1 / 68

New (BSM) Physics Searches with Single Jets: Jet Substructure and Jet Grooming

Steve Ellis (See also Jesse Thaler last week). New (BSM) Physics Searches with Single Jets: Jet Substructure and Jet Grooming. LPC @ Fermilab 11.18.10. Background/Assumptions:. LHC is intended to find new “stuff” (BSM physics) Hadronic “stuff” will be organized into jets

kin
Télécharger la présentation

New (BSM) Physics Searches with Single Jets: Jet Substructure and Jet Grooming

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Steve Ellis (See also Jesse Thaler last week) New (BSM) Physics Searches with Single Jets: Jet Substructure and Jet Grooming LPC @ Fermilab 11.18.10

  2. Background/Assumptions: • LHC is intended to find new “stuff” (BSM physics) • Hadronic “stuff” will be organized into jets • At 14 (7) TeV many interesting particles (t, W, Z, Susy, ..) will be boosted enough to be in a single jet  Want to use single jets in search for BSM physics • Want to distinguish jets with decays from QCD jets • Want to use jet substructure for this purpose LPC @ Fermilab S.D. Ellis 11/18/10

  3. Outline & Issues • Brief review of (QCD) jets - defined by algorithms (no intrinsic definition) - jets have substructure, including masses (not just 1 parton, 1 jet) • Typically use Recombination (kT) jets  natural substructure, but also algorithm systematics (shaping of distributions) contributions from (uncorrelated) ISR, FSR, UE and Pile-up • But can use Cone or Anti-kT as jet finder Warning! LPC @ Fermilab S.D. Ellis 11/18/10

  4. Outline & Issues (cont’d) • Search for BSM physics in SINGLE jets at the LHC by separating from QCD; want generic techniques, which are robust under systematic effects– Consider a variety of jet substructure measures: 1) masses (simply look for bumps) 2) angularities (single jet version of thrust-like) distribution 3) subjets found with jet algorithm with smaller size parameter, e.g., “taggers” top q, Higgs, or “groomers” that delete all but the leading subjets LPC @ Fermilab S.D. Ellis 11/18/10

  5. Outline & Issues (cont’d) • QCD yields a large but Smooth (limited structure) QCD background • But substructure (mass bumps) in signal can degraded by • algorithm systematics • uncorrelated UE and Pile-Up contributions • Need to “clean-up” or “Groom” the jets, e.g., PRUNING - remove large angle, soft branchings • Validate with studies of boosted top q’s, W’s, Z’s at Tevatron and LHC LPC @ Fermilab S.D. Ellis 11/18/10

  6. Jet Substructure – a new tool for the LHC! Two general approaches – 1) Taggers that use specific decay properties, e.g., top quarks, W/Zs, Higgs, … to predict jet substructure (subjets) 2) More generic jet grooming to let the underlying structure, e.g., mass, be manifest (even when don’t know what it is, i.e., searches) and reduce impact of UE, PU and algorithm details See the overview in the (immediately) forthcoming Proceedings of Boost 2010! LPC @ Fermilab S.D. Ellis 11/18/10

  7. Defining Jets • Map the observed (hadronic) final states onto the (short-distance) partons by summing up all the approximately collinear stuff, ideally on an event-by-event basis. • Need rules for summing  jet algorithm Start with list of particles/towers End with list of jets (and stuff not in jets) • E.g., • Cone Algorithms, based on fixed geometry – focus on core of jetSimple, “well” suited to hadron colliders with Underlying Events (UE),but found jets can/do overlap • Recombination (or kT) Algorithm, based on pairwisemerging to undo shower Tends to “vacuum up” soft particles, “well” suited to e+e- colliders LPC @ Fermilab S.D. Ellis 11/18/10

  8. Recombination Algorithm – focus on undoing the shower pairwise,  Natural definition of substructure Merge partons, particles or towers pairwise based on “closeness” defined by minimum value ofkT, i.e. make list of metric values(rapidity y and azimuth , pTtransverse to beam) If kT,(ij)is the minimum, merge pair (add 4-vectors), replace pair with sum in list and redo list; IfkT,iis the minimum →i is a jet! (no more merging for i, it is isolated by D), 1 angular size parameterD (NLO, equals Cone for D = R, Rsep = 1), plus = 1, ordinary kT, recombinesoft stuff first = 0, Cambridge/Aachen (CA),controlled by angles only  = -1, Anti-kT, justrecombinestuff around hard guys – cone-like (with seeds) LPC @ Fermilab S.D. Ellis 11/18/10

