Implemention of ACTS into sPHENIX Track Reconstruction Joe Osborn May 17, 2021 Next Generation of QCD at RHIC sPHENIX The Relativistic Heavy Ion Collider (RHIC) at Brookhaven National Laboratory Joe Osborn 2 Collaboration Timeline DOE “Mission need” PD-2/3, CD-0 construction begins You are here Start physics Collaboration DOE CD-1/3A: cost, Installation and formed schedule, advance commissioning purchase approval Joe Osborn 3 sPHENIX • Study QCD matter at varying temperatures for direct comparisons to LHC • Study partonic structure of protons and nuclei Joe Osborn 4 • Primary tracking detectors: • Micro vertexing (MVTX) - 3 layers of MAPS staves • Intermediate silicon tracker (INTT) - 2 layers of silicon strips (fast integration time) • Compact GEM-based TPC - continuous readout sPHENIX Detector • sPHENIX detector designed for high precision tracking and jet measurements at RHIC • Large, hermetic acceptance • Hadronic calorimetery (first at RHIC) • Large offline data rate of ∼100 Gbit/s for collecting large minimum bias sample Joe Osborn 5 sPHENIX Detector • sPHENIX detector designed for high precision tracking and jet measurements at RHIC • Large, hermetic acceptance • Hadronic calorimetery (first at RHIC) • Large offline data rate of ∼100 Gbit/s for collecting large minimum bias sample • Primary tracking detectors: • Micro vertexing (MVTX) - 3 layers of MAPS staves • Intermediate silicon tracker (INTT) - 2 layers of silicon strips (fast integration time) • Compact GEM-based TPC - continuous readout Joe Osborn 5 • Additionally, TPC will contain charge from 2-3 Au+Au collisions at a given time • Hit occupancies of O(100,000) expected • Track reconstruction must be robust to high occupancies from pile up charge sPHENIX Computing Challenges • RHIC will deliver Au+Au collisions at 50 kHz, while sPHENIX will record at 15 kHz • In a 3 year, ∼24 cryo-week per year data taking campaign, sPHENIX will collect ∼ 250 PB of data • Data will be processed on a fixed size computational farm at BNL • Necessitates fast, efficient track reconstruction • Goal is a CPU budget of 5 seconds-per-event on a single tracking pass Joe Osborn 6 • Track reconstruction must be robust to high occupancies from pile up charge sPHENIX Computing Challenges • RHIC will deliver Au+Au collisions at 50 kHz, while sPHENIX will record at 15 kHz • In a 3 year, ∼24 cryo-week per year data taking campaign, sPHENIX will collect ∼ 250 PB of data • Data will be processed on a fixed size computational farm at BNL • Necessitates fast, efficient track reconstruction • Goal is a CPU budget of 5 seconds-per-event on a single tracking pass • Additionally, TPC will contain charge from 2-3 Au+Au collisions at a given time • Hit occupancies of O(100,000) expected Joe Osborn 6 • Track reconstruction must be robust to high occupancies from pile up charge sPHENIX Computing Challenges • RHIC will deliver Au+Au collisions at 50 kHz, while sPHENIX will record at 15 kHz • In a 3 year, ∼24 cryo-week per year data taking campaign, sPHENIX will collect ∼ 250 PB of data • Data will be processed on a fixed size computational farm at BNL • Necessitates fast, efficient track reconstruction • Goal is a CPU budget of 5 seconds-per-event on a single tracking pass • Additionally, TPC will contain charge from 2-3 Au+Au collisions at a given time • Hit occupancies of O(100,000) expected Joe Osborn 6 sPHENIX Computing Challenges • RHIC will deliver Au+Au collisions at 50 kHz, while sPHENIX will record at 15 kHz • In a 3 year, ∼24 cryo-week per year data taking campaign, sPHENIX will collect ∼ 250 PB of data • Data will be processed on a fixed size computational farm at BNL • Necessitates fast, efficient track reconstruction • Goal is a CPU budget of 5 seconds-per-event on a single tracking pass • Additionally, TPC will contain charge from 2-3 Au+Au collisions at a given time • Hit occupancies of O(100,000) expected • Track reconstruction must be robust to high occupancies from pile up charge Joe Osborn 6 sPHENIX-ACTS Track Reconstruction • To work towards meeting these goals, sPHENIX has implemented the A Common Tracking Software (ACTS) toolkit into our software stack • ACTS is intended to be a modern, performant, flexible track reconstruction toolkit that is experiment independent • Largely developed by ATLAS tracking experts; however, user/developer base has grown • ACTS has modern development practices, e.g. • Semantic versioning/releases • Full CI/CD implemented in Github Actions ACTS Github link • Issue tracking • Documentation • Unit testing • ... Joe Osborn 7 • Eventually plan to move to a paradigm where sPHENIX objects == ACTS objects, for saving memory and time • Fun4All-sPHENIX code available on Github - code is open source and containerized with Singularity. Feel free to ask for more details! ACTS