Bridging theory and practice: scalable prediction of alloy nanoparticle morphology and stability via machine-learned moment tensor potentials

Alloy nanoparticles exhibit unique properties that significantly differentiate them from their bulk counterparts, making them crucial for catalytic applications. However, the design of functional nanoparticles is often constrained by the challenge of accurately predicting their stability, especially at large scales, where traditional ab initio methods are limited by computational demands. To overcome this, we introduce a novel method for predicting the thermodynamic stability and chemical ordering of alloy nanoparticles, utilizing the machine-learned moment tensor potential (MTP) integrated with active learning techniques. This approach allows precise assessment of the energetics of alloy nanoparticles regardless of morphology and remains scalable to experimental sizes, maintaining an optimal balance between computational efficiency and accuracy. The effectiveness of our method was demonstrated through its ability to predict the ground-state configurations of alloy spherical nanoparticles (SNPs), which is consistent with experimental results. In particular, the machine-learned MTP model enables a thorough investigation of the thermodynamic stability of SNPs and accurate prediction of the chemical ordering of SNPs depending on the type and composition ratio of the binary alloy systems. Consequently, this work not only highlights the superior performance of our approach over traditional computational methods for modeling large nanoparticles but also establishes a systematic approach for evaluating the stability of nanoparticles at full scale. We believe that our findings will pave the way for more reliable and scalable predictions in nanoparticle research and potentially accelerate the development of novel nanoparticle-based technologies.