methods
CGWGAN
What it does
CGWGAN explores symmetry-aware crystal generation through Wyckoff-based representations, connecting compact descriptions of crystals with generative models.
Approach & capabilities
Represents crystals through asymmetric units, space groups, lattice vectors, and elements.
Uses Wyckoff-aware generative modeling to propose crystal structures.
Studies structure generation while retaining crystallographic symmetry.
Read more
Diffraction patterns and crystal structures are closely related concepts. Therefore, my research interest lies in crystal representation.
In this survey (https://arxiv.org/pdf/2505.16379), we provide a comprehensive overview of crystal generation, summarizing and organizing various types of materials while illustrating multiple representations of crystalline structures. We then present a detailed summary and taxonomy of current AI-driven materials generation approaches. Furthermore, we discuss commonly used evaluation metrics and highlight open-source codebases and benchmark datasets.
One of the projects we have worked on involves embedding crystals using asymmetry units (ASUs), space groups, lattice vectors, and the minimal element set to inversely generate stable, novel crystal structures (CGWGAN, JMI, 2024), achieving good results while preserving high symmetry.
In another project, we introduced powder XRD to provide additional insights from reciprocal space, enhancing the model's understanding of crystals (ASUGNN, J. Appl. Cryst.), which shows great potential.
I am currently working on deriving a universal pre-trained model for crystals and hope to share more soon!