Effect of unit cell size for convergence in atomic calculation

Message

itomasa · #1 Post by **itomasa** » Mon Sep 12, 2005 11:49 pm

Dear everyone,

I have problem in atomic calculation.
For Ni atom, I calculated with following input files
and it was successful for 10 A cubic unit cell.

However for the case of 15, 20 and 25 A,
I obtained possitive values in free energy.
Furthermore at non-self-consistent calculation,
the order of degeneracy is different between the cases of
20 and 25 A and the cases of 10 and 15 A.

If you know the reason or solution,
please answer.

Sincerely
itomasa

=INCAR for - non-self-consistent calculation====

#Ionic part
IBRION = 2 !CG
POTIM = 0.5 !delta in CG
NSW = 0 !Ionic step
EDIFFG = -0.004 !If all forces are smaller than |EDIFFG|,
!Ionic step will be finished.
#Electronic part
ISYM = 0
#LCHARG = .FALSE.
LWAVE = .TRUE.
PREC = Med
#ISTART = 0 !for new calculation
ISTART = 1 !for continued calculation
#IALGO = 8 !CG for small system calculation
IALGO = 48 !RMMDIIS for large system caluculation
#LREAL = .TRUE. !For rough caluculation
LREAL = .FALSE. !For final caluculation
#ISPIN = 1 !spin unpolarized calculation
ISPIN = 2 !spin polarized calculation
#NUPDOWN = 2
#EMIN = -13.0
#EMAX = 0.0
#NEDOS = 1300 !width of grids in energy is 0.01 eV
#RWIGS = 1.286
#SIGMA = 0.02 !For final calculation
SIGMA = 0.1
#ISMEAR = 0
ISMEAR = -2
#LDIAG = .FALSE.
ICHARG = 12
EDIFF = 0.0001
GGA = 91
#VOSKOWN = 1
NPAR=1
MAGMOM=2
#AMIX = 0.2
#BMIX =0.0001
#FERWE = 5*1 1*1 4*0
#FERDO = 5*0.8 1*0 4*0
===========================

===INCAR for self consistent calculation===========

#Ionic part
IBRION = 2 !CG
POTIM = 0.5 !delta in CG
NSW = 0 !Ionic step
EDIFFG = -0.004 !If all forces are smaller than |EDIFFG|,
!Ionic step will be finished.
#Electronic part
ISYM = 0
#LCHARG = .FALSE.
LWAVE = .TRUE.
PREC = Med
#ISTART = 0 !for new calculation
ISTART = 1 !for continued calculation
#IALGO = 8 !CG for small system calculation
IALGO = 48 !RMMDIIS for large system caluculation
#LREAL = .TRUE. !For rough caluculation
LREAL = .FALSE. !For final caluculation
#ISPIN = 1 !spin unpolarized calculation
ISPIN = 2 !spin polarized calculation
#NUPDOWN = 2
#EMIN = -13.0
#EMAX = 0.0
#NEDOS = 1300 !width of grids in energy is 0.01 eV
#RWIGS = 1.286
#SIGMA = 0.02 !For final calculation
SIGMA = 0.1
#ISMEAR = 0
ISMEAR = -2
#LDIAG = .FALSE.
#ICHARG = 12
EDIFF = 0.0001
GGA = 91
#VOSKOWN = 1
NPAR=1
MAGMOM=2
AMIX = 0.2
BMIX =0.0001
FERWE = 5*1 1*1 4*0
FERDO = 5*0.8 1*0 4*0
=============================

====POSCAR===================
POSCAR
20.0000000000000000
1.0000000000000000 0.0000000000000000 0.0000000000000000
0.0000000000000000 1.0000000000000000 0.0000000000000000
0.0000000000000000 0.0000000000000000 1.0000000000000000
1
Direct
0.5000000000000000 0.5000000000000000 0.5000000000000000
===================================

==KPOINTS===========================
atom
0
Monkhorst-Pack
1 1 1
0.0 0.0 0.0
===================================

POTCAR-Ni-GGA-USPP

Veronika · #2 Post by **Veronika** » Tue Sep 13, 2005 6:35 am

Set the atom in the origin (0 0 0 in the POSCAR in direct coordinates) and also try PREC = HIGH and set EDIFF to a smaller value (eg. 0.000001).

tjf · #3 Post by **tjf** » Tue Sep 13, 2005 3:13 pm

Ergh! What nastiness occurs that makes the location og the atom in the box important?

