Efficiently Listing Combinatorial Patterns in Graphs

Abstract: Graphs are extremely versatile and ubiquitous mathematical structures with potential to model a wide range of domains. For this reason, graph problems have been of interest since the early days of computer science. Some of these problems consider substructures of a graph that have certain properties. These substructures of interest, generally called patterns, are often meaningful in the domain being modeled. Classic examples of patterns include spanning trees, cycles and subgraphs. This thesis focuses on the topic of explicitly listing all the patterns existing in an input graph. One of the defining features of this problem is that the number of patterns is frequently exponential on the size of the input graph. Thus, the time complexity of listing algorithms is parameterized by the size of the output. The main contribution of this work is the presentation of optimal algorithms for four different problems of listing patterns in graphs, namely the listing of k-subtrees, k-subgraphs, st-paths and cycles. The algorithms presented are framed within the same generic approach, based in a recursive partition of the search space that divides the problem into subproblems. The key to an efficient implementation of this approach is to avoid recursing into subproblems that do not list any patterns. With this goal in sight, a dynamic data structure, called the certificate, is introduced and maintained throughout the recursion. Moreover, properties of the recursion tree and lower bounds on the number of patterns are used to amortize the cost of the algorithm on the size of the output.