HKBU Leevan Ling homepage Department of Mathematics

Research Interests:

My research lies in numerical analysis and kernel-based approximation. My recent focus is on developing inverse inequalities for kernel trial spaces and applying them to the stability and convergence analysis of numerical methods for partial differential equations. Related interests include least-squares and collocation methods, adaptive approximation, and local smoothness estimation. My earlier work focused extensively on meshfree methods for PDEs on surfaces and manifolds.

Current Research Team:

Research Staff

Former Research Team Members:

Research Postgraduate Students

Former Research Postgraduate Students


General Research Fund (GRF):
Abstracts & Details

  1. GRF HKBU 12301526 Meshfree Regularity Detection and Its Applications

    Funding amount: $1,051,200

    Period: 01/2027-12/2029

    Abstract: In many fields of science, engineering, and data-driven modeling, we face the challenge of reconstructing or predicting a function from scattered measurements. These functions can be smooth in some regions but may also have sharp changes, abrupt transitions, or irregular features in others. Our goal goes beyond standard edge detection and seeks to determine, solely from data, whether the underlying function is continuously differentiable up to order $k\geq0$ (that is, $C^k$ regular) at a given location, but fails to be differentiable at order $k+1$.

    In the first phase of the project, we design robust algorithms for such pointwise regularity detection in the noise-free setting. While many established approaches, such as those related to WENO, require structured grid data, recent samplet methods extend to scattered data by constructing polynomial-based bases with vanishing moments. This enables regularity detection via basis coefficients. In this project, we focus on developing additional alternatives within the meshfree kernel-based framework. Our approach adaptively combines two strategies: smoothness ladder techniques, which assess smoothness by varying the kernel order on a fixed dataset, and nested fill approaches, which analyze a sequence of increasingly dense data subsets. The smoothness ladder is suitable for general scattered data, while the nested fill method is most effective when denser nested samples are available. On the practical side, we use the detected regularity information to guide meshfree interpolation and surrogate modeling. We allow kernel parameters such as smoothness and anisotropy to vary across the domain according to detected regularity. This spatial adaptivity is rarely used in the kernel approximation community, although it is familiar in Gaussian process literature.

    In the next phase, we further extend our methods to handle moderate observational noise in real-world data. Theoretical connections between deterministic regularized kernel interpolation and stochastic Gaussian process regression allow us to incorporate kernel-based regularization into regularity detection. This approach allows us to assign a statistical confidence to the detection of $C^k$ but not $C^{k+1}$ regularity at each location based on the observed data. Finally, we address streaming or sequential data scenarios by developing memory-efficient algorithms that update regularity estimates in real time and preserve a compact summary of regions with low regularity for use by downstream models.

  2. GRF HKBU 12301824 Advancements in meshfree methods for hyperbolic conservation laws on high-dimensional manifolds

    Funding amount: $910,742

    Period: 1/2025-12/2027

    Co-I: W. S. Don

    Abstract: Hyperbolic conservation laws on higher-dimensional forms are crucial in understanding various physical phenomena. These mathematical structures help us describe and predict the behavior of different systems, such as traffic flow, fluid dynamics, and wave propagation. Improving our knowledge and computational methods in this field could lead to advancements in engineering, meteorology, and physics, which is the core motivation for our project.

    We begin our research by generalizing an embedding method (explained below) used for solving scalar hyperbolic conservation laws on one-dimensional manifolds to handle more complex two-dimensional systems on a manifold of codimension two (which includes one-dimensional cases as a special case). One significant challenge we face is understanding how hyperbolic characteristics interact with the manifold's shape and the conservative variables' small- and large-scale structures while still maintaining positivity, conserving certain quantities, preserving physical bounds, and mitigating/reducing oscillations. Our project is eager to tackle this challenge and clarify this complex topic.

    Next, we plan to develop an alternative solver based on an intrinsic formulation. The goal is to reduce computational complexity by working in the parameter space. However, this necessitates working with patches of the surface. Our first objective is to adapt essentially non-oscillatory schemes to work with data points stored in patches. We will explore how the placement of high-gradient and fine-scale solutions across these patches impacts our solutions' accuracy and numerical stability. We will then identify an effective data structure for storing the solutions of hyperbolic conservation laws in the local parameter space.

    Finally, we plan to examine how surfaces with added roughness affect the solutions of scalar hyperbolic conservation laws. We will test the effectiveness of our developed embedding and intrinsic methods in capturing these changes. We are particularly interested in gaining insights into hyperbolic solutions' behaviors as the roughness amplitude increases. We will then simulate the more complex hyperbolic conservation laws, such as Euler, Navier-Stokes, and shallow water-wave equations, on rough surfaces. The results from these simulations will be used to test our models and improve their ability to simulate outcomes that align with real-world expectations and fundamental physics principles. Our research could provide a more grounded and reliable understanding of the physical phenomena under study.

