Jets and E reconstruc1on, calibra1on and performance with the ALAS detector at LHC CHEF2013 Paris, April 2013 Silvia Resconi INFN Milano (on behalf of the ALAS Collabora1on)
Outline Atlas calorimeter pile- up sensi1vity Input signals to Jets and E Jet reconstruc1on and calibra1on: Pile- up correc1on: Effect on Jet Energy Resolu1on Jet Energy Scale and uncertainty E reconstruc1on and calibra1on: Pile- up correc1on of SoQ erm: Effect on E Resolu1on E uncertainty CHEF2013 Silvia Resconi 2
Jets: Jets and E play a major role for the physics @ LHC: Measurements of hadronic s provide tests of strong interac1ons. hey can be backgrounds and/or signals for many New Physics searches. he uncertainty on the calibra1on dominates the experimental uncertainty for many physics results. E : Measurement of E is crucial for the search of Higgs boson in the decay channels H WW lνlν and H ττ where a good measurement of E allows for the accurate reconstruc1on of the Higgs boson mass. Large E is a key signature for searches for New Physics processes such as SUSY and extra dimensions. CHEF2013 Silvia Resconi 3
Calorimeter pile- up sensi1vity è Pile- up is one of the main challenges for s and E at LHC In- $me pile- up: addi1onal pp collisions in the same bunch crossing Adds energy in calorimeters mainly from minimum bias interac1ons (MB) Es1mated by the number of reconstructed primary ver1ces (N PV ) Out- of $me pile- up: residual contribu1ons from collisions in preceding bunch crossings LAr calorimeters sensi1ve to bunch crossing history: Long charge collec1on 1me ( 400 ns) Bi- polar shaped signal (600 ns) with zero integral over 1me does not permit full pile- up cancella1on: op1mized for 25 ns spacing (currently 50 ns ) large fluctua1ons of number of interac1ons from bunch crossing to bunch crossing. Es1mated by the average number of interac1ons per bunch crossing <µ> over 1me window >> 600 ns: <µ> = L σ inel / N bunch f LHC CHEF2013 Silvia Resconi 4
Input signals to Jets and E opo- Clusters: group of calorimeter cells topologically connected op1mized for electronic noise and pile- up suppression: Cluster cells in 3D via noise- driven thresholds: Seed: E cell >4 σ noise Neighbours: E cell >2 σ noise Perimeter cells E cell > 0 σ noise = (σ noise electronic ) 2 + ( σ noise pile- up ) 2 Calibra1on using the local cluster weigh1ng (LCW) : Classifica1ons as em- like or hadron- like clusters based on cluster shape variables: energy density and depth. Hadronic weights, derived from pion MC simula1on, applied to hadron- like clusters. Correc1ons for dead material and out of cluster Improve correspondence between clusters and stable par1cles è Contribu1on from pile- up fluctua1ons can survive and overlap on large signal from hard scaoering process, more pile- up suppression techniques needed CHEF2013 Silvia Resconi 5
Jet reconstruc1on and calibra1on Reconstruc1on of calorimeter s: Reconstructed from EM or LCW calibrated topo- clusters Using an1- k t algorithm with distance parameters R = 0.4 and 0.6 Calibra1on of calorimeter s: Mul1- step approach that combines in- situ and MC methods CHEF2013 Silvia Resconi 6
