Generative Design of Inorganic Compounds Using Deep Diffusion Language Models.

Area of Science:

• Materials Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Discovering new materials with desired functions is hindered by the vast chemical space and complex constraints.
• Existing methods struggle to efficiently navigate chemical possibilities and predict stable material structures.

Purpose of the Study:

• To develop and validate a deep-learning-based generative model for designing novel material compositions and structures.
• To accelerate the discovery of stable and functional materials.

Main Methods:

• Utilized deep diffusion language models for generating chemical compositions.
• Employed template-based crystal structure prediction and graph neural network (GNN)-based potential for structure relaxation.
• Validated material stability using Density Functional Theory (DFT) calculations for formation energies and energy-above-hull analysis.

Main Results:

• Generated six new stable materials, including Ti2HfO5, TaNbP, YMoN2, TaReO4, HfTiO2, and HfMnO2, with negative formation energies.
• Four of these materials (Ti2HfO5, TaNbP, YMoN2, TaReO4) demonstrated excellent stability with energy-above-hull values below 0.3 eV.
• Confirmed the effectiveness of the deep learning approach in predicting stable material candidates.

Conclusions:

• The developed deep learning pipeline significantly enhances the efficiency of discovering novel, stable materials.
• This AI-driven approach offers a powerful tool for navigating complex chemical spaces in materials science.
• The identified materials represent promising candidates for further experimental investigation and application.