Character expansion method for the first order asymptotics of a matrix integral (Q2487020)

The authors show how to use the character expansion method to estimate rigorously specific matrix integrals. Namely let \(A_N\) and \(B_N\) be two \(N\times N\) Hermitian matrices. The partition function is defined as \[ Z_N(\Phi)=\int dM e^{-\frac N2 \text{tr} M^2+\sum_{k\geq 1}\text{tr}(B_N^k)\text{tr}((\Phi(M)A_N)^k)}, \] where \(dM\) is the Lebesgue measure over the set of Hermitian matrices of size \(N\), tr denotes the usual trace of a matrix and function \(\Phi:\mathbb{R}\to\mathbb{R}\) is continuous and \(\Phi(M)=U \text{diag}(\Phi(\lambda_1),\ldots,\Phi(\lambda_N))U^{*}\) whenever \(M=U\text{diag}(\lambda_1,\ldots,\lambda_N)U^{*}\). Under appropriated boundness conditions, for function \(\Phi\), the authors are capable to obtain the asymptotic behavior of the partition function. Their two main results are established in Theorem 1: firstly under certain hypotheses \(F_N(\Phi)=\frac1{N^2}\log Z_N(\Phi)\) converges as \(N\) tends to infinity and in Theorem 5 an explicitly formula is derived for this limit and secondly they can also give a weak characterization of the limit points of the spectral measure of \(M\) under the Gibbs measure associated to \(Z_N(\Phi)\).