1. 20 September 2005Fabrizio Cei1 Status and Perspectives of MEG Software Fabrizio Cei INFN & University of Pisa INFN Scientific Committee I Napoli, 20 September 2005

2. 20 September 2005Fabrizio Cei2 Outline  MC status and perspectives  Offline status, perspectives and milestones and milestones  First Estimate of CPU/data storage needs Contributions: Tokyo, Pisa, Lecce, PSI, Roma and Pavia groups. Tokyo, Pisa, Lecce, PSI, Roma and Pavia groups.

3. 20 September 2005Fabrizio Cei3 MC status & perspectives MEGEVE - Event Generator; MEGEVE - Event Generator; GEM – The detector simulator: GEM – The detector simulator: – Liquid Xenon Calorimeter; – Drift Chamber; – Timing Counter; – Magnet and Target. ZEBRA Output and Analysis Tools; ZEBRA Output and Analysis Tools; Event Cocktail: the MEG “Bartender”; Event Cocktail: the MEG “Bartender”; LP/Beam Test fully simulated (not discussed); LP/Beam Test fully simulated (not discussed); MC code almost ready for preproduction tests. MC code almost ready for preproduction tests.

4. 20 September 2005Fabrizio Cei4 Energy release in LXe Positron track Hits on TC

5. 20 September 2005Fabrizio Cei5 MEGEVE: the event generator StatusStatus –Signal events; –Michel positrons; –Radiative decay (RD): Exact formulae for unpolarized muon;Exact formulae for unpolarized muon; Approximation for polarized muon (back-to-back, P e  P   52.8 MeV).Approximation for polarized muon (back-to-back, P e  P   52.8 MeV). –Positron annihilation in flight (AIF): Preliminary AIF within target;Preliminary AIF within target; Started study for realistic AIF: magnet, DCH, TC and Target.Started study for realistic AIF: magnet, DCH, TC and Target. –Scheme to generate pile-up events: (Michel + RD, Michel + AIF, AIF + RD, RD + RD) + additional Michel decays; more than two events can be overlaid; RD + RD) + additional Michel decays; more than two events can be overlaid; -CPU time: 2  6 sec/event, dominated by scintillation photon tracking (maybe it can be improved). (maybe it can be improved). NextNext –Exact formulae for polarized muon’s radiative decay; –Realistic AIF and background studies (under way); –Study of online/offline pre-selection and calibrations (under way).

6. 20 September 2005Fabrizio Cei6 Background studies under way 1) Accidental background from superimposition of a Michel positron and a  from radiative decay. Simulation of trigger conditions quoted in the Proposal – Photon energy cut E  > 45 MeV – Time window e + -  △ T = 10 ns – e + -  direction matching Preliminary Work E (MeV)

7. 20 September 2005Fabrizio Cei7 Background studies under way 2) Realistic studies of Michel positron annihilation in flight. Almost complete detector simulation (not the target only). Main contributions from target and drift chambers. Annihilation  energy spectrum in LXe Preliminary Work

8. 20 September 2005Fabrizio Cei8 Detector simulation 1) LXe StatusStatus –Geometry: revised shape for vacuum vessel, PMT holders and honeycomb; –Implemented decay curve and wavelength spectrum of LXe scintillation; –GEANT based scintillation photon tracking: Reflection on PMT quartz window and PMT holders;Reflection on PMT quartz window and PMT holders; PMT quartz window transmittance;PMT quartz window transmittance; Absorption and scattering in Liquid Xenon.Absorption and scattering in Liquid Xenon. –Outputs: Energy deposit, position and timing in Liquid Xenon;Energy deposit, position and timing in Liquid Xenon; Preliminary waveform output: hit timing of scintillation photonsPreliminary waveform output: hit timing of scintillation photons for each PMT ( ～ 3 x 10 4 photoelectrons @ 50 MeV). for each PMT ( ～ 3 x 10 4 photoelectrons @ 50 MeV). NextNext –Update geometry to match the final design (almost done); –Implement support structure; –“Fast” scintillation photon tracking; multiple options ? (GNEXT/hand made).

