From: Alexander Conley (AJConley@lbl.gov)
Date: Tue Aug 31 2004 - 13:18:56 PDT
Hi Ariel,
Thank you for the comments. Here are some responses.
> 1) As I understand it, you correct for MW extinction. Have you
> considered
> the idea of using the "known" MW extinction for tests the CMAGIC
> method
> versus the standard analysis? Eg, one could perhaps look for what
> the
> MW dust does to the low-z dispersion in both methods and thus check
> that the lower sensitivity to extinction for CMAGIC holds. Maybe
> this
> could be extended to the best measured high-z SNe.
That is an excellent idea. Thank you Ariel!
I will check this for the low-z SNe right away. I don't expect the
hi-z SNe to
be of much use for this because their E(B-V)MW values are quite low --
most
of the high redshift fields were chosen for their low levels of
E(B-V)MW.
However, there are a number of low-z SN with high amounts of
foreground, MW extinction, and it would be interesting to see if they
have worse residuals
(on average) in m_B than in B_BV0.6. I should make a plot of residual
from
the Hubble line vs. E(B-V)MW for the two methods.
> 2) It does seem a bit awkward that you are very systematic about your
> data cuts
> but yet make "case by case" decisions about what kind of LC-fits you
> use (i.e B&V together or fitted separately). Especially, as you
> point
> out, the "choice" will not work the same way for low and high-z
> fits. I suggest you stick to one strategy for all lightcurves
> and estimate the associated systematic error based on what you would
> have found if you would have taken the "other" track. I am sure you
> have considered this. Why did you decide not to?
Yes, this is awkward, and I'm really not happy about it. I think the
issue is one
of internal consistency. I am asking people to accept that CMAGIC
works, and
if somebody has done this, and read Wang '03, then they will expect
that a single
stretch just wouldn't work for SNe with bumps. So if I proceeded with
the single
stretch approach I would be sending contradictory messages.
Note that this really doesn't affect the CMAG fits, since they depend
on stretch
weakly and date of maximum rather indirectly.
The single stretch fits to most of the bump SNe are absolutely terrible.
It's not just a case of the lightcurve fits looking a little better if
I do it this way --
the maximum magnitudes for some SNe are clearly way off if I force s_B
= s_V.
> 3) The two populations in the high-z residuals seem like a real concern
> to me. If the explanation has indeed to do with host galaxy
> extinction
> (as I think you advocate), shouldn't you be able to see hints of
> that in
> the residual dispersion in CMAG vs standard, i.e from graphs like in
> Fig 31 in your method doc? I guess additional tests could be
> provided
> with simulations as in Commins et al.
Yes, it is of great concern to me as well.
This effect does show up as expected in the matched residuals.
Fig 31 (the correlated residuals plot) doesn't work as well here as one
would like
because the two samples are so large that they really drag the fit
around.
The whole premise of looking at this plot is that you are considering a
single
SN at a time, which doesn't have the power to move the best fit around.
When
you look at a whole sample (Knop or Barris) this isn't true. The thing
that
gets messed up is the distance from the best fit -- you can no longer
expect
the m_B resid to be twice the size of the Bbvzs one becuase the set of
SNe
has dragged the whole fit over.
However, looking at the residuals individually, a pattern does appear.
The m_B residuals for the Knop sample are generally larger than the
Bbvzs sample by a significant amount, which is a possible sign of
extinction.
When the residual is positive (the SNe is too dim), the m_B residual is
roughly
twice the size of the B_BV0.6 residual. There aren't enough negative
residual SNe in the Knop sample to say much about this side.
For the Barris sample, the residuals between the two methods are
essentially
the same, suggesting no extinction. Taken together, these are
consistent with the observed bifurcation of the high-z residuals.
The best I can say at this point is that the story seems to be
consistent with
the hypothesis that the Knop sample suffers from higher mean extinction
than the Barris sample, but I certainly can't rule out other
explanations.
And that scares me. What I really worry about is that it's some
systematic
difference in the photometric reduction. In that case I'm basically
screwed.
To summarize:
i) The high redshift residuals for CMAGIC roughly split into two
samples:
Knop and Barris. The Knop SNe are dimmer by about 0.3 mags, which
is about a 2 sigma effect.
ii) The same is true in m_B, without extinction correction. The
difference
may be slighly larger, but the distribution of residuals is far less
smooth
than in B_BV0.6, so it's hard to say.
iii) If I take the A_B values quoted by Knop '03 and Barris '04 as
correct,
then the Knop sample has significantly higher mean extinction. If I
furthermore apply the given extinction corrections to the sample, the
bifurcation becomes much less significant (~1 sigma).
iv) The residuals from the best fit cosmology for the CMAGIC and maximum
magnitude analysis appear consistent with the hypothesis that the Knop
sample is more extinguished.
If you have any other ideas about how I can test this, I would love to
hear
them.
> 4) Have you considered quantifying the bias from possible
> intergalactic dust?
> If you want, I can help you with synthetic data sets.
What do you have in mind?
Alex
This archive was generated by hypermail 2.1.4 : Tue Aug 31 2004 - 17:24:48 PDT