GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations

Abstract

We present our latest advancements of machine-learned potentials (MLPs) based on the neuroevolution potential (NEP) framework introduced in [Fan et al., Phys. Rev. B 104, 104309 (2021)] and their implementation in the open-source package GPUMD.We increase the accuracy of NEP models both by improving the radial functions in the atomic-environment descriptor using a linear combination of Chebyshev basis functions and by extending the angular descriptor with some four-body and five-body contributions as in the atomic cluster expansion approach.We also detail our efficient implementation of the NEP approach in graphics processing units as well as our workflow for the construction of NEP models, and we demonstrate their application in large-scale atomistic simulations.By comparing to state-of-the-art MLPs, we show that the NEP approach not only achieves above-average accuracy but also is far more computationally efficient.These results demonstrate that the GPUMD package is a promising tool for solving challenging problems requiring highly accurate, large-scale atomistic simulations.To enable the construction of MLPs using a minimal training set, we propose an active-learning scheme based on the latent space of a pre-trained NEP model.Finally, we introduce three separate Python packages, GPYUMD, CALORINE, and PYNEP, which enable the integration of GPUMD into Python workflows.

Quantum-corrected thickness-dependent thermal conductivity in amorphous silicon predicted by machine learning molecular dynamics simulations

“…However, thermal transport usually involves large length and long timescales and GAP18 is not currently efficient enough for this purpose. The NEP model as implemented in the GPUMD package [27,28], on the other hand, can reach a computational speed of about 5 × 10 6 atom step per second for a-Si by using a single GPU card such as Tesla V100, which is about three orders of magnitude faster than GAP18 using 72 Xeon-Gold 6240 central processing unit (CPU) cores [18].…”

Investigation of the mechanical and transport properties of InGeX3 (X = S, Se and Te) monolayers using density functional theory and machine learning

“…U 0 (n), F 0 i and W 0 mn (n) are the corresponding quantities calculated from first-principles calculations based on DFT, respectively, and are also considered as target values for machine learning. In this study, NEP generation and related machine learning simulations (phonon dispersion, mechanical behavior and lattice thermal conductivity) are implemented in the open-source GPUMD-3.3.1 package 45,46 The training data are prepared using the first-principles calculations program in the QUANTUM ESPRESSO 6.8 package. 42 AIMD within the NVT ensemble is applied to produce atomic displacement, which is performed in a supercell containing 40 atoms.…”

“…The NEP still needs to be further developed for complex systems. 45 So far, the HNEMD method has been widely used to investigate the heat transport properties of materials, in which the heat flow circulates under a driving force. To eliminate the size effect in the thermal conductivity simulations, a big rectangular supercell is established, as shown in Fig.…”

“…U 0 ( n ), F 0 i and W 0 μν ( n ) are the corresponding quantities calculated from first-principles calculations based on DFT, respectively, and are also considered as target values for machine learning. In this study, NEP generation and related machine learning simulations (phonon dispersion, mechanical behavior and lattice thermal conductivity) are implemented in the open-source GPUMD-3.3.1 package 45,46…”

Atomistic modeling of the mechanical properties: the rise of machine learning interatomic potentials

“…MLIPs are commonly trained using DFT-based datasets, and their performance with respect to flexibility and accuracy 15 is close to that of the DFT method, but with substantially accelerated computational time due to bypassing the expensive and also unnecessary electronic structure calculations. MLIPs nowadays can be directly employed to conduct molecular dynamics simulations over highly computationally efficient massively parallel processors and graphics processing units, using various packages, such as LAMMPS, 16 TorchMD, 17 GPUMD 18 and MLatom. 17…”