Efficient Algorithms for Partitioning Circulant Graphs with Optimal Spectral Approximation

The Marcus-Spielman-Srivastava theorem (Annals of Mathematics, 2015) for the Kadison-Singer conjecture implies the following result in spectral graph theory: For any undirected graph $G = (V,E)$ with a maximum edge effective resistance at most $α$, there exists a partition of its edge set $E$ into $E_1 \cup E_2$ such that the two edge-induced subgraphs of $G$ spectrally approximates $(1/2)G$ with a relative error $O(\sqrtα)$. However, the proof of this theorem is non-constructive. It remains an open question whether such a partition can be found in polynomial time, even for special classes of graphs. In this paper, we explore polynomial-time algorithms for partitioning circulant graphs via partitioning their generators. We develop an efficient algorithm that partitions a circulant graph whose generators form an arithmetic progression, with an error matching that in the Marcus-Spielman-Srivastava theorem and optimal, up to a constant. On the other hand, we prove that if the generators of a circulant graph are ``far" from an arithmetic progression, no partition of the generators can yield two circulant subgraphs with an error matching that in the Marcus-Spielman-Srivastava theorem. In addition, we extend our algorithm to Cayley graphs whose generators are from a product of multiple arithmetic progressions.