Frontiers in Planetary Dynamos Through High-Performance Computing

English Abstract:

High-performance computing has enabled scientific breakthroughs in our understanding of planetary dynamos. While current numerical models still fall far short of reaching realistic values for the viscosity of liquid metal, the goal of producing fully turbulent dynamos may be within reach on the world’s largest computers. To achieve this goal we have undertaken a series of performance and accuracy benchmarks on a large supercomputer (XSEDE Stampede). We have used the results to identify approaches that enable numerical dynamo models to run efficiently on 10^5 compute cores. Dynamo codes included in the benchmark exercise include a variety of numerical methods (spectral, finite difference, finite element, and hybrid methods) and rely on different strategies for domain decomposition. Our analysis of the results suggests that spectral methods can achieve the required efficiency using a two-dimensional domain decomposition. Estimates for scalability of a new community dynamo code called {\it Rayleigh} are presented and used to define the parameter regime that should be feasible on present-day computers. We also describe an ongoing project through the DOE INCITE program to run planetary dynamo models on Mira, the fifth fastest computer in the world.