The Closest Point Method for Surface PDEs with General Boundary Conditions

Tony WongColin B. Macdonald, Byungjoon Lee · 2025

Read The Closest Point Method for Surface PDEs with General Boundary Conditions by Tony WongColin B. Macdonald, Byungjoon Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-17820.txt.

Introduction

Tony Wong, Colin Macdonald, and Byungjoon Lee extend the closest point method, a numerical technique for solving equations posed on curved surfaces. The method embeds a surface in ordinary space, but handling what happens at a surface edge has limited it to simple boundary rules.

Their extrapolation approach handles a broad class of Neumann and Robin conditions, which mix a function's value and its slope at the edge, within one framework. Tests cover a basic elliptic problem, eigenvalue problems, and a nonlinear reaction and diffusion system. Accuracy and robustness in those tests support using the method for surface problems with realistic boundaries.

Catalog record

CREATORTony WongColin B. Macdonald, Byungjoon Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-17820

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Closest Point Method for Surface PDEs with General Boundary Conditions Tony WongColin B. Macdonald, Byungjoon Lee Abstract We generalize the closest point method (CPM) to solve surface partial differential equations with general boundary conditions. The proposed extrapolation method provides a unified framework for treating a broad class of inhomogeneous Neumann and Robin boundary conditions within the framework of CPM. The accuracy and robustness of the method are demonstrated through numerical convergence studies of an elliptic problem, Steklov eigenvalue problems, and a nonlinear reaction–diffusion system. 5 5 footnotetext: Corresponding author. Email: [email protected] Keywords— Closest point method, surface computation, implicit surfaces, partial differential equations, boundary conditions 1 Introduction Many partial differential equations (PDEs) arising in scientific and engineering problems are defined on curved surfaces, which requires robust and accurate numerical methods to resolve the geometry. The closest point method (CPM) [ 18 ] is a versatile embedding method for numerically solving PDEs on curved geometry, with applications such as pattern formation [ 13 , 4 , 16 , 17 ] , geometry and image processing [ 19 , 2 , 3 , 10 , 1 ] , and first-passage time problems of diffusion [ 9 , 8 ] . However, there is limited discussion on the uses of CPM for open surfaces. It remains an open challenge to extend the CPM to surface PDEs with general boundary conditions beyond the homogeneous Neumann and Dirichlet types [ 12 ] . In this paper, we provide a simple and accurate extrapolation technique for the CPM to tackle a broad class of boundary conditions. Our work is particularly relevant for applications where boundary effects are critical, such as fluid dynamics on curved interfaces and biological membrane modeling. 