L / Real.log m    m₀ ≤ m → 0 < θ → m / ((2 : ℕ) : ℝ) ≤ A → (A : ℝ) ≤ ((2 : ℕ) : ℝ) * m →    m / ((2 : ℕ) : ℝ) ≤ D → (D : ℝ) ≤ ((2 : ℕ) : ℝ) * m →    ∀ (w : Fin A → Equiv.Perm (Fin D) → ℝ)      (v : Fin D → Equiv.Perm (Fin A) → ℝ),    (∀ i σ, 0 ≤ w i σ) → (∀ k σ, 0 ≤ v k σ) →    weightedCompatibility A D w v ≤      Real.exp (C₀ * n ^ (((54 : ℕ) : ℝ) / ((100 : ℕ) : ℝ))) *        (∏ i, marginalFactor θ (w i)) * (∏ k, marginalFactor θ (v k))theorem weightedCompatibility_main : WeightedCompatibilityTheorem := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

WeightedSweepMoments

The paper states

Theorem 1.1 (Weighted sweep moments)

In Row–column symmetry and contraction of coordinate sweeps

There are positive weights W_\lambda, an absolute finite p_*, and real exponents 2\le p_d\le p_* such that, for every d\ge1 and every \lambda\vdash2^d,

D_\lambda^{3/4}\le W_\lambda\le D_\lambda, \qquad W_\lambda\|K_d(\lambda)\|_{p_d}^{p_d}\le1.

Consequently \|K_d(\lambda)\|_{\rm op}\le D_\lambda^{-3/(4p_*)}.

The words they share: 3/4, weighted, finite, real, moments, sweep, row, exponents, consequently, absolute.

Possible Lean counterpart

Selected from shared wording. No review of whether the statements mean the same thing is recorded.

Weighted Schatten moments and operator contraction for coordinate sweeps

WeightedSweepMoments.lean, the release lists it as a supporting statement.

theorem weighted_sweep_moments :    ∃ (W : YoungDiagram → ℝ) (pStar : ℝ) (p : ℕ → ℝ),      (∀ μ : YoungDiagram, 0 < W μ) ∧      2 ≤ pStar ∧      (∀ d : ℕ, 1 ≤ d → 2 ≤ p d ∧ p d ≤ pStar) ∧      ∀ d : ℕ, 1 ≤ d → ∀ μ : Shapes (2 ^ d),        dimension μ.1 ^ (3 / 4 : ℝ) ≤ W μ.1 ∧        W μ.1 ≤ dimension μ.1 ∧        W μ.1 *          schattenNorm (V := hilbertSpace μ.1) (fourierSweep d μ) (p d) ^ (p d) ≤ 1 ∧        ‖fourierSweep d μ‖ ≤ dimension μ.1 ^ (-3 / (4 * pStar)) := by  sorry

This file states the theorem with sorry, a placeholder, in place of its proof. OpenAI supplies a separate proof file. Comparator tests whether that proof proves the statement here.

three independent groups report that Comparator accepted runs for this Lean statement file: cert-machine; MathVet; Neal McBurnett, openai-math-lean-tooling. Who ran the proof-verification program

  • cert-machine records it as “CERTIFIED”.
  • MathVet records it as “comparator-sandboxed-pass”.
  • Neal McBurnett, openai-math-lean-tooling records it as “accepted”.

Read the complete Lean statement

Other mathematical statements in OpenAI’s Lean library

Atlas found these propositions by scanning the text of the project’s Model.lean file. Each name links to its definition, with any associated statement file beside it. This list does not record running Lean to interpret or verify those definitions.

Independent comparisons and proof-verification reports

Does the Lean express the advertised result?

MathVet and cert-machine agree on how much of the advertised result the Lean statements express.

MathVet and Carlos Toledo’s cert-machine independently assessed whether this family’s Lean statements express the whole result OpenAI describes in its catalogue. These are assessments of the family as a whole, not of each paper theorem separately.

MathVet finds that the Lean statements express or imply the whole advertised result (their term: “full”).

