[Wien] P and H differences

[Peter Blaha]

>    I want to compare energies between different lattice structures, and
> I need high accuracy. But when I cross-check the *same* lattice structure
> with different but equivalent case.struct-files, I find differences that
> are already larger than I can tolerate.  I append a sample struct file:
> running it either with the P lattice or with the H lattice produces
> energy differences of 7 to 10 mRy !
>
>    I tested this extensively for k-mesh and basis set size.
> Is there another parameter to consider? Or is this an unsurmountable
> accuracy limit, which would be very bad news indeed ....

Make sure you are using the latest WIEN2k_04 version! (I remember that
there were some GGA bugs in early non-orthogonal lattices? Please check
the "update" comments on the web)

Wien uses different coordinate systems for P and H lattices. In particular
for H the hexagonal coordinate system is used and and everything is
described by integer numbers (K-vectors, symmetry operations,...)

Anyway, usually significantly differences appear only in GGA calculations,
not in LDA! Please verify this first!

If it is due to GGA, it is because the various gradients of rho are
calculated numerically on the "FFT" grid, and this grid is different for
different lattices (this holds also for B of F,C centered lattices). You
may increase the accuracy with a higher GMAX (case.in2, up to GMAX=24),
which is fairly "cheap". Nevertheless, most likely you must do the final
comparison of different structures with the same "lattice type".

Regards
                                      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/
--------------------------------------------------------------------------