Implementation Strategy ACTS info Call tool sPHENIX Object sPHENIX-ACTS Module ACTS Tool Update ACTS result • ACTS requires geometry and measurement objects (that’s all) • sPHENIX objects store necessary information for ACTS objects • Modules act as wrappers for calling ACTS tools and updating sPHENIX objects Joe Osborn 8 • Fun4All-sPHENIX code available on Github - code is open source and containerized with Singularity. Feel free to ask for more details! ACTS Implementation Strategy ACTS info Call tool sPHENIX Object sPHENIX-ACTS Module ACTS Tool Update ACTS result • ACTS requires geometry and measurement objects (that’s all) • sPHENIX objects store necessary information for ACTS objects • Modules act as wrappers for calling ACTS tools and updating sPHENIX objects • Eventually plan to move to a paradigm where sPHENIX objects == ACTS objects, for saving memory and time Joe Osborn 8 ACTS Implementation Strategy ACTS info Call tool sPHENIX Object sPHENIX-ACTS Module ACTS Tool Update ACTS result • ACTS requires geometry and measurement objects (that’s all) • sPHENIX objects store necessary information for ACTS objects • Modules act as wrappers for calling ACTS tools and updating sPHENIX objects • Eventually plan to move to a paradigm where sPHENIX objects == ACTS objects, for saving memory and time • Fun4All-sPHENIX code available on Github - code is open source and containerized with Singularity. Feel free to ask for more details! Joe Osborn 8 ACTS Geometry - Silicon INTT • ACTS is able to perform material calculations quickly due to a simplified geometry model • ACTS contains an available MVTX TGeometry plugin which takes TGeoNodes and builds Acts::Surfaces • Any changes to sPHENIX GEANT 4 silicon surfaces are then reflected in ACTS transparently Joe Osborn 9 ACTS Geometry - TPC • ACTS geometry model not immediately suited to TPC geometries, since surfaces are required • With TPC, charge can exist anywhere in 3D volume • Side note: ongoing development within ACTS to allow for 3D fitting • In place, create planar surfaces that mock cylindrical surfaces 48 TPC layers • Surfaces are set at readout layers, so there is a direct mapping from a TPC readout module to n planar surfaces Joe Osborn 1010 Track Reconstruction Strategy Joe Osborn 11 Track Reconstruction Strategy Joe Osborn 11 Silicon Seeding 1 • ACTS track seeding tool takes a list 0.98 of space points and creates 0.96 three-measurement seeds 0.94 sPHENIX simulation • Ideally suited for MVTX, which has 0.92 100 pions 3 layers 0.9 |φ-φ |<0.02 truth • ACTS seeding efficiency shown in Acts Seeding Efficiency 0.88 |η-η |<0.004 truth 100 pion events 0.86 nMVTX>2 • MVTX triplets are propagated to 0.84 INTT to find additional 0.82 measurements 0.8 0 5 10 15 20 25 30 • Final output is a set of 3-5 p [GeV] T measurement silicon seeds Joe Osborn 12 ACTS Vertex Finding m] 20 µ ) [ 18 sPHENIX simulation vtx • Silicon track seeds have excellent (z σ 16 Initial vertex finding spatial resolution, and almost 14 entirely define the event vertex 12 • Acts::IterativeVertexFinder is used to assign silicon seeds a 10 vertex position 8 • Seeds are clustered to identify 6 outliers and remaining seeds are 4 provided to the ACTS tool 2 0 20 40 60 80 100 Ntracks Joe Osborn 13 CA Seeding in TPC 1 • Independently (from ACTS) developed Cellular Automaton 0.9 Efficiency seeder is deployed in TPC 0.8 sPHENIX simulation 100 pions • Measurement gathering algorithm 0.7 |φ-φ |<0.02 is a reimplementation of concept truth 0.6 |η-η |<0.006 underpinning ALICE TPC tracking truth software 0.5 • Seeder is efficient - continuing 0.4 development on producing high 0.3 quality seeds encompassing entire 0 2 4 6 8 10 12 14 16 18 20 22 p [GeV] TPC T Joe Osborn 14 Track Matching • Track seeds in silicon and TPC are matched to one another with φ/η 9 windows match si 8 N 103 • Windows tuned to limit duplicate 7 matches while ensuring real tracks 6 5 are matched 102 4 • Predominantly have 1 silicon 3 tracklet per TPC tracklet. 2 10 Continuing development to reduce 1 number of multiple matches 0 1 • ACTS fitter (next page) is very good 0 2 4 6 8 10 12 14 16 18 20 sPHENIX simulation p [GeV] at identifying bad seeds in T sPHENIX geometry Joe Osborn 15 • Single upsilon mass spectrum from first mock data challenge meets sPHENIX physics goals • Continued improvement on TPC seeds expected Track Fitting T 0.03 )/p T sPHENIX simulation (p 0.025 • Use tool to σ Acts::KalmanFitter 100 pions fit fully assembled track seeds 0.02 nMVTX>2, nTPC>20 • Tool performs a full fit outwards, then smooths result inwards to 0.015 vertex 0.01 0.005 0 0 2 4 6 8 10 12 14 16 18 20 p [GeV] T Joe Osborn 16 Track Fitting 1000 • Use Acts::KalmanFitter tool to sPHENIX simulation fit fully assembled track seeds Υ Arb. Units 800 Single (1S) • Tool performs