Characteristic polynomials of sparse non-Hermitian random matrices (Q7032432)

For an \(n \times n\) random matrix \(X_n\), the second correlation function of characteristic polynomials is defined as\N\[\N\mathbf{f}_2(z_1,z_2) =\mathbb{E} \left[ \prod_{s=1}^2 \det (X_n-z_s)\det (X_n-z_s)^* \right].\N\]\NThis paper examines the asymptotic behavior of \(\mathbf{f}_2(z_1,z_2)\) for the random matrix \(X_n=(x_{jk})_{1 \le j,k \le n}\). Each element of \(X_n\) is defined as \(x_{jk}= d_{jk} w_{jk}\), where \(\{w_{jk}\}\) are i.i.d. complex random variables with \(\mathbb{E}[w_{jk}] =\mathbb{E}[w_{jk}^2]=0\) and \(\mathbb{E} [|w_{jk}|^2]=1\). The \(\{d_{jk}\}\) are i.i.d. normalized Bernoulli variables, independent of \(\{w_{jk}\}\), where \(d_{jk}= \frac{1}{\sqrt{p}}\) with probability \(\frac{p}{n}\) and \(d_{jk}=0\) otherwise. Such random matrices are called sparse non-Hermitian random matrices.\N\NThe asymptotic behavior is examined under the following conditions:\N\[\Nb= \sqrt{\frac{2(n-p)}{np}}, \qquad z_j=z_0+ \frac{\zeta_j}{\sqrt{n}}, \quad j=1,2.\N\]\NWhen \(p\) is fixed or goes to infinity, and then \(n\) goes to infinity, the asymptotic behavior of \(\mathbf{f}_2(z_1,z_2)\) significantly varies depending on the relationship between \(b\) and \(|z_0|^2\). In some cases, it can be expressed by a determinant. The proof relies on the integral representation of \(\mathbf{f}_2\) and saddle-point analysis.