Efficient eigenvalue counts for tree-like networks

Estimating the number of eigenvalues $\mu_{[a,b]}$ of a network’s adjacency matrix in a given interval $[a,b]$ is essential in several fields. The straightforward approach consists of calculating all the eigenvalues in $O(n^3)$ (where $n$ is the number of nodes in the network) and then counting the ones that belong to the interval $[a,b]$. Another approach is to use Sylvester’s law of inertia, which also requires $O(n^3)$. Although both methods provide the exact number of eigenvalues in $[a,b]$, their application for large networks is computationally infeasible. Sometimes, an approximation of $\mu_{[a,b]}$ is enough. In this case, Chebyshev’s method approximates $\mu_{[a,b]}$ in $O(|E|)$ (where $|E|$ is the number of edges). This study presents two alternatives to compute $\mu_{[a,b]}$ for locally tree-like networks: edge- and degree-based algorithms. The former presented a better accuracy than Chebyshev’s method. It runs in $O(d|E|)$, where $d$ is the number of iterations. The latter presented slightly lower accuracy but ran linearly ($O(n)$).