betaFeedback

WorksMathematics

A Cut Finite Element Method with Boundary Value Correction for Robin, Neumann and Interface Conditions

Abstract

We extend the cut finite element method with boundary value correction of Burman, Hansbo and Larson [Math.\ Comp.\ 87 (2018) 633–657] to Robin and Neumann conditions and to interface problems with imperfect bonding. The formulation retains a piecewise affine boundary for all integration, while using the exact geometry only at boundary quadrature points. The correction uses Taylor expansions of the solution and its gradient along the discrete normal, with the exact normal used to evaluate the transported flux. At an interface, a further correction to the normal flux jump accounts for the difference between these two normals. Under the stated trace and Poincaré bounds, the corrected form is an O(h1/2)O(h^{1/2}) perturbation of a symmetric form, whose matrix can therefore be used as a preconditioner. For a fixed finite Robin coefficient, we obtain the energy error bound Cβ(hp+δhm+1)C_β(h^p+δ_h^{m+1}), whereas the bound uniform in the Robin coefficient is C(hp+h−1/2δhm+1)C(h^p+h^{-1/2}δ_h^{m+1}), provided the Poincaré constant is uniform. With δh=O(h2)δ_h=O(h^2), the sufficient Taylor orders are therefore 2m+2≥p2m+2\ge p and 2m+32≥p2m+\frac32\ge p, respectively. Numerical examples for p=1,2,3p=1,2,3 illustrate these rates and the effect of the flux-jump correction in the interface problem.

Introduction

Unfitted finite element methods use meshes that need not align with the boundary or interface of the physical domain, which considerably simplifies mesh generation. The geometric approximation must nevertheless be sufficiently accurate to support the desired convergence rate. If the boundary is replaced by a piecewise affine surface Γh\Gamma_h, imposing the boundary condition directly on Γh\Gamma_h limits this rate, regardless of the polynomial degree.

There are two main approaches to this problem. A higher order geometric representation can be used, with curved elements or quadrature on the exact domain. Alternatively, the affine approximation can be retained and the boundary values corrected through a Taylor expansion from Γ\Gamma to Γh\Gamma_h. This idea was introduced by Bramble, Dupont and Thomée [6] for polygonal approximations of curved domains. Burman, Hansbo and Larson [9] subsequently developed it for unfitted methods with Nitsche boundary conditions and ghost penalty stabilisation. Their analysis shows that a correction of order mm gives the optimal energy error for elements of degree pp when 2m+32≥p2m+\frac{3}{2}\ge p. Thus one Taylor term suffices for p=2,3p=2,3 and two for p=4,5p=4,5 [9 Tab. 1]. Without correction, the geometric distance must satisfy δh≲hp+1/2\delta_h\lesssim h^{p+1/2} for the energy estimate and δh≲hp+1\delta_h\lesssim h^{p+1} for the L2L^2 estimate [9 Rem. 3.5]. With one correction term, a piecewise affine boundary with δh=O(h2)\delta_h=O(h^2) suffices for degrees up to three.

The shifted boundary method of Main and Scovazzi [20, 21] uses the same principle, with a surrogate boundary formed by mesh faces. The computation then takes place on the resulting fitted submesh, and we refer to [1] for the analysis. Related methods on polytopal approximations were developed in [12, 14], while the virtual element methods in [4, 2] also allow staircase boundaries from pixel tessellations. For unfitted meshes, boundary value correction has also been combined with Lagrange multipliers [10], a penalty-free Nitsche method [5], and a mixed formulation [17]. An alternative based on isoparametric mappings of cut elements is given in [19]. These developments mainly concern Dirichlet boundary conditions.

For Neumann conditions, Bertoluzza, Montardini and Prada [3] develop a high order virtual element method on approximate domains. Their formulation allows non-aligned physical and numerical normals, with optimal error estimates under suitable assumptions for arbitrary approximation order. This provides a related treatment of the geometric difficulties that arise when the boundary condition involves the flux.

A Dirichlet condition prescribes the solution trace, so its transport requires only a Taylor expansion of that trace. The normal flux enters through integration by parts on Ωh\Omega_h, where the resulting terms use the normal of Γh\Gamma_h and are exact on the discrete domain.

For a Robin condition, the boundary data also involve the flux, and we consider

κ∂nu+βu=gon Γ,(1)\kappa\partial_n u+\beta u=g \quad\text{on }\Gamma, \tag*{(1)}

where nn is the exact normal. Since the discrete normal nhn_h generally differs from nn by O(h)O(h), transporting the evaluation point while retaining nhn_h leaves an O(h)O(h) error in the flux. We remove this error by transporting the vector field κ∇u\kappa\nabla u and then taking its scalar product with the exact normal n∗n^* at the image point. The resulting operator, denoted by Λm\Lambda_m, has an error of O(δhm+1)O(\delta_h^{m+1}), as for the trace correction (Lemma 3.2). The construction uses the polynomial gradient on each cut element together with pointwise evaluations of the exact normal, without requiring curved quadrature or additional unknowns.

We combine this flux correction with a formulation that is uniform in the Robin coefficient β\beta, connecting the Neumann problem at β=0\beta=0 to the Dirichlet limit as β→∞\beta\to\infty. Following Juntunen and Stenberg [18], we set ρ=γκ/h\rho=\gamma\kappa/h and introduce the weights

θ:=ββ+ρ∈[0,1],s:=1β+ρ,(2)\theta:=\frac{\beta}{\beta+\rho}\in[0,1], \qquad s:=\frac{1}{\beta+\rho}, \tag*{(2)}

which give a single formulation for the whole family. At θ=1\theta=1 and s=0s=0, it reduces to the Dirichlet method with boundary value correction from [9], while θ=0\theta=0 gives a corrected Neumann formulation. The pair (θ,s)(\theta,s) is equivalent to the weights in [18 (2.5)–(2.6)], but makes the roles of the trace and flux corrections explicit. The estimates have constants independent of β\beta under the stated coercivity and regularity assumptions.

We also consider an elliptic interface problem, where the prescribed flux jump is taken with respect to the exact normal, [κ∂nu]=gN[\kappa\partial_n u]=g_N. Integration by parts on the two discrete subdomains instead gives the jump with respect to nhn_h. Their difference contains

[κ∇u]⋅(nh−n∗).(3)[\kappa\nabla u]\cdot(n_h-n^*). \tag*{(3)}

This term depends on the tangential flux jump, which is not prescribed by the interface data and therefore cannot be removed by transporting those data alone. We retain the corresponding integration-by-parts term, −⟨[κ∂nhu],{v}⟩Γh-\langle[\kappa\partial_{n_h}u],\{v\}\rangle_{\Gamma_h}, in the formulation on Γh\Gamma_h. Although this term vanishes on the exact interface for a homogeneous normal flux jump, it need not vanish on the surrogate interface. We control it using two different trace estimates (Lemma 5.2). The elementwise gradient requires a discrete trace estimate with a factor h−1/2h^{-1/2}, whereas the test function belongs to H1H^1 and satisfies a continuous trace estimate without this factor. Together with ∥nh−n∗∥=O(h)\lVert n_h-n^*\rVert=O(h), this gives an O(h1/2)O(h^{1/2}) perturbation.

Finally, we consider imperfect bonding, where a spring layer with stiffness βΓ\beta_\Gamma joins the two phases, as in [16, 15]. For this interface counterpart of the Robin condition, the weight θΓ=βΓ/(βΓ+ρ)\theta_\Gamma=\beta_\Gamma/(\beta_\Gamma+\rho) gives a single formulation for all bonding stiffnesses. Uniform estimates additionally require the phasewise Poincaré bounds stated below. In §7, we interpret the boundary value correction as an artificial compliance of order h2h^2, given by the geometric offset weighted by the jump of the inverse diffusivity. This correction adds to the compliance in the physical bonding law.

We give the main arguments in Lemma 3.2, Lemma 5.2, Theorem 5.2 and Theorem 6.1, while proofs that follow from standard arguments are sketched.

Throughout, a≲ba\lesssim b means a≤Cba\leq Cb, where CC is independent of the mesh size, the position of the boundary relative to the mesh, β\beta, and the contrast, unless stated otherwise. It may depend on the polynomial degree, the shape regularity, the curvature of Γ\Gamma, and the stated Poincaré bounds. Uniformity in the physical parameters is therefore conditional on uniform control of these bounds and of the solution norms.

The Robin problem and its two limits

Let Ω⊂Rd\Omega\subset\mathbb{R}^d, d=2,3d=2,3, be a bounded domain with a C2C^2 boundary Γ=∂Ω\Gamma=\partial\Omega and outward unit normal nn. For constant κ>0\kappa>0, reaction coefficient c≥0c\geq0, and given ff, gg, and β∈[0,∞]\beta\in[0,\infty], we consider

−∇⋅(κ∇u)+cu=fin Ω,κ∂nu+βu=gon Γ.(4)\begin{aligned} -\nabla\cdot(\kappa\nabla u)+cu&=f &&\text{in }\Omega,\\ \kappa\partial_nu+\beta u&=g &&\text{on }\Gamma. \tag*{(4)} \end{aligned}

The case β=0\beta=0 gives the Neumann problem, while finite positive β\beta gives the Robin problem. The Dirichlet limit is obtained by setting g=βgDg=\beta g_D and letting β→∞\beta\to\infty with gDg_D fixed. We assume c≥c0>0c\geq c_0>0 or β≥β0>0\beta\geq\beta_0>0. For the pure Neumann problem (c=β=0c=\beta=0), the continuous solution is defined modulo constants and the data must satisfy the usual compatibility condition. We restrict both the discrete trial and test spaces to functions with zero mean on Ωh\Omega_h, and normalise the exact solution to have the same mean on Ωh\Omega_h in the error analysis. This restriction is needed because flux transport need not preserve the discrete compatibility condition for constant test functions. In [18], the condition is written in terms of the compliance ϵ=1/β\epsilon=1/\beta. We use β\beta so that the Neumann condition corresponds to the finite value β=0\beta=0 and the notation agrees with the bonding stiffness in §6.

