Space-time parallel iterative solvers for the integration of parabolic problems

Abstract

In view of the existing limitations of sequential computing, parallelization has emerged as an alternative in order to improve the speedup of numerical simulations. In the framework of evolutionary problems, space-time parallel methods offer the possibility to optimize parallelization. In the present paper, we propose a new family of these methods, built as a combination of the well-known parareal algorithm and suitable splitting techniques which permit us to parallelize in space. In particular, dimensional and domain decomposition splittings are considered for partitioning the elliptic operator, and first-order splitting time integrators are chosen as the propagators of the parareal algorithm to solve the resulting split problem. The major contribution of these methods is that, not only does the fine propagator perform in parallel, but also the coarse propagator. Unlike the classical version of the parareal algorithm, where all processors remain idle during the coarse propagator computations, the newly proposed schemes utilize the computational cores for both integrators. A convergence analysis of the methods is provided, and several numerical experiments are performed to test the solvers under consideration.