Plectis Atlas
Search

↑ and ↓ move, Enter opens, Esc closes.

Show on the map

Reflexive midpoint convexity and diamond distortion

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

Constructs a real reflexive Banach space with an asymptotically midpoint uniformly convex norm but no equivalent asymptotically uniformly convex norm, extending Baudier’s separation to reflexive spaces. In the same space, depth-k countably branching diamonds require distortion at least \(\sqrt{1+k/12}\), so midpoint uniform convexity does not force uniformly bounded diamond distortion even under reflexivity. (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

What the papers state

Theorem 1.1

In Asymptotic midpoint uniform convexity and unbounded diamond distortion in a reflexive tree space

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

The space X=J^* defined by (1.4) is infinite dimensional, separable, and reflexive, and has no equivalent AUC norm. Its given norm satisfies

\widehat\delta_X(t)\ge\sqrt{1+t^2/12}-1\qquad(t>0).

For every integer k\ge0, every embedding of D_k into X has distortion at least \sqrt{1+k/12}.

Theorem 2.1

In Midpoint convexity from bounded tree potentials and path costs

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

Each of the four completed spaces X_K is a separable infinite-dimensional real Banach space. For K\in\{K_{\rm s},K_{\rm p},K_{\rm a},K_{\rm g}\} and 0<t<1,

\widehat\delta_{X_K}(t)\ge\sqrt{1+t^2/4}-1.

None of these spaces admits an equivalent AUC norm.

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.

Nine more possible matches

  • Theorem 1.1 in Exact asymptotic moduli in a Daugavet subspace of L1.
  • Theorem 4.1 in Midpoint convexity from bounded tree potentials and path costs.
  • Theorem 1.1 in Independent products in real L1.
  • Theorem 2.1 in Midpoint convexity from two recursive potentials.
  • Theorem 1.1 in Distortion of countably branching diamonds from midpoint and tree energies.
  • Theorem 5.4 (Clipping estimate on a head and its complement) in Midpoint convexity from bounded tree potentials and path costs.
  • Theorem 6.3 (Separated midpoint families) in Midpoint convexity from two recursive potentials.
  • Theorem 7.4 in Midpoint convexity from two recursive potentials.
  • Theorem 1.1 in Midpoint lenses in segment spaces.
Reported proof verification
Seven 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.

Asymptotic midpoint uniform convexity and unbounded diamond distortion in a reflexive tree space

Undated, current edition.

Read it herePDF

We construct a separable reflexive real Banach space whose given norm is asymptotically midpoint uniformly convex but which admits no asymptotically uniformly convex equivalent norm. Its averaged midpoint modulus is at least \(\sqrt{1+t^2/12}-1\), and the countably branching diamond of depth k has distortion at least \(\sqrt{1+k/12}\) in this space. This gives a negative answer to the reflexive diamond converse for asymptotic uniform convexifiability.

Midpoint lenses in segment spaces

Undated, current edition.

Read it herePDF

For a real segment-forest dual with unbounded finite component heights and the real infinite-height coordinate predual, we bound the tail of an arbitrary displacement in a symmetric lens by \(2\sqrt{R^2-\|x\|^2}\), where R is the lens radius and x is its finitely supported center. Both given norms are asymptotically midpoint uniformly convex, although neither space admits an equivalent asymptotically uniformly convex norm. The finite-height forest dual is reflexive.

Distortion of countably branching diamonds from midpoint and tree energies

Undated, current edition.

Read it herePDF

We derive quantitative distortion bounds for countably branching diamond graphs from midpoint estimates and direct tree energies. In the dual of a finite-height segment forest, every distortion-D embedding of the depth-k diamond satisfies \(D^2\ge1+k/4\). The same bound holds in the infinite-height coordinate predual. We also obtain power-type distortion bounds for path-cost and recursive tree norms.

Exact asymptotic moduli in a Daugavet subspace of L1

Undated, current edition.

Read it herePDF

For an infinite-dimensional real subspace of L1 whose unit ball is totally bounded in measure and whose norm has the Daugavet property, we compute the averaged midpoint and one-sided asymptotic moduli at every unit center. They are \(\max\{t/2,t-1\}\) and \(\max\{0,t-2\}\), respectively. The same weak-neighborhood geometry excludes every equivalent asymptotically uniformly convex norm. A quantitative realization of the Kadets–Werner construction supplies a space with both hypotheses.

Midpoint convexity from bounded tree potentials and path costs

Undated, current edition.

