arXiv subjects

Kamron Fazel

Publications and source records attributed to Kamron Fazel.

4 recordsLinked to original sources

arXiv2025

Development of machine-learned interatomic potentials to predict structure, transport, and reactivity in platinum-based fuel cells

Machine-learned interatomic potentials (MLIPs) have rapidly progressed in accuracy, speed, and data efficiency in recent years. However, training robust MLIPs in multicomponent systems still remains a challenge. In this work, we train a MLIP to describe hydrated Nafion ionomers and platinum catalysts, which are important components of fuel cells, by constructing a diverse training set to describe the bulk polymer and interfacial catalyst-polymer interactions well. We find that active learning improves the initial dataset little in terms of reducing uncertainty and error, pointing towards a need for more effective methods to efficiently explore the relevant interactions in complex, multicomponent systems. We use our trained MLIP to study the properties of the platinum-Nafion system, including polymer structure, proton mobility in a bulk Nafion polymer and near a platinum-Nafion interface, and reactions near and far from the interface, finding excellent results for structure and reactions contained within our training set. Transport seems well described, with both vehicular transport and Grotthuss hopping captured, although converged calculations of diffusivities were not computed because they require calculations of tens of nanoseconds that are challenging with current state-of-the-art MLIPs. The combined insights that this model provides can be leveraged to optimize fuel cell performance, and the approach can be applied to other chemical processes and devices where structure, transport, and reactivity all contribute to overall observed performance.

physics.chem-ph↗

arXiv2024

Bridging electronic and classical density-functional theory using universal machine-learned functional approximations

The accuracy of density-functional theory (DFT) is determined by the quality of the approximate functionals, such as exchange-correlation in electronic DFT and the excess functional in the classical DFT formalism of fluids. The exact functional is highly nonlocal for both electrons and fluids, yet most approximate functionals are semi-local or nonlocal in a limited weighted-density form. Machine-learned (ML) nonlocal density-functional approximations are promising in both electronic and classical DFT, but have so far employed disparate approaches with limited generality. Here, we formulate a universal approximation framework and training protocol for nonlocal ML functionals, combining features of equivariant convolutional neural networks and the weighted-density approximation. We prototype this approach for several 1D and quasi-1D problems and demonstrate that a functional with exactly the same hyperparameters achieves excellent accuracy for the hard-rod fluid, the inhomogeneous Ising model, the exact exchange functional for electrons, the electron kinetic energy functional for orbital-free DFT, as well as for liquid water with 1D inhomogeneities. These results lay the foundation for a universal ML approach to exact 3D functionals spanning electronic and classical DFT.

physics.chem-ph↗

arXiv2023

First-principles molten salt phase diagrams through thermodynamic integration

Precise prediction of phase diagrams in molecular dynamics (MD) simulations is challenging due to the simultaneous need for long time scales, large length scales and accurate interatomic potentials. We show that thermodynamic integration (TI) from low-cost force fields to neural network potentials (NNPs) trained using density-functional theory (DFT) enables rapid first-principles prediction of the solid-liquid phase boundary in the model salt NaCl. We use this technique to compare the accuracy of several DFT exchange-correlation functionals for predicting the NaCl phase boundary, and find that the inclusion of dispersion interactions is critical to obtain good agreement with experiment. Importantly, our approach introduces a method to predict solid-liquid phase boundaries for any material at an ab-initio level of accuracy, with the majority of the computational cost at the level of classical potentials.

cond-mat.mtrl-sci↗

arXiv2023

Improving the reliability of machine learned potentials for modeling inhomogenous liquids

The atomic-scale response of inhomogeneous fluids at interfaces and surrounding solute particles plays a critical role in governing chemical, electrochemical and biological processes at such interfaces. Classical molecular dynamics simulations have been applied extensively to simulate the response of inhomogeneous fluids directly, and as inputs to classical density functional theory, but are limited by the accuracy of the underlying empirical force fields. Here, we deploy neural network potentials (NNPs) trained to ab initio simulations to accurately predict the inhomogeneous response of two widely different fluids: liquid water and molten NaCl. Although NNPs can be readily trained to model complex bulk systems across a range of state points, in order to appropriately model a fluid's response at an interface, inhomogeneous configurations must be included in the training data. We establish protocols based on molecular dynamics simulations in external atomic potentials in order to sufficiently sample the correct configurations of inhomogeneous fluids. We show that NNPs trained to inhomogeneous fluid configurations can predict several properties such as the density response, surface tension and size-dependent cavitation free energies in water and molten NaCl corresponding to ab initio interactions more accurately than empirical force fields. This work therefore provides a first demonstration and framework for extracting the response of inhomogeneous fluids from first principles for classical density-functional treatment of fluids free from empirical potentials.

cond-mat.mtrl-sci↗