We denote the normal flux by

λ(u):=κ∂nu=κn⋅∇uon Γ.(5)\lambda(u):=\kappa\partial_nu=\kappa n\cdot\nabla u \quad\text{on }\Gamma. \tag*{(5)}

The boundary condition then reads βu+λ(u)=g\beta u+\lambda(u)=g, while multiplication of the equation by a test function and integration by parts give

(κ∇u,∇v)Ω+(cu,v)Ω−⟨λ(u),v⟩Γ=(f,v)Ω.(6)(\kappa\nabla u,\nabla v)_\Omega+(cu,v)_\Omega-\langle\lambda(u),v\rangle_\Gamma=(f,v)_\Omega. \tag*{(6)}

The method combines this identity with the boundary condition through the following substitution.

Lemma 2.1 (The uniform substitution). Let ρ>0\rho> 0, let β∈[0,∞)\beta\in[0,\infty), and let θ\theta and ss be given by (2). Then θ=βs\theta= \beta s and 1−θ=ρs1-\theta= \rho s. For any UU, Λ\Lambda, and GG satisfying βU+Λ=G\beta U+\Lambda=G, we have

Λ=ρsG−θρU+θΛ,θU+sΛ=sG.(7)\Lambda=\rho sG-\theta\rho U+\theta\Lambda,\qquad\theta U+s\Lambda=sG. \tag*{(7)}

Proof. The first identity follows from (1−θ)Λ=ρsΛ=ρs(G−βU)=ρsG−θρU(1-\theta)\Lambda=\rho s\Lambda=\rho s(G-\beta U)=\rho sG-\theta\rho U. The second is s(βU+Λ)=sGs(\beta U+\Lambda)=sG. ◻

The first identity replaces part of the flux in (6) by the trace and the data. The second supplies a residual that may be tested without changing the exact solution. At β=0\beta=0, the identities become Λ=G\Lambda=G and sΛ=sGs\Lambda=sG, while in the Dirichlet limit they become Λ=Λ\Lambda=\Lambda and U=gDU=g_D.

Remark 2.1 (Relation to the method of Juntunen and Stenberg). Applying Lemma 2.1 on the exact domain with U=uU=u and Λ=λ(u)\Lambda=\lambda(u), without geometric correction, gives the symmetric formulation (κ∇u,∇v)Ω+(cu,v)Ω−θ(⟨λ(u),v⟩Γ+⟨u,λ(v)⟩Γ)+θρ⟨u,v⟩Γ−s⟨λ(u),λ(v)⟩Γ=(f,v)Ω+ρs⟨g,v⟩Γ−s⟨g,λ(v)⟩Γ(\kappa\nabla u,\nabla v)_\Omega+(cu,v)_\Omega-\theta(\langle\lambda(u),v\rangle_\Gamma+\langle u,\lambda(v)\rangle_\Gamma)+\theta\rho\langle u,v\rangle_\Gamma-s\langle\lambda(u),\lambda(v)\rangle_\Gamma=(f,v)_\Omega+\rho s\langle g,v\rangle_\Gamma-s\langle g,\lambda(v)\rangle_\Gamma. It agrees with [18 (2.5)–(2.6)] under the correspondence ϵ=1/β\epsilon=1/\beta and γJShE=1/ρ\gamma_{\mathrm{JShE}}=1/\rho. Their stability condition γJS<1/CI\gamma_{\mathrm{JS}}<1/C_I [18 Thm. 3.2] requires sufficiently large γ\gamma in our notation. We extend this form by introducing a surrogate boundary, a ghost penalty, and the two transport operators.

The discrete geometry and the two Taylor operators.

The surrogate boundary.

Let Th\mathcal{T}_h be a quasi-uniform, shape-regular simplicial mesh of a fixed polytopal domain containing Ω‾\overline{\Omega}, with mesh size h=max⁡T∈ThhTh=\max_{T\in\mathcal{T}_h}h_T. We describe the boundary by a level set function Φ∈C2\Phi\in C^2, with ∣∇Φ∣>0|\nabla\Phi|>0 near Γ\Gamma, such that

Ω={Φ<0},Γ={Φ=0},n=∇Φ∣∇Φ∣ on Γ.(8)\Omega=\{\Phi<0\},\qquad\Gamma=\{\Phi=0\},\qquad n=\frac{\nabla\Phi}{|\nabla\Phi|}\text{ on }\Gamma. \tag*{(8)}

The cut elements are ThΓ:={T∈Th:T∩Γ≠∅}\mathcal{T}_h^\Gamma:=\{T\in\mathcal{T}_h:T\cap\Gamma\ne\varnothing\}. In two dimensions, we assume that each cut element meets Γ\Gamma in one connected arc with endpoints on two element edges, and replace this arc by the chord ΓhT\Gamma_h^T joining its endpoints. In three dimensions we use a piecewise planar surface and assume the geometric estimates stated below. We set

Γh:=⋃T∈ThΓΓhT.(9)\Gamma_h:=\bigcup_{T\in\mathcal{T}_h^\Gamma}\Gamma_h^T. \tag*{(9)}

We assume that Γh\Gamma_h is a closed Lipschitz surface bounding a polytopal domain Ωh\Omega_h, with uniformly bounded Lipschitz character as h→0h\to0. The active mesh Thact\mathcal{T}_h^{\mathrm{act}} consists of the elements for which ∣T∩Ωh∣>0|T\cap\Omega_h|>0, and on their union Ωhact\Omega_h^{\mathrm{act}} we define

Vh:={v∈C0(Ωhact):v∣T∈Pp(T) ∀T∈Thact}.(10)V_h:=\left\{v\in C^0(\Omega_h^{\mathrm{act}}):v|_T\in P_p(T)\ \forall T\in\mathcal{T}_h^{\mathrm{act}}\right\}. \tag*{(10)}

The outward unit normal of Ωh\Omega_h, denoted by nhn_h, is constant on each affine piece of Γh\Gamma_h. Since all integrals are taken over Ωh\Omega_h, Γh\Gamma_h, or faces of the active mesh, the method requires quadrature only on polytopal sets.

The three pointwise maps.

Let x∈Γhx\in\Gamma_h be a point where nhn_h is defined. For sufficiently small hh, the equation Φ(x+rnh(x))=0\Phi(x+r n_h(x))=0 has a unique root near zero. We define

ϱh(x):=that root,ph(x):=x+ϱh(x)nh(x)∈Γ,n∗(x):=∇Φ(ph(x))∣∇Φ(ph(x))∣.(11)\varrho_h(x):=\text{that root},\qquad p_h(x):=x+\varrho_h(x)n_h(x)\in\Gamma,\qquad n^*(x):=\frac{\nabla\Phi(p_h(x))}{|\nabla\Phi(p_h(x))|}. \tag*{(11)}

Thus php_h transfers points from Γh\Gamma_h to Γ\Gamma along the discrete normal, while n∗(x)n^*(x) gives the exact normal at the image point ph(x)p_h(x). The geometric errors are measured by

δh:=∥ϱh∥L∞(Γh),θh:=∥nh−n∗∥L∞(Γh).(12)\delta_h := \|\varrho_h\|_{L^\infty(\Gamma_h)}, \qquad\theta_h := \|n_h-n^*\|_{L^\infty(\Gamma_h)}. \tag*{(12)}

Lemma 3.1 (Geometric consistency). In two dimensions, let Γ\Gamma be of class C2C^2 with curvature bounded by cΓc_\Gamma. For sufficiently small cΓhc_\Gamma h, the chord construction satisfies

δh≤cΓh2,θh≤2cΓh,∣Ωh△Ω∣≤cΓ∣Γ∣h2.(13)\delta_h \le c_\Gamma h^2, \qquad\theta_h \le2c_\Gamma h, \qquad|\Omega_h\triangle\Omega| \le c_\Gamma|\Gamma|h^2. \tag*{(13)}

The surface Jacobian JphJ_{p_h} of php_h satisfies ∣Jph−1∣≲h2|J_{p_h}-1|\lesssim h^2 on each affine piece of Γh\Gamma_h. In three dimensions, these estimates are assumptions on the piecewise planar approximation.

Proof. For a chord of length ℓ≤hT\ell\le h_T with endpoints on Γ\Gamma, consider a second-order expansion of the corresponding arc. This gives a distance of order cΓℓ2c_\Gamma\ell^2 and a normal error of order cΓℓc_\Gamma\ell, from which the first two estimates follow for sufficiently small hh. Integration of the distance estimate over Γh\Gamma_h then gives the bound on the symmetric difference. On each chord, the surface Jacobian differs from one by O(cΓδh+θh2)O(c_\Gamma\delta_h+\theta_h^2), which is O(h2)O(h^2) with a constant depending on the curvature.

Other expansion directions are possible, and [9] allows a field νh\nu_h satisfying ∥νh−n∘p∥=O(1)\|\nu_h-n\circ p\|=O(1). With our choice νh=nh\nu_h=n_h, the offset ϱh\varrho_h can be computed by Newton iteration for Φ(x+rnh(x))=0\Phi(x+r n_h(x))=0, starting at r=0r=0. Since php_h is not the closest point map, we evaluate n∗n^* separately from ∇Φ\nabla\Phi for use in the flux transport (15). All three maps are needed only at quadrature nodes on Γh\Gamma_h.

Transporting a trace and transporting a flux

Write ∂nh:=nh⋅∇\partial_{n_h}:=n_h\cdot\nabla for differentiation in the discrete normal direction. For m≥0m\ge0 and a function ww smooth near x∈Γhx\in\Gamma_h, define

Tm(w)(x):=∑j=0mϱh(x)jj!∂nhjw(x).(14)\mathcal{T}_m(w)(x):=\sum_{j=0}^{m}\frac{\varrho_h(x)^j}{j!}\partial_{n_h}^j w(x). \tag*{(14)}

This is the Taylor polynomial along the segment from xx to ph(x)p_h(x), evaluated at ph(x)p_h(x), with T0=Id⁡\mathcal{T}_0=\operatorname{Id}. The operator transports the trace, as in [6, 9], and we also apply it componentwise to vector fields.

