The automorphism-group conjecture for algorithms computing matrix multiplication

About 3 years old · traced to

Let A\mathcal{A} be an algorithm of length rr computing the matrix multiplication tensor ⟨m,n,p⟩\langle m,n,p\rangle. Its automorphism group is denoted by Aut(A)Aut(\mathcal{A}). Let Pm\mathcal{P}_m, Pn\mathcal{P}_n, and Pp\mathcal{P}_p be the corresponding groups of generalized permutation matrices, and let Q(m,n,p)Q(m,n,p) be the group appearing in the natural semidirect-product action on algorithms. Automorphism-group conjecture. Suppose that A\mathcal{A} is an algorithm of any length rr computing ⟨m,n,p⟩\langle m,n,p\rangle. Then

Aut(A)≤(Pm×Pn×Pp)⋊Q(m,n,p).Aut(\mathcal{A})\leq(\mathcal{P}_m\times\mathcal{P}_n\times\mathcal{P}_p)\rtimes Q(m,n,p).

The conjecture asserts that the DD-property and weak DD-property used in the preceding partial results are unnecessary. It is intended to apply to algorithms of arbitrary length, whether rr is less than, equal to, or greater than mnpmnp; the supplied source does not state a resolution.

References

Primary source

Xin Li, Yixin Bao and Liping Zhang, “On the local dimensions of solutions of Brent equations”, arXiv:2303.09754 (2024).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.