1 of 1

New Operator Splitting Capabilities released in the SUNDIALS Library

Scientific Achievement

SUNDIALS version 7.2.0 introduced an operator splitting module with arbitrary order methods.
As a first application, we demonstrated speed and flexibility of the new operator splitting capabilities in a cloud microphysics model developed in the PAESCAL BER SciDAC.

Significance and Impact

SUNDIALS’ new splitting and forcing capabilities provide up to an 8x speedup and better asymptotic accuracy than the first order splitting commonly used in climate simulators.
This module expands SUNDIALS’ multiphysics capabilities and offers applications a stepping stone to adopting more sophisticated methods, e.g., multirate.

Technical Approach

The splitting methods provide low and high order accuracy and flexible operator evaluation order to accurately support multiphysics simulation.
We leveraged SUNDIALS’ infrastructure for vectors, (non)linear solvers, and integrators to support execution on parallel and GPU hardware.

PI: Carol S. Woodward (Lawrence Livermore National Laboratory)

Collaborating Institutions: Pacific Northwest National Laboratory

ASCR Program: SciDAC Institutes

ASCR PM: Steve Lee

Code: https://github.com/LLNL/sundials

Documentation: https://sundials.readthedocs.io/en/latest/arkode/Mathematics_link.htm

LOCAL LAB POC:

Carol S. Woodward (LLNL)

TALKING POINTS:

Operator splitting is a popular approach used to divide complex multiphysics systems into smaller sub-models that can be integrated independently
The SUNDIALS library provides advanced multiphysics time-stepping methods, and as of release 7.2.0, now includes operator splitting methods and a “forcing method” inspired by a coupling approach used in E3SM
The new methods allow flexible integration of each sub-model and are equipped to run on high performance computing systems
A primary motivation and first application of this work was a cloud microphysics model under development in the PAECSAL BER SciDAC
Initial experiments show high order operator splitting schemes from SUNDIALS attain a 0.1% error in the rain mass mixing ratio ~8x faster than a low order splitting method currently used by E3SM

METADATA:

Name of the associated awarded project: FASTMath SciDAC Institute Supplemental Effort: Operator Splitting Package for Use in SciDAC Applications with First Application to Atmospheric Physics

PI name(s): Carol S. Woodward (LLNL)

Name of the program manager: Steve Lee

CITATIONS:

Code: https://github.com/LLNL/sundials

Documentation: https://sundials.readthedocs.io/en/latest/arkode/Mathematics_link.htm

AWARDS:

If this work has received any awards at conference or otherwise, please note them here.

REPRODUCIBILITY:

If this work has received any badges or recognitions for reproducibility, please note them here. (Alternatively, can associated these with specific publications listed above)

BACKGROUND AND CONTEXT INFORMATION:

You may include further background or other information that would be helpful in understanding and delivering the full impact of this research.