Supporting information for “Predicting with confidence the efficiency of new dyes in dye sensitized solar cells”

Free energy of oxidation and reorganization energy were approximated from the total energy differences computed using the B3LYP functional[1] and the 3-21G* basis set (all structures were optimized at this level). Polarized Continuum Model (PCM)[2] was included to mimic solvent effects in DSSCs, using the solvent parameters appropriate for acetonitrile. The model was built with a solvent excluding surface (SES), where the overlapping index between two interlocking spheres was 0.8 and the minimum radius was 0.5 Å. Construction of SES was based on the GePol method and the set of atomic radii was defined according to the UAKS model.[3] All quantum chemical calculations of this work were performed with Gaussian03.[4] The computational approach to calculate ΔG and λ was identical to that adopted by Maggio et. al. [5]. The largest dye considered (Dye 50) has 699 basis functions and the set of calculation cannot be performed on a semi-automatic fashion with a much larger basis set due to very slow convergence of the computational procedure. For this reason the 7 dyes from the original data set with more than 760 basis function (at the 3-21G* level) have not been included in the analysis. We have also excluded 2 dyes from the original work because they did not have a single carboxylic anchoring group and the single dye with efficiency 0.05% (clearly an outlier).