User Tools

Site Tools


Les Houches

2023 Session

Wikis of Previous sessions

Les Houches Themes



Quark/Gluon Enrichment Studies

a.k.a. Hunting the White Whale of Jet Substructure

If you are interested in contributing please subscribe to the mailing list. You will be asked to log in with your CERN account. If you don't have a CERN account, please fill this form to create a lightweight account.

Link to GitHub repository:

Slides from Saturday

Plan of attack for Wednesday Les Houches Presentation

  • Write our quark/gluon manifesto/history.
  • Catalog quark gluon discriminants and performance measures
  • e+e- → u uubar , gg
  • Baseline: Q = 200 GeV, R = 0.6, all Monte Carlo programs optimal
  • Quark distribution, gluon distribution, separation, integrated separation
  • Pythia variations: noME, nogqq
  • Herwig variations: …
  • Sherpa variations: …
  • Vincia variations: noME
  • Q = 50, 100, 200, 400, 800 GeV (everything else baseline)
  • R = 0.2, 0.4, 0.6, 0.8, 1.0 (everything else baseline)
  • delta alphas / alphas = -0.2, -0.1, 0.0, +0.1, +0.2 (everything else baseline)


Quark/gluon discrimination well-defined?

  • Of course, at the hadron level, you can't define a quark jet vs. a gluon jet unambiguously.
  • That said, one can talk about quark/gluon enriched samples, where restrictions are placed on the final state to preferentially select quark- or gluon-initiated jets (e.g. gluon enrichment in dijets, quark enrichment in vector boson plus jet).
  • In fixed-order QCD, there is an ambiguity from soft gluon splitting to wide-angle quark/anti-quark. However, in the eikonal limit, there is no ambiguity (up to power corrections), so quark/gluon calculations can be done at the parton level in the eikonal limit (relevant for resummed calculations).
  • If needed, we can use flavored jet algorithms to give an IRC safe definition of jet flavor at the parton level.

How to isolate quark vs. gluon samples?

  • Ultimately, we need an operational definition of quark- and gluon-enriched samples (e.g. event type, rapidity correlations, event shapes).
  • This will allow us to separate the measurement of jet properties from the interpretation of those properties in the context of discrimination/enrichment studies.
  • One has to be aware of process dependence, since a quark in one context may not look like a quark in another context (color correlations).
  • Ultimately, need MC studies to compare to behavior in data.

"Discrimination" really the right word?

  • Probably better to talk about “quark/gluon enrichment”.
  • For physics applications, we want to achieve S/sqrt{B} improvement, which isn't really the same as quark/gluon discrimination.
  • Similar issues arise in how to define a “hadronic W”.
  • Quark/gluon enrichment should be a piece of a more refined analysis.
  • We can provide general recipes, but should not aim for optimal analyses, which are only sensible in the context of specific physics goals.

What is the killer app of quark/gluon enrichment?

  • VBF tagging
  • Rejecting (stochastic) pileup jets (important for VBF)
  • SUSY multi-jet tends to be quark-enriched
  • Enhancing W/Z/t/H in moderately boosted regime (where we can quark-tag the subjets.

Physics Issues

Separating final state from initial state effects

  • Different jet shapes probe different phase space regions. For example, jet mass is more sensitive to wide angle physics while multiplicity is more sensitive to collinear physics.
  • Differences between MC programs appear in multiplicity-like observables, so most likely a final state effect.
  • We can probe different physics by looking at hard core (collinear, FSR, beta → 0) vs. wide angle (soft, ISR, beta → infty).

FSR effects

  • FSR effects should be dominant at small angles, yielding universal properties.
  • Tuning of gluon final state shower can affect jet shapes.
  • Examples: g → q qbar vs. g → gg, including spin-polarization information
  • Do beyond-LL effects help or hurt quark/gluon enrichment?
  • What about the impact of heavy flavor?

