guest ::
| Home > Publications database > Adaptive-precision interatomic potentials for large-scale atomistic simulations |
| Home > Publications database > Adaptive-precision interatomic potentials for large-scale atomistic simulations |
| Book/Dissertation / PhD Thesis | FZJ-2026-02758 |
2026
Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag
Jülich
ISBN: 978-3-95806-935-0
This record in other databases:
Please use a persistent id in citations: doi:10.34734/FZJ-2026-02758
Abstract: Large-scale atomistic simulations rely on interatomic potentials providing an efficient representation of atomic energies and forces. Modern machine-learning (ML) potentials provide the most precise representation compared to electronic structure calculations while traditional empirical potentials provide a less precise, but computationally much faster representation and, thus, allow simulations of larger systems. In the present work, a novel method is developed to combine a traditional and a ML potential to a multi-resolution description, leading to an adaptive-precision interatomic potential (APIP) with an optimum of performance and precision in large complex atomistic systems. This new method is, in contrast to previous methods, appropriate for microcanonical ensembles as both momentum and energy are conserved by the energy-based coupling approach. The required precision is determined per atom by a local structure analysis and updated automatically during simulation. The approach is implemented as APIP package for the molecular-dynamics simulator LAMMPS and includes a per-atom computational load estimation so that a load-balancer can prevent computational-time problems arising due to the atom dependent force-calculation times, which makes APIP suitable for large-scale atomistic simulations. The approach is demonstrated in nanoindentation simulations for both the prototypical facecentred cubic metal copper and the body-centred cubic metal tungsten: a computationally efficient embedded atom method (EAM) potential is combined with a precise but computationally less efficient ML potential based on the atomic cluster expansion (ACE) into an adaptive-precision interatomic potential tailored for the nanoindentation. The numerically more expensive ACE potential is employed selectively only in regions of the computational cell where high precision is required. The comparison with pure EAM and pure ACE simulations shows that for copper, all potentials yield similar dislocation morphologies under the indenter with only small quantitative differences. In contrast, markedly different plasticity mechanisms are observed for tungsten in simulations performed with the central-force EAM potential compared to results obtained using the many-body ACE potential, that is able to describe accurately angular characteristics of bonding. All ACE-specific mechanisms are reproduced in the APIP nanoindentation simulations, however, with a significant speedup of 20-30 times compared to the pure ACE simulations. Hence, adaptiveprecision interatomic potentials overcome the performance gap between the precise ACE and the fast EAM potential by combining the advantages of both potentials. Finally, the novel energy and momentum conserving and dynamically self-adapting coupling approach is further improved by local averaging to a conservative adaptive-precision potential, which, by design, guarantees both energy and momentum conservation and is also included in the APIP LAMMPS package. As the force precision improves with the averaging radius, force-coupling approaches, which usually cannot define a potential energy, are not required any more in order to obtain accurate forces on specific atoms. Hence, conservative adaptive-precision potentials can be used regularly in the future – without the currently existing problems caused by the violation of conservativity. By coupling a fast EAM tungsten potential to a highly accurate ACE tungsten potential, the conservation properties are verified numerically and it is show that one can achieve – dependent on both the potential and the atomistic system – a speedup of one or two orders of magnitude compared to a pure ACE simulation.
; 
|
The record appears in these collections: |
PDF