Pith. sign in

REVIEW 2 major objections 6 minor 92 references

This paper reports the first ATLAS search for a heavy scalar decaying into a Higgs boson plus a lighter scalar, both decaying to b-quark pairs; no excess is found and 95% confidence-level limits are set.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 15:17 UTC pith:TOM24BNO

load-bearing objection First ATLAS X→SH→4b search, a careful null result with a genuinely new GP+flow background model; referees should push for quantitative support on the one-sentence dismissal of resonant subleading backgrounds. the 2 major comments →

arxiv 2607.18484 v1 pith:TOM24BNO submitted 2026-07-20 hep-ex

Search for new scalars via X rightarrow SH rightarrow bbar{b}bbar{b} in proton-proton collisions at sqrt{s} = 13 TeV with the ATLAS detector

ATLAS Collaboration This is my paper
classification hep-ex
keywords X→SH→4bheavy scalar resonanceHiggs bosonbottom-quark jetsGaussian processnormalizing flowdata-driven background estimationLHC Run 2
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish whether a heavy scalar resonance X, produced by gluon fusion, decays into a Standard Model Higgs boson H and a new lighter scalar S, with each decaying into a bottom-antibottom pair. It finds no significant excess above the expected background and sets 95% confidence-level upper limits on the production rate, from 0.7 fb to 2.6 pb, scanning X masses from 300 GeV to 6 TeV and S masses from 70 GeV to 5 TeV. A sympathetic reader should care because extended Higgs sectors predict this asymmetric decay chain, and these are the first ATLAS constraints on it, reaching the maximum rate allowed by the Two-Real-Singlet benchmark for the first time in this final state. The analysis also introduces a fully data-driven background estimate, using Gaussian processes for the Higgs mass shape and a normalizing flow for the correlated kinematics, which reduces the dominant background uncertainty by about a factor of two relative to earlier control-region approaches.

Core claim

The central claim is that, using 126–136 fb^-1 of 13 TeV proton–proton collisions, the process pp→X→SH→4b is not observed; the data are consistent with the Standard Model background. The search is split into a resolved channel, where both bosons are paired from small-radius b-tagged jets, and a mixed channel, where the boosted Higgs is a single large-radius jet and S is resolved. For the first time in ATLAS, asymmetric scalar decays in this final state are constrained, with the strongest limits (a few fb) at high X masses, where the multijet background falls rapidly. The observed limits disagree with an excess previously reported by CMS near (mX, mS) = (600, 400) GeV: the closest ATLAS point

What carries the argument

The load-bearing object is the joint density p(x, mH) = p_GP(mH) · p_flow(x | mH). A Gaussian process with a squared-exponential kernel is fitted to the reconstructed Higgs mass in sideband control regions, giving the smooth background shape and normalization. A rational-quadratic neural spline flow, conditioned on mH and data-taking year, learns the correlated kinematics of the S and H candidates (transverse momenta, pseudorapidities, azimuthal separation, reconstructed S mass) and transfers those correlations into the blinded signal region. Sampling the GP posterior and conditioning the flow on those samples produces the final discriminant, the reconstructed X mass. The validity of the int

Load-bearing premise

The background in the blinded signal region is smooth enough that a Gaussian process fitted to the Higgs-mass sidebands and a normalizing flow conditioned on that mass can extrapolate without bias; the paper states that this interpolation validity is the key assumption of the method.

What would settle it

Inject a narrow resonant background component (for example Z→bb or single-Higgs production) inside the blinded Higgs-mass window of a background-only simulation, train the GP plus flow on the sidebands only, and check whether a spurious peak appears in the reconstructed X mass distribution at the level of the quoted limits.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Any model predicting a production rate above the quoted 95% CL contour in the covered (mX, mS) grid is excluded.
  • The maximally allowed Two-Real-Singlet-Model rate is approached for the first time in ATLAS; where the prediction lies above the limit, corresponding parameter points are excluded.
  • The excess previously reported by CMS near (600, 400) GeV is disfavored: at the nearest ATLAS mass hypothesis, with about 70% of the dedicated sensitivity, the observed limit is stronger.
  • The largest ATLAS fluctuation, at (300, 170) GeV with local significance 1.8 sigma, is consistent with background and would need more data to be assessed.
  • The interpolation-based background method cuts the non-closure uncertainty by roughly a factor of two relative to earlier control-region approaches, making the four-b final state competitive for future searches.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The fully boosted topology, where both H and S are reconstructed as large-radius jets, is explicitly left for future work; a dedicated search there could extend coverage into the high-mX, high-mS region where the present limits are weakest.
  • The GP-plus-flow machinery is transferable to other hadronic resonance searches that need sideband-to-signal interpolation, but its value depends on whether validation regions capture all differences between sidebands and the signal region.
  • With Run 3 data at higher luminosity, the same pipeline should either confirm or refute the (300, 170) GeV fluctuation and push limits below the Two-Real-Singlet benchmark across the plane; the present analysis lays out that test.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper presents a search for a heavy scalar resonance X decaying into a Standard Model Higgs boson H and an additional scalar S, with both bosons reconstructed in the bbbar bbbar final state. The analysis uses ATLAS Run 2 pp collision data at sqrt(s)=13 TeV, with an integrated luminosity of 126-136 fb^-1, and covers mX from 300 GeV to 6 TeV and mS from 70 GeV to 5 TeV in two topologies: resolved (two small-R b-jets for each boson) and mixed (boosted large-R jet for H, resolved b-jets for S). The dominant multijet and ttbar backgrounds are estimated with a data-driven hierarchical model in which a Gaussian process interpolates the mH_reco sidebands into the blinded signal region and a normalizing flow conditioned on mH_reco generates the remaining event kinematics. Validation is performed in orthogonal 3b/1b and dedicated kinematic validation regions, and a non-closure uncertainty is propagated. No significant excess over the background-only hypothesis is found; the largest local excess is 1.8 sigma at (mX,mS)=(300,170) GeV. Upper limits at 95% CL on sigma x B(pp->X->SH->4b) are set, ranging from 0.7 fb to 2.6 pb, and are compared with maximal TRSM benchmark predictions.

