Searcharxiv⌕ Search

arXiv subjects

Zhihong Xia

Publications and source records attributed to Zhihong Xia.

At least 19 recordsLinked to original sources
arXiv2026

LRX-PINN: A Layer-Resolving XNet Physics-Informed Neural Network with Integrated Cauchy Activations for Convection-Dominated Problems

Convection-dominated convection-diffusion problems often develop thin layers, where the solution has sharp transition profiles and its derivatives are highly localized. This creates a structural mismatch for standard physics-informed neural networks (PINNs), whose trial spaces are not designed to match the value--derivative structure of such layers. We propose a Layer-Resolving XNet Physics-Informed Neural Network (LRX-PINN) based on integrated Cauchy activations. The proposed basis is transition-type at the solution level, while its derivative recovers a localized Cauchy kernel. We show that this structure matches the scaling of convection-dominated layers, inherits the Cauchy approximation mechanism at the derivative-profile level, and identifies \(d/\|w\|\) as the effective physical width of a ridge neuron. For analytic layer profiles, this yields derivative-stable exponential approximation in the stretched coordinate and a layer-scaled estimate for the strong residual of the singularly perturbed operator. Numerical experiments on several convection-dominated benchmarks show that LRX-PINN achieves higher accuracy than PIKAN and Fourier-feature PINNs while using less than \(30\%\) of their trainable parameters. On more challenging benchmarks, embedding the proposed representation into hp-VPINN-based frameworks further improves the best results obtained by existing hp-VPINN-based baselines without changing their original loss functionals or stabilization strategies. These results show that neural representations aligned with layer structure provide a compact and effective approach for convection-dominated problems.

math.AP↗
arXiv2026

XNet-Enhanced Deep BSDE Method and Numerical Analysis

Semilinear parabolic partial differential equations (PDEs) are fundamental to modeling complex dynamical systems across scientific domains. The Deep Backward Stochastic Differential Equation (BSDE) method is a promising approach for high-dimensional PDEs; however, existing convergence results apply only to globally Lipschitz generators, excluding important cases such as Allen--Cahn and Hamilton--Jacobi--Bellman (HJB) equations. This paper presents both a theoretical and a computational advance for Deep BSDE methods. Theoretically, we establish the convergence theory for non--Lipschitz generators--covering Allen--Cahn equations with cubic nonlinearity and HJB equations with quadratic gradient growth--based on a bounded double--well lemma and a truncated-BSDE analysis within the Bouchard--Touzi--Zhang theory. Computationally, we instantiate the framework with XNet, a shallow architecture with $\mathcal O(L)$ parameters that preserves strong approximation while substantially reducing optimization and computational cost. Numerical experiments on 100--dimensional PDEs corroborate the predicted convergence behavior and demonstrate significant efficiency gains over standard feedforward implementations.

cs.CE↗
arXiv2026

On the Degeneracy of the Central Configuration Formed by a Regular n-Gon with a Central Mass

We investigate the degeneracy of the central configuration formed by a regular $n$-gon of equal masses together with an additional mass at the center. While degeneracy of such configurations has traditionally been studied through direct spectral computations, a systematic structural understanding of the origin and multiplicity of degeneracy values has remained incomplete. Exploiting the dihedral symmetry $D_n$, we develop a representation-theoretic framework that decomposes the Hessian of $\sqrt{IU}$ into invariant blocks associated with irreducible symmetry modes, reducing the degeneracy problem to a finite collection of low-dimensional determinants. In particular, this decomposition reveals a distinguished $3 \times 3$ block arising from the coupling between the central mass and the first Fourier mode. Within this framework, degeneracy is organized mode by mode: for each admissible Fourier mode $l \geq 2$, there exists at most one critical value of the central mass parameter at which degeneracy occurs, while the mode $l = 1$ exhibits a qualitatively different behavior. As a consequence, all degeneracy values can be determined explicitly, and their number increases with $n$, reflecting the growing number of independent symmetry modes. Our results provide a structural explanation for the multiplicity of degeneracy values and show that degeneracy is not an isolated phenomenon, but a consequence of the underlying symmetry. The approach also suggests a general framework for analyzing degeneracy in symmetric central configurations.

math.DS↗
arXiv2025

CauchyNet: Compact and Data-Efficient Learning using Holomorphic Activation Functions

