DFPT convergence criteria

[tania_sandoval]

Hello everybody,

Currently I am computing DFPT calculations to obtain born effective charges. I am concerned about the convergence of different cycles during the calculation.
My input is the following:

Code: Select all

!   ###   Electronic   ###
ALGO = Fast
PREC = Accurate
ENCUT = 400
EDIFF = 1E-8
ISMEAR = 0
SIGMA = 0.05
IVDW = 12
ISPIN = 1
LREAL = .FALSE.
!   ###   GEOMETRY   ###
IBRION = -1
ISIF = 2
NSW = 0
!   ###   RESPONSE   ###
LEPSILON = .TRUE.
LRPA = .FALSE.

I am using a convergence criteria of 1E-8 for tight convergence calculation. This criteria is achieved for the electronic SCF initial calculation. But for the following cycles (Linear Response G (3 directions), Linear response to external field (no local field effect) (3 directions), Linear response to external field (3 directions)), the cycles CONVERGE, but did not achieve the desired criteria. Mainly, the full field effect on the Linear response to external field did not be satisfactory IMHO:

Code: Select all

Entering main loop
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1     0.827768530909E+02    0.82777E+02   -0.37289E+03    96   0.222E+02
...
RMM:  34    -0.635354747403E+02   -0.29256E-07   -0.89351E-11    83   0.597E-05    0.182E-05
RMM:  35    -0.635354747499E+02   -0.95891E-08   -0.11781E-10    82   0.633E-05
   1 F= -.63747406E+02 E0= -.63747406E+02  d E =-.460711E-13
 Linear response reoptimize wavefunctions to high precision
DAV:   1    -0.635354747561E+02    0.21193E+00    0.35393E-08    80   0.423E-05
DAV:   2    -0.635354747566E+02   -0.46089E-09    0.35591E-08    96   0.419E-06
DAV:   3    -0.635354747564E+02    0.23851E-09    0.35581E-08    96   0.782E-07
 Linear response G [H, r] |phi>, progress :
  Direction:   1
       N       E                     dE             d eps       ncg     rms
RMM:   1    -0.918976332341E+02   -0.91898E+02    0.11799E+02   154   0.448E+01
...
RMM:   9    -0.893369002996E+02    0.80117E-06   -0.68177E-04   157   0.354E-02
 Linear response G [H, r] |phi>, progress :
  Direction:   2
       N       E                     dE             d eps       ncg     rms
RMM:   1    -0.919183278107E+02   -0.91918E+02    0.11783E+02   150   0.448E+01
...
RMM:   9    -0.893429474054E+02    0.26500E-05   -0.60698E-04   158   0.298E-02
 Linear response G [H, r] |phi>, progress :
  Direction:   3
       N       E                     dE             d eps       ncg     rms
RMM:   1    -0.853916058547E+02   -0.85392E+02    0.12973E+02   133   0.415E+01
...
RMM:   9    -0.891791457158E+02   -0.84460E-07   -0.38656E-04   150   0.235E-02
 Linear response to external field (no local field effect), progress :
  Direction:   1
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.567460138539E+00   -0.56746E+00   -0.12415E+00   148   0.201E+00
..
RMM:   8    -0.576946592500E+00   -0.67951E-08   -0.36534E-06   150   0.214E-03
 change of polarisation eV/A/(eV/A) component  1 :     1.154     0.000     0.000
 dielectric tensor                  component  1 :     1.008     0.000     0.000
 Linear response to external field (no local field effect), progress :
  Direction:   2
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.567654419809E+00   -0.56765E+00   -0.12425E+00   148   0.201E+00
...
RMM:   9    -0.577144624316E+00   -0.12221E-08   -0.25083E-06   149   0.179E-03
 change of polarisation eV/A/(eV/A) component  2 :     0.000     1.154     0.000
 dielectric tensor                  component  2 :     0.000     1.008     0.000
 Linear response to external field (no local field effect), progress :
  Direction:   3
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.515097938259E+00   -0.51510E+00   -0.78025E-01   156   0.192E+00
...
