My Community

Hi!

I am completely new to VASP (for bulk systems), but I am quite experienced with electronic structure theory for molecules. I am doing some initial calculations, and I find that even for doing no-frills geometry optimizations (using input files from https://nomad-lab.eu/) VASP seems to have a hard time reaching convergence of the electronic wave function. I am using VASP on a cluster, so I know that this isn't a problem with my installation.

The slowness of the convergence, and also the fact that it is non monotonic would usually lead me to believe that there is something severally wrong with my model geometry. Like that the wave function is trying to optimize the wrong state, or that there are some eigenstates very close in energy that it is oscillating between. But I feel like, apart from the VASP tutorial cases, none of my calculations seem well behaved.

What I am wondering is whether this type of behaviour is considered normal. Does one need to tweak a lot of the parameters in order for VASP to do "the right thing" for a geometry optimization?

Example of a calculation for cubic C15 CoPr2:

N E dE d eps ncg rms rms(c)
DAV: 1 0.618220140789E+03 0.61822E+03 -0.23357E+04 6216 0.272E+03
DAV: 2 0.775624071653E+02 -0.54066E+03 -0.50469E+03 6248 0.414E+02
DAV: 3 -0.314976855759E+02 -0.10906E+03 -0.97902E+02 8424 0.217E+02
DAV: 4 -0.460702657768E+02 -0.14573E+02 -0.13692E+02 10616 0.813E+01
DAV: 5 -0.480818858669E+02 -0.20116E+01 -0.18883E+01 10680 0.276E+01 0.901E+01
RMM: 6 -0.687546525490E+02 -0.20673E+02 -0.24321E+02 7175 0.135E+02 0.654E+01
RMM: 7 -0.134938353445E+03 -0.66184E+02 -0.11654E+02 6533 0.109E+02 0.163E+02
RMM: 8 -0.435177749389E+02 0.91421E+02 -0.26473E+01 6107 0.839E+01 0.403E+01
RMM: 9 -0.424219184708E+02 0.10959E+01 -0.16681E+01 7335 0.335E+01 0.245E+01
RMM: 10 -0.399280768384E+02 0.24938E+01 -0.92649E+00 7275 0.243E+01 0.183E+01
RMM: 11 -0.394154165125E+02 0.51266E+00 -0.36171E+00 8739 0.744E+00 0.131E+01
RMM: 12 -0.394140802087E+02 0.13363E-02 -0.14277E+00 9504 0.385E+00 0.159E+01
RMM: 13 -0.400473899597E+02 -0.63331E+00 -0.53736E-01 9118 0.413E+00 0.197E+01
RMM: 14 -0.401098143817E+02 -0.62424E-01 -0.36434E-01 9357 0.203E+00 0.182E+01
RMM: 15 -0.403352142550E+02 -0.22540E+00 -0.17021E-01 8895 0.240E+00 0.189E+01
RMM: 16 -0.404303429690E+02 -0.95129E-01 -0.18267E-01 8541 0.268E+00 0.159E+01
RMM: 17 -0.403331678236E+02 0.97175E-01 -0.19439E-01 8610 0.284E+00 0.882E+00
RMM: 18 -0.403690037677E+02 -0.35836E-01 -0.98865E-02 9125 0.145E+00 0.916E+00
RMM: 19 -0.402971768026E+02 0.71827E-01 -0.91740E-02 8688 0.223E+00 0.361E+00
RMM: 20 -0.402432297441E+02 0.53947E-01 -0.24244E-01 8819 0.331E+00 0.312E+00
RMM: 21 -0.402630458033E+02 -0.19816E-01 -0.32972E-02 8764 0.107E+00 0.236E+00
RMM: 22 -0.402921110010E+02 -0.29065E-01 -0.45542E-02 8423 0.117E+00 0.109E+00
RMM: 23 -0.402965510256E+02 -0.44400E-02 -0.17602E-02 8954 0.555E-01 0.722E-01
RMM: 24 -0.402988453172E+02 -0.22943E-02 -0.78829E-03 9046 0.432E-01 0.619E-01
RMM: 25 -0.403002312138E+02 -0.13859E-02 -0.69930E-03 8758 0.410E-01 0.253E-01
RMM: 26 -0.403003217295E+02 -0.90516E-04 -0.71943E-03 8895 0.346E-01 0.332E-01
RMM: 27 -0.403008285260E+02 -0.50680E-03 -0.31358E-03 9218 0.232E-01 0.506E-01
RMM: 28 -0.403015442299E+02 -0.71570E-03 -0.39811E-03 8797 0.214E-01 0.419E-01
RMM: 29 -0.403018559206E+02 -0.31169E-03 -0.30422E-03 8879 0.201E-01 0.211E-01
WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 8
RMM: 30 -0.403020656323E+02 -0.20971E-03 -0.20024E-03 8917 0.146E-01 0.172E-01
WARNING in EDDRMM: call to ZHEGV failed, returncode = 8 4 8
RMM: 31 -0.403022394274E+02 -0.17380E-03 -0.18506E-03 8391 0.134E-01 0.111E-01
RMM: 32 -0.403023811686E+02 -0.14174E-03 -0.15222E-03 8468 0.123E-01 0.268E-02
RMM: 33 -0.403025197292E+02 -0.13856E-03 -0.13924E-03 7877 0.109E-01 0.426E-02
WARNING in EDDRMM: call to ZHEGV failed, returncode = 8 4 6
RMM: 34 -0.403026261181E+02 -0.10639E-03 -0.11037E-03 7356 0.981E-02 0.342E-02
RMM: 35 -0.403027296768E+02 -0.10356E-03 -0.10667E-03 7267 0.921E-02 0.329E-02
WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 7
WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 9
RMM: 36 -0.403028098105E+02 -0.80134E-04 -0.84136E-04 7090 0.869E-02 0.986E-03
RMM: 37 -0.403028920250E+02 -0.82215E-04 -0.84820E-04 6983 0.810E-02 0.134E-02
RMM: 38 -0.403029502073E+02 -0.58182E-04 -0.60691E-04 6792 0.761E-02 0.157E-02
RMM: 39 -0.403030152825E+02 -0.65075E-04 -0.67334E-04 6799 0.711E-02 0.179E-02
RMM: 40 -0.403030586062E+02 -0.43324E-04 -0.44881E-04 6637 0.665E-02 0.870E-03
RMM: 41 -0.403031095304E+02 -0.50924E-04 -0.52447E-04 6497 0.613E-02 0.675E-03
RMM: 42 -0.403031413269E+02 -0.31797E-04 -0.33244E-04 6347 0.580E-02 0.663E-03
RMM: 43 -0.403031804621E+02 -0.39135E-04 -0.40372E-04 6306 0.550E-02 0.724E-03
RMM: 44 -0.403032040638E+02 -0.23602E-04 -0.24644E-04 6202 0.506E-02 0.611E-03
RMM: 45 -0.403032341159E+02 -0.30052E-04 -0.31048E-04 6096 0.471E-02 0.731E-03
RMM: 46 -0.403032516086E+02 -0.17493E-04 -0.18341E-04 5973 0.448E-02 0.670E-03
RMM: 47 -0.403032744093E+02 -0.22801E-04 -0.23649E-04 5943 0.420E-02 0.910E-03
RMM: 48 -0.403032876587E+02 -0.13249E-04 -0.13902E-04 5810 0.394E-02 0.876E-03
RMM: 49 -0.403033052046E+02 -0.17546E-04 -0.18220E-04 5746 0.373E-02 0.111E-02
RMM: 50 -0.403033155988E+02 -0.10394E-04 -0.10939E-04 5635 0.354E-02 0.121E-02
RMM: 51 -0.403033286242E+02 -0.13025E-04 -0.13486E-04 5589 0.340E-02 0.110E-02
RMM: 52 -0.403033365659E+02 -0.79417E-05 -0.83449E-05 5483 0.316E-02 0.112E-02
WARNING in EDDRMM: call to ZHEGV failed, returncode = 6 3 6
RMM: 53 -0.403033464679E+02 -0.99020E-05 -0.10271E-04 5391 0.298E-02 0.119E-02

