Structural constraint integration in a generative model for the discovery of quantum materials.

Area of Science:

• Materials Science
• Computational Chemistry
• Quantum Physics

Background:

• Generating functional inorganic materials computationally is challenging due to data scarcity and structural complexity.
• Existing generative models struggle to incorporate specific structural requirements for novel materials.

Purpose of the Study:

• To introduce a novel generative framework, SCIGEN, for discovering stable quantum materials candidates.
• To enforce geometric constraints within generative models for targeted material design.
• To accelerate the discovery of inorganic materials with specific lattice structures.

Main Methods:

• Developed SCIGEN (Structural Constraint Integration in a GENerative model), a diffusion-based generative model incorporating geometric constraints.
• Applied SCIGEN to generate inorganic compounds with Archimedean and Lieb lattices.
• Utilized multistage stability screening and high-throughput density functional theory (DFT) calculations.
• Employed a graph neural network classifier to predict magnetic ordering.

Main Results:

• Generated ten million inorganic compounds with specified lattice geometries.
• Over 10% of generated compounds passed initial stability screening.
• DFT calculations confirmed high convergence (95%) and 53% structural stability for 26,000 candidates.
• Identified magnetic ordering in 41% of relaxed structures.
• Successfully synthesized and characterized two predicted materials, TiPd0.22Bi0.88 and Ti0.5Pd1.5Sb.

Conclusions:

• SCIGEN offers a scalable and effective approach for generating quantum materials guided by lattice geometry.
• The framework successfully integrates structural constraints into generative models, overcoming limitations of previous methods.
• The study demonstrates a significant advancement in computational materials discovery, paving the way for new functional materials.