Fast and stable deep-learning predictions of material properties for solid solution alloys.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Predicting macroscopic physical properties of alloys is crucial for materials design.
• Traditional methods like density functional theory are computationally expensive.
• Developing accurate and efficient predictive models for alloys remains a challenge.

Purpose of the Study:

• To develop a novel deep learning approach for accurate prediction of alloy properties.
• To leverage correlations between physical properties for enhanced neural network accuracy.
• To accelerate material property prediction compared to first-principles calculations.

Main Methods:

• Utilized multitasking neural network (NN) models to simultaneously predict total energy, charge density, and magnetic moment.
• Trained NNs using inter-property correlations as constraints for improved reliability.
• Applied the approach to binary alloys copper-gold (CuAu) and iron-platinum (FePt).

Main Results:

• Multitasking NNs achieved highly accurate predictions of macroscopic physical properties.
• The models predicted material properties hundreds of times faster than density functional theory.
• Inclusion of charge density and magnetic moment constraints improved model stability, accuracy, and reduced energy prediction uncertainty.

Conclusions:

• The proposed deep learning approach offers a computationally efficient alternative for alloy property prediction.
• Multitasking NNs enhance prediction reliability by incorporating multiple physical constraints.
• This method holds promise for accelerating materials discovery and design.