Eigenvalues in the large sieve inequality (Q2479902)

The ``arithmetic'' large sieve can be viewed as bounding the largest eigenvalue of the Hermitian matrix \((X_{mn})\), with \[ X_{mn}= \sum_{q\leq Q}\;\sum_{(a,q)= 1} e((m- n)a/q),\quad\text{for }m,n\leq N. \] We know that the largest eigenvalue is at most \(N+ Q^2\). In this paper the author is interested in the distribution of the eigenvalues of \((X_{mn})\). If \(M(k)\) is the sum of the \(k\)th powers of the eigenvalues it is easy to see that \[ M(1)= N \sum_{q\leq Q} \varphi(q). \] The principal theoretical result of the paper is the asymptotic formula \[ M(2)= NM(1)+ N^2 Q^2 f(N/Q^2)+ O(NQ^3\log^7 Q), \] where \(f\) is an explicit but complicated \(\mathbb C^2\) function. Numerical evidence is presented which suggests that the eigenvalues, normalized by dividing by \(N\), have a limiting distribution, at least in the case in which \(N= \sum_{q\leq Q}\varphi(q)\). For this value of \(N\) the average of the logarithms of the normalized eigenvalues is also computed asymptotically. The result implies that the smallest eigenvalue is between \(N^{1/2}\) and \(N^{-({1\over 2}+ o(1))\log N}\). It would be interesting to narrow this range.