[Q-e-developers] Problems converging anions in vacuum

[Oliviero Andreussi]

Dear Lorenzo,

Thanks a lot for your answer. We also thought about some physical origin 
of the problem and the vacuum delocalization was also one of our main 
guesses. We want to use larger cells because even with pcb correction 
schemes, the energy is not converged enough for the smaller cells. We 
tried the smearing, but it didn't work. We looked at the density plots 
exactly to try to figure out where the electrons are going, but it does 
not seem like they are going into the vacuum (see the pictures attached 
in the first post). I admit that on this point I do not understand those 
plots enough to make any judgement, the mail was sent with this purpose. 
VdW-DF is an option, as well as EXX, we will try those is nothing works 
with normal functionals.
Thanks,

Oliviero


On 05/11/2012 12:21 PM, Lorenzo Paulatto wrote:
> On Fri, 11 May 2012 12:04:30 +0200, Oliviero Andreussi
> <oliviero.andreussi at epfl.ch>  wrote:
>> Just to make things more reproducible, I rerun some tests with clean
>> versions of QE, both the svn and the 4.3.2, the latter also using the
>> gamma only algorithm, you can find the input and outputs attached. There
>> are some small differences in the numbers, but the overall effect is the
>> same, the SCF goes nuts. I would like to stress that the problem is not
>> system specific, the same happens for the input atomic positions
>> translated or in larger cells, with or without martyna-tuckerman
>> correction, and the same happens for many anions (10 out of 60,
>> including OH-),
> Dear Oliviero,
> I'm not 100% sure that what I-m about to say is right, but I think it is
> the explanation of your problems. I think that one electronic state may be
> very little bound, or not bound at all, in DFT hence the electron prefers
> to delocalize instead on staying on the molecule. When the delocalization
> happens, it is very hard to make the system converge, and even if it does
> the result will depend strongly on the cell size and other factors that
> should not really matter.
>
> Anyway, this would not be a bug in the code, but a limit of the functional
> that is causing a pathological case. I guess Martina-Tuckerman would not
> help as it does not act on the bands, on the other hand vdw-DF or EXX
> could help. Less radical "solutions" may include: increase the number of
> bands, add a small smearing and, most effectively, reduce the size of the
> unit cell. I reckon that the latter could also introduce spurious
> interaction between periodic images, but there are corrections for that.
>
> bests
>
>