Efficient Application of Low Mach Number Hydrodynamics Code to Stellar Flows
Authors: Duoming Fan (Lawrence Berkeley National Laboratory)
Abstract: Astrophysical phenomena that occur in the low Mach number regime are often computationally expensive because the time step is constrained by both the characteristic fluid velocity and, by comparison, the much larger speed of sound. In addition, astrophysical flows are generally highly turbulent, and considerable computational cost is required to resolve the local flow in smaller regions of interest. We introduce MAESTROeX, a low Mach number hydrodynamics code for computing stellar flows that uses the AMReX C++/F90 libraries and software structure. By filtering out the acoustic waves from the governing equations, MAESTROeX uses a time step constraint that is based only on the fluid velocity rather than the sound speed. It also incorporates adaptive mesh refinement (AMR) to efficiently increase spatial resolutions by locally refining the grid in the regions of interest. Additionally, to allow the proper capture of the effects of an expanding star, MAESTROeX uses a novel one-dimensional radial base state whose evolution is coupled to the evolution of the full state. In our latest attempt to improve the local hydrodynamic equilibrium errors caused by the radial base states, the base states are now computed at uneven intervals as to directly map to the Cartesian grid. The performance and scalability of MAESTROeX are evaluated on NERSC and OLCF systems scalable to over 100K cores. Future work includes improving the current algorithm and offloading simple subroutines to GPUs to increase computational efficiency, and incorporating new physics into the system such as those needed for rotating stars.
Back to Women in HPC: Diversifying the HPC Community Archive Listing
Back to Full Workshop Archive Listing