Upper bounds for the second largest eigenvalue of symmetric nonnegative matrices (Q357320)

Let \(A=(a_{ij})\) be a symmetric nonnegative \(n\times n\) matrix, \(n\geq 2\), with largest eigenvalue~\(\lambda_1\) and second largest~\(\lambda_2\). \textit{B.~J.~Ebrahimi} et al. [Linear Algebra Appl. 429, No. 11--12, 2781--2787 (2008; Zbl 1148.05047)] proved that if \(A\) has all entries at most one and all row sums equal to~\(r\), then \(\lambda_2\leq n-r\) and \(\lambda_1+\lambda_2\leq n\). Extending these results to the case when the row sums are allowed to be unequal, and further to all symmetric nonnegative matrices, the present author provides several upper bounds for \(\lambda_2\) and \(\lambda_1+\lambda_2\). She pays special attention to matrices with diagonal zero and entries at most one, so obtaining corresponding bounds also for graph eigenvalues. One example follows. If \(\mathrm{diag}\,A=0\) and \(a_{ij}\leq 1\) for all~\(i,j\), then \[ \lambda_2\leq\rho(J-I-A)-1, \] where \(I\) is the identity matrix, \(J\) is the matrix of ones, and \(\rho\) denotes the Perron root. In the general case, this implies \[ \lambda_2\leq a-2M+\rho(MJ-(A-D_A)), \] where \(a\geq\max_ia_{ii}\), \(M\geq\max_{i\neq j}a_{ij}\), and \(D_A=\mathrm{diag}\,(a_{11},\dots,a_{nn})\).