STAR StiVmc V. Perevoztchikov Brookhaven National Laboratory,USA.

Slides:



Advertisements
Similar presentations
Transformations We want to be able to make changes to the image larger/smaller rotate move This can be efficiently achieved through mathematical operations.
Advertisements

Combined tracking based on MIP. Proposal Marian Ivanov.
N. Saoulidou Fermilab 1 Update on track reconstruction in the Near Detector N. Saoulidou, Fermilab
Pattern Recognition in OPERA Tracking A.Chukanov, S.Dmitrievsky, Yu.Gornushkin OPERA collaboration meeting, Ankara, Turkey, 1-4 of April 2009 JINR, Dubna.
31 March 2004 Geneva University MICE TPG Reconstruction Cluster finding Space point reconstruction Track recognition Track fit GENEVA UNIVERSITY GENEVA.
Bill Atwood, Core Meeting, 9-Oct GLAS T 1 Finding and Fitting A Recast of Traditional GLAST Finding: Combo A Recast of the Kalman Filter Setting.
STAR Collaboration Meeting, Nantes, July2002 SVT Analysis/Status Update Jun Takahashi – University of Sao Paulo.
The LiC Detector Toy M. Valentan, M. Regler, R. Frühwirth Austrian Academy of Sciences Institute of High Energy Physics, Vienna InputSimulation ReconstructionOutput.
STAR C OMPUTING Maker and I/O Model in STAR Victor Perevoztchikov.
9/26/11HFT soft meeting, BNL1 Chain analysis fz file MuDst.root minimc.root geant.root event.root McEvent.root StMiniMcMaker StAssociationMaker : STAR.
1 Tracking Reconstruction Norman A. Graf SLAC July 19, 2006.
Tracking at LHCb Introduction: Tracking Performance at LHCb Kalman Filter Technique Speed Optimization Status & Plans.
Track Reconstruction: the trf & ftf toolkits Norman Graf (SLAC) ILD Software Meeting, DESY July 6, 2010.
STS track recognition by 3D track-following method Gennady Ososkov, A.Airiyan, A.Lebedev, S.Lebedev, E.Litvinenko Laboratory of Information Technologies.
Event Data History David Adams BNL Atlas Software Week December 2001.
Material budget, energy losses and multiple scattering.
Optimising Cuts for HLT George Talbot Supervisor: Stewart Martin-Haugh.
LAV Software Status Emanuele Leonardi – Tommaso Spadaro Photon Veto WG meeting – 2015/03/24.
STAR Sti, main features V. Perevoztchikov Brookhaven National Laboratory,USA.
Some Thoughts about Hits, Geometry etc Rob Kutschke, Hans Wenzel Fermilab March 13, 2007.
Status of Pattern Recognition for the T-Tracker Hans Wenzel, Hogan Nguyen March 12 th, 2011 Introduction Hans implemented stereo hits, formed by the intersection.
Progress report on Muon Reconstruction based on Kalman filter Y. Fisyak, BNL.
Y.Fisyak, BNL - STAR Upgrade workshop, 12/2/ Integrated Tracker – STAR tracking framework of the future update on  status and  perspective IT(TF)
1 Request for MuDst revision Y.Fisyak, 03/05/08 S&C meeting.
Track extrapolation to TOF with Kalman filter F. Pierella for the TOF-Offline Group INFN & Bologna University PPR Meeting, January 2003.
STAR STAR VMC tracker V. Perevoztchikov Brookhaven National Laboratory,USA.
STAR Kalman Track Fit V. Perevoztchikov Brookhaven National Laboratory,USA.
What is in my contribution area Nick Sinev, University of Oregon.
Pattern Recognition in OPERA Tracking A.Chukanov, S.Dmitrievsky, Yu.Gornushkin OPERA collaboration meeting, Mizunami, Japan, of January 2009 JINR,
G. Sirri - Bologna videoconf /9 Lateral X-rays with ESS  Goal: implementation of the lateral mark finding in the ESS software  test version.
LCWS 06 Bangalore, India, March Track fitting using weight matrix Nick Sinev, University of Oregon.
