Eigenvalue distributions of large Hermitian matrices; Wigner's semi- circle law and a theorem of Kac, Murdock, and Szegö (Q2266756)

To each sequence \(h=(h_ j)^{\infty}_{j=-\infty}\in L^ 2([0,1])\) with \(\sum \| h_ j\|^ 2<\infty\) one can associate a function \(\sigma (h)\in L^ 2([0,1]^ 2)\) defined by \(\sigma (h)(x,t):=\sum h_ j(x)e^{2\pi ijt}.\) On the other hand, if the \(h_ j\) are continuous functions which satisfy \(h_{-j}=\bar h_ j\) one can form the Hermitean matrices \(A_ n\) with elements \(a_{\nu \mu}:=h_{\nu -\mu}((\nu +\mu)/2n),\) \(\nu,\mu =1,...,n\). \textit{M. Kac}, \textit{W. L. Murdock} and \textit{G. Szegö} [J. Rat. Mech. Analysis 2, 767-800 (1953; Zbl 0051.303)] proved that for \(n\to \infty\) the distribution of the eigenvalues of \(A_ n\) is asymptotically the same as that of the random variable \(\sigma\) (h) on \([0,1]^ 2\). In the special case of constant functions \(h_ j(x)\equiv c_ j\) the \(A_ n\) are Toeplitz matrices. For these the statement about the eigenvalue distribution was proved by Szegö in 1913. In the paper under review a very general \(L^ 2\)-version of the theorem of Kac, Murdock, and Szegö is presented and proved by rather elementary means, which use only \(L^ 2\)-estimates. This theorem is then applied to give a new proof of Wigner's semi-circle law for random Hermitean matrices.