Multiple track-fit solutions for improved reconstruction of event topology

[cherylepatrick]

How things work now

Individual clusters of tracker hits are fitted independently by the TrackFit module. No information other than the hits in that cluster is used in the fit (so - nothing about the rest of the tracker topology, no calorimeter information etc) As we know radius but not direction of particles passing through a cell, tracker tracks are fitted by joining a series of circles. This gives some ambiguity in the set of possible tracks a particle can have taken. The TrackFit module currently reports several solutions (from different initial conditions) for fitted helical and linear tracks, with a chi-squared value to indicate the goodness of fit for each. These are written to the TTD bank, with a marker for a preferred default solution - the one with the lowest chi-square.

Only the default solution for each track is used when constructing the final event topology, and used to generate the particle records written to the PTD bank - vertex extrapolation to calorimeter / source foil, charge ID etc. If two tracks have the same chi-square, one will be arbitrarily chosen as the default and used from then on. The PTD bank presents a single topological solution, with alternatives, and with no indication of the likelihood of the selection over others.

The problem

In some cases, the rest of the event topology can make it very clear that the "most likely" default solution selected by TrackFit does not correspond to the most likely track of the particle. Typically two tracks are equally, or almost equally, plausible (as in the first example shown below, a track containing hits all in the same column of cells will always have two symmetrical solutions, at the top and bottom of the track, which are exactly equivalent). However an alternative solution (either with the same or a slightly higher chi-square than the selected default) may be favourable if one considers factors like:

• A shared vertex with another track (typically a shared foil vertex). Here, the top track option for the left-hand particle shares a vertex with another track, while the bottom one (same chi-square) does not:

• A vertex on a fired calorimeter block (if extrapolating the outermost vertex of two tracks to the calorimeter wall, does one hit a calorimeter that detected a particle, and the other hit a calorimeter that did not?)

• Particle charges: in the case of two equally-likely tracks, the opposing curvatures of the two solutions will identify one as a positron and the other as an electron, affecting whether or not they pass our charge cuts.

** Possible solutions **

When developing an event topology, other track-fitting solutions should be considered as well as the default solution. In the first case, this would require extending vertex extrapolation to include non-default solutions. An improved version could either continue to present a single topology solution, but with more sophisticated testing of alternatives (but how would we rank low chi-square vs common vertices vs charge considerations etc...?) or to provide several solutions, along with some kind of confidence ranking (similar issues).

[cherylepatrick]

Matteo's presentation on this
Notes from discussion of that presentation