Les Houches SM and NLO multi-leg group: experimental introduction and charge

1. Les Houches SM and NLO multi-leg group: experimental introduction and charge G. Heinrich, J. Huston, J. Maestre, D. Maitre, R. Pittau, G. Soyez (jet liason)

2. Understanding cross sections at the LHC LO, NLO and NNLO calculations K-factors benchmark cross sections and pdf correlations PDF’s, PDF luminosities and PDF uncertainties underlying event and minimum bias events Sudakov form factors jet algorithms and jet reconstruction We’ll be dealing with all of these topics in this session, in the NLM group, in the Tools/MC group and in overlap.

3. Understanding cross sections at the LHC • We’re all looking for BSM physics at the LHC • Before we publish BSM discoveries from the early running of the LHC, we want to make sure that we measure/understand SM cross sections • detector and reconstruction algorithms operating properly • SM physics understood properly • especially the effects of higher order corrections • SM backgrounds to BSM physics correctly taken into account • This is the first Les Houches at which we have LHC data to test • in addition to a plethora of data from the Tevatron, including some mysteries

4. List of topics (from web page) • Higher order calculations and techniques->Roberto’s talk • Public computational tools/templates->(mostly) Roberto’s talk • NLO/PS matching (joint with Tools/MC WG)->(mostly) Roberto’s talk • Jetology • jet observables • boosted object tagging • connections/differences between LO, NLO and MC jet clustering for complex n-parton final states • variations in NLO multi-parton cross sections with jet algorithms, jet sizes and scales • Higgs observables • To this, I would add PDFs (also in parallel with Tools/MC group I would say)

5. We had 3 evo pre-meetings • May 6: HO computations and techniques • May 13: Jetology • May 20: Higgs observables

6. Start with the wishlist • Began in 2005, added to in 2007 and 2009 • only process 12 left among NLO • Are there other motivated needs for NLO multi-parton final states? • from dedicated calculation or automatic calculation? • one thing we promised to do last Les Houches is provide a table of the needed accuracy for each final state • Should we move on to expanding the NNLO list? • There’s also the issue of how experimentalists can use these calculations • aMC@NLO: but what is the learning curve to get to say W + 3,4 jets at NLO • ntuples more practical for immediate future, i.e. before next Les Houches?

7. Calculations • Once we have the calculations, how do we (experimentalists) use them? • If a theoretical calculation is done, but it can not be used by any experimentalists, does it make a sound? • We need public programs and/or public ntuples

8. Example: Blackhat+Sherpantuples Born loop: lc and fmlc real vsub so this is not Sherpa the parton shower, but Sherpa used as a (very efficient) fixed order matrix element generator

9. How it’s put together Born loop: lc and fmlc real vsub for W+3 jets, W+3 parton tree-level matrix elements the dipole subtraction terms evaluated in n-body phase space; to make matters more complex, vsub can be either + or -, compensated by other terms in the total cross section; note the sum over all quarks and antiquarks; makes matters more complex when coming to scale uncertainties all of the real emission terms, (W+4 partons for W + 3 jets), modified by the dipole subtraction terms; divergences are gone all of the virtual terms, both leading color and full-minus- leading color; the latter is typically a few % effect, but much of the complexity of the calculation

10. ROOT ntuples • More complex to use than MCFM • no manual for example • and you don’t produce the events yourself • my student Brian Martin and I are the beta users • ntuples produced separately by Blackhat + Sherpa for • No jet clustering has been performed; that’s up to the user • a difference from MCFM, where the program has to be re-run for each jet size/algorithm • What algorithms/jet sizes that can be run depends on how the files were generated • i.e. whether the right counter-events are present • For the files on the right at 7 TeV (for W+ + 3 jets), one can use kT, antikT, siscone (f=0.75) for jet sizes of 0.4, 0.5, 0.6 and 0.7 • bornLO (stands alone for pure LO comparisons; not to be added with other contributions below) • 20 files, 5M events/file, 780 MB/file • Born • 18 files, 5M events/file, 750 MB/file • loop-lc (leading color loop corrections) • 398 files, 100K events/file, 19 MB/file • loop-fmlc (needed for full color loop corrections) • 399 files, 15K events/file, 3 MB/file • real (real emission terms) • 169 files, 2.5 M event/file, 5 GB/file • vsub (subtraction terms) • 18 files, 10M events/file, 2.8 GB/file