ISR effects

  • ISR effects should dominate at large angles
  • Highly process dependent, depends on color corrections of jet with ISR
  • We will attempt to deemphasize these in our study, if possible

Experimental Results

  • ATLAS paper suggests that beta → 0 (i.e. hard core) is not as effective as NLL calculations suggest. (see Appendix A.)
  • CMS finds ptD (an example of a beta → 0 observable) is quite effective.
  • ATLAS sees considerable process dependence, whereas CMS has not emphasized this issue. Is this connected to ISR in some way?
  • ATLAS A14 tune already uses jet shapes, and finds that alpha_s has to be tuned downward in Pythia 8. This, however, has a detrimental effect on LEP measurements, so one has to be cautious about this.
  • Is there a tuning flat direction?

Hemisphere quark/gluon definitions in e+e-

  • Consider the case of e+ e- → q qbar. Partition event into (thrust) hemisphere, define hemisphere flavor by summing over flavors of hemisphere constituents.
  • At LO, we can unambiguously define hemisphere flavors.
  • At NLO, we can also unambiguously define flavor via hemisphere, though there is now a small gluon fraction from gluon recoiling against q qbar pair.
  • At NNLO, things are more complicated.
    • Can have soft gluon splitting into q-qbar in different hemispheres, creates IRC safety issue.
    • One can use a flavored algorithm (BSZ) to define the flavour of two flavor-kt jets
  • Ultimately, want to give an operational definition of flavor based on the Born-level operator contributing to the process.
    • Claim: all subtleties are formally power suppressed.
    • Use case, VBF, two jets with a third jet veto, q/g well-defined in the exclusive limit.

Flavored Jet Algorithms

  • This is a topic worthy of its own Les Houches study.
  • For pp collisions, multiple possible uses of flavored jet algorithms.
  • One can just run flavor-kT
  • Or one can run flavor-kT to define flavor ghosts, and run standard anti-kT.
  • Or one can run flavor-kT for deflavoring constituents, and then run standard anti-kT.

Ultimate Goal for Les Houches Study

  • Recommendation to ATLAS and CMS for observables that should be measured which carry quark/gluon information.
  • These observables must be defined on the final state alone (i.e. fiducial cross section).
  • These observables should help enrich quarks over gluons (or vice versa).
  • Eventually, these observables should be useful for MC tuning, with controllable systematics.
  • Make recommendation about robustness vs. performance. We will likely emphasize robustness, since performance depends strongly on process dependence, pileup.
  • Question: How should we discuss low pT vs. high pT

Initial Les Houches Study

Key Question

  • Do we understand FSR modeling by workhorse parton showers?
  • Start with the clean case of e+e-, move to pp later.

Basic Plan

  • Take e+e- → q qbar, and e+e- → g g
  • Vary collision energy, jet radius
  • Choose a core set of jet shapes
  • Use as many MC options as possible.
  • Question: use ROC curves or mutual information (I(T;A)) to quantify discrimination power?
    • Answer: doesn't really matter, probably I(T;A) is easier to begin with.
    • Better answer: Use separation (S-B)^2 / (2 (S + B)).

Core Jet Shapes

  • Generalized angularities (kappa, beta)
    • (1,0.5)
    • (1, 1) – jet width
    • (1,2) – jet mass
    • (0,0) – multiplicity
    • (2,0) – ptD
  • Question: Apply on full events or just tracks
    • Answer: Apply on full events.
  • Question: Choice of axes? (issue of recoil)
    • Answer: WTA recombination axes from anti-kT
  • Question: Sum over particles (angularity-style) vs. sum over pairs (ECF-style)
    • Answer: Sum over particles (angularity-style)
  • Question: Plot linear or log scale?
    • Answer: Do both if it makes sense, better for angularities to have log scale.

Supplemental Jet Shapes

  • More generalized angularities (kappa, beta)
    • (0.5,0.5)
    • (0.5,1.5) – should give bad performance
    • limit (1+epsilon,0) / epsilon – should give good perfomance
  • Ellipticity/eccentricity
  • Covariance matrix observables
  • Pull
  • Psi® – the jet shape
  • Check Gallicchio and Schwarz catalog
  • tau21, or ECF(2,3)
  • Generalized angularities with soft-drop jets, varying beta_SD
  • Do sum over pairs version of angularities (i.e. ECF-style)

