Product expansions of \(q\)-character polynomials (Q6100550)

Character polynomials are an algebra of class functions, defined simultaneously on an infinite collection of groups, whose statistical properties relate the combinatorics and representation theory of the different groups, thus appearing in the context of representation stability. The present paper studies the \textit{ring of \(q\)-character polynomials} -- which is by definition a collection of class functions on \(\mathrm{GL}_n(\mathbb{F}_q)\) (the set of all \(n\times n\) invertible matrices with entry from \(\mathbb F _q\) a field of order \(q\)), \(n\) varying over all positive integers. These are \(q\)-analogs of character polynomials of symmetric groups, which arise as the stable characters of a finitely generated sequence of representations via \(VI\)-modules. These functions exhibit uniform behaviour when considered as statistics on matrix groups. Fix an \(m\times m\) matrix \(A\). The \textit{\(q\)-character polynomial} associated with \(A\) is the statistic \(X_A:\bigsqcup\limits_{n\geq 1} \mathrm{End}(\mathbb {F} _q ^n)\longrightarrow \mathbb Q\) defined simultaneously on square matrices \(B\in\mathrm{End}(\mathbb F _q ^n)\) as \[ X_A(B,q)=\left\vert\left\{W\leq \mathbb F_q ^n\mid \dim W=d,~ B(W)\subseteq W,~ B|_W\sim A\right\} \right\vert, \] where the symbol \(\sim\) denotes the matrix similarity. A \textit{\(q\)-character polynomial} is a finite \(\mathbb Q\)-linear combination of functions of the form \(X_A\). In this paper, the authors prove the following main results. \begin{itemize} \item[1.] For two matrices \(A\) and \(B\), the statistics \(X_A(B,q^m)\) coincide with finitely many integer polynomials \(P_1(t),\dots, P_d(t)\in\mathbb Z[t]\) evaluated at \(t=q^m\), in a periodic manner (the number \(d\) depends on the degree of the splitting field of \(\mathrm{Char}(A)\) over \(\mathbb {F}_q\)). Furthermore, if for two partitions \(\lambda\) and \(\mu\), \(J_\lambda\) and \(J_\mu\) are unipotent matrices in Jordan form, then there exists a single polynomial \(P_{\lambda,\mu}(t)\) satisfying \[ X_{J_{\lambda}}(J_\mu,p^m)=P_{\lambda,\mu}(p^m) \] for all primes \(p\) and \(m\geq 1\). \item[2.] Fix two square matrices \(A\) and \(B\) such that their eigenvalues are contained in \(\mathbb F_q\). Let \(\{\alpha_1,\alpha_2,\dots,\alpha_k\}\) be the set of distinct eigenvalues of \(A\) and \(B\). Also, assume \(\lambda_i\) and \(\mu_i\) denote the partitions enumerating the Jordan block sizes with eigenvalue \(\alpha_i\) in \(A\) and \(B\) respectively. Then there exists integer polynomials \(R^{\nu_i}_{\lambda_i,\mu_i}\in\mathbb Z[t]\) labelled by triples of partitions such that \[ X_A\cdot X_B=\displaystyle\sum\limits_{\nu_1,\dots,\nu_k}\left(\prod\limits_{i=1}^k R^{\nu_i}_{\lambda_i,\mu_i}(q)\right)X_{J_{\nu_1}(\alpha_1)\oplus \dots\oplus J_{\nu_1}(\alpha_1)}, \] where the sum runs over all \(k\)-tuples of partitions \(\nu_1,\dots,\nu_k\) satisfying \(\sum||\nu_i||\leq \dim A+\dim B\). \item[3.] For every pair of integer partitions \(\lambda\) and \(\mu\), the polynomials \(R^{\nu}_{\lambda,\mu}\) satisfy the recursion, \[ R^{\nu}_{\lambda,\mu}(t)=P_{\lambda,\nu}(t)\cdot P_{\mu,\nu}(t)-\sum\limits_{||\nu'||<||\nu||}R^{\nu'}_{\lambda,\mu}(t)P_{\nu'.\nu}(t), \] where the sum goes over all partitions \(\nu'\) of numbers \(<||\nu||\). \end{itemize} It is worth noting that an important implication is that, for \(b\geq a\), the correlation between random variables \(X_{J_a}\) and \(X_{J_b}\) in the uniform probability space \(\mathrm{GL}_n(\mathbb F_q)\) is \(\sqrt{\dfrac{q^a-1}{q^b-1}}\) whenever \(n\geq 2b\).