[Wien] When to add a U.

[Stefaan Cottenier]

> The main question is when to add a U:
> a) Always - with some guess/estimate?
> b) Only after estimating it from a supercell calculation (not so
> simple and not always possible)?
> c) Only when you can't explain experimental data, with the value
> of U chosen to "agree with experiment" - in which case it is really a
fudge.

I would say: c)

But with these precautions:

1) Often you know right from the start whether for a particular compound you
could need LDA+U, and what the U-value should be more or less. From the past
decade of LDA+U literature, several classes of materials can be identified
for which U has been shown to make sense (lanthanide compounds, correlated
transition metal oxides, ...), and often an interval of reasonable U-values
has been identified for each such class. If your material does not belong to
such a known class or if you need an unreasonable value of U in order to get
agreement with experiment, you should probably not use LDA+U.

2) If a reasonable U restores agreement with experiment, then there is
usually a good reason for that in terms of the band structure. Only with a
proper interpretation of the band structure differences between LDA and
LDA+U for your compound, a LDA+U calculation makes sense. Otherwise, it's
just fitting, indeed.

Stefaan