Abstract:The emergence of artificial intelligence has profoundly impacted computational chemistry, particularly through machine-learned potentials (MLPs), which offer a balance of accuracy and efficiency in calculating atomic energies and forces to be used in molecular dynamics simulations. These MLPs have significantly advanced molecular dynamics simulations across various applications, including large-scale simulations of materials, interfaces, and chemical reactions. Despite these advances, the construction of training datasets - a critical component for the accuracy of MLPs - has not received proportional attention. This is particularly critical for chemical reactivity which depends on rare barrier-crossing events. Here we address this gap by introducing ArcaNN, a comprehensive framework designed for generating training datasets for reactive MLPs. ArcaNN employs a concurrent learning approach combined with advanced sampling techniques to ensure accurate representation of high-energy geometries. The framework integrates automated processes for iterative training, exploration, new configuration selection, and energy and force labeling, while ensuring reproducibility and documentation. We demonstrate ArcaNN's capabilities through a paradigm nucleophilic substitution reaction in solution, showcasing its effectiveness, the uniformly low error of the resulting MLP everywhere along the chemical reaction coordinate, and its potential for broad applications in reactive molecular dynamics. We also provide guidelines on how to assess the quality of a NNP for a reactive system.

Training Machine Learning Potentials for Reactive Systems: A Colab Tutorial on Basic Models.

Machine Learning of Reactive Potentials

Tutorial: How to Train a Neural Network Potential

Developing General Reactive Element-Based Machine Learning Potentials as the Main Computational Engine for Heterogeneous Catalysis

Active learning meets metadynamics: Automated workflow for reactive machine learning potentials

chemtrain: Learning Deep Potential Models via Automatic Differentiation and Statistical Physics

Efficient Training of Neural Network Potentials for Chemical and Enzymatic Reactions by Continual Learning

Training Transferable Interatomic Neural Network Potentials for Reactive Chemistry: Improved Chemical Space Sampling

Neural network potentials for exploring condensed phase chemical reactivity

Force Training Neural Network Potential Energy Surface Models

Molecular Dynamics with Neural-Network Potentials

ArcaNN: automated enhanced sampling generation of training sets for chemically reactive machine learning interatomic potentials

Exploring the frontiers of condensed-phase chemistry with a general reactive machine learning potential

Strategies for the Construction of Machine-Learning Potentials for Accurate and Efficient Atomic-Scale Simulations

Peering inside the black box: Learning the relevance of many-body functions in Neural Network potentials

Machine learning for molecular simulation

Beyond potential energy surface benchmarking: a complete application of machine learning to chemical reactivity

Transition1x - a dataset for building generalizable reactive machine learning potentials

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials

Introduction to machine learning potentials for atomistic simulations