  3. GRF HKBU 12300922 Kernel-based hybrid methods for convection-dominated equations on surfaces

    Funding amount: $741,520

    Period: 1/2023-12/2025

    Abstract: By coupling appropriate high-resolution techniques, we aim to develop meshless numerical methods for solving partial differential equations (PDEs) on smooth surfaces whose solutions may contain shocks and/or discontinuities. We focus on both viscous and inviscid scalar hyperbolic conservation laws and study second-order quasilinear diffusion-convection PDEs with zero or very slow diffusive transport, i.e., large P\'{e}clet number, as in many engineering applications. Both Meshless method of lines and various types of elliptic PDEs solvers (with finite difference discretization in time) are known to work well on PDEs with moderate P\'{e}clet number. Meshless semi-Lagrangian methods also show promising in solving pure advection PDEs on surfaces. This project begins with a literature review on existing meshless PDEs solvers. We are interested in numerical performance when solving PDEs on open surfaces with boundary conditions in order to identify good candidates to be used on smooth region. Note that all available error estimates and convergence analysis for (some kernel-based) surface PDE solvers only apply to closed surfaces. On the theoretical front, we focus on the meshless semi-Lagrangian methods in order to ensure that we use (domain type) integrator (on surface) correctly. The second part of this project will focus on the development time-dependent embedding methods. We conjecture that there exists some extensions to the velocity field in the advection and/or diffusion-convection PDEs so that the constant-along-normal solutions solves the corresponding embedded PDEs. After completing this analytical task, the embedded PDEs are readily solvable with ENO/WENO-type shock capturing algorithms in literature. As a proof of concept, the preliminary two dimensional results presented in this proposal show promising results in capturing shock and discontinuity in solution. The final task is to assemble a hybrid algorithm that run meshless and ENO/WENO solvers on different parts of the surface adaptively based on the regularities of the numerical solutions. The project will be concluded with numerical simulations of various benchmark PDEs on surfaces.

  4. GRF HKBU 12301021 An intrinsic formulation for radial basis function finite difference methods of partial differential equations on manifolds

    Funding amount: $598,015

    Period: 1/2022-12/2024

    Abstract: This project aims to develop a class of meshless algorithm for solving partial differential equations (PDEs) on surfaces. These equations are used in modeling physical, simulating flow in cell membrane biology, creating computer graphics, and helping engineers to process data from 3D scanners. Due to the curved nature of surfaces, it is one of the fronts for meshless algorithms, i.e., ones that do not require data connectivity to run, to excel. In the past few years, meshless methods for surface PDEs are under active development. There are different varieties, say local vs global methods, and they are all working on extrinsic coordinates. The first part of this project focuses on algorithms development; we will fill in the gap by developing some local methods, namely least-squares (LS) radial basis function finite difference (RBF-FD) methods, that works on intrinsic formulations. All algorithms developed in the project enjoy higher flexibility to work with different metric tensors instead of just the one defined by the surface. The second phase emphasizes on applications including anisotropic diffusion (for MRI processing), pattern formations (for computer graphics), evolving surface (for point cloud processing) problems on surfaces specified by point clouds.

  5. GRF HKBU 12301520 Kernel-based approximation methods for curvatures estimations and applications on raw point clouds

    Funding amount: $576,337

    Period: 1/2021-12/2023

    Abstract: The first part of this project focuses on numerical differentiation problems required in analysing raw point clouds. Input data are sets of three dimensional coordinates that model surfaces of objects and error in data is expected. Our goal is to apply the well-developed kernel-based approximation methods and theories to this particular setting in order to derive new algorithms for computing normal vectors and curvatures. For normal information, we will apply some $W_2^\mu(\Omega)$-convergent ($\mu\geq0$) kernel-based discrete least-squares methods to cases where localized subset of the point cloud can be modeled by some level-set or parameterized functions. We will also develop adaptive data-dependent local neighborhood search algorithms by focusing on error estimates solely at the point of interest. Using the connection between normal vector and Laplacian Beltrami, a refinement scheme will be developed for better accuracy. Next, our developed algorithms will be extended to approximate curvatures of the surfaces by using higher order numerical derivatives. Aiming to improve robustness, we will explore various kernel-based approaches to estimate curvatures using only first order numerical derivatives. This project will be concluded by applying the developed algorithms with two applications on geometry processing.

