Title: Ghost Trackers
1Ghost Trackers
Dave Jackson Oxford University / RAL
- If theres something strange (or charm or
bottom) in your neighbourhood
LCFI Collaboration 28th March 2006
2ZVRES
ZVKIN
In the conventional ZVTOP algorithm secondary
vertices are found first. The linearity of the
B-D decay chain is then used for L/D track
attachment
In the Ghost Track algorithm the straight
IP-B-D topology is exploited first, to estimate
the B/D flight direction before vertex finding
3The straight ghost track is anchored at the IP,
initially given a 25µm width then moved in ? and
f to minimise ? ?2 (sum over jet tracks). Final
fitted width of ghost track (minimum 25µm)
calculated for compatibility with jet tracks
each has ?2 1.0 with fitted ghost track
SLD
4The SLD Ghost Track used straight Gaussian tubes
for fast analytic track fits this required
reparametrising the track near a vertex
location. This should not be so much of an issue
for the LCFI C code in which the approximation
is not made
5Care with errors and ?2s important since vertex
finding relies on probability of vertex fit
calculation
SLD
PROB(??2,2N-3) For N tracks in jet from same MC B
vertex (spike at zero due to non-Gaussian tails)
PROB(??2,2N-4) Where N now includes the Ghost
Track in the fit so less free than the jet
tracks alone. Probability is now a measure of a
good secondary vertex fit AND compatibility with
the B direction (ghost track).
6SLD
The algorithm proceeds to build vertices
according to the highest Probability while
Prob1. So this distribution needs to be fairly
flat for genuine vertices and to be flat for a
range of track multiplicity (shown here), decay
length, etc. The pre-requisite for this is a
fitter that has the right properties for jet
tracks alone.
Probability of ghost B tracks fit
7At SLD the Bs came in back-to-back pairs. The
EVENT would be tagged with ZVRES before running
ZVKIN for analysis of each jet
For LCFI generally would like to consider each
jet independently design flavour/charge tagger
for each jet with ZVTOP3 in C
SLD
8BACKUP SLIDES
9L/D for non-Seed tracks passing T Cut at L/D 0.3 to attach tracks from B decay
chain to Seed
10- Highly boosted B kinematics IP?B?D straight to
1 (for Z0) - B,D vertex locations are not independent in 3D
space - Ghost Track Algorithm builds in this
information from the beginning
11- Stage 1
- Pivot straight ghost track at IP, initially
along jet axis direction - Give ghost track a 25µm width and calculate ?2
of ghost to each track in jet - Swivel ghost track in ? and f to minimise ? ?2
(sum over jet tracks)
12- Stage 2
- PROB(?2,ndof) for jet track(s) to be consistent
with each other and with the ghost track or IP
ellipsoid is constructed - For jet with N tracks, initially N1 candidate
vertices
1
2
N1
Ghost
IP
3
-
- Calculate fit probability for all pairs of
objects (if IP is not included, then ghost track
is added) - If maximim PROB PCUT (typically 1) then
combine the two objects and iterate - Else vertex reconstruction is complete
N
PRI
SEC
TER
Allows reconstruction of 1-prong vertices
13The Topologies
For a B decay to a single cascade charm
D
lD
B
lB
IP
Measured lB
Measured lD
True MC lB
True MC lD
cm
cm
14Compare ZVTOP with Ghost Track Algorithm
ZVTOP
GHOST
Number of Found Vertices
ZVTOP
GHOST
GeV/c2
B Decay Invariant Mass
15Options
- The Tidy cuts
- For SLD 20 of jets contained 1 high impact
parameter track from Ks, ?, detector interaction
etc. - Tidy cuts are applied first to prevent the Ghost
Track direction being distorted (half background
tracks removed at SLD) - For each algorithm 4 tunable parameters that
effect efficiency vrs purity of vertex
reconstruction - also
- ZVTOP can guide vertex finding with V(r)
weighting - Ghost Track
can
force the topology to find fixed number of
vertices Momentum factor
16World BS mixing sensitivity
B0 b?c D or D0
July 2003