T. Verstraelen

ReaxFF is a computationally efficient model for reactive molecular dynamics simulations that has been applied to a wide variety of chemical systems. When ReaxFF parameters are not yet available for a chemistry of interest, they must be (re)optimized, for which one defines a set of training data that the new ReaxFF parameters should reproduce. ReaxFF training sets typically contain diverse properties with different units, some of which are more abundant (by orders of magnitude) than others. To find the best parameters, one conventionally minimizes a weighted sum of squared errors over all of the data in the training set. One of the challenges in such numerical optimizations is to assign weights so that the optimized parameters represent a good compromise among all the requirements defined in the training set. This work introduces a new loss function, called Balanced Loss, and a workflow that replaces weight assignment with a more manageable procedure. The training data are divided into categories with corresponding “tolerances”, i.e., acceptable root-mean-square errors for the categories, which define the expectations for the optimized ReaxFF parameters. Through the Log-Sum-Exp form of Balanced Loss, the parameter optimization is also a validation of one’s expectations, providing meaningful feedback that can be used to reconfigure the tolerances if needed. The new methodology is demonstrated with a nontrivial parametrization of ReaxFF for water adsorption on alumina. This results in a new force field that reproduces both the rare and frequent properties of a validation set not used for training. We also demonstrate the robustness of the new force field with a molecular dynamics simulation of water desorption from a γ-Al₂O₃ slab model.

This publication is licensed under the terms of your institutional subscription. Request reuse permissions.

A novel force-field parameterization procedure[1] is proposed that surmounts well-known difficulties of the conventional least squares parameterization. The multidimensional ab initio training data are first transformed into individual one-dimensional data sets, each associated with one term in the force-field model. In the second step conventional methods call be used to fit each energy term separately to its corresponding data set. The first step call be completed without any knowledge of the analytical expressions for the energy terms. Moreover the transformed data sets dictate the form of these expressions, which makes the method very suitable for deriving valence force fields. During the transformation in the first step, continuity and least-norm criteria are imposed. The latter facilitate the intuitive physical interpretation of the energy terms that are fitted to the transformed data sets, a prerequisite for transferable force fields. Benchmark parameterizations have been performed oil three small molecules, showing that the new method results in physically intuitive energy terms, exactly when a conventional parameterization would suffer from parameter correlations, i.e. when the number of redundant internal coordinates in the force-field model increases.