Read it herePDF

Four real Banach spaces defined by bounded tree potentials satisfy the averaged asymptotic midpoint bound \(\widehat\delta(t)\ge\sqrt{1+t^2/4}-1\) for \(0\lt t\lt 1\), while none admits an asymptotically uniformly convex (AUC) renorming. On finite-height trees, the globally constrained norm equals the least additive cost of a Hilbert vector and root paths. The quadratic path and segment-start outer Hilbert sums are reflexive and asymptotically midpoint uniformly convex, and admit no equivalent AUC norm.

Midpoint convexity from two recursive potentials

Undated, current edition.

Read it herePDF

We study tree norms computed by two least nonnegative fields whose difference is the vector. For Euclidean child aggregation, two root-sum spaces have an averaged asymptotic midpoint modulus of at least \(t^3/128\) for \(0\lt t\lt 1\), including a reflexive joining-root space. A reflexive construction with a fixed zero root also satisfies a homogeneous cubic estimate. These spaces admit no equivalent asymptotically uniformly convex norm. We also obtain sixth-power and cubic estimates when the aggregation exponent depends on the height of a finite component.

Seven 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.

BoundedTreePotentials

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 2.1 in Midpoint convexity from bounded tree potentials and path costs.
  • Proposition 5.1 (Exact duality for the linear cost) in Midpoint convexity from bounded tree potentials and path costs.
  • Proposition 5.2 (Duality and branching in the outer sum) in Midpoint convexity from bounded tree potentials and path costs.
  • Theorem 4.1 in Midpoint convexity from bounded tree potentials and path costs.
  • Proposition 7.1 in Midpoint convexity from bounded tree potentials and path costs.
  • Theorem 1.1 in Asymptotic midpoint uniform convexity and unbounded diamond distortion in a reflexive tree space.
  • Proposition 9.1 (The bounded-path obstruction) in Midpoint lenses in segment spaces.
  • Proposition 8.2 in Midpoint convexity from bounded tree potentials and path costs.

The Lean statement

No paper theorem is paired with it.

Tree-potential midpoint convexity and renorming obstruction

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

theorem TreeCalculus.main_counterexample (r : Bool) (k : QuadraticKind) :    (CompleteSpace (TestCompletion (treeTestFamily r k)) ∧      TopologicalSpace.SeparableSpace (TestCompletion (treeTestFamily r k)) ∧      ¬Module.Finite ℝ (TestCompletion (treeTestFamily r k))) ∧    (∀ t : ℝ, 0 < t → t < 1 →      Real.sqrt (1 + t ^ 2 / 4) - 1 ≤        averagedModulusReal (TestCompletion (treeTestFamily r k)) t) ∧    (∀ (Y : Type u) [NormedAddCommGroup Y] [NormedSpace ℝ Y],      (TestCompletion (treeTestFamily r k) ≃L[ℝ] Y) → ¬IsAUCReal Y) := 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

DaugavetModuli

The paper states

Theorem 1.1

In Exact asymptotic moduli in a Daugavet subspace of L1

Let E be an infinite-dimensional closed real subspace of L^1(\Omega,\mathbb P), where \mathbb P is a probability measure. Suppose its closed unit ball is totally bounded for convergence in measure and its inherited norm has the Daugavet property. Then every x\in S_E and t>0 satisfy

H_E(x,t)=\max\{t/2,t-1\},\qquad D_E(x,t)=\max\{0,t-2\}.

Consequently these are also the formulas for \widehat\delta_E(t) and \overline\delta_E(t). The space E admits no equivalent AUC norm.

The words they share: convergence, t/2, probability, daugavet, omega, measure, property, ball, auc, dimensional, subspace, real, unit, equivalent, closed, max, infinite, space, norm, formulas.

Possible Lean counterpart

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

Exact asymptotic moduli and a Daugavet example

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

theorem exact_main {Ω : Type u_1} [MeasurableSpace Ω] (μ : Measure Ω)    [IsProbabilityMeasure μ] (E : Submodule ℝ (Lp ℝ 1 μ))    (_hclosed : IsClosed (E : Set (Lp ℝ 1 μ))) (hdim : ¬ FiniteDimensional ℝ E)    (hprecompact : MeasurePrecompactBall μ E) (hdaugavet : Daugavet E) :    (∀ (x : E), ‖x‖ = 1 → ∀ t : ℝ, 0 < t →      H (normSeminorm ℝ E) x t = max (t / 2) (t - 1) ∧      D (normSeminorm ℝ E) x t = max 0 (t - 2)) ∧    (∀ t : ℝ, 0 < t →      averagedModulus (normSeminorm ℝ E) t = max (t / 2) (t - 1) ∧      oneSidedModulus (normSeminorm ℝ E) t = max 0 (t - 2)) ∧    (∀ N : Seminorm ℝ E, EquivalentNorm N → ¬ AUC N) := 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 two more theorems: kadets_werner_existence, exact_example.

