Particle Flow at 40 MHz with the CMS L1 Trigger Christian Herwig, - PowerPoint PPT Presentation

Particle Flow at 40 MHz with the CMS L1 Trigger Christian Herwig, for the CMS L1PF Team CPAD Instrumentation Frontier Workshop December 8-10, 2019

Outline • Motivation and the High-luminosity LHC • Particle Flow reconstruction • PUPPI Pileup subtraction • The Phase-II Upgrade to the L1 CMS Trigger • Progress of PF+PUPPI implementation C. Herwig — CPAD Instrumentation Frontier Workshop 2 Dec. 9, 2019

We are here C. Herwig — CPAD Instrumentation Frontier Workshop 3 Dec. 9, 2019

We are here Phase-II upgrades 10x dataset increase C. Herwig — CPAD Instrumentation Frontier Workshop 4 Dec. 9, 2019

Discover Higgs! C. Herwig — CPAD Instrumentation Frontier Workshop 5 Dec. 9, 2019

-1 s = 13 TeV, 36.1-139 fb July 2019 700 ) [GeV] Observed limits Expected limits ATLAS Preliminary 70 ~ ~ 60 600 t t production -1 139.0 fb 1 1 + m W ) = 0 50 ~ ∼ 0 1L, t Wb 0 1 → χ Limits at 95% CL 0 ) = m ∼ 1 ∼ χ b 1 χ 1 ~ , 40 m( t 0 [ATLAS-CONF-2019-17] 1 m( ∼ χ , 1 ~ Δ t m( 1 500 ) = m 30 t Δ 0 ∼ -1 χ 36.1 fb 1 ~ , 200 210 220 230 t m( ~ ~ 1 ∼ 0 ∼ 0 0L, t → t χ / t → Wb χ Δ 1 1 Constraints on BSM Physics 1 1 [1709.04183] 400 ~ ~ ~ ∼ 0 ∼ 0 ∼ 0 1L, t t / t Wb / t bff' → χ → χ → χ 1 1 1 1 1 1 [1711.11520] ~ ~ ~ ∼ ∼ ∼ 0 0 0 2L, t → t χ / t → Wb χ / t → bff' χ 300 1 1 1 1 1 1 [1708.03247] (especially strongly produced) ~ ∼ 0 monojet, t → bff' χ 1 1 [1711.03301] 200 ~ ∼ 0 t t , t → t χ 1 1 [1903.07570] ~ ∼ 0 c0L, t c → χ 1 1 100 [1805.01649] ~ ∼ 0 monojet, t → c χ 1 1 [1711.03301] 200 300 400 500 600 700 800 9001000 -1 Run 1, s = 8 TeV, 20 fb [1506.08616] ~ m( t ) [GeV] 1 C. Herwig — CPAD Instrumentation Frontier Workshop 6 Dec. 9, 2019

HL-LHC 14 TeV Higgsino-like EWK processes 30 CMS Phase-2 -1 (%) 3000 fb (14 TeV) m(NLSP, LSP) [GeV] CMS Phase-2 Loss in signal significance [%] HL-LHC 3/ab, 14 TeV (soft-lepton A) Simulation Preliminary HL-LHC monojet 30 Simulation Preliminary SM HL-LHC 3/ab, 14 TeV (soft-lepton B) HE-LHC 15/ab, 27 TeV (soft-lepton B) LHeC monojet-like (proj) σ 25 FCC-hh (HE-LHC approx. rescaling) -1 L = 300 fb inv)/ ILC , 0.5/ab HE-LHC monojet data 500 HH → b b b b ILC , 1/ab 25 1000 -1 FCC-eh monojet-like CLIC / FCC-ee L = 1000 fb 2 380 380 10 data CLIC , 2.5/ab 1500 FCC-hh monojet → CLIC , 5/ab 20 -1 3000 L = 3000 fb B(H data 20 × 15 σ 15 Δ 95% CL upper limit on 10 10 10 5 CLIC: extrapolated below 5 GeV 5 Monojet reach in Δ m(NLSP,LSP) not displayed 1 0 0 200 400 600 800 1000 1200 1400 45 50 55 60 65 70 75 80 150 200 250 300 350 400 miss m(NLSP) Minimum threshold on E (GeV) Minimum jet p threshold [GeV] T T Rare+Exotic Higgs EWK BSM SM hh C. Herwig — CPAD Instrumentation Frontier Workshop 7 Dec. 9, 2019

