Eq. (70) — degree identities; is a probability and a reversible Markov matrix
ProvedROLDescent.eq_70hilbert-space-operatorsp2o-gran-per-chapterp2o-plan-paperp2o-v1simplicial-complexesspectral-gap
Let be a weighted pure complex with facet size and let be an internal face (), with link vertices and remaining rank . Then:
- ;
- for every , ;
- for and ;
- for every , for and ;
- for distinct ,
- for all (reversibility).
These identities make a reversible Markov chain with stationary law , which is what allows the Markov-chain spectral gap to be used in the descent.
Differences from the source:
- reading: Σ_i W_τi and Σ_{j≠i} W_τij (index range not printed) → sums over the link vertices N_τ (and N_τ ∖ {i}); reason: the walk lives on N_τ; for j ∈ τ the literal set τij equals τi, so a sum over all of V is not what is meant
- encoding: π_τ is a probability distribution and P_τ is a Markov matrix → nonnegativity and sums equal to 1 on N_τ, at faces with τ.card + 2 ≤ d; reason: the definitions; π_τ and P_τ are defined at internal nodes
Preamble
import Mathlib import Definitions.Def_ROLDescent_WeightedComplex import Definitions.Def_ROLDescent_facetMass import Definitions.Def_HDXConc_linkWeight import Definitions.Def_ROLDescent_remRank import Definitions.Def_ROLDescent_outProb import Definitions.Def_ROLDescent_linkWalk
Formal statement
namespace ROLDescent
/-- **Eq. (70): degree identities, π_τ a probability, P_τ a reversible Markov matrix.**
Hayakawa, *Recursive Operator Localization and Spectral Gap Descent via the Geometry of Local
Spectral Expansion*, arXiv:2610.23695 (held version, sha256 702bd512…20bdd545), Section 5.1.1, eq. (70), p.30 (PDF 30): "Counting the remaining vertices in each facet
gives Σ_i W_τi = m_τ W_τ, Σ_{j≠i} W_τij = (m_τ − 1) W_τi. Consequently π_τ is a probability
distribution and P_τ is a Markov matrix. It is reversible, since
π_τ(i) P_τ(i, j) = W_τij / (m_τ(m_τ − 1) W_τ) = π_τ(j) P_τ(j, i)."
**Formalization Note** At an internal face (τ.card + 2 ≤ d, i.e. m_τ ≥ 2). The sums range over the link
vertices `N_τ` (the sum over j ≠ i is over `N_τ \ {i}`; other vertices contribute `W_τij = 0` or are
in `τ`). The middle formula of the reversibility display is stated for `i ≠ j`; the reversibility
identity itself for all `i, j ∈ N_τ`. -/
theorem eq_70 {V : Type*} [Fintype V] [DecidableEq V]
(X : Finset (Finset V)) (d : ℕ) (w : Finset V → ℝ) (hX : WeightedComplex X d w)
(τ : Finset V) (hτ : τ ∈ X) (hτd : τ.card + 2 ≤ d) :
(∑ i ∈ HDXConc.linkVerts X τ, facetMass X d w (insert i τ) =
(remRank d τ : ℝ) * facetMass X d w τ) ∧
(∀ i ∈ HDXConc.linkVerts X τ, ∑ j ∈ (HDXConc.linkVerts X τ).erase i, facetMass X d w (insert j (insert i τ)) =
((remRank d τ : ℝ) - 1) * facetMass X d w (insert i τ)) ∧
((∀ i ∈ HDXConc.linkVerts X τ, 0 ≤ outProb X d w τ i) ∧
∑ i ∈ HDXConc.linkVerts X τ, outProb X d w τ i = 1) ∧
(∀ i ∈ HDXConc.linkVerts X τ, (∀ j ∈ HDXConc.linkVerts X τ, 0 ≤ linkWalk X d w τ i j) ∧
∑ j ∈ HDXConc.linkVerts X τ, linkWalk X d w τ i j = 1) ∧
(∀ i ∈ HDXConc.linkVerts X τ, ∀ j ∈ HDXConc.linkVerts X τ, i ≠ j →
outProb X d w τ i * linkWalk X d w τ i j =
facetMass X d w (insert j (insert i τ)) /
((remRank d τ : ℝ) * ((remRank d τ : ℝ) - 1) * facetMass X d w τ)) ∧
(∀ i ∈ HDXConc.linkVerts X τ, ∀ j ∈ HDXConc.linkVerts X τ,
outProb X d w τ i * linkWalk X d w τ i j = outProb X d w τ j * linkWalk X d w τ j i) := by sorry
end ROLDescent
Source
Hayakawa, Recursive Operator Localization and Spectral Gap Descent via the Geometry of Local Spectral Expansion, arXiv:2610.23695 (held version, sha256 702bd512…20bdd545), Section 5.1.1, eq. (70), p.30 (PDF 30)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.