Comparing energies of calculations with different ISMEAR

[smattsson]

Hello

I want to calculate with binding energy of a hydrogen fluoride molecule on a Ni surface and hence need to do basically 3 calculations to achieve this. I have learned to employ different smearing techniques for different systems - e.g. Gaussian smearing for molecules, and the Methfessel-Paxton method for metals - but I am now curious about possible errors if I use different smearing techniques when comparing absolute energies.

The techniques I would like to use are for the 3 calculations:

Ni slab: ISMEAR = 1; SIGMA = 0.20
Ni slab + HF: ISMEAR = 1; SIGMA = 0.20
HF molecule in vacuum: ISMEAR = 0, SIGMA = 0.20

Could I expect any intrinsic errors in the binding energy, E(slab+HF)-[E(slab)+E(HF)], using these methods?

For all the slab calculations, the [free energy-E0] is around 1 meV per atom.

In the molecular calculation of HF, using ISMEAR = 0 gives a lower E0 of -8 meV per atom, in comparison to the E0 using ISMEAR = 1 (SIGMA = 0.20 for both cases). Would this mean Gaussian smearing is to be preferred over M-P smearing, simply due to its lower absolute energy?

[admin]

Smearing is a numerical tool for improving the convergence.
When using small values of smearing energies should be the same.
Your value of 0.2, however, is very large.
In calculations of the binding energy you should use much smaller
values of ISMEAR.

[cpp6f]

I don't know, SIGMA=0.2 is quite normal for the calculations we typically run on metals. I think it's a pretty common value, at least in the catalysis and surface science literature. Be sure to use E0 (the energy extrapolated to 0K) rather than F

[smattsson]

First of all, thank you very much for your replies.

I decreased the SIGMA value to 0.1 (still using M-P smearing) and the F-E(sigma->0) difference is always significantly smaller than 1 meV per atom for the Ni surface.

When calculating a single HF in vacuum using Gaussian or M-P smearing, I get following SIGMA dependencies:


(20x20x20 Ang cell, 1 k-point, ENCUT=600, Prec = Medium)

Code: Select all

ISMEAR = 0
SIGMA	Free energy	E(sigma->0)
0.00		-7.8712001	-7.8712001
0.01		-7.8712001	-7.8712001
0.10		-7.8712001	-7.8712001

ISMEAR = 1
SIGMA	Free energy	E(sigma->0)	F-E0 [meV per atom]
0.00		-7.8712001	-7.8712001	0.0
0.01		-7.8712001	-7.8712001	0.0
0.025	  -7.8683012	-7.8692675	0.5
0.05		-7.8654022	-7.8673348	1.0
0.10		-7.8596043	-7.8634696	1.9

If I need to calculate HF on Ni using M-P smearing, how can I be sure my calculations are accurate? Should I do single point calculations with the tetrahedron method for HF (with 4 arbitrary k-points), Ni and HF/Ni instead?