9. 20 September 2005Fabrizio Cei9 Detector simulation 2) DCH StatusStatus –Geometry: DCH geometry completed; wires simulated;DCH geometry completed; wires simulated; Helium bag included, but turned off by default; Helium bag included, but turned off by default; Implementation of the Vernier pad.Implementation of the Vernier pad. –Isochrones tables for various B-field; –Outputs: Entrance/Exit position from cells;Entrance/Exit position from cells; Energy, timing and direction for each hit;Energy, timing and direction for each hit; Drift time in cells.Drift time in cells. NextNext - Update geometry for latest DCH support and peripherals; - Update geometry for latest DCH support and peripherals; - Implement charge and time signal simulation on wires & pads; - Implement charge and time signal simulation on wires & pads; - Implement waveform digitization. - Implement waveform digitization.

10. 20 September 2005Fabrizio Cei10 Detector simulation 3) TC StatusStatus –Geometry: scintillation bars/fibers, PMTs, APDs and light guides. –Outputs: Hit position, timing and energy;Hit position, timing and energy; Energy release and step length for each GEANT hit;Energy release and step length for each GEANT hit; Waveform outputs for scintillation bars.Waveform outputs for scintillation bars. –Photon propagation inside the curved fibers recently implemented. NextNext –Implement support structure; –Waveform for scintillation fibers; - A standalone MC for comparison with scintillation bar beam test.

11. 20 September 2005Fabrizio Cei11 Detector simulation 4) Target and Magnet StatusStatus - Preliminary approach outside GEM; - Preliminary approach outside GEM; - Realistic treatment of target geometry included in GEM. - Realistic treatment of target geometry included in GEM. NextNext –Implement target support; –Beam transport within the detector; –Create and implement the module “beam” within GEM.

12. 20 September 2005Fabrizio Cei12 ZEBRA Output & Analysis Tools HitsHits –Energy deposit, position and timing for LXe/DCH/TC; –Scintillation photon hits for LXe and TC PMTs. Preliminary waveform outputPreliminary waveform output - Brute force approach: record all hit timing for each PMT (LXe) - Brute force approach: record all hit timing for each PMT (LXe) 3 x 10 4 photons @ 500 PMTs/828 PMTs for 52.8 MeV gamma 3 x 10 4 photons @ 500 PMTs/828 PMTs for 52.8 MeV gamma ⇒ 120 kB/event (w/ zero suppression); few kB/event from TC; ⇒ 120 kB/event (w/ zero suppression); few kB/event from TC; - Need binned waveforms to save disk space. - Need binned waveforms to save disk space. Analysis toolsAnalysis tools - ZEBRA2NTUPLE - ZEBRA2NTUPLE - ZEBRA2ROOT – based on ROME framework; - ZEBRA2ROOT – based on ROME framework; temporary software waiting for MegRoot temporary software waiting for MegRoot

13. 20 September 2005Fabrizio Cei13 The MEG “Bartender” 1) ROOT based program for ROOT based program for Event Cocktail; Event Cocktail; Read experimental and simulation data; Read experimental and simulation data; Make mixture of several MC sub events; Make mixture of several MC sub events; Simulation of pulse shape of MC data (digitization); Simulation of pulse shape of MC data (digitization); Rearrange channels of experimental data to make them as MC. Rearrange channels of experimental data to make them as MC. Possible simple calibration; Possible simple calibration; Possible trigger simulation. Possible trigger simulation. (R. Sawada, S. Yamada) S. Yamada)

14. 20 September 2005Fabrizio Cei14 The MEG “Bartender” 2) += Nphe = 620 Nphe = 123 Nphe = 743 Example: waveform pile-up  three possible models of single waveform;  gaussian, sinusoidal or constant noise can be added;  event rate can be specified;  relative timing is extracted randomly.

15. 20 September 2005Fabrizio Cei15 Offline status 1) MegRoot Main framework,  Main framework, based on ROOT; based on ROOT;  Under development, mainly in Lecce. mainly in Lecce. ROME Backup framework,  Backup framework, also based on ROOT; also based on ROOT;  Developed at PSI;  Already used for beam test analysis. test analysis.

