Machine Learning Classification of Local Environments in Molecular Crystals

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Characterizing local structural motifs and packing patterns in molecular solids is crucial but challenging for simulations and experiments.
• Understanding polymorphism is key to controlling material properties and predicting solid-state behavior.

Purpose of the Study:

• To develop and validate novel machine learning approaches for characterizing local environments in molecular crystals.
• To enable accurate classification of different polymorphs and analysis of crystal dynamics.

Main Methods:

• Utilized two machine learning models: an atomistic graph convolutional network with molecule-wise aggregation for flexible learning, and a novel descriptor set with symmetry functions and point-vector representation for handcrafted learning.
• Applied these models to urea and nicotinamide crystal polymorphs, and to dynamical trajectory data of nanocrystals and solid-solid interfaces.

Main Results:

• Achieved very high classification accuracy for both developed models on diverse crystal polymorphs.
• Demonstrated practical applicability in analyzing complex dynamical trajectory data, including nanocrystals and interfaces.

Conclusions:

• The proposed learning models effectively characterize local environments in molecular crystals, offering a significant advancement for exploring condensed matter phenomena.
• Both flexible and handcrafted representation approaches are versatile, applicable to various molecules and crystal topologies.