Transferable Neural Network Potentials and Condensed Phase Properties

Transferable neural network potentials (NNP) are undergoing rapid development. Many practical applications of NNPs focus on single molecules; e.g., using NNPs as a fast replacement for quantum chemical methods for dihedral angle scans in force field development. Similarly, the reference data on which most transferable NNPs have been trained are single molecule properties. As NNPs are beginning to be used to simulate more complex systems, such as solute-solvent simulations, the question arises whether the current generation of transferable NNPs is accurate enough to reproduce condensed phase properties, which in most cases are outside the training domain of the models. Here we present a first analysis of how well two transferable NNPs (ANI-2x, MACE-OFF23(S/M)) perform in reproducing properties such as density, heat of vaporization, heat capacity, and isothermal compressibility of several pure liquids (water, methanol, acetone, benzene, n-hexane at room temperature, and N-methylacetamide at 100 C). In addition, we examine selected radial distribution functions and the self-diffusion constant. We find specific weaknesses for each of the models, and seemingly small flaws lead to poor performance when applied to condensed phase simulations. The varied outcomes observed with the machine learning potentials suggest that, currently, selecting an architecture or model for all-NNP simulations of real-world applications requires careful consideration and testing.