RMM:   8    -0.519427088710E+00   -0.18201E-08   -0.24448E-06   151   0.166E-03
 change of polarisation eV/A/(eV/A) component  3 :     0.000     0.000     1.039
 dielectric tensor                  component  3 :     0.000     0.000     1.007
 Linear response to external field, progress :
  Direction:   1
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.567460138539E+00   -0.56746E+00   -0.12415E+00   148   0.201E+00
RMM:   2    -0.576728346163E+00   -0.92682E-02   -0.10743E-02   134   0.351E-01    0.130E+00
RMM:   3    -0.226232663330E+00    0.35050E+00    0.14631E-01   133   0.703E-01    0.679E-01
RMM:   4     0.749555101135E+00    0.97579E+00    0.23777E+00   138   0.769E-01    0.226E-01
RMM:   5     0.166197002797E+01    0.91241E+00    0.98273E+00   148   0.300E-01    0.107E-01
RMM:   6     0.228706790562E+01    0.62510E+00    0.16155E+01   148   0.209E-01    0.103E-01
RMM:   7     0.239241851595E+01    0.10535E+00    0.16717E+01   149   0.154E-01    0.438E-02
RMM:   8     0.287515979254E+01    0.48274E+00    0.21204E+01   154   0.165E-01    0.293E-02
RMM:   9     0.223896186431E+01   -0.63620E+00    0.15079E+01   160   0.130E-01    0.158E-02
RMM:  10     0.233487766022E+01    0.95916E-01    0.16809E+01   161   0.126E-01    0.109E-02
RMM:  11     0.253341946396E+01    0.19854E+00    0.18794E+01   166   0.105E-01    0.594E-03
RMM:  12     0.255572234719E+01    0.22303E-01    0.19089E+01   172   0.101E-01    0.501E-03
RMM:  13     0.254324301252E+01   -0.12479E-01    0.18982E+01   175   0.958E-02    0.183E-03
RMM:  14     0.255642517585E+01    0.13182E-01    0.19116E+01   185   0.937E-02    0.120E-03
RMM:  15     0.259767001887E+01    0.41245E-01    0.19515E+01   181   0.939E-02    0.680E-04
RMM:  16     0.260737840312E+01    0.97084E-02    0.19629E+01   187   0.931E-02    0.452E-04
RMM:  17     0.262848588913E+01    0.21107E-01    0.19824E+01   184   0.931E-02    0.244E-04
RMM:  18     0.263377801443E+01    0.52921E-02    0.19898E+01   186   0.929E-02    0.198E-04
RMM:  19     0.265790853766E+01    0.24131E-01    0.20123E+01   186   0.929E-02    0.107E-04
RMM:  20     0.265110618313E+01   -0.68024E-02    0.20069E+01   184   0.929E-02    0.109E-04
RMM:  21     0.267061653125E+01    0.19510E-01    0.20255E+01   185   0.929E-02    0.637E-05
RMM:  22     0.265639504984E+01   -0.14221E-01    0.20115E+01   186   0.929E-02    0.937E-05
RMM:  23     0.267168300469E+01    0.15288E-01    0.20269E+01   187   0.928E-02    0.737E-05
RMM:  24     0.265987466449E+01   -0.11808E-01    0.20131E+01   185   0.928E-02    0.188E-04
RMM:  25     0.266684761451E+01    0.69730E-02    0.20136E+01   187   0.926E-02    0.235E-04
RMM:  26     0.266792850989E+01    0.10809E-02    0.20075E+01   184   0.926E-02    0.253E-04
RMM:  27     0.265992044856E+01   -0.80081E-02    0.19987E+01   183   0.925E-02    0.235E-04
 change of polarisation eV/A/(eV/A) component  1 :     0.795     0.000     0.000
 dielectric tensor                  component  1 :     1.005     0.000     0.000
 Linear response to external field, progress :
  Direction:   2
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.567654419809E+00   -0.56765E+00   -0.12425E+00   148   0.201E+00
RMM:   2    -0.576919132733E+00   -0.92647E-02   -0.10678E-02   138   0.353E-01    0.130E+00
RMM:   3    -0.183312366723E+00    0.39361E+00    0.56802E-01   133   0.703E-01    0.678E-01
RMM:   4     0.115377276276E+01    0.13371E+01    0.63427E+00   139   0.781E-01    0.228E-01
RMM:   5     0.357301834028E+01    0.24192E+01    0.28751E+01   148   0.332E-01    0.112E-01
RMM:   6     0.412569511990E+01    0.55268E+00    0.34498E+01   152   0.226E-01    0.950E-02
RMM:   7     0.524621494205E+01    0.11205E+01    0.45235E+01   151   0.187E-01    0.429E-02
