Data-driven Design of High Pressure Hydride Superconductors using DFT and Deep Learning

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Superconductivity in hydride materials under high pressure has garnered significant interest.
• Discovering new superconductors often relies on experimental synthesis and characterization, which can be time-consuming and resource-intensive.
• A data-driven approach is needed to accelerate the search for novel high-pressure hydride superconductors.

Purpose of the Study:

• To computationally predict the critical temperature (Tc) of over 900 hydride materials under high pressures.
• To identify dynamically stable hydride structures with Tc values exceeding that of MgB2 (39 K).
• To develop and apply machine learning models for accelerated screening of potential superconductors.

Main Methods:

• Density Functional Theory (DFT) calculations were employed to predict critical temperatures (Tc) for numerous hydride materials.
• Graph Neural Network (GNN) models were trained to predict Tc, enabling rapid screening.
• A universal machine-learned force field was utilized for efficient structural relaxation under varying pressures.

Main Results:

• 122 dynamically stable hydride structures were identified with predicted Tc values above 39 K across a pressure range of 0 to 500 GPa.
• The GNN model demonstrated effectiveness in accelerating the prediction of Tc.
• Machine learning significantly reduced the computational cost associated with structural relaxation.

Conclusions:

• The combination of DFT and GNNs provides a powerful and efficient approach for mapping high-pressure hydride superconductors.
• This data-driven strategy accelerates the discovery process for new materials with potential superconducting properties.
• The study establishes a more comprehensive understanding of hydride behavior under pressure for superconductivity research.