Analysis of RHESSI Variance
As a test of the models fitted to RHESSI countrate profiles using
hsi_unpix_fit.pro, we have made histograms of the countrates binned
according to the model counts and compared them with histograms based
on Poisson statistics. In this comparison if the model predicts a
value of c (within a range of deltac), then all the time bins (=1 ms)
where the model has that value should form a Poisson distribution with
mean c.
The model countrates (shown in red) and the observed countrates are
shown below. (countrates in ms-1)
For detectors 6-9 and countrates 0.25-0.85 ms-1, the histograms are
shown below. The predicted Poisson distribution with
mean equal to the model mean FMEAN is shown in red. The histograms are
normalized to unity at countrates of 0.
For detectors 3-6 and countrates of 0.25-0.85 ms-1,
the histograms are
shown below. The predicted Poisson distribution with
mean equal to the model mean FMEAN is shown in red.
The good agreement between the predicted Poisson distributions and
the histograms of observed counts indicates that
(a) the model is a good fit and
(b) the countrates are Poisson distributed.
Additional tests were made by computing a simulated Poisson-distributed
time series using the model countrates. This was done 15 times for each
detector (3-9), so that the range of scatter could be seen in histograms.
The seven plots below show histograms of the observed and simulated count rates
with integer counts of 0 to 8.
In general, the histograms from the
simulated counts (red) mimic the histograms of the observed counts (black),
but there are notable differences:
- For all detectors there are more 0s than 1s, but the simulated counts
have a higher proportion of 1s.
- Higher counts (>3) are more frequent in the observed time series than the simulated series..
- The differences become greater going into the tails of the distributions.
- The differences between the histograms are least for detector 5
The differences are not large enough to affect the derived parameters
(such as reative amplitude) by more than a few percent. Possible
origins of the differences are:
- use of the fundamental without harmonics
- errors in calibration
- asymmetries in the sources
- time variability
Numerical experiments with the use of higher harmonics seem to give
very nearly the same chisquare values for the fits, and the same
results for the histograms described above.
Ed Schmahl
Last modified: Mon Jun 10 16:39:23 EDT 2002