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

DiamondDistortion

The paper states

The words they share: eq, integer, segment, infinite, space, norm, sqrt, uniform, tree, diamond, distortion, embedding.

Possible Lean counterpart

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

All-pairs diamond distortion bound

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

theorem headline_all_pairs :    (∀ (k : ℕ) (f : Diamond.Vertex k → ForestDual) (D s : ℝ),      0 < s → (∀ u v,s*Diamond.distance k u v ≤ ‖f u-f v‖ ∧        ‖f u-f v‖ ≤ D*s*Diamond.distance k u v) → 1+(k:ℝ)/4 ≤ D^2) ∧    (∀ (k : ℕ) (f : Diamond.Vertex k → InfinitePredual) (D s : ℝ),      0 < s → (∀ u v,s*Diamond.distance k u v ≤ ‖f u-f v‖ ∧        ‖f u-f v‖ ≤ D*s*Diamond.distance k u v) → 1+(k:ℝ)/4 ≤ D^2) := 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

ForestSpace

The paper states

The words they share: k/12, eq, 2/12, integer, separable, auc, dimensional, reflexive, distortion, segment, embedding, equivalent, space, norm, sqrt, uniform, convexity, diamond, asymptotic, midpoint, least, infinite.

Possible Lean counterpart

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

Reflexive tree space, midpoint modulus, and diamond distortion

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

theorem main_theorem :    (¬FiniteDimensional ℝ (FullDual Vertex)) ∧    TopologicalSpace.SeparableSpace (FullDual Vertex) ∧    Reflexive (FullDual Vertex) ∧    (∀ (α β : ℝ) (R : EquivalentNorm (FullDual Vertex) α β), ¬IsAUC R.N) ∧    (∀ t : ℝ, 0 < t →      Real.sqrt (1+t^2/12)-1 ≤ averageModulus (FullDual Vertex) norm t) ∧    (∀ (k : ℕ) (f : (Diamond.diamond k).V → FullDual Vertex) (C : ℝ),      Diamond.EmbeddingBound k f C → Real.sqrt (1+(k:ℝ)/12) ≤ C) := 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

IndependentProducts

The paper states

Theorem 1.1

In Independent products in real L1: asymptotic midpoint convexity without AUC renormings

For every multiplier W as above, the Banach space X_W is infinite dimensional and has no equivalent AUC norm. For the exponential and Gaussian-square laws, the inherited norms satisfy

\begin{aligned}\whdelta_{X_{\mathrm{exp}}}(t) &\ge \frac{7t}{30}c(10/t)>0 &&(t>0),\\ \whdelta_{X_{\mathrm{sq}}}(t) &\ge\frac{\PP\bigl(|G|\ge4\sqrt2(2+\sqrt2)/t\bigr)}{8}>0 &&(t>0),\\ \whdelta_{X_{\mathrm{sq}}}(t) &\ge\frac{\PP\bigl(|G|\ge8\sqrt6/t\bigr)}{4}>0 &&(t>0).\end{aligned}

The words they share: 7t/30, inherited, exponential, multiplier, square, gaussian, laws, exp, auc, dimensional, equivalent, infinite, norm, sqrt, independent, products, asymptotic, midpoint, real, space.

Possible Lean counterpart

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

Independent-product spaces and positive midpoint moduli

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

theorem main_general (μ : Measure ℝ) (hμ : MultiplierHypotheses μ) :    ¬ FiniteDimensional ℝ (productSpan μ) ∧    ∀ N : Seminorm ℝ (productSpan μ),      IsEquivalentNorm (productSpan μ) N → ¬ IsAUC (productSpan μ) N := 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 three more theorems: main_exponential, main_gaussian_24, main_gaussian_12.

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

MidpointLenses

The paper states

Theorem 1.1

In Midpoint lenses in segment spaces

Let Y be either the finite-height forest dual X=J^* or the infinite-height coordinate predual B_\infty. For every finite ancestral H, every x,y\in Y with x=P_Hx, and every R\ge0 with \|x+y\|,\|x-y\|\le R,

\|Q_Hy\|\le2\sqrt{R^2-\|x\|^2}.

