arXiv subjects

Lorenzo Rossi

Publications and source records attributed to Lorenzo Rossi.

At least 19 recordsLinked to original sources

arXiv2026

Symbolic Extraction of Non-Perturbative Transverse-Momentum-Dependent Distributions from Drell-Yan Data

We present an analytical parametrization of the non-perturbative transverse-momentum-dependent (TMD) parton distribution function of unpolarized quarks, extracted from Drell-Yan data using a combination of neural-network fitting and symbolic regression. A factorized neural network is trained directly against experimental cross-section data from fixed-target, Tevatron, RHIC, and LHC experiments at next-to-next-to-next-to-leading logarithmic accuracy, and symbolic regression is subsequently applied to each network component to discover compact analytical expressions. The final formula is selected from a Pareto front in the space of expression complexity and experimental $\chi^2$, yielding a closed-form non-perturbative function with 9 free numerical constants that achieves $\chi^2/\mathrm{ndf}=1.040$ over 482 data points. A non-trivial $x$-$b_T$ cross term is retained even under a sparsity prior that biases it toward zero, indicating a mild but genuine correlation between the longitudinal momentum fraction and the transverse momentum. This work demonstrates that symbolic regression is a viable tool for bridging flexible machine-learning fits and interpretable analytical TMD parametrizations, and opens a systematic path toward data-driven discovery of specific features of non-perturbative QCD.

hep-ph↗

arXiv2026

Transverse-momentum resummation effects on angular coefficients in Z and W boson hadroproduction

We present a comprehensive analysis of the angular coefficients of $Z$ and $W$ boson production at hadron colliders in different kinematical ranges, using data from the ATLAS, LHCb, and CMS Collaborations at the LHC, as well as CDF data at the Tevatron. We provide theoretical predictions obtained by consistently combining the resummation of logarithmically enhanced QCD corrections at small transverse momenta $q_T$ up to next-to-next-to-leading logarithmic accuracy (NNLL) with fixed-order calculations at next-to-leading order (NLO), valid at large $q_T$. We quantify the impact of transverse-momentum resummation on the angular coefficients. We find that the inclusion of resummation effects leads to a moderate and systematic improvement in the description of the data in the intermediate $q_T$ region, $q_T \sim 20-50$~GeV, for several angular coefficients, while in the remaining cases it does not degrade the agreement of fixed-order QCD predictions with experimental measurements.

hep-ph↗

arXiv2026

Factorization of the Energy-Energy Correlation in the two-jet limit in the massive case

We consider non-logarithmic heavy-quark mass effects in the factorization and resummation of the Energy-Energy-Correlation (EEC) function, in the two-jet limit. We define a new, "partial" event fraction, restricted to the two-jet region and excluding the forward region, whose calculation at first order requires to consider real emission diagrams only, in $D=4$ space-time dimensions (no need to consider virtual diagrams or take $D \ne 4$). In order to determine explicitly the next-to-leading order coefficient function and the remainder function (both entering the standard resummation formula), we evaluate numerically the EEC spectrum at first-order in $\alpha_S$, finding good agreement with previous calculations. To have a smooth massless limit, a new, improved factorization scheme is proposed, in which the coefficient function also depends on the correlation angle $\chi$.

hep-ph↗

arXiv2026

Natural Identifiers for Privacy and Data Audits in Large Language Models

Assessing the privacy of large language models (LLMs) presents significant challenges. In particular, most existing methods for auditing differential privacy require the insertion of specially crafted canary data during training, making them impractical for auditing already-trained models without costly retraining. Additionally, dataset inference, which audits whether a suspect dataset was used to train a model, is infeasible without access to a private non-member held-out dataset. Yet, such held-out datasets are often unavailable or difficult to construct for real-world cases since they have to be from the same distribution (IID) as the suspect data. These limitations severely hinder the ability to conduct scalable, post-hoc audits. To enable such audits, this work introduces natural identifiers (NIDs) as a novel solution to the above-mentioned challenges. NIDs are structured random strings, such as cryptographic hashes and shortened URLs, naturally occurring in common LLM training datasets. Their format enables the generation of unlimited additional random strings from the same distribution, which can act as alternative canaries for audits and as same-distribution held-out data for dataset inference. Our evaluation highlights that indeed, using NIDs, we can facilitate post-hoc differential privacy auditing without any retraining and enable dataset inference for any suspect dataset containing NIDs without the need for a private non-member held-out dataset.