RMM:   8     0.582597450630E+01    0.57976E+00    0.50540E+01   151   0.203E-01    0.287E-02
RMM:   9     0.648992415168E+01    0.66395E+00    0.59196E+01   158   0.180E-01    0.216E-02
RMM:  10     0.596799627279E+01   -0.52193E+00    0.51975E+01   162   0.155E-01    0.951E-03
RMM:  11     0.622900694691E+01    0.26101E+00    0.54687E+01   166   0.148E-01    0.745E-03
RMM:  12     0.580134662589E+01   -0.42766E+00    0.50476E+01   175   0.138E-01    0.311E-03
RMM:  13     0.602102551240E+01    0.21968E+00    0.52642E+01   176   0.134E-01    0.202E-03
RMM:  14     0.583105846533E+01   -0.18997E+00    0.50805E+01   178   0.133E-01    0.126E-03
RMM:  15     0.599092818994E+01    0.15987E+00    0.52390E+01   179   0.133E-01    0.104E-03
RMM:  16     0.584533005947E+01   -0.14560E+00    0.50953E+01   183   0.132E-01    0.856E-04
RMM:  17     0.596796530214E+01    0.12264E+00    0.52166E+01   187   0.132E-01    0.749E-04
RMM:  18     0.584981528301E+01   -0.11815E+00    0.51001E+01   185   0.132E-01    0.315E-04
RMM:  19     0.594036601690E+01    0.90551E-01    0.51891E+01   185   0.132E-01    0.400E-04
RMM:  20     0.629727609174E+01    0.35691E+00    0.55477E+01   186   0.132E-01    0.404E-04
RMM:  21     0.613773044902E+01   -0.15955E+00    0.53860E+01   187   0.131E-01    0.434E-04
 change of polarisation eV/A/(eV/A) component  2 :     0.000     0.795     0.000
 dielectric tensor                  component  2 :     0.000     1.005     0.000
 Linear response to external field, progress :
  Direction:   3
       N       E                     dE             d eps       ncg     rms          rms(c)
RMM:   1    -0.515097938259E+00   -0.51510E+00   -0.78025E-01   156   0.192E+00
RMM:   2    -0.519302437681E+00   -0.42045E-02   -0.68161E-03   152   0.291E-01    0.135E+00
RMM:   3    -0.116681091292E+00    0.40262E+00   -0.17634E-01   147   0.885E-01    0.619E-01
RMM:   4     0.512193954952E+00    0.62888E+00   -0.12184E-01   141   0.826E-01    0.220E-01
RMM:   5     0.758843268250E+00    0.24665E+00   -0.24307E-02   140   0.307E-01    0.976E-02
RMM:   6     0.673933111343E+00   -0.84910E-01   -0.22649E-03   141   0.984E-02    0.631E-02
RMM:   7     0.644818385230E+00   -0.29115E-01   -0.43250E-03   137   0.132E-01    0.478E-02
RMM:   8     0.708684099843E+00    0.63866E-01   -0.66866E-04   148   0.496E-02    0.228E-02
RMM:   9     0.674364560091E+00   -0.34320E-01   -0.25343E-03   142   0.937E-02    0.115E-02
RMM:  10     0.680315651578E+00    0.59511E-02   -0.35767E-04   142   0.355E-02    0.976E-03
RMM:  11     0.664238924829E+00   -0.16077E-01   -0.26194E-04   143   0.296E-02    0.310E-03
RMM:  12     0.664453710883E+00    0.21479E-03   -0.97849E-06   150   0.563E-03    0.252E-03
RMM:  13     0.662250158998E+00   -0.22036E-02   -0.17386E-05   148   0.858E-03    0.115E-03
RMM:  14     0.663639857564E+00    0.13897E-02   -0.22302E-06   153   0.284E-03    0.565E-04
RMM:  15     0.663291061505E+00   -0.34880E-03   -0.12916E-06   152   0.219E-03    0.378E-04
RMM:  16     0.663468681402E+00    0.17762E-03   -0.14240E-06   150   0.243E-03    0.228E-04
RMM:  17     0.663532976206E+00    0.64295E-04   -0.79124E-07   149   0.107E-03    0.153E-04
RMM:  18     0.663324773960E+00   -0.20820E-03   -0.99720E-07   150   0.139E-03    0.616E-05
RMM:  19     0.663343412325E+00    0.18638E-04   -0.13953E-06   149   0.105E-03    0.359E-05
RMM:  20     0.663349291903E+00    0.58796E-05   -0.32603E-06   149   0.107E-03    0.138E-05
RMM:  21     0.663350599752E+00    0.13078E-05   -0.99049E-06   149   0.101E-03    0.904E-06
 change of polarisation eV/A/(eV/A) component  3 :     0.000     0.000     0.597
 dielectric tensor                  component  3 :     0.000     0.000     1.004
Born effective charges drift removed