Jet pile- up correc1on Signal added from pile- up: Distorts the energy reconstructed in s: Modifies p, smears resolu1on, adds mass, shiq axis in η, φ Jets from addi1onal pp collisions 50 First method: Offset correc$on MC based correc1on derived from response to pile- up Parametriza1on in N PV and <µ> : Δp = α (N PV 1) + β <µ> α, β obtained from the fits (slopes) Dependent on type and η 2 4 6 8 Number of primary vertices (N ) PV Limita1ons: Average correc1on, does not account for event fluctua1ons in pile- up ac1vity. [GeV] EM Jet p 45 40 35 30 25 20 15 ALAS Preliminary truth 20 p < 25 GeV 25 p 30 p 35 p truth truth truth truth < 30 GeV < 35 GeV < 40 GeV Simulation s = 7 ev Pythia Di, anti-k R=0.4 t < 2.1, 7.5 µ < 8.5 40 p < 45 GeV Average Slope = 0.288±0.003 GeV/N PV CHEF2013 Silvia Resconi 7
Second method: area More sophis1cated correc1on based on the idea that noise (pile- up) 0.14 has a lower density (ρ) than signal: ALAS Simulation ρ = median Jet pile- up correc1on p,i A i ρ calculated for k t s with R=0.4 A i = area calculated with FastJet ac1ve area based on ghost par1cles: Normalised entries 0.12 0.1 0.08 0.06 0.04 0.02 20 µ < 21 N = 6 = 0 0 5 15 20 25 30 [GeV] PV N PV N = 14 = 18 PV N PV Pythia Di 2012, s = 8 ev LCW opoclusters Randomly generated distribu1on of par1cles with infinitesimally small p added to the event and clustered with real signal in finding process. Number of ghost par1cles associated to a is a measure of the area. Correc1on: p,i corr = p,i - ρ A i Advantages: Captures event- by- event fluctua1ons not described by N PV, <µ> Use of area accounts for by varia1ons in sensi1vity Data driven method: no dependence on pile- up modelling CHEF2013 Silvia Resconi 8
Effect of pile- up correc1on on Jet Energy Resolu1on (JER) )/p (p Improvements of RMS (p reco p truth ) for area over the offset correc1on s1ll some residual dependence on <µ> Frac1onal energy resolu1on vs p truth degrades with increasing pile- up condi1ons: noise term increases with <µ> and dominates at low p 0.4 0.3 0.2 0.1 ALAS Preliminary Simulation s = 7eV Anti-k t R=0.4, EM+JES < 0.8 opo cluster Noise hreshold = 30 µ =0 µ = µ = 20 µ = 30 µ = 40 30 40 0 200 300 00 µ CHEF2013 truth p [GeV] Silvia Resconi 9 Noise erm [GeV] 9 8 7 6 5 4 3 2 1 true ) [GeV] - p reco ALAS Preliminary Simulation s = 7eV RMS(p Anti-k t R=0.4, EM+JES < 0.8 14 13 12 11 Pile up suppression method offset corrected x area corrected 0-0 20 30 40 50 9 8 7 6 5 ALAS Simulation Pythia Di s=8 ev anti-k t LCW R=0.6 true 20 p < 30 GeV < 2.4 uncorrected f( µ, N ) correction PV A correction 5 15 20 25 30 35 40 µ Noise term (N) vs <μ> obtained from the fits to the energy resolu1on. N is reduced by applying area correc1on
rack- based pile- up suppression Jet Vertex Frac$on (JVF): rack based quan1ty to filter pile- up s: JVF = ( Σ p track PV / Σ p track ) Probability for s with matched tracks to come from hard- scaoer vertex selected with the highest Σ tracks (p 2 ) > 20 GeV, p N 3.2 ALAS Preliminary MC, No JVF Cut Data, No JVF Cut Z µµ + s MC, JVF > 0.25 3 Data, JVF > 0.25 anti-k t LCW+JES R = 0.4 MC, JVF > 0.50 Data, JVF > 0.50 0.0 2.1 JVF Uncertainty 2.8 2.6 2.4 2.2 2 0 5 15 20 25 30 35 40 hree different JVF regions: JVF=- 1: no matched tracks JVF= 0: all matched tracks from pile- up 0 < JVF 1: JVF closer to 0 for with significant pile- up contribu1on JVF cut reduces pile- up s: Mean mul1plicity in Z+s improves data/mc agreement CHEF2013 µ Silvia Resconi