cs.LG↗

arXiv2026

Benchmarking Empirical Privacy Protection for Adaptations of Large Language Models

Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining, where overlaps and interdependencies with adaptation data can undermine privacy despite DP efforts. To analyze this issue in practice, we investigate privacy risks under DP adaptations in LLMs using state-of-the-art attacks such as robust membership inference and canary data extraction. We benchmark these risks by systematically varying the adaptation data distribution, from exact overlaps with pretraining data, through in-distribution (IID) cases, to entirely out-of-distribution (OOD) examples. Additionally, we evaluate how different adaptation methods and different privacy regimes impact the vulnerability. Our results show that distribution shifts strongly influence privacy vulnerability: the closer the adaptation data is to the pretraining distribution, the higher the practical privacy risk at the same theoretical guarantee, even without direct data overlap. We find that parameter-efficient fine-tuning methods, such as LoRA, achieve the highest empirical privacy protection for OOD data. Our benchmark identifies key factors for achieving practical privacy in DP LLM adaptation, providing actionable insights for deploying customized models in sensitive settings. Looking forward, we propose a structured framework for holistic privacy assessment beyond adaptation privacy, to identify and evaluate risks across the full pretrain-adapt pipeline of LLMs.

cs.LG↗

arXiv2026

Collective decay of interacting bosons

We study a bosonic analog of the paradigmatic Dicke model of superradiance, comprising interacting bosonic modes subject to fully symmetric collective decay. Depending on the interaction strength, we uncover qualitatively distinct regimes of emission. For strong interactions, the emission closely resembles Dicke superradiance, with perturbative corrections arising from the presence of additional levels. For weaker interactions, the bosonic statistics qualitatively changes the dynamics, leading to a crossover to subradiant emission. Remarkably, we show that the dynamics in this regime can be described by rate equations analogous to those of the Dicke model despite the large accessible bosonic Hilbert space. Our findings are based on a combination of analytical arguments and large-scale numerics enabled by the permutational symmetry of the problem and may be probed in circuit QED experiments.

quant-ph↗

arXiv2026

Neural network modeling of many-body super- and sub-radiant dynamics

There is significant interest in exploring novel phenomena in quantum light-matter interfaces, which are driven by the combination of structured dissipation and long-range interactions that are typical in such systems. To this end, it is important to develop new general numerical simulation techniques, which can access large system sizes and are not based on semi-classical approaches. Here, we report the first application of neural quantum states to obtain the dissipative dynamics of light-matter-coupled systems beyond what is accessible with exact and tensor-network calculations. We specifically apply this method to simulate the many-body emission dynamics of approximately 40 atoms, arranged in dense arrays in one and two dimensions. These systems have been chosen because they can support prominent subradiant dynamics at late times and could be realized with cold atomic quantum simulators.

quant-ph↗

arXiv2026

A global analysis of Energy-Energy Correlation data: determination of $α_S$ and non-perturbative QCD parameters

We present a comprehensive global analysis of Energy-Energy Correlation (EEC) data in electron-positron annihilation into hadrons, spanning a wide range of center-of-mass energies ($7.7\,\,\text{GeV}\!\leq\!\sqrt{s}\!\leq\! 91.2\,\,\text{GeV})$. In the back-to-back (two-jet) region, we resum to all orders the logarithmically-enhanced contributions up to next-to-next-to-next-to-leading logarithmic (N$^3$LL) accuracy. The resummed results are consistently matched to fixed-order calculations up to $\mathcal{O}(α_S^3)$. Our resummation formalism also incorporates dominant heavy-quark mass effects and models non-perturbative power corrections by means of an analytic dispersive approach. A simultaneous fit yields an excellent description of experimental data across all energies, enabling a precise determination of the strong coupling, $α_S(m_Z^2) = 0.119 \pm 0.002$, as well as the non-perturbative parameters, including those characterizing the Collins--Soper evolution kernel. Our analysis includes, for the first time in a global fit, datasets from the ALEPH and AMY collaborations.