The words they share: predual, dual, forest, coordinate, ancestral, infinite, sqrt, segment, midpoint, spaces, height, finite.

Possible Lean counterpart

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

Midpoint-lens tail estimate

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

theorem midpoint_lens_main :    (∀ (H : Finset ForestVertex) (_ : Ancestral ForestAncestor H)      (x y : X) (_ : x = head H x) (R : ℝ) (_ : 0 ≤ R)      (_ : ‖x + y‖ ≤ R) (_ : ‖x - y‖ ≤ R),      ‖tail H y‖ ≤ 2 * Real.sqrt (R ^ 2 - ‖x‖ ^ 2)) ∧    (∀ (H : Finset InfiniteVertex) (_ : Ancestral List.IsPrefix H)      (x y : BInfinite) (_ : x.val = head H x.val) (R : ℝ) (_ : 0 ≤ R)      (_ : ‖x + y‖ ≤ R) (_ : ‖x - y‖ ≤ R),      ‖tail H y.val‖ ≤ 2 * Real.sqrt (R ^ 2 - ‖x‖ ^ 2)) :=  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

RecursivePotentials

The paper states

It is paired with Theorem 2.1 and Theorem 6.3, each shown below.

Theorem 2.1

In Midpoint convexity from two recursive potentials

For N=N_\Sigma on X_\Sigma, or N=N_J on X_J, one has

\widehat\delta_N(t)\ge t^3/128\qquad(0<t<1).

For X_0, a finite initial set H, a vector x supported in H, and z\in F_H\setminus\{0\}, one has

\frac{N_0(x+z)+N_0(x-z)}2\ge N_0(x) +\frac{N_0(z)^3}{8(2N_0(x)+N_0(z))^2}.

Each of X_\Sigma,X_0,X_J admits no equivalent AUC norm.

Theorem 6.3 (Separated midpoint families)

In Midpoint convexity from two recursive potentials

For every \varepsilon>0 there is \eta(\varepsilon)\in(0,1) such that, for x,z_1,z_2,\ldots\in X_J,

\left. \begin{gathered} N_J(x+z_j),N_J(x-z_j)\le1\quad(j\ge1),\\ N_J(z_i-z_j)\ge\varepsilon\quad(i\ne j) \end{gathered} \right\} \quad\Longrightarrow\quad N_J(x)\le1-\eta(\varepsilon).

For 0<\varepsilon\le2, one can take \eta(\varepsilon)=\min\{1/2,\gamma(2,\varepsilon/16)\}.

The words it shares with Theorem 2.1: 3/128, sigma, initial, auc, equivalent, vector, supported, one, norm, finite, recursive, potentials; with Theorem 6.3: min, gamma, 1/2, separated, families, one, midpoint.

Possible Lean counterpart

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

Recursive-potential spaces and midpoint convexity

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

def MainClaim : Prop :=  ∃ h0 : FiniteRootedTree.Heads.{0}, letI := h0;  ∃ h1 : TreeVector.Construction.{0}, letI := h1;  ∃ h2 : ZeroTreeVector.Construction.{0}, letI := h2;  ∃ h3 : AggregationVector.Construction.{0}, letI := h3;  ∃ h4 : AggregationVector.LpConstruction.{0}, letI := h4;  rec_mainStatement ∧  rec_completionStatement ∧  stop_all_radiiStatement ∧  XJoined_separated_familiesStatement ∧  Variable.spaceStatement ∧  Variable.lp_model_exactStatement ∧  Variable.one_sided_modulus_zeroStatement ∧  Variable.no_equivalent_AUCStatement ∧  Variable.sixth_stabilityStatement ∧  Variable.cubic_stabilityStatement ∧

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: ForestSpace.main_theorem (unconditional) states for FullDual Vertex (continuous dual of the completed disjoint-segment-norm space over a forest): infinite-dimensional, separable, canonically reflexive, no equivalent norm is AUC, averaged midpoint modulus ≥ √(1+t²/12) - 1 for all t > 0 (this implies positivity of the usual max-version midpoint modulus, since max ≥ average: the one routine step), and every embedding of the depth-k countably branching diamond has distortion ≥ √(1+k/12). The other six challenges are companion results of the seven papers (lens tail bound, all-pairs diamond bound 1+k/4 ≤ D², exact Daugavet-subspace moduli, tree-potential, independent-product and recursive-potential spaces), not needed for the headline.