Jet energy calibra1on Simple correc1on based on MC rela1ng the reconstructed energy to the truth energy: Correc1on factor = E truth / E EM/LCW dependent on E, η Much smaller correc1on for LCW- scale s wrt EM- scale s Residual in situ calibra1on From in situ techniques that exploit the p balance between the p and the p of a reference object: Correc1on factor = < p / p ref > data / < p / p ref > MC Mul1ple methods to cover large kinema1c phase space: Di η- intercalibra1on: equalize p between central and forward γ/z + direct balance: compare p γ or p Ζ with recoiling p Mul1 balance: low p system recoils against a high p used to calibrate s in the ev regime. CHEF2013 Silvia Resconi 11
Jet Energy Scale (JES) otal in situ JES correc1on as a func1on of p obtained from the combined in situ techniques: Z+ allows to probe the energy scale down to p 17 GeV γ + relevant for 0 p 800 GeV Mul1 balance provides results at high p ( 1.5 ev) è he residual calibra1on is minimal ( 1%) with a maximum uncertainty 3%. è Adding addi1onal uncertain1es due to pile- up, flavour response, close- by s frac1onal uncertainty < 2.5 % for central s with p > 0 GeV / Response Response MC Data 1.1 1.08 1.06 1.04 1.02 1 0.98 0.96 0.94 0.92 anti-k t R = 0.4, LCW+JES Data 2012 + Z + Multi ALAS Preliminary s = 8 ev, < 0.8 otal in situ uncertainty Statistical component Fractional JES uncertainty 0.9 0 2 2 3 2 2 3 3 20 30 40 2 20 30 40 2 2 p [GeV] p [GeV] CHEF2013 Silvia Resconi 12 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 anti-k t R = 0.4, LCW+JES + in situ correction Data 2012, = 0.0 s = 8 ev otal uncertainty Absolute in situ JES Relative in situ JES Flav. composition, inclusive s Flav. response, inclusive s Pileup, average 2012 conditions Close-by s, inclusive s ALAS Preliminary
E reconstruc1on and calibra1on Missing transverse momentum (E ) is a complex event quan1ty: Adding significant signals from all detectors Asking for momentum conserva1on in the transverse plane. Reconstructed physics objects: e, γ, τ, s, muons SoQ energy: topo- clusters and tracks not associated to physics objects Decomposi1on into cons1tuent topo- clusters to veto mul1ple contribu1ons and avoid energy double coun1ng Keep separate contribu1ons calculated from the nega1ve sum of calibrated p x(y) of physics objects and soq energy E x(y) = E x(y),e +E x(y),γ +E x(y),τ +E x(y),s +E x(y), SoQ erm + E x(y), µ E = ( E x ) 2 + ( E y ) 2 13
Effect of pile- up on E Resolu1on E highly affected by pile- up: largest acceptance (coverage area) of any given reconstructed quan1ty. Drama1c effect on E x,y resolu1on with increasing pile- up condi1ons: Resolution [GeV],E y E x 30 25 20 15 5 <µ>=0: fit 0.40 <µ>=9: fit 0.68 <µ>=20: fit 0.81 <µ>=30: fit 0.88 <µ>=40: fit 0.94 E E E E E No pile-up suppression MC Z µµ Simulation s = 7 ev ALAS Preliminary 0 0 0 200 300 400 500 600 700 E x,y resolu1on = k Σ E E x,y resolu1on doubles from <µ> = 0 to 20 (2012 pile- up condi1ons) Large impact on analyses with E in final state (SUSY searches, H ττ invariant mass reconstruc1on) Jet term and SoQ erm are the most affected by pile- up: E (event) [GeV] Jet term pileup correc1on with area method plus JVF cut. SoQ erm is very similar to pile- up, so any correc1on should be based or on PV associa1on or on exploi1ng the small difference between signal and pile- up è very challenging CHEF2013 Silvia Resconi 14