hep-ph↗

arXiv2026

Drell--Yan lepton pair production at low invariant masses: transverse-momentum resummation and non-perturbative effects in QCD

We consider the transverse-momentum ($q_T$) distribution of Drell-Yan lepton pairs produced with invariant masses ($M$) from low values up to the $Z$-boson peak ($4\leq M \leq 116$~GeV). We present perturbative predictions obtained by consistently combining the resummation of logarithmically enhanced QCD corrections at small $q_T$ ($q_T \ll M$) up to next-to-next-to-next-to-next-to-leading logarithmic accuracy with the available fixed-order calculations at next-to-next-to-leading order (i.e. $\mathcal{O}(α_S^3)$) valid at large $q_T$. For very low $q_T$ ($q_T\sim Λ_{\mathrm{QCD}}$), non-perturbative (NP) QCD effects become dominant and have been included through a NP form factor with a small number of free-parameters. We compare our results with multiple experimental datasets from hadron colliders, finding excellent agreement between theory and data. By fitting the NP parameters, we achieve a precise extraction of the NP form factor and the so-called Collins-Soper kernel.

hep-ph↗

arXiv2025

Study of SIDIS Unpolarized Cross Sections from a $^3$He Target with the Solenoidal Large Intensity Device at JLab

In this paper we present a detailed impact study of semi-inclusive deep inelastic scattering unpolarized cross sections' measurements using the proposed SoLID apparatus at Jefferson Lab. This type of data, collected at large Bjorken $x_{bj}$, moderate values of $Q^2$ and small values of the transverse momentum of produced hadrons, $P_{hT}$, allows to study transverse momentum dependent (TMD) parton distribution and fragmentation functions in a still poorly explored region. We present the projected results for charged light mesons based on simulated data. For the azimuthal-angle integrated cross sections we adopt the TMD framework up to the next-to-next-to-next-to-leading-logarithmic (N3LL) accuracy, while a simpler TMD parton model is employed for the study of azimuthal angular dependencies.

nucl-ex↗

arXiv2025

New insights from the flavor dependence of quark transverse momentum distributions in the pion

We update our previous extraction of transverse momentum distributions of unpolarized quarks in the pion by implementing a more comprehensive description of theoretical uncertainties and, for the first time, by exploring possible differences among quark flavors. We extract such distributions from all available data for unpolarized pion-nucleus Drell-Yan processes, where the cross section is differential in the transverse momentum of the final lepton pair. The cross section involves transverse momentum distributions in the nucleon, that we consistently take from our previous studies.

hep-ph↗

arXiv2025

Non-linear instability of slowly rotating Kerr-AdS black holes

Generic scalar perturbations on a fixed slowly rotating Kerr-AdS black hole background exhibit stable trapping, that is, the scalar field remains in a region between the exterior of the black hole and the AdS boundary for a very long time, decaying only inverse logarithmically in time. We study this effect employing fully general simulations that take into account the non-linear backreaction of the scalar field on the geometry. We find that the stable trapping of generic perturbations of Kerr-AdS persists at the non-linear level. Furthermore, the spacetime settles into a time-dependant and non-axisymmetric black hole which differs from Kerr-AdS. Since our perturbations are generic, our results indicate that slowly rotating Kerr-AdS black holes are non-linearly unstable.

hep-th↗

arXiv2025

Extraction of Dihadron Fragmentation Functions at NNLO with and without Neural Networks

