Random matrices: The distribution of the smallest singular values (Q987365)

Let \(\xi\) be a (real or complex) random variable. The authors are concerned with the distribution of the smallest singular value \(\sigma_n(M_n(\xi))\) of the \(n\times n\) random matrix whose entries are i.i.d. copies of \(\xi\). \textit{A. Edelman} [SIAM J. Matrix Anal. Appl. 9, No. 4, 543--560 (1988; Zbl 0678.15019)] proved that for \(\xi\) Gaussian one has \[ \mathbb{P}(n\,\sigma_n(M_n(\xi))^2\leq t)=\int_0^t\,\frac{1+\sqrt x}{2\sqrt x}\,e^{-(x/2+\sqrt x)}\,dx +o(1) \] in the real case, and \[ \mathbb{P}(n\,\sigma_n(M_n(\xi))^2\leq t)=\int_0^t\,e^{-x}\,dx \] in the complex case. As expected, Edelman's proof relies heavily on the symmetry and special properties of the Gaussian ensemble. In the paper under review, the authors set out to work on the same problem, but without setting any restrictions on the distribution of \(\xi\) but for a finite moment condition. Concretely, they prove that if \(\xi\) is real valued with mean zero and variance 1, one has \[ \mathbb{P}(n\,\sigma_n(M_n(\xi))^2\leq t)=\int_0^t\,\frac{1+\sqrt x}{2\sqrt x}\,e^{-(x/2+\sqrt x)}\,dx +O(n^{-c}), \] and if \(\xi\) is complex valued with mean zero and real and imaginary parts having variance \(1/2\) and covariance zero, \[ \mathbb{P}(n\,\sigma_n(M_n(\xi))^2\leq t)=\int_0^t\,e^{-x}\,dx. \] The authors also show that the result still holds if the matrix entries are not identically distributed, provided that they are independent and that the matrix still satisfies the finite large moment condition. They also use their results to generalize other results by Edelman from the paper cited above.