[Ifeffit] Re: Ifeffit Digest, Vol 20, Issue 22

[John J. Rehr]

    I tend to agree with Scott that a SCF E0 may be important in EXAFS
fits, since changes in E0 give phase errors, leading to distance
errors of order [delta R/R] =  [delta E0/ E_max]. Thus for a 1000 eV
data range a 1 eV error in E0 gives a .003 Ang error in a 3 Ang
bond length.

   However, I don't disagree with Matt's view that FEFF8 is not significantly
better for EXAFS than FEFF6.  The reason is that the SCF potentials in FEFF8
typically yield E0 with an accuracy of 1-2 eV, while FEFF6 is typically
off by over 3 eV. Such a difference is thus marginal for EXAFS analysis.
However, it could be important in complex systems where the FEFF6 estimate
is very far off. I often try to align the edge spectra calculated with
FEFF8 with experiment to improve on the E0.

   Both FEFF6 and 8 estimates tend to be high, due in FEFF8 to the muffin-tin
potential and limited basis set, and in FEFF6 to the electron gas model.
These numbers are obviously case dependent. In systems with band gaps,
even very small errors in the charge counts (in the SCF procedure)
give big errors in the Fermi energy! In my view improvements in both the
SCF potentials (especially muffin tin corrections) and the self energy
are needed. Efforts along these lines are in progress here at UW.

   Systematic studies of all these effects would be useful. In the past
both Matt and I have done number of comparisons of FEFF6, 7 and 8
e.g., on test cases e.g., Cu, Ag and Au. More would be desirable.

    John Rehr