[Pw_forum] vc-relax bug?

Konstantin Kudin konstantin_kudin at yahoo.com
Wed Apr 4 22:40:09 CEST 2007


 Javier,

 First of all, I recommend NOT suggesting that there is a "bug" in the
code right away when something does not seem to be clear to you
personally. And even after one's expertise is vast, a significant
percentage of seemingly weird related to the code things still turn out
to be not actual bugs, but *intended* (though still weird!) features.

 With this important philosophical issue out of the way :-) , it looks
like what you get in your calculations is an *intended* feature. With
vc-relax the basis set is changing during the simulation, so then if
the number of plane waves is kept fixed the cutoff effectively "floats"
(that is what you seem to have been doing). Then, when you restart your
calculation freshly from the input you get a *different* basis set.
Hence the difference in energies, forces, and stresses.

 The basis set issue is an important and well-known one, and there is a
mechanism to compensate for it. If for fixed cell calculations you use,
for example:
    ecutwfc =80.0
then for vc-relax you need to use more parameters such as:
    ecutwfc = 95.0
    ecfixed = 80.0
    qcutz = 80.0
    q2sigma = 9.0

[to developers: these values need better defaults than those in the
manual:
ecfixed        REAL ( default = 0.0 )
qcutz          REAL ( default = 0.0 )
q2sigma        REAL ( default = 0.1 )
like ecutwfc, qcutz and q2sigma could all be certain percentages of
ecfixed, triggered by "vc-relax".]

 These parameters add a buffer zone with planewaves, such that during a
vc-relax run planewaves will be smoothly added or removed as needed,
although still not giving the exact match with the fixed cutoff runs.

 I hope this helps.

 Kostya 


--- Javier Antonio Montoya <j.antonio.montoya at gmail.com> wrote:

> Hello dear all,
> 
> In recent days I have been trying to get a triclinic cell to have a
> diagonal
> stress by using vc-relax. The pw.x program reads data in the correct
> way,
> calculates the right starting volume for the cell given as input and
> also
> seems to do right the first electronic minimization but, as soon as
> the
> Wentzcovitch Damped Cell-Dynamics Minimization starts, it gives cell
> parameters and ion positions that are not consistent with the
> energies and
> stress that it prints (i.e. if you take those cells and ions to an
> independent scf run the final energy and stress differ from the ones
> in the
> vc-output) the problem is that in the end, when I get a nice diagonal
> stress
> after 60 or more dynamic steps, the difference between the reported
> energy
> in the vc-output and the energy of the system that you get with the
> cell and
> ion parameters shown is of almost 1 Ry and the pressures differ by
> more than
> 10GPa. Any idea? by the way the ph.x complain and doesn't get
> anything from
> such restart files.
> The output files are huge for sending by e-mail because my system is
> 72
> atoms big, but I can provide them.
> 
> cheers,
> 
> JAvier.
> 



 
____________________________________________________________________________________
Don't pick lemons.
See all the new 2007 cars at Yahoo! Autos.
http://autos.yahoo.com/new_cars.html 



More information about the users mailing list