Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Definitions bundle OAIQ9Tr001 of the split of OpenAI, Quantitative lower bounds for trace reconstruction

Definition
OAIQ9Tr001

by dbenbenn · Oct 26, 2026 · Mathlib 0df444a (Lean v4.33.1)

openai-math

A definitions bundle of the split of OpenAI's proof of OAI.TraceReconstruction.quantitative_sample_lower_bound, OAI.TraceReconstruction.superpolynomial_one_trace and OAI.TraceReconstruction.superpolynomial_sample_complexity, carved from OpenAI's openai/math development (OpenAI, Quantitative lower bounds for trace reconstruction, OpenAI Math Release, September 24, 2026, result family 122, github.com/openai/math at commit adc7f12, Apache License 2.0) and ported to Lean 4.33.1 and Mathlib 0df444a. The declarations and their proofs are OpenAI's.

Contents. It holds 15 declarations of OpenAI's Lean: 7 definitions, 4 instances and 4 theorems. The definitions and types are OAI.TraceReconstruction.sourceDepth, OAI.TraceReconstruction.sourceFitTail, OAI.TraceReconstruction.sourceGapFloor, OAI.TraceReconstruction.sourceInitialOffset, OAI.TraceReconstruction.sourceInitialWeight, OAI.TraceReconstruction.sourceScaleInverseEnvelope, OAI.TraceReconstruction.workingDeletion.

Sources. They come from these modules of lean/OAI/Probability/TraceReconstruction, in whole or in part: TraceReconstruction_ErrorRates, TraceReconstruction_InitialRates, TraceReconstruction_Metadata, TraceReconstruction_Propagation, TraceReconstruction_Scales.

Imports. It imports no bundle or statement, only Mathlib.

Imported by (directly, in this split): the bundle OAIQ9Tr002; the statement OAI.TraceReconstruction.source_all_levels_eventually.

