arXiv subjects

Daohan Wang

Publications and source records attributed to Daohan Wang.

At least 19 recordsLinked to original sources

arXiv2026

Efficient Event Generation for High-Multiplicity LHC Processes: An End-to-End GPU Workflow with Normalizing Flows

Producing very large unweighted event samples for high-multiplicity processes is limited by expensive matrix-element evaluations and low unweighting efficiencies. We present the first end-to-end GPU-resident event-generation workflow that integrates normalizing-flow proposals with the parton-level event generator Pepper. Helicity-conditioned coupling flows are trained using online updates supplemented by sample replay and deployed across all subprocesses of complete proton--proton collision processes with many final-state jets. In this workflow, a Python-based control layer and Pepper exchange flow-generated phase-space points and the corresponding target-density evaluations directly in device memory. The control layer performs flow sampling, proposal-density evaluation, and unweighting, while Pepper evaluates the matrix elements, PDFs, and phase-space factors defining the target density and writes the accepted events in standard formats. We compare subprocess-specific flows, with one flow per partonic subprocess, to grouped conditional flows that share parameters among subprocesses with related parton content. The workflow is benchmarked for $pp \to e^+e^- + 4j$, $pp \to e^+e^- + 5j$, $pp \to t \bar t + 4j$, $pp \to 4j$, and $pp \to 5j$ production. On four H100 GPUs, we generate $10^9$ unweighted events for each benchmark process. Including the cost of flow training, the workflow achieves end-to-end speedups of up to two orders of magnitude over standalone Pepper event generation and turns a multi-week task into a sub-day computation. It thereby makes billion-event production more practical and offers a pathway to alleviating the Monte Carlo statistics bottleneck in high-multiplicity collider physics.

hep-ph↗

arXiv2026

Proton Structure from Neural Simulation-Based Inference at the LHC

The precise determination of the parton distribution functions (PDFs) of the proton is an essential ingredient for LHC analyses, including for those at the upcoming High-Luminosity LHC. So far, PDFs are determined from global fits to binned low-dimensional data obtained from unfolded hard-scattering cross section measurements. In this work we demonstrate for the first time the feasibility of neural simulation-based inference (NSBI) for constraining the proton PDFs using a high-dimensional unbinned data set. Exploiting the full statistical power of unbinned data removes the loss of information inherited by the binning procedure. As a proof-of-concept, we determine the gluon PDF from simulated data of top quark pair production at the LHC with $\sqrt{s}=13$ TeV. Taking into account both experimental and theoretical systematic uncertainties in the detector-level features, we demonstrate how the NSBI pipeline achieves significant improvements in precision compared to existing low-dimensional binned analyses. Our results illustrate the potential of unbinned inference to reduce the reliance on coarse approximations of uncertainties and their correlations entering PDF determinations, hence contributing to a new paradigm of unbinned detector-level ML-assisted measurements at the LHC.

hep-ph↗

arXiv2025

Higgs Signal Strength Estimation with Machine Learning under Systematic Uncertainties

We present a dedicated graph neural network (GNN)-based methodology for the extraction of the Higgs boson signal strength $\mu$, incorporating systematic uncertainties. The architecture features two branches: a deterministic GNN that processes kinematic variables unaffected by nuisance parameters, and an uncertainty-aware GNN that handles inputs modulated by systematic effects through gated attention-based message passing. Their outputs are fused to produce classification scores for signal-background discrimination. During training we sample nuisance-parameter configurations and aggregate the loss across them, promoting stability of the classifier under systematic shifts and effectively decorrelating its outputs from nuisance variations. The resulting binned classifier outputs are used to construct a Poisson likelihood, which enables profile likelihood scans over signal strength, with nuisance parameters profiled out via numerical optimization. We validate this framework on the FAIR Universe Higgs Uncertainty Challenge dataset, yielding accurate estimation of signal strength $\mu$ and its 68.27\% confidence interval, achieving competitive coverage and interval widths in large-scale pseudo-experiments. Our code "Systematics-Aware Graph Estimator" (SAGE) is publicly available.

hep-ph↗

arXiv2025

Unbinned inclusive cross-section measurements with machine-learned systematic uncertainties

