A new family of \((I+S)\)-type preconditioner with some applications (Q747196)

To construct a preconditioner \(P\) for a matrix \(A=I+U+L\) (\(U\) and \(L\) are the strict upper- and lower triangular parts, respectively), one has to find an approximation for \(A^{-1}\). These are often of the form \(P=I+S\). Approximations for \(A^{-1}\) can be obtained by solving \(X^{-1}-A=0\) using Newton's method or a higher-order method. The matrix \(I+S\) can be used as a starting value. Newton's method gives for example \(X_{k+1}=X_k(2I-AX_n)\), thus \(P=I+K=(I+S)[(I-S)+(L+U)(I+S)]\). It is proved and illustrated by numerical examples, and under certain conditions, a speed-up is obtained with such an iterated preconditioner.