Transportation Process For BaBar

Transcription

1 Transportation Process For BaBar David C. Williams University of California, Santa Cruz Geant4 User s Workshop Stanford Linear Accelerator Center February 21, 2002 Outline: History and Motivation Design Implementation Interface to Geant4 Future Plans Conclusion

2 Why Start Such a Project? History and Motivation Originally began purely as an experiment. BaBar, summer 1999: Geant4 based simulation not robust: Geometry problems Tracking issues in a field Debugging native transportation in geant4 is not simple Some progress in understanding solids Worries about tracking precision in native transportation Interesting as a research project for geant4 Why not? Transportation Process For BaBar February 21, 2002 David C. Williams 2

3 Timeline History and Motivation December 1999: first simple implementation February 2000: installed in BaBar code repository April 2000: replicas supported May 2000 Jan 2001: test, test, test, test optimize, optimize July 2001: Large scale Monte Carlo production begins Transportation Process For BaBar February 21, 2002 David C. Williams 3

4 Tracking Precision History and Motivation BaBar is a high statistics experiment Tracking accuracy: ε δ N, δ = 25 microns, N = 100,000 Tolerance Transportation Process For BaBar February 21, 2002 David C. Williams 4

5 Tracking Precision History and Motivation Test Example Situation today with geant4 navigator is much better. Transportation Process For BaBar February 21, 2002 David C. Williams 5

6 Design Principles Design Specific solution for BaBar Simplicity Speed Precise tracking Robust Require as little modification to geant4 toolkit as possible Transportation Process For BaBar February 21, 2002 David C. Williams 6

7 Encapsulation Design What is a volume? Something that: Understands its geometric boundaries Given an internal point, can tell you Which daughter volume (if any) it is inside of The closest daughter The daughter that is intersected by a trajectory Can supply a daughter volume upon request Transportation Process For BaBar February 21, 2002 David C. Williams 7

8 Encapsulation and Inheritance Design The code that tracks particles and finds intersections deals with generic volumes (i.e. does not care about how the volume methods are implemented). Implementation of a specific type of volume through inheritance: Plain volumes Volumes with voxelized search optimization Replicas Parameterized volumes Is performed independently of the tracking code Transportation Process For BaBar February 21, 2002 David C. Williams 8

9 Tracking in a Magnetic Field Design General design: encapsulation in a generic trajectory object Implementation: cheap, easy, fast helical segmentation Transportation Process For BaBar February 21, 2002 David C. Williams 9

10 Track Intersections Implementation Push ahead approach Tolerance = 0.5 µm Tolerance Transportation Process For BaBar February 21, 2002 David C. Williams 10

11 Tracking Precision Implementation Double-precision is preserved Transportation Process For BaBar February 21, 2002 David C. Williams 12

12 Tracking Intersection Implementation Optimizing the Search for a Solution After first iteration, one can do better than making a linear guess! s = a l a + b a b Transportation Process For BaBar February 21, 2002 David C. Williams 22

13 Track Intersection Implementation Analogy: search for the zero of a function Method: inverse parabolic extrapolation (Brent s method) f(x) x Numerical Recipes Copyright (c) Cambridge University Press Transportation Process For BaBar February 21, 2002 David C. Williams 23

14 Track Intersection Implementation Some problems are just plain difficult! Transportation Process For BaBar February 21, 2002 David C. Williams 24

15 Track Intersection Implementation Be careful of your assumptions!? Transportation Process For BaBar February 21, 2002 David C. Williams 25

16 Track Logic Implementation Optimization requires some tradeoffs r B Transportation Process For BaBar February 21, 2002 David C. Williams 26

17 Round off is Your Enemy Implementation Machine precision is finite Intent Reality Transportation Process For BaBar February 21, 2002 David C. Williams 27

18 Error Tolerance is Important Implementation One intersection, or two?? Transportation Process For BaBar February 21, 2002 David C. Williams 28

19 Implementation Beware Infinite Loops Keep pushing ahead?????? Transportation Process For BaBar February 21, 2002 David C. Williams 29

20 Trust is Essential Implementation What happens if a Solid has a bug? Can you tell the difference between a solid problem and a tracking problem? Nice error messages can help a little BgsLocation::TrapEscapedTrajectory -- tracking error encountered Linear distance to exit >= kinfinity Global location: e e e+01 cm Global direction: e e e+00 cm Position is on the tolerance surface of the current volume Geometry history: Local Position Distance x y z To Out (cm) Volume Name [copy] e e e e+00 B1S11_phys[0] e e e e+00 B1Shell_phys1[1] e e e e+00 Pep[0] e e e e+02 BFAC[0] Assuming true distance to exit is zero Paranoid option Transportation Process For BaBar February 21, 2002 David C. Williams 20

21 Speed Implementation Difficult to quantify Depends on detector geometry and physics BaBar transportation tolerances somewhat higher For many lower energy particles: BaBar transportation faster For single, high momentum particle: speeds comparable Remember: BaBar transportation optimized for one application Transportation Process For BaBar February 21, 2002 David C. Williams 22

22 Robustness Implementation Not always clear when a problem is due to a solid Failure rate at the level of one part out of a million events Transportation Process For BaBar February 21, 2002 David C. Williams 22

23 Modularity is a Good Thing Interface to Geant4 Replace G4Transportation Turn off multiple scattering displacement (standard option) Modify energy loss code to use internal safety calculation Replace initial step touchable at first step Not a general solution! The few code modifications must be redone for each Geant4 release. G4Transportation remains a switchable option for BaBar. Transportation Process For BaBar February 21, 2002 David C. Williams 23

24 Improving the Toolkit Future Plans Some useful ideas have been demonstrated: Allowing applications to replace one or more pieces of transportation Encapsulation makes code clearer and simpler Track intersection algorithm can be application specific How about incorporating some of these ideas into the toolkit? Transportation Process For BaBar February 21, 2002 David C. Williams 24

25 Generalized Navigator Future Plans New design for the geant4 toolkit Main goal: encapsulation Generic volumes Generic trajectories Generic track intersection methods Uniform navigator interface for processes Design being developed many issues to be resolved A big project! Transportation Process For BaBar February 21, 2002 David C. Williams 25

26 BaBar Transportation Conclusion Started as an experiment, now used successfully by BaBar in large scale Monte Carlo production Encapsulation as a design principle has been very useful Testing and optimization has uncovered (and still uncovering) complex geometric issues, each of which had to be addressed Lessons learned will be used to improve the geant4 toolkit Transportation Process For BaBar February 21, 2002 David C. Williams 26