Jonathan Butterworth MC4LHC/MCnet/HEPTOOLS/Artemis training event 5 th & 6 th Jan 20 09

1. Experimental Techniques and their Impact on Phenomenological Analyses Part 1 Jonathan Butterworth MC4LHC/MCnet/HEPTOOLS/Artemis training event 5th & 6thJan 2009 Thanks to K.McFarland, E. Nurse, P.Renton, F. Siegert, R.Thorne…

2. Outline • Introduction • Fundamental or observable? • Corrections, corrections • Jets and Jet Algorithms • Leptons and photons • Inclusive or exclusive • Monte Carlos and tuning • Inconclusion

3. Introduction • The aim of these lectures is to discuss how choices which are made by experimentalists have an impact on the usefulness of the data for phenomenological analysis. • Seemingly arbitrary choices can sometimes dramatically affect the shelf-life and impact of a measurement!

4. Fundamental or Observable “Fundamental” things (top, W, H masses, couplings etc) are often only defined within the theory, and so are often rather model dependent, whereas “observable” things (proton mass, charged particle multiplicity, inclusive lepton cross section etc) are well defined but difficult to interpret.

5. Pete Renton Jul/ Aug 2008 Precision (pseudo-)observables Z lineshape (5) tau polarisation - Ae(1) left-right asymm -Ae (1) Z (b,c) properties (6) Total of 17 at high Q2 (assuming lepton universality) from > 1000 measurements with (correlated) uncertainties W properties (2) top quark mass (1)

6. ICHEP08 Pete Renton Jul/ Aug 2008 progress vs. time Pete Renton 2008 2008 direct mt and mw 2008 indirect mt and mw 1995 indirect mt and mw 1995 direct mt and mw

7. Fundamental or Observable

8. Fundamental or Observable • Note that H1 and ZEUS do not even plot and fit F2 here; • The reduced cross section is an obserable (modulo electroweak corrections), F2 is not.

9. Fundamental or Observable • Parton densities are transportable to other experiments within the SM (QCD, factorization). • Good example of experiments making measurements and doing the phenomenological analysis themselves, but also making the data available to others (MSTW, CTEQ, NNPDF etc…)

10. Fundamental or Observable • So the biggest headline derived from a measurement is not the same thing as the measurement itself • It is often not even really a measurement. • Ask yourself: if • (e.g.) the LHC proves that the SM is wrong in some possibly bizarre way (no Higgs in loops, lots of KK particles in loops, low mass gluino or gravitino) • Or some theorist makes a much better calculation of my process how does a given result need to be modified? • For a real measurement, the answer should be “it doesn’t” or at least “it doesn’t beyond the systematic uncertainties”.

11. Corrections, corrections • Many corrections are or may be applied on the way to publishing a measurement; • Dead time, trigger acceptance, detector geometry, energy scale, dead material, pile-up, electroweak radiative corrections, FL, “underlying event”, “hadronisation”… • If a correction requires intimate knowledge of the detector, then apply it. • After our experiment finishes, no-one is going to be able to (or want to!) correct for our trigger efficiencies, energy scale and resolution etc

12. Corrections, corrections • If a correction requires intimate knowledge of a developing theory, or a Monte Carlo model of some poorly-known physics, do not apply it!* • After your experiment finishes, if someone does a NNLO calculation, or really understands underlying events, hopefully your data remain useful! • There’s obviously a sliding scale, and a grey area… • Dead time, trigger acceptance, detector geometry, energy scale, dead material, pile-up, electroweak radiative corrections, FL, “underlying event”, “hadronisation”… *or at very least, present the uncorrected version first!

13. Corrections, corrections • If a correction requires intimate knowledge of a developing theory, or a Monte Carlo model of some poorly-known physics, do not apply it! • After your experiment finishes, if someone does a NNLO calculation, or really understands underlying events, hopefully your data remain useful! • There’s obviously a sliding scale, and a grey area*… • Dead time, trigger acceptance, detector geometry, energy scale, dead material, pile-up, electroweak radiative corrections, FL, “underlying event”, “hadronisation”… *although personally I don’t think it’s very big!

14. Corrections, corrections • So it is best to haveboth (a la HERA DIS: cross sections and PDFs), but if you are only going to have one it must be the observable! • Few, if any, measurements are completely model-independent. • There will probably be some model dependence in our detector corrections (e.g different hadronisation models affect how we understand calorimeter response) • Minimise it and include in systematic error bars.

15. Corrections, corrections • An example of something in the grey area • W-asymmetry or lepton asymmetry at Tevatron? • next slides slides from Kevin McFarland (Rochester) DIS08

26. Corrections, corrections • W-asymmetry or lepton asymmetry at Tevatron? • Previous slides slides from Kevin McFarland (Rochester) DIS08 • W asymmetry has a more direct relationship to the distribution of particular partons • And it contains real extra information (from the missing pT) • BUT sensitivity to sea quarks is lost and replaced by a systematic error • Easier for PDF fitters to fit the lepton asymmetry. • Similar issues at LHC (less severe if one assumes anti-u and anti-d converge at low x.