And finally, My concern is if this "non converged" cycles affect on the computed born effective charges.
Any explanation or help, thanks.

Regards

[marie-therese.huebsch]

Hi, thank you for posting your question.

The first thing that comes to mind is that the starting point is not good because you use ALGO=Fast. Have you tried a different ALGO?

But maybe the reason is something else. Could you post a minimal reproducible, so I can analyze where the calculation fails?

[tania_sandoval]

Hi Marie,

I tested different ALGO (Fast, Normal, Damped, Conjugate), and all of them converge to the same energy (up to 1E-10 approx), so I think this is not the problem... The convergence of the first SCF electronic structure is pretty weel behavied.
Also I checked the behavoir of following cycles with the different ALGO, and in all cases the RMM is used. It appears that ALGO and EDIFF do not control the DFPT SCF cycles at all.

I added a simple input/ouput files

DFPT.zip

where a see this problem. During many test I did, some of them with the exact same input testing reproducibility, and others with different ALGO, the resulting Born effective charges, oscilates a little bit (~0.5%) which does not affect in real terms for my porpouse, but I am still aware if I am obtaining the correct result of this calculation.

Regards

[marie-therese.huebsch]

Hi Tania,

just a small update. I can reproduce the behaviour and indeed the convergence is abborted prematurely (before reaching EDIFF) because the rms of the charge density is stagnating. I think this is really reason to be cautious (as you also implied in your question). Let me discuss with my colleagues and get back to you. Sorry for the inconvenience.

[tania_sandoval]

Hi Marie,

Just to check if you have news about this issue?

I tried computing with the exact same input many times, then compare the resulting Born effective charges from different launches, and there is just small changes among them, and also DFPT cycles "converge" to different accuracies. It appears Born effective charges are not too dependent on tight convergence... but I am still concerned.

Regards!

[marie-therese.huebsch]

Hi Tania,

I am so sorry for the late response. There is really no excuse. After consulting with a collegue, we gathered the k-point density may be the reason for the bad convergence.

Could you try to perform a convergence study of the born effective charges with respect to the k-point density?

Best regards and again I am terribly sorry. (Even your reminder went unnoticed because there was a system change on the forum which resulted in an issue that notifications diod not get send via email.)

[tania_sandoval]

Hi Marie, do not worry.

Regarding the kpoint mesh, as I am simulating a single molecule in "vacuum" I used a Gamma Kpoint. But to check it, I launched calculations with Gamma centered 2x2x2 and 3x3x3 kpoints. I observed small changes (higher than the last case with different algorithms, around one order higher), but the resulting charges are practically the same, and if I use them I obtain just small numerical changes. Here are the resulting charges for Kpoints 1x1x1, 2x2x2 and 3x3x3, respectively:

