DATE:

Friday, September 23, 2022

TIME:

3:30 PM

LOCATION:

GMCS 314

SPEAKER:

Dr. Ronald Caplan, Computational Science, Predictive Science Inc

ABSTRACT:

Solar storms, including coronal mass ejections (CMEs), are large explosive events on the Sun that are capable of ejecting billions of tons of magnetized, million-degree plasma into space. When these CMEs reach Earth, they can disrupt radio transmissions, damage satellites, and severely impact power transmission grids, leading to extended large-scale power outages. Due to their inherent complexity, the mechanisms that control CME properties are poorly understood and challenging to model empirically. Data-constrained magnetohydrodynamic (MHD) simulations provide a powerful tool for improving our understanding of CMEs and our predictive capabilities. Such simulations must incorporate the large-scale solar corona and inner heliosphere and, therefore, require the use of multiple numerical techniques and high-performance computing platforms. In this talk, we will highlight two of these components.

First, we will discuss the use of stabilized extended Runge-Kutta methods (also known as `super time-stepping schemes’) to integrate parabolic operators (such as heat diffusion) at very large time-steps. These unconditionally stable explicit schemes are compared to an implicit Krylov subspace preconditioned conjugate gradient method. We show that they can be more scalable across large compute clusters and can be comparable or better in overall performance.

Second, we will discuss the use of accelerated computing on GPUs to allow for moderate-sized simulations to be performed on in-house workstations, as well as for very large simulations to utilize dozens of GPUs on supercomputers. The process of adapting our legacy code to run on GPUs is described, where we make use of the OpenACC programming model, as well as standard Fortran parallelism using `do concurrent’.

An example of a full Sun-to-Earth simulation of a CME is shown, utilizing our new web-based interactive interface called CORHEL-AMCG. This interface allows non-experts to set up and run solar storm simulations, and will soon be delivered to NASA’s Community Coordinated Modeling Center for public use.

HOST:

Jose Castillo

DOWNLOAD: