Search for Vector-Like Singlet Top (T) Quark in a Future Muon-Proton (mu p) Collider at sqrt s = 5.29, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: High Energy Physics: Phenomenology. The day in particle phenomenology: models, signatures and constraints.
Kai: We're looking at the paper: "Search for Vector-Like Singlet Top (T) Quark in a Future Muon-Proton (mu p) Collider at sqrt s = 5.29, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures".
Mira: Here's the gist of the paper. We study the sensitivity of a future muon-proton collider to singly produced singlet vector-like T quarks in the T → W b decay mode across both fully hadronic (bjj) and leptonic (blν) final states at √s = 5.29, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures The single most informative sentence about the paper's subject and finding is: We study the sensitivity of a future µp collider to singly produced singlet vector-like T quarks in the T → W b decay mode across both fully hadronic (bjj) and leptonic (blν) final states at √s = 5.29, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures Vector-like quarks (VLQs) are a natural and compelling extension in numerous Beyond the Standard Model (BSM) frameworks designed to address these deficiencies, including composite Higgs models, little Higgs constructions, extra-dimensional scenarios, and superstring-inspired E6 models Unlike chiral quarks, VLQs are defined by the fact that their left- and right-handed components transform identically under the SM SU(2)L × U(1)Y gauge group, allowing for gauge-invariant mass terms that are not strictly tied to the electroweak symmetry breaking scale In many of these theoretical setups, such as the Randall-Sundrum models
4: or the SO(5)/SO(4) composite Higgs framework
17, 18: , the top-partner T quark with electric charge +2/3 plays a critical role in stabilizing the Higgs mass by canceling quadratic divergences stemming from top-quark loops
19–21> How the training signal is built The analysis utilizes TMVA-based multivariate classifiers because conventional cut-based selections do not fully exploit the correlated kinematic structure of these boosted final states The input observables chosen for the classifiers correspond to identifiable features of the T → W b topology, such as a resonant mass structure, hard recoil against the spectator system, and angular separation patterns The eight physically motivated observables used include Reconstructed Mass (Mblν), Transverse Mass (MT (l, MET)), Helicity Angle (cos θ∗), pT Ratio and MET, and others These features are used to separate the signal from the background ensemble The classifiers employed are Boosted Decision Trees (BDT), Gradient Boosted Decision Trees (BDTG), Multilayer Perceptrons (MLP), and Likelihood [34> Where the gains come from The observed quantitative gains originate primarily from the higher-purity multivariate signal regions rather than from modifications of the collider configuration itself In the fully hadronic channel, for a benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, the present MLP working point gives S/√S + B = 273.5 and ZA ≈ 42.2 for εB = 20% background-normalization uncertainty Similarly, in the leptonic channel at this benchmark, the present MLP working point reaches S/√S + B = 113.9 and ZA ≈ 21.5 The crossover between hadronic and leptonic dominance occurs as a function of mT, with the hadronic channel providing the strongest sensitivity at intermediate masses (typically between 2000 and 3000 GeV) due to its superior production rate Interpretation of Results and Discussion The statistical landscape is governed by a non-trivial interplay between integrated luminosity, center-of-mass energy, and the intrinsic characteristics of the hadronic and leptonic decay channels The comparison with the Asimov significance ZA reveals a qualitatively different behavior at high statistics, where further luminosity accumulation no longer translates into proportional gains in projected sensitivity For instance, at √s = 9.16 TeV in the hadronic channel for mT = 3000 GeV, the Gaussian significance reaches 273.51σ while the corresponding ZA saturates at 42.23σ, indicating that a 20% background systematic uncertainty sets the ultimate sensitivity ceiling The leptonic channel remains more robust toward the mass frontier because of its cleaner final state and higher signal purity in the low-statistics regime At √s = 9.16 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0.20, 0.50: up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈
0.30, 0.5: up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈
0.10, 0.5: up to mT ≈ 5.0 TeV The hadronic mode benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity The results should be interpreted as phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions Benchmark Scenarios for the Search The analysis compares the present MVA working points to a cut-based baseline at a common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1 In this comparison, the reference cut-based treatment yields S/√S + B ≈ 18.7 in the hadronic channel and S/√S + B ≈ 3.6 in the leptonic channel once a 20% background uncertainty is included The present MVA working points for these channels become 273.5 and 42.2 in the hadronic channel, and 113.9 and 21.5 in the leptonic channel This comparison shows that the main improvement is the large increase in post-selection purity under a common significance prescription The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states The hadronic channel with the MLP classifier excludes regions with g∗ ∈
0.20, 0.5: up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈
0.3, 0.5: up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈
0.1, 0.5: up to mT ≈ 5.0 TeV The hadronic mode therefore benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity Sensitivity Projections in Parameter Space The dependence on center-of-mass energy further highlights the critical role of the √s = 9.16 TeV configuration, as sensitivity rapidly deteriorates beyond mT ∼ 3000 GeV at √s = 5.29 TeV By contrast, the 9.16 TeV benchmark sustains projected significance above the discovery threshold over a much broader mass range under the stated assumptions The hadronic-channel MLP sensitivity map demonstrates that, despite its large signal yield driven by the dominant branching fraction, residual QCD backgrounds and resolved-reconstruction losses push the required luminosity contours outward In contrast, the leptonic-channel BDT sensitivity map exhibits substantially lower luminosity requirements across a wide region of parameter space The crossover between hadronic and leptonic dominance occurs as a function of mT, with the hadronic channel providing the strongest sensitivity at intermediate masses (typically between 2000 and 3000 GeV) Reconstruction Methods for Mass Determination In the hadronic decay channel, the T quark mass is reconstructed using the invariant mass of the three-jet system, denoted as Mbjj The reconstruction uses the anti-kt jet clustering algorithm with a radius parameter of R = 0.4 to define jets In the leptonic decay channel, the T quark mass is reconstructed from a final state consisting of a high-pT lepton, a b-jet, and missing transverse energy (Emiss T) The neutrino longitudinal momentum is reconstructed by imposing a W-boson mass constraint, Mlν = MW This process yields resonant peaks for the benchmark mass points ranging from 2000 GeV to 5000 GeV in both channels The reconstructed leptonic mass distributions exhibit clear resonant peaks for the benchmark mass points ranging from 2000 GeV to 5.43 TeV The leptonic channel provides a much cleaner signal-to-background ratio compared to the hadronic channel Advanced Machine Learning Techniques To classify and separate signal events, four TMVA classifiers are used: Boosted Decision Trees (BDT), Gradient Boosted Decision Trees (BDTG), Multilayer Perceptrons (MLP), and Likelihood
34> These techniques exploit high-dimensional and non-linear correlations among kinematic observables that are only partially accessible in a sequential rectangular-cut analysis The input variables used for the classifiers include kinematic properties of reconstructed objects, b-tagging information, and angular separation patterns The performance of these models is evaluated through the signal-efficiency versus background-rejection tradeoff and the area under the ROC curve (AUC) The analysis confirms that no evidence of over-training was found for all classifiers, indicating stable generalization The tuned models are then applied separately to the datasets at different center-of-mass energies to assess how reliably the classifier performance extrapolates across the benchmark scan Final Conclusion on Sensitivity The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline At the common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, these values become 273.5 and 42.2 in the hadronic channel, and 113.9 and 21.5 in the leptonic channel The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states The hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0.20, 0.5: up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈
0.3, 0.5: up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈
0.1, 0.5: up to mT ≈ 5.0 TeV Any experimental interpretation would still depend on detector-specific systematics and data-driven background constraints The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline, while preserving a transparent benchmark-level comparison to the existing literature Benchmark Scenarios for Parameter Space The benchmark scenarios for the singlet vector-like T-quark used in this analysis are summarized in Table XIII These benchmarks cover various masses and coupling strengths, such as BP1 with mT = 2000 GeV and V4b = 0.1 (1,1,1) W b: 50, Zt: 25, Ht: 25 The sensitivity maps allow a concise quantitative summary of the achievable reach at the highest-energy and highest-luminosity benchmark of √s = 9.16 TeV with L = 3000 fb−1 The hadronic channel using the MLP classifier excludes regions with g∗ ∈