Code: Select all

 BORN EFFECTIVE CHARGES (including local field effects) (in |e|, cummulative output)
 ---------------------------------------------------------------------------------
 ion    1
    1     1.48305    -0.00630     0.00001	    1     1.47897    -0.00633     0.00002	    1     1.48023    -0.00629     0.00001
    2     0.00650     1.48338    -0.00025	    2     0.00622     1.48250    -0.00024	    2     0.00625     1.48247    -0.00024
    3     0.00002    -0.00013     0.80194	    3     0.00002    -0.00013     0.80195	    3     0.00002    -0.00013     0.80196
 ion    2
    1    -0.46712     0.09918    -0.00033	    1    -0.46943     0.09714    -0.00016	    1    -0.46876     0.09772    -0.00022
    2     0.11467    -0.50096     0.00008	    2     0.11387    -0.50424    -0.00006	    2     0.11413    -0.50288    -0.00004
    3    -0.00003    -0.00001    -0.24115	    3    -0.00006     0.00003    -0.24123	    3    -0.00005     0.00001    -0.24120
 ion    3
    1    -0.40046    -0.07480     0.00006	    1    -0.40253    -0.07675     0.00020	    1    -0.40194    -0.07615     0.00015
    2    -0.06009    -0.56830    -0.00019	    2    -0.06230    -0.57016     0.00003	    2    -0.06169    -0.56966     0.00002
    3    -0.00004    -0.00002    -0.24139	    3    -0.00004    -0.00003    -0.24142	    3    -0.00004    -0.00002    -0.24140
 ion    4
    1    -0.58423    -0.04718     0.00026	    1    -0.58167    -0.04342    -0.00011	    1    -0.58243    -0.04457     0.00003
    2    -0.03214    -0.38458     0.00002	    2    -0.02971    -0.38351    -0.00013	    2    -0.03091    -0.38353    -0.00016
    3    -0.00000     0.00001    -0.24130	    3    -0.00005    -0.00001    -0.24137	    3    -0.00004    -0.00001    -0.24134
 ion    5
    1     0.06809    -0.10147     0.00015	    1     0.06644    -0.09905     0.00008	    1     0.06695    -0.09977     0.00010
    2    -0.03966    -0.15100     0.00041	    2    -0.03419    -0.14445     0.00040	    2    -0.03573    -0.14641     0.00041
    3     0.00008     0.00042     0.05269	    3     0.00006     0.00038     0.05282	    3     0.00007     0.00039     0.05277
 ion    6
    1    -0.03411     0.03466     0.08931	    1    -0.03716     0.03523     0.08982	    1    -0.03623     0.03507     0.08966
    2    -0.00609     0.06568    -0.01942	    2    -0.00705     0.06508    -0.01839	    2    -0.00674     0.06524    -0.01861
    3     0.09785    -0.03616    -0.03973	    3     0.09786    -0.03618    -0.03985	    3     0.09786    -0.03617    -0.03982
 ion    7
    1    -0.03494     0.03467    -0.08934	    1    -0.03798     0.03527    -0.08983	    1    -0.03706     0.03510    -0.08968
    2    -0.00641     0.06593     0.01906	    2    -0.00727     0.06536     0.01816	    2    -0.00699     0.06551     0.01835
    3    -0.09793     0.03579    -0.03907	    3    -0.09787     0.03581    -0.03905	    3    -0.09788     0.03581    -0.03906
 ion    8
    1     0.05307     0.05594     0.02721	    1     0.05184     0.05752     0.02867	    1     0.05222     0.05704     0.02822
    2     0.01532    -0.02075    -0.08634	    2     0.01523    -0.02258    -0.08666	    2     0.01531    -0.02206    -0.08652
    3     0.01765    -0.10285    -0.03988	    3     0.01763    -0.10285    -0.03991	    3     0.01764    -0.10285    -0.03990
 ion    9
    1    -0.15716    -0.09106     0.00051	    1    -0.14602    -0.09191     0.00055	    1    -0.14939    -0.09165     0.00054
    2    -0.02821     0.07461     0.00015	    2    -0.02660     0.07307     0.00021	    2    -0.02697     0.07347     0.00019
    3     0.00055     0.00021     0.05278	    3     0.00054     0.00020     0.05289	    3     0.00054     0.00021     0.05284
 ion   10
    1     0.05284     0.05698    -0.02768	    1     0.05148     0.05874    -0.02937	    1     0.05190     0.05821    -0.02886
    2     0.01601    -0.02172     0.08649	    2     0.01571    -0.02333     0.08650	    2     0.01585    -0.02290     0.08647
    3    -0.01815     0.10273    -0.03888	    3    -0.01813     0.10271    -0.03888	    3    -0.01814     0.10272    -0.03888
 ion   11
    1     0.02833    -0.03022    -0.06161	    1     0.02806    -0.02894    -0.06037	    1     0.02813    -0.02933    -0.06074
    2    -0.07068     0.00266    -0.06727	    2    -0.07168     0.00169    -0.06835	    2    -0.07133     0.00192    -0.06807
    3    -0.07991    -0.06706    -0.03936	    3    -0.07991    -0.06700    -0.03934	    3    -0.07991    -0.06701    -0.03934
 ion   12
    1     0.02835    -0.03012     0.06149	    1     0.02796    -0.02902     0.06054	    1     0.02807    -0.02936     0.06083
    2    -0.07068     0.00276     0.06718	    2    -0.07172     0.00160     0.06848	    2    -0.07135     0.00189     0.06813
    3     0.07994     0.06704    -0.03942	    3     0.07996     0.06704    -0.03952	    3     0.07995     0.06704    -0.03949
 ion   13
    1    -0.03572     0.09972    -0.00005	    1    -0.02994     0.09152    -0.00004	    1    -0.03169     0.09399    -0.00005
    2     0.16147    -0.04770     0.00009	    2     0.15950    -0.04103     0.00005	    2     0.16018    -0.04308     0.00007
    3    -0.00003     0.00004     0.05278	    3    -0.00001     0.00000     0.05290	    3    -0.00001     0.00001     0.05286

