[Pw_forum] density of states of graphene

[Giovanni Cantele]

> Dear all, I have got results for the density of states of graphene using
> both scf and nscf techniques and both gave different results. The SCF
> calculation was with the C.pz-rrkjus.UPF and with a k points grid of
> 30x30x1. The NSCF calculation was with C.pz-vbc.UPF nad with 700 points
> (crystal_b)..I think the nscf converges more towards a zero DOS at some
> point. I am attaching both results and I would be greatful if someone can
> let me know if any of them looks like the DOS of graphene and why that
> difference.
> Regards.
> Elie Moujaes
> University of Nottingham
> Physics department
> NG7 2RD
> nOTTINGHAM,
> UK

Dear Elie,
I think there is something wrong with the procedure you are following. It
has no meaning (generally speaking) to have a DOS built using an "scf
technique" and a DOS built using an "nscf technique".

First try to understand what is the difference between scf and nscf: they
are not alternative techniques to get the same answer, but just two
different things.
An "scf" calculation consists in starting from an initial wave function
guess, building the initial Hamiltonian, diagonalizing it, getting the new
wave functions, then the new Hamiltonian. The process is iterated until
convergence is achieved. The final result is the ground state charge
density (GSCD), which is stored in some format within your outdir.

The "nscf" calculation means that you ALREADY have the GSCD, so you do not
need to iterate as described above, to obtain the self-consistent
Hamiltonian / GSCD. In this sense you say non-self-consistent-field=nscf.
Using this procedure you can get, for example, the eigenvalues along
special lines in the Brilluoin zone (BZ), namely, the band structure.
Else, you can calculate (by specifying the nbnd variable) the "unoccupied"
eigenvalues, that are not needed of course to build the GSCD.

Let's come now to the DOS calculation: what do you need for it? You need a
very fine sampling of the BZ, much denser than the one needed to
self-consistently calculate the GSCD. Because it would be very expensive
to do the scf calculation with such dense grids (the scf converges with
respect to the BZ sampling much faster than the DOS), one usually makes
FIRST the scf calculation with a reasonable k-point grid, THEN the nscf
calculation where you specify a much denser grid.

For example, a safe choice in the case of graphene would be:
- scf
K_POINTS { automatic }
16  16  1    0  0  0

-nscf
K_POINTS { automatic }
64  64  1    0  0  0
(128  128 1    0  0  0 might be checcked as well!)

That means that by sampling the graphene BZ with a uniform 16x16x1 grid
should be enough to get a converged GSCD, but if you want to calculate the
DOS you need a sampling with a 64x64x1 or 128x128x1 grid.