[Ifeffit] Dodgy Edge Steps

[Matt Newville]

Lachlan,

Sorry for jumping in late on this discussion, but I was out of town.
I did want to add something to this discussion:

I think an edge step as large as 3.4 could very well give fine data.
I've seen many cases of good data on standards with edge jumps > 3.
The issues addressed so far (that so many x-rays are absorbed that
you'd get noisy data from counting statistics, and thickness / pinhole
effects) are good points, but not necessarily fatal.

If you were really running out of x-ray intensity,  the data would
probably look very noisy anyway.  You didn't mention the detectors you
used, so I'll assume it was a standard ion chamber.  These tend to not
work very well when the x-ray shot noise is an important noise source
(even for fluorescence "Lytle" chambers), so I'd doubt this is a
problem and assume you had enough x-rays.

The rest of the counting chain could cause problems.  Generally,
current amplifiers aren't linear over many decades, so we're often
fighting trying to find the "best" gain settings for a particular
*total absorption*.   An edge jump of 3.4 (~0.03 transmitted intensity
above edge as below edge) could be OK, but it's starting to get close
to amplifier limits.  The signal from current amplifiers are often
sent through a V2F (Voltage to Frequency Converter) which can have
remarkably low resolution, so that interdigitation becomes a problem
for small signals:

<side track> (AKA I learned this the hard way)
The system I use (and common), uses a current-to-voltage amplifier
that saturates around ~6 Volts, so I try to not have voltages higher
than 5 Volts.  This signal goes to a V2F converter where 1 V gives
100000 counts per sec (this is the integer recorded as Intensity).  If
the voltage is 0.01Volts, I'd get 1000 counts per second.  At that
voltage, it doesn't matter what the x-ray intensity is or the current
amplifier settings, the "noise level" WILL BE 1e-3 (I can record 999,
1000, or 1001, but not anything in between.  I could count longer than
a second, but that isn't a huge win).    So, to be safe, I really want
the voltage (never mind the x-ray flux) to be between 0.05 and 5
Volts.  That's only two decades (so, a max edge step of 4.6 =
ln(100)), and I have to fiddle with the amplifier gains so it doesn't
saturate below the edge or go too low above the edge.
</side track>

But: if the data looks OK, the counting chain is probably not killing your data.

The main issues with large edge steps are a) pinholes and b)
harmonics.  If the sample has pinholes and you measure an edge step of
3.4, it's possible that your sample is much much thicker, and that all
the transmitted intensity you do see is going through the pinholes.
This might be trickier to diagnosis after the fact, but if the data
looks like real EXAFS data, it is probably OK, though it may need
amplitude corrections.

Similarly, if the x-ray beam had significant harmonics (which are seen
fairly efficiently by gas-filled ion chambers),  it's possible that
your sample is much much thicker and all the transmitted intensity you
do see is harmonics getting through your sample.  Again, if this was
the case, I'd expect the data to look completely useless.

So, my advice would be:  if data with an edge step of 3.4 *looks* OK,
it probably is OK.   There may be pinhole / thickness effects.  If
you're using the standard to extract S02 or amplitude factors, you may
want to check the literature on how to correct for this.   If other
standards can give you S02, and you're using this one to see if you
can model the distances, I'd guess the data was still useful.

--Matt