Accurate prediction of structure and stability of molecular crystals is crucial in materials science and requires reliable modeling of long-range dispersion interactions. Semiempirical electronic structure methods are computationally more efficient than their ab initio counterparts, allowing structure sampling with significant speedups. We combine the Tkatchenko−Scheffler van der Waals method (TS) and the many-body dispersion method (MBD) with third-order density functional tight-binding (DFTB3) via a charge population-based method. We find an overall good performance for the X23 benchmark database of molecular crystals, despite an underestimation of crystal volume that can be traced to the DFTB parametrization. We achieve accurate lattice energy predictions with DFT+MBD energetics on top of vdW-inclusive DFTB3 structures, resulting in a speedup of up to 3000 times compared with a full DFT treatment. This suggests that vdW-inclusive DFTB3 can serve as a viable structural prescreening tool in crystal structure prediction.
■ INTRODUCTIONStability and structure prediction of molecular materials from first-principles electronic structure calculations bears significance to a wide range of problems ranging from pharmaceutical activity of drugs to optical properties of modern organic materials.1−3 The rugged and complex energy landscapes of molecular crystals give rise to the phenomenon of polymorphismthe ability of molecules to form different crystalpacking motifswhich is a crucial aspect in drug design, food chemistry, and organic semiconductor materials.4−7 Polymorphic materials exhibit many energetically close-lying minima, which can easily coexist and transform into each other at time scales that are inaccessible by conventional molecular dynamic simulations. Rigorous computational polymorph screening followed by correct stability ranking is therefore a crucial aspect for molecular crystal structure prediction (CSP).
8−11In recent years, DFT methods have become more reliable in predicting and ranking polymorphic systems due to the incorporation of efficient dispersion correction methods that, 9,12−15 at the same time, ensure computational feasibility. 10,16−19 In particular, the inclusion of beyond-pairwise dispersion interactions through the many-body dispersion (MBD) method coupled to semilocal DFT functionals has proven to be successful in this context. 10,15,20 To address larger length and time scales and more efficient structure prediction, several approximate electronic structure methods have been highly successful including semiempirical quantum chemical methods such as AM1, PM7, or the DFT-based densityfunctional tight binding (DFTB). 21−23 DFTB has been significantly improved recently, particularly in its description of charge polarization via third-order charge fluctuation corrections (DFTB3) 24,25 or its description of hydrogen bonding.26 Nevertheless, several shortcomings still persist that prohibit its use as standard tool in structure and stability prediction for molecular crystals. 22,2...