Page 1 of 1

Memory usage by VASP-5.4 DFPT calculations

Posted: Sat Apr 08, 2017 6:18 am
by deepashri
Dear All,

I am performing DFPT calculations with VASP-5.4.1 for a monolayer with vacuum in z-direction. The POSCAR contains 108 atoms with 788 electrons. Following are the KPOINTS and INCAR files,

%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Monkhorst-Pack
0
Monkhorst-Pack
10 10 1
0 0 0
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
ISTART = 0
INIWAV = 1
NWRITE = 2
LWAVE = .TRUE.
PREC = accurate
EDIFF = 1.0E-6
EDIFFG = -0.001
ISMEAR = 1
SIGMA = 0.001
ICHARG = 2
LCHARG = .TRUE.
ISYM = 1
NELMIN = 4
NELM = 100
NELMDL = 0

IBRION = 8
ISIF = 3
NFREE = 2
NSW = 1
POTIM = 0.02
IALGO = 38
LDIAG = .TRUE.
LPETIM = .FALSE.
LREAL = .FALSE.
LORBIT = 1
# NPAR = 1
# NSIM = 16
LPLANE = .TRUE.
LRPA = .FALSE.
LELF = .FALSE.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

The calculations demand enormous amount of memory. They use entire memory on every node, even if the memory per node is 256GB. Is this job supposed to use so much memory? How to handle this issue? Could we somehow restrict the memory usage through the input or during VASP installation?

Thanks in advance.
Best,
Deepashri.

Re: Memory usage by VASP-5.4 DFPT calculations

Posted: Mon Apr 10, 2017 11:55 am
by admin
In contrast to local basis sets, which increase with increasing number of atoms,
the basis set of planewaves fills the space. I.e. Basis set increases with
increasing cell volume. Because of this, in 2D calculations, which use the vacuum layer,
the cell volume can be prohibitively large.
Check the number of planewaves and the cell volume.

Re: Memory usage by VASP-5.4 DFPT calculations

Posted: Wed Apr 12, 2017 12:24 pm
by deepashri
Thanks for the reply.
As you correctly mentioned, the number of plane waves is too large. Following are the required outputs,

total plane-waves NPLWV = ******
volume of cell : 5093.29

The vacuum the I have used is the minimum possible. Since, these are phonon calculations, the accuracy is also high. Is there any other solution for this?

Re: Memory usage by VASP-5.4 DFPT calculations

Posted: Thu Apr 13, 2017 9:34 am
by alex
Hi deepashri,

I could see two ways to reduce memory:
a) reduce you Monkhorst mesh. Even for metals it might go down to sth. like 2x2x1. Consider your big cell. You have covered probably all interactions already using the gamma-point only. Check!
b) use cheaper PPs (or PAWs) to reduce cut-off energy. (I can't see it in the INCAR, so I can't judge if it is low enough already.)

Cheers,

alex