Transcript ppt

Efficiencies and backgrounds
re-evaluation
ANKARA CM 2/4/2009
D.Autiero, S.Dusini
1
The re-evaluation of OPERA background and efficiencies is asked by the
scientific committees who would like to know what we have learnt with one year
of real data.
For years we have been showing efficiencies and backgrounds estimated at the
time of the proposal or during the year after the proposal submission.
This work can be based on the following handles which were not available in
the past:
Availability of real data and possibility of measuring directly on them
backgrounds and efficiencies
New MC production with final geometry of the experiment, detailed simulation
of the events at emulsion level with the same reconstruction as for data
Possibility of validating the MC simulation with real data in order to have a
credible knowledge of the variables and the effects of the cuts
Improvements in the analysis: more sophisticated analysis including the merging
of scanning data + electronic detector data (many things at the proposal level
where estimated almost at the generator level) and taking into account the
correlations among different steps. New analysis approaches with optimized cuts
aiming at maximizing the sensitivity of the experiment. Multivariable likelihood2
approaches.
Going from the conservative cuts of the proposal to these last two points is
possible only if we show that we have under control MC vs Data the variables
used for the analysis and we can squeeze their discriminating power without
creating artefacts and believing in corners of the phase-space which are not
really existing in real data.
In OPERA we have a factor 10 between tau interactions in the bricks and
candidates selected at the end of the analysis. We should see how this could
be optimized given the experience from real data.
The re-evaluation implies:
 New MC production with state of the art knowledge
 Data ED and MC reconstruction with final version
 Reconstruction of MC emulsion data with OpEmurec
 Availability of emulsion data and merging with ED data by using OpEmurec
In particular there are requirements on the quality and the kind of samples of
emulsion data which should be made available for the analysis
3
Goal: Timescale to complete the re-evaluation work: at latest the end of the year
Data availability:
OpEmurec is almost in final shape. Succesful first reconstruction test down to
vertex, dedicated meeting last week on first steps Alignment  Luca
During after Mitzunami we got a sample event of Japanese data. An interface was
built in order to put it in the standard DB format and this event has been rereconstructed offline by Cristiano using sysal. A second event was provided last
week to complete testing. Tools have been provided to Nagoya in order to handle
the bricks publication.
So we are also going towards the full integration of data from Japan. 
Elisabetta
It is a first priority job to get it interfaced also to the MC emulsion data (+
background from real data) in order to perform full MC studies as for real data
 Luca
The new MC production was started at CCINP23 at the time of the last PC
Data reprocessing will be following  Elisabetta + Stefano
4
Bulk efficiencies:
Trigger
Target event selection
Brick finding
CS tagging
Geometrical efficiency
Vertex location
B2B
Long and Short decays
5
Trigger efficiencies:
During 2008 a cut at 10 hits was left in the DAQ manager. It penalizes the
tau>e QE events.
The situation has to be improved for 2009 with a smarter cut based on pulse
height and relaxing the 10 hits.
A special run has been taken with 4 hits threshold  factor 2 larger data flow,
random coincidences with veto (affordable), timeouts of DAQ manager (being
investigated)
 re-evaluation with MC of trigger efficiency, optimization of cuts. (T.
Brugiere)
Caveat: in data there are spurious hits and x-talk not reproduced by MC, this can
increase the real efficiency. Try to estimate this bias.
Checks on data (also useful for other aspects of this re-evaluation): check with a
sample of rock muons that the number of p.e. is well reproduced DATA/MC,
check on these tracks the inefficiency: how many times there are missing hits
( Cecile final results)
Implementation by Stefano in OpFilter of a package simulating for MC data the
6
trigger conditions in DAQ and removing hits not satisfying the trigger conditions.
Fiducial volume selection efficiencies:
This efficiency was never taken into account so far.
2007-2008
The algorithm of Alessandro/Tiem is mature now and it can be used for automatic
selection on real data. It has a few per mille inefficiency on low energy NC
events with respect to visual scanning. These events are at the borderline and
partially doubtful
 Re-evaluation with MC samples of these efficiencies with final version of the
algorithm (OpCarac)
Show with Data vs MC that we have under control the background from external
interactions
7
Brick finding and CS tagging
Getting data from second bricks to complete efficiency estimation from real
data. At the moment still poor statistics of 2nd bricks (data from EU only)
Last data:
First bricks raw eff: 467/715 = 65.3%
Second bricks raw eff: 29/65 = 44.6%
Removal of interactions in dead materials 715 - 3.8% = 688
 1st brick corrected eff: 467/688 / (1 -6.8%) = 72.8%
(72% MC)
Correlation with 2nd brick:
Unfound raw after first brick (100% - 65.3%)
Unfound related to BF inefficiency (100% - 72.8%) foundable in 2nd brick
 Second brick corrected eff: = 29/65 * (100 – 65.3) / ( 100 - 72.8) = 56.9%
1st brick BF eff = 72.8% +- 1.7%
2nd brick BF eff = 56.9% +- 6.1%
Total efficiency 1st + 2nd: 72.8% + (100% - 72.8%)*56.9% = 88.3 % +- 5%
(80% MC)
Still too large uncertainty on 2nd bricks
8
Comments and warnings:
1) Predictions were made time ago for these events. Taking the latest
version of the reconstruction for about 1/3 (9 events) now the brick is
predicted directly as first brick, reconstruction problems fixed
2) For 8 events the second brick suffers from the underestimation of
tracking errors. This has to be fixed in order to predict correctly the
lateral probability
3) For 8 events the second brick was in another wall with large probability
(30-40%), this looks higher (~ a factor 2) than on average prediction
for second bricks. It is a fluctuation in this first sample
4) Too few NC
 We need a larger sample to draw some more solid conclusions
We should get a large sample from the last couple of weeks of extractions
9
We should have a clearer view with the increase in statistics of the last
weeks (gather all possible second brick results, also from Japan)
 Re-evaluation of last version of the reconstruction on all events
 A deeper comparison per events categories is needed (QE,NC,CC,
electromagnetic-like)
 So far the algorithms have been kept untouched, the main work has been
concerning the debugging of the reconstruction. On the basis of these results
it will be possible to have a second version
BF efficiency was optimized for tau events (the efficiency is about 5%
larger than for standard numu NC or CC events) , in order to check it we
should look for events which are tau like CC QE and with low energy muons,
events with dominant electromagnetic activity, like taue and taurho.
