2011 - Sti, CA and Stv tracking component review

This page will keep information related to the 2011 tracking component review. The review will cover the state of the Cellular Automaton (CA) seed finding component as well as the Virtual Monte-Carlo based tracker (Stv) and their relevance to STAR's future need in terms of tracking capabilities.

Project goals

After a successful review of the ITTF/Sti tracker in 2004, the STAR collaboration have approved the move to the new framework bringing at the time unprecedented new capabilities to the experiment and physics analysis.Sti allowed the STAR reconstruction approach to integrate to its tracking other detector sub-systems by providing method to integrate simple geometry models and allow to extrapolate track to the non-TPC detector planes therefore, correlating information across detector sub-systems. In 2005, the STAR production switched to a Sti based production and we have run in this mode ever since.

However, careful architecture considerations revealed a few areas where improvements seemed needed. Those are:

  • The need to maintain two different geometry models (one for reconstruction, one for simulation) increasing workforce load at a time when STAR is both active and ambitious in its future program as well as running thin on detector sub-system code developer. Beyond workforce considerations
    • The two separate geometries have consequences on embedding and simulation hence, our ability to bring efficiency corrections to the next level of accuracy.
    • Material budgets were found to be ill-accounted in reconstruction (dead-material were not properly modeled in the Sti framework). The use of a common geometry model would have removed this issue
  • Sti has some tracking restrictions - geometries made of planes and volumes perpendicular to the beam cannot be treated due to a technical choice (detector elements are organized in planes // to the beam, sub-systems assumed to be composed of elements replicated in Phi). This would preclude tracking in detectors such as the FGT.
    • Our goal was to create an extended set of functionalities providing a truly complete integrated tracking approach, allowing the inclusion of hit information from other detectors (a key goal the inclusion of detector hits placed in the forward direction)
  • The use Monte-Carlo based propagators would allow better access to Eloss, better predictors and track swimming allowing for tracking in non constant B field (this is also not possible in Sti)

Additional considerations for the future of STAR were

  • A single yet flexible geometry model would allow STAR to be ready for GeantX (5 and TGeo based)
  • A flexible geometry model would allow STAR to better cope with STAR to eSTAR migration (geometry morphing)
  • A revitalize framework would allow addressing long standing issues of event mode in simulation
    • While STAR has a FORtran based approach allowing integration of some event generators, many have appeared pure C++ based, making their integration to the STAR simulation framework difficult. A new generic model would allow a "plug-and-play" approach.
    • The use of non-perfect geometries (miss-aligned) have been lacking in the simulation framework and would be advisable
  • Novel algorithm have appeared in the community, leveraging and harvesting the power of multi-core and many-core architectures. Investigating speed and efficient gains and evaluate the best use of resources is necessary for STAR demanding physics program. Equally important, those new algorithm (Cellular Automaton being one) are opening to online tracking algorithm (GPU based).

 

Based on those considerations, several projects were launch and encouraged

  • CA based tracking - the sudy of the CBM/Alice Cellular Automaton algorithm for seed finding was launched in collaboration with our GSI colleagues.  Multi-core aware, the simple algorithm is thought to provide speed gains over the seed finding. Further work could spurse from this evaluation (online HLT) if successful. The algorithm was showed to be portable to STAR, thanks to Yuri Fisyak and Ivan Kisel team, and the product of this evaluation to be tested.
  • The VMC project - a three part project (VMC tracking, VMC geometry, VMC simulation framework), the VMC geometry (a.k.a. aGML) has rapidly matured under the care of Jason Webb. The VMC trakcing (a.k.a. Stv) has been developed by Victor Perevoztchikov and thought to provide equal efficiency than Sti (as well as implement all the features listed above).

We propose to review the aGML, CA and Stv components of our framework reshape.

 

NB: Beyond the scope of this review, a key goal for VMC is to allow the inclusion of newer Geant version and hence, getting ready to step away from Geant3 (barely maintainable), the FORtran baggage (Zebra and portability issues on 64 bits architectures) and remove the need for a special verison of root (root4star) hard-binding root and STAR specific needed runtime non-dynamic libraries.

 

Why a review?

  • All R&D projects are reviewed in STAR
    • Initial approach was to proceed with a "soft" PWG evaluation but (on second thoughts) not really an options …
    • An internal STAR review process should (and will) be established
  • Advantages
    • A review process provides strong and independent backing  of the projects
    • A review process provides  an independent set of guidance to management (S&C and PWG) on path forward
    • Collaboration wide scrutiny and endorsement across PWG lessen the risks of  finding problems later
  • Reminder: ITTF / Sti was not carried without problems
    • Sti review missed  the UPC PWG’s feedback –problems found a-posteriori diverted attention and workforce in solving it
    • Problem are seen in HBT and fluctuation analysis when Run 4 is compared to Run 10
      • HBT issues were not seen at Sti evaluation – Is it an analysis problem? Something else?
  • A review will also provide a good time to re-establish a solid baseline and get feedback from the PWG on opened issues if any
    • This is even more so important that STAR is moving forward to a new set of detectors and high-precision measurements

 

 

Review charges

See attachment at the bottom of this page.

