Closing the gap for pseudo-polynomial strip packing

TL;DR

This paper presents a new pseudo-polynomial time algorithm for strip packing that achieves an approximation ratio of (5/4 + ε), closing the gap between the known inapproximability bound and the best previous algorithms.

Contribution

It introduces a structural approach that transforms optimal solutions into a limited set of forms, enabling improved approximation algorithms for strip packing and related problems.

Findings

Achieves an approximation ratio of (5/4 + ε) for pseudo-polynomial strip packing.

Provides a structural theorem applicable to multiple problem variants.

Enables algorithms with improved ratios for related scheduling problems.

Abstract

The set of 2-dimensional packing problems builds an important class of optimization problems and Strip Packing together with 2-dimensional Bin Packing and 2-dimensional Knapsack is one of the most famous of these problems. Given a set of rectangular axis parallel items and a strip with bounded width and infinite height the objective is to find a packing of the items into the strip which minimizes the packing height. We speak of pseudo-polynomial Strip Packing if we consider algorithms with pseudo-polynomial running time with respect to the width of the strip. It is known that there is no pseudo-polynomial algorithm for Strip Packing with a ratio better than $5/4$ unless $\mathrm{P} = \mathrm{NP}$. The best algorithm so far has a ratio of $(4/3 + \varepsilon)$. In this paper, we close this gap between inapproximability result and best known algorithm by presenting an algorithm with approximation ratio $(5/4 + \varepsilon)$ and thus categorize the problem accurately. The algorithm uses a structural result which states that each optimal solution can be transformed such that it has one of a polynomial number of different forms. The strength of this structural result is that it applies to other problem settings as well for example to Strip Packing with rotations (90 degrees) and Contiguous Moldable Task Scheduling. This fact enabled us to present algorithms with approximation ratio $(5/4 + \varepsilon)$ for these problems as well.