arXiv subjects

Fu-Peng Li

Publications and source records attributed to Fu-Peng Li.

13 recordsLinked to original sources

arXiv2026

Bottomonium suppression with a machine-learning-informed Debye mass

Motivated by recent progress in data-driven approaches, we introduce a machine-learning (ML)-informed Debye mass, extracted from lattice-informed inputs, exclusively in the complex-valued heavy-quark Kent State University (KSU) potential. The resulting complex potential is used to solve the real-time Schr\"odinger equation within the quantum trajectories (QTraj) framework for the evolution of bottomonium in the quark-gluon plasma. We then compute the nuclear modification factors and double ratios for bottomonium $\Upsilon(1S)$, $\Upsilon(2S)$, and $\Upsilon(3S)$ states in Pb-Pb collisions at $\sqrt{s_{NN}} = 5.02$ TeV. We compare our ML-induced results with those from the original KSU model and with experimental measurements from ALICE, ATLAS, and CMS collaborations. We find that the machine-learned Debye mass leads to improved agreement with data, particularly for excited states, highlighting the utility of machine learning in modeling in-medium QCD effects.

hep-ph↗

arXiv2026

Bayesian inference of event-by-event collision geometry from charged-particle multiplicity in heavy-ion collisions

We propose the Inference-driven Participant Determination (IPD) method, a Bayesian framework for inferring event-by-event posterior distributions of the number of participants ($N_{\text{part}}$) and binary collisions ($N_{\text{coll}}$) from final-state charged-particle multiplicities in relativistic heavy-ion collisions. The joint distribution of $(N_{\text{part}}, N_{\text{coll}})$ obtained from the Monte-Carlo Glauber model is used as the prior, while negative binomial distributions calibrated to charged-particle multiplicity fluctuations define the likelihood. This approach replaces conventional hard-cut centrality classification with a probabilistic assignment based on $N_{\text{part}}$, making the multiplicity--geometry smearing explicit and reducing the impact of volume fluctuations on downstream observables. A closure test using an UrQMD-MCG hybrid model at $\sqrt{s_{NN}} = 19.6$~GeV shows that the method yields well-calibrated posterior distributions with negligible bias and improves the reconstruction of net-proton cumulants relative to conventional multiplicity-based centrality selection.

nucl-th↗

arXiv2026

Neural network maximum entropy framework for distribution reconstruction in heavy-ion collisions

We develop a neural-network maximum-entropy (NN+MaxEnt) framework for reconstructing probability distributions from limited observables in heavy-ion collisions. The method combines flexible neural-network representations with Shannon-entropy regularization, preserving positivity and normalization without assuming a fixed analytic form. After validation with Gaussian, Poisson, and mixed-Poisson closure tests, we apply the framework to two physics-motivated inverse problems: an effective multiplicity reconstruction constrained by functional renormalization group cumulants, used as a closure test, and the conditional jet-energy-loss distribution extracted from single-inclusive jet $R_{AA}$ data in Pb+Pb collisions at $\sqrt{s_{NN}}=2.76$~TeV. For the fRG closure test, NN+MaxEnt accurately reproduces the imposed cumulants and yields distributions consistent with conventional MaxEnt solutions. For jets, the reconstructed energy-loss distributions reproduce the measured $R_{AA}$; at an initial jet momentum $x=50~\mathrm{GeV}$, the conditional mean energy loss is $\langle\Delta p_T\rangle\simeq11.8~\mathrm{GeV}$, with a central $16\text{--}84\%$ interval of $9.0\text{--}15.0~\mathrm{GeV}$. The extracted energy-loss profile is qualitatively consistent with Bayesian MCMC and LBT results. NN+MaxEnt thus provides a flexible, less ansatz-dependent framework for regularized distribution reconstruction from observables connected to the underlying distribution through differentiable forward maps.

nucl-th↗

arXiv2026

Physics-Informed Neural Network with Squeeze-Excitation-like Attention

We introduce SEA-PINN, a novel architecture that incorporates a Squeeze-Excitation-like attention mechanism into physics-informed neural networks to dynamically recalibrate the importance of neurons across layers. A key feature of SEA-PINN is its highly stable initialization. On 17 out of 20 benchmark problems, SEA-PINN exhibit nearly negligible variance and significantly reduced initial loss, establishing a quasi-deterministic and favorable starting point for optimization. Notably, without employing Fourier feature embeddings or periodic activation functions, SEA-PINN attained competitive accuracy (83\% vs. 90\% improvement relative to FNN-PINN on the high-frequency case 7) as compared with TSA-PINN-a model specifically engineered for high-frequency problems via learnable frequencies in sinusoidal activations. Furthermore, integrating SEA-PINN into TSA-PINN boosted performance by 42.49\%. These results underscore SEA-PINN as a lightweight plug-in module that enhances nonlinear representation power, promotes more robust and efficient convergence, and strengthens the overall reliability of physics-informed learning.

cs.LG↗

arXiv2026

Four-dimensional QCD equation of state from a quasi-parton model with physics-informed neural networks

