Collaborative Research: Enabling Multi-Scale Studies of Magnetic Reconnection with Interpretable Data-Driven Models

This project will explore the multiscale physics of magnetic energy release in plasmas. Most of the visible matter in the universe is in the state of plasma and is magnetized. The magnetic energy stored in the plasma can be explosively released by magnetic reconnection –– a fundamental process that plays a key role in laboratory and astrophysical systems, from disruptions in fusion experiments, to spectacular solar flare events, to, potentially, the acceleration of very high-energy cosmic rays.

The understanding of magnetic reconnection is challenging due to the complex interplay of different processes at many scales: from detailed physics of electron motion at very small scales, to plasma heating and flow generation at large scales, to high energy photon and particle acceleration that can carry away a large part of the available plasma energy. The goal of this project is to use machine learning techniques to unravel the connection between physics processes at small and large scales, and develop better multi-scale models of magnetic reconnection.

In doing so, it will contribute to the goals of NSF's "Windows on the Universe: The Era of Multi-Messenger Astrophysics" Big Idea. The project will provide students and postdocs, including those from traditionally under-represented groups, with advanced training in basic plasma physics, computational physics, and machine learning, empowering them with a unique set of tools to address emerging scientific opportunities.

The holistic understanding of magnetic reconnection requires the development of new coarse-grained models that can describe the macroscopic consequences of the essential kinetic physics of reconnection and particle acceleration. This is often referred to as the problem of finding good "closures"; that is, a reduced set of equations that capture the essential processes occurring on unresolved scales as a function of resolved quantities, and that can be solved in a computationally efficient way.

Techniques from the field of machine learning are providing unique opportunities to harness the increasingly abundant data from experiments and high-fidelity simulations to accelerate the development of the required reduced physics models. The goal of this project is to develop and apply novel machine learning tools based on sparse and symbolic regression techniques to extract interpretable and generalizable reduced models from data of first-principles plasma simulations.

Specifically, the project aims to construct better kinetic closures for magnetic reconnection; to derive better models of particle injection and acceleration by this fundamental plasma process; and to use this understanding to accelerate the development of multi-scale plasma algorithms. While the immediate focus will be on the problem of magnetic reconnection, the tools that will be developed are general and applicable to other areas of plasma physics, and more broadly to many-body phenomena.

The development of these multiscale models can have a significant impact across different areas of plasma science, from fusion to space and astrophysical plasmas.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.