To transport the flux in the Robin condition, we define

Λm(w)(x):=Tm(κ∇w)(x)⋅n∗(x)=κ∑j=0mϱh(x)jj!(∂nhj∇w)(x)⋅n∗(x).(15)\Lambda_m(w)(x):=\mathcal{T}_m(\kappa\nabla w)(x)\cdot n^*(x)=\kappa\sum_{j=0}^{m}\frac{\varrho_h(x)^j}{j!}(\partial_{n_h}^j\nabla w)(x)\cdot n^*(x). \tag*{(15)}

We first expand κ∇w\kappa\nabla w along nhn_h and then take its scalar product with the exact normal at the image point. Since n∗(x)n^*(x) is a fixed vector at each quadrature node, this is equivalent to expanding the scalar function y↦κn∗(x)⋅∇w(y)y\mapsto\kappa n^*(x)\cdot\nabla w(y). In particular, no derivative of the normal is required. For a discrete function, Tm\mathcal{T}_m uses derivatives up to order mm and Λm\Lambda_m uses derivatives up to order m+1m+1, evaluated from the polynomial on the cut element. We therefore take p≥m+1p\ge m+1.

Lemma 3.2 (Transport error). Let w∈Wm+2,∞(U)w\in W^{m+2,\infty}(U), where UU is a neighbourhood of the segment [x,ph(x)][x,p_h(x)]. Then

∣Tm(w)(x)−w(ph(x))∣≤δhm+1(m+1)!∥∂nhm+1w∥L∞,(16)|\mathcal{T}_m(w)(x)-w(p_h(x))|\le\frac{\delta_h^{m+1}}{(m+1)!}\|\partial_{n_h}^{m+1}w\|_{L^\infty}, \tag*{(16)}
∣Λm(w)(x)−λ(w)(ph(x))∣≤κδhm+1(m+1)!∥∂nhm+1∇w∥L∞.|\Lambda_m(w)(x)-\lambda(w)(p_h(x))|\le\frac{\kappa\delta_h^{m+1}}{(m+1)!}\|\partial_{n_h}^{m+1}\nabla w\|_{L^\infty}.

Proof. The first estimate follows by applying Taylor’s theorem with remainder to t↦w(x+tnh(x))t \mapsto w(x+tn_h(x)) between 00 and ϱh(x)\varrho_h(x). For the second, apply the same argument to t↦κn∗(x)⋅∇w(x+tnh(x))t \mapsto\kappa n^*(x) \cdot\nabla w(x+tn_h(x)). Its value at t=ϱh(x)t=\varrho_h(x) is λ(w)(ph(x))\lambda(w)(p_h(x)), since n∗(x)n^*(x) is the exact normal at ph(x)p_h(x).

The discrete flux κ∂nhw(x)\kappa\partial_{n_h}w(x) differs from the exact flux at the image point by κ∇w⋅(nh−n∗)+O(δh)\kappa\nabla w\cdot(n_h-n^*)+O(\delta_h), which is generally O(h)O(h) for an affine boundary approximation. A Taylor expansion that retains nhn_h in the scalar product leaves this directional error. By using n∗n^* in Λm\Lambda_m, we retain only the transport error O(δhm+1)O(\delta_h^{m+1}). This construction uses the elementwise polynomial gradient and does not require gradient continuity between elements.

Assumption 3.1 (Extension). The solution uu of (4) and the datum ff extend to a fixed neighbourhood of Ω‾\overline{\Omega}, with u∈Wm+2,∞∩Hp+1u \in W^{m+2,\infty} \cap H^{p+1} there. The extensions satisfy −∇⋅(κ∇u)+cu=f-\nabla\cdot(\kappa\nabla u)+cu=f pointwise in Ωh∪Ω\Omega_h\cup\Omega. The datum gg also extends to a neighbourhood of Γ\Gamma.

This assumption allows us to integrate by parts on Ωh\Omega_h without a volume consistency term. Under the weaker extension assumption in [9], the residual outside Ω\Omega is retained and estimated on the strip Ωh∖Ω\Omega_h\setminus\Omega [9 (3.14), (3.49)]. We use Assumption 3.1, which holds exactly for all test cases in §8, to focus on the transport of the boundary condition.

The method

The form

Let γ>0\gamma>0 be a Nitsche parameter and set

ρ:=γκh,θ:=ββ+ρ,s:=1β+ρ.(17)\rho:= \frac{\gamma\kappa}{h}, \qquad\theta:= \frac{\beta}{\beta+\rho}, \qquad s := \frac{1}{\beta+\rho}. \tag*{(17)}

We use a ghost penalty on the interior faces of the active mesh adjacent to cut elements,

jh(v,w):=∑F∈FhΓ∑j=1pγgκh2j−1⟨[∂nFjv],[∂nFjw]⟩F.(18)j_h(v,w) := \sum_{F\in\mathcal{F}_h^\Gamma}\sum_{j=1}^{p}\gamma_g\kappa h^{2j-1}\left\langle[\partial_{n_F}^{j}v],[\partial_{n_F}^{j}w]\right\rangle_F. \tag*{(18)}

The discrete problem is to find uh∈Vhu_h\in V_h such that Ah(uh,v)=Lh(v)A_h(u_h,v)=L_h(v) for all v∈Vhv\in V_h, where

Ah(u,v):=(κ∇u,∇v)Ωh+(cu,v)Ωh+jh(u,v)−⟨κ∂nhu,v⟩Γh+ρs⟨Λm(u),v⟩Γh+θρ⟨Tm(u),v⟩Γh−θ⟨Tm(u),κ∂nhv⟩Γh−s⟨Λm(u),κ∂nhv⟩Γh.(19)\begin{aligned} A_h(u,v) :={}&(\kappa\nabla u,\nabla v)_{\Omega_h}+(cu,v)_{\Omega_h}+j_h(u,v)-\langle\kappa\partial_{n_h}u,v\rangle_{\Gamma_h}\\ &+\rho s\langle\Lambda_m(u),v\rangle_{\Gamma_h}+\theta\rho\langle T_m(u),v\rangle_{\Gamma_h}\\ &-\theta\langle T_m(u),\kappa\partial_{n_h}v\rangle_{\Gamma_h}-s\langle\Lambda_m(u),\kappa\partial_{n_h}v\rangle_{\Gamma_h}. \tag*{(19)} \end{aligned}
Lh(v):=(f,v)Ωh+ρs⟨g∘ph,v⟩Γh−s⟨g∘ph,κ∂nhv⟩Γh.(20)L_h(v):=(f,v)_{\Omega_h}+\rho s\langle g\circ p_h,v\rangle_{\Gamma_h}-s\langle g\circ p_h,\kappa\partial_{n_h}v\rangle_{\Gamma_h}. \tag*{(20)}

If c=β=0c=\beta=0, both trial and test functions are restricted to the subspace {v∈Vh:(v,1)Ωh=0}\{v\in V_h:(v,1)_{\Omega_h}=0\}. The datum is evaluated at ph(x)∈Γp_h(x)\in\Gamma, while TmT_m and Λm\Lambda_m approximate the corresponding solution trace and flux from the polynomial on the cut element.

The term −⟨κ∂nhu,v⟩Γh-\langle\kappa\partial_{n_h}u,v\rangle_{\Gamma_h} comes from integration by parts on Ωh\Omega_h and therefore uses the normal nhn_h of that domain. It is present for every β\beta, while the terms involving TmT_m and Λm\Lambda_m impose the transported boundary condition through Lemma 2.1. The ghost penalty provides uniform control of the active mesh norm with respect to the boundary cuts.

The two limits and the Taylor order.

Let g=βgDg = \beta g_D and keep gDg_D fixed as β→∞\beta\to\infty. Then θ→1\theta\to1, s→0s \to0, ρsg→ρgD\rho s g \to\rho g_D, and sg→gDs g \to g_D. The method tends to Nitsche’s method with the boundary value correction of [9], acting on the trial function only. The flux transport Λm\Lambda_m disappears because the Dirichlet condition prescribes only the trace.

At β=0\beta= 0, we have θ=0\theta= 0, s=1/ρs = 1/\rho, and ρs=1\rho s = 1. Equations (19)–(20) become

(κ∇u,∇v)Ωh+(cu,v)Ωh+jh(u,v)−⟨κ∂nhu−Λm(u),v⟩Γh−1ρ⟨Λm(u),κ∂nhv⟩Γh=(f,v)Ωh+⟨g∘ph,v⟩Γh−1ρ⟨g∘ph,κ∂nhv⟩Γh.(\kappa\nabla u,\nabla v)_{\Omega_h} + (cu,v)_{\Omega_h} + j_h(u,v) - \langle\kappa\partial_{n_h}u - \Lambda_m(u),v\rangle_{\Gamma_h} - \frac{1}{\rho}\langle\Lambda_m(u),\kappa\partial_{n_h}v\rangle_{\Gamma_h} = (f,v)_{\Omega_h} + \langle g\circ p_h,v\rangle_{\Gamma_h} - \frac{1}{\rho}\langle g\circ p_h,\kappa\partial_{n_h}v\rangle_{\Gamma_h}.

The two terms with coefficient 1/ρ1/\rho form a Barbosa–Hughes stabilisation whose residual vanishes at the exact solution on the exact geometry, as in [18 (2.14)]. On Γh\Gamma_h, the residual instead has the Taylor transport error from Lemma 3.2. Dropping these two terms gives a natural Neumann formulation with a flux correction: the flux from integration by parts is replaced by Λm(u)\Lambda_m(u) before the datum is imposed.

Remark 4.1 (Where the Robin coefficient acts). The trace correction TmT_m is multiplied by θ\theta, while the flux correction Λm\Lambda_m is multiplied by ρs=1−θ\rho s = 1-\theta or by s=(1−θ)/ρs = (1-\theta)/\rho. Thus the Dirichlet limit uses only the trace correction and the Neumann case only the flux correction, with both entering for finite positive β\beta.

The uniform consistency bound in Corollary 5.1 is O(h−1/2δhm+1)O(h^{-1/2}\delta_h^{m+1}), which becomes O(h2m+3/2)O(h^{2m+3/2}) under δh≲cΓh2\delta_h \lesssim c_{\Gamma}h^2. It is therefore bounded by the approximation error O(hp)O(h^p) when

2m+32≥p,i.e.m=0 (p=1),m=1 (p=2,3),m=2 (p=4,5).(21)2m + \frac{3}{2} \ge p, \qquad\text{i.e.}\qquad m = 0\ (p = 1),\qquad m = 1\ (p = 2,3),\qquad m = 2\ (p = 4,5). \tag*{(21)}

This is the rule in [9 Tab. 1] for the Dirichlet problem, and also suffices for the expected L2L^2 estimate (Remark 5.2 and [9 Tab. 2]). The estimate can be sharpened when β\beta is fixed and finite, as shown below.

