Unsupervised discovery of solid-state lithium ion conductors.

Area of Science:

• Materials Science
• Computational Materials Science
• Machine Learning in Materials Discovery

Background:

• Machine learning shows promise for functional materials discovery.
• Reliable model development is hindered by limited materials property data.
• Data scarcity poses a significant challenge in accelerating materials innovation.

Purpose of the Study:

• To propose and demonstrate an unsupervised learning approach for materials discovery.
• To alleviate the data scarcity challenge in machine learning for materials science.
• To discover novel fast lithium-ion conductors using limited conductivity data.

Main Methods:

• Employed unsupervised machine learning, requiring no labeled data.
• Utilized a limited set of conductivity data for prioritization.
• Applied ab initio molecular dynamics simulations for property prediction.

Main Results:

• Discovered 16 new fast lithium-ion conductors.
• Predicted conductivities range from 10^-4 to 10^-1 S/cm.
• Identified compounds with distinct structures and chemistries compared to known systems.

Conclusions:

• Unsupervised learning effectively discovers new functional materials despite data limitations.
• This method enables exploration of a wide materials space.
• The approach demonstrates significant potential for accelerating the discovery of advanced materials.