Core Concepts
For any pattern graph H, we determine whether it admits subquadratic algorithms for minimum-weight subgraph, subgraph listing, and subgraph enumeration, and if so, we provide the optimal time complexity.
Abstract
The paper studies the subgraph isomorphism problem, where the pattern graph H is fixed and the task is to find or list all occurrences of H in a given host graph G. The authors focus on the setting where the pattern H is small and the goal is to obtain algorithms with time complexity that is subquadratic in the number of edges m of the host graph.
The key insights are:
The authors introduce the concept of "P-graphs", a class of patterns that can be constructed by stitching together certain basic building blocks along cliques. They show that all patterns with subquadratic complexity can be decomposed into P-graphs.
For each P-graph, the authors define a "savings function" F(α,β,γ) that captures the time complexity of algorithms for that pattern. The overall complexity is then determined by the maximum value of 2-1/F(α,β,γ) over all the P-graph components.
The authors provide fine-grained conditional lower bounds, showing that patterns not decomposable into P-graphs have complexity at least quadratic. For patterns that are decomposable into P-graphs, they design algorithms matching the lower bounds.
The results provide a complete characterization of pattern graphs that admit subquadratic algorithms for minimum-weight subgraph, subgraph listing, and subgraph enumeration, and determine the optimal time complexity for each such pattern.