Comment on “surface characterization of cerium oxide catalysts using deep learning with infrared spectroscopy of CO”

We demonstrate that the predicted distribution of ceria facets in CeO2 nanorods, derived from the analysis of the measured IR spectrum of CO by fitting it using experimentally or DFT(HSE) calculated CO vibrational frequencies collected on all three low-index ceria surfaces, does not align with that derived using deep learning methods [1]. The latter utilizes a dataset based on extensive DFT(PBE) + U calculations, which are erroneous and have thus been scaled to the experimental values.