Large-angle convergent-beam electron diffraction patterns via conditional generative adversarial networks.

Area of Science:

• Materials Science
• Computational Physics
• Machine Learning

Background:

• Electron diffraction is crucial for crystal structure analysis.
• Simulating dynamical electron diffraction is computationally intensive.
• Large-angle convergent-beam electron diffraction (LACBED) provides detailed structural information.

Purpose of the Study:

• To develop a rapid method for computing dynamical electron diffraction patterns using machine learning.
• To explore the application of generative models in analyzing crystal structures.
• To enable faster and potentially more accurate crystal structure determination.

Main Methods:

• Utilized a conditional generative adversarial network (cGAN).
• Trained the cGAN to learn the relationship between crystal structure potentials and LACBED patterns.
• Employed GPU acceleration for pattern generation.

Main Results:

• The cGAN model successfully learned the connection between projected potentials and LACBED patterns.
• Generated diffraction patterns orders of magnitude faster than traditional simulation methods.
• Demonstrated accurate retrieval of projected potentials from diffraction patterns.

Conclusions:

• Generative machine learning offers a highly efficient approach for dynamical electron diffraction simulation.
• This method significantly accelerates the computation of LACBED patterns.
• The developed technique provides a novel pathway for solving the inverse problem of crystal structure determination.