From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Wed May 14 2003 - 07:01:33 PDT
On Tue, May 13, 2003 at 11:15:03AM -0700, Greg Aldering wrote:
> Let me start by discussing the new analysis with R_B = 2.X. You of
> course realize that the minimization I performed is valid only if all
> the other steps - the photometry, K-corrections, fiducial color vs
> stretch - are correct. While they may be, I believe the read may
> equally well take our modified R_B as a signal that those other steps
> are causing systematic errors that force a low R_B. The fact that
> a change in R_B covaries with alpha (as can be seen in looking at your
> fits with different R_B's) also means that the minimization that I
> performed is only very approximate. It also makes me concerned about
> the fiducial color vs stretch. We have not yet seen your color vs
> stretch relation - it seems prudent to show that relation (internally)
> so it can be compared with the relation used in P99. It would also
> help to see that relation with data only from E/S0 hosts (which is
> how the P99 relation was set).
The way I did the fiducial colors, it won't be possible to show it just
for specific hosts. The whole thing was fit to the entire lightcurve as
described in the paper; the peak fiducial relationship is just what the
fit happens to return at day 0. (To do it just with E/S0 hosts would
mean redoing the whole process.) Note that I *did* omit systematically
red supenovae from the fit, even with the asymmetric error bars,
although I did that by looking at the supernova color curves and not by
looking at the host types.
The expression I'm using is:
B-V = -0.076 + 0.0446 * (1/s - 1)
I think the numbers you used in P99 were -0.11 and 0.243 respectively.
This is a redder central color and a milder slope, but looking at the
fits I couldn't justify the bluer color, and the slope wasn't as clear
as all that. (Indeed, the uncertainty on the slope is 0.037, nearly the
full size of the slope.)
> a) Follow Phillips 1999 in throwing out very red objects even though
> you are going to extinction correct. I believe that Phillips
> threw-out objects with Bmax-Vmax > 0.2 mag. (Talk about dirty
> laundry!)
>
> b) Use the Phillips 1999 R_B, and note that that value is consistent
> with both our low- and high-redshift SNe.
I'm not sure (b) is really true-- if we through out the most reddened
objects, then technically it's true, but the most reddened objects is
the only place where the signal overcomes the error bars.
However, I will take this approach-- I agree that it's the easier way to
get through it than trying to justify a whole new R_B. (We can save
that for a separate paper, which will show that the new R_B doesn't
destroy the results.)
> Next, on a related front, there are a few unsettling changes wrt to
> earlier draft. Whereas in the early version OM was lower by 0.02 for
> the extinction-corrected HST sample *relative to* the uncorrected HST
> sample, it is now relatively 0.04 higher for the HST
> extinction-corrected sample (see Table 8).
This is the effect of that 7% photometry error, which tended to make the
HST supernovae redder. The reason is that the lightcurve didn't just
purely scale up; there was also a shift in the peak of the lightcurve as
the points moved about, plus the R-band tended to have more of a
contribution from the ground-basd points than did the I-band
> In addition, in Table 7 one now sees that different SN subset have
> become redder by different amounts following the change in the
> fiducial color by 0.02 mag.
Note that different supernovae are thrown out in the low-extinction set
in this case. Additionally, the HST supernovae went through the
aperture correction change. Finally, it's all complicated by the fact
that we're not just talking a simple shift of the fiducial color at
peak, but of the whole assumed fiducial color curve-- which will affect
K-corrections a little bit.
> (By the way, could you let us know how you performed the shift in the
> fiducial color?)
By redoing the color curves as described in the text of the paper. What
I'd done in the past was generate color curves, use those to make an
uberspectrum and a stretch coloring of the uberspectrum, and then
iterate back (as the uberspectrum is used for K-corrections). What I
did was go back and iterate again with the uberspectrum I'd been using
in the previous paper.
The whole thing is a little messy because of the way I throw out
supernovae. There wasn't a hard cut. The problem was that some
supernovae were fine for much of their lightcurve, but screwy at one
point (e.g. near max). If they were screwy at one point, they could
pull off the color curve there, so I'd throw them out. The goal was to
get the supernovae which were all yielding a consistent blue ridgeline
result. Out of 53 low-redshfit supernovae, I threw out 12 in generating
the B-V color curve. The practical upshot of this was that supernovae
which were systematically red (at least around maximum light) did get
thrown out, in addition to 1992bp (whose B-V color is too blue compared
to everything else at max).
> This has broadened the gap between H96 and HST from 0.013 to 0.024 in
> E(B-V), which is a *change* in the difference by 0.045 mag in A_B.
That's that change in the intrinsic HST colors noted above as a result
of the aperture correction; it's not just the fiducial color shift that
did that. Even with the old fiducial colors, the Hamuy-HST color
difference would have gone up.
-Rob
-- --Prof. Robert Knop Department of Physics & Astronomy, Vanderbilt University robert.a.knop@vanderbilt.edu
This archive was generated by hypermail 2.1.4 : Wed May 14 2003 - 07:01:34 PDT