Definitions the group asks readers to compare with the paper: aucModulus, averageModulus and Cofinite are hand-built (inf_{N x=1} sup_{F cofinite closed} inf_{y∈F, N y=1} (N(x+ty)-1), resp. with the average of N(x±ty) minus 1), matching the standard asymptotic moduli; IsAUC N := ∀ t > 0, 0 < aucModulus N t and EquivalentNorm as a seminorm with α‖x‖ ≤ N x ≤ β‖x‖; Diamond.diamond k := base.iterate k (each edge replaced by ℕ-many 2-edge paths, graph distance via SimpleGraph.dist) and EmbeddingBound (∃ scale s > 0, s·d ≤ ‖f u - f v‖ ≤ C·s·d): standard distortion; docs note 22 companion inputs are assumed in some graph deductions, not in the headline theorem

“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 paper constructs a Banach space whose midpoint geometry is asymptotically uniformly convex even though no equivalent norm is asymptotically uniformly convex in the usual one-sided sense. The formalization proves that the full dual of the specified segment-norm completion is infinite-dimensional, separable, and reflexive; its averaged midpoint modulus is at least $\sqrt{1+t^2/12}-1$ for every $t>0$, while every equivalent norm fails asymptotic uniform convexity. It also proves that an embedding of a depth-$k$ countably branching diamond has distortion at least $\sqrt{1+k/12}$.

The general-forest and word-forest results use the closed span of the coordinate functionals. Their conclusions retain the finite-ancestor and equivalent-norm hypotheses, including the scale $\alpha/(2\beta)$ when $0<\alpha\le\beta$ and the new norm lies between $\alpha$ and $\beta$ times the original norm; they do not identify that span with the full dual or assert reflexivity for every forest. The paper controls midpoint lenses in Banach spaces defined by tree-segment norms. The formalization proves the following estimate in both the full dual of the finite-height forest space and the coordinate predual of the infinite-tree space. If $x$ is supported on a finite ancestral set $H$, $R\ge0$, and $\|x+y\|,\|x-y\|\le R$, then the part of $y$ outside $H$ has norm at most $2\sqrt{R^2-\|x\|^2}$. The formalization also covers selected consequences for averaged midpoint moduli and separated families of vectors, together with reflexivity and renorming obstructions. This scope concerns the segment-space results; the companion results on diamond energies remain separate. The paper studies embeddings of countably branching diamond graphs into Banach spaces built from tree segments. The formalization proves that every embedding of the depth-$k$ diamond with distortion $D$, at any positive scale, satisfies $1+k/4\le D^2$ in either of two spaces: the full dual of the completed finite-height forest space and the closed coordinate span in the dual of the infinite-tree space. The bound applies to all pairs of vertices and is unconditional. The formalization also contains selected graph deductions under explicit companion assumptions. Twenty-two of those companion inputs are assumed in these deductions; the diamond distortion bound above does not require them. The paper exhibits a real $L_1$ subspace with positive averaged midpoint convexity but no equivalent asymptotically uniformly convex norm. For every closed infinite-dimensional subspace whose unit ball is precompact in measure and which has the Daugavet property, the formalization computes the moduli at every unit vector: the averaged midpoint modulus is $\max(t/2,t-1)$ and the usual one-sided asymptotic modulus is $\max(0,t-2)$ for every $t>0$. The same formulas hold after taking the infimum over unit vectors, and every equivalent norm fails asymptotic uniform convexity. The formalization also constructs such a subspace on a countable product of unit intervals and proves that it has the stated measure-precompactness and Daugavet properties, so the modulus formulas apply to an actual example. The paper constructs Banach spaces from bounded tree potentials and path costs to separate averaged midpoint convexity from asymptotic uniform convexity. For the specified tree-potential completions, the formalization proves completeness, separability, infinite dimension, and an averaged midpoint modulus of at least $\sqrt{1+t^2/4}-1$ for $0<t<1$. No space linearly isomorphic to one of these completions is asymptotically uniformly convex. The formalization also covers selected path-cost duality and comparison results, clipping and energy estimates, and further renorming obstructions. These retain their stated support, head, and tail hypotheses; the remaining auxiliary assertions of the paper are outside this scope. The paper forms a real $L_1$ space from the closed span of products of independent multipliers along tree paths. For a positive, nonconstant multiplier of mean one and finite second moment, the formalization proves that this space is infinite-dimensional and has no equivalent asymptotically uniformly convex norm. For every $t>0$, the averaged midpoint modulus nevertheless has explicit positive lower bounds. If $G$ is a standard real Gaussian, the exponential-multiplier bound is $\frac{7t}{480}\mathbb E[(|G|-20/t)_+]$, where $u_+=\max(u,0)$. For squared-Gaussian multipliers, the bounds are $\frac18\Pr(|G|\ge4\sqrt2(2+\sqrt2)/t)$ and $\frac14\Pr(|G|\ge8\sqrt6/t)$. The paper constructs three Banach spaces from recursive tree potentials: a root-sum space, a zero-root space, and a joined space. The formalization proves that all three are infinite-dimensional and have no equivalent asymptotically uniformly convex norm. The root-sum and joined spaces have averaged midpoint modulus at least $t^3/128$ for $0<t<1$, and the root-sum modulus is positive for every $t>0$. The zero-root estimate retains its finite ancestral support and tail hypotheses, and the zero-root and joined spaces are reflexive. The formalization also covers a separated-midpoint result for the joined space and a variable-exponent model with its recursive norm, duality, projections, renorming obstruction, and sixth- and third-order stability estimates. The separation, sequence, support, and head assumptions in these results are essential parts of their statements.

