Artificial Intelligence Helps Reveal Hidden Atomic Structures of Crystals
In a breakthrough study, scientists at Columbia Engineering have developed an artificial intelligence model capable of solving a century-old problem in materials science—determining the atomic structure of nanocrystals using X-ray diffraction. This discovery, published in the prestigious journal Nature Materials, represents a significant step forward in the development of new materials, particularly for technologies such as advanced batteries, solar cells, and catalysts.
Led by Professor Simon J. L. Billinge, the research team created a machine-learning model called PXRDnet, which can interpret complex powder X-ray diffraction (PXRD) patterns and convert them into precise atomic arrangements. This process has traditionally been considered an almost impossible task because of what scientists call the “phase problem”—the loss of critical information during the measurement process. According to Professor Billinge, it is like trying to reconstruct a three-dimensional object from only its two-dimensional shadow, which represents a mathematically underdetermined problem. PXRDnet uses training based on thousands of known crystal structures to decipher new, previously unknown structures. This capability is particularly important for nanocrystals, which are too small for traditional crystallography methods. “We taught the model to recognize patterns from experimental data and link them to the corresponding atomic structures,” Billinge explains. “That is something that would be extremely difficult, if not impossible, for human experts, especially with complex structures.” The significance of this discovery extends far beyond academic research. Nanocrystals play a key role in many modern technologies, from catalysis to electronics. They are particularly important for the development of next-generation batteries, where a precise understanding of their atomic structure could lead to significant improvements in performance and lifespan. Columbia Technology Ventures has already filed a patent application for this technology, indicating its significant commercial potential.
This work is part of a broader trend of applying artificial intelligence to complex scientific problems. Unlike traditional analytical methods, which require precise mathematical models, AI can identify patterns and connections in data that would otherwise remain hidden. “We are using artificial intelligence to solve problems that were traditionally considered unsolvable,” says study co-author Dr. Fang Ren. “It is about connecting experimental data with theoretical models in a way that was not previously possible.” The researchers demonstrated the effectiveness of their method on several nanocrystals, including gold nanoparticles and complex structures used in batteries. In all cases, PXRDnet was able to accurately determine the atomic arrangement, confirming its robustness and versatility. In addition, the tool is designed to be accessible to the scientific community—the team plans to make the model available online, enabling researchers around the world to analyze their own materials.
The discovery comes at a time of growing interest in advanced materials for sustainable technologies. Better batteries are crucial for the transition to renewable energy sources and electric mobility, while more efficient catalysts can reduce energy consumption in industrial processes. The ability to accurately characterize nanomaterials is essential in these fields, making the work of the Columbia Engineering team particularly timely. Professor Billinge emphasizes that although AI is a powerful tool, it does not replace human intuition and expertise. “Our model is most effective when it works in synergy with human scientists,” he explains. “It provides fast and accurate analyses, but interpreting the results and applying them requires a deep understanding of materials science.” This symbiosis between AI and human intelligence represents the future of scientific research—machines process data and identify patterns, while people provide context and direction. As nanotechnology and advanced materials become increasingly important in addressing global challenges, tools such as PXRDnet will play a key role in accelerating the pace of discovery. The ability to characterize new materials quickly and accurately can shorten the path from laboratory research to practical applications, potentially accelerating the development of cleaner and more efficient technologies.
