[Wien] Using LSDA+U for CeO2/NiO system

[Stefaan Cottenier]

> 1)  Which of the three LSDA+U methods would be best suited for
> describing the Ce(f) and Ni(d) electrons?

Looking at the description in the UG (sec. 7.2), it looks like the version
indicated as SIC (and now often called FLL) is the most appropriate one: you
have 'strong' correlations for Ce and Ni (although, for Ce one could debate
how strong they really are). There are papers who compare the various
methods for NiO (using wien2k), check the papers of Pavel Novak.

> 2)  SO interaction will have negligilble effect on the properties that I
> am interested in, but I have read that it is needed in order to converge
> the SCF calculation when LSDA+U is used.  Should I include it for either
> of the two types of electrons?

This convergence issue will probably be system dependent. Magnetite
converges fine with U and without SO, hence, this is not a general rule. If
your calculation time allows it, including SO on Ce might be worthwhile as a
matter of precaution?

> 3)  From what I have read, the J value is not a very sensitive parameter
> as long as it is between about 0.5 - 1.0 ryd.  Is the U parameter just
> the Hubbard U that is used extensively in the literature and is the same
> value used for all three of the LSDA+U methods?

Forget about the J, you can use an effective U that is U-J (see UG and
probably also Shick PRB 60 (1999) 10763). Then, what is the value of that
effective U? The eternal question... U=0.6 Ry for Ce will probably not very
far off (my experience). For Ni, look at the literature (a.o. the NiO paper
mentioned above) to get an idea of a reasonable value. The bare Hubbard U is
not necessarily a good choice. The same value will probably not give you
exactly the same result for all methods, but most likely there will be no
qualitative differences.

Note: use a recent wien2k version (04.11 or the new 5.0), to avoid a
potentially harmful LDA+U bug.

Stefaan