Significance. If correct, this result provides the first ATLAS constraints on the X->SH->4b signature, covering a broad region of the (mX,mS) plane and approaching the maximally allowed TRSM rate for the first time in this final state. The analysis has clear strengths: the signal region was blinded during background estimation; the background prediction is an extrapolation from sidebands with no fit to signal; validation uses orthogonal 3b/1b regions feeding a non-closure uncertainty; signal injection tests reportedly retain the injected signal; the 25-model flow ensembles propagate training uncertainty; and the systematic menu is complete, with background modeling dominating the resolved-channel limits (44-73% impact). The GP+normalizing-flow interpolation is a novel and potentially useful technique. The principal risk is openly acknowledged in Sec. 6: the validity of the interpolation is the key assumption of the method. The main weakness is that one resonant subleading background category is dismissed with an unsupported statement, which is load-bearing for the interpolation assumption.

major comments (2)
  1. [Section 6, subleading-background paragraph] The paper dismisses Z->bb and single-Higgs->bb backgrounds with the sentence: "Studies with MC simulation show that these processes contribute only at the percent level and their resonant shape does not introduce any localized bias in the background estimate, hence their overall impact is negligible." This claim is load-bearing because the GP is trained only on the mH_reco sidebands (59.3-107.8 and 148.8-190.7 GeV resolved; 70-110 and 140-200 GeV mixed), while a resonant single-H or Z->bb component would peak in exactly the blinded SR window around 125 GeV and would be invisible to the training data. The paper needs to provide quantitative support: MC event yields after the full selection for these processes in the CRs and the SR, and a demonstration that the non-closure uncertainty covers such a localized component. A closure test with an injected H-like component would directly address
  2. [Sections 7.2 and 7.3, validation regions and non-closure scaling] The non-closure uncertainty is derived from the 3b/1b validation regions (Sec. 5) and from the Delta-eta/GN2X-loose kinematic VRs (Sec. 7.3), then propagated to the SR by scaling by the ratio of expected yields. This assumes the fractional mismodeling is the same in the VRs and the SR. The 3b/1b regions alter the b-jet content, and the kinematic VRs either invert Delta-eta or use a different large-R-jet tag; neither directly reproduces the resonant single-H or Z->bb contribution in the true 4b SR. Please justify the similarity quantitatively, e.g., by comparing MC background compositions in the VRs and SR, or add an additional uncertainty based on an orthogonal closure test. As written, the claim that the key interpolation assumption is covered by the assigned uncertainties is not fully demonstrated.
minor comments (6)
  1. [Section 7.3] The section says "Each analysis regime employs a VR" and describes only the Delta-eta and GN2X-loose regions, which appears inconsistent with the 3b/1b validation regions defined in Secs. 5.1 and 5.2 and used for the non-closure uncertainty in Sec. 7.2. Please clarify the relationship between the two sets of validation regions.
  2. [Section 3.1] In the trigger descriptions, "seeded with L1 jets within eta<2.5" and similar expressions should read |eta|<2.5 for consistency with the rest of the paper.
  3. [Table 3 caption] The caption states "The vacuum expectation values are called vS and vS"; the second should be vX.
  4. [Section 7.3] Typo: "discrepencies" should be "discrepancies."
  5. [Section 6.1] The statement "The only hyperparameter is the length scale ell" refers only to the GP; since the normalizing flow has many hyperparameters, please rephrase to avoid ambiguity.
  6. [Figure 5] The caption states that lighter colors indicate the CR and darker colors the SR, but the legend and shading in the figure are not easy to distinguish. Please make the CR/SR distinction visually explicit.

