GLAST Large Area Telescope

1. Gamma-ray Large Area Space Telescope GLAST Large Area Telescope AGILE - GLAST Workshop July 2, 2007 Simulation & Reconstruction Overview Francesco Longo University and INFN, Trieste on behalf of the GLAST LAT collaboration

2. Goals of this Talk • Overview of Simulation/Reconstruction: • Introduce the simulation and reconstruction for LAT events • Focus on general structure of the framework • Performance Studies and Analysis based on Reconstruction • Provide background for tomorrow’s talks: • Background rejection, IRF studies, Service Challenge simulations

3. Introduction

4. EGRET and GLAST EGRET (1991-2000) Phases 1-5 • Spark chamber • sense electrode spacing ~mm • sensitive layer depth ~cm LAT (2007- >2012) 1-yr simulation • Si-strip detectors • sense electrode spacing ~0.2mm • better single hit resolution • sensitive layer depth ~0.4mm • converter proximity to minimize MCS Cygnus region (150 x 150), Eg > 1 GeV

5. e– e+ Pair Conversion Technique  The anti-coincidence shield vetos incoming charged particles. photon converts to an e+e- pair in one of the conversion foils The directions of the charged particles are recorded by particle tracking detectors, the measured tracks point back to the source. The energy is measured in the calorimeter Tracker: angular resolution is determined by: multiple scattering (at low energies) => Many thin layers position resolution (at high energies) => fine pitch detectors Calorimeter: Enough X0 to contain shower, shower leakage correction. Anti-coincidence detector: Must have high efficiency for rejecting charged particles, but not veto gamma-rays

6. design of the experimental set-up evaluation and definition of the potential physics output of the project evaluation of potentialrisks to the project assessment of the performance of the experiment development, test and optimisation of reconstruction and physics analysis software contribution to the calculation and validation of physics results The scope of Geant4 encompasses the simulation of the passage of particles through matter there are other kinds of simulation components, such as physics event generators, detector/electronics response generators, etc. often the simulation of a complex experiment consists of several of these components interfaced to one another The role of simulation Simulation plays a fundamental role in various domains and phases of an experimental physics project

7. gamma ray proton Simulation development Detailed detector model includes gaps, support material, thermal blanket, simple spacecraft, noise, sensor responses… • The LAT design is based on detailed Monte Carlo simulations. • Integral part of the project from the start. • Background rejection • Calculate effective area and resolutions (computer models now verified by beam tests). Current reconstruction algorithms are existence proofs -- many further improvements under development. • Trigger design. • Overall design optimization. • Simulations and analyses are all C++, based on standard HEP packages. Instrument naturally distinguishes gammas from backgrounds, but details matter.

8. Simulation and Analysis Simulation Real LAT Data gamma-ray sky model background fluxes Particle Generation and Tracking Instrument Response(Digitization), Formatting Trigger and OnboardFilter (wrapped FSW) Event Reconstruction Detector Calibration High-level Science Analysis Event Classification Performance

9. GLEAM

10. Framework Gaudi environment Simulation Overview of the flow of simulation LAT Geometry as used in simulation and reconstruction Simulating events with Geant4 Digitization Trigger Reconstruction Overview and flow of the reconstruction Calorimeter Reconstruction Clustering/Moments Analysis Energy correction methods MIP Finder Tracker Reconstruction The basic steps ACD Recon “Analysis” and Output Simulation and Reconstruction

11. Simulation

12. Simulation Overview • Goal of simulation is to produce “raw data”, representing the response of the LAT to a given event, which can be fed to the reconstruction • Logically breaks into four steps: • Event generation – “source particle” • Simulation of LAT response to the generated event • Digitization: the conversion of simulated LAT response to “raw data” • Simulation of Trigger

