A Geometrical Modeller for HEP

1
A Geometrical Modeller for HEP

• René Brun, Andrei Gheata, Mihaela Gheata

2
General introduction

• Intended as a toolkit to provide geometrical
  description of an experiment, full navigation
  functionality and additional tools to ease-up
  building, checking and debugging a geometry.
• Its development is a common ALICE/ ROOT effort
  that started 1.5 years ago.
• Driven by ALICE specific needs related to the
  simulation/ reconstruction framework, it is
  however designed as an experiment-independent
  package.
• This component is being integrated in a general
  Virtual Monte Carlo scheme enabling running
  transparently several simulation MCs starting
  from the same user code (see talk of Ivana
  Hrivnacova) and using the same geometry.
• Based on a GEANT-like architecture, it is able to
  map and confirmed to optimize the geometry
  performance of several HEP experiments. The gain
  in speed for navigation tasks ranges from 20
  (ATLAS) to 800 (CDF) compared to G3

3
Why a new geometrical modeller

• The idea would be to be able to provide a unique
  geometry description

4
Design overview

• Based on some initial requirements
• Provide basic navigation features Where am
  I?, How far from next boundary ?,
• Map Geant3 geometries -gt smooth transition
• Scalability we deal with big geometries
• Performance it rather be faster than existing
  modelers
• Interactivity should be at highest possible
  level - users should easily build, access and
  debug their geometry

• GEANT-like flavor (CSG based on
  container-contained concept)
• main elements volumes and nodes
• Extensible set of 16 primitives composite
  (boolean) parametrized shapes support for
  MANY concept
• Volumetric pixelized navigation
• Make use of symmetries - divisions
• Geometry checking interactive tool

5
Navigation features

• These make the difference between a geometrical
  modeller and a transient store providing
  geometrical input for other applications.
• Long way from implementing to validating/optimizin
  g/ tuning these features gt took most of the time
• Luckily we had a gold mine of G3 geometries to
  test upon.
• ''Where am I ?'' gt up to 2000 performance gain
  compared to GEANT3
• Computing the distance to next boundary gt up to
  800 gain
• Safety gt computed when needed
• Normals to crossed surfaces gt on demand (ongoing
  work)

6
Automatic conversion from G3
7
Collecting samples validation
8
Benchmarks performance for Where am I? vs.
GEANT3
9
User interface other features

• Quite simple API Material(), Volume(), Node(),
  Division(), managed by a single class.
• Browsable geometry with functionality in the
  context menus ray-tracing, lego plots, weight
  estimation, geometry checking tool,
• Perspective geometry viewing allowing picking,
  zooming, animation.
• ROOT I/O size of geometry and time to load are
  very important during geometry design/testing

10
Geometry checker

• It has to be able to check (in reasonable time)
  for illegal extrusions/overlaps in the geometry
  definition.
• Extrusions
• Overlaps

Detected
Not detected
Geometry Overlaps/Extrusions Overlaps/Extrusions Overlaps/Extrusions
Geometry gt 1 mm gt100 m gt10 m
ALICE 154 764 1460
No error-free geometry found
(up to 20K gt 1mm)
11
The Virtual MC integration

• Allows running several MC's starting from the
  same code and having the same geometry for
  simulation, reconstruction and event display

User code
12
Conclusions

• A new geometrical modeller able to represent a
  large number of HEP experiments is being
  developed by ALICE and ROOT teams.
• This will provide an unique representation of
  ALICE geometry and will allow running
  transparently several MCs starting from the same
  user-code.
• Performance was the highest priority during the
  development and this is reflected by the
  benchmarks.

• The code is available in ROOT and we
  welcome everybody to use it !

13
Benchmarks performance for Where am I? vs. G3
14
Benchmarks performance for distance to next
vs. G3