[Halld-offline] Kinematic fitting and error matrices

[Will Levine]

Huh, I just figured out that I'm supposed to use
DTrackFitter=KalmanSIMD and not DTrackFitter=Kalman. That seems to fix
the problem of the absurdly large error matrices.

Will

On Thu, Jan 20, 2011 at 1:52 PM, Will Levine <wilevine at andrew.cmu.edu> wrote:
> Hi,
>
> I've been looking into kinematic fitting and just thought I should share a
> couple issues I've been running into, in case anyone has any ideas or
> thoughts.
>
> Using the error matrices from the Kalman fitter, if I run a kinematic fit,
> and cut out events with confidence level<.01, it seems to do its basic job
> of cutting out most "bad" events and keeping some good events.
>
> However, there are a few problems, which lead to too few events passing the
> cut.
>
> Pulls indicate that the error matrices are somewhat off, in particular, that
> the errors are too small. Simply scaling the error matrix (as given in
> px,py,pz) does not solve the problem, so it will be necessary to look at the
> errors of the tracking variables.
>
> A few of the error matrices are totally bogus. i.e.
>
> 7x7 matrix is as follows
>
>    |      0    |      1    |      2    |      3    |      4    |
> ------------------------------------------------------------
>  0 | 9.425e+141 -4.773e+142  2.402e+142 -6.569e+140  1.786e+143
>  1 |-4.773e+142  2.379e+143 -1.197e+143  3.273e+141 -9.047e+143
>  2 | 2.402e+142 -1.197e+143  6.031e+142 -1.649e+141  4.555e+143
>  3 |-6.569e+140  3.273e+141 -1.649e+141  4.507e+139 -1.245e+142
>  4 | 1.786e+143 -9.047e+143  4.555e+143 -1.245e+142  3.473e+144
>  5 | -8.99e+143  4.554e+144 -2.293e+144  6.266e+142 -1.748e+145
>  6 |-4.053e+141  2.053e+142 -1.033e+142  2.826e+140 -7.805e+142
>
> I'll look into whether these tracks are identifiable or bogus in any other
> way.
>
> If any one has any insight into these issues, I'd be happy to hear it.
>
> Will
>