0.2, 0.5: with mT ∈
2000 GeV, 4000 GeV: at the 2σ level The leptonic channel employing a BDT classifier enables exclusion and discovery sensitivity over a significantly wider mass range, with correlated regions g∗ ∈
0.1, 0.5: remaining accessible for both exclusion and discovery up to mT ≈ 5000 GeV The hadronic reconstruction at the highest masses is a resolved-object baseline rather than a dedicated boosted-W analysis The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions Final Summary of Performance At the common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, the cut-based reference gives S/√S + B ≈ 18.7 in the hadronic channel and S/√S + B ≈ 3.6 in the leptonic channel once a 20% background uncertainty is included For the corresponding MVA working points in the present analysis, these values become 273.5 and 42.2 in the hadronic channel, and 113.9 and 21.5 in the leptonic channel The main improvement is the large increase in post-selection purity under a common significance prescription The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states At √s = 9.16 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈
0.2, 0.5: up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈
0.3, 0.5: up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈
0.1, 0.5: up to mT ≈ 5.0 TeV The hadronic mode therefore benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions Comparison with Existing Literature The comparison with Ref.
62: shows that the present resolved-object MVA study gives Z = 30.20 at mT = 4.0 TeV and Z = 21.28 at mT = 4.5 TeV in the hadronic channel at the same collider energy This supports numerically competitive sensitivity through the 4.0–4.5 TeV region under the present assumptions, while the comparison is not strictly one-to-one because of how Ref.
62: reconstructs the hadronic W in a boosted fat-jet strategy and with a resolved small-R strategy here The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline, while preserving a transparent benchmark-level comparison to the existing literature The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline Final Statistical Significance Summary The Gaussian significance follows the expected Z ∝ √L scaling, but ZA reveals a qualitatively different behavior at high statistics where systematics dominate This effect is clearly visible at √s = 9.16 TeV in the hadronic channel for mT = 3000 GeV, where the Gaussian significance reaches 273.51σ while the corresponding ZA saturates at 42.23σ, indicating that a 20% background systematic uncertainty sets the ultimate sensitivity ceiling The leptonic channel BDT sensitivity map exhibits substantially lower luminosity requirements across a wide region of parameter space, reflecting the cleaner experimental signature associated with an isolated high-pT lepton and missing transverse energy At √s = 9.16 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈
0.2, 0.5: up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈
0.3, 0.5: up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈
0.1, 0.5: up to mT ≈ 5.
Title and authors: Kai: So, focusing on the title and authors of this paper, we’re looking at "Search for Vector-Like Singlet Top (T) Quark in a Future Muon-Proton (µp) Collider at sqrt s = five point two nine, six point four eight, and nine point one six TeV using Advanced Machine Learning Architectures <ref:2602.01010#pg1,Search for Vector-Like Singlet Top (T) Quark in a Future Muon>." The authors are Haroon Sagheer and others from institutions like Riphah International University and Konkuk University.
Mira: The title is quite descriptive; it immediately tells you the target—the vector-like singlet top quark—the machine it’s being searched for, the specific collision energies, and crucially, the method they are using: advanced machine learning architectures. It sets a very clear picture of what they did without getting bogged down in technical jargon right away.
Lev: From an engineering standpoint, having that clear energy range specified like five point two nine tera electron volts versus nine point one six tera electron volts gives us a concrete scale to think about when we design the experimental setup or plan the necessary simulation campaigns.
Kai: Right. It’s not just theoretical; they’ve tied this search directly to specific, achievable operational points at a future collider, which makes it feel more grounded than some of these purely theoretical BSM searches you see out there.
Mira: And that ties into the paper's main goal: comparing their machine learning approach against the closest cut-based baseline, which is essentially how we usually do things when we first start looking for new physics signals at a collider.
