CASPT2 molecular geometries of Fe(II) spin-crossover complexes.

Using fully internally contracted (FIC)-CASPT2 analytical gradients, geometry optimizations of spin-crossover complexes are reported. This approach is tested on a series of Fe(II) complexes with different sizes, ranging from 13 to 61 atoms. A combination of active space and basis set choices are employed to investigate their role in determining reliable molecular geometries. The reported strategy demonstrates that a wave function-based level of theory can be used to optimize the geometries of metal complexes in reasonable times and enables one to treat the molecular geometry and electronic structure of the complexes using the same level of theory. For a series of smaller Fe(II) SCO complexes, strong field ligands in the LS state result in geometries with the largest differences between DFT and CASPT2; however, good agreement overall is observed between DFT and CASPT2. For the larger complexes, moderate sized basis sets yield geometries that compare well with DFT and available experimental data. We recommend using the (10e,12o) active space since convergence to a minimum structure was more efficient than with truncated active spaces despite having similar Fe-ligand bond distances.

Ligand-Centered Borenium Reactivity in Triaminoborane-Bridged Diphosphine Complexes.

Borenium ions (i.e., three-coordinate boron cations) are known to promote a wide variety of stoichiometric and catalytic reactions because of their high Lewis acidity. As demonstrated by the growing number of chemically reactive borane ligands, there is considerable interest in developing ligands with highly electrophilic boron sites that promote multisite reactivity in metal complexes. However, there are currently few examples of ligand-centered borenium ions, especially with ligands that form coordination complexes with a wide range of metals. Here we report borenium-like reactivity on a highly versatile diphosphine ligand. Treating (PhTBDPhos)NiCl2 (1) with strong Bronsted acids such as HBF4·Et2O, HOTf, or HNTf2 resulted in fluoride or chloride abstraction from BF4- or NiCl2, respectively, to form trans N-H and B-X bonds on the ligand backbone. HCl addition to the bridgehead N-B bond is reversible, and the reactivity depends on the identity of the supporting counteranions, as observed when treating [(PhTBDPhos)NiCl]2X2, where X = NTf2- (3), OTf- (4), or BArF4- (5), with HCl. The reaction of 4 with HNTf2 instead of HCl yielded NMR evidence of the latent borenium cation in solution and showed how poor nucleophiles such as triflate bind to the borenium ion in the solid state. Remarkably, replacing the chloride ligands in 1 with chelating and less-labile thiolates or catecholates, or changing the phosphorus substituents (phenyl to isopropyl), attenuates the reactivity on the ligand backbone. Density functional theory was used to quantify the reaction free energies, and the theoretical results corroborate the experimental observations. Given the broad utility of diphosphines in homogeneous catalysis and the known benefits of strong Lewis acid promotors in many catalytic reactions, we anticipate that the results will provide new opportunities for dual-site reactivity involving boron ligands and metals.

CatEmbed: A Machine-Learned Representation Obtained via Categorical Entity Embedding for Predicting Adsorption and Reaction Energies on Bimetallic Alloy Surfaces.

Machine-learning models for predicting adsorption energies on metallic surfaces often rely on basic elemental properties and electronic and geometric descriptors. Here, we apply categorical entity embedding, a featurization method inspired by natural language processing techniques, to predict adsorption energies on bimetallic alloy surfaces using categorical descriptors. Using this method, we develop a machine-learned representation from categorical descriptors (e.g., surface composition, adsorbate type, and site type) of the slab/adsorbate complex. By combining this representation with numerical features (e.g., slab metal stoichiometric ratios), we create the CatEmbed representation. Remarkably, decision tree models trained using CatEmbed, which includes no explicit geometric information, achieve a Mean Absolute Error (MAE) of 0.12 eV. Additionally, we extend this technique to predict reaction energies on bimetallic surfaces, creating the CatEmbed-React representation, which achieves an MAE of 0.08 eV. These findings highlight the effectiveness of categorical entity embedding for predicting adsorption and reaction energies on bimetallic alloy surfaces.