Pfaffian point process for the Gaussian real generalised eigenvalue problem (Q1930862)

Let \(X_1\) and \(X_2\) be two independent \(n\times n\) random matrices from the real Ginibre ensemble. That is, the entries of \(X_1\) and \(X_2\) are independent real standard Gaussian random variables and no symmetry constraints are imposed on \(X_1\) and \(X_2\). The aim of the paper is to study the generalised eigenvalues of the pair \((X_1,X_2)\). These are defined as the solutions of the equation \(\det (X_1-\lambda X_2)=0\) or, equivalently, for \(X_2\) invertible, as the eigenvalues of the matrix \(Y:=X_2^{-1}X_1\). The joint density of the elements of the random matrix \(Y\) is some explicit constant times \(\det (\mathbf{1}_n + Y Y^T)^{-n}\). This is a matrix generalization of the Cauchy distribution. The authors compute explicitly the joint probability density of the eigenvalues of \(Y\), the probability of finding exactly \(k\) real eigenvalues, the average densities of real and complex eigenvalues, and give explicit Pfaffian formulas for the correlation functions \(\rho_{k_1,k_2}\) specifying the probability of finding exactly \(k_1\) real eigenvalues as some given points of the real line and exactly \(k_2\) pairs of complex conjugate eigenvalues at some given points outside the real line. A local limit theorem with a normal limiting distribution for the number of real eigenvalues is obtained. The scaling limit of the correlation functions \(\rho_{k_1,k_2}\) in the vicinity of the origin is shown to be of the same form as in the case of the real Ginibre ensemble. The limiting correlations are shown to satisfy sum rules characteristic of the underlying two-component Coulomb gas.