The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi Kim Duckyoung

Since Matthieu has posted this nice discussion to the forum I think some comments are expected although Matthieu did already a great tour of the question. So below are my few remarks.

On Dec 12, 2007 3:31 PM, Matthieu Verstraete <mjv500@york.ac.uk> wrote:

Hello Duckyoung, I have forwarded this to the forum as well. Probably will
interest others. Also, others may be able to answer your questions. This
really is what to forum is for, and you should send your queries there.

>As everybody does, I am calculating electronic strucutures with VASP or
>abinit or something else. The interesting thing with abinit is that I can
>calculate phonon dispersion w.r.t DFPT. But while I am doing that, I met
>one question about smearing scheme.

>As far as I know, abinit use two smearing schemes, one is broading of
>kpoints meshing and the other one is finite temperature mimicking.
No: there are several smearing schemes, but they can serve 2 purposes, and
act differently depending on the smearing function you use (occopt). The
physical temperature is the FD smearing (occopt 3) and the others are
intended to improve kpoint convergence. FD smearing also improves kpoint
convergence.

>With my system, I am using occopt = 3 and use tsmear. In the manual,
>abinit(or) says the default is 0.04 Ha for almost free electron type
>metal such as Al. But the corresponding physical temperature is already
>too high enough to melt the system away ( ~ 11000 K ). Even in your
>paper, you have used 3000 K smearing which is above melting temperature
>of Al. Thus I'd like to ask you what is the real meaning of tsmear = 0.04
>Ha when occopt = 3. Could you make some analogy between Fermi-Dirac type
>physical smearing value and "real system temperature"
No: the tsmear only gives the electron temperature. For fixed ions the
crystal obviously won't melt. If you allow them to move (MD) you will find
a transfer of energy to the ions, and it may melt if enough energy is
transferred (the electrons are effectively on a thermostat, so T_e is
constant and will keep feeding energy to the ions unless you give them a
thermostat as well - see the different ionmov options). In a real
equilibrated system the ions vibrate and the has a FD
distribution, both with the same temperature.

Beware that the transfer of energy Matthieu is telling you about is not at all thermal equilibration. It would require (at least) a special piece of code implementing the transfer of energy from the electrons to the lattice via electron phonon coupling (has Matthieu explained later in his answer). If you run MD with high temperature for e-, all what you will get is equipartiion of the extra potential energy associated with the electronic temperature between the ionic kinetic and potential energy. That is you can find initial system configuration such that no energy is transfered from you heated (and nonetheless groundstate) e- to your ions.
 

>I am calculating a system with tsmear = 0.04 Ha.
even with cold smearing this is quite a high value (as you have said
above) and will definitely not reflect the ground state properties.

>With less smearing value, the phonon goes to instable regime. Then, to
>explain the instalbility and stability change w.r.t tsmear, I should know
>the physical meaning of it. smaller would be better ?
The physical result is the limit for small smearing. In principle (as in
the tutorial) you converge your kpoints for a given smearing, then
decrease and converge kpoints again, until you get a value which is
converged wrt tsmear as well. So I would say your phase is unstable.
However you may be interested in a stabilizing effect due to electronic
temperature. This is similar to Kohn anomalies or other T related sh
instabilities, and could be due to the nesting of the Fermi Surface once
the temperature is low enough and not too smeared out.

You may want to have a look at 
http://anglade.googlepages.com/anglade-abinit-hands-on.pdf page 7-8 for a basic example of convergency with respect to tsmear. It show how much you and Matthieu are right when studying convergence of your inspected property with respect to tsmear. Indeed, one sees that quite not all quantities behave as nicely as lattice parameter when tsmear is increasded...



Regards,

-- 
Pierre-Matthieu Anglade