Their note: ThorpResults.remaining_main (ThorpRemaining) is a conjunction whose last conjunct OptimalOrderMain := IsTheta atTop (mixingTime d) (log (2^d)) states exactly the Θ(log N) mixing for N = 2^d cards, with mixingTime d := sInf {t | distance d t ≤ 1/4}, distance d t := tv (law d t) (uniform d) the total-variation distance of the full permutation law (and InformationMain records that lawFrom from any starting deck has the same distance, plus distance d (2048 d) → 0); the 12 other challenge files are supporting Fourier/representation estimates (regular_trace_smoothing, binary_sweep_contraction_and_mixing, robust_permanent,...). All of them state estimates for the same hand-built Thorp shuffle.

Definitions the group asks readers to compare with the paper: Custom Thorp shuffle and ‘mixing time’: step (d+1) c := (pairSwitch d c on the head bit, then rotate the coordinate order) on State d := Equiv.Perm (Fin d → Bool) with Coins (d+1) = (Fin d → Bool) → Bool iid fair coins, fairMass, tv, mixingTime d := sInf {t | distance d t ≤ 1/4}; this is the standard pair-swap-and-interleave Thorp shuffle and TV-from-uniform with ε = 1/4, but it is a hand-written model and the Θ-statement hides explicit constants (1600d, 2048d,...). OccupiedOverlap shows an axiom_files hit (HookModel.lean) that is a doc-comment false positive.

“full”, in their words: The challenge statements imply the headline claim after unpacking definitions or one routine step.

Their method: An independent, model-assisted, pre-referee reading of each family’s challenge statements against its headline claim under a published rubric (PROTOCOL.md), with its own Comparator runs per challenge.

cert-machine finds that the Lean statements express or imply the whole advertised result (their term: “MATCHES”).

“MATCHES”, in their words: The challenge statements state the headline as OpenAI wrote it.

Their method: Every Comparator challenge re-run on Linux with nanoda switched on, decided by a preregistered rule.

OpenAI’s own note on what its Lean covers

The formalization proves optimal-order mixing of the Thorp shuffle on $2^d$ cards. After $1600d$ complete shuffles, the full permutation law converges in total variation to uniform as $d\to\infty$, uniformly over initial decks. Together with the support lower bound of $2d-O(1)$, this gives mixing time $\Theta(d)=\Theta(\log(2^d))$.

