NPAR error in DOS calculation

Message

ccccc · #1 Post by **ccccc** » Mon Nov 11, 2013 7:06 pm

Hi,
I am trying to perform DFT+U calculations on an ionic compound.

I run the calculations on 8 nodes and for DOS calculation, when i dont specify NPAR tag the result is 1.8eV indirect band gap. But with exactly the same input files, when i use NPAR=2 the result is 1.3eV direct band gap.

Do you know the reason of it?
Also there is a big difference between results. Which result is reliable?

Many thanks

#2 Post by **admin** » Tue Nov 12, 2013 11:50 am

NPAR should not change the results, please check for other differences in the two runs.

ccccc · #3 Post by **ccccc** » Tue Nov 12, 2013 12:44 pm

Dear Admin

I checked the input files and results twice. The input files are completely same. The only difference is the NPAR=2 tag in one of them. Although starting with same geometry and same convergence parameters, the final geometries and charge densities are different in the relaxation. So DOS calculations give different results.

alex · #4 Post by **alex** » Wed Nov 13, 2013 5:42 pm

NPAR settings affect the number of _empty_ bands, if I'm not terribly wrong. Maybe here is your trouble.

Cheers,

alex

ccccc · #5 Post by **ccccc** » Fri Nov 15, 2013 2:09 pm

Dear Alex
As far as I know, NPAR is used for parallelisation to improve the performance of the job. It should not change the results. So do you have any advice to me?

alex · #6 Post by **alex** » Fri Nov 15, 2013 4:37 pm

I agree, it should not change a thing. But well, you might guess that it might be more complicated in the real world. Maybe your NBANDS are simply not sufficient?

Try typing

grep NBANDS OUTCAR

for both NPAR settings and tell us the result. What is the number of your last occupied band?

Cheers,

alex

ccccc · #7 Post by **ccccc** » Fri Nov 15, 2013 8:00 pm

I checked the results. The number of last occupied bands in both calculations are same.

PS: Now i submitted a new calculation with increased number of bands and will see what happens.

ccccc · #8 Post by **ccccc** » Mon Nov 18, 2013 11:52 am

Nothing has changed!

Yue-Wen Fang · #9 Post by **Yue-Wen Fang** » Thu Mar 13, 2014 12:28 pm

Based on my experience, I think NPAR isn't a purely computational setting.
For some calculations involving particular materials or systems, NPAR gives us very different physical quantities as you said. It may also causes some convergence problems in some cases.