Managing Expectations and Imbalanced Training Data in Reactive Force Field Development: an Application to Water Adsorption on Alumina

TL;DR

This paper introduces a Balanced Loss function and workflow for ReaxFF parameter optimization, effectively managing imbalanced training data and expectations, demonstrated on water adsorption on alumina.

Contribution

A novel Balanced Loss function and workflow that replace weight assignment in ReaxFF training, improving parameter optimization for imbalanced data sets.

Findings

Successfully optimized ReaxFF for water on alumina

Reproduces both common and rare properties accurately

Demonstrates robustness in molecular dynamics simulations

Abstract

ReaxFF is a computationally efficient model for reactive molecular dynamics simulations, which 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 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 between 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 is 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 non-trivial parameterization of ReaxFF for water adsorption on alumina. This results in a new force field that reproduces both 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 $\gamma$-Al$_2$O$_3$ slab model.