Low-z stability test

From: Alex Conley (aconley@panisse.lbl.gov)
Date: Thu Mar 13 2003 - 20:58:03 PST

  • Next message: Robert A. Knop Jr.: "Re: sn types, templates, and other comments on the Mar 13 draft"

    I have finished checking the Hamuy and Riess SNe for fit stability. For
    the most part they are pretty stable, with a few exceptions mostly having
    to do with floating offset cases. In general, I think that sn1992ag is
    the only SNe which floating offsets helped. All of the others mentioned
    as needing the offsets don't IMHO. In fact, some of them are made much
    worse by letting the offset float, both in terms of the visual quality of
    the fit and that they go a bit unstable.

    sn1995bd is a piece of crap, and I'm glad it's excluded.

    My fit strategy was to do fit I with the defaults, fit II with the initial
    conditions matching the fit values from I fairly closely (I find that this
    frequently leads to better error bar estimates), and in fit III I try to
    'shake things up' to see if anything changes. For the SNe listed as
    needing floating offsets, there are fits IV and V (and sometimes VI) where
    I continue the process with floating offsets.

    These labels are also attached to all of the output plots.

    Note that the magnitudes in Rob's tables have been corrected for
    extinction, but mine haven't, so it takes a little work to compare them.
    The stretches are directly comparable. All of the fits were done with
    the latest version of snminuit. The corrected magnitudes are known to
    have bad error bars, but Rob doesn't use those since he is fitting his
    own alpha anyways. In fact, I can't imagine any reason why anybody would
    ever use them directly.

    Attached are three things:
     (i) a gnumeric spreadsheet giving details of many of the fit parameters
       for each SNe. The fit description is either 'basic fit' or gives
       changes relative to the basic fit. The basic fit is detailed below.
     (ii) A tar file with all of the snminuit control files used to do these
       fits.
     (iii) A tar file with all of the output from snmin02.

    These will dump a bunch of files in your current directory if you untar
    them, so you might want to do this in a temporary directory.

    The basic fit uses the defaults

     1 't of B max' 0.0 0.2
     2 'Stretch ' 1.0 0.02 0.45 5
     3 'Vert scl 1' 0.7 0.05 0.1 10
     4 'Vert off 1' 0.0 0.05
     5 'Vert scl 2' 0.7 0.05 0.1 10
     6 'Vert off 2' 0.0 0.05

    Now, on to the 42 SNe. I expect these to be much more unstable.

    Alex

    Here are SN by SN comments:

    sn1990o -- solid
    sn1990af -- solid
    sn1992p -- solid
    sn1992ae -- solid
    sn1992ag -- ugly fit, moves around. Fitting with floating offsets (fit
    IV)
      makes it much better
    sn1992al -- solid
    sn1992aq -- solid
    sn1992bg -- values solid. Improving initial guess inflates error bars
    slightly
    sn1992bh -- solid
    sn1992bl -- solid
    sn1992bo -- solid. Rob has this as one that requires floating offsets.
      Fits IV and V are floating, I-III are not. The differences are pretty
      minor (0.03 mag, 0.02 stretch in the direction that they will mostly
      cancel each other).
    sn1992bp -- solid. Another floating offset. This time the effect is a
      little larger. As before I-III are fixed, IV and V are floated.
    sn1992br -- solid
    sn1992bs -- solid
    sn1993b -- pretty solid. Very minor (3rd decimal place) shifts in the
    error
      bars when you tweak. Rob has this as an SNe which require floating
    offsets,
      but I don't really see it. Nonetheless, fits IV and V show what
    floating
      offsets do. Floating offsets make a substantial difference
    sn1993o -- solid BUT not a great fit to B peak. Changing it to floating
     offsets doesn't help. Either way, the fit parameters are pretty
    insensitive
     to the initial values.
    sn1993ag -- solid
    sn1994M -- solid
    sn1994S -- pretty solid, but the V peak seems slightly off. Changing to
     floating offsets doesn't really improve this. In any case, changing the
     initial conditions changes the error bars a bit, but not the central
    values.
    sn1995ac -- solid
    sn1995bd -- a piece of junk. The fits move around a bit when the initial
      conditions change, but mostly this just misses the peak badly, even when
      a floating offset is used. I'm glad this one gets excluded from the
    fits.
    sn1996C -- solid. There are some points around max (one point especially
    in B)
      which aren't fit so well, but playing around with the fit doesn't seem
    to
      help.
    sn1996ab -- solid as fixed, not solid as float. This is one that Rob
    claims
     needs a floating offset. I don't agree. Furthermore, I think that the
     floating fits suck, and it was a mistake to use them. The fixed floats
    are
     quite stable, and the floating ones move around a decent amount when you
     adjust your initial conditions.
    sn1996bl -- solid
    sn1996bo -- solid, both fixed and floating. However, the floating fit is
     worse. It seems to miss the peak. Again, using floating offsets has
     ruined a perfectly good fit.









    This archive was generated by hypermail 2.1.4 : Thu Mar 13 2003 - 20:58:36 PST