Hi Oleksandr,
The answer is to use as little smearing as you can get away with.
Only if you are having trouble converging to a ground state
because of oscillations in the total energy due to the varying of the
occupancy of the lowest unocc. orbitals from the iteration (but
also check the density mixing parameter) should you increase the
smearing.
The inaccuracies are, in the final density (because it will
include some contributions from empty states) as well as
in the total energy in case you are comparing two systems
for example...as you know the free energu E-TS should be minimised
at finite T. BUT you can correct for this in a series expansion
for T..in some paper but I don't remember which.
Yes, metals tend to have a lot of states at the fermi level both
above and below and crossing which is why it's necessary to use a higher
value.Semicond. surfaces and other things like impurities in semicond.
will introduce some surf states but not too many. Use a k point
sampling if you feel that the states cross the fermi level.
For you I would start with the 300K value, get a good ground state and
then reduce it to around 0.05 eV but again I don't know exactly
what you are dealing with so you have to experiment. The above
worked for me.
Cheers
Demetra Psiachos
On Tue, 16 Jan 2007, Oleksandr Voznyy wrote:
Hi everybody,
I've plot the MP occupation function for different temperatures and checked
the amount of smearing obtained (for zeroth order, for higher orders it would
be slightly less smeared):
300K - 0.12eV
1200K - 0.4eV
3500K - 1.2eV - this corresponds to sigma=0.3eV=3500K, the smallest value
used by Kresse, Furthmuller, i.e. for most materials even bigger values can
be used.
My question is whether it is correct at all to use such a big smearing?
The problem that in case of a big smearing, very high levels become populated
and total energy is calculated based on that, but that high levels are badly
reproduced in DFT.
I'm not sure if this is a problem for metals, but for semiconductors we know
about 50% underestimation of bandgap.
So for my case of states in the gap of semiconductor, which might be
partially or completely filled, i.e. have metallic character, would it be OK
to use big smearing so that conduction band will become significantly
occupied (even though that entropy is still very small)?