16. 20 September 2005Fabrizio Cei16 Offline status 2)  MegRoot: - First version recently released; - Backbone from AliRoot; - Copyrights issues solved with F. Carminati; - Included algorithms: Alice, GEM or proprietary. - 4 Modules, 344 Classes, > 4000 Methods.  Other topics: - Promising pattern recognition and tracking - Promising pattern recognition and tracking algorithms under development. algorithms under development.

17. 20 September 2005Fabrizio Cei17 MegRoot Contributors Main Architecture: Alice collaboration, V. Di BenedettoMain Architecture: Alice collaboration, V. Di Benedetto Magnet: E. Cavallo, W. OotaniMagnet: E. Cavallo, W. Ootani DCH: C. Chiri, F. Ignatov, M. Schneebeli, S. Spagnolo,DCH: C. Chiri, F. Ignatov, M. Schneebeli, S. Spagnolo, G. Tassielli, H. Nishiguchi G. Tassielli, H. Nishiguchi LXe: V. Di Benedetto, S. Mihara, R. Pazzi, R. Sawada,LXe: V. Di Benedetto, S. Mihara, R. Pazzi, R. Sawada, G. Signorelli, G. Terracciano G. Signorelli, G. Terracciano TC: G. SiragusaTC: G. Siragusa Event Generator for VMC: F. Cei, A. MazzacaneEvent Generator for VMC: F. Cei, A. Mazzacane Database: D. Barbareschi, R. SawadaDatabase: D. Barbareschi, R. Sawada Blue: young people who recently joined the collaboration

18. 20 September 2005Fabrizio Cei18 MegRoot Architecture MegRoot FrameworkMegRoot Framework –Based on ROOT –User code in C++ –Usage of FORTRAN libraries Geant3, event generators, “microcernlib”Geant3, event generators, “microcernlib” Integrates reconstruction and analysis softwareIntegrates reconstruction and analysis software Each detector subsystem has one single package (oneEach detector subsystem has one single package (one directory, one library); detector geometry is hard-coded directory, one library); detector geometry is hard-coded Calibration constants read via MySQL interfaceCalibration constants read via MySQL interface (to be unified with MC). (to be unified with MC).

19. 20 September 2005Fabrizio Cei19 MegRoot status Structure of all Classes complete up to the Digitization stepStructure of all Classes complete up to the Digitization step (main purpose is to test the Reconstruction Modules); (main purpose is to test the Reconstruction Modules); All Methods designed; most algorithms implemented;All Methods designed; most algorithms implemented; The Reconstruction Class is currently being designed:The Reconstruction Class is currently being designed: - DCH reconstruction & Kalman filter almost completed; - TC & LXe: under development. - TC & LXe: under development. Reconstruction Class will read out ROOT ObjectsReconstruction Class will read out ROOT Objects (MC Digits) or streamer objects (raw data); (MC Digits) or streamer objects (raw data); Intense collaboration between Detector Experts andIntense collaboration between Detector Experts and Core Offline Group under way; Core Offline Group under way; Lecce people are joining the effort to bring the LP code withinLecce people are joining the effort to bring the LP code within MegRoot and analyze the data with the rest of the collaboration. MegRoot and analyze the data with the rest of the collaboration.

20. 20 September 2005Fabrizio Cei20 What’s missing in MegRoot Waveform decoding: studies under way.Waveform decoding: studies under way. Detailed structure of the Reconstruction classDetailed structure of the Reconstruction class MySQL interface needs to be extended to the final Detector (presently it only works for LP)MySQL interface needs to be extended to the final Detector (presently it only works for LP) Calibration ModuleCalibration Module Farm Manager Code (Global/Local)Farm Manager Code (Global/Local) Not final geometry (to be borrowed from GEM)Not final geometry (to be borrowed from GEM) It could be one of the most CPU consuming processes

21. 20 September 2005Fabrizio Cei21 Offline Milestones LP in MegRoot:LP in MegRoot: geometry & event display completed; geometry & event display completed; data decoding: OK for MC, under way for data; data decoding: OK for MC, under way for data; analysis within October analysis within October Code implementation in CVS: September 2005Code implementation in CVS: September 2005 Reconstruction Class: October 2005Reconstruction Class: October 2005 Estimate LXe analysis CPU load: October 2005Estimate LXe analysis CPU load: October 2005 Calibration Module: it will start in November 2005 (after completion of Reconstruction Class)Calibration Module: it will start in November 2005 (after completion of Reconstruction Class)

