I’ve been trying for days to converge the symmetry broken ground state of a single isolated Nickel atom to no avail. I’ve managed to converge Cr, Fe, and Cu but for some reason Ni doesn’t want to converge to the ground state. The closest I’ve been able to get is with the INCAR below, but the occupation levels are NOT 0 and 1. All other changes I've made seem to make the situation worse. The following website helped me get this far
http://cms.mpi.univie.ac.at/vasp/vasp/Determining_groundstate_energ_atoms.html
I believe the problem is that bands 4 and 5 in the spin 2 component are so close in energy that the electron gets distributed 50/50 between them. Any advice on how to solve this problem would be must appreciated.
For my purposes I need the ground state energy at both 500 eV cut off and 800 eV cut off. I am using PAW_PBE Ni 02Aug2007 pseudo-potentials.
INCAR
SYSTEM = Ni Atom
ENCUT = 800
EAUG = 600
NCORE = 4
#KPAR = 8
ISPIN = 2
ISYM = 0
MAGMOM = 2
LMAXMIX = 4
NELMDL = -5
EDIFF = 1E-6
ICHARG = 1
#### DOS ####
NEDOS = 3001 # DEFAULT IS 301
ISMEAR = 0
SIGMA = 0.0005
#SIGMA = 0.0001 # WARNING: DENMP: can't reach specified precision
#SIGMA = 0.001
AMIX = 0.134
#AMIX = 0.0956
#AMIX = 0.25
BMIX = 0.0001
#NELM = 100
ALGO = A
VOSKOWN = 1 # disregarded by PBE but why not
LSUBROT = .FALSE.
NELM = 500
TIME = 0.05
#TIME = 0.1
#TIME = 0.2
Occupation Numbers
spin component 1
k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -6.0108 1.00000
2 -5.8331 1.00000
3 -5.3853 1.00000
4 -5.3719 1.00000
5 -5.2918 1.00000
6 -4.9831 1.00000
7 -1.0926 0.00000
8 -0.9899 0.00000
9 -0.9816 0.00000
10 -0.0856 0.00000
11 0.2160 0.00000
12 0.3188 0.00000
13 0.3195 0.00000
14 0.3529 0.00000
15 0.3549 0.00000
16 0.3898 0.00000
17 0.5899 0.00000
18 0.6172 0.00000
19 0.6180 0.00000
20 0.6433 0.00000
21 0.6556 0.00000
22 0.6610 0.00000
23 0.6654 0.00000
24 0.7045 0.00000
spin component 2
k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -4.5370 1.00000
2 -4.5261 1.00000
3 -4.5085 1.00000
4 -4.0966 0.53488
5 -4.1000 0.46512
6 -3.9704 0.00000
7 -0.4068 0.00000
8 -0.4008 0.00000
9 -0.2585 0.00000
10 -0.0064 0.00000
11 0.3031 0.00000
12 0.3441 0.00000
13 0.3640 0.00000
14 0.3778 0.00000
15 0.3861 0.00000
16 0.4209 0.00000
17 0.6335 0.00000
18 0.6543 0.00000
19 0.6657 0.00000
20 0.6803 0.00000
21 0.6855 0.00000
22 0.6879 0.00000
23 0.7001 0.00000
24 0.7177 0.00000
You can also try to set (and fix) the occupation numbers by hand (brute force method if you know what should come out). Try for example
ISMEAR = -2 ! this reads and fixes occupations from FERWE/FERDO
FERWE = 6*1. 18*0. ! upspin occupations
FERDO = 4*1. 20*0. ! downspin occupations
you can try to use NUPDOWN to fix the difference of spin up and down, this is not as drastic as the suggestion above, and if it leads to a correct ground state than you can relax if later once you have the correct WAVECAR.
also, i did not understand what is your k-mesh? is it just the single k-point? what is the unit cell size? one k-point can sometimes be a bit tricky for calculations, and you are also using a very small sigma which is needed for this calculation, but usually requires denser k-meshes. I'm not sure what the procedure is for just the one atom, but its something to consider.
