[Ifeffit] Other programs than FEFF - GNXAS and MXAN

[Matt Newville]

Hi Matthew,

> One reason I can think of for GNXAS to be less popular than
> FEFF is that it hasn't been compiled for Windows.

Oh, I think you're right.  FEFF made a conscious effort to be portable,
and easy to use.  In John's defense of this approach, by aiming for a
wide and diverse user base, bugs and mistakes are found quickly,
and features important to the experimentalists are added.  I think
there's an additional point that Feff had a larger group of people working
on it, and a more sustained and diverse development.  That does not
make it the best theory code.  But Wu's assertion was about preference,
not correctness ;).

Feff deliberately did not include graphics or analysis, with the hope /
expectation that lots of other people would do this. This turned out to
be true, and very many analysis programs use Feff.   It is perhaps
interesting that this has not happened for GNXAS.  In contrast, it is
uninteresting for why this did not happen for EXCURVE: they gave
an exclusive license to a company, so that regardless of how good
that theory is, they chose for it to never be in wide use.

I don't know why GNXAS hasn't been ported to Windows.  I'm sure
it has graphics (it used to be topdrawer?), but I do not know how
tightly it is integrated into the code.   In any event, GNXAS seems to
not aim at being user-friendly.

> What is this 'interesting approach'?  [ of GNXAS]

Bruce discussed this, and I may get some of the details wrong (hopefully
someone will correct me) but:   GNXAS models g(R) directly and applies
this in an integral of the EXAFS equation.  It does not use a sum over
paths with a set of cumulants for each path.  I do  not know what
limitations are put on g(R), but I think they may have a few lineshapes
used to model g(R) (I do have the documentation somewhere!).    If I
remember correctly, GNXAS also fits the full mu(E) spectra,  including
the background and EXAFS, but this is not what I consider "interesting"
about GNXAS.

The direct modeling of g(R) is definitely a useful approach for highly
disordered systems, and the GNXAS authors and collaborators have
tended to study glasses, melts, and ion in solutions and biomolecules,
which are the sorts of systems where a simple cumulant expansion
is most likely to break down. In fairness, this is possible with a sum of
paths  by using a "histograms" of paths  with varying amplitudes.   The
spatial resolution of EXAFS is such that the "True" g(R) will only be
sparsely sampled, and that sampling can be modeled in a variety of
ways: it is not the analysis programs which limit the ability to model g(R).
For what its' worth, Bruce has a nice  demo  in one of his many tutorials
that uses a histogram to mimic what GNXAS does.

That's not about "theory" as analysis. I don't know what theoretical
advantages GNXAS might have.  I believe it includes curved wave effects,
and that it uses the Hedin-Lunqvist exchange (perhaps the only option?).
I think that it has a simple polarization model  (perhaps only dipole??).  I
believe it does multiple-scattering only out to four-leg paths  (good enough
for almost  all cases), and don't know how it does this.  I recall from one of
the GNXAS papers that they talked about correlating Debye-Waller factors
for MS paths with those of SS paths, but I don't know that works in practice.

For background subtraction, I believe it uses a simple cubic spline with
the possibility to add ad-hoc arc-tangents for multi-electron excitations.
Again, I could be wrong on these details -- I'm going from memory of the
papers and documentation.   But I would conclude that GNXAS has no
*theoretical" advantages over Feff, but might have a small analytical
advantages over Feffit/Artemis for highly disordered systems.  OTOH,
GNXAS is by all accounts hard to use, so it might be easier to set up a
histogram model with Artemis than to run GNXAS at all.

> MXAN seems to have a fit module which allows you to start with a
> cluster of atoms and automagically move them around to improve the
> XANES fit.  By using XANES instead of EXAFS, you can apparently see
> quite far out from the absorber.

Sure, though I'm not sure that MXAN has been demonstrated to be able
to do this better than other XANES approaches.  I may have missed some
papers, but I've mostly seen it applied to small molecules.

> A quick Google search didn't reveal any obvious way to acquire this
> package, so it may not be publically available.  Anyone know where/how
> to get it?  It might make a nice compliment to FEFF.

I got a copy in 2001 by asking Benfatto (Maurizio.Benfatto at lnf.infn.it).

> That reminds me - I'm a co-author on a paper, which was submitted to
> PRL, in which we used FEFF8.1 to simulate EXAFS.  A referee
> complained that FEFF8.1 was for XANES, not EXAFS.  What was that
> about?

That's silly.   I would (and often do!) say that Feff8 is not necessary for
most EXAFS. But Feff8 is certainly sufficient for EXAFS!!  There are a few
cases where  it's better than Feff6 for EXAFS,  but I don't know of any cases
where Feff8 is worse.

--Matt