From: Serena Nobili (serena@physto.se)
Date: Fri Dec 12 2003 - 08:29:59 PST
Hi Rob
thank you very much for your e-mail. It was especially good to read you
"positive" comments. I agree with your main doubts. (1) The disagreement
we get on the K-corrections should be understood better, (2) the
lightcurve fit of the high-z SNe seems to be a "hot" topic. However I
have
confidence in the tests we made and I don't think we can improve this in
any way given the data we have at the moment.
I try to answer very quickly to each of your comments here:
On Thu, 11 Dec 2003, Robert A. Knop Jr. wrote:
> I think that Serena's paper is ready to go out to the collaboration.
> By
> and large, I think it's a good piece of work. It's too bad that the
> grey dust conclusion is "we can't conclude anything", but I like to see
> this out there. Two big worries:
>
> (1) The punt on the difference between Serena's K-correction and the
> Tonry/Riess K-correction. Puting is probably the right thing to
> do, but this will raise huge red flags and objections from referees
> and from our competitors.
>
In this point, I think it would be nice if you could check the values
for
the K-corrections you would get using your own templates. This could
give
us more - or less - confidence in one of the directions (they are doing
it
wrong, we are doing it wrong). I put the various J filters in
attachment.
> (2) The lightcurve fitting method for high-z supernovae that Chris is
> debating. See my comments below. I think that the method used is
> probably fine as it is, but that a lot more supporting
> documentation and cross checks are needed (many of which I suspect
> Serena has already done, but just hasn't written into the paper).
>
> In particular, I'd like to see the answer to two questions. (1)
> How much different is the fit I-band magnitude when a different
> low-z supernova's template are used for a given high-z supernova?
> Perhaps some statistics and a plot showing the dispersion you get
> when you use different templates.
Check out this plot (linked in the iband web page):
http://www.physto.se/~snova/internal/papers/iband/figs/read_fitlog.ps
it shows the Imax resulted from the fit to the 42 templates vs. the
chisq
of the fit. The bottom panel is limited in the region chi2_min+3, (the
vertical dashed lines are chi2_min+1 and chi2_min+2. For both 2000fr
(Beethoven) and 1999ff the second best fit template gives a chi2 that is
greater than chi2_min+1. For 1999Q only 1 template is within the
chi2_min+1 value. What I do (I hope this is clear) is to use the results
from the fit within chi2_min+3 to estimate the systematic uncertainty on
my fitting procedure. I believe this is a very conservative approach.
Note
that the systematics is very small for Beethoven (only 0.04), and quite
large for the other 2 SNe.
I am trying to fit 2-to-4 data points here, I don't think anything
"robust" can be done about it, apart from estimating the uncertainties
introduced by the low statistics in the fitting method.
Now, the question whether this should go into the paper or no. I don't
think this kind of plots tell us more than it is said in the text, on
how
the systematic uncertainty coming from the fitting procedure is
calculated. But if you have a different opinion, or suggestions about
what kind of plots or discussion could convince people more about this
fit, I would be happy to discuss it with you.
> (2) Do the templates chosen
> match the stretch of the high-redshift supernova where that
> information is available? I understand you can't *choose*
> templates based on this, as one high-z supernova doesn't have a
> stretch availble, but you could comment on how well the stretch
> matches. Given that there is a correlation between stretch and
> time of second peak, it is a major warning sign if the stretch of
> the low-z supernova used as a template doesn't match the B stretch
> of a high-z supernova.
>
The same SN 1992bc template (sB=1.076) fits both 2000fr (sB=1.06) and
1999ff (sB=0.82). I am afraid this is not supporting us.
> This one thing-- more documentation on consistency and variation
> with templates-- might be worth putting in before the paper goes
> out to the whole collaboration. If you want to go on to the next
> step now, I won't object, but I think that the next step might go
> more smootly if that analysis is put into the paper before the
> paper is released to the ravaging hoards of collaborators.
>
> General comments on the paper:
>
> * Section 2.2: Is there any correlation between the time offset from
> Bmax to Imax and Bmax to t2? (I probably asked this before... If
> there is no correlation, mention that.) (Does the last sentence of
> section 2 refer to this?)
>
You mean, if I tried to plot t1 vs t2, or something like that, right?
If I understand right your question, you picked probably the only
correlation I did not try to look at.
This plot:
http://www.physto.se/~snova/internal/papers/iband/figs/select.comp_ib.ps
shows some correlation I have tried instead. I can look at t1 vs t2 and
let you know.
> * Section 2.3: I think this could be clarified a bit further. Before
> the sentence starting "However, on a total of 42 supernovae, 2
> cases..." add the following sentence: "This lends confidence that
> the fitting procedure is robust, and given the model of the
> lightcurve template will not yeild biased estimations of the
> parmeters."
>
done.
> * Section 3, second paragraph: what was alpha_I? I think you may have
> mentioned it earlier (1.19 if memory serves), but it woudl be worht
> putting it in again here and referring back to the earlier section.