The paper's summary: Kai: The summary of this paper explains that they study the sensitivity of a future muon-proton collider to singly produced vector-like T quarks in the T to W b decay mode across both fully hadronic and leptonic final states at those three collision energies. They are using resolved-object reconstruction and TMVA classifiers for this search.
Mira: Essentially, they’re looking at how well they can find these T quarks by analyzing the specific kinematic features of their decays—the mass structure, the hard recoil against the spectator system—using these multivariate classifiers to quantify projected sensitivity.
Lev: So what I’m hearing is that they aren't just saying "we expect to see something"; they are mapping out exactly where the signal should appear in terms of coupling strength and mass, which is a much more useful thing for experimental planning than just a general statement about discovery potential.
Kai: And that mapping involves comparing their new multivariate selection strategy directly against the established cut-based baseline at a benchmark point of three tera electron volts, nine point one six tera electron volts, and three thousand femtobarns of luminosity.
Mira: The key finding they highlight is that the improvements they see come primarily from these higher purity multivariate signal regions rather than just changing the machine itself. They’re showing how much better the selection gets when you use these sophisticated tools to cut down on background noise.
The paper's improvements: Kai: The paper points out several improvements, and one big thing is their enhanced signal extraction capability using TMVA classifiers like BDT, BDTG, and MLP. This lets the AI systems classify events based on those physically motivated handles they mentioned earlier—like the resonant mass structure or b-jet activity.
Mira: That’s significant because it means they get a more stable significance once you factor in the background systematics, which is a big problem in these kinds of searches where you have to account for all those uncertainties. They found that no evidence of overtraining was found for all the classifiers they used, which suggests stable generalization.
Lev: From my side, when I look at this improvement, I see it translates directly to how much cleaner the final selection boundary is; if you can separate signal from background with better purity using these variables like transverse mass or pT ratios, that reduces the effective noise level we have to deal with in the actual data.
Kai: And they also mention improved mass reconstruction precision, which they achieved by using global chi square minimization fits for things like the reconstructed mass of three jets, Mbjj, and they noted that at lower mass points, these peaks are quite narrow.
Mira: That narrowness is a good sign; it implies high reconstruction efficiency and excellent energy resolution for those hadronic final states, which is something we need to see if we try to build detectors capable of resolving those structures.
Conclusion: Kai: So wrapping up the paper "Search for Vector-Like Singlet Top (T) Quark in a Future Muon-Proton (µp) Collider at sqrt s = five point two nine, six point four eight, and nine point one six TeV using Advanced Machine Learning Architectures," the main conclusion is that the systematic multivariate treatment materially strengthens the projected µp sensitivity to T to W b relative to the available cut-based baseline <ref:2602.01010#pg1,Search for Vector-Like Singlet Top (T) Quark in a Future Muon>.
Mira: They show that at their benchmark point of three tera electron volts, nine point one six tera electron volts, and three thousand femtobarns of luminosity, these new working points give you twenty seven thirty five and forty two point two in the hadronic channel and one hundred thirteen point nine and twenty one point five in the leptonic channel.
Lev: And what that means for us running real hardware is that we have a much clearer picture of where we need to focus our efforts; it shows that the improvement comes from a large increase in post-selection purity under a common significance prescription, which is something you can actually build toward.
Kai: The full scan over the mass two to five tera electron volts and the g star times mass plane reveals a clear complementarity between those two final states, meaning you have different strengths depending on whether you look at hadronic or leptonic channels.
Mira: Specifically, the hadronic channel with their MLP classifier excludes regions with g star in zero point two to zero point five up to three tera electron volts and reaches five sigma discovery for g star in zero point three to zero point five up to about three point five tera electron volts.
Lev: The paper also notes that the leptonic channel, using its best-performing classifier, retains both exclusion and discovery sensitivity for g star in zero point one to zero point five up to about five tera electron volts.
Kai: It’s interesting how the hadronic mode benefits from the larger branching fraction at intermediate masses, while the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity.