13. Output Data New Event EventGenerator (FluxAlg)(PointingAlg) Event Simulation(G4Generator) Digitization(ACDDigiAlg)(CalDigiAlg) (TkrDigiAlg) Trigger / Filter(TriggerAlg)(OnboardFilter) Output of digis given to Trigger and Onboard Filter Events MUST pass Trigger in order to get reconstructed Onboard Filter results recorded, events not rejected Sources, Orbit Orientation, Etc. Geant4Event/DetectorSimulation Convert simulated detector response to “raw data” (calibration effects enter here) LAT Geometry(G4GeometrySvc)(also accessed by the reconstruction) Flow of Simulation Algorithm Structure for DC-2

14. Background Flux DC2 (2006) • Background Flux Review • J. Ormes et al., LAT-TD-08316-01 • Albedo e+e- flux a factor >3 larger than for PDR. • Primary cosmic proton flux is higher • New Albedogflux Updated integrated flux 13000 Hz/m2 PDR flux ~4200 Hz/m2 CDR & PDR (2000)

15. Astrophysical Sources • DC2 sky in galactic coordinates Plot by Seth Digel

16. Structure for celestial source simulators • The development of a “celestial source” model for GLAST/LAT software implies: • Capability to use the models within the full Monte Carlo of the LAT instrument (GLEAM) • Capability to use the models within “observationSim”, the “science simulator”. • Source simulators • the GRB physical model • GRB phenomenological model • High level Pulsars simulations (“polar cap” - “outer gap model”) • Sources have been using by the GLAST community and are included in the next Data Challenges.

17. LAT GeometrySimulation and Reconstruction

18. LAT GeometrySimulation and Reconstruction

19. Geometry Repository • A complex instrument: GLAST geometry is formed by more than 40000 different kind of volumes. • A typical problem in HEP experiments (and GLAST is not so big) • The geometry database in GLAST is in XML, a quite common choice (LHCb, Atlas and more) today • detModel is a set of C++ classes to parse and represent in memory such XML description • Used by various clients (reconstruction algorithms, MonteCarlo simulation, graphics etc. etc)

20. Simulation: Based on GEANT4 High-energy g interacts in LAT Geometry Detail Over 45,000 volumes, and growing! Includes: tracker electronics boards mounting holes in ACD tiles spacecraft details and much more Interaction Physics QED: derived from GEANT3 with extensions to higher and lower energies (alternate models available) Hadronic: based on GEISHA (alternate models available) Propagation Full treatment of multiple scattering Medium-dependent range cut-off Surface-to-surface ray tracing. Includes information from actual LAT tests detailed instrument response dead channels noise etc. Overall Deadtime Effects Black: Charged particles White: Photons Red: Deposited energy Blue: Reconstructed tracks Yellow: Inferred γ direction

21. Geant4 • Geant4 is the successor of GEANT3, the world-standard toolkit for HEP detector simulation. • Geant4 is one of the first successful attempt to re-design a major package of HEP software for the next generation of experiments using an Object-Oriented environment. • A variety of requirements have also taken into account from heavy ion physics, CP violation physics, cosmic ray physics, astrophysics, space science and medical applications. • In order to meet such requirements, a large degree of functionality and flexibility are provided.

22. Instrument Response • We turn the energy deposit given by GEANT4 into the signals that we would record in the detectors: • Tracker: • tower triggers • hits strips when energy is above threshold • time-over-threshold ORs with correct gains • Calorimeter • correct sharing of signal between two ends of crystals (attenuation) • signals in small and large diodes, each with two ranges • Anticoincidence Detector • signals from tiles to both phototubes • correct sharing of signals between two ends of ribbons (attenuation)

23. Digitization • Digitization: Convert Monte Carlo information to detector response • Simulates the response of the detector to the simulated event • TKR • Input: McPositionHits – position/energy in active silicon • Output: hit strips and ToT (Time over Threshold) • Some Details: • Implementation of “bad” (hot/dead) strips • Implementation of cable controller buffer limit • Truncation of data in “big” events (64 strips/plane end, 128 strips/cable) • CAL • Input: McIntegratingHits – Energy deposited in cubes subdividing the Calorimeter crystals • Output: ADC counts for both diodes at both ends • ACD • Input: McPositionHits – position/energy in ACD Tiles/ribbons • Output: PMT ADC counts per end

24. Calibration Electronic response • thresholds, gains, non-linearity, efficiency, etc. • ~900k tracker strip time-over-thresholds • ~12k calorimeter channels • ~200 ACD channels • Dead channels • Noise σ ≈ 250μrad Relative alignment of tracker planes σ ≈ 40μm Alignment (Important for tracker!)

