arXiv subjects

Darius A. Faroughy

Publications and source records attributed to Darius A. Faroughy.

At least 19 recordsLinked to original sources

arXiv2026

Neural Scaling Laws for Jet Generation

Recently observed empirical scaling laws describe the performance of foundation-type models as three independent key quantities -- dataset size, compute, and model parameters -- are modified. Extracting these scaling laws informs the training of large complex models for which the tuning of hyperparameters in traditional ways is not feasible. This work for the first time explores if scaling laws can also be observed for the task of particle jet generation -- both relevant as a pre-training objective for foundation models and as in-situ simulation by itself. We indeed replicate the key logarithmic scaling law behavior for model-size scaling. Beyond studying the next token prediction validation loss of the generative model, we also study the sliced Wasserstein distance of five physical quantities that are not immediately available to the model during training. Our study shows that this quantity is monotonically related to the next token prediction validation loss, meaning that this loss is indeed a good proxy for the physics performance. For the scaling with dataset size and compute, we observe substantially weaker scaling behavior of both the loss and the sliced Wasserstein distance. We analyze this behavior by introducing the concept of a learnable window, and argue that autoregressive next token prediction on jet constituents exhibits comparatively rapid saturation relative to language-model studies. We discuss possible origins of this behavior, including the stochastic nature of QCD radiation and differences between generative and supervised learning tasks in collider physics.

hep-ph↗

arXiv2026

My Chemical Harness: Evolutionary Molecular Design over Synthetic Pathways with Large Language Model Agents

Designing molecules with target properties is most useful when candidate structures are accompanied by feasible synthetic routes. We introduce My Chemical Harness, a route-native evolutionary framework for goal-directed molecular design in which the search population consists of executable synthetic pathways rather than isolated molecular graphs. Each route is built from purchasable building blocks and reaction templates, executed by deterministic chemistry tools, and scored through task-specific molecular oracles. Large language models (LLMs) are used only as strategy controllers that select high-level preferences over route length, move type, reaction families, motifs, and exploration pressure, while local code performs route construction, validation, deduplication, scoring, selection, and memory updates. This separation lets the LLM guide exploration without allowing it to introduce hallucinated products or unsupported reaction steps. On a soluble epoxide hydrolase proxy task, our LLM agent improves over single pass LLM and deterministic controllers, reaching state-of-the-art performance across the sEH score, synthetic accessibility score, and AiZynthFinder success rate metrics. These results suggest that constrained LLM agents can play a significant role in molecular discovery without requiring training, fine-tuning, or dedicated generative models.

physics.chem-ph↗

arXiv2026

Collider-Bench: Benchmarking AI Agents with Particle Physics Analysis Reproduction

Autonomous language-model agents are increasingly evaluated on long-horizon tool-use tasks, but existing benchmarks rarely capture the complexity and nuance of real scientific work. To address this gap, we introduce Collider-Bench, a benchmark for evaluating whether LLM agents can reproduce experimental analyses from the Large Hadron Collider (LHC) using only public papers and open scientific software. Such analyses are often difficult to reproduce because the public toolchain only approximates the software used internally by the experimental collaborations, while the published papers inevitably omit implementation details needed for a faithful reconstruction. Agents must therefore rely on physical reasoning, domain knowledge, and trial-and-error to fill these gaps. Each task requires the agent to turn a published analysis into an executable simulation-and-selection pipeline and submit predicted collision event yields in specified signal regions. These predictions are evaluated with standard histogram metrics that provide continuous fidelity scores without a hand-written rubric. We also report the computational cost incurred by each agent per task. Finally, we evaluate the codebase and full session trace using an LLM judge to catch qualitative failure modes such as fabrications, hallucinations and duplications. We release an initial set of tasks drawn from LHC searches, together with a containerized sandbox and event simulation tools. We evaluate across a capability ladder of general purpose coding agents. Our results show that on average no agent reliably beats the physicist-in-the-loop solution.

cs.LG↗

arXiv2026

Prior-Fitted Functional Flow: In-Context Generative Models for Pharmacokinetics

We introduce Prior-Fitted Functional Flows, a generative foundation model for pharmacokinetics that enables zero-shot population synthesis and individual forecasting without manual parameter tuning. We learn functional vector fields, explicitly conditioned on the sparse, irregular data of an entire study population. This enables the generation of coherent virtual cohorts as well as forecasting of partially observed patient trajectories with calibrated uncertainty. We construct a new open-access literature corpus to inform our priors, and demonstrate state-of-the-art predictive accuracy on extensive real-world datasets.