Mira: This paper, "Search for Vector-Like Singlet Top (T) Quark in a Future Muon-Proton (µp) Collider at sqrt s = five point two nine, six point four eight, and nine point one six TeV using Advanced Machine Learning Architectures," gives us a very transparent benchmark-level comparison to existing literature while showing how powerful multivariate techniques can push the sensitivity limits for these BSM searches <ref:2602.01010#pg1,Search for Vector-Like Singlet Top (T) Quark in a Future Muon>.
Lev: It’s a solid projection based on their assumptions about the generator and detector, so it’s really useful for setting realistic targets for where we need to spend our time developing those high-fidelity simulation tools.
Haroon Sagheer, M. Tayyab Javaid, Mudassar Hussain, M Danial Farooq, Ijaz Ahmed, ܵ Jamil Muhammad
Riphah International University Islamabad · Federal Urdu University of Arts Science and Technology Islamabad Pakistan · Sang-Ho College Department of Physics Konkuk University Seoul South Korea
hep-ph
Submitted: 2026-02-01
Updated: 2026-11-03
Comments: 34 pages,11 figures, 13 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 77/100
The gist: We study the sensitivity of a future muon-proton collider to singly produced singlet vector-like T quarks in the T → W b decay mode across both fully hadronic (bjj) and leptonic (blν) final states…
Key concepts
- Vector-Like Quarks (VLQs)
- These are hypothetical heavy quarks that extend the Standard Model. Unlike standard quarks, their left- and right-handed components transform identically under the weak force. They are introduced in various Beyond the Standard Model theories to stabilize the Higgs mass or address other theoretical issues.
- Multivariate Analysis (MVA)
- This technique uses multiple kinematic variables—such as reconstructed mass, angular separation, and transverse momentum ratios—to classify events. Machine learning tools like Boosted Decision Trees (BDT) and Multilayer Perceptrons (MLP) are trained on these features to distinguish the rare signal from the overwhelming background noise.
- Hadronic vs. Leptonic Channels
- These refer to two different ways the T quark decay products are observed. The hadronic channel involves detecting all jets (e.g., bjj), while the leptonic channel involves detecting a lepton, a b-jet, and missing energy (bℓν). Each channel has different signal yields and background characteristics.
- Sensitivity Ceiling
- This refers to the ultimate limit of how much new physics can be discovered given the experimental setup. The analysis found that for high statistics at 9.16 TeV, this ceiling is set by systematic uncertainties (like a 20% uncertainty in background normalization) rather than just raw integrated luminosity.
Terminology
Summary
We study the sensitivity of a future muon-proton collider to singly produced singlet vector-like T quarks in the T → W b decay mode across both fully hadronic (bjj) and leptonic (blν) final states at √s = 5. p. 029, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures The single most informative sentence about the paper's subject and finding is: We study the sensitivity of a future µp collider to singly produced singlet vector-like T quarks in the T → W b decay mode across both fully hadronic (bjj) and leptonic (blν) final states at √s = 5.29, 6.48, and 9.16 TeV using Advanced Machine Learning Architectures
Introduction to the Theoretical Framework
Vector-like quarks (VLQs) are a natural and compelling extension in numerous Beyond the Standard Model (BSM) frameworks designed to address these deficiencies, including composite Higgs models, little Higgs constructions, extra-dimensional scenarios, and superstring-inspired E6 models Unlike chiral quarks, VLQs are defined by the fact that their left- and right-handed components transform identically under the SM SU(2)L × U(1)Y gauge group, allowing for gauge-invariant mass terms that are not strictly tied to the electroweak symmetry breaking scale In many of these theoretical setups, such as the Randall-Sundrum models [4] or the SO(5)/SO(4) composite Higgs framework [17, 18], the top-partner T quark with electric charge +2/3 plays a critical role in stabilizing the Higgs mass by canceling quadratic divergences stemming from top-quark loops [19–21>
How the training signal is built
The analysis utilizes TMVA-based multivariate classifiers because conventional cut-based selections do not fully exploit the correlated kinematic structure of these boosted final states The input observables chosen for the classifiers correspond to identifiable features of the T → W b topology, such as a resonant mass structure, hard recoil against the spectator system, and angular separation patterns The eight physically motivated observables used include Reconstructed Mass (Mblν), Transverse Mass (MT (l, MET)), Helicity Angle (cos θ∗), pT Ratio and MET, and others These features are used to separate the signal from the background ensemble The classifiers employed are Boosted Decision Trees (BDT), Gradient Boosted Decision Trees (BDTG), Multilayer Perceptrons (MLP), and Likelihood [34>
