stability fits redux

From: Alex Conley (aconley@panisse.lbl.gov)
Date: Tue May 20 2003 - 15:10:52 PDT

  • Next message: Robert A. Knop Jr.: "Re: stability fits redux"

    Hi Rob,

      What was decided about the SNe that I thought were poorly behaved
    (that is, the SNMINUIT fits were unstable to initial conditions) when
    I refit all of them? There were three in the 42 sample that were
    exceedingly bad, and I thought were not suitable for cosmology fits.
    Two of them no longer appear to be used, but one of them (1997aj = 9794)
    is still present.

      There was a message thread about this

    http://panisse.lbl.gov/collab/archive/hstpaper/0121.html

    which ended up with you tossing all SNe with NO color information
    (which took care of the other two that I had problems with), but didn't
    affect sn1997aj.

    The reason I thought that sn1997aj was bad is that there were two
    solutions with the same chisquare but very different fit values.
    The first one had s=0.954, mR=22.587, and the second had
    s=1.504, mR=22.859. The both had chisquare of 120/70. Admittedly
    I am more inclined to believe in the s=0.954 fit, but this is a prior
    that probably shouldn't be implicitly included.

    I think we could use this as a formal critereon: If it was found that
    changing the initial conditions of the fit produced more than one solution
    with similar chisquare but fit parameters that varied by more than x sigma
    (where x could be 5 or more), the supernova was excluded. This would
    eliminate only 97aj from the currently existing 42 sample.

    Alex



    This archive was generated by hypermail 2.1.4 : Tue May 20 2003 - 15:10:53 PDT