[Fwd: Marginalize or intersect? (for Flat Omega_M)]

From: Tony Spadafora (ALSpadafora@lbl.gov)
Date: Tue Feb 25 2003 - 14:58:34 PST

  • Next message: Gerson Goldhaber: "sn92ag"

    Note from Rob below:

    Rob,
    Please continue to send a copy of all e-mail to hstpaper@panisse.lbl.gov
    and check the archive:
    http://panisse.lbl.gov/collab/archive/hstpaper

    but feel free to also email deepnews or whomever when you think
    appropriate. (the archive is to just so we have a complete set and avoid
    deepnews-saturation-syndrome.)

    -Tony

    -------- Original Message --------
    Subject: Marginalize or intersect? (for Flat Omega_M)
    Date: Tue, 25 Feb 2003 16:36:24 -0600
    From: "Robert A. Knop Jr." <robert.a.knop@vanderbilt.edu>
    To: Tony Spadafora <ALSpadafora@lbl.gov>
    CC: hstpaper@panisse.lbl.gov
    References: <20030225184405.GD27745@brahms.phy.vanderbilt.edu> <3E5BBE4D.638735D@lbl.gov>

    (Tony -- could you please alert the relevant people that this has shown
    up on the archive page?)

    The argument was made during the group meeting, and at the time I
    believed it, that I should be marginalizing rather than intersecting to
    get the best value of Omega_M for a flat universe.

    I no longer think this is correct.

    Omega_M for a flat universe means, if you assume Omega_M+Omega_Lambda=1,
    what is the best fit value you get for Omega_M? The way to address this
    is to do a chi-square fit of the cosmology, allowing Omega_M to vary.
    Specifically, I would calcualte chisquare for lots of values of Omega_M,
    and at each value use 1-Omega_M for Omega_L in the luminosity distance
    calculation. What this would produce is *exactly* the *intersection* of
    our confidence regions with the Omega_M + Omega_L = 1 line.

    Marginalizing is something weird. To get the value in question, one
    would Marginalize over the parameter Omega_M+Omega_L to find the best
    value of Omega_L-Omega_M; call that result x. The "best fit Omega_M
    assuming flat" would them be (1-x)/2. However, that's not really what
    that is; that's just another way of saying what is the best fit value of
    Omega_L-Omega_M, where Omega_L+Omega_M can have any value. That's
    something perverse that isn't really directly what we care about in this
    case.What we're *really* interested in is what is the best fit value of
    Omega_M, given that Omega_M+Omega_L is constrained to be equal to 1.

    Thus, we want to intersect, and not marginalize, in order to determine
    the "assumed flat universe" value of Omega_M.

    -Rob



    This archive was generated by hypermail 2.1.4 : Tue Feb 25 2003 - 14:58:50 PST