FAQ on pseudopotentials:
Q:
I'm confused about how to select a pseudopotential type (Troullier-Martins,
Goedecker, etc.). What makes one choose say a Goedecker potential instead
of a Vanderbilt potential?
A:
The choice of a pseudopotential in CPMD calculations depends on
needs, available resources and taste.
Troullier-Martins norm-conserving pseudopotentials are probably the
most-commonly used type of pseudopotentials in CPMD calculations.
They work very well for non-problematic elements and they are quite
easy to create (note, that it is also easy to create a bad pseudopotential).
When using the Kleinman-Bylander separation, one also has to be
careful to avoid so called ghost states (e.g. many transition metals
need LOC=
with
being an angular momentum smaller than default value
which is highest).
Goedecker pseudopotentials are stored in an analytical form, that ensures the separability, but they usually need a higher (sometimes much higher) plane wave cutoff than the corresponding Troullier-Martins counterparts. Also the creation procedure is more complicated, but there is a very large library of tested pseudopotentials (mostly LDA but also some GGA pseudopotentials).
Vanderbilt pseudopotentials have the advantage of needing a much reduced plane wave cutoff. The drawback is, that only a limited subset of the functionality in CPMD is actually implemented with uspps (MD, wavefunction/geometry optimization and related stuff and only at the gamma point and you have to make sure, that your real space grid is tight enough). Also due to sacrificing norm-conservation for softer pseudopotentials, your wavefunction has very limited meaning, so that not all features available for norm-conserving pseudopotentials can actually be easily implemented or implemented at all.
For some elements it can be rather difficult to generate good (i.e. transferable) pseudopotentials, so you should always check out the available literature.
Q:
How do I choose the correct value of LMAX?
A:
If you use a Vanderbilt or Goedecker type potential only the format of
the LMAX-line has to be valid. The actual value is read from the
pseudopotential http and the value in the input file will be ignored.
It is highly recommended to still use values that make sense, in case you want
to do a quick test with a numerical (Troullier-Martins) pseudopotential.
Generally, the highest possible value of LMAX depends on the highest angular momentum for which a ``channel'' was created in the pseudopotential. In the pseudopotential http you can see this from the number of coloumns in the &POTENTIAL section. The first is the radius, the next ones are the orbital angular momenta (s, p, d, f,...). As an example you can determine a potential for carbon using f-electrons and set LMAX=F. Since the f-state is not occupied in this case there is very little advantage but it costs calculation time. In short, you can use values as high as there is data in the pseudopotential http, but you don't have to if it is not needed by the physics of the problem.
A fact that causes confusion is that Hamann's code for pseudopotential
generation always produces output for the d-channel, even if you only
request channels s and p. You should be cautious if 
and the
energy eigenvalue of the p- and d-channels are equal. In most of these
cases LMAX=P should be used.