 

Review committee

Status:

  • 2011/08/12 Intent of a review brought to management (charges to be written).
                        Action items was to suggest a set of names for the committee set.
  • 2011/08/18 Committee members suggestions provided at management meeting. Spokesperson decides he will contact chair.
  • 2011/09/02 Charges sent to management for comments along a note that the charges may be long (text is both for committee and reviewee). No feedback outside the provided self-provided note.
  • 2011/10/07 Chair contacted - process of selecting committee being worked out (Spokesperson or)
  • 2011/10/13 Spokesperson delegate committee forming to review Chair (Olga Evdokimov), S&C Leader (Jerome Lauret) and PAC (Xin Dong)
  • 2011/10/15 Committee assembled
  • 2011/10/31 Draft agenda made
  • 2011/11/01 Agenda presented and feedback requested
  • 2011/11/08 Final agenda crystalized

Members:

  • Olga Evdokimov (chair)
  • Claude Pruneau
  • Jim Thomas
  • Renee Fatemi                [EVO]
  • Aihong Tang
  • Thomas Ullrich              [EVO]
  • Jan Balewski                 [EVO]
  • Anselm Vossen

The agenda is ready and available at You do not have access to view this node.

Material

Below is a list of cross-references to other documents:

Follow-up of Stv performances

This page will list material for the follow-up review of Stv performances.

Material

Other information

Review findings and key points/problems

The review recommendations for focii were

  • Hit residuals and track chi2 should be the first problems to be addressed
  • Tracking inefficiencies, pT and charge dependences for reconstructed tracks, and charge separation for high pT tracks need to be reevaluated after Stv tuning and optimization.
  • Implementation of the track extensions to other detector volumes is necessary for the Stv to become the truly integrated STAR tracker.
  • Extension to the TOF detector is recommended to be next on the priority list

 

The review finding details included

  • No documentation is available on the method (in general), and its specific implementation - this needs to be addressed
    • Appropriate documentation needs to be created for the track predictors to other detector elements, with detailed workflow of the algorithm to be worked out with the relevant subsystem groups
  • CPU resources: Stv uses twice the CPU resources - recommends addressing this matter before a full deployment is attempted (and along development)
     
  • The ability of the Stv to properly account for the material budget is unclear at the moment, as performance tests uncovered significant losses in the dead material
  • The importance of optimizing PPV VF within Stv (work toward optimization should be pursued by the developers from both, S&C and VF teams)
  • The implementation of the track extension method into volumes other than the TPC was not reviewed by this committee. The validity of an extrapolated covariance matrix was not discussed. Integration of all sub-systems, for example FGT and TOF, into the tracking model should become a major focus of the project
  • Integration of forward detector geometries has not been showed
  • The possibility to have track fits using different mass assumptions has not been presented
     
  • Performance were not all understood:
    • The pT dependence of the tracking reconstruction efficiency is not entirely understood (degraded resolution of charge sign reconstruction)
    • The ψ resolutions and pulls were found to be slightly worse for Stv compared to Sti.
    • Track by track comparison of the Stv and Sti trackers on the real data have shown similar momentum resolution and pulls for inclusive tracks: Systematic and opposite shifts for positive and negative tracks, were observed between tracks found by the two trackers. The discrepancies are shown to grow with the transverse momenta
  • Insufficient supporting evidence were presented with regard to Stv’s ability to handle low multiplicity high pile-up events - this issue be studied in details and presented for the next review before Stv deployment
  • Physics evaluation specifics
    • Stv and StvCA have been reported to have lower efficiencies and unusual phi distributions by the UPC
    • The efficiencies of Stv and StvCA look significantly lower than those from Sti from Jetcorr correlation studies
    • The widths of Lambda and AntiLambda become broader in Stv and StvCA data samples, compared with Sti samples
    • Mass shifts were observed between the Stv and Sti tracker family
    • Spin group reported major losses in jet reconstruction efficiency (factor of 2) due to the loss of low pT tracks in Stv
    • The charge sign discrimination at high pT deteriorates significantly while switching from Sti to Stv.
    • Jetcorr tests uncovered Stv problems with saving of global tracks to MuDst format; also they report incorrect results for pT and yT correlations in Stv

Specific example of the DCA distribution problem (reminded 4/13/2012)