AUTHOR=Karimabadi Homa , Wilkes Jason , Roberts D. Aaron 

TITLE=The need for adoption of neural HPC (NeuHPC) in space sciences

JOURNAL=Frontiers in Astronomy and Space Sciences

VOLUME=Volume 10 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/astronomy-and-space-sciences/articles/10.3389/fspas.2023.1120389

DOI=10.3389/fspas.2023.1120389

ISSN=2296-987X

ABSTRACT=This paper advocates for the use and combining modern artificial intelligence (AI) approaches and HPC, which we refer to as NeuHPC, for space plasma simulations.   AI is poised to transform all aspect of HPC pipeline, from knowledge discovery from massive datasets to computational steering, to new approaches to solving PDEs such as derivation of closure models from data, learning based techniques such as zero-shot super resolution, improvements in existing solvers, among others.  While AI techniques for data analysis (e.g., classification, image segmentation, detection, tracking) are in a more mature state, application of AI to solving PDEs is in a nascent stage, and a wide variety of solutions are being explored across different disciplines.  Given the pace and extent of research in this area across multiple disciplines, any review article is expected to soon become outdated and is also beyond the scope of a perspective paper. Rather, here our goal is to bring attention to some of the most promising recent algorithmic developments using AI in addressing extreme scales in simulations. Many of these works have used CFD as their testbed and their applicability to space plasma simulations remains to be determined. As such, we propose several proof-of-concept studies involving fluid and kinetic plasma simulations as a starting point for further exploration. This exposition is not meant to be comprehensive and important topics such as inverse problem or mesh generation are not covered.