The MLS group has published a new paper in the top journal Nature NPJ Computational Materials in collaboration with NEC Laboratories Europe and Prof. Johannes Kaestner.

May 7, 2024

The work proposes and analyzes methods for intelligent data generation for machine learning models for atomistic systems. These methods are essential for accelerating molecular simulations and can be used to explore new materials and drugs. The SimTech cluster of excellence facilitates this collaboration, which highlights the cluster’s interdisciplinary character, bringing together researchers from computer science, computational chemistry, and industry labs.

Uncertainty-biased molecular dynamics for learning uniformly accurate interatomic potentials

Authors: Viktor Zaverkin, David Holzmüller, Henrik Christiansen, Federico Errica, Francesco Alesiani, Makoto Takamoto, Mathias Niepert, Johannes Kästner


Efficiently creating a concise but comprehensive data set for training machine-learned interatomic potentials (MLIPs) is an under-explored problem. Active learning, which uses biased or unbiased molecular dynamics (MD) to generate candidate pools, aims to address this objective. However, existing biased and unbiased MD-simulation methods are prone to miss either rare events or extrapolative regions—areas of the configurational space where unreliable predictions are made. This work demonstrates that MD, when biased by the MLIP’s energy uncertainty, simultaneously captures extrapolative regions and rare events, which is crucial for developing uniformly accurate MLIPs. Furthermore, exploiting automatic differentiation, we enhance bias-forces-driven MD with the concept of bias stress. We employ calibrated gradient-based uncertainties to yield MLIPs with similar or, sometimes, better accuracy than ensemble-based methods at a lower computational cost. Finally, we apply uncertainty-biased MD to alanine dipeptide and MIL-53(Al), generating MLIPs representing both configurational spaces more accurately than models trained with conventional MD.

Link to paper:

To the top of the page