-----------------------------
Referee Report

The use of Type Ia Supernovae for cosmology has produced remarkable
results. The checking of these results is of utmost importance and the
paper 'Restframe I-band Hubble diagram for type Ia supernovae up to z
sim~0.5' by Nobili et al. is trying to present such a check.

The paper has two main topics: The necessary establishment of the
I-band light curves of Type Ia Supernovae and the construction of an
I-band Hubble diagram. The former is a precondition for the latter. The
paper has to remain exploratory, since there are not enough data to
populate the Hubble diagram at redshifts z>0.3 where the cosmological
effects start to become significant.

There are several shortcomings with this paper and it should be modified
considerably before it can be recommended for publication. However,
the paper can be brought to a stage where it can be published with some
effort.

RRR) The light curve fitting is rather ad hoc. The motivation to use B-band
light curves to construct a surrogate I-band light curve is not
explained anywhere. The authors should give a reason, why they chose
this route and not another. There are obvious shortcomings from this
approach that are not described in the paper. E.g. the late decline of
the I band, which is typically faster than in B can not be described
with this method at all. This means that after about 40 days past
maximum, the I light curves can not be fit any longer. The paper does
not use these late-phase data, but it should be made clear that the
proposed model can only fit the maximum phase. By the way, the approach
by Contardo et al. allowed them to fit the light curves, within the
stated limitations, over a much longer time than what is proposed here.


AAA) As correctly pointed out by the referee, the choice of the B-band
template is arbitrary. The B-band template has been chosen only by the
fact to be able to describe the overall data, and that has been
already published (Nugent et al.). Different templates could be used
(such as a different band or any other functions), that could probably
work as well or better than the one we decided to use. We have
modified the text of Section 2 to point out the insignificance of the
template chosen. Also following the suggestions of the referee, we
have pointed out the limitation on epoch range of the method, as
opposed to the one of Contardo et al., developed with the aim of
fitting only data up to day +40, by rephrasing part of the text in the
end of Section 2.





RRR) Another issue are the rise times. The paper states that it is 'assumed
that the rising part of the light curve in I-band is the same as in
B-band' to then on the next page say 'the trend in the residuals' in the
fit to SN 1994D 'shows the limitations of the model.' An inspection of
Figure 1 shows that this is evident in all cases where data in the rise
is available. In all cases the fit has strong systematic deviations,
when the peak is defined through pre- and post-maximum data. SN 1994D is
just the case in point. It seems important to make the paper consistent
in this respect.
The above draws the result of Figure 2 into question. This figure is
remarkably different from what was found in Contardo et al. (their
Figure 4). This discrepancy should be discussed in the paper.

AAA) Only six supernovae, less than half of the SNe of our sample with
data that cover the rising part of the light curve, are not well
fitted by the model and present a systematic trend in the residuals of
the fit. We already pointed out that this problem does not affect the
Hubble diagram. However, the referee is concerned about the possible
consequences on the fitted time of maximum, especially due to the
disagreement of our Figure 2 with Figure 4 of Contardo et al. We note
that if the fit is forced through the pre-maximum data points, the
date of maximum would be shifted at slightly earlier times.  We
compared the histogram of figure 2 with the histogram obtained by
selecting the sub-sample of 17 SNe that we have in common with
Contardo et al. (out of 22 SNe). We note that only 4 of the 6 SNe
under discussion are in common. The comparison histogram (see attached
file) show how the sub-sample used by Contardo et al., shown in red
shaded regions, is more spread over broad values of the time of
maximum and generally before B-maximum, while the whole sample is more
picked. We conclude that the differences in the two distribution are
only due to statistical fluctuations.





RRR) Another example is the exclusion of two (SNe 1997br and
1998ab; both 'spectroscopically peculiar') supernovae due to the Monte
Carlo fits. They happen to be the two with the smallest ratios between
peak and dip in the light curves. The fact that the Monte Carlo fits
fail to reproduce these light curves consistently is probably an
indication that the model fails here as well.

