Speaker
Description
Forecasting and understanding space weather remains a fundamental challenge due to the inherently multi-scale nature of plasma dynamics in the Sun-Earth connection. Traditional first-principles models like fully kinetic Particle-in-Cell (PIC) simulations are highly accurate but computationally prohibitive for operational or ensemble forecasting. In this work, we present a novel hybrid modelling strategy that bridges this gap by combining the rigor of physics-based simulation with the efficiency of machine-learned surrogates.
We focus on turbulent magnetosheath, a key region controlling solar wind energy transfer into the magnetosphere. We develop and validate a neural surrogate model which consists of physics based equations for ions and a neural surrogate for modelling pressure response of the electrons. This neural surrogate component which represents electron pressure dependence on lower order moments is trained on fully kinetic PIC simulations performed with ECsim (Energy Conserving Semi-Implicit PIC code). This represents the first instance of online (a posteriori) coupling and testing between a physics-based kinetic ion model and data-driven fluid electron closure that retains dynamic consistency implemented in efficient hybrid PIC code Menura that is ported on a GPU.
Our results demonstrate that the surrogate model generalizes across multiple ECsim runs. As a result, it contributes to the broader agenda of physics-informed AI, showcasing hybrid physics-AI methodologies for space weather. This parametrization of unresolved scales provides a surrogate that can capture complex plasma behavior, opening possibilities for computationally efficient high fidelity models, which is necessary for future operational viability of such space weather forecasting frameworks.
| Do you plan to attend in-person or online? | In-person |
|---|
