Pith. Sign in

REVIEW

W-jet Tagging: Optimizing the Identification of Boosted Hadronically-Decaying W Bosons

PaperPDF from arXiv

data files

ReviewThe objections and verdict, as JSON Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Yanou Cui, Zhenyu Han, Matthew D. Schwartz

High Energy Physics - Phenomenology · Submitted 9 Dec 2010 · 43 references

Pith's verdict: A real tool paper: the R-core observables are new and the multivariate gain over filtering is genuine, but the headline ~5x is a Pythia-8-internal number and the cross-generator check is a train/test domain shift, not a systematic.

T0 · AI review

How this review was made: An AI referee (selfhosted/deepseek-v4.1-flash) read the paper on November 10, 2026 and answered a fixed list of questions about it. T0 is the starting level every paper gets. The levels describe how deeply Pith has checked the paper, never the authors.

  1. T0 AI review An AI read the paper and answered a fixed list of questions.
  2. T1 Audit Automatic checks that the references and files the paper relies on exist and match.
  3. T2 Compute The paper's code or calculations re-run with the same results.
  4. T3 Formal statement The claims written out in a formal math language a computer can read.
  5. T4 Proof Those claims proved, and the proof checked by computer.

How Pith checks papers →

· 3 major objections · 9 minor · Ask about this paper · The compiled record · Read the full review →
Challenge this review Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID1012.2077 (version 2)
Pith IDMDPOK76U, Pith's permanent record of this paper and its review
CitationMatthew D. Schwartz, Yanou Cui, Zhenyu Han. "W-jet Tagging: Optimizing the Identification of Boosted Hadronically-Decaying W Bosons." arXiv:1012.2077 (2010). https://pith.science/paper/MDPOK76U
Cite this reviewBibTeX for Pith's review of this paper
Claim paper

Why Pith says this

Highlight any text to comment on it.

Referee report

How Pith checks papers → · Full audit of this paper →

The paper

The paper claims

A multivariate tagger built from 25 jet observables raises the significance of boosted hadronically decaying W bosons by roughly a factor of five over QCD-jet backgrounds, about twice what jet grooming alone delivers.

The reading

The paper asks how much of a boosted, hadronically decaying W boson can be recovered from the single fat jet it leaves in the calorimeter, and how far that jet can be separated from an ordinary quark or gluon jet of the same transverse momentum. The authors' answer is that grooming a fat jet down to its hard subjets — filtering with a mass-drop, trimming, or pruning — recovers only about a factor of two in $S/√B$, and that most of the remaining discriminating information lies in the radiation pattern outside and around those subjets. They assemble 25 observables that describe that pattern, combine them with boosted decision trees, and report a factor of roughly five in $S/√B$ for jets from 200 to 1000 GeV, rising with $p_T$. If this holds, searches for heavy resonances decaying to a W pair, and attempts to pick a hadronic W out of the QCD dijet background, gain reach comparable to a large increase in luminosity. The authors also release the trained tagger code so that the gain can be reproduced and reused.

What carries the argument

The new ingredient is the set of $p_T$ R-cores: recluster the fat jet with a smaller radius $R$, take the hardest subjet, and record the ratio $c_{p_T}(R) = p_T(R)/p_T(R_{fat})$, with $R$ scanned from 0.2 to 1.1. The scan interpolates between two regimes — at large $R$ it measures how much of the jet's momentum stays inside a cone, which reflects whether the radiation is confined as it is for a color-singlet decay or spread out as it is for a jet color-connected to the beam; at small $R$ it degenerates into a hard-splitting energy fraction. The transition sits near $∆R_{sub} ∼ 2m_W/p_T$, about 0.3 at 500 GeV, and no single R-core does much, but ten of them together give a 40 percent improvement in significance where the best single one gives 15 percent. The second ingredient is a set of grooming sensitivities, ratios such as $sens_{m}^{filt} = m_{filt}/m$ and the analogous trimming and pruning ratios and $p_T$ versions, which record how much of the jet each grooming procedure throws away. These, plus fat-jet and subjet masses, subjet $p_T$ fractions, $∆R_{sub}$, subjet multiplicity, and planar flow $P_f$ (a measure of whether energy deposition in the jet is linear or isotropic, defined as $P_f = 4λ_1λ_2/(λ_1+λ_2)^2$ from the eigenvalues of the transverse momentum tensor), form the 25-variable input. The bookkeeping device throughout is the significance improvement characteristic $SIC = ε_S/√{ε_B}$, which lets the authors compare variables and cut choices independent of luminosity.