AAA) We checked whether the shape of the light curve for SN 1997br and
SN 1998ab could be responsible for the few cases in which the MC
simulation failed. However, a close look at the simulated light curves
showed that the real cause of the failure is in the poorly determined
first peak, defined by only one point for SN1998ab and two points for
SN 1997br. The Monte Carlo succesfully reproduces the results in the
majority of the cases. For this reason we decided NOT to exclude these
two SNe from the analysis.





RRR) Table 2 should list the number of data points used in the fit. This
would help the reader to assess the goodness of the fit as well.
Especially, since these light curves are then later used to fit the
distant supernovae, it would be good to restrict the construction of 
light curve 'templates' to only objects that have an over-constrained 
fit, i.e. more than let's say 10 phases spread over at least over the 
two peaks. Establishing a template with 6 data points appears unreliable 
given the model.

AAA) Following what suggested by the referee we have added a column in
Table 2 with the number of degrees of freedom. The number of points
used together with the chisq reported in Table 2, should give
the reader enough information to established the goodness of the fit,
e.g. an over-constrained fit will result in a very small chisq
value. We understand the concern of the referee about the propagation
of the uncertainty due to an over-constrained fit on the nearby SNe
into the high redshift light curve fit, through the use of all 42
templates. We believe the systematic uncertainty on the fitted I_max,
as estimated in Section 4.4, takes into account also this
contribution. Moreover, we note that the template fitting the
high-redshift SNe with lesser points is the one of SN1996bl,
determined by 9 data points.





RRR) Light curves are stretched according to values derived from the B light
curves. What would be the rationale to apply the same stretch in I as
well?

AAA) We infer from this comment that it was not very clear in the text
that the fit of the B-band and I-band light curves are done
separately. The confusion was probably introduced by the use of a
B-band template to fit the I-band one. However, the value of the
B-band stretch does not enter in any way in the I-band light curve
fit. We expect this to be more evident now, after the modification
made to Section 2.






RRR) The fits are done in flux units. Does that mean the magnitudes are
converted into flux, fit and then converted back into magnitudes? Why is
this done like this? How are the errors treated? This is an important
detail that needs to be explained.


AAA) Since the fitting procedure uses symmetric uncertainties, the
fits are done in flux units. The parameter of the fit, i.e. the peak
fluxes, are converted in magnitudes and reported in Table 2. The
uncertainties are treated as gaussian for the fit.








RRR) The authors should decide what they want to do with SN 1999Q. As it is
right now, they introduce the data for this supernova to quickly reject
them as non-consistent. That would be OK, if they then would refrain
from presenting all the analyses as with the other two objects. I
suggest to either accept the published (Riess et al.) data as they are
or, if the data can not be used, reject them and leave it at that. It
does not make sense to fit a light curve, calculate a brightness and put
the point into the Hubble diagram, if the data can not be trusted. The
authors should indicate, why the data can not be trusted. Just saying 'a
re-analysis of the publicly available SofI data suggest that the
published J-band magnitude may be too faint' is not good enough. The
public data contain all the information necessary to derive a reliable
value. If it differs from the published result the difference should be
stated explicitly.


AAA) The importance of this supernova is given by the fact that has
been used by Riess et al. in 2000 for a similar study of intergalactic
dust. Unfortunatelly, the difference in the re-analysed data point
with the published result by Riess et al. is quite large.  Following
the indication of the referee, we now discuss the disagreement found
in Section 4.3 of the paper, and exclude it from the rest of the
analysis.





RRR) In several places the paper is confusing in the way it presents
the samples. The abstract quotes 26 local supernovae, but the reader
is confronted with different numbers (42, 28) in the text. It might be
useful to state the sample selection (redshift range and exclusion of
spectroscopically peculiar supernovae) in the abstract so that this is
clear from the beginning. Maybe the selection criteria could be
summarised again at the beginning of section 3.


AAA) We fit I-band light curves of 42 nearby SNe, and 28 of these are
in the Hubble flow. Two SNe are further excluded from the Hubble
diagram as explained in Section 3, reducing the number of SNe in the
Hubble diagram to 26. Following the suggestion of the referee we added
in the abstract the total number of SNe (42) on which the light curve
fit is performed. We leave at Section 3 the task of explaining to the
reader the exclusion of 2 SNe from the sample in the Hubble diagram.






