Re: Fwd: Kfit with timing

[Matt <bellis@ernest.phys.cmu.edu>]

Hi Eugene,

 One may treat it as a superposition of several Gaussians,
with no way to separate them on the track to track basis. Perhaps, by now
someone has figured out how to treat such things consistently.

   GlueX-doc-818 details Curtis and Mike Williams' experience with kinematic fitting in CLAS and the parametrized corrections they had to make to the error matrix to get good pulls and confidence levels. The errors, as you say, tend to be non-Gaussian and so the tails mean that a 10% CL cut may remove more than 10% of the signal. However, if this is understood and handled in a consistent way between real world data and Monte Carlo, this does not adversely affect the analysis. This worked very well in CLAS and papers will be coming out shortly having used this procedure for PID. 


-- 
-- 
----------------------------
Matt Bellis
Carnegie Mellon University
(office) 412-268-6949
(cell) 412-310-4586
----------------------------