I'm interested in the change in occupation of the BZ for specific, small (tens of meV) shifts in Ef, for which I need an accurate Ef (to ~ meV).
If I integrate the TDOS (from fine DOS calculations) numerically (trapezoidal or Simpson's method), I find a different Ef than the SCF EFERMI (obviously, I realise integrating the DOS is not trivial). I've tested with Al using version 1.4.22, and without going to a large k-mesh in the SCF cycle, my numerical integration is ~ 100 meV different than the SCF EFERMI. How accurate is the SCF EFERMI?
Thanks in advance for any help,
I am not sure you are considering the "broadening" of the states when you do the numerical integration. As standard the DOS in the SCF cycle is multiplied with a distribution function, either Gaussian or Fermi-Dirac. You should compare your integration withe the Fermi energy you get by using a vanishing broadening width, swidth, in a one iteration ELK calculation, preferably from a SC STATE.OUT. Or alternatively by explicitly include a (temperature-dependent) FD distribution in your numerical integration.