arxiv: 2604.25407 · v1 · submitted 2026-04-28 · ✦ hep-ph

Recognition: unknown

Big Dipper, Help Me Find A Way -- Dip-hunting at hadron colliders

Diego A. Baron Moreno , Christoph Englert , Yvonne Peters

Authors on Pith no claims yet

Pith reviewed 2026-05-07 16:02 UTC · model grok-4.3

classification ✦ hep-ph

keywords dip-huntinginterference effectsparametric neural networkslikelihood ratiotop-philic scalarsBSM parameter extractionhadron collidersscalar resonances

0 comments

The pith

Parametric neural networks extract BSM parameters from interference dips in collider data.

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

The paper aims to show that destructive interference between a new scalar resonance and standard model background, which creates dips rather than bumps in distributions, can be turned into a tool for parameter inference instead of a barrier. Traditional bump-hunting fails when the narrow-width approximation breaks down across much of the parameter space for top-philic scalars. The authors train parametric neural networks to compute the likelihood ratio that depends simultaneously on background and beyond-standard-model parameters, using a ratio-of-signed-mixtures approach. They then scan the parameter space of a minimal scalar extension of the Standard Model and test which points are compatible with an observed data pattern. A sympathetic reader would care because this offers a way to diagnose or constrain new physics even in regimes where conventional searches lose all sensitivity.

Core claim

We employ parametric neural networks to learn the likelihood ratio as a function of both background and key BSM parameters, based on a ratio-of-signed-mixtures framework. We perform inference by testing the compatibility of observed data with a scan over the parameter space of a minimal scalar extension of the Standard Model. While BSM parameter extraction remains inherently model-dependent, our approach provides a robust diagnostic in perturbative regimes and motivates a complementary strategy of dip-hunting that extends traditional bump-hunts.

What carries the argument

Parametric neural networks that learn the likelihood ratio as a function of background and BSM parameters within a ratio-of-signed-mixtures framework, enabling scans over scalar extension parameter space from observed interference patterns.

If this is right

Parameter extraction from data remains feasible in regions where interference invalidates the narrow-width approximation and suppresses visible signals.
The method functions as a diagnostic tool that can flag consistency or inconsistency with a chosen BSM model in perturbative regimes.
Analyses can extend bump-hunting strategies by also searching for and interpreting dips caused by destructive interference.
The approach complements rather than replaces detailed simulation modeling throughout the experimental analysis chain.

Where Pith is reading between the lines

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

The same likelihood-ratio learning technique could be tested on other BSM scenarios that feature strong signal-background interference at hadron colliders.
If the method holds up under real-data conditions, it could reduce the need for repeated full simulations for every point in a parameter scan during interpretation.
Future experimental searches might incorporate dip-hunting as a standard complementary channel alongside traditional bump searches when designing analysis strategies.

Load-bearing premise

Neural networks trained on simulated events can faithfully reproduce the true likelihood ratios that would appear in real collider data, including all interference effects, without large unaccounted model dependence or systematics.

What would settle it

Apply the trained networks to large sets of simulated events generated with known injected BSM parameter values and check whether the recovered best-fit parameters and uncertainties systematically fail to contain the true injected values.

read the original abstract

Destructive interference between signal and background processes poses a fundamental challenge in searches for top-philic scalar resonances, significantly reducing experimental sensitivity to well-motivated extensions of the Higgs sector. Traditional bump-hunting strategies fail in this instance because interference effects invalidate the narrow-width approximation across large regions of the BSM parameter space. As a result, experimental analyses typically rely on detailed simulations to accurately model these effects throughout the full analysis chain. In this work, we consider the inverse problem in a proof-of-principle study: given an observed pattern in a discriminating distribution, what is the likelihood that it originates from a BSM scalar? To address this, we employ parametric neural networks to learn the likelihood ratio as a function of both background and key BSM parameters, based on a ratio-of-signed-mixtures framework. We perform inference by testing the compatibility of observed data with a scan over the parameter space of a minimal scalar extension of the Standard Model. While BSM parameter extraction remains inherently model-dependent, our approach provides a robust diagnostic in perturbative regimes and motivates a complementary strategy of `dip-hunting'. This strategy extends traditional bump-hunts and could point the way as we navigate towards future discoveries.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper gives a proof-of-principle for parametric neural networks that learn a likelihood ratio over signed mixtures to hunt dips from interfering top-philic scalars instead of bumps.

