Co-evolutionary search for Cu-Pd-Ag nanoparticle ground states accelerated with neural network potentials

Unconstrained optimization of nanoparticles requires advanced search methods capable of locating global minima in large configuration spaces. In this study, we demonstrate that algorithm efficiency can be improved substantially if ground state searches are performed across a range of nanoparticle sizes simultaneously. In this symbiotic co-evolutionary approach implemented in our MAISE package [1], stable motifs are periodically exchanged among tribes with neighboring nanoparticle sizes. The algorithm was extensively tested on elemental Cu, Pd, and Ag nanoparticles up to 80 atoms using both traditional classical potentials and our neural network models. Examination of the lowest-energy configurations revealed that the neural network set was consistently more stable at the density functional theory level. Lastly, we used our Cu-Pd-Ag neural network model to identify stability regions in binary and ternary systems.

[1]: https://github.com/maise-guide/maise