We introduce a novel methodology for addressing systematic uncertainties in unbinned inclusive cross-section measurements and related collider-based inference problems. Our approach incorporates known analytic dependencies on parameters of interest, including signal strengths and nuisance parameters. When these dependencies are unknown, as is frequently the case for systematic uncertainties, dedicated neural network parametrizations provide an approximation that is trained on simulated data. The resulting machine-learned surrogate captures the complete parameter dependence of the likelihood ratio, providing a near-optimal test statistic. As a case study, we perform a first-principles inclusive cross-section measurement of $\textrm{H}\rightarrow\tau\tau$ in the single-lepton channel, utilizing simulated data from the FAIR Universe Higgs Uncertainty Challenge. Results in Asimov data, from large-scale toy studies, and using the Fisher information demonstrate significant improvements over traditional binned methods. Our computer code ``Guaranteed Optimal Log-Likelihood-based Unbinned Method'' (GOLLUM) for machine-learning and inference is publicly available.

hep-ph↗

arXiv2025

BitHEP -- The Limits of Low-Precision ML in HEP

The increasing complexity of modern neural network architectures demands fast and memory-efficient implementations to mitigate computational bottlenecks. In this work, we evaluate the recently proposed BitNet architecture in HEP applications, assessing its performance in classification, regression, and generative modeling tasks. Specifically, we investigate its suitability for quark-gluon discrimination, SMEFT parameter estimation, and detector simulation, comparing its efficiency and accuracy to state-of-the-art methods. Our results show that while BitNet consistently performs competitively in classification tasks, its performance in regression and generation varies with the size and type of the network, highlighting key limitations and potential areas for improvement.

hep-ph↗

arXiv2024

FAIR Universe HiggsML Uncertainty Dataset and Competition

The FAIR Universe HiggsML Uncertainty Challenge focused on measuring the physical properties of elementary particles with imperfect simulators. Participants were required to compute and report confidence intervals for a parameter of interest regarding the Higgs boson while accounting for various systematic (epistemic) uncertainties. The dataset is a tabular dataset of 28 features and 280 million instances. Each instance represents a simulated proton-proton collision as observed at CERN's Large Hadron Collider in Geneva, Switzerland. The features of these simulations were chosen to capture key characteristics of different types of particles. These include primary attributes, such as the energy and three-dimensional momentum of the particles, as well as derived attributes, which are calculated from the primary ones using domain-specific knowledge. Additionally, a label feature designates each instance's type of proton-proton collision, distinguishing the Higgs boson events of interest from three background sources. As outlined in this paper, the permanent release of the dataset allows long-term benchmarking of new techniques. The leading submissions, including Contrastive Normalising Flows and Density Ratios estimation through classification, are described. Our challenge has brought together the physics and machine learning communities to advance our understanding and methodologies in handling systematic uncertainties within AI techniques.

hep-ph↗

arXiv2024

Discovery Prospects for the Light Charged Higgs Boson Decay to an Off-Shell Top Quark and a Bottom Quark at Future High-Energy Colliders

The charged Higgs boson ($H^\pm$) with a mass below the top quark mass remains a viable possibility within the Type-I two-Higgs-doublet model under current constraints. While previous LHC searches have primarily focused on the $H^\pm\to\tau^\pm\nu$ decay mode, the decay channel into an off-shell top quark and a bottom quark, $H^\pm \rightarrow t^*b$, is leading or subleading for $H^\pm$ masses between 130 and 170 GeV. This study investigates the discovery potential of future colliders for this off-shell decay mode through pair-produced charged Higgs bosons decaying via $H^+H^-\rightarrow t^*b\tau\nu\rightarrow bbjj\tau\nu$. We perform signal-to-background analyses at the HL-LHC and a prospective 100 TeV proton-proton collider, employing cut-flow strategies and the Boosted Decision Tree method. However, due to the softness of the $b$ jets, signal significances fall below detection thresholds at these facilities. Extending our study to a multi-TeV muon collider (MuC), we demonstrate that a 3 TeV MuC achieves high signal significance, surpassing the $5\sigma$ threshold with an integrated luminosity of 1 ab$^{-1}$ and a 10\% background uncertainty. Specifically, for $M_{H^\pm} = 130$, 150, and 170 GeV, the significances are 13.7, 13.5, and 6.06, respectively. In contrast, a 10 TeV MuC requires 10 ab$^{-1}$ to achieve similar results. Our findings highlight the critical role of the MuC in probing the new signal channel $H^\pm\rightarrow t^*b$, offering a promising avenue for future charged Higgs boson searches involving off-shell top quarks.