read the letter

The core contribution is a method that trains parametric networks on simulated background-plus-signal samples to extract the likelihood ratio as a function of both SM background and BSM scalar parameters. Inference then proceeds by scanning the minimal scalar model and testing compatibility with an observed distribution. This directly targets the regime where destructive interference invalidates narrow-width bump searches, which is a recurring issue in top-philic scalar models at hadron colliders. The ratio-of-signed-mixtures framing and the parametric network choice appear new relative to standard bump-hunting literature, and the proof-of-principle framing is appropriately modest about model dependence in perturbative regimes. The paper does a clean job laying out why traditional strategies fall short and how the diagnostic could complement them. Credit is due for keeping the focus on an independent test rather than a fitted prediction from the same data. The main soft spot is the leap from simulation to real data. The approach assumes that parton shower, hadronization, and detector effects are modeled well enough that the learned ratio remains unbiased when applied to observed events, yet the description does not include explicit closure tests or systematic-variation studies that would quantify residual mismodeling. Without those, it is difficult to judge how much the extracted parameters or compatibility statistic could shift under realistic conditions. This is a genuine limitation for a method meant to guide experimental analyses. The work is aimed at phenomenologists and LHC analysts who already deal with interference in resonance searches and are open to ML-based diagnostics. A reader who wants concrete implementation details or validation plots will find the current version thin, but the idea itself is worth seeing in print. It deserves a serious referee because the problem is real, the proposed tool is distinct from prior work, and the gaps are fixable with additional studies rather than fatal to the central claim. I would send it to peer review.

Referee Report

1 major / 2 minor

Summary. The manuscript presents a proof-of-principle study for 'dip-hunting' at hadron colliders targeting top-philic scalar resonances. Traditional bump-hunting fails due to destructive interference between signal and background that invalidates the narrow-width approximation over much of the BSM parameter space. The authors train parametric neural networks on signed-mixture simulations to learn the likelihood ratio as a function of both background and key BSM parameters, then perform inference by testing the compatibility of an observed distribution against a scan over the parameter space of a minimal scalar extension of the Standard Model. The work frames this as a robust diagnostic in perturbative regimes and motivates dip-hunting as a complementary strategy to bump-hunting.

Significance. If the central methodological claim holds, the approach could provide a useful new diagnostic for BSM searches in which interference effects dominate, extending sensitivity beyond standard bump-hunting techniques. The ratio-of-signed-mixtures framework combined with parametric networks is a concrete technical contribution that could be adapted to other interference-dominated channels. However, the significance is currently limited by the absence of any reported validation that would demonstrate reliable transfer from simulation to data.

major comments (1)

The central inference procedure relies on the learned likelihood ratio accurately reproducing interference-induced dip shapes when applied to real data. No closure tests, toy-data validation, or systematic-variation studies are described that would quantify the impact of unmodeled effects (parton shower, hadronization, detector response) on the compatibility test statistic or extracted parameters. This is load-bearing for the claim that the method provides a 'robust diagnostic'.

minor comments (2)

The abstract states that 'BSM parameter extraction remains inherently model-dependent' but does not clarify how this model dependence propagates into the reported compatibility tests or whether the scan is performed under fixed assumptions about other parameters.
Notation for the signed-mixture framework and the precise definition of the parametric network inputs/outputs should be introduced earlier and used consistently to aid readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback. We appreciate the recognition of the potential utility of the dip-hunting approach and the technical contribution of the signed-mixture parametric network framework. We address the major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: The central inference procedure relies on the learned likelihood ratio accurately reproducing interference-induced dip shapes when applied to real data. No closure tests, toy-data validation, or systematic-variation studies are described that would quantify the impact of unmodeled effects (parton shower, hadronization, detector response) on the compatibility test statistic or extracted parameters. This is load-bearing for the claim that the method provides a 'robust diagnostic'.

