When running the L2 alignment with TRT-Only tracks (A and C side together) in 3 DoF we find that the combined resiudals are miminized.
However we find that when looking at the residuals on side A and C seperately there are residual discreapencies.
Currently it is not for seen to have module level alignment constants seperately for side A and C.
During reconstruction the detector positions are stored in GeoModel and retrieve via, the InDetReadOutGeometry package.
GeoModel only has one alignable-transform per barrel module.
Talking to the experts Grant and Kittleman, GeoModel changes are not something one does lightly and can take years to test and validate.
It turns out that the Pixels have had a similair problem. It was found that there was significant bowing in the pixel detector and it needed to be accounted for offline when running the reconstruction.
This however was not forseen to be in the GeoModel.
The fix was in the InDetReadOutGeometry, which retrieved the constants from GeoModel, and then applied a correction to account for the pixel bowing before passing the numbers to the reconstruction.
It seems like this is the way to go for the module level A/C side differences, however if we are going to allow for general single straw movements (currently the straws can only move in the plane of the straw layer) in the future, it might be better to do this now, and then put in module level differences as coherent movements of single straws.
Work-around.
By masking of hits from a given side of the detector during recostruction (How to do this is explained here. ) one can effectively align the detector one side at a time.
This has been done using TRT-Only track in 3DoF. (it was put in the db as TRT_Cosmic_2008_08 and TRT_Cosmic_2008_09).
Result are sumarized here.
The following shows the differences in the alignment parameters when aligning sides A and C seperately.
