Difference in autoparal 1 vs autoparal 0+paral_kgb 0

cennanaradipa · Post by **cennanaradipa** » Tue Apr 25, 2017 6:57 am

Hello everyone,

I'm currently trying to find the EELS and Dielectric function of La2CuO4 (LCO). The input file is below:

Code: Select all

ndtset 3
autoparal 1
#paral_kgb 0

# Structural  Parameters
ntypat     3
znucl      57 29 8
natom      14
typat      1 1 1 1 2 2 3 3 3 3 3 3 3 3
acell      3*3.808 Angstrom
chkprim    0
rprim
  1.0  0.000  0.000
  0.000  1.0  0.000
  0.000  0.000  3.45851
xred 0.500000 0.500000 0.138998
  0.000000 0.000000 0.361002
  0.000000 0.000000 0.638998
  0.500000 0.500000 0.861002
  0.000000 0.000000 0.000000
  0.500000 0.500000 0.500000
  0.500000 0.500000 0.313883
  0.000000 0.000000 0.186117
  0.000000 0.500000 0.000000
  0.500000 0.000000 0.000000
  0.000000 0.000000 0.813883
  0.500000 0.500000 0.686117
  0.500000 0.000000 0.500000
  0.000000 0.500000 0.500000

# Spins on atoms
spinat 0.0  0.0 -1.0
0.0  0.0 -1.0
0.0  0.0 -1.0
0.0  0.0 -1.0
0.0  0.0  1.0
0.0  0.0  1.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
0.0  0.0  0.0
nspinor 1
nsppol  2
nspden  2

# Kpoint grid, XC,  and smearing
ixc    -101130
occopt  3
diemac  13
tsmear  0.005

# Parallelization hacks 
#npfft   1
#npkpt   2
#npband  144

# Convergence parameters
nstep      200
ecut       20
ecutsm     0.5
pawecutdg  50
pawovlp    10

# GGA-PBE + U (1=FLL, 2=AMF)
usepawu    1
lpawu      -1 2 -1
upawu      0.0 8.8 0.0 eV
jpawu      0.0 1.2 0.0 eV

# K-points
ngkpt      8 8 8
nshiftk    1
shiftk     0.5 0.5 0.5

### DATASETS ###
# Ground State 
   iscf1    17
 tolvrs1    1.d-6
 kptopt1    1
#prtcif1    1
 prtden1    1

# DOS 
  iscf2   -3
getden2    1
 nband2    90
prtdos2    2
tolwfr2    1.0d-10

# Band structure 
     iscf3    -2
   kptopt3    -4
   getden3     1
    nband3     90
   ndivsm3     20
   enunit3     1
   tolwfr3     1.0d-10
kptbounds3
  0.0  0.0  0.0 # Gamma 
  0.0  0.5  0.0 # X
  0.5  0.5  0.0 # M
  0.0  0.0  0.5 # Z
  0.0  0.0  0.0 # Gamma

I was using 8.0.8 at first, but I found discrepancies when using autoparal 1 and autoparal 0 with paral_kgb 0. I initially ran into an error when running with autoparal 1, but then realized the kpoints are different from each dataset. I applied paral_kgb 0, and it works without changing the position of the kpoints in the dataset. Then, I moved the kpoints information to apply to all datasets and autoparal 1 works fine.

The things is that the results between autoparal 0 and autoparal 1 is vastly different. Here are thes results for the screening calculation:

autoparal1vs0.pdf: EELS and Dielectric Function; (224.26 KiB) Downloaded 197 times

I varied the ecuteps and ecutwfn in autoparal 0 in previous calculations, but it only changes the intensity of the peaks (i.e. not drastic). When using autoparal 1, I get a huge difference in both imaginary and real part of the dielectric function, with additional peaks appearing in EELS.

Analysing the band structure, I also get this for autoparal 1:

bands_dos_autoparal1.pdf: Autoparal 1; (116.11 KiB) Downloaded 222 times

and this for autoparal 0 + paral_kgb 0:

full_bands_dos_autoparal0kgb0.pdf: Autoparal 0 + paral_kgb 0; (112.92 KiB) Downloaded 213 times

The Fermi energy was different as well (0.22788 Ha for autoparal 1, 0.24696 Ha for autoparal 0). I was wondering:

Is this a bug in ABINIT 8.0.8? I have tried 7.10 and the results are the same. Exact same Fermi energy, but I haven't checked the results on the dielectric and EELS function. I assume it's the same
Does this have to do with the structure/DFT+U incorporated in the calculation? I tried to print the CIF file, and it matches the i4mmm structure of LCO from materials.org
The autoparal 0 and paral_kgb 0 looks a bit more realistic. Should this be the preferred results?

Any help is appreciated. Thank you

Sincerely,
Cenna