Machine-learning-assisted Synthesis of Polar Racemates

The dataset used here consists of 51 experiments transcribed from laboratory notebook records, using the ESCALATE "entity, materials, actions, observations" ontology. The composition spaces of the (CuO, MO2)/bpy/HF(aq) (M = Ti, Zr, Hf) systems were explored by varying the amounts of bpy and HF(aq) used in each reaction, while the amounts of CuO and MO2 (M = Ti, Zr, Hf) were held constant. In addition to this raw experimental data, additional calculated stoichiometric properties and computed electronic structure properties were added. Stoichiometric features, such as molar amounts and molar ratios, were calculated directly from the experimental observations. Electronic structure calculations were performed on the [TiF6]2−, [ZrF6]2−, and [HfF6]2− anionic building units to provide data on geometry, energetics, and charges using Gaussian 09, The B3LYP/LANL2DZ model chemistry was used as it provides good estimations (+-10 pm) for bond lengths of transition metal oxides and halides. Atomic charges were assessed for the optimized geometries using Mulliken, Hirshfeld, CM5, Natural Bond Orbital (NBO), and electrostatic potential fitting (Merz–Singh–Kollman using UFF radii, MKUFF) methods.