Title:Optimal transport distances to characterise electronic excitations

Abstract:Understanding the character of electronic excitations is important in computational and mechanistic studies, but their classification from numerical simulations remains an open problem despite significant progress in density-based and exciton wavefunction-based descriptors. We propose and investigate a new diagnostic based on the Sinkhorn divergence from optimal transport, which is highly sensitive to translations and hence promising for the identification of charge transfer excitations. In spite of this, we show through numerical simulations on a representative set of molecules that the new diagnostic is not able to separate charge transfer from Rydberg excitations in practice, which can be explained by its inability to distinguish between translational and diffusive processes. We trained a $k$-NN classification algorithm on the optimal transport diagonistic, the popular $\Lambda$ diagnostic as well as their combination, and assessed its performance in labelling excitations, finding that (i) The combination improves the classification, (ii) Rydberg excitations are not separated well in any setting, suggesting that key information on the diffusivity of the excited state is missing and (iii) the optimal transport diagnostic breaks down for charge transfer in small molecules.