Provisionary fix for ITS fit with B=0 - #10261
Conversation
Sorry, something went wrong.
|
Hi @shahor02, it looks good to me. I copied the code from the Cooked matrix long ago, so I cannot exclude that I did some mistakes in the process. However, we can even remove this part of the code and do a full fit of the candidate by providing a reasonable set of parameters and a very large covariance as starting conditions. |
Sorry, something went wrong.
|
@mpuccio starting from large diagonal cov matrix would be the right choice later you fit through the all clusters, but I was not sure if you do this: I see that you make a piecewise fit AliceO2/Detectors/ITSMFT/ITS/tracking/src/TrackerTraits.cxx Lines 537 to 544 in 4d49728 Does it fit through all clusters contiguously, including the cluster3 ? |
Sorry, something went wrong.
Hi @mpuccio @davidrohr
The ITS tracking with B=0 produces a weird q/pT spectrum instead of fixed (modulo e.loss) "most probable" q2pt:
The reason is that in the
AliceO2/Detectors/ITSMFT/ITS/tracking/src/TrackerTraits.cxx
Line 913 in d869925
This PR attempts to fix this, but I don't understand the way you build the cov. matrix in general: should not it correspond to the matrix updated only at cluster3, i.e. only diag. errors in Y and Z set to resolution^2, other diagonal errors "as large as possible" and non-diagonal ones set to 0?
With this fix the result of the fit becomes:
and the final covariances of q/pT with other parameters are 0 within ~10^-7
@davidrohr, for the TPC I see something different: the mean q/pt corresponds to conventional "most probable" but then there are huge tails on both sides
and also I see a significant correlation between the q/pT error and e.g. y, this cannot happen with B=0.
Could you please check if you are not initializing with the covariance with some non-diagonal elements?
It seems that currently all fits with B=0 are wrong.