Convergence Problem using VASP 5.2

Message

hpaudel · #1 Post by **hpaudel** » Tue Oct 08, 2013 12:18 am

Hello Vasp users,

I recently started using VASP 5.3 and now try to optimize the structure of Bi2Te3 in bulk. It is running very slow. To test I also ran same system in VASP 4.6 taking everything same (also same computational resources). There it was quite fast.
Following is my INCAR file.

SYSTEM = Bi2Te3
ISTART = 0

ICHARG = 2
LREAL=.FALSE.
ISPIN = 1
#LOPTICS = .TRUE.
#LSORBIT = .TRUE.

#LMAXMIX = 4
#SAXIS = 0 0 1
EDIFF = 0.0001
EDIFFG = -0.01
ENCUT = 400
LWAVE = .TRUE.
LCHARG = .TRUE.
NELM = 100
NSW = 100
#NGX=24
#NGY=24
#NGZ=24

#GGA = PE
IBRION = 2
#ISIF = 2
IALGO = 48
NSIM = 4
#ALGO = GW0
POTIM = 0.10
TEIN = 0.0
TEBEG = 0.0
TEEND = 0.0
ISMEAR = -5
SIGMA = 0.2

ISYM = 0

This is the self consistent calculation where I am optimizing my geometry. How can I enhance the speed for the convergence in Vasp 5.3 ?

Thank you,
hpaudel

#2 Post by **admin** » Wed Oct 09, 2013 10:53 am

Did you have a look at old recipes how to improve convergence:
1) if you use PAW potentials: set LMAXMIX = 6 in INCAR to ensure that the
high-l components are mixed properly
2) reduce the mixing parameters until a certain pre-convergence is reached:
I suggest to do a few steps with linear mixing first, keep WAVECAR and
then continue with the default mixing parameters:
a) set
AMIX = 0.2
BMIX = 0.0001
(AMIX_MAG = 0.

(BMIX_MAG = 0.0001)
LWAVE =.True.
for the first steps and,
b) comment the mixing - tags and set
ISTART = 1
in addition to continue from this run. Do not forget to copy WAVECAR to the
working directory in the continuation run.

hpaudel · #3 Post by **hpaudel** » Wed Oct 09, 2013 6:50 pm

I hope this will work for me.
I will update the improvement.
Thank you very much.

hpaudel · #4 Post by **hpaudel** » Thu Nov 07, 2013 12:51 am

Hello
I tried the way you suggested. It stops after few electronic iteration cycles. So the job is not completed. I am stagnated at the moment. The end of the OUTCAR file is posted below.

----------------------------------------- Iteration 1( 26) ---------------------------------------

POTLOK: cpu time 0.75: real time 0.37
SETDIJ: cpu time 5.34: real time 3.38

The wall-time already reached. It could not complete even in few days in supercomputer cluster with 12 nodes, ppn:8, peme:6000 M.
How to circumvent this problem of convergence in GW calculation ?

Thank you.

<span class='smallblacktext'>[ Edited Thu Nov 07 2013, 12:53AM ]</span>