Analysis.

Norms and trace inequalities.

We use the energy norm

∥∣v∣∥h2:=∥κ∇v∥Ωh2+c∥v∥Ωh2+jh(v,v)+θρ∥v∥Γh2.(22)\lVert\lvert v\rvert\rVert_h^2 := \lVert\sqrt{\kappa}\nabla v\rVert_{\Omega_h}^2 + c\lVert v\rVert_{\Omega_h}^2 + j_h(v,v) + \theta\rho\lVert v\rVert_{\Gamma_h}^2. \tag*{(22)}

The boundary weight is ρ=γκ/h\rho= \gamma\kappa/h at the Dirichlet limit and vanishes at the Neumann limit. For fixed 0<β<∞0 < \beta< \infty, θρ=βρ/(β+ρ)→β\theta\rho= \beta\rho/(\beta+\rho) \to\beta as h→0h \to0, so the norm approaches the natural Robin energy norm. When c=β=0c = \beta= 0, we restrict the discrete space to functions with zero mean on Ωh\Omega_h.

The ghost penalty controls the gradient on the full active elements. Combined with the polynomial inverse inequality, this gives

κ2∥∇v∥Γh2≲CIκh(∥κ∇v∥Ωh2+jh(v,v)),\kappa^2\lVert\nabla v\rVert_{\Gamma_h}^2 \lesssim\frac{C_I\kappa}{h}\left(\lVert\sqrt{\kappa}\nabla v\rVert_{\Omega_h}^2 + j_h(v,v)\right),
∥v∥Γh2≲h−1∥v∥Ωh2+h(∥∇v∥Ωh2+κ−1jh(v,v)),v∈Vh.(23)\lVert v\rVert_{\Gamma_h}^2 \lesssim h^{-1}\lVert v\rVert_{\Omega_h}^2 + h\left(\lVert\nabla v\rVert_{\Omega_h}^2 + \kappa^{-1}j_h(v,v)\right), \qquad v \in V_h. \tag*{(23)}

The constants are independent of the cut position, and analogous estimates hold on active-mesh faces. These are standard consequences of the ghost penalty, for which we refer to [7, 9].

For the trace of the function itself, we also use

∥v∥L2(Γh)≤Ctr∥v∥H1(Ωh),v∈H1(Ωh).(24)\lVert v\rVert_{L^2(\Gamma_h)} \le C_{\mathrm{tr}} \lVert v\rVert_{H^1(\Omega_h)}, \qquad v \in H^1(\Omega_h). \tag*{(24)}

The constant CtrC_{\mathrm{tr}} is independent of hh because the domains Ωh\Omega_h have a uniform Lipschitz character. For a resolved chordal approximation of a C2C^2 curve, neighbouring segments meet at an angle π+O(h)\pi+ O(h), which allows local boundary charts to be chosen with uniform bounds. In three dimensions we assume the same uniform domain property.

We make the Poincaré bound explicit:

κ∥v∥Ωh2≤Λ2∥ ⁣∣v∣ ⁣∥h2,v∈Vh.(25)\kappa\lVert v\rVert_{\Omega_h}^2 \le\Lambda^2 \lVert\!\lvert v\rvert\!\rVert_h^2, \qquad v \in V_h. \tag*{(25)}

We assume that Λ\Lambda is independent of hh and make its role in the analysis explicit. If c≥c0>0c \ge c_0 > 0, we may take Λ2=κ/c0\Lambda^2 = \kappa/c_0. For β≥β0>0\beta\ge\beta_0 > 0, a Poincaré inequality with a boundary term also gives a uniform bound for sufficiently small hh. For c=β=0c = \beta= 0, the zero-mean restriction gives the usual Poincaré bound. Since the constants below may depend on Λ\Lambda, uniformity in β\beta requires a uniform bound on Λ\Lambda. In particular, such a bound does not hold on the full space as β↓0\beta\downarrow0 when c=0c = 0.

Combining (5.3) and (5.4) gives

∥v∥Γh≤CΛκ−1/2∥ ⁣∣v∣ ⁣∥h,CΛ:=Ctr(1+Λ2)1/2.(26)\lVert v\rVert_{\Gamma_h} \le C_{\Lambda}\kappa^{-1/2} \lVert\!\lvert v\rvert\!\rVert_h, \qquad C_{\Lambda} := C_{\mathrm{tr}}(1+\Lambda^2)^{1/2}. \tag*{(26)}

This estimate has no negative power of hh and will be used in both the stability and consistency arguments.

Lemma 5.1 (Taylor operators on discrete functions). Let 0≤m≤p−10 \le m \le p-1 and assume δh/h\delta_h/h is bounded. Then

∥Tm(v)−v∥Γh≲δhκh∥ ⁣∣v∣ ⁣∥h,\lVert T_m(v)-v\rVert_{\Gamma_h} \lesssim\frac{\delta_h}{\sqrt{\kappa h}}\lVert\!\lvert v\rvert\!\rVert_h,
∥Λm(v)−κn∗⋅∇v∥Γh≲δhh(κh)1/2∥ ⁣∣v∣ ⁣∥h.(27)\lVert\Lambda_m(v)-\kappa n^* \cdot\nabla v\rVert_{\Gamma_h} \lesssim\frac{\delta_h}{h}\left(\frac{\kappa}{h}\right)^{1/2}\lVert\!\lvert v\rvert\!\rVert_h. \tag*{(27)}

Both left-hand sides vanish when m=0m=0.

Proof. On each full active element, we apply the polynomial inverse inequality and use the ghost penalty to control the gradient. The jjth trace and flux terms are then bounded, respectively, by

Cδhjh−(j−1)(κh)−1/2∥ ⁣∣v∣ ⁣∥hC\delta_h^j h^{-(j-1)}(\kappa h)^{-1/2}\lVert\!\lvert v\rvert\!\rVert_h

and

C(δh/h)j(κ/h)1/2∥ ⁣∣v∣ ⁣∥h.C(\delta_h/h)^j(\kappa/h)^{1/2}\lVert\!\lvert v\rvert\!\rVert_h.

Summing over j=1,…,mj=1,\ldots,m proves the result. □\square

The normal mismatch

Integration by parts uses the normal nhn_h, whereas the boundary condition uses n∗n^*. For m=0m=0, the resulting flux difference is

E(v):=κ(n∗−nh)⋅∇v,∣E(v)∣≤θhκ∣∇v∣.(28)E(v) := \kappa(n^*-n_h)\cdot\nabla v, \qquad|E(v)| \le\theta_h\kappa|\nabla v|. \tag*{(28)}

The term ⟨E(v),v⟩Γh\langle E(v),v\rangle_{\Gamma_h} occurs in the stability estimate.

Lemma 5.2 (Bound for the normal mismatch). Under (5.4),

∣⟨E(v),v⟩Γh∣≤CCΓh1/2∥ ⁣∣v∣ ⁣∥h2,v∈Vh.(29)|\langle E(v),v\rangle_{\Gamma_h}| \le C_C\Gamma_h^{1/2}\lVert\!\lvert v\rvert\!\rVert_h^2, \qquad v \in V_h. \tag*{(29)}

The constant depends on CIC_I and CΛC_{\Lambda} but is independent of hh and of the cut position.

Proof. By the Cauchy–Schwarz inequality,

∣⟨E(v),v⟩Γh∣≤θhκ∥∇v∥Γh∥v∥Γh.\left|\langle E(v),v\rangle_{\Gamma_h}\right| \le\theta_h\kappa\lVert\nabla v\rVert_{\Gamma_h}\lVert v\rVert_{\Gamma_h}.

We apply the discrete trace estimate (23) to the gradient and the continuous estimate (26) to the function itself. Together they give

∣⟨E(v),v⟩Γh∣≤CI1/2CΛθhh−1/2∥ ⁣∥v∥ ⁣∥h2.\left|\langle E(v),v\rangle_{\Gamma_h}\right| \le C_I^{1/2}C_\Lambda\theta_h h^{-1/2}\lVert\!\lVert v\rVert\!\rVert_h^2.

The result follows from θh≤2cΓh\theta_h\le2c_{\Gamma_h}.

Remark 5.1 (The Neumann case). The volume terms and (5.4) control the H1H^1 norm without a boundary penalty. Hence (5.5) and the argument above also apply when β=0\beta=0. If c=0c=0, both trial and test functions must satisfy the zero-mean restriction specified above. The corrected form need not annihilate constant test functions, so a quotient-space statement alone is not enough.

Stability

Theorem 5.1 (Stability). Let γ\gamma be sufficiently large and let 0≤m≤p−10\le m\le p-1. Under (25), there are h0>0h_0>0 and C>0C>0 such that

Ah(v,v)≥C∥ ⁣∥v∥ ⁣∥h2,v∈Vh,h≤h0.(30)A_h(v,v)\ge C\lVert\!\lVert v\rVert\!\rVert_h^2,\qquad v\in V_h,\quad h\le h_0. \tag*{(30)}

The constants may depend on γ\gamma, the mesh regularity, the geometry and Λ\Lambda, but are independent of the cut position. They are also independent of β\beta whenever Λ\Lambda is uniformly bounded.

Proof. For m=0m=0, we have Tm(v)=v\mathcal{T}_m(v)=v and Λm(v)=κ∂nhv+E(v)\Lambda_m(v)=\kappa\partial_{n_h}v+E(v). Using ρs=1−θ\rho s=1-\theta, we can therefore write the form as

Ah(v,v)=∥κ∇v∥Ωh2+c∥v∥Ωh2+jh(v,v)+θρ∥v∥Γh2−2θ⟨κ∂nhv,v⟩Γh−s∥κ∂nhv∥Γh2+R(v).(31)\begin{aligned} A_h(v,v)={}&\lVert\sqrt{\kappa}\nabla v\rVert_{\Omega_h}^2+c\lVert v\rVert_{\Omega_h}^2+j_h(v,v)+\theta\rho\lVert v\rVert_{\Gamma_h}^2\\ &-2\theta\langle\kappa\partial_{n_h}v,v\rangle_{\Gamma_h}-s\lVert\kappa\partial_{n_h}v\rVert_{\Gamma_h}^2+\mathcal{R}(v). \tag*{(31)} \end{aligned}

where

R(v)=(1−θ)⟨E(v),v⟩Γh−s⟨E(v),κ∂nhv⟩Γh.\mathcal{R}(v)=(1-\theta)\langle E(v),v\rangle_{\Gamma_h}-s\langle E(v),\kappa\partial_{n_h}v\rangle_{\Gamma_h}.

