From: Gabriele Garavini (garavini@in2p3.fr)
Date: Tue Dec 21 2004 - 06:45:27 PST
Dear All,
I apologize for the length of this mail but I hope this clarifies some of
the issues.
As Chris pointed out the studies on the possible systematic effects on the
EW measurement technique has been carried out in Gaston's paper and are
described in its section 3.3. I've also performed some test and I would
like to resume what came out:
First I would like you to remember that the definition of what Gaston
called EW measurement is probably somewhat different from the usual one:
the pseudo continuum is fitted taking into account two small regions
(defined in table 3 of his paper) on the maxima around the absorption
features. The possible systematic related to this definition are the
following
-The size and position of the region selected for the fit can be somewhat
arbitrary (always within the limits defined in table 3) and could be a
kind of 'systematic of the researcher' error. This was found (see
section 3.3 of EW paper) to be
the dominant source of error in the cases where S/N > 10 per
resolution element. The VLT high-z spectra we selected for the evolution
paper have typically S/N = 10 per 20A
resolution. This source of uncertainties is then not a big concern for my
case since it is sub-dominant. Moreover, I recall Gaston doing
these tests on the high-z spectra and finding the error being
sub-dominant with respect to the statistical one.
-The effect of resolution was also studied and found completely marginal
already at high signal to noise ratios. Gaston tried to smooth the data
and redo the measurements with the lowest resolution of 10A/pix. I did a
test lowering the resolution even more, down to 20-25A/pix, and the
systematic was at most 1%. This was just to check out if for some reason
Gaston did the actual integration on the smoothed data. The result is
that if he did it wouldn't matter.
-Signal to noise was also a concern. The possible systematic arising from
a low signal to noise ratio is not significant, the statistical error
is dominant.
All these tests can be found also in page 108 of Gaston's thesis, fig
5.10.
I personally was concern of whether the definition of the continuum
endpoints was in some way affected by smoothing. I recall Gaston not
smoothing the data for selecting the regions on the maxima, but I did not
want to trust only my memory so I tried on few spectra. I tested the
effect of fitting the continuum on the smoothed spectrum. The result is
that this has a small effect pushing EW toward low values (around 1% for
resolution down to 20A/pix). This is however sub-dominant with respect to
the statistical error.
Last point to address is why Eric found some systematic difference:
On his document he describes the methodology used to perform the
measurements. On the first page he says (point 2) that it does select the
endpoints (both lambda and flux) on the smoothed spectra and that he then
computes the EW using 'splot' and those exact points. This is not the
standard procedure Gaston identified. If I understand correctly what Eric
is doing he is looking at the smoothed spectrum with 'splot' and writing
down were he thinks is the maximum on each side of the absorption.
Gaston, I remind you, was selecting a small region around each maximum and
was fitting the continuum in between those regions. Selecting the maxima
by eye and using 'splot' for performing the measurements on a noisy
spectrum I think is not recommended and can be the source of big
systematics, bigger than fitting the continuum on the smoothed regions (as
the test I've done).
To sum up I think the origin of the difference is just that a different
measurement technique than that used for the low-z SN. The technique
introduced by Gaston in the EW paper was tested and found robust for
possible systematics involved in the measurements and I strongly recommend
everybody to follow it. If the S/N ratio is too low to identify the shape
of the maxima just drop the spectrum. Unfortunately the EW are a tricky
thing to measure and it should not be attempted to measure spectra with
S/N lower than 3 per 20A pixel. This was the limit we set while measuring
the VLT data.
I hope this clarifies a bit.
Speak to you later.
Thank you very much
Cheers
Gabriele
On Sun, 19 Dec 2004, Chris Lidman wrote:
> Dear All,
> I think that Rob and Eric have raised as issue that needs to be
> discussed on Tuesday.
>
> I attach the most recent version of Gaston's paper (version Jan. 22
> 2004). A description of the method that was used to determine the
> bounds over which features are measured is given in section 3.2.
>
> Gaston worked with spectra that had good S/N ratios and the
> description given in section 3.2 might be sufficient for such spectra.
> The spectra with the lowest S/N ratio in Gaston's paper has a S/N
> ratio of 5. Gaston did not find any systematic effects when the
> S/N ratio of the best spectra were reduced to such a level. This is
> reasonable.
>
> However, such a S/N ratio is probably higher than S/N ratios of
> most of the high-z SNe in Gabriele's paper. Hence, it seems reasonable
> to ask Gabriele or Gaston to check for systematic errors when the
> S/N ratio is reduced to the levels that are typical for high z SNe.
>
> For such low S/N spectra, it would seem that some sort of rebinning is
> necessary before features can be identified and measured. Gaston, were
> the high-z spectra rebinned before the EWs were measured?
>
> Cheers, Chris.
>
> PS The last message in the e-mail archive is dated January 26th, 2004.
> This e-mail contained the minutes of a meeting we had on Gaston's paper
> and it contained suggestions on how the collaboration thought the paper
> could be improved. Gaston, let us know if you are still out there. I
> might send a search party to La Serena to find you :)
>
>
>
> On Fri, 2004-12-17 at 17:10, Robert A. Knop Jr. wrote:
> > On Fri, Dec 17, 2004 at 11:59:45AM -0800, Tony Spadafora wrote:
> > > I've put more time for the evolution paper because this paper has been
> > > before the collaboration since July. The discussion here should be to
> > > confirm that all issues have been resolved (e.g. the systematics of the
> > > equivalent width determination) and, if so, that a submission-candidate
> > > version should be finalized and submitted soon.
> >
> > For the record: I'm not convinced at all that the systematics in the EqW
> > determination are resolved. So far, all that's happened is that the
> > issue has been raised. More basically, the method for determining
> > them from noisy spectra needs to be documented in the paper; all that
> > Eric and I have heard so far is that "Gaston did it" is the method,
> > which obviously isn't something one writes in a published paper.
> >
> > Gaston probably really needs to be at this meeting since he is the one
> > who knows how the measurements were made.
> >
> > There's also the issue that this builds on the nearby EqW stuff that
> > Gaston has done. If *that* paper isn't published first, then that stuff
> > will need to get incorporated into this paper for this paper to make
> > sense.
> >
> > Please include Eric on mailings for this, since obvously he is involved
> > with this. Also please re-send the schedule so he gets it.
> >
> > -Rob
>
This archive was generated by hypermail 2.1.4 : Tue Dec 21 2004 - 06:45:45 PST