Dresden 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 9: Topical Session: Physics-driven Artificial Intelligence for Materials II

MM 9.8: Talk

Monday, March 9, 2026, 18:00–18:15, SCH/A251

Integrating FlashMD into LAMMPS for Efficient Long-Timestep Molecular Dynamics — •Johannes Spies, Filippo Bigi, and Michele Ceriotti — Laboratory of Computational Science and Modeling, Institut des Matériaux, École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland

FlashMD [1] is a machine-learning model that predicts future molecular configurations from the current state, reducing the need to call machine-learning interatomic potentials at every timestep and enabling larger effective integration steps.

I present its integration into LAMMPS. The implementation acts as a drop-in replacement for standard integrators and makes FlashMD directly available to the molecular simulation community. The interface is modular and extensible through the metatomic ecosystem, allowing new predictor models to be added with minimal effort.

The contribution outlines the integration strategy and initial performance results, focusing on usability, extensibility, and compatibility with existing MLIP workflows. The approach provides a practical route to accelerating large-scale atomistic simulations by reducing the frequency of expensive potential evaluations while maintaining physical reliability.

[1] Filippo Bigi, Sanggyu Chong, Agustinus Kristiadi & Michele Ceriotti, arXiv:2505.19350 (2025).

Keywords: machine learning; molecular dynamics; simulation