On Computing Centroids According to the $p$-Norms of Hamming Distance Vectors

TL;DR

This paper investigates the computational complexity of the $p$-Norm Hamming Centroid problem, establishing NP-hardness for all fixed rational $p > 1$, and provides tight lower and upper bounds on algorithmic running times, along with approximation and fixed-parameter algorithms.

Contribution

It proves NP-hardness for all fixed rational $p > 1$, closing the complexity gap, and offers tight bounds, along with practical algorithms for the problem.

Findings

NP-hardness for all fixed rational p > 1

Tight lower bounds on running time based on complexity assumptions

A fixed-parameter and a factor-2 approximation algorithm

Abstract

In this paper we consider the $p$-Norm Hamming Centroid problem which asks to determine whether some given binary strings have a centroid with a bound on the $p$-norm of its Hamming distances to the strings. Specifically, given a set of strings $S$ and a real $k$, we consider the problem of determining whether there exists a string $s^*$ with $\big(\sum_{s \in S}d^p(s^*,s)\big)^{1/p} \leq k$, where $d(,)$ denotes the Hamming distance metric. This problem has important applications in data clustering, and is a generalization of the well-known polynomial-time solvable \textsc{Consensus String} $(p=1)$ problem, as well as the NP-hard \textsc{Closest String} $(p=\infty)$ problem. Our main result shows that the problem is NP-hard for all fixed rational $p > 1$, closing the gap for all rational values of $p$ between $1$ and $\infty$. Under standard complexity assumptions the reduction also implies that the problem has no $2^{o(n+m)}$-time or $2^{o(k^{\frac{p}{(p+1)}})}$-time algorithm, where $m$ denotes the number of input strings and $n$ denotes the length of each string, for any fixed $p > 1$. Both running time lower bounds are tight. In particular, we provide a $2^{k^{\frac{p}{(p+1)}+\varepsilon}}$-time algorithm for each fixed $\varepsilon > 0$. In the last part of the paper, we complement our hardness result by presenting a fixed-parameter algorithm and a factor-$2$ approximation algorithm for the problem.