arXiv subjects

Yonglin Li

Publications and source records attributed to Yonglin Li.

11 recordsLinked to original sources

arXiv2026

High-order finite element method for perfect conductivity and linear elasticity with nearly touching inclusions

In perfect conductivity and linear elasticity problems, the electric field and stress always become highly concentrated within narrow regions between adjacent perfect (rigid) inclusions, and blow up as the distance between inclusions approaches zero. The design of high-order numerical methods with rigorous error analysis for such concentration problems remains open. In this paper, we present the first high-order finite element method for solving these problems. Our approach is based on asymptotic estimates of high-order derivatives of solutions, employing a graded mesh and auxiliary basis functions specifically designed from these derivative estimates. We prove that the proposed method converges with an $H^1$-error bound of $O(h^p)$ and the error bound is independent of the distance (possibly approaching zero) between inclusions, where $h$ is the mesh size and $p$ is the degree of finite elements. Numerical examples in both two and three dimensions are presented to demonstrate the convergence rates of the numerical solutions. In particular, the blow-up behaviors of the gradients of solutions are demonstrated when the inclusions approach each other.

math.NA↗

arXiv2026

JuZhou 1.0 Technical Report: The First Edge-Native Text-to-Image Foundation Model Trained Entirely on China-Developed AI Accelerators

Text-to-image (T2I) diffusion models typically require substantial computational resources and cloud infrastructure, posing significant challenges for edge deployment in terms of latency, cost, and user privacy. We present JuZhou 1.0, an ultra-lightweight T2I foundation model designed for fully offline, on-device execution. JuZhou 1.0 achieves its efficiency through four key designs: (1) a compact image-generation backbone consisting of a 0.385B-parameter denoising U-Net and a 1.90M-parameter distilled decoder, totaling approximately 0.387B parameters; (2) Rectified Flow training combined with DMD2 distillation, reducing inference to 4 sampling steps; (3) Chinese semantic alignment trained on 9M curated image-text pairs, enabling direct Chinese prompting without external translation at inference time; and (4) a training and distillation pipeline completed on domestically developed Sugon K100 AI accelerators without relying on NVIDIA GPUs for training or distillation. Despite its compact scale, the 28-step base model of JuZhou 1.0 achieves an overall GenEval score of 0.69, outperforming published baselines including SDXL (2.6B, 0.55), SD3-Medium (2B, 0.62), and IF-XL (4.3B, 0.61). We further validate the full poetry-to-image pipeline on Android and the core CLIP-U-Net-VAE generation branch on iOS. On a smartphone powered by the Snapdragon 8 Elite Gen 5 Mobile Platform, the 4-step U-Net denoising branch runs in approximately 1.6 seconds, while the full Android poetry-to-image pipeline takes 4.5 seconds with on-device prompt refinement on Xiaomi 17 Pro Max. These results position JuZhou 1.0 as a practical approach to mobile text-to-image generation and provide a concrete reference for Chinese-native generation, domestic-compute training, and fully offline on-device deployment after one-time installation.

cs.CV↗

arXiv2025

A discrete Perfectly Matched Layer for peridynamic scalar waves in two-dimensional viscous media

In this paper, we propose a discrete perfectly matched layer (PML) for the peridynamic scalar wave-type problems in viscous media. Constructing PMLs for nonlocal models is often challenging, mainly due to the fact that nonlocal operators are usually associated with various kernels. We first convert the continua model to a spatial semi-discretized version by adopting quadrature-based finite difference scheme, and then derive the PML equations from the semi-discretized equations using discrete analytic continuation. The harmonic exponential fundamental solutions (plane wave modes) of the semi-discretized equations are absorbed by the PML layer without reflection and are exponentially damped. The excellent efficiency and stability of discrete PML are demonstrated in numerical tests by comparison with exact absorbing boundary conditions.

