TF Overview current alpha release The end Backup o status update for the low momentum TF - F2F oct 12 Jakob Lettenbichler, Rudolf Fr¨ uhwirth Institute of High Energy Physics Austrian Academy of Sciences October 29, 2012 Jakob Lettenbichler, Rudolf Fr¨ uhwirth 1 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup The silicon vertex detector of Belle II Jakob Lettenbichler, Rudolf Fr¨ uhwirth 2 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Before we start... Definitions sector : subunit of a sensor friends : two compatible sectors combined segment : two compatible hits combined neighbours (NBs): two compatible segments combined efficiency: number of TCs recovered / number of tracks got by mcTrackfinder (mcTF records track, if at least 3 hits found within SVD or filters by momentum) track is recovered, when: at least 3 hits of track has been found again by any single TC more than 70% of hits are of assigned track clean TC has no foreign hits contaminated TC carries foreign hits Jakob Lettenbichler, Rudolf Fr¨ uhwirth 3 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Overlapping sensors for low momentum track finding aspects relevant for SVD-only tracking slanted sensors with overlapping regions are huge advantage compared to SVD2-design (next to fully usable 4 layers) overlapping regions are a bonus, but not essentially needed low momentum tracks need 3 hits for reconstruction, therefore: overlapping sensors strongly appreciated for layer 3 and 4 (one extra hit there gives opportunity to search for tracks not reaching layer 5&6) overlapping sensors are a bonus, but not essentially needed for layer 5 and 6 (here we have already at least 3 hits) Jakob Lettenbichler, Rudolf Fr¨ uhwirth 4 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Approach for reducing combinatorics Sector of interest Schematic view of the low momentum track finder in Belle II Unsorted hits from tracks, background, ghost coming from an event Friends (compatible Sectors) O L Sector setup - 1-hit filter filters by set of compatible sectors, allows momentum dependent setups distance O L Segment finder - 2-hit filter filters by distance, min&max, including virtual Segment - The arrows represent a O L Neighbour finder - 3-hit filter angle schematic interpretation of the filters by angle and Δ -distance min&max, pT possible number of combinations of hits at that point - Filters marked with an O use L Cellular Automaton external information generated by evolving states, includes TC-collector simulation - Steps marked with an L cycle O L Post 4-hit filter through several passes filters by zigZag, ΔpT, ... zigg- zagg Clean Kalman filter Hopfield Network TC's calculates QI's uses QI's to find best subset among overlapping TC's Jakob Lettenbichler, Rudolf Fr¨ uhwirth 5 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Alpha release changes since July: multipass without hit-removal to cover broad momentum ranges within one event Kalman filter finally implemented first SVD-only BG-check for pGun and evtGen-events finished cleaning of code external analysis tool works (not released) (tons of) bugfixes Jakob Lettenbichler, Rudolf Fr¨ uhwirth 6 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Test setup settings pGun: 1T/10T pions with 45MeV/ c ≤ p T ≤ 250MeV/ c with and without BGmixer (see Peters talk) TF: minimal filters (dist3D, angles3D, deltapT, ziggzagg) with and without KF+NN evtGen: with and without BG. Efficiencies calculated only for 50-200 MeV/ c p T all runs: SVD only 3 pass: low: 3 layers, 45MeV/ c ≤ p T ≤ 80MeV/ c , std: 4 layers, 75MeV/ c ≤ p T ≤ 175MeV/ c , high: 4 layers, 140MeV/ c ≤ p T ≤ 1100MeV/ c 16 ➦ ≤ θ ≤ 151 ➦ 6 sectors per layer - u-boundaries (normed to sensor size): 0, 0.5, 1. v-boundaries: 0, 0.33, 0.67, 1 Jakob Lettenbichler, Rudolf Fr¨ uhwirth 7 