Multiscale simulation of plasma flows using active learning.

Area of Science:

• Computational Physics
• Plasma Physics
• Multiscale Modeling

Background:

• High-energy-density experiments involve complex plasma flows deviating from equilibrium conditions.
• Kinetic theory (KT) models these phenomena but requires closure information from microscale simulations.
• Traditional methods for obtaining this information are computationally expensive.

Purpose of the Study:

• To develop a computationally efficient multiscale approach coupling KT with molecular dynamics (MD).
• To reduce the computational cost associated with gathering closure information from MD simulations.
• To enable the study of plasma interfacial mixing and Coulomb coupling physics in warm dense matter.

Main Methods:

• A concurrent heterogeneous multiscale approach coupling KT with MD for near-equilibrium flows.
• Active learning is employed to train neural networks on a small subset of MD data, reducing computational overhead.
• The method is applied to a plasma interfacial mixing problem relevant to warm dense matter.

Main Results:

• Demonstrated considerable computational gains compared to the full kinetic-MD approach.
• Successfully applied the method to a relevant plasma interfacial mixing problem.
• Achieved significant reductions in the cost of gathering information from MD simulations.

Conclusions:

• The developed multiscale approach offers a computationally efficient way to simulate nonequilibrium plasma flows.
• This method enables the exploration of Coulomb coupling physics across a wide range of temperatures and densities.
• The findings provide a pathway to investigate phenomena previously inaccessible with current theoretical models.