The equation of state (EoS) of strongly interacting matter at finite temperature and chemical potentials (baryon, charge, and strangeness) is a crucial input for hydrodynamic simulations of relativistic heavy-ion collisions. We construct a four-dimensional EoS using a deep-learning-assisted quasi-particle model (DLQPM) within a physics-informed neural network (PINN) framework, in which the masses of light quarks, strange quarks, and gluons are parameterized as functions of temperature and chemical potentials ($T, \mu_B, \mu_Q, \mu_S$). The model is constrained by lattice QCD data at vanishing chemical potentials and provides a thermodynamically consistent extrapolation to finite $\mu_{B,Q,S}$. The DLQPM accurately reproduces the lattice-calculated cumulants $\chi^{B,Q,S}_{i,j,k}$ at $\mu_{B,Q,S}=0$, and its predicted EoS at various chemical potentials agrees well with results from the generalized $T'$-expansion method in lattice QCD. Furthermore, the calculated baryon-strangeness correlation $C_{BS}$ is consistent, within uncertainties, with preliminary STAR data. This work offers a reliable EoS for exploring the QCD phase structure in the beam energy scan region.

nucl-th↗

arXiv2026

Physics-Informed Global Extraction of the Universal Small-$x$ Dipole Amplitude

We extract the universal small-$x$ dipole scattering amplitude $N(r,x_B)$ from a global analysis based on a physics-informed neural network (PINN), without imposing a priori MV-type parametrization of the initial condition. The network provides a smooth and differentiable surrogate for $N(r,x_B)$, whose rapidity dependence is constrained by the collinearly improved Balitsky--Kovchegov evolution equation, while its functional form is simultaneously constrained by Deep Inelastic Scattering (DIS) data for the reduced total and charm cross sections, exclusive $J/\psi$ photoproduction measurements, and a positivity requirement for the momentum-space dipole amplitude. The resulting single universal amplitude consistently describes all fitted observables within a unified framework, alleviating the long-standing tension between total and charm channels encountered in conventional small-$x$ fits based on rigid parametric ans\"atze. Within the fitted kinematic domain, the best extracted PINN solution yields a smooth, non-negative momentum-space dipole over the full transverse-momentum range examined. Our results provide a robust and well-behaved input for Color Glass Condensate phenomenology across a broad class of high-energy processes.

hep-ph↗

arXiv2026

Parton Fragmentation Functions Extracted with a Physics-Informed Neural Network

Reliable predictions of many high-energy strong interaction processes rely heavily on the non-perturbative parton fragmentation functions (FFs) extracted from existing experimental data. Conventional methods often require parameterized forms of FFs and additional scale evolution according to the Dokshitzer-Gribov-Lipatov-Altarelli-Parisi (DGLAP) evolution equations. We introduce a novel approach to determining parton FFs using a Physics-Informed Neural Network (PINN). Unlike traditional methods, our approach does not require prior parameterized forms and directly integrates the DGLAP evolution equations into the neural network architecture, allowing the FFs to automatically satisfy these equations. We present new sets of parton FFs extracted from hadron spectra in electron-positron annihilation processes at next-to-leading order (NLO) in pQCD using this new technique. To validate our approach, we calculate charged hadron spectra in proton-(anti)proton collisions using the extracted FFs and demonstrate that the results align well with experimental data across a large range of colliding energies ($\sqrt{s}$ = 130, 200, 500, 630, 900, 1800, 2760, 5020, 5440, 7000 GeV). Our findings indicate that the PINN method not only simplifies the extraction process but also enhances the universal applicability of FFs across different energy scales. By eliminating the need for parameterized forms and additional DGLAP evolution, our approach represents a significant step forward toward fast and accurate extractions of non-perturbative quantities such as parton fragmentations functions and parton distribution functions.

hep-ph↗

arXiv2025

Melting of heavy quarkonia in QGP using deep neural networks

Machine learning techniques have emerged as powerful tools for tackling non-perturbative challenges in quantum chromodynamics. In this study, we introduce a data-driven framework employing deep neural networks to systematically predict the temperature-dependent behavior of the screening mass $m_D(T)$ and the strong coupling constant $\alpha_s(T)$ within a quark-gluon plasma medium. These medium-sensitive quantities are subsequently employed to compute the thermal widths $\Gamma_{\text{n}}(T)$ and binding energies $E_B(T)$ of heavy quarkonia states, specifically charmonia and bottomonia, by numerically solving the Schr\"odinger equation with medium-modified heavy quark potentials. To estimate the dissociation temperatures $T_d$ of various quarkonia states, we employ two complementary dissociation criteria: the conventional one, where $2E_B(T_d) = \Gamma_{\text{n}}(T_d)$, and an additional lower bound criterion defined by $E_B(T_d) = 3T_d$. This dual-criterion approach provides a more constrained and physically motivated estimate of the temperature range over which quarkonia states dissolve in the QGP environment. Our machine learning-enhanced predictions show excellent agreement with available lattice QCD results, especially for the ground states $\Upsilon(1S)$ and $J/\psi$, and offer new perspectives on the sequential suppression pattern detected in relativistic heavy-ion collision experiments. Overall, this work advances the quantitative description of quarkonium suppression and demonstrates the prospect of modern machine learning methods to bridge theoretical predictions and experimental observations, thereby contributing significantly to QGP tomography.

