The generation of plausible crystal structures is often the first step in predicting the structure and properties of a material from its chemical composition. However, most current methods for crystal structure prediction are computationally expensive, slowing the pace of innovation. Seeding structure prediction algorithms with quality generated candidates can overcome a major bottleneck. Here, we introduce CrystaLLM, a methodology for the versatile generation of crystal structures, based on the autoregressive large language modeling (LLM) of the Crystallographic Information File (CIF) format. Trained on millions of CIF files, CrystaLLM focuses on modeling crystal structures through text. CrystaLLM can produce plausible crystal structures for a wide range of inorganic compounds unseen in training, as demonstrated by ab initio simulations. Our approach challenges conventional representations of crystals, and demonstrates the potential of LLMs for learning effective models of crystal chemistry, which will lead to accelerated discovery and innovation in materials science.

The carbon–carbon coupling via electrochemical reduction of carbon dioxide represents the biggest challenge for using this route as platform for chemicals synthesis. Here we show that nanostructured iron (III) oxyhydroxide on nitrogen-doped carbon enables high Faraday efficiency (97.4%) and selectivity to acetic acid (61%) at very-low potential (−0.5 V vs silver/silver chloride). Using a combination of electron microscopy, operando X-ray spectroscopy techniques and density functional theory simulations, we correlate the activity to acetic acid at this potential to the formation of nitrogen-coordinated iron (II) sites as single atoms or polyatomic species at the interface between iron oxyhydroxide and the nitrogen-doped carbon. The evolution of hydrogen is correlated to the formation of metallic iron and observed as dominant reaction path over iron oxyhydroxide on oxygen-doped carbon in the overall range of negative potential investigated, whereas over iron oxyhydroxide on nitrogen-doped carbon it becomes important only at more negative potentials.

Redox behaviour of magnetite Fe₃O₄nanoparticles in thermodynamic equilibrium conditions enclosed by non-dipolar reconstructed surfaces.

The crystal structure of the iron oxide γ-Fe₂O₃ is usually reported in either the cubic system (space group P4(3)32) with partial Fe vacancy disorder or in the tetragonal system (space group P4(1)2(1)2) with full site ordering and c/a≈3. Using a supercell of the cubic structure, we obtain the spectrum of energies of all the ordered configurations which contribute to the partially disordered P4(3)32 cubic structure. Our results show that the configuration with space group P4(1)2(1)2 is indeed much more stable than the others, and that this stability arises from a favourable electrostatic contribution, as this configuration exhibits the maximum possible homogeneity in the distribution of iron cations and vacancies. Maghemite is therefore expected to be fully ordered in equilibrium, and deviations from this behaviour should be associated with metastable growth, extended anti-site defects and surface effects in the case of small nanoparticles. The confirmation of the ordered tetragonal structure allows us to investigate the electronic structure of the material using density functional theory (DFT) calculations. The inclusion of a Hubbard (DFT + U) correction allows the calculation of a band gap in good agreement with experiment. The value of the gap is dependent on the electron spin, which is the basis for the spin-filtering properties of maghemite.