Definition code
import Mathlib

section
-- module Solutions.OAIQ9_TR.Compat
namespace OAI

/-! Compatibility with the newer Lean core OpenAI's code was written against: lemmas it uses by
names this environment (Lean 4.33.1, Mathlib 0df444a) does not have yet. Each is the
corresponding `if_pos` / `if_neg` / `dif_pos` / `dif_neg`. The `dite` forms are stated as
`dite p a b`, not with the `if hp : p then …` notation: its eta-expanded lambdas form is a higher-order pattern
that `rw` cannot find (DepthThree, PeriodicTiling 2026-10-25). -/
section OAICompat434
/-- Mathlib's newer `TensorProduct.inductionOn` (no `zero` case: it is the `tmul` case at
`0 ⊗ₜ 0`); ours has only `TensorProduct.induction_on`, which asks for one. Not named
`TensorProduct.inductionOn`: inside `namespace OAI` that would make `open TensorProduct` open
`OAI.TensorProduct` instead of Mathlib's. -/
@[elab_as_elim]
theorem tensorInductionOn {R M N : Type*} [CommSemiring R] [AddCommMonoid M] [AddCommMonoid N]
    [Module R M] [Module R N] {motive : TensorProduct R M N → Prop} (z : TensorProduct R M N)
    (tmul : ∀ x y, motive (TensorProduct.tmul R x y))
    (add : ∀ x y, motive x → motive y → motive (x + y)) : motive z :=
  TensorProduct.induction_on z (by simpa using tmul 0 0) tmul add
/-- Mathlib's newer case names for `FreeGroup.induction_on` (`one`, where ours says `C1`). -/
@[elab_as_elim]
theorem freeGroupInductionOn {α : Type*} {C : FreeGroup α → Prop} (z : FreeGroup α) (one : C 1)
    (of : ∀ x, C (FreeGroup.of x)) (inv_of : ∀ x, C (FreeGroup.of x) → C (FreeGroup.of x)⁻¹)
    (mul : ∀ x y, C x → C y → C (x * y)) : C z :=
  FreeGroup.induction_on z one of inv_of mul
/-- The first projection of a probability measure on a product: found by instance search in d13f23b
(total `Measure.map`), here a measurable map. -/
instance isProbabilityMeasure_map_fst {α β : Type*} [MeasurableSpace α] [MeasurableSpace β]
    {μ : MeasureTheory.Measure (α × β)} [MeasureTheory.IsProbabilityMeasure μ] :
    MeasureTheory.IsProbabilityMeasure (μ.map Prod.fst) :=
  MeasureTheory.Measure.isProbabilityMeasure_map measurable_fst.aemeasurable
instance isProbabilityMeasure_map_snd {α β : Type*} [MeasurableSpace α] [MeasurableSpace β]
    {μ : MeasureTheory.Measure (α × β)} [MeasureTheory.IsProbabilityMeasure μ] :
    MeasureTheory.IsProbabilityMeasure (μ.map Prod.snd) :=
  MeasureTheory.Measure.isProbabilityMeasure_map measurable_snd.aemeasurable
instance isProbabilityMeasure_map_subtype_val {α : Type*} [MeasurableSpace α] {p : α → Prop}
    {μ : MeasureTheory.Measure {a // p a}} [MeasureTheory.IsProbabilityMeasure μ] :
    MeasureTheory.IsProbabilityMeasure (μ.map Subtype.val) :=
  MeasureTheory.Measure.isProbabilityMeasure_map measurable_subtype_coe.aemeasurable
/-- Mathlib d13f23b's `YoungDiagram.card_transpose`. -/
@[simp]
theorem _root_.YoungDiagram.card_transpose (μ : YoungDiagram) : μ.transpose.card = μ.card := by
  unfold YoungDiagram.card
  apply Finset.card_bij (fun c _ => c.swap)
  · intro c hc
    have : c ∈ μ.transpose := hc
    rw [YoungDiagram.mem_transpose] at this
    exact this
  · intro a _ b _ h
    simpa using h
  · intro b hb
    refine ⟨b.swap, ?_, by simp⟩
    show b.swap ∈ μ.transpose
    rw [YoungDiagram.mem_transpose]
    simpa using hb
/-- For a measurable equivalence `e`, `IsProbabilityMeasure (μ.map e)` is found by instance search
(Mathlib d13f23b finds it for every map `f`: there `Measure.map` of a non-measurable map is a Dirac
mass, here it is `0`, so only the measurable case can be shimmed). -/
instance isProbabilityMeasure_map_measurableEquiv {α β : Type*} [MeasurableSpace α]
    [MeasurableSpace β] {μ : MeasureTheory.Measure α} [MeasureTheory.IsProbabilityMeasure μ]
    (e : α ≃ᵐ β) : MeasureTheory.IsProbabilityMeasure (μ.map e) :=
  MeasureTheory.Measure.isProbabilityMeasure_map e.measurable.aemeasurable
end OAICompat434

end OAI
end

section
-- module Solutions.OAIQ9_TR.TraceReconstruction_Scales
set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

/-!
# Scales

Exact block exponents, deletion reduction, induction depth, backward target scales, and uniform power absorption.
-/

open scoped ENNReal
noncomputable section

open Filter
open scoped Topology


open Filter
open scoped Topology
namespace TraceReconstruction

def workingDeletion (q : ℝ) : ℝ := min q (1/2)
def sourceDepth (B : ℕ) (ζ H : ℝ) : ℕ := ⌊(1-ζ)*Real.log H/((B:ℝ)*Real.log 2)⌋₊

end TraceReconstruction

open Filter
open scoped Topology


end

end OAI
end

section
-- module Solutions.OAIQ9_TR.TraceReconstruction_Metadata
set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