We present a new extraction of unpolarized Dihadron Fragmentation Functions, which describe the probability density for an unpolarized parton to fragment into a $π^+ π^-$ pair. Our analysis is based on data from the BELLE collaboration. We improve on previous determinations in several key aspects: we employ state-of-the-art perturbative QCD calculations up to next-to-next-to-leading order (NNLO); we limit the use of Monte Carlo event generators to estimating the relative contributions of different flavors, a necessary input due to the limited flavor sensitivity of the available data; and, in addition to a traditional fit based on a physics-informed functional form, we explore a Neural Network parametrization. This latter approach paves the way for more robust and flexible determinations of Dihadron Fragmentation Functions using machine learning techniques.

hep-ph↗

arXiv2025

Membership Inference Attacks on Sequence Models

Sequence models, such as Large Language Models (LLMs) and autoregressive image generators, have a tendency to memorize and inadvertently leak sensitive information. While this tendency has critical legal implications, existing tools are insufficient to audit the resulting risks. We hypothesize that those tools' shortcomings are due to mismatched assumptions. Thus, we argue that effectively measuring privacy leakage in sequence models requires leveraging the correlations inherent in sequential generation. To illustrate this, we adapt a state-of-the-art membership inference attack to explicitly model within-sequence correlations, thereby demonstrating how a strong existing attack can be naturally extended to suit the structure of sequence models. Through a case study, we show that our adaptations consistently improve the effectiveness of memorization audits without introducing additional computational costs. Our work hence serves as an important stepping stone toward reliable memorization audits for large sequence models.

cs.CR↗

arXiv2025

A Neural-Network Extraction of Unpolarised Transverse-Momentum-Dependent Distributions

We present the first extraction of transverse-momentum-dependent distributions of unpolarised quarks from experimental Drell-Yan data using neural networks to parametrise their nonperturbative part. We show that neural networks outperform traditional parametrisations providing a more accurate description of data. This work establishes the feasibility of using neural networks to explore the multi-dimensional partonic structure of hadrons and paves the way for more accurate determinations based on machine-learning techniques.

hep-ph↗

arXiv2024

Flavor dependence of unpolarized quark Transverse Momentum Distributions from a global fit

We present an extraction of the unpolarized transverse-momentum-dependent parton distribution and fragmentation functions that takes into account possible differences between quark flavors and final-state hadrons. The extraction is based on experimental measurements from Drell-Yan processes and semi-inclusive deep-inelastic scattering, whose combination is essential to distinguish flavor differences. The analysis is carried out at N$^3$LL accuracy. The extracted flavor-dependent distributions give a very good description of the data ($χ^2/N_{\rm dat} = 1.08$). The resulting uncertainties take fully into account also the uncertainties in the determination of the corresponding collinear distributions.

hep-ph↗

arXiv2024

Exploring the three-dimensional momentum distribution of longitudinally polarized quarks in the proton

By analyzing experimental data on semi-inclusive deep inelastic scattering off longitudinally polarized targets, we extract the transverse momentum dependence of the quark helicity distribution, i.e., the difference between the three-dimensional motion of quarks with polarization parallel or antiparallel to the longitudinal polarization of the parent hadron. We perform the analysis at next-to-leading (NLL) and next-to-next-to-leading (NNLL) perturbative accuracy. The quality of the fit is very good for both cases, reaching a $χ^2$ per number of data points equal to $1.11$ and $1.09$, respectively. Although the limited number of data points leads to significant uncertainties, the data are consistent with an interpretation in which the helicity distribution is narrower in transverse momentum than the unpolarized distribution.

hep-ph↗

arXiv2024

Numerical implementation of evolution equations for twist-3 collinear PDFs

Twist-3 collinear parton distribution functions (PDFs) are matrix elements of quark-gluon-quark or three-gluons light-cone operators. They depend on three momentum fraction variables, which are restricted to a hexagon region, and the evolution kernels are defined via two-dimensional convolution in these variables. We present the numerical realisation of the twist-3 evolution equations at leading order in the strong coupling for all kinds of twist-3 PDF (quark, gluon, chiral-even/odd, etc). We provide two independent codes (in C and Fortran) that have been extensively cross-checked, and are ready-to-use. We supplement the paper with a review of known properties of twist-3 PDFs.

hep-ph↗