AI-drafted comparisons, with no recorded human review

  • Tree-potential midpoint convexity and renorming obstruction BoundedTreePotentials

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the tree-potential paper’s main theorem: complete separable infinite-dimensional spaces with averaged midpoint modulus at least √(1 + t²/4) − 1 and no AUC renorming.

  • Exact asymptotic moduli and a Daugavet example DaugavetModuli

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the Daugavet paper’s main theorem (exact averaged midpoint and asymptotic moduli for the stated subspaces of L¹) and constructs an example satisfying its hypotheses.

  • All-pairs diamond distortion bound DiamondDistortion

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the diamond-distortion paper’s main theorem: every distortion-D embedding of the depth-k diamond satisfies D² ≥ 1 + k/4 in either space.

  • Reflexive tree space, midpoint modulus, and diamond distortion ForestSpace

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the main theorem of the reflexive tree-space paper: the dual J* is infinite-dimensional, separable and reflexive with no AUC renorming, with the midpoint modulus bound and diamond distortion at least √(1 + k/12).

  • Independent-product spaces and positive midpoint moduli IndependentProducts

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the independent-products paper’s main theorem: the spaces X_W have no equivalent AUC norm, with the explicit midpoint-modulus lower bounds for exponential and Gaussian-square multipliers.

  • Midpoint-lens tail estimate MidpointLenses

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the midpoint-lens paper’s main tail estimate in both the finite-height forest dual and the infinite-tree predual.

  • Recursive-potential spaces and midpoint convexity RecursivePotentials

    Unreviewed AI draft: states a paper’s theorem as the paper does. States the recursive-potential paper’s main theorem (midpoint modulus at least t³/128 for the root-sum and joined spaces, with no AUC renorming) together with its separated-family and variable-exponent results.

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.

Tree-potential midpoint convexity and renorming obstruction BoundedTreePotentials

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 253 s; peak memory 6.3 GiB).11 October 2026Original report

Exact asymptotic moduli and a Daugavet example DaugavetModuli

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 401 s; peak memory 6.3 GiB).10 October 2026Original report

All-pairs diamond distortion bound DiamondDistortion

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 73 s; peak memory 6.2 GiB).11 October 2026Original report

Reflexive tree space, midpoint modulus, and diamond distortion ForestSpace

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 102 s; peak memory 6.3 GiB).10 October 2026Original report

Independent-product spaces and positive midpoint moduli IndependentProducts

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 235 s; peak memory 6.4 GiB).10 October 2026Original report

Midpoint-lens tail estimate MidpointLenses

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 229 s; peak memory 6.2 GiB).10 October 2026Original report

Recursive-potential spaces and midpoint convexity RecursivePotentials

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 176 s; peak memory 6.3 GiB).10 October 2026Original report

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.

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: ForestSpace, MidpointLenses, DiamondDistortion, DaugavetModuli, BoundedTreePotentials, IndependentProducts, RecursivePotentialsSuggest 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 counterexample and rates its significance, if it holds up, as “Field-specific journal”. 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 doc lists further results as formalised that no Comparator theorem name covers.

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

    The doc lists further results as formalised that no Comparator theorem name covers.

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

Identify the theorem the catalogue describes

  • Family 331 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 331 Atlas records no review of whether each linked Lean statement and paper theorem mean the same thing. Open the form