The linked supporting results include frame and conditional-list estimates, regular trace and spectral bounds, signed moments, and full-density $L^2$ control. They also bound reciprocal Specht-dimension sums and give the eight-block Fourier estimate used in the mixing argument. The formalized supporting result turns bounds on partial-permutation cosets into Fourier bounds. Partition a finite set into $b\ge1$ nonempty blocks, and let $f\ge0$ be a subprobability weight on its symmetric group whose left-coset masses for each block subgroup are at most $B$. For every irreducible unitary representation of dimension $D$ and every $u>0$, both the squared Hilbert–Schmidt norm and squared operator norm of its Fourier transform are at most $bB\,C(u)D^{-1+(u+2)/b}$, where $C(u)$ is the stated symmetric-group degree constant. The paper’s asymptotic conclusion about the product of two random permutations is outside this selected finite estimate. The formalized results give averaged conditional mixing for half-permutations in the revealed-path model and full-deck mixing for the Thorp shuffle. In particular, the full-deck total-variation distance tends to zero after $16040400d$ steps on $2^d$ cards, uniformly over deterministic initial decks. The conditional estimate is averaged over the actual outside-path law, rather than asserted for each individual environment. The formalization also includes the associated overlay, reset minorization, and two-color conditional constructions. The formalization covers nine routing and representation estimates for Thorp sweeps: adaptive operator and fourth-trace bounds, Casimir moments, dense truncation, sparse contact, harmonic-cycle contraction, smoothing, and three tail or sparse-saving estimates. They bound routing-density deviations and Fourier operators in dense and sparse regimes, including exponential savings in the level scale and representation dimension. The conditional high-height estimate requires a sufficiently large cutoff height $J$ that exceeds twice the number $s$ of coordinates in the fixed low block, so $J>2s$. These are the earlier nine statements associated with the paper; its changed later statements and six other result blocks are outside this scope. The formalization proves uniform representation bounds for one coordinate sweep of the Thorp shuffle. For each irreducible representation of dimension $D$, it chooses a weight between $D^{3/4}$ and $D$ and a Schatten exponent bounded by one absolute constant so that the weighted Schatten moment is at most one. The sweep’s operator norm is consequently at most $D^{-c}$ for one absolute $c>0$. The supporting row–column estimate bounds the total squared overlap over all multiplicity copies on an occupied board, with explicit dependence on the row and column representation dimensions, the global dimension, the number of missing cells, and the number of rows or columns containing the global Young diagram. The paper’s full physical-mixing conclusion is outside these selected representation estimates. The formalization proves a uniform trace-smoothing estimate for coordinate sweeps of the Thorp shuffle on $2^d$ cards. One absolute sweep parameter works for every $d\ge1$: the regular trace is at most $1+(2^d)^{-10}$, and the full permutation law after the corresponding fixed number of sweeps is within $\tfrac12(2^d)^{-5}$ of uniform in total variation, for every initial deck. Thus the selected upper bound uses $O(d)$ physical shuffles. The formalized result is the weighted compatibility theorem for balanced $A\times D$ rectangles, with $n=AD$ and $\sqrt n/2\le A,D\le2\sqrt n$. For arbitrary nonnegative weights on the row and column permutation groups, the normalized compatibility average is bounded by $\exp(C_0n^{54/100})$ times the product of the marginal $L^{1/\theta}$ factors, where $\theta=1-L/\log\sqrt n>0$ and the constants are absolute. Compatibility means injectivity in every original column. The bounded regular-moment and other moment/rank conclusions are not included. The formalization gives signed representation estimates for coordinate sweeps of the Thorp shuffle. For every signed occurrence of a Young diagram of size $2^d$, it bounds the logarithm of a fixed-order weighted sweep moment by a small multiple of the signed partition entropy plus an explicit remainder-size budget. The moment order is uniform over the diagrams and decompositions after the two small coefficients are fixed. The linked supporting angle bound controls the overlap of row, column, and global signed-type projections on an occupied rectangle, with explicit entropy, dimension, and missing-cell factors. These selected estimates support the paper’s full-density argument; the full $L^2$ and total-variation mixing conclusion is outside them. The formalized binary-sweep theorem gives a universal power contraction in every irreducible representation: for sufficiently large $d$, the averaged sweep on $2^d$ slots has operator norm at most $D^{-g}$ in representation dimension $D$, for some $g>0$. The sign expectation is zero, and a fixed number of independent sweeps approaches uniform in total variation from every initial deck. A separate conditional moment estimate covers allowed power-of-two grids and disjoint coordinate-respecting trajectories, with its stated feasibility premise. Physical-shuffle law identification is not included. The formalization proves the paper’s strict four-row permanent inequality. There are constants $4/3<p<2$ and $\varepsilon>0$ such that every probability law on $S_4$ within total-variation distance $\varepsilon$ of uniform, and with exactly uniform coordinate marginals, satisfies the permanent bound by the product of the four $L^p$ row norms for every nonnegative matrix. The exponent and neighborhood are uniform over those laws. The Thorp mixing consequence is outside this selected statement.

AI-drafted comparisons, with no recorded human review

  • Binary sweep contraction and mixing BinarySweep

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the binary sweep theorem: operator norm at most D^{-g} in every irreducible representation, with mixing after a fixed number of sweeps. Identification with the physical shuffle law is not included.

  • Conditional coordinate-sweep moment estimate CoordinateSweeps

    Unreviewed AI draft: states a supporting result. States the conditional moment estimate for grids and prescribed trajectories under its feasibility premise, a theorem in Section 3 of the paper rather than its headline sweep theorem.

  • Regular trace smoothing and full-permutation mixing CoordinateTrace

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the trace-smoothing theorem: regular trace at most 1 + (2^d)^{-10}, and the permutation law after a fixed number of sweeps within ½(2^d)^{-5} of uniform.