1.1 Terminology and notations Let \mathcal{S} be an open surface — a smooth surface embedded in \mathbb{R}^{3} , with a co-dimension one boundary \partial\mathcal{S} . For \bm{x}\in\mathcal{S} , let \mathbf{N}(\bm{x}) be a unit normal to \mathcal{S} . For \bm{y}\in\partial\mathcal{S} , let \mathbf{T}(\bm{y}) a unit tangent to \partial\mathcal{S} , and \mathbf{n}(\bm{y}) the unit outward co-normal to \partial\mathcal{S} , as in Figure 1 . Let \nabla_{S} be the surface gradient on \mathcal{S} and \partial_{n}=\mathbf{n}\cdot\nabla_{\mathcal{S}} be the co-normal derivative defined on \partial\mathcal{S} . In this paper, we consider the following boundary condition for a function u defined on \mathcal{S} : \partial_{n}u\big|_{\bm{y}}=j\left(\bm{y},u(\bm{y})\right),\hskip 18.49988pt\bm{y}\in\partial\mathcal{S}. (1.1) The form ( 1.1 ) includes many commonly used boundary conditions. For example, the homogeneous Neumann condition corresponds to j=0 , while the Robin condition corresponds to j=-\kappa u . The closest point function of \mathcal{S} is defined by \mathrm{cp}(\bm{x}):=\argmin_{\bm{y}\in\mathcal{S}}\|\bm{x}-\bm{y}\| , where \|\cdot\| denotes the Euclidean norm. Provided that \mathcal{S} is smooth, there exists a tubular neighbourhood \mathcal{B}(\mathcal{S}) where the closest point function \mathrm{cp}(\bm{x}) is well-defined [ 7 ] . Within \mathcal{B}(\mathcal{S}) , we define the set of interior points as \mathcal{I}(\mathcal{S})=\{\bm{x}\in\mathcal{B}(\mathcal{S}):\mathrm{cp}(\bm{x})\notin\partial\mathcal{S}\} and the set of exterior points as \mathcal{E}(\mathcal{S})=\{\bm{x}\in\mathcal{B}(\mathcal{S}):\mathrm{cp}(\bm{x})\in\partial\mathcal{S}\} (see Figure 1 (b)). The closest point extension operator E extends a surface function u to \mathcal{B}(\mathcal{S}) by Eu(\bm{x}):=u(\mathrm{cp}(\bm{x})) . The operator E propagates the surface values along the normal direction in \mathcal{I}(\mathcal{S}) and the boundary values to \mathcal{E}(\mathcal{S}) . Following [ 12 ] , we define the modified closest point function \bar{\mathrm{cp}}(\bm{x}):=\mathrm{cp}(r(\bm{x})) , where r(\bm{x}) is the reflection of \bm{x} across S , i.e., r(\bm{x}):=2\,\mathrm{cp}(\bm{x})-\bm{x} . The modified closest point extension operator \bar{E} is defined by \bar{E}u(\bm{x}):=u(\bar{\mathrm{cp}}(\bm{x})) . If \bm{x}\in\mathcal{I}(\mathcal{S}) , we have \bar{\mathrm{cp}}(\bm{x})=\mathrm{cp}(\bm{x}) . Therefore, the extensions \bar{E}u and Eu coincide in \mathcal{I}(\mathcal{E}) . For \bm{x}\in\mathcal{E}(\mathcal{S}) , the value of \bar{E}u(\bm{x}) is \bar{E}u(r(\bm{x})) . In other words, \bar{E} mirrors the interior values to the exterior points, and indeed this is sufficient to implement homogeneous Neumann conditions [ 12 ] , but is not enough for a smooth implementation of ( 1.1 ). Figure 1: (a) Hemisphere \mathcal{S} with boundary \partial\mathcal{S} , normal \mathbf{N} , tangent \mathbf{T} , and outward co-normal \mathbf{n} . (b) An interior point \bm{x}_{1}\in\mathcal{I}(\mathcal{S}) , we have \bar{\mathrm{cp}}\left(\bm{x}_{1}\right)={\mathrm{cp}}\left(\bm{x}_{1}\right) . An exterior point \bm{x}_{2}\in\mathcal{E}(\mathcal{S}) , which satisfies \mathrm{cp}\left(\bm{x}_{j}\right)\in\partial\mathcal{S} and \bar{\mathrm{cp}}\left(\bm{x}_{j}\right)\neq\mathrm{cp}\left(\bm{x}_{j}\right) . (c) The smooth extension \bar{u} defined in ( 2.4 ). 