Adaptation of parallel in time integration algorithm to groundwater simulations

Abstract

Computation performance has rapidly increased by equipping a large number of cores. To use effectively these high-performance computers, the domain decomposition method, which divides a space domain into sub-domains, has been widely employed in large-scale simulations, in which each core carries out parallel computing. On the other hand, the parallel in time integration has been proposed to further utilize massive cores in high performance computer more efficiently. Although theoretical studies on the method have been reported, its adaptability for engineering applications has still been under investigation. To investigate this, the parareal method, one of the methods of the parallel in time integration methods, is adapted into a general-purpose groundwater simulator TOUGH2 in this study. The developed code is applied to groundwater simulations with water injection in a radially symmetrical model. The results show that the parareal method can reduce the computation time down to one-eighth, compared with that from serial in time method.