Recommendations to treat 4f electrons (valence) of Yb3+ in magnetic systems in VASP

Question

I need help with: I would like to know what is the most common approach to treat valence 4f electrons of lanthanides (3+ charge) in VASP that circumvents the calculation problems shown below. (or more recommendations)

System: I am having trouble with SCF calculations of a system that contains $\ce{2Yb^{3+}}$ as dopants when including 4f electrons of Yb.

Setup: I am using the parameters below in the INCAR file:

NWRITE = 2          
ISTART = 0  
ICHARG = 2    
LORBIT = 11
LPLANE = .FALSE. 
NPAR = 4   
PREC  = Accurate  (Precision level)  
LREAL = AUTO  (Projection operators: automatic) 
ALGO  = N         (Elect. algorithm: 38/48)  
ENCUT = 500 eV

!GGA + U correction for Yb ion 
LDAU = .TRUE. 
LDAUTYPE = 2  
LDAUL = -1 -1 -1 3 
LDAUU = 0 0 0 7 
EDIFF  = 1.0e-05  
EDIFFG = -1.0e-01      
NBLOCK = 1       
IBRION = -1 
NSW    = 0  
ISYM   = 0    
ISMEAR = 0   
SIGMA  = 0.01  
POTIM  = 0.90 
LWAVE = .TRUE. 
LCHARG = .TRUE.
!Magnetisation   
ISPIN  = 2  
LMAXMIX = 6 ! FM ORDER 
MAGMOM = 27*0 24*0 81*0 1.0 1.0 
NUPDOWN = 2

!Mixer  
AMIX     = 0.2 
BMIX     = 0.00001 
AMIX_MAG = 0.8 
BMIX_MAG = 0.00001

Also, I have varied several of them ENCUT, AMIX and BMIX as well as the U correction from 3 to 7. Have also tried different electronic algorithms to no avail.

This is a 3x3x3 supercell so I am testing runs only on the Gamma point and also with automatic Gamma 2 2 2 grid. I have tried HSE0 functional as well and the same problem persists.

Errors: Using ALGO=N I can get this error:

Error EDDDAV: Call to ZHEGV failed. Returncode =   7 1   8
and some other trials the energies keep decreasing indefinitely to really big negative numbers: 
free energy    TOTEN  =      -131.40843111 eV 
free energy    TOTEN  =      -120.45157042 eV  
free energy    TOTEN  =      -103.28925863 eV  
free energy    TOTEN  =    -12481.39444222 eV 
free energy    TOTEN  = -67096650.34128196 eV 
free energy    TOTEN  = ****************** eV 
free energy    TOTEN  = ****************** eV 

Workarounds?: I have found papers where a "\Delta SCF" approach is used to move one 4f electron to 5d orbital and reach desired orbital occupancy ($\ce{Yb^{3+}}$) but to know which band's occupation one needs to modify, a previously converged solution (probably without spin polarization: ISPIN=1) is needed which is not possible in my case due to charge unbalance.

Other resources: Similar issues have been found in other systems with lanthanides and specifically with $\ce{Yb^{3+}}$: See post here. and also here.

Note: When turning off spin polarization and using Yb_3 (valence : 5p6 6s2 5d1) PAW pseudopotential with frozen f electrons the calculations converge just fine.

Answer 1

There is a discussion on this limitation on VASP page. Maybe the approach of a frozen f valence PAW pseudopotential could help you. Although magnetic properties might not be as well represented. I'm quantum espresso user and when their PAW potentials show problems (not so rare) I go to a norm-conserving solution like pseudodojo library, if you can do this in VASP i think it may help. Also about the charge treatment, in QuantumESPRESSO in many cases I just implement a total charge on the supercell instead of specific atom as I think you're doing. If you have a dopant scf will usually bring charge localization towards the dopant in the end without need to explicitly set it in the atom, it may help convergence too.