Re: Problems with DFT for reaction barriers

[JOHNSONB -x- at -x- CPWSCA.PSC.EDU]

 Dear Dr. Creve,
 >From my own work and that of others, is has emerged that DFT has problems
 >in investigating transition structures for H-radical reactions (additions
 >and eliminations) with a low energy barrier. Often, the lack of exact
 >Hartree-Fock exchange in some functionals is blamed for this.
 We published a study on this type of problem after finding that
 regular Kohn-Sham DFT is generally a complete failure for radical H
 abstractions.  We took a step back and examined the simplest such
 reaction, H + H2 -> H2 + H, and found that much of the problem can be
 traced to the spurious self-interaction of the electrons in
 approximate density functionals.  The effect can be dramatic: for the
 LSDA, regular KS theory predicts H3 to be *stable* with respect to
 H + H2 (!) but after a simple approximate self-interaction correction
 (SIC) to the potential surface the transition structure was restored.
 The reference is
 B.G. Johnson, C.A. Gonzalez, P.M.W. Gill and J.A. Pople,
 Chem. Phys. Lett. 221, 100 (1994).
 I don't really agree with the lack of HF exchange as the explanation
 for the poor performance of DFT in the case of reaction barriers.
 "Hybrid" functionals like B3LYP have been shown to give good results
 on some reaction barriers in practice, but it is worrisome that the
 methods being mixed in these hybrids, i.e. Hartree-Fock and pure DFT,
 give reaction barriers which are significantly higher than experiment
 in one case (HF) and significantly lower than experiment in the other
 (DFT).  This leads one to wonder whether the good hybrid results
 simply come from fortuitous cancellation of errors -- the effect of
 the spurious residual self-energy in DFT roughly cancelling that of
 the neglect of electron correlation in Hartree-Fock.
 Having said this, SIC is certainly more difficult and expensive to do
 rigorously in practice than to use a hybrid HF-DFT functional; I do
 think, though, that one should be cautious of the results when using
 hybrid methods for this type of problem, for the reasons mentioned
 above.
 Regards,
 Benny Johnson
 -------------------------------------------------------------------------------
 Benny G. Johnson, Ph.D.		Phone:  (412) 828-7106
 President			Fax:    (412) 828-0483
 Q-Chem, Inc.			E-mail: johnsonb -x- at -x- psc.edu
 317 Whipple St.
 Pittsburgh, PA  15218
 USA
 -------------------------------------------------------------------------------