I am noticing a possible convergence problem in a magnetic $U$ system I am studying. In the OSZICAR file I often see the rms(c) value go to about 0.2e+3 and stabilize. Can anyone suggest what may be wrong in my calculation that can cause this?

I suspect the mixing parameters are the issue and I have tried the defaults as well as the pseudolinear mixing suggested by VASP of AMIX=0.3, BMIX=0.001.

Here is the start of the OSZICAR output. I am working on tungsten oxides using the materials project default parameters in terms of $U$, pseudopotentials, and general energy cutoffs.

       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1     0.996952880030E+03    0.99695E+03   -0.39229E+04  3360   0.140E+03
DAV:   2     0.420320154504E+02   -0.95492E+03   -0.89279E+03  3588   0.430E+02
DAV:   3    -0.925122529562E+02   -0.13454E+03   -0.13109E+03  3832   0.172E+02
DAV:   4    -0.981421469589E+02   -0.56299E+01   -0.55757E+01  3984   0.375E+01
DAV:   5    -0.983539220900E+02   -0.21178E+00   -0.21087E+00  5016   0.665E+00
DAV:   6    -0.983570555376E+02   -0.31334E-02   -0.31033E-02  4224   0.800E-01
DAV:   7    -0.983571847233E+02   -0.12919E-03   -0.12837E-03  4920   0.162E-01
DAV:   8    -0.983571860469E+02   -0.13235E-05   -0.13192E-05  2656   0.166E-02    0.662E+01
DAV:   9    -0.804838944016E+02    0.17873E+02   -0.44298E+01  4544   0.495E+01    0.381E+01
DAV:  10    -0.811208958861E+02   -0.63700E+00   -0.38276E+01  4936   0.568E+01    0.188E+03
DAV:  11    -0.814674048119E+02   -0.34651E+00   -0.42624E+00  3756   0.964E+00    0.188E+03
DAV:  12    -0.803474022983E+02    0.11200E+01   -0.27246E-01  4264   0.268E+00    0.184E+03
DAV:  13    -0.803447815420E+02    0.26208E-02   -0.15007E-02  3696   0.671E-01    0.184E+03
DAV:  14    -0.827481331115E+02   -0.24034E+01   -0.11705E+00  5324   0.101E+01    0.188E+03
DAV:  15    -0.830941535514E+02   -0.34602E+00   -0.19499E-01  4256   0.245E+00    0.188E+03
DAV:  16    -0.827076829935E+02    0.38647E+00   -0.19109E-01  4028   0.362E+00    0.188E+03
DAV:  17    -0.839071530160E+02   -0.11995E+01   -0.18920E+00  4072   0.775E+00    0.188E+03
DAV:  18    -0.852362012098E+02   -0.13290E+01   -0.19289E+00  4272   0.700E+00    0.188E+03
DAV:  19    -0.850163602711E+02    0.21984E+00   -0.43203E-01  3960   0.207E+00    0.188E+03
DAV:  20    -0.853790409329E+02   -0.36268E+00   -0.21727E+01  4452   0.694E+00
   1 F= -.85379041E+02 E0= -.85379041E+02  d E =-.853790E+02  mag=    -0.3688
quench:  g(F)=  0.000E+00 g(S)=  0.000E+00 dE (1.order)= 0.000E+00
 bond charge predicted
  • 1
    $\begingroup$ Others might be able to provide more insight if you add in what class of materials you're studying here, broadly speaking. $\endgroup$ Dec 30 '20 at 6:56
  • $\begingroup$ @AndrewRosen I have included some extra details. My suspected fix of raising kpoints seems to have only slightly fixed the problem, I am still having trouble with this so I will be trying some of your suggestions on discord as well. $\endgroup$ Dec 30 '20 at 18:41
  • $\begingroup$ @TristanMaxson Did you use LMAXMIX? For systems with d/f electrons when performing spin polarized +U calculations, you necessarily need to include LMAXMIX. $\endgroup$
    – Xivi76
    Jan 3 '21 at 19:46
  • $\begingroup$ @TristanMaxson Are you using the W_pv POTCAR? That one is really dodgy (which is used in the Material Project) and was later removed in the PBE 54 set (cms.mpi.univie.ac.at/marsweb/index.php/news/36-highlights/…). From my experience using W_sv gives better convergence for the electronic loops. $\endgroup$
    – bzbzbz
    Apr 20 '21 at 8:27
  • $\begingroup$ I am using the W_sv POTCAR not the W_pv, the W_pv POTCAR was almost completely unusable. $\endgroup$ Apr 20 '21 at 11:52

Judging by your initial energies, it looks like you're starting a calculation from scratch. I've had good luck so far converging spin-polarized calculations by first doing a non-spin-polarized calculation, writing the CHG, CHGCAR, and WAVECAR files, and using those to start a spin-polarized calculation. This is the recommended method if such calculations aren't converging reliably according to the wiki.

In addition, I've found ALGO=All to converge sometimes when the default doesn't.

  • $\begingroup$ +1 Welcome to our community, and thank you for contributing this answer to a long-standing unanswered question! We hope to see much more of you in the future! I just made some slight edits that it might be helpful for you to look at. $\endgroup$ Jun 23 '21 at 18:32
  • $\begingroup$ Welcome, @Eric Taw! I just wanted to say I agree with your answer. :) $\endgroup$ Jun 24 '21 at 2:43

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.