From: Chris Lidman (clidman@eso.org)
Date: Tue Jan 27 2004 - 00:38:26 PST
Dear Colleagues,
Please find attached the minutes to Monday's meeting. Please check
what has been written and report any errors or omissions.
Cheers, Chris.
Minutes of the SCP meeting discussing Gaston's EW paper
26/01/2004.
1) The systematic error related to the region over which
features are measured.
This concerns the first paragraph of section 3. CLi thought that
it was the entire region that was shifted by l/4 and not the regions
that are used to define the upper and lower limits of the feature.
After reading through the paragraph again, the meaning is clear.
However, the following suggestions were made:
- Add a clause which indicates the typical magnitude of the
systematic error. During the discussion, GFo mentioned that this
error is typically 1%.
- In the footnote use l/sqrt(12) and remove l/4, there is no need
to approximate.
2) The error in the mean.
The concerns column 3 in table 5 and other columns in this and other
tables where this error is reported.
After some discussion, it was agreed that we leave the reporting of
this error as it is, but that we edit the footnote to the table to make
it clear how this error is derived. Additionally the text of the
footnote to the caption should be added and the footnote removed.
3) Plots showing the correlation between absolute magnitude and
a spectral indicator which is made up of one or more EWs that have been
derived from a single spectrum.
It was agreed that we should have one or more plots showing some
examples. The suggestion of showing several correlations in a multi-panel
plot was made.
We should try to compute the probability of getting a correlation as good as
the one we find for alpha(2+3) if the true dispersion about the fit is more
similar to the dispersion seen in other plots. I attach a description of
how this might be done as an addendum to these minutes.
4) A plot showing how alpha(2+3) is derived.
It was agreed that this would compliment the description in
section 5.1.3. We did not discuss whether all SNe or a representative sample
of those that are used in figure 12 should be plotted. I'd suggest that
we first include all SNe that are in figure 12 and then reduce that number
if we think that the plot looks too busy.
5) The paragraph in section 5.1.3 concerning the effect of host
contamination on the computation of alpha(2+3) and figure 13
are not clear.
The following changes were suggested:
- Remove figure 13 as this applies to a spectrum with a particular alpha(2+3)
- Rephrase the paragraph so that it is clear that magnitude error induced
by incorrectly removing the host is a function of the SNe subtype and that
for Branch normal SNe and error of x magnitudes occurs for y% of
contamination.
- Mention that a parameter like alpha(2+3) is less likely to be affected
by poor host galaxy subtraction than parameters that are based on the sum
of one or more EWs.
6) Separate section concerning the applicability of these parameters to
high redshift SNe.
The following changes were suggested:
- In order to emphasise the applicability of these parameters to high z
studies, the final paragraphs in section 5.1.2 and section 5.1.3 should
be moved to their own section.
- Mention Bruch explicitly
- Provide an example for a feature which requires only one spectrum.
7) Delta m15 and alpha(2+3)
There was a discussion as to how close Delta m15 and alpha(2+3) are
related. It was agreed that some additional plots should be produced.
These plots are not for the paper, but for the collaboration to use
as a staring point for further discussion.
- Delta m15 versus alpha(2+3)
- Delta m15 versus M_B - It was noted that the method that is used
to compute Delta_m15 in this paper is different to that used in the
Phillips' 99 paper.
8) Linear fit to figure 7.
As in other fits a spline fit was suggested.
9) An additional point from LWa.
Increase the size of figure 1.
Addendum
========
I'd like to expand upon a point discussed during Monday's discussion.
Saul mentioned that it would be good to estimate the probably of
getting a correlation as good as the one we get for alpha(2+3) because
of the limited sample size and because of the fact that we have
searched many correlations. If we make some assumptions, this should
not be too difficult.
The problem can be rephrased in the following way.
Let's assume that we are drawing from a distribution with dispersion
sigma. If we draw n times, what is the probability of inferring a
dispersion which is significantly less than the true dispersion of the
distribution. To ease the calculation we can assume that the
distribution in Gaussian.
As an example.
Let's assume that the true dispersion of the distribution is 0.3
magnitudes and let's assume that we take 10 samples (this corresponds
to the number of SNe in figure 12). What is the probability of
measuring a dispersion of 0.14 magnitudes. The calculation can be done
via Monte-Carlo, or, alternatively, there may actually be a formula we can
use, since we have assumed that the distribution is Gaussian. If we
call this probability P, then probability of not getting a dispersion
as low as 0.14 magnitudes is (1-P). This number will probably be close
to 1. If this experiment was repeated m times (where m is the number
of correlations that were searched), then the probability of finding
one with dispersion 0.14 magnitudes is 1-(1-P)^m.
This archive was generated by hypermail 2.1.4 : Tue Jan 27 2004 - 00:46:48 PST