I am facing the exact same problem right now. I already tried using ISMEAR -2, but the energy keeps jumping around.
Did you find any solution jet?
So first off I want to thank Jürgen Furthmüller and Ofer Neufeld for their input because it was very helpful in solving this problem. The reason I didn't comment sooner is because I simply became too frustrated to deal with this problem and went to work on other problems. However, this problem is back on my plate and I because of you two I have managed to at least solve the 800 eV convergence issue.
Ni Atom 500 threw an error and gave garbage results when I reduced the smearing
"WARNING: DENTET: can't reach specified precision."
Looking online it seems that the DOS was not accurate enough and lead to discrepancies in the total number of electrons. I wasn't sure if this might have been why we couldn't resolve the ground state so I resubmitted both the 500 eV and 800 eV jobs with higher
NEDOS = 3001 instead of the standard 301
This seems to have fixed the occupancy issues for the 800 eV Ni Atom calculation. The NUPDOWN = 2, ENCUT = 800 job converged quickly and nicely to 0 or 1 occupation states with the standard charge mixing algorithm.
The 500 eV job however wouldn't converge electronically so I switched to using ISMEAR = -2
I am still waiting on the fixed smearing 500 eV Ni Atom job, but I will report back when it finishes.
Below is the INCAR I used for the 800 eV calculation.
SYSTEM = Ni Atom
ENCUT = 800
EAUG = 600
NCORE = 4
#KPAR = 8
ISPIN = 2
ISYM = 0
NUPDOWN = 2 # I AM FORCING A SPIN 2 STATE
MAGMOM = 2
LMAXMIX = 4
#### DOS ####
NEDOS = 3001 # DEFAULT IS 301
#EMIN = -7
#ENMAX =0
NELMDL = -5 # Makes sure the initial charge density is well converged before updating it.
ISMEAR = 0
SIGMA = 0.001
AMIX = 0.233
#AMIX = 0.275
#AMIX = 0.25
BMIX = 0.0001
NELM = 500
Thanks for you input Erik.
Was wondering how your input works so smooth and just takes less than 30 SCF steps. It was related to the EDIFF tag which for you is the default: 1E-4 eV, while I wanted to converge it to 1E-5 eV.
It looks like ENCUT is very important. The following converges relatively (when compared to ENCUT=450eV) "smoothly" for ENCUT=800 and takes around 160 steps.
INCAR created by Atomic Simulation Environment
ENCUT = 800.000000 #(replace with 450 and it converged for me in around 450 steps - where the energies oscillated for a while)
BMIX = 0.000100
AMIX = 0.200000
SIGMA = 0.002000
EDIFF = 1.00e-05
PREC = Normal
ALGO = 38
ISYM = 0
ISPIN = 2
ISMEAR = 0
ISTART = 0
ICHARG = 1
NELM = 500
NSW = 0
NPAR = 2
IVDW = 12 #I needed this for my purposes
LCHARG = .TRUE.
LWAVE = .TRUE.
NUPDOWN=2
I also encountered the partial occupancy issue of single Ni atom at the PBE level with ENCUT=500. Jürgen Furthmüller 's suggestion (manually set and fix occupancies using ISMEAR=-2) turns out to work well. But the calculated total energy is higher than that associated with the partial occupancy configuration. Including the Hubbard U term can fix the issue at ISMEAR=0, but it gives a positive total energy.
Fictitious partial occupancies may arise due to the choice of which orbitals are treated as core in the pseudopotential. You could try to experiment with the Ni_pv pseudopotential. You could also try including the aspherical contributions (LASPH = .TRUE., relevant for 3d-4f and some magnetic systems) and using a more dense augmentation grid with PREC = Accurate (to eliminate wrap-around errors).
Also, some convergence testing regarding the box size would be necessary, in order to ensure that periodic images don't interact with each other (if it hasn't already been done).
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts