Implicitly restarted global FOM and GMRES for nonsymmetric matrix equations and Sylvester equations (Q2570769): Difference between revisions

Many applications require the solution of several sparse systems of linear equations \(Ax^i=b^i\), for \(i=1,\ldots, s,\) with the same coefficient matrix and different right-hand sides. When all the \(b^i\) are available simultaneously, we have a matrix equation \(AX=B\), where \(A\) is an \(n\times n\) real nonsingular matrix, \(B\) and \(X\) are \(n\times s\) rectangular matrices whose columns are \(b^i\) and \(x^i\), respectively. Recently, global full orthogonal method (FOM) and generalized minimal residual (GMRES) algorithms for nonsymmetric matrix equations presented by \textit{K. Jbilou, A. Messauudi} and \textit{H. Sadok} [Appl. Numer. Math. 31, No.~1, 49--63 (1999; Zbl 0935.65024)]. In general, they must be restarted which slows the convergence. However, some information can be retained at the time of restart and used it in the next cycle. In this paper, the author presents algorithms that use the implicitly restarted technique and deflation in order to retain this information. In addition, it is proved that the global FOM and GMRES methods for matrix equations are equivalent with the corresponding methods for linear equations. In this case, the author proposes implicitly restarted global FOM and GMRES.