From: Chris Lidman (clidman@eso.org)
Date: Fri Feb 25 2005 - 17:28:54 PST
Dear Greg et al.,
Serena, Tony, Saul and I had a meeting to discuss the points you
raise. For the sake of brevity, I will discuss only those
issues that are still open.
Cheers, Chris
1) Are we underselling the result in the I-band Hubble diagram.
Greg's comment.
Yes, I am arguing that we are underselling our data and
results despite the fact that the work in the paper has
been done carefully and cautiously. The significance of these
results would be more obvious if Table 7 listed chi^2 and
chi^2 - chi^2_of_concordance_model, rather than chi^2/dof.
Table 7 modified to include delta chi^2:
fit a fit b fit c fit
dd
(OM,OL) chi^2 del_chi^2 chi^2 del_chi^2 chi^2 del_chi^2
chi^2 del_chi^2
-----------------------------------------------------------------------------------
(0.25,0.75) 2.18 0.00 2.36 0.00 1.16 0.00
1.44 0.00
(1.00,0.00) 7.56 5.38 8.44 6.08 15.56 14.40
18.30 16.86
(1.00,0.00)_dust 3.52 1.34 3.96 1.60 5.36 4.20
6.48 5.04
a, b - no stretch correction with and without systematic uncertainties
c, d - stretch correction with and without systematic uncertainties
For 2 DOF, (1.00,0.00) is ruled out a high confidence for cases c
and d. (1.00,0.00)_dust is disfavored at more than 90% confidence
for case d (my favored case).
So, I'd like something less wimpy such as: "These new I-band
measurements are in excellent agreement with the concordance
model, while even this small sample eliminates a pure
Einstein-de Sitter model. Even an Einstein-de Sitter model
with dimming due to grey dust is disfavored at roughly 90\%
confidence."
The words "suggests" and "disfavored" are sufficiently cautious,
but this way it sounds like we do have some interesting results.
(If I see "not statistically significant" in an abstract I'm
unlikely to read the paper unless it is said in the context of
disproving a previous (spurious) result.)
This text could be placed at the end of the last full paragraph on
page 12 and then be reinforced in the bold paragraph at the start
of page 17.
Figure 11 could show cases "b" and "d"
The first full paragraph of on page 17 would have to be
strengthened too.
In the abstract "... consistent with ..." would become
"... in excellent agreement with ..."
"However, the uncertainties ... drawn." would be
deleted.
This is an hour at most to implement, once approved, and another hour to
double-check and correct for spelling and TeX errors.
Conclusion from phone conference: This needs more study as there is a
clear disagreement here. Serena asserts that although the concordance
model has a smaller chi-sq. value, it is not statistically
significant. Note that this point related to point 3.
2) Did the calculation that led to the ellipses drawn in figure 13
include
or exclude the measurement error?
Conclusion from the phone conference. The ellipses include the
measurement error.
3) Colour correcting the high z SNe.
Greg: Yes, but what is the point of testing for grey dust if you know
you
haven't corrrected for regular dust? If one corrects for dust
using
B-V, can't one still test for grey dust using B-I? If there were
grey dust, E(B-V) would underestimate the correction needed in the
I-band.
By not correcting for dust, we are underselling the best part of
the
paper (IMHO) - the I-band Hubble diagram.
Conclusion from the phone conference: At the meeting, we propose to
add a sentence indicating how the points would move if an extinction
correction was made. Before I try to do this, I would like us to
consider, for arguments sake, two hypotheses.
Hypothesis 1) There is no grey dust.
Correcting for host extinction is then strightforward.
In fit c, the high z SNe would, on average, be slightly closer to the
concordance model if an extinction correction was applied. SN 1999ff
would move closer and SN 2000fr would move away. However, the
uncertainties would increase.
Hypothesis 2) There is grey dust which dims SNe in the B-band by 0.5
magnitudes at z~0.5.
Correcting for host galaxy extinction is more complex.
To work out the extinction by the host, one would have to remove the
extinction caused by the grey dust first. Here one would have to
assume an R_V for the grey dust. Serena, what was the R_V used for the
dust model that is plotted in the Hubble diagram? This does not seem
to be in the paper.
Let's assume R_V=9.5. The correction to the excess B-V colour is 0.06
magnitudes (I've taken this from figure 13). Note that there is a
slight redshift dependence that I have ignored here. Hence, the
excess B-V colour of SN 1999ff becomes ~0.08 and the excess B-V colour
of
SN2000fr becomes ~ -0.16.
If one then corrects for host galaxy extinction and applies it to both
high z SNe, even though the excess colour of SN 2000fr is now very
negative, SN 1999fr would move down by 0.14 magnitudes and SN 2000fr
would move up by 0.29 magnitudes.
I imagine that the reduced chi-sq. for the dust model would now be
considerably worse, mainly because of the large correction that
is applied to SN 2000fr, whereas the reduced chi-sq for the
non dust model probably be slightly better.
So, one would get a better result by correcting for host galaxy
extinction. At this stage of the paper, I am against this, mainly
because one could argue that we are choosing the analysis that gives
the best result - I just read Alex's wonderfully written draft. Can we
capsulate some of this in a single sentence. In a single sentence, no.
I will try to do it in a few.
"The restframe I-band magnitudes of SN 2000fr and SN 1999ff can be
corrected for host galaxy extinction by examining the colour excesses
(See figure 13). For the no grey dust case, this is a simple
correction, and would result in a better agreement between the data
and the concordance model. The correction in a universe with grey dust
is more complex as the colour excesses would have to be corrected for
the redenning that is caused by the grey dust before host galaxy
extinction is estimated and such a correction is model dependent. For
the R_V=9.5 case, the agreement between the data and the model would
become worse."
... or something along those lines.
4) the bottom of page 5 you indicate that extinction values
from Phillips 1999 were used. Do these already have the Bayesian
prior? Are Phillips fiducial colors the same as yours? It seems that
either of these could be problems with the analysis later in the
paper.
Concl: I left the meeting at this stage, so I do not
know the conclusion here.
This archive was generated by hypermail 2.1.4 : Fri Feb 25 2005 - 17:29:39 PST