Automatic structural elucidation of vacancies in materials by active learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence in Materials Discovery

Background:

• Determining optimal structures for non-stoichiometric materials with defects, like vacancies, is computationally challenging due to vast search spaces.
• Traditional methods often struggle with the complexity and scale of exploring all possible atomic arrangements for vacancies.
• Understanding vacancy structures is crucial for applications in catalysis and environmental science.

Purpose of the Study:

• To develop an automated artificial intelligence (AI) method based on active learning (AL) for efficient structural elucidation of vacancies in solids and nanoparticles.
• To enhance the probability of finding global minimum energy structures with significantly fewer computational calculations.
• To implement and validate the AL method for vacancy discovery in various material systems.

Main Methods:

• Developed an active learning (AL) framework integrating machine learning regression algorithms and uncertainty quantification to guide the search for optimal structures.
• Implemented the AL method within the QMLMaterial software, including acquisition functions like Expected Improvement (EI), Lower Confidence Bound (LCB), and Probability of Improvement (PI).
• Utilized Density Functional Tight Binding (DFTB) and Density Functional Theory (DFT) calculations to generate potential energy surfaces and validate results for graphite, fullerene, and perovskite systems.

Main Results:

• Successfully applied the AL method for automated structural searches of carbon vacancies in graphite (C36) and C60 fullerene.
• Elucidated the optimal oxygen vacancy distribution in CaTiO3 perovskite, revealing its semiconductor behavior.
• Demonstrated superior performance of the AL method compared to random search and genetic algorithms in finding low-energy vacancy structures.

Conclusions:

• The developed AI-driven active learning methodology provides an accurate and automated approach for structural elucidation of vacancies in materials.
• This method significantly accelerates the discovery of stable structures for non-stoichiometric materials, crucial for understanding chemical processes.
• The QMLMaterial software with AL capabilities offers a powerful tool for advancing materials design and discovery in fields like catalysis and environmental science.