hep-ph↗

arXiv2023

Probing Light Fermiophobic Higgs Boson via diphoton jets at the HL-LHC

In this study, we explore the phenomenological signatures associated with a light fermiophobic Higgs boson, $h_{\rm f}$, within the type-I two-Higgs-doublet model at the HL-LHC. Our meticulous parameter scan illuminates an intriguing mass range for $m_{h_{\rm f}}$, spanning $[1,10]{\;{\rm GeV}}$. This mass range owes its viability to substantial parameter points, largely due to the inherent challenges of detecting the soft decay products of $h_{\rm f}$ at contemporary high-energy colliders. Given that this light $h_{\rm f}$ ensures $Br(h_{\rm f}\to\gamma\gamma)\simeq 1$, $Br(H^\pm \to h_{\rm f} W^\pm)\simeq 1$, and $M_{H^\pm}\lesssim 330{\;{\rm GeV}}$, we propose a golden discovery channel: $pp\to h_{\rm f}H^\pm\to \gamma\gamma\gamma\gamma \,l^\pm\nu$, where $l^\pm$ includes $e^\pm$ and $\mu^\pm$. However, a significant obstacle arises as the two photons from the $h_{\rm f}$ decay mostly merge into a single jet due to their proximity within $\Delta R<0.4$. This results in a final state characterized by two jets, rather than four isolated photons, thus intensifying the QCD backgrounds. To tackle this, we devise a strategy within \textsc{Delphes} to identify jets with two leading subparticles as photons, termed diphoton jets. Our thorough detector-level simulations across 18 benchmark points predominantly show signal significances exceeding the $5\sigma$ threshold at an integrated luminosity of $3{\;{\rm ab}^{-1}}$. Furthermore, our approach facilitates accurate mass reconstructions for both $m_{h_{\rm f}}$ and $M_{H^\pm}$. Notably, in the intricate scenarios with heavy charged Higgs bosons, our application of machine learning techniques provides a significant boost in significance.

hep-ph↗

arXiv2023

Hierarchical High-Point Energy Flow Network for Jet Tagging

Jet substructure observable basis is a systematic and powerful tool for analyzing the internal energy distribution of constituent particles within a jet. In this work, we propose a novel method to insert neural networks into jet substructure basis as a simple yet efficient interpretable IRC-safe deep learning framework to discover discriminative jet observables. The Energy Flow Polynomial (EFP) could be computed with a certain summation order, resulting in a reorganized form which exhibits hierarchical IRC-safety. Thus inserting non-linear functions after the separate summation could significantly extend the scope of IRC-safe jet substructure observables, where neural networks can come into play as an important role. Based on the structure of the simplest class of EFPs which corresponds to path graphs, we propose the Hierarchical Energy Flow Networks and the Local Hierarchical Energy Flow Networks. These two architectures exhibit remarkable discrimination performance on the top tagging dataset and quark-gluon dataset compared to other benchmark algorithms even only utilizing the kinematic information of constituent particles.

hep-ph↗

arXiv2023

Quark/Gluon Discrimination and Top Tagging with Dual Attention Transformer

Jet tagging is a crucial classification task in high energy physics. Recently the performance of jet tagging has been significantly improved by the application of deep learning techniques. In this study, we introduce a new architecture for jet tagging: the Particle Dual Attention Transformer (P-DAT). This novel transformer architecture stands out by concurrently capturing both global and local information, while maintaining computational efficiency. Regarding the self attention mechanism, we have extended the established attention mechanism between particles to encompass the attention mechanism between particle features. The particle attention module computes particle level interactions across all the particles, while the channel attention module computes attention scores between particle features, which naturally captures jet level interactions by taking all particles into account. These two kinds of attention mechanisms can complement each other. Further, we incorporate both the pairwise particle interactions and the pairwise jet feature interactions in the attention mechanism. We demonstrate the effectiveness of the P-DAT architecture in classic top tagging and quark-gluon discrimination tasks, achieving competitive performance compared to other benchmark strategies.

hep-ph↗

arXiv2023

Exploring lepton flavor violation phenomena of the $Z$ and Higgs bosons at electron-proton colliders

We comprehensively study the potential for discovering lepton flavor violation (LFV) phenomena associated with the $Z$ and Higgs bosons at the LHeC and FCC-he. Our meticulous investigation reveals the remarkable suitability of electron-proton colliders for probing these rare new physics signals. This is due to the distinct advantages they offer, including negligible pileups, minimal QCD backgrounds, electron-beam polarization $P_e$, and the capability of distinguishing the charged-current from neutral-current processes. In our pursuit of LFV of the $Z$ boson, we employ an innovative indirect probe, utilizing the $t$-channel mediation of the $Z$ boson in the process $p e^- \to j \tau^-$. For LFV in the Higgs sector, we scrutinize direct observations of the on-shell decays of $H\to e^+\tau^-$ and $H\to \mu^\pm\tau^\mp$ through the charged-current production of $H$. Focusing on $H\to e^+\tau^-$ proves highly efficient due to the absence of positron-related backgrounds in the charged-current modes at electron-proton colliders. Through a dedicated signal-to-background analysis with the boosted decision tree algorithm, we demonstrate that the LHeC with the total integrated luminosity of $1{\,{\rm ab}^{-1}}$ can put significantly lower $2\sigma$ bounds than the HL-LHC with $3{\,{\rm ab}^{-1}}$. Specifically, we find ${\rm{Br}}(Z\to e\tau)< 2.2 \times 10^{-7}$, ${\rm{Br}}(H\to e\tau) <1.7 \times 10^{-4} $, and ${\rm{Br}}(H\to \mu\tau) < 1.0 \times 10^{-4}$. Furthermore, our study uncovers the exceptional precision of the FCC-he in measuring the LFV signatures of the $Z$ and Higgs bosons, which indicates the potential for future discoveries in this captivating field.