Young’s inequality gives

2θ∣⟨κ∂nhv,v⟩Γh∣≤2θρ∥κ∂nhv∥Γh2+θρ2∥v∥Γh2.2\theta\left|\langle\kappa\partial_{n_h}v,v\rangle_{\Gamma_h}\right| \le\frac{2\theta}{\rho}\lVert\kappa\partial_{n_h}v\rVert_{\Gamma_h}^2+\frac{\theta\rho}{2}\lVert v\rVert_{\Gamma_h}^2.

Since (2θ/ρ)+s≤2/ρ(2\theta/\rho)+s\le2/\rho, the first estimate in (23) absorbs the flux terms when γ>4CI\gamma>4C_I. The remaining positive terms control 12∥ ⁣∥v∥ ⁣∥h2\frac{1}{2}\lVert\!\lVert v\rVert\!\rVert_h^2.

By Lemma 5.2, the first term in R\mathcal{R} is bounded by CcΓhh1/2∥ ⁣∥v∥ ⁣∥h2C c_{\Gamma_h}h^{1/2}\lVert\!\lVert v\rVert\!\rVert_h^2. The bound s≤h/(γκ)s\le h/(\gamma\kappa) gives CcΓhh∥ ⁣∥v∥ ⁣∥h2C c_{\Gamma_h}h\lVert\!\lVert v\rVert\!\rVert_h^2 for the second term, so both can be absorbed for small hh.

For m≥1m\ge1, Lemma 5.1 bounds the trace-correction terms and the flux-flux term by C(δh/h)∥ ⁣∥v∥ ⁣∥h2C(\delta_h/h)\lVert\!\lVert v\rVert\!\rVert_h^2. Applying (5.5) to the remaining flux-correction term gives the bound CCΛδhh−3/2∥ ⁣∥v∥ ⁣∥h2C C_\Lambda\delta_h h^{-3/2}\lVert\!\lVert v\rVert\!\rVert_h^2. Since δh≲h2\delta_h\lesssim h^2, these terms can also be absorbed. □\square

Consistency

Define the transported boundary residual by

R(w):=βTm(w)+Λm(w)−g∘phon Γh.(32)R(w):=\beta\mathcal{T}_m(w)+\Lambda_m(w)-g\circ p_h\quad\text{on }\Gamma_h. \tag*{(32)}

Theorem 5.2 (Consistency). For 0≤β<∞0 \le\beta< \infty, let uu solve (4) and satisfy Assumption 3.1. Then

Ah(u,v)−Lh(v)=ρs⟨R(u),v⟩Γh−s⟨R(u),κ∂nhv⟩Γh,v∈Vh,(33)A_h(u,v)-L_h(v)=\rho s\langle R(u),v\rangle_{\Gamma_h}-s\langle R(u),\kappa\partial_{n_h}v\rangle_{\Gamma_h},\qquad v\in V_h, \tag*{(33)}

and

∥R(u)∥Γh≲δhm+1(m+1)!(β∥∂nhm+1u∥L∞+κ∥∂nhm+1∇u∥L∞)∣Γh∣1/2.(34)\lVert R(u)\rVert_{\Gamma_h}\lesssim\frac{\delta_h^{m+1}}{(m+1)!}\left(\beta\lVert\partial_{n_h}^{m+1}u\rVert_{L^\infty}+\kappa\lVert\partial_{n_h}^{m+1}\nabla u\rVert_{L^\infty}\right)|\Gamma_h|^{1/2}. \tag*{(34)}

Proof. The extension assumption allows integration by parts on Ωh\Omega_h, giving

(κ∇u,∇v)Ωh+(cu,v)Ωh−⟨κ∂nhu,v⟩Γh=(f,v)Ωh.(\kappa\nabla u,\nabla v)_{\Omega_h}+(cu,v)_{\Omega_h}-\langle\kappa\partial_{n_h}u,v\rangle_{\Gamma_h}=(f,v)_{\Omega_h}.

Since the extension is smooth across mesh faces, jh(u,v)=0j_h(u,v)=0. Subtracting the load from (19) therefore leaves

ρs⟨Λm(u)−g∘ph,v⟩Γh+θρ⟨Tm(u),v⟩Γh−⟨θTm(u)+sΛm(u)−sg∘ph,κ∂nhv⟩Γh.\rho s\langle\Lambda_m(u)-g\circ p_h,v\rangle_{\Gamma_h}+\theta\rho\langle\mathcal{T}_m(u),v\rangle_{\Gamma_h}-\langle\theta\mathcal{T}_m(u)+s\Lambda_m(u)-sg\circ p_h,\kappa\partial_{n_h}v\rangle_{\Gamma_h}.

Using θ=βs\theta=\beta s gives (33), while the boundary condition at ph(x)p_h(x) gives

R(u)=β(Tm(u)−u∘ph)+Λm(u)−λ(u)∘ph.R(u)=\beta\left(\mathcal{T}_m(u)-u\circ p_h\right)+\Lambda_m(u)-\lambda(u)\circ p_h.

The residual bound follows by applying Lemma 3.2 to these two remainders. □\square

Corollary 5.1 (Weighted consistency). Let CuC_u denote the Wm+2,∞W^{m+2,\infty} norm of the extension of uu and set CΛ:=Ctr(1+Λ2)1/2C_\Lambda:=C_{\mathrm{tr}}(1+\Lambda^2)^{1/2}. Then

∣Ah(u,v)−Lh(v)∣≲δhm+1Cuκ[(θh)1/2+(1−θ)(CΛ+h1/2)]∥ ⁣∥v∥ ⁣∥h.(35)\lvert A_h(u,v)-L_h(v)\rvert\lesssim\delta_h^{m+1}C_u\sqrt{\kappa}\left[\left(\frac{\theta}{h}\right)^{1/2}+(1-\theta)(C_\Lambda+h^{1/2})\right]\lVert\!\lVert v\rVert\!\rVert_h. \tag*{(35)}

The constant is independent of hh, β\beta, and the position of the cut, with the dependence on the Poincaré estimate displayed through CΛC_\Lambda. For each fixed finite β\beta, the right-hand side is O(δhm+1)∥ ⁣∥v∥ ⁣∥hO(\delta_h^{m+1})\lVert\!\lVert v\rVert\!\rVert_h. At the Dirichlet limit it is O(h−1/2δhm+1)∥ ⁣∥v∥ ⁣∥hO(h^{-1/2}\delta_h^{m+1})\lVert\!\lVert v\rVert\!\rVert_h.

Proof. Write

rT:=Tm(u)−u∘ph,rΛ:=Λm(u)−λ(u)∘ph.r_T:=\mathcal{T}_m(u)-u\circ p_h,\qquad r_\Lambda:=\Lambda_m(u)-\lambda(u)\circ p_h.

This gives the decomposition R(u)=βrT+rΛR(u)=\beta r_T+r_\Lambda, with the transport estimates ∥rT∥Γh≲δhm+1Cu\lVert r_T\rVert_{\Gamma_h}\lesssim\delta_h^{m+1}C_u and ∥rΛ∥Γh≲κδhm+1Cu\lVert r_\Lambda\rVert_{\Gamma_h}\lesssim\kappa\delta_h^{m+1}C_u. We estimate the two residuals separately, starting with the trace residual. The boundary term in the energy norm gives

θρ∥v∥Γh≤(θρ)1/2∥ ⁣∥v∥ ⁣∥h.\theta\rho\lVert v\rVert_{\Gamma_h}\le(\theta\rho)^{1/2}\lVert\!\lVert v\rVert\!\rVert_h.

This inequality also holds at θ=0\theta=0, when both sides vanish. The discrete flux estimate gives θ∥κ∂nhv∥Γh≲θ(κ/h)1/2∥ ⁣∥v∥ ⁣∥h\theta\lVert\kappa\partial_{n_h}v\rVert_{\Gamma_h}\lesssim\theta(\kappa/h)^{1/2}\lVert\!\lVert v\rVert\!\rVert_h. Since βρs=θρ\beta\rho s=\theta\rho, βs=θ\beta s=\theta, and θ≤θ\theta\le\sqrt{\theta}, these estimates bound the part containing rTr_T by

Cδhm+1Cu(κθ/h)1/2∥ ⁣∥v∥ ⁣∥h.C\delta_h^{m+1}C_u(\kappa\theta/h)^{1/2}\lVert\!\lVert v\rVert\!\rVert_h.

For the flux residual, the continuous trace estimate and (25) give ∥v∥Γh≤CΛκ−1/2∥ ⁣∥v∥ ⁣∥h\lVert v\rVert_{\Gamma_h}\le C_\Lambda\kappa^{-1/2}\lVert\!\lVert v\rVert\!\rVert_h. Combining this with s=(1−θ)h/(γκ)s=(1-\theta)h/(\gamma\kappa) yields

ρs∥v∥Γh+s∥κ∂nhv∥Γh≲(1−θ)κ−1/2(CΛ+h1/2)∥ ⁣∥v∥ ⁣∥h.\rho s\lVert v\rVert_{\Gamma_h}+s\lVert\kappa\partial_{n_h}v\rVert_{\Gamma_h}\lesssim(1-\theta)\kappa^{-1/2}(C_\Lambda+h^{1/2})\lVert\!\lVert v\rVert\!\rVert_h.

Multiplication by ∥rΛ∥Γh\lVert r_\Lambda\rVert_{\Gamma_h} proves (5.14). Finally,

θh=ββh+γκ≤βγκ(β<∞).\frac{\theta}{h}=\frac{\beta}{\beta h+\gamma\kappa}\le\frac{\beta}{\gamma\kappa}\qquad(\beta<\infty).

This proves the estimate for fixed finite β\beta, with constants that may depend on β\beta and Λ.□\Lambda. \quad\square

The energy error.