A novel neural network inspired by Cauchy's integral formula, is proposed for function approximation tasks that include time series forecasting, missing data imputation, etc. Hence, the novel neural network is named CauchyNet. By embedding real-valued data into the complex plane, CauchyNet efficiently captures complex temporal dependencies, surpassing traditional real-valued models in both predictive performance and computational efficiency. Grounded in Cauchy's integral formula and supported by the universal approximation theorem, CauchyNet offers strong theoretical guarantees for function approximation. The architecture incorporates complex-valued activation functions, enabling robust learning from incomplete data while maintaining a compact parameter footprint and reducing computational overhead. Through extensive experiments in diverse domains, including transportation, energy consumption, and epidemiological data, CauchyNet consistently outperforms state-of-the-art models in predictive accuracy, often achieving a 50% lower mean absolute error with fewer parameters. These findings highlight CauchyNet's potential as an effective and efficient tool for data-driven predictive modeling, particularly in resource-constrained and data-scarce environments.

cs.LG↗
arXiv2025

Complex Physics-Informed Neural Network

We propose compleX-PINN, a novel physics-informed neural network (PINN) architecture incorporating a learnable activation function inspired by the Cauchy integral theorem. By optimizing the activation parameters, compleX-PINN achieves high accuracy with just a single hidden layer. Empirically, we demonstrate that compleX-PINN solves high-dimensional problems that pose significant challenges for PINNs. Our results show that compleX-PINN consistently achieves substantially greater precision, often improving accuracy by an order of magnitude, on these complex tasks.

cs.LG↗
arXiv2025

Enhancing Neural Function Approximation: The XNet Outperforming KAN

XNet is a single-layer neural network architecture that leverages Cauchy integral-based activation functions for high-order function approximation. Through theoretical analysis, we show that the Cauchy activation functions used in XNet can achieve arbitrary-order polynomial convergence, fundamentally outperforming traditional MLPs and Kolmogorov-Arnold Networks (KANs) that rely on increased depth or B-spline activations. Our extensive experiments on function approximation, PDE solving, and reinforcement learning demonstrate XNet's superior performance - reducing approximation error by up to 50000 times and accelerating training by up to 10 times compared to existing approaches. These results establish XNet as a highly efficient architecture for both scientific computing and AI applications.

cs.LG↗
arXiv2025

Cauchy activation function and XNet

We have developed a novel activation function, named the Cauchy Activation Function. This function is derived from the Cauchy Integral Theorem in complex analysis and is specifically tailored for problems requiring high precision. This innovation has led to the creation of a new class of neural networks, which we call (Comple)XNet, or simply XNet. We will demonstrate that XNet is particularly effective for high-dimensional challenges such as image classification and solving Partial Differential Equations (PDEs). Our evaluations show that XNet significantly outperforms established benchmarks like MNIST and CIFAR-10 in computer vision, and offers substantial advantages over Physics-Informed Neural Networks (PINNs) in both low-dimensional and high-dimensional PDE scenarios.

cs.LG↗
arXiv2024

A Simple Example of Pathological Foliations in Skew-Product Diffeomorphisms

Inspired by examples of Katok and Milnor \cite{Milnor1997}, we construct simple examples of skew-product volume preserving diffeomorphism where the center foliation is pathological in the sense that, there is a full measure set whose intersection with any center leaf contains at most one point. Comparing with other examples in literature, our mechanism for producing such pathological foliation is simple and elementary.

math.DS↗
arXiv2024

Central Configurations with Dihedral Symmetry

As an application of the representation theory for the dihedral groups, we study the symmetric central configurations in the n-body problem where $n$ equal masses are placed at the vertices of a regular $n$-gon. Since the Hessian matrices at these configurations are typically very large, particularly when $n$ is large, computations of their eigenvalues present a challenging problem. However, by decomposing the action of the dihedral groups into irreducible representations, we show that the Hessians can be simplified to a block-diagonal matrix with small blocks, of the sizes at most 2*2. This is due to the fact that the action of a dihedral group can be represented as a block-diagonal matrix with small irreducible blocks. In the end, the eigenvalues can be explicitly obtained by simply computing eigenvalues of these small block matrices.

math.DS↗
arXiv2024

Homotopical Minimal Measures for Geodesic flows on Surfaces of Higher Genus

We study the homotopical minimal measures for positive definite autonomous Lagrangian systems. Homotopical minimal measures are action-minimizers in their homotopy classes, while the classical minimal measures (Mather measures) are action-minimizers in homology classes. Homotopical minimal measures are much more general, they are not necessarily homological action-minimizers. However, some of them can be obtained from the classical ones by lifting them to finite-fold covering spaces. We apply this idea of finite covering to the geodesic flows on surfaces of higher genus. Let $(M,G)$ be a compact closed surface with genus $g>1$, where $G$ is a complete Riemannian metric on $M$. Consider the positive definite autonomous Lagrangian $L(x,v)=G_x(v,v)$, whose Lagrangian system $ϕ_t: TM\rightarrow TM$ is exactly the complete geodesic flow on $TM$. We show that for each homotopical minimal ergodic measure $μ$ that is supported on a nontrivial simple closed periodic trajectory, there is a finite-fold covering space $M'$ such that each ergodic preimage of $μ$ on $TM'$ is a minimal measure in the classic Mather theory for the Lagrangian system on $TM'$.

