140 likes | 346 Views
Tracker Alignment with MillePede. Why we need alignment? Alignment with MillePede Algorithm Alignment with TIF trigger configuration on simulated cosmics data Preliminary results Status of the alignment with TIF real data Conclusions and outlook. Outline. Why we need alignment?.
E N D
Why we need alignment? Alignment with MillePede Algorithm Alignment with TIF trigger configuration on simulated cosmics data Preliminary results Status of the alignment with TIF real data Conclusions and outlook Outline R.Castello
Why we need alignment? • The CMS tracker is build in order to optimize the particle momenta resolution. • It depends on two factors: C1 is geometry -dependent C2 depends on MCS B = magnetic field L = track length n = # hit of the track σx = resolution on the measured point Systematic error can be minimized by a correct alignment ~ 10 μm (Si) R.Castello
Survey measurements (during assembly) s = 100-200 mm Laser Alignment System (LAS) : alignment of TIB vs TOB etc ... Track Based Alignment (cosmics, Z→mm, etc...) s = 10 mm The problem 16k microstrip modules 6 d.o.f per module O (100k) unknowns Complex system of equation to solve efficient and fast algorithm For CMS tracker alignment 3 algorithms: HIPMillePedeKalman Filter (Helsinki, Milano, Perugia)(Hamburg, Torino)(Wien) The tracker alignment R.Castello
MillePede Algorithm MILLE PEDE ( V. Blobel ) • A measurement can be written as • Linearised least square combined fit of alignment parameters (global) and track parameters (local) • MIllePede uses the Method of the Least Squares residuum, where mk is the measurement with uncertainty sk and dk is the coefficient vector. the idea : minimize square of residuum. global local R.Castello
Interested to the nglobal parameters. For a set of N measurements: from (n+Nn) equations to n C’ matrix inversion (Computational time ~ n3 ) With MillePede you can align @ different levels (Detectors, String, Layer level, etc…) 6 degrees of freedom for each alignable structure (6 parameters): 3 shifts (respectively along local x,y, z) and 3 rotations (around x,y,z) the aim it’s to find a with good uncertainty Alignment with MillePede R.Castello
- CMSSW 1_3_1 version - Alignment/MillePedeAlignmentAlgorithm (tag branchV00-07-0X-01) - Cfg file: AlignmentTrackSelectorModule:- selection of Tracking algorithm - set APE - set cut for tk selection (pt, #hits,..) AlignmentProducer: - selection of alignment parameters - geometrical selection in eta, phi,z - selection of misalignment scenario - set solving method ( inversion, etc.) - set c2 range acceptance PoolSource: - selection of dataset - set number of events MillePede Algorithm in CMSSW(implemented by G. Flucke, Hamburg) R.Castello
Cosmics simulation @TIF (Tracker Integration Facility) Since Feb.’07 the 25% of the tracker system is under commissioning Old scintillator configuration (36k events) New scintillator configuration (20k events) R.Castello
Results on TIF cosmics simulated sample (old trigger configuration) with Short Term misalignment scenario (~100 pb-1) RMS =151 mm RMS =263 mrad Rod Level (79 alignment parameters) R.Castello
Du (Strings level) Good correlation between Du at the start (misaligned) and at the end (after the alignment) Global correlation : largest correlation of a parameter with any linear combination of all other parameters. A value close to 1 means solution not well determined. R.Castello
Old and new MC configuration agreement for Du To be understood.. Simulated cosmic sample with new scintillator configuration (y axis) Simulated cosmic sample with old scintillator configuration (x axis) R.Castello
Du and g for real data (10th March run) Too precise if compared with the “expected” results RMS =21 mm RMS =79 mrad Rod Level (71 alignment parameters) under studying… R.Castello
Conclusions & outlook • Preliminary results obtained running MIllePede alignment algorithm on TIF simulated sample for Du and g parameters at two different hierarchical levels (Strings & Layer ) • No relevant differences between new and old scintillator configuration on simulated data (a first indication of algorithm stability) • The exercise on real data is going on • Good impression and relevant hints from recent Hamburg workshop: http://indico.cern.ch/conferenceDisplay.py?confId=16095 R.Castello