From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Thu Mar 20 2003 - 05:36:05 PST
(I haven't sent this to deepnews because I really don't know who's on
that list any more, and I have to admit I don't feel fully comfortable
discussing this paper on that list.)
Here's the basic problem: why are my results coming out significantly
different from the old results in places? (Note that the color
differences really aren't significant individually, but in total I agree
with Greg that there's a trend which is significant.)
Note that if you compare Mr and R-I you don't see nearly as big a
difference, nor do you see as big a trend. Therefore, the differences
are primarily because of the K-corrections (and the assumed "standar"
colors used in generating E(B-V) -- these are entangled with the
K-corrections). There are only small differences due to the new
template, and only a few SNe suffer from my not having started snminuit
at the right spot. (I'm basing this also on my refit using the old fit
method but new K-corrections, the results of which I haven't put out yet
as it's not fully done.)
I did my K-corrections using Peter's new method, but using his
uberspectrum modified to reproduce the colors I measured (BVRI as a
function of t and s) from the H96 and R99 paper. These K-corrections
are giving different results than the spline files we used in the 42SNe
paper. *Peter's* uberspectrum gives different results from the spline
files we used in the 42SNe paper.
The result is different supernova fit parameters, and different
cosmologies.
Now, we might say, well, to avoid having to move our confidence limits
by more than the previously quoted K-correction undertainty, *and* to be
able to use K-corrections which we have believed before, we should *just
use* the old K-corrections and be done with it. I grant you that's an
appealing approach from the point of view of getting it done. The
problem is, it's impossible. We have the old K-corrections for the
Bessel filters, and we can correct the ground-based filters to the
Bessel filters using linear color terms (and, indeed, it basically
doesn't matter if we don't apply those terms, they're so small).
However, the HST filter is enough different that we really have to worry
about the K-corrections from those filters, or (equivalently) the color
corrections to the Bessell filters. What's more, I don't to "just
trust" Dolphin's transformations to Bessel filters, as those are based
on obsevations of stars. We need something like an uberspectrum to
generate the HST K-corrections. (Or, equivalently, to generate color
corrections to Bessell filters, and then K-corrections between Bessell
filters. Both have to be done with the same method for consistency.)
And, then, for consistency, we should do the other supernovae with the
same K-corrections.
Note that I'm also doing K-corrections more carefully than
before... previously, we had assumed that K was only a function of
t', where t'=t*(1+z)*s. Now, I'm doing K-corrections which takes into
account the change in the spectrum due to host and Milky Way reddening,
*and* the different colors as a function of stretch. (What I'm not
doing is propogating uncertainties on all of these things into the
K-correction.)
What do we do? There are four ways to proceed, as I see it:
(1) Just use the old K-corrections and templates, meaning the old
analysis for P99, and the standard transformations for the HST
filters to Bessel filters ignoring that stars and SNe don't have
the same spectrum. This is really not quite right and will
introduce systematic biases to the HST supernova fits and colors
and hence the cosmology. (Indeed, doing it this way a long time
ago was what originally gave us the "too blue" problem for HST
supernovae... although there I was comparing the colors to the
colors from Peter's uberspectrum, so it wasn't even a
self-consistent analysis, I now know. However, the fact that this
came up is the reason this paper wasn't published a year and a
half ago.)
(2) I could make an uberspectrum designed to reproduce the
K-corrections from the old paper, and use that to see what happens
when we fit the old data. I predict it will be a lot closer to
our old results, though not exactly the same if I still try to
adapt the K-corrections for stretch color and extinction. This
will take me several weeks to complete at least, *and* I believe
it won't be using the "right" supernova colors.
(3) We can move forward and say that we know the K-corrections have
changed and that we think we underestimtaed the size of that
systematic in the previous paper.
(4) We can decide that we don't trust what I did and redo absolutely
everything right down to the beginning, and keep repeating this
exercise until we stumble upon a way of doing more modern
K-corrections which lands the cosmology within the P99 quoted
systematic. (Or, to be less caustic and cynical, at least we could
redo it enough times so that it's self-consistent and we no longer
have doubts.) For example, instead of my "ridgeline" fits we
could trust the Phillips E(B-V) values (picking your favorite one
from the many he gives for each supernova) and I could make an
uberspectrum in a manner similar to the way Serena did it, and see
how things look. This is the "publish the February paper in
November" approach, but we have had success with publishing the
February paper in November in the past, and perhaps that's what we
should do again.
-Rob
-- --Prof. Robert Knop Department of Physics & Astronomy, Vanderbilt University robert.a.knop@vanderbilt.edu
This archive was generated by hypermail 2.1.4 : Thu Mar 20 2003 - 05:36:17 PST