Lean statement

Variance bounds for the random $k$-SAT hitting time

Lean is a language for mathematical statements and proofs that a computer can verify. This page shows one statement file from OpenAI’s release, possible paper matches and independent reports. The code-like names below identify the definitions or theorems selected by the proof-verification settings.

Theorem selected for verification: OAI.RandomKSAT.variance_main

Lean statement file: SATVariance.leanShow on the mapOpenAI’s Lean catalogue does not list this as a main result; the release’s Lean notes for family 235 link it.

See the candidate theorem Who ran the proof-verification program Report a comparison with the paper (GitHub issue)

The Lean statement

The lines of the Lean statement that state the theorem, then the definitions in the file that it uses.

theorem variance_main (k : ℕ) (hk : 3 ≤ k) :    (∃ C : ℝ, 0 ≤ C ∧ ∀ n : ℕ, k ≤ n →      variance (H : Stream n k → ℝ) (streamLaw n k) ≤ C * n * ell k n) ∧    (∀ B : ℝ, 0 < B → ∃ C : ℝ, 0 ≤ C ∧ ∀ n : ℕ, k ≤ n →      variance (T B : Stream n k → ℝ) (streamLaw n k) ≤ C * n * ell k n) ∧    (∃ c : ℝ, 0 < c ∧      (∃ N : ℕ, ∀ n : ℕ, N ≤ n → k ≤ n →        c * n ≤ variance (H : Stream n k → ℝ) (streamLaw n k)) ∧      (∀ B : ℝ, U k < B → ∃ N : ℕ, ∀ n : ℕ, N ≤ n → k ≤ n →        c * n ≤ variance (T B : Stream n k → ℝ) (streamLaw n k))) := by  sorry

A Lean statement states the theorem and leaves its proof as sorry on purpose. OpenAI’s proof is in a separate solution module, and a program called Comparator checks that proof against this statement.

Definitions needed to read it

Local definitions the statement names, followed through the definitions they name in turn. The match is by name in the source text, so a name reached only through notation can be missed; the full file is below.

abbreviation OAI.RandomKSAT.Assignment line 11

abbrev Assignment (n : ℕ) := Fin n → Bool

definition OAI.RandomKSAT.Clause line 13

def Clause (n k : ℕ) :=  (s : {s : Finset (Fin n) // s.card = k}) × (s.1 → Bool)

definition OAI.RandomKSAT.Satisfies line 26

def Satisfies {n k : ℕ} (c : Clause n k) (a : Assignment n) : Prop :=  ∃ v : c.1.1, a v = c.2 v

abbreviation OAI.RandomKSAT.Stream line 29

abbrev Stream (n k : ℕ) := ℕ → Clause n k

definition OAI.RandomKSAT.clauseLaw line 31

def clauseLaw (n k : ℕ) : Measure (Clause n k) := uniformOn Set.univ

definition OAI.RandomKSAT.streamLaw line 33

def streamLaw (n k : ℕ) : Measure (Stream n k) :=  Measure.infinitePi (fun _ : ℕ => clauseLaw n k)

definition OAI.RandomKSAT.PrefixSAT line 36

def PrefixSAT {n k : ℕ} (ω : Stream n k) (m : ℕ) : Prop :=  ∃ a : Assignment n, ∀ i < m, Satisfies (ω i) a

definition OAI.RandomKSAT.firstFailure line 39

def firstFailure {n k : ℕ} (ω : Stream n k) : ℕ∞ :=  sInf ((fun m : ℕ => (m : ℕ∞)) '' {m | ¬ PrefixSAT ω m})

definition OAI.RandomKSAT.H line 42

def H {n k : ℕ} (ω : Stream n k) : ℝ := (firstFailure ω).toNat

definition OAI.RandomKSAT.cap line 44

def cap (n : ℕ) (B : ℝ) : ℕ := ⌊B * n⌋₊

definition OAI.RandomKSAT.T line 46

def T {n k : ℕ} (B : ℝ) (ω : Stream n k) : ℝ :=  (min (firstFailure ω) (cap n B : ℕ∞)).toNat

definition OAI.RandomKSAT.ell line 49

def ell (k n : ℕ) : ℝ :=  if k = 3 then Real.log (Real.exp 1 * n) else 1

definition OAI.RandomKSAT.U line 52

def U (k : ℕ) : ℝ := Real.log 2 / (-Real.log (1 - ((2 : ℝ) ^ k)⁻¹))

Possible paper match, with no recorded review

Atlas selected this possible match by comparing words in the Lean statement with every statement in the family. No review of whether the statements mean the same thing is recorded.