22. 20 September 2005Fabrizio Cei22 Pattern Recognition Studies  First algorithm developed in Pisa (2003): combined use of detector segmentation, fast momentum combined use of detector segmentation, fast momentum reconstruction and geometrical correlations between hits in a track; reconstruction and geometrical correlations between hits in a track;  New algorithm developed in Lecce (2005): - complete simulation of time measurement; - complete simulation of time measurement; - definition of signal sensitive parameters for discriminating - definition of signal sensitive parameters for discriminating tracks related to physical trigger from background; tracks related to physical trigger from background; - space point reconstruction by using timing information; - space point reconstruction by using timing information; - technical note in progress. - technical note in progress.  Typical background: 5 Michel positron tracks for both algorithms.  Similar performances:  95 % efficiency in reconstructing good tracks, with few per cent contamination of spurious hits; tracks, with few per cent contamination of spurious hits;  Complementary techniques; possible merging ?  Timing counter information to be inserted (possible improvement in discrimination power). in discrimination power).

23. 20 September 2005Fabrizio Cei23 Track fitting studies Kalman filter based algorithm under development; preliminary results. Reconstructed tracks (5 e + ’s) Momentum reconstruction:  p/p  1.4 % Resolutions: Resolutions: Target position Angle Target position Angle  R ~ 1.7 mm   = 8 mrad  R ~ 1.7 mm   = 8 mrad  z ~ 1.9 mm   = 6 mrad  z ~ 1.9 mm   = 6 mrad

24. 20 September 2005Fabrizio Cei24 First estimate of CPU/data storage needs IngredientsIngredients - MC production GEANT3 simulation GEANT3 simulation Hit production & digitizationHit production & digitization - Data/MC reconstruction (partial) Computations based on Pentium III, 1.4 GHz for dataComputations based on Pentium III, 1.4 GHz for data (faster computers available), Pentium IV, 3.0 GHz for MC (faster computers available), Pentium IV, 3.0 GHz for MC Data reconstructionData reconstruction – LXe: LP beam test data; QSUM, Linear Fit and MINUIT fit – DCH: Kalman Filter to 5 simulated charged tracks & Track extrapolation to TC (1.2 x storage and 1.5 x CPU) Track extrapolation to TC (1.2 x storage and 1.5 x CPU) - TC: Not implemented - TC: Not implemented Waveform decoding not implementedWaveform decoding not implemented

25. 20 September 2005Fabrizio Cei25 CPU estimates (msec/evt) 1) Hits Production DigitsProductionReconstruction Montecarlo event generation 2000 – 6000 Bartender (no noise) 230 DCH reconstruction 200 (MC) LXe reconstruction 40 – 80 (LP data) TC reconstruction not included not included Total 2000 - 6000 230 250 – 300

26. 20 September 2005Fabrizio Cei26 CPU estimates 2) Assumptions:Assumptions: –Trigger rate: 20 Hz (physical events) - MC: - 10 12 accidental events obtained by “Event Cocktail” (10 6 Michel positrons & 10 6 AIF/RD photons)/year; (10 6 Michel positrons & 10 6 AIF/RD photons)/year; Reconstruct ~ 10 8 events & use others for occupancy checks; Reconstruct ~ 10 8 events & use others for occupancy checks; - 2 x 10 7 correlated events in the signal region/year. Results:Results: - Montecarlo production & event cocktail: 10  30 CPU - Montecarlo production & event cocktail: 10  30 CPU - MC/Raw data Reconstruction :  20 CPU/proc. - MC/Raw data Reconstruction :  20 CPU/proc. - WF decoding not implemented; studies under way; - WF decoding not implemented; studies under way; - Calibrations to be estimated; - Calibrations to be estimated; - No pre-filtering (it could reduce CPU time). - No pre-filtering (it could reduce CPU time).

