Blazing a Trail via Matrix Multiplications: A Faster Algorithm for Non-shortest Induced Paths
Yung-Chung Chiu, Hsueh-I Lu
TL;DR
The paper addresses the problem of finding non-shortest induced $uv$-paths (uv-trails) in undirected graphs, building on a prior $O(n^{18})$-time algorithm. It introduces a guess-and-verify framework based on $uv$-straight graphs, wings, and winged quadruples, and then exploits poly-logarithmic matrix multiplications and a witness-matrix data structure to dramatically accelerate preprocessing. The main contribution is an $O(n^{4.75})$-time algorithm achieved through $O(n^ ext{ω})$ preprocessing to obtain a $uv$-straight instance, a $ ilde{O}(n^{2 ext{ω}})$ winged-quadruple data structure, and a degree-two trailblazer made feasible by dynamic connectivity, culminating in substantial speedups for related recognition tasks. This work broadens the practical tractability of induced-subgraph problems and raises open questions about shortest uv-trails and all-pairs variants, with implications for problems like hole-length recognition.
Abstract
For vertices $u$ and $v$ of an $n$-vertex graph $G$, a $uv$-trail of $G$ is an induced $uv$-path of $G$ that is not a shortest $uv$-path of $G$. Berger, Seymour, and Spirkl [Discrete Mathematics 2021] gave the previously only known polynomial-time algorithm, running in $O(n^{18})$ time, to either output a $uv$-trail of $G$ or ensure that $G$ admits no $uv$-trail. We reduce the complexity to the time required to perform a poly-logarithmic number of multiplications of $n^2\times n^2$ Boolean matrices, leading to a largely improved $O(n^{4.75})$-time algorithm.