math.NA↗

arXiv2025

Supernova constraints on lepton flavor violating ALPs

Supernovae offer a unique hot and dense environment to probe new physics beyond the Standard Model. We investigate supernova cooling constraints on lepton-flavor-violating (LFV) axions and axion-like particles (ALPs) that couple to electrons and muons. For LFV-ALP production in supernovae, muon decay and lepton bremsstrahlung have been considered previously. In this work, we identify the electron-muon coalescence channel as an efficient new production mechanism in the high-mass regime. We also include the semi-Compton scattering process, which has recently been shown to provide sizable contributions for electron-coupled ALPs. We find that muon decay dominates in the low-mass regime, electron-muon coalescence becomes the leading channel at high masses, and semi-Compton scattering provides the dominant contribution in the intermediate mass range. We find that the electron-muon coalescence process yields the strongest constraints in the mass range of $\sim (115,280)$ MeV, probing the ALP-electron-muon coupling down to $\sim 4\times 10^{-10}$ for an ALP mass of $\sim200$ MeV.

hep-ph↗

arXiv2024

Uncover the Dynamic Community Structure of Instant Delivery Network

The rise of instant delivery services has reshaped urban spatial structures through the interaction between suppliers and consumers. However, limited research has explored the spatiotemporal dynamics of delivery network structures. This study constructs a time-dependent, multi-layer instant delivery network in the case city of Beijing using a large-scale dataset from Eleme, organized into 500m grid units. A dynamic community detection method identifies evolving community structures over time. The results reveal 309 dynamic communities, with an average size of 13.78 square kilometers. Communities form in the morning, expand, stabilize, then contract, and disappear by night. Key factors influencing stability include building area and residential population, while online retail and service facilities contribute to instability. These findings offer insights into the spatial structure of instant delivery networks and the factors driving their dynamics, with practical implications for optimizing platform strategies, resource allocation, and urban transportation planning.

physics.soc-ph↗

arXiv2024

XQC and CSR constraints on strongly interacting dark matter with spin and velocity dependent cross sections

Dark matter that interacts strongly with baryons can avoid the stringent dark matter direct detection constraints, because, like baryons, they are likely to be absorbed when traversing the rocks, leading to a suppressed flux in deep underground labs. Such strongly interacting dark matter, however, can be probed by dark matter experiments or other experiments operated on the ground level or in the atmosphere. In this paper we carry out systematic analysis of two of these experiments, XQC and CSR, to compute the experimental constraints on the strongly interacting dark matter in the following three scenarios: (1) spin-independent and spin-dependent interactions; (2) different velocity dependent cross sections; (3) different dark matter mass fractions. Some of the scenarios are first analyzed in the literature. We find that the XQC exclusion region has some non-trivial dependencies on the various parameters and the limits in the spin-dependent case is quite different from the spin-independent case. A peculiar region in the parameter space, where the XQC constraint disappears, is also found in our Monte Carlo simulations. This occurs in the case where the interaction cross section is proportional to the square of the velocity. We further compare our XQC and CSR limits to other experimental constraints, and find that a large parameter space is allowed by various experiments if the dark matter mass fraction is sufficiently small, $f_χ\lesssim 10^{-4}$.

hep-ph↗

arXiv2023

Higher-order FEM and CIP-FEM for Helmholtz equation with high wave number and perfectly matched layer truncation

The high-frequency Helmholtz equation on the entire space is truncated into a bounded domain using the perfectly matched layer (PML) technique and subsequently, discretized by the higher-order finite element method (FEM) and the continuous interior penalty finite element method (CIP-FEM). By formulating an elliptic problem involving a linear combination of a finite number of eigenfunctions related to the PML differential operator, a wave-number-explicit decomposition lemma is proved for the PML problem, which implies that the PML solution can be decomposed into a non-oscillating elliptic part and an oscillating but analytic part. The preasymptotic error estimates in the energy norm for both the $p$-th order CIP-FEM and FEM are proved to be $C_1(kh)^p + C_2k(kh)^{2p} +C_3 E^{\rm PML}$ under the mesh condition that $k^{2p+1}h^{2p}$ is sufficiently small, where $k$ is the wave number, $h$ is the mesh size, and $E^{\rm PML}$ is the PML truncation error which is exponentially small. In particular, the dependences of coefficients $C_j~(j=1,2)$ on the source $f$ are improved. Numerical experiments are presented to validate the theoretical findings, illustrating that the higher-order CIP-FEM can greatly reduce the pollution errors.