27. Jets and Jet Algorithms • What is a jet? • Some requirements on jet definitions • Available algorithms

28. What is a Jet? • Protons are made up of quarks and gluons. • Quarks and gluons are coloured and confined – we only ever see hadrons. • A jet of hadrons is the signature of a quark or gluon in the final state. • The gross properties (energy, momentum) reflect the properties of the quark or gluon, and stand out above the rest of the event. • Jets have a complex substructure. Picture from Rick Field

29. What is a Jet? • Evolution from a hard parton to a jet of partons takes place in a regime where: • Energy scale is high enough to use perturbation theory • x is not very small • Collinear logarithms are large • Multiplicities can be large • This is largely understood QCD, and can be calculated • Hadronisation(non-perturbative) stage has a small effect (sub-GeV level) and is well modelled by tuned Monte Carlo simulation (e.g. Lund string)

30. What is a Jet? • Jets are not just less-well-measured leptons or “smeared” partons. • Hard radiation interference at amplitude level • Matching at high scales with Matrix element • Matching at low scales with parton densities and hadronisation model • potentially useful information in the internal jet structure, and in particle/energy flow between the jets • Jets have no existence independent of the algorithm • even if the “algorithm” = event display + physicist

31. What is a Jet? • So jet algorithms don’t so much find a pre-existing jet as define one. • A “jet” (or a pattern of jets) is a complex QCD event shape, designed to reflect as closely as possible the short distance degrees of freedom (quarks, gluons, H, Z, W…) • The degrees of freedom themselves are generally not physical observables, but can only be extracted within some theory or model • The cross section for quark production in the final state at LHC is zero (unless we find something very exciting…)

32. What is a Jet? (continued) • We want to connect what we measure back to the fundamental degrees of freedom of our (Standard) model, so we can publish for example: • Parton densities • Top, W, (Higgs?) masses • Deviations from the Standard Model (or limits derived from their absence) • (Usually) requires comparison to state-of-the-art theory, so our jet finding procedure had better be something the theory/model can replicate • Clear separation between detector corrections (model independent) and interpretation (model dependent). • “Truth” is algorithm+final state, not the calorimeter and not “partons”

33. Example (from my own experience) • Jet photoproduction at HERA • Was a completely new energy regime: Photoproduction of jets never before observed (some analogy to what we are about to embark on) • ZEUS tried kTcluster, plus two implementations of “Snowmass” cone algorithms • One based on a sliding window, one based on a seed. Both respectable at the time (the seeded one was based on the CDF algorithm of the time) • NLO (Klasen, Kramer) drew cones around partons (max three per event) • Two partons separated by R ~ 2 may still be in a single cone of R=1, but may not be found if there is no seed in between them. • Introduce Rsep, the maximum separation allowed between two partons (Ellis, Kunszt & Soper 1992).

34. Example • Rsep has a large effect on the NLO theory (left) • This has no analogy in a final state jet algorithm. • No common jet definition possible. • Difference between cone algorithms has large effect on the experiment (right) • NLO theory cannot use either of ZEUS’s IR unsafe cone algorithms From JMB, L.Feld, G.Kramer, M.Klasen, hep-ph/9608481

35. Example • Neither theory and experiment knows how reproduce the other • Actually they can be made to agree by tweaking RSEP • Predictive power of NLO QCD practically gone • Not a problem if both use the same algorithm • in the ZEUS case we went with kT From JMB, L.Feld, G.Kramer, M.Klasen, hep-ph/9608481

36. What is a Jet? • One good place to look: Les Houches 2007 accords (arXiv:0803.0678) • Jet definition specifies all details of the procedure by which an arbitrary set of four-momenta is mapped into a set of jets. Composed of: • Jet algorithm (e.g inclusive, longitudinally boost-invariant kT). • All the parameters of the jet algorithm (e.g. R=0.7) • The recombination scheme (e.g four-vector recombination, or E-scheme, or…)

37. What is a Jet? • One good place to look: Les Houches 2007 accords (arXiv:0803.0678) • Final-statetruth-level specification • What is the input (at the truth level). e.g. all final state particles with lifetimes longer than some cut… • Theory and experiment can correct to this level, within some controlled systematic error.

38. Jets for different applications • If you are doing simple searches at high pT, jets are obvious • …maybe • Certainly even the “event display + physicist” algorithm should be pretty reproducible for counting ~1 TeV jets at LHC • (really? How precise is that 1 TeV cut?) • If you are making precision measurements of jet cross sections (energy scale ~1%) then you will need to compare at least to NLO QCD. • In this case it is mandatory to have a jet definition which can also be applied to NLO QCD • Consequence: This means a detector-independent, infrared & collinear safe algorithm

39. Infrared Safety? • Adding an arbitrarily soft gluon to the event should not change the jets. • Non-infrared-safe jet algorithms cannot be used in higher order calculations. • so if we use them in the experiment we cannot compare to the best theory (at least without making some intermediate model-dependent correction). • Actually, infrared instabilities undermine the claim of a jet algorithm to be telling us about the short distance physics. • We should also worry about sensitivity to arbitrarily low noise, or arbitrarily soft pions, for example. • An example of an infrared instability is the “seed” in old cone algorithm. But there are others (e.g. in the splitting/merging)

