AI-assisted prediction of catalytically reactive hotspots in nanoalloys

TL;DR

This paper introduces an AI-assisted framework combining machine learning, Monte Carlo simulations, and multiscale modeling to predict reactive hotspots in nanoalloys, enabling efficient design of catalytically active materials.

Contribution

It presents a novel integrated approach that combines ML, Monte Carlo, and multiscale methods to accurately predict catalytic activity in nanoalloys without extensive DFT calculations.

Findings

Successfully predicts stable nanoparticle structures with core-shell geometries.

Accurately maps d-band centers across various nanoalloy compositions.

Validated on Pt-Ni nanoparticles relevant to experiments.

Abstract

Nanoalloys offer a unique opportunity to tailor chemical properties through changes in composition, shape, and size. However, this flexibility introduces complexity that challenges both experimental and conventional theoretical methods. In this work, we present an AI-assisted framework for predicting reactive hotspots in nanoalloys. First, we use a Metropolis Monte Carlo method with a lattice-based machine learning potential, trained on 2NN-MEAM data, to rapidly identify thermodynamically stable nanoparticle structures, demonstrated for Pt-Ni homotops. This approach yields core-shell geometries with Ni-rich cores and Pt-enriched surfaces. To predict catalytic activity, we exploit the correlation between reactivity and d-band centers. Rather than relying on costly DFT calculations, we employ a multiscale method using SCC-DFTB and machine learning to efficiently and accurately map d-band centers across a wide range of nanoalloy sizes and compositions. The framework is validated on Pt-Ni nanoparticles of experimental relevance and is readily extendable to other nanoalloy systems.