Limit theorems for orthogonal polynomials related to circular ensembles (Q501816): Difference between revisions

For \(n\geq1\), \(\beta >0\) and \(\delta\), a complex number with \(\Re \delta >-1/2\), let us consider a random probability measure \(\mu= \sum_{j=1}^{n} \pi_{j}\delta_{e^{i\theta_{j}}}\) supported on \(n\) points of the unit circle. \(\mu\) is said to belong to the family \(CJ^{(n)}_{\beta,\delta}\) if {\parindent=0.6cm\begin{itemize}\item[--] the joint density of the law of \((\theta_{1}, \dots, \theta_{n})\) with respect to the Lebesgue measure on \([0, 2\pi)^{n}\) is given by \(c^{(n)}_{\delta, \beta} |\triangle (\theta_{1}, \dots, \theta_{n})|^{\beta} \prod_{j=1}^{n} (1- e^{-i\theta_{j}})^{\delta}(1- e^{i\theta_{j}})^{\delta}\), where \(\triangle\) is the Vandermonde determinant and \(c^{(n)}_{\delta, \beta}\) is a normalization constant; \item[--] the weights \((\pi_{1}, \dots, \pi_{n})\) follow a Dirichlet law of parameter \((\beta', \dots, \beta')\) on the simplex \(\pi_{1} + \cdots + \pi_{n}= 1\), \(\pi_{j}>0\) and \(\beta'= \beta/2\); \item[--] the tuples \((\theta_{1}, \dots, \theta_{n})\) and \((\pi_{1}, \dots, \pi_{n})\) are independent. \end{itemize}} Let us consider the family of monic orthogonal polynomials \(\{\Phi_{k,n}(z)\}_{k=0}^{n}\) associated with \(\mu\). Notice that \(\Phi_{n,n}(z) = \prod_{j=1}^{n} (z- e^{i\theta_{j}})\). If \(U_{n}\) denotes the \(n\times n\) unitary matrix representing the multiplication operator by \(z\) in terms of the above orthogonal basis and \(G_{k}(U_{n})\), \(1\leq k \leq n\), is the \(k\times k\) leading principal submatrix of \(U_{n}\), then \(\Phi_{k,n}(1)= \det (I_{k} - G_{k}(U_{n}))\); see [\textit{B. Simon}, J. Comput. Appl. Math. 208, No. 1, 120--154 (2007; Zbl 1125.15027)]. In the paper under review, the asymptotic properties of the process \((\Phi_{\lfloor nt\rfloor,n}(1), t\in[0,1])\) when \(n\rightarrow\infty\) and \(\Phi_{k,n}(1)\neq 1\), a fact that occurs almost surely when \(\mu \in CJ^{(n)}_{\beta,\delta}\), are studied under two regimes: First regime: \(\delta\) is fixed and \(\Re \delta>-1/2\). Second regime: \(\delta= \beta' d n\), with \(\Re d>0\). The variables log \(\Phi_{\lfloor nt\rfloor,n}(1)\) are analyzed for both regimes in such a way that their expectations converge to some explicit deterministic function of \(t\). Next, the fluctuations of \((\log \Phi_{\lfloor nt\rfloor,n}(1))\) are studied as a stochastic process on the space of càdlàg \(\mathbb{R}^{2}\)-valued functions. In the first regime, the cases \(0<t<1\) and \(t=1\) must be distinguished taking into account that at \(t=1\) some transition appears since the normalization is changed. Let \(\xi_{n}(t) + i \eta_{n}(t) = \log \Phi_{\lfloor nt\rfloor,n}(1)- \operatorname{E} \log \Phi_{\lfloor nt\rfloor,n}(1)\) and \(\zeta_{n}(t)= (\xi_{n}(t), \eta_{n}(t))\). In the first regime, \(\{\zeta_{n}(t): t<1\}\) converges to some explicit Gaussian diffusion \(\{\zeta^{0}_{d}(t): t<1\}\) while \(\frac{\zeta_{n}(1)}{\sqrt{n}}\) converges to a Gaussian random variable which is independent of the above diffusion. In the second regime, \(\{\zeta_{n}(t): 0\leq t \leq 1\}\) converges to some explicit Gaussian diffusion and there is no normalization to perform. The next step is focused on the large deviation principle (LDP) for the distributions of the real and imaginary parts of \(\log \Phi_{\lfloor nt\rfloor,n}(1)\) as a two-dimensional random vector with values in the Skorokhod space endowed with the weak topology. The case of the first regime where \(\delta=0\) is studied in detail. Finally, the connections between the above results and existing results on the LDP for the empirical spectral distributions in the circular Jacobi ensemble are studied based on the fact that \(\log \Phi_{n,n}(1)\) is a functional of the empirical spectral distribution defined by \(\mu^{(n)}_{\mathrm{ESD}}= \frac{1}{n} \sum_{j=1}^{n} \delta_{e^{i\theta_{j}}}\). Notice that, for \(\delta=0\) and \(\beta>0\), \textit{R. Killip} and \textit{I. Nenciu} [Int. Math. Res. Not. 2004, No. 50, 2665--2701 (2004; Zbl 1255.82004)] provided a model of random unitary matrices such that the spectral measure has the distribution \(CJ^{(n)}_{\beta,0}\). In the case \(\delta \neq0\), \textit{P. Bourgade} et al. [Int. Math. Res. Not. 2009, No. 23, 4357--4394 (2009; Zbl 1183.33013)] presented a matrix model where the matrices are products of \(n\) independent complex reflections. These results rely on the theory of orthogonal polynomials on the unit circle.