  6. GRF HKBU 12301419 On the Asymptotics of Least-squares Kernel Collocation Methods

    Funding amount: $332,261

    Period: 1/2020-12/2022

    Abstract: Kernel collocation methods are meshfree and enjoy advantages in that they do not require construction of mesh grids. Since its inception, meshless methods have been welcomed by many researchers and successfully applied to solve a vast amount of applications and simulations in irregular domains. The aim of this research project is to develop some \emph{continuous least-squares} kernel collocation methods and numerical techniques capable of yielding efficient solvers for elliptic partial differential equations with low regularity.

    In \emph{discrete} least-squares kernel collocation methods, a fixed weight depending on some denseness measures of data points is imposed on all boundary collocation conditions. We proved that this discrete approach is convergent for a class of kernels that reproduce Sobolev spaces [SIAM Numer. Anal.,~56(1):614-633.~2018]. However, using a weight that ignores local geometries of data points is not numerically sound when data points are highly nonuniform; but, this is exactly what we expect from any adaptive method for solving equations with non-smooth coefficients, forcing functions, and/or boundary conditions. We propose to look into the asymptotic of the discrete approach as refined collocation points and consider continuous least-squares formulations. The first task of this project is to numerically identify some collocation-pointwise weights that allow kernel collocation methods to solve problems with discontinuous coefficients.

    The next challenging task is to come up with a rigid convergence analysis for the continuous least-squares kernel collocation methods. We note that regularity estimates for elliptic problems do not naturally yield upper bounds in $L^2$ norms and leave a gap between the continuous least-squares formulations. We conjecture an inverse inequality for trial functions. Convergence proof will be built on top of this result.

    In the last part of this project, we will develop some boundary-type kernel collocation methods that will be an alternative to the method of fundamental solutions but without the restriction on the differential operators. We propose to extend the integration domain for the interior residual to some larger domains that allow either analytic or easy numerical integrations. The former is computationally efficient, but restricted by our knowledge of analytical solutions to some integrals. The latter is more general, but comes with an overhead of numerical quadrature. All developed algorithms will adopt a fast direct matrix factorization algorithm and be subjected to thorough numerical experiments for understanding their limitations.

  7. GRF HKBU 12303818 Meshfree Methods for Coupled Bulk-Surface Problems

    Funding amount: $456,452

    Period: 1/2019-06/2022

    Abstract: In his 1952 paper ``The Chemical Basis of Morphogenesis'', Alan Turing put forth a groundbreaking concept for developmental biology. He proposed that particular patterns seen in nature can be modelled as the steady state solution of some reaction-diffusion processes of two morphogens. This is nowadays commonly known as the Turing pattern, which can be found in real-life on the skin of an angelfish. Founded on what Turing originally outlined, reaction-diffusion models remain a subject undergoing intense study. In the recent past, mathematical models for cell biology were proposed to account for the chemical processes inside the cell (i.e., bulk domain) and on the cell membrane (i.e., surface). This project considers numerical methods for solving the resulting coupled bulk-surface systems of partial differential equations (PDEs).

    The first challenging task is to come up with a sensible choice of trial spaces for solving coupled elliptic bulk-surface systems and provide a rigid convergence analysis for it. Coupled bulk-surface kernel-based collocation methods built upon some appropriate trial spaces would benefit from a convergence rate that is determined by the employed kernel and the smoothness of the solution but not the solutions of the coupled systems. To define such a coupled trial space, we use the standard kernel-based trial space in the bulk domain. Besides our previously proposed embedded kernel-based meshfree collocation methods for elliptic PDEs on surfaces, we plan to develop and analyze an intrinsic version to compete against each other. The next objective is to provide easy-to-implement and efficient numerical methods for coupled bulk-surface reaction-diffusion models. First, we focus on a model system that couples diffusion in the bulk domain and reaction-diffusion on the surface. Individually, kernel-based collocation methods were proven to be highly successful on these problems. We then take what we learn from the model system to design algorithms in time-dependent settings. We measure success in terms of agreements between simulated results and the prediction made by the symmetry breaking instability theories in literatures. After identifying and verifying some suitable numerical formulations, we will start to address the issue of computational efficiency and adaptivity by studying bulk-surface reaction-diffusion systems, in which we have four unknown functions to solve.

