Problems with ALGO = CHI step

[nicholas_dimakis1]

Hello,

I am trying to calculate the optical properties for Y2O2S. The POSCAR file only contains 10 atoms. The structure is optimized and then I run ALGO = Exact, which runs without any issues. However, I cannot get an output for the ALGO = CHI step.

I use VASP 6.2.1 under lodestar 6 supercomputer using 10 nodes and 1280 CPUs. Even by increasing to 20 nodes and 2560 CPUs I still get the same errors on the ALGO = CHI step.

login1.ls6(4391)$ more KPOINTS
KPT-Resolved Value to Generate K-Mesh: 0.060
0
G
4 4 4
0.0 0.0 0.0

For ALGO = Exact step the INCAR is

login1.ls6(4394)$ more INCAR
System = Y2SO2
ISTART=1
ICHARG =1
ISMEAR = 0;
EDIFF = 0.1E-08
EDIFFG = -0.0001
ADDGRID = TRUE
LMAXMIX = 6
LOPTICS = .TRUE.
CSHIFT = 0.100
NBANDS = 320
NEDOS = 2000
#
#ISYM = 0
IVDW = 11
NPAR = 10
ALGO= Exact
PREC=Accurate
NELM = 1
ENCUT = 525
NWRITE=4
#LWAVE = .FALSE.
#LCHARG = .FALSE.
ISPIN = 2

For For ALGO = CHI step the INCAR is

System = Y2SO2
ISTART=1
ICHARG =1
ISMEAR = 0;
EDIFF = 0.1E-08
EDIFFG = -0.0001
ADDGRID = TRUE
LMAXMIX = 6
#LOPTICS = .TRUE.
#CSHIFT = 0.100
NBANDS = 320
#NEDOS = 2000
#
#ISYM = 0
IVDW = 11
#NPAR = 10
ALGO= CHI
PREC=Accurate
#NELM = 1
ENCUT = 525
NWRITE=4
LWAVE = .FALSE.
LCHARG = .FALSE.
ISPIN = 2

The OUTCAR files and the POSCAR are attached.

The error I get is that I need to remove NPAR. However, if I do this VASP crashes.

Thank you,

Nick

optical_properties.zip

[ferenc_karsai]

So I ran your calculation on 64 cores and it takes of course long but it finished without any problems.
Your problem is you ran on too many cores. Please run the calculation on a more reasonable number of cores (not larger than your NBANDS).

Since you set 1280 cores the number of bands are changed:

| The number of bands has been changed from the values supplied in |
| the INCAR file. This is a result of running the parallel version. |
| The orbitals not found in the WAVECAR file will be initialized with |
| random numbers, which is usually adequate. For correlated |
| calculations, however, you should redo the groundstate calculation. |
| I found NBANDS = 320. Now, NBANDS = 1280. |

With these number of bands the calculation takes forever. If you double the number of cores it would make the number of bands also double. It needs at least one band per core.

Also you forgot to set NBANDS in your ALGO=Exact step. This way random wave functions are used for many bands.

Please always check if NBANDS in your OUTCAR file is the same in all of your calculation steps. It can be automatically changed if NBANDS cannot be evenly divided by NCORE/NPAR and the number of cores used (so best is don't set NCORE/NPAR in the previous steps too).