The number of \(2\times 2\) integer matrices having a prescribed integer eigenvalue (Q1005894)

For each integer \(\lambda\) let \(M_{2}^{\lambda}(k)\) denote the set of \(2\times2\) matrices with integer entries from the interval \([-k,k]\) having \(\lambda\) as an eigenvalue. The authors show that \(\left| M_{2}^{\lambda }(k)\right| =24V(\lambda/k)\pi^{-2}k^{2}\log k+O(k^{2})\) where \[ V(\delta)=\begin{cases} 4-2\delta-\delta^{2}+\delta^{2}\log(1+\delta)-2(1-\delta)\log(1-\delta) & \text{for }0\leq\delta<1\\ 1+\log2 &\text{for }\delta=1\\ 4-2\delta-\delta^{2}+\delta^{2}\log(\delta+1)+2(\delta-1)\log(\delta-1) & \text{for }1<\delta\leq\sqrt{2}\\ \delta^{2}-2\delta-(\delta^{2}-2\delta+2)\log(\delta-1) & \text{for }\sqrt {2}<\delta\leq2 \end{cases} \] and \(V(\delta)=0\) otherwise. In particular, the probability that a random \(2\times2\) integer matrix with entries in \([-k,k]\) has integer eigenvalues is asymptotic to \((C\log k)/k\) where \(C=(7\sqrt{2}+4+3\log(\sqrt{2}+1))/(3\pi ^{2})=0.5587...\). The proof is elementary, but quite complicated, and it seems unlikely that it can be generalized to obtain corresponding results for matrices of larger order. The authors point out that the more powerful but less elementary method of \textit{Y. R. Katznelson} [Duke Math. J. 69, 121--136 (1993; Zbl 0785.11050)] can also be used to prove their main theorem, and that Katznelson's results apply more generally to \(n\times n\) matrices.