Active Learning-Guided Accelerated Discovery of Ultra-Efficient High-Entropy Thermoelectrics

Area of Science:

• Materials Science
• Solid State Physics
• Thermoelectrics

Background:

• High-entropy alloys (HEAs) show promise as efficient thermoelectric materials.
• Vast compositional spaces in HEAs challenge traditional material discovery methods.
• Machine learning approaches face difficulties in exploring HEA compositional complexity.

Purpose of the Study:

• To develop an active learning framework for efficient discovery of high-entropy chalcogenides (HECs).
• To identify novel HECs with superior thermoelectric performance (zT > 2).
• To enable non-experts to design new thermoelectric systems.

Main Methods:

• Leveraging sparse experimental data (80 samples) from a large dataset (16206 compositions).
• Integrating physics-informed descriptors with uncertainty-aware sampling.
• Employing an active learning framework to guide material discovery.

Main Results:

• Identification of three new high-entropy chalcogenides (HECs) with thermoelectric figure of merit (zT) > 2.
• Efficient assimilation of structure-property relationships from limited data.
• Systematic exclusion of unfavorable chemical compositions.

Conclusions:

• The active learning framework efficiently identifies high-performance HECs.
• Novel atomic arrangements and transport properties contribute to superior performance in HECs.
• The study advances the understanding of physical phenomena in disorder-rich thermoelectric systems.