Optimization of small Ru4 cluster with N2 yields nonsensical structure

[julien_steffen]

Dear VASP community,

I'm currently attempting to calculate the structure of adsorbed N2 on a small Ru4 cluster.
Optimizing the structure of Ru4 was no problem, however adding N2 somewhat closely and performing a new optimization results
in the all atoms, both the cluster and N2, breaking apart.
I also noticed that after the first SCF cycle, the following cycles don't converge within the given 200 steps.
For the Ru4 cluster, VASP gave me a magnetization of 4, which I use as the magnetization of the four Ru atoms in the new INCAR.

I'm using vasp.6.4.1.
In the INCAR, I added the AMIX and BMIX options because I previously had some SCF convergence issues which I could fix using them.
I also added the OUTCAR and vasp.out in the attached zip.

INCAR:

Code: Select all

System=Ru4
PREC=Normal
GGA=PE
NCORE=26
ISPIN=2
MAGMOM=4*1 2*0
NELM=200
NELMIN=8
ENCUT=450
EDIFF=1e-7
LREAL=.FALSE.
ALGO=fast
ISMEAR=0
SIGMA=0.04
IVDW=12
LWAVE=.False.
LCHARG=.false.
ISYM=-1
IBRION=1
ISIF=2
EDIFFG=-0.01
NSW=2000
POTIM=0.5
NFREE=15
ICHARG=2
AMIX=0.1
BMIX=0.01

POSCAR:

Code: Select all

Converted_POSCAR
1.0
25.0 0.0 0.0
0.0 25.0 0.0
0.0 0.0 25.0
Ru N
4 2
Cartesian
22.164740000000002 27.04158 24.79714
24.705489999999998 26.20118 24.53167
24.038800000000002 28.21321 25.40812
23.3315 25.78318 26.31942
23.90916 26.96917 27.12981
24.15339 27.96538 26.735300000000002

CONTCAR:

Code: Select all

Converted_POSCAR
   1.00000000000000     
    25.0000000000000000    0.0000000000000000    0.0000000000000000
     0.0000000000000000   25.0000000000000000    0.0000000000000000
     0.0000000000000000    0.0000000000000000   25.0000000000000000
   Ru   N
     4     2
Direct
  0.8276274718819474 -0.0011647557682060  0.9941303277647526
  0.9799753243528740 -0.0711104976542719  0.9791456438929194
  1.0015156241781316  0.3122549846799526 -0.1403885798963147
  0.9028610149955586 -0.0378973221813510 -0.0129534276977092
  0.9668207811990601  0.0528704596467234  0.2009450572723029
  1.0133229833926694  0.2319951312771190  0.1759793786632861

Best regards

[henrique_miranda]

If you are running an ionic relaxation and you observe that some of the electronic minimization do not converge then that is the first thing you need to fix.
Here are a few things you can try:
1. You are using ISMEAR=0 but during the ionic minimization, you get some partially filled states (look for 'occupation' in the OUTCAR file). Try changing to ISMEAR=-1 or 1
2. Change from ALGO=fast to ALGO=Normal (default)

Normally the SCF should converge in less than 60 steps (normally less than 30 steps).
If it does not then you should review your settings.

We are collecting this type of general recommendations about the electronic minimization in this page:
https://www.vasp.at/wiki/index.php/Trou ... onvergence

[julien_steffen]

Thank you for the pointers, but they weren't able to solve my issue.

I had a look at the POSCAR again and after visualization noticed that the N2 molecule
was fairly close to the Ru cluster (1.3 angstrom, I think), so I moved it away a bit.

Rerunning the calculation with the N2 further away has solved my issue.
Most likely the close proximity of the N2 caused some issues with my input magnetization,
resulting in a large force being applied to the cluster after the first relaxation step, breaking it apart and messing up the rest of the optimization.

Best regards