Circularity Check

0 steps flagged

No significant circularity: the background is a blinded sideband extrapolation, signal shapes come from independent MC, and the TRSM benchmark is external; the interpolation assumption is a stated limitation, not a circular reduction.

full rationale

The derivation chain is self-contained with respect to the central claim. The signal region is blinded during background estimation: the GP is fit only to the mH_reco sidebands (59.3–107.8 and 148.8–190.7 GeV resolved; 70–110 and 140–200 GeV mixed), and the normalizing flow is trained on CR kinematics conditioned on mH_reco and year. The SR yield is then obtained by sampling the GP posterior in the blinded region and normalizing with n_SR = n_CR^obs * n_SR^pred / n_CR^pred. This is a standard sideband interpolation, not a fit to signal: the GP length scale and flow weights are background-calibration parameters, and no signal-strength or signal-template parameter is extracted from the SR before the final profile-likelihood fit. The SR definitions (m'_S, r_f,S from the X_SH variable) are determined from simulated signal events, not from data. The signal templates are independent MC, and the limits are set by fitting these templates plus the extrapolated background to the mX_reco distribution. The paper explicitly flags the interpolation as an assumption ('The validity of the interpolation constitutes the key assumption of the method') and derives non-closure uncertainties from the 3b/1b validation regions and signal-injection tests; whether that assumption is adequate for localized subleading backgrounds (single-Higgs, Z->bb) is a modeling/closure risk, not a circularity. Self-citations to earlier ATLAS analyses (e.g., Refs. [11,85]) support the GP technique and the X_SH-inspired SR definition, but they are not load-bearing: the present closure checks stand independently. The TRSM benchmark (Ref. [48]) is an external parameter scan used only for comparison with the limits, and the limits do not depend on it. No prediction reduces by construction to a fitted input, so no circular step is identified.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

No new entities are postulated: X and S are the standard benchmark scalars of TRSM/2HDM/NMSSM (Refs. [6,16,17]), and the TRSM scan itself is external (Ref. [48]). The central load-bearing inputs are the background-smoothness/interpolation assumption, the simplified signal MC templates, and calibrated detector inputs. Free parameters are background-model and signal-region definitions rather than physics constants.

free parameters (6)
  • GP length scale ℓ = maximized log-likelihood per (year, background estimate)
    Squared-exponential kernel length scale in the mH_reco interpolation (Sec. 6.1); fit to control-region data per year and background estimate; sets how smooth the extrapolation into the SR must be.
  • Normalizing-flow hyperparameters = 10 layers, K=4 spline bins, 32 hidden ResNet units, dropout 0.1, L2 1e-6, lr 1e-3, 25 trainings
    Hand-chosen in Sec. 6.2 to balance expressivity and overfitting; the ensemble spread drives the dominant resolved-channel systematic (36–63% impact on limits).
  • Signal-region resolution factors (m'S, rf,S; m'H and rf,H fixed) = m'H = 125 GeV, rf,H = 0.1; m'S and rf,S derived from signal MC per mS hypothesis
    Define the XrecoSH < 1.6 SR contour (Eq. 1, Sec. 5.1). Derived from simulated signal, not data, but they determine which events enter the fit.
  • Online b-tagging normalization uncertainty = 20%
    Ad hoc conservative uncertainty justified by 'checks for a few representative signal samples' (Sec. 7.1); covers preliminary online b-tagging calibration differences.
  • Mixed-channel mS_reco SR windows = 99% probability interval of a skew-normal fit to signal mS_reco per truth mS
    Per-(mX, mS) selection windows (Sec. 5.2) fit to signal simulation and applied to data; affect which events are counted for each hypothesis.
  • Autocorrelation threshold for PCA background-shape split = 0.6
    Hand-chosen cutoff (Sec. 7.2) separating correlated PCA shape components from bin-by-bin statistical components in the resolved-channel background templates.
axioms (6)
  • domain assumption The background in the SR is a smooth function of mH_reco and the flow inputs, so GP+NF trained on CR sidebands extrapolates without bias; the interpolation is 'the key assumption of the method'.
    Sec. 6, explicitly stated by the authors; guarded only by VR non-closure uncertainties (Sec. 7.2).
  • domain assumption Signal model: narrow-width scalar X (Γ = 10 MeV) produced by gluon–gluon fusion with B(X→SH→4b) = 1; LO Pythia8.306 + AtlFast3 fast simulation provides unbiased template shapes.
    Sec. 3.2; a simplified-model assumption standard for reinterpretation but a real restriction on the generality of the limits.
  • domain assumption SM inputs: BR(H→bb) ≈ 59%, ΓH ≈ 3.8 MeV (Appendix A), background composition (multijet ≈ 90%, ttbar, Z→bb and single-Higgs sub-percent and non-localized).
    Sec. 6 intro and Appendix A; the Z→bb/single-H claim is asserted from MC studies without shown detail.
  • standard math Asymptotic profile-likelihood / CLs formalism is valid with ≥ 10 expected events per bin.
    Sec. 8.1; rebinning enforces the ≥10 events/bin condition to justify Cowan–Cranmer–Gross–Vitells asymptotics.
  • standard math Gaussian-process conditioning and normalizing-flow change-of-variables provide valid conditional density estimation with tractable uncertainties.
    Secs. 6.1–6.2; the ML background is taken from the cited literature (Rasmussen–Williams; Durkan et al.; Rezende–Mohamed).
  • domain assumption Calibration transfers (trigger SFs from di-leptonic ttbar, GN2/GN2X tagger calibrations, jet energy/mass calibrations) are unbiased for SR kinematics.
    Sec. 7.1; standard ATLAS calibration chain applied to signal templates and validation-region transfers.

