From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Sat Mar 15 2003 - 07:35:26 PST
All this debate about how one should or should not fit lightcurves leads
to one question:
What is the One True Way to fit lightcurves?
More specifically, here's my worry. As we keep bringing up questions
about lightcurves and what is the very best way to fit them, I could
spend a lot of time dicking around redoing the entire set (which takes
about a day each time it happens) under the latest theory of what's
right. None of this, I'm very sure, will make much difference in the
final results.
As such, in the interests of actually getting the paper published, I'd
like to redo the whole set only *one* more time. Before I do that, I
want to know the way that's going to satisfy everybody so that we don't
get second guessed again.
The stuff Alex has done was very useful, and tells me which SNe need to
have their initial conditions carefully checked. I will do that next
time through.
As for the rest, should I:
* Add an "error floor" of (what value?), so that any errors which come
out *smaller* than that value get replaced by that value. This will
tend to puff up the error bars on the later points.
* Eliminate all points after (say) day 40(1+z) on the low redshift
supernovae. Note that some high redshift supernovae do go later
than that, and even some HST points which have fairly good error
bars.
* What else?
Whatever it is, at some point I am going to ask for help in shouting
down people who insist that it wasn't done right and must be redone yet
again... unless there really is true evidence of a problem (as opposed
to evidence that there is a "slightly more optimized or philosophically
pure" way of doing nit).
Some things I won't do : fit with floating offsets generally (that
majorly screws up some of the HST fits; without final references, they
get sad) ; do a two-step fitting procedure to fit R for stretch and fit
I for color (I know we did that last time, but it's the wrong thing to
do for some supernove (esp. some higher redshift HST SNe), I'm not
convinced it will make that much difference, it's more of a pain to do,
and it's conceptually less clean than just fitting the data all at
once).
A few other notes when it comes to comparing to P99 data: I'm not
surprised there are some systematic offsets. I am not going to spend my
life trying to fully characterize and eliminate them; if that is what
the collaboration thinks is necessary, then that is equivalent to our
having decided not to publish this paper. The mere fact that the
K-corrections are different will give you different values. Also, I
suspect that the K-corrections are *more* different some time after max
than they are at max, which will tend to give you systematically
different stretches. How to diagnose if this is true?
* Look for correlations between stretch residual (i.e. my stretch
minues P99 stretch) and redshift. You might see something if the
residuals are due to K-corrections.
* Harder: You can compare the K-corrections I have to the old
K-corrections. (You can generate the old K-corrections by fitting
via an old version of snmniuit WITHOUT the "kcorr=" parameters, and
looking at the kcorr_1 and kcorr_2 files that get written to the
directory.) Some will be quite different; I've only looked at one,
and it was really quite different as you approached the knee. This
is not surprising; check out this document I wrote a long time ago:
http://brahms.phy.vanderbilt.edu/~rknop/scp/fidcolors.ps
This is very out of date, but even then I was seeing colors in the
Nugent uberspectrum I thought systematically wrong in comparison to
what I was getting from looking at H96 and R99 low-z supernovae.
I'm not planning on doing either of these, but if people are worried
about it, then at least look at the first one.
-Rob
-- --Prof. Robert Knop Department of Physics & Astronomy, Vanderbilt University robert.a.knop@vanderbilt.edu
This archive was generated by hypermail 2.1.4 : Sat Mar 15 2003 - 07:35:48 PST