Most computational materials science studies perform or reference some sort of baseline process for choosing the DFT functional that they will employ to model their system. It is unlikely that a functional reproduces all properties of a material, but we try to choose one that gets it mostly right. Here I am looking for salient examples of this choice being critical to correctly predict some properties that do not immediately pop out of a relaxed structure. In other words, lattice constants are too easy.

A good answer could be a reference comparing functionals, or a series of references where perhaps an earlier one was acceptable but superceded, or especially where a change in functional gave opposing predictions of some trend or behavior. Ideally less obvious than the phase diagram of water and ices.

The opposite question asking for examples of agreement is here.

  • 1
    $\begingroup$ Does this answer your question? Examples of low sensitivity of observable behavior to the choice of DFT functional $\endgroup$ Commented May 13 at 16:38
  • 1
    $\begingroup$ @AndreaPellegrini I'm not sure why you bring up the linked question. The question you linked is also mine, and I set up both questions at the same time to link to each other saying that they are looking for the opposite answers. Could you tell me more about why you suggested it? $\endgroup$ Commented May 13 at 17:34
  • 1
    $\begingroup$ I'm okay with having two separate questions here! $\endgroup$ Commented May 16 at 15:24
  • 1
    $\begingroup$ Regarding frequencies, I know of a solid-state example that is not just 'quantitative error'. For rutile TiO2, LDA predicts it to be unstable (imaginary modes) while GGA gets it right-ish (all modes real, but there are deviations from expt. for other non-trivial reasons). Dunno if this helps. $\endgroup$ Commented May 21 at 21:00
  • 1
    $\begingroup$ @TylerSterling that seems like a nice example $\endgroup$ Commented May 23 at 14:54


You must log in to answer this question.