pith-pipeline@v1.3.0-alltime-deepseek · 59973 in / 25272 out tokens · 230912 ms · 2026-08-01T15:17:58.838538+00:00 · methodology

0 comments
read the original abstract

A search is performed for a scalar resonance X decaying into a Standard Model Higgs boson H and an additional scalar S, each decaying into a bottom-antibottom-quark pair, giving a $b\bar{b}b\bar{b}$ final state. Two event topologies are considered, depending on whether the Higgs boson is reconstructed from a single large-radius jet or from a pair of small-radius jets. The analysis uses proton-proton collision data at $\sqrt{s} = 13$ TeV, collected by the ATLAS experiment at the Large Hadron Collider during Run 2, corresponding to an integrated luminosity of 126-136 fb$^{-1}$ depending on the topology and the associated triggers. The search covers masses of 300 GeV-6 TeV for X and 70 GeV-5 TeV for S. The dominant multijet and $t\bar{t}$ backgrounds are estimated with a hierarchical model using Gaussian processes and normalizing flows. No significant excess over the Standard Model background expectation is observed. Upper limits at 95% confidence level are set on the production cross-section times branching ratio, ranging from 0.7 fb to 2.6 pb.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

92 extracted references · 1 canonical work pages

  1. [1]

    ATLAS Collaboration,Observation of a new particle in the search for the Standard Model Higgs boson with the ATLAS detector at the LHC, Phys. Lett. B716(2012) 1, arXiv:1207.7214 [hep-ex]

    Pith/arXiv arXiv 2012

  2. [2]

    CMS Collaboration, Observation of a new boson at a mass of 125 GeV with the CMS experiment at the LHC, Phys. Lett. B716(2012) 30, arXiv:1207.7235 [hep-ex]

    Pith/arXiv arXiv 2012

  3. [3]

    ATLAS Collaboration, A detailed map of Higgs boson interactions by the ATLAS experiment ten years after the discovery, Nature607(2022) 52, arXiv:2207.00092 [hep-ex], Erratum: Nature612(2022) E24

    Pith/arXiv arXiv 2022

  4. [4]

    CMS Collaboration, A portrait of the Higgs boson by the CMS experiment ten years after the discovery, Nature607(2022) 60, arXiv:2207.00043 [hep-ex], Erratum: Nature623(2023) E4

    Pith/arXiv arXiv 2022

  5. [5]

    Bosse et al.,Constraining the real scalar singlet extension of the SM, (2026), arXiv:2606.12533 [hep-ph]

    M. Bosse et al.,Constraining the real scalar singlet extension of the SM, (2026), arXiv:2606.12533 [hep-ph]

    Pith/arXiv arXiv 2026

  6. [6]

    Branco et al.,Theory and phenomenology of two-Higgs-doublet models, Phys

    G. Branco et al.,Theory and phenomenology of two-Higgs-doublet models, Phys. Rept.516(2012) 1, arXiv:1106.0034 [hep-ph]

    Pith/arXiv arXiv 2012

  7. [7]

    Belyaev, G

    A. Belyaev, G. Cacciapaglia, I. P. Ivanov, F. Rojas-Abatte, and M. Thomas,Anatomy of the inert two-Higgs-doublet model in the light of the LHC and non-LHC dark matter searches, Phys. Rev. D97(2018) 035011, arXiv:1612.00511 [hep-ph]

    Pith/arXiv arXiv 2018

  8. [8]

    Arhrib, R

    A. Arhrib, R. Benbrik, M. El Kacimi, L. Rahili, and S. Semlali, Extended Higgs sector of 2HDM with real singlet facing LHC data, Eur. Phys. J. C80(2020) 13, arXiv:1811.12431 [hep-ph]

    Pith/arXiv arXiv 2020

  9. [9]

    Darvishi, M

    N. Darvishi, M. R. Masouminia, and A. Pilaftsis,Maximally symmetric three-Higgs-doublet model, Phys. Rev. D104(2021) 115017, arXiv:2106.03159 [hep-ph]

    Pith/arXiv arXiv 2021

  10. [10]

    Georgi and M

    H. Georgi and M. Machacek,Doubly charged Higgs bosons, Nucl. Phys. B262(1985) 463

    1985

  11. [11]

    ATLAS Collaboration,Search for resonant pair production of Higgs bosons in the𝑏 ¯𝑏𝑏 ¯𝑏 final state using𝑝𝑝collisions at √𝑠=13TeV with the ATLAS detector, Phys. Rev. D105(2022) 092002, arXiv:2202.07288 [hep-ex]

    Pith/arXiv arXiv 2022

  12. [12]

