Benchmarking Computational Approaches to Predict the UV-Vis Spectra of Organic Photocatalysts

In this work, we present a new benchmark strategy and use it to evaluate the predictions given by several DFT functionals for the UV-vis absorption spectra of recently developed organic photocatalyst molecules with sizes about 50-100 atoms. Our approach is based on a spectral fitting procedure that takes excitation energies and oscillator strengths, then it uses Gaussians with two parameters to reproduce the experimental spectra. One parameter is responsible for broadening, while the other is a scaling factor to stretch or compress the calculated excitation spectra. These parameters are optimized to achieve the best possible fit of full spectral shapes in the UV-vis region. The goodness of the fit can be evaluated using any common error metric (i.e., MAE, MSE, R2, etc.) but we found that all of them yield very similar results. The functionals that are best able to match the experiments are the B2PLYP, ω-B2PLYP, CAM-B3LYP, ω-B97XD, M06-2X and M06 in approximately this order (the order somewhat depends on the error metric). This result reflects the hierarchy of the DFT approximations, however, the range separated ω-B2PLYP, CAM-B3LYP, ω-B97XD functionals and M06-2X require a significant scaling of excitation energies, making them less suited for real-world predictions. Also, we obtained a highly unexpected, overall bad performance from the DSD-BLYP double hybrid. The B2PLYP and the M06 are the only functionals that are accurate enough to consistently predict spectra within a ±10% scaling window. B2PLYP is exceptionally good in this regard with almost 90% of the fitted spectra falling into the ±10% scaling window. This is the same level of accuracy as what the more demanding STEOM-DLPNO-CCSD method provides, which we have also investigated with the aim to determine if its predictions can be used as theoretical best estimates for real-life molecules such as those in our benchmark set. We can also conclude that the B2PLYP and M06 functionals are reliable DFT methods to calculate spectra of organic photocatalysts with central chromophores included in this benchmark even without any wavelength adjustment.