converging M06L calculations
Moderators: Global Moderator, Moderator
-
WolverBean
- Newbie
- Posts: 27
- Joined: Wed Dec 28, 2011 7:13 pm
converging M06L calculations
I have recently started using the M06L metagga functional, and have found that it gives good results for calculations on bulk metal oxides. However, I find that calculations containing vacuum space fail to converge, even when reading in a pre-converged PBE wavefunction as an initial guess (as recommended by the Truhlar group). Has anyone else experienced this problem? Can anyone recommend a solution?
Last edited by WolverBean on Thu Nov 08, 2012 7:58 pm, edited 1 time in total.
-
WolverBean
- Newbie
- Posts: 27
- Joined: Wed Dec 28, 2011 7:13 pm
converging M06L calculations
I have found the solution to this problem. When using metaGGA functionals in VASP, often the default integration grid is not fine enough to accurately evaluate the kinetic energy density. Manually increasing the values of NGX, NGY, NGZ to produce a finer integration grid solves this particular convergence problem.
Last edited by WolverBean on Fri Nov 16, 2012 1:09 am, edited 1 time in total.
-
fish
- Newbie
- Posts: 12
- Joined: Tue Jun 14, 2005 1:13 pm
- License Nr.: 198
- Location: Argonne National Lab
converging M06L calculations
Could you share the INCAR and POSCAR you used with your M06L calculation?
Last edited by fish on Thu Jan 24, 2013 3:07 am, edited 1 time in total.
-
WolverBean
- Newbie
- Posts: 27
- Joined: Wed Dec 28, 2011 7:13 pm
converging M06L calculations
fish,
To run M06L calculations, I first run the same calculation (same ENCUT, KPOINT configuration, etc) using GGA=PE, and save the CHGCAR and WAVECAR. I then run a second calculation with the following terms in the INCAR file:
ISTART=1 !read in preconverged PBE wavefunction
ICHARG=1 !read in preconverged PBE charge density
METAGGA=M06L
ISYM=2 !using PAW cores
LASPH=T !this is essential!
NELMDL=0 !no non-selfconsistent steps because I'm starting the calculation reading in an already self-consistent wavefunction
ALGO=A !Initial testing suggests ALGO=Damped also works but is 2x slower for my system. I have not had success with ALGO=N,F, or V
I've found that good settings for NGX,Y,Z seem to be something like 12 points per angstrom of lattice vector. This works out to be roughly double the number VASP would use by default. So for example, if I wanted to run a calculation on litharge (PbO) with a 3.96 x 3.96 x 5.00 A unit cell, I'd use
NGX = 48
NGY = 48
NGZ = 60
If I then wanted to create a slab with 12 A of vacuum space in the c lattice direction, I'd increase NGZ to 204. (These calculations get expensive very quickly!)
Nothing special is required in the POSCAR file. The POTCAR file must be built from the new VASP 5.2 cores, as the old cores do not contain kinetic energy density information.
I hope that answers your question.
<span class='smallblacktext'>[ Edited Sat Feb 02 2013, 03:22AM ]</span>
To run M06L calculations, I first run the same calculation (same ENCUT, KPOINT configuration, etc) using GGA=PE, and save the CHGCAR and WAVECAR. I then run a second calculation with the following terms in the INCAR file:
ISTART=1 !read in preconverged PBE wavefunction
ICHARG=1 !read in preconverged PBE charge density
METAGGA=M06L
ISYM=2 !using PAW cores
LASPH=T !this is essential!
NELMDL=0 !no non-selfconsistent steps because I'm starting the calculation reading in an already self-consistent wavefunction
ALGO=A !Initial testing suggests ALGO=Damped also works but is 2x slower for my system. I have not had success with ALGO=N,F, or V
I've found that good settings for NGX,Y,Z seem to be something like 12 points per angstrom of lattice vector. This works out to be roughly double the number VASP would use by default. So for example, if I wanted to run a calculation on litharge (PbO) with a 3.96 x 3.96 x 5.00 A unit cell, I'd use
NGX = 48
NGY = 48
NGZ = 60
If I then wanted to create a slab with 12 A of vacuum space in the c lattice direction, I'd increase NGZ to 204. (These calculations get expensive very quickly!)
Nothing special is required in the POSCAR file. The POTCAR file must be built from the new VASP 5.2 cores, as the old cores do not contain kinetic energy density information.
I hope that answers your question.
<span class='smallblacktext'>[ Edited Sat Feb 02 2013, 03:22AM ]</span>
Last edited by WolverBean on Sat Feb 02, 2013 3:20 am, edited 1 time in total.