The Limits of Local Search for the Maximum Weight Independent Set Problem in d-Claw Free Graphs

TL;DR

This paper investigates the limitations of local search algorithms for the Maximum Weight Independent Set problem in d-claw free graphs, establishing new approximation bounds and demonstrating inherent barriers of local improvements.

Contribution

It provides asymptotic approximation ratios for local search in d-claw free graphs and shows fundamental limits of local improvements for weighted k-Set Packing.

Findings

Approximation ratio of (d-1+ε_d)/2 for d-claw free graphs with local search

Polynomial time (k+ε_{k+1})/2-approximation for weighted k-Set Packing

Examples showing no local improvement algorithm can surpass k/2 approximation for k-Set Packing

Abstract

We consider the Maximum Weight Independent Set Problem (MWIS) in $d$-claw free graphs, i.e. the task of computing an independent set of maximum weight in a given $d$-claw free graph $G=(V,E)$ equipped with a positive weight function $w:V\rightarrow\mathbb{R}_{>0}$. For $k\geq 1$, the MWIS in $k+1$-claw free graphs generalizes the weighted $k$-Set Packing Problem. Given that for $k\geq 3$, this problem does not permit a polynomial time $o(\frac{k}{\log k})$-approximation unless $P=NP$, most previous algorithms for both weighted $k$-Set Packing and the MWIS in $d$-claw free graphs rely on local search. For the last twenty years, Berman's algorithm SquareImp, which yields a $\frac{d}{2}+\epsilon$-approximation for the MWIS in $d$-claw free graphs, has remained unchallenged for both problems. Recently, it was improved by Neuwohner, obtaining an approximation guarantee slightly below $\frac{d}{2}$, and inevitably raising the question of how far one can get by using local search. In this paper, we finally answer this question asymptotically in the following sense: By considering local improvements of logarithmic size, we obtain approximation ratios of $\frac{d-1+\epsilon_d}{2}$ for the MWIS in $d$-claw free graphs for $d\geq 3$ in quasi-polynomial time, where $0\leq \epsilon_d\leq 1$ and $\lim_{d\rightarrow\infty}\epsilon_d = 0$. By employing the color coding technique, we can use the previous result to obtain a polynomial time $\frac{k+\epsilon_{k+1}}{2}$-approximation for weighted $k$-Set Packing. On the other hand, we provide examples showing that no local improvement algorithm considering local improvements of size $\mathcal{O}(\log(|\mathcal{S}|))$ with respect to some power $w^\alpha$ of the weight function, where $\alpha\in\mathbb{R}$ is chosen arbitrarily, but fixed, can yield an approximation guarantee better than $\frac{k}{2}$ for the weighted $k$-Set Packing Problem with $k\geq 3$.