A Bayesian Committee Machine Potential for Organic Nitrogen Compounds (2402.17132v1)

Abstract: Large-scale computer simulations of chemical atoms are used in a wide range of applications, including batteries, drugs, and more. However, there is a problem with efficiency as it takes a long time due to the large amount of calculation. To solve these problems, machine learning interatomic potential (ML-IAP) technology is attracting attention as an alternative. ML-IAP not only has high accuracy by faithfully expressing the density functional theory (DFT), but also has the advantage of low computational cost. However, there is a problem that the potential energy changes significantly depending on the environment of each atom, and expansion to a wide range of compounds within a single model is still difficult to build in the case of a kernel-based model. To solve this problem, we would like to develop a universal ML-IAP using this active Bayesian Committee Machine (BCM) potential methodology for carbon-nitrogen-hydrogen (CNH) with various compositions. ML models are trained and generated through first-principles calculations and molecular dynamics simulations for molecules with only CNH. Using long amine structures to test an ML model trained only with short chains, the results show excellent consistency with DFT calculations. Consequently, machine learning-based models for organic molecules not only demonstrate the ability to accurately describe various physical properties but also hold promise for investigating a broad spectrum of diverse materials systems.