[WIEN] DOS question

[Peter Blaha]

> > (Or if they are included, it's even worse:
> > the integration of a nearly zero-width peak is of course very inaccurate).
>
> Which seems somewhat ironic, since the tetrahdron method that is used to do
> the integration in calculating the DOS is a linear scheme (LATM), in that it
> linearizes the band structure to use "analytic" expressions for the integral
> over the tetrahedra of the Brillouin Zone.  For a given k-mesh one expects
> the tetrahedron integration routine to work better for semi-core states since
> they are flatter than conduction/valence bands (that is, semi-core bands are
> better approximated by linear fits).  So, the resulting contribution from the
> semi-core states to the DOS spectra, is in a sense more numerically accurate
> than the contribution from the higher states.

Actually this comment is correct. Semicore states are well described by
the tetrahedron method and also the analytical integral IS accurate.

The problem with semicore states: If they are very narrow and your E-grid
(case.int) is coarse, then you might even miss them completely!!

Suppose the SC-band is located at -5.999 to -5.997 Ry  (so very sharp bands,
just 2 mRy wide). If you select in case.int an energy grid of:

-6., 1.5, 0.005 ; i.e. a grid with 5 mRy spacing, then you search for the DOS
at  -6.000 -5.995, -5.990, ....

Of coarse on all those energies the DOS is zero, so you miss the SC states.

Anyway, I think it is quite useless to plot a "DOS" of those states. If
you really want to show their position on such a wide E-range, a "delta"
function peak can be plotted "by hand" at the correct energy.

Or you produce an extra SC-DOS using:    -6.,-5.99,0.0001
(The "band-limits" are given in case.output2 and case.outputt.)


A common mistake which could eplain some of your "integration" problems:

Suppose you have a SC-state from -1.2 to -0.9 Ry.

In case.int you specify the energy grid:    -1., 1.5, 0.003

Thus you start with your DOS right in the middle of that SC band. Of course
in that case you get a "non-integer" integral and also the number of
electrons up to EF is not what you expect (and some odd fractional number).

                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------