Status report - Andrew Bennett 2002Feb

[Andrew Bennett <andrew.bennett@ns.sympatico.ca>]

I am still beavering away at the PSF fit/Ensemble photometry
stuff.

I'm glad it's called a "Status Report" and not a "Progress
Report" as most of the progress has been, at best, lateral.

The PSF fit program reached its "final" form and the code
got posted on my web site. This presumption was duly
punished: when I reran Data Set 20 yet one more time,
the same code came up, for some images, with completely
different fits starting from different initial guesses.
This is a known problem in this sort of optimisation but
this is the first time I have seen it in this investigation.
The problem, I think, is that I am now using the sum of
3 elliptical components in the PSF: this gives fit residuals
as low as 5% which is a significant improvement over 1 or 2
(see various posts in the archives.) The three components
ara allowed to change in size, shape and relative position
across the image and the two different "optimum" solutions
differ wildly in the relative motion ascribed to the
components. In other words, the program is labelling
components differently across the image in the different
cases and having started off with an incorrect labelling
is unable to correct this as the iteration "converges."
Back to the drawing board.

The reason for rerunning Dat Set 20 was to accommodate
"a few minor changes" in the ensemble photometry. What
was supposed to be an entirely trivial change was to
carry the Tycho2 designation of identified sources right
through the process in a human-readable form: the previous
version boiled the ID down to a single integer which was
unique for that processing run but an absolute pain to
turn back into the full Tycho2 identifier. While putting
in this simple change, it looked as though I could easily
stick in a unique identifier for the sources added during
the processing. Things escalated from there ... it's
nearly working but along the way I found some new bugs
both in the PSF fitting code and in the Ensemble Photometry.

As I said: progress was mostly lateral.

Andrew Bennett, Avondale Vineyard