Upper-Bound Energy Minimization to Search for Stable Functional Materials with Graph Neural Networks.

Area of Science:

• Materials Science
• Computational Materials Science
• Solid-State Chemistry

Background:

• Discovering novel materials requires efficient prediction of thermodynamic stability.
• Density functional theory (DFT) is a standard but computationally intensive method for stability assessment.
• Accelerated materials discovery demands faster and accurate predictive tools.

Purpose of the Study:

• To develop a rapid and accurate computational approach for predicting the thermodynamic stability of inorganic materials.
• To establish an efficient search strategy for identifying new stable functional materials in unexplored chemical spaces.
• To demonstrate the utility of the developed method in discovering promising solid-state battery electrolytes.

Main Methods:

• Defined an upper bound energy using constrained DFT optimization (cell volume only).
• Developed a scale-invariant graph neural network (GNN) to predict this upper bound energy.
• Trained the GNN on a large dataset of 128,000 DFT calculations, including fully relaxed and volume-only relaxed structures.
• Generated new candidate structures via ionic substitution of known prototypes.

Main Results:

• The GNN accurately predicts the upper bound energy, enabling rapid stability assessment.
• The method achieves over 99% accuracy in predicting stable structures compared to full DFT calculations.
• Successfully identified new, stable inorganic material candidates, including promising solid-state battery electrolytes.
• Discovered electrolytes with desirable properties like high ionic conductivity and wide electrochemical stability windows.

Conclusions:

• The developed GNN-based framework significantly accelerates the discovery of stable inorganic materials.
• This approach offers a highly accurate and efficient alternative to traditional DFT methods for materials screening.
• The method holds broad applicability for various materials design challenges, particularly in energy storage applications.