Exact Distributed Sampling

Fast distributed algorithms that output a feasible solution for constraint satisfaction problems, such as maximal independent sets, have been heavily studied. There has been much less research on distributed sampling problems, where one wants to sample from a distribution over all feasible solutions (e.g., uniformly sampling a feasible solution). Recent work (Feng, Sun, Yin PODC 2017; Fischer and Ghaffari DISC 2018; Feng, Hayes, and Yin arXiv 2018) has shown that for some constraint satisfaction problems there are distributed Markov chains that mix in $O(\log n)$ rounds in the classical LOCAL model of distributed computation. However, these methods return samples from a distribution close to the desired distribution, but with some small amount of error. In this paper, we focus on the problem of exact distributed sampling. Our main contribution is to show that these distributed Markov chains in tandem with techniques from the sequential setting, namely coupling from the past and bounding chains, can be used to design $O(\log n)$-round LOCAL model exact sampling algorithms for a class of weighted local constraint satisfaction problems. This general result leads to $O(\log n)$-round exact sampling algorithms that use small messages (i.e., run in the CONGEST model) and polynomial-time local computation for some important special cases, such as sampling weighted independent sets (aka the hardcore model) and weighted dominating sets.