Integrated Hamming influence bound for a monotone output map
ProvedStabilityDichotomies.Approximability.integrated_monotone_hamming_influencesconstraint-satisfactionprobability
Let out be a monotone map from finite Boolean occurrence masks to Boolean assignments, with at most B ones in every output, and let 0 ≤ ε ≤ 1. The integral over [1−ε,1] of p times the sum of the Bernoulli-p expected Hamming distances between the outputs with each coordinate forced present and absent is at most B. This is the summed Hamming influence bound in the source sensitivity proof. Applying it to the deleted CSP instance and its actual output law is a separate step.
Preamble
import Mathlib noncomputable section open scoped Classical BigOperators Interval open MeasureTheory
Formal statement
theorem StabilityDichotomies.Approximability.integrated_monotone_hamming_influences {ι V : Type} [Fintype ι] [Fintype V]
(out : (ι → Bool) → V → Bool) (hmono : Monotone out) (B ε : ℝ)
(hbound : ∀ x, (∑ v : V, if out x v then (1 : ℝ) else 0) ≤ B)
(hε0 : 0 ≤ ε) (hε1 : ε ≤ 1) :
(∫ p in (1-ε)..1, p * ∑ i : ι, ∑ x : ι → Bool,
(∏ j : ι, if x j then p else 1-p) *
(hammingDist (out (Function.update x i true)) (out (Function.update x i false)) : ℝ)) ≤ B := by sorrySource
Wang and Yoshida, Stability Dichotomies for Boolean Constraint Satisfaction Problems, STOC 2027 submission #535, proof of Theorem 4.3, p. 17, https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/535.pdf