Found in the paper’s title, which its main theorem states: sat and variance. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: theta and var.

Statements in Linear Variance of the Random 3-SAT Hitting Time come first because the AI draft reading of OpenAI’s note shares words with that paper’s title.

Theorem 1.1 of Linear Variance of the Random 3-SAT Hitting Time

Theorem

There is an absolute constant C<\infty such that

\Var(H_n)\le Cn\qquad(n\ge3).

There are also absolute constants c>0 and n_0 such that \Var(H_n)\ge cn for n\ge n_0. In particular, \Var(H_n)=\Theta(n) as n\to\infty.

Section 1, Introduction Read it in the paper TeX source, line 61

Show the TeX

There is an absolute constant $C<\infty$ such that
\[
 \Var(H_n)\le Cn\qquad(n\ge3).
\]
There are also absolute constants $c>0$ and $n_0$ such that
$\Var(H_n)\ge cn$ for $n\ge n_0$. In particular,
$\Var(H_n)=\Theta(n)$ as $n\to\infty$.

Other possible paper matches, closest wording first

Proposition 6.2 of Variance of the Random k-SAT Hitting Time. Words of the Lean statement, its names split into words, that also appear in the theorem: caps, variances, log, min and integer. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: multiples, theta, linear, positive, lower, bound and two more.
Theorem 1.1 of Variance of the Random k-SAT Hitting Time. Words of the Lean statement, its names split into words, that also appear in the theorem: ell, log and integer. Found in the paper’s title, which its main theorem states: sat and variance. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: var and sufficiently.
Corollary 5.2 of Variance of the Random k-SAT Hitting Time. Words of the Lean statement, its names split into words, that also appear in the theorem: ell, uniform, log, variance, min and clauses. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: upper, bounds, var and sufficiently.
Theorem C.1 of Variance of the Random k-SAT Hitting Time. Words of the Lean statement, its names split into words, that also appear in the theorem: prefix, variance, log and integers. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: var.
Proposition 4.6 of Computing the Random 3-SAT Threshold. Words of the Lean statement, its names split into words, that also appear in the theorem: assignment, real, clauses and integers. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: time and linear.
Corollary C.10 of Variance of the Random k-SAT Hitting Time. Words of the Lean statement, its names split into words, that also appear in the theorem: prefix, variance, min and clause. Shared only with the release’s name for the Lean statement or OpenAI’s note on it: bounds and sufficiently.

How Atlas selects possible matches: it compares words in the Lean statement with every statement in the family. It also splits Lean names into words. The release’s label and note receive less weight; a main theorem wins a close match. If the Lean wording yields no clear match, Atlas uses the catalogue description. This selects passages to compare; it does not establish that they mean the same thing.

Reported Comparator runs

three independent groups report that Comparator accepted runs for this Lean statement file: Carlos Toledo’s cert-machine; MathVet; Neal McBurnett’s openai-math-lean-tooling. Each group reports its own Comparator run for this statement file. Atlas has not repeated the runs.

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

Carlos Toledo’s cert-machine

How they ran it, in their words

Every Comparator challenge re-run on Linux with nanoda switched on, decided by a preregistered rule; every family’s challenge statements read against OpenAI’s headline, clause by clause. Decider: Comparator (leanprover/comparator d03acab1 = tag v4.34.0, built with the release’s toolchain v4.34.1), lean4export 076e8e57, landrun 811cfff5, nanoda_lib 3a240721 — on GitHub’s Linux runners (.github/workflows/openai-math-kernel.yml), one j…

Accepted Their term: “CERTIFIED”.
Re-run through Comparator with the independent nanoda kernel switched on: Lean’s kernel and nanoda both accept OpenAI’s solution.
Reported 10 October 2026, at the release version shown here. Original verification report

MathVet

How they ran it, in their words

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. Its protocol describes a planned error-rate study; this import does not establish that the study has been completed.

Accepted Their term: “comparator-sandboxed-pass”.
Re-run through Comparator in a sandboxed virtual machine: accepted.
Reported 8 October 2026, at the release version shown here. Original verification report

Neal McBurnett’s openai-math-lean-tooling

How they ran it, in their words

Each challenge’s solution is built and checked by Comparator at the pinned release, on Lean leanprover/lean4:v4.34.1, recording the verdict, the check time and the peak memory; the run was still in progress at this snapshot (192 accepted, 224 pending).

