Mixing ReaxFF parameters for transition metal oxides using force-matching method

Abstract

We present the development of the method for the refitting the ReaxFF parameters for a system consisting of the mixed transition metal oxides. We have tested several methods allowing to calculate the differences between the vectors of the forces acting on atoms obtained from the reference DFT simulation and the parameters-dependent ReaxFF. We conclude that the footrule method yields the best parameters among the investigated options. We then validate the parameters using the system consisting of the hematite supported (TiO2)n clusters. The results indicate the refitted parameters allow to obtain acceptable geometries of the clusters upon MD simulation on the ReaxFF level, and despite the short timescale lead to the stable structures.

The formation mechanism of Sc-based metallofullerenes: a molecular dynamics simulation study

“…Therefore, the development and testing of force field parameters often precedes ReaxFF MD simulations. 32–42…”

“…Therefore, the development and testing of force field parameters often precedes ReaxFF MD simulations. [32][33][34][35][36][37][38][39][40][41][42] To the best of our knowledge, no reactive force fields tailored for Sc exist to date. We used the parameters of Zr 43 to form the initial parameter set.…”

ReaxFF molecular dynamics of graphene oxide/NaCl aqueous solution interfaces

“…This force field was originally developed and trained by Kim et al 20 using several previously available force fields [21][22][23][24][25][26][27] and its applicability and reliability have been affirmed by multiple studies. [28][29][30][31] The Berendsen thermostat controlled the temperature at 300 K with a relaxation time of 100 fs. During the 1 ns equilibration run, the NPT ensemble set the pressure to 1 atm using the Berendsen barostat.…”

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

“…It naturally takes into account the data imbalance and the inherent strengths and weaknesses of the model being trained. A ReaxFF parametrization is used as a case study in this paper because we believe ReaxFF can greatly benefit from Balanced Loss, but the methodology is general enough to be applied to other (even nonchemical) parametrizations with similar challenges. ,− As software tools and algorithms for (re)­parametrization (reactive) force fields improve, ,− ,, we expect that more practitioners to face the challenge of data imbalance, also for machine learning potentials that are trained on increasingly large and diverse datasets. − …”