Convergence with PREC=High

Message

tholme · #1 Post by **tholme** » Wed Jan 25, 2012 4:39 am

I'm having trouble obtaining SCF convergence with PREC=High in a number of systems, even if I get good convergence with PREC=Accurate. I've tried IALGO=38 and 48, and ALGO=Normal, there is no significant difference between those choices in terms of convergence. I've set AMIX so that my GAMMA values are near 1, I've also tried setting BMIX to near zero, this doesn't help.

My KPOINT meshes are modest, like 2x2x2 or 3x3x1.

I have noticed the worst convergence for larger systems with a defect.

Here is a sample INCAR:

Code: Select all

#Electronic Static Optimization---------------
PREC = High
NELM = 100
NELMIN =6
EDIFF=1E-6
AMIX=0.2
#BMIX=0.1
ISMEAR=-5
SIGMA=0.2
IALGO=38
LREAL=A
NSIM=4
ISMEAR?=?-5

#Ionic Relaxtion------------------------------
NSW= 0 #0 ionic relaxation steps
IBRION = -1


LCHARG=.FALSE.


NPAR=4

Any ideas? Thanks very much for your help.

<span class='smallblacktext'>[ Edited ]</span>

tholme · #2 Post by **tholme** » Sat Feb 18, 2012 6:08 pm

Nobody has this problem?

tholme · #3 Post by **tholme** » Mon Feb 27, 2012 6:55 pm

For anyone else who may have noticed this problem, I seem to have found a solution. The solution is changing the kpoint integration from ISMEAR = -5 (tetrahedron method) to ISMEAR = 0 (Gaussian method). My convergence (for insulating systems) then has no problems.