Accepted Their term: “accepted”.
Re-run through Comparator with Lean’s kernel: accepted (the check took 251 s; peak memory 6.3 GiB).
Reported 11 October 2026, at the release version shown here. Original verification report

What an accepted Comparator run would establish

If Comparator is run under this configuration and accepts, the solution proves the configured theorem using only the allowed axioms, and the Lean kernel accepts every step.

This configuration lists no definition holes, so every definition the statement uses is compared as it is written in this file.

Allowed axioms: propext, Quot.sound, Classical.choice

Acceptance leaves open whether the Lean statement means the same thing as the paper’s theorem, whether the paper’s other claims are correct, and whether the result is new. Those questions require comparing the definitions and statements, reviewing the paper’s arguments, and reading the earlier literature.

Do family 235’s Lean statements express the advertised result?

These assessments compare all the family’s Lean statements with the whole result OpenAI describes in its catalogue. They do not assess this statement alone. Each group uses its own published rules.

Neither group finds that the Lean statements express or imply the whole advertised result. They describe what the statements express in different terms.

MathVet
Finds part of it stated, or a narrower or weaker statement Their term: “partial”.
Only a special case is stated, or a co-equal headline claim has no challenge statement. Their record
Their note: Stated: FixedClauseThreshold.main (k) (hk : 3 ≤ k) : HasLimitingThreshold k (∃ α > 0, P(SAT) → 1 for c < α and → 0 for c > α with m = ⌊c n⌋ i.i.d. uniform proper clauses), SATComputability.main (3-SAT threshold α with a Nat.Partrec.Code producing rationals q r, |q r - α| ≤ 2^-r) and variance_main. But the variance claim is only Θ(n) for k >= 4: for k = 3 the Lean gives Var(H) ≤ C n ell with ell = log(e n) (an n log n upper bound) and a linear lower bound, so the summary's 'hitting-time variance Θ_k(n) for every fixed k ≥ 3' is not stated for k = 3 (docs: 'They do not show that the logarithmic factor in the k=3 variance bound is necessary').
Carlos Toledo’s cert-machine
Finds part of it stated, or a narrower or weaker statement Their term: “NARROWER — DECLARED”.
The challenge statements state less than the headline, and OpenAI’s Lean note says so. Their record
Their note: the k = 3 linear upper bound Var(H_n) = O(n) (Lean has O(n log n))

Read the complete assessments on the family page.

OpenAI’s own note on what this Lean statement covers

“The formalization proves $\mathrm{Var}(H_n)=\Theta_k(n)$ for every fixed $k\ge4$. For $k=3$, it proves a positive linear lower bound and an $O(n\log n)$ upper bound. The same upper bounds hold after clipping at any fixed positive multiple of $n$, and linear lower bounds hold for sufficiently high clipping levels.”
Quoted from OpenAI’s Lean note for this family

An AI agent drafted an interpretation of that note, classifying this statement as “states a paper’s theorem as the paper does”. No human review of the draft is recorded.

Supporting material

The complete Lean statement file

70 lines, copied from release version fd4aeeb. A name defined in this file links to its line; pointing at the name shows its definition.

Declared in this fileOpenAI’s library at fd4aeebOther linked names open the current documentation for Mathlib, Lean’s mathematical library.

Links in the file: 35 to its own declarations, 24 to Mathlib’s documentation. 17 declared names also link to the same declaration in OpenAI’s library. OpenAI uses Mathlib version d13f23b. The linked documentation describes the current version, so names that have since changed may be missing there.

Definitions and lemmas named by the statement

If Comparator accepts a proof, it proves variance_main exactly as written here. To compare that statement with the paper, also read the definitions and lemmas listed below. Atlas found them by following names in the source text and lists each after the names it uses. For x.foo, Atlas includes every foo defined here for a type, even when the statement may use only one of them.

It reaches 13 of the file’s 16 other declarations; with the theorem itself they span 31 lines. 6 of them are named in the statement itself.

