Convergence issue with U ramping

[william_watson]

I am applying U-value ramping to a plutonium mononitride structure between U=0 and U=5eV in intervals of 0.5eV. I am doing structure relaxations for both ferromagnetic and 1-k antiferromagnetic configurations of this structure. I achieve convergence for all U values except 3.5, 4 and 5eV - I have attached my files for 3.5eV (PuN AFM). I have checked my structures and started calcs with non spin-polarised CHGCAR and WAVECAR but with no success. I have now started to change the mixing parameters but I wanted to check firstly whether I need to do this and if so, whether I am taking the right steps. In terms of optimising the parameters, should I:

• 1. Keep AMIX_MAG and BMIX_MAG values as they are?
  2. Calculate optimal AMIX based on equation AMIX_optimal=AMIX_current*Γ_mean (which comes out to be ~0.3 for this calculation)?
  3. Slowly increase BMIX given that Γ_mean<1?

But please let me know whether I am prematurely changing mixing parameters and have missed something else in my input parameters.
Thanks!

PuN-Uramp.zip

[jonathan_lahnsteiner2]

Dear william_watson,

I looked at your calculation. I was doing a structure/cell shape optimization. The input structure that you supply is far from equilibrium. Your input lattice in the POSCAR is

4.9807167846658933 -0.0000023955861988 0.0000189991261648
-0.0000023953156486 4.9808243240385668 -0.0000732554913398
0.0000190083560749 -0.0000732685890912 4.9807117720123957

And after doing the cell optimization with the following INCAR tags ( which are your tags without magnetism and spin orbit coupling):

Code: Select all

GGA    = PE      
ISYM = 0
PREGGA    = PE      
ISYM = 0
PREC = ACCURATE
ENCUT = 500     
ISMEAR = 0      
SIGMA = 0.05    
ALGO = Normal   
LREAL = .FALSE. 
ISIF = 3 
IBRION = 2        
NSW = 200         
NELM = 100        
NELMIN  =  6      
NBANDS = 512      
EDIFF = 1.0E-6
LDAU = .TRUE.      
LDAUTYPE = 2       
LDAUL = 3 -1       
: no on site terms 
LDAUU = 3.50 0.00  
LDAUJ = 0.00 0.00  
LDAUPRINT = 1     

I get the following lattice:

8.0246444590301582 0.4824507745305144 0.0827355941910482
0.4635716200390327 7.7999531177771342 -1.5833379371509162
0.1038220930390076 -1.5755822672247557 6.2819356456196278

This calculation converged without problems.
Since your structure is so far from equilibrium the optimization does not converge in your case. So I would recommend, first do a cheaper calculation to relax your structure to equilibrium positions and lattice. And then switch on spin orbit coupling and magnetism and further relax your structure. I hope this helps. Otherwise don't hesitate to contact us again.

All the best Jonathan

[william_watson]

The input lattice in the POSCAR is actually the result (CONTCAR) of an initial run without magnetism and SOC. The original POSCAR that I used in the non-magnetic calculation is as follows:

Code: Select all

Plutonium(III) nitride
1.0
        4.9050002098         0.0000000000         0.0000000000
        0.0000000000         4.9050002098         0.0000000000
        0.0000000000         0.0000000000         4.9050002098
   Pu    N
    4    4
Direct
     0.000000000         0.000000000         0.000000000
     0.500000000         0.500000000         0.000000000
     0.000000000         0.500000000         0.500000000
     0.500000000         0.000000000         0.500000000
     0.500000000         0.500000000         0.500000000
     0.500000000         0.000000000         0.000000000
     0.000000000         0.500000000         0.000000000
     0.000000000         0.000000000         0.500000000

Given this, would you recommend a constant volume relaxation (ISIF=2) then followed by another full relaxation to achieve better convergence; and then follow on with calculations including magnetism and SOC?

[jonathan_lahnsteiner2]

Dear william_watson,

Sorry for my previous post the lattice which I was reporting was wrong. I tried to do the calculation with
a single KPOINT to have a faster test. Nevertheless, the problem you are reporting hints to unlucky settings
in the electronic minimization.

I tried your computation again and I am getting the WARNING: Sub-Space-Matrix is not hermitian in DAV already without spin-orbit coupling and without switching on magnetism. I would not focus on changing mixing parameters. I would try to do the optimization with a smaller POTIM. You are using the default POTIM which is 0.5. Setting a smaller value for POTIM might
resolve your issue. Another thing you can try is to use a different IBRION=1,2 or 3 for structure minimization.

Another thing I recognized is that you are setting ISTART = 1 , ICHARG = 1 in your INCAR file. As you told me you did
the structure optimization first without magnetism and without spin-orbit coupling. When switching from a spin un-polarized calculation
to a spin polarized calculation you should either set ISTART=0 or you have to remove the WAVECAR file. Otherwise you might run into
problems during electronic minimization.

Setting a different ALGO tag might also resolve your issue.

What might also be of help for you is to check other posts which are related to the problem you are running into.
https://www.vasp.at/forum/viewtopic.php?t=5894
https://www.vasp.at/forum/viewtopic.php?t=1192


If you need further help. I would like to ask you to supply me with the input files of your structure optimization without magnetism and spin orbit coupling. And another set of input files showing only the tags you were using for the spin-orbit coupling and spin-polarized case. Otherwise it is very difficult for me to disentangle the single steps of your workflow.


If you need further advise don't hesitate to contact us again.

All the best Jonathan