[Pw_forum] epsilon.f90 skips transitions involving similarly occupied states

[Roberto Gaspari]

Dear pwscf community,

I was reading through the epsilon.f90 code and I noticed that (5.2.1 version) at
line 420 we have the following condition

IF (abs(focc(iband2,ik)-focc(iband1,ik))< 1e-3) CYCLE

this basically means that the contribution  to the dielectric function due to the transitions between
iband2 and iband1 is neglected if the two states have nearly the same occupation. I removed this
condition and noticed that, in some test case, states close to the Fermi level and closely spaced in energy would otherwise contribute a huge amount. This seems to be due to the fact that the denominator of eq. 10 ( epsilon.x manual ) becomes close to zero.

I would like to ask experts of pwscf and epsilon.x, in particular, if the CYCLE condition was introduced for any specific physical or numerical reason and why it is safe to proceed this way.

Thank you very much for your attention and have a nice day.

Roberto Gaspari
Ph.D.
Compunet-Istituto Italiano di Tecnologia


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20160513/1e34ab6e/attachment.html>