  • Strict four-row permanent inequality near the uniform law FourRowPermanent

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the paper’s robust four-row permanent inequality near the uniform law on S₄. The Thorp mixing consequence is outside the statement.

  • Row–column overlap on occupied boards OccupiedOverlap

    Unreviewed AI draft: states a supporting result. States the supporting row–column overlap estimate on occupied boards. Five definitions are Comparator holes; the library body of the endpoint statement OccupiedOverlapEndpoint matches the challenge file.

  • Fourier bounds from complementary coset caps PartialPermutation

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the main theorem of the partial-permutation paper: coset-mass bounds for a block partition give Fourier bounds bB·C(u)·D^{-1+(u+2)/b}. The paper’s asymptotic conclusion about products of random permutations is outside it.

  • Uniform signed-occurrence moment bound for coordinate sweeps SignedSweepMoment

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the paper’s moment bound for every signed occurrence of a Young diagram, uniform over diagrams. The full-deck convergence corollary is not part of the statement.

  • One-sided angle bound for signed spin types SpinAngle

    Unreviewed AI draft: states a supporting result. States the supporting angle bound for signed spin types on occupied rectangles. Its three Comparator holes have library bodies matching the challenge file.

  • Reciprocal Specht degrees and eight-block Fourier bound ThorpFirstReciprocal

    Unreviewed AI draft: states a supporting result. States supporting estimates for the mixing argument: bounds on reciprocal Specht-dimension sums and the eight-block Fourier estimate.

  • Frame, spectral, and density estimates for Thorp mixing ThorpRemaining

    Unreviewed AI draft: states a weaker form of a paper’s theorem. Bundles supporting estimates with full-deck mixing after 32800d steps and mixing time Θ(log 2^d). This states the optimal order of the main theorem with a worse constant: the theorem states mixing after 1600d shuffles.

  • Nine routing-density and representation estimates ThorpRouting

    Unreviewed AI draft: unclear. States nine routing and representation estimates that the note calls the earlier statements associated with the paper, with its changed later statements outside. Whether any of the nine is the current main theorem on routing density and representation contraction would need a line-by-line comparison.

  • Weighted row–column compatibility bound ThorpWeightedCompatibility

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the paper’s weighted compatibility theorem for balanced rectangles. The bounded regular-moment theorem and the rank and mixing consequences are not included.

  • Weighted Schatten moments and operator contraction for coordinate sweeps WeightedSweepMoments

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the paper’s weighted sweep-moment theorem: for each irreducible representation, a weighted Schatten moment at most one, hence operator norm at most D^{-c}. The mixing corollary is outside the statement.

Who ran the proof-verification program?

These groups report running Comparator with the release’s Lean statement files. For a configured theorem, it verifies OpenAI’s separate proof against the exact statement; some files instead configure definitions for comparison by name and type. Each row is the named group’s report. Atlas has not repeated the runs.

These reports do not list the software versions, proof-verification programs, axioms, time or memory in separate fields. Any such details in a report’s prose are reproduced below.

Binary sweep contraction and mixing BinarySweep

Two reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report

Conditional coordinate-sweep moment estimate CoordinateSweeps

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 618 s; peak memory 6.6 GiB).11 October 2026Original report

Regular trace smoothing and full-permutation mixing CoordinateTrace

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 650 s; peak memory 6.5 GiB).10 October 2026Original report

Strict four-row permanent inequality near the uniform law FourRowPermanent

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 55 s; peak memory 6.1 GiB).10 October 2026Original report

Row–column overlap on occupied boards OccupiedOverlap

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 744 s; peak memory 6.6 GiB).11 October 2026Original report

Fourier bounds from complementary coset caps PartialPermutation

Two reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report

Uniform signed-occurrence moment bound for coordinate sweeps SignedSweepMoment

Two reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report

One-sided angle bound for signed spin types SpinAngle

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 439 s).10 October 2026Original report

Reciprocal Specht degrees and eight-block Fourier bound ThorpFirstReciprocal

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 458 s; peak memory 6.4 GiB).11 October 2026Original report

Frame, spectral, and density estimates for Thorp mixing ThorpRemaining

Two reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report

Nine routing-density and representation estimates ThorpRouting

One report, at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report

Weighted row–column compatibility bound ThorpWeightedCompatibility

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 93 s; peak memory 6.3 GiB).11 October 2026Original report

Weighted Schatten moments and operator contraction for coordinate sweeps WeightedSweepMoments

Three reports, all at the release’s recorded version.

