Παρουσίαση του PowerPoint

Download Report

Transcript Παρουσίαση του PowerPoint

Thoughts on ECAL ‘calibration
stream’
D. del Re, P.Meridiani, R. Paramatti,
Sh. Rahatlou, F. Santanastasio
Introduction
Calibration is a key issue for ECAL
it affects directly the energy resolution
can limit the ECAL physics performance
Several methods are foreseen to work in conjuction to obtain the final
calibration coefficients:
•-simmetry
•0/0
•We (10Hz @ 2x1033 after HLT )
•Z → ee (1 Hz @ 2x1033)
Laser monitoring is not treated here since it is considered a special/separate
monitoring task
Methods can be divided in
• online calibration methods (which needs to access information before
HLT decision)
• offline methods (which can work on standard HLT streams)
Paolo Meridiani – INFN Roma1
2
Motivation for a calibration stream
Typically ECAL calibration tasks need access only to calorimeter
hits in a particular area of interest (around the candidate
electron/photon/pi0)
Also access to fitted tracks/hits in a cone around candidate particle
is needed
Both offline and online calibration tasks need acces to a data
format which should be be as light as possible
In particular online calibration methods needs a filter application
able which should run on the filter farm. Constraints on the readout
rate / computation timings that CMS filter farm can devote to this
type of job should be understood (question to filter farm experts)
Paolo Meridiani – INFN Roma1
3
ORCA electron calibration
stream
The ORCA calibration stream implementation was
EgammaCalibObject,Level1Trigger,(EgammaMC)
Package: ElectronPhoton/EgammaCalibObject (G. Daskalakis, P. Meridiani, P. Govoni)
The EgammaCalibObject is constructed starting from each ElectronCandidate (L3 Candidate) and contains:
SC information (E,PT,,)
Associated track information (E,PT,,, number of valid hits, charge, trans. & longitud. impact parameter,…)
vector<CaloRecHit> contained into an array (21- x 11-) around the max. containment crystal
MC truth (formatted as an EgammaMC object) was also available
Paolo Meridiani – INFN Roma1
4
CaloRecHit map
11 x 21 matrix
Paolo Meridiani – INFN Roma1
5
Past experience with ORCA



In ORCA for the Physics TDR studies it was
needed to further dump the information from
the calibration stream to another format (ROOT
or ASCII). This is infact one of the known
problems of the past EDM
Further data reduction means that afterwards
you have no possibility of rereconstruction
Physics TDR calibration studies on single
electron and Z calibration however
demonstrated that a single calibration stream
type can be used for offline electron calibration
Paolo Meridiani – INFN Roma1
6
What will be in CMSSW

Guideline (also following proposal from Y. Gerstein)
should be to write just a subset of the existing collection





e.g. a collection of EcalUncalibratedRecHits in a certain
calorimeter area
In principle no need to have a further data reduction step,
since CMSSW data formats are lighter than ORCA and EDM
is root based
Clear advantage of reusing standard reconstruction
algorithms (clustering, electron...) on reduced collection
to process the calibration stream
We only need filter modules which put into the event
reduced collection of hits
Conceptually the filter modules would be similar for
calibration types.
Paolo Meridiani – INFN Roma1
7
Proposal for pi0

Currently understudy use of L1 objects to
identify regions of interest for isolated pi0
sample


Results to be shown at future egamma meetings
Goal




Select very small regions of ECAL containing pi0
Store only interesting crystals in output
Perform clustering and pi0 selection on subset of
crystal
Perform in situ calibration with precision of ~1%
Paolo Meridiani – INFN Roma1
8
Pi0 Selection Based on L1 Objects

Look at 4 L1 EM objects in barrel (||<1.4)


Looking for hits inside a matrix of
20x20 around L1 object



So far using QCD jets with 30<pT<170
Non-overlapping by construction
About 0.5 0/matrix
with E(g)>0.5 GeV
Find 3x3 clusters within each
matrix with seed xtal with
E>0.5 GeV
Paolo Meridiani – INFN Roma1
9
Current Rate Estimate for Pi0

