Researchers at Lawrence Livermore National Laboratory (LLNL) have developed a new approach that combines generative artificial intelligence (AI) and first-principles simulations to predict three-dimensional (3D) atomic structures of highly complex materials.
This research highlights LLNL's efforts in advancing machine learning for materials science research and supporting the Lab's mission to develop innovative technological solutions for energy and sustainability.
The study, recently published in Machine Learning: Science and Technology, represents a potential leap forward in the application of AI for materials characterization and inverse design.
The approach uses X-ray absorption near edge structure (XANES) spectroscopy. Accurately determining atomic structures from spectroscopic data has long posed a challenge, particularly for complex systems, such as shapeless materials. In response, LLNL scientists have introduced a generative framework based on diffusion models, which are an emerging machine learning technique. The authors demonstrate how this framework enables the prediction of 3D atomic arrangements from XANES spectra.
"Our method bridges a crucial gap between spectroscopic analysis and precise structure determination," said Hyuna Kwon, a materials scientist in LLNL's Quantum Simulations Group, Materials Science Division. "By conditioning the generative model on XANES data, we can reconstruct atomic structures that align closely with the target spectra, offering a powerful tool for material analysis and custom design."
The project was a collaborative effort, with Kwon and Tim Hsu from LLNL's Center for Applied Scientific Computing contributing equally. The team demonstrated that their AI model also scales effectively from small datasets for generating realistic, large-scale structures. This scale-agnostic property demonstrates the model's ability to bridge scales from nanoscale to microscale, enabling detailed atomic structure generation even at complex features like grain boundaries and phase interfaces.
"This approach can be leveraged beyond just structural analysis," said Anh Pham, the principal investigator of the project. "It can be extended to inverse design - where we start from a desired material property and engineer the corresponding atomic structure - accelerating the discovery of materials with tailored functionalities."
Other LLNL co-authors include Wenyu Sun, Wonseok Jeong, Fikret Aydin, Xiao Chen, Vincenzo Lordi and Fei Zhou. The work was supported by LLNL's Laboratory Directed Research and Development program. Collaborators include Brookhaven National Laboratory and Boston University.