Authors: We agree that the absence of explicit closure tests, toy-data validations, or systematic studies on unmodeled effects limits the strength of the 'robust diagnostic' claim for application to real experimental data. Our manuscript is framed as a proof-of-principle study using parton-level simulations to isolate and demonstrate the ability of parametric networks to learn likelihood ratios from signed mixtures and recover interference-induced dip shapes in a controlled perturbative setting. No such validation studies were performed because the primary objective was to establish the methodological framework and inference procedure rather than to simulate a full experimental analysis chain. We will revise the manuscript by adding a dedicated subsection on limitations and assumptions. This will explicitly discuss the potential impact of parton shower, hadronization, and detector response on the test statistic, clarify that current results are simulation-based, and outline how future work could incorporate these effects to test transfer to data. This will appropriately scope our claims without overstating robustness. revision: partial

Circularity Check

0 steps flagged

No circularity: method is an independent diagnostic proposal

full rationale

The paper proposes a parametric NN approach to learn likelihood ratios from simulated signed-mixture samples and then tests compatibility of observed data against a BSM parameter scan. This is framed as a proof-of-principle diagnostic extending bump-hunting, relying on standard simulation-based inference techniques without any derivation step that reduces a claimed prediction or result to a fitted input, self-citation chain, or definitional tautology. No equations or load-bearing claims in the abstract or described framework equate outputs to inputs by construction; the central claim remains an independent methodological suggestion whose validity hinges on external validation (e.g., closure tests) rather than internal reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only information limits the ledger; the central claim rests on the validity of the ratio-of-signed-mixtures framework and the ability of parametric networks to generalize across BSM parameters.

axioms (1)

domain assumption The ratio-of-signed-mixtures framework accurately captures the interference effects between signal and background.
Invoked as the basis for the likelihood ratio learning in the method description.

pith-pipeline@v0.9.0 · 5513 in / 1195 out tokens · 57806 ms · 2026-05-07T16:02:53.075199+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 28 canonical work pages · 3 internal anchors

[1]

K. J. F. Gaemers and F. Hoogeveen,Higgs Production and Decay Into Heavy Flavors With the Gluon Fusion Mechanism,Phys. Lett. B146(1984) 347–349

1984
[2]

Dicus, A

D. Dicus, A. Stange, and S. Willenbrock,Higgs decay to top quarks at hadron colliders,Phys. Lett. B333 (1994) 126–131, [hep-ph/9404359]

work page arXiv 1994
[3]

Basler, S

P. Basler, S. Dawson, C. Englert, and M. M¨ uhlleitner,Di-Higgs boson peaks and top valleys: Interference effects in Higgs sector extensions,Phys. Rev. D101(2020), no. 1 015019, [arXiv:1909.09987]

work page arXiv 2020
[4]

S. Jung, J. Song, and Y. W. Yoon,Dip or nothingness of a Higgs resonance from the interference with a complex phase,Phys. Rev. D92(2015), no. 5 055009, [arXiv:1505.00291]

work page arXiv 2015
[5]

Frederix and F

R. Frederix and F. Maltoni,Top pair invariant mass distribution: A Window on new physics,JHEP01 (2009) 047, [arXiv:0712.2355]

work page arXiv 2009
[6]

Djouadi, J

A. Djouadi, J. Ellis, A. Popov, and J. Quevillon,Interference effects int tproduction at the LHC as a window on new physics,JHEP03(2019) 119, [arXiv:1901.03417]

work page arXiv 2019
[7]

Challenges and opportunities for heavy scalar searches in the $t\bar t$ channel at the LHC

M. Carena and Z. Liu,Challenges and opportunities for heavy scalar searches in the tt channel at the LHC, JHEP11(2016) 159, [arXiv:1608.07282]

work page Pith review arXiv 2016
[8]

