Re: Greg's comments on my paper draft

From: Alex Conley (AJConley@lbl.gov)
Date: Tue Sep 07 2004 - 15:29:10 PDT

  • Next message: Vitaliy Fadeyev: "CMAGIC cosmology paper comments"

    Hi Greg,

       Thanks for the comments. They were very useful. Some
    will take me a while to answer (not that I'm complaining;
    these were exactly the sort of comments I was hoping for
    before I unblinded), but I can try to respond to some of them.

    Analysis questions:

      \sigma_int comes only from the low redshift SNe, although
       the high redshift SNe are consistent with the same value.

      Does more dust imply hihger R_B? Maybe, maybe not. But
       R_B is usually larger in dense clouds, which is thought to
       arise because of UV shielding. So
       more dust->UV shielding->different grain size distribution,
       which suggests to me that more dust -> different R_B.

      When points with large uncertainties are allowed (the
       magerror cut is relaxed) the contours on the cosmological
       parameters are still well behaved.

      I did not exclude SN1997O in the first cosmology fits that
       I did. At the time I was not aware that it had been
       excluded from P99.

      I have not gone back and investigated the SNe that fail the
        probcut test in any detail.

      I have not looked at the P(chisq|N) distribution for the CMAG
        fits, but the expectation is that it will be far too good.
        This is what led Wang '03 to rescale the photometric errors --
        the error bars on the low redshift sample are almost certainly
        both overstated and correlated. However, I will try to put
        this plot together for you. I'll have to think about how
        to deal with the fact that N varies over the sample, so the
        expected width of the probability distribution is different
        for different SN. Perhaps I can scale by the width of the
        distribution, kind of like dividing by sigma.

      The value of the blindness point (om = 1, ol = 1.1) was chosen
       for two reasons. My initial blindness point was 1, 1.5.
       Reynald pointed out that this was very similar to the expected
       result, so there was psychological advantage to moving away
       from the expected value and forcing readers of my draft to
       psychologically confront the blindness scheme. On the other
       hand, the size of the contours depends somewhat critically on
       the om, ol values, so if I chose extremely different values the
       contours would look huge, and people would dismiss the analysis
       as not providing meaningful constraints. I will add a footnote
       to this effect.

      Values of S -- is it different from 1, etc. : There is a
       slightly newer version of the paper than the one you looked
       at (1.27, as opposed to 1.26 which I think is the one you saw)
       which includes error bars in the residual plots. When I
       performed fits including the error bars I get S=1.11+-0.22 (low-z)
       and 0.96+-0.19 (high-z). I am going to try and improve this
       by attempting to include some covariance information. Mark
       Strovink also pointed out that I should include the errors in
       the Pearson's correlation coefficients, which I have not done.

      Figuring out if SN2001fo is a Ia:
       Type Ic SN have CMAG diagrams that look like low stretch Ia
       CMAG diagrams, so unfortunately it can't easily be used to
       test for Ia-ness.

      I have tried to investigate differences between the Riess/Hamuy/Jha
        sample. I refer you to sections 7.1, 7.5, and 8.4 and figure 14 of
        the method document linked to off of my web site. To summarize,
        however, the Riess and Hamuy samples seem consistent, but the
        Jha sample has a larger dispersion and a mean offset from the
        other samples. When the Jha sample is removed the intrinsic
        scatter drops to 0.08. However, the high redshift sample does
        not seem to be consistent with this small of a value of sigma_int.
        The effects on the final cosmological fit are not large.

      I will look into breaking up the sample into various subsets. The
        one you suggest (R->B, I->B, Z->B) is essentially by redshift.

      Color at B max (fig 4): The newer version of the paper has a much
        nicer version of this figure thanks to a suggestion by Chris.
        I haven't performed an F-test, but I would certainly expect the
        samples to fail the KS test. The KS test effectively compares
        means and variances of two populations, and the high redshift
        sample will have a higher variance because of measurement errors.
        The same can be said for the beta (cmag slope) fits. I will
        add a note about the mean values of the colors.

      The presence of a bump is not correlated with the slope, nor is
        the stretch. These plots are in my thesis, but not in the
        paper. The second (at least) was shown in Wang '03.

      The first three high redshift SNe being high on the Hubble diagram:
        These are SN1995ba (P99), SN1998as (K03) and SN1996K (Riess '98).
        I have no idea what the galaxy density around these SNe was.
        The most likely explanation is dust -- these are the lowest
        redshift end of whatever survey they came from, so are the least
        affected by dust related Malmquist bias, and therefore should
        have higher mean extinction values than the rest of the sample.
        I will look into it.

      Matched residuals decreasing with faint resids: I will investigate.

    Other questions:

      The script form of R_B is used by some authors and not by
       others. There doesn't seem to be any clear standard.

      On the qustion of what plausibly a Ia means in section 3:
       Yes, that wording was very vague. I'm not sure how to
       make it much more precise. Chris also commented on this,
       so I changed the text in version 1.26 to read:
        'must be at least plausibly a Ia based on either lightcurve
         shape, spectroscopic ID, or host galaxy morphology.'
       Basically the criteria here is meant to be something like
       that we use when deciding whether or not to follow a
       SNe candidate. Any suggestions for a better way to
       explain this are welcome, but I don't think I could ever
       put a percentage number on it.

      Covariance matricies: No, I didn't ask for anybody else's
       covariance matricies. Ours have never been
       published. I will try to think of a good way to reword
       my statement to avoid the implication that we are the only
       people who bother with calculating them.

      Script K in equation (4) (luminosity distance equation):
       Goobar and Perlmutter '95 used the script K. Also,
       people frequently restrict k to -1,0,+1 by a coordinate
       rescaling. What you may mean is 'Should script k be
       \Omega_k?', which is true except when \Omega_k = 0.

      Infrared extinction laws of LMC/SMC: In the far IR
        the LMC and SMC have substantially more extinction
        than one might expect. In the MW people usually assume
        that the number of graphite and silicate grains is about
        the same, but the MCs seem to require more silicates.
        See Pei ApJ 395, 130 (1992).



    This archive was generated by hypermail 2.1.4 : Tue Sep 07 2004 - 15:29:13 PDT