    ATLAS Collaboration,Search for pair production of boosted Higgs bosons via vector-boson fusion in the𝑏 ¯𝑏𝑏 ¯𝑏final state using𝑝𝑝collisions at√𝑠=13TeV with the ATLAS detector, Phys. Lett. B858(2024) 139007, arXiv:2404.17193 [hep-ex]

    Pith/arXiv arXiv 2024

  13. [13]

    CMS Collaboration,Search for resonant pair production of Higgs bosons in the𝑏¯𝑏𝑏 ¯𝑏final state using large-area jets in proton–proton collisions at√𝑠=13TeV, JHEP02(2025) 040, arXiv:2407.13872 [hep-ex]

    Pith/arXiv arXiv 2025

  14. [14]

    CMS Collaboration,Search for a massive scalar resonance decaying to a light scalar and a Higgs boson in the four𝑏quarks final state with boosted topology, Phys. Lett. B842(2023) 137392, arXiv:2204.12413 [hep-ex]

    Pith/arXiv arXiv 2023

  15. [15]

    CMS Collaboration,Improved results on Higgs boson pair production in the4𝑏 final state, (2026), arXiv:2604.27044 [hep-ex]. 29

    Pith/arXiv arXiv 2026

  16. [16]

    Robens, T

    T. Robens, T. Stefaniak, and J. Wittbrodt, Two-real-scalar-singlet extension of the SM: LHC phenomenology and benchmark scenarios, Eur. Phys. J. C80(2020) 151, arXiv:1908.08554 [hep-ph]

    Pith/arXiv arXiv 2020

  17. [17]

    Ellwanger, C

    U. Ellwanger, C. Hugonie, and A. M. Teixeira, The Next-to-Minimal Supersymmetric Standard Model, Phys. Rept.496(2010) 1, arXiv:0910.1785 [hep-ph]

    Pith/arXiv arXiv 2010

  18. [18]

    CMS Collaboration,Search for a new heavy scalar resonance decaying into the Higgs boson and a new scalar particle in the𝑏¯𝑏𝑏 ¯𝑏final state using proton–proton collisions at√𝑠=13TeV, (2026), arXiv:2605.02848 [hep-ex]

    Pith/arXiv arXiv 2026

  19. [19]

    CMS Collaboration,Search for heavy scalar resonances decaying to Lorentz-boosted Higgs and Higgs-like bosons in the𝑏¯𝑏4𝑞final state at√𝑠=13TeV, (2026), arXiv:2602.00273 [hep-ex]

    arXiv 2026

  20. [20]

    CMS Collaboration, Search for a new scalar resonance decaying to a Higgs boson and another new scalar particle in the final state with two bottom quarks and two photons in proton–proton collisions at√𝑠=13TeV, JHEP12(2025) 178, arXiv:2508.11494 [hep-ex]

    arXiv 2025

  21. [21]

    CMS Collaboration,Search for a new resonance decaying to a Higgs boson and a scalar boson in events with two𝑏jets and two𝑍bosons in proton–proton collisions at√𝑠=13.6TeV, (2026), arXiv:2602.18223 [hep-ex]

    arXiv 2026

  22. [22]

    ATLAS Collaboration,Search for triple Higgs boson production in the6𝑏final state using𝑝𝑝 collisions at√𝑠=13TeV with the ATLAS detector, Phys. Rev. D111(2025) 032006, arXiv:2411.02040 [hep-ex]

    Pith/arXiv arXiv 2025

  23. [23]

    ATLAS Collaboration, Search for a new heavy scalar particle decaying into a Higgs boson and a new scalar singlet in final states with one or two light leptons and a pair of𝜏-leptons with the ATLAS detector, JHEP10(2023) 009, arXiv:2307.11120 [hep-ex]

    Pith/arXiv arXiv 2023

  24. [24]

    ATLAS Collaboration,Search for a resonance decaying into a scalar particle and a Higgs boson in the final state with two bottom quarks and two photons with199fb−1 of data collected at√𝑠=13 and13.6TeV with the ATLAS detector, Phys. Lett. B877(2026) 140425, arXiv:2510.02857 [hep-ex]

    Pith/arXiv arXiv 2026

  25. [25]

    ATLAS Collaboration, Search for a resonance decaying into a scalar particle and a Higgs boson in final states with leptons and two photons in proton–proton collisions at√𝑠=13TeV with the ATLAS detector, JHEP10(2024) 104, arXiv:2405.20926 [hep-ex]

    Pith/arXiv arXiv 2024

  26. [26]

    Vaswani et al.,Attention Is All You Need, 2017, arXiv:1706.03762 [cs.CL]

    A. Vaswani et al.,Attention Is All You Need, 2017, arXiv:1706.03762 [cs.CL]

    Pith/arXiv arXiv 2017

  27. [27]

    ATLAS Collaboration,Transforming jet flavour tagging at ATLAS, Nature Commun.17(2026) 541, arXiv:2505.19689 [hep-ex]

    arXiv 2026

  28. [28]