Names in the namespace OAI.RandomKSAT are listed without it.

abbrev Assignmentline 11library
def Clauseline 13library
def Satisfiesline 26library
abbrev Streamline 29in the statementlibrary
def clauseLawline 31library
def streamLawline 33in the statementlibrary
def PrefixSATline 36library
def firstFailureline 39library
def Hline 42in the statementlibrary
def capline 44library
def Tline 46in the statementlibrary
def ellline 49in the statementlibrary
def Uline 52in the statementlibrary
theorem variance_mainline 54selected theoremlibrary

Other Lean rules that may affect the statement 3

Lean can use instances, notation and attributes without naming them in the statement. These are rules for interpreting or finding mathematical objects. Atlas’s scan of names cannot follow all such uses; read the listed source as well.

instance clauseFintype line 16
instance clauseMeasurableSpace line 21
instance clauseMeasurableSingletonClass line 23

import Mathlibnamespace OAInoncomputable sectionopen MeasureTheory ProbabilityTheorynamespace RandomKSATabbrev Assignment (n : ℕ) := Fin n → Booldef Clause (n k : ℕ) :=  (s : {s : Finset (Fin n) // s.card = k}) × (s.1 → Bool)instance clauseFintype (n k : ℕ) : Fintype (Clause n k) := by  classical  unfold Clause  infer_instanceinstance clauseMeasurableSpace (n k : ℕ) : MeasurableSpace (Clause n k) := ⊤instance clauseMeasurableSingletonClass (n k : ℕ) :    MeasurableSingletonClass (Clause n k) := ⟨fun _ => trivial⟩def Satisfies {n k : ℕ} (c : Clause n k) (a : Assignment n) : Prop :=  ∃ v : c.1.1, a v = c.2 vabbrev Stream (n k : ℕ) := ℕ → Clause n kdef clauseLaw (n k : ℕ) : Measure (Clause n k) := uniformOn Set.univdef streamLaw (n k : ℕ) : Measure (Stream n k) :=  Measure.infinitePi (fun _ : ℕ => clauseLaw n k)def PrefixSAT {n k : ℕ} (ω : Stream n k) (m : ℕ) : Prop :=  ∃ a : Assignment n, ∀ i < m, Satisfies (ω i) adef firstFailure {n k : ℕ} (ω : Stream n k) : ℕ∞ :=  sInf ((fun m : ℕ => (m : ℕ∞)) '' {m | ¬ PrefixSAT ω m})def H {n k : ℕ} (ω : Stream n k) : ℝ := (firstFailure ω).toNatdef cap (n : ℕ) (B : ℝ) : ℕ := ⌊B * n⌋₊def T {n k : ℕ} (B : ℝ) (ω : Stream n k) : ℝ :=  (min (firstFailure ω) (cap n B : ℕ∞)).toNatdef ell (k n : ℕ) : ℝ :=  if k = 3 then Real.log (Real.exp 1 * n) else 1def U (k : ℕ) : ℝ := Real.log 2 / (-Real.log (1 - ((2 : ℝ) ^ k)⁻¹))theorem variance_main (k : ℕ) (hk : 3 ≤ k) :    (∃ C : ℝ, 0 ≤ C ∧ ∀ n : ℕ, k ≤ n →      variance (H : Stream n k → ℝ) (streamLaw n k) ≤ C * n * ell k n) ∧    (∀ B : ℝ, 0 < B → ∃ C : ℝ, 0 ≤ C ∧ ∀ n : ℕ, k ≤ n →      variance (T B : Stream n k → ℝ) (streamLaw n k) ≤ C * n * ell k n) ∧    (∃ c : ℝ, 0 < c ∧      (∃ N : ℕ, ∀ n : ℕ, N ≤ n → k ≤ n →        c * n ≤ variance (H : Stream n k → ℝ) (streamLaw n k)) ∧      (∀ B : ℝ, U k < B → ∃ N : ℕ, ∀ n : ℕ, N ≤ n → k ≤ n →        c * n ≤ variance (T B : Stream n k → ℝ) (streamLaw n k))) := by  sorryend RandomKSATendend OAI

Every extracted statement of the accompanying papers

The statements the atlas extracted automatically from each accompanying paper’s TeX source, with links to the full sources.

