Overlapping Schwarz domain decomposition method for two-dimensional magnetotelluric forward modeling

Dan LI,Hui CHEN,Ju-zhi DENG
DOI: https://doi.org/10.6038/pg20170304
2017-01-01
Abstract:Domain decomposition algorithm can transform the large-scale problems into small problems to solve with reducing computational scale greatly and saving memory space,and has become a powerful means to solve the large-scale complex numerical problems.In this paper,the overlapping domain decomposition method (Schwarz alternating algorithm) is used for 2-D magnetotelluric (MT) forward modeling.The model domain is divided into several overlapping sub-domains.Finite difference method is used for discreting sub-domain problems,and LU-factorization is used in solving the problems of sub-domains independently.Then Schwarz alternating algorithm is applied to transmit and update solutions of overlapping domains,so as to realize 2-D MT forward numerical simulation.The typical low resistivity geoelectric model is used for numerical simulation.We compared the results with the traditional algorithm.The comparison results show that the algorithm is accurate and feasible.Memory usage and CPU time of overlapping decomposition method are decreased obviously when compared to the traditional whole domain algorithm.Furthermore,Research shows that memory usage decreases with the increasing of sub-domain number,and CPU time increases first and then decreases with the increasing of sub-domain number.The combination of overlapping sub-domains and solving methods have certain influence on the computational efficiency,which need to be optimized.Therefore,the proposed algorithm provides a new idea for forward and inverse computation of 3-D large-scale problems.
What problem does this paper attempt to address?