A Branch-and-Price Framework for Decomposing Graphs into Relaxed Cliques

Abstract

We study the family of problems of partitioning and covering a graph into/with a minimum number of relaxed cliques. Relaxed cliques are subsets of vertices of a graph for which a clique-defining property—for example, the degree of the vertices, the distance between the vertices, the density of the edges, or the connectivity between the vertices—is relaxed. These graph partitioning and covering problems have important applications in many areas such as social network analysis, biology, and disease-spread prevention. We propose a unified framework based on branch-and-price techniques to compute optimal decompositions. For this purpose, new, effective pricing algorithms are developed, and new branching schemes are invented. In extensive computational studies, we compare several algorithmic designs, such as structure-preserving versus dichotomous branching, and their interplay with different pricing algorithms. The final chosen branch-and-price setup produces results that demonstrate the effectiveness of all components of the newly developed framework and the validity of our approach when applied to social network instances.