40. Precision Application • ZEUS Jet measurements • 1% energy scale, kT algorithm • Compared to NLO QCD, used in NLO PDF fits Extrapolation: A. Cooper-Sarkar, C.Gwenlan, C.Targett-Adams, HERA-LHC Workshop, hep-ph/0509220

41. Jets for different applications • Our “hard” events will be complicated. Many hard jets, minijetvetos and rapidity gaps. Several different hard scales in the event (pT, MW, Mt… Mplanck???) • This is when QCD background calculations get really hard, and sensitivity to soft effects and higher orders becomes critical. • Clearly we will take as much background information as we can from the data, but equally clearly there will still be a need for stable jet definitions, and for some modelling and comparison with theory • e.g. extrapolating the W pT from the Z pT at the Tevatron • Consequence: Reiterates need for a reproducible & stable definition • Consequence: Our jet algorithms must be pretty fast even for very high multiplicities

42. Jets for different applications • Our the environment our “hard” events are in will be complicated. • Underlying event and pile-up, not to mention experimental noise & efficiency • Consequence: Our jet algorithms need to be good at picking out the “hard” physics. • Not just “reproducible” underlying event corrections, but small ones

43. Jets for different applications • Our high pT jets will sometimes have stuff in them (High pT top, W, Z, Higgs, SUSY particles…) • Consequence: Useful to have an algorithm which can be “unpicked”, has a recombination scheme which conserves energy & momentum, and is stable against soft physics.

44. Available Algorithms • “Cone” algorithms • Generally seek to find geometric regions which maximise the momentum in a given region (often not actually a cone!) • Sort of mimics the “event display + eye” method • “Cluster” algorithms • Generally start from the smallest objects available, and perform an iterative pair-wise clustering to build larger objects (using either geometric or kinematic properties of the objects) • Sort of inverts the QCD parton shower idea

45. Recombination Schemes • In principle can choose how to calculate the momentum of a (pseudo) jet from its constituents • In practice, two main possibilities used: • Snowmass/ET scheme • ET = S(ETi) ; h = S(Etihi)/S(ET) • Forces massless jets from massless particles (pseudorapidity=rapidity, ET = pT) • Four-vector addition • Preserves mass information, although normally starts with massless objects • Care needed (pT or ET, h or y?)

46. Cone algorithms • “Simple” (snowmass) cone algorithm not fully specified • Many subtly different algorithms go under the name “cone algorithm” (with or without the Snowmass badge). • Often re-implemented and changed within each experiment, which is bad for inter-experiment comparison and for reproducibility. • Solved by using an external package, e.g. fastJet (Cacciari & Salam) • Iterative • Find a cone, • Calculate the centroid • Redraw cone around the new centre • Recalculate centroid • Etc (look for stable solutions)

47. Cone algorithms • If they rely on a seed to start cone finding, and many older ones did, they are not infrared safe. • Using midpoints of seeds as well helps, but only postpones the problem to the next highest order. • Seedless cone algorithms (e.g. SiSCone, Salam & Soyez) now available. • “Progressive Removal” type (as current in CMS) • start with the highest momentum seed (not colinear or IR safe!) • Find a stable cone and remove it • Continue • Split/merge (as is the current “ATLAS cone”) • Find all stable cones. • If two stable cones overlap, merge if the overlap is > X (~ 50-75%) . Otherwise assign energy to the higher pT one.

48. Cone algorithms • Geometrical localisation is appealing when mapping on to a real detector (which is of course geometrically localised!) • Although note, the overlap issue means that the “cone” is not in fact regular, especially in high multiplicity events. • (Can be very large, in fact, depending on the split/merge treatment) • But its edges are fairly sharp. • SiSCone does now provide a practical (i.e. fast enough) infrared safe & seedless cone algorithm in public code • Used a sliding window idea to find all possible stable cones • Has a split-merge step • arXiv:0704.0292[hep-ph], Salam & Soyez

49. Cluster algorithms • Again, many possibilities • Jade, kT, Cambridge/Aachen, Anti-kT, … • Each has a distance measure, and merges the “closest” objects by this measure until some criteria is reached (could be a specified multiplicity, or “distance”) • Modern ones (kT, Cambridge, anti-kT) belong to a general class where the distance parameter is given as • p=1 for kT, 0 for Cam/Aachen, -1 for anti-kT • Can be very slow at high multiplicities • recently solved infastJet (Cacciari & Salam, Phys.Lett.B641:57-61,2006) • No seed needed • All objects assigned uniquely to one jet (no separate split/merge stage)

50. kT algorithm • Catani et al Phys Lett B269 (1991); Nucl. Phys. B406 (1993); Ellis and Soper Phys Rev D48 (1993). • p=1 • Successively merge objects with low relative kT • If the kT2 of an object w.r.t the beam is lower than kT2w.r.t anything else in the event divided by R2, don’t merge any more; call it a jet. • Mimics (inverts) the QCD parton shower. • Soft stuff merged into the nearest hard stuff. • Can undo merging (Y splitter). Last merge is the hardest.