A zero fine block annihilates the full-square extension of a selected component map
Provedmme_dwz_canonicalComponent_selected_fullMap_zero_of_fine_blockalgebraic-complexitybasis-projectionmatrix-multiplicationtensor-restriction
If the fine block selected by three lifted canonical component labels is zero, then the tensor of the full square is annihilated by the three corresponding component singleton maps extended through their canonical coarse-block projectors, even after arbitrary linear postprocessing.
Preamble
import Definitions.Def_mme_dwz_step1_source_address_projectors import Definitions.Def_mme_dwz_basis_label_projection import Definitions.Def_mme_dwz_cw_square_fine_split_grading open MME Module PiTensorProduct open MME.DWZStep1Support open MME.DWZSourceAligned universe u set_option autoImplicit false
Formal statement
theorem mme_dwz_canonicalComponent_selected_fullMap_zero_of_fine_block
{K : Type u} [Field K] (s : Fin 15)
(selected : ∀ i : Fin 3,
MME.DWZComponentRestriction.LiftedCoarsePair.{u} 6
(cwSquareBlockType
(MME.DWZSquare.shapeX s)
(MME.DWZSquare.shapeY s)
(MME.DWZSquare.shapeZ s) i))
{W : Fin 3 → Type u}
[∀ i, AddCommGroup (W i)] [∀ i, Module K (W i)]
(post : ∀ i,
(MME.DWZComponentRestriction.canonicalComponentBlock K s).V i →ₗ[K]
W i)
(hzero : (cwSquareFineSplitGrading K 6).blockTensor
(fun i ↦ fineSplitGrade
(selected i).leftGrade (selected i).rightGrade) = 0) :
PiTensorProduct.map
(fun i ↦ ((post i).comp
(MME.DWZComponentRestriction.basisLabelProjection
(canonicalComponentModeBasis K s i) id {selected i})).comp
((cwSquareCanonicalGrading K 6).blockProj i
(cwSquareBlockType
(MME.DWZSquare.shapeX s)
(MME.DWZSquare.shapeY s)
(MME.DWZSquare.shapeZ s) i)))
(TensorObj.kron (CWObj K 6) (CWObj K 6)).t = 0 := by
sorry
Source
Duan--Wu--Zhou, Faster Matrix Multiplication via Asymmetric Hashing, Section 6, canonical component and fine-block source decomposition.