Deep learning generative model for conditional crystal structure prediction of sodium amide.

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Sodium amide (NaNH₂) exhibits complex high-pressure behavior crucial for hydrogen storage and synthesis.
• Traditional structure prediction methods face challenges with NaNH₂'s complex energy surface and atomic rearrangements.
• Accurate determination of high-pressure phases is vital for understanding material properties.

Purpose of the Study:

• To develop an experimental-informed deep learning framework for crystal structure determination of NaNH₂ under high pressure.
• To accurately identify and validate the structure of the high-pressure gamma phase of NaNH₂.
• To elucidate the mechanisms behind pressure-induced phase transitions in NaNH₂.

Main Methods:

• An experimental-informed deep learning generative framework using crystallographic constraints.
• Conditional structure determination based on lattice parameters and space-group symmetry.
• Energy-guided diffusion sampling and validation via synchrotron X-ray diffraction.

Main Results:

• Successfully identified the high-pressure gamma phase of NaNH₂ as a P2₁/c structure (Z=16, 64 atoms).
• The determined structure remained stable up to 14.0 GPa, confirmed by experimental validation.
• Charge-density analysis revealed mechanisms driving phase transitions and stabilizing the gamma phase.

Conclusions:

• The deep learning framework provides a general strategy for solving experimentally observed but structurally unknown phases in complex ionic materials.
• This approach is effective when lattice metrics and symmetry constraints are available.
• The findings enhance understanding of sodium amide's high-pressure behavior and its potential applications.