Computing the k Densest Subgraphs of a Graph

TL;DR

This paper explores the computational complexity of finding multiple densest subgraphs in a graph under various overlap constraints, revealing NP-hardness and tractability results for different scenarios.

Contribution

It analyzes three variants of the k densest subgraphs problem, providing complexity classifications and algorithms for each case.

Findings

NP-hard for no overlap when k >= 3

Fixed-parameter tractable for unrestricted overlap

PTAS available for constant k with no overlap restrictions

Abstract

Computing cohesive subgraphs is a central problem in graph theory. While many formulations of cohesive subgraphs lead to NP-hard problems, finding a densest subgraph can be done in polynomial time. As such, the densest subgraph model has emerged as the most popular notion of cohesiveness. Recently, the data mining community has started looking into the problem of computing k densest subgraphs in a given graph, rather than one, with various restrictions on the possible overlap between the subgraphs. However, there seems to be very little known on this important and natural generalization from a theoretical perspective. In this paper we hope to remedy this situation by analyzing three natural variants of the k densest subgraphs problem. Each variant differs depending on the amount of overlap that is allowed between the subgraphs. In one extreme, when no overlap is allowed, we prove that the problem is NP-hard for k >= 3. On the other extreme, when overlap is allowed without any restrictions and the solution subgraphs only have to be distinct, we show that the problem is fixed-parameter tractable with respect to k, and admits a PTAS for constant k. Finally, when a limited of overlap is allowed between the subgraphs, we prove that the problem is NP-hard for k = 2.