English

Statistical mechanics of the maximum-average submatrix problem

Disordered Systems and Neural Networks 2024-01-24 v2 Information Theory math.IT

Abstract

We study the maximum-average submatrix problem, in which given an N×NN \times N matrix JJ one needs to find the k×kk \times k submatrix with the largest average of entries. We study the problem for random matrices JJ whose entries are i.i.d. random variables by mapping it to a variant of the Sherrington-Kirkpatrick spin-glass model at fixed magnetization. We characterize analytically the phase diagram of the model as a function of the submatrix average and the size of the submatrix kk in the limit NN\to\infty. We consider submatrices of size k=mNk = m N with 0<m<10 < m < 1. We find a rich phase diagram, including dynamical, static one-step replica symmetry breaking and full-step replica symmetry breaking. In the limit of m0m \to 0, we find a simpler phase diagram featuring a frozen 1-RSB phase, where the Gibbs measure is composed of exponentially many pure states each with zero entropy. We discover an interesting phenomenon, reminiscent of the phenomenology of the binary perceptron: there exist efficient algorithms that provably work in the frozen 1-RSB phase.

Keywords

Cite

@article{arxiv.2303.05237,
  title  = {Statistical mechanics of the maximum-average submatrix problem},
  author = {Vittorio Erba and Florent Krzakala and Rodrigo Pérez and Lenka Zdeborová},
  journal= {arXiv preprint arXiv:2303.05237},
  year   = {2024}
}
R2 v1 2026-06-28T09:09:12.209Z