Multiobjective genetic training and uncertainty quantification of reactive force fields

Abstract

The ReaxFF reactive force-field approach has significantly extended the applicability of reactive molecular dynamics simulations to a wide range of material properties and processes. ReaxFF parameters are commonly trained to fit a predefined set of quantummechanical data, but it remains uncertain how accurately the quantities of interest are described when applied to complex chemical reactions. Here, we present a dynamic approach based on multiobjective genetic algorithm for the training of ReaxFF parameters and uncertainty quantification of simulated quantities of interest. ReaxFF parameters are trained by directly fitting reactive molecular dynamics trajectories against quantum molecular dynamics trajectories on the fly, where the Pareto optimal front for the multiple quantities of interest provides an ensemble of ReaxFF models for uncertainty quantification. Our in situ multiobjective genetic algorithm workflow achieves scalability by eliminating the file I/O bottleneck using interprocess communications. The in situ multiobjective genetic algorithm workflow has been applied to high-temperature sulfidation of MoO 3 by H 2 S precursor, which is an essential reaction step for chemical vapor deposition synthesis of MoS 2 layers. Our work suggests a new reactive molecular dynamics simulation approach for far-from-equilibrium chemical processes, which quantitatively reproduces quantum molecular dynamics simulations while providing error bars.

Evolutionary multi-objective optimization and Pareto-frontal uncertainty quantification of interatomic forcefields for thermal conductivity simulations

“…This Pareto-frontal breakdown of different forcefields for each epoch provides a natural way to establish one of the primary sources of uncertainty in molecular dynamics simulations -namely the uncertainty in forcefield parameters. This Pareto-frontal uncertainty quantification approach offers an alternative method to estimate the errors in forcefield parameters [11,[30][31][32][33], to complement the predominantly Bayesian approaches employed in prior studies [34].…”

“…These shortcomings are partially addressed in recent multi-objective schemes like GARFfield [9], which use evolutionary algorithms to perform global minimization of a weighted sum of multiple objectives, using an a priori user-provided weighting scheme. Other schemes such as Multi-objective evolutionary strategies [10] and MOGA [11] Rotation-invariant Particle Swarm Optimization with isotropic Gaussian Mutation (RIPSOGM) [6] have been developed that evolve the entire Pareto Frontier of multiple forcefield populations at once, without the need to specify a priori weights for the different objectives. Existing software frameworks for forcefield optimization are also commonly limited to the parameterization of a single predefined functional forcefield form, such as the Forcefield Toolkit (ffTK) [12] and general automated atomic model parameterization (GAAMP) [13] frameworks for the CHARMM forcefield, Paramfit [14] for AMBER forcefields and MOGA [11] for ReaxFF forcefields.…”

Uncertainty Quantification in Atomistic Modeling of Metals and Its Effect on Mesoscale and Continuum Modeling: A Review

“…The quantification of parametric uncertainty for single potentials has been undertaken in several cases [52,53] while Bayesian frameworks have also been proposed for a variety of interatomic models and force fields [54,55,56]. Furthermore, quantification of uncertainty due to the potential fitting reference set [57] was augmented by propagation of parametric uncertainties to MD outputs [58].…”

Unveiling oxidation mechanism of bulk ZrS2

“…In this work, we performed first principles-informed reactive molecular dynamics (RMD) simulations [6] to study atomistic oxidation processes in ZrS 2 . We first developed new reactive forcefield (ReaxFF) parameters for Zr/O/S using multi-objective genetic algorithm (MOGA) [7][8][9]. The optimized forcefield is able to reproduce quantummechanically computed charge values and bond-population dynamics.…”