color cuts for 1st epoch search

From: Greg Aldering (aldering@panisse.lbl.gov)
Date: Sun Sep 07 2003 - 12:54:09 PDT

  • Next message: Mamoru Doi: "Re: fall 2003 search with Subaru?"

    As you know, the 1st epoch of the spring 2004 HST search will be done
    with a very large gap, ~ 200 days, between new and refs. This means
    that some fraction of candidates will be old Type Ia's, whose value for
    follow-up is dubious. In addition, we have the issue of trying to
    suppress Type II's for all searches.

    I asked Peter to run Type Ia and Type II simulations with a 180 gap,
    and generating the detected SN counts in r, i, and z for ACS on HST. I
    have used the results to calculate S/N values using the expected search
    exposure times, using 4/5 orbit in z and 1/5 orbit in r or i. These S/N
    values were used to insert noise into the color measurements. I also
    used a scatter of 0.1 in the redshifts, assuming that most of the
    redshifts will come from photo-z's. I required S/N > 6 in z for an
    object to be considered a candidate. I then experimented with a range
    of linear color cuts of the form color_cut = a*z + b for z > 1.2 to see
    whether Type II's or old Type Ia's can be suppressed using the color
    information. Again, remember that all cuts are made using the measured
    (noisy) values in order to simulate the information actually available
    from the search.

    As a baseline, out of all the candidates with phot-z > 1.2, the
    simulation finds that 61% will be good Type Ia's, 19% will be Type II's,
    and 20% will be late Ia's.

    Now, whenever a cut is made, some good Type Ia's will be lost. The
    fraction of good Ia's retained relative to the total number of Ia's is
    the "efficiency". Of the Ia's that pass the cuts, the fraction which
    are late is the "late fraction". Without cuts the late fraction is 24%.
    Likewise, I define the "Type II fraction" as the number of Type II's
    that pass the cuts, relative to the sample of good Ia's plus the number
    of Type II's. (Here one could have constructed alternative indicators
    of Type II contamination by including the late Ia's or excluding the
    Type II's from the denominator.) Without cuts the Type II fraction is
    also 24%.

    What I find is that for i-z the Type II fraction runs at about 23% for
    any color cut that gives a reasonable efficiency. This is because for a
    fraction of Type II's, color at a single epoch doesn't tell you much.
    For r-i the Type II fraction decreases as the efficiency increases, and
    so is never better than the limiting no-cut value of 24% Thus, for
    candidates with z_phot > 1.2, color cuts don't help in rejecting Type II's.

    As for rejecting late Type Ia's, recall that the baseline late fraction
    is 24% for 100% efficiency. Making a color cut of course leads to a
    loss of efficiency, so I tried to balance efficiency and late
    fraction. When there is no noise included, most late Ia's can be
    rejected while retaining very high efficiency. (Indeed, my observation
    of this situation in Peter's simulation is what motiviated the present
    investigation.) When noise was included, I found that r-i was not
    effective in eliminating late Type Ia's. This is probably because there
    is little flux in r-band to work with for z > 1.2. For noisy i-z,
    there is a shallow minimum in late fraction over efficiency when the
    efficiency is about 80%. Here the late fraction is about 16%.

    In this case one has thrown out 20% of the good Ia's in order to
    achieve a reduction from 24% to 16% in the fraction of late objects you
    would follow. This can be a successful strategy when there are SNe to
    spare, however, when there are no spare SNe, as in our HDFN search, an
    80% efficiency is too low. At 90% efficiency, using i-z the late
    fraction is 19%; at 95% efficiency the late fraction is 21% for i-z.
    The bottom line is that some of these populations overlap a lot in
    color-redshift space when the color and redshift uncertainties are
    considered, so they can't be effectively separated.

    The big danger is that these results are model-dependent. One could
    choose a color cut based on the models which then leads to a poor
    efficiency for real data. Also, the uncertainties play a key part and
    thus must be well known. Moreover, there appears to be little gain in
    applying a color cut since Type II's can't be rejected very well and
    the gains in cutting late Ia's are modest. If a color cut is to be
    used, i-z is better than r-z; from this, I would not expect V-z to help
    at all.

    Note that these cuts may help much more at lower redshift where the
    S/N of the color measurements is better and where there is a larger
    fraction of observable Type II's (according to the models). This would
    matter if we want to work at lower redshift to increase the search
    yield, or in understanding what can be done when we don't have photo-z's.

    Overall, for the 1st epoch search we are likely to have about 40%
    contamination from undesirable objects, even if we have photo-z's. Our
    follow-up plan should be conceived to try to detect such objects using
    changes in color and/or brightness as well as color and brightness
    themselves.

    - Greg



    This archive was generated by hypermail 2.1.4 : Sun Sep 07 2003 - 12:54:10 PDT