math.NA↗

arXiv2023

RefSAM: Efficiently Adapting Segmenting Anything Model for Referring Video Object Segmentation

The Segment Anything Model (SAM) has gained significant attention for its impressive performance in image segmentation. However, it lacks proficiency in referring video object segmentation (RVOS) due to the need for precise user-interactive prompts and a limited understanding of different modalities, such as language and vision. This paper presents the RefSAM model, which explores the potential of SAM for RVOS by incorporating multi-view information from diverse modalities and successive frames at different timestamps in an online manner. Our proposed approach adapts the original SAM model to enhance cross-modality learning by employing a lightweight Cross-Modal MLP that projects the text embedding of the referring expression into sparse and dense embeddings, serving as user-interactive prompts. Additionally, we have introduced the hierarchical dense attention module to fuse hierarchical visual semantic information with sparse embeddings to obtain fine-grained dense embeddings, and an implicit tracking module to generate a tracking token and provide historical information for the mask decoder. Furthermore, we employ a parameter-efficient tuning strategy to align and fuse the language and vision features effectively. Through comprehensive ablation studies, we demonstrate our model's practical and effective design choices. Extensive experiments conducted on Refer-Youtube-VOS, Ref-DAVIS17, and three referring image segmentation datasets validate the superiority and effectiveness of our RefSAM model over existing methods.

cs.CV↗

arXiv2022

1st Workshop on Maritime Computer Vision (MaCVi) 2023: Challenge Results

The 1$^{\text{st}}$ Workshop on Maritime Computer Vision (MaCVi) 2023 focused on maritime computer vision for Unmanned Aerial Vehicles (UAV) and Unmanned Surface Vehicle (USV), and organized several subchallenges in this domain: (i) UAV-based Maritime Object Detection, (ii) UAV-based Maritime Object Tracking, (iii) USV-based Maritime Obstacle Segmentation and (iv) USV-based Maritime Obstacle Detection. The subchallenges were based on the SeaDronesSee and MODS benchmarks. This report summarizes the main findings of the individual subchallenges and introduces a new benchmark, called SeaDronesSee Object Detection v2, which extends the previous benchmark by including more classes and footage. We provide statistical and qualitative analyses, and assess trends in the best-performing methodologies of over 130 submissions. The methods are summarized in the appendix. The datasets, evaluation code and the leaderboard are publicly available at https://seadronessee.cs.uni-tuebingen.de/macvi.

cs.CV↗

arXiv2022

Finite Element Method for a Nonlinear PML Helmholtz Equation with High Wave Number

A nonlinear Helmholtz equation (NLH) with high wave number and Sommerfeld radiation condition is approximated by the perfectly matched layer (PML) technique and then discretized by the linear finite element method (FEM). Wave-number-explicit stability and regularity estimates and the exponential convergence are proved for the nonlinear truncated PML problem. Preasymptotic error estimates are obtained for the FEM, where the logarithmic factors in h required by the previous results for the NLH with impedance boundary condition are removed in the case of two dimensions. Moreover, local quadratic convergences of the Newton's methods are derived for both the NLH with PML and its FEM. Numerical examples are presented to verify the accuracy of the FEM, which demonstrate that the pollution errors may be greatly reduced by applying the interior penalty technique with proper penalty parameters to the FEM. The nonlinear phenomenon of optical bistability can be successfully simulated.

math.NA↗

arXiv2018

FEM and CIP-FEM for Helmholtz Equation with High Wave Number and PML truncation

The Helmholtz scattering problem with high wave number is truncated by the perfectly matched layer (PML) technique and then discretized by the linear continuous interior penalty finite element method (CIP-FEM). It is proved that the truncated PML problem satisfies the inf--sup condition with inf--sup constant of order $O(k^{-1})$. Stability and convergence of the truncated PML problem are discussed. In particular, the convergence rate is twice of the previous result. The preasymptotic error estimates in the energy norm of the linear CIP-FEM as well as FEM are proved to be $C_1kh+C_2k^3h^2$ under the mesh condition that $k^3h^2$ is sufficiently small. Numerical tests are provided to illustrate the preasymptotic error estimates and show that the penalty parameter in the CIP-FEM may be tuned to reduce greatly the pollution error.

math.NA↗