[Pw_forum] General pseudopotential generation questions (using ld1)

[Tyler Grassman]

Hello all.  I'm currently trying to generate some norm-conserving
pseudopotentials for use in some calculations I want to try (and they are
not available for download).  I've read Paolo Giannozzi's guide a few dozen
times, in addition to the basic pseudopotential literature, and it's been
extremely helpful so far, but I still have some questions.  I've managed to
generate a number of NC-PPs (RRKJ, PBE) for Ge, H, and O so far, and have
been trying to thoroughly test them...  For testing I'm comparing the AE vs
PS wavefunctions, the logarithmic derivatives, the transferability tests
using different electronic configurations using the ld1 code, as well as the
Bessell function basis tests.  After the PPs pass these tests I'm performing
bulk convergence tests and checking for lattice paramter and density of
states and that sort of thing (for Ge), and diatomic bond lengths and such
for H2 and O2.  Pretty standard stuff, I would think.  And I think I've
managed to generate some decent (I think) PPs from a combination of my own
experimentation, as well as utilizing some of the input values used in the
VASP pseudopotential library (the NC parts of the US-PPs).  However, at this
point, I've run into a few questions regarding some of the parameters
available for PP generation.

First, the potential mixing paramter, "beta."  The default is 0.2, but I
found in some of the examples in the atomic_doc directory values of 0.5.
How does this paramter effect the PP?  I tried playing with it a little with
my O potentials, but didn't really see a difference.  So, what does it do,
exactly, and how important is it?

Second...  Choice of local potential (i.e. lloc=-1 vs. lloc>-1).  I think I
understand, in principle, what the difference is between these two
approaches, but I'm not quite sure that I understand the end result.  So,
for the O potentials, nothing weird happens, but using lloc=-1 tends to
produce PPs that give me an O2 bond length closer to the experimental value
with a smaller cutoff energy than do similar potentials (e.g. 2s2 2p4 with
lloc=1, 2s2 2p4 3d-1 with lloc=2).  With Ge, however, the results *is*
weird.  The potentials produced with lloc=-1 (4s2 4p2, no semicore d) give a
lattice parameter that is actually smaller than experimental (10.60 vs 10.69
Bohr, which definitely seems funny for a GGA calc), while potentials using
4s2 4p2 4d-1 (lloc=2) and 4s2 4p2 4d-1 4f-1 (lloc=3), give values larger
than experimental, as would be expected.  The lloc=3 case, which is the same
used in the VASP US-PPs, gives the best results (10.91 Bohr).  Any
suggestions as to why this might be occuring?  And does anyone have any
pointers as to when using the lloc=-1 vs lloc>-1 is or is not appropriate?

Finally... rcore values for non-linear core correction...  I find that the
core charge calculation depends upon what I set rcore to (as well it
should).  If rcore is set in the input file, the local potential and core
charge radii are equal.  However, if rcore is not explicitly set in the
input file the core radius chosen for the local potential vs. that chosen
for the core pseudo-charge are different; the local potential rcore is
chosen as the largest rcut, and the core pseudo-charge radius is chosen by
some other method (and, at least for Ge, it tends to be quite small compared
to the rcut values).  Is this perhaps a bug?  It seems like it would be
safest just to make sure to set rcore by hand, but I just wanted to check.

Well, sorry for such a large e-mail, but hopefully someone will be willing
to answer these questions (and hopefully such answers will be helpful to
more than just myself).  I realize that I did not attach the
pseudo-generation inputs, mostly just because I felt these questions were of
a more general nature, but I would be happy to provide the inputs should
anyone feel they are necessary.  Thank you very much!

Regards,

Tyler Grassman
Dept. of Chemistry/Biochemistry, Materials Science and Engineering
University of California, San Diego