Tamuz-Mazeh-Zucker paper

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

 Michael Sallman <msallman@pro-ns.net> wrote
>Date: Fri, 04 Feb 2005 17:24:17 -0600
>
>Thought this might be of interest:
>http://arxiv.org/astro-ph/0502056
>
>Mike

I was less than enthusiastic when I first read this
paper but - what the hell - I dug out my old ensemble
code and programmed it in.

As presented, their algorithm is for correcting
multiple shots of exactly the same piece of sky
with every star appearing on every image. With
some trepidation, I tried it on the set of MkIV
archive data that I downloaded a long time ago
which, of course, bears no resemblance to these
requirements. Some selected results for
final processed standard deviation for the
16x16 degree square in V (I is similar)

0.0586 mag Raw
0.0329 my old code: 2nd order spatial
0.0277 2nd order spatial + colour term
0.0212 2nd order spatial, 3rd order TMZ
0.0199 2nd ord spatial, 3rd TMZ + colour

So the T-M-Z algorithm does chip away at the
standard deviation. It does burn away rather a
lot of degrees of freedom too. Starting from
1869487 raw data points on 1322 images
with 60322 stars, T-M-Z uses up
61644 df per order. By the end, I am down to
something like 1614979 degrees of freedom ...

The T-M-Z fitted coefficients are wildly
non-Gaussian, just like TASS errors. A few
spot checks showed that at least one of the
large resulting corrections pulled a wild
point back into place! Don't get too
excited: it isn't consistent. As T-M-Z remark
on p2, one should be using the SVD technique
which is easier to say than to do (SVD works
by comparing the different coefficients which
are fitted simultaneously: T-M-Z fits each
coefficient independently. It's easy enough to
compare 11 things with each other but rather hard
to compare just one ...)  but might tame some of 
the excesses.

T-M-Z originated as an extension of
colour correction provided every star was seen on
every image. I haven't the faintest idea what it
is really doing to the totally different data set to
which I am applying it.

Andrew Bennett, Avondale Vineyard, NS, Canada