Particle Flow Calorimetry Technology Requirements and Opportunities François Corriveau Institute of Particle Physics of Canada and McGill University (Montréal) American Workshop for Linear Colliders 2020 October 21 st , 2020 2020.10.21 F.Corriveau (IPP/McGill) - AWLC 2020 - Particle Flow Calorimetry 1
26
Embed
Particle Flow Calorimetry Technology Requirements and ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Particle Flow Calorimetry Technology Requirements and Opportunities
François Corriveau
Institute of Particle Physics of Canada and McGill University (Montréal)
1) Identify and follow each particle in the detector 2) Optimize the event reconstruction for energy/momentum/position/(time) 3) Use the best available information for each particle, such as:
• tracker information for charged particles • electromagnetic calorimeter for photons • hadronic calorimeter for neutral hadrons • .. and their inter-correlations
This then requires from the detectors:
• accurate tracking with high efficiency • high calorimeter granularity (particle separation) • maximum hermiticity in containing particles
PFA is actually routinely applied to single particles in most detectors. The technique however hits limitations and breaks down:
• when the particle density saturates the detector granularity, or • in large background events, such as at hadron colliders, or • for hadronic “jets” typical of event final states at LCs
e+e- colliders provide a unique opportunity to exploit the full potential of PFA:
• relatively less background • design of the detectors around PFA techniques • use of the latest reconstruction methods
However, other effects come to play and worsen the achievable resolution:
• tracking and support material in front of the electromagnetic calorimeter • “confusion” term when particle association/separation is ambiguous • tracking efficiency, missing energy, ..
ECAL must be a compact highly granular calorimeter optimized for electron and photon reconstruction, as well as separation from hadrons Main requirements:
• High-Z (Pb or W) absorber needed to keep shower radius small • Cell sizes of the order of 5 × 5 mm2 • There are easily millions of channels: cost and data volume challenges • Readout via silicon pads or scintillator strips (price vs size)
Alternatives?
• Monolithic Active Pixel Sensors (MAPS) technology offers extreme granularity but at a yet prohibitive cost.
24 layers, 30 µm pitch, 4×4×12 cm3, 39 M pixels! Nuclear and Particle Physics Proceedings 273–275 (2016) 1090–1095
new technological prototype with tungsten absorber Si pads: 5 × 5 mm2 (ILD design) 15 Si layers × 1024 channels/layer ≈ 15000 cells
going to test beams again at the end of 2020 components could be installed in a e+e- collider!
Previous: Development of a track-finding algorithm • removal of interaction region: hits ≥ 6 neighboring pads • clusterisation of energy deposits: seeds from downstream • track-like clustering: minimal length and limited curvature
99.7% efficiency for muons, ≤10% agreement between MC and π data Numerous results: energy fraction in core, lateral size, Nclusters, Ntracks, Nhits per track, angle distribution, .. Use secondary (~MIP) tracks for in situ calibration? Insights to be incorporated into PFA optimization.
With relaxed geometrical constraints, 8 modules were used to build a “long slab” typical of what is needed for the ILD barrel. Successful feasibility studies: mechanical structure cosmics electronics tested with sources MIP response DESY beams
arXiv:2004.13791v1
arXiv:1909.04329
Also in the works: development of an ultra-thin PCB called Chip-on-Board (COB) that is equipped with wirebonded ASICs and pixelated silicon wafers to form the basic unit of detection.
long slab: 144 cm × 18 cm
Design configuration: “(20+10)”, i.e. 20 thin W layers (2.5 mm) 10 thick W layers (5.0 mm) 1.25 mm readout gap
Energy leakage of electromagnetic particles estimated by analyzing the patterns in total energy deposition in each layer using neural networks. (18+6) vs (60+0) GEANT4 models, with:
HCAL must be very large in order to contain extended hadronic showers at LC energies. However the granularity does not need to be as extreme as for electromagnetic calorimeters. Main requirements:
• Traditional approach: Fe or Pb absorbers can be used Scintillator as affordable active medium
• Cell sizes of the order of 30 × 30 mm2 • Readout is now possible via silicon photo-multipliers (SiPM)
advantages: small, tile-integrable, low voltage (~60-100 V) disadvantages: saturation leads to non-linearities (can be handled)
• There are also millions of channels: cost and data volume challenges Several approaches are investigated in analog or digital modes.
