Jet Substructure

From Wiki Les Houches 09

(Difference between revisions)
Jump to: navigation, search
(Coloured heavy objects (top,...): added link to session 2 ATLAS material)
Line 56: Line 56:
* [ Almeida+] and [ Almeida+] top event shapes
* [ Almeida+] and [ Almeida+] top event shapes
* Top ID using kT y scales [ this Les Houches report] (p106)
* Top ID using kT y scales [ this Les Houches report] (p106)
* [] ATLAS note on top reconstruction with substructure, shown in session 2.
* Comparison of tagging for different jet algorithms on top (Matt Schwartz):
* Comparison of tagging for different jet algorithms on top (Matt Schwartz):

Revision as of 11:56, 23 June 2009

Back to Tools_and_Monte_Carlo_Session_1_(SM)#Jet_Physics.

Boosted hadronic decays of massive particles (W,Z,Top,H,BSM..., with session 2) and jet mass/shape studies with QCD jets in early data.

Interested parties (Session 1)

Matt Schwartz, Giacinto Piacquadio, Mario Campanelli, Paulo Francvilla, Jon Butterworth, Peter Loch, Ezio Maina, Leif Lonnblad, Keith Hamilton, Simon Dean, Rohini Godbole, ...

Examine/discuss the different regions of validity for calculations of variables like jet mass, jet width, subjet multiplicity, using e.g. (N)LO ME, PS, matched and resummed calculations etc... (see also A Collection of Matching Benchmarks). Review, compare, critique literature, think about future ideas.


Types of object one might use substructure on

  • Substructure tagging minireview. (Schwartz...)

QCD jets (quark gluon separation)

  • Wjj provides a sample of quark jets at LHC?
  • SUSY cascades are rich in quark jets. Could use quark ID to simplify decay chains.

Colour singlet heavy objects, two body decay (W,Z,H...)

Plot from Matt showing higgs pT spectrum in ttH events.

Image:Pt of higgs.png

Shows that 17% of ttH events contain a Higgs with pT>200 GeV, and ~50% have one with pT>100 GeV. Shows that subjet analyses could have a big impact on ttH.

Simon/Jan/Jon to look at using the Higgs substructure analysis on ttH events from Sherpa (particle level).

Colour singlet heavy objects, three body decay (Neutralino,...)

Coloured heavy objects (top,...)

  • Comparison of tagging for different jet algorithms on top (Matt Schwartz):

The minimum invariant mass of the sum of any 2 subjets of the top jet (which can contain 4 subjets), as identified by various algorithms. This variable is sensitive to the QCD singularity in the background (solid,red) and has a W mass peak in the signal (blue). Note that anti-KT declustering does not provide strong discrimination, but finding the jets with anti-kT, and then declustering with C/A works well.

MC issues

Differences in heavy object decays; different parton showers, matrix element corrections in some MC, not in others; in Herwig, Herwig++, PYTHIA, SHERPA. POWHEG improvement soon. See talk from Carlo Oleari Friday afternoon for some discussion of MC@NLO and POWHEG. Dead cones etc. Keith Hamilton, Giacinto Piacquadio, Matt Schwartz, Leif Lonnblad, Jan Winter to produce a short summary of the effects implemented in different MCs.

How sensitive are the various subjet methods to the differences?

Would be nice to have some truth level comparisons, of simple variables (jet mass etc) and also of some of the various boosted decay methods (esp top?) if possible. Matt will look into this if people will send him events.

Plots from GP showing the difference between PYTHIA and Herwig when applying the subjet clustering technique implemented for the high pT WH to lnubb analysis:

  • H to bb invariant mass distribution at true level
  • H to bb invariant mass distribution with expected detector smearing
  • pT(third subjet)/(pT(second subjet)+pT(first subjet) looking into two different windows

Image:FilteredMassJetComp.png Image:PlotMassSigSmeared.png Image:ThirdSubjet.png

In the third plot, it can be clearly seen that Herwig fills almost no events with pT(third subjet) > pT(second b-subjet), while PYTHIA does, due to the ME corrections implemented in the decay. The radiation spectrum is therefore softer in HERWIG with respect to PYTHIA. This can be a reason for the fact that one sees a degradation in the mass resolution at hadron level in HERWIG with respect to PYTHIA. However, once the detector smearing is taken into account, the difference gets essentially smeared out and the mass distributions are again quite comparable.

MC references:

Related, non-MC, theoretical activity on resummation and QCD radiation patterns:

Top mass:

  • Adrian Signer Definition of mtop and colour connection effects / non-factorisable contributions (motivation for the multiscale shower).

Detector issues

Pile-up, calorimeter noise, granularity, acceptance: The reconstruction quality for the various jet shape variables need to be understood. A first look at jet masses and y-scale in ATLAS can be found at (transparencies), S.D.Ellis+, and at ATLAS Collaboration.

Studies from GP.

Proposed cuts for studying this (highest priority ones in bold):

  • pT cut on constituents (MeV) 0,100,500,1000,2000.
  • QCD jets in pT bins between 17,35,70,140,280,560,1120 GeV. Look at the leading jet. Radipity +/- 5.
  • Pile-up. 0,4,8,16-20 interactions per signal event (Poisson). (Think about doing this as a function of number vertices in full sim).
  • Jet algorithms. Use Anti-kT R=0.4, R=0.6. And also C/A, kT etc where appropriate, and also studies of using e.g. C/A to recluster anti-kT jets

Sarah Allwood's thesis can be found in the Manchester thesis archive. This includes studies of the effect of pile-up on W reconstruction using the y-scale method (e.g. p103).

Find the jet first, then cut on constituents, or cut on constituents first then find the jet?

Constituents = final state particles, or pre-clustered final state particles, or towers, or topoclusters...

Slides from Peter and Paulo on the effect of pile up on jet mass and substructure variables.

Miscellaneous References

* Seymour+ clustering (kT algorithm)
* Dokshitzer+ clustering
* S. Ellis+ jet pruning, using top as an example
* Krohn+ variable R
Personal tools