Year: 2026
Type: preprint
Abstract: Weighted $k$-set packing asks for a maximum-weight collection of pairwise disjoint sets, each of cardinality at most $k$. For every fixed $k\in\mathbb Z_{\geq1}$, we give a polynomial-time algorithm with approximation factor $\left(1- 1/{\sqrt3} \right)k+O(\log k) \approx 0.4226k + O(\log k)$, improving the previous best factor of $0.4621k + O(\sqrt{k})$ through a simple two-phase approach. In a first phase, our algorithm constructs a packing by processing random weight buckets of varying widths with an unweighted packing algorithm. We show that a small expected weight of this packing relative to the optimum forces a large expected total surplus in the optimum. Surplus quantifies the potential for improving a packing through star swaps. In the second phase, we solve and round a configuration LP to turn sufficiently large surplus into an increase in expected weight. Our procedure can easily be derandomized.
Source: arXiv (Cornell University)
Authors: Taziki, Mehrshad, Zenklusen, Rico
Institution: ETH Zurich
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7235298082