/-!
# Metadata

Hidden metadata, fair jitter, heterogeneous emissions, independent window trials, concentration, and fit.
-/

open scoped ENNReal
noncomputable section

open Filter
open scoped Topology



noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section
namespace TraceReconstruction
open scoped Classical

/-- Explicit finite tree union bound. The source's scale estimates will make
 this number exponentially small; no independence of sibling shifts is used. -/
def sourceFitTail (T n : ℕ → ℕ) : ℕ → ℝ
  | 0 => 0
  | 1 => 0
  | b+2 => (T (b+1):ℝ)*(2*Real.exp (-2*((T b:ℝ)/8)^2/(((T (b+1):ℝ)+1)*(n b:ℝ)))+
      sourceFitTail T n (b+1))

end TraceReconstruction

end
end
end
end
end
end
end

end OAI
end

section
-- module Solutions.OAIQ9_TR.TraceReconstruction_InitialRates
set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

/-!
# InitialRates

Initial geometric parameters, moment depths, error and overlap rates, and production of the base induction pair.
-/

open scoped ENNReal
noncomputable section

open Filter
open scoped Topology


noncomputable section


noncomputable section


noncomputable section
namespace TraceReconstruction
open Filter
open scoped Topology Classical

/-- The manuscript's one-based middle defect, translated to a list index. -/
def sourceInitialOffset (m : ℕ) (hm : 4 ≤ m) : Fin m :=
  ⟨m/2-1,by omega⟩

theorem sourceInitialOffset_interior (m : ℕ) (hm : 4 ≤ m) :
    1 ≤ (sourceInitialOffset m hm).val ∧ (sourceInitialOffset m hm).val+1 < m := by
  simp only [sourceInitialOffset]; omega

end TraceReconstruction


noncomputable section


noncomputable section


noncomputable section
namespace TraceReconstruction
open Filter
open scoped Topology Classical

/-- Actual primitive-score weight at the base scale. -/
def sourceInitialWeight (q : ℝ) (k : ℕ) : ℝ := (6*(2*k:ℕ):ℝ)*(2/q)

end TraceReconstruction


end
end
end
end
end
end
end

end OAI
end

section
-- module Solutions.OAIQ9_TR.TraceReconstruction_ErrorRates
set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

/-!
# ErrorRates

Realization of block caps, sparse scales, sharp and smoothing budgets, tree tails, fit errors, and uniform polynomial/exponential bounds.
-/

open scoped ENNReal
noncomputable section

open Filter
open scoped Topology


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section


noncomputable section
namespace TraceReconstruction
open Filter
open scoped Topology Classical

def sourceScaleInverseEnvelope (B D : ℕ) : ℝ := 2*(2:ℝ)^(B*D)

end TraceReconstruction


end
end
end
end
end
end
end
end
end
end
end
end

end OAI
end

section
-- module Solutions.OAIQ9_TR.TraceReconstruction_Propagation
set_option linter.all false
set_option linter.unusedVariables false
set_option linter.unusedTactic false
set_option linter.unreachableTactic false

namespace OAI

/-!
# Propagation

Moment restriction, scale margins, sharp budget closure, clipping and error propagation, and dense/warmup overlap.
-/

open scoped ENNReal
noncomputable section

open Filter
open scoped Topology


noncomputable section


noncomputable section
namespace TraceReconstruction
open scoped Classical

def sourceGapFloor (B : ℕ) : ℝ := 1/(128*(2:ℝ)^(2*B))

end TraceReconstruction

end
end
end

end OAI
end
Source
OpenAI, Quantitative lower bounds for trace reconstruction, OpenAI Math Release, September 24, 2026, https://github.com/openai/math (result family 122; paper: https://github.com/openai/math/blob/adc7f1241b42e322a6451854ab7e4b4c146bf78a/preprints/quantitative-lower-bounds-for-trace-reconstruction-September-24-2026/paper.pdf; Apache License 2.0); definitions carved from lean/OAI/Probability/TraceReconstruction at commit adc7f12

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me