The X3LYP extended density functional accurately describes H-bonding but fails completely for stacking.

The performance of the recently introduced X3LYP density functional which was claimed to significantly improve the accuracy for H-bonded and van der Waals complexes was tested for extended H-bonded and stacked complexes (nucleic acid base pairs and amino acid pairs). In the case of planar H-bonded complexes (guanine...cytosine, adenine...thymine) the DFT results nicely agree with accurate correlated ab initio results. For the stacked pairs (uracil dimer, cytosine dimer, adenine...thymine and guanine...cytosine) the DFT fails completely and it was even not able to localize any minimum at the stacked subspace of the potential energy surface. The geometry optimization of all these stacked clusters leads systematically to the planar H-bonded pairs. The amino acid pairs were investigated in the crystal geometry. DFT again strongly underestimates the accurate correlated ab initio stabilization energies and usually it was not able to describe the stabilization of a pair. The X3LYP functional thus behaves similarly to other current functionals. Stacking of nucleic acid bases as well as interaction of amino acids was described satisfactorily by using the tight-binding DFT method, which explicitly covers the London dispersion energy.