Extended reading notes

Core claim

The central claim is that a boosted decision tree fed with a broad set of jet-substructure observables separates W-jets from QCD-jets far better than any single grooming algorithm, and that the improvement is large enough to change the reach of LHC analyses. Starting from Cambridge/Aachen jets with $R=1.2$ that pass a filtered-mass window of $(60,100)$ GeV, the authors find that filtering with mass-drop, trimming, and pruning each give a significance improvement characteristic $SIC = ε_S/√{ε_B}$ of about $2$ in the 500–550 GeV bin, and that simply using narrow $R=0.4$ jets does nearly as well. On top of the filtered sample, individual shape and substructure variables add at most about 20 percent each, but the combination reaches a maximal $SIC$ of $3.4$ to $6.7$ across $p_T$ from 200 to 1000 GeV. The practical statement is that at a signal efficiency of 40 percent, the multivariate tagger rejects roughly four times more background than filtering alone. The authors also report that longitudinally polarized W bosons are tagged better than transverse ones — maximal $SIC$ about $7.0$ versus $5.3$ in the 500–550 GeV bin — because their decay products carry more balanced momenta, and they demonstrate the tagger on a 1.5 TeV $Z'$ decaying to $W^+W^-$ and on hadronic W plus jet in dijet events.

Load-bearing premise

The gain depends on the event generator's description of soft radiation and the underlying event inside large-radius jets, and the paper's own cross-check with a second generator costs about a quarter of the maximal significance.

Editorial extensions

If this is right

  • Diboson resonance searches in the semileptonic channel gain substantially: in the authors' 1.5 TeV $Z'$ example at 2 fb$^{-1}$, the multivariate tagger gives $S/√B$ of about 7.6 against about 5.0 for filtering and 4.3 for a conventional $R=0.4$ mass cut.
  • Hadronic W plus jet can be separated from the QCD dijet background at the 7 TeV LHC: the authors report $S/√B$ rising from 2.2 before tagging to 5.1 after the multivariate stage at 1 fb$^{-1}$.
  • Because the tagger responds differently to longitudinal and transverse polarizations, the same variables can in principle be used to measure the polarization of a W-jet sample rather than only to select it.
  • The released weight files mean the tagger can be applied directly to other processes with boosted W-jets, and the same procedure should transfer to boosted Z and Higgs jets with re-optimized cuts.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The variables carrying most of the gain — the large-$R$ R-cores, the unfiltered $R=1.2$ mass, the $R=1.2$ planar flow, and the grooming ratios — all integrate over the whole fat-jet area, so the tagger's performance is tied to how well a shower model describes soft radiation at large angles. That suggests the gain will shrink further under real pileup, and that a pileup-corrected or smaller-radius
  • The transition radius where the R-core scan is most informative is set by $2m/p_T$, so the same construction should work for any two-body resonance once the mass is substituted; a Z or Higgs tagger would want the scan concentrated near its own $2m/p_T$ rather than the fixed 0.2–1.1 grid used here.
  • Because the discriminating power depends on polarization, a sample selected by this tagger is biased toward longitudinal W's; any cross-section measurement made after tagging would need an unfolding or a polarization-dependent efficiency correction.
  • The disagreement between shower models points to an in-situ calibration path: a tag-and-probe measurement in a sample with a leptonic W on the other side could fix the tagger's efficiency on data and absorb the underlying-event uncertainty directly.
Original abstract

A method is proposed for distinguishing highly boosted hadronically decaying W's (W-jets) from QCD-jets using jet substructure. Previous methods, such as the filtering/mass-drop method, can give a factor of ~2 improvement in S/sqrt(B) for jet pT > 200 GeV. In contrast, a multivariate approach including new discriminants such as R-cores, which characterize the shape of the W-jet, subjet planar flow, and grooming-sensitivities is shown to provide a much larger factor of ~5 improvement in S/sqrt(B). For longitudinally polarized W's, such as those coming from many new physics models, the discrimination is even better. Comparing different Monte Carlo simulations, we observe a sensitivity of some variables to the underlying event; however, even with a conservative estimates, the multivariate approach is very powerful. Applications to semileptonic WW resonance searches and all-hadronic W+jet searches at the LHC are also discussed. Code implementing our W-jet tagging algorithm is publicly available at http://jets.physics.harvard.edu/wtag

The compiled record

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Compiled by classifier-1.13-20260917 via OpenRouter, question bank abstract-p1-20260923, 2026-09-23: 1 request, 5,690 input tokens, $0.0002 reported.

What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.87 · human participants 0.01 · controlled comparison 0.67 · randomized assignment 0.01 · states quantitative result 0.98 · compares with prior results 0.98 · mentions code or data availability 0.98.