Veronika · #4 Post by **Veronika** » Wed Sep 14, 2005 6:29 am

Usually, if you have just one k-point, you set it on the same position as the atom's ...
btw, did ilt work?

itomasa · #5 Post by **itomasa** » Fri Sep 16, 2005 9:18 pm

Thank you for your kind advice. => Veronika

I calculated with changing only PREC, EDIFF and
position of atom as you wrote.
But the calculation results are as below...

=self consistent calc. in 15 A cubic unit cell====
Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 0.36950667
Ewald energy TEWEN = -136.18826670
-1/2 Hartree DENC = -739.78499421
-V(xc)+E(xc) XCENC = -253.11116352
PAW double counting = 0.00000000 0.00000000
entropy T*S EENTRO = 0.00000000
eigenvalues EBANDS = -75.50587709
atomic energy EATOM = 1204.51786711
---------------------------------------------------
free energy TOTEN = 0.29707225 eV

energy without entropy = 0.29707225 energy(sigma->0) = 0.297072
25

E-fermi : -5.5546 XC(G=0): -0.1912 alpha+bet : -0.0356
add alpha+bet to get absolut eigen values

spin component 1

k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -9.9134 1.00000
2 -8.9654 1.00000
3 -8.5079 1.00000
4 -8.2144 1.00000
5 -8.1813 1.00000
6 -5.5727 1.00000
7 -1.2693 0.00000
8 -1.1884 0.00000
9 -1.0605 0.00000
10 -0.0053 0.00000

spin component 2

k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -7.4693 0.80000
2 -6.9109 0.80000
3 -6.6658 0.80000
4 -6.5195 0.80000
5 -5.1230 0.80000
6 -1.0100 0.00000
7 -0.9823 0.00000
8 -0.9088 0.00000
9 0.0284 0.00000
10 0.6663 0.00000
==================================

What's wrong ?

itomasa · #6 Post by **itomasa** » Fri Sep 16, 2005 9:26 pm

Sorry, maybe enegy is not still converged...
I will report again.

itomasa · #7 Post by **itomasa** » Sat Sep 17, 2005 10:41 pm

I show the result after 104 electronic step
for self consistent calculation for Ni in 15 A
cubic unit cell.

==================================
Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 0.36950667
Ewald energy TEWEN = -136.18826670
-1/2 Hartree DENC = -739.68303769
-V(xc)+E(xc) XCENC = -253.26871169
PAW double counting = 0.00000000 0.00000000
entropy T*S EENTRO = 0.00000000
eigenvalues EBANDS = -75.63592018
atomic energy EATOM = 1204.51786711
---------------------------------------------------
free energy TOTEN = 0.11143751 eV

energy without entropy = 0.11143751 energy(sigma->0) = 0.111437
51

E-fermi : -5.5833 XC(G=0): -0.2453 alpha+bet : -0.0356
add alpha+bet to get absolut eigen values

spin component 1

k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -9.8774 1.00000
2 -8.9747 1.00000
3 -8.3692 1.00000
4 -8.3639 1.00000
5 -8.2323 1.00000
6 -5.6026 1.00000
7 -1.2941 0.00000
8 -1.2922 0.00000
9 -1.1285 0.00000
10 -0.0741 0.00000

spin component 2

k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -7.4974 0.80000
2 -6.8033 0.80000
3 -6.7990 0.80000
4 -6.5386 0.80000
5 -5.1314 0.80000
6 -1.0508 0.00000
7 -1.0507 0.00000
8 -0.9710 0.00000
9 -0.0632 0.00000
10 0.6389 0.00000
===================================

Free Energy = 0.111437 eV

is well converged within 0.000001 eV error.
Positive value...

Occupancy should be changed ? ...

<span class='smallblacktext'>[ Edited Sun Sep 18 2005, 12:42AM ]</span>

Veronika · #8 Post by **Veronika** » Thu Sep 22, 2005 7:50 am

Well, after rereading your INCAR file, I saw that you specified ISYM = 0 and the FERWE and FERDO tag. I am not familiar with the latter two, but I guess, that your problems stem from this settings? The occupance of the levels do not seem to be very reasonable to me (only 0.8 for all bands of spin 2??)... Try to do a most simple calculation omitting all tags just to compare and for a start.

#9 Post by **admin** » Mon Sep 26, 2005 3:46 pm

1) GGA potentials break the symmetry, therefore the cell shape
should be slightly different from cubic (a \= b \= c)
2) both, the settings
-- ISPIN =2 only
-- ISPIN =2, FERWE=6*1.0; FERDO = 4*1, LDIAG=F, ISMEAR = -2
yield negative total energies then,
setting the occupancies manually increases the convergence
considerably