  8. GRF HKBU 12301917 Meshfree methods for parabolic equations based on weighted least-squares

    Funding amount: $314,900

    Period: 1/2018-06/2021

    Abstract: This project considers kernel-based meshfree methods for solving parabolic partial differential equations. For time-{\bf in}dependent problems, these are widely used and well understood, but this project plans to extend them to time-dependent ``parabolic'' problems that arise in Science and Engineering whenever diffusion or dispersion effects come into play, e.g. in air or water pollution or heat flux. We plan to develop novel numerical methods and provide a rigid theoretical analysis for them. After defining a general framework for the standard (non-nodal) and nodal notions of kernel-based trial spaces that have proven to be very useful in time-independent settings, we first show that there exist comparison functions in trial spaces that are very good approximations to the true solution of the problem. Their convergence rates are determined by the employed kernel and the smoothness of the solution but not the parabolic partial differential equation. The next challenging task is to come up with numerically stable methods to calculate a trial function from the given data that converges as fast as the comparison function.

    Our methods will apply {\em collocation} and {\em overtesting}, because there are plenty of applications and analysis for time-independent elliptic partial differential equations in the literature hinting at the benefits of overtesting by using more collocation conditions than the number of basis functions. Overtesting not only reduces the problem size but is also known to be necessary for stability and convergence of widely used unsymmetric Kansa-type strong-form collocation methods. The overtesting technique has not yet been applied to parabolic partial differential equations.

    Via overtested strong collocation, we can obtain various semi-discretized solutions of parabolic partial differential equations, in which the time variable remains continuous. For additional stability and for reducing computational complexity, we shall focus on least-squares methods to handle the overtesting numerically. Using the heat equation as an example, we want to show that overtesting yields a very good approximation of the spectrum of the analytical non-discretized evolution operator. In particular, preliminary numerical tests showed that the numerical performance of a weighted least-squares technique for handling overtesting combined with a sensible choice of trial spaces stands out from competing standard methods, and it is the main objective of this project to provide a mathematical foundation for this observation.

    Summarizing, the purpose of this project is to provide a new class of numerical techniques for solving parabolic partial differential equations and to establish a solid mathematical foundation for it.

  9. GRF HKBU 12304114 Error-optimal methods for PDEs arise from meshless theories

    Funding amount: $614,810

    Period: 01/2015-12/2017

    Abstract: Many problems in science and engineering require solving partial differential equations (PDEs). The area of meshless methods, including radial basis function (RBF) collocation or the Kansa's method, is an important area of contemporary research into numerical PDEs; most works focus on elliptic and parabolic types. Meshless methods enjoy an advantages that they do not require construction and possible manipulation of grids. Hence, the rather short history of meshless methods already features plenty of nicely calculated practical applications. On the other hand, the theories of the Kansa's method are very limited. Some researchers may not be persuaded that the Kansa's method is the right approach at all.

    Providing any theoretical result for the Kansa's method, we look at the issue that motivated Kansa back then, the RBF interpolation problem which has far more complete theoretical supports, and use the generic error estimates for (generalized) RBF interpolation to derive new numerical methods for solving PDEs. To be more specific, a generic error estimate of RBF interpolation (analogous to that for polynomial interpolations) says that the ($L^\infty$) difference between the RBF interpolant and the function is bounded by the (Reproducing Kernel Hilbert Space, RKHS) norm of the function times some constant independent of the function. This constant, commonly known as the power function, can then be further bounded in terms of the fill distance of data to complete the error analysis.

    In this project, we focus on utilizing the power function, which depends on both the RKHS, and the distribution of the data, to design new algorithms. Once the kernel and hence its RKHS are fixed, the function or PDE solution and its norm are also specified. Moreover, using the Mat\'{e}rn kernels, which reproduce the standard Sobolev spaces, we are able to evaluate error (up to a factor of the norm of the solution) in the familiar Sobolev sense, rather than some abstract RKHS. Based on information extracted from the corresponding power function, we can design algorithms to minimize the power function (over some Sobolev norms) under various constraints for the sake of numerical efficiency; hence, we title this project as Error-Optimal methods.

  10. GRF HKBU 201813 Convergence theories and adaptive algorithms for Kansa Methods

    Funding amount: $592,987

    Period: 01/2014-12/2016

    Abstract: To solve and investigate a real-life problem, a mathematical model is often constructed, which may involve differential equations and certain conditions. The traditional numerical solution procedure for such a model decomposes the region of interest into small sub-regions defined by a mesh, from relationships between these sub-regions. This research project in applied and computational mathematics is focused on whether a mathematical model can be solved without the need for meshing?

    Although a meshed landscape, car or airplane may have aesthetic appeal for some, unfortunately the construction of a mesh is a ``bottleneck" in many computations. Thus, although the actual computations may take few hours to complete, the mesh may take days or weeks to detail for complex geometries. So, there has been considerable recent interest in a new class of computational methods that do not require meshes -- i.e., meshless methods. Since 2000, the PI for this project has been working intensively on developing the relevant algorithmic theory and on various applications of meshless methods. The aim of this research project is to develop a new algorithm that can avoid any mesh construction and exploit adaptive processes, for the convenience of the user who seeks a robust and efficient computational solution of a mathematical model.

