Transcript LCD Analysis with JAS and ROOT

Full Event
Reconstruction in
Java
(“From 0 to analysis in 15 minutes”)
Norman Graf (SLAC)
for the LCD development team
CHEP '04
Interlaken, Switzerland
Sep. 30, 2004
International Linear Collider (ILC)
 Detectors
designed to exploit the physics
discovery potential of e+e- collisions at
s~1TeV.
 Will perform precision measurements of
complex final states.
 Require:




Exceptional momentum resolution.
Excellent vertexing capabilities.
“Particle Flow” calorimetry.
Hermeticity.
Mission Statement

Provide full simulation capabilities for ILC physics program:




Need flexibility for:






Physics simulations.
Detector designs.
Machine-Detector interface & backgrounds.
New detector geometries/technologies.
Innovative reconstruction algorithms.
Limited resources demand efficient solutions, focused effort.
Desire a common simulation environment used in ILC studies
which allows sharing of detectors, algorithms, and code.
The system should be flexible, powerful, yet simple to install,
maintain and use.
Metric of performance:
Ease with which physicist having an idea can implement and test its effect!
Most important tool of all!
Why Java?

Java is a pure Object Oriented Language.

Simpler to learn and use than C++ :




Language design emphasizes ease-of-use over
performance.
Garbage collector takes care of freeing unused objects.
New language, no historical baggage.
Very powerful standard library.
– Large number of open-source libraries including libraries for
scientific computing.


Physicist gets to concentrate on writing clean OO
code to perform analysis tasks.


Platform independent, compile once just runs
everywhere (Linux, Windows, Mac OSX, Solaris, …).
Not understanding core dumps and learning difference
between a pointer and a reference.
Performance of Java code is close to that of C++.

In real life maybe 20-30% overhead typical.
Why Java?

Several very powerful,
free, IDE’s now
available:




e.g. Netbeans, Eclipse.
Support editing, code
completion, GUI
building, debugging,
performance profiling,
refactoring, CVS, etc…
Multiplatform.
Mantra “I prefer
EMACS” is becoming
hard to justify!
Why a full reconstruction now?

ILC  LEP
Much higher luminosity, energy, precision.
 More complex events (not on a resonance).


ILC  LHC

Detector R&D focused on orthogonal issues.

e.g. rad hardness not an issue, tracker material is!
Desired precision and resolutions require a
new detector paradigm, intimately connected
with the reconstruction: “Particle Flow”.
 Aim for very tight loop

 design  simulate  reconstruct  analyze 

Particle Flow Motivation
Measure momenta of charged tracks in the tracker
with superb resolution.
 Measure photons in highly segmented EM calorimeter
with reasonable resolution.
 Remaining neutral hadrons measured in hadron
calorimeter.

E jet  Echarged  Ephotons  Eneut.had.
2
 E2jet   E2charged   E2photons   E2neut.had.   confusion
2confusion
is the hardest term, cannot be correctly
simulated with fast 4-vector smearing. Needs detailed
calorimeter shower simulation and reconstruction to
understand.
Imaging Calorimeter
EMCal:
dense, small Moliere radius
highly segmented:
laterally (~30-40 layers)
longitudinally (~5x5 mm2)
HadCal:
emphasize segmentation
& granularity over
intrinsic resolution.
Digital readout?
ILC Detector Simulations
1.
2.
Use simplified volumes to investigate larger
number of designs, allowing dynamic
topologies using single executable.
Describe “optimal” detectors in great detail.
TPC Tracker, Si Disks, CCD VTX
All Si Tracker, CCD VTX
Java Reconstruction/Analysis Overview


Runs standalone or inside Java Analysis Studio (JAS3).
Fast MC provides target (“best you can do”).

Smeared tracks and calorimeter clusters, fiducial acceptance.
Beam & Physics background overlays at hit level.
 Hit digitization allows detector readout schemes to be varied,
provides more realistic hits (merged, ghost, pos ).
 Full Reconstruction:







ab initio track finding and fitting.
calorimeter clustering.
Individual Particle reconstruction (cluster-track association).
Java Neural Net available for training & use.
Physics Tools (Vertex Finding, Jet Finding, Flavor Tagging).
Analysis Tools (including LCD-specific WIRED event display).
Event Reconstruction Flow
Sequential Record Loop
LCEvent
Reconstructors implemented
Bckgnd
G4 Detector Simulation Input +
SimTrackerHits, SimCalorimeterHits, MCParticles
Events as Drivers with hooks, e.g.:
add(Processor p)
Detector Digitization/ Hit Finding processEvent(EventData evt)
CCD Pixels, Si strips*, TPC waveform*
Tracker hit clustering, uncertainty assignment
beforeFirstEvent()
afterLastEvent()
Track Finding/Fitting
Standalone 3D trackfinding (e.g. TPC, CCD)
setDetector(Detector det)
Track merging, fitting
…
Calorimeter Clustering
“Nearest” Neighbor, MST*, Fuzzy*
Drivers implement Processor
Cluster Particle ID (e.g. , , h , h )
interface, so can be nested.
Reconstructed Particle
Track-Cluster association,
Event contains some
predefined collection hooks,
Jet-Finding, Flavor Tagging
evt.getCalorimeterHits()
ZVTop vertexing, NN Flavor Tagging (b, c, uds)*
Arbitrary objects can be added
Reconstruction Output
to/retrieved from Event
LCIO (Tracks, Clusters, ReconstructedParticles…)
AIDA (tuples, histograms)
evt.put(“myStuff”, myStuff)
±
0
Detector Hit Digitization

SimCalorimeter quantized into cells in Geant4.


