Unsupervised Exploration of MoS₂ Nanocluster Configurations: Structures, Energetics, and Electronic Properties

Abstract

We propose an efficient method to explore the configuration space of nanoclusters by combining together ab initio molecular dynamics, metadynamics and data clustering algorithms. On one side, we employ collective variables sensitive to topological changes in the network of interatomic connections to map the configuration space; on the other, we introduce an automatic approach to select, in such space, representative structures to be optimized. In this way, we show that it is possible to sample thoroughly the set of relevant nanocluster geometries, at a limited computational cost. We apply our method to explore MoS 2 clusters, that recently raised a sizable interest due to their remarkable electronic and catalytic properties. We demonstrate that the unsupervised algorithm is able to find a large number of low-energy structures at different cluster sizes, including both bulk-like geometries and very different topologies. We are thus able to recapitulate, in a single computational study on technologically-relevant MoS 2 clusters, the results of all previous works that employed distinct techniques like genetic algorithms or heuristic hypotheses. Furthermore, we found several new structures not previously reported. The ensemble of MoS 2 cluster structures is deposited in a publicly accessible database.

Machine Learning-Assisted Discovery of Hidden States in Expanded Free Energy Space

“…We follow a previous work on the development of SPRINT ( S i ) coordinates (see the section S1.2 of the Supporting Information), which were originally established based on the use of spectral graph theory and have been used as a CV in the search for simple multistep reactions. , Nonetheless, given a reference atom in the system, we found that chemical information on its surrounding atoms that are too far away is missing and not included in the SPRINT coordinates of the reference atom, which may lead to unsatisfactory performance of SPRINT coordinates in the case of multiple-step or complicated reactions. A clear example supporting the existence of this problem is, for example, the case of intermediate or product molecules that are composed of atoms belonging to different reactant molecules.…”