Presentation on theme: "ATLAS Experiment Silicon Detector Alignment 1 MKU IPRD06 Müge KARAGÖZ ÜNEL the University of Oxford for the ATLAS Silicon Alignment Community IPRD06, Siena,"— Presentation transcript:
ATLAS Experiment Silicon Detector Alignment 1 MKU IPRD06 Müge KARAGÖZ ÜNEL the University of Oxford for the ATLAS Silicon Alignment Community IPRD06, Siena, Italy October 1-5, 2006
2 MKU IPRD06 The Prelude z y x 6 PIXEL modules 8 SCT modules All results are preliminary. There will be a bias will towards cosmic results and global chi2 algorithm. For information on the SCT+Pixel hardware, installation and performance, see Sergio Sevilla, Harald Fox and Andreas Korn’s talks. Combined Test Beam Surface Cosmic ATLAS Pit
3 MKU IPRD06 The Challenge: Atlas Silicon is BIG! SCT barrels: 4 layers SCT endcaps: 9 disks Pixels (3 layers+3 disks) TRT BarrelForward DetectorPIXSCTPIXSCT # of layers/disks342x32x9 # of modules x1442x988 sub Total Total ,992 In total we have to deal with 34,992 DoF’s! 3 translations & 3 rotations of each module
4 MKU IPRD06 The Strategy: Various Approaches Prerequisite: need to cope with the demands of ATLAS physics requirements –intrinsic alignment of Si (offline, online, survey) –Si+TRT (help sagitta, etc..) Methods: all rely on track residual information Global 2 : –minimization of 2 fit to track and alignment parameters –6 DoF, correlations managed, small number of iterations –Inherent challenge of large matrix handling and solving Local 2 : –similar to global 2, but inversion of 6x6 matrix per module –6 DoF, no inter-module or MCS correlations, i.e. diagonal covariance matrix – large number of iterations Robust Alignment: –use weighted residuals, z and rphi overlap residuals of neighboring modules –2-3 DoF, many iterations, no minimization Valencia Alignment (used mainly for CTB): –Numerical 2 minimization –6 DoF, many iterations
5 MKU IPRD06 Algorithms’ Functionalities DigitsReconstruction Alignment Algorithm Alignment Constants Tracks in StoreGate Final Alignment Constants Iteration until convergence, N_iter different for all algorithms All algorithms are implemented within ATLAS framework and able to use the common tracking tool and database Functionalities exist to add constraints from physics& external data (vertex, mass requirements, online alignment and survey data) for the global and/or local chi2 minimization algorithms
6 MKU IPRD06 Global 2 approach The method consists of minimizing the giant 2 resulting from a simultaneous fit of all particle trajectories and alignment parameters: Let us consequently use the linear expansion (we assume all second order derivatives are negligible). The track fit is solved by: while the alignment parameters are given by: track hit residual Key relation! Intrinsic measurement error + MCS Equivalent to Millepede approach from V. Blobel
7 MKU IPRD06 Combined Test Beam (2004) z y x 6 PIXEL modules 8 (-1) SCT Modules (1 dead) First real data from ID subsystems! Abundant statistics of e+/e- and pions (2-180 GeV) (O(10 5 ) tracks/module/E), Magnetic field on/off runs. Limited setup and layout, creating systematic effects in modes A good start to test algorithms for more realistic upcoming data! Algorithms use different approaches in extracting alignment constants: various DoF, (un/)biased residuals, etc.. All algorithms improve residuals and quality of the track parameters significantly. Consistent results within slight differences (likely to be attributed to global transformations) Ongoing efforts to combine/compare results of the algorithms.
8 MKU IPRD06 CTB – Performances - I Before After PIXEL SCT 3D unbiased residuals – run Iteration 1 and 15 Robust: converge after about 15 iterations only detector plane alignment (X,Y) after alignment, pixel residuals 0(10 m) Local chi2: converge after about 15 iterations after alignment, SCT residuals are O(25 m) and pixels are 0(10 m)
9 MKU IPRD06 CTB – Performances - II =19.1 μm Global chi2: on 3DoF (detector plane, TX,TY,RZ), quick convergence results with anchorage or no anchorage overall pixel residual mean is 10.7 mum and SCT is 19.1 mum SCT before SCT after Valencia: align pixel first, then SCT, then all first pixel as anchor, many iterations
10 MKU IPRD06 ID Cosmic Run (2006) Scint 3 Scint 2 Scint 1 Data-taking at surface (SR1) in spring 2006: ~400k events recorded! 22% of SCT Barrel: 467 instrumented modules 13% of TRT Barrel 3 (or 2) Scintillators to trigger 144cmx40cmx2.5cm Trigger Time resolution ~ 0.5ns: ~ MeV cutoff for alignment studies No B-field! No momentum! MSC important ~<10 GeV, need to deal with large residuals floor acts as absorber for p< 170 MeV Loss of efficiency for small d0 and large phi0 Use CTB Straightlinefitter
11 MKU IPRD06 SR1 Cosmics: global 2 - real data TX TY TZ RX RY RZ [mm] [mm] [mm] [mrad] [mrad] [mrad] Barrel Barrel Barrel Barrel first iteration Corrections due to modes >1500 Alignment corrections for rigid barrels =31.9 μm =46 μm Corrections x100 Indication of very good assembly precision! Better than misaligned simulation! 2808 DoF’s, ~250k tracks Before alignmentAfter alignment
12 SR1 Cosmics: global 2 - real data MKU IPRD06 First iteration Third iteration Correction x100 systematic effects visible when only 6 modes removed before after
13 MKU IPRD06 SR1 Cosmics: global 2 - simulated data =31.2 μm Perfectly aligned detector, but still minor bugs in detector descriptions Traces of systematics??? Beware: Alignment is ultimately sensitive to own bugs and approximations but also to slightest upstream imperfections. Sometimes hard to tell apart!
