Parallel-in-Time with SUNDIALS and XBraid

David Gardner (Lawrence Livermore National Laboratory)

SUNDIALS is a suite of robust and scalable integrators and solvers for systems of ordinary differential equations, differential-algebraic equations, and nonlinear equations used in numerous applications codes for research and industry. The suite consists of six packages, CVODE(S), ARKode, IDA(S), and KINSOL, all built on shared vector, matrix, linear solver, and nonlinear solver APIs allowing for user-defined/application-specific data structures and solvers, encapsulated parallelism, and algorithmic flexibility.

Presently the numerical methods in SUNDIALS utilize sequential time marching schemes with parallelization only in the spatial dimension. Parallel-in-time methods introduce an additional dimension of parallelism to better leverage the increased concurrency available on massively parallel systems. In particular recent work utilizing multigrid-reduction-in-time (MGRIT) has shown significant speedups over sequential time stepping and can be implemented in a non-intrusive manner. In this talk we will present results from recent efforts to combine the adaptive-step explicit, implicit, and IMEX time integration methods from the SUNDIALS ARKode library with the XBraid MGRIT library to provide parallel-in-time integration with SUNDIALS.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-ABS-810233