    ATLAS Collaboration,Transformer Neural Networks for Identifying Boosted Higgs Bosons decaying into𝑏 ¯𝑏and𝑐¯𝑐in ATLAS, ATL-PHYS-PUB-2023-021, 2023, url:https://cds.cern.ch/record/2866601

    arXiv 2023

  29. [29]

    C. E. Rasmussen and C. K. I. Williams,Gaussian Processes for Machine Learning, The MIT Press, 2005,isbn: 9780262256834. 30

    2005

  30. [30]

    Daumann et al.,One Flow to Correct Them all: Improving Simulations in High-Energy Physics with a Single Normalising Flow and a Switch, Comput

    C. Daumann et al.,One Flow to Correct Them all: Improving Simulations in High-Energy Physics with a Single Normalising Flow and a Switch, Comput. Softw. Big Sci.8(2024), arXiv:2403.18582 [hep-ph]

    Pith/arXiv arXiv 2024

  31. [31]

    ATLAS Collaboration,The ATLAS Experiment at the CERN Large Hadron Collider, JINST3(2008) S08003

    2008

  32. [32]

    ATLAS Collaboration,ATLAS Insertable B-Layer: Technical Design Report, ATLAS-TDR-19; CERN-LHCC-2010-013, 2010, url:https://cds.cern.ch/record/1291633, Addendum: ATLAS-TDR-19-ADD-1; CERN-LHCC-2012-009, 2012,url:https://cds.cern.ch/record/1451888

    arXiv 2010

  33. [33]

    Abbott et al.,Production and integration of the ATLAS Insertable B-Layer, JINST13(2018) T05008, arXiv:1803.00844 [physics.ins-det]

    B. Abbott et al.,Production and integration of the ATLAS Insertable B-Layer, JINST13(2018) T05008, arXiv:1803.00844 [physics.ins-det]

    Pith/arXiv arXiv 2018

  34. [34]

    Avoni et al.,The new LUCID-2 detector for luminosity measurement and monitoring in ATLAS, JINST13(2018) P07017

    G. Avoni et al.,The new LUCID-2 detector for luminosity measurement and monitoring in ATLAS, JINST13(2018) P07017

    2018

  35. [35]

    ATLAS Collaboration,Performance of the ATLAS trigger system in 2015, Eur. Phys. J. C77(2017) 317, arXiv:1611.09661 [hep-ex]

    Pith/arXiv arXiv 2015

  36. [36]

    ATLAS Collaboration,Software and computing for Run 3 of the ATLAS experiment at the LHC, Eur. Phys. J. C85(2025) 234, arXiv:2404.06335 [hep-ex], Erratum: Eur. Phys. J. C85(2025) 907

    Pith/arXiv arXiv 2025

  37. [37]

    ATLAS Collaboration, Luminosity determination in𝑝𝑝collisions at√𝑠=13TeV using the ATLAS detector at the LHC, Eur. Phys. J. C83(2023) 982, arXiv:2212.09379 [hep-ex]

    Pith/arXiv arXiv 2023

  38. [38]

    ATLAS Collaboration,ATLAS data quality operations and performance for 2015–2018 data-taking, JINST15(2020) P04003, arXiv:1911.04632 [physics.ins-det]

    Pith/arXiv arXiv 2015

  39. [39]

    Cacciari, G

    M. Cacciari, G. P. Salam, and G. Soyez,The anti-𝑘𝑡 jet clustering algorithm, JHEP04(2008) 063, arXiv:0802.1189 [hep-ph]

    Pith/arXiv arXiv 2008

  40. [40]

    Cacciari, G

    M. Cacciari, G. P. Salam, and G. Soyez,FastJet user manual, Eur. Phys. J. C72(2012) 1896, arXiv:1111.6097 [hep-ph]

    Pith/arXiv arXiv 2012

  41. [41]

    ATLAS Collaboration,Configuration and performance of the ATLAS𝑏-jet triggers in Run 2, Eur. Phys. J. C81(2021) 1087, arXiv:2106.03584 [hep-ex]

    Pith/arXiv arXiv 2021

  42. [42]

    Bierlich et al.,A comprehensive guide to the physics and usage of PYTHIA 8.3, SciPost Phys

    C. Bierlich et al.,A comprehensive guide to the physics and usage of PYTHIA 8.3, SciPost Phys. Codebases (2022) 8, arXiv:2203.11601 [hep-ph]

    Pith/arXiv arXiv 2022

  43. [43]

    D. J. Lange,The EvtGen particle decay simulation package, Nucl. Instrum. Meth. A462(2001) 152

    2001

  44. [44]

    NNPDF Collaboration, R. D. Ball, et al.,Parton distributions with LHC data, Nucl. Phys. B867(2013) 244, arXiv:1207.1303 [hep-ph]

    Pith/arXiv arXiv 2013

  45. [45]

    ATLAS Collaboration,ATLAS Pythia 8 tunes to7TeV data, ATL-PHYS-PUB-2014-021, 2014, url:https://cds.cern.ch/record/1966419

    arXiv 2014

  46. [46]

    ATLAS Collaboration,AtlFast3: The Next Generation of Fast Simulation in ATLAS, Comput. Softw. Big Sci.6(2022) 7, arXiv:2109.02551 [hep-ex]