STAR Event data storage and management in STAR V. Perevoztchikov Brookhaven National Laboratory,USA.
Pandora calorimetry and leakage correction Peter Speckmayer 2010/09/011Peter Speckmayer, WG2 meeting.
ALICE Offline Week, CERN, Andrea Dainese 1 Primary vertex with TPC-only tracks Andrea Dainese INFN Legnaro Motivation: TPC stand-alone analyses.
ROBUSTIFICATION of the Belle Vertex Fitter April 14 th 2003Johannes Rindhauser Hephy Vienna Belle Weekly Meeting (AdaptiveVtxFitter)
Tracking in High Density Environment
V0 analytical selection Marian Ivanov, Alexander Kalweit.
Integrated Tracker (progress, status, plans) Y. Fisyak.
LM Feb SSD status and Plans for Year 5 Lilian Martin - SUBATECH STAR Collaboration Meeting BNL - February 2005.
1 A first look at the KEK tracker data with G4MICE Malcolm Ellis 2 nd December 2005.
Secondary Vertex reconstruction for the D + Elena Bruna University of Torino ALICE Physics Week Erice, Dec. 6 th 2005.
05/04/06Predrag Krstonosic - Cambridge True particle flow and performance of recent particle flow algorithms.
Implementation Highlights Mike Miller Yale University.
Fast Tracking of Strip and MAPS Detectors Joachim Gläß Computer Engineering, University of Mannheim Target application is trigger  1. do it fast  2.
HLT Kalman Filter Implementation of a Kalman Filter in the ALICE High Level Trigger. Thomas Vik, UiO.
STAR Analysis Meeting, BNL – oct 2002 Alexandre A. P. Suaide Wayne State University Slide 1 EMC update Status of EMC analysis –Calibration –Transverse.
STAR StiVmc V. Perevoztchikov Brookhaven National Laboratory,USA.
Adam Blake, June 9 th Results Quick Review Look at Some Data In Depth Look at One Anomalous Event Conclusion.
Development of the parallel TPC tracking Marian Ivanov CERN.
D0 analysis in HFT era with (  Vertex) code LBL May 12, 2010 BNL: Y. Fisyak, V. Perevoztchikov [A. Kisel (guest)] Kent : J. Bouchet, J. Joseph, S. Margetis,
Info Read SEGY Wavelet estimation New Project Correlate near offset far offset Display Well Tie Elog Strata Geoview Hampson-Russell References Create New.
STAR SVT Self Alignment V. Perevoztchikov Brookhaven National Laboratory,USA.
STAR Simulation. Status and plans V. Perevoztchikov Brookhaven National Laboratory,USA.
AliRoot survey: Reconstruction P.Hristov 11/06/2013.
LHCb Alignment Strategy 26 th September 2007 S. Viret 1. Introduction 2. The alignment challenge 3. Conclusions.
Tracking software of the BESIII drift chamber Linghui WU For the BESIII MDC software group.
Geant4 Simulation for KM3 Georgios Stavropoulos NESTOR Institute WP2 meeting, Paris December 2008.
CHEP ’06 GEANT4E 1 GEANT4E: Error propagation for track reconstruction inside the GEANT4 framework Pedro Arce (CIEMAT) CHEP 2006, Mumbai, 13-17th February.
Track Reconstruction in MUCH and TRD Andrey Lebedev 1,2 Gennady Ososkov 2 1 Gesellschaft für Schwerionenforschung, Darmstadt, Germany 2 Laboratory of Information.
Track Reconstruction: the ftf and trf toolkits Norman Graf (SLAC) Common Software Working Meeting CERN, January 31, 2013.
IPHC, Strasbourg / GSI, Darmstadt
New TRD (&TOF) tracking algorithm
C.Cheshkov 15/09/2005 Weekly Offline Meeting
M. Kuhn, P. Hopchev, M. Ferro-Luzzi
HARPO Analysis.
Where did we stop? The Bayes decision rule guarantees an optimal classification… … But it requires the knowledge of P(ci|x) (or p(x|ci) and P(ci)) We.
Feature space tansformation methods
LHCb Alignment Strategy
Presentation transcript:

