Random integral matrices and the Cohen Lenstra Heuristics

TL;DR

This paper demonstrates that large random integral matrices with certain entry distributions have cokernels whose distribution matches the Cohen-Lenstra heuristics for class groups, extending previous results on matrix ranks.

Contribution

It refines the understanding of cokernel distributions of random matrices, aligning them with Cohen-Lenstra heuristics for class groups, including cases with rectangular matrices.

Findings

Cokernels of random matrices follow Cohen-Lenstra distribution asymptotically.

Results apply to matrices with entries in residue classes modulo primes.

Extension to rectangular matrices with size n by n+u.

Abstract

We prove that given any $\epsilon>0$, random integral $n\times n$ matrices with independent entries that lie in any residue class modulo a prime with probability at most $1-\epsilon$ have cokernels asymptotically (as $n\rightarrow\infty$) distributed as in the distribution on finite abelian groups that Cohen and Lenstra conjecture as the distribution for class groups of imaginary quadratic fields. This is a refinement of a result on the distribution of ranks of random matrices with independent entries in $\mathbb{Z}/p\mathbb{Z}$. This is interesting especially in light of the fact that these class groups are naturally cokernels of square matrices. We also prove the analogue for $n\times (n+u)$ matrices.