It is well-established that within the DFT+U framework, one can achieve several self-consistent solutions with greatly varying energies for the same exact magnetic ordering. In 2010, Meredig et al. showed that a relatively simple "U-ramping" method could be employed to more easily identify the true ground state solution for troublesome DFT+U cases.

Since it has been over a decade since the aforementioned paper, is this still considered to be the optimal approach when carrying out DFT+U calculations? Are there any other methods commonly used to identify the ground state solution when carrying out DFT+U calculations?

  • 1
    $\begingroup$ As it's been close to a year since this question was asked, I wonder if you learned anything about this problem over the last 11 months that you could share with us in an update? $\endgroup$ Jan 26, 2022 at 3:42
  • $\begingroup$ My experience is with Hartree-Fock and post-HF methods like coupled-cluster and CI, much more than with DFT, but the method that I use to find lower and lower self-consistent HF solutions is to form the orbital Hessian and diagonalize it to find any negative eigenvalues, then use something like HFSTABILITY=FOLLOW in CFOUR or Aces2 to "follow" the eigenvectors to the lowest eigenvalue of the orbital Hessian. $\endgroup$ Jan 29, 2022 at 22:52
  • $\begingroup$ Table 1 of my 2021 paper shows that doing this makes a big difference not only in the HF energy but also for the post-HF. Gaussian is able to do this type of calculation to find lower and lower self-consistent solutions in DFT, using the keyword Stable (click on Options here then see Opt which does repeated reoptimizations until the lowest-energy solution is found). Is this something that can be done with DFT+U? The paper you mentioned has 247 citations on GS, anything useful there? $\endgroup$ Jan 29, 2022 at 23:00
  • 3
    $\begingroup$ My consensus from casual conversations with people is that this is an oft-neglected topic. Perhaps it's something that is convenient to "sweep under the rug." Most plane-wave DFT codes, which is where most DFT+U is done, unfortunately do not come with wavefunction stability analyses like in Gaussian, but perhaps an analogous idea could be applied here. $\endgroup$ Jan 30, 2022 at 0:46
  • $\begingroup$ Anecdotal comment as someone running too much DFT+U, I find that sometimes calculations converge to the same exact magnetic state and identical positions down to some precision, but the only time I have had problems is when the SCF cycles run unstable. If you have stable convergence behavior, then I don't see this occurring at least qualitatively. $\endgroup$ Jan 30, 2022 at 7:45


You must log in to answer this question.