STAR StiVmc V. Perevoztchikov Brookhaven National Laboratory,USA

STAR Victor Perevoztchikov, BNL StarSoft local StiTGeo helper StiVmc is based on TGeo representation of geometry. But there is no direct dependency between them. All the communication goes thru StiTGeo helper class. This class contains: Simple array of pointers to StVoluInfo objects. TGeoVolume::GetNumber() used as an index. These objects keep specific Sti information, related to each volume. Contains global StiHit container, only for hits used in SeedFinder. Map container of StHitPlane(s) with HardwarePosition of StHit as a key

STAR Victor Perevoztchikov, BNL StarSoft local For each TGeoVolume could be related one StVoluInfo object which contains: Some flags, like: It is a module, like TPCE,SVTT; It is a Module with Hits; Active module; It is a HitPlane : StVoluInfo class

STAR Victor Perevoztchikov, BNL StarSoft local StHitPlaneInfo inherited from StVoluInfo class and used instead of it for sensitive volumes : Center of Hit plane in mother system; Direction of Hit plane in mother system; Hit container, simultaneously sorted by two variables, like Phi/Z Phi/Rxy etc…; Pointer to functor, which provides activity of sensitive sub volume. Map container of StHitPlane objects. These ones related to each TGeoVolume instantiation. It is sorted by TGeo path of founded object. : StHitPlaneInfo class

STAR Victor Perevoztchikov, BNL StarSoft local This class represents concrete sensitive detector. It’s functionality close to StiDetector. Contains: Center of Hit plane in global system; Direction of Hit plane in global system; Hit container, simultaneously sorted by two variables, like Phi/Z Phi/Rxy etc…; Pointer to functor, which provides activity of sensitive sub volume. Container of StHitPlane and StHitTube classes. These classes : StiHitPlane class

STAR Victor Perevoztchikov, BNL StarSoft local Loading of hits is made maximally independent of hit nature and origin. To add into StiVmc new detector there is no need to modify even one line of code. No more StiSsd, StiSvt, StiTpc, StiRnD and StiXXX classes. StiHitLoader does the following:  Iterate via StEventHitIter iterator;  Adds it to StTGeoHelper object;  StTGeoHelper tries to find according StHitPlane in StHitPlane map container by StHit hardware position.  If not found, search volume it in TGeoManager by x,y,z. l If not found, hit ignored; l Get volume id and get according StHitPlaneInfo; l Get StHitPlane by path and save it into StHitPlane map.  Save hit into StHitPlane object. : StiHitLoader

STAR Victor Perevoztchikov, BNL StarSoft local If geometry is not aligned, some hits could not find their volume. How many? StiMaker:INFO - StTGeoHelper::AddHit : Hit in /HALL_1/CAVE_1/TPCE_1/TPGV_1/TPSS_3 Not Found StiMaker:INFO - StTGeoHelper::AddHit : Hit in /HALL_1/CAVE_1/TPCE_1/TPGV_1/TPSS_9 Not Found StiMaker:INFO - StiHitLoader::LoadHits : Loaded good and failed 25 Tpc hits StiMaker:INFO - StiHitLoader::LoadHits : Loaded good and failed 25 of all hits So in case of pp2009a 0.2% of hits lost. So not aligned TPC geometry does not affect reconstruction. : StiHitLoader(2)

STAR Victor Perevoztchikov, BNL StarSoft local StiSeedFinder in StiVmc used similar to Sti algorithm. The difference only different hit container. In StiVmc used 3 keys map container. In the following pictures:  Seeds + unused TPC outer hits;  Only unused TPC outer hits; You can see that new StiSeedFinder is rather good. On the picture with only unused hits you can not notice the real tracks. : StiSeedFinder