  9. Recombination Algorithm– in action, here CA algorithm on QCD jet Think of starting with calorimeter cells, recombine “closest” pair at each step leading to larger pT For CA close in quantity (0.05 x 0.05) Cells with E > 1 GeV low pTtohigh pT LPC @ Fermilab S.D. Ellis 11/18/10

  10. Note: the details of the substructure (at each step) depend on the algorithm kT(pT and angle) CA (just angle) LPC @ Fermilab S.D. Ellis 11/18/10

  11. Jet Masses in QCD: A Brief Review • In NLO PertThy Phase space from pdfs, f ~ 1 & const Dimensions Jet Size, D = R ~ , determined by jet algorithm Peaked at low mass(log(m)/m behavior),cuts off for (M/P)2 > 0.25 ~ D2/4 (M/P > 0.5) large mass can’t fit in fixed size jet, QCD suppressed for M/P > 0.3 (~  < 3) Want heavy particle boosted enough to be in a jet (use large-ish D ~1), but not so much to be QCD like (~ 2 <  < 5) Soft – Collinear pole version Useful QCD “Rule-of-Thumb” LPC @ Fermilab S.D. Ellis 11/18/10

  12. Jet Mass in PYTHIA (matched set)D = 1, 500 GeV/c < pT < 700 GeV/c Algorithm matters Tails depends (somewhat) on being matched set Turns over LPC @ Fermilab S.D. Ellis 11/18/10

  13. Jet Mass – CDF Data (CDF/PUB/JET/PUBLIC/10199 7/19/10) At least qualitatively the expected shape – masses slightly larger than MC – need the true hard emissions (as in matched sets) Large mass tail grows, as expected, with jet size parameter in the algorithm -You find what you look for! LPC @ Fermilab S.D. Ellis 11/18/10

  14. Angularity – a jet shape measure(introduced by G. Sterman and collaborators) • (Full) Event shapes in e+e- - (from early days) First define Thrust direction and value viawrt the Thrust direction define angularity with parameter aNote for a = 0 just (1-T). For massless particles LPC @ Fermilab S.D. Ellis 11/18/10

  15. Angularity – Apply to single jetFind jet components with algorithm, jet direction replaces Thrust direction • Single Jet Angularity – some choice involved, e.g., in CDF note Public/10199 (following Sterman, et al.) where this is applied to jets binned in both pT,J (EJ) and mJ This results in a limited range for general a LPC @ Fermilab S.D. Ellis 11/18/10

  16. CDF data Cone, R = 0.4 Cone, R = 0.7 Theory = minimal double pole expression LPC @ Fermilab S.D. Ellis 11/18/10

  17. Angularity – Apply to single jet - IIFind jet components with algorithm, jet direction replaces Thrust direction • Single Jet Angularity – a different choice, Q 2E(SDE, Hornig, Lee, Vermilion & Walsh, 1001.0014, applied with SCET in e+e-) now applied to jets binned only in pT,J (EJ) and we have ie., factor of m/2E compared to earlier. Also A narrow jet Recall earlier LPC @ Fermilab S.D. Ellis 11/18/10

  18. SCET Results – NLO + NLL (in global logs) no hadronization General features: Damped at small  (Sudakov) Peak and then falls off Shoulder at large  Distributions: Broader as R (~D) increases Broader as a increases Broader for gluons LPC @ Fermilab S.D. Ellis 11/18/10

  19. More SCET: Compare to MC, look at peak Distribution qualitatively like MC, but hadronization is relevant (broader) Peak moves with R and a LPC @ Fermilab S.D. Ellis 11/18/10

  20. The Future for Angularities • Data – plot for other definition either without mass binning or in multiple mass bins (even with top contamination) – test for systematics noted above • Theory – evaluate the double distribution -3 Pythia Version With a = -3 0 LPC @ Fermilab S.D. Ellis 11/18/10

