Dear Nicola,
Thank you for your answer and sorry for my late reply. The argument with
the Gibbs phenomena would make sense to me. Your argument, however, that
this should have become obvious in the pseudopotential generation does
not become clear to me yet: Would the Vanderbilt code not only complain,
if it were to suggest 'real' negative densities? ('real' meaning "before
interpolating the augmentation charges to a grid and then evaluating it
on an even finer grid").
I guess the question is not the most pressing one - for the time being I
simply switched to a different pseudopotential. It is more my own
curiosity that drives me to wanting to understand the issue. Also, it
seems to me as if the presence of negative densities weren't really
hurting the calculations (except for the fact that they are somewhat
tricky to converge), suggesting that one may be able to ignore the
warning, but I am not willing to do so before I haven't understood its
origin and can judge on a more scientific basis than "from the few
calculations I did it seems as if...".
Thank you and best regards,
Katharina
On 04/01/2019 21:20, Nicola Marzari wrote:
Dear Katharina,
good reminder, and I'm not sure what the consensus is.
I vaguely recall that if the augmentation charges are "hard"
(very localized), the pseudization of the q_ij function
ifqopt,nqf,qtryc,nfix (see discussion in
http://physics.rutgers.edu/~dhv/uspp/uspp-736/Doc/INPUT_GEN)
can give rise to a Gibbs phenomenon (if you try to interpolate
something hard like a step, you go negative).
For this reason, the CP code (that resricts to the small boxes nr1b
nr2b nr3b with fine resolution the action of the uspp) doesn't suffer
from this problem, that happens away from the core (PWscf has fine
resolution everywhere, in the action of the uspp).
Now, this should also have been seen in the pseudopotential generation
(see "negative densities" in
http://physics.rutgers.edu/~dhv/uspp/uspp-736/Doc/TUTORIAL), so I'm
the first one at a loss (of density).
nicola
Dear QE users and developers,
I performed DFT calculations using the grvb-1.4 uspp for Pt (just a
single atom in a cell) and I get rather large negative values for rho
even after scf convergence:
-----snippet of the pwscf log
file-----------------------------------------------------
negative rho (up, down): 1.759E-01 0.000E+00
total cpu time spent up to now is 41.4 secs
End of self-consistent calculation
Number of k-points >= 100: set verbosity='high' to print the
bands.
the Fermi energy is 22.2850 ev
! total energy = -210.69743345 Ry
------end of
snippet---------------------------------------------------------------------
Usually, I would have thought that this is a sign of a too low
ecutwfc (or actually ecutrho). But, while very similar results were
obtained for ecutwfc=35Ry and ecutrho=280Ry, the above results were
obtained for ecutwfc=100Ry and ecutrho=1000Ry, which should be
crazily large for the PP in question.
I guess, one might say that -0.17e can be safely ignored (see the
pwscf faq), and the lattice constant and bulk modulus are indeed
totally reasonable. However, the error scales with number of atoms,
so once I go to a Pt slab, I get negative charges around -3e and I am
hesitant to accept that without thinking twice.
A very similar question has been posed before
(https://www.mail-archive.com/[email protected]/msg18381.html),
but has not been answered.
Any hints?
Thanks in advance
Katharina Doblhoff-Dier
_______________________________________________
users mailing list
[email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
