Parallel computations for solving 3D Helmholtz problem by using direct solver with low-rank approximation and HSS technique
(ИНГГ СО РАН)
(ИНГГ СО РАН)
дата публикации: 2017
The modern methods of processing the geophysical data, such as Reverse Time Migration (RTM) and Full Waveform Inversion (FWI) require solving series of forward problems where the main step is solution of Systems of Linear Algebraic Equations (SLAE) of big size. For big sizes, it is time and memory consuming problem. In this paper, we present a parallel direct algorithm to solve boundary value problems for 3D Helmholtz equation discretized with help of finite differences. The memory consumption has been resolved due to Nested Dissection approach, low-rank approximation technique and HSS format. OpenMP parallelization is based on standard BLAS and LAPACK functionality. For MPI parallelization, we propose a novel algorithm that uses dynamical distribution of the elimination tree nodes across cluster nodes. Numerical experiments show performance benefits of the proposed cluster algorithm compared to the not parallel version and demonstrate significant memory advantages over direct solvers without low-rank approximation.
первоисточник: Numerical Analysis and Its Applications. 6th International Conference, NAA 2016 (Lozenetz, Bulgaria, June 15-22, 2016): Revised Selected Papers. (Lecture Notes in Computer Science 10187)