Compositional descriptor-based recommender system for the materials discovery.

Area of Science:

• Materials Science
• Inorganic Chemistry
• Computational Chemistry

Background:

• Historical data on inorganic compounds cover a limited fraction of potential crystal structures.
• Numerous unknown inorganic compounds likely exist, necessitating advanced discovery methods.

Purpose of the Study:

• To develop a machine learning strategy for discovering new inorganic compounds.
• To estimate the chemical relevance of compositions for crystal formation.

Main Methods:

• A descriptor-based recommender system was developed to predict chemically relevant compositions (CRCs).
• The system incorporates compositional similarity and prior knowledge from compositional descriptors.
• Validation involved using separate databases for model construction and testing, alongside density functional theory (DFT) for phase stability analysis.

Main Results:

• The recommender system effectively identifies potentially stable inorganic compounds.
• Incorporating compositional descriptors improved the prediction accuracy for novel compounds.
• DFT calculations confirmed the phase stability of predicted CRCs.

Conclusions:

• Descriptor-based recommender systems offer a powerful tool for accelerating inorganic materials discovery.
• This approach expands the search space for new compounds beyond traditional methods.
• The combination of machine learning and DFT provides a robust framework for identifying novel inorganic materials.