E SoQ erm pile- up correc1on Resolution [GeV],E y E x First method: SoG erm Vertex Frac$on (SVF) rack based quan1ty to scale SoQ erm: 25 20 15 SVF = ( Σ p track SoQ erm PV / Σ p track SoQ erm ) Uses tracks not associated to physics objects and matched to PV to provide a reliable es1mate of pile- condi1ons. Restore E x,y resolu$on closer to that observed in absence of pileup Limita1ons: calculated in limited coverage (ALAS ID η < 2.5) and does not take into account neutral contribu1ons 30 Data 2012 default Data 2012 Pile-up suppression MC default MC Pile-up suppression ALAS Preliminary Z µµ s = 8 ev Ldt=20 fb -1 Resolution [GeV],E y E x 35 30 25 20 15 5 Data 2012 default MC default Data 2012 Pile-up suppression MC Pile-up suppression ALAS Preliminary MC Z µµ s = 8 ev Ldt=20 fb -1 0 0 5 15 20 25 0 200 400 600 800 00 1200 N pv CHEF2013 Silvia Resconi E (event) before pile-up suppression [GeV] 15
Resolution [GeV],E y E x E SoQ erm pile- up correc1on Second method: area Similar approach as the one described for s but more challenging when applied to SoQ erm. Procedure: Reconstruct k t s with R=0.4 from topo- clusters and tracks of SoQ erm Correct each for pile- up applying a filter and recalculate SoQ erm E x(y), SoQ erm = - Σ p x(y),i corr, p,i corr = 24 22 20 18 16 14 12 8 ALAS Preliminary Data 2011 s = 7 ev -1 Ldt=4.2fb Z ee 0 s p >20GeV default 2011 Pile-up suppression Jet Area 6 0 2 4 6 8 12 14 16 18 20 p i - ρ A i, p i > ρ A i 0, p i ρ A i CHEF2013 Silvia Resconi 16 N pv Jet area method: Improves the resolu1on but some residual dependence on N PV s1ll present Can be combined with a track based filter (JVF) on k t s similar to what explained for s
Z µµ E data/mc agreement and uncertainty W eν Events / 2 GeV Data / MC 6 5 4 3 2 1 Ldt=20 fb s= 8 ev -1 ALAS Preliminary Data 2012 MC Z µ µ MC ttbar MC WZ MC ZZ MC WW 1.6 1.4 1.2 1 0.8 0.6 0 50 0 150 200 250 E after pile-up suppression [GeV] uncertainty Fractional E 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 No pile-up suppression,softerm Resol E on E,Softerm Scale E on E,JES E,JER E,e E on E on E on E,terms All E on E ALAS Preliminary Simulation W e s = 7 ev 0 200 300 400 500 600 700 800 E [GeV] Good E data/mc agreement in Z µµ in 2012 data ( 20 { - 1 ) otal E frac1onal uncertainty obtained combining uncertain1es of all physics objects (e, γ, τ, s, µ) used to calculate E terms and the SoQ erm: Depends on event topology: on average 3% in W eν (in 2011), increasing with ΣE CHEF2013 Silvia Resconi 17
Conclusions ALAS has developed and comioned several techniques to mi1gate pile- up effects coherently on s and E : opological clustering, intrinsically noise and pile- up suppressed. Jet area method to es1mate event- by- event the p density from pile- up. Use of tracks to filter pile- up s and scale the SoQ erm: Jet mul1plicity and E resolu1on closer to that observed in absence of pile- up. A high precision obtained for JES in 2012 with in- situ techniques: JES uncertainty smaller than 2.5% for s with p > 0 GeV è Dedicated op1miza1on of all these techniques needed to face the new challenge in 2015 at very high luminosity: New techniques, like grooming, have been developed to reduce pile- up sensi1vity in boosted topologies CHEF2013 Silvia Resconi 18
