x,y position in HST fits

From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Thu Feb 27 2003 - 07:08:27 PST

  • Next message: Robert A. Knop Jr.: "missing greek letters"

    Greg said:

    > Do you say that you marginalized over the photometry nuisqnce
    > parameters, like x0,y0. We should comment that we did enforce the
    > position to be fixed with respect to the image coordinates due to the
    > lack of many reference sources and the complex and time-dependent
    > nature of the PC coordinate transformations (cite Anderson & King in a
    > recent issue of PASP).

    This statement is incorrect. We did *not* enforce the position to be
    fixed. The result of Alex Conley's extensive work on calculating
    transformations between images using reference objects in the WF fields
    was that you can only get transformations good to ~1 pixel, which is
    insufficient for PSF fitting.

    As such, I allowed x0,y0 to vary separately in every image. I used
    Alex's transformations to calculate a rotation (or, where unavailable, I
    used a pair of well-separated objects on the PC field to calculate a
    rotation); this rotation was *only* used to rotate the background
    model-- I didn't actually rotate the images. Then, I allowed the psf
    fit model shift the position, so the effectively the supernova point
    source itself was used as an alignment reference. This was clearly the
    right thing to do, since in some cases the position was ~1 pixel off,
    and a fit that insisted on not moving would have been terrible. (This
    does complicate use of final reference images; I don't think any of
    those went into this paper, but where I've used them I just have to
    assume that the Conleyesque transformations are perfect, and fix the x0
    and y0 for those specific images.)

    I marginalized over x0 and y0 implicitly, in that I didn't calculate a
    chisquare everywhere in this n-dimensional space, but rather used a
    standard lmfit procedure (outlined in Numerical Recipes). This
    procedure does produce a covariance matrix for the fit parameters; the
    total errors quoted do implicitly include the effects of marginalizing
    over all the other parameters. (In the end, all I really cared about
    were the fluxes for each image.)

    -Rob

    -- 
    --Prof. Robert Knop
      Department of Physics & Astronomy, Vanderbilt University
      robert.a.knop@vanderbilt.edu
    


    This archive was generated by hypermail 2.1.4 : Thu Feb 27 2003 - 07:08:28 PST