nbdbuf: a problem using this variable w/ an insulator?

brehmj · Post by **brehmj** » Tue Jul 10, 2012 10:44 pm

I have an insulator, PbTiO3. I set it up two ways: in a cubic fashion and a tetragonal fashion. I have pseudopotential in which there are 14 val. electrons for Pb, 12 for Ti, and 6 for O. So, in my 5 atom unit cell I have 44 electrons. since this is an insulator, I need 22 bands.
Ok, so far.

So, playing around with the phonon codes from the tutorials,
I run a phonon calc. in 4 ways:

1. With nband 22
2. With nband 24
3. With no nband specified.
4. With nband 24 AND nbdbuf 4.

The first 3 ways give similar results with phonons at gamma in the 10E2 cm-1 range.

But the 4th way gives me results in the 10E3 range -- too high by a factor of 10.

What could be going on?

Thanks,

John B.

jzwanzig · Post by **jzwanzig** » Wed Jul 11, 2012 11:38 am

I think what could be going on is that you are using nbdbuf in a situation where it should remain at its default, namely zero--in the nband 24 case with nbdbuf 4, you are treating 4 bands as "fixed" during the computations but they should not be, 2 of them are the top two valence bands. The documentation for nbdbuf indicates that it is of use in certain situations involving metaillic systems or finite electric fields, which are not your case, and that the defaults are changed automatically for those cases. In your system it should be left at its default, namely zero.

maxim · Post by **maxim** » Fri Sep 14, 2012 2:19 pm

Please help!

I perform a geometry optimization of KNO3 with the purpose of subsequent RF calculation. Could you tell me which option of occopt should I put (it is an insulator)? And should I also specify the nband variable? Or maybe additionally use nbdbuf.
Here is my input:

Code: Select all

#gamma-KNO3

ndtset 8 #REMOVE if it's not necessary
#getwfk -1 #this is not possible since printing wfk file not allowed
prtden 0 #to reduce output size
prtwf 0 #to reduce output size
prteig 0 #to reduce output size

#Parameters
nstep  1000
ecut 60
toldfe1 1.0d-5
toldfe2 1.0d-9
toldfe3 1.0d-10
toldfe4 1.0d-11
toldfe5 1.0d-12
toldfe6 1.0d-13
toldfe7 1.0d-14
toldfe8 1.0d-18

#Definition of occupation numbers
occopt 1

#Variables needed for geometry optim. - this
#will be retained for consistency throughout
dilatmx 1.01
ecutsm  0.5

#chkprim 0 #mozno tak delatj, PC perestaet rugatsa
nsym 1 #break the symmetries
natom 15
ntypat 3
typat 1 1 1 2 2 2 3 3 3 3 3 3 3 3 3
znucl 19 7 8
xred 0.0000000000 0.0000000000 0.0000000000
0.6666666667 0.3333333333 0.3333333333
0.3333333333 0.6666666667 0.6666666667
0.0000000000 0.0000000000 0.4050000000
0.3333333333 0.6666666667 0.0716700000
0.6666666667 0.3333333333 0.7383300000
0.0713500000 0.5356400000 0.1006700000
0.4643600000 0.5356400000 0.1006700000
0.4643600000 0.9286500000 0.1006700000
0.1309800000 0.2620200000 0.4340000000
0.1309800000 0.8690200000 0.4340000000
0.7379800000 0.8690200000 0.4340000000
0.4046600000 0.2023300000 0.7673300000
0.7976700000 0.2023300000 0.7673300000
0.7976700000 0.5953400000 0.7673300000

rprim 1.0000000000E+00 0.0000000000E+00 0.0000000000E+00 #obtained from
-5.0000000000E-01 8.6602540378E-01 0.0000000000E+00 #output of input file
0.0000000000E+00  0.0000000000E+00  1.0000000000E+00 #with angdeg
#scalecart 10.3689272478 10.3689272478 17.3023324004 #see descr. of rprim
acell 10.3689272478 10.3689272478 17.3023324004 #see descr. of rprim

#Kpoint Grid
#prtkpt 1
#kptrlen 40
kptopt 1
nshiftk 1
shiftk 0.0 0.0 0.5
ngkpt 6 6 6

#uncomment this only for paw psp test as jzwanzig said you in forum:
#it will show you if you are getting occupations and or D_ij's that are too 
#big for the code to work well
#pawprtvol -3

diemac 4.0

#Berry phase calculation of the polarization
#berryopt -1
#rfdir 1 1 1 #as jzwanzig said in forum: always use rfdir 1 1 1,
#not e.g. 0 0 1... It will calculate polarization in all three cartesian
#directions. The computational overhead of doing this is negligible
#symmorphi 0

thanks for any help,
Maxim.

jzwanzig · Post by **jzwanzig** » Fri Sep 14, 2012 2:26 pm

I would use occopt 1, so all bands are filled with two electrons; and set nband explicitly to the number of filled bands (so no empty bands in the calculation). Don't set nbdbuf, leave it at its default.

maxim · Post by **maxim** » Fri Sep 14, 2012 6:56 pm

Thank you very much, you saved me a lot of effort!

And in order to understand for the future - empty bands should be turned off to save computer time? And the more conductuing system is, the more empty bands and nbdbuf should be included?

Best regards,
Maxim

maxim · Post by **maxim** » Mon Sep 17, 2012 8:47 am

Thank you very much for all!

I forgot to clarify my question - after the geometry optimization I'm going to investigate finite electric field calculations, polarization, as in tutorial "lesson on polarization and finite electric field". So how I understand from the description of the nbdbuf variable, the value of this variable should be set? And leave occopt 1, and increase nband value above the number of filled bands?

Thank you again

jzwanzig · Post by **jzwanzig** » Tue Sep 18, 2012 6:11 pm

For finite electric fields, nband should be set to exactly the number of filled bands, and nbdbuf should not be set (it will default to zero).

maxim · Post by **maxim** » Thu Sep 20, 2012 1:36 pm

Thank you very much for your time and quick response!

Best regards,
Maxim

ABINIT Discussion Forums

nbdbuf: a problem using this variable w/ an insulator?

nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?

Re: nbdbuf: a problem using this variable w/ an insulator?