hep-ph↗

arXiv2023

Probing the electroweak $4b + \ell + {\rlap{\,/}{E}_T}$ final state in type I 2HDM at the LHC

Most of the experimental searches of the non-Standard Model Higgs boson(s) at the LHC rely on the QCD induced production modes. However, in some beyond Standard Model frameworks, the additional Higgs bosons can have fermiophobic behaviour. The type I two Higgs doublet model considered here is a perfect example where all the additional Higgs bosons exhibit fermiophobic nature over a large region of parameter space. Thus the electroweak productions of these new Higgs bosons are more dominant over the QCD induced processes. In scenarios with light pseuodoscalar ($A$) which is bound to decay dominantly to $b\bar{b}$, even being fermiophobic, the $4b + W$ state via $p p \to H^\pm A \to (AW)A \to 4b + W$ and followed by the leptonic decay of $W$ boson can surpass the QCD initiated $4b$ final state. However, the signal gets overshadowed by large $t\bar{t}+$jets background and hence constructing a suitable discriminator based on the signal hypothesis and signal topology is necessary. We devised a $\chi^2$ variable as the most suitable signal-background discrimintor to reduce the background by a sizable amount and showed the discovery reach ( $>3\sigma$) of the electroweak initiated $4b+ \ell + {\rlap{\,/}{E}_T}$ final state at the LHC.

hep-ph↗

arXiv2023

$\tau^\pm \nu \gamma\gamma$ and $\ell^\pm \ell^\pm \gamma \gamma {\rlap{\,/}{E}_T} X$ to probe the fermiophobic Higgs boson with high cutoff scales

The light fermiophobic Higgs boson $h_{\rm f}$ in the type-I two-Higgs-doublet model can evade the current search programs at the LHC since its production through the quark-antiquark annihilation and gluon fusion is not feasible. The particle can be more elusive if the model retains stability up to the Planck scale because the efficient discovery channels are missing from the existing search chart. Through the comprehensive scanning, we show that all the viable parameter points with the Planck cutoff scale require $ m_{h_{\rm f}} \in[80,\, 120]{\;{\rm GeV}}$ and $M_{A/H^\pm} \in [90,\,150]{\;{\rm GeV}}$. Since $h_{\rm f}h_{\rm f}\to \gamma\gamma W^+ W^-$ and $H^\pm \to \tau^\pm \nu/h_{\rm f}W^\pm$ are dominant in this case, two final states are more efficient to probe $h_{\rm f}$ than the conventional search mode of $4\gamma+W^\pm/Z$. One is $\tau^\pm\nu \gamma\gamma$ from $pp \to H^\pm(\to\tau^\pm\nu) h_{\rm f}(\to \gamma\gamma)$ and the other is $\ell^\pm \ell^\pm \gamma\gamma {\rlap{\,/}{E}_T} X$ ($\ell^\pm=e^\pm,\mu^\pm$) from $pp \to H^\pm(\to h_{\rm f}W^\pm) h_{\rm f} \to \gamma\gamma W^+ W^-W^\pm $, $pp \to H^\pm(\to h_{\rm f} W^\pm) A(\to h_{\rm f} Z) \to \gamma\gamma W^+ W^- W^\pm Z $, and $pp \to H^+(\to h_{\rm f} W^+)H^-(\to h_{\rm f} W^-)\to \gamma\gamma W^+ W^- W^+ W^-$. The inclusive $\ell^\pm \ell^\pm \gamma\gamma {\rlap{\,/}{E}_T} X$ consists of a same-sign dilepton, two prompt photons, and missing transverse energy. We perform the signal-background analysis at the detector level. With the total integrated luminosity of $300\;{\rm fb}^{-1}$ and the 5\% background uncertainty, two proposed channels at the 14 TeV LHC yield signal significances above five in the entire viable parameter space of the fermiophobic type-I with a high cutoff scale.

