The unique hybrid-kinetic Vlasiator is regularly deployed on the world’s largest supercomputers and function as is the most accurate near-Earth space simulation. This study is a approach to further enhance Vlasiator’s efficiency and computational efficacy through the development and implementation of a lossy restart algorithm. Leveraging neural network-based artificial intelligence, the research is focussing on a prototype for simplifying and reconstructing velocity distribution functions (VDFs), aiming to minimize computational load, optimize memory usage, and optimize the usage of computational grants to their fullest extent. This algorithm is able to improve the time-to-solution and reduces disk usage. Furthermore, it has the potential to lead to real-time solutions based on ion-kinetic descriptions and facilitates the transition to exascale computations through its reliance on graphics processing units (GPUs).
The researching method centers the prototyping, evaluation, and implementation of a lossy restart algorithm tailored specifically for Vlasiator. Central to this approach is the utilization of neural network-based AI techniques to simplify and reconstruct velocity distribution functions (VDFs). By minimizing the computational load and memory usage, the aim is to maximize the efficient utilization of computational grants in production runs while maintaining the quality of solutions comparable to the noise-free VDFs native to Vlasiator. ASTERIX is able to enhance the time-to-solution by accelerating data processing and analysis. It reduces disk usage for checkpointing, optimizing resource allocation and enhancing computational efficiency. Moreover, the algorithm increases the resilience of simulations running on supercomputers, ensuring robustness and reliability in complex computational environments. The potential of ASTERIX to lead to an AI-based representation of velocity space opens new opportunities for real-time solutions based on ion-kinetic descriptions. The development and implementation of the code as a lossy restart algorithm represents a milestone for the near-Earth space simulation technology.