Accelerating the discovery of athermal deep-ultraviolet nonlinear optical crystals via physics-informed herarchical...

Area of Science:

• Materials Science
• Optics
• Artificial Intelligence

Background:

• Thermal instability in deep-ultraviolet (DUV) coherent sources limits their performance.
• Discovering novel nonlinear optical (NLO) crystals is crucial for stable DUV frequency comb generation.

Purpose of the Study:

• To propose a passive intrinsic control strategy for discovering athermal NLO crystals.
• To develop a physics-informed hierarchical graph learning framework for accelerated materials discovery.

Main Methods:

• Integrating anisotropic attention and space-group embeddings with physics-informed feature injection.
• Utilizing a hierarchical graph learning framework for crystal property prediction.
• Performing virtual screening on the Novel Opto-Electronics Materials Database (NOEMD).

Main Results:

• Achieved R-squared > 0.95 for electronic bandgap and formation enthalpy predictions.
• Obtained Spearman's rank correlation coefficient (ρ) > 0.90 for second-harmonic generation (SHG) response.
• Demonstrated a 7.91-fold enrichment factor and a 73.3% effective discovery rate for athermal NLO crystals.

Conclusions:

• The framework efficiently identifies multi-functional photonic materials for stable DUV frequency comb generation.
• The model captures structural features and physical regularities consistent with chemical heuristics.
• This data-driven approach provides an efficient paradigm for designing advanced photonic materials.