The pure and representative types of disordered platinum nanoparticles from machine learning

Area of Science:

• Nanotechnology
• Materials Science
• Computational Chemistry

Background:

• Interpretable structure/property relationships are crucial in nanoscience.
• Characterizing imperfect, disordered, and amorphous nanoparticles is challenging due to structural complexity.
• Disordered platinum nanoparticles show enhanced catalytic performance, necessitating systematic description methods.

Purpose of the Study:

• To develop a systematic method for describing disordered platinum nanoparticles.
• To identify representative and pure structures within a diverse set of disordered platinum nanoparticles.
• To establish a theoretical basis for understanding structure/property relationships in non-ideal platinum nanoparticles.

Main Methods:

• Utilized machine learning algorithms to analyze a diverse set of disordered platinum nanoparticles.
• Employed a 121-dimensional feature space to assess structural similarity.
• Identified prototype and archetype structures representing distinct classes and pure forms.

Main Results:

• Identified two prototypes representing separable classes of disordered platinum nanoparticles.
• Discovered seven archetypes representing pure structures on the convex hull.
• These nine representative nanoparticles explain all observed structural variance and can describe various morphologies (single crystal, twinned, spherical, branched, roughened surfaces).

Conclusions:

• A robust subset of nine representative platinum nanoparticles has been identified.
• These findings provide a theoretical framework for discussing structure/property relationships in geometrically imperfect platinum nanoparticles.
• This systematic approach facilitates further research on disordered nanomaterials.