    Pith/arXiv arXiv 2022

  47. [47]

    Higgs Properties, (2013), arXiv:1307.1347 [hep-ph]

    LHC Higgs Cross Section Working Group, Handbook of LHC Higgs Cross Sections: 3. Higgs Properties, (2013), arXiv:1307.1347 [hep-ph]. 31

    Pith/arXiv arXiv 2013

  48. [48]

    Robens, R

    T. Robens, R. Rodriguez, M. Samaniego, and J. Veatch,TRSMScans, 2026, arXiv:2607.11571 [hep-ph]

    Pith/arXiv arXiv 2026

  49. [49]

    Bahl et al., HiggsTools: BSM scalar phenomenology with new versions of HiggsBounds and HiggsSignals, Comput

    H. Bahl et al., HiggsTools: BSM scalar phenomenology with new versions of HiggsBounds and HiggsSignals, Comput. Phys. Commun.291(2023) 108803, arXiv:2210.09332 [hep-ph]

    Pith/arXiv arXiv 2023

  50. [50]

    ATLAS Collaboration,Reconstruction of primary vertices at the ATLAS experiment in Run 1 proton–proton collisions at the LHC, Eur. Phys. J. C77(2017) 332, arXiv:1611.10235 [physics.ins-det]

    Pith/arXiv arXiv 2017

  51. [51]

    ATLAS Collaboration, Jet reconstruction and performance using particle flow with the ATLAS Detector, Eur. Phys. J. C77(2017) 466, arXiv:1703.10485 [hep-ex]

    Pith/arXiv arXiv 2017

  52. [52]

    ATLAS Collaboration, Topological cell clustering in the ATLAS calorimeters and its performance in LHC Run 1, Eur. Phys. J. C77(2017) 490, arXiv:1603.02934 [hep-ex]

    Pith/arXiv arXiv 2017

  53. [53]

    ATLAS Collaboration,Jet energy scale and resolution measured in proton–proton collisions at√𝑠=13TeV with the ATLAS detector, Eur. Phys. J. C81(2021) 689, arXiv:2007.02645 [hep-ex]

    Pith/arXiv arXiv 2021

  54. [54]

    ATLAS Collaboration, Selection of jets produced in13TeV proton–proton collisions with the ATLAS detector, ATLAS-CONF-2015-029, 2015,url:https://cds.cern.ch/record/2037702

    arXiv 2015

  55. [55]

    ATLAS Collaboration,Performance of pile-up mitigation techniques for jets in𝑝𝑝collisions at√𝑠=8TeV using the ATLAS detector, Eur. Phys. J. C76(2016) 581, arXiv:1510.03823 [hep-ex]

    Pith/arXiv arXiv 2016

  56. [56]

    ATLAS Collaboration,ATLAS flavour-tagging algorithms for the LHC Run 2𝑝𝑝 collision dataset, Eur. Phys. J. C83(2023) 681, arXiv:2211.16345 [physics.data-an]

    Pith/arXiv arXiv 2023

  57. [57]

    ATLAS Collaboration, Calibration of the light-flavour jet mistagging efficiency of the𝑏-tagging algorithms with𝑍+jets events using139fb−1 of ATLAS proton–proton collision data at√𝑠=13TeV, Eur. Phys. J. C83(2023) 728, arXiv:2301.06319 [hep-ex]

    Pith/arXiv arXiv 2023

  58. [58]

    ATLAS Collaboration,Measurement of the𝑐-jet mistagging efficiency in𝑡¯𝑡events using𝑝𝑝 collision data at√𝑠=13TeV collected with the ATLAS detector, Eur. Phys. J. C82(2022) 95, arXiv:2109.10627 [hep-ex]

    Pith/arXiv arXiv 2022

  59. [59]

    ATLAS Collaboration,ATLAS𝑏-jet identification performance and efficiency measurement with𝑡¯𝑡 events in𝑝𝑝collisions at√𝑠=13TeV, Eur. Phys. J. C79(2019) 970, arXiv:1907.05120 [hep-ex]

    Pith/arXiv arXiv 2019

  60. [60]

    ATLAS Collaboration,Optimisation of large-radius jet reconstruction for the ATLAS detector in 13TeV proton–proton collisions, Eur. Phys. J. C81(2021) 334, arXiv:2009.04986 [hep-ex]

    Pith/arXiv arXiv 2021

  61. [61]

    ATLAS Collaboration,Improving jet substructure performance in ATLAS using Track-CaloClusters, ATL-PHYS-PUB-2017-015, 2017,url:https://cds.cern.ch/record/2275636

    arXiv 2017

  62. [62]

    A. J. Larkoski, S. Marzani, G. Soyez, and J. Thaler,Soft Drop, JHEP05(2014) 146, arXiv:1402.2657 [hep-ph]. 32

    Pith/arXiv arXiv 2014

  63. [63]

    ATLAS Collaboration,In situ calibration of large-radius jet energy and mass in13TeV proton–proton collisions with the ATLAS detector, Eur. Phys. J. C79(2019) 135, arXiv:1807.09477 [hep-ex]

