Orbital-dependent functionals: Hartree-Fock, Kohn-Sham exact-exchange and beyond

Prof Evert Jan Baerends, Vrije Universiteit, Amsterdam, Pays-Bas


Jeudi 23 juin 2005, 11h00, salle 2/3, Bâtiment St.Raphael, 3 rue Galilée, 94200 Ivry-sur-Seine
After the local-density approximation and the very successful generalized gradient approximations, it is often stated that using an exact exchange (EXX) functional is the next step up our ladder of DFT approximations. It would notably solve the self-interaction problem that plagues DFT. It only leaves the much smaller correlation energy to be treated with some approximate functional.
We will argue that the EXX approach cannot be considered a step forward compared to the GGAs. It reintroduces a very important flaw present in the Hartree-Fock approximation. Arguably the great success of DFT is a consequence of avoiding the typical Hartree-Fock error, obviating the distinction between exchange and correlation that is so ingrained in quantum chemistry after decades of the reigning paradigm: first Hartree-Fock, and then beyond.
We will argue that, once one is prepared to accept the computational complications and the price of dealing with orbital dependent functionals, it is still not necessary, and maybe even against the origin of the success of DFT, to go back to the exact-exchange starting point. It is possible to go back to the statistical definition of correlation in terms of two-electron probability distributions, or exchange and correlation holes,1 treating exchange and correlation jointly rather than separately, and devise orbital dependent functionals that incorporate exchange and correlation on equal footing.2,3 This is closely connected to density-matrix functional theory. We will demonstrate that such full exchange-correlation functionals can be formulated successfully, and present applications in the context of density-matrix functional theory.4 If such full exchange-correlation functionals will also be successful in Kohn-Sham DFT they may ultimately justify the increased computational expense of dealing with orbital-dependent functionals.

1. M. A. Buijse and E. J. Baerends, Mol. Phys. 100 (2002) 401.
2. E. J. Baerends, Phys. Rev. Lett. 87 (2001) 133004.
3. M. Grüning, O. V. Gritsenko and E. J. Baerends, J. Chem. Phys. 118 (2003) 7183.
4. O. Gritsenko, K. Pernal, E. J. Baerends, J. Chem. Phys., to be published.