Compilation··3:07·Probability and statistical mechanics

Six-Vertex Model, Lipschitz Heights, The O(3) Model Continuum Limit and 6 More

Compilation · 3:07 · 15 October 2026Watch on YouTube →

Watch on YouTubeAbout the problem →

About this video

9 results from OpenAI's public mathematics catalogue, one per chapter. Each chapter names the problem, shows the real objects, and states what OpenAI claims; none of it is peer reviewed yet.

Chapters:
0:00 Six-Vertex Model
0:21 Lipschitz Heights
0:41 Critical Ashkin–Teller Model
1:02 Orthogonally Invariant Spin Glass
1:24 Factor-of-IID Ising Threshold on Trees
1:45 Random Planar Maps
2:06 The O(3) Model Continuum Limit
2:27 GOE Universality for Random Regular Graphs
2:47 Perceptron Free Energy and Jamming

Six-Vertex Model: Square ice: at every corner of a grid, two arrows point in and two point out.
Explained: gradientdamage.com/p/225
Paper: github.com/openai/math/blob/main/preprints/The-Gaussian-free-…

Lipschitz Heights: Put a random whole-number height on every site of a triangular lattice, with odd values only and neighbours differing by 0 or 2, and fix the boundary at +1 on one arc and −1 on the other.
Explained: gradientdamage.com/p/232
Paper: github.com/openai/math/blob/main/preprints/Gaussian-free-fiel…

Critical Ashkin–Teller Model: The Ashkin–Teller model puts two coupled Ising magnets on every site of a grid.
Explained: gradientdamage.com/p/233
Paper: github.com/openai/math/blob/main/preprints/The-joint-scaling-…

Orthogonally Invariant Spin Glass: An Ising spin glass is a set of up/down spins whose pairwise couplings are random, some pulling neighbours together and some pushing them apart.
Explained: gradientdamage.com/p/234
Paper: github.com/openai/math/blob/main/preprints/All-temperature-pr…

Factor-of-IID Ising Threshold on Trees: Put a tiny magnet on every dot of an endless tree where each dot has d links, and let linked magnets prefer to agree.
Explained: gradientdamage.com/p/236
Paper: github.com/openai/math/blob/main/preprints/The-sharp-factor-o…

Random Planar Maps: Glue triangles together at random and you get a random planar map, a discrete random sphere.
Explained: gradientdamage.com/p/211
Paper: github.com/openai/math/blob/main/preprints/Random-Walks-on-Cr…

The O(3) Model Continuum Limit: Put a tiny compass needle that can point anywhere in 3D on every square of a grid, and let neighbours prefer to agree: that is the two-dimensional O(3) model, a famous toy for how quantum field theories get mass.
Explained: gradientdamage.com/p/215
Paper: github.com/openai/math/blob/main/preprints/The-canonical-mass…

GOE Universality for Random Regular Graphs: Take a random graph where every dot has exactly 3 links and list the eigenvalues of its adjacency matrix.
Explained: gradientdamage.com/p/219
Paper: github.com/openai/math/blob/main/preprints/GOE-bulk-universal…

Perceptron Free Energy and Jamming: The perceptron is the simplest neural network: one weight vector that must satisfy many random rules.
Explained: gradientdamage.com/p/222
Paper: github.com/openai/math/blob/main/preprints/The-free-energy-of…

The animations illustrate the results; they are not proofs.

Gradient Damage: mathematics you can see, equations you can hear.
youtube.com/@gradientdamage

Chapters

One problem per chapter. Each name links to the result's own page, each time to that moment on YouTube.

  1. Six-Vertex Model Probability and statistical mechanics
  2. Lipschitz Heights Probability and statistical mechanics
  3. Critical Ashkin–Teller Model Probability and statistical mechanics
  4. Orthogonally Invariant Spin Glass Probability and statistical mechanics
  5. Factor-of-IID Ising Threshold on Trees Probability and statistical mechanics
  6. Random Planar Maps Probability and statistical mechanics
  7. The O(3) Model Continuum Limit Probability and statistical mechanics
  8. GOE Universality for Random Regular Graphs Probability and statistical mechanics
  9. Perceptron Free Energy and Jamming Probability and statistical mechanics

Chapter 1: about the problem

Probability and statistical mechanics·OpenAI Math·Result 225

Gaussian free field limits throughout the balanced six-vertex regime

The balanced square-lattice six-vertex height field with a=b=1 and 0 < c ≤ 2 converges to a Gaussian free field, including at the endpoint c = 2. The plane state is defined by balanced-torus limits. For unit height increments and Green kernel -(2π)-1 log|x-y|, the exact variance multiplier is 1/arcsin(c/2).

