An improved density matrix functional ͓correction to Buijse and Baerends functional ͑BBC͔͒ is proposed, in which a hierarchy of physically motivated repulsive corrections is employed to the strongly overbinding functional of Buijse and Baerends ͑BB͒. The first correction C1 restores the repulsive exchange-correlation ͑xc͒ interaction between electrons in weakly occupied natural orbitals ͑NOs͒ as it appears in the exact electron pair density 2 for the limiting two-electron case. The second correction C2 reduces the xc interaction of the BB functional between electrons in strongly occupied NOs to an exchange-type interaction. The third correction C3 employs a similar reduction for the interaction of the antibonding orbital of a dissociating molecular bond. In addition, C3 applies a selective cancellation of diagonal terms in the Coulomb and xc energies ͑not for the frontier orbitals͒. With these corrections, BBC still retains a correct description of strong nondynamical correlation for the dissociating electron pair bond. BBC greatly improves the quality of the BB potential energy curves for the prototype few-electron molecules and in several cases BBC reproduces very well the benchmark ab initio potential curves. The average error of the self-consistent correlation energies obtained with BBC3 for prototype atomic systems and molecular systems at the equilibrium geometry is only ca. 6%.

A new density functional (DF) method is proposed for calculations of intermolecular interaction energies. The exchange-correlation functional was optimized in such a way that the method recovers the interaction energies with the dispersion (including exchange-dispersion) component subtracted and therefore our approach is named the dispersionless DF (dlDF) method. The dlDF method is shown to predict very well the dispersionless part of the interaction energy for all types of intermolecular interactions. Thus, if combined with a dispersion component, computed ab initio or from a simple function fitted to ab initio values, it provides accurate and physically justified interaction energies in the whole range of intermolecular separations. Our dispersion function is significantly more accurate than the published ones.