Tuning Parameters in the Kalman Filter fit
In principle the Kalman filter should contain no fudge factors. In the past two fudge factors have been introduced:
- the multiple scattering fudge factor
- the average energy loss
Only if all items below are fulfilled, there will be no fudge factors and the covariance matrix should be correct.
- The amount of material seen by a particle traversing the detector must be included in all detail in the reconstruction; i.e. the transportservice must be used with all details.
- The non-Gaussian multiple scattering tail can be neglected
- The magnetic field tracking algorithm has high numerical precision; i.e. a high order stepper should be used
- The covariances (or "weights") of the detector measurement should be correct. This can be a difficult criterion to meet. The error of a Si clusters with 1 or 2 hit strips is not Gaussian. The error depends on the track angle, the Phi strip error depends on R, etc.
- The Kalman filter must be seeded with an initial state which either:
- is obtained form other measurements as the one used inside the fit,
- has a covariance of infinite magnitude (in other words: a weight that is negligible in the final fit).
For a fast fit it is conceivable that a fudgefactor is introduced, eg. to compensate for a non precise material description
--
MarcelMerk - 11 Apr 2005