Parameterized Inapproximability of Exact Cover and Nearest Codeword

Abstract

The $k$-ExactCover problem is a parameterized version of the ExactCover problem, in which we are given a universe $U$, a collection $S$ of subsets of $U$, and an integer $k$, and the task is to determine whether $U$ can be partitioned into $k$ sets in $S$. This is a natural extension of the well-studied SetCover problem; though in the parameterized regime we know it to be $W[1]$-complete in the exact case, its parameterized complexity with respect to approximability is not well understood. We prove that, assuming ETH, for some $\gamma > 0$ there is no time $f(k) \cdot N^{\gamma k}$ algorithm that can, given a $k$-ExactCover instance $I$, distinguish between the case where $I$ has an exact cover of size $k$ and the case where every set cover of $I$ has size at least $\frac14 \sqrt[k]{\frac{\log N}{\log \log N}}$. This rules out even more than FPT algorithms, and additionally rules out any algorithm whose approximation ratio depends only on the parameter $k$. By assuming SETH, we instead improve the lower bound to requiring time $f(k) \cdot N^{k - \varepsilon}$, for any $\varepsilon > 0$. In this work we also extend the inapproximability result to the $k$-Nearest-Codeword ($k$-NCP) problem. Specifically, given a generator matrix $A \in \mathbb{F}_2^{m \times n}$, a vector $y \in \mathbb{F}_2^m$, and the parameter $k$, we show that it is hard to distinguish between the case where there exists a codeword with distance at most $k$ from $y$ and the case where every codeword has distance at least $\frac18 \sqrt[k]{\frac{\log N}{\log \log N}}$ from $y$. This improves the best known parameterized inapproximability result, which rules out approximations with a factor of $\text{poly} (\log k)$, but requires us to assume ETH instead of $W[1] \neq FPT$.