The convergence is "reached" as OUTCAR writes, but the situation is the same, the criterion is not reached at all. Also, conversely for the differents algorithms test, where all calculation have similar energies for the DFPT cycles (with just different convergence values), the DFPT energies are completely different for the different kpoint mesh, but this does not changes abruptly the Born charges.

IMHO, the kpoints are not the problem here, they affect the calculation, but the non convergent behaviour on DFPT cycles is present in all cases... it appears more related with the code internal procedure into the DFPT cycles and equations. I did not find any way to control these DFPT cycles at all.

Regards

[tania_sandoval]

Some news about this?
I did many test, and the unique results are small variations on born charges... but I would be nice to control and tighten the DFPT cycles for good convergence and accuracy.

Regards

[jonathan_lahnsteiner2]

Dear Tania Sandoval,

I was checking the input files which you supplied. You are using a vacuum molecule in a perfectly cubic box. You could try to impose some small orthorhombic distortions onto your simulation box while keeping the molecule constant. You might modify the lattice constants in your POSCAR in the following way:

Code: Select all

     30.0000000000000000    0.0000000000000000    0.0000000000000000
      0.0000000000000000   30.0000001000000000    0.0000000000000000
      0.0000000000000000    0.0000000000000000   30.0000002000000000

Please note that the your coordinates are in direct coordinates. So just modifying the lattice in the POSCAR file will alter the geometry of your molecule. Going to slightly orthorhombic boxes is a common trick when dealing with convergence issues for vacuum molecules. This will destroy degeneracies of the molecular orbitals and might improve convergence. Additionally you can reduce the EDIFF tag in your INCAR file. By tightening the stopping criterion for the electronic energy also the born effective charges will be relaxed more tightly.
I hope this is of help to you. Otherwise please contact us again.

All the Best Jonathan

[tania_sandoval]

Hi Jonathan,

I tested both. the orthorombic and the tighter criterium. Both gives similar behavoiur and results.
The orthorombic case gives exactly the same results for convergence and born charges until the convergence value for each cycle (non converged in terms of the EDIFF selected in INCAR), and small changes in born charges in the same range as observed for multiple launches with the same input.
In the case of tighter criterium, I changed EDIFF from 10E-8 to 10E-10, the convergence problem maintains, only those cycles with already converges at 10E-8, but now converging at 10E-10.

The problem related with the stenheimer cycles and the non RPA ones persist... the cycles "converge" with accuracies too bigger than the EDIFF value... and any TAG in INCAR can change this behaviour or control it for the moment.

Regards

[jonathan_lahnsteiner2]

Dear Tania Sandoval,

The EDIFF tag in the INCAR file does control the convergence criterion for the electronic minimization, as mentioned in the VASP wiki. It does NOT control the convergence criterion for linear response calculations. So the expectation that linear response will use the same convergence criterion as the electronic ground state minimization is not given. The convergence criterion for the linear response calculations can not be set by the user but is hard coded in the source file src/linear_response.F.

I was suggesting to you to decrease the EDIFF tag in the hope that a more tightly converged electronic ground state will result automatically in more tightly converged linear response calculations. I did tests for EDIFF = 1E-8 1E-9 1E-10 1E-11 1E-12. I was able to observe a small trend that the linear response convergence improves slightly with decreasing EDIFF. But as you point out the values are still larger than the set EDIFF. Which is expected.

If you want to set the convergence criterion differently you have to modify the linear_response.F file of the VASP source code and recompile. You can do this by changing line 1241

Code: Select all

EDIFF = 1E-10

Note the units of the convergence criterion are internal units and not those written into the output files. If you modify the source code please do careful testing before relying that the code works. I hope this helps you resolve your issue otherwise please contact us again.

All the Best Jonathan