10kHz L1 input
x 4 L1 EM objects
x 0.5 [require N(xtal with energy in matrix) < 50]
x 25 <xtals> read in the 20 x 20 matrix (SR)
x size of EcalUncalibratedRecHit (20 bytes)
 10 MB /sec

Estimate IF we want to include only Ecal
information in the same format used by
standard reconstruction
Paolo Meridiani – INFN Roma1
10
Additional Considerations for Pi0
Estimate

No cluster level selection applied in current
scheme before storing output


Tracker information



If applied we could further reduce output rate
For the moment NO TRACKER information used to
clean up or veto matrices
Storing slice of tracker corresponding to L1 object
desirable IF fits within allocated bandwith and disk
budget
Calib Format

We strongly prefer using the reco data objects rather
than new specialized ones
Paolo Meridiani – INFN Roma1
11
Electron calibration stream

Logically electron calibration stream application seems to be
well defined




Calibration stream is produced offline starting from HLT
single/double electron stream
It will look at offline calibration electrons, having the possibility to
apply electron id requirements to further reduce jet background
Mantain calorimeter hits in a certain window around electron.
Window size for calorimeter hits (in ORCA 11() x 21() ) can be
further optimized.
Need to answer which tracker hits to store
• Only hits associated to the electron track (allows simply refitting when
new alignments are available)
• Hits in a “small” area around electron candidate (allows also to redo track
building step with new alignments)


Event size would be around 1-2 Kbyte x electron. Not critical as for
pi0 calibration stream
The same stream can be used both for single electron and Z
calibration
Paolo Meridiani – INFN Roma1
12
Calibration application in
CMSSW

A calibration application needs typically to run
over the calibration stream for more than one
loop in one single job

This possibility has been already taken into
account by CMSSW framework developers,
following a question on hypernews from P.
Govoni
Paolo Meridiani – INFN Roma1
13
New EDLooper object

Proposal from Chris Jones for an EDLooper
(https://twiki.cern.ch/twiki/bin/view/CMS/FwEdmProposalForMultiplePassesPerJob)
As per the agreement from our conference meeting today, here is an API proposal for
a new class of EDM 'module' which can be used to control 'multi-pass' looping over
an input sources data.
class EDLooper {
enum Status {kContinue, kStop}
//called once per job just before the first processing
virtual void beginOfJob(const edm::EventSetup&);
//called each time just before starting to loop over the events in the source
// the argument is the loop count, starting at '0'
virtual void startingNewLoop(unsigned int ) = 0;
//called for each event in the source
virtual Status duringLoop(const edm::Event&, const edm::EventSetup&) = 0;
//called after the last event in the source has been processed
virtual Status endOfLoop(const edm::EventSetup&) = 0;
//called just before the job is going to end
virtual void endOfJob(); };
The EDLooper would also have the ability to add data to the edm::EventSetup, just
like an ESProducer. This would allow one to change calibrations/alignments in the
'startingNewLoop' phase
I also propose extending the configuration language to allow loading and configuring
such EDLoopers
looper = TrackBasedAlignmentLooper { .... }
So classes inheriting from EDLooper would take an 'const edm::ParameterSet&' in
their constructor which can be used to configure the looper.
Paolo Meridiani – INFN Roma1
14
Conclusions

Guideline for CMSSW implementation of Ecal calibration
stream would be to write filtered/reduced collection of
hits

Filter modules can start to be modeled. Electron stream is
almost well defined, more thoughts are needed for
pi0/eta0

We need to understand which are the constraints for data
volume and computation time for a filter module running
on the filter farm

We have to also define which are the objectives for ECAL
calibration in the CSA 2006. For example which datasets
do we want to be simulated? Which exercises do we want
to play?
Paolo Meridiani – INFN Roma1
15