[Pw_forum] Problem with nbnd in clusters

Nicola Marzari marzari at MIT.EDU
Wed Nov 21 15:00:10 CET 2007


Cyrille Barreteau wrote:
> Stefano,
> 
> I did various checks on the smearing (mv,mp,fd..)
> and the broadening. For nbnd=20 I systematically
> find the same 'odd' result if diagonalization='david'
> and I recover a 'reasonable' result if diagonalization='cg'
> (similar to the one obtained with nbnd=12,25).
> Let me also mention that the "odd" result gives a
> lower energy than the "reasonable" one.
> 
> cyrille
> 


Dear Cyrille,

1) I wouldn't focus on cg vs. david . They both converge
to a scf state - the fact that it's different it's
just related to the fact that both might be scf state, and
one recipe takes a path that lead to one, and another to the
other.

2) it looks like the difference might be driven by the fact
that in the 12 or 25 bands calculation, you get 3 spin
down states around -3.75 to -3.72 eV , close enough that they
will have fractional occupations. In fact, -TS is more
significant in the 12/25 bands calculation, than the 20 bands
calculation

3) tightening the scf accuracy, and decreasing the smearing, in my
opinion would ultimately lead the system to occupy fully the lower
of the 3 states, and empty the others

4) note that the energy that pwscf writes is what I'd call E-TS,
and for these systems you want to subtract -TS (i.e. add TS)
to obtain E. Otherwise, if the system at zero T has fractional 
occupations (say because two states are exactly degenerate) and you
increase T (and any other empty states are much higher) the energy
that PWSCF writes as total energy does change, while in principle
nothing is happening (i.e. only T is changeing in E-TS, while
E and S do not change).

5) the case still remains - for the finite broadening, do you have 
really two very similar scf states ? My wilde guess that in reality the
system wants to break symmetry, and occupy fully one of the three
states, but is living on a very shallow potential energy surface, that
become a tight less shallow if you decrease the smearing

6) this open up my favourite question - should B be spherical or
cylindrical in DFT ? I.e. should we look like for occupation 1 0 0
for the 2p levels, or 1/3 1/3 1/3 ?

7) Last, but important - even using Gamma sampling, PWSCF finds
for your geometry 16 symmetry operations, and thus heavily
symmetrizes the system. That means that in 6) even 1 0 0 would
become spherical. I.e. the symmetrization has a similar effect
to temperature, in creating a charge density that is more symmetrical 
/degenerate occupations than it could be.

I would be *Very curious* to see what happens if you were to
put the dimer in a very random position, so the code misses most
of the symmetry operations, and also you specify no_sym, so no 
symmetrization is performed.

			nicola




-- 
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu



More information about the users mailing list