STAR Victor Perevoztchikov, BNL StarSoft local : Seeds + all hits

STAR Victor Perevoztchikov, BNL StarSoft local. : Unused hits

STAR Victor Perevoztchikov, BNL StarSoft local Hit Errors D w

STAR Track direction T. Victor Perevoztchikov, BNL StarSoft local Hit error parameterization

STAR Victor Perevoztchikov, BNL StarSoft local Hit error + diffusion Electron cloud around the track increased due to diffusion. Diffusion along Z and in X,Y directions is different and additional dispersion could be written as: Resulting parameterization is:

STAR Comparing this with the previous it is unfortunately evident that old Sti error parameterization is completely wrong. All our production was made with wrong errors. Fortunately it does not mean that results are wrong, but accuracy is not such good as it could be. It could affect only such very precise physics as D0. Victor Perevoztchikov, BNL StarSoft local Old Sti Hit errors

STAR It will be shown later that the best plane for a fit is DCA plane. DCA plane is a plane orthogonal track direction and includes hit point and DCA point to this hit. Projecting hit errors on detector plane into DCA plane along the track direction gives errors: You see, these formulas are much simpler than in detector plane and there is no any bad divisions by zero, like in previous case. So DCA plane is much better for fit. All these formulas were tested by Monte-Carlo, which gave exactly same result. Victor Perevoztchikov, BNL StarSoft local DCA plane hit errors

STAR StiNavigate class provides navigation thru geometry volumes. In current version it is based on VMC navigation but in future probably directly on Tgeo. StiNavigate: Initialized by current space point, direction and momentum; Navigates thru volumes. On each volume: Updates matrix of derivatives, using helix approximation; Calculates energy loss and multiple scattering; If volume is sensitive and active, navigation is stopped; Then get from StiHitPlane double sorted StiHit container, StiHitPlane direction, passed length, accumulated material info, position, track parameters and errors, volume path, derivative matrix. Fitting started. Victor Perevoztchikov, BNL StarSoft local Navigation

STAR Victor Perevoztchikov, BNL StarSoft local Fitting Let consider simplified, 2d case. Hitting plane along Y axis, track crossed Y axis with angle α wrt X axis. So :  global track parameter α projected into local Y and proportional to sin(α).  when α << 1, then sin(α ) = α – α**3/6, second term is very small and projection from global to local system is linear.  Projections of error matrix is also linear. In local frame fit is linear. Back transformation of fitted parameters and errors is linear too. When α is small then curvature ρ is also projected into local Y linearly. So life is good. But when α >1, life is bad. Linearization is wrong, linear fit is wrong, Backward transformation into global is also wrong. All the times, when I saw unstable fit in Sti, it was α >1

STAR Victor Perevoztchikov, BNL StarSoft local Fit continue Could we do something with bad fit with α > 1? Yes, we can! Why we fit in local frame? There are only two reasons:  We know that track is crossing our hit plane;  We know hit errors on this plane; Let invent another local frame, where linearization is working always. The evident candidate is DCA frame. This frame is: Plane perpendicular to the track and crossing the hit point. Look the following picture.

STAR Victor Perevoztchikov, BNL StarSoft local DCA Frame α α δdδd D δDδD d β

STAR Victor Perevoztchikov, BNL StarSoft local Picture caption  Vertical bold line: Hit Plane;  Green star: Hit on hit plane;  Red star: DCA point of track wrt hit (green star). Arrow is a predicted track with the parameters in DCA point.  Blue star is a new position of track with the modified parameters in this point;  α is a crossing angle of the predicted track;  β is an angle between predicted track and fitted o α ne;  α+β is a crossing angle of the fitted track;  D is the distance between hit and crossing point of the predicted track;  D+δD is the distance between hit and crossing point of the fitted track;

STAR Victor Perevoztchikov, BNL StarSoft local Picture caption continue  d is the distance between hit and DCA point of the predicted track;  d+δd is the distance between hit and fitted track point in DCA plane;

