A new inorganic homologous series emerges, offering a potentially endless ladder of related solid structures with predictable unit cells. The team from the United States behind this finding argues that mapping these structure–composition relationships could empower machine learning models to uncover novel inorganic materials more effectively.
Homologous series are a familiar concept in organic chemistry, where sequences of compounds built from repeating units can be described by a general formula. Classic examples include straight-chain alkanes and alkenes. Such orderly sequences can also appear in solid-state inorganic materials, though they are less common. Notable cases include non-stoichiometric titanium oxides and two-dimensional halide perovskites used in solar cells.
In this study, inorganic chemist Mercouri Kanatzidis of Northwestern University and collaborators identified a chain of barium-based compounds that reorganize into closely related solid structures as a single parameter shifts. “That means if you have one member and you know you’re in a sequence, you know what the next member is going to be,” Kanatzidis explains.
The team began with barium antimony telluride (BaSbTe3) and progressively replaced more of the tellurium with sulfur, a neighboring element in Group 16. While one might expect the sulfur and telluride to mix randomly on the anion sites, the researchers showed theoretically that sulfur, being more electronegative, preferentially occupies electron-rich sites in the crystal lattice. This preference triggers secondary effects on the overall structure, making the telluride ions become progressively electron-poor as sulfur content rises. “A simple solid solution doesn’t occur,” Kanatzidis notes. “Instead, the material responds by rearranging in a different ordered pattern, creating another member of the homology. That’s the part that’s truly extraordinary.”
Laying down chemical principles
The researchers synthesized ten members of this homologous series, each with increasing structural complexity. The series’ capstone, BaSbSTe2, exhibits an electronic instability known as a charge density wave. In many materials with charge density waves, superconductivity can emerge at high temperatures or under low-pressure conditions. The team now aims to leverage these findings to design, in a rational way, new superconductors—an achievement that currently eludes straightforward prediction.
Kanatzidis and colleagues also highlight a broader point about machine learning in materials science. While AI is increasingly used to design new structures, its strength lies in predicting materials within established structure types. As a result, these tools have found more success in fields like organic chemistry, where deeper, well-developed chemical principles exist, and have struggled to generate genuinely novel solid-state materials. Kanatzidis suggests that the concept of a phase homology could supply valuable training data, boosting AI’s predictive power for inorganic systems.
Materials scientist Leslie Schoop of Princeton University calls the work “very solid.” She adds that researchers will need to probe the new materials for any properties worth pursuing in detail. Schoop, who has previously voiced concerns about autonomous materials discovery, also believes that the links uncovered by Kanatzidis and colleagues could help AI tackle hard problems in inorganic chemistry. “We need to start embedding this kind of reasoning into AI algorithms to drive real material discoveries,” she says.
Would you agree that defining clear structure–composition relationships could unlock AI-driven breakthroughs in inorganic materials, or do you think the risks of overreliance on patterns could hinder true novelty? Share your thoughts in the comments.