11. Jet Clustering • For jet clustering, we use SpartyJet, and store the jet results in SJ ntuples • and they tend to be big since we store the results for multiple jet algorithms/sizes • Then we friend the Blackhat+Sherpantuples with the SpartyJetntuples producing analysis ntuples (histograms with cuts) for each of the event categories • Add all event category histograms together to get the plots of relevant physical observables http://projects.hepforge.org/spartyjet/ If interested, please contact Brian.thomas.martin@cern.ch

12. Logistics • So total file disk space is quite large, multi-TB (and there are many events to be processed) • I bought a 20TB disk specifically for this purpose • But they’re divided into few GB files (Blackhat+SJ) • So we can make our analysis parallel using 350 nodes at MSU • Possible to run through W + 3 jet NLO analysis in few days (much faster without the scale variations) • somewhat longer with more variations included

13. …so for example • W+ + 3 jets at 7 TeV for standard cuts (plus for electron cuts) • |ym|<2.4 • pTm>20 GeV/c • pTn> 25 GeV/c • PTjet>20 GeV/c • |yjet|<2.8 • mT(m,n)>40 GeV • New cuts or histograms means re-running through the ntuples • For antikT4 • born: 22.69 pb • loop-lc: -0.69 pb • loop-fmlc: 0.39 pb • vsub: 27.16 pb • real: -17.34 pb • Total: 32.21 pb

14. Predictions • From Blackhat+Sherpa, we have ntuples (in same format) for W + 1,2, 3,4 jets • Makes it easy to make plots for different jet multiplicities and/or combined jet multiplicities • including PDF uncertainties • including scale uncertainties • would like to explore a CKKW-like scale at NLO at Les Houches • examining dependence on jet size/algorithm

15. Lead jet pT

16. Scale dependence • Factorization and renormalization scale dependence for any cross section can be calculated (relatively easily) independent of the evaluation of the full matrix element, if you’re careful to collect the relevant terms • In new version of Blackhat+Sherpantuples, they were careful to collect the relevant terms

17. Reweighting can reweight each event to new -PDF -factorization scale -renormalization scale -as (tied to the relevant PDFs) based on weights stored in ntuple (and linking with LHAPDF) so, for example, the events were generated with CTEQ6, and were re-weighted to CTEQ6.6

18. Reweighting, cont. complex: carry both single and double logs 9 we run into the sum over quarks and antiquarks again

19. PDF Errors Better than what is done in MCFM (as far as disk space is concerned); PDF errors are generated on-the-fly through calls to LHAPDF. But then don’t store information for individual eigenvectors.

20. Example scale/PDF uncertainty …calculated using ntuples LO at this point for 4 jets

21. LO/NLO predictions for jet cross sections • Don’t believe (fixed) LO predictions for jet cross sections • Let’s look at predictions for W+ + 3 jets for two different jet algorithms as a function of jet size at the LHC (7 TeV) • At LO, both antikT and SISCone show a marked decrease in cross section as the jet size increases • because of the log(1/DR) terms • But at NLO, the two cross sections show little dependence on the jet size, and are similar to each other • due to addition of extra gluon in jet possible at NLO • You’ll see the same thing in ATLAS Monte Carlo Blackhat + Sherpa note NLO~LO because a scale of HT has been used; if a scale like mW2+pTW2 is used K-factor <<1

22. Predictions for jet cross sections Compare to ATLAS ALPGEN+ PYTHIA samples for jet sizes of 0.7 At parton level, antikT is ~25% higher than SISCone (same as we observe here at LO) At topocluster level, antikT is ~2% higher than SISCone (not the 7% observed here) Why 2%, not 7%? Some of the W + 3 parton events reconstructed as 2 jets at the parton level for SISCone are reconstructed as 3 jets at the hadron. The cross section for 3 jets increases.