cs.LG↗

arXiv2025

Multimodal Generative Flows for LHC Jets

Generative modeling of high-energy collisions at the Large Hadron Collider (LHC) offers a data-driven route to simulations, anomaly detection, among other applications. A central challenge lies in the hybrid nature of particle-cloud data: each particle carries continuous kinematic features and discrete quantum numbers such as charge and flavor. We introduce a transformer-based multimodal flow that extends flow-matching with a continuous-time Markov jump bridge to jointly model LHC jets with both modalities. Trained on CMS Open Data, our model can generate high fidelity jets with realistic kinematics, jet substructure and flavor composition.

hep-ph↗

arXiv2025

SURFing to the Fundamental Limit of Jet Tagging

Beyond the practical goal of improving search and measurement sensitivity through better jet tagging algorithms, there is a deeper question: what are their upper performance limits? Generative surrogate models with learned likelihood functions offer a new approach to this problem, provided the surrogate correctly captures the underlying data distribution. In this work, we introduce the SUrrogate ReFerence (SURF) method, a new approach to validating generative models. This framework enables exact Neyman-Pearson tests by training the target model on samples from another tractable surrogate, which is itself trained on real data. We argue that the EPiC-FM generative model is a valid surrogate reference for JetClass jets and apply SURF to show that modern jet taggers may already be operating close to the true statistical limit. By contrast, we find that autoregressive GPT models unphysically exaggerate top vs. QCD separation power encoded in the surrogate reference, implying that they are giving a misleading picture of the fundamental limit.

hep-ph↗

arXiv2025

Aspen Open Jets: Unlocking LHC Data for Foundation Models in Particle Physics

Foundation models are deep learning models pre-trained on large amounts of data which are capable of generalizing to multiple datasets and/or downstream tasks. This work demonstrates how data collected by the CMS experiment at the Large Hadron Collider can be useful in pre-training foundation models for HEP. Specifically, we introduce the AspenOpenJets dataset, consisting of approximately 178M high $p_T$ jets derived from CMS 2016 Open Data. We show how pre-training the OmniJet-$α$ foundation model on AspenOpenJets improves performance on generative tasks with significant domain shift: generating boosted top and QCD jets from the simulated JetClass dataset. In addition to demonstrating the power of pre-training of a jet-based foundation model on actual proton-proton collision data, we provide the ML-ready derived AspenOpenJets dataset for further public use.

hep-ph↗

arXiv2025

On the EFT validity for Drell-Yan tails at the LHC

In this article, we examine the validity range of the Effective Field Theory (EFT) description of high-energy Drell-Yan processes at the LHC. To this purpose, we consider explicit mediators that contribute to these processes in the $s$- and $t$-channels, comparing their effects in Drell-Yan distributions with the ones obtained by matching onto the corresponding EFT. We determine the conditions for the EFT results to accurately describe these scenarios. In particular, we explore the impact of including dimension-eight $(d=8)$ operators in the faster convergence of the EFT series, at the analytical and numerical level, considering contributions to the cross section up to the square of $d=8$ EFT operator insertions. Moreover, we discuss the possible implications of clipping LHC data and illustrate results for a specific New-Physics scenario motivated by low-energy flavor data.

hep-ph↗

arXiv2024

Probing Lepton Flavor Violation in Meson Decays with LHC Data

In this letter, we use LHC data from the Drell-Yan processes $pp\to\ell_i\ell_j$ (with $i\neq j$) to derive model-independent upper limits on lepton-flavor-violating meson decays. Our analysis is based on an Effective Field Theory (EFT) approach and it does not require a specific assumption regarding the basis of effective operators. We find that current LHC data (140~$\mathrm{fb}^{-1}$) already provides competitive limits on $B(B\to πe τ)$ and $B(B\to πμτ)$ with respect to the ones obtained through experimental searches at the $B$-factories. Moreover, we derive upper limits on several decays that have not been searched for experimentally yet, such as $D^0\to eτ$ in the charm sector, and various semileptonic decays such as $B\to ρμτ$, $B_s\to K μτ$ and $B_s\toϕμτ$. Lastly, we discuss the validity of the EFT description of LHC data and the impact of loop corrections in our analysis.

hep-ph↗

arXiv2023

EPiC-ly Fast Particle Cloud Generation with Flow-Matching and Diffusion

Jets at the LHC, typically consisting of a large number of highly correlated particles, are a fascinating laboratory for deep generative modeling. In this paper, we present two novel methods that generate LHC jets as point clouds efficiently and accurately. We introduce \epcjedi, which combines score-matching diffusion models with the Equivariant Point Cloud (EPiC) architecture based on the deep sets framework. This model offers a much faster alternative to previous transformer-based diffusion models without reducing the quality of the generated jets. In addition, we introduce \epcfm, the first permutation equivariant continuous normalizing flow (CNF) for particle cloud generation. This model is trained with {\it flow-matching}, a scalable and easy-to-train objective based on optimal transport that directly regresses the vector fields connecting the Gaussian noise prior to the data distribution. Our experiments demonstrate that \epcjedi and \epcfm both achieve state-of-the-art performance on the top-quark JetNet datasets whilst maintaining fast generation speed. Most notably, we find that the \epcfm model consistently outperforms all the other generative models considered here across every metric. Finally, we also introduce two new particle cloud performance metrics: the first based on the Kullback-Leibler divergence between feature distributions, the second is the negative log-posterior of a multi-model ParticleNet classifier.