SimTrackerHit has full MC info: x, t, dE/dx…


Study readout segmentation by ganging cells.
Digitized at reco level: e.g. strip pitch, pixel size, charge
sharing, electronic noise, …
Provides more realistic simulation of effects of
backgrounds, efficiencies and noise.


Nearby hits merged, ghosts simulated.
Hit-specific measurement uncertainties.
Can add Random Hits (salt & pepper backgrounds).
 Efficiencies simulated by dropping hits, channels.

CCD Digitization
VXD hits from simulated events, finds charge
deposited in each pixel, adds electronics noise and
digitizes signal.
 Finds CCD clusters, splitting if necessary.
 Coordinates of found centers are used to replace
TrackerHits in the events. Further event processing
(track finding, fitting, and so on) proceeds the same
way as it was before.
 Can set CCD parameters (like thickness, depleted
layer depth, epitaxial layer thickness and so on),
electronics parameters (noise, ADC conversion scale,
pixel and cluster thresholds), processing parameters
(like cluster center calculation method).

Tiling Forward Disks
Wedges
Stereo Angle
Hexagons
Track Finding/Fitting
Full pattern recognition in 3D detectors.
 Extrapolations of found tracks to 2D det’s.
 Tuned for Large and Silicon Detectors.



Working on generalizing this.*
Track Fitters:
SLD Weight Matrix Fitter.
 Single Detector or Combined fit (e.g. VTX+TPC).
 Kalman Fitter*


MIP Tracking in Cal (’s, track starting).
Calorimeter Clustering

Cluster Interface defined, several clustering
algorithms currently implemented.

MC “Cheater”, “Nearest” Neighbor, Fixed-cone, …
Cluster QA package developed.
 Cluster Refinements:

Combining HAD + EM clusters.
 Endcap + Barrel overlap region.
 Identify cluster as originating from charged or
neutral hadron, EM.

Reconstructed Particle/ Jet Finding
Particle Flow algorithms being developed with
minimal coupling to specific detector designs.
 Photon and muon reconstruction fairly mature.
 Emphasis on track-following for charged hadrons.


MIP reconstruction quite promising.
Canonical data samples identified and will be used
to characterize detector response.
 Systematic investigation of jet-jet mass as a function of
BnRmaplq (B-field, Cal radius, Cal cell area, Cal
longitudinal segmentation), material and readout
technology employing a Particle Flow paradigm
being undertaken.

JAS3 IDE
Java, Python, Pnuts editor
Dataset, Program, AIDA Explorer
Event
Scripting Console
Browser
Histogram &
ntuple analysis,
fitting

Tutorial: http://jas.freehep.org/jas3/Tutorial/index.html
WIRED Event Display
Tree controls
visibility. (HepRep2
support in WIRED3
is preliminary)
Picking supports
viewing physics
attributes
Next Generation (org.lcsim) Goals

Retain core functionality from hep.lcd package.

Full suite of reconstruction and analysis tools available to
all LCIO users.
Update to use LCIO for I/O and as basis for simulation,
raw data and reconstruction event formats.
 Update/simplify framework using experience from
hep.lcd.
 Internationalization:




Update to Java 1.5.


Try to make package independent of detector & geometry
assumptions so can work with any detector.
Read properties of detectors at runtime.
Many improvements since hep.lcd framework was created.
Revitalize work on reconstruction algorithms.
Conclusions
ILC Detector design is an area of active
development with many design choices,
requiring close coupling of design, simulation,
and reconstruction.
 Ease of use and speed of development are
essential for physicists conducting ILC studies
part-time & in a heterogeneous environment.
 A fairly complete suite of simulation tools
written in Java exists and is being successfully
used for these studies.

Links
linearcollider.org
 ILC Simulations: http://lcsim.org
 hep.lcd: http://www-sldnt.slac.stanford.edu/jas/Documentation/lcd/


Tutorials:







http://jas.freehep.org/jas3/Tutorial/index.html
http://www-sldnt.slac.stanford.edu/snowmass/Welcome.html
JAS3: http://jas.freehep.org/jas3
WIRED: http://wired.freehep.org/
HepRep: http://heprep.freehep.org/
LCIO: http://lcio.desy.de
Discussion Forums:


http://forum.linearcollider.org (linear collider specific)
http://forum.freehep.org/ (general JAS, Wired etc)
Related Talks
248 - LCIO - A persistency framework and data
model for the linear collider, Frank Gaede
395 - WIRED 4 - A generic Event Display plugin
for JAS 3, Mark Dönszelmann
397 - AIDA, JAIDA and AIDAJNI: Data Analysis
using interfaces, Victor Serbo
401 - Interactive Data Analysis on the Grid
using Globus 3 and JAS3, Tony Johnson
402 - Writing Extension Modules (Plug-ins) for
JAS3, Mark Dönszelmann