> Did you use an error on alpha_I? Or was alpha_I a fit parameter?
>
yes, alphaI = 1.13 +- 0.19
I add this, as you suggested.
> * Section 3, top of page 8: mentioning the RMS is 0.19 is, I believe,
> redundant with putting in 0.19 as the scatter; didn't the latter come
> from the former prveiously?
>
Yes, true. I am working this out. It is a bit tricky because I fit the
alphaI in the previous section and then I already know the intrinsic
dispersion to be add in the uncertainty in the plot. I have to find the
way for it not to look redundant.
> * Figure 10 is pretty
Thanks!
> * End of section 4.1, where does this 0.05 magnitude estimate come
> from?
>
Chris. See his email.
> * Figure 11 : if it doesn't make the figure too complicated, it might
> be nice to add a supernova spectrum at some relevant redshift here,
> pointing out the Ca triplet feature. This might help lend confidence
> in the K-corrections for those who worry about it.
>
I will try to do it.
> * Section 4.4 : did you allow for any intrinsic dispersion in the date
> of I max relative to B max, given that earlier in the paper you
> report that low-redshift supernovae show such a thing? I'd like to
> see more details on the fitting method-- explicitly stating what was
> not done or what was assumed, rather than leaving it unstated.
>
The details are very simple: I only fit the maximum value: 1 parameter.
I
don't allow for any dispersion for the date. The main reason for that is
that 1999ff has only 2 data points: I cannot do better than a 1
parameter fit.
> * Section 4.5: "randomlly distributed around the data points". Does
> this mean at the time of the data points, and starting with the data
> fluxes distributed by a gaussian with a sigma of the data errors?
> Please be more explicit about this.
>
Yes. It means exactly what you are saying.
> It might be worth reporting more results of this. E.g., given that
> you've fit all high-redshift supernovae with all 42 low-redshift
> templates, what is the typical dispersion in Imax you get from the
> different fits? I.e., how robust is your parameter estimation to
> choice of template? It might be worth discussing this in some depth,
> and perhaps shoring it up with a plot or two.
>
> Also, comment on how similar the low-z supernova you chose for a
> template is to the high-z supernova based on stretch and spectrum,
> where each is avilable. Especially given Figure 8 (stretch is
> correlated with time delay to second peak), use of the stretch ought
> to allow you to constrain which low-redshift supernovae are
> reasonable for selecting a template for the high-redshift supernovae.
>
I think I already answered to these comments (see above).
> * I'm very happy to see you taking Riess to task for not publishing his
> B-band data. Just because I'm nasty and mean and tired of him taking
> us to task for that same thing.
>
> * Consider doing Figure 15 in color.
>
> * Section 5 : "possible systematic differeences cannot be excluded and
> could result in measuring an artifically large dispersion". Might
> this be worth some more analysis/discussion?
>
Actually, I think I am going to take away this sentence, since it is a
repetition of something already discussed at th end of section 3.
> * Table 12 : clarify the first column:
>
> no dust, Omega_M=0.3, Omega_Lambda=0.7
> dust Rv=9.5, Omega_M=1, Omega_Lambda=0
> dust Rv=4.5, Oemga_M=1, Omega_Lambda=0
>
> (I belive that's what you're saying!) This will make the table more
> immediately transparent.
>
ok.
> * Last sentence of paragraph 6.0: "These results do not allow to reach
> any definitive conclusions". (I think there's an "us" missing.)
> Perhaps expand on this a bit? Is there any hope that further
> statistics in rest-frame I-band will allow us to limit grey dust, or
> are the intrinsic uncertainties and systematics big enough that it's
> hopeless at the moment?
>
ok. I will add some discussion.
> * Figure 6.1: the first data point of B-I color lies well above the
> other models. Does it? It only looks like 2-sigma to me in the
> plot, which really isn't all that huge. I do agree that the B-I
> colors appear to have correlations (too red early, too blue late)
> compared to the models.
>
It is true that the point is not more than 2 sigmas away, but it still
look too red, and I thought a comment on this was needed. The problem is
of course in the fact that we don't have enough data.
For the correlation you talk about, it is not real. If you look at the
single SN you realize that the impression of the correlation is given by
the first epoch of 99ff (being too red). If you exclude it, you don't
see
the correlation anymore.
> -Rob
>
> --
> --Prof. Robert Knop
> Department of Physics & Astronomy, Vanderbilt University
> robert.a.knop@vanderbilt.edu
I will be ready with a new draft by Monday.
Thanks again for your comments.
Cheers
Serena
----------------------------------------------------------------------
www.physto.se/~serena
Tel +46 8 55378661
Give free food at:
http://www.porloschicos.com/
http://www.thehungersite.com/
This archive was generated by hypermail 2.1.4 : Mon Dec 15 2003 - 17:09:46 PST