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Preliminary results - september 2012 3-pass-efficiency for single/10T pions depending on momentum range 100 90 80 70 efficiency in % 60 1T 50 10T w/o Hopfield 10T w Hopfield 40 30 20 10 40 60 100 250 500 1000 transverse momentum [MeV/c] Jakob Lettenbichler, Rudolf Fr¨ uhwirth 8 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Preliminary results - F2F oct 2012 3-pass-efficiency for single/10T pions depending on momentum range 100 90 80 70 efficiency in % 60 1T wNN 1T w/oNN 50 10T wNN 10T wNNwBG 40 10T w/oNN 30 20 10 40 6075 100 250 500 transverse momentum [MeV/c] Jakob Lettenbichler, Rudolf Fr¨ uhwirth 9 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Preliminary results - F2F oct 2012 closeup 3-pass-efficiency for single/10T pions depending on momentum range 100 95 90 efficiency in % 1T wNN 1T w/oNN 85 10T wNN 10T wNNwBG 10T w/oNN 80 75 70 40 50 60 75 100 250 transverse momentum [MeV/c] Jakob Lettenbichler, Rudolf Fr¨ uhwirth 10 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup evtGen ghost rate not easy to calculate because of limited momentum range considered by efficiency w/o BG: 84.8% within 50-200 MeV/ c p T with BG: 83.3% within same momentum range 10T with KF+NN: 400ms/event (some single events take several seconds, combinatorics in KF too) tested on a 4 yeaar old C2D with 3GHz, singlethreaded evtGen: w/o BG: 1100ms*, withBG: 1200ms* (*: tested on 7 year old C1D with 1,66GHz, therefore not comparable) Jakob Lettenbichler, Rudolf Fr¨ uhwirth 11 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Time Consumption 1T w/o KF+NN: 1ms/event 1T with KF+NN: 15ms/event (KF is slow) 10T w/o KF+NN: 15ms/event (combinatorics takes effect) 10T with KF+NN: 400ms/event (some single events take several seconds, combinatorics in KF too) tested on a 4 yeaar old C2D with 3GHz, singlethreaded evtGen: w/o BG: 1100ms*, withBG: 1200ms* (*: tested on 7 year old C1D with 1,66GHz, therefore not comparable) Jakob Lettenbichler, Rudolf Fr¨ uhwirth 12 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Outlook ToDo improve analysis capabilities (long term: auto-test with pretty plots) checking performance under realistic event- and background conditions (next goal: PXD) support curling tracks A.S.A.P (after BG-check) optimize sectorMaps for each pass (long term task) bugfixing developing online version (project starts 2013Q1) Jakob Lettenbichler, Rudolf Fr¨ uhwirth 13 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup that’s all, folks! Any suggestions, ideas or requests? Jakob.Lettenbichler@oeaw.ac.at Jakob Lettenbichler, Rudolf Fr¨ uhwirth 14 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Results of CHEP2012 Setting Low: p T = 60 – 70MeV/ c , 3 layers (w/o PXD) Low: p T = 60 – 70MeV/ c , 5 layers (with PXD) High: p T = 70 – 100MeV/ c , 4 layers (w/o PXD) High: p T = 70 – 100MeV/ c , 6 layers (with PXD) p T range # of layers # of tracks results post TCC clean cont. lost rec tot Low 3 10 000 88.9% 0.48% 10.7% 89.3% 20 000 88.2% 1.1% 10.8% 89.2% Low 5 10 000 99.6% 0.1% 0.3% 99.7% 20 000 99.1% 0.3% 0.6% 99.4% High 4 10 000 99.6% 0.1% 0.4% 99.6% 20 000 99.5% 0.1% 0.4% 99.6% High 6 10 000 99.6% 0.1% 0.3% 99.7% 20 000 99.4% 0.2% 0.4% 99.6% Jakob Lettenbichler, Rudolf Fr¨ uhwirth 15 HEPHY Wien & BELLE Collaboration
TF Overview current alpha release The end Backup Features some highlights input: PXD/SVD-Clusters, output: GFTrackCand (same as McTrackFinder, be careful when using both) recovery strongly depends on sector maps (XML-Files), which cover p T range of ∼ 50 − 1000MeV/ c Sector maps for low (50-75MeV/ c ), med(70-170MeV/ c ) and high (150-1000MeV/ c ) momenta available (SVD-only, PXD-next week) Jakob Lettenbichler, Rudolf Fr¨ uhwirth 16 HEPHY Wien & BELLE Collaboration
Recommend
More recommend
