I-band paper comments

From: VAFadeyev@lbl.gov
Date: Wed Jan 14 2004 - 01:15:56 PST

  • Next message: Robert A. Knop Jr.: "link"

    Hi Serena,

    sorry for belated comments. The paper looks very
    good to me. I have one concern and a few passing
    remarks:

    1) You have found some correlations between
    the template fitting parameters (Figs. 7-9).
    Did you checked the Monte-Carlo tests data
    (section 2.3) for the existence of the same
    features, by any chance? Or, alternatively,
    did you obtain the correlation matrix of
    the fitted parameters?

    My concern is that I would actually *expect*
    some correlations (e.g. between the stretch
    and the time of the second peak), simply
    because of the way the fitting procedure
    is constructed. For example, if there is
    a spurious upward fluctuation on the rising
    slope, that should push the stretch to higher
    value; then the righ-hand-side slope of the first
    peak would also be higher, and the position of
    the 2nd peak shifted to a later time.

    This issue may or may not be important,
    it depends on how correlated the fitted parameters
    are. If they are fairly independent, then
    the observed correlations are due to the real
    SN features. If not, then you might be seeing
    an artifact of your procedure.

    2) Do you have a rough estimate of how much
    of the time-of-the-maximum dispersion shown in
    Fig. 6 are intrinsic, i.e. not due to the
    measurement errors?

    3) You have alluded to the systematic trends
    in the lightcurve fit residuals. Well, the artifact of rising slope in the pre-1st-maximum residuals
    is very obvious. It shows up in 7 out of 9 SNe
    which have an adequate pre-max coverage.
    Not that I have a constructive suggestion
    about this.

    4) Many lightcurve fits have rather large Chisq/Dof.
    Did you try to account for that when obtaining
    the fit parameter errors and later on the magnitude
    dispersion in the Hubble diagram? There is a
    "recipy" when one artificially scales the data
    errors to make Chisq/Doc = 1. (I'm just not sure
    if you used this or any other way of "massaging"
    the data in such cases.)

    Cheers,
    vitaliy



    This archive was generated by hypermail 2.1.4 : Wed Jan 14 2004 - 01:16:03 PST