DFT-D4 goes periodic

London-dispersion effects are of great relevance to almost all aspects of materials science. Here, we present the adaptation and implementation of the DFT-D4 model [1] explicitly tailored for periodic systems. To correct a notorious overbinding of the original model in dense periodic systems, we introduce new reference polarizabilities for high coordination numbers, mostly for groups 1-5 in the Periodic Table. These are obtained by generalizing the reference systems and using the periodic electrostatic embedded cluster method. The method features analytical gradients and stress tensors to allow fast and reliable geometry optimizations and cell relaxations in periodic systems. Evaluating the new method, we consider a benchmark set of 16 solid state polarizabilities, for which the model achieves an unprecedented accuracy with a mean relative deviation of 9.8% (29.6% for D3). In addition mass densities of molecular crystals with diverse interaction motifs are much improved. Even compared to the computationally more demanding many-body dispersion expansion of Tkatchenko and coworkers we observe systematically improved results. Ultimately, we consider adsorption energies of organic molecules on non-polar as well as on ionic surfaces, for which we observe the largest improvements to previous models. We suggest DFT-D4 as a physically improved and more sophisticated dispersion model for periodic DFT calculations as well as other low-cost approaches like semi-empirical models employing periodic boundary conditions.

References:

1. E. Caldeweyher et al., J. Chem. Phys. 150, 154122 (2019). doi:10.1063/1.5090222