Progress in Speeding Up Multiscale Simulations of Coupled Toroidal ITG/ETG Turbulence

POSTER

Abstract

Simulating cross-scale interactions in ITG/ETG turbulence using direct gyrokinetic simulations requires extremely long computation times, posing a barrier to parameter scans and practical applications. Multi-rate algorithms offer an approach for reducing this computational load, separating the relevant differential equation into “fast” and “slow” components which can be addressed with different timesteps. MuSHrooM1 is a multiscale gyrofluid model of 2D toroidal ITG/ETG turbulence which aims to test these multi-rate algorithms. MuSHrooM, while operating as a single rate pseudo-spectral code, has performed multiscale toroidal ITG/ETG simulations using the same resources and time as GENE requires for a single scale simulation. More extensive parallelization of the FFT module of the code has provided a further 2x speedup and improves scaling to effectively utilize more processors. Using the ARKODE library, a multi-rate approach has been implemented in MuSHrooM, currently under testing to confirm the accuracy of its results against single rate operation, as well as to determine the speedup that the multi-rate stepping can provide. The verification of this multi-rate approach will provide a path for a significant speedup of standard gyrokinetic codes in multiscale simulations.

1D. R. Ernst, M. Francisquez, D. Reynolds, C. Balos, and C. Woodward, Reduced Model and Algorithmic Test-bed for Cross-Scale Interactions in Multi-Scale ITG/ETG Turbulence, Sherwood Conference, 2022.

Presenters

  • Ian Gill

    Yale University

Authors

  • Ian Gill

    Yale University

  • Darin R Ernst

    Massachusetts Institute of Technology MIT, Massachusetts Institute of Technology

  • Daniel Reynolds

    Southern Methodist University

  • Manaure Francisquez

    Princeton Plasma Physics Laboratory, Princeton Plasma Physics Laboratory (PPPL)

  • Cody J Balos

    Lawrence Livermore National Laboratory

  • Carol S Woodward

    Lawrence Livermore National Laboratory