RRR) The use of references is sometimes unclear and appears also occasionally
arbitrary. It would be good to carefully go through the paper again and
sort these out. Suggestions are made below as well.

AAA) ????






Specific comments:

1. 'spectroscopic peculiar' should be changed to 'spectroscopically
peculiar' throughout the text.

OK

2. add Schmidt et al. (1998) to the reference list to the first sentence
in the introduction.

OK

3. add '(similar to Riess et al. 2000)' as a reference to the first
sentence of the last paragraph of section 1.

OK

4. The references for SN 1991T (section 2.2) are rather strange. It
appears as if there is a mix-up with the following sentence. The
references for SN 1991T should be Filippenko et al. 1991, Ruiz-Lapuente
et al. 1991 and Phillips et al. 1992. The references for Li et al. 2001,
Howell 2001 and Garavini et al. 2004 presumably refer to SNe 1995bd,
1997br, etc.

The sentence starting with 'However, not all of these show ...' is
unclear. Which supernovae are meant with 'these'? SN 1986G, SN 1993H or
others?

OK


5. In one paragraph Type Ia Supernovae are called 'standard candles' and
the next (only two lines down) they are described to show 'a variety of
properties.' It is passages like these that confuse the reader.

OK 


6. What does 'somewhat deviant' mean (caption of figure 6)? This
statement should be quantified.

AAA) Following the suggestion, figure 6 has been changed. The data
have been fitted by a straight line and 4 SNe are identified as more
then 2 standard deviation away from the ensemble.


7. There is a change from observed redshift to a redshift corrected to
the CMB frame between Tables 2 and 3. This should be explicitly stated
either in the text or the captions to Table 3.

OK


8. Table 4 and its discussion in the text are meaningless. Since the
fits in the Hubble diagram is done with all data sets the smallest
sample one will have the least weight and hence largest uncertainty.
With the larger uncertainty the 'offset' automatically can be larger.
Not surprisingly, all samples are consistent with each other within the
errors. This table and the corresponding section should be dropped.

OK

9. The first sentence of the third paragraph in section 5 should start
with 'The low statistics of the high redshifts sample is insufficient
...' No statement can be made whether the concordance model is favoured
or not from the presented data.

OK



10. In the same paragraph there is a garbled sentence, the meaning of
which did not become clear. However, as suggested above the discussion
of SN 1999Q should be dropped here anyway.

OK


11. It appears from the fits that a dust model with R_V=9.5 is the most
favoured case, given the data. For the Monte Carlo simulation to check
for the sample size to reject a dust hypothesis the number of individual
data points per individual supernova must be important as well. This
should be described.

AAA) The number of points for individual supernova is certainly
important, however, the major contribution is given by the
uncertainties on the individual points. In the simplest case of
uncorrelated data, the least squares reduces to the weighted average,
and its uncertainty will scale roughly as the average uncertainty and
the inverse of the square root of the number of points. The main
limitation is thus given by the current typical uncertainty on J-band
measurements, each obtained in about one hour time at a 8-m telescope,
e.g. the case of SN2000fr. Our simulation is based on colour
measurements of 20 SNe, whose final weighted average is known with a
precision of 0.05 mag. Increasing the number of points for each
supernova, however would help keeping systematic uncertainties under
control.

12. The fourth paragraph of section 8 should start: 'J-band measurements
of one high-redshift supernova plus published data of one more were used
to extend ...'
Also, instead of saying 'seemingly inconsistent information' on SN 1999Q
the authors should be more explicit. It is alright to say that the authors do
not trust the data and hence discarded this object.

OK


13. There is an interesting mismatch in the interpretation between
concordance model and the analysis of potential dust in the conclusions.
The 'concordance model ... is found in better agreement with the data
than other models' but the dust analysis yielded 'no firm limits on the
presence of grey dust.'

AAA) ???