A TraceFEM C 0 C^{0} Interior Penalty Method for the Surface Biharmonic Equation
Michael Neilan, Hongzhi Wan · 2025
Read A TraceFEM C 0 C^{0} Interior Penalty Method for the Surface Biharmonic Equation by Michael Neilan, Hongzhi Wan (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-2512-18949.txt.
Introduction
Michael Neilan and Hongzhi Wan build a finite element method for fourth-order problems living on curved surfaces, like bending a thin shell that exists only as a surface in space. The surface biharmonic equation is the model case. Their TraceFEM approach, posted to arXiv in December 2025, uses ordinary quadratic elements on a three-dimensional background mesh and reads the solution off a polyhedral approximation of the surface.
Stability comes from penalties: surface edge terms plus bulk penalties on jumps in the gradient and Hessian. The authors prove optimal first-order convergence in a discrete H2 norm and quadratic convergence in the L2 norm, the best rates the spaces allow. For computational scientists modeling membranes, interfaces, and surface flows, a proven method on background meshes avoids the pain of meshing the surface itself.
Catalog record
| CREATOR | Michael Neilan, Hongzhi Wan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 84 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-18949
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A TraceFEM C 0 C^{0} Interior Penalty Method for the Surface Biharmonic Equation Thanks: The authors were supported in part through the NSF, grant DMS-2309425. Michael Neilan, Hongzhi Wan Abstract . We construct and analyze a TraceFEM discretization for the surface biharmonic problem. The method utilizes standard quadratic Lagrange finite element spaces defined on a three-dimensional background mesh and a symmetric C 0 C^{0} interior penalty formulation posed on a second-order polyhedral approximation of the surface. Stability is achieved through a combination of surface edge penalties and bulk-facet penalization of gradient and Hessian jumps. We prove optimal first-order convergence in a discrete H 2 H^{2} norm and quadratic convergence in the L 2 L^{2} norm. 1. Introduction TraceFEM has become a widely used unfitted method for surface PDEs due to its conceptual simplicity and robustness with respect to surface geometry. In contrast to surface finite element methods (SFEMs), which build discrete spaces directly on a surface triangulation, TraceFEM employs standard volumetric finite element spaces defined on a fixed three-dimensional background mesh and poses the discrete formulation on the induced two-dimensional surface mesh. This approach enables the use of conventional finite element spaces, and it is particularly well-suited for problems involving evolving surfaces as well as problems with coupling conditions with the surrounding bulk. This framework is now relatively established and has been successfully applied for second-order elliptic problems, as well as for the surface Navier-Stokes equations (cf., e.g., [ 20 , 21 , 5 , 14 ] ). Although several TraceFEM (and SFEM) discretizations have been developed in the past few decades, substantially fewer results exist for finite element methods for fourth-order surface PDEs. Existing contributions include SFEMs based on C^{0} interior penalty (IP) methods in [ 15 , 18 ] , non-conforming SFEMs using Zienkiewicz-type elements [ 27 ] , and SFEMs built on either Ciarlet-Raviart or Hellan–Herrmann–Johnson mixed formulations [ 19 , 1 , 26 , 25 ] . In contrast, to the best of our knowledge, the only TraceFEM discretization for fourth-order surface PDEs is given in [ 23 , 2 ] which proposes and analyzes TraceFEM discretizations for the surface stream function formulation of the Stokes problem, a fourth-order PDE whose principal operator is the surface biharmonic. Their method introduces a Ciarlet-Raviart-type mixed finite element formulation, where the Laplace-Beltrami of the solution is introduced as an additional unknown. While this construction is relatively straightforward, using standard Lagrange finite element spaces, it increases the number of degrees of freedom and leads to a relatively large saddle-point system. In this paper, we introduce and analyze a TraceFEM for the surface biharmonic problem based on a symmetric C^{0} IP formulation [ 3 , 10 ] . The discrete space is the standard quadratic Lagrange finite element space defined on a three-dimensional background mesh, with consistency enforced through edge-based integral terms. In addition to penalization of (surface) gradients over surface edges, analogous to the Euclidean setting, the proposed method also includes penalization of the gradient and Hessian operators across facets in the background mesh. We show that these bulk-penalty terms ensure stability of the method, independent to how the mesh intersects the surface. We prove optimal first-order convergence in a discrete H^{2} norm and quadratic convergence in the L^{2} norm, with both estimates taking geometric errors into account. The proposed method offers several advantages, foremost is its simplicity: similar