High-pressure and temperature neural network reactive force field for energetic materials

Abstract

Reactive force fields for molecular dynamics have enabled a wide range of studies in numerous material classes. These force fields are computationally inexpensive compared with electronic structure calculations and allow for simulations of millions of atoms. However, the accuracy of traditional force fields is limited by their functional forms, preventing continual refinement and improvement. Therefore, we develop a neural network-based reactive interatomic potential for the prediction of the mechanical, thermal, and chemical responses of energetic materials at extreme conditions. The training set is expanded in an automatic iterative approach and consists of various CHNO materials and their reactions under ambient and shock-loading conditions. This new potential shows improved accuracy over the current state-of-the-art force fields for a wide range of properties such as detonation performance, decomposition product formation, and vibrational spectra under ambient and shock-loading conditions.

Surface Melting Promoted by Hydrogenation: Guidelines for Superhydride Synthesis

“…Machine learning potentials were constructed using DeepMD-Kit (25,35). The mean absolute error (MAE) of atomic forces between DFT and MLP calculations was around 0.3 eV/Å for training and testing data below 3000 K and 50 GPa (see supplemental materials for detailed discussion), which is comparable to those in previous studies (29,33). In addition, local structures of the CaH2/H2 interface during the MLP-MD simulation described below were consistent with those of bulks, such as CaH2, CaH4, and H2, showing that these MD simulations were performed in the interpolated region of the constructed MLP or close to the interpolated region (Figs.…”

“…The recent developments of machine-learning potentials (23)(24)(25)(26)(27) have accelerated and facilitated the potential construction of new materials. These developments have enabled MD simulations to be applied to a wider variety of materials (28)(29)(30)(31)(32)(33)(34). In fact, machine-learning potential (MLP) MD simulations have accurately reproduced the phase transition behavior of high-pressure H2, which is strongly affected by the simulation cell size (29).…”

Multiscale Reactive Model for 1,3,5-Triamino-2,4,6-trinitrobenzene Inferred by Reactive MD Simulations and Unsupervised Learning

“…This force field has been previously employed to study shock initiation in TATB, , as well as to widely investigate the thermochemistry of various energetic materials at extreme conditions of temperature and pressure. , We, however, emphasize the inherent limitations of using a classical reactive force field for which the underlying functional forms limit its accuracy. Machine learning potentials offer a promising alternative to classical reactive force fields for the decomposition of high-energy materials, , but our primary aim is to develop and validate a multiscale modeling approach. All the MD simulations were performed using the parallel MD software LAMMPS from Sandia National Laboratories , with 3D periodic boundary conditions.…”

Molecular Modeling of Stabilization during Processing of Polyacrylonitrile-Based Carbon Fibers

“…Electronic-structure methods can describe chemical reactions using quantum mechanics but are too computationally intensive to describe the complex molecular structure of condensed phase polymers. Reactive force fields, such as the ReaxFF and neural network reactive force field (NNRF), , can also describe chemistry at a fraction of the cost of electronic structure calculations. This method has been used to describe complex chemistry under extreme conditions − and the carbonization step in the processing of CFs. − The main drawback of this approach is that the simulations are restricted to the nanosecond regime; orders of magnitude shorter than those characteristic of stabilization.…”