Theorem 5.3 (Energy error estimate). Under the hypotheses of Theorem 5.1 and Assumption 3.1, let m≤p−1m \le p-1 and h≤h0h \le h_0. Then

∥ ⁣∥u−uh∥ ⁣∥h≲hp∥u∥Hp+1+δhm+1Cuκ[(θh)1/2+(1−θ)(CΛ+h1/2)].(36)\lVert\!\lVert u-u_h\rVert\!\rVert_h \lesssim h^p\lVert u\rVert_{H^{p+1}}+\delta_h^{m+1}C_u\sqrt{\kappa}\left[\left(\frac{\theta}{h}\right)^{1/2}+(1-\theta)(C_\Lambda+h^{1/2})\right]. \tag*{(36)}

Constants are independent of hh, β\beta, and the position of the cut when the Poincaré constant Λ\Lambda is uniformly bounded. For fixed 0≤β<∞0 \le\beta< \infty this implies

∥ ⁣∥u−uh∥ ⁣∥h≲βhp∥u∥Hp+1+δhm+1Cu.(37)\lVert\!\lVert u-u_h\rVert\!\rVert_h \lesssim_\beta h^p\lVert u\rVert_{H^{p+1}}+\delta_h^{m+1}C_u. \tag*{(37)}

Thus 2m+2≥p2m+2 \ge p gives the optimal energy order for fixed finite β\beta, while 2m+32≥p2m+\frac{3}{2} \ge p remains sufficient uniformly up to the Dirichlet limit.

Proof. Let IhuI_hu be an interpolant on the active mesh and set η=u−Ihu\eta=u-I_hu and eh=Ihu−uhe_h=I_hu-u_h. Standard interpolation and trace estimates on the full elements give

∥Dkη∥Γh≲hp+1/2−k∥u∥Hp+1(0≤k≤p).\lVert D^k\eta\rVert_{\Gamma_h}\lesssim h^{p+1/2-k}\lVert u\rVert_{H^{p+1}}\qquad(0\le k\le p).

They also give ∥ ⁣∥η∥ ⁣∥h≲hp∥u∥Hp+1\lVert\!\lVert\eta\rVert\!\rVert_h\lesssim h^p\lVert u\rVert_{H^{p+1}}, with the extension used in all these norms. For a zero-mean space, we adjust the interpolant by a constant without changing the approximation order.

We split Ah=Ah0+BhA_h=A_h^0+B_h as in (40) and first consider Ah0(η,v)A_h^0(\eta,v). The volume and ghost-penalty terms have the usual interpolation bounds, while the boundary terms are controlled by

θ∥v∥Γh≤(θ/ρ)1/2∥ ⁣∥v∥ ⁣∥h,∥κ∂nhv∥Γh≲(κ/h)1/2∥ ⁣∥v∥ ⁣∥h,s≤1/ρ.\theta\lVert v\rVert_{\Gamma_h}\le(\theta/\rho)^{1/2}\lVert\!\lVert v\rVert\!\rVert_h,\qquad \lVert\kappa\partial_{n_h}v\rVert_{\Gamma_h}\lesssim(\kappa/h)^{1/2}\lVert\!\lVert v\rVert\!\rVert_h,\qquad s\le1/\rho.

The first estimate is interpreted as zero at θ=0\theta=0. Together with the interpolation bounds, these estimates give

∣Ah0(η,v)∣≲hp∥u∥Hp+1∥ ⁣∥v∥ ⁣∥h.\lvert A_h^0(\eta,v)\rvert\lesssim h^p\lVert u\rVert_{H^{p+1}}\lVert\!\lVert v\rVert\!\rVert_h.

In particular, no inverse power of θ\theta is needed.

To estimate BhB_h, write DT(η)=Tm(η)−ηD_T(\eta)=\mathcal{T}_m(\eta)-\eta and DΛ(η)=Λm(η)−κ∂nhηD_\Lambda(\eta)=\Lambda_m(\eta)-\kappa\partial_{n_h}\eta. Since m≤p−1m\le p-1 and δh≲h2\delta_h\lesssim h^2, the same trace estimates imply

∥DT(η)∥Γh≲δhhp−1/2∥u∥Hp+1,\lVert D_T(\eta)\rVert_{\Gamma_h}\lesssim\delta_hh^{p-1/2}\lVert u\rVert_{H^{p+1}},
∥DΛ(η)∥Γh≲κ(θhhp−1/2+δhhp−3/2)∥u∥Hp+1≲κhp+1/2∥u∥Hp+1.\lVert D_\Lambda(\eta)\rVert_{\Gamma_h}\lesssim\kappa\left(\theta_hh^{p-1/2}+\delta_hh^{p-3/2}\right)\lVert u\rVert_{H^{p+1}}\lesssim\kappa h^{p+1/2}\lVert u\rVert_{H^{p+1}}.

For m=0m=0, the first expression vanishes and the δh\delta_h term in the second expression is absent. We apply the continuous trace estimate to vv in the first term of (41) and use the boundary penalty and the discrete flux estimate in the other three terms. This gives

∣Bh(η,v)∣≲(hp+1/2+hp+1)∥u∥Hp+1∥ ⁣∥v∥ ⁣∥h.\lvert B_h(\eta,v)\rvert\lesssim\left(h^{p+1/2}+h^{p+1}\right)\lVert u\rVert_{H^{p+1}}\lVert\!\lVert v\rVert\!\rVert_h.

Its constant depends on the stated geometric and Poincaré constants. Consequently, ∣Ah(η,v)∣≲hp∥u∥Hp+1∥ ⁣∥v∥ ⁣∥h\lvert A_h(\eta,v)\rvert\lesssim h^p\lVert u\rVert_{H^{p+1}}\lVert\!\lVert v\rVert\!\rVert_h.

Coercivity and the discrete equation now give

∥ ⁣∥eh∥ ⁣∥h2≲∣Ah(η,eh)∣+∣Ah(u,eh)−Lh(eh)∣.\lVert\!\lVert e_h\rVert\!\rVert_h^2\lesssim\lvert A_h(\eta,e_h)\rvert+\lvert A_h(u,e_h)-L_h(e_h)\rvert.

Applying Corollary 5.1 and then the triangle inequality proves (5.15). The last assertions follow from δh≲h2\delta_h\lesssim h^2 and θ/h≤β/(γκ)\theta/h\le\beta/(\gamma\kappa) for finite β\beta. □\square

The L2L^2 error.

Remark 5.2 (Expected L2L^2 estimate). The expected estimate is

∥u−uh∥Ωh≲hp+1∥u∥Hp+1+δhm+1Cu.(38)\lVert u-u_h\rVert_{\Omega_h} \lesssim h^{p+1}\lVert u\rVert_{H^{p+1}}+\delta_h^{m+1}C_u. \tag*{(38)}

For the Dirichlet problem, the corresponding estimate is proved in [9 Thm. 3.2] under the assumptions made there. We do not prove (38) for finite β\beta in this paper. Since the trial-side corrections make the form non-symmetric, a duality proof requires a separate adjoint-consistency estimate. It must also control the geometric remainders using the regularity available for the dual solution, usually H2H^2. The Wm+2,∞W^{m+2,\infty} remainder bound above does not provide this directly.

These points remain open after the improvement of the energy estimate. For example, p=2p=2 and m=0m=0 give order two in energy at fixed finite β\beta, but the geometric term in (38) would still be of order two. An additional Taylor term is therefore needed to obtain the expected third order in L2L^2, in agreement with the numerical results.

A symmetric preconditioner.

Setting m=0m=0 and replacing n∗\mathbf{n}^* by nh\mathbf{n}_h in (4.3) gives the symmetric form

Ah0(u,v):=(κ∇u,∇v)Ωh+(cu,v)Ωh+jh(u,v)−θ(⟨κ∂nhu,v⟩Γh+⟨u,κ∂nhv⟩Γh)+θρ⟨u,v⟩Γh−s⟨κ∂nhu,κ∂nhv⟩Γh.(39)A_h^0(u,v) := (\kappa\nabla u,\nabla v)_{\Omega_h}+(cu,v)_{\Omega_h}+j_h(u,v) -\theta\left(\langle\kappa\partial_{\mathbf{n}_h}u,v\rangle_{\Gamma_h}+\langle u,\kappa\partial_{\mathbf{n}_h}v\rangle_{\Gamma_h}\right) +\theta\rho\langle u,v\rangle_{\Gamma_h}-s\langle\kappa\partial_{\mathbf{n}_h}u,\kappa\partial_{\mathbf{n}_h}v\rangle_{\Gamma_h}. \tag*{(39)}

This is the uncorrected Juntunen–Stenberg form on Ωh\Omega_h, with the ghost penalty added. Define

Ah=Ah0+Bh,DT(u):=Tm(u)−u,DΛ(u):=Λm(u)−κ∂nhu.(40)A_h=A_h^0+B_h,\qquad D_T(u):=T_m(u)-u,\qquad D_\Lambda(u):=\Lambda_m(u)-\kappa\partial_{\mathbf{n}_h}u. \tag*{(40)}

Then

Bh(u,v)=ρs⟨DΛ(u),v⟩Γh+θρ⟨DT(u),v⟩Γh−θ⟨DT(u),κ∂nhv⟩Γh−s⟨DΛ(u),κ∂nhv⟩Γh.(41)B_h(u,v)=\rho s\langle D_\Lambda(u),v\rangle_{\Gamma_h} +\theta\rho\langle D_T(u),v\rangle_{\Gamma_h} -\theta\langle D_T(u),\kappa\partial_{\mathbf{n}_h}v\rangle_{\Gamma_h} -s\langle D_\Lambda(u),\kappa\partial_{\mathbf{n}_h}v\rangle_{\Gamma_h}. \tag*{(41)}

Proposition 5.1 (A small perturbation of a symmetric form). Let γ>4CI\gamma>4C_I and 0≤m≤p−10\leq m\leq p-1. Under the mesh assumptions, the symmetric form satisfies

Ah0(v,v)≥C0∥∣v∣∥h2,∣Ah0(u,v)∣≤C1∥∣u∣∥h∥∣v∣∥h.(42)A_h^0(v,v)\geq C_0\lVert\lvert v\rvert\rVert_h^2,\qquad\lvert A_h^0(u,v)\rvert\leq C_1\lVert\lvert u\rvert\rVert_h\lVert\lvert v\rvert\rVert_h. \tag*{(42)}

Under (5.4),

∣Bh(u,v)∣≤CBCΓh1/2∥∣u∣∥h∥∣v∣∥h.(43)\lvert B_h(u,v)\rvert\leq C_B C_\Gamma h^{1/2}\lVert\lvert u\rvert\rVert_h\lVert\lvert v\rvert\rVert_h. \tag*{(43)}

The constants are independent of the cut position, although CBC_B may depend on Λ\Lambda.

