[Pw_forum] negative densities

[Stefano de Gironcoli]

When using US pseudopotentials negative densities can appears because
the cutoff used is not sufficient to include all relevant Fourier component
of the augmentation charges.
When this happens the augmentation charge do not decays with an
exponential tail as one expect but has an oscillating tail that does not 
decay.
If your system has "empty" regions where no other electron penetrates
significantly then this oscillatin g tail could be the only contribution 
to the
charge density and this could therefore result to be negative.

This is your case: a single iron atom in a rather open structure
Other systems where this happens are isolated atoms, molecules in a box, 
slabs
simulating surfaces and similar. In bulk metallic systems this problem 
should
not appear or involve only a minority of sites.

After all, this negative points should not be harmful. They appear in 
regions
ofmotherwise very low density that contribute very little to the total 
energy
and other properties. But it is annoying to see them.

The solution should be to increase the cutoff (or at least the cutoff 
for the
density ecutrho). It seems that this does not improve in your case but
probably you need a larger value of ecutrho/ecutwfc.
One more thing to look at would be the lowest value of the negative charge.
It could be that the fraction of negative points does not decrease (because
the tail is still oscillating) but they are approaching zero. This would be
a signal that things are improving after all. You need to edit 
PW/v_of_rho.f90
in order to compute and output this information but should be easy.

Stefano dce Gironcoli


Noah_Park wrote:

>Dear All;
>When I calculate Iron atoms in the interstitial of graphite,
>I found the following messages are persistent over the self-consistency steps.
>   > npt with |zeta| > 1 : 53321, npt tot 878400, 6.10 %
>   > npt with rhoup <0 : 266028, npt tot 878400, 30.41%
>   > npt with rhodw<0 : 266028, npt tot 878400, 27.81%
>
>Assuming it is due to incompleteness of FFT mesh, I increased the energy cutoff.
>But such "npt"s do not decrease measurably.
>My question is that,
>    1. what is the origin for such unphysical(?) charge density on a certain point ?
>    2. Is it really no problem, as the code does not produce warning message ?
>
>When I tested Iron bulk(FCC, and BCC), very small amount of  such npt points appear,
>    > npt with rhoup <0 : 1, npt tot 5832, 0.02%.
>However, appearing physical results, for example, 
>the energy difference between nonmagnetic FCC and magnetic BCC looks decent.
>
>I really appreciate any of comments.
>
>Noah
>
> 
>ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÏÃ÷è®é¦j)bž	b²Óðýú+ºoéÂÇþŠà†ÛiÿÿðÃÝzj
®+h³ø­þf¢–f§þX¬¶)ߣúpýú+
>  
>