The block smoothing method for estimating the spectral radius of a nonnegative matrix (Q870159)

The author presents a new estimate for the spectral radius of a nonnegative matrix, using only the elements of a series of block matrices. The first part consists in an introduction regarding the spectrum of a nonnegative matrix \(A\), of order \(n\), with eigenvalues \(\lambda_1,\lambda_2,\dots, \lambda_n\) and the bounds for \(\rho(A)\): \(r(A)=\min_i\,R_i\), the lower one, and \(R(A)= \max_i\,R_i\), the upper one, where \(R_i= \sum^n_{j=1} a_{ij}\), \(1\leq i\leq n\). The second part introduces the block smoothing method partitionning a matrix and provides some notations, which will be used in the following. The third section states the relationship between \(r(A^{(k-1)})\) and \(R(A^{(k-1)})\) \[ A^{(p)}= {1\over n^2_p- 1} \sum^{n_p-1}_{i,j} A^{(i-1)}_{ij}= (a^{(i-1)}_{ij})_{n_p\times n_p},\quad p= 1,2,\dots, n_k- 1, \] where \(n= n_1\times n_2\times\cdots\times n_k\), \(n_i\), \(i= 1,\dots, k\), being prime numbers. \(A^{(k-1)}\) being the last matrix of the sequence of new matrices which preserve the spectral radius. Two formulas are also given which allow for estimating the spectral radius of the matrix \(A\): \[ \begin{gathered} \rho(A)\sim n_1n_2\cdots n_{k-1}\rho(A^{(k-1)}),\\ \rho(A)\sim n_1n_2\cdots n_{k-1} r(A^{(k-1)})\sim n_1n_2\cdots n_{k-1} R(A^{(k-1)}).\end{gathered} \] By using them one can transform a large matrix to several smaller matrices and, in fact, we only need to calculate the spectral radius of the \(n_k\times n_k\) matrix \(A^{(k-1)}\), even its row sum \(r(A^{(k-1)})\) or \(R(A^{(k-1)})\). The fourth section concerns the computation of the matrices \(A^{(i)}\) \((i= 1,2,\dots, n_{k-1})\) by utilizing the identity matrix. Some numerical examples are presented in the last section. Thus, there are analysed a \(3\times 3\) nonnegative irreductible matrix, five \(4\times 4\) matrices generated randomly by Matlab, having all the elements range between 1 and 10 and a \(120\times 120\) nonnegative matrix, randomly generated by Matlab with 0-1 average distribution.