[Sane-analysis] cer_h

[O. A. Rondon]

Hi,

Using James' plots and tables of the asymmetries for different
combinations of cuts, I have summarized some results which indicate the
need to better understand the way cer_h codes are computed and to modify
them accordingly.

Hovhannes has been working on the issue, but there are some unexpected
features which I think need to be reviewed in more detail. We need to
thoroughly understand what is going at every stage if we are ever going
to figure out the perp data.

The first summary is on the mean asymmetries for 4.7 GeV 80 deg runs.
The cer_h=0 cut for single low energy clusters is inconsistent with the
idea that cer_h = 0 would correspond to accidentals. The asymmetry is
well correlated with the Pb*Pt product and it is twice as large as the
asymmetry for low energy cer_h = 1 events, which itself is non-zero at
the 8 sigmas level, see p. 2 (the x axis is labeled with the applied
cuts: trigger type, No. of clusters, cer_h)
http://www.jlab.org/~rondon/sane/analysis/asym/cuts.pdf

I would also expect that the cer_h = 0 asymmetry should be very small,
uncorrelated with Pb*Pt, and it would disappear as the cuts become
tighter, but it actually grows.

We have concluded with Hovhannes that cer_h = 0 includes electrons,
explaining in part the asymmetry we see. What remains to be understood
is why the cer_h = 1 asymmetry for the same low energy electrons is only
one half of that with cer_h = 0 (different by many sigmas), see p. 1.

The problem I see is that the cer_h = 1 events pass the constraint of 5
or more cells in the cluster, but the cer_h = 0 don't. But, since both
cer_h = 0 and 1 events have the same E<0.8 GeV and single cluster cuts,
one cannot conclude that the difference is somehow due to the cer_h 1
events corresponding to higher energy clusters with more cells.

Why are there low energy clusters with 5 or more cells? and why is the
asymmetry of clusters with 4 or fewer cells less than 1/2 that of those
with 5 or  more cells? We must understand this before making any changes
to cer_h.  Don't forget that we are talking about low energy single
clusters in both cases.

Next, I explored the asymmetry for low energy cer_h = 0 for all
configurations:
http://www.jlab.org/~rondon/sane/analysis/asym/cer/cuts2a.pdf

This is a further mystery. To some extent, Hovhannes reasoning about the
asymmetry for cer_h =0 being due to low energy DIS electrons explains
part of what we see, for example, the smaller asymmetry for 5.9 vs 4.7
perp is consistent with Q2 being larger for 5.9 GeV data.

But the equality and opposite signs of the para and perp asymmetries for
each energy is not expected, nor there is any reason for the 5.9 GeV
parallel to be smaller than the 4.7 GeV one. In fact, if cer_h = 0 are
electrons, equality between parallel and perp would be excluded, because
the parallel DIS asymmetry is known to be much larger than the perp one,
especially for the high W region that corresponds to the low x-low Q2
events with E<0.8 GeV. See also the discussion below about the last plot
of this report.

Next, there is the question of false asymmetries plaguing the 5.9 GeV 80
deg. data, which can clearly be seen in Illustration 4. But the other
three configurations seem free of the false asymmetry. Just to make
sure, I replotted James' data for the parallel runs using a visible
vertical scale, because the original plots are inconclusive (presenting
data using appropriate scales is mandatory for all analysis group
members). It seems that only the 5.9 GeV perp data have the problem.

Before making any conclusions, we need quantitative and qualitative
evidence of this:
- we need new plots, with the same cuts, for Illustrations 3 and 5,
showing A_phys and Pb*Pt, and we need to plot and analyze the
distribution of asymmetries about their means for the four cases, to
test whether they correspond to a single population or to two
populations. I suggest James to do this for para and Jon for perp.

Finally, on the last page of the last report, I've plotted the physics
asymmetry comparing single cluster and any number of clusters for all
low E cer_h = 0 asymmetries. There is a trend for the single cluster cut
to increase the absolute asymmetry relative to no cut, especially for
perp data (different at 2 sigmas), Illustration 6.

But what is very interesting is the final plot, which compares low e
single cluster with cer_h = 0 vs no cer_h cut. There is some sensitivity
for the perp data, but there is a giant difference between the two cuts
for parallel: a change of sign and a factor of three or four with the
cut vs without.

This change for parallel data can be explained in terms of the no cut
events including all values of cer_h. As we know, the single clusters
with cer_h>0 data are dominated by real DIS electrons that have a huge
40-50% asymmetry, which is combined here with the negative asymmetry of
the events with cer_h = 0.

On the other hand, for perp data we may be looking at a significant
negative asymmetry of single clusters with cer_h>0 that is canceled by
the positive asymmetry of the far more numerous cer_h = 0 data. The
difference is not insignificant: cer_h=0 vs all cer_h asymmetries differ
by more than 2 (combined) sigmas.

Another possibility for perp is that the asymmetry with cer_h>0 is also
positive, and we are seeing the average of about equal numbers with
cer_h = 0 and with cer_h >0, with the latter having a slightly smaller
asymmetry as the combination, but this is inconsistent with the actual
number of cer_h > 0 events.

In either case, an improved cer_h definition could increase our sample
of good DIS electrons considerably, moving them from the cer_h=0 sample
to the >0 one, assuming we figure out the two questions I asked above,
about low energy clusters with more than 4 cells, and why their
asymmetry is half of that with less than 5 cells.

So everyone is encouraged to investigate the issues further. Specific
projects are:

- make new plots of all asymmetries with the current cer_h definition
using the cut cer_h>0, for all combinations of cuts on No of clusters
and E<0.8 or E>0.8 (but always trigger type 4). This should improve a
bit our sample of what we now take as good DIS electrons (I say DIS
because the asymmetries have a clear correlation with Pb*Pt). No charge
or livetime normalizations should be applied yet.

- replay all parallel and 4.7 GeV 80 deg data with the same analyzer
used for the latest 5.9 GeV 80 deg replay. No need to replay again the
latter, but check that the old and new engines (with the BCM fixes)
produce the same raw asymmetries.

- compare the distributions and asymmetries for one or a few runs after
systematically commenting out each of the conditions tested when
calculating the cer_h code. In particular, compare cer_h = 0 without any
ncell condition (i.e. cer_h=0 fails only the timing and geometry cuts)
to cer_h as is, with single cluster and E<0.8 cuts. According to
Hovhannes, ncell = something is needed to find clusters, and/or the
ntuple size may become much larger without an ncell condition. Both
possibilities should be confirmed.

- Hovhannes has already replayed runs 72487 to 72531 with ncell=2 in the
cer_h definition. The ntuples are on the /lacie disk on octavian. I
suggest Jon and James to run their asymmetry building codes on these
ntuples, and apply the different cut combinations, to compare with the
standard cer_h.

- replay and post plots and tables of the raw asymmetries for C, C+He,
He runs for all energies and field angles.

I suggest not waiting until just before the weekly meeting to circulate
or post your results. Discussing work in progress is useful and helpful,
saves time and wasted efforts. Share it with whoever may have useful
feedback.

I would also encourage group members not working on the BETA analysis,
both doctoral students and senior collaborators, to contribute by
investigating possible additional approaches to the issue. Anyone with
paw or paw++ knowledge can participate. All the replayed data are
available in ntuple form and text scaler reports on the octavian server,
which is part of JLab's CUE system. The folder is
/mnt/raid/exp/SANE/analysis/replay


Cheers,

Oscar