STAR Victor Perevoztchikov, BNL StarSoft local Fitting in DCA frame In DCA frame, linearization with β and d is correct, track is perpendicular to DCA plane. In hit plane we minimizing the: ((D+δD) / σ )**2 where σ is a hit error. Let express it in DCA parameters. Looking on the previous picture and using sine rule: (D+δD) / cos(β) = (d+δd) / cos(α+β) In linearization β is small and cos(β) = 1 Then: ((D+δD) / σ )**2 == ((d+δd) / ( σ * cos(α+β) )**2

STAR Victor Perevoztchikov, BNL StarSoft local Fitting in DCA frame(2) ((D+δD) / σ )**2 == ((d+δd) / ( σ * cos(α+β) ))**2 So now we can minimize: ((d+δd) / ( σ * cos(α+β) ))**2 It still does not look like much better. Still cos(α+β) when α is not small could not be linearized. But now is a time to remember that for TPC and for all other detectors: σ = σ0 / cos(α+β) So cos(α+β) is canceled out. We minimize: ((d+δd) / σ0 )**2 without any loss of accuracy.

STAR Victor Perevoztchikov, BNL StarSoft local Local or DCA frame? From the discussion above it is clear that fit in DCA frame always is better than in local one. In addition, fit in DCA frame is irrelevant to hit plane orientation. It is proper for TPC, FTPC and all other detectors. But amount of computation in DCA is slightly bigger. I propose to use strategy:  TPC like detector: |α| < 0.5 use local frame fit. It is about 90% of all tracks;  TPC like detector: |α| > 0.5 use DCA frame fit.  Not a TPC like detector: use DCA frame fit.

STAR Victor Perevoztchikov, BNL StarSoft local The role of alignment We see, that for fit, we need only:  To know the space position of the hits;  Hit errors. These errors are calculated from hit plane orientation and predicted track parameters; Hit plane position is not used at all. Hit errors are proportional to 1/cos(α). So for small α to know precise orientation of hit plane is not too important. Some inaccuracy arise only for |α| > 1. But even here it is not dangerous. Even completely wrong errors do not lead to the shifted estimate. Only estimation error are increased in this case. So some inaccuracy in the hit errors is not too critical. Precise alignment of hit planes is not important for fit.

STAR Victor Perevoztchikov, BNL StarSoft local The role of alignment(2) The correct geometry description is important for accounting energy loss and multiple scattering. We see it Sti, which does not work properly for UPC, due to wrong Sti geometry. In the same time alignment of dead material volumes also not important. The correct, “ideal” geometry is absolutely enough for the material accounting. Precise Geometry alignment is not important for reconstruction.

STAR Victor Perevoztchikov, BNL StarSoft local What is already done(Old)  StEvent StHit iterator;  TGeoVolume iterator;  StiTGeo helper class;  Multi key (2 & 3) sorted container. 3 key global hit container used in seed finder. 2 key hit container in each hit plane.  SeedFinder based on 3 key sorted container.  Helix derivatives matrix  StiNavigate is ready.  Multiple scattering and energy loss accounting implemented and working also for geometry comparison.

STAR Victor Perevoztchikov, BNL StarSoft local What is already done(Now)  StiHitLoader, no dependency of detectors ;  StiDefault Seed Finder. Now tuned for TPC but only tuned;  StiGeometryLoader, with no detector dependency;  Propagation thru TGeo using TGeant3 + derivatives and errors  HitErrorCalculator with two modes: l Standard detector plane errors l DCA plane errors

STAR Victor Perevoztchikov, BNL StarSoft local Loading of hits is made maximally independent of hit nature and origin. To add into StiVmc new detector there is no need to modify even one line of code. No more StiSsd, StiSvt, StiTpc, StiRnD and StiXXX classes. StiHitLoader does the following:  Iterate via StEventHitIter iterator;  Adds it to StTGeoHelper object;  StTGeoHelper searches according StHitPlane in StHitPlane map container by StHit hardware position.  If not found, search volume it in TGeoManager by x,y,z. l If not found, hit ignored; l Get volume id and get according StHitPlaneInfo; l Get StHitPlane by path and save it into StHitPlane map. : StiHitLoader