Analysis Workflow

  • Rivet analysis in place which computes from a HepMC event sample the various generalised angularity distributions.
  • Processes to consider:
    • mu+mu- → spin1 → q qbar take photons
    • mu+mu- → spin0 → g g take Higgs
    • for tests of universality: mu+mu- → spin0 → q qbar
  • Energies
    • Q=sqrt{s} = 50, 200, 800 GeV
    • Optionally: Q = 100, 400 GeV
  • Jet definition:
    • ee-antikt [genkt, p=-1], WTA_modp recomb scheme
    • R = 0.3, 0.6, 0.9
  • Add thrust from thrust hemispheres for anticipated analytic comparisons
  • Add multiplicity (event-wide) in bins of thrust:
    • T < 5 GeV/sqrt(S)
    • 5 GeV/sqrt(S) < T < 0.1
    • 0.1 < T < 0.2
    • 0.2 < T

Preliminary plots for meeting on Thursday


  • Is discrimination power (e.g. for width) coming from the hadronization regime?
    • Possibility: Isolate hadronization regime (thrust ~ LambdaQCD/Q) and shower regime (thrust ~ 0.1-0.2) and optionally hard jet regime (thrust >~ 0.25). Study scaling of, e.g., multiplicity as a function of Q in each of these regimes.
  • By testing pythia vs. herwig, can we test string vs. cluster hadronization?
  • Is there jet radius dependence?
  • Does matching help in controlling quark/gluon uncertainties?
  • Universality/process dependence of conclusions?
    • Related to whether the discrimination power comes from the core or the periphery of jet.

Next Les Houches Study (for after LH)

  • Above study at hadron colliders, using dijets, W/Z/gamma + j, and maybe t tbar samples

Analytic Les Houches Study?

  • Analytic predictions known/available/straightforward for:
    • Quark thrust: N^3LL' + N^3L0
    • Gluon thrust: N^2LL' + N^2L0
    • ang (kappa =1): NLL'
  • Can we do useful quark/gluon study from analytic results?

Notes from Tuesday Meeting

* original discussion from 2015-06-02

Generic commentes:

- ill-defined but OK in the eikonal limit
  one speaks of quark or gluon-enriched samples

- we need an operational definition defined by final-state.
  e.g. rapidity correlation or shape
  separate measurement v. interpretation

- be worried about the process dependence and the detector effects (MC
  study v. behaviour in data)

- try to speak of a S/\sqrt{B} improvements rather than q/g discrimination
  [this is also related to the definition of a hadronic W]

- can we find a better naming convention?
  Suggestion: quark-gluon enrichment

- Why?
  Is there a killer app?
  use q/g as a piece of an analysis
  general recipe v. optimal analysis

- can we use flavoured jet algorithms?

- FSR issues (pointed out by Jesse in the introductory talk)
  related to tuning gluon jets in MC
  related to colour correlations with ISR
  [ISR should be sensitive to large angle => see some process-dependence]
  [FSR should be fimen at small angle => see universality]

- Look at hard core v. wide angle.

- Experimental results
   . ATLAS v. CMS: 
      ATLAS suggests beta->0 is not effective
      CMS says pTD is good
   . analytics suggest low beta works best
   . ATLAS sees a large process dependence
   . is there a connection with ISR?

     see Appendix A.

- Spin information.
   . is it in the MC? [yes]
   . does it help or hurt?

- g \to qqbar v. g \to gg
   does it help or hurt

- quark v. heavy flavour?

Use cases:

- VBF tagging
- SUSY multijet is q-enriched
- pileup jet rejection (stochastic?)
- enhance W/Z/t/H in boosted regime

Concrete study(ies) for Les Houches?

Ideal result: recommendation for experiment for observables that carry
  info, defined in the final state and define q/g and eventually
  useful for MC tuning (check the systematics)

- A14 tune uses jet shape 
  this brings alphas down
  Q: what happens to LEP?

- 1st study: do we understand FSR modelling?

   . take e+e- to qqbar and gg
   . vary energy and jet radius
   . vary shapes
   . ROC v. mutual info I(T;A)
   . use as many MC options as possible

shapes: (kappa,beta)=(1  ,0.5)
                     (1  ,1)
                     (1  ,2)
                     (0  ,0)
                     (2  ,0)

        + ellipticity
       (+ pull)?
        + Psi(r)
        + check Gallicchio and Schwarz
        . tau21 of ECF(2,3)
        + ??
      on full event or only on tracks?

     questions: choice of axes (recoil)
                sum or sum over pairs?

