WIEN2k-FAQ: How to select RMT radii?
©2001 by P. Blaha, K. Schwarz and J. Luitz
The determination of Rmt's is in principal not so crucial when you take care
that all expansions are properly converged (RKmax, LM expansion, Lvns, GMAX). However, in connection with other
defaults of the code and with respect of the required computer time, proper
settings of RMTs can be VERY IMPORTANT. Let me try to make a few statements, but it is important that you
consider all points, not just one or two.
Unless you are very experienced and test a specific situation carefully
yourself, it is highly recommended that you use the setrmt_lapw
script, and accept the RMTs as set there. Since an automated script can never
be perfect, small variations of these suggestions are possible and might improve
the performance a bit.
- In order to make the calculations as fast as possible, choose the radii as
large as possible, but consider possible structural changes (lattice
parameter or force-minimization), such that you can do all calculations with
the same RMTs.
- Choosing them smaller will make the calculations more expensive (it will
leed to more plane waves, i.e. the matrices become bigger), but eventually
also more accurate when you stay with the default inputs (PWs are more
general basis functions, the linearization error gets smaller (unless you use
HDLOs), better series convergence (unless you increase Lvns or GMAX,
...). However, if core-leakage starts due to smaller spheres, the results
- If core charge leaks out of the spheres (check :NEC01 in
case.scf whether too much charge is missing) you can either:
- include some high lying states as valence states by putting an
extra LO for that state (a good example is Si 2p, which is by default
in the core, but for small spheres should go into valence (set a lower
Ecut in init_lapw),
- or "touch .lcore" to form a proper superposition of the core densities
leaking out of the sphere during scf (done automatically during init_lapw).
- Do not make your Rmt's too different!!! (even when geometry would allow)
Some rules of thumb may be:
- atoms with d (f) states may be 20 (30)% larger,
- when space permits, 30% larger spheres are still ok, but
when the spheres are too different (eg. 1.3 and 3.0 a.u.) this will lead
to quite different "effective" RKmax values for these two atoms and
eventually lead to approximative linear dependencies and
ghostbands (QTL-B error).
- When your RMT are larger than 2.5 (2.2 for highest precission) bohr,
choose LVNS=6-8 and consider setting of HDLOs for d and in particular f-states.
- For hydrogen in very short C-H, O-H or similar bonds: Select the H sphere
radius about half of it's partner (typically C: 1.2, H: 0.6 bohr).
Reduce RKMAX (3-4 is sufficient; check!), accept the automatically increased
GMAX (case.in2) or use an even bigger one.
- Identical atoms (elements) should have identical RMTs.
- The standard ionic radii are NOT important in determination of Rmt.