Primary q=6 hash capacity with one common balanced halving
Provedmme_CW_q6_primary_hash_sqrt_capacity_common_halvingasymmetric-hashingcoppersmith-winogradlaser-methodmatrix-multiplicationpigeonhole-principletensor-restriction
Fix with , and put
for the even lengths . There is a constant such that, for every sufficiently large , one can choose a primary hash family with outer fibers and common inner size whose entries all admit one and the same balanced split of the coupled positions. For
the retained family satisfies
The family-wide common split balances the X word on the first source half and the Y word on the second source half for every retained entry. This is the compatibility needed to realize rows 121 and 211 on the literal paired source without independently reindexing addresses. The exponential rate is identical to the ordinary primary-hash capacity; only the subexponential constant changes.
Preamble
import Mathlib.Analysis.SpecificLimits.Basic import Mathlib.Analysis.SpecialFunctions.Exp import Mathlib.Analysis.SpecialFunctions.Pow.Real import Definitions.Def_mme_CW_q6_common_paired_halving open MME BigOperators Filter set_option autoImplicit false
Formal statement
theorem mme_CW_q6_primary_hash_sqrt_capacity_common_halving
(tau : ℝ) (htau : 2 ≤ 3 * tau) :
∃ C : ℝ, 0 ≤ C ∧
∀ᶠ n : ℕ in atTop,
let N : ℕ := 2 * n
let lambda : ℝ := 2 / ((6 : ℝ) ^ (3 * tau) + 2)
let L : ℕ := ⌊lambda * (N : ℝ)⌋₊
let G : ℕ := N - L
let side : ℕ := 36 ^ (2 * G) * 6 ^ (2 * L)
let raw : ℝ :=
4 * (6 : ℝ) ^ (3 * tau) * ((6 : ℝ) ^ (3 * tau) + 2)
∃ A H : ℕ,
∃ family : CWQ6PrimaryHashFamily N L G A H,
∃ _halving : family.CommonBalancedXYHalving,
raw ^ (2 * N) *
Real.exp (-C * Real.sqrt (((N + 1 : ℕ) : ℝ))) ≤
(((A ^ 3 : ℕ) : ℝ) * (H : ℝ) ^ 2) *
((((side * side * side : ℕ) : ℝ)) ^ tau) := by
sorrySource
Ran Duan, Hongxun Wu, and Renfei Zhou, Faster Matrix Multiplication via Asymmetric Hashing, arXiv:2210.10173v5, Sections 5 and 6.3; the common-halving refinement is the source-faithful paired-row extraction obtained by double-counting balanced bipartitions and absorbing the polynomial thinning loss.