Prove2Me
Navigate
DiscoverFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Positive integer frames for the compact released regions

Definition
mme_released_positive_integer_frame_data

by BrunoDCDO · Sep 24, 2026 · Mathlib 777aaa6 (Lean v4.29.0-rc3)

assemblyinteger-profilesmatrix-multiplicationmore-asymmetry

A positive integer frame stores the finite geometry and exact integer constraints for one released compact region at scale kkk. Let D=1012D=10^{12}D=1012 and let BBB be that common region's number of parent occurrences. The parent grades and the three integer-count families are fixed to the published compact data:

p=p3,n=kn3,m=km3,μ=kμ3.p=p_3,\qquad n=k n_3,\qquad m=k m_3,\qquad \mu=k\mu_3.p=p3​,n=kn3​,m=km3​,μ=kμ3​.

The frame stores an enumeration of the parent occurrences, an enumeration of their two child positions, and one reference address realizing the split counts mmm. It records exact child masses, grade support, the three boundary-reflection identities, the bound kD2≤n(r)kD^2\le n(r)kD2≤n(r) for every compact label, and divisibility kD2∣m(r,c)kD^2\mid m(r,c)kD2∣m(r,c). The physical enumeration identifies parent grading and every positive-tolerance parent band with the canonical common-coordinate predicates on the same fine word.

The transparent constructor keeps these data unchanged. Given k>0k>0k>0, a repair scale d>1d>1d>1, a tolerance ε>0\varepsilon>0ε>0, and

8d⋅25⋅88⋅92≤kD2ε2,8d\cdot25\cdot88\cdot9^2\le kD^2\varepsilon^2,8d⋅25⋅88⋅92≤kD2ε2,

it builds an ordinary IntegerStep at level two on the canonical common parent band of radius ε\varepsilonε. Its minimum is kD2kD^2kD2, its repair scale is ddd, and it uses the frame's same reference and positions. Existence of a frame is a separate theorem. This definition supplies no output-rate budget, nearby-profile coverage, or recursive matrix-multiplication construction.

Definition code
import Definitions.Def_mme_released_recursive_stage_data
import Definitions.Def_mme_released_joint_interior_frame
import Definitions.Def_mme_graded_integer_regional_step_data
import Mathlib.Tactic.NormNum

set_option autoImplicit false

open scoped BigOperators
open MME MME.RecursiveYZ MME.RegionRealization MME.CompleteSplit
open MME.MoreAsymmetryExactSeed

namespace MME.ReleasedPositiveInteger

/-- Integer geometry over abstract profiles. Keeping these functions as
parameters avoids generating record eliminators over the released tables. -/
structure FrameData {half ell R M B L : ℕ}
    (parent : Fin R → Fin 3 → ℕ) (n : Fin R → ℕ)
    (total : ∀ r, parent r 0 + parent r 1 + parent r 2 = 2 * half)
    (m : ∀ r, RecursiveThinSplit.Split half (parent r) → ℕ)
    (mu : Fin 3 → Cell half R parent → CompleteWord ell → ℕ)
    (length : L * 2 ^ (ell - 1) = M) (minimum : ℕ)
    (grading : ProfiledCW.Predicate M) (source : ℝ → ProfiledCW.Predicate M) where
  hashPositions : Fin (B - 1 + 1) ≃ (r : Fin R) × Fin (n r)
  positions : Fin L ≃ Position n
  reference : RecursiveXHash.Address half R parent n
  reference_target : reference ∈ RecursiveXHash.target m
  mass : ∀ i c, (∑ w, mu i c w) = m c.1 c.2 + m c.1 (complement (total c.1) c.2)
  support : ∀ i c w, 0 < mu i c w →
    ∑ h, (w h).val = (c.2.val i).val
  boundary : BoundaryProfiles mu
  parent_size : ∀ r, minimum ≤ n r
  split_divisible : ∀ r c, minimum ∣ m r c
  parent_graded : ∀ i x,
    ParentGraded parent n i (ProfiledCW.split positions length x) ↔ grading i x
  typical : ∀ (eta : ℝ), 0 < eta → ∀ i x,
    parentTypical total n m (mu i) eta (ProfiledCW.split positions length x) ↔ source eta i x

