A Pragmatic Transfer Learning Approach for Oxygen Vacancy Formation Energies in Oxidic Ceramics

Area of Science:

• Materials Science
• Computational Chemistry
• Energy Storage

Background:

• Lower oxygen vacancy formation energy is crucial for air electrode materials in solid oxide cells.
• Developing efficient prediction methods is key to discovering novel energy materials.

Purpose of the Study:

• To introduce a transfer learning approach for predicting oxygen vacancy formation energy in ABO3 perovskites.
• To adapt predictive models from two-species-doped to four-species-doped systems.

Main Methods:

• Utilized an artificial neural network for predictive modeling.
• Defined a formally similar feature space for transfer learning.
• Validated model transferability using statistical analysis of residual distributions.

Main Results:

• Successfully trained predictive models for oxygen vacancy formation energy using a two-species-doping training dataset.
• Demonstrated the transferability of these models to predict energies in four-species-doped perovskites.
• Validated the predictive accuracy and reliability of the transfer learning approach.

Conclusions:

• The proposed transfer learning approach is an effective tool for accelerating the search for novel energy materials.
• This method enhances the prediction of oxygen vacancy formation energy in complex doped perovskite systems.
• The findings support the development of advanced materials for solid oxide cell applications.