14 MKU SR1 Cosmics: local 2 Flow of alignment parameter a x for all modules of barrel layer 2 through the iterations Alignment of SCT barrel cosmic test setup with 36k tracks. BEFORE AFTER = > μm =99.5 -> μm =96.1 -> μm = > μm rnd misal: IPRD06
15 MKU IPRD06 SR1 Cosmics: Robust alignment After 8 iterations: RMS residual 61µm → 47µm After 8 iterations: RMS overlap residual 81µm → 69µm The method works on the cosmic setup, ongoing studies with refitted tracks. 10k tracks
16 MKU IPRD06 Nominal Detector Simulation: Global 2 1M muon sample (~8M tracks in full fiducial) generated for ATLAS Computing Challenge. The largest sample looked currently! Align 2172 Barrel SCT+PIX modules in |eta|<1 (“Double Cone”), DoF Solution using ScaLAPACK on an AMD Opteron ||-cluster (< ½ hour). Systematic effects in Pixels which was not visible to earlier studies: statistics washed out or sth new? Disappear after cut on low freq. modes (i.e., global effect) Special misalignments can also be introduced in three levels: –module level, layers&disks, Silicon ID subsystems –Data analysis of this is ongoing.. x100
17 MKU IPRD06 x100 Pulls of alignment corrections Typical errors < 10μm PIXEL Pulls in diagonal base Nominal Detector Simulation: Global 2
18 MKU IPRD06 Including online alignment: FSI Barrel SCT End-cap SCT 165x2= (3x[80+16])+(2x72)=512 On-detector geodetic grid of 842 simultaneous length measurements (precision <1 m ) between nodes on SCT support structure. Grid shape changes determined to < 10 m in 3D. Time + spatial frequency sensitivity of FSI complements track based alignment: –Track alignment average over ~24hrs+. high spatial frequency eigenmodes, “long” timescales. –FSI timescale (~10mins) low spatial frequency distortion eigenmodes (e.g. sagitta). Need both! First principles studied, implementation work is to be performed! Spatial frequency eigenmode FSI Time seconds minutes hours days months Tracks
19 MKU Various algorithms are adapted to optimally align ATLAS ID. Algorithms have proved proof of principles. Codes have been continuously implemented and heavily tested. We have been looking at real data already! At SR1 data taking, SCT barrel was found to be built very well! CTB efforts are almost finalized. Cosmic alignment progressing rapidly. We are on our way to understand and tackle many systematic issues, both in real data and simulation, upstream and downstream of alignment algorithms… Pixels and endcaps are getting ready for cosmic data. FSI is getting ready for monitoring during pixel insertion in pit. Stay tuned! IPRD06 Summary & Conclusions
20 MKU IPRD06 We are… The Silicon alignment team –Pawel Bruckman, Stephen Gibson, Tobias Golling, Carlos Escobar, Richard Hawkings, Roland Heartel, Florian Heinemann, Adlene Hicheur (ex), MKU, Stefan Kluth, Salva Marti Garcia, Bjarte Mohn, Ola K. Oye, Sergio Gonzalez Sevilla, Jochen Schiek, (am surely missing a few, please help me finalize this list!) And –The hardware and DAQ teams of all setups –Reconstruction groups –CERN crew for installation and survey data More information: –https://uimon.cern.ch/twiki/bin/view/Atlas/InDetAlignmenthttps://uimon.cern.ch/twiki/bin/view/Atlas/InDetAlignment
21 MKU IPRD06 BACKUP Si modules CTB+SR1 Details Global chi2 details Full barrel solution Robust align. details FSI Details Survey Data
22 MKU IPRD06 Building blocks of SCT ~140mm ~70mm ASICs PIXel detectors provide real 2-D readout (size 50 400 m resulting in 14 115 m resolution). SCT modules are double-sided strip detectors with 1-D RO/side (768 instrumented strips). Sensitive strips have pitch of 80 m giving 23 m resolution. Stereo-angle of 40 mrad gives 580 m resolution in the other direction. Entire Si tracker is equipped with binary RO chips. SCT end-cap modules are different in shape: wedged structure TPG+BeO Baseboard Si Sensor Hybrid with ASICsSi Sensors
23 MKU IPRD06 Weak distortional modes (ex.) “clocking” R (VTX constraint) “telescope” z~R radial distortions (various) dependent sagitta X a bR cR 2 dependent sagitta “Global twist” Rcot( ) We need extra handles in order to tackle these. Candidates: Requirement of a common vertex for a group of tracks (VTX constraint), Constraints on track parameters or vertex position (external tracking (TRT, Muons?), calorimetery, resonant mass, etc.) Cosmic events, External constraints on alignment parameters (hardware systems, mechanical constraints, etc).