The abstract, sentence by sentence

arXiv listing abstract v2. Results and the main claim are underlined; .

A method is proposed for distinguishing highly boosted hadronically decaying W's (W-jets) from QCD-jets using jet substructure. Previous methods, such as the filtering/mass-drop method, can give a factor of ~2 improvement in S/sqrt(B) for jet pT > 200 GeV. In contrast, a multivariate approach including new discriminants such as R-cores, which characterize the shape of the W-jet, subjet planar flow, and grooming-sensitivities is shown to provide a much larger factor of ~5 improvement in S/sqrt(B). For longitudinally polarized W's, such as those coming from many new physics models, the discrimination is even better. Comparing different Monte Carlo simulations, we observe a sensitivity of some variables to the underlying event; however, even with a conservative estimates, the multivariate approach is very powerful. Applications to semileptonic WW resonance searches and all-hadronic W+jet searches at the LHC are also discussed. Code implementing our W-jet tagging algorithm is publicly available at http://jets.physics.harvard.edu/wtag

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 A method is proposed for distinguishing highly boosted hadronically decaying W's (W-jets) from QCD-jets using jet substructure. contribution 0.97 0.65 0.14 asserted 0.81
s1 Previous methods, such as the filtering/mass-drop method, can give a factor of ~2 improvement in S/sqrt(B) for jet pT > 200 GeV. background 1.00 0.05 0.04 asserted 0.57
s2 In contrast, a multivariate approach including new discriminants such as R-cores, which characterize the shape of the W-jet, subjet planar flow, and grooming-sensitivities is shown to provide a much larger factor of ~5 improvement in S/sqrt(B). result 0.84 0.90 0.83 asserted 1.00
s3 For longitudinally polarized W's, such as those coming from many new physics models, the discrimination is even better. result 0.90 0.70 0.07 asserted 0.99
s4 Comparing different Monte Carlo simulations, we observe a sensitivity of some variables to the underlying event; however, even with a conservative estimates, the multivariate approach is very powerful. limitation 0.51 0.94 0.08 asserted 0.97
s5 Applications to semileptonic WW resonance searches and all-hadronic W+jet searches at the LHC are also discussed. contribution 0.78 0.46 0.03 asserted 0.88
s6 Code implementing our W-jet tagging algorithm is publicly available at http://jets.physics.harvard.edu/wtag contribution 0.92 0.68 0.02 asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from simulation or computation 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v2
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

Figures
Figure 1. FIG. 1: Significance Improvement Characteristics (
Figure 2. FIG. 2: Jet masses before and after filtering/mass-drop for
Figure 3. FIG. 3: The significance improvement characteristic (SIC
Figure 4. FIG. 4: Tuning of filtering parameters for
Figure 5. FIG. 5: Distributions for the fat-jet mass and hardest subje
Figure 6. FIG. 6: Representative
Figure 7. FIG. 7: The average values of
Figure 8. FIG. 8: Maximal SIC as a function of
Figure 9. FIG. 9: Gradual significance gain when adding
Figure 10. FIG. 10: Maximal SICs for the whole set of
Figure 11. FIG. 11: Distributions of grooming sensitivities, sens
Figure 12. FIG. 12: Signal vs. background planar flow (
Figure 13. FIG. 13: Signal and background distributions of
Figure 14. FIG. 14: Significance gain from the multivariate analysis fo
Figure 15. FIG. 15: The maximal SIC with MVA using all 25 principle varia
Figure 16. FIG. 16: Ratio of the
Figure 17. FIG. 17: Tuning of filtering parameters for longitudinally p
Figure 18. FIG. 18: The SIC using BDTs as a function of signal efficiency fo
Figure 19. FIG. 19: Signficance improvements resulting from a boosted d
Figure 20. FIG. 20: Simulation dependence of jet masses, for
Figure 21. FIG. 21: Planar flow for
Figure 22. FIG. 22: sens
Figure 23. FIG. 23: SIC curves obtained using a smaller set of variables
Figure 24. FIG. 24: Invariant mass distributions for signal (
Figure 25. FIG. 25: Application of
Figure 26. FIG. 26: For jet

Figures reproduced from arXiv:1012.2077 by the authors.

References
How later papers used it

45 later papers cite it

Evidence: substantial. 31 of the citing papers share no author with it, and 6 of those carry a confident label from Pith's citation classifier.