  21. Finding Heavy Particles with Jets - Issues • QCD multijet production rate >> production rate for heavy particles • In the jet mass spectrum, production of non-QCD jets may appear as local excesses (bumps!) but must be enhanced using analyses • Use jet substructure as defined by recombination algorithms to refine jets •  Algorithm will systematically shape distributions • Use top quark as surrogate new particle. σttbar≈ 10-3σjj ttbar QCD dijet shaped by the jet algorithm arb. units arb. units falling, no intrinsic large mass scale LPC @ Fermilab S.D. Ellis 11/18/10 mJ (GeV/c2) mJ (GeV/c2)

  22. Reconstruction of Jet Substructure – QCD vs Heavy Particle • Want to identify a heavy particle reconstructed in a single jet via substructure • Need correct ordering in the substructure and accurate reconstruction (to obtain masses accurately) • Need to understand how decays and QCD differ in their expected substructure, e.g., distributions at branchings.  But jet substructure affected by the systematics of the algorithm, and by kinematics when jet masses/subjet masses are fixed. The algorithm metric affects the substructure - introduces bias interpret last recombinations as a heavy particle or QCD jet LPC @ Fermilab S.D. Ellis 11/18/10

  23. Systematics of the Jet Algorithm • Consider generic recombination step: i,j➜p • Useful variables: (Lab frame) • Merging metrics: • Daughter masses (scaled by jet mass) : • In terms of z, R, the algorithms will give different kinematic distributions: • CA orders only in R: z is unconstrained • kTorders in z· R: z and R are both regulated • The metrics of kT and CA will shape the jet substructure. LPC @ Fermilab S.D. Ellis 11/18/10

  24. 2 is softer Phase Space for 1→2 a1= 0, a2= 0 a1= 0.46, a2= 0 • The allowed phase space in R, zfor a fixed  (mJ and pT) is nearly one-dimensional • QCD and decays will weight the phase space differently • Cutoffs on variables set by the kinematics, not the dynamics • Sample of phase space slices for different subjet masses, 1 is softer a1= 0.9, a2= 0 a1= 0.3, a2= 0.1 1 never softer LPC @ Fermilab S.D. Ellis 11/18/10

  25. 1→2 Decay in a Jet • Goal is to identify jets reconstructing a heavy particle and separate them from QCD jets • Consider a 1→2 decay (J 1,2) reconstructed in a jet, massless daughters (a1= a2= 0) for now • Requirement to be in a jet: R12 < D- algorithm independent • Look at the decay in terms of the algorithm variables Large D is needed to reconstruct jets with a lower boost - use D = 1.0,sweet spot  ~ 3, pT ~3 mJ, xJ ~ 0.1 Recall for QCD jet pT ~5 <mJ > LPC @ Fermilab S.D. Ellis 11/18/10

  26. 1→2 Decay in a Jet (unpolarized) No enhancement at the lower limit in - unlike QCD Enhancement at the lower limit for - like QCD Decays not reconstructed: small , large J rest frame lab frame J (boost to the lab) J 2 2 1 beam direction LPC @ Fermilab S.D. Ellis 11/18/10 1

  27. 1→2 Decay in a Jet a1 = 0, a2 = 0 Cutoffs are set by the kinematics - same between QCD and decay with fixed Enhancement at the lower limit for - like QCD No enhancement at the lower limit in - unlike QCD LPC @ Fermilab S.D. Ellis 11/18/10

  28. QCD Splittings Take a leading-log approximation of QCD: For small angles - good approximation for a splitting in a jet: This lets us fix xJ (or  ). Distribution in xJ : LPC @ Fermilab S.D. Ellis 11/18/10

  29. QCD Splittings: R12 and z a1 = 0, a2 = 0 Fix  (xJ), find distributions in R12 and z Limits set by the kinematics QCD will have many more soft (small z) splittings than decays do - QCD splittings are small z, small xJenhanced Enhancement at the lower limit in - like decays Enhancement at the lower limit in z- unlike decays LPC @ Fermilab S.D. Ellis 11/18/10