25. x x • TKR3x•ypair layers in a row workhorsegtrigger x • CAL: LO – independent check, energy info. HI – indicates high energy event: Instrument Triggering and Onboard Data Flow Hardware Trigger On-board Processing Onboard filters: reduce data to fit within downlink, provide samples for systematic studies. Hardware trigger based on special signals from each tower; initiates readout Function: • “did anything happen?” • keep as simple as possible • • flexible, loose cuts • The FSW filter code is wrapped and embedded in the full detector simulation • leak a fraction of otherwise-rejected events to the ground for diagnostics, along with events ID for calibration • signal/background can be tuned  rate: a few Hz Combinations of trigger primitives: Total Downlink Rate: <~400 Hz> ** On-board science analysis: transient detection (bursts) Upon a trigger, all subsystems are read out in ~27ms Spacecraft Instrument Total Rate: <3 kHz>* **current best estimate, assumes compression, 1.2 Mbps allocation. *using ACD veto in hardware trigger

26. Trigger and Filter Rates Summary Trigger Filter • Operating daily-average rate is 2.9kHz • Peak rate is 6 kHz (watch deadtime) • For this simulated day, 201 minutes spent in SAA (14%). • Gamma filter rate in this configuration is 360 Hz • Pass any event w/ E>20 GeV: +40 Hz • Plus other filters for mips and heavy ions • Handles to reduce this rate significantly if needed

27. Reconstruction & Analysis

28. Reconstruction Overview • Goal of the reconstruction: • Determine the incident gamma direction and energy • Provide tools for suppressing background • Follow a two pass approach for Cal and Tracker • Best energy from Calorimeter needs tracking information • Best tracks from Tracker needs best energy reconstruction • Solve by breaking into two passes • Background suppression • ACD Recon • MIP Finder • Analysis step at end of reconstruction • Create output ntuple for detailed performance studies • Event Classification • Create FT1 output for analysis

29. Key Not “Recon” Existing Algorithms New Algorithms Flow of ReconstructionGaudi Algorithm Structure: DC2 “Raw data” Root tree/ntuple CalReconPass I TkrReconPass I CalReconPass II TkrReconPass II AcdRecon Event“Analysis” CalXtalRecAlg TkrClusterAlg TkrTrackFitAlg meritAlg(output ntuple) CalClusterAlg TkrFilterAlg (inactive for DC2) TkrVertexAlg FT1Alg (FT1 tuple) TkrFindAlg CalEventEnergyAlg ClassifyAlg (Classification Analysis) CalMipFinderAlg TkrTrackFitAlg CalEventEnergyAlg(parametric only) AcdReconAlg TkrVertexAlg

30. Use now-known tracks to correct the calorimeter energy Use corrected energy to properly weight the track hits in the fit Event Reconstruction Add up the energy in all the crystals(can be an underestimate) “Raw” Calorimeter Response Track Pattern Recognition and Fitting (Kalman Filter) Use calorimeter cluster energy and position to help find the tracks Refined Calorimeter Response Combine tracks to find gamma candidates Track Refitting ACD Analysis Vertex Finding

31. Conversion point Conversion point Conversion point Conversion point Conversion point Vector resultingfrom vertexing offit track pair forthis event Vector resultingfrom vertexing offit track pair forthis event e+ and e-trajectories Pattern Track trajectories Fit TrackTrajectories 1 GeV Gamma 1 GeV Gamma 1 GeV Gamma 1 GeV Gamma 1 GeV Gamma The Illustrated Tracker Reconstruction Tkr Clustering: Associate adjacent hit strips to form strip clusters (corrected for hot and dead strips) Pattern Recognition: Associate strips to form candidate tracks (Combinatoric pattern recognition using Kalman Filter to accept/reject candidate hits) Track Fit: Take candidate tracks and track energy estimate and fit using a Kalman Filter Vertexing: Combine fit tracks to form common vertex point- If one track, vertex is head of track - If two tracks don’t “combine” well, keep as separate vertices

