My Community

Hello,

I am running into a persistent problem with reading the WAVEDER file for a GW0 calculation. I write out a WAVEDER file from a self-consistent calculation in the first step and then try to run a GW0 calculation, which crashes with an error message saying that the WAVEDER file cannot be found even though the file is clearly there -- I just generated it in the previous step! My INCARs and KPOINTS files are below. Any suggestions are most appreciated!

INCAR for step 1:

ISTART = 0
ENCUT = 500
ISMEAR = 0
SIGMA = 0.05
NPAR = 4
NSIM = 4
LPLANE=.TRUE.
LORBIT = 11
NBANDS = 32
NGX = 118
NGY = 20
NGZ = 86
LOPTICS = .TRUE.

INCAR for step 2:

ENCUT = 500
ISMEAR = 0
SIGMA = 0.05
LORBIT = 11
NBANDS = 32
NGX = 118
NGY = 20
NGZ = 86
LOPTICS = .TRUE.
ALGO = GW0
NOMEGA = 48
NELM = 4
ENCUTGW = 150

KPOINTS for both steps:

Automatic mesh
0
Gamma
1 35 1
0 0 0

hmm, I have never met tis problem. please include a ls -l in your job script, just to check in the job.stdout file that WAVEDER is present in the directory in which the second job runs, and that it has non-zero filesize.

Thanks, I got it to run. But now the calculation runs for a while and produces a new error:

calculate QP shifts <psi_nk| G(iteration)W_0 |psi_nk>: iteration 1
calculate exact exchange contribution
allocating 2 responsefunctions rank= 3264
allocating 3 responsefunctions rank= 3264
Doing 2 frequencies in blocks of 3
NQ= 1 0.0000 0.0000 0.0000,
error in XI_LOCAL_FIELD: ZGETRF returns 1

I have gotten around Lapack related errors in the past (Vasp 4.6) by simply recompiling mpi.F with lower optimization. Does the same solution work for Vasp 5.2 or is there a problem elsewhere?

I have not had this problem but we compiled using the Intel mkl libraries. Have you checked to make sure that your memory is sufficient b/c these tmp files are huge

I'm also using the Intel MKL libraries (no scalapack though). I have 2GB available per core... is this generally sufficient in your experience?

Also, head admin, is there any way to continue a GW job from the tmp files if it terminates midway due to walltime limits?

Yes that should be completely sufficient, the way I have done the calculations is to store all of my vasp files in a /home2 directory and then wrote a small run script to access the files from that directory, it seems to work for me.

Do you face this problem with all k-point samplings? For example, what about if you give a smaller sampling, i.e., 1x1x1? The same?

Best,
Giacomo

I mean, in my opinion it is a problem related to the symmetry of your cell. what about it? can you also add the POSCAR, please?
Very best,
Giacomo

Hi Giacomo,

Here's my POSCAR:

C O
1.00000000000000
16.4362176406276355 0.0000000000000000 0.0000000000000000
0.0000000000000000 2.5199321854421943 0.0000000000000000
0.0000000000000000 0.0000000000000000 11.9857473261969947
C O
4 2
Selective dynamics
Direct
0.3052150904082325 0.0000000000000000 0.5000000000000000 T T T
0.3532231167432727 0.5000000000000000 0.5000000000000000 T T T
0.4364447799356910 0.5000000000000000 0.5000000000000000 T T T
0.4844528062707383 0.0000000000000000 0.5000000000000000 T T T
0.2303011338429855 0.0000000000000000 0.5000000000000000 T T T
0.5593667628359853 0.0000000000000000 0.5000000000000000 T T T

I have not tried other k-point samplings. Maybe I need to try that next (and possibly set ISYM=0 as well)

best,
-Ashwin.

Is there a symmetry reduced representation? What is your kpoint sampling? Have you tried with removing the Selective dynamics tag? Pardon my ignorance but what are the TTT at the end of the file? I never put those but have seen others use this (I am assuming true or false) in their poscar. Thanks.

Selective dynamics is for relaxing the atoms and the Ts (or Fs) allow (forbid) relaxation for the relevant atomic degrees of freedom. This is irrelevant for the moment since ISIF and NSW are not set in the INCAR.

OK, I will know this for surface relaxation for vacancy or defect simulation in the future, thank you. Have you had any more luck? I have only run the GWo routine for small systems, I am trying it on a large system now and am running into paging issues, so I am going to try running it on a larger machine. Anyway keep us informed and hopefully things are working out, they will eventually, don't give up!