ATLASCollaboration, G. Aad et al.,Search for heavy neutral Higgs bosons decaying into a top quark pair in 140 fb −1 of proton-proton collision data at √s = 13 TeV with the ATLAS detector,JHEP08(2024) 013, [arXiv:2404.18986]. 4As always, the ‘correct’ parameters of the underlying model, however, depend on the BSM model assumptions (just as is the case for ...

work page arXiv 2024
[9]

Drnevich, S

M. Drnevich, S. Jiggins, J. Katzy, and K. Cranmer,Neural quasiprobabilistic likelihood ratio estimation with negatively weighted data,Mach. Learn. Sci. Tech.6(2025), no. 4 045023, [arXiv:2410.10216]. [13]ATLASCollaboration, G. Aad et al.,Search for ttbar resonances in final states with exactly one or two leptons using 140 fb −1 of pp collision data at √s ...

work page arXiv 2025
[10]

Bessidskaia Bylund, F

O. Bessidskaia Bylund, F. Maltoni, I. Tsinikos, E. Vryonidou, and C. Zhang,Probing top quark neutral couplings in the Standard Model Effective Field Theory at NLO in QCD,JHEP05(2016) 052, [arXiv:1601.08193]

work page arXiv 2016
[11]

Englert, P

C. Englert, P. Galler, and C. D. White,Effective field theory and scalar extensions of the top quark sector, Phys. Rev. D101(2020), no. 3 035035, [arXiv:1908.05588]

work page arXiv 2020
[12]

H. M. Georgi, S. L. Glashow, M. E. Machacek, and D. V. Nanopoulos,Higgs Bosons from Two Gluon Annihilation in Proton Proton Collisions,Phys. Rev. Lett.40(1978) 692

1978
[13]

Djouadi, The anatomy of Electro weak symmetry breaking I: the Higgs boson in the Standard Model, Phys

A. Djouadi,The Anatomy of electro-weak symmetry breaking. I: The Higgs boson in the standard model, Phys. Rept.457(2008) 1–216, [hep-ph/0503172]

work page arXiv 2008
[14]

Plehn,Lectures on LHC Physics,Lect

T. Plehn,Lectures on LHC Physics,Lect. Notes Phys.844(2012) 1–193, [arXiv:0910.4182]

work page arXiv 2012
[15]

Goria, G

S. Goria, G. Passarino, and D. Rosco,The Higgs Boson Lineshape,Nucl. Phys. B864(2012) 530–579, [arXiv:1112.5517]

work page arXiv 2012
[16]

M. H. Seymour,The Higgs boson line shape and perturbative unitarity,Phys. Lett. B354(1995) 409–414, [hep-ph/9505211]

work page arXiv 1995
[17]

Papavassiliou and A

J. Papavassiliou and A. Pilaftsis,Gauge invariant resummation formalism for two point correlation functions, Phys. Rev. D54(1996) 5315–5335, [hep-ph/9605385]

work page arXiv 1996
[18]

Papavassiliou and A

J. Papavassiliou and A. Pilaftsis,Effective charge of the Higgs boson,Phys. Rev. Lett.80(1998) 2785–2788, [hep-ph/9710380]

work page arXiv 1998
[19]

Englert, I

C. Englert, I. Low, and M. Spannowsky,On-shell interference effects in Higgs boson final states,Phys. Rev. D91(2015), no. 7 074029, [arXiv:1502.04678]

work page arXiv 2015
[20]

Murayama, I

H. Murayama, I. Watanabe, and K. Hagiwara,HELAS: HELicity amplitude subroutines for Feynman diagram evaluations,
[21]

UFO - The Universal FeynRules Output

C. Degrande, C. Duhr, B. Fuks, D. Grellscheid, O. Mattelaer, and T. Reiter,UFO - The Universal FeynRules Output,Comput. Phys. Commun.183(2012) 1201–1214, [arXiv:1108.2040]

work page Pith review arXiv 2012
[22]

