Neural network reactive force field for C, H, N, and O systems

Abstract

Reactive force fields have enabled an atomic level description of a wide range of phenomena, from chemistry at extreme conditions to the operation of electrochemical devices and catalysis. While significant insight and semi-quantitative understanding have been drawn from such work, the accuracy of reactive force fields limits quantitative predictions. We developed a neural network reactive force field (NNRF) for CHNO systems to describe the decomposition and reaction of the high-energy nitramine 1,3,5-trinitroperhydro-1,3,5-triazine (RDX). NNRF was trained using energies and forces of a total of 3100 molecules (11,941 geometries) and 15 condensed matter systems (32,973 geometries) obtained from density functional theory calculations with semi-empirical corrections to dispersion interactions. The training set is generated via a semi-automated iterative procedure that enables refinement of the NNRF until a desired accuracy is attained. The root mean square (RMS) error of NNRF on a testing set of configurations describing the reaction of RDX is one order of magnitude lower than current state of the art potentials.

Progress towards machine learning reaction rate constants

“…Recently, Yoo et al 115 used a DNN to generate a reactive forcefield for C, H, N and O systems. Their train and test dataset consisted in DFT/PBE-D2 computed energies for 3100 molecular structures part of the decomposition of crystalline nitramine RDX.…”

Molecular Forcefield Methods for Describing Energetic Molecular Crystals: A Review

“…The higher accuracy of reactive FFs is pursued for various molecular crystals, as well as extension to more elements [ 137 ]; thus, ReaxFF can be continuously improved using more quantum-mechanics-based calculations [ 138 , 139 ] and even machine learning methods [ 140 , 141 , 142 ]. Yoo et al [ 143 ] have developed a neural network reactive forcefield (NNRF) for studying the reactivity of RDX crystals under a wide range of temperatures, showing higher accuracy for predicting vibration spectra, thermal decomposition, and reactions ( Figure 6 ).…”

“… Fitting procedure of NNRF ( a ) and the predicted infrared spectra of RDX using NNRF ( b ). Reprinted by permission from [ 143 ]; copyright 2021 Springer Nature Limited. Permission conveyed through Copyright Clearance Center, Inc. …”

Active learning and molecular dynamics simulations to find high melting temperature alloys

“…AL has been used for a wide range of applications from natural language processing, 16 reaction screening for pharmaceutical applications, 1 and multiscale modeling. 2,17 In materials science, AL has been used to accelerate discovery of materials with desired properties by coupling it with experiments 4,[18][19][20][21][22] and physics-based simulations. [5][6][7] In addition, AL workflows paired with existing closed data sets has shown the ability of these models to reduce the number of queries needed to identify the best candidate.…”

“…Examples range from crystal plasticity models 10,24 to interatomic potentials from ab-initio simulations. 17,[25][26][27][28][29] Physics-based materials models across scales open the possibility of significantly expanding the reach of AL approaches.…”