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 2: about the problem

Probability and statistical mechanics·OpenAI Math·Result 232·from 0:21

Gaussian fields and interfaces for triangular-lattice Lipschitz heights

Proves Gaussian free field limits on bounded smooth simply connected domains for triangular-lattice height models: uniform odd heights with increments 0,±2 and two-arc boundary values ±1, and zero-boundary integer Lipschitz heights weighted by fixed x ∈ [1/√2,1]. Uniform real Lipschitz heights also converge to a Gaussian field; at a tuned opposite-boundary amplitude, their interface converges to chordal SLE4, establishing Schramm’s real-field/interface predictions.

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 3: about the problem

Probability and statistical mechanics·OpenAI Math·Result 233·from 0:41

The joint critical Ashkin–Teller current limit

Identifies the joint scaling limit of Ashkin–Teller heights and both complete current-cluster collections throughout the critical line, including the four-state Potts endpoint. In bounded Jordan domains with admissible lattice approximations and wired primal/free dual boundaries, the height converges to the predicted Gaussian free field, and the clusters to canonical recursive sets of that same field, retaining every nesting depth.

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 4: about the problem

Probability and statistical mechanics·OpenAI Math·Result 234·from 1:02

All-temperature pressure of orthogonally invariant Ising spin glasses

Gives an exact variational formula for the limiting pressure of orthogonally invariant Ising spin glasses at every fixed temperature, both almost surely and in expectation. The coupling matrix is a Haar-random rotation of a deterministic spectrum converging to a compactly supported law, with extreme eigenvalues converging to its support edges. The zero-field ground-state energy follows as temperature tends to zero. (Lean)

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 5: about the problem

Probability and statistical mechanics·OpenAI Math·Result 236·from 1:24

The exact factor-of-IID threshold for free Ising spins on trees

Determines when the free zero-field ferromagnetic Ising state on the infinite d-regular tree is a factor of independent vertex labels: exactly when tanh β ≤ (d-1)-1/2, including equality, for d ≥ 3 and β ≥ 0. The construction uses no root and is almost surely equivariant for each fixed tree automorphism, resolving the ferromagnetic case of Lyons's question. (Lean)

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 6: about the problem

Probability and statistical mechanics·OpenAI Math·Result 211·from 1:45

The geometric phase diagram, diffusion, and spectra of random planar maps

Critical Fortuin–Kasteleyn planar maps converge to Liouville quantum gravity spheres for 0 < q ≤ 4 and to the Brownian continuum random tree for q > 4, establishing the surface-to-tree geometric transition. For FK–Ising and spanning-tree-weighted maps, stationary random walks converge to Liouville Brownian motion on the limiting sphere. The FK–Ising spectral result also gives convergence of eigenvalues and heat traces, using the stated Brownian/LQG inputs. (Lean)

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 7: about the problem

Probability and statistical mechanics·OpenAI Math·Result 215·from 2:06

Canonical O(3) continuum limit and exact O(4) mass asymptotics

Constructs the canonical continuum limit of the two-dimensional nearest-neighbor O(3) model: a non-Gaussian local relativistic theory with a unique vacuum and a positive mass gap. For the square-lattice O(4) model, determines the exact leading asymptotic of the full transfer gap, mlat(β) ∼ 32eπ/4-1/2√β e-πβ. The family also proves exponential spin-correlation decay for two-dimensional nearest-neighbor O(n) models with n ≥ 3 at every positive temperature. (Lean)

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 8: about the problem

Probability and statistical mechanics·OpenAI Math·Result 219·from 2:27

GOE bulk universality for regular graphs with weak Anderson disorder

For every fixed degree d ≥ 3, the bulk adjacency-eigenvalue point process of a uniform simple random d-regular graph converges to the Gaussian orthogonal ensemble law, including for cubic graphs. The same fixed-energy universality persists under sufficiently weak fixed iid uniform diagonal disorder, throughout compact bands strictly inside the clean spectral edges.

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Chapter 9: about the problem

Probability and statistical mechanics·OpenAI Math·Result 222·from 2:47

Perceptron free energies and microscopic jamming exponents

Determines finite-temperature variational free energies for Gaussian Ising perceptrons with bounded Borel log-potentials and Gaussian spherical perceptrons with bounded continuous potentials, at every positive pattern density. A spherical extension treats bi-orthogonally invariant disorder with compact limiting singular-value distributions and no outliers. At margin −1, the quadratic-penalty spherical model has a sharp feasibility threshold and limiting gap and force laws, with system size, zero temperature, and critical density taken in that order. (Lean)

OpenAI, in its public catalogue of maths results · preprint, not peer reviewed

The papers and the claim →

Topics

six-vertex modelLipschitz height functionsAshkin-Teller modelspin glassfactor of IIDrandom planar mapsO(3) modelGOE universalityperceptronProbability and statistical mechanicsOpenAI mathAI proofAI mathematicsmathsGradient Damage