Accurate energy barriers for catalytic reaction pathways: an automatic training protocol for machine learning force fields

Abstract

We introduce a training protocol for developing machine learning force fields (MLFFs), capable of accurately determining energy barriers in catalytic reaction pathways. The protocol is validated on the extensively explored hydrogenation of carbon dioxide to methanol over indium oxide. With the help of active learning, the final force field obtains energy barriers within 0.05 eV of Density Functional Theory. Thanks to the computational speedup, not only do we reduce the cost of routine in-silico catalytic tasks, but also find an alternative path for the previously established rate-limiting step, with a 40% reduction in activation energy. Furthermore, we illustrate the importance of finite temperature effects and compute free energy barriers. The transferability of the protocol is demonstrated on the experimentally relevant, yet unexplored, top-layer reduced indium oxide surface. The ability of MLFFs to enhance our understanding of extensively studied catalysts underscores the need for fast and accurate alternatives to direct ab-initio simulations.

Transferable Machine Learning Interatomic Potential for Bond Dissociation Energy Prediction of Drug-like Molecules

“…learning curves of Figure ) and nonequilibrium reactive structures from along the full hydrogen abstraction path. Active learning approaches, where new structures are added based on some model uncertainty or predicted error criterion, would be helpful for faster model error convergence. ,− Active learning has already been applied to iteratively improve MACE models which were used to calculate NEB-based reaction barriers, however, in the context of heterogeneous catalysis (e.g., for hydrogenation of carbon dioxide to methanol over indium oxide) and not gas-phase small molecule reactions.…”

“…In recent years, machine learning interatomic potentials (MLIPs) have emerged as successful tools for accurately approximating ab initio potential energy surfaces at a significantly reduced computational cost: roughly milliseconds instead of minutes per single-point evaluation of a typical structure found in this work. − MLIPs have found applications in various computational chemistry problems, ranging from highly accurate spectra prediction long MD simulations of electrolytes, , path-integral MD of supramolecular complexes, to crystal structure prediction, heterogeneous catalysis, , and radical reaction networks . MLIPs offer an efficiency advantage by avoiding the need to solve the all-electron Schrödinger equation; instead, they predict energy and forces from the atom positions directly.…”

Quasi-Classical Trajectory Calculation of Rate Constants Using an Ab Initio Trained Machine Learning Model (aML-MD) with Multifidelity Data

“…There are several different ML-MD options available, each with their own ML approaches. − The open-source DeePMD-kit framework, − which utilizes deep neural networks to model the PES, was used in this work. The trained deep neural network takes atom configurations as inputs and predicts the energies and forces in the system.…”

The design and optimization of heterogeneous catalysts using computational methods

“…These workflows are automated and iterative where the training set eventually expands using different adaptive sampling and/or query strategies. 154,161 The accuracy of reaction energetics predicted via these MLFFs is claimed to be as low as 0.05 eV (ref. 161) of those obtained through density functional which is quite remarkable.…”

“…154,161 The accuracy of reaction energetics predicted via these MLFFs is claimed to be as low as 0.05 eV (ref. 161) of those obtained through density functional which is quite remarkable.…”