L1 HLT 40 mhz 100 khz 1 khz 35 pp/event (400x rej) (100x rej) Typically limited to information from a single sub-detector (calorimeter, muons) C. Herwig — CPAD Instrumentation Frontier Workshop 8 Dec. 9, 2019

L1 HLT 40 mhz 750 khz 7.5 khz 200 pp/event (50x rej) (100x rej) Naively scales with luminosity C. Herwig — CPAD Instrumentation Frontier Workshop 9 Dec. 9, 2019

Challenges to Phase-II L1 Trigger • L1 Accept rate scales ~ linearly with luminosity increase • Must maintain performance in hostile environment! C. Herwig — CPAD Instrumentation Frontier Workshop 10 Dec. 9, 2019

� � Challenges to Phase-II L1 Trigger • L1 Accept rate scales ~ linearly with luminosity increase • Must maintain performance in hostile environment! Take hh production in 4 b (or bb ττ ) decay mode 0.14 Normalized entries Higher pileup ATLAS Simulation 0.12 = 6 = 10 20 ≤ < 21 N N → Extra stochastic energy PV PV = 14 = 18 N N 0.1 PV PV enters into the jet cone Pythia8 dijets, √ s = 8 TeV 0.08 � from LCW topo-clusters 0.06 More low-p T jets to "measure 0.04 high" than vice versa 0.02 → Higher trigger rate 0 0 5 10 15 20 25 30 [GeV] C. Herwig — CPAD Instrumentation Frontier Workshop 11 Dec. 9, 2019

Challenges to Phase-II L1 Trigger • L1 Accept rate scales ~ linearly with luminosity increase • Must maintain performance in hostile environment! It gets worse !! Background (uncorrelated coincidences) ~ (lumi) 2 beamspot "cigar"~30cm Not new problems, solved offline with Particle Flow Reco+ C. Herwig — CPAD Instrumentation Frontier Workshop 12 Dec. 9, 2019

Particle Flow Reconstruction • Idea: combine measurements across all sub-detectors to achieve best possible resolution per object • Algorithm returns a list of single-particle candidates Muons Tracks Electrons Muon segments (Isolated) photons ECal Charged hadrons HCal Neutral hadrons C. Herwig — CPAD Instrumentation Frontier Workshop 13 Dec. 9, 2019

Particle Flow Reconstruction • Idea: combine measurements across all sub-detectors to achieve best possible resolution per object 0.5 Energy resolution resolution Anti-k , R = 0.4 Calo CMS CMS 0.6 T 0.45 • Algorithm returns a list of single-particle candidates Calo Simulation Ref PF | | < 1.3 Simulation η PF 0.4 0.35 miss Muons 0.4 0.3 T Tracks Relative p 0.25 Electrons 0.2 Muon 0.2 0.15 0.1 segments (Isolated) photons 0.05 0 0 50 100 150 200 250 20 100 200 1000 ECal miss p (GeV) Ref p (GeV) T,Ref T Charged hadrons improved jet pT resolution improved missing pT resolution Improved Jet p T resolution Improved p T -miss resolution HCal Neutral hadrons C. Herwig — CPAD Instrumentation Frontier Workshop 14 Dec. 9, 2019

