ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Since each step can be implemented efficiently, the fundamental question is how long to simulate the chain. We say µ is η-spectrally independent if \lambdamax (Ψµ ) \leq 1 + η. Note that Ψµ = Dµ−1 Σµ , where Σµ (i, j) = Cov\sigma\simµ (\sigmai , \sigmaj ) is the usual covari- ance matrix, and Dµ is a diagonal matrix with entries given by the variance of each coordinate; in particular, the eigenvalues of Ψµ are all real. If µ is a product measure (e.g. uniform over {\pm1}n), then µ is 0-spectrally independent. At the other extreme, if µ = 12 \delta+1 + 12 \delta−1 , then Ψµ = 11⊤ and so µ is (n − 1)-spectrally independent. This property of the distribution µ was originally distilled from the recently emerging theory of high-dimensional expanders. We have the following local-to-global theorems connecting spectral independence with the theory of mix- ing times. Theorem 1 (Informal; [1, 3] building on [25, 21, 26]). Suppose there exists η \leq O(1) such that for every S \subseteq [n] with |S| \leq n − 2 and every partial assignment \tau : S \to {\pm1}, the distribution of \sigma \sim µ conditioned on agreeing with \tau on S is η-spectrally independent. Then Glauber dynamics mixes in O(n2+η )-steps. If we impose an additional graphical assumption on the structure of µ, then we can improve the mixing time to the optimal O(n log n). More specifically, we say µ sat

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.