2 The closest point method We introduce two lemmas to support our main theorem characterizing a smooth extension which propagates surface values constantly along the normal direction in \mathcal{I}(\mathcal{S}) and into \mathcal{E}(\mathcal{S}) according to the boundary condition ( 1.1 ). We then describe the numerical discretization. 2.1 Extrapolation for the general boundary condition Lemma 1 . Suppose \bm{x}\in\mathcal{E}(\mathcal{S}) . Let \mathbf{N}=\mathbf{N}(\mathrm{cp}(\bm{x})) and \mathbf{n}=\mathbf{n}(\mathrm{cp}(\bm{x})) . Then \bm{x}-\mathrm{cp}(\bm{x}) lies in \mathrm{span}\{\mathbf{N},\mathbf{n}\} . Proof. Near \mathrm{cp}(\bm{x}) , there exists a local chart (U,\phi) mapping a neighborhood U\subset\mathcal{S} to the half-plane \{(\xi,\eta)\in\mathbb{R}^{2}:\eta\geq 0\} , with \partial\mathcal{S}\cap U corresponding to \eta=0 . The boundary curve admits the 1D parameterization \gamma(t)=\phi^{-1}(t,0) for t in some interval I , and \gamma is smooth and regular. Let \mathrm{cp}(\bm{x})=\gamma(t_{*}) with t_{*}=\argmin_{t\in I}\|\gamma(t)-\bm{x}\|^{2} . The first-order condition gives \gamma^{\prime}(t_{*})^{T}[\gamma(t_{*})-\bm{x}]=0 , so \bm{x}-\mathrm{cp}(\bm{x}) is orthogonal to \gamma^{\prime}(t_{*}) , the tangent to \partial\mathcal{S} at \mathrm{cp}(\bm{x}) . Since \{\mathbf{N},\mathbf{T},\mathbf{n}\} is an orthogonal basis, it follows that \bm{x}-\mathrm{cp}(\bm{x})\in\mathrm{span}\{\mathbf{N},\mathbf{n}\} . ∎ With Lemma 1 , we characterize how a smooth extension \bar{u} propagates the boundary values into \mathcal{E}(\mathcal{S}) , according to the ( 1.1 ). Lemma 2 . Let u be a smooth function defined on \mathcal{S} that satisfies ( 1.1 ). Let \bar{u} be a smooth extension of u in \mathcal{B}(\mathcal{S}) such that \bar{u}(\bm{x})=u(\mathrm{cp}(\bm{x})) for \bm{x}\in\mathcal{I}(\mathcal{S}) . Suppose \bm{x}\in\mathcal{E}(\mathcal{S}) . Let \mathbf{w}=(\bm{x}-\mathrm{cp}(\bm{x}))/\|\bm{x}-\mathrm{cp}(\bm{x})\| . Denote the unit outward co-normal \mathbf{n}(\mathrm{cp}(\bm{x})) by \mathbf{n} . Then, we have \partial_{w}\bar{u}\,|_{\mathrm{cp}(\bm{x})}=\langle\mathbf{w},\mathbf{n}\rangle\,j\big(\mathrm{cp}(\bm{x}),\bar{u}(\mathrm{cp}(\bm{x}))\big)\,,\hskip 18.49988pt\mbox{where $\langle\cdot,\cdot\rangle$ denotes the Euclidean inner product.} (2.1) Proof. By Lemma 1 , \mathbf{w}\in\mathrm{span}\{\mathbf{N},\mathbf{n}\} , where \mathbf{N} is a unit normal at \mathrm{cp}(\bm{x}) . We apply the orthogonal decomposition \mathbf{w}=\langle\mathbf{w},\mathbf{n}\rangle\mathbf{n}+\langle\mathbf{w},\mathbf{N}\rangle\mathbf{N} and the linearity of directional derivatives to obtain \partial_{w}\bar{u}\,|_{\mathrm{cp}(\bm{x})}=\langle\mathbf{w},\mathbf{n}\rangle\,\partial_{n}\bar{u}\,|_{\mathrm{cp}(\bm{x})}+\langle\mathbf{w},\mathbf{N}\rangle\,\partial_{N}\bar{u}\,|_{\mathrm{cp}(\bm{x})}\,. Now \bar{u} is constant along the normal \mathbf{N} so \partial_{N}\bar{u}|_{\mathrm{cp}(\bm{x})}=0 . Secondly, we