Pileup Per Particle Identification • Idea: get probability that a neutral PF candidate is pileup based on local activity from the leading vertex -1 -1 0.36 fb 0.36 fb (13 TeV) (13 TeV) fraction of particles weight>0.01 3 10 charged LV CMS CMS Neutral Particles 2 charged PU 10 Data Preliminary neutrals LV Pileup 10 0.06 MC neutrals PU N/ N 1 p T ,i 1 − 10 X α ∼ 0.04 Δ 2 − Leading 10 ∆ R i 3 − i ∈ cone 10 Vertex − 4 10 0.02 5 − 10 2 Data/MC Weight 1407.6013 1 0 -5 0 5 10 15 0 0 0.2 0.4 0.6 0.8 1 C α Weight i C. Herwig — CPAD Instrumentation Frontier Workshop 15 Dec. 9, 2019

Pileup Per Particle Identification • Idea: get probability that a neutral PF candidate is pileup based on local activity from the leading vertex -1 35.9 fb (13 TeV) 40 ) [GeV] CMS 35 Preliminary ( u 30 σ 25 Improved p T -miss 20 resolution 15 Response-corrected miss PF p Z ee → 10 T miss PUPPI p Z ee → T 5 JME-18-001 Uncertainty 0.7 0 0.6 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 Number of vertices C. Herwig — CPAD Instrumentation Frontier Workshop 16 Dec. 9, 2019

Architecture of the Phase-II L1 Trigger C. Herwig — CPAD Instrumentation Frontier Workshop 17 Dec. 9, 2019

Architecture of the Phase-II L1 Trigger 2-3 GeV tracks | η |<2.5 9 ɸ sectors vertices C. Herwig — CPAD Instrumentation Frontier Workshop 18 Dec. 9, 2019

Architecture of the Phase-II L1 Trigger Layer 1: Run the PF+PUPPI algorithm itself Layer 2: Algorithms using PF+PUPPI inputs C. Herwig — CPAD Instrumentation Frontier Workshop 19 Dec. 9, 2019

Strategy for L1 Implementation • Take advantage of the inherent locality of PF+PUPPI • Distribute computation across many processing units • Processing is divided into three main steps: Layer 1 • Regionalization (VHDL) • PF+PUPPI calculation (High Level Synthesis C++) • Algorithms using PF+PUPPI inputs (HLS C++) Layer 2 • HLS: no expertise required! • Fast prototyping, debugging, comparison of alg variants C. Herwig — CPAD Instrumentation Frontier Workshop 20 Dec. 9, 2019

Inputs versus η , PF+PUPPI regions TMUX 18 → 6 C. Herwig — CPAD Instrumentation Frontier Workshop 21 Dec. 9, 2019

100% Regionizer validation match! # objects CMS Internal 100 Simulation Tracks EM calo Emulation Calo Muons 80 VHDL algorithm validated with simulated data inputs 60 40 # objects CMS Internal 80 20 Tracks EM calo Calo Muons 70 ttbar events 0 60 0 2 4 6 8 10 12 14 16 μ ~200 Region index 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 Input link index C. Herwig — CPAD Instrumentation Frontier Workshop 22 Dec. 9, 2019

HW Particle Flow + PUPPI • Regionalization → small # of objects to link (truncation) • Cluster input pre-processing: exploit shapes • PUPPI 'linearized'; smaller cone size Work in Progress • Classify cluster: • Hadronic or EM-like? • Remove pileup deposits • Less work for PUPPI! C. Herwig — CPAD Instrumentation Frontier Workshop 23 Dec. 9, 2019

Resource drivers • Many Δ R calculations for track-calo linking drives DSP • Scales as (#tracks)*(#calo clusters) • PUPPI weights drive BRAM usage • To compute p T / Δ R quickly requires division tables • DSPs also used to map (p T , Δ R) → PUPPI weights Resource LUT FF BRAM DSP Usage 528k 785k 871 1020 % VU9P 45% 33% 40% 15% PF+PUPPI resources for 22 tracks, 15+13 calo clusters C. Herwig — CPAD Instrumentation Frontier Workshop 24 Dec. 9, 2019