Underlying Event measurements with first LHC data

Transcription

1 Underlying Event measurements with first LHC data Holger Schulz Graduiertenkolleg Masse, Spektrum, Symmetrie Berlin, Septemer 29, 2009

2 Holger Schulz Underlying Event measurements with first LHC data 1 / 13 Introduction LHC is a QCD machine hard to find interesting signals QCD perturatively calculale in hard processes Need models for soft physics (α s 1) to understand ackground Large ackground at LHC is Underlying Event (UE) UE everything except the hard scattering of interest Have different models/generators: Herwig, Pythia, Phojet, Sherpa... LHC-predictions differ vastly need measurements to tune generators

3 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Incoming eams, parton density functions (pdfs) & primordial k

4 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) The hard su-process, the matrix element

5 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Resonance decays correlated with the hard su-process

6 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Initial-state radiation (ISR), parton shower (ackward evolution)

7 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Final-state radiation (FSR), parton shower (forward evolution)

8 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Multiple parton-parton interactions soft, semi-hard or hard scatterings

9 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Initial-/Final state showers of ISR-particles

10 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Formation of colour strings, outgoing partons & eam remnants

11 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Hadronisation

12 Holger Schulz Underlying Event measurements with first LHC data 2 / 13 Hadron collisions (orrowed from Leif Lönnlad) Decay of unstale particles, this is what hits the detector

13 Holger Schulz Underlying Event measurements with first LHC data 3 / 13 Extrapolations to the LHC Drastically different predictions for LHC Different UE energy-scaling: Phojet ln s Pythia ln 2 s Generators were tuned to data at different s Will need retuning of UE-parameters to LHC data plot y A. Moraes

14 UE measurements at the Tevatron Z p from q q Z: α S in ISR, primordial k Multiplicity distriutions: numer of particles produced p vs. N ch : numer and p of particles produced Exploiting the event topology - p sum, N ch vs. p,leading jet in jet events: almost everything 2π Transverse Toward Φ Transverse Φ 0 Leading jet direction Φ Toward TransMIN TransMAX Away Away Third jet 0 η 1 1 Second hardest jet Holger Schulz Underlying Event measurements with first LHC data 4 / 13

15 Holger Schulz Underlying Event measurements with first LHC data 5 / 13 Disadvantages of first LHC data Jet-energy caliration not very precise in the eginning rather use tracks and lepton-id Cross-section at s = 7 TeV smaller than at 10 or 14 TeV Expect integrated luminosity of O(100 p 1 )

16 Advantages of first LHC data Measurements at s = 7 TeV give another energy point for extrapolations to 10, 14 TeV Lower luminosity means reduced pile-up L = 1033 cm 2 s 1 Holger Schulz L = 1034 cm 2 s 1 Underlying Event measurements with first LHC data 6 / 13

17 Holger Schulz Underlying Event measurements with first LHC data 7 / 13 Measurement strategy with ATLAS Use inner detector for track-p measurements + electron-id from ECAL + muon-id from muon chamers

18 Holger Schulz Underlying Event measurements with first LHC data 8 / 13 Measuring p of Z-Bosons Use only tracks from leptonic Z-decays Clean signal, look for two leptons of opposite sign within a Z-mass window dσ/dp Z p ( η < 2.5, p > 1.0 GeV) Z e + e & Z µ + µ 100 p 1 after detector cuts: events remain on generator level Statistics might e too low for a tuning p /GeV

19 Holger Schulz Underlying Event measurements with first LHC data 9 / 13 Leading track Measure track-p using only inner detector Identify leading track = largest p in event defines φ 0 Define transverse region, measure N tracks, scalar p -sum as function of p, leading track 2π Transverse Direction of leading track Φ Toward Φ Transverse Φ 0 TransMIN Toward TransMAX Away 0 η Away

20 Holger Schulz Underlying Event measurements with first LHC data 10 / 13 N ch /dηdφ Transverse region charged pt sum density Pythia p T (leading track) / GeV p track T / GeV Transverse region charged particle density Pythia p T (leading track) / GeV Plateau is a measure for Underlying Event activity Data will e taken with Minimum Bias trigger no statistics prolem

