Cohen-Lenstra distributions via random matrices over complete discrete valuation rings with finite residue fields (Q2039738)

Motivated by conjectures of Cohen and Lenstra on \(p\)-parts of class groups of number fields, Friedman and Washington determined the distribution of the cokernel of \(n \times n\) matrices with entries in \(\mathbb{Z}_p\) chosen with respect to Haar measure [\textit{E. Friedman} and \textit{L. C. Washington}, in: Théorie des nombres, C. R. Conf. Int., Québec/Can. 1987, 227--239 (1989; Zbl 0693.12013)]. More precisely, they show that for any finite abelian \(p\)-group \(H\), \[ \lim_{n \rightarrow \infty} \operatorname{Prob}_{A \in \operatorname{Mat}_n(\mathbb{Z}_p)}(\operatorname{coker}(A) \simeq H) = \frac{1}{|\operatorname{Aut}_{\mathbb{Z}}(H)|} \prod_{i=1}^\infty(1-p^{-i}). \] Function field analogues of the Cohen-Lenstra conjectures led Friedman and Washington to also prove that \[ \lim_{n \rightarrow \infty} \operatorname{Prob}_{A \in \operatorname{GL}_n(\mathbb{Z}_p)}(\operatorname{coker}(A-\operatorname{id}) \simeq H) = \frac{1}{|\operatorname{Aut}_{\mathbb{Z}}(H)|} \prod_{i=1}^\infty(1-p^{-i}). \] That is, they determine the probability that a random \(A \in \operatorname{Mat}_n(\mathbb{Z}_p)\) satisfies both \(\operatorname{coker}(A) = 0\) and \(\operatorname{coker}(A-\operatorname{id}) \simeq H\). The main results of this paper generalize this earlier work. Let \(r \ge 1\) and \(P_1(t),\ldots, P_r(t) \in \mathbb{Z}_p[t]\) be monic irreducible polynomials such that their reductions modulo \(p\) are distinct and irreducible. Suppose that \(\operatorname{deg}(P_r) = 1\). The authors prove that for any finite abelian \(p\)-group \(H\), \[ \lim_{n \rightarrow \infty} \operatorname{Prob}_{A \in \operatorname{Mat}_n(\mathbb{Z}_p)} \left( \begin{array}{c} \operatorname{coker}(P_1(A)) = \cdots = \operatorname{coker}(P_{r-1}(A)) =0\\ \text{ and } \operatorname{coker}(P_r(A)) \simeq H. \end{array} \right) \] \[ = \frac{1}{|\operatorname{Aut}_{\mathbb{Z}}(H)|} \prod_{j=1}^r \prod_{i=1}^\infty (1-p^{-i \operatorname{deg}(P_j)}).\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \] In fact, the authors prove something more general where \(\mathbb{Z}_p\) can be replaced by any complete DVR with finite residue field. A main step in the arguments is to prove a result about the distribution of the \(\mathbb{F}_q[t]\)-module structure of random matrices over \(\mathbb{F}_q\). An important idea in the proofs is to use a combinatorial tool, the cycle index, and study generating functions that encode information about similarity classes in \(\operatorname{Mat}_n(\mathbb{F}_q)\). The authors also give an interesting general conjecture about the simultaneous behavior of \(\operatorname{coker}(P_1(A)), \ldots, \operatorname{coker}(P_{r}(A))\) that has already motivated further work in this area.