Hartree-Fock density is free of self-interaction but lacks electron correlation effects, while the density from KS-DFT (using an xc functional or potential, both which are explicitly density-dependent and gradient-dependent) suffers from self-interaction errors but does incorporate electron correlation effects.

When is the HF density better, and when is the KS density better?

  • 2
    $\begingroup$ opinion is a "bad" word on SE. Therefore, consider refining your question to ask for factual evidence $\endgroup$ – Cody Aldaz Jun 9 at 4:37
  • $\begingroup$ +1. I have reworded the sentence so that it asks an objective question rather than an opinion-based one, then I also deleted the second question, about how to analyze them, since that should be asked as a separate question. To keep the answers focused, each question should ask only one question. $\endgroup$ – Nike Dattani Jun 9 at 4:39

Kieron Burke and co-workers have shown that in many cases one can get better results by using the HF density as input to a DFT-XC evaluation of the energy, as opposed to using the DFT-XC to generate a self-consistent density. This has been termed density-driven error, as opposed to functional-error. SE is especially problematic in cases where an electron is weakly bound, as in many (most) anions. Even small SE in these cases may lead to an unbound electron, as is visible by a positive HOMO energy. Any standard basis set, however, will not allow the electron to escape, and thus produces an energy corresponding to the DFT-XC evaluated on an artificially constrained density. This leads to the energy being sensitive to the size of the basis set, especially including more and more diffuse functions will lead to an energy drift.

| cite | improve this answer | |
  • 1
    $\begingroup$ But I think the cases that have been published are strictly systems where DFAs get the density wrong.. It's possible that systems where DFAs are unproblematic give the opposite result. $\endgroup$ – Susi Lehtola Jun 9 at 17:44
  • $\begingroup$ There is conceptually related work using Optimised Effective Potential (OEP) and more recently Local Fock Exchange: "A local Fock-exchange potential in Kohn–Sham equations", T.W. Hollins, S.J. Clark, K. Refson and N.I. Gidopoulos, Journal of Physics: Condensed Matter, Volume 29, Number 4 (2016); doi.org/10.1088/1361-648X/29/4/04LT01 $\endgroup$ – Phil Hasnip Jun 29 at 2:19

I guess this will go in the answer slot, it is a bit long for a comment.

DFT typically has quite a bit less spin contamination than HF (attributed to the inclusion of correlation).

One issue however is that DFT tends to favor electron delocalization, and conveniently HF tends to the opposite. This is part of why hybrid DFT methods get improvements, the over/under electron delocalization is "averaged out". Hybrids also benefit from the HF exchange not suffering from self-interaction, however, range separation methods also correct for self-interaction errors if you don't want to use a hybrid (or apply range-separation to a hybrid for more of a win).

In Hartree-Fock virtual orbitals are akin to electron affinity(determined in the field of N electrons), whereas in DFT they are akin to exciting an electron(determined in a field of N-1 electrons).

Actually saying when HF is better or worse than DFT seems quite difficult to me since you must say what functional you are comparing HF to. There are many ways to fix-up a given DFT functional to overcome deficiencies, two well known corrections being range-separation and dispersion corrections.

To give some sort of summary to what I have said, hopefully others will be more prolific in their answers my instincts say that


  1. HF would possibly be better for electron affinity when using a frozen orbital assumption (assuming a low quality bare-bones DFT functional).
  2. DFT provides a better density for systems where spin-contamination is an issue.

I look forward to more answers on this one.

| cite | improve this answer | |
  • 1
    $\begingroup$ Restricted HF should have zero spin-contamination right? $\endgroup$ – Nike Dattani Jun 9 at 15:10
  • $\begingroup$ Yep, although, uses for RHF can be somewhat restricted. $\endgroup$ – Charlie Crown Jun 9 at 17:10
  • 1
    $\begingroup$ IIRC virtual orbitals are quite bad in DFT... $\endgroup$ – Susi Lehtola Jun 9 at 17:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.