Darmé et al.,UFO 2.0: the ‘Universal Feynman Output’ format,Eur

L. Darm´ e et al.,UFO 2.0: the ‘Universal Feynman Output’ format,Eur. Phys. J. C83(2023), no. 7 631, [arXiv:2304.09883]. – 21 –

work page arXiv 2023
[23]

The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations

J. Alwall, R. Frederix, S. Frixione, V. Hirschi, F. Maltoni, O. Mattelaer, H. S. Shao, T. Stelzer, P. Torrielli, and M. Zaro,The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations,JHEP07(2014) 079, [arXiv:1405.0301]

work page internal anchor Pith review arXiv 2014
[24]

B. A. Kniehl and M. Spira,Low-energy theorems in Higgs physics,Z. Phys. C69(1995) 77–88, [hep-ph/9505225]. [29]ATLASCollaboration, M. Aaboud et al.,Measurements of top quark spin observables int tevents using dilepton final states in √s= 8TeV pp collisions with the ATLAS detector,JHEP03(2017) 113, [arXiv:1612.07004]

work page internal anchor Pith review arXiv 1995
[25]

Aad et al.,Differential tt cross-section measurements using boosted top quarks in the all-hadronic final state with 139 fb −1 of ATLAS data,JHEP04(2023) 080, [arXiv:2205.02817]

ATLASCollaboration, G. Aad et al.,Differential tt cross-section measurements using boosted top quarks in the all-hadronic final state with 139 fb −1 of ATLAS data,JHEP04(2023) 080, [arXiv:2205.02817]

work page arXiv 2023
[26]

Buckley, C

A. Buckley, C. Englert, J. Ferrando, D. J. Miller, L. Moore, M. Russell, and C. D. White,Constraining top quark effective theory in the LHC Run II era,JHEP04(2016) 015, [arXiv:1512.03360]

work page arXiv 2016
[27]

Brivio, S

I. Brivio, S. Bruggisser, F. Maltoni, R. Moutafis, T. Plehn, E. Vryonidou, S. Westhoff, and C. Zhang,O new physics, where art thou? A global search in the top sector,JHEP02(2020) 131, [arXiv:1910.03606]

work page arXiv 2020
[28]

D. A. B. Moreno, C. Englert, and Y. Peters,in preparation,
[29]

PyTorch: An Imperative Style, High-Performance Deep Learning Library

A. Paszke et al.,PyTorch: An Imperative Style, High-Performance Deep Learning Library, arXiv:1912.01703

work page internal anchor Pith review arXiv 1912
[30]

Scikit-learn: Machine Learning in Python

F. Pedregosa et al.,Scikit-learn: Machine Learning in Python,J. Machine Learning Res.12(2011) 2825–2830, [arXiv:1201.0490]

work page Pith review arXiv 2011
[31]

Falcon et al.,Pytorchlightning/pytorch-lightning: 0.7.6 release, May, 2020

W. Falcon et al.,Pytorchlightning/pytorch-lightning: 0.7.6 release, May, 2020

2020
[32]

H. Bahl, R. Kumar, and G. Weiglein,Impact of interference effects on Higgs-boson searches in the di-top final state at the LHC,JHEP05(2025) 098, [arXiv:2503.02705]

work page arXiv 2025
[33]

Robens,Interference effects in new physics searches,arXiv:2602.00256

T. Robens,Interference effects in new physics searches,arXiv:2602.00256

work page arXiv
[34]

Baron,diegobaronm/ttbarresonancestudies: v1.0 - deep hunting paper version., Apr., 2026

D. Baron,diegobaronm/ttbarresonancestudies: v1.0 - deep hunting paper version., Apr., 2026

2026
[35]

Baron,diegobaronm/peakdeepmaster: v1.0 - deep hunting paper version., Apr., 2026

D. Baron,diegobaronm/peakdeepmaster: v1.0 - deep hunting paper version., Apr., 2026. – 22 –

2026