Track finding algorithms
V. Kuznetsov
Dec 3, 2001
Introduction
Object oriented approach (C++ language) and modular design of D0
software allows development of many tracking algorithms.
They may be run simultaneously/separately
to achieve best performance for different physics tasks.
Currently, four algorithms are under investigation:
- Road approach (default algorithm):
based on TRF++ uses specific paths (roads) during
track finding
- HTF (histogramming track finder):
divides D0 detector into slices in (phi, rho) and uses Hough transform
to reduce initial number of combinations. Existing algorithm can
use either CFT, SMT hits or combinations of them to construct tracks.
- Elastic reco (elastic-template algorithm): can
use existing tracks (from other track finders) as initial seeds,
existing vertex (run after of GTR and vertex code)
and/or run in stand-alone mode (construct own seeds and vertices).
- SBF (silicon barrel finder): based on random
search. Track candidates are formed in SMT detector and than extended to CFT.
Default algorithm (road-following)
The default algorithm it is based on TRF++ package and use specific
paths (roads) during track finding.
TRF++ is a software system for finding and fitting tracks in any
particle physics detector. Design is object-oriented with
considerable emphasis on reusability and extensibility.
D0 acceptance is divided into four regions:
central - full CFT extended into SMT
forward - forward SMT with three F-disks
overlap - partial CFT extended into SMT
gap - between overlap and forward.
The track finder consists of particular paths for given region.
Currently the following paths (roads) are ready:
- CFT central -
includes axial and axial+stereo) steps. It searches only tracks
in CFT central region requiring to have 16 CFT hits.
- CFT overlap looks for track which may cross only part of
CFT layer (8-15 layers).
- SMT forward is designed to work only for tracks in forward region.
Track finder requires that track candidate should cross closest
to barrel F-disks. Minimal number of required hits equal 5.
- gap tries to find tracks in a gap region. Algorithm is started with
all combinations in two different F-disks (inside of barrels).
The eta cut is applied to select only candidates in a gap region.
Additional hits are added to the track from barrels.
- SMT overlap is alternative approach for overlap tracking. It starts
from SMT. Tracks are selected using particular eta, phi cuts and
than may be extended into CFT.
- CFT to SMT extension is additional step to extend collection of
tracks found by all CFT track finders into SMT region.
- SMT to CFT extension is additional step to extend collection of
tracks found by all SMT track finders into CFT region.
- H-disk extenstion is additional step to extend collection of
forward and gap tracks up to H-disks. At this step hits from
H-disks may be added to the track. If track cross H-disks two
hits are required.
Full documentation about elastic track finder can be found
here.
HTF (histogramming track finder)
HTF (histogramming track finder) is a package for reconstructing tracks in
the central D0 tracker (SMT + CFT). It begins with the hit preselection
using histogramming technique based on Hough transform which helps to
heavily reduce combinatorics. The output of the preselection stage is a
set of templates. Each template is a track candidate consisting of several
hits, with approximately known track parameters. The templates are futher
processed applying the Kalman filtering. During this stage, fake templates
are discarded, wrong hits removed, and track parameters accurately
calculated.
The present implementation of the algorithm makes use of two strategies:
a) begin with SMT hits, find tracks in SMT, then extrapolate them to CFT,
and b) begin with CFT hits, find tracks in CFT, then extrapolate them to
SMT. Another possibility (silicon barrel track finder, SBF) is to begin
with all compatible 4-hit combinations in the SMT barrels, thus bypassing
the histogramming step.
Full documentation about elastic track finder can be found
here.
Elastic arms
elastic_reco is an offline package for reconstructing tracks with pT > 100
MeV from the full CFT and SMT detectors (|eta| < ~3.5). It has shown decent
performance in reasonable time for many types of MC events as well as the real
data. There also is an integrated graphical display for viewing/debugging the
tracking results and a root tree output for measuring performance.
The algorithm starts by selecting a set of primary vertices in each event. If
none are available, it can make its own, using other information in the event).
Tracking begins with "seeds" which are filtered from a histogramming technique.
These seeds are then fit "elastically", meaning they continuously change their
probabilities of being connected with each hit as their track-parameters
evolve. In particular, hits that were not originally associated with the track
seed can become associated through the elastic fitting process.
Full documentation about elastic track finder can be found
here.
vkuznet@fnal.gov