Tools for optimising and assessing the performance

1
Simulation Software Meeting DESY, 27
/ 28 June 2005
Tools for optimising and assessing the
performance of the vertex detector

• from MIPS to physics
• high-level reconstruction tools
• outlook plans for Snowmass

Sonja Hillert (Oxford) on behalf of the LCFI
collaboration
2
Typical event processing at the ILC
3
From MIPS to physics

• To optimise design of vertex detector and
  evaluate its physics performance need
• 1) sufficiently accurate reconstruction (from
  MIPS to tracks)
• 2) high-level reconstruction tools,
• e.g. flavour tagging, vertex charge
  reconstruction, (see previous page)
• 3) study of benchmark physics channels based on
  these tools
• Step 1 comprises, e.g., simulation of
• signals from the sensors charge
  generation/collection, multiple scattering
• data sparsification signal background hit
  densities, edge of acceptance
• Other parameters to be determined from the
  results obtained with the entire chain
• overall detector design radial positions (inner
  radius!) and length of detector layers,
• arrangement of sensors in layers, overlap of
  barrel staves (alignment), strength of B-field
• material budget beam pipe, sensors,
  electronics, support structure (material at large
  cos q)

4
Current status

• so far focused on high-level reconstruction
  tools
• (in particular flavour-tag, vertex charge)
  using mainly fast MC simulation SGV
• and (for part of the studies) JAS3
• SGV core of the program well tested (? DELPHI),
  allows fast change of geometry
• lacks accurate description of processes in
  sensors readout chain, and of
• multiple scattering
• JAS3 full MC under development, but not ready /
  robust for the time being
• tracking used in the fast MC available under
  JAS3 less precise than SGV
• (SGV Billoir algorithm)

5
Current status contd

• LCFI proposed independent development of full
  GEANT4-based description
• of processes in vertex detector sensors and
  readout chain to UK funding agencies
• (see also 59th DESY-PRC, May 05)
• while such an approach is in principle
  appreciated, the current funding situation
• in the UK does not allow an effort in this
  field at the level needed to implement
• the full MIPS to physics programme
• looking for ideas how to form international
  effort to develop the essential
• simulation and reconstruction tools

6
Future plans

• future programme will depend on further
  negotiations outline of plans preliminary!
• envisage top-down approach select physics
  channels requiring variety of
• higher level reconstruction tools
  develop/improve and assess those in parallel
• processes to be studied (both requiring flavour
  tagging, vertex charge reconstruction)
• Higgs self-coupling
• might profit from improving track-jet
  association using vertex information
• Left-right forward-backward asymmetry in ee- ?
  b bbar, c cbar
• sensitive to polar angle dependence, decays
  outside the vertex detector (at high energy),
• could be used to assess performance of charge
  dipole reconstruction
• (yielding quark charge measurement for
  neutral hadrons)
• use these processes as benchmarks to determine
  sensitivity to detector design
• parameters on a timescale of 2-3 years

7
Visualisation tools
Purpose flavour tagging vertex charge
reconstruction can be improved by looking at
cases, where the reconstruction fails, on an
event by event basis

• top written in Python with Coin/HEPVis
• wrappers input read from XML file (D.
  Bailey)
• right root-based tool so far MC tracks only
• reconstruction level to be added (B Jeffery)

8
OO reference version of ZVTOP

• ZVTOP in use for 10 years, several versions
  (SLD ? LEP ? ILC
• variable transformations differences in what
  is included in the different versions)
• LCFI therefore decided to develop an object
  oriented (C) version of ZVTOP,
• and to check it against the SLD code (a
  Java-based version is being developed at SLAC)
• latest version of ZVTOP (ZVTOP3) comprises two
  branches
• ZVRES and ZVKIN (also known as the ghost
  track algorithm)
• ghost track algorithm should
• cope with cases with a 1-prong B decay followed
  by a 1-prong D decay
• allow reconstruction of the charge dipole
  (information on neutral Bs)
• at the ILC improve flavour tagging capabilities
• development of the class structure in progress
• estimated timescale for development and
  verification 1 year

9
Neural Network Tool

• neural nets used for flavour tagging, vertex
  charge reconstruction,
• C based code developed in Bristol allows
  implementation of
• feed-forward nets of arbitrary topologies
• 3 response functions available sigmoid,
  tansigmoid, linear, can be combined
• (i.e. different neurons in same net can have
  different response functions)
• 4 training algorithms 3 based on
  back-propagation, 1 genetic algorithm
• networks generated with this tool can be
  serialised as plain text or in XML format
• for retrieval from a web server
• tar-file available at
• http//www.phy.bris.ac.uk/research/pppages/Dav
  eB/NeuralNet.tar.gz

10
Vertex charge reconstruction
Procedure find vertices and vertex axis
(ZVTOP) assign tracks to B decay chain sum
their charge
can either use a neural net or assign all tracks
found in inner vertices (methods work equally
well at ECM 200 GeV)

• Status
• extending study to range of centre of mass
  energies
• larger fraction of B hadrons decay outside
  vertex detector
• find steep drop in 2D seed vertex decay length
• at the vertex detector edge ? drop of
  efficiency
• indications that this is due to faulty track
  selection
• Plans
• extend study to ccbar events, combine with
  flavour tagging

11
Flavour tagging

• Study ee- ? qqbar events (all flavours except
  for ttbar), so far using JAS3 framework
• neural net used for flavour tagging including
  the primary vertex momentum (left) as input
• variable, in addition to secondary vertex
  parameters, improves b/c jet separation by 10

blue use only secondary vertex
parameters magenta also use primary vertex
momentum
12
Plans for Snowmass

• presentation of results on vertex charge
  reconstruction over range of ECM values
• comparison of SiD detector concept and formerly
  European concept
• in terms of vertex charge reconstruction
  using SGV
• in particular look at performance at edge of
  polar angle range,
• where difference between the detectors is
  expected
• (SiD vertex detector includes forward disks,
  LCFI-detector does not)

13
Additional Material
14
Vertex charge reconstruction
Additional Material Additional Material
Additional Material
Additional Material Additional Material
Additional Material
15
Changes since LCWS 2004

• between LCWS04 and ECFA workshop (Durham)
• optimised cut on L/D, masked KS and L
• dropped ISR while studying vertex charge
  reconstruction for fixed jet energy
• (otherwise lose 85 of generated events
  through back-to-back cut on jets)
• include information from inner vertices seed
  vertex is ZVTOP vertex furthest from IP
• assigning tracks contained in inner
  vertices to B decay chain regardless of their
• L/D value improves vertex charge
  reconstruction (for large distances of seed
  vertex
• from IP, L/D cut is much larger than required
  to remove IP tracks)

Additional Material Additional Material
Additional Material
Additional Material Additional Material
Additional Material
an atypical event with a large distance of the
seed vertex from the IP
16
Improvement of reconstructed vertex charge
Additional Material Additional Material
Additional Material
Additional Material Additional Material
Additional Material