A Limiting Satisfiability Threshold for Every Fixed Clause Size

Theorem

For every fixed integer k\ge3, there is a constant \alpha_k\in(0,\infty) such that, for independent proper uniformly signed k-clauses sampled with replacement,

\lim_{n\to\infty}P_n(\lfloor cn\rfloor) = \begin{cases} 1,&0\le c<\alpha_k,\\ 0,&c>\alpha_k. \end{cases}

Section 1, The limiting-threshold problem Read it in the paper TeX source, line 31

Show the TeX

For every fixed integer \(k\ge3\), there is a constant
\(\alpha_k\in(0,\infty)\) such that, for independent proper uniformly
signed \(k\)-clauses sampled with replacement,
\[
 \lim_{n\to\infty}P_n(\lfloor cn\rfloor)
 =
 \begin{cases}
  1,&0\le c<\alpha_k,\\
  0,&c>\alpha_k.
 \end{cases}
\]

Variance of the Random k-SAT Hitting Time

Theorem

Fix an integer k\ge3, and write

\ell_k(n)=\begin{cases}\log(en),&k=3,\\1,&k\ge4,\end{cases} \qquad U_k=\frac{\log2}{-\log(1-2^{-k})}.

There are constants C_k,c_k>0 such that

\Var(H_n)\le C_kn\ell_k(n)\quad(n\ge k),\qquad \Var(H_n)\ge c_kn\quad\text{for all sufficiently large }n.

For every fixed B>0 there is C_{k,B}<\infty such that

\Var(T_{n,B})\le C_{k,B}n\ell_k(n)\quad(n\ge k).

If B>U_k, then also \Var(T_{n,B})\ge c_kn for all sufficiently large n, with the size cutoff allowed to depend on k and B.

Section 1, Fluctuations of the satisfiability transition Read it in the paper TeX source, line 27

Show the TeX

Fix an integer $k\ge3$, and write
\[
 \ell_k(n)=\begin{cases}\log(en),&k=3,\\1,&k\ge4,\end{cases}
 \qquad U_k=\frac{\log2}{-\log(1-2^{-k})}.
\]
There are constants $C_k,c_k>0$ such that
\[
 \Var(H_n)\le C_kn\ell_k(n)\quad(n\ge k),\qquad
 \Var(H_n)\ge c_kn\quad\text{for all sufficiently large }n.
\]
For every fixed $B>0$ there is $C_{k,B}<\infty$ such that
\[
 \Var(T_{n,B})\le C_{k,B}n\ell_k(n)\quad(n\ge k).
\]
If $B>U_k$, then also $\Var(T_{n,B})\ge c_kn$ for all sufficiently
large $n$, with the size cutoff allowed to depend on $k$ and $B$.

Computing the Random 3-SAT Threshold

Theorem

There is a real number \alpha_3\in(0,\infty) such that, for every fixed real a\ge0,

\lim_{n\to\infty}p(n,\lfloor an\rfloor)= \begin{cases}1,&a<\alpha_3,\\0,&a>\alpha_3.\end{cases}

There is one finite deterministic Turing machine which, on input 1^r for any integer r\ge1, halts with a rational number q_r satisfying |q_r-\alpha_3|\le2^{-r}. Its only input is the unary precision 1^r; it uses no oracle, advice, or noncomputable real constant.

Section 1, Introduction Read it in the paper TeX source, line 22

Show the TeX

There is a real number $\alpha_3\in(0,\infty)$ such that, for every fixed
real $a\ge0$,
\[
 \lim_{n\to\infty}p(n,\lfloor an\rfloor)=
 \begin{cases}1,&a<\alpha_3,\\0,&a>\alpha_3.\end{cases}
\]
There is one finite deterministic Turing machine which, on input $1^r$
for any integer $r\ge1$, halts with a rational number $q_r$ satisfying
$|q_r-\alpha_3|\le2^{-r}$. Its only input is the unary precision $1^r$;
it uses no oracle, advice, or noncomputable real constant.

Configuration and file facts

File: lean/ComparatorChallenges/SATVariance.lean
OpenAI’s Lean catalogue: OpenAI’s Lean catalogue does not list this as a main result; the release’s Lean notes for family 235 link it.
Solution module: OAI.Probability.SATVariance.Main
Allowed axioms: propext, Quot.sound, Classical.choiceThe axioms this configuration allows a proof to use. A replay shows which of them a proof uses.
Imports: Mathlib
Imports from outside Mathlib: None
Local vocabulary: 16 local vocabulary declarations
Lines of Lean, without comments or blank lines: 38
Second proof-verification program (nanoda): Off in OpenAI's configuration.

A shareable summary of this page: the proposed paper match, what the Lean statement expresses, and which reviews are recorded.

Plectis Atlas