# Data-driven construction of a generalized kinetic collision operator from molecular dynamics

Source: [Data-driven construction of a generalized kinetic collision operator from molecular dynamics](<https://www.sota2.com/research/paper/68fa49db1e209131ce93a147>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T07:31:36.980Z

Authors: Yue Zhao, Joshua W. Burby, Andrew Christlieb, Huan Lei

Published: 2025-03-31T15:26:06.000Z

## Abstract

We introduce a data-driven approach to learn a generalized kinetic collision operator directly from molecular dynamics. Unlike the conventional (e.g., Landau) models, the present operator takes an anisotropic form that accounts for a second energy transfer arising from the collective interactions between the pair of collision particles and the environment. Numerical results show that preserving the broadly overlooked anisotropic nature of the collision energy transfer is crucial for predicting the plasma kinetics with non-negligible correlations, where the Landau model shows limitations.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2503.24208>)

[PDF](<https://arxiv.org/pdf/2503.24208.pdf>)