hep-ph↗

arXiv2022

Deep Learning Jet Image as a Probe of Light Higgsino Dark Matter at the LHC

Higgsino in supersymmetric standard models can play the role of dark matter particle. In conjunction with the naturalness criterion, the higgsino mass parameter is expected to be around the electroweak scale. In this work, we explore the potential of probing the nearly degenerate light higgsinos with machine learning at the LHC. By analyzing jet images and other jet substructure information, we use the Convolutional Neural Network(CNN) to enhance the signal significance. We find that our deep learning jet image method can improve the previous result based on the conventional cut-flow by about a factor of two at the High-Luminosity LHC.

hep-ph↗

arXiv2022

Enhanced Higgs pair production from higgsino decay at the HL-LHC

The scenario of multi-sector SUSY breaking predicts pseudo-goldstinos which are not absorbed by the gravitino and their mass can be as low as ${{\cal O} (0.1)}$ GeV. Since the interactions of pseudo-goldstinos are not so weak as gravitino, a produced higgsino can decay to a pseudo-goldstino plus a Higgs boson insider the detector at the LHC, and thus the higgsino pair production can lead to the signal of Higgs pair plus missing energy. For the scenario of natural SUSY which requires rather light higgsinos, such events may sizably outnumber the Higgs pair events predicted by the SM and be accessible at the HL-LHC (14 TeV with a luminosity of 3~$\rm{ab}^{-1}$). In this work we examine the observability of such Higgs pair plus missing energy from the decay of light higgsinos produced at the HL-LHC. Considering three channels of the Higgs-pair decay ($bbWW^*$, $bb\gamma\gamma$, $bbbb$), our detailed Monte Carlo simulations for the signal and backgrounds show that the best channel is $bbbb+\textrm{E\!\!\!\! \!\slash}_T$, whose statistical significance can reach $2\sigma$ level for a light higgsino allowed by current experiments. This is over the SM Higgs pair result which is about $1.8\sigma$.

hep-ph↗

arXiv2021

Detecting an axion-like particle with machine learning at the LHC

Axion-like particles (ALPs) appear in various new physics models with spontaneous global symmetry breaking. When the ALP mass is in the range of MeV to GeV, the cosmology and astrophysics bounds are so far quite weak. In this work, we investigate such light ALPs through the ALP-strahlung production processes $pp \to W^\pm a, Z a$ with the sequential decay $a \to \gamma\gamma$ at the 14 TeV LHC with an integrated luminosity of 3000 fb$^{-1}$ (HL-LHC). Building on the concept of jet image which uses calorimeter towers as the pixels of the image and measures a jet as an image, we investigate the potential of machine learning techniques based on convolutional neural network (CNN) to identify the highly boosted ALPs which decay to a pair of highly collimated photons. With the CNN tagging algorithm, we demonstrate that our approach can extend current LHC sensitivity and probe the ALP mass range from 0.3~GeV to 5~GeV. The obtained bounds are stronger than the existing limits on the ALP-photon coupling.

hep-ph↗

arXiv2021

Heavy Bino and Slepton for Muon g-2 Anomaly

In light of very recent E989 experimental result, we investigate the possibility that heavy sparticles explain the muon g-2 anomaly. We focus on the bino-smuon loop in an effective SUSY scenario, where a light gravitino plays the role of dark matter and other sparticles are heavy. Due to the enhancement of left-right mixing of smuons by heavy higgsinos, the contribution of bino-smuon loop can sizably increase the prediction of muon g-2 to the experimental value. Under collider and vacuum stability constraints, we find that TeV scale bino and smuon can still account for the new muon g-2 anomaly. The implications for LHC phenomenology are also discussed.

hep-ph↗

arXiv2021

Photon-jet events as a probe of axion-like particles at the LHC

Axion-like particles (ALPs) are predicted by many extensions of the Standard Model (SM). When ALP mass lies in the range of MeV to GeV, the cosmology and astrophysics will be largely irrelevant. In this work, we investigate such light ALPs through the ALP-strahlung process $pp \to V a (\to \gamma\gamma)$ at the 14 TeV LHC with an integrated luminosity of 3000 fb$^{-1}$ (HL-LHC). With the photon-jet algorithm, we demonstrate that our approach can probe the mass range of ALPs, which is inaccessible to previous LHC experiments. The obtained result can surpass the existing limits on ALP-photon coupling in the ALP mass range from 0.3 GeV to 10 GeV.

hep-ph↗