Proof. Coercivity follows from the argument preceding R\mathcal{R} in (31), and continuity follows from (23) and the bounds θ≤1\theta\leq1 and s≤1/ρs\leq1/\rho. By (27) and (28),

∥DT(u)∥Γh≲δhκh∥∣u∣∥h,∥DΛ(u)∥Γh≲(θh+δhh)(κh)1/2∥∣u∣∥h.\lVert D_T(u)\rVert_{\Gamma_h}\lesssim\frac{\delta_h}{\sqrt{\kappa h}}\lVert\lvert u\rvert\rVert_h,\qquad \lVert D_\Lambda(u)\rVert_{\Gamma_h}\lesssim\left(\theta_h+\frac{\delta_h}{h}\right)\left(\frac{\kappa}{h}\right)^{1/2}\lVert\lvert u\rvert\rVert_h.

Substitution in (41) bounds the two trace-correction terms by C(δh/h)∥∣u∣∥h∥∣v∣∥hC(\delta_h/h)\lVert\lvert u\rvert\rVert_h\lVert\lvert v\rvert\rVert_h and the flux-flux term by C(θh+δh/h)∥∣u∣∥h∥∣v∣∥hC(\theta_h+\delta_h/h)\lVert\lvert u\rvert\rVert_h\lVert\lvert v\rvert\rVert_h. The first term is bounded by CCΛ(θh+δh/h)h−1/2∥∣u∣∥h∥∣v∣∥hCC_\Lambda(\theta_h+\delta_h/h)h^{-1/2}\lVert\lvert u\rvert\rVert_h\lVert\lvert v\rvert\rVert_h using (26). The geometric estimates now give (43). □\square

Corollary 5.2 (Preconditioning by the uncorrected form). Let q(h):=CBCΓh1/2/C0<1q(h) := C_B C_\Gamma h^{1/2}/C_0 < 1. In the inner product induced by Ah0A_h^0,

∥I−(Ah0)−1Ah∥Ah0≤q(h).(44)\left\|I-(A_h^0)^{-1}A_h\right\|_{A_h^0} \le q(h). \tag*{(44)}

Thus preconditioned Richardson iteration converges with contraction factor at most q(h)q(h), and the operator condition number in this norm is at most (1+q)/(1−q)(1+q)/(1-q).

Proof. For a discrete function xx,

∥(Ah0)−1Bhx∥Ah0=sup⁡y≠0∣Bh(x,y)∣∥y∥Ah0≤CBCΓh1/2C0∥x∥Ah0.\left\|(A_h^0)^{-1}B_hx\right\|_{A_h^0} =\sup_{y\ne0}\frac{|B_h(x,y)|}{\|y\|_{A_h^0}} \le\frac{C_B C_\Gamma h^{1/2}}{C_0}\|x\|_{A_h^0}.

This proves the operator bound. When the preconditioned operator acts on a unit vector, the resulting norm lies between 1−q1-q and 1+q1+q, which gives the condition-number estimate. □

The symmetric matrix uses the same quadrature points as the corrected method but does not require ϱh\varrho_h or n∗n^*. The correction can then be applied through matrix–vector products with BhB_h, as illustrated by the numerical results in § 8.3. An analogous splitting is available for the interface form, subject to the phasewise Poincaré bounds used in § 6.

The interface problem with imperfect bonding.

The problem.

Let Γ\Gamma be a closed C2C^2 interface in a fixed domain Ω\Omega, dividing Ω\Omega into an enclosed subdomain Ω1\Omega_1 and a surrounding subdomain Ω2\Omega_2. We take the normal n\mathbf{n} to point from Ω1\Omega_1 into Ω2\Omega_2 and let κ1,κ2>0\kappa_1,\kappa_2>0. For a function vv defined on both sides, set [v]:=v1−v2[v] := v_1-v_2. Given weights w1+w2=1w_1+w_2=1, we also define {v}:=w1v1+w2v2\{v\} := w_1v_1+w_2v_2 and the complementary average {v}′:=w2v1+w1v2\{v\}' := w_2v_1+w_1v_2. With this notation, we consider the problem

−∇⋅(κi∇u)=fin Ωi, i=1,2,-\nabla\cdot(\kappa_i\nabla u)=f \quad\text{in }\Omega_i,\ i=1,2,
[κ∂nu]=gNon Γ,[\kappa\partial_nu]=g_N \quad\text{on }\Gamma,
βΓ[u]+{κ∂nu}=gΓon Γ,(45)\beta_\Gamma[u]+\{\kappa\partial_nu\}=g_\Gamma\quad\text{on }\Gamma, \tag*{(45)}

with a Dirichlet condition on the fitted outer boundary ∂Ω\partial\Omega. The third equation describes imperfect bonding through a spring layer with stiffness βΓ∈[0,∞]\beta_\Gamma\in[0,\infty]. For homogeneous data, the flux is proportional to the jump in uu. The limit βΓ=∞\beta_\Gamma=\infty gives perfect bonding, with a prescribed jump [u]=gD[u]=g_D obtained by scaling the data as in the boundary Robin problem. At βΓ=0\beta_\Gamma=0, the two fluxes are prescribed independently and vanish when gN=gΓ=0g_N=g_\Gamma=0. This is the scalar counterpart of the spring-layer law of Zhong and Meguid [22], which A. Hansbo and P. Hansbo [16] analysed and discretised in elasticity. We return to this relation in § 7.

The bonding law has the form of the Robin condition, with [u][u] replacing the boundary value and {κ∂nu}\{\kappa\partial_nu\} replacing the boundary flux. We can therefore apply Lemma 2.1 directly, although the flux-jump condition requires an additional correction.

Three transports instead of two.

We construct Γh\Gamma_h as in § 3, so that it divides Ω\Omega into two polytopal subdomains Ω1h\Omega_1^h and Ω2h\Omega_2^h. Let ThΓ\mathcal{T}_h^\Gamma denote the cut elements. We use the Hansbo–Hansbo space, with separate unknowns on the two sides,

Vh:=Vh1×Vh2,Vhi:={v∈C0:v∣T∈Pp(T) ∀T∈Thact,i}.(46)V_h:=V_h^1\times V_h^2,\qquad V_h^i:=\left\{v\in C^0:v|_T\in\mathbb{P}_p(T)\ \forall T\in\mathcal{T}_h^{\mathrm{act},i}\right\}. \tag*{(46)}

where each factor is defined on the elements meeting its subdomain. Thus each cut element carries two independent sets of degrees of freedom. We choose the harmonic weights and mean coefficient

w1=κ2κ1+κ2,w2=κ1κ1+κ2,κ‾=2κ1κ2κ1+κ2,ρ=γκ‾h,(47)w_1=\frac{\kappa_2}{\kappa_1+\kappa_2},\quad w_2=\frac{\kappa_1}{\kappa_1+\kappa_2},\quad\overline{\kappa}=\frac{2\kappa_1\kappa_2}{\kappa_1+\kappa_2},\quad\rho=\frac{\gamma\overline{\kappa}}{h}, \tag*{(47)}

following Dryja [13] and Burman and Zunino [11, (2.22)], with a related discussion in [8 §2]. With ω:=w1κ1=w2κ2=κ‾/2\omega:=w_1\kappa_1=w_2\kappa_2=\overline{\kappa}/2, we have {κ∇v⋅n}=ω(∇v1+∇v2)⋅n\{\kappa\nabla v\cdot n\}=\omega(\nabla v_1+\nabla v_2)\cdot n. Thus the weighted flux average is κ‾\overline{\kappa} times the arithmetic average of the normal gradients [11 §2.2]. Although other weights with w1+w2=1w_1+w_2=1 can be used in the analysis [11 Rem. 2.13], the present choice controls the dependence on the coefficient contrast. This property is used in Lemma 6.1. Integration by parts on the polygonal subdomains gives

∑i(κi∇u,∇v)Ωhi−⟨{κ∂nhu},[v]⟩Γh−⟨[κ∂nhu],{v}⟩Γh=(f,v)Ω,(48)\sum_i(\kappa_i\nabla u,\nabla v)_{\Omega_h^i}-\langle\{\kappa\partial_{n_h}u\},[v]\rangle_{\Gamma_h}-\langle[\kappa\partial_{n_h}u],\{v\}\rangle_{\Gamma_h}=(f,v)_{\Omega}, \tag*{(48)}

where we used the identity a1b1−a2b2={a}[b]+[a]{b}′a_1b_1-a_2b_2=\{a\}[b]+[a]\{b\}'. In addition to Tm\mathcal{T}_m and Λm\Lambda_m, we need a transport of the flux jump,

Πm(w)(x):=Tm([κ∇w])(x)⋅n∗(x),(49)\Pi_m(w)(x):=\mathcal{T}_m\bigl([\kappa\nabla w]\bigr)(x)\cdot n^*(x), \tag*{(49)}

which is the analogue of (15) for the flux jump. We expand [κ∇w][\kappa\nabla w] componentwise along nhn_h and contract the result with the fixed vector n∗(x)n^*(x), so that only ww is differentiated. Applying Lemma 3.2 on each side then shows that Πm(u)\Pi_m(u) approximates [κ∂nu](ph(x))[\kappa\partial_nu](p_h(x)) with an error of order δhm+1\delta_h^{m+1}.

