How many materials can fit in one chemical formula? Argonne National Laboratory and Northwestern University researchers gave a surprisingly specific answer to this question recently: up to ten crystal compounds with identical nominal formulas (BaSbQ3, Q = sulfur and tellurium) due to slight modifications of their occupancy.
What is the significance of this discovery? Superconductors, magnets, and quantum materials are generally found empirically and with some theoretical basis, sometimes by pure luck. In this case, however, the researchers maintained a constant ratio between barium, antimony, and chalcogens, but changed their exact ratio and configuration. Instead of creating a usual mixed solid solution, the system formed a number of completely separate structures: “The surprise was that as we added more sulfur, almost every sample turned out to be a different compound,” according to Xiuquan Zhou. “Every structure was different, but when we looked closer, we realized they were related by a mathematical relationship that put them all into the same family, called a homologous series.”
Such behavior belongs to the well-known phenomenon of multiple crystal structures at a given chemical composition, although the new study significantly extends this concept. Instead of discovering another phase, researchers managed to generate a whole homologous series of structures related to each other. The meaning is obvious: if it becomes possible to control structure formation so reliably, then it will be much easier to control properties.
For confirmation, the team applied such an extensive range of methods as synchrotron scattering and diffraction, electron microscopy, and spectroscopy. Why? Complex chalcogenides are known to show drastic variations in the electronic properties of the system depending on a small difference in atomic stacking or ordering. “Each compound is new, so each is worth investigating for the original goal of superconductivity, quantum phenomena and other exotic effects,” Hengdi Zhao commented. This is a chemically simple but methodologically significant case.
The most relevant aspect of this work relates to superconductors. Crystallographic structure plays a significant role in determining properties for high-temperature oxide superconductors, including critical current density. Moreover, recent research has shown that superconducting films are capable of maintaining stability at increased temperature and magnetic field strength if designed on the nanoscale level with an interface. Although there are no superconductors among the obtained crystals, this discovery provides a wider space for finding them without switching chemistry to the random method but searching for the desired effect in the family of similar structures.
In this sense, this finding comes at an interesting time for materials science since it is actively trying to incorporate human intuition into generative algorithms. Review articles devoted to AI models aimed at predicting crystal structure are constantly reminding us that modern machine learning systems are increasingly able to propose plausible material compositions and structures, but their training is limited by existing experience. Mercouri Kanatzidis puts it very concisely: Our goal is to discover new families. We want to stay ahead of AI so that if we succeed, we can train it on our knowledge.
And this is the actual lesson that can be learned from this study: it is not simply about generating several crystal structures. It concerns the possibility of forming a family of crystals based on compositional parameters that can be controlled experimentally and used for generating data needed for training machines.