Where the gains come from
The observed quantitative gains originate primarily from the higher-purity multivariate signal regions rather than from modifications of the collider configuration itself In the fully hadronic channel, for a benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, the present MLP working point gives S/√S + B = 273. p. 55 and ZA ≈ 42.2 for εB = 20% background-normalization uncertainty Similarly, in the leptonic channel at this benchmark, the present MLP working point reaches S/√S + B = 113. p. 69 and ZA ≈ 21.5 The crossover between hadronic and leptonic dominance occurs as a function of mT, with the hadronic channel providing the strongest sensitivity at intermediate masses (typically between 2000 and 3000 GeV) due to its superior production rate
Interpretation of Results and Discussion
The statistical landscape is governed by a non-trivial interplay between integrated luminosity, center-of-mass energy, and the intrinsic characteristics of the hadronic and leptonic decay channels The comparison with the Asimov significance ZA reveals a qualitatively different behavior at high statistics, where further luminosity accumulation no longer translates into proportional gains in projected sensitivity For instance, at √s = 9.16 TeV in the hadronic channel for mT = 3000 GeV, the Gaussian significance reaches 273. p. 051σ while the corresponding ZA saturates at 42. p. 4 p. 423σ, indicating that a 20% background systematic uncertainty sets the ultimate sensitivity ceiling The leptonic channel remains more robust toward the mass frontier because of its cleaner final state and higher signal purity in the low-statistics regime
Conclusion on Search Reach
At √s = 9.16 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0. p. 0 p. 0 p. 020, 0.50] up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈ [0. p. 4 p. 4 p. 4 p. 4 p. 430, 0.5] up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈ [0. p. 5 p. 5 p. 5 p. 5 p. 510, 0.5] up to mT ≈ 5.0 TeV The hadronic mode benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity The results should be interpreted as phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions
Benchmark Scenarios for the Search
The analysis compares the present MVA working points to a cut-based baseline at a common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1 In this comparison, the reference cut-based treatment yields S/√S + B ≈ 18. p. 67 in the hadronic channel and S/√S + B ≈ 3. p. 0 p. 06 in the leptonic channel once a 20% background uncertainty is included The present MVA working points for these channels become 273. p. 55 and 42.2 in the hadronic channel, and 113. p. 6 p. 6 p. 69 and 21.5 in the leptonic channel This comparison shows that the main improvement is the large increase in post-selection purity under a common significance prescription The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states The hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0. p. 520, 0.5] up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈ [0.3, 0.5] up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈ [0.1, 0.5] up to mT ≈ 5.0 TeV The hadronic mode therefore benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity
Sensitivity Projections in Parameter Space
The dependence on center-of-mass energy further highlights the critical role of the √s = 9.16 TeV configuration, as sensitivity rapidly deteriorates beyond mT ∼ 3000 GeV at √s = 5. p. 029 TeV By contrast, the 9. p. 016 TeV benchmark sustains projected significance above the discovery threshold over a much broader mass range under the stated assumptions The hadronic-channel MLP sensitivity map demonstrates that, despite its large signal yield driven by the dominant branching fraction, residual QCD backgrounds and resolved-reconstruction losses push the required luminosity contours outward In contrast, the leptonic-channel BDT sensitivity map exhibits substantially lower luminosity requirements across a wide region of parameter space The crossover between hadronic and leptonic dominance occurs as a function of mT, with the hadronic channel providing the strongest sensitivity at intermediate masses (typically between 2000 and 3000 GeV)
Reconstruction Methods for Mass Determination