/-- Construct an integer step without changing the finite geometry. -/
noncomputable def FrameData.step {half ell R M B L : ℕ}
    {parent : Fin R → Fin 3 → ℕ} {n : Fin R → ℕ}
    {total : ∀ r, parent r 0 + parent r 1 + parent r 2 = 2 * half}
    {m : ∀ r, RecursiveThinSplit.Split half (parent r) → ℕ}
    {mu : Fin 3 → Cell half R parent → CompleteWord ell → ℕ}
    {length : L * 2 ^ (ell - 1) = M} {minimum : ℕ}
    {grading : ProfiledCW.Predicate M} {source : ℝ → ProfiledCW.Predicate M}
    (frame : FrameData (B := B) parent n total m mu length minimum grading source)
    (half_eq : half = 2 * 2 ^ (ell - 1)) (minimum_pos : 0 < minimum)
    (repairScale : ℕ) (hrepair : 1 < repairScale)
    (epsilon : ℝ) (hepsilon : 0 < epsilon)
    (hsize : (8 * repairScale : ℝ) *
      (25 * R * (Fintype.card (CompleteWord ell) : ℝ) ^ 2) ≤
        (minimum : ℝ) * epsilon ^ 2) : IntegerStep ell M (source epsilon) where
  half := half
  R := R
  parent := parent
  n := n
  total := total
  half_eq := half_eq
  m := m
  N := B - 1
  hashPositions := frame.hashPositions
  L := L
  positions := frame.positions
  length := length
  mu := mu
  mass := frame.mass
  support := frame.support
  boundary := frame.boundary
  reference := frame.reference
  reference_target := frame.reference_target
  minimum := minimum
  repairScale := repairScale
  minimum_pos := minimum_pos
  repairScale_gt_one := hrepair
  parent_size := frame.parent_size
  split_divisible := frame.split_divisible
  epsilon := epsilon
  epsilon_pos := hepsilon
  size_test := hsize
  source_inside := fun i x hx => (frame.typical epsilon hepsilon i x).mp hx

/-- The compact released specialization keeps all data functions literal.
Its reference is fixed across every step constructed from the frame. -/
abbrev Frame (region : Fin 6) (k : ℕ) :=
  FrameData (half := 4) (ell := 2) (R := 88)
    (M := ReleasedJointInterior.blocks region k * 4)
    (B := ReleasedJointInterior.blocks region k)
    (L := ReleasedJointInterior.blocks region k * 2)
    (RecStage.parent3 region) (fun r => k * RecStage.n3 region r) (RecStage.htotal3 region)
    (fun r c => k * RecStage.m3 region r c) (fun i c w => k * RecStage.mu3 region i c w)
    (ReleasedJointInterior.positions_length region k) (k * denominator ^ 2)
    (fun i x => ParentGraded (ReleasedJointInterior.parent region)
      (ReleasedJointInterior.size region k) i
      (ProfiledCW.split (ell := 2) (ReleasedJointInterior.positions region k)
        (ReleasedJointInterior.positions_length region k) x))
    (ReleasedJointInterior.source region k)

/-- Construct the central integer step with its geometry and profiles unchanged.
The scalar size test is explicit; no copy-rate bound is assumed. -/
noncomputable def Frame.step {region : Fin 6} {k : ℕ}
    (frame : Frame region k) (hk : 0 < k)
    (repairScale : ℕ) (hrepair : 1 < repairScale)
    (epsilon : ℝ) (hepsilon : 0 < epsilon)
    (hsize : (8 * repairScale : ℝ) *
      (25 * 88 * (Fintype.card (CompleteWord 2) : ℝ) ^ 2) ≤
        (k * denominator ^ 2 : ℕ) * epsilon ^ 2) :
    IntegerStep 2 (ReleasedJointInterior.blocks region k * 4)
      (ReleasedJointInterior.source region k epsilon) :=
  FrameData.step frame (by norm_num) (Nat.mul_pos hk (by norm_num [denominator]))
    repairScale hrepair epsilon hepsilon hsize

end MME.ReleasedPositiveInteger
Source
Finite-data assembly interface for the published RecStage tables (p2m:theorem/60610bd3-0675-4be4-a731-ca71c173a5bc, marwahaha), common profiles and position frame (p2m:theorem/3d489536-019f-46c1-b6c3-52ee24f948d0 and p2m:theorem/8bcbc67f-e01e-4bd0-ad82-deb2f4e5b904, Robertboy18), and released exact seed (p2m:theorem/cb80ec03-0b0a-4b6c-a75e-ca788b94d914, raresbuhai). Underlying regional construction: Alman, Duan, Vassilevska Williams, Xu, Xu and Zhou, More Asymmetry Yields Faster Matrix Multiplication, https://arxiv.org/pdf/2404.16349v2, Section 6.1 and Claim 6.5, printed page 32. This interface is not stated verbatim in the paper; its nontrivial existence is a separate theorem.

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