    Pith/arXiv arXiv 2019

  64. [64]

    ATLAS Collaboration,Multijet simulation for13TeVATLAS Analyses, ATL-PHYS-PUB-2019-017, 2019,url:https://cds.cern.ch/record/2672252

    arXiv 2019

  65. [65]

    ATLAS Collaboration,Efficiency corrections for a tagger for boosted𝐻→𝑏¯𝑏decays in𝑝𝑝 collisions at√𝑠=13TeV with the ATLAS detector, ATL-PHYS-PUB-2021-035, 2021, url:https://cds.cern.ch/record/2777811

    arXiv 2021

  66. [66]

    Cacciari and G

    M. Cacciari and G. P. Salam,Pileup subtraction using jet areas, Phys. Lett. B659(2008) 119, arXiv:0707.1378 [hep-ph]

    Pith/arXiv arXiv 2008

  67. [67]

    ATLAS Collaboration,easyjet analysis framework, version 1.0.0, 2025, url:https://doi.org/10.5281/zenodo.15927813

    work page doi:10.5281/zenodo.15927813 2025

  68. [68]

    Hoecker et al.,TMVA - Toolkit for Multivariate Data Analysis, 2009, arXiv:physics/0703039 [physics.data-an]

    A. Hoecker et al.,TMVA - Toolkit for Multivariate Data Analysis, 2009, arXiv:physics/0703039 [physics.data-an]

    Pith/arXiv arXiv 2009

  69. [69]

    ATLAS Collaboration, Measurements of the Higgs boson inclusive and differential fiducial cross-sections in the diphoton decay channel with𝑝𝑝collisions at√𝑠=13TeV with the ATLAS detector, JHEP08(2022) 027, arXiv:2202.00487 [hep-ex]

    Pith/arXiv arXiv 2022

  70. [70]

    ATLAS Collaboration,Recommendations for the Modeling of Smooth Backgrounds, ATL-PHYS-PUB-2020-028, 2020,url:https://cds.cern.ch/record/2743717

    arXiv 2020

  71. [71]

    ATLAS Collaboration,Search for the associated production of charm quarks and a Higgs boson decaying into a photon pair with the ATLAS detector, JHEP02(2025) 045, arXiv:2407.15550 [hep-ex]

    Pith/arXiv arXiv 2025

  72. [72]

    ATLAS Collaboration,Search for boosted diphoton resonances in the10to70GeV mass range using138fb −1 of13TeV𝑝𝑝collisions with the ATLAS detector, JHEP07(2023) 155, arXiv:2211.04172 [hep-ex]

    Pith/arXiv arXiv 2023

  73. [73]

    ATLAS Collaboration,Search for dimuon resonance in the 35 to 75 GeV mass range using140fb−1 of13TeV𝑝𝑝collisions with the ATLAS detector, JHEP06(2026) 165, arXiv:2601.21361 [hep-ex]

    Pith/arXiv arXiv 2026

  74. [74]

    Pedregosa et al.,Scikit-learn: Machine Learning in Python, J

    F. Pedregosa et al.,Scikit-learn: Machine Learning in Python, J. Machine Learning Res.12(2011) 2825, arXiv:1201.0490 [cs.LG]. [75]Inverse transform sampling, wikipedia, url:https://en.wikipedia.org/wiki/Inverse_transform_sampling

    Pith/arXiv arXiv 2011

  75. [76]

    Jimenez Rezende and S

    D. Jimenez Rezende and S. Mohamed,Variational Inference with Normalizing Flows, 2015, arXiv:1505.05770 [stat.ML]

    Pith/arXiv arXiv 2015

  76. [77]

    Papamakarios,Neural Density Estimation and Likelihood-free Inference, 2019, arXiv:1910.13233 [stat.ML]

    G. Papamakarios,Neural Density Estimation and Likelihood-free Inference, 2019, arXiv:1910.13233 [stat.ML]

    Pith/arXiv arXiv 2019

  77. [78]

    Papamakarios, E

    G. Papamakarios, E. Nalisnick, D. Jimenez Rezende, S. Mohamed, and B. Lakshminarayanan, Normalizing Flows for Probabilistic Modeling and Inference, J. Machine Learning Res.22(2021) 2617, arXiv:1912.02762 [stat.ML]

    Pith/arXiv arXiv 2021

  78. [79]

    Durkan, A

    C. Durkan, A. Bekasov, I. Murray, and G. Papamakarios,Neural Spline Flows, 2019, arXiv:1906.04032 [stat.ML]. 33

    Pith/arXiv arXiv 2019

  79. [80]

    K. He, X. Zhang, S. Ren, and J. Sun,Deep Residual Learning for Image Recognition, 2015, arXiv:1512.03385 [cs.CV]

    Pith/arXiv arXiv 2015

  80. [81]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, Dropout: A Simple Way to Prevent Neural Networks from Overfitting, J. Machine Learning Res.15(2014) 1929

    2014

Showing first 80 references.