Project Details
Projekt Print View

Current Challenges in Isomorphism Testing and Graph Canonization

Subject Area Theoretical Computer Science
Term from 2017 to 2024
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 383863674
 
Final Report Year 2024

Final Report Abstract

The computational complexity of the graph isomorphism and canonization problems is one of the most significant open questions in theoretical computer science. Babai’s isomorphism algorithm (2016), a major achievement in this field, raises the challenging question of whether a superpolynomial running time is intrinsic to isomorphism testing or merely a result of the limitations of currently available methods. This project primarily focuses on combinatorial refinement techniques, which are a basic ingredient of Babai’s approach. Our overarching goal is to identify, as precisely as possible, the “part” of the problem tractable by these techniques. It is well known that a relatively simple combinatorial refinement method is an efficient tool for the canonization of almost all graphs. Specifically, if a graph is chosen at random, this method with high probability produces a canonical isomorphic copy of the graph. The success of combinatorial refinement in the average case relies heavily on the fact that a random graph is very likely to be asymmetric. We demonstrate that combinatorial refinement can also be effective for graphs with symmetries (i.e., those admitting nontrivial automorphisms). Our analysis encompasses two distinct cases: sparse random graphs in the evolutional Erdős-Rényi model and vertex-transitive graphs. Another objective of our research is to obtain an efficiently checkable characterization of graphs that are “easy instances” for a specified isomorphism-testing approach. We address this problem for the 2-dimensional Weisfeiler-Leman (WL) algorithm in the context of graphs with small color multiplicity. This case is particularly interesting because it includes the famous Cai-Fürer-Immerman graphs, which present a significant challenge to the WL algorithm at any dimension. A substantial portion of our work is devoted to WL-invariant graph parameters, which are parameters that remain equal for any two graphs indistinguishable by the WL algorithm. The results obtained in this direction are relevant to several other areas, including the design of graph neural networks, chemical graph theory, the analysis of LP-relaxations of NP-hard optimization problems, and combinatorics.

Link to the final report

https://doi.org/10.18452/33141

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung