SCF convergence issue for a nanoparticle

Queries about input and output files, running specific calculations, etc.


Moderators: Global Moderator, Moderator

Post Reply
Message
Author
asif_ali28
Newbie
Newbie
Posts: 9
Joined: Tue Apr 05, 2022 12:18 pm

SCF convergence issue for a nanoparticle

#1 Post by asif_ali28 » Thu Sep 26, 2024 7:10 am

Hello everyone,

I am performing a SCF calculation for a nanoparticle of size 9x10x11Ang^3 (72 atoms) and 5 Ang vacuum in all three directions. Even after long SCF run the convergence does not reach below 0.1E0. I have tried increasing NBANDS and ENCUT, but could not reach the convergence even after few hundreds SCF cycles. I would be grateful for your kind help and suggestions.

INCAR:

Code: Select all

ISTART =  1            
ISPIN  =  1       
LREAL  = .FALSE.      
ENCUT  =  320      
LWAVE  = .TRUE.      
LCHARG = .TRUE.      
KPAR   = 1           
NCORE = 6
ISMEAR =  0          
SIGMA  =  0.05         
NELM   =  200       
EDIFF  =  1E-05 

K-point: 1x1x1

Thankyou
Regards,
Asif Ali


martin.schlipf
Global Moderator
Global Moderator
Posts: 542
Joined: Fri Nov 08, 2019 7:18 am

Re: SCF convergence issue for a nanoparticle

#2 Post by martin.schlipf » Thu Sep 26, 2024 2:16 pm

There are a couple of things you can try:

  • use a different electronic optimizer (ALGO), sometimes CG methods can proceed where Davidson fails

  • try to play around with the mixing parameters in particular (BMIX, this in particular relevant if you see that a optimization at constant charge (ICHARG=12) works

  • you mentioned that you considered a different ENCUT, please check that it aligns with the ENMAX value(s) in your POTCAR

  • visualize the structure or look at the neighbor list in the OUTCAR file to make sure the structure is set up correctly

  • it is also a good idea to try a simpler system (less vacuum, smaller nanoparticle) which should converge simpler and may help you to learn techniques to improve convergence at a smaller computational cost

If none of this advice helps, please get back to me.

Martin Schlipf
VASP developer


Post Reply