Hi Paul,
I was curious to tap the collective wisdom of those on the list
regarding the "true meaning" of S02, the passive electron reduction
factor, at least with regards to fitting experimental data to feff
calculated paths.
It may be (?) that the "meaning" of S02 from fits may differ from
the "true meaning" from a theoretical perspective. Luke Campbell
looked at this question in his thesis work [see "Interference between
Extrinsic and Intrinsic Losses in XAFS," L. Campbell, L. Hedin, J. J. Rehr,
and W. Bardyszewski, Phys. Rev. B 65, 064107 (2002)], and found
that, as had been anticipated, S02 is only weakly energy and path dependent.
The work also found that the intrinsic loss estimate for S02, i.e. the
many body overlap integral ||**2 which is typically 0.7
is too small, due to interference between intrinsic and extrinsic losses.
Instead S02 is more like 0.9. This also means that there is about 10%
of the spectral weight in multi-electron excitations, which can be
modeled roughly as an additional broadening term (and hence a shorter
mean free path).
The current estimates of S02 in FEFF are not very satisfactory.
FEFF will attempt to calculate this overlap integral (i.e., intrinsic
losses only) if one uses the card S02 0 in feff.inp. But, the
Hedin-Lundqvist self energy has too much loss, which can only be
corrected for by increasing the value of S02. Fortunately these errors partly
compensate each other.
Doing better requires better many body coding. Luke Campbell's codes
are not yet implemented yet - but we hope to add them in one of the next
releases. We are also working on better self-energies. Stay tuned
on this.
Thus at present it's still better to fit but not to let S02 vary too much.
In my view S02 might be best modelled as a constant about 0.9 +/- 0.1 plus
an additional broadening term (that is by allowing the variable ei to
vary by about an eV or so in feffit), i.e.,
S02= 0.9 exp(-2R/lambda(ei)
I seem to recall that Matt Newville has tested such fits.
A further problem with fitting S02 in practice is that fits to experimental
data depend critically on the edge jump. To minimize this, I would
recommend adjusting the near edge structure to theoretical calculations (e.g.,
from FEFF8) to fix the edge jump as well as possible, rather than simply
fitting it.
standards. How about the expected temperature dependence of S02? I do
I think the temperature dependence should be very weak. Mosly
the path dependence is in a correction to the mean free path.
I'll be curious to see what others say about S02 especially
fixing the edge jump or fitting ei rather than S02 or both ei and S02.
Cheers,
John Rehr
