Machine Learning Methodologies Applied to Magnetocaloric Perovskites Discovery.

Area of Science:

• Materials Science
• Computational Materials Science
• Condensed Matter Physics

Background:

• Traditional materials design relies on inefficient trial-and-error methods, limiting exploration of vast chemical spaces.
• Discovering new materials with specific properties, like magnetocaloric effects, is crucial for technological advancements.

Purpose of the Study:

• To develop a machine learning (ML) based methodology for accelerated discovery of magnetocaloric perovskites.
• To computationally screen over 1.2 million potential material compositions.
• To identify promising materials for room-temperature refrigeration applications.

Main Methods:

• Created two databases with 1227 inputs from published literature.
• Trained four ML models using 517 compositional features derived from 58 atomic properties.
• Predicted key magnetocaloric properties: Curie temperature (T_C), magnetic entropy change (ME), and relative cooling power (RCP).

Main Results:

• Identified optimal ML model-feature combinations for predicting magnetocaloric properties.
• Explored the chemical space of lanthanum, praseodymium, and neodymium manganites.
• Highlighted composition trends and suitable doping elements for room-temperature refrigeration.

Conclusions:

• The ML methodology significantly enhances the efficiency of discovering novel magnetocaloric materials.
• The study provides valuable guidelines for future research in magnetocaloric materials.
• The approach is transferable to other perovskite-based materials for applications like catalysis and solar cells.