math.DS↗
arXiv2021

Symmetry in n-body problem via group representations

We introduce an algebraic method to study local stability in the Newtonian $n$-body problem when certain symmetries are present. We use representation theory of groups to simplify the calculations of certain eigenvalue problems. The method should be applicable in many cases, we give two main examples here: the square central configurations with four equal masses, and the equilateral triangular configurations with three equal masses plus an additional mass of arbitrary size at the center. We explicitly found the eigenvalues of certain 8x8 Hessians in these examples, with only some simple calculations of traces. We also studied the local stability properties of corresponding relative equilibria in the four-body problems.

math.DS↗
arXiv2021

A $C^{\infty}$ closing lemma on torus

Asaoka & Irie recently proved a $C^{\infty}$ closing lemma of Hamiltonian diffeomorphisms of closed surfaces. We reformulated their techniques into a more general perturbation lemma for area-preserving diffeomorphism and proved a $C^{\infty}$ closing lemma for area-preserving diffeomorphisms on a torus that is isotopic to identity. i.e., we show that the set of periodic orbits is dense for a generic diffeomorphism isotopic to identity area-preserving diffeomorphism on torus. The main tool is the flux vector of area-preserving diffeomorphisms which is, different from Hamiltonian cases, non-zero in general.

math.DS↗
arXiv2021

Action and periodic orbits on annulus

We consider the classical problem of area-preserving maps on annulus $\mathbb{A} = S^1 \times [0, 1]$ . Let $\mathcal{M}_f$ be the set of all invariant probability measures of an area-preserving, orientation preserving diffeomorphism $f$ on $\mathbb{A}$. Given any $μ_1$ and $μ_2$ in $\mathcal{M}_f$, Franks \cite{Franks1988}\cite{Franks1992}, generalizing Poincaré's last geometric theorem (Birkhoff \cite{Birkhoff1913}), showed that if their rotation numbers are different, then $f$ has infinitely many periodic orbits. In this paper, we show that if $μ_1$ and $μ_2$ have different actions, even if they have the same rotation number, then $f$ has infinitely many periodic orbits. In particular, if the action difference is larger than one, then $f$ has at least two fixed points. The same result is also true for area-preserving diffeomorphisms on unit disk, where no rotation number is available.

math.DS↗
arXiv2021

A fixed point theorem for twist maps

Poincare's last geometric theorem (Poincare-Birkhoff Theorem) states that any area-preserving twist map of annulus has at least two fixed points. We replace the area-preserving condition with a weaker intersection property, which states that any essential simple closed curve intersects its image under $f$ at least at one point. The conclusion is that any such map has at least one fixed point. Besides providing a new proof to Poincare's geometric theorem, our result also has some applications to reversible systems.

math.DS↗
arXiv2013

Homoclinic points for convex billiards

In this paper we investigate some generic properties of a billiard system on a convex table. We show that generically, every hyperbolic periodic point admits some homoclinic orbit.

math.DS↗
arXiv2006

Geometric Expansions, Lyapunov Exponents and Foliations

We consider hyperbolic and partially hyperbolic diffeomorphisms on compact manifolds. Associated with invariant foliation of these systems, we define some topological invariants and show certain relationships between these topological invariants and the geometric and Lyapunov growths of these foliations. As an application, we show examples of systems with persistent non- absolute continuous center and weak unstable foliations. This generalizes the remarkable results of Shub and Wilkinson to cases where the center manifolds are not compact.

math.DS↗
arXiv2006

Topological Entropy and Partially Hyperbolic Diffeomorphisms

We consider partially hyperbolic diffeomorphisms on compact manifolds where the unstable and stable foliations stably carry some unique non-trivial homologies. We prove the following two results: if the center foliation is one dimensional, then the topological entropy is locally a constant; and if the center foliation is two dimensional, then the topological entropy is continuous on the set of all $C^\8$ diffeomorphisms. The proof uses a topological invariant we introduced; Yomdin's theorem on upper semi-continuity; Katok's theorem on lower semi-continuity for two dimensional systems and a refined Pesin-Ruelle inequality we proved for partially hyperbolic diffeomorphisms.

math.DS↗