hep-ph↗

arXiv2023

Drell-Yan Tails Beyond the Standard Model

We investigate the high-$p_T$ tails of the $pp\to \ell ν$ and $pp \to \ell \ell$ Drell-Yan processes as probes of New Physics in semileptonic interactions with an arbitrary flavor structure. For this purpose, we provide a general decomposition of the $2\to2$ scattering amplitudes in terms of form-factors that we match to specific scenarios, such as the Standard Model Effective Field Theory (SMEFT), including all relevant operators up to dimension-$8$, as well as ultraviolet scenarios giving rise to tree-level exchange of new bosonic mediators with masses at the TeV scale. By using the latest LHC run-II data in the monolepton ($eν$, $μν$, $τν$) and dilepton ($ee$, $μμ$, $ττ$, $eμ$, $eτ$, $μτ$) production channels, we derive constraints on the SMEFT Wilson coefficients for semileptonic four-fermion and dipole operators with the most general flavor structure, as well as on all possible leptoquark models. For the SMEFT, we discuss the range of validity of the EFT description, the relevance of $\mathcal{O}(1/Λ^2)$ and $\mathcal{O}(1/Λ^4)$ truncations, the impact of $d=8$ operators and the effects of different quark-flavor alignments. Finally, as a highlight, we extract for several New Physics scenarios the combined limits from high-$p_T$ processes, electroweak pole measurements and low-energy flavor data for the $b\to cτν$ transition, showing the complementarity between these different observables. Our results are compiled in {\tt HighPT}, a package in {\tt Mathematica} which provides a simple way for users to extract the Drell-Yan tails likelihoods for semileptonic effective operators and for leptoquark models.

hep-ph↗

arXiv2022

Accessing CKM suppressed top decays at the LHC

We propose an extension of the existing experimental strategy for measuring branching fractions of top quark decays, targeting specifically $t\to j_q W$, where $j_q$ is a light quark jet. The improved strategy uses orthogonal $b$- and $q$-taggers, and adds a new observable, the number of light-quark-tagged jets, to the already commonly used observable, the fraction of $b$-tagged jets in an event. Careful inclusion of the additional complementary observable significantly increases the expected statistical power of the analysis, with the possibility of excluding $|V_{tb}|=1$ at $95\%$ C.L. at the HL-LHC, and accessing directly the standard model value of $|V_{td}|^2+|V_{ts}|^2$.

hep-ph↗

arXiv2022

Semileptonics at High-$p_T$

We consider the Drell-Yan processes $pp \to \ell \ell^{(\prime)}$ and $pp\to \ell \barν$ at high-$p_T$ as probes of New Physics effects in semileptonic transitions. For this purpose, we describe the $2\to2$ scattering amplitudes in terms of general form-factors which we match to the SMEFT at $\mathcal{O}(1/Λ^{4})$ including corrections from dimension-$8$ operators and to tree-level mediators with $\mathcal{O}$(TeV) masses arising from ultraviolet models. By using the latest LHC run-II data from monolepton and dilepton production channels, we derive constraints on the SMEFT and on leptoquark models, with the most general flavor structures. Our results are compiled into the Mathematica package 'HighPT', which provides a simple way to perform high-$p_T$ collider analyses of Drell-Yan production Beyond the Standard Model.

hep-ph↗

arXiv2022

On a model with two scalar leptoquarks -- $R_2$ and $S_3$

We discuss a model that can accommodate the $B$-physics anomalies, based on combining two scalar leptoquarks, $R_2$ and $S_3$, of mass $\mathcal{O}(1\,\mathrm{TeV})$, and that we proposed in our previous paper. We update the analysis of its parameter space and show that a model remains viable and consistent with a number of low energy and high energy flavor physics constraints. Since the model predicts a non-zero New Physics phase, we discuss the possibility to test its contribution to the neutron electric dipole moment and to the angular distributions of the exclusive $b\to cτ\bar ν$ decays. We find that the model can provide a significant enhancement to $\mathcal{B}(B\to K^{(\ast)}νν)$ and provides both the upper and lower bounds to $\mathcal{B}(B\to K^{(\ast)}μτ)$.

