Core Concepts
Tensor Isomorphism can be reduced to other algebraic isomorphism problems, such as Group Isomorphism and Cubic Form Equivalence, with only linear-length blow-up in the input sizes, leading to faster algorithms for these problems.
Abstract
The paper presents new techniques for reducing Tensor Isomorphism (TI) to other algebraic isomorphism problems, such as Group Isomorphism (GpI) and Cubic Form Equivalence (CFE), with only linear-length blow-up in the input sizes. This is a significant improvement over previous reductions that incurred quadratic blow-ups.
The key contributions are:
A new gadget construction for reducing Partitioned Tensor Isomorphism (part-TI) to plain TI, with only linear-length blow-up in the input sizes. This improves upon the previous quadratic blow-up.
Using the new part-TI gadget, the authors show that if Graph Isomorphism (GI) is in P, then CFE over finite fields can be solved in quasilinear time, improving upon the previous brute-force upper bound.
The authors devise a reduction from GpI for p-groups of class c and exponent p, where c < p, to GpI for the class c = 2 and exponent p. This, combined with a recent breakthrough algorithm for the class c = 2 case, yields faster algorithms for the more general class c < p.
Polynomial-time search- and counting-to-decision reductions are presented for GpI of p-groups of class 2 and exponent p, answering a question of Arvind and Tóran.
The techniques developed for TI and its connections to other algebraic isomorphism problems provide a unifying framework for tackling long-standing complexity-theoretic questions about these problems.