Faster Randomized Branching Algorithms for $r$-SAT

TL;DR

This paper introduces faster randomized branching algorithms for $r$-SAT and related problems, improving upon previous deterministic methods and matching or surpassing existing randomized algorithms in efficiency.

Contribution

It presents a simple randomized branching algorithm for $r$-SAT with improved running time, and extends these techniques to parameterized hitting set and vertex cover problems.

Findings

Randomized $r$-SAT algorithm runs in $O^*({(rac{r+1}{2})}^d)$ time.

Achieves faster algorithms for hitting set and vertex cover problems.

Matches or improves upon the best known randomized algorithms for these problems.

Abstract

The problem of determining if an $r$-CNF boolean formula $F$ over $n$ variables is satisifiable reduces to the problem of determining if $F$ has a satisfying assignment with a Hamming distance of at most $d$ from a fixed assignment $\alpha$. This problem is also a very important subproblem in Schoning's local search algorithm for $r$-SAT. While Schoning described a randomized algorithm solves this subproblem in $O((r-1)^d)$ time, Dantsin et al. presented a deterministic branching algorithm with $O^*(r^d)$ running time. In this paper we present a simple randomized branching algorithm that runs in time $O^*({(\frac{r+1}{2})}^d)$. As a consequence we get a randomized algorithm for $r$-SAT that runs in $O^*({(\frac{2(r+1)}{r+3})}^n)$ time. This algorithm matches the running time of Schoning's algorithm for 3-SAT and is an improvement over Schoning's algorithm for all $r \geq 4$. For $r$-uniform hitting set parameterized by solution size $k$, we describe a randomized FPT algorithm with a running time of $O^*({(\frac{r+1}{2})}^k)$. For the above LP guarantee parameterization of vertex cover, we have a randomized FPT algorithm to find a vertex cover of size $k$ in a running time of $O^*(2.25^{k-vc^*})$, where $vc^*$ is the LP optimum of the natural LP relaxation of vertex cover. In both the cases, these randomized algorithms have a better running time than the current best deterministic algorithms.