On-the-fly Singular Value Decomposition for Aitken's Acceleration of the Schwarz Domain Decomposition Method

This paper discusses the low-rank approximation of the Aitken's acceleration of the convergence of the Schwarz domain decomposition method. The singular value decomposition (SVD) of the solution on interfaces at successive Schwarz iterations gives a basis of a low-dimensional space. The error operator of the Schwarz process, involved in the Aitken's formula, is approximated in this space. The efficiency of the Aitken's acceleration depends on the quality of this basis to approximate the exact solution, which is related to the number of iterations involved in the SVD. We use an iterative procedure to compute the SVD simultaneously with the Schwarz process in order to perform the Aitken's acceleration when the SVD satisfies a mathematical criterion.