One zero fine coordinate kills a selected broken-owner word
Provedmme_dwz_broken_owner_selected_three_words_zero_of_coordinateasymmetric-hashingcoordinate-supportmatrix-multiplicationstep-1-zeroing
For one broken owner, if the square-CW fine block of selected X, Y, and Z address words vanishes in one coordinate, then the corresponding three singleton-selected and broken-block-projected tensor map is zero.
Preamble
import Definitions.Def_mme_dwz_step1_projector_basis_api import Definitions.Def_mme_dwz_cw_square_fine_split_grading import Theorems.Thm_mme_dwz_coarseAddress_selected_mode_words_postmap_eq_zero_of_coordinate open MME Module PiTensorProduct open MME.DWZStep1Support open MME.DWZSourceAligned universe u set_option autoImplicit false
Formal statement
theorem mme_dwz_broken_owner_selected_three_words_zero_of_coordinate
{K : Type u} [Field K]
(m : ℕ) {N : ℕ} (outer : Fin N → Fin 15)
(copy : DWZSquare.BrokenBlockCopy
(DWZTable2StandardForm.UsefulBlock m outer))
(x : AddressModeWord outer 0)
(y : AddressModeWord outer 1)
(z : AddressZWord outer)
(hzero : ∃ r : Fin N,
(cwSquareFineSplitGrading K 6).blockTensor
(fun i ↦ ![
MME.DWZStep1Support.fineSplitGrade
(addressModeLeftGrade x r)
(addressModeRightGrade x r),
MME.DWZStep1Support.fineSplitGrade
(addressModeLeftGrade y r)
(addressModeRightGrade y r),
MME.DWZStep1Support.fineSplitGrade
(z r).leftGrade (z r).rightGrade] i) = 0) :
let G := brokenAddressGrading K m outer copy
let sx := DWZComponentRestriction.basisLabelProjection
(coarseAddressModeBasis K outer 0) id {x}
let sy := DWZComponentRestriction.basisLabelProjection
(coarseAddressModeBasis K outer 1) id {y}
let sz := DWZComponentRestriction.basisLabelProjection
(coarseAddressZBasis K outer) id {z}
let selected : ∀ i : Fin 3,
(coarseAddressObj K outer).V i →ₗ[K]
(coarseAddressObj K outer).V i :=
Function.update
(Function.update
(Function.update (fun _ ↦ LinearMap.id) 0 sx) 1 sy) 2 sz
PiTensorProduct.map
(fun i ↦ (G.blockProj i 0).comp (selected i))
(coarseAddressObj K outer).t = 0 := by
sorrySource
Duan--Wu--Zhou, Faster Matrix Multiplication via Asymmetric Hashing, arXiv:2210.10173v5, Section 6, Additional Zeroing-Out Step 1; https://arxiv.org/abs/2210.10173