A triple-parameter modified SSOR method for solving singular saddle point problems (Q291896)

Let \(A\) be a positive definite matrix with real coefficients and \(B\) be a rectangular matrix with the same number of rows of \(A\) and whose columns are linearly dependent. The authors consider the saddle point problem of solving a linear system whose coefficient matrix is the \(2 \times 2\) block matrix \(\begin{pmatrix} A & B \\ -B^T & 0\end{pmatrix}\). They propose a triple-parameter modified symmetric successive overrelaxation (SSOR) method for this problem. A frame work is given for the method to be semi-convergent, which is to guarantee that the pseudo-spectral radius is strictly less than one. A local optimal parameter of the proposed technique is given. Numerical examples are presented to illustrate the efficiency of the method.