Data set "Tom" noise floor

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

We had a blizzard last week so I was allowed a
day off pruning vines ...

I have at last found the bug that was garbling my
colour- and magnitude-dependent corrections: the
noise floor immediately dropped from 0.010 mags to
0.008 mags.

That's with a simple one-coefficient linear
correction proportional to V-I. Unfortunately, the
magnitude dependent correction that was supposed to
deal with saturation of bright sources got lost on the
way to finding the bug. When I put it back, the whole
system died claiming that my UPS mains voltage monitor
was doing nasty things to the floating point processor.
Windows wins that round!

I'll have another go the next time it snows - I'll have
that noise floor down to 0.005 to match Michael
Richmond's results. Or else!

For the technically minded: I am running a suite of 3
programs on my star lists.

Program 1 selects the objects which match a finding list,
have no error flags and have both V and I magnitudes. It
performs magnitude corrections using the current
coefficients, if any - there aren't any the first time 
around. It rejects stars with "very large" (adjustable
fiddle!) Welch-Stetson parameters. It sorts as many pairs as 
memory will hold into order of image number and writes the 
magnitude differences from their means plus the magnitude
estimates themselves (needed for the magnitude and colour 
corrections) to files.

Program 2 reads these files, accumulating all the pairs
from one image. A least-squares fit (with elimination of
outliers) is used to update the coefficients.

Programs 1 and 2 can be run iteratively to improve the
correction ... this does appear to work but I have fixed
so many bugs that I will have to run the whole mess again
to put a number on how much iteration improves things.

Program 3 applies the correction, computes new Welch-Stetson
statistics and dumps all the data (not just the clean pairs)
for apparently variable sources.

With the noise floor problem, a number of bright
sources show up as variables when, probably, they
should not. But the majority of selected sources
seem to be real variables! Progress!!

Andrew Bennett, Avondale Vineyard