  30. Summary of Dynamics of QCD Vs Decays: • Distributions in R very similar (for fixed boost) • QCD enhanced at small z, xJ • Will these be represented in the last recombinations of a jet? LPC @ Fermilab S.D. Ellis 11/18/10

  31. Effects of the Jet Algorithm – Algorithm Bias • Recombination metrics: • Recombinations arealmost always monotonic in the metric • The algorithm cuts out phase space in (z, Rij) as it proceeds • Certain decays will be reconstructed earlier in the algorithm, or not at all kT CA pTp dependent boundaries late late intermediate intermediate early early LPC @ Fermilab S.D. Ellis 11/18/10

  32. Typical Recombinations • Late recombinations are set by the available phase space • For CA, Rmust be near D, and the phase space tends to create small zrecombinations • For kT, z Rwill be larger, with a pT dependent cut • The soft (small z) radiation is recombined earlier in kT, meaning it is harder to identify - leads to poorer mass resolution Matched QCD sample (2, 3, 4 partons) from MadGraph/Pythia, jet pT between 500-700 GeV last recombination last recombination LPC @ Fermilab S.D. Ellis 11/18/10

  33. Matched QCD, jet pT between 500-700 GeV Comparing CA and kT: • Final recombinations for CA not QCD-like - No enhancement at small R • Final recombinations for kT more QCD-like - Enhanced at small zandR • kT has poorer mass resolution - Soft objects recombined early in algorithm - more merged tt sample from MadGraph/Pythia jet pT between 500-700 GeV - LPC @ Fermilab S.D. Ellis 11/18/10

  34. Summary: Identifying Reconstructed Decays in Jets • Reconstruction of a decay can be hidden in the substructure • Small z recombination unlikely to accurately give decay • Small z recombinations also arise from UE and pile-up •  The jet algorithm significantly shapes the jet substructure – less so for kTbut has poorer mass resolution •  Proposing amethod to deal with these issues: modify the jet substructure to reduce algorithm effects and improve mass resolution, background rejection, and heavy particle identification - pruning LPC @ Fermilab S.D. Ellis 11/18/10

  35. SDE, Jon Walsh and Chris Vermilion, 0903.5081, 0912.0033- go to tinyurl.com/jetpruning Pruning : Procedure: • Start with the objects (e.g. towers) forming a jet found with a recombination algorithm (kT, CA, Anti-kT) • Rerun with kT or CA algorithm, but at each recombination test whether soft – large angle: • z < zcut and ΔRij > Dcut • If true (a soft, large angle recombination), prune the softer branch by NOT doing the recombination and discarding the softer branch (bottom up approach) • Proceed with the algorithm  The resulting jet is the pruned jet LPC @ Fermilab S.D. Ellis 11/18/10

  36. Choices of the pruning parameters no pruning After studies we choose: CA: zcut = 0.1 and Dcut = mJ/PT,J optimal pruning kT: zcut = 0.15 and Dcut = mJ/PT,J no pruning over-pruning  mJ/PT,J is IR safe measureof opening angle of found jet and “adaptive” to the specific jet over-pruning shower pruning LPC @ Fermilab S.D. Ellis 11/18/10

  37. Prune Fixed Order Result “Large” mass tail unchanged(symmetric splitting) Move “small” masses to zero mass(asymmetric splitting) LPC @ Fermilab S.D. Ellis 11/18/10

  38. Pruning in Action Pruning of a QCD jet near the top mass with the CA algorithm a typical jet (see above) Red is higher pT Blue is lower pT Green X is a pruning Start with cells with energy > 1GeV R Prune pT: 600 → 590 GeV mass: 170 → 160 GeV z LPC @ Fermilab S.D. Ellis 11/18/10

  39. Pruning in Action Pruning of a QCD jet near the top mass with the CA algorithm atypical jet Red is higher pT Blue is lower pT Green X is a pruning Start with cells with energy > 1GeV R Prune pT: 600 →550 GeV mass: 180 →30 GeV z LPC @ Fermilab S.D. Ellis 11/18/10

