Computing exponentials of essentially non-negative matrices entrywise to high relative accuracy (Q2840622)

This authors consider the computation of exponentials of essentially non-negative matrices, or of matrices whith non-negative nondiagonal elements. Three algorithms to compute the exponentials entrywise are presented. The first algorithm is a new version of the Taylor series method with a truncation criterion and an entrywise error analysis included. The second and the third are based in the polynomial method applied to the characteristic polynomial and the matrix exponential, respectively, where the entrywise relative accuracy is the main feature. Some examples validate the presented methods.