development of the parallel TPC tracking Marian Ivanov

development of the parallel TPC tracking Marian Ivanov CERN

Changes in TPC tracking(1) n Preparation for the PARALEL combined tracking new functionality added n n forward and backward propagation Set. IO function to specify input and output for tracking – only place in the code related to IO n n algorithm independent of IO ESD input and output enabled in parallel possible to write “standard” output for tracking –Tree. T… …

Changes in TPC tracking(2) n Motivated by V 0 studies in TPC n Increase tracking efficiency for secondary particles n n track primary particles decaying deep inside of the TPC n n continuous seeding in TPC added improve momentum and position resolution for secondary particles n n new (combinatorial) seeding implemented eliminate systematic shifts due to the vertex constrain controversially - speed-up tracking code n n 2. 2 min for full event - -g option 1. 2 min –o 2 option

Changes in TPC tracking(3) n Ali. Helix implemented n special class for geometrical calculation n n track propagation DCA calculation current momenta calculation interfaced to Ali. Kalman. Track, TParticle and Ali. Track. Reference n easier to compare reconstruction with MC data

Changes in TPC tracking(4) n necessary to implement new TPC comparison n n correlation analysis with user defined cuts enabled many x many problem solved n n curling track are multiple reconstructed (properly or improperly) generated output – TTree with branches for track MC and reconstructed information

Changes in TPC tracking(5) n new classes implemented n Ali. TPCGen. Info n n Ali. TPCRec. Info n n contain relevant MC information for given track: TParticle, container with track references, digit information (map of padrows which were hitted by track +queries –first … last pad row, number hitted padrows…), mean Nprim (~d. Edx) Ali. TPCtrack + derived preprocessed information necessary for easier correlation study Ali. TPCV 0 Info n n contain Ali. TPCGen. Info for mother and daughter particle characteristic of the vertex

New seeding with vertex constrain n goals: n n don’t seed ‘evidently’ secondary particles reduce N 2 problem speed-up factor 10 for d. Ndy 8000 before loop over clusters in layer 2 geometrical transformation coefficient calculated n shift, rotation, shrink n n fast cuts implemented n n n vertex [0, 0, 0], X 1 [1, 0, z 1] z 2 coordinate of cluster 2 2 given by position of vertex not used point near intersection of “hypothetical” track with middle pad-row required additional cut after kalman tracking between layer 1 -2 n if track does not point to z vertex founded clusters are reused by fast Make. Seed without vertex constrain

New seeding without vertex constrain n old seeding n n n solution – combinatorial seeding to minimize correlation n fast but … low efficiency for strongly inclined tracks due to the angular effect correlating errors between neighboring pad-rows distance between seeding pad-rows small (tested with 7 padrows) hypothetical required cluster at the middle calculated using linear aproximation more efficient but slower than old seeding n used only after “fast” seeding with vertex constrain

New tracking strategy (2) n loop over different seeding region n seeding with vertex constrain tracking of seeds down to the innermost sector updating statistical information n n mean track quantities and their dispersions (number of accepted clusters, cluster density, chi 2) goal - to have unique cuts for different multiplicities sign clusters belonging to tracks with acceptable quality (n-sigma cut, with n as parameter) similar loop over different seeding region – seeding without vertex constrain

Efficiency (d. Ndy=2000) n n n left side – efficiency for tracking of primaries decayed in TPC at radius r right side– efficiency for tracking of secondaries created in TPC at radius r integral efficiency according old criteria (defined in Ali. TPCComparison. C) n 99. 9% for primaries

Kink and V 0 finding strategy n step 1: tracking n n n looking for all possible – even very short track candidates several seeding in different region of TPC necessary to find both mother and also daughter particles for step 2: combinatorial search for Kink and V 0 n fiducial volume – given by tracking efficiency, track parameter precision and track density n n minimal DCA n n n kink 120 -220 cm cut on n (currently 6) sigma N 2 problem causality cut n n n probability that primary track continue after DCA point and that secondary has prolongation even before DCA based on the track - cluster density before, respectively after DCA should be optimized for different track densities

Kink fiducial volume n n volume given by seeding and tracking efficiency for “short” track better seeding and tracking strategy

Kink vertex resolution(1) n better r resolution (0. 18 cm comparing to 0. 3 reported during last offline week)+ non systemetic effects

Kink vertex resolution (2) n OK, but: n n n improvement because we stop tracks with high chi 2 and non acceptable space resolution to don’t take clusters from other tracks also non secondary tracks can be stopped not sufficient information about the track overlaps n n worse d. Edx resolution for high multiplicity event after kink and finding – the tracks have to be post processed n kink and V 0 finding in the TPC volume has to be performed during TPC tracking

New kink finder - strategy n N 2 problem with combinatorial search n n Linear loop: n n very fast cut necessary Ali. Helix defined during linear track preprocessing N 2 loop: n n n n fast analytical calculation of track intersection or DCA in rφ projection rough cut on nearest point radii in rφ projection analytical calculation of DCA in two or one local minima from rφ direction – calculated in 3 dimension stronger cut applied on R and distance DCA calculation using hessian approximation final cuts on DCA Kink properties calculation

Ali. Helix n n N 2 problem with combinatorial search of V 0 and kink finder Ali. Helix n n definition during sequential loop track preprocessing – or reading used for all DCA geometrical calculation data layout optimized for fast computation of DCA global coordinate system used – no transformation - rotation needed during time critical combinatorial search

DCA calculation in rφ projection x 1, y 1, t 1, p 1 x 2, y 2, t 2, p 2 n n n three considered situation x, y – global position of the DCA in rφ ti, pi – time - phase of the helix in DCA

Linear versus Hessian DCA calculation n n started directly from the two local minima linear DCA approximation faster the resolution on the level of slower Hessian calculation – (three iteration used) both are implemented in Ali. Helix

Parallel incremental tracking – Ali. Barrel. Tracker n n Tracking using information from different detectors Requirements n n n as fast as possible as efficient as possible as “good” as possible as modular as possible all other criteria (backward compatibility, dependency problems) lower priority – taken only as technical complication in TPC tracker – already implemented some of the basic functionality

Conclusion n Ali. TPCtracker strongly updated n n n Ali. Helix – stand alone class n n ready to commit now new comparison n n cvsa diff Ali. TPCtracker. MI 4000 lines improvement in efficiency and pt resolution for secondary particles speedup of the code (seeding, error parametrization, faster navigation through the clusters using look-up table, …) because of reported problems with d. Edx, commit planned only after implementation of V 0 finder during TPC tracking planned commit after conversion to the new IO V 0 finder – to be committed together with TPC tracker as integral part