34 5. Orbital Dependent <strong>Functional</strong>sBesides KLI, and LHF there exists several other ways of determ<strong>in</strong><strong>in</strong>g approximations of thelocal exchange potential, see e.g. [54, 52] for nice reviews, or [50, 53] for specific methods. Recently(2003) Perdew and Kümmel proposed an iterative approach for solv<strong>in</strong>g the OEP equation [55, 56].Start<strong>in</strong>g from e.g. the KLI potential, they claim that a converged solution of the OEP equationcan be reached with<strong>in</strong> 4-5 iterations of their scheme. This makes the OEP potential obta<strong>in</strong>able forpractical calculations, and has even revealed some surpris<strong>in</strong>g behaviors of the exact OEP potentialon nodal surfaces of the HOMO orbital.5.4 Screened <strong>Exchange</strong>A fundamental problem with <strong>in</strong>clud<strong>in</strong>g exact exchange <strong>in</strong>stead of the local approximations <strong>in</strong> thexc-functional, is that it is <strong>in</strong>compatible with local correlation approximations. Because of thesuccess of local xc-functionals, we know that there must be a large degree of cancellation betweenthe long range effects of exchange and correlation. This can not be exploited when us<strong>in</strong>g a nonlocalexchange functional and a local correlation functional. This problem can either be resolved bymak<strong>in</strong>g a proper non-local approximation for the correlation, or to screen the exchange, such that itbecomes local. The last procedure is obviously the simplest way of handl<strong>in</strong>g the problem, althoughit will remove some of the required features of exact exchange, i.e. the self-<strong>in</strong>teraction correctionis no longer complete, the 1/r decay of the exchange potential is no longer captured correctly,etc. Nevertheless functionals us<strong>in</strong>g screened exchange, seems to perform remarkably better thanord<strong>in</strong>ary exact exchange calculations <strong>in</strong> a wide range of areas. In addition, the computational costof evaluat<strong>in</strong>g exact exchange is improved by screen<strong>in</strong>g, as fewer k po<strong>in</strong>ts are needed for convergence,and <strong>in</strong> calculations us<strong>in</strong>g localized basis sets, the locality can be exploited to neglect the exchange<strong>in</strong>tegrals <strong>in</strong>volv<strong>in</strong>g states which are distant <strong>in</strong> space.The most successful implementation of screened exchange, is that of Heyd, Scuseria, andErnzerhof [38, 39]. In their approach, the Coulomb kernel is decomposed <strong>in</strong> a short range (SR)and a long range (LR) part us<strong>in</strong>g the errorfunction:1r = erfc(ωr) + erf(ωr)} {{r} } {{r}SRLRwhere ω is an adjustable screen<strong>in</strong>g length. Next, the hybrid methodE GGAx→ aE exactx+ (1 − a)E GGAxis only performed on the short ranged part of the separated xc-functionalE HSExcexact, SR= aExGGA, SR+ (1 − a)ExGGA, LR+ Ex+ EcGGAthis method reduces to the orig<strong>in</strong>al GGA <strong>in</strong> the ω → ∞ limit, and the hybrid version <strong>in</strong> the ω = 0limit. The only complication is that <strong>in</strong> order to screen the Coulomb kernel of the GGA exchangefunctional, it has to be put <strong>in</strong> the form∫Ex GGA =∫drn(r)dr ′ nGGA xc (r, r ′ )|r − r ′ |such that there is a potential which can be screened. This is not necessarily the standard formof the considered GGA functional, so further approximations has to be made, i.e. numericalexpansions of the analytic form of the functional. The authors have applied their method to thePBE functional, and the result<strong>in</strong>g functional is called HSE03. This screened hybrid functional hasbeen tested extensively <strong>in</strong> the articles [40, 41], and shown to give good results.
5.5 Conclusion 355.5 ConclusionTo <strong>in</strong>clude exact exchange self-consistently <strong>in</strong> the SCF, one must <strong>in</strong> some way have access tothe exchange potential. In the hybrid HF-KS schemes of section 3.2.1, the exchange potential issimply the non-local Fock potential of Hartree-Fock theory, this however redef<strong>in</strong>es the nature ofthe correlation functional, mak<strong>in</strong>g established approximations less usable. In Kohn-Sham theory,the exchange-correlation potential must be a local multiplicative potential. Construct<strong>in</strong>g a localpotential from an orbital dependent functional is a quite complicated, i.e. computationally timeconsum<strong>in</strong>g, process and approximations are needed to make the approach feasible <strong>in</strong> practice. Constructionof the local exchange potential can be achieved at several levels, climb<strong>in</strong>g the follow<strong>in</strong>gladder of accuracy (and complexity):• Slater potential: v x (r) = vxSla (r)• KLI: v x (r) = v Slax• LHF: v x (r) = v Slax(r) + ∑ i f i|φ i (r)| 2 〈i|ˆv x − ˆv NL |i〉/n(r)(r) + ∑ ij f if j φ ∗ i (r)φ j(r)〈j|ˆv x − ˆv xNL |i〉/n(r))φi (r ′ ) + c.c. = 0• OEP: ∑ i f i∫dr ′ φ ∗ i (r)G i(r, r ′ ) ( v x (r ′ ) − ˆv NLxwhere the OEP level can also be reached by an iterative scheme from one of the lower rungs ofthe ladder [55].x