Tridiagonal splittings in the conditioning and parallel solution of banded linear systems (Q5961702): Difference between revisions

Consider the linear system \(Au=b\) where \(A\) is a real square nonsingular matrix and \(u\) and \(b\) are vectors of size \(n\), and assume the splitting of \(A\) of the form \(A=T-N\) where \(T\) is a nonsingular tridiagonal matrix and \(-N\) is the remaining part of \(A\). The matrix \(T\) may be an arbitrary tridiagonal matrix, in particular the tridiagonal part of \(A\). Such a tridiagonal splitting provides the iterative method \(Tu^{(m+1)} = Nu^{(m)} + {\mathbf b}\) for \(m=0,1,2,...\), where \(u^{(0)}\) is the initial vector. The authors study sufficient conditions for the convergence of the iteration as above. They introduce the definition of a tridiagonal dominant matrix \(A\) for which the method is convergent. In this case the conditioning of \(A\) is also examined. Some numerical results illustrate the conditioning of several classes of band matrices. The above iterative method may be implemented in the form of a numerical procedure for the parallel solution of banded linear systems. Some parallel numerical tests are also presented.