  40. Impact of Pruning – qualitatively just what we want! Win Win The mass resolution of prunedtop jets is narrower  Pruned QCD jets have lower mass, sometimes much lower QCD jets Top jets CA kT LPC @ Fermilab S.D. Ellis 11/18/10 500 < pT < 700 GeV

  41. Test Pruning in more detail: • Study of top reconstruction: • Hadronic top decay as a surrogate for a massive particle produced at the LHC • Use a QCD multijet background based on matched samples from 2, 3, and 4 hard parton MEs • ME from MadGraph, showered and hadronized in Pythia, jets found with FastJet • Look at several quantities before/after pruning: Mass resolution of reconstructed tops (width of bump),small width means smaller background contribution • pTdependence of pruning effect (bin in pT) • Dependence on choice of jet algorithm and angular parameter D • UE dependence LPC @ Fermilab S.D. Ellis 11/18/10

  42. Defining Reconstructed Tops – Search Mode • A jet reconstructing a top will have a mass within the top mass window, and a primary subjet mass within the W mass window - call these jets top jets • Defining the top, W mass windows: • Fit the observed jet mass and subjet mass distributions with (asymmetric) Breit-Wigner plus continuum  widths of the peaks • The top and W windows are defined separately for pruned and not pruned - test whether pruning is narrowing the mass distribution pruned unpruned sample mass fit LPC @ Fermilab S.D. Ellis 11/18/10

  43. Defining Reconstructed Tops fit mass windows to identify a reconstructed top quark peak function: skewed Breit-Wigner fit top jet mass peak width Γjet plus continuum background distribution 2Γjet LPC @ Fermilab S.D. Ellis 11/18/10

  44. Defining Reconstructed Tops fit mass windows to identify a reconstructed top quark cut on masses of jet (top mass) and subjet(W mass) fit top jet mass peak width Γjet fit W subjet mass 2Γ1 2Γjet LPC @ Fermilab S.D. Ellis 11/18/10

  45. Defining Reconstructed Tops fit mass windows to identify a reconstructed top quark cut on masses of jet (top mass) and subjet (W mass) fit top jet mass fit W subjet mass window widths for pruned (pX) and unpruned jets LPC @ Fermilab S.D. Ellis 11/18/10

  46. Mass Windows and Pruning - Summary • Fit the top and W mass peaks, look at window widths for unpruned and pruned (pX) cases in (200 - 300 GeV wide) pT bins  Pruned windows narrower, meaning better mass bump resolution - better heavy particle ID  Pruned window widths fairly consistent between algorithms (not true of unpruned), over the full range in pT LPC @ Fermilab S.D. Ellis 11/18/10

  47. Statistical Measures: • Count top jets in signal and background samples in fitted bins • Have compared pruned and unpruned samples with 3 measures: • ε, R, S - efficiency, Sig/Bkg, and Sig/Bkg1/2 Here focus on S D = 1 S > 1 (improved likelihood to see bump if prune), all pT, all bkgs, both algorithms LPC @ Fermilab S.D. Ellis 11/18/10

  48. Heavy Particle Decays and D See also Krohn, Thaler & Wang (0903.0392) • Heavy particle ID with the unpruned algorithm is improved when D is matched to the expected average decay angle • Rule of thumb (as above): R = 2m/pT • Two cases: D R D R • D > R • lets in extra radiation • QCD jet masses larger • D < R • particle will not be reconstructed LPC @ Fermilab S.D. Ellis 11/18/10

  49. Improvements in Pruning • Optimize D for each pT bin: D = min(2m/pTmin, 1.0)  (1.0,0.7,0.5,0.4) for our pT bins • Pruning still shows improvements • How does pruning compare between fixed D = 1.0 and D optimized for each pTbin  SD = SD opt/SD=1? • Little further improvement obtained by varying D • SD = 1 in first bin • Pruning with Fixed D does most of the work LPC @ Fermilab S.D. Ellis 11/18/10

  50. Underlying Event Rejection with Pruning The mass resolution of pruned jets is (essentially) unchanged with or without the underlying event top study no pruning pruning CA kT LPC @ Fermilab S.D. Ellis 11/18/10 500 < pT < 700 GeV

More Related