Advancements in neural machinery have led to a wide range of algorithmic solutions for molecular property prediction. Two classes of models in particular have yielded promising results: neural networks applied to computed molecular fingerprints or expert-crafted descriptors and graph convolutional neural networks that construct a learned molecular representation by operating on the graph structure of the molecule. However, recent literature has yet to clearly determine which of these two methods is superior when generalizing to new chemical space. Furthermore, prior research has rarely examined these new models in industry research settings in comparison to existing employed models. In this paper, we benchmark models extensively on 19 public and 16 proprietary industrial data sets spanning a wide variety of chemical end points. In addition, we introduce a graph convolutional model that consistently matches or outperforms models using fixed molecular descriptors as well as previous graph neural architectures on both public and proprietary data sets. Our empirical findings indicate that while approaches based on these representations have yet to reach the level of experimental reproducibility, our proposed model nevertheless offers significant improvements over models currently used in industrial workflows.

Femtosecond time-resolved experiments demonstrate that the photoexcited state of perylene tetracarboxylic acid bisimide (PBI) aggregates in solution decays nonradiatively on a time-scale of 215 fs. High-level electronic structure calculations on dimers point toward the importance of an excited state intermolecular geometry distortion along a reaction coordinate that induces energy shifts and couplings between various electronic states. Time-dependent wave packet calculations incorporating a simple dissipation mechanism indicate that the fast energy quenching results from a doorway state with a charge-transfer character that is only transiently populated. The identified relaxation mechanism corresponds to a possible exciton trap in molecular materials.

Quantum chemical protocols explaining the crystal structures and the visible light absorption properties of 3,4:9,10-perylene tetracarboxylic acid bisimide (PBI) derivates are proposed. Dispersion-corrected density functional theory has provided an intermolecular potential energy of PBI dimers showing several energetically low-lying minima, which corresponds well with the packing of different PBI dyes in the solid state. While the dispersion interaction is found to be crucial for the binding strength, the minimum structures of the PESs are best explained by electrostatic interactions. Furthermore, a method is introduced, which reproduces the photon energies at the absorption maxima of PBI pigments within 0.1 eV. It is based on time-dependent Hartree-Fock (TD-HF) excitation energies calculated for PBI dimers with the next-neighbor arrangement in the pigment and incorporates crystal packing effects. This success provides clear evidence that the electronically excited states, which determine the color of these pigments, have no significant charge-transfer character. The developed protocols can be applied in a routine manner to understand and to predict the properties of such pigments, which are important materials for organic solar cells and (opto-)electronic devices.

The reliability of linear response approaches such as time-dependent Hartree-Fock (TD-HF) and time-dependent density functional theory (TD-DFT) for the prediction of the excited state properties of 3,4;9,10-tetracarboxylic-perylene-bisimide (PBI) aggregates is investigated. A dimer model of PBI is investigated as a function of a torsional motion of the monomers, which was shown before to be an important intermolecular coordinate in these aggregates. The potential energy curves of the ground state and the two energetically lowest neutral excited and charge-transfer (CT) states were obtained with the spin-component scaling modification of the approximate coupled-cluster singles-and-doubles (SCS-CC2) method as a benchmark for dispersion corrected TD-HF and a range of TD-DFT approaches. The highly accurate SCS-CC2 results are used to assess the other, computationally less demanding methods. TD-HF predicts similar potential energy curves and transition dipole moments as SCS-CC2, as well as the correct order of neutral and CT states. This supports an exciton trapping mechanism, which was found on the basis of TD-HF data. However, the investigated TD-DFT methods provide generally the opposite character for the excited states. As a consequence, these TD-DFT results have unacceptably large errors for optical properties of these dye aggregates.

The exciton diffusion length (LD) is a key parameter for the efficiency of organic optoelectronic devices. Its limitation to the nm length scale causes the need of complex bulk-heterojunction solar cells incorporating difficulties in long-term stability and reproducibility. A comprehensive model providing an atomistic understanding of processes that limit exciton trasport is therefore highly desirable and will be proposed here for perylene-based materials. Our model is based on simulations with a hybrid approach which combines high-level ab initio computations for the part of the system directly involved in the described processes with a force field to include environmental effects. The adequacy of the model is shown by detailed comparison with available experimental results. The model indicates that the short exciton diffusion lengths of α-perylene tetracarboxylicdianhydride (PTCDA) are due to ultrafast relaxation processes of the optical excitation via intermolecular motions leading to a state from which further exciton diffusion is hampered. As the efficiency of this mechanism depends strongly on molecular arrangement and environment, the model explains the strong dependence of LD on the morphology of the materials, for example, the differences between α-PTCDA and diindenoperylene. Our findings indicate how relaxation processes can be diminished in perylene-based materials. This model can be generalized to other organic compounds.