   . for hadron colliders: 
        look at dijets

   . use groomed jets
        use soft-drop beta 

   . question of robustness v. performance 
     (including process dependence, pileup, low pt v. high pt)


 - tool writing (jet shapes)
     use FJ, Rivet
   [Jesse, Gregory, Deepak]  [+Andy B??]
 - MC for e+e-
     HERWIG, Pythia, SHERPA [hook Franck, Mark?], VINCIA, ...
     + options
     [Peter S, Andrzej]
 - MC for pp
     above, +PU, VBF
     [Peter L.]

Notes from Thursday Meeting

Meeting in Les-Houches

Presentation of the wiki notes: list of contributors, ...

Presentation of the status of the software: 
  start w e+e- and do pp later
  Rivet analysis in place which computes from a HepMC event sample the various generalised angularity distributions

Reminder: what we mean by a q and a g is e+e-\to qq and e+e-\to gg
  If we want to do something more refined:
    - at LO we can unambiguously sum flavours in hemispheres defined by thrust
    - at NLO we can unambiguously sum flavours in hemispheres defined by thrust
      we get a quark and a small gluon fraction
    - at NNLO things are more complicated. We can use a flavoured
      algorithm (BSZ) to define the flavour of each hemisphere
    - for pp collisions, we should use a flavoured algorithm to
      determine flavour, and then find a way (e.g. using ghosts) to
      run anti-kt jets. This would deserve a topic per se (a LH accord)?
    - Question: can we match to the Born and find an operatiroal
      definition up t power corrections?
      Use case: VBF, two jets with a third jet veto. q/g well-defined
                in the exclusive limit

Questions to look into:
 - is the discrimination power (e.g. for width) coming from the hadronisation regime?
 - plotting in log binninb?
 - pythia v. herwig important to test string v. cluster hadronisation
 - isolate hadronisation regime. Study the scaling in different bins
   of one angularity (e.g. thrust). Take a hadronisation region
   (T\propto LQCD/Q) and a shower region (T~0.1-0.2) plus optionally a
   "hard jet region" (T >~ 0.25)
 - does mathing help?
 - jet radius dependence (edit analysis and recompile)
 - analytic predictions?
    for thrust: ee->qq known at N^3LL' + N^3LO
                ee->qq known at N^2LL' + N^2LO
        ang(bkappa=1): NLL'
 - question of the universality/process dependence of the conclusions?
   Related to whether the power comes from the core or the periphery?

 - process to consider: 
     mu+mu- -> spin1 -> qq  take photons
     mu+mu- -> spin0 -> gg  take Higgs
   for tests of universality
     mu+mu- -> spin0 -> qq

 - Energies Q=sqrt = 50, 200, 800 GeV
   jetdef: ee-antikt [genkt, p=-1], WTA_modp recomb scheme
   radii: 0.3, 0.6, 0.9

 - add thrust from thrust hemispheres for analytic purpose

 - add multiplicity (event-wide) in bins of thrust:
     T < 5 GeV/sqrt(S)
     5 GeV/sqrt(S) < T < 0.1
     0.1 < T < 0.2
     0.2 < T

Notes for Jesse for Preparing Summary Talk

  • Quark is more of an adjective than a noun.
  • Pseudo-quark? (That language doesn't go over very well.)

What is a Quark Jet?

(From ill-defined to well-defined)

  • A quark parton
  • A Born-level quark parton
  • The initiating quark parton in a final state shower
  • An eikonal line with baryon number 1/3 and carrying triplet color charge
  • A quark operator that appears in a hard matrix element in the context of a factorization theorem.
  • A parton-level jet object that has been tagged as a quark using a soft-safe flavored jet algorithm (automatically collinear safe if you sum constituent flavors).
  • A phase space region (as defined by an unambiguous hadronic fiducial cross section measurement) that yields an enriched sample of quarks (as interpreted by some suitable, though fundamentally ambiguous, criterion).

(Sometimes people think we care about the top of the list while we are really focused entirely on the bottom.)

2015/groups/sm/qg.txt · Last modified: 2015/07/14 08:27 by philippe.gras