Back- up CHEF2013 Silvia Resconi 19
References opoclusters and LCW calibra1on: AL- LARG- PUB- 2009-001 Jet Area method: M. Cacciari, G. P. Salam, Pileup subtrac0on using areas, Phys.Leo.B659(2008), arxiv:0707.1378 [hep- ph] Jet pile- up correc1on: ALAS- CONF- 2012-064 Updated performance plots for 2012 data: h6ps://twiki.cern.ch/twiki/bin/view/atlaspublic/jetetapproved2013pileup1 Jet Energy Scale and its systema1c uncertainty: ALAS- CONF- 2013-004 Updated performance plots for 2012 data: h6ps://twiki.cern.ch/twiki/bin/view/atlaspublic/jetetapproved2013jesuncertainty Performance of E : ALAS- CONF- 2012-1 Updated performance plots for 2012 data: h6ps://twiki.cern.ch/twiki/bin/view/atlaspublic/jetetapproved2013etmiss Jet substructure: ALAS- CONF- 2012-065, ALAS- CONF- 2012-066 CHEF2013 Silvia Resconi 20
Jet substructure he large centre- of- mass energy of LHC enables the produc1on of boosted heavy par1cles whose decay products can be reconstructed as one large radius. Needed new techniques: Jet grooming Emphasize hard substructure removing soq radia1on Reduce their sensi1vity to pile- up hree algorithms: rimming, pruning, filtering rimming algorithm: Recluster k t subjects, remove those with p i / p < f cut to form the trimmed MC based calibra1on to restore a uniform mass response over the full η range and validated in situ Arbitrary units 0.22 0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 ALAS Preliminary - Simulation anti-k t LCW s with R=1.0, 600 p No grooming applied Z qq No grooming applied Dis (POWHEG+Pythia) rimmed (f =0.05, R =0.3) Z qq cut sub < 800 GeV rimmed (f =0.05, R =0.3) Dis (POWHEG+Pythia) cut sub 0 0 50 0 150 200 250 300 350 400 Jet mass [GeV] rimming Improve mass resolu1on for signal (Z qq) rimming CHEF2013 Silvia Resconi Remove dependence of Jet mass vs N PV 21
Jet Energy Scale uncertainty Frac1onal JES uncertainty as a func1on of p and η : è In central region: uncertainty < 2.5 % for central s with p > 0 GeV è In forward region: uncertainty increases in the forward region up to 7% due to difference in the modelling of the parton showering between PYHIA and HERWIG++ in the di η- intercalibra1on method. Fractional JES uncertainty 0.1 0.09 0.08 0.07 0.06 anti-k t R = 0.4, LCW+JES + in situ correction Data 2012, = 0.0 s = 8 ev otal uncertainty Absolute in situ JES Relative in situ JES Flav. composition, inclusive s Flav. response, inclusive s Pileup, average 2012 conditions Close-by s, inclusive s ALAS Preliminary Fractional JES uncertainty 0.05 0.05 0.04 0.04 0.03 0.03 0.02 0.02 0.01 0.01 0 2 2 3 3 20 30 40 2 2 0-4 -3-2 -1 0 1 2 3 4 p [GeV] CHEF2013 Silvia Resconi 22 0.1 0.09 0.08 0.07 0.06 R = 0.4, LCW+JES + in situ anti-k t Data 2012, p = 40 GeV s = 8 ev correction otal uncertainty Absolute in situ JES Relative in situ JES Flav. composition, inclusive s Flav. response, inclusive s Pileup, average 2012 conditions Close-by s, inclusive s ALAS Preliminary
Effect of pile- up correc1on on E Scale 2011 2011 m Τ = 2(p lept * E )(1 cos Δφ lept, E ) Check the effect of pileup correc1on for SoQ erm on E scale in W eν events without s : both SVF and area methods improves the agreement of E and the transverse mass with the ruth. CHEF2013 Silvia Resconi 23