Pith read 13 of the 80 citing sentences, choosing only those whose citation markers point at this paper alone, and wrote the paraphrases from them. The summary, the counts, the evidence level, the links and the quotes come from code; the summary counts the papers listed below. A later paper is listed under a label when Pith's citation classifier gives one of its sentences that label at probability 0.8 or more (0.95 for support), Pith's reading finds that the citing paper itself did it, and a second check confirms the paraphrase against the sentence. This section lists uses only. It does not list criticism or disagreement, so their absence here says nothing about how the paper has been received. Papers that share an author with this one are listed apart, under "By the same authors". A quote is the citing authors' own sentence, shown when their paper's licence is CC BY, CC BY-SA or CC0. JSON

Uses its method or tool 2 of 5 papers with this label · 2 by the same authors

By the same authors

How the literature uses it

45 arXiv papers cite this in 80 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 37 sentences without a confident label

Uses its method or tool 5 papers
  • We have shown that, assisted by the recently developed jet substructure techniques, in particular the multivariate $W$ jet tagging method [ref] , we can further greatly reduce the persisting backgrounds from $W$+jets and $tt̅$+jets, while keeping a significant amount of signals.

    New light on WW scattering at the LHC with W jet tagging · Conclusion · 0.98

  • In particular, using the model of Latent Dirichlet Allocation (LDA) [ref] , which can be solved efficiently using e.g. Variational Inference (VI) [ref] techniques, we were able to define robust parametric jet and event classifiers.[Some related ML techniques have been previously studied in Refs. [ref] .]

    Learning the latent structure of collider events · Introduction · 0.97

  • Here, $c_P_t$'s are $P_t$ $R$-cores [ref] , from 0.2 to 1.1 by 0.1 and $sens^m,P_t_filt,trim,prun$ represent 6 grooming [ref] sensitivities. $P_f$ and $P_f(0.4)$ are the planar flow parameters for the original jet and for the highest $P_t$ subjet from reculstering with $R=0.4$, respectively. $ΔR_sub$ is the distance between the two leading subjets and $n_sub$ is the total number of subjets after the filtering process.

    Hadronic $b^\prime$ search at the LHC with top and W taggers · W-taggers · 0.99

Uses its data or values 1 paper
Compares with it 2 papers
  • We find that in the range $0.1 <μ^cut< 0.4$, the tagging efficiency is mildly sensitive to changes in $y^cut$, but that changes in $μ^cut$ result in large changes of performance[ The authors of Ref. [ref] considered $μ^cut> 0.4$, and found that in this operating regime, there is greater sensitivity to $y^cut$ than to $μ^cut$.].

    Tagging Boosted Ws with Wavelets · Mass Drop Tagging · 0.81

  • This approach simplifies the multivariate method in Ref. [ref] that utilizes as many as 25 variables, and may find applications at the LHC especially at its early stages.

    Tracking the Identities of Boosted Particles · Conclusion · 0.86

Cites it as support 2 papers
Cites it as background 22 papers
  • We study boosted gluinos using jet-substructure techniques [ref] , which have matured enormously in recent years and are being verified experimentally [ref] (for some recent reviews see [ref] ).

    Boosted Multijet Resonances and New Color-Flow Variables · Introduction · 0.85

  • Along these lines, many sophisticated W-tagging tools has been developed to pick out the highly boosted W bosons from backgrounds [ref] .

    Hadronic $b^\prime$ search at the LHC with top and W taggers · W-taggers · 0.84

  • These hard subjets are the key to distinguishing hadronically decaying $W$ bosons from the QCD background, and their exploitation has formed a very productive industry in collider physics over the past few years, with strategies having been developed [ref] for tagging top quarks [ref] , Higgs bosons [ref] , heavy gauge bosons [ref] , and even hypothetical particles [ref] .

    Tagging Boosted Ws with Wavelets · Introduction and motivation · 0.93

What it cites · 29 sentences without a confident label

Uses the method or tool of 8 papers
  • Starting from a jet with relatively large size $R$, the jet grooming algorithms act on the fat jet as follows * Filtering with mass drop [ref] : For a given jet found with recombination parameter $R$, we first look for a significant “mass drop” by the following procedure: * Undo the last step of jet clustering for jet $j$.

    Jet substructure as a new Higgs search channel at the LHC · Filtering/trimming/pruning · 0.99

  • To quantify and compare variables, we use the Significance Improvement Characteristic (SIC) [ref] , defined as the ratio of the signal efficiency to the square root of the background efficiency, $ε_S/√(ε_B)$.

    Multivariate discrimination and the Higgs + W/Z search · Introduction · 0.94

  • We simulate the hard $WW$ process in $pp$ collisions at 14 TeV center of mass energy with both $W$'s decayed using Madgraph/Madevent v4.4.32 [ref] , which includes the full $2→4$ matrix elements.

    MadGraph/MadEvent v4: The New Web Generation · Event Samples · 1.00

Cites as background 25 papers

Discussion

Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.