In the hadronic decay channel, the T quark mass is reconstructed using the invariant mass of the three-jet system, denoted as Mbjj The reconstruction uses the anti-kt jet clustering algorithm with a radius parameter of R = 0.4 to define jets In the leptonic decay channel, the T quark mass is reconstructed from a final state consisting of a high-pT lepton, a b-jet, and missing transverse energy (Emiss T) The neutrino longitudinal momentum is reconstructed by imposing a W-boson mass constraint, Mlν = MW This process yields resonant peaks for the benchmark mass points ranging from 2000 GeV to 5000 GeV in both channels The reconstructed leptonic mass distributions exhibit clear resonant peaks for the benchmark mass points ranging from 2000 GeV to 5. p. 643 TeV The leptonic channel provides a much cleaner signal-to-background ratio compared to the hadronic channel
Advanced Machine Learning Techniques
To classify and separate signal events, four TMVA classifiers are used: Boosted Decision Trees (BDT), Gradient Boosted Decision Trees (BDTG), Multilayer Perceptrons (MLP), and Likelihood [34> These techniques exploit high-dimensional and non-linear correlations among kinematic observables that are only partially accessible in a sequential rectangular-cut analysis The input variables used for the classifiers include kinematic properties of reconstructed objects, b-tagging information, and angular separation patterns The performance of these models is evaluated through the signal-efficiency versus background-rejection tradeoff and the area under the ROC curve (AUC) The analysis confirms that no evidence of over-training was found for all classifiers, indicating stable generalization The tuned models are then applied separately to the datasets at different center-of-mass energies to assess how reliably the classifier performance extrapolates across the benchmark scan
Final Conclusion on Sensitivity
The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline At the common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, these values become 273. p. 15 and 42.2 in the hadronic channel, and 113.9 and 21.5 in the leptonic channel The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states The hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0. p. 520, 0.5] up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈ [0.3, 0.5] up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈ [0.1, 0.5] up to mT ≈ 5.0 TeV Any experimental interpretation would still depend on detector-specific systematics and data-driven background constraints The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline, while preserving a transparent benchmark-level comparison to the existing literature
Benchmark Scenarios for Parameter Space
The benchmark scenarios for the singlet vector-like T-quark used in this analysis are summarized in Table XIII These benchmarks cover various masses and coupling strengths, such as BP1 with mT = 2000 GeV and V4b = 0.1 (1,1,1) W b: 50, Zt: 25, Ht: 25 The sensitivity maps allow a concise quantitative summary of the achievable reach at the highest-energy and highest-luminosity benchmark of √s = 9. p. 416 TeV with L = 3000 fb−1 The hadronic channel using the MLP classifier excludes regions with g∗ ∈ [0. p. 02, 0.5] with mT ∈ [2000 GeV, 4000 GeV] at the 2σ level The leptonic channel employing a BDT classifier enables exclusion and discovery sensitivity over a significantly wider mass range, with correlated regions g∗ ∈ [0. p. 81, 0.5] remaining accessible for both exclusion and discovery up to mT ≈ 5000 GeV The hadronic reconstruction at the highest masses is a resolved-object baseline rather than a dedicated boosted-W analysis The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions
Final Summary of Performance
At the common benchmark point mT = 3 TeV, √s = 9.16 TeV, and L = 3000 fb−1, the cut-based reference gives S/√S + B ≈ 18. p. 67 in the hadronic channel and S/√S + B ≈ 3.6 in the leptonic channel once a 20% background uncertainty is included For the corresponding MVA working points in the present analysis, these values become 273. p. 55 and 42.2 in the hadronic channel, and 113.9 and 21.5 in the leptonic channel The main improvement is the large increase in post-selection purity under a common significance prescription The full scan over mT = 2–5 TeV and the (g∗, mT) plane shows a clear complementarity between the two final states At √s = 9. p. 216 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0.2, 0.5] up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈ [0.3, 0.5] up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈ [0.1, 0.5] up to mT ≈ 5.0 TeV The hadronic mode therefore benefits from the larger branching fraction at intermediate masses, whereas the leptonic mode becomes more robust toward the mass frontier because of its cleaner final state and higher signal purity The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions
