GW0 and GW with vasp5.2.2

[Vahid Askarpour]

Dear VASP Users,

I have been attempting to reproduce the GW0 and GW corrected Si bandgaps reported in PRB75, 235102, 2007. I have run into two errors:

1. GW0 run
I generate WAVEDER using these inputs

INCAR

System = Si
ISMEAR = 0 ; SIGMA = 0.05
PREC=Accurate
ENCUT=300
NBANDS = 100
LREAL=.FALSE.
LOPTICS=.TRUE.
NPAR=16

KPOINTS (gamma-centered)

K-Points
0
Monkhorst Pack
4 4 4
0.5 0.5 0.5

Then use the WAVEDER, the same KPOINTS and the following INCAR

System = Si
ISMEAR = 0 ; SIGMA = 0.05
PREC=Accurate
ENCUT=300
NBANDS = 100
LREAL=.FALSE.
LSPECTRAL=.TRUE.
ALGO = GW0 ; NOMEGA = 56
NELM=4
NPAR=16

But I get this error after the first iteration:

calculate QP shifts <psi_nk| G(iteration)W_0 |psi_nk>: iteration 2
calculate exact exchange contribution
allocating 4 responsefunctions rank= 288
allocating 56 responsefunctions rank= 288
Doing 4 frequencies in blocks of 56
NQ= 1 0.0000 0.0000 0.0000,
error in XI_LOCAL_FIELD: ZGETRF returns 2

2. GW run
I use the same WAVEDER as the first part and then use the following INCAR

System = Si
ISMEAR = 0 ; SIGMA = 0.05
PREC=Accurate
ENCUT=300
NBANDS = 100
LREAL=.FALSE.
ALGO = GW ; NOMEGA = 56
NELM=4
LWAVE=.TRUE.
NPAR=16

But I get the following error:

calculate QP shifts <psi_nk| G(iteration)W_0 |psi_nk>: iteration 1
calculate exact exchange contribution
allocating 4 responsefunctions rank= 288
allocating 56 responsefunctions rank= 288
Doing 4 frequencies in blocks of 56
NQ= 1 0.0000 0.0000 0.0000,
internal error in CDER_BETWEEN_STATES_ROTATED: optical matrix elements not calculated
1 41 36
internal error in CDER_BETWEEN_STATES_ROTATED: optical matrix elements not calculated
1 60 33


Any suggestions will be appreciated.

Thanks,

Vahid

[sankh]

Hi Vahid,

Although I quite a new user of VASP, I did several GW (especially G0W0) calculations into VASP without any crash. So, I could amend a few suggestions to try out if you have got time to play around a bit. Here are my suggestions:

1. Increase NEDOS to about 3000-3500.
2. Set NPAR carefully according to your comput cluster; Since I am using LINUX IBM clusters, I am using NPAR = No. Nodes with NSIM = 4.
3. Try the spectral routine: LSPECTRAL = T
4. You might try to set higher LMAXFOCKAE
5. If you are using VASP 2.5.8 (<2.8.11), try ENCUTFOCK = 0, but if 2.8.11, try with PRECFOCK = F/N.
6. Finally, try to set ENCUTGW less than 300ev (say 200 ev), if 1-5 did not improve.

Please let me the outcomes here if you have tried these options.

Cheers,
Sankh

[Vahid Askarpour]

Hi Sankh,

Thanks for your suggestions. I tried them but still got the same errors. Then I removed the two lines that specify PREC and ENCUT from my INCAR files. THE G0W0, GW0, and GW runs finished successfully with no errors.

Thanks again,

Vahid