calculate the dipole moment of a slab

[hanhanfudan]

Dear All,

I was trying to calculate the dipole moment of an absorbed surface with

IDIPOL = 3 (the z direction is normal to the surface)
DIPOL = 0.5 0.5 0.3 (the center of the slab)

then read the dipole value from OUTCAR(search for the string 'dipolmoment' there)

But I find the dipole value in OUTCAR seems not correct? Because I also calculated the dipole by intergating CHGCAR, the result(0.123eA) is different from the value in OUTCAR(0.018eA).

I also tried dipole correction by adding LDIPOL=.True. But the dipole value read from OUTCAR and calculated from CHGCAR by myself still differ.

However, I think they should be the same, and I think my calculation from CHGCAR is right(the charge of the ions is also considered)

Any of your suggestion will be highly appreciated. Thanks!



<span class='smallblacktext'>[ Edited ]</span>

[dhfphysics]

I have a similar problem, although I do not use any DIPOL features and my surfaces/interfaces are abrupt.

By adding Gaussian charges to account for the cores, and by either averaging either the lattice period (for a surface) or averaging twice, over the period of both interfaces (for and interface) [Baroni et al., in "Spectroscopy of Semiconductor Microstructures", ed. G. Fasol et al., Plenum Press (1989) 251-271] I obtain a smoothed total charge, based on the CHGCAR, in a symmetric supercell.
The same averaging is done for the potential (LOCPOT).

Two examples illustrate large differences in the physical dipole and the work function (surface of unrelaxed Be metal slab) or the local vacuum offset (S-Si normal-dangling-bond (100) interface between Si and ZnS in a sufficiently long, relaxed, symmetric supercell).

The calculated Fermi energy in bulk Be is used with the potential jump from LOCPOT to determine a work function of 4.8 eV (experiment ranges 4.9 to 5.1).
Likewise bulk Si and ZnS calculations yield E_V - V_ref values which are used to determine the valence band offset -2.1 eV, which is then used with experimental band gaps and electron affinities to obtain a local vacuum offset of 0.5 eV (Si into ZnS).

On the other hand, integration of the smoothed total charge yields dipoles of 0.18 eA and 0.79 eA for the Be surface and the Si/ZnS interface, respectively. These dipoles translate to potential offsets of 7.2 eV and 9.6 eV, in strong disagreement with the (above) potential-based values of 4.8 eV and 0.5 eV.

This discrepancy seems larger than could be produced by effects of exchange-correlation potential (cf. Wang et al., Chem. Phys. Lett., 522 (2012) 83-85) or any other effect I can think of.

I have performed the Be calculation with Be_GW and Be_sv_GW POTCAR at increasing ENCUT, PREC=Accurate, LADDGRID=T, LASPH=T, GGA_COMPAT=F, LREAL=FALSE!!, MAXMIX=4, ALGO=Normal, AMIX=0.1,AMIN=0.03. In scripts, I have used linear and cubic interpolation in conjuction with the charge and dipole integrals. As an alternative to Gaussians, I have also used delta functions for the nuclei and CHGCAR+AECCAR0 for the valence charge (works better than AECCAR2 which doesn't have a very accurate total cell charge). These two approaches give similar results. The supercell has 10 Be unit cells and 4 vacuum (Be-size) unit cells. I have converged KPOINT density with bulk, so it is 30x30x1 for the supercell. ISMEAR=1, SIGMA=0.15. To determine the dipole I integrate (against a line) from the generalized mirror symmetry point in the center of the Be slab to the center of the vacuum region (net charge in this interval is 0 by symmetry). I see a significant assymptotic change going from high ENCUT to very high ENCUT (for the Be_sv_GW POTCAR one can go to 3100 eV before warnings appear, giving a charge grid with about 40 points per linear angstrom). The result is thus very dependent on having a very find charge mesh. This makes sense considering it takes a fine mesh to avoid a dipole error of 0.1 eA. So I believe my Be calcs, but I do not believe my Si/ZnS calcs anymore, as I have nowhere near the spatial resolution.

So, here is the VASP physics question:

Do CHGCAR charges (plus core and nuclear charge), performed at sufficiently high accuracy, really give an accurate electrostatic dipole? If so, what can account for the large discrepancy between the charge based and the LOCPOT/Fermi Energy based estimates of the Be work function? The latter agrees well with experiment.


Thank you greatly,
David
<span class='smallblacktext'>[ Edited Thu Dec 19 2013, 08:43AM ]</span>