Recipe to obtain polarization information for the 2002 run 

Attendees: Les, Johanna, Jeff, Yuji, Abhay, Sandro, Gerry, Osamu

We met to discuss our plan to prepare a recipe to obtain polarizations
for the 2002 run.  We decided to develop a plan, then to send the plan
to the spinners, and get feedback.  We need to have polarization that
we can use for our results by mid-August (August 10?).

1. spin-sorted information vs. bunch-sorted information from the
polarimeters.  The spin-sorted information contains histograms by
energy for each strip.  The bunch sorted info. has scaler numbers
in a preselected carbon "banana" for each strip and each bunch,
with no energy information.  With the spin-sorted histograms, it is
possible to correct the energy calibration through the run (it was
found to change by as much as 30%(?) over the 2002 data run).  With
the bunch-sorted data, it is possible to eliminate "bad" bunches which
show anomalous behavior (unphysical ratios of counts across the 6
silicon detectors, for example).

We looked at the change in measured asymmetry before and after
energy correction, and found changes of only .3sigma, with sigma=2.5x10^-3.
Since the analyzing power is about 10^-2, this corresponds to about
2% change in the polarization for energy corrected vs. uncorrected.
We decided that this is small compared to observed systematic errors
(from false asymmetries) which are of order 5%.

For the bunch sorted data, we have numerous cases of very large chi^2
for sin phi fits to the detectors, where large asymmetries are reported.
We decided to use the bunch sorted data.

2. We need a plan to select bad bunches.  We decided on the following:
(note that this is tentative, to go to the collaborations before deciding
on any final plan)

  a. throw out bunches 1 and 11 for all runs (bunch 1 is always
  anomalous; bunch 11 is regularly kicked to measure the tune and is
  always anomalous)

  b. identify bunches with bad chi^2 for a sin phi fit to the 6
  detectors. For this fit, two "good" bunches are identified as comparison
  bunches. Ratios are calculated of the number of counts in each of the 6
  detectors to the number of counts in the comparison bunches, for each
  detector, for each bunch. The 6 ratios for the 6 detectors, for each
  bunch, are fitted to a constant + sin phi curve, thus allowing for
  vertical polarization. The chi^2 clearly shows separation between bad
  bunches (typically with chi^2/NDF > 10), and normal bunches (chi^2/ndf
  < 10).  To decide on the cut, we will generate histograms of chi^2/ndf
  for all of the bunches in each polarization measurement run.

     Chi^2/ndf plots for every polarimeter runs
        - projections (Ps format)(Pdf format)
        - function of bid (Ps format)(Pdf format)
        - sum up all the runs into one plot (pdf)
     Performance plots
        - asymmetry as a function of time
          o physics asymmetry (1.1-1.25/2002) (12.19-12.31/2001)
          o acceptance asymmetry (1.1-1.25/2002) (12.19-12.31/2001)
          o luminosity asymmetry (1.1-1.25/2002) (12.19-12.31/2001)
        - projection to Y axis (ps format)
     Note for the plots (txt file)

  c. also create a histogram of specific luminosity at the
  polarimeters for each bunch, and look for a cut for bad bunches for
  each polarimeter measurement. (Specific luminosity at polarimeter =
  N(total counts for all detectors)/current in bunch (wall current monitor).)

     Web page for the specific luminosity at CNI polarimeter (link)
     Note for the plots (txt file)


  d. generate a list of the bad bunches by a,b,c for each polarimeter
  measurement, and distribute to experiments.  Decide whether the experiments
  would add any additional bunches to this list, from their specific
  luminosities for the bunches at the experiments.  (Specific luminosity at
  the experiment is counts(BBC or whatever)/(wall current monitor values for
  yellow x blue).)

     lists of the bunch selection from several cut conditions
        Format of the files:
        Fill RUN RING NDISC BFLAG1 .... BFLAG60

        FILL : fill#
        RUN : CNI run# (note: only flattop measurements are shown)
        RING : ring information 1=blue 2=yellow
        NDISC :number of discarded bunches with flag=1
        BFLAG1-60 : bunch selection flags for bunch 1-60
            1 = irregular discarding (satisfy the cut condition)
            2 = regular discarding i.e. bxing=1,11,16-20,36-40
            3 = abort gap
        cut condition = Chisq/ndf > 3 (dat)
        cut condition = Chisq/ndf > 5 (dat)
        cut condition = Chisq/ndf >10 (dat)

     Web page for the specific luminosity at STAR (by Joanna Kiryluk) (link)
     Web page for the specific luminosity at PHENIX (by Takahiro Kawabata) (link)

  e. throw out polarization measurements with a number of bad bunches
  above some number, to be determined from a distribution.

     Plots showing the number of irregular bunch discarding (other than the
     regular ones 1st,11th,16-20th,36-40th). CNI measurements are only from
     flat top and three types of Chi square cut conditions are displayed. (pdf)
     n over 8 will be discarded for chisq/ndf > 5 

     (pdf) The chi square distributions compared to the Chi square functions for ndf=3,4,5

  f. throw out fills based on a number of bad polarization measurements
  (e) > 2 in one fill.

     CNI run and fill selections (txt)

3. We still need to decide how to proceed at this point.  We are
assuming, based on at least one correlation between bad bunches and
differing measured asymmetry in a fill, that the bad measurements are
just that, and should be rejected, and that the polarization for the fill
should be calculated without them.  This selection of bad measurements
must not be done looking at the asymmetry!
POLARIZATION INFO with tentative format discussions through emails