27. 20 September 2005Fabrizio Cei27 Data Storage Estimate 1) A) REAL DATA  Assume 10 7 sec/year & an overall trigger rate of 100 Hz, 20 Hz of physical data and 80 Hz of calibrations ; 20 Hz of physical data and 80 Hz of calibrations ;  Waveform data (channel occupancy assumed: 50 % for LXe, 10 % for DCH & TC): 1.2 Mb/event  120 Mb/sec; LXe, 10 % for DCH & TC): 1.2 Mb/event  120 Mb/sec;  Compression factors: 10 for true events (obtained in PIBETA); 10 for true events (obtained in PIBETA); 100 for calibrations  0.12 Mb/event for physical 100 for calibrations  0.12 Mb/event for physical events, 0.012 Mb/event for calibrations events, 0.012 Mb/event for calibrations  Data storage: (20 x 0.12 + 80 x 0.012) Mb/sec = 3.4 Mb/sec  3.4 Mb/sec x 10 7 sec/year = 34 Tbyte/year  3.4 Mb/sec x 10 7 sec/year = 34 Tbyte/year

28. 20 September 2005Fabrizio Cei28 Data Storage Estimate 2) B) MONTE CARLO  Assume a total production of 10 12 accidental events (dominant background in the signal window) and 2 x 10 7 correlated background in the signal window) and 2 x 10 7 correlated events/year (first sample obtained by merging two 10 6 events/year (first sample obtained by merging two 10 6 independent samples of positrons & photons). To reduce problems independent samples of positrons & photons). To reduce problems of multiple disk accesses, MC events must be duplicated of multiple disk accesses, MC events must be duplicated (x 2 correlated, x 20 accidental). (x 2 correlated, x 20 accidental).  Event size based on LXe (photon arrival times) and TC information: 200 kb/event; waveforms not simulated and noise not included. 200 kb/event; waveforms not simulated and noise not included.  Data storage: - (200 kb/event x 2 x 10 7 x 2) = 8 Tb/year (correlated events); - (200 kb/event x 2 x 10 7 x 2) = 8 Tb/year (correlated events); - (200 kb/event x 2 x 10 6 x 20) = 8 Tb/year (uncorrelated - (200 kb/event x 2 x 10 6 x 20) = 8 Tb/year (uncorrelated events); + factor 3 for digitization: events); + factor 3 for digitization: TOTAL ~ 50 Tb/year TOTAL ~ 50 Tb/year

29. 20 September 2005Fabrizio Cei29 Data Storage Estimate 3) Summary Real data: 34 Tbytes/year Monte Carlo: ~ 50 Tbytes/year Overhead (DSTs, reconstructed info …) ~ 15 Tbytes/year __________________________________________ Total ~ 100 Tbytes/year

30. 20 September 2005Fabrizio Cei30 Conclusions  Relevant progresses were obtained in MC and offline framework;  About 10 young people joined the collaboration in the offline core group; the MC group is also in expansion; offline core group; the MC group is also in expansion;  A first version of Offline framework was released; important milestones were defined for the Offline group; important milestones were defined for the Offline group;  Other software jobs are under way: tracking and pattern recognition algorithms, studies of possible calibration procedures … recognition algorithms, studies of possible calibration procedures …  Thanks to the experience gained in the beam test and to the good level of sophistication reached by the MC code we are good level of sophistication reached by the MC code we are in a position to perform a first estimate of CPU power and in a position to perform a first estimate of CPU power and disk storage needs of our experiment. disk storage needs of our experiment.  We are starting to think at physics analysis (likelihood, blind …)

31. 20 September 2005Fabrizio Cei31 Backup slides

32. 20 September 2005Fabrizio Cei32 MegRoot structure MegRoot run management interface classes detector base classes data structure base classes Detectors DCHLXE FMDTC STEER PYTHIA6 Geant3 MICROCERN Geant4 ROOT HIJING…EVGEN Geant3 VMC Geant4 VMC External packages VMC

33. 20 September 2005Fabrizio Cei33 ROME features ROME is a framework generator.ROME is a framework generator. Only 6 different objects with up to 6 access methods.Only 6 different objects with up to 6 access methods. All classes are generated, only event methods have to be written.All classes are generated, only event methods have to be written. No knowledge about object oriented programming is needed.No knowledge about object oriented programming is needed. Folders and Tasks support a very clear program structure.Folders and Tasks support a very clear program structure. Modularity: tasks can be exchanged even at runtime.Modularity: tasks can be exchanged even at runtime.