SiPM
38 layers 72×72×2.5 cm3 / layer 22,000 tiles SiPM under the tiles for better uniformity and light collection each cell also provides time information with ~1ns resolution a true 5D “pixel” detector: x,y,z,E,t
50 layers, based on cheap/tested resistive plate chamber technology. 96 × 96 channels per layer, i.e. 460,800 1×1 cm2 readout channels. Energies are not measured, but hits are counted → simple, fast readout. Principle demonstrated, still issues.
48 layers × 26 mm, also made of glass RPC. 96 × 96 channels per layer, i.e. 442,368 1×1 cm2 readout channels. Energies are not measured per se, but hits are counted with 3 thresholds coded into 2 bits → pads with few, many or lots of hits. Optimize hadronic shower reconstruction via choice of thresholds. Better linearity response, improved energy resolution.
“Implement a large number of ‘decoupled’ pattern-recognition algorithms, each of which looks to reconstruct specific particle topologies, whilst carefully avoiding causing confusion”
working outwards topological merging tracker-calorimeter improve match neutral vs charged particle flow objects particle identification CLIC Workshop 2013
ILC: Tested with ILD-model Monte-Carlo Z’→jj events produced at rest at 4 energies
(θ = polar angle)
100-250 GeV jets: resolution ~constant (barrel) 45 GeV jets: limited by intrinsic term high energy jets: limited by confusion term PFA robust wrt shower parameters
CLIC (higher energies and larger backgrounds): e.g. W vs Z separation (pT, PID) “traditional” “novel” e+e-→WW→µνqq
e+e-→ZZ→ννqq W/Z energies: 125-1000 GeV overlaid γγ→hadrons background (BX=beam crossing) 2σ separation without background ~1.7σ with 60BX background
Shower development topology in an imaging calorimeter reminds of a tree structure. Backward approach, from leaf to branches to tree with seeds often in the last layers
April APRIL ≈ Arbor PFA with modified cluster merging for SDHCAL
Garlic Gamma reconstruction at a Linear Collider arXiv:1203.0774
High luminosity LHC will have radiation background conditions such that the current CMS endcap calorimeter will no longer be efficient: an upgrade opportunity to reach for new physics with the HGCAL. High granularity to distinguish very narrow VBF jets + Timing for an effective pile-up rejection → complex and ambitious new detector
TICL – The Iterative Clustering
modular framework for particle reconstruction
1) pattern recognition: from hits to tracksters 2) GPU-friendly 2D clustering 3) NN → Particle ID score and energy regression
~340 physicists and engineers 58 institutes 18 countries 4 regions
R&D international collaboration towards highly granular calorimetry optimized for particle flow event reconstruction for future detectors focusing on ILC and CLIC
several technologies are studied prototypes tested in particle beams investigate performances in details
Shower development topology in an imaging calorimeter reminds of a tree structure. Step 1: initial hit cleaning if necessary (e.g. noise) close pairs of hits are connected a connector is the outgoing vector between them Step 2: a reference direction calculated from a hit position and the directions of its outgoing connectors the most likely incoming connector is kept → tree structure this structure can be iterated. Tree means no loop. Step 3: some hits are seeds (no ingoing connector) or leafs (no outgoing connector) tracing from leaf to seed → branches → tree ideal case, a tree = a particle shower intuitive, effective is separating nearby showers
The algorithm is next applied to jets Reconstructed energy for e.g. Higgs decay events Jet energy resolution comparable to Pandora’s
arXiv:1403.4784
APRIL: Algorithm for Particle Reconstruction at ILC from Lyon
APRIL ≈ Arbor PFA with modified cluster merging for SDHCAL
(Pandora PFA assumes linear responses as in AHCAL case) SDHCAL energy reconstruction: Ereco = α1N1 + α2N2 + α3N3 where Ni are the number of hits for each threshold
Tracks: 1) clustering done by Arbor with parameters set to avoid big clusters 2) remaining hits merged by efficient Nearest Neighbour clustering (mlpack) 3) keep only one backward connection per hit (minimal angles × distance) Clusters: 1) cluster merging similar to above hit clustering 2) function of cluster orientations and distances 3) (work in progress, e.g. splitting/reclustering) Results: jet energy resolution in barrel at MZ APRIL: 4.2% → competitive with Pandora (<60 GeV) Pandora: 4.1% “ideal PFA”: 3.3% CHEF 2019