Page 1 of 1

Identical inputs, different results at different cpu cluster

PostPosted: Wed May 09, 2018 12:07 pm
by smattsson
Hello everyone

I am currently calculating the bulk structure of CoF3 using PBE+U according to the Dudarev formulation. The cell is a rhombohedral one with two units of CoF3 in it. I have optimized the structure at my computer cluster #1 and wanted to do a frequency calculation at another cluster #2 for reasons of convenience. But when I use the optimized structure as the POSCAR and run a single-point SCF calculation at cluster #2, the E0 is higher: -.33626837E+02 at #2 vs. the reference calculation at #1: -.35146116E+02 - i.e. a difference of a whole 1.5 eV. By looking at the DOS, the #2 calculation falsely describes the system as a conductor. #1 gives a band gap of 2.8 eV.

I tested the two single-point calculations using identical input files INCAR, KPOINTS, POSCAR and POTCAR (by copy-pasting them). The only difference is the computer cluster they are running on and the core/node distribution. Does anyone have an idea what might be going wrong here? I am thankful for any replies

Best regards,
Stefan

Cpu details:
#1: aprun protocol, 8 nodes with 8 cores each
#2: mpirun protocol, 2 nodes with 8 cores each

Both systems run vasp 5.4.1 (vasp_std executable, i.e. non-collinear)


INCAR:
Code: Select all
SYSTEM = CoF3 PBE
NCORE = 8

ISTART = 0
LWAVE = .FALSE.

ISMEAR = 0 ! -5=linear tetrahedron (semiconductor/insulator), 0=gaussian
SIGMA = 0.10 ! temperature

PREC = Medium

!NSW = 100 ! ionic steps
IBRION = -1 ! 1=use quasi-newton optimization 2=conjugate gradient, -1=no optimization
ISIF = 3 ! 3=relax ions, cell shape and cell volume
ENMAX = 850 ! use together with ISIF
!ISYM = -1
POTIM = 0.25

EDIFF = 1.0E-5
EDIFFG = -0.01

ENCUT = 700
ICHARG = 2 ! 1=restart, 2=from atom solutions, 11=don't change density

ISPIN =  2 ! 2=spin polarized calculation
LORBIT = 11
MAGMOM = 4 -4 0 0 0 0 0 0

LAECHG = .TRUE.


LDAU = .TRUE.
LDAUTYPE = 2 ! 2=Dudarev
LDAUL = 2 -1 ! angular quantum number of orbitals 1=p, 2=d, 3=f
LDAUU = 6.3 0.0
LDAUJ = 1.0 0.0
LDAUPRINT = 2

LMAXMIX = 4 ! use with Hubbard


KPOINTS:
Code: Select all
K-Points
 0
Monkhorst Pack
 8 8 8
 0  0  0


POSCAR:
Code: Select all
CoF3                                   
   1.00000000000000     
     2.9515852336660924    0.0000000000000000    4.4095306744468550
    -1.4757926168330462    2.5561477937949211    4.4095306744468550
    -1.4757926168330462   -2.5561477937949211    4.4095306744468550
   Co   F
     2     6
Direct
  0.0000000000000000  0.0000000000000000  0.0000000000000000
  0.5000000000000000  0.5000000000000000  0.5000000000000000
  0.8486837912520073  0.6513162087479927  0.2500000000000000
  0.2500000000000000  0.8486837912520073  0.6513162087479927
  0.6513162087479927  0.2500000000000000  0.8486837912520073
  0.1513162087479927  0.3486837912520073  0.7500000000000000
  0.7500000000000000  0.1513162087479927  0.3486837912520073
  0.3486837912520073  0.7500000000000000  0.1513162087479927
 
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00
  0.00000000E+00  0.00000000E+00  0.00000000E+00


I spare you the POTCAR but it is constructed from the standard PBE PAW's

Re: Identical inputs, different results at different cpu clu

PostPosted: Fri Aug 31, 2018 10:34 am
by admin
sorry, but we cannot deal with this issue based on the information you give
we need a FULL error report of both runs (including the OUTCAR,OSZICAR,job.out and job.err files)