hep-ph↗

arXiv2025

Nuclear equation of state at finite $\mu_B$ using deep learning assisted quasi-parton model

To accurately determine the nuclear equation of state (EoS) at finite baryon chemical potential ($\mu_B$) remains a challenging yet essential goal in the study of QCD matter under extreme conditions. In this study, we develop a deep learning assisted quasi-parton model, which utilizes three deep neural networks, to reconstruct the QCD EoS at zero $\mu_B$ and predict the EoS and transport coefficient $\eta/s$ at finite $\mu_B$. The EoS derived from our quasi-parton model shows excellent agreement with lattice QCD results obtained using Taylor expansion techniques. The minimum value of $\eta/s$ is found to be approximately 175 MeV and decreases with increasing chemical potential within the confidence interval. This model not only provides a robust framework for understanding the properties of the QCD EoS at finite $\mu_B$ but also offers critical input for relativistic hydrodynamic simulations of nuclear matter produced in heavy-ion collisions by the RHIC beam energy scan program.

nucl-th↗

arXiv2024

Is AI Robust Enough for Scientific Research?

We uncover a phenomenon largely overlooked by the scientific community utilizing AI: neural networks exhibit high susceptibility to minute perturbations, resulting in significant deviations in their outputs. Through an analysis of five diverse application areas -- weather forecasting, chemical energy and force calculations, fluid dynamics, quantum chromodynamics, and wireless communication -- we demonstrate that this vulnerability is a broad and general characteristic of AI systems. This revelation exposes a hidden risk in relying on neural networks for essential scientific computations, calling further studies on their reliability and security.

cs.LG↗

arXiv2024

Symmetry Breaking in Neural Network Optimization: Insights from Input Dimension Expansion

Understanding the mechanisms behind neural network optimization is crucial for improving network design and performance. While various optimization techniques have been developed, a comprehensive understanding of the underlying principles that govern these techniques remains elusive. Specifically, the role of symmetry breaking, a fundamental concept in physics, has not been fully explored in neural network optimization. This gap in knowledge limits our ability to design networks that are both efficient and effective. Here, we propose the symmetry breaking hypothesis to elucidate the significance of symmetry breaking in enhancing neural network optimization. We demonstrate that a simple input expansion can significantly improve network performance across various tasks, and we show that this improvement can be attributed to the underlying symmetry breaking mechanism. We further develop a metric to quantify the degree of symmetry breaking in neural networks, providing a practical approach to evaluate and guide network design. Our findings confirm that symmetry breaking is a fundamental principle that underpins various optimization techniques, including dropout, batch normalization, and equivariance. By quantifying the degree of symmetry breaking, our work offers a practical technique for performance enhancement and a metric to guide network design without the need for complete datasets and extensive training processes.

cs.LG↗

arXiv2024

Neural Network Modeling of Heavy-Quark Potential from Holography

Using Multi-Layer Perceptrons (MLP) and Kolmogorov-Arnold Networks (KAN), we construct a holographic model based on lattice QCD data for the heavy-quark potential in the 2+1 system. The deformation factor $w(r)$ in the metric is obtained using the two types of neural network. First, we numerically obtain $w(r)$ using MLP, accurately reproducing the QCD results of the lattice, and calculate the heavy quark potential at finite temperature and the chemical potential. Subsequently, we employ KAN within the Andreev-Zakharov model for validation purpose, which can analytically reconstruct $w(r)$, matching the Andreev-Zakharov model exactly and confirming the validity of MLP. Finally, we construct an analytical holographic model using KAN and study the heavy-quark potential at finite temperature and chemical potential using the KAN-based holographic model. This work demonstrates the potential of KAN to derive analytical expressions for high-energy physics applications.

hep-ph↗

arXiv2022

Deep-learning quasi-particle masses from QCD equation of state

The interactions of quarks and gluons are strong at non-perturbative region. The equation of state (EoS) of a strongly-interacting quantum chromodynamics (QCD) medium can only be studied using the first-principle lattice QCD calculations. However, the complicated QCD EoS can be reproduced using simple statistical formula by treating the medium as a free parton gas whose fundamental degree of freedoms are dressed quarks and gluons called quasi-particles, with temperature-dependent masses. We use deep neural network and auto differentiation to solve this variational problem in which the masses of quasi gluons, up/down and strange quarks are three unknown functions, whose forms are represented by deep neural network. We reproduce the QCD EoS using these machine learned quasi-particle masses, and calculate the shear viscosity over entropy density ($\eta/s$) as a function of temperature of the hot QCD matter.

hep-ph↗