to [ 23 , 2 ] , it uses standard Lagrange finite element spaces which are supported in existing software programs (e.g., NGSolve [ 24 ] ). In addition, the proposed scheme is a primal formulation involving a single unknown. The resulting algebraic system is symmetric positive definite, in line with the underlying structure of the elliptic PDE. As far as we are aware, this is the first primal TraceFEM for fourth-order surface PDEs. The rest of the paper is organized as follows. Section 2 introduces the relevant surface differential operators as well as the surface biharmonic problem. Section 3 describes the surface approximation, triangulations, scaling estimates and norm equivalences with respect to the exact and discrete surfaces. In Section 4 , we present the C^{0} IP method and establish coercivity, proving that the discrete problem is well posed. Section 5 gives the error analysis, including optimal first-order convergence in a discrete H^{2} norm and second-order L^{2} estimates obtained via a duality argument. Finally, in Section 6 we perform some numerical experiments which support the theoretical results. 2. The surface biharmonic problem We assume \Gamma is a simply connected compact oriented smooth surface without boundary in \mathbb{R}^{3} . We assume that \Gamma is the level set of a smooth function \phi , i.e., \Gamma=\{x\in\mathbb{R}^{3}:\ \phi(x)=0\} . Let U_{\delta}(\Gamma)=\{x\in\mathbb{R}^{3}:{\rm dist}(x,\Gamma)<\delta\} be a neighborhood of \Gamma with \delta>0 sufficiently small such that the signed distance function d (with d<0 in the interior of \Gamma ) is well defined. Set {\bm{n}}=\nabla d , a column vector, so that {\bm{n}}|_{\Gamma} is the outward unit normal of \Gamma . We let {\bf P} be the tangential projection, i.e., {\bf P}={\bf I}-{\bm{n}}\otimes{\bm{n}}, where {\bf I} denotes the 3\times 3 identity matrix. The closest point projection {\bm{p}}:U_{\delta}\to\Gamma is {\bm{p}}(x)=x-d(x){\bm{n}}(x). The Weingarten map is denoted by {\bf H}=\nabla^{2}d , and the principal curvatures of \Gamma are denoted by \kappa_{1} and \kappa_{2} . For a function \psi defined on the surface \Gamma , its extension \psi^{e}:U_{\delta}\rightarrow\mathbb{R} is given by \psi^{e}=\psi\circ{\bm{p}}. On the other hand, a function \psi defined on some \tilde{\Gamma}\subset U_{\delta}(\Gamma) , can be lifted to \Gamma via \psi^{\ell}\circ{\bm{p}}=\psi\quad\text{on }\tilde{\Gamma}. We will often drop the subscripts e and \ell for the extensions and lifts if the context is clear. The surface gradient of a scalar function \psi:\Gamma\to\mathbb{R} is given by \nabla_{\Gamma}\psi={\bf P}\nabla\psi^{e}. For a vector-valued function {\bm{v}}=[v_{1},v_{2},v_{3}]^{\intercal} , the extension is defined componentwise so that ({\bm{v}}^{e})_{j}=v_{j}^{e} and its Jacobian \nabla\!\!\!\!\nabla{\bm{v}} satisfies (\nabla\!\!\!\!\nabla{\bm{v}})_{i,j}=\frac{\partial v_{i}}{\partial x_{j}} for i,j=1,2,3 . The surface Jacobian and the surface divergence are as follows: \nabla\!\!\!\!\nabla_{\Gamma}{\bm{v}}={\bf P}\nabla\!\!\!\!\nabla{\bm{v}}^{e}{\bf P},\qquad{\rm div}_{\Gamma}{\bm{v}}={\rm tr}(\nabla\!\!\!\!\nabla_{\Gamma}{\bm{v}}). Combining the above definitions, we define the projected surface Hessian \nabla_{\Gamma}^{2}\psi:\Gamma\rightarrow\mathbb{R}^{3\times 3} and Laplace-Beltrami operator \Delta_{\Gamma}\psi:\Gamma\to\mathbb{R} as \nabla_{\Gamma}^{2}\psi=\nabla\!\!\!\!\nabla_{\Gamma}\nabla_{\Gamma}\psi,\quad\Delta_{\Gamma}\psi={\rm div}_{\Gamma}\nabla_{\Gamma}\psi={\rm tr}(\nabla_{\Gamma}^{2}\psi). Given a source function f:\Gamma\to\mathbb{R} with vanishing mean value \int_{\Gamma}f=0 , we consider the problem of finding u:\Gamma\rightarrow\mathbb{R}^{3} satisfying \int_{\Gamma}u=0 such that (2.1) \Delta_{\Gamma}^{2}u=f\quad\text{on }\Gamma, where \Delta_{\Gamma}^{2}u:=\Delta_{\Gamma}(\Delta_{\Gamma}u) . 3. Preliminaries 3.1. Approximate Geometries and Meshes Let \mathcal{T}^{\rm ext}_{h} be a quasi-uniform mesh of shape-regular simplices of a polygonal domain containing U_{\delta}(\Gamma) with h\approx{\rm diam}(T)\ \forall T\in\mathcal{T}^{ext} . Let \phi_{h} be the piecewise linear Lagrange interpolant of the level set \phi , and let \Gamma_{h}=\{x\in\mathbb{R}^{3}:\ \phi_{h}(x)=0\} be a polyhedral approximation to \Gamma . By standard approximation properties of the Lagrange interpolant, we have (3.1) \|d\|_{L^{\infty}(\Gamma_{h})}\lesssim h^{2},\qquad\|{\bm{n}}-{\bm{n}}_{h}\|_{L^{\infty}(\Gamma_{h})}\lesssim h, where {\bm{n}}_{h} is the outward unit normal of \Gamma_{h} . Here, we use the notation A\lesssim B to mean A\leq cB for some constant c>0 independent of h as well as A\approx B to mean A\lesssim B and B\lesssim A . The constant c is also independent on how the surface \Gamma_{h} intersects the background mesh \mathcal{T}^{\rm ext}_{h} . Above and throughout this paper, we assume h is sufficiently small so that \Gamma_{h}\subset U_{\delta} , in particular, the closest point projection is well defined on \Gamma_{h} . We define the tangential projection with respect to \Gamma_{h} as {\bf P}_{h}={\bf I}-{\bm{n}}_{h}\otimes{\bm{n}}_{h} and