24 MKU IPRD06 Global chi2: more on modes Example “lowest modes” in PIX+SCT Global Freedom have been ignored (only one Z slice shown) The above “weak modes” contribute to the lowest part of the eigen-spectrum. Consequently they dominate the overall error on the alignment parameters. More importantly, these deformations lead directly to biases on physics (systematic effects).
25 MKU IPRD06 Robust alignment: constants Sum over neighbours, take correlations into account Correct for change in radius Sum over all modules in a ring For a perfect detector:
26 MKU IPRD06 Solving the full system ● Global 2 formalism requires solving large number of LE and handling large sized-matrices. ● Limiting factors: ● Size: * Solving full ID needs 9.8GB. By default, Atlas software infrastructure allows only 2GB of memory/job on 32-bit machines. ● Precision: Conventional 32-bit libraries are not fully adequate for full size solution accuracy. Alignment matrices can have large condition numbers (compete with machine precision.) ● Execution time: Single-CPU machines with unoptimized libraries can take number of hours to solve large-size problems. ● Currently using 64-bit //-computing: a huge improvement. ● Solving full pixel (2112 modules, 12.5k DoF) on 16 nodes takes 10mins compared with 7hrs on Intel P4 ● And solving full system was not even possible until last year! ● Work ongoing on other methods for further improvement: ● Various investigations on iterative methods ● Plan to implement one such method (MA27) in athena ● plan to port code on 64-bit to do processing+solving all at once.
27 MKU IPRD06 Nominal Detector Simulation: Global chi2 Full barrel i.e DoF’s. pdsyevd of ScaLAPACK on a 64bit ||-cluster AMD Opteron. Using 16 CPU grid N=21k system diagonalised within ~1h! Pulls of alignment corrections Pulls of all corrections of ~unit width and centred at zero Pulls in diagonal base =0.94
28 MKU IPRD06 CTB Setup
29 MKU IPRD06 SR1 Cosmics: Readout SCT inserted into the TRT on17/02/2006 Physics mode data coming in end of april until june SCT read 3 time bins (75nsec) and accept any hit in these three time bins TRT: read every 3.125nsec and 24 time bins (75 nsec).
30 MKU IPRD06 SR1 Cosmics : Hitmap From simulation
31 MKU IPRD06 Online alignment: FSI
32 MKU IPRD06 On-detector FSI System FSI grid nodes attached to inner surface of SCT carbon-fibre cylinder Distance measurements between grid nodes precise to <1 micron
33 MKU IPRD06 FSI Analysis Chain FSI Grid Lengths Reconstruction Module corrections Interpolation Grid Node DoF FSI interferometer data Length extraction ~400MB per scan (one scan upto every 10 minutes) Model configuration data (TBD) (192 kB per scan) ~<9.6kB eigenmode amplitudes +2MB laser configuration data -> either configuration db or some local file Most CPU intensive Nodes + errors 9.6kB meta data ~300 kB Lengths + errors: 12.6kB Model configuration data (TBD) only every Nth scan archived
34 MKU IPRD06 FSI + Track alignment How to include time dependency? 1.FSI provides low spatial frequency module corrections at time t i, t 0
35 MKU IPRD06 SR1 SCT+TRT photogrammetry SCT Barrel photogrammetry survey was completed early this year. SCT-TRT relative position survey also performed.
36 MKU IPRD06 Photogrammetry: Deformations Circles (colored) are fits, black curves are guidelines for ellipses using the scaled up differences of data points (col) to the circles B3B4 B5B6 Measurements performed before and after insertion into TRT. Detailed measurements exist only before the insertion O(20μm) in XY. After insertion only coordinate system transfer was measured. The individual cylinder interlink data showed deformations consistent with tilted ellipses. Face A and face C appear to be rotated in opposite directions, hinting at twists of the complete barrel. Deformations are order of 100 μm.