Hi,

Could you please provide a zip file of the input (INCAR, KPOINTS, POSCAR) and OUTCAR files, as mentioned at https://www.vasp.at/forum/viewtopic.php?t=17928?

Absolutely! I've added the input files, the OUTCAR and the CONTCAR.

As the ionization steps progress, it takes fewer and fewer steps to reach convergence. The initial slow convergence is probably due to a poor starting geometry?
Another thing: The final geometry I arrived at is not the same as the one I found on nomad-lab. This is somewhat unsettling.

Is there any common procedure for finding a starting geometry close enough to the ground state geometry before initiating a geometry optimization? I know that Avogadro is often used for this purpose when it comes to molecules.

In your calculation, the number of iterations for the SCF convergence is not unusually large, even at the initial geometry. At the initial geometry, it is the largest probably because the calculation started without WAVECAR file present in the directory (see ISTART). Overall, your calculation seems to have run properly without any particular problem.

How was the geometry obtained on nomad-lab obtained? With the same code and setting?

Suggestions of programs that could provide a starting geometry would be CHGNet and MACE (https://wiki.fysik.dtu.dk/ase/ase/calcu ... ators.html).

The SCF convergence can sometimes be very difficult or even impossible to achieve with the default setting. This can be particularly the case for calculations with non-collinear magnetism. Using ALGO=Conjugate may help, but this can still be slow.