Lower bounds for the Perron root of a nonnegative matrix (Q5905669)

Some new lower bounds for the Perron root \(r(A)\) of a nonnegative square matrix \(A\) (i.e. the eigenvalue of \(A\) which is not less than \(| \lambda|\) for any \(\lambda \in \sigma(A)\)) are determined. Firstly the inequality \(r(S(A))\leq r(A)\) is discussed, where \(S(A) = [s_{ij}]\) is the geometric symmetrization of \(A\) defined by \(s_{ij} = (a_{ij}a_{ji})^{1/2}\); and a necessary and sufficient algebraic condition for the equality to occur is given. The lower bound \((1/n)e^ TS(A)e\) is derived, where \(e = (1,1,\dots,1)^ T\) and it is shown that it is equal to \(r(A)\) if and only if up to a common positive constant all irreducible components of \(A\) are diagonally similar to some symmetric stochastic matrices. Using the Rayleigh quotient for the geometric symmetrizations of \(A^{2^ k}\) a convergent and increasing sequence of lower bounds for \(r(A)\) are constructed. Two simple ways to obtain more accurate lower bounds are described by symmetrizing \(A^ 2\) and by squaring instead of \(A\) the shifted matrix \(A - (\min_ ia_{ii})I\). It is shown that the new introduced bounds are frequently better than known bounds and readily computable, this fact is emphasized by two numerical examples.