Machine Learning Assisted Clustering of Nanoparticle Structures.

Area of Science:

• Materials Science
• Computational Chemistry
• Nanotechnology

Background:

• Atomistic simulations generate large datasets of nanoparticle (NP) structures.
• Manual inspection of these simulation outputs is time-consuming and challenging.
• Identifying diverse structural motifs within these datasets requires robust analytical methods.

Purpose of the Study:

• To develop an automated scheme for clustering and analyzing nanoparticle structures.
• To leverage Machine Learning (ML) for efficient data separation from atomistic simulations.
• To improve the identification of various structural motifs in nanoparticle systems.

Main Methods:

• Utilizing Machine Learning techniques, specifically unsupervised learning algorithms like K-Means and Gaussian mixture models.
• Describing nanoparticles based on their local atomic environment for feature extraction.
• Applying the developed scheme to datasets from global optimization and molecular dynamics simulations.

Main Results:

• Successfully distinguished between different structural motifs including icosahedra, decahedra, polyicosahedra, fcc fragments, and twins.
• Demonstrated improved performance over previous methods, especially for complex systems with varied and disordered structures.
• Validated the effectiveness of a detailed local atomic environment description for NP analysis.

Conclusions:

• The proposed ML-based scheme provides an efficient and accurate method for automated nanoparticle structure analysis.
• This approach significantly aids in the interpretation of large-scale atomistic simulation data.
• The method offers enhanced capabilities for identifying structural diversity in nanoparticles, including amorphous configurations.