hep-ph↗

arXiv2022

Bayesian Probabilistic Modelling for Four-Tops at the LHC

Monte Carlo (MC) generators are crucial for analyzing data in particle collider experiments. However, often even a small mismatch between the MC simulations and the measurements can undermine the interpretation of the results. This is particularly important in the context of LHC searches for rare physics processes within and beyond the standard model (SM). One of the ultimate rare processes in the SM currently being explored at the LHC, $pp\to t\bar tt \bar t$ with its large multi-dimensional phase-space is an ideal testing ground to explore new ways to reduce the impact of potential MC mismodelling on experimental results. We propose a novel statistical method capable of disentangling the 4-top signal from the dominant backgrounds in the same-sign dilepton channel, while simultaneously correcting for possible MC imperfections in modelling of the most relevant discriminating observables -- the jet multiplicity distributions. A Bayesian mixture of multinomials is used to model the light-jet and $b$-jet multiplicities under the assumption of their conditional independence. The signal and background distributions generated from a deliberately mistuned MC simulator are used as model priors. The posterior distributions, as well as the signal and background fractions, are then learned from the data using Bayesian inference. We demonstrate that our method can mitigate the effects of large MC mismodellings in the context of a realistic $t\bar tt\bar t$ search, leading to corrected posterior distributions that better approximate the underlying truth-level spectra.

hep-ph↗

arXiv2021

Collider Signatures of Coannihilating Dark Matter in Light of the B-Physics Anomalies

Motivated by UV explanations of the $B$-physics anomalies, we study a dark sector containing a Majorana dark matter candidate and a coloured coannihilation partner, connected to the Standard Model predominantly via a $U_1$ vector leptoquark. A TeV scale $U_1$ leptoquark, which couples mostly to third generation fermions, is the only successful single-mediator description of the $B$-physics anomalies. After calculating the dark matter relic surface, we focus on the most promising experimental avenue: LHC searches for the coloured coannihilation partner. We find that the coloured partner hadronizes and forms meson-like bound states leading to resonant signatures at colliders reminiscent of the quarkonia decay modes in the Standard Model. By recasting existing dilepton and monojet searches we exclude coannihilation partner masses less than 280 GeV and 400 GeV, respectively. Since other existing collider searches do not significantly probe the parameter space, we propose a new dedicated search strategy for pair production of the coloured partner decaying into $bbττ$ final states and dark matter particles. This search is expected to probe the model up to dark matter masses around 600 GeV with current luminosity.

hep-ph↗

arXiv2021

Reading the footprints of the B-meson flavor anomalies

Motivated by the recent LHCb announcement of a $3.1σ$ violation of lepton-flavor universality in the ratio $R_K=Γ(B\to Kμ^+μ^-)/Γ(B\to K e^+ e^-)$, we present an updated, comprehensive analysis of the flavor anomalies seen in both neutral-current ($b\to s\ell^+\ell^-$) and charged-current ($b\to cτ\barν$) decays of $B$ mesons. Our study starts from a model-independent effective field-theory approach and then considers both a simplified model and a UV-complete extension of the Standard Model featuring a vector leptoquark $U_1$ as the main mediator of the anomalies. We show that the new LHCb data corroborate the emerging pattern of a new, predominantly left-handed, semileptonic current-current interaction with a flavor structure respecting a (minimally) broken $U(2)^5$ flavor symmetry. New aspects of our analysis include a combined analysis of the semileptonic operators involving tau leptons, including in particular the important constraint from $B_s$--$\bar B_s$ mixing, a systematic study of the effects of right-handed leptoquark couplings and of deviations from minimal flavor-symmetry breaking, a detailed analysis of various rare $B$-decay modes which would provide smoking-gun signatures of this non-standard framework (LFV decays, di-tau modes, and $B\to K^{(*)}ν\barν$), and finally an updated analysis of collider bounds on the leptoquark mass and couplings.

hep-ph↗

arXiv2021

On the single leptoquark solutions to the $B$-physics anomalies

We revisit the possibilities of accommodating the experimental indications of the lepton flavor universality violation in $b$-hadron decays in the minimal scenarios in which the Standard Model is extended by the presence of a single $\mathcal{O}(1\,\mathrm{TeV})$ leptoquark state. To do so we combine the most recent low energy flavor physics constraints, including $R_{K^{(\ast)}}^\mathrm{exp}$ and $R_{D^{(\ast)}}^\mathrm{exp}$, and combine them with the bounds on the leptoquark masses and their couplings to quarks and leptons as inferred from the direct searches at the LHC and the studies of the large $p_T$ tails of the $pp\to \ell\ell$ differential cross section. We find that none of the scalar leptoquarks of $m_\mathrm{LQ} \simeq 1÷2$ TeV can accommodate the $B$-anomalies alone. Only the vector leptoquark, known as $U_1$, can provide a viable solution which, in the minimal setup, provides an interesting prediction, i.e. a lower bound to the lepton flavor violating $b\to sμ^\pmτ^\mp$ decay modes, such as $\mathcal{B}(B\to Kμτ) \gtrsim 0.7\times 10^{-7}$.

hep-ph↗