Event/Track Selection for Alignment – S. Gonzalez Study of different samples with pile-up and muons in the final state (Z→  (ee), W→  b→  X and Minimum bias First version of InDetTrackSelTool available First look at the matching Muons/ID Results of W →  are very similar to those obtained with Z →  (ee)  N shared cut has small impact in efficiency and fake rate  N holes cut become significative if request N holes (MAX) = 0 Lots of tracks without truth association in the cases b→  X and Min.bias  Look into more detail how association is done  Tune parameters ?  Something wrong we did ? Next:  Run with misaligned setup and test effect of cuts  Proper definition of efficiency and fake rate

Efficiency vs fake rate p T > 2 p T > 4 p T > 6 p T > 8 p T > 10 p T > 20 n shared  7n shared  6 n shared  5 n shared  4

Correlations among variables Correlation between isolation and number of shared hits Requesting that the number of shared hits is 1 removes the great majority of non-isolated tracks

Other possibilites ? Try cut on p instead of p T ? MS depends on p, not p T  barrel : p  p T  endcaps : p > p T If cut on p is performed instead of p T :  small impact on barrel  Endcaps: recover tracks with high p, otherwise lost if cut on p T  increase statistics in endcaps  MS do not get worse in this region with high X 0 endcap region barrel region Minimum bias

Summary on out of plane distortions/tilts – R. Hawkings Radial misaligments –Need to control pixel and SCT barrel radial misaligments to around 25  m –For SCT endcaps, 100  m ‘radial’ (actually Z) misalignments is more than enough Module bows –Observed level of bows in pixel barrel modules dominates resolution for high p T track parameters – need to measure this and improve by at least a factor 2 –Observed levels of bows in SCT endcap modules do not contribute significantly to tracking resolution –Need to try more realistic models for pixel endcap and (perhaps) SCT barrel Module tilts –Need to control module tilts to around 1mrad in both pixels and SCT, tilts around short axis (i.e end of SCT strips up/down) are more important

A comment on the W mass For making a useful measurement of m W, require two things: 1.Absolute momentum scale understood to 0.02% (e.g. at 50 GeV) 2.Momentum resolution understood to 1% of value Requirement (2) is easier –Residual non-understood random bows/tilts etc should be at a level of 14% of resolution or better (  (1+0.14^2)=1.01) – say 10% for safety Not so far from that at 50 GeV in p T resolution for most effects Requirement (1) is very hard – requires residual systematic effects common to all modules must be very tightly controlled – preliminary calculations: –Levels like  m for transverse alignment, 1  m for radial alignment –Short axis residual systematic tilts to 0.05 mrad (pixel), 0.03 mrad (SCT) –Module bow parameters to mm -1 – i.e. uncertainty must be around 50 times better than observed values… –SCT module V-bends to 4  m This seems just as scary as for the transverse alignment –Latter already known to be very hard

Conditions database and COOL COOL is beginning to be usable in Athena in release ; some features still missing… –Hope for implementation usable for ID alignment in (COOL 1.2.1) Need tagging functionality (should be there soon) and possibly support for multiple Storegate keys in one folder With introduction of COOL, plan several other things: –Remove separate transforms for side 1 of SCT modules (couple r-  and stereo sides) –Possibly condense current set of 34 folders used to determine position of all layers into one folder with multiple keys –Review reading/writing scripts, try to rationalise and make more robust –Move to production Oracle COOL CERN, centralise all test ‘tags’ there Also start to introduce ‘real’ misalignment data from surveys –Use new DDM system to centrally manage POOL data files Maybe in release , certainly for September (release 11) Also need to add in description of module distortions – common format?

TRT Calibration and Alignment – P.Hansen Migrating to Athena Strictly a technical exercise at this point, i.e., no pretty plots shown at workshop

ס ס ס B A Robust Alignment using overlap residuals – F. Heinemann Main Principle Use overlap residuals for determining relative module to module misalignment Overlap residual = inner hit residual – outer hit residual Mean of A – B ≈ M M

Sort into groups StavesRingsModules

With misalignment Example SCT barrel 5 and 6 Sector 2 RPhiRPhi Ring OR in mm 0.03 mm ∆mean = ∆mean = RPhiRPhi Ring Barrel 5 Barrel 6 OR in mm

