The Fourth Grand Challenge
The solar wind, a continuous stream of plasma emanating from the Sun, interacts with the Earth’s magnetosphere resulting in adverse space weather. During geomagnetic storms, systems like power grids and satellite positioning systems are at risk due to enhanced particle radiation and disturbances of the geomagnetic field, potentially leading to large-scale blackouts and global disruption of communications. The strongest historical storms even resulted in aurorae visible in the Caribbean.
The grand challenge in predicting space weather risks and modelling powerful geomagnetic storms is to encompass the whole of near-Earth space with an ion-kinetic description, while simulating the whole system for the duration of a typical geomagnetic storm, lasting from a few hours to several days, with up to extreme driving conditions (order-of-magnitude increases in solar wind speed, density, and magnetic field magnitude). To properly model the geomagnetic response, the ion-kinetic plasma model must be coupled to a model capable of describing ionospheric currents and feedback.
Vlasiator is the world’s first simulation model using the 6D Vlasov equation to model the Earth’s plasma environment in its global context. Vlasiator has been developed with two European Research Council grants, the starting grant 200141-QuESpace and consolidator grant 682068-PRESTISSIMO, and has been used to pilot several supercomputers such as Sisu, Mahti, and LUMI-C at CSC, Finland. Vlasiator is also the first global kinetic model of the magnetosphere coupling to a realistic model of the ionised upper-atmosphere layers called the ionosphere, as illustrated in the figure.
Vlasiator uses the hybrid-kinetic approach, modelling ions through their particle distribution functions (in 3D-3V) and electrons as a massless fluid. This improves significantly over the older standard fluid-like models such as magnetohydrodynamics (MHD) and does not suffer from statistical sampling noise present in particle-in-cell (PIC) models. The use of hybrid parallelism, efficient AVX-vectorised operations, adaptive mesh refinement (AMR) for the 3D spatial domain and a sparse representation for the 3D velocity space have made limited 6D geospace simulations possible, capable of resolving mesoscale dynamics. A typical Vlasiator run consists of an order of 105 timesteps for an order of 1011 phase-space cells with outputs requiring tens of terabytes of disk space. The figure-of-merit (FOMVlasiator) performance metric (timestep * memory bandwidth / bytes of data to propagate) for current state-of-the-art Vlasiator runs is ~700.
How can exascale simulations solve this grand challenge? The grand challenge of simulating the whole of near-Earth space under space storm conditions for hours and at ion-kinetic resolutions in regions of interest requires the advent of exascale computational capacities for several reasons. The increase in total simulated time (order of 107 timesteps) and the increased number of sample points on the AMR mesh (order of 1013 phase-space cells), are to be addressed in part by algorithmic improvements in load balancing and adaptive time-stepping. The major leap ahead will be taken by porting all solvers and memory use to modern accelerators such as GPUs. Vlasiator has led the effort in global kinetic geospace modelling for a decade and is uniquely poised to take advantage of exascale resources in making this a reality.