Regensburg 2025 – scientific programme

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

KFM: Fachverband Kristalline Festkörper und deren Mikrostruktur

KFM 14: Poster

KFM 14.1: Poster

Wednesday, March 19, 2025, 17:00–18:30, P1

Stability of Machine-Learned Interatomic Potentials in Molecular Dynamics Simulations for Organic Semiconductors and Metal-Organic Frameworks — •Martin Tritthart and Egbert Zojer — Institute for Solid State Physics, Graz, Austria

Organic semiconductors (OSCs) and metal-organic frameworks (MOFs) are two classes of materials that have garnered significant interest in materials science. To optimize their performance, it is crucial to understand the physical properties of these crystalline polymer materials, such as heat transport and mechanical stability. This understanding can lead to improvements in properties like thermal stability. Molecular dynamics (MD) simulations are commonly used for this purpose, as they are orders of magnitude less computationally expensive than first-principles calculations. While machine learning interatomic potentials (MLIPs) are much faster than ab initio methods, they approximate the true potential energy surface, which can result in significant errors for atomic configurations outside the training data space. Such shortcomings lead to incorrect predictions of forces and energies in MD simulations, potentially causing molecular instability during simulations. To address this issue and improve the robustness of MLIPs, a reliable estimation of their uncertainty is necessary. This enables the identification of uncertain structures, which can then be incorporated into the training set to enhance accuracy. With this iterative approach, larger and more complex molecules can be simulated with relatively efficient computational effort.

Keywords: Machine learned interatomic potential; Ab-initio