21 Does early data improve generator tuning? Tool for systematic generator tuning: Professor (arxiv: ) Professor in three lines 1 Parameterisation of generator response to shifts in parameter space 2 Add experimental data construct goodness of fit (g.o.f.) 3 Minimise g.o.f. to get est parameter setting (tuning) Question: If we add data corresponding to 50, 60, p 1 does this improve the tuning? need meaningful error-definition on tuned parameters use covariance matrix (work in progress) to get error-ands measurement worthwile, if error decreases Holger Schulz Underlying Event measurements with first LHC data 11 / 13

22 Holger Schulz Underlying Event measurements with first LHC data 12 / 13 Errorands for generator tuning Sample points from contour of hyper (error) ellipsis Run generator with these points, construct envelope Add fake Monte Carlo data see if e.g. plateau is constrained p Transverse region charged particle density LHC, 7TeV Bin-heights Bin-errors p 1 p T(leading track) / GeV

23 Holger Schulz Underlying Event measurements with first LHC data 13 / 13 Summary and Outlook UE measurements at LHC essential for proper generator tuning Need to identify reasonale oservales for first data UE as function of leading track p looks promising Proaly not enough statistics for Z-osons W-osons might e an option ATLAS CMS co-operation on Minimum Bias & UE Include UA5 data at s = 200 and 900 GeV Thank you!

24 Backup

25 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2

26 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

27 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

28 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

29 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

30 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

31 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

32 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 p

33 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 in interpolation p

34 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 data in in interpolation p

35 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 data in in interpolation p

36 Tuning procedure in Professor 1 random sampling: N parameter points in n-dimensional space 2 run generator and fill histograms 3 for each in: use N points to fit interpolation (2 nd or 3 rd order polynomial) 4 construct overall (now trivial) χ 2 (interpolation data) = 2 ins 5 and numerically minimize pyminuit, SciPy error 2 data in in interpolation est p p

37 2nd order polynomial includes lowest-order correlations etween parameters MC ( p ) f () ( p ) = α () 0 + i β () i p i + i j γ () ij p i p j Now use N generator runs, i.e. N different parameter sets x,y: v 1 v N v 2. }{{} v (N values, i.e. N in contents) = 1 x 1 y 1 x1 2 x 1 y 1 y x 2 y 2 x2 2 x 2 y 2 y x N y N xn 2 x Ny N yn 2 } {{ } P (N sampled parameter sets) α 0 β x β y γ xx γ xy γ yy }{{} c (coeffs) Therefore: c = Ĩ[ P] v where Ĩ is the pseudoinverse operator.

38 c = Ĩ[ P] v Use Singular Value Decomposition (SVD), a general diagonalisation for all normal matrices M:M = UΣV Method availale in SciPy.linalg Minimal numer of runs = numer of coefficients in c : N (n) min = 1 + n + n(n + 1)/2 + (n + 1)(n + 2)/6 }{{} cuic only Oversampling y a factor of three has proven to e much etter Num params, P N (P) 2 (2nd order) N (P) 3 (3rd order)

39 c = Ĩ[ P] v Use Singular Value Decomposition (SVD), a general diagonalisation for all normal matrices M:M = UΣV Method availale in SciPy.linalg Minimal numer of runs = numer of coefficients in c : N (n) min = 1 + n + n(n + 1)/2 + (n + 1)(n + 2)/6 }{{} cuic only Oversampling y a factor of three has proven to e much etter Num params, P N (P) 2 (2nd order) N (P) 3 (3rd order)

40 c = Ĩ[ P] v Use Singular Value Decomposition (SVD), a general diagonalisation for all normal matrices M:M = UΣV Method availale in SciPy.linalg Minimal numer of runs = numer of coefficients in c : N (n) min = 1 + n + n(n + 1)/2 + (n + 1)(n + 2)/6 }{{} cuic only Oversampling y a factor of three has proven to e much etter Num params, P N (P) 2 (2nd order) N (P) 3 (3rd order)