eng
Schloss Dagstuhl – Leibniz-Zentrum für Informatik
Leibniz International Proceedings in Informatics
1868-8969
2020-08-11
3:1
3:21
10.4230/LIPIcs.APPROX/RANDOM.2020.3
article
Iterated Decomposition of Biased Permutations via New Bounds on the Spectral Gap of Markov Chains
Miracle, Sarah
1
Streib, Amanda Pascoe
2
Streib, Noah
2
University of St. Thomas, St. Paul, MN, USA
Center for Computing Sciences, Bowie, MD, USA
In this paper, we address a conjecture of Fill [Fill03] about the spectral gap of a nearest-neighbor transposition Markov chain ℳ_nn over biased permutations of [n]. Suppose we are given a set of input probabilities 𝒫 = {p_{i,j}} for all 1 ≤ i, j ≤ n with p_{i, j} = 1-p_{j, i}. The Markov chain ℳ_nn operates by uniformly choosing a pair of adjacent elements, i and j, and putting i ahead of j with probability p_{i,j} and j ahead of i with probability p_{j,i}, independent of their current ordering.
We build on previous work [S. Miracle and A.P. Streib, 2018] that analyzed the spectral gap of ℳ_nn when the particles in [n] fall into k classes. There, the authors iteratively decomposed ℳ_nn into simpler chains, but incurred a multiplicative penalty of n^-2 for each application of the decomposition theorem of [Martin and Randall, 2000], leading to an exponentially small lower bound on the gap. We make progress by introducing a new complementary decomposition theorem. We introduce the notion of ε-orthogonality, and show that for ε-orthogonal chains, the complementary decomposition theorem may be iterated O(1/√ε) times while only giving away a constant multiplicative factor on the overall spectral gap. We show the decomposition given in [S. Miracle and A.P. Streib, 2018] of a related Markov chain ℳ_pp over k-class particle systems is 1/n²-orthogonal when the number of particles in each class is at least C log n, where C is a constant not depending on n. We then apply the complementary decomposition theorem iteratively n times to prove nearly optimal bounds on the spectral gap of ℳ_pp and to further prove the first inverse-polynomial bound on the spectral gap of ℳ_nn when k is as large as Θ(n/log n). The previous best known bound assumed k was at most a constant.
https://drops.dagstuhl.de/storage/00lipics/lipics-vol176-approx-random2020/LIPIcs.APPROX-RANDOM.2020.3/LIPIcs.APPROX-RANDOM.2020.3.pdf
Markov chains
Permutations
Decomposition
Spectral Gap
Iterated Decomposition