 First priority: increase the efficiency as much as possible, There are ideas
about that on the side of BF and also on the possibility of concentrating the
efforts on a tau signal enriched sample. We should first advance with the
analysis of the event sample and understand why it goes wrong
10
(reconstruction, 2nd bricks, …)
Re-evaluation campaign (brick finding CS tagging part):
Integrate in the simulation the CS tagging absent in the past: acceptance
effect with respect to the total surface behind the bricks (7.4% uncovered
area)
+ CS base track efficiencies measured from real data, implement also the
3/4)
 Complete efficiency evaluation from real data (convolution of BF + CS)
 Evaluate CS tagging efficiency from full MC (including the CS intrinsic
efficiency as a function of the angle measured in real data, provided by
Giovanni for base tracks)
11
 Better errors evaluation in tracking, important for probability maps
Re-evaluate MC BF efficiency per categories of event including the interplay
of CS tagging efficiency, compare to data per categories of events: DIS, QE,
NC, NC with e.m. component.
 Possible tau enriched sample to maximize the efficiency
 Estimate possible bias of CS tagging inefficiency (~ 25%) on muon matching:
are all muons found at the vertex even if not followed by scan-back ? even a
residual inefficiency of 2% could change a lot the charm background.
12
Geometrical efficiency
At the proposal level it was simply assumed that 1mm along the border was
inefficient at the level of emulsion scanning a 3.6% loss.
Giovanni thinks that the dimension of the inefficient region is actually reasonably
close to 0.5 mm.
We have also to take into account the fact that the lead plates seem to have a
different geometry with respect to the nominal one: 137.9 g/plate vs 142.57
g/plates (nominal) a loss of 3.3%. Where is the missing mass ? The reduction
in surface of the plates (124.96 x 99.35 vs 125.5x100: -1.16%) does not justify
the effect. In The MC prod we applied also a thickness reduction of 20
microns (Stefano)
 we have concluded at the last PC meeting that the thickness reduction of 20
microns was real (BAM and lead production presentations)
 update also the target mass and number of interactions (in progress)
 evaluate with the full MC the real loss of efficiency for events at the borders
(need to include in MC final lead geometry) Emulsion dimensions 124.6 x 99.0
13
B2B connection
Now we should have experience from real data. The statistics is still poor,
in Europe we have 32 events with frontal connection, very few events
with lateral connection
 re-evaluate with full MC
 increase statistics with data measurement
14
Vertex localization (Giovanni)
This is another point where people expect some experience return from real
data. At the moment the statistics is poor and can provide just a range for
the efficiency:
CC: 84-94%
NC: 71-92%
The lower limit of the range is computed assuming that the actually pending
events are not localized.
 Improve the statistics with data
 Launch the evaluation of efficiency (quite relevant for NC like) with full MC
by performing detailed reconstruction
 Study of multiplicity distributions at the primary
 How to prove that we understand the efficiency for events with
electromagnetic showers
 How to proove that we understand the efficiency for secondary vertices ?
15
Try a K0s sample with one track detected in CS, rates well known
How to prove that we understand the efficiency for NC events
Fake NC removing the muon
CS
How to prove that we understand the efficiency for events with
electromagnetic showers
Tagging in the CS of known showers
from photon conversions
CS
 How to proove that we understand the efficiency for secondary
vertices ? Try a K0s sample with one track detected in CS, rates well
known
16
Tracks slopes at primary vertex for CC events
Tracks at large angle
Slope zx
Slope zy
File provided by Giovanni with 196 vertices, 166 CC
Beam angle
17
Track multiplicity at primary vertex for numu cc
Single track events 24%
Expected contribution from QE+RES
10%
 To be quantified losses of tracks for
DIS (efficiency for large angles, range for
soft particles)
18
Muon slope in xz plane
Data
MC
19
20
Muliplicity at primary vertex
Data
QE+RES
(10%) not
included but
taken into
account in
normalization
MC
Protons <300 MeV/c (mainly from nuclear rescattering removed)
21
22
All tracks slope in zx plane
Data
Angular efficiency
MC
23
Distributions already spoiled by deficit of tracks seen at the multiplicity level
24