Notes on Fe feature centroid energy
Introduction
This follows up on the discovery by Richard and Brian of a detector
gain dependence on livetime fraction.
The question is, of course, can we calibrate this well enough to use
the data for observing subtle solar effects such as ionization state?
I have now checked a couple of the important points and almost have
enough information to be able to make a calibration and parameter
fits for the shutter-out data at least.
Variation with livetime
The plot below shows G9 data for a two-minute integration on the
early phase of the April 21 event.
The fitting is as I have been doing it, namely a polynomial fit
to the background under the Fe feature, then a Gaussian fit to the
feature itself.
A couple of points are worth making.
First, the variation shows a good linear fit.
The raw data here were binned from 10-msec integrations and so
resolve the whole modulation.
A view of 5 sec of data at 50 msec binning is in the plot below,
so that you can see how the data gaps and livetime variations look.
I think that the linear dependence is a good thing, because in practice
we must deal with 4-sec integrations for which the modulation will do
whatever it wants to do.
The degree of linearity here looks good enough so that there may not
be much bias even early on when we had more dead time than we needed.
A second point is that the statistics are good, bearing out the
relatively low scatter that I was seeing before in all those 2-sec
and 4-sec integrations.
Variation with time
We do not know what the feature energy really should be, because it
is model-dependent.
Thus there is every reason to expect variations as a function of time.
Also, of course, the apparent energy of the Fe feature will depend on
the gain calibration, which is subject to variations as it is
maintained on the basis of incoming data.
The plot below shows a set of fits starting with the April 21 event
and going on until May 2003.
The error bars are a bit hokey.
The points refer to the intercept at zero dead time for each fit.
Once I have a robust fitting program this could be automated.
Some points:
The error bars are pretty small for April 21 and one or two other
events.
This seems to depend mainly upon counting statistics, and it is much
worse during decimation level 2 or worse.
There does not seem to be much long-term variation resolved here.
The points with small error bars all agree pretty well on 6.75 keV.
Ignoring error bars completely, the mean is 6.743 keV and the sigma is
0.037 keV.
Things to be done
Look at G8, G6, G4, G3... At first glance there is not much
perceptible difference.
Look at shutter-in data.
I can do this but I want to make sure my fitting is suitable.
Recall that we thought that there might be differences in the way
the electronics handle shutter-out spectra, where the peak in the
pulse-height spectrum is below the fast analysis threshold, and
shutter-in data where it is not.
(morbid interest) Look at some shutter-out data with huge deadtimes
during major flares.
Hugh Hudson, 13-jul-2003