Reported byTheir wordTheir accountDateOriginal report
cert-machine“CERTIFIED”Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.10 October 2026Original report
MathVet“comparator-sandboxed-pass”Re-run through Comparator in a sandboxed virtual machine: accepted.8 October 2026Original report
Neal McBurnett, openai-math-lean-tooling“accepted”Re-run through Comparator with Lean’s kernel: accepted (the check took 183 s; peak memory 6.5 GiB).11 October 2026Original report

Other reports of proof verification

Other people’s accounts of running proof-verification software, with links to their reports.

  • Jack McCarthy, Re-running Comparator with the definition holes pinned, 9 October 2026. Source

    With the definition holes pinned, Comparator stops on a mismatch between challenge and solution at the endpoint definition; the author suspects a dependent-type technicality with a trivial fix.

  • Jack McCarthy, Re-running Comparator with the definition holes pinned, 9 October 2026. Source

    With the definition holes pinned, Comparator stops on a universe difference in a variable block of the challenge file; the author says they are fixing it and re-running.

  • From github.com/BenKnill/openai-math-hol-cards, 8 October 2026. Source

    Passed its release configuration; a strict re-check without holes was refused because one hole differs in a proof argument for the instance behind the literal 2, which Lean’s proof irrelevance makes definitionally equal. Explained, not resolved: strict comparison still refuses.

  • From github.com/BenKnill/openai-math-hol-cards, 8 October 2026. Source

    Passed its release configuration; a strict re-check without holes was refused because two definitions agree only after renaming universe parameters by position. The three holes themselves match structurally.

  • Robert Joseph George, Do the Lean statements in openai/math match their papers?, 7 October 2026. Source

    The review reports that Comparator accepted OpenAI’s solution for this challenge, run end to end on Lean 4.34.1 with the default kernel and only the three standard axioms.

  • Robert Joseph George, Do the Lean statements in openai/math match their papers?, 7 October 2026. Source

    The review reports that Comparator accepted OpenAI’s solution for this challenge, run end to end on Lean 4.34.1 with the default kernel and only the three standard axioms.

  • Lean community, OpenAI releases solutions to hundreds of problems (117 messages, 6-10 Oct 2026), 6 October 2026. Source

    With definition_names removed, SpinAngle and OccupiedOverlap initially failed with a constant mismatch (a universe difference and a dependent-type difference); their fork fixes both, along with DefocusingNLS (371) and ElementaryPositivity (169).

Other proofs and exact calculations

These reports examine claims in the papers using exact calculations for finite claims, independently of the release’s Lean proofs. Read each report for the exact claim it covers.

  • Routing densities and representation contraction for Thorp sweeps

    Reported by
    cert-machine
    Method
    Exact computation
    Their verdict
    “CERTIFIED” Exact code written independently from the paper confirms that the published object has the claimed property.
    What it covers
    “a finite component: the five rational inequalities of Section 4 (H(1/8), H(13/50), H(1) and the two entropy brackets at 7/25 and 1) with every supporting finite fact the text states (closed form of h_j from its definition, h_j <= sqrt(2/M), increasing ratios, the 169/70 tail cap, slopes 7/3 and 3/2, the 25/24 slope bound) — not the probabilistic/asymptotic arguments, and not the headline, which rests on the factorial proof of Section 3; this section is the paper’s independent second route”
    Original report
    Their report

Problems reported by others

Reports of errors and corrections from outside readers, as they state them. The atlas has not judged them.

  • Jack McCarthy, OpenAI math: Comparator fixes (mirror of openai/math at fd4aeeb), 9 October 2026. Source

    Removing the five definition holes broke the check on an auxiliary-lemma mismatch; a small fix to the solution makes it pass with no holes (Lean kernel only, no sandbox).

