The seventh blind test of crystal structure prediction: structure ranking methods

Area of Science:

• Crystallography
• Computational Chemistry
• Materials Science

Background:

• Crystal structure prediction is crucial for understanding material properties.
• Evaluating the accuracy of different prediction methods is essential for advancing the field.
• The Cambridge Crystallographic Data Centre organizes blind tests to assess crystal structure prediction techniques.

Purpose of the Study:

• To evaluate and rank various computational methods for predicting crystal structure stability.
• To compare the performance of periodic DFT-D methods, machine learning potentials, and force fields.
• To identify promising and efficient alternatives to traditional density functional theory (DFT) methods.

Main Methods:

• Standardized sets of crystal structures were generated using diverse methods.
• Participants applied periodic DFT-D methods, machine learned potentials (including AIMnet), and empirical force fields.
• Energy rankings were compared against experimental data and a non-energy-based scoring function was also utilized.

Main Results:

• Periodic DFT-D methods generally agreed well with experimental data.
• A machine learning model using AIMnet potentials demonstrated high accuracy and efficiency.
• For target XXXII, a consensus indicated that a more stable polymorph may exist beyond the global minimum.
• Free energy calculations provided improved predictions for specific targets but not universally.

Conclusions:

• Periodic DFT-D methods remain reliable for crystal structure stability ranking.
• Machine learning potentials, particularly AIMnet, show significant promise as efficient alternatives.
• Further research is needed to improve the efficiency of crystal structure prediction methods due to high computational costs.