Effect of pile- up correc1on on E Scale E scale in Z µµ events: Mean value of the projec1on of E onto the longitudinal axis defined by the vectorial sum of the 2 leptons momenta: è sensi1ve to the balance between the leptons and the hadronic recoil. è aqer pileup suppression slight increase of the bias If the 2 leptons from Z perfectly balance the hadronic recoil the projec1on of E along Z direc1on should be zero. CHEF2013 Silvia Resconi 24
Jet pile- up correc1on true ) [GeV] - p reco RMS(p 14 13 12 11 9 8 7 6 5 ALAS Simulation Pythia Di s=8 ev anti-k t LCW R=0.6 true 20 p < 30 GeV < 2.4 uncorrected f( µ, N ) correction PV A correction 5 15 20 25 30 35 40 µ Resolu1on: improvements of area over the offset correc1on Jet area provides % improvement wrt offset correc1on S1ll some residual dependence on <µ>. Scale: shown dependence of p on N PV vs η Needed a residual correc1on very similar to the offset correc1on (in par1cular in Forward calorimeter) [GeV] p / N PV 0 0.5 1 1.5 2 2.5 3 3.5 4 CHEF2013 Silvia Resconi 25 1 0.8 0.6 0.4 0.2 0-0.2-0.4 ALAS Simulation Pythia Dis 2012 anti-k t LCW R=0.4 Before any correction After A subtraction After residual correction
Jet energy resolu1on (JER) Frac1onal energy resolu1on vs p : measured in situ using di balance and bisector techniques. Slight beoer data- MC agreement obtained with the bisector method Jets with LCW calibra1on provides improved resolu1on up to 40% at 1 ev wrt EM+JES calibra1on )/p (p 0.25 0.2 0.15 ALAS Preliminary Data 2011 s = 7 ev -1 L dt ~ 950 pb R = 0.6 cluster s Anti-k t LC+JES calibration 0.0< y <0.8 ref 0.0< y <0.8 probe )/p (p 0.3 0.25 0.2 Data 2011 s = 7 ev anti-k t R = 0.6 cluster s 0.0< y <0.8 ALAS Preliminary -1 L dt = 950 pb 0.15 Diff (%) (Data-MC) 0.1 0.05 Di Balance: Monte Carlo (PYHIA) Di Balance: Data Bisector: Monte Carlo (PYHIA) Bisector: Data 0 30 40 50 60 70 80 0 200 300 400 500 00 20 0-20 30 40 50 60 70 80 0 200 300 400 500 00 p (GeV) Rel. Improvement (%) 0.1 0.05 EM+JES LCW 0 60 30 40 50 60 70 80 EM+JES 0 200 300 400 500 00 40 LCW 20 0 30 40 50 60 70 80 0 200 300 400 500 00 p (GeV) CHEF2013 Silvia Resconi 26
CHEF2013 ALAS calorimeters Main features for and E reconstruc1on and calibra1on: Heterogeneous system with fine granularity calorimeters and many transi1on regions: Crack regions: η 1.4, 3.2 Non compensa1ng (e/h) > 1: Response to hadrons is lower than that to electrons and photons. è Developed specific calibra1ons Dead material: Energy loss before EM calorimeter and between EM and HAD barrel calorimeters: è Dead material correc1ons ALAS Fiducial Regions: Hadronic Calorimeter: Barrel (ile) η <1.7 Endcap (LAr- Cu) 1.5 < η <3.2 Electromagne1c Calorimeters: Barrel (LAr- Pb) η <1.4 Endcap (LAr- Pb) 1.375 < η <3.2 Forward (LAr) : 3.2 < η <4.9 Silvia Resconi 27
he SoQ erm track- cluster matching algorithm (1) rack selec$on All tracks from rackcontainer Apply quality criteria Veto on tracks associated to high physics objects Add good tracks to E calcula1on (2) Cluster removal All opoclusters not associated to physics objects Veto on opoclusters associated to good tracks Add remained opoclusters to E calcula1on è Improve calcula1on of the low contribu1on to SoQ erm è racks are added to recover the contribu1on from low- p par1cles which do not reach the calorimeter or do not seed a opocluster. è No associa1on with PV => no pile- up suppression at this level 28