DeePMD-kit v2: A software package for deep potential models

Abstract

DeePMD-kit is a powerful open-source software package that facilitates molecular dynamics simulations using machine learning potentials known as Deep Potential (DP) models. This package, which was released in 2017, has been widely used in the fields of physics, chemistry, biology, and material science for studying atomistic systems. The current version of DeePMD-kit offers numerous advanced features, such as DeepPot-SE, attention-based and hybrid descriptors, the ability to fit tensile properties, type embedding, model deviation, DP-range correction, DP long range, graphics processing unit support for customized operators, model compression, non-von Neumann molecular dynamics, and improved usability, including documentation, compiled binary packages, graphical user interfaces, and application programming interfaces. This article presents an overview of the current major version of the DeePMD-kit package, highlighting its features and technical details. Additionally, this article presents a comprehensive procedure for conducting molecular dynamics as a representative application, benchmarks the accuracy and efficiency of different models, and discusses ongoing developments.

Understanding the Anomalous Diffusion of Water in Aqueous Electrolytes Using Machine Learned Potentials

“…To this end, two MLP force fields, one each for NaCl and CsI system, were constructed using the deep potential (DP) framework − trained on condensed phase DFT data using the revPBE-D3 functional. , A comparison of several thermodynamic, structural, and transport properties computed from DP based MD simulations (DPMD) with corresponding experimental data demonstrates its efficacy (Figure , Table , and SI Section S2). Figure a shows the temperature dependence of the density of liquid water and ice.…”

Data Efficient and Stability Indicated Sampling for Developing Reactive Machine Learning Potential to Achieve Ultralong Simulation in Lithium-Metal Batteries

“…The two methods have been proved to be accurate and efficient for modeling a diverse range of chemical systems and properties, including alloys, NMR shift of paramagnetic materials, and organic–inorganic perovskites . In particular, their compatibility have been proved for those systems that are similar to this study, i.e., complex reaction networks of bond-breaking and bond-forming events and multicomponent battery systems. ,,,– In Note S1, we briefly introduce related theories of DeepMD and DP-GEN, and a more comprehensive introduction of them and MLPs can be found in refs , , , , and .…”

Multiscale Modeling of Aqueous Electric Double Layers

“…An added complication is that whereas the optimization targets for the ion–water and ion–ion interactions are clear, experimental reference quantities for the parametrization of force fields for the substrates are less readily available. If it turns out to be necessary, alternative force field parametrizations in regions of varying dielectric properties and near interfaces are being developed. , In the past decade, new simulation methods based on machine-learned (ML) potentials developed rapidly because they reach DFT accuracy at a fraction of the computational costs. − Originally, ML potentials were purely short-range, but recent extensions have made them also able to cover long-range effects, such as electrostatics, − which are crucial for simulating the electric double layer. First studies already show that ML potentials mitigate some of the force field accuracy problems and DFT MD time-length scale issues of aqueous systems. , Similarly to force field parametrizations, however, the choice of exchange and correlation functionals when using DFT MD simulations is essential for the success of the approach reviewed here. , …”