Lemma 2.3 — Gram norm and projection sum: ,
ProvedROLDescent.lemma_2_3hilbert-space-operatorsp2o-gran-per-chapterp2o-plan-paperp2o-v1simplicial-complexesspectral-gap
Let be subspaces of a finite-dimensional complex Hilbert space with isometric inclusions , summation map on the orthogonal direct sum, and Gram operator . Then is the orthogonal projection onto , and
This converts the norm of a projection sum into a bound for sums of vectors taken from the subspaces; it is used, with weights, inside the one-step Gram estimate.
Formalization Note The direct sum is PiLp 2 (fun i => U i) over a finite index type; is Mathlib's IsPositive (self-adjoint with nonnegative quadratic form).
Differences from the source:
- encoding: orthogonal direct sum ⊕_i U_i of U_1, …, U_m → PiLp 2 (fun i => U i) over an arbitrary finite index type ι; reason: the ℓ²-product is the orthogonal direct sum; ι plays the role of {1, …, m}, and the empty family (m = 0, S = 0) satisfies every clause trivially
Preamble
import Mathlib import Definitions.Def_ROLDescent_sumMap
Formal statement
namespace ROLDescent
/-- **Lemma 2.3 (Gram norm and projection sum).** Hayakawa, *Recursive Operator Localization and Spectral Gap Descent via the Geometry of Local
Spectral Expansion*, arXiv:2610.23695 (held version, sha256 702bd512…20bdd545), Lemma 2.3, eq. (8), p.10 (PDF 10):
"For the data of Definition 2.2, let E_i = J_i J_i* be the orthogonal projection onto U_i. Then
C ⪰ 0, ‖C‖ = ‖S‖² = ‖Σ_i E_i‖, ‖Σ_i J_i u_i‖² ≤ ‖C‖ Σ_i ‖u_i‖²."
**Formalization Note** `J_i` is `(U i).subtypeL`, `S` is `sumMap U` on `PiLp 2 (fun i => U i)`,
`C = S† S`. The first conjunct is the clause "E_i = J_i J_i* is the orthogonal projection onto U_i". -/
theorem lemma_2_3 {ι : Type*} [Fintype ι]
{H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [FiniteDimensional ℂ H]
(U : ι → Submodule ℂ H) :
(∀ i, (U i).subtypeL ∘L ContinuousLinearMap.adjoint (U i).subtypeL = (U i).starProjection) ∧
(ContinuousLinearMap.adjoint (sumMap U) ∘L sumMap U).IsPositive ∧
‖ContinuousLinearMap.adjoint (sumMap U) ∘L sumMap U‖ = ‖sumMap U‖ ^ 2 ∧
‖sumMap U‖ ^ 2 = ‖∑ i, (U i).starProjection‖ ∧
(∀ u : PiLp 2 (fun i => ↥(U i)),
‖∑ i, ((u i : ↥(U i)) : H)‖ ^ 2 ≤
‖ContinuousLinearMap.adjoint (sumMap U) ∘L sumMap U‖ * ∑ i, ‖u i‖ ^ 2) := 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), Lemma 2.3, eq. (8), p.10 (PDF 10)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.