[Halld-offline] error matrices

Will Levine wilevine at andrew.cmu.edu
Thu Mar 3 14:13:11 EST 2011


As I mentioned in the physics meeting, the 5x5 tracking error matrices
provided by reconstruction are not equivalent to the 7x7 error
matrices, which are used in kinematic fitting. I think I tracked down
the sources of this problem in DTrackFitterKalmanSIMD.cc.

1) The first problem is simpler and is due to a bewildering feature of
ROOT. TMatrixDSym is supposed to be a symmetric matrix, but this
symmetry is not enforced. The matrix must be filled completely to work
properly.

2) The 7x7 matrix is set (line 354) via cov (line 350). cov is a 5x5
error matrix in the central parametrization. cov is set from Cc (line
2404). Cc comes from Cbest (line 2381), which is a 5x5 error matrix in
the forward parametrization. The problem is that Cbest is not the same
5x5 matrix used to setTrackingErrorMatrix() (line 373). That one comes
from Clast (line 2410), which is not equal to Cbest. I'm not sure
which of these is what we want. To my ear, Cbest sounds better, but if
someone (simon?) could confirm this, that would be helpful, then I
could check in a fix.

I've attached my attempted fix and also a couple of functions to
compare the 5x5 and 7x7 matrices in a post-reconstruction
DKinematicData.

Will
-------------- next part --------------
A non-text attachment was scrubbed...
Name: DTrackFitterKalmanSIMD.cc.diff
Type: text/x-patch
Size: 1311 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110303/1f315041/attachment-0006.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: compare_matrices.h
Type: text/x-chdr
Size: 3774 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110303/1f315041/attachment-0007.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: TMatrixDSym_test.C
Type: text/x-csrc
Size: 1223 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110303/1f315041/attachment-0008.bin>


More information about the Halld-offline mailing list