Re: HST paper U-B, etc.

From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Sat Apr 12 2003 - 13:53:37 PDT

  • Next message: Greg Aldering: "Re: HST paper U-B, etc."

    On Sat, Apr 12, 2003 at 12:28:21PM -0700, Greg Aldering wrote:
    > Now, regarding the U-B error to quote. First, I would suggest that for
    > the sole purpose of determining the extinction correction that you *do*
    > use a stretch-dependent U-B at max even though you did not do this for
    > the K-corrections. Since the K-corrections are tilted to match the
    > observed colors, most of the apparent inconsistency is eliminated.

    I don't want to patch in something like this to part of the analysis.
    First, specific hacks will spaghetti-ize the code worse than it already
    is, which runs all the usual risks. More importantly, it's just less
    clean-- I'd really rather if we did have a closed loop, instead of
    hacking in a specific patch in one place which isn't consistent with the
    rest of the analysis. It's not even exactly obvious how to use a tilted
    U-B "just" for E(B-V). The K-corrections and E(B-V) estimations are
    convolved together. The whole color tilting you talk about is done by
    applying a host-based E(B-V), which in turn comes from the estimated
    colors. If you're going to change your E(B-V) based on a tilted U-B,
    then you also need to change the K-corrections-- or do the analysis in a
    manner completely different from what I've done. (And, as I noted
    above, one which won't be internally consistent.) If we really want a
    tilted U-B in there, it means going back and making an appropriate
    template spectrum and refitting all of the lightcurves and redoing all
    the K-corrections.

    If what the collaboration wants is for me just to use Jha's thesis for
    these things, I wish we'd just say that rather than patching in one
    piece, then the next piece, and then the next piece. In the long run,
    it will be *less* work to just go back and do it from beginning than to
    put it in piecemeal as we decide that the next bit of it needs to be in
    the paper; and, putting it in piecemeal could potentially build a house
    of cards which isn't internally self-consistent.

    A general note on this issue: I was under the impression a month or so
    ago that we had agreed that we shouldn't use Jha's thesis, since it's
    unpublished, and as such I wasn't too worried about it. However, we've
    gotten to the point that we're going to cite it, and to the point that
    we're going to use it in pieces... so it seriously begs the question as
    to why we aren't using it for real. I do have to admit to being a bit
    annoyed that we're reaching this decision now, and didn't just bite the
    bullet and reach this decision a month ago. If I was wrong in thinking
    it shouldn't be used, I wish the collaboration would just have told me
    so. If I was right, I would ask those who agreed with me then to
    question how much we're going to bring it in now, and whether we we all
    wrong in deciding that back then. If we've changed our mind, fine, I'll
    go back and use it. We need to decide now, once and for all, not to be
    changed the way the month-ago decision was changed, if we need to use
    Jha's thesis in this paper's analysis. We need to decide this soon,
    too, because if we really decide that we're going to use it, it's going
    to take quite a bit of time for me over the next week or so to get it
    integrated in there.

    > (By the way, why don't you quote E(U-B) in Table 3? Is it that you are
    > taking E(U-B) and converting to E(B-V)? Do you think this is clear to
    > the reader? Perhaps a footnote for the afffected SNe is in order. Or
    > maybe I missed something entirely!)

    The way I estimate E(B-V) is not by subtracting observed B-V from an
    intrinsic B-V, but rather by figuring out what E(B-V) I need to process
    the appropriate stretch spectrum through (in addition to redshifting it
    and then putting it through Milky Way extinction) in order to get the
    observed integrated R-I color. E(B-V) is as such a perhaps better
    thought of as a parameter; it's also the usual parameter used when
    people talk about extinction, and as such I believe is the right thing
    to quote.

    > But, can you tell us how does the cosmology change?

    There was an attached postscript file; did you see that? If not, I can
    stick it somewhere for people to grab. The error ellipses puff out
    along th emajor axis.

    > In particular, one point I haven't seen discussed (or maybe I've
    > forgotten!) is that since U-band maps to F675W at z ~ 0.88, only
    > SN1997ek is nearly 100% U-B in the F675W-F814W frame. For the other
    > SNe, you still have B-band light in F675W, so the intrinsic dispersion
    > used should be a combination of the U-B and B-V intrinsic dispersions,
    > weighted by the relative fractions of U and B light in the filter.

    You have to be careful about this, because if you're subtracting U-B
    color, the dispersion in the intrinsic color will enter your calculation
    there too. However, as I note above, that's not how I do it, so you're
    right, the degree to which the dispersion affects us is based on the
    degree to which U-band light "contaminates" the observed R filter.
    Really, then, to be consistent we would need to add a variable
    dispersion for supernovae at all redshifts above about z=0.5, once U
    starts to creep into R-band (even though R is primarly K-corrected to B
    for z<0.7).

    There is a complication that the filter measurements aren't points, but
    are integrated ranges, and that it's not obvious that the dispersion in
    U-B is due evenly to the whole range of the U-band filter (probably the
    bluer side of the U-band filter has more dispersion than the redder
    side). It gets more complicated to really do it right; a linear
    estimation of what fraction of the observed filter overlaps part of the
    U filter probably isn't really right. (This is further complicated
    since the ground-baed data and the HST data use different filters.)
    Given that, I'm very hesitant to implement something seemingly
    sophisticated which is too sophisticated for the real knowledge it's
    based on. Better to implement something simple and reasonable and
    doable (e.g. the estimate that "those at z>0.7 are closer to U-B than
    B-V, so we add a color dispersion jsut for those"), unless we really can
    come up with a better estimate which we believe of what is the "right"
    dispersion to use at a given redshift. (And, then, which somehow
    manages to correctly subtract out the 0.11 of intrinsic dispersion we're
    using for extinction corrected data.)

    > In Table 8, could you add P(Omega_Lambda > 0) for each of your fits?

    Doing this might be a little gratuitous, because (if memory serves)
    they're all pretty much at least 0.999. What makes it complicated is
    that by the time you get to the fourth or fifth digit it gets sensitive
    to the resolution of the fit (although by the time we finally converge
    on how to do all of this, I'll do very high resolution fits for
    everything). Can I get away with putting this in the text somewhere?

    > P.S. BTW, I invite everyone to scan the paper for missing references and
    > then provide Rob with the bibtex entry (from ADS), as currently the
    > scholarship is a bit light in a few places. Those types of things
    > annoy some reviewers, and may make them question our analysis because
    > they may think we haven't done our homework.

    Yes, please do! There are a few places that have in all caps a request
    for the right reference on things; look at those, but also let me know
    anywhere in there where something ought to be cited which currently
    isn't.

    -Rob

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


    This archive was generated by hypermail 2.1.4 : Sat Apr 12 2003 - 13:54:00 PDT