23. Try this out in ATLAS/CMS Monte Carlo • Take W + 2 parton events (ALPGEN+PYTHIA), run SISCone 0.7 algorithm on parton level, hadron level (not shown) and topocluster level • Plot the probability for the two sub-jets to merge as a function of the separation of the original two partons in DR • Color code: • red: high probability for merging • blue: low probability for merging • everything for DR<0.7 is merged for SISCone (and antikT) • Parton level reconstruction agrees with naïve expectation • Topocluster level reconstruction agrees with need for Rsep • I’d like to come to some resolution/better understanding on this issue at Les Houches, using a standardized file of W + jets events

24. Choosing jet size • Experimentally • in complex final states, such as W + n jets, it is useful to have jet sizes smaller so as to be able to resolve the n jet structure • this can also reduce the impact of pileup/underlying event • Theoretically • hadronization effects become larger as R decreases • for small R, the ln R perturbative terms referred to previously can become noticeable • this restriction in the gluon phase space can affect the scale dependence, i.e. the scale uncertainty for an n-jet final state can depend on the jet size, • …to be investigated Another motivation for the use of multiple jet algorithms/parameters in LHC analyses. Can we explore this further?

25. Jet sizes and scale uncertainties: the Goldilocks theorm • Take inclusive jet production at the LHC for transverse momenta of the order of 50 GeV • Look at the theory uncertainty due to scale dependence as a function of jet size • It appears to be a minimum for cone sizes of the order of 0.7 • i.e. if you use a cone size of 0.4, there are residual un-cancelled virtual effects • if you use a cone size of 1.0, you are adding too much tree level information with its intrinsically larger scale uncertainty • This effect becomes smaller for jet pT values on the order of 100 GeV/c • how does it translate for multi-parton final states? • …good subject for investigation here

26. Scale choices • Take inclusive jet production at the LHC • Canonical scale choice is mr=mf=1.0*pT • Close to saddle point for low pT • But saddle point moves down for higher pT • Can we think about recommendations for scale choices (and ranges) for the LHC? • I know there is worry about typical scale choices that can lead to negative cross sections, for example at very forward rapidities • Rather than look for some magic formula, we should try to understand what is going on the kinematic/scale point-of-view R=0.4 antikT

27. Scale dependence also depends on jet size R=0.4 antikT R=0.6 antikT

28. One scheme • F. Olness and D. Soper, arXiv:0907.5052 • Define x1 and x2 • Make a circle of radius |x|=2 around a central scale (could be saddle point, or could be some canonical scale) and evaluate the scale uncertainty col Fred is here, so maybe we can explore this further, comparing to the LHC data AJ and MJK carry information on the scale dependence beyond NLO

29. Another scheme Higgs Cross Section Working Group arXiv:1101.0593

30. Scale dependence: jet algorithms • Look at results for SISCone/antikT; antikT cross sections larger than SISCone, smaller scale dependence? H. Ita, SLAC Hadronic Final State Forum

31. Z + 3 jets: scale dependence Note that peak cross sections are actually quite close; the cross sections just peak at different scales. 1004.1659 Can we understand/quantify this better? For LHC cross sections.

32. Scales: CKKW and NLO • Applying a CKKW-like scale at LO also leads to better agreement for shapes of kinematic distributions • (Partially) investigated at last Les Houches; needs more work at this Les Houches See review of W + 3 jets in Les Houches 2009 NLM proceedings 0910.3671 Melnikov, Zanderighi

33. Jet vetos • For some cross sections, the scale dependence improves with a jet veto, and in others the scale dependence worsens • I think it would be worthwhile to collect this information • And of course, these conclusions are drawn from using fixed order predictions only WWjet tTbB

34. Uncertainties for Higgs production with jet binning …large logs result from jet vetoing naïve scale variation may provide too small an estimate of scale uncertainty have to resum these logs; can re-weight MC@NLO or Powheg using this information maybe we can generalize to other processes at the LHC F. Tackmann May 20 evo

