No convergence in big supercell calculations

Ubiq · Post by **Ubiq** » Fri Aug 26, 2011 12:51 pm

Hi everyone,

I have convergence problem in GS calculations of fcc Al with 32, 64 atoms in basis. It is OK, when I set 1 or 4 atoms in basis and it takes about 10 iteration for good convergence. But when the number of atoms is increased (32, 64), there is no convergence even for nstep=300. The only difference between the inputs for small and big number of atoms is that I put some very small distortion in coordinates of atoms for the big number case to prevent symfind Error. Nevertheless, I have the same problem with liquid phase, so I think this parameter don't relate with the problem. Could you please help to solve this problem?

My input for 32 atoms in basis:

Code: Select all

# Crystalline aluminum

#Definition of occupation numbers
nband 80
occopt 4
tsmear 0.05

#Definition of the k-point grid
ngkpt  2 2 2
nshiftk 4
shiftk  0.5 0.5 0.5
         0.5 0.0 0.0
         0.0 0.5 0.0
         0.0 0.0 0.5


#Definition of the unit cell
chkprim 0
acell 3*7.5056           # Lattice parameters of bulk aluminum
rprim  2.0  0.0  0.0    # The lattice vector along the z direction
         0.0  2.0  0.0     # is doubled with respect to previous run.
         0.0  0.0  2.0     #

#Definition of the atom types
ntypat 1          # There is only one type of atom
znucl 13          # The keyword "znucl" refers to the atomic number of the
                     # possible type(s) of atom. The pseudopotential(s)
                     # mentioned in the "files" file must correspond
                     # to the type(s) of atom. Here, the only type is Aluminum


#Definition of the atoms
natom 32          #
typat 32*1        # These atoms are of type 1, that is, Aluminum
xred              # This keyword indicate that the location of the atom
   0.000000000000000   0.000000000000000   0.000000000000000
   0.500000000000000   0.000000000000000   0.000000000000000
   0.000000000000000   0.500000000000000   0.000000000000000
   0.500000000000000   0.500000000000000   0.000000000000000
   0.000000000000000   0.000000000000000   0.500000000000000
   0.500000000000000   0.000000000000000   0.500000000000000
   0.000000000000000   0.500000000000000   0.500000000000000
   0.500000000000000   0.500000000000000   0.500000000000000
   0.250000000000000   0.250000000000000   0.000000000000000
   0.750000000000000   0.250000000000000   0.000000000000000
   0.250000000000000   0.750000000000000   0.000000000000000
   0.750000000000000   0.750000000000000   0.000000000000000
   0.250000000000000   0.250000000000000   0.500000000000000
   0.750000000000000   0.250000000000000   0.500000000000000
   0.250000000000000   0.750000000000000   0.500000000000000
   0.750000000000000   0.750000000000000   0.500000000000000
   0.000000000000000   0.250000000000000   0.250000000000000
   0.500000000000000   0.250000000000000   0.250000000000000
   0.000000000000000   0.750000000000000   0.250000000000000
   0.500000000000000   0.750000000000000   0.250000000000000
   0.000000000000000   0.250000000000000   0.750000000000000
   0.500000000000000   0.250000000000000   0.750000000000000
   0.000000000000000   0.750000000000000   0.750000000000000
   0.500000000000000   0.750000000000000   0.750000000000000
   0.250000000000000   0.000000000000000   0.250000000000000
   0.750000000000000   0.000000000000000   0.250000000000000
   0.250000000000000   0.500000000000000   0.250000000000000
   0.750000000000000   0.500000000000000   0.250000000000000
   0.250000000000000   0.000000000000000   0.750000000000000
   0.750000000000000   0.000000000000000   0.750000000000000
   0.250000000000000   0.500000000000000   0.750000000000000
   0.750000000000000   0.500000000000000   0.750000000000000
   0.250000000000000   0.500000000000000   0.750000000000000
   0.750100000000000   0.500000000000000   0.750000000000000 #distortion
#The relaxation
#ionmov 3
#tolmxf 5.0d-4
#ntime 10

#Exchange-correlation functional
ixc 1             # LDA Teter Pade parametrization

#Definition of the planewave basis set
ecut  6.0         # Maximal kinetic energy cut-off, in Hartree

#Definition of the SCF procedure
nstep 20          # Maximal number of SCF cycles
toldfe 1.0d-8

I use intel 12.0 and openmpi-1.4.
Thanks in advance!

Peter

blackburn · Post by **blackburn** » Fri Aug 26, 2011 6:32 pm

Hi,

Your ecut value is very low. Your calculations will probably not be accurate. If increasing ecut to a 'reasonable' value doesn't solve your problem, you can set nline to 8 (or higher if it doesn't work).

FYI, you can use chkprim 0 to bypass the symfind error.

Simon

Ubiq · Post by **Ubiq** » Mon Aug 29, 2011 3:24 pm

Thanks a lot for your answer Simon, but unfornutaly after I have tried all your suggestion (increasing ecut and nline), the result remains the same. For the same input parameters the system with small basis converges very well and on the contrary there is no convergence for systems with big basis. I will be highly appreciated for further suggestions. This problem embarrasses me very much.

By the way, chkprim can't help avoid symfind error, maybe I haven't used that option in a proper way?

Boris · Post by **Boris** » Thu Sep 01, 2011 7:51 pm

Hi

Check if you have enough bands to accomodate all electrons (though I think 80 is way enough). Also try with a slightly larger kpoint mesh. Sometimes if the kpoint mesh is not fine enough, the convergence is harder, especially for metals.

You can also try to first optimize atomic positions only and leave the cell volume constant.

Finally you can try with a different value of occopt. Try a smearing of Methfessel and Paxton (occopt=6) and see if that changes anything

Good luck

Boris

ABINIT Discussion Forums

No convergence in big supercell calculations

No convergence in big supercell calculations

Re: No convergence in big supercell calculations

Re: No convergence in big supercell calculations

Re: No convergence in big supercell calculations