Comparison with Existing Literature
The comparison with Ref. [62] shows that the present resolved-object MVA study gives Z = 30. p. 420 at mT = 4.0 TeV and Z = 21. p. 028 at mT = 4.5 TeV in the hadronic channel at the same collider energy This supports numerically competitive sensitivity through the 4. p. 40–4.5 TeV region under the present assumptions, while the comparison is not strictly one-to-one because of how Ref. p. 4 [62] reconstructs the hadronic W in a boosted fat-jet strategy and with a resolved small-R strategy here The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline, while preserving a transparent benchmark-level comparison to the existing literature The results presented are phenomenological projections under the stated generator, detector, resolved-reconstruction, and background-uncertainty assumptions The main conclusion of the revised manuscript is that a systematic multivariate treatment materially strengthens the projected µp sensitivity to T → W b relative to the available cut-based baseline
Final Statistical Significance Summary
The Gaussian significance follows the expected Z ∝ √L scaling, but ZA reveals a qualitatively different behavior at high statistics where systematics dominate This effect is clearly visible at √s = 9.16 TeV in the hadronic channel for mT = 3000 GeV, where the Gaussian significance reaches 273. p. 051σ while the corresponding ZA saturates at 42.23σ, indicating that a 20% background systematic uncertainty sets the ultimate sensitivity ceiling The leptonic channel BDT sensitivity map exhibits substantially lower luminosity requirements across a wide region of parameter space, reflecting the cleaner experimental signature associated with an isolated high-pT lepton and missing transverse energy At √s = 9. p. 616 TeV and L = 3000 fb−1, the hadronic channel with the MLP classifier excludes regions with g∗ ∈ [0.2, 0.5] up to mT ≈ 4.0 TeV and reaches 5σ discovery for g∗ ∈ [0.3, 0.5] up to about 3.5 TeV The leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g∗ ∈ [0.1, 0.5] up to mT ≈ 5.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved signal extraction capability for Vector-Like Top quark searches using TMVA classifiers (BDT, BDTG, MLP). This allows AI systems to classify events based on
physically motivated handles
likea resonant mass structure, hard recoil against the spectator system, b-jet activity,
leading to a more stable significance once background systematics are included. -
Enhanced ability to map sensitivity across the parameter space using fast simulation scans over mT and the g star mT plane. This enables AI systems to project
exclusion and discovery contours
showing thatthe leptonic channel with the best-performing classifier retains both exclusion and discovery sensitivity for g star in [0.10, 0.50] up to mT approximately 5000 GeV.
-
Improved mass reconstruction precision for reconstructed signal candidates using global chi square minimization fits. This allows AI systems to reconstruct the parent top partner mass from final states like
Mbjj
and provides a measure of resolution, noting thatthe narrowness of these peaks at lower mass points indicates high reconstruction efficiency and excellent energy resolution for the hadronic final states.
-
Robust discrimination between signal and background using kinematic balance variables like Transverse Mass MT (lepton, MET) or pT Ratio. This allows AI systems to exploit how
the products follow a characteristic energy-sharing profile
in VLT decays versus W+jets events to achievea clean separation boundary.
Sources
- Vector-like Singlet Quarks: a Roadmap
- Pair production of the singlet vector-like B quark at the CLIC
- Largest temperature of the radiation era and its cosmological implications
- Phase Transitions in QCD
- Multiphoton Bloch-Siegert shifts and level-splittings in a three-level system
- TMVA - Toolkit for Multivariate Data Analysis
- XGBoost: A Scalable Tree Boosting System
- A comprehensive guide to the physics and usage of PYTHIA 8.3
- Searching for single production of vectorlike quarks decaying into Wb at a future muon-proton collider
- Probing Vector-Like Quarks at a future Muon-Proton Collider
Related papers
- MadNIS at NLO
- The NANOGrav 15-year data set: A Limit on Continuous Spin Gravitons
- DIS-secting the universe with high-energy neutrinos
- Crowdsourcing Gravitational Waves from Superradiant Axions
- Classification of g-modes for neutron stars with a strong transition: Novel universal relation including slow stable hybrid stars
- Scalar axion field of toroidal electromagnetic pulses