[Wien] Fermi broadening and optics

[Igor I Mazin]

I found a curious inconsistency in optical calculations. If I use TETRA 
in case.in2 for the lapw2 -fermi step, or TEMP with a very small 
broadening (1 mRy or less), or Gauss, I get more or less consistent 
results for jdos (and, of course, for epsilon2). However, if I start 
increasing the broadening parameter, jdos in TEMP rapidly becomes 
unphysical, with the absorption threshold shifting higher and higher 
with the broadening. At T=0.01 Ry it has shiften by 3 eV(!). With 
Gaussian broadening, this does not happen. The broadening of 0.01 Ry 
still gives jdos rather similar to TETRA.

Anybody experienced that before and/or has a clue as to why that may be 
happening?