GW calculation

[Newlife1982]

Dear all,

I am calculating the band gap of the crystal TiO2. There is a error as follows:

screening : taking advantage of time-reversal symmetry
-P-0000 Maximum band index for partially occupied states nbvw = 24
-P-0000 Remaining bands to be divided among processors nbcw = 76
-P-0000 Number of bands treated by each node ~ 9

screening.F90:381:ERROR
Only sequential version or band parallelism are compatible with time-reversal trick or spectral method
-P-0000
-P-0000 leave_new : decision taken to exit ...

Could you give me some suggestions? Thank you in advance.

Wang

[david.waroquiers]

Hello,

Try to set istwfk variable to 1 for each k-point (so that you dont use the time reversal properties). The error message is quite clear : the time reversal is not yet coded for k-point parallelism. I guess you used gwpara 1. You can also try gwpara 2 (parallelisation over bands).

So try :
gwpara 2
and/or
istwfk *1

David

[Newlife1982]

Dear David

Thank you very much.

Best regards,

Wang

[Newlife1982]

Hello,

I used gwpara 2. Unfortunately, it still does not work.

screening : taking advantage of time-reversal symmetry
-P-0000 Maximum band index for partially occupied states nbvw = 189
-P-0000 Remaining bands to be divided among processors nbcw = 11
-P-0000 Number of bands treated by each node ~ 2

loop over bands done in parallel (assuming time reversal!)
-P-0000 treating 3 bands from 190 up to 192 by node 0
-P-0000 Memory required for wfg= 64.5 [Mb]
-P-0000 init_Wfs: Wavefunctions in real space are stored in memory.
-P-0000 Memory required for wfr= 1345.5 [Mb]
-P-0000 Memory required for wfg= 4064.3 [Mb]
-P-0000 init_Wfs: Wavefunctions in real space are stored in memory.
-P-0000 Memory required for wfr= -13536.4 [Mb]
2 total processes killed (some possibly by mpirun during cleanup)
sending incremental file list
tgw2_2.outA
tgw2t_LOG_0001
tgw2t_LOG_0002
tgw2t_LOG_0003
tgw2t_STATUS
tgw2t_STATUS_P-0001
tgw2t_STATUS_P-0002
tgw2t_STATUS_P-0003


The calculation stops here. Anyone has some ideas about this problem? Thank you in advance for replying.

Wang

[david.waroquiers]

Hello,

It might be that you need more memory :
-P-0000 Memory required for wfr= 1345.5 [Mb]
-P-0000 Memory required for wfg= 4064.3 [Mb]
The wavefunctions need about 1.3 Gb in real space and 4 Gb in reciprocal space.
How much memory do you have on your machines ?

David

[Newlife1982]

Hello,

Thank you for your replying.

For my machines, 8 cores with Mem = 16080M totally. Usually, I use 4 cores. I think that it should be fine. What is your opinion?

Regards,

Wang

[bruneval]

By the way, the number of empty bands you have is extremely low. Please perform a convergence study with respect to nband.
bye.

Fabien

[Newlife1982]

Thank you very much.

Best Regards,

Wang