35. CDF Wjj • Potentially an important discovery, but are current tools capable of modelling the W + jets background precisely enough • Session on Saturday afternoon • You know that it’s important when it makes it to prime-time TV

37. …if you paid close attention CDF Wjj analysis cuts

38. LHC jets • ATLAS and CMS are both using an IR-safe jet algorithm (anti-kT) • Unfortunately no common sizes • 0.4 and 0.6 for ATLAS • 0.5 and 0.7 for CMS • It would be nice to • have at least one common jet size • exploit any capability to perform analyses with multiple jet sizes/algorithms • ATLAS topoclusters have the potential to allow for more flexibility in jet analyses • Should be similar potential in CMS with particle flow, etc

39. UE/pileup corrections: Jet areas determined by clustering ghost particles of vanishing energy; see jet references note that the kT algorithm has the largest jet areas, SISCone the smallest and anti-kT the most regular; one of the reasons we like the antikt

40. Jets: area-based correction: Cacciari/Salam/Soyez Used by both ATLAS and CMS. Can we understand what works/what needs improvement in the light of LHC data with significant pileup?

41. Aside: Photon isolation at the LHC • From a theoretical perspective, it’s best to apply a Frixione-style isolation criterion, in which the amount of energy allowed depends on the distance from the photon; this has the advantage of removing the fragmentation contribution for photon production, as well as discriminating against backgrounds from jet fragmentation • But most of the energy in an isolation cone is from underlying event/pileup • At Les Houches, we started to develop (being continued by Mike Hance, Brian,…in ATLAS): • (1) an implementation of the Frixione isolation appropriate for segmented calorimeters • (2) a hybrid technique that separates the UE/pileup energy from fragmentation contributions using the jet density approach more development at this Les Houches?

42. Jets at parton level and in (NLO) MC …from Jet Pair Production in Powheg, arXiv:1012.3380 note that theory/data has a slope not evident with fixed order comparisons (NLO corrected by UE/hadronization) also observed in ATLAS comparisons; differences observed when using Pythia as shower instead of Herwig an effect we need to understand; this will affect all global PDF fits, for example; Les Houches is a good place to do it

43. PDFs • We’ve learned a lot from the PDF4LHC exercises • In particular, we’ve seen where the PDFs agree and where they don’t • The exercise was at NLO; now we are in a position to continue it at NNLO Plots by G. Watt

44. …as well as to start adding LHC data …and already seeing differences between Experiments Note that resummed predictions are important

45. PDF correlations • Consider a cross section X(a), a function of the Hessian eigenvectors • ith component of gradient of X is • Now take 2 cross sections X and Y • or one or both can be pdf’s • Consider the projection of gradients of X and Y onto a circle of radius 1 in the plane of the gradients in the parton parameter space • The circle maps onto an ellipse in the XY plane • The angle f between the gradients of X and Y is given by • The ellipse itself is given by • If two cross sections are very • correlated, then cosf~1 • …uncorrelated, then cosf~0 • …anti-correlated, then cosf~-1

46. …from PDF4LHC report

47. Correlations, continued… one interesting angle to calculate is the angle between the gradient for a particular physics process and the hyperplane formed by the first n eigenvectors take gg->Higgs (120 GeV) eigenvector cosf =1 0.028 <=2 0.077 <=3 0.077 <=4 0.534 <=5 0.551 <=6 0.553 <=7 0.602 <=8 0.604 <=9 0.609 <=10 0.808 <=11 0.808 so very strong correlation (0.8) between the Higgs cross section and the hyperplane formed by the first 11 (of 22) eigenvectors in CTEQ6.6 low number eigenvectors have quadratic c2 behavior

48. Being used by Higgs combination groups • Can we extend this use?

49. Summary • Due to lack of time, haven’t mentioned boosted jets/analyses, but clearly this is an important aspect of this workshop • some people are coming straight from BOOST2011 • There are a lot of interesting physics topics at this Les Houches, as well as LHC data (for the first time) and greatly improved NLO technology • It should be an interesting week and a half