Dresden 2026 – scientific programme

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

DY: Fachverband Dynamik und Statistische Physik

DY 14: Machine Learning in Dynamics and Statistical Physics II

DY 14.9: Talk

Monday, March 9, 2026, 17:15–17:30, HÜL/S186

Machine-learned Potentials for Vibrational Properties of Acene-based Molecular Crystals — •Shubham Sharma, Burak Gurlek, Paolo Lazzaroni, and Mariana Rossi — MPI for the Structure and Dynamics of Matter, Hamburg, Germany

Machine-learning potentials (MLPs) have enabled efficient modelling of complex atomistic systems with ab-initio accuracy. A major challenge, however, is the construction of sufficiently large and diverse reference datasets using first-principles calculations. To mitigate this, several active-learning strategies have been proposed to improve training efficiency, especially when combined with molecular-dynamics sampling. In this work, we develop protocols for building training sets of MACE potentials [1], targeting an accurate description of the vibrational properties of weakly-bound condensed-phase systems [2]. We assess the performance of MACE against the VASP-ML framework [3], highlighting differences in predictive accuracy for energies, forces, and vibrational properties. We also propagate committee-based uncertainties to estimate errors in dynamical quantities coming from imperfect force predictions. Finally, we demonstrate the generalisation capability of the acene-based potential by applying it to host-guest systems, enabling the identification of distinct vibrational modes within the complex dynamical spectra. [1] I. Batatia et. al., Nat Mach Intell 7, 56-67 (2025); [2] B. Gurlek, S. Sharma et. al., npj Comput Mater 11, 318 (2025); [3] R. Jinnouchi et. al., PRB 100, 014105 (2019).

Keywords: Machine-learned potential; Active learning; Uncertainty propagation; Molecular dynamics