In this talk, we shall discuss recent advances in entrywise perturbation theory and high accuracy algorithms for diagonally dominant M-matrices and present their applications to high accuracy computations in solving queuing problems. Diagonally dominant M-matrices arise in a large variety of applications, including the queuing models. In the recent works [1,2], we have shown that, if each off-diagonal entry and the diagonally dominant part of a diagonally dominant M-matrix is determined to high relative accuracy, then its smallest eigenvalue (corresponding to the Perron root of the inverse) and each entry of its inverse are all determined to the same relative accuracy. New algorithms have also been developed that compute these quantities with relative errors in the magnitude of the machine precision. These more accurate algorithms can be used, for example, to compute h to high accuracy where h is the decay rate for queue length in GI/M/1 queuing systems. Note that the quantity delta = 1-h is also of interest in such applications and hence high accuracy h is necessary when h is close to 1.
Here, we shall consider functional iteration algorithms for solving nonlinear matrix equations arising in queuing models. Specifically, to solve
|