PROFiT-Net: Property-Networking Deep Learning Model for Materials.

Area of Science:

• Materials Science
• Artificial Intelligence
• Computational Chemistry

Background:

• Accurate prediction of material properties is crucial for developing new technologies.
• Existing material databases and deep learning (DL) models face limitations with high-fidelity data.
• Developing advanced AI for materials science requires models trained on sparse, high-quality datasets.

Purpose of the Study:

• To develop a novel deep learning model for predicting material properties.
• To enhance material property prediction accuracy using crystal structure representations.
• To create an AI model capable of learning from limited high-fidelity material data.

Main Methods:

• Developed a deep learning model named PRoperty-networking Orbital Field maTrix-convolutional neural Network (PROFiT-Net).
• Utilized a modified orbital field matrix (OFM) representation incorporating elemental properties and valence electron configurations.
• Trained the model to capture interrelations between elemental properties within crystal structures.

Main Results:

• PROFiT-Net achieved high accuracy in predicting dielectric constants, experimental band gaps, and formation enthalpies.
• The model demonstrated superior performance compared to other leading deep learning models.
• PROFiT-Net successfully identified physical patterns, avoiding unphysical predictions and maintaining scalability.

Conclusions:

• PROFiT-Net offers a scalable and accurate approach for predicting material properties.
• The model's ability to learn from limited data addresses a key challenge in materials informatics.
• PROFiT-Net is expected to significantly accelerate the discovery and development of functional materials.