Conclusion / Outlook  Algorithm works down to module level  Get local x and y alignment for PIXEL barrel  Get local x alignment for SCT barrel  Version 1.4 in CVS  Need more statistics Next steps:  Further tests with misalignment  Start looking at endcaps  Start looking at iteration process Still a lot to do …

Robust Alignment using Chi2 fitting – R. Härtel Work done mostly for SCT Correctly clusters hits Capable of doing iterations Alignment results dependent on fitting package –XKalman (see pp 8-12) shows what can be achieved –KalmanFitter (refitting package; see pp 13-15) still produces some funny results (distributions too wide?) –Should improve once tracker and ESD are synchronised in Athena Still need to do large-scale validation

Robust chi2 alignment for pixels – T. Göttfert Toy MC of a single pixel module Use robust chi2 to determine 6 alignment parameters Pull distributions have correct shape, but are too broad (see p 7) –Need to improve error calculation (p 3)? Zero alignment clearly seen Misalignment under development

Global chi2 alignment – A. Hicheur What have we learnt? The SiAlignment code does a good job and correct biases that were present before Both eigenmodes and alignment constants pulls are Ok –Very strong point for validation The eigenvalues spectrum is sensible Code is running smoothly with a reasonable timing (~ 5 ms/track)

Conclusion – next steps Code behaves correctly and almost ready for Tag Collector and release –Validation done with tracking (to allow for comparison with the old exercises) –To be done: check with new tracking (format) in 10.x.y, provide feedback to tracking experts Further tests foreseen with misaligned layouts (at the reconstruction level) –First step: redo the exercise with the same sample Further developments foreseen, e.g.: –Constraints –Endcap alignment –etc… Get ready for coming challenges –Extensive testing with “DC3” samples in Autumn

Alignment constant pull – Tx no vtx fit With old prototype and 45k evts Current SiAlignment and 20k evts Pixel bias

Alignment constant pull – Tx vtx fit With old prototype and 45k evts Current SiAlignment tag > and 20k evts Pixel “bias” removed

Chi2 Alignment Status – P. Brückmann New graduate student – reinforcement of the team I’ve had a very first look into the production code Something is wrong in rel (technical – not solved yet)  Very good news: I seem to understand what SiAlignment does and ‘m able to ~reproduce the results with the prototype code. All differences (as long as the basic algebra is correct) is down to the definition of the derivatives (predominantly de/dp) – has always been like that! However, I still believe the prototype implementation of derivatives is correct – puzzling!?

Unpleasant surprise… multimuring_2.rz ~20000 ev Processed with:  ATHENA rel  InDetAlignment  SiAlignment Pull plot still looks deceivingly alright Problem: why does it look this bad?

Problem: why does it look this good?  Looking through SiAlignment code could not see any obvious reason (difference relative to what the prototype does).  I made one step back and loked into InDetAlignment for the calculation of derivatives w.r.t. track parameters.  I found differences in three out of five: phi0, cot(theta), Q/p_T  Implemented the “InDetAlignment” calculation in my code  Processed ~20,000 events on old ntuples made with 8.7 (corresponding to multimuring_2.rz).

Prototype a’la Adlene using ntuples from ATHENA 8.7 Adlene’s direct result from ATHENA 8.7 The two are not massively different. Residual differences may result from minor differences in treatment of missing hits and scattering angles. Will have a closer check into this. Encouraging comparison… Ring 2

…and the pull distributions

Parallel Computing Status – M. Ünel Looking to add quadruple precision to use 64-bit processors (?) Two main options: –AMD-Scalapack – optimised for AMD Opteron Less promising since we have high condition numbers –NAG libraries – requires license More promising, but implementing quadruple is non-trivial Case studies also done

Multiple Coulomb scattering with CTB04 data – S. Martí Inserted 10mm Al plate between pixel detector and SCT to simulate service feedthrough Pixel and SCT residuals naively expected to depend as 1/p 2 CTB04 data does not follow this dependence (see pp 9-10 of talk) Due to hard scatters in Al layers rather than gradual scatters throughout material (see p 8)

Analysis of Endcap Pixel Alignment Survey Data – A. Korn for T. Golling Bow in barrel pixel modules described by circular shape (Vadim Kostyukhin) For endcap pixels, third-order polynomial gives good description –Bows in endcap (4-8  m) smaller than for barrel –Bows comparable to precision of data points (3-10  m), so negligible effect for alignment? “Twist” (tilt along long axis) also visible in survey data See pp 11, 12, 16 from talk