  • Jack McCarthy, OpenAI math: Comparator fixes (mirror of openai/math at fd4aeeb), 9 October 2026. Source

    Removing the three definition holes broke the check on a universe-name mismatch; a small fix to the solution makes it pass with no holes (Lean kernel only, no sandbox).

  • Alex Meiburg, Several of the comparator challenges in the OpenAI release are broken and hackable, 8 October 2026. Source

    Listed as broken in the same way, without further detail in the post.

  • Alex Meiburg, Several of the comparator challenges in the OpenAI release are broken and hackable, 8 October 2026. Source

    The post notes a sorry inside a definition that is a proof obligation, and that Comparator lets the rest of the definition change.

What these papers cite in the release, and what cites them

These links come from the papers’ bibliographies. A citation records that one paper cites another; how each cited result is used has not been examined here.

Within this family

Papers elsewhere in the release that this family’s papers cite

This family’s papers cite no paper elsewhere in the release.

Papers elsewhere in the release that cite this family’s papers

No paper elsewhere in the release cites this family’s papers.

Follow-up work by others

  • Palomar Registry, Palomar registry listing (api/v1/results), read 2026-10-11, 11 October 2026. Source

    PALOMAR-2026-10-08-000007 (JD Jones) adapts the four-row permanent inequality manuscript filed under family 238.

  • JD Jones, Sharp robustness and stability of the four-row permanent inequality (PALOMAR-2026-10-08-000007), 8 October 2026. Source

    Registered 2026-10-08T20:26:30Z; the OpenAI paper is recorded with relationship adapts.

Explain a result, compare statements or propose a correction

Atlas has no contributed explanation for this family yet. You can offer one with a GitHub account. You receive named credit and disclose any AI assistance.

Each button opens a GitHub form with this page and its source version filled in. Opening the form does not reserve the work.

Offer to write an explanationCompare one Lean statement with its paper: ThorpFirstReciprocal, ThorpRemaining, PartialPermutation, ThorpRouting, OccupiedOverlap, WeightedSweepMoments, CoordinateTrace, ThorpWeightedCompatibility, SpinAngle, SignedSweepMoment, BinarySweep, CoordinateSweeps, FourRowPermanentSuggest a source attributionSuggest a change

What people have said

Commentary on this family by others, dated and credited. It is commentary, not a finding.

  • An unsigned public tier list, the OpenAI Math Tier List, classes it as a proof and rates its significance, if it holds up, as “Solid publication, could lead to tenure”. The page says that nothing on it is verified.

    Author not stated. Source

  • Robert Joseph George, Do the Lean statements in openai/math match their papers?, 7 October 2026. Source

    The Lean statement covers only powers of two, where the paper’s Fourier bound holds for every n divisible by eight.

  • Robert Joseph George, Do the Lean statements in openai/math match their papers?, 7 October 2026. Source

    The constant in the Lean statement depends on the exact density, so there is no uniform rate over an interval as in the paper.

Questions awaiting review

Each item identifies a statement, definition or report that needs further review. Its link opens a GitHub form with the source version already recorded.

Read the Lean statement against the paper

  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form
  • An AI agent drafted an assessment of which part of the paper this Lean statement expresses. No named human review is recorded. Open the form

Read the definitions the solution supplies

  • OpenAI’s proof file supplies some definitions. Atlas records no comparison confirming that they mean the same objects as the paper’s definitions. IsUnitary, hilbertSpace, unitaryRepresentation, relabelledUnitary, OccupiedOverlapEndpoint Open the form
  • OpenAI’s proof file supplies some definitions. Atlas records no comparison confirming that they mean the same objects as the paper’s definitions. row, column, phi Open the form

Identify the theorem the catalogue describes

  • Family 238 No single paper statement clearly matches OpenAI’s description by wording, and no review identifying the intended theorem is recorded. Open the form

Address a question raised in an independent assessment

Compare the proposed paper and Lean matches

  • Family 238 Atlas records no review of whether each linked Lean statement and paper theorem mean the same thing. Open the form