Asymptotics for products of characteristic polynomials in classical \(\beta\)-ensembles (Q485304)

The authors deal with three classical \(\beta\)-ensembles studied in random matrix theory whose eigenvalue probability densities, up to a normalization factors, are (i) the Gaussian (Hermite) \(\beta\)-ensemble: \(\prod_{1\leq i \leq N} e^{-\beta \frac{x_{i}^{2}}{2}} \prod_{1\leq j < k \leq N}|x_{j}- x_{k}|^{\beta}\), \(x_{i} \in \mathbb R\). (ii) the Chiral (Laguerre) \(\beta\)-ensemble: \(\prod_{1\leq i \leq N} x_{i} ^{\lambda_{1}} e^{-\beta \frac{x_{i}}{2}} \prod_{1\leq j < k \leq N}|x_{j}- x_{k}|^{\beta}\), \(x_{i}\in \mathbb R_{+} \). (iii) the Jacobi \(\beta\)-ensemble: \(\prod_{1\leq i \leq N} x_{i} ^{\lambda_{1}} (1- x_{i}) ^{\lambda_{2}} \prod_{1\leq j < k \leq N}|x_{j}- x_{k}|^{\beta}\), \(x_{i}\in (0,1)\). \(\beta\) is the so-called Dyson number and for \(\beta=1,2,4\), the ensembles of random matrices whose respective probability measures exhibit orthogonal, unitary or symplectic symmetry appear. Many interesting probabilistic quantities (the global fluctuations, the gap probabilities, the distribution of the largest eigenvalues) have been studied when the size \(N\) of the matrix tends to infinity. On the other hand, some universality results concerning the case \(\beta >0\) have been recently obtained. When \(N\) tends to infinity, it was shown that the eigenvalues in the bulk of the spectrum of any \(\beta\)-ensemble defined by the density \(\prod_{1\leq i \leq N} e^{-\beta V(x_{i})} \prod_{1 \leq j < k \leq N}|x_{j} - x_{k}|^{\beta}\), where \(V\) is a real-valued function on the real line, are correlated as the eigenvalues of the Gaussian \(\beta\)-ensemble, when an appropriate rescaling and a small averaging are introduced. Let \(X\) be a random matrix of size \(N\) in some \(\beta\)-ensemble. In the paper under review, exact and explicit expressions, when \(N\) tends to infinity, of the expectation value of \(\prod_{j=1}^{n} \det (X - s_{j})\), for \(\beta\)-classical ensembles are obtained. More precisely, the authors focus the attention in the weighted quantity \(\varphi_{N}(s_{1},\dots, s_{n}) = e^{-\frac{1}{2} \sum_{j=1}^{n} V(s_{j})} K_{N} (s_{1},\dots ,s_{n})\). Here \(K_{N} (s_{1},\dots, s_{n})= \langle \prod_{i=1}^{N} \prod_{j=1}^{n} (x_{i} - s_{j})\rangle,\) where \((x_{1}, ...x_{N}\) are the eigenvalues of the random matrix \(X\), the angle brackets stand for the expectation value, and \(V(x_{j})\) equals \(x_{j}^{2}\) (Gaussian case), \(x_{j}- (2/\beta) \lambda_{1} \ln (x_{j})\) (Chiral case) and \(-2/\beta (\lambda_{1} \ln (x_{j}) + \lambda_{2} \ln (1- x_{j}))\) (Jacobi case). At the soft edge, the expectation value of products of the characteristic polynomials in the Gaussian and Chiral \(\beta\)-ensembles leads to the same multivariate Airy function. This result suggests that for any \(\beta\)-ensemble characterized by a potential \(V\), the average of the product of \(n\) characteristic polynomials, with an appropriate rescaling and re-centered at the soft edge, should become independent of \(V\) and proportional to the multivariate Airy function \(Ai ^{(\beta/2)} (s_{1}, \dots , s_{n})\) when \(N\) tends to infinity. This would constitute a universality property. The conjecture for universality is even stronger in the bulk of the spectrum. The authors prove that the three classical \(\beta\)-ensembles have the same asymptotic limit for the weighted expectation value \(\varphi_{N}\) in the bulk when \(N\) tends to infinity. Indeed, it is a multivariate hypergeometric function of exponential type \(_{1}F_{1}^{(\beta/2)}\) when \(n\) is an even number. Notice that if \(n\) is an odd number, the combination of \(\varphi_{N}\) and \(\varphi_{N-1}\) exhibits a universal pattern in terms of the generalized exponential \(E_{k}^{(\alpha)}\) defined in terms of the \((0,0)\)-type hypergeometric series, see [\textit{J. Kaneko}, SIAM J. Math. Anal. 24, No. 4, 1086--1110 (1993; Zbl 0783.33008)]. The calculation of scaling limits in the bulk and at the soft edge requires the asymptotic evaluation of Selberg-type integrals. The authors generalize the steepest descent method for high-dimensional integrals containing the absolute value of the Vandermonde determinant. The hard edge expectation for \(K_{N}\) also involves a single hypergeometric series \(_{0}F_{1}^{(\beta/2)}\) that can be seen as a multivariate Bessel function. This result is proved for Chiral and Jacobi \(\beta\)-ensembles. The universality of the asymptotic expectation is also conjectured. The asymptotic behavior, when \(N\) tends to infinity, of the soft-edge (resp. bulk) correlation is given when \(\beta\) is even and \(n=\beta k\) for Gaussian and Chiral (resp. the three classical) \(\beta\)-ensembles. Notice that the hard-edge case is well known in the Chiral and Jacobi cases (see [\textit{P. J. Forrester}, Log-gases and random matrices. Princeton, NJ: Princeton University Press (2010; Zbl 1217.82003)]).