    An important missing aspect of the meshless asymmetric technique introduced by Kansa, designed for models involving time-independent linear partial differential equations (PDE), is a solid mathematical basis for solution convergence. In this project, it is intended to provide the scientific and engineering communities with stable numerical algorithms for this method. The primary aim is to develop both theory and algorithms for the constrained least-squares Kansa method. The theoretical part is to determine how good a numerical solution could be, due to features such as the approximation power and convergence rate of these meshless approaches. On the practical side, proper implementations and efficient algorithms to realize the theoretically proven advantages will be provided. Further appropriate techniques such as adaptive data points refinement, adaptive greedy algorithms in the meshless numerical differentiation scheme will play an important role.

  11. GRF HKBU 201509 Regularization method for detecting corrosion in pipelines

    Funding amount: $486,000

    Period: 09/2009-08/2011

    Abstract: One of the most important topics in the industry is the detection of corrosion inside a pipe. Detecting corrosion by electrical field can be modeled by a Cauchy problem of Laplace's equation in annulus domain under the assumption that the thickness of the pipe is relatively small compared with the radius of the pipe. The interior surface of the pipe is inaccessible and the nondestructive detection is solely based on measurements from the outer layer. Two cases are considered: first, measurements are available on the whole outer boundary. Second, measurements are available only on part of the outer boundary. The Cauchy problem for an elliptic equation is a typical ill-posed problem whose solution does not depend continuously on the boundary data. The theoretical goal of this work is the stabilities of these two cases by imposing reasonable assumptions. Numerical algorithms will be derived directly from the theoretical results. These algorithms will be compared with common methods, including the regularized method of fundamental solutions, regularized boundary element methods and boundary control methods, for solving Cauchy problems.

  12. GRF HKBU 201408 Convergence Analysis for Meshless Unsymmetric Collocation Method with Multiscale Property

    Funding amount: $293,700

    Period: 09/2008-08/2010

    Abstract: The history of meshless collocation methods featured plenty of nicely calculated practical solutions, but a solid mathematical basis was long missing for the most popular asymmetric technique introduced by E. Kansa. Thus the impact of this work will be to supply a lasting mathematical foundation which will also improve our general understanding of such technique. Our previous research gave a convergent algorithm with solid theoretical background; however, its implementation is not trivial. In this project, we continue to analyze the convergent theories to other easy-to-implement algorithms, to explore the pre-wavelet property of certain radial basis functions, and to further improve the efficiency of our previously proposed adaptive algorithm.

Other Research Grants:

  1. Guangdong-Hong Kong Universities “1+1+1” Joint Funding Programme Data-driven scientific and engineering computing algorithms and software

    Funding amount: $800,000 + RMB¥1,600,000

    Period: 01/05/2025-30/04/2028

    BNBU Co-PI: Y. Di

    HKBU Co-I: Y.S. Hon, K.F. Lam, & X. Tang

  2. RMGS 2023-16-06 Marine Analytics: Theory, Algorithms and Applications

    Funding amount: $2,154,774

    Period: 12/6/2023-11/6/2025

  3. RMGS 2023-15-05 Physics-Informed Regression and PDE solvers

    Funding amount: $4,991,758

    Period: 11/2/2023-10/2/2025

  4. RMGS 2022-11-09 Meshfree approximations for high dimensional manifolds and machine learning

    Funding amount: $4,998,157

    Period: 21/3/2022-20/3/2024

  5. NSFC 1516A020317 A study on numerical modeling of large-scale anomalous diffusion

    Funding amount: RMB¥200,000

    Period: 2016-2017

    Collaborative Institution: Hohai University

  6. NSC 94-2119-M-110-015 Theoretical Analysis and Computational Methods for Inverse Heat Source

    Funding amount: TWD$374,000

    Period: 2005-2006

Scholarships & Fellowships:

  1. JSPS Postdoctoral Fellowship Multivariate numerical differentiation for noisy scattered data

    Funding amount: ¥5,384,000

    Period: 2004-2005

  2. Natural Sciences and Engineering Research Council of Canada PGS-B

    Funding amount: C$21,000 per year

    Period: 2001-2003

  3. Natural Sciences and Engineering Research Council of Canada PGS-A

    Funding amount: C$17,000 per year

    Period: 1998-2000

Editorial & Research Service:

Conferences & Workshops:

Earlier events