Lemma 4.3, pp. 33–34 — exponentially weighted row sums of D = Σ Cⁿ are at most 1/(1 − c)
ProvedLocalPF.Block.lemma_4_3decay-of-correlationsmatrix-seriesp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a finite set, a pseudometric on (real valued, , symmetric, satisfying the triangle inequality), and a matrix with nonnegative entries such that
Then the series converges, and
In particular, for every nonempty and every ,
The lemma turns exponential decay of Dobrushin's matrix into exponential decay of , which is how decay of correlations enters the comparison theorem.
Formalization Note Convergence of is part of the conclusion (as the existence of with HasSum). The case of the last bound is trivial () and is omitted.
Preamble
import Mathlib import Definitions.Def_LocalPF_Block_Setting open MeasureTheory open scoped ENNReal
Formal statement
namespace LocalPF.Block
/-- Lemma 4.3 (pp. 33–34). -/
theorem lemma_4_3 {I : Type*} [Fintype I] [DecidableEq I] (m : I → I → ℝ) (hm_self : ∀ i, m i i = 0)
(hm_symm : ∀ i j, m i j = m j i) (hm_triangle : ∀ i j k, m i k ≤ m i j + m j k)
(C : Matrix I I ℝ) (hC : ∀ i j, 0 ≤ C i j) (c : ℝ) (hc : c < 1)
(hCc : ∀ i, ∑ j, Real.exp (m i j) * C i j ≤ c) :
∃ D : Matrix I I ℝ, HasSum (fun n : ℕ => C ^ n) D ∧
(∀ i, ∑ j, Real.exp (m i j) * D i j ≤ 1 / (1 - c)) ∧
∀ (i : I) (J : Finset I) (hJ : J.Nonempty),
∑ j ∈ J, D i j ≤ Real.exp (-(J.inf' hJ (m i))) / (1 - c) := by sorry
end LocalPF.Block
Source
Rebeschini & van Handel, Can Local Particle Filters Beat the Curse of Dimensionality?, arXiv:1301.6585v2 (reprint of Ann. Appl. Probab. 25(5), 2015), pp. 33–34, Lemma 4.3
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.