have \partial_{n}\bar{u}|_{\mathrm{cp}(\bm{x})}=\partial_{n}u|_{\mathrm{cp}(\bm{x})}=j\big(\mathrm{cp}(\bm{x}),u(\mathrm{cp}(\bm{x}))\big)=j\big(\mathrm{cp}(\bm{x}),\bar{u}(\mathrm{cp}(\bm{x}))\big) . Hence, we obtain ( 2.1 ). ∎ Theorem 1 . Suppose \mathcal{S} is a smooth surface with a tubular neighbourhood \mathcal{B}(\mathcal{S}) . Suppose \bar{u}:\mathcal{B}(\mathcal{S})\to\mathbb{R} satisfies the extrapolation formula for boundary points \displaystyle\bar{u}(\bm{x})=\bar{u}(\bar{\mathrm{cp}}(\bm{x})) \displaystyle+2\big\langle\bm{x}-\mathrm{cp}(\bm{x}),\mathbf{n}(\mathrm{cp}(\bm{x}))\big\rangle\,j\big(\mathrm{cp}(\bm{x}),\bar{u}(\mathrm{cp}(\bm{x}))\big) (2.2) \displaystyle=\bar{E}\bar{u}(\bm{x}) \displaystyle+2\big\langle\bm{x}-\mathrm{cp}(\bm{x}),\mathbf{n}(\mathrm{cp}(\bm{x}))\big\rangle\,j\big(\mathrm{cp}(\bm{x}),E\bar{u}(\bm{x})\big),\hskip 18.49988pt\bm{x}\in\mathcal{E}(\mathcal{S}), and \bar{u}(\bm{x})=\bar{u}(\bar{\mathrm{cp}}(\bm{x})) for interior points \bm{x}\in\mathcal{I}(\mathcal{S}) . If \bar{u} is a sufficiently-smooth function when restricted to surfaces points \bm{y}\in\mathcal{S} , then \bar{u}|_{\partial S} satisfies the boundary condition ( 1.1 ) to within O(\|\bm{x}-\mathrm{cp}(\bm{x})\|^{3}) . Proof. Let \bm{x}\in\mathcal{E}(\mathcal{S}) . Applying the central difference formula to \partial_{w}\bar{u}|_{\mathrm{cp}(\bm{x})} with the stencil \{\bm{x},r(\bm{x})\} , we have \partial_{w}\bar{u}|_{\mathrm{cp}(\bm{x})}=\frac{\bar{u}(\bm{x})-\bar{u}(r(\bm{x}))}{\|\bm{x}-r(\bm{x})\|}+O(\|\bm{x}-\mathrm{cp}(\bm{x})\|^{2})=\frac{\bar{u}(\bm{x})-\bar{u}(\bar{\mathrm{cp}}(\bm{x}))}{2\|\bm{x}-\mathrm{cp}(\bm{x})\|}+O(\|\bm{x}-\mathrm{cp}(\bm{x})\|^{2}), (2.3) where the second equality follows by \bar{u}(r(\bm{x}))=\bar{u}(\bar{\mathrm{cp}}(\bm{x})) and that \mathrm{cp}(\bm{x}) is the mid-point of the segment connecting \bm{x} and r(\bm{x}) . Substituting ( 2.3 ) into ( 2.1 ), we obtain the extrapolation formula and the error term. ∎ Remark 1 . In practice, \|\bm{x}-\mathrm{cp}(\bm{x})\| is bounded by the radius of tubular neighbourhood, which is a small multiple of the mesh size \Delta x [ 18 , 14 ] . Hence, the error term in Theorem 1 is O(\Delta x^{3}) . We can combine the extension condition \bar{u}=\bar{E}\bar{u} in \mathcal{I}(\mathcal{S}) with the extrapolation formula ( 2.2 ) to define an extension in all of \mathcal{B}(\mathcal{S}) : \bar{u}(\bm{x})=\bar{E}\,\bar{u}(\bm{x})+2\,\chi_{\mathcal{E}(\mathcal{S})}(\bm{x})\,\langle\bm{x}-\mathrm{cp}(\bm{x}),\mathbf{n}\rangle\,j\big(\mathrm{cp}(\bm{x}),E\,\bar{u}(\bm{x})\big),\hskip 9.24994pt\bm{x}\in\mathcal{B}(\mathcal{S}), (2.4) where \chi_{\mathcal{E}(\mathcal{S})} is the indicator function of \mathcal{E}(\mathcal{S}) . 2.2 Embedding formulation and discretization We illustrate how to incorporate the extrapolation method into the CPM through an elliptic example: \Delta_{\mathcal{S}}u-c\,u=f\hskip 9.24994pt\text{in $\mathcal{S}$,}\hskip 18.49988pt\mbox{subject to \eqref{eqn:gbc} in $\partial\mathcal{S}$,} (2.5) where \Delta_{\mathcal{S}} is the Laplace–Beltrami operator on

Read the complete text