32. Measuring the Event Energy 1-GeV g Thin-converter hits Gap & dead material between tracker towers Thick-converter hits Blank-converter hits Energy lost in tracker Gap between CAL towers Calorimeter crystals Energy lost in CAL Leakage out the back of the CAL

33. Measuring the Energy Deposit in the Calorimeter • Three methods • Parametric Correction (can be used for any track) • Use the tracks to characterize the shower • Position, angle • radiation lengths traversed • Proximity to gaps • Correct “raw” energy • “Likelihood” (limited energy and angular range) • uses relation between energy deposit in last layer and in the rest of the shower. Below about 50 GeV, last-layer energy is proportional to the leaked energy. • Profile Fitting (limited angular range) • Fit layer-by-layer deposit to shower shape • Best if shower peak is contained in CAL • Choose best answer among available methods

34. ACD Analysis Dots show intersection of tracks with planes of ACD tiles. The ACD has been measured to be ~99.97% efficient for minimum-ionizing particles. So what’s most interesting about the ACD is where it isn’t! Dots show intersection of tracks with planes of ACD tiles. Because of gaps in the ACD coverage, charged tracks may fail to produce a signal in any tile. The ACD analysis identifies these gaps to remove sources of background. x x x x x x x x Because of backsplash, there may be struck tiles that are not associated with the tracks. Segmentation of the ACD allows us to salvage such events. We project the track back to the tiles, and ask how close it comes to the nearest struck tile, if any. x

35. Post Reconstruction ProcessingThe Final Steps • meritAlg • Controls the creation of the merit output ntuple • Used, for example, in various event classification analyses • ClassifyAlg • Runs the classification analysis: • Energy Selection • PSF Tail suppression • Background Rejection • Classification Tree analysis • Possibility to use alternate analyses as well • FT1Alg • Produce the Fits Level 1 output

36. Thin Thick Estimated LAT PSF, Aeff PerformanceFrom Detailed Monte Carlo Studies http://www-glast.slac.stanford.edu/software/IS/glast_lat_performance.htm

37. Simulated LAT (>100 MeV, 1 yr) Simulated LAT (>1 GeV, 1 yr) EGRET (>100 MeV) The Gamma-Ray Sky • Comparing EGRET to GLAST: • Illustrating the anticipated improvement in our knowledge of the sky

38. Simulation Checks

39. Data/MC agreement : agreements • Event processing steps : • Trigger and On Board Filter • Reconstruction algorithms • Classification trees ( -> Instrument Response Functions and Background rejection ) • Different kinds of agreements : • For raw signals • Local and usually not too complex • For reconstruction variables • Local or global, can be complex but still a concrete quantity • For classification variables • Global, very high correlation level • For the whole phase space and all types of particles…

40. LAT ground muons and CERN beam tests • LAT ground muons • Testing the real LAT • With a lot of muons but no gammas, no protons • CERN beam tests • Muons, gammas, electrons, positrons,protons, pions • From few MeV to 280 GeV • Testing the Calibration Unit • Most of the events are within two towers but the CU is not the real LAT

41. Beam Simulation CU volume Cerenkov Trigger Scintillators Magnet Dump Si Detectors 2nd arm of Si Detectors EM NaI Cal Overall Dimension Exp Area ~ 7 m

42. The interface

43. CU Simulation

44. Mass simulation • Simulation of the beam upstream the CU (Geant4) • Trigger/veto scintillators & cerenkov • Electron tagger at PS • LAT simulation (GLEAM-Geant4) • Automatic generation of the run configuration • Allowing the best comparison • Efficient processing through the pipeline at SLAC • Geant4 optimization • Simulation parameters • Physics lists • Also a very dedicated team !