[Ifeffit] EXAFS Divination Set results
scalvin at slc.edu
scalvin at slc.edu
Sun Aug 12 21:00:56 CDT 2007
I thought you might want to know about the first results from the "EXAFS
Divination Set" project I've been plugging. I don't think the results will
be any surprised to the veterans on this list, but might give novices an
idea of what's possible.
To briefly recap, my students made up mixtures of various iron compounds
in various proportions, and then I analyzed them "blind." The idea was to
see how well XAFS analysis as practiced by an "expert'" can do in that
kind of case. Unfortunately, unbeknownst to me, the standards were poorly
prepped (i.e. they were very uneven). So these results show what can be
done with poor samples. I'm planning a follow-up for this fall using an
entirely new set of samples and standards prepped as carefully as we can.
Presumably better samples will generate more accurate results.
Nonetheless, it's nice to know what can be done under conditions that are
far from ideal.
The first two samples I analyzed were mixtures of known compounds in
unknown amounts (iron and hematite in one case; iron, goethite, and
humboltine in the other). My analysis gave the correct percent composition
to within 15 percentage points in all cases.
The next four samples were mixtures of two or three compounds from a list
(iron, hematite, goethite, humboltine, FeO, lepidocrocite, and magnetite).
I didn't know which compounds were chosen, whether there were two or
three, and how much of each. In two of the four cases I correctly
identified the compounds present. In one case I identified two of the
three compounds, and noted that there might be a third present in a small
concentration but that I couldn't confirm that; the three compounds proved
to be present. In the last case I correctly identified the majority
compound but misidentified the minor contributor. In three cases I found
the correct percent compositions to within 10 percentage points; the last
was off badly.
The next three samples involved a compound completely unknown to me mixed
with one or two compounds known to me. I correctly identified the class of
the completely unknown compound. In two cases the percent compositions I
found were correct within 10 percentage points; the other within 16.
The last sample was a mixture of the complete unknown with one or two
compounds chosen from the list, but without me knowing which ones. As
already indicated, I correctly identified the class of the complete
unknown. I also identified one of the two other compounds as definitely
present and the other as most likely if there were a third compound
involved. Percent compositions for this sample were not accurate.
You could therefore argue that this shows that bad data yields bad
results. But I'd tend to look at it another way--even questionable data
can yield pretty good results. In all cases but one the substances were
correctly identified (twice provisionally as a possible minor
contributor), despite the fact that many of the candidate substances are
structurally similar in that they involve octahedrally coordinated iron.
Percent compositions were usually good rough guides, although on occasion
they went astray.
I'll leave these data up at
and let you know when I get the high-quality follow-up project posted...
Sarah Lawrence College
More information about the Ifeffit