STAR Victor Perevoztchikov, BNL StarSoft local : Seeds + all hits

STAR Victor Perevoztchikov, BNL StarSoft local. : Unused hits

STAR Victor Perevoztchikov, BNL StarSoft local : Geant track propagation

STAR Victor Perevoztchikov, BNL StarSoft local What is needed to be done  StiTrackFinder (copy & paste from old Sti);  StiTrackFitter (copy & paste from old Sti);  StiRefit (copy & paste from old Sti);  Total reshape. When: with free hands, beginning of June. But free hands is impossible, so +1 month. :

STAR Victor Perevoztchikov, BNL StarSoft local : Fitted tracks propagation

STAR Victor Perevoztchikov, BNL StarSoft local Status Jul 2010 To avoid clashes with Sti new Sti is renamed to Stv. Now alpha version of Stv is ready. What is included?  Hit loader;  Seed Finder;  Track finder, based on Kalman fitter/filter;  Vertex finder (not PPV);  StEvent filler;  Pull filler What is not:  No backward track finder;  No smoother or refit;  No primary track fitter;

STAR Victor Perevoztchikov, BNL StarSoft local Comparison and testing Right now real comparison is not possible yet. It is related to new hit error model. Before testing, we must obtain hit error parameters. For Sti StiFiterr application is used. There is no such application for Stv yet. During this week I hope the new StvFitErr will be ready. Then the end of Jul and Aug I will perform the testing and comparison. In parallel all the “NO” points above will be implemented. If above will be successful, in September, the massive testing will be performed

STAR Victor Perevoztchikov, BNL StarSoft local Local Y pull distribution

STAR Victor Perevoztchikov, BNL StarSoft local lZPull

STAR Victor Perevoztchikov, BNL StarSoft local lYHit

STAR Victor Perevoztchikov, BNL StarSoft local lZHit

STAR Victor Perevoztchikov, BNL StarSoft local Status Nov 2010 Geometry loader:  Loads TGeo geometry. Automatically recognize sensitive volumes, their orientations, positions. Creates structures to support hits Hit Loader:  Loads hits. Automatically recognition to which volume each hit belongs. No need for RND developer to create specific software for each version of detector. Geant geometry is enough. Seed finder & Track reconstruction is ready. Track smoother or refit is ready; Primary track assignment and fitting in process. Right now is mainly tuning and testing

STAR Victor Perevoztchikov, BNL StarSoft local : Fitted tracks propagation

STAR Victor Perevoztchikov, BNL StarSoft local. : Unused hits

STAR Victor Perevoztchikov, BNL StarSoft local Status Feb 2011 Alpha version is finished. Now testing. For test was selected file st_physics_ _raw_ daq First events was used. Sti has 260 good globals per event. Stv % more Sti has 10 good primary per event. Stv % more Sti has 4% primaries per global. Stv 5% Sti 28 hits per global. Stv 21. Stv 25% less Sti 32 hits per Primary. Stv 22. Stv 32% less (???) Sti Dca00, DcaXY & DcaZ are much worse then Stv ones(???)

STAR Victor Perevoztchikov, BNL StarSoft local Sti/Stv: good globals & primaries

STAR Victor Perevoztchikov, BNL StarSoft local Sti/Stv: Hits globals & primaries.

STAR Victor Perevoztchikov, BNL StarSoft local Sti/Stv: Dca globals of primaries.

STAR Victor Perevoztchikov, BNL StarSoft local Status Feb Discussion Stv has more globals but less hits per global. Why? Let us count total amount of global hits. Sti = 260*28 = 7280 Stv= 349*22 = It looks like Sti steals hits from other tracks. Then these robbed tracks became bad (<15 hits) Now the same for primaries: Sti=10*32 = 320 Stv=17*22 = 374. The previous guess works here as well Now why Dca’s for Sti are such bad? The same explanation fits. Wrong hits stolen by Sti deteriorate Dca’s Of course, it is only guess.