The method

As in the boundary problem, we introduce the weights

θΓ:=βΓβΓ+ρ∈[0,1],sΓ:=1βΓ+ρ,ρ=γκ‾h,(50)\theta_\Gamma:=\frac{\beta_\Gamma}{\beta_\Gamma+\rho}\in[0,1],\quad s_\Gamma:=\frac{1}{\beta_\Gamma+\rho},\quad\rho=\frac{\gamma\overline{\kappa}}{h}, \tag*{(50)}

which satisfy θΓ=βΓsΓ\theta_\Gamma=\beta_\Gamma s_\Gamma and 1−θΓ=ρsΓ1-\theta_\Gamma=\rho s_\Gamma. We can therefore apply Lemma 2.1 with U=[u]U=[u], Λ={κ∂nu}\Lambda=\{\kappa\partial_nu\}, G=gΓG=g_\Gamma and β=βΓ\beta=\beta_\Gamma. The discrete problem is to find uh∈Vhu_h\in V_h such that AhΓ(uh,v)=LhΓ(v)A_h^\Gamma(u_h,v)=L_h^\Gamma(v) for all v∈Vhv\in V_h, where

AhΓ(u,v):=∑i(κi∇u,∇v)Ωhi+jh(u,v)−⟨{κ∂nhu},[v]⟩Γh−⟨[κ∂nhu]−Πm(u),{v}⟩Γh+ρsΓ⟨Λm(u),[v]⟩Γh+θΓρ⟨[Tm(u)],[v]⟩Γh−θΓ⟨[Tm(u)],{κ∂nhv}⟩Γh−sΓ⟨Λm(u),{κ∂nhv}⟩Γh,(51)\begin{aligned} A_h^\Gamma(u,v):={}&\sum_i(\kappa_i\nabla u,\nabla v)_{\Omega_h^i}+j_h(u,v)-\langle\{\kappa\partial_{n_h}u\},[v]\rangle_{\Gamma_h}\\ &-\langle[\kappa\partial_{n_h}u]-\Pi_m(u),\{v\}\rangle_{\Gamma_h}\\ &+\rho s_\Gamma\langle\Lambda_m(u),[v]\rangle_{\Gamma_h}+\theta_\Gamma\rho\langle[\mathcal{T}_m(u)],[v]\rangle_{\Gamma_h}\\ &-\theta_\Gamma\langle[\mathcal{T}_m(u)],\{\kappa\partial_{n_h}v\}\rangle_{\Gamma_h}-s_\Gamma\langle\Lambda_m(u),\{\kappa\partial_{n_h}v\}\rangle_{\Gamma_h}, \tag*{(51)} \end{aligned}
LhΓ(v):=(f,v)Ω+⟨gN∘ph,{v}⟩Γh+ρsΓ⟨gΓ∘ph,[v]⟩Γh−sΓ⟨gΓ∘ph,{κ∂nhv}⟩Γh,(52)L_h^\Gamma(v):=(f,v)_\Omega+\langle g_N\circ p_h,\{v\}\rangle_{\Gamma_h}+\rho s_\Gamma\langle g_\Gamma\circ p_h,[v]\rangle_{\Gamma_h}-s_\Gamma\langle g_\Gamma\circ p_h,\{\kappa\partial_{n_h}v\}\rangle_{\Gamma_h}, \tag*{(52)}

Here Λm(u):=Tm({κn∗⋅∇u})\Lambda_m(u):=\mathcal{T}_m(\{\kappa n^*\cdot\nabla u\}) transports the weighted average flux, while the ghost penalty jhj_h acts separately on the two active meshes. On each side, it is applied to faces belonging to cut elements.

The second line of (6.7) corrects the flux jump, while the remaining terms follow from the Robin method in § 4. The value of βΓ\beta_\Gamma enters only through θΓ\theta_\Gamma and sΓs_\Gamma. At βΓ=∞\beta_\Gamma=\infty, we have θΓ=1\theta_\Gamma=1 and sΓ=ρsΓ=0s_\Gamma=\rho s_\Gamma=0, so that the two Λm\Lambda_m terms vanish. With the corresponding scaling of the data, this gives the Hansbo–Hansbo method [15] for [u]=gD[u]=g_D, with the jump corrected by Tm\mathcal{T}_m and the flux jump by Πm\Pi_m. At βΓ=0\beta_\Gamma=0, the values θΓ=0\theta_\Gamma=0, sΓ=1/ρs_\Gamma=1/\rho and ρsΓ=1\rho s_\Gamma=1 remove the Tm\mathcal{T}_m terms, and the two flux conditions prescribe the one-sided fluxes. For 0<βΓ<∞0<\beta_\Gamma<\infty, both transports contribute with weights θΓ\theta_\Gamma and 1−θΓ1-\theta_\Gamma. The experiments in §8.2 cover nine orders of magnitude in βΓ\beta_\Gamma.

Remark 6.1 (The flux-jump term is the misalignment correction). On the exact interface, the term ⟨[κ∂nu],{v}⟩Γ\langle[\kappa\partial_n u],\{v\}\rangle_\Gamma vanishes when the flux is continuous. Its counterpart on Γh\Gamma_h must nevertheless be retained, as shown by the exact identity on Γh\Gamma_h,

[κ∂nhu]=Π0(u)+[κ∇u]⋅(nh−n∗).(53)[\kappa\partial_{n_h}u]=\Pi_0(u)+[\kappa\nabla u]\cdot(n_h-n^*). \tag*{(53)}

Here Π0(u)\Pi_0(u) approximates gN∘phg_N\circ p_h with an error of order δh\delta_h, while the second term contains the tangential flux jump and a normal mismatch of size θh=O(h)\theta_h=O(h). The tangential flux jump is generally non-zero, since on Γ\Gamma it satisfies [κ∇u]⋅t=(κ1−κ2)∂tu1+κ2∂t[u][\kappa\nabla u]\cdot t=(\kappa_1-\kappa_2)\partial_tu_1+\kappa_2\partial_t[u]. This term therefore depends on the solution and cannot be replaced by data. A fixed subdivision of each chord reduces its size but does not improve its order in hh. We instead retain the computable flux jump [κ∂nhu][\kappa\partial_{n_h}u] in the method. After integration by parts, the correction by Πm(u)\Pi_m(u) leaves only the residual Πm(u)−gN∘ph\Pi_m(u)-g_N\circ p_h, which is of order δhm+1\delta_h^{m+1} at the exact solution.

Consistency and error estimate

In this section we use the interface energy norm

∥ ⁣∣v∣ ⁣∥h2:=∑i=12κi∥∇v∥Ωih2+jh(v,v)+θΓρ∥[v]∥Γh2.(54)\lVert\!\lvert v\rvert\!\rVert_h^2:=\sum_{i=1}^{2}\kappa_i\lVert\nabla v\rVert_{\Omega_i^h}^2+j_h(v,v)+\theta_\Gamma\rho\lVert[v]\rVert_{\Gamma_h}^2. \tag*{(54)}

Here jh=∑ijh,ij_h=\sum_i j_{h,i} is the sum of the two ghost penalties. At βΓ=0\beta_\Gamma=0, we fix the mean on the enclosed phase by imposing zero mean there on both the trial and test spaces. For the error estimate, we use the same normalization for the exact solution.

We define the two transported residuals

RΓ(u):=βΓ[Tm(u)]+Λm(u)−gΓ∘ph,RN(u):=Πm(u)−gN∘ph.(55)R_\Gamma(u):=\beta_\Gamma[T_m(u)]+\Lambda_m(u)-g_\Gamma\circ p_h,\qquad R_N(u):=\Pi_m(u)-g_N\circ p_h. \tag*{(55)}

Theorem 6.1 (Consistency). Let 0≤βΓ<∞0\le\beta_\Gamma<\infty and let uu solve (45), with each uiu_i satisfying Assumption 3.1 on its side. Then, for all v∈Vhv\in V_h,

AhΓ(u,v)−LhΓ(v)=ρsΓ⟨RΓ(u),[v]⟩Γh−sΓ⟨RΓ(u),{κ∂nhv}⟩Γh+⟨RN(u),{v}⟩Γh,(56)A_h^\Gamma(u,v)-L_h^\Gamma(v)=\rho s_\Gamma\langle R_\Gamma(u),[v]\rangle_{\Gamma_h}-s_\Gamma\langle R_\Gamma(u),\{\kappa\partial_{n_h}v\}\rangle_{\Gamma_h}+\langle R_N(u),\{v\}\rangle_{\Gamma_h}, \tag*{(56)}

and ∥RΓ(u)∥Γh+∥RN(u)∥Γh≲δhm+1(βΓ+κ1+κ2)Cu\lVert R_\Gamma(u)\rVert_{\Gamma_h}+\lVert R_N(u)\rVert_{\Gamma_h}\lesssim\delta_h^{m+1}(\beta_\Gamma+\kappa_1+\kappa_2)C_u.

Proof. As in the proof of Theorem 5.2, we use (6.4) to cancel the volume terms and the two boundary terms. The flux-jump correction and its datum then leave ⟨RN(u),{v}⟩\langle R_N(u),\{v\}\rangle. Applying Lemma 2.1 to the remaining terms, with U=[u]U=[u] and Λ={κ∂nu}\Lambda=\{\kappa\partial_nu\}, gives the first two terms of (6.12). The residual bound follows from Lemma 3.2 on each side. □\square

Theorem 6.2 (Energy error estimate). Assume the phasewise Poincaré bounds of Lemma 6.1 and Assumption 3.1 on each side, and let 0≤m≤p−10\le m\le p-1. For γ\gamma sufficiently large and h≤h0h\le h_0, the discrete problem has a unique solution when βΓ>0\beta_\Gamma>0. When βΓ=0\beta_\Gamma=0, impose the Neumann compatibility condition and fix the mean in the enclosed phase, restricting both trial and test spaces as specified above. Then

(∑i∥κi∇(u−uh)∥Ωih2+θΓρ∥[u−uh]∥Γh2)1/2≲hp(∑iκi∥ui∥Hp+1(Ui)2)1/2+h−1/2δhm+1Cu,(57)\left(\sum_i\lVert\sqrt{\kappa_i}\nabla(u-u_h)\rVert_{\Omega_i^h}^2+\theta_\Gamma\rho\lVert[u-u_h]\rVert_{\Gamma_h}^2\right)^{1/2}\lesssim h^p\left(\sum_i\kappa_i\lVert u_i\rVert_{H^{p+1}(U_i)}^2\right)^{1/2}+h^{-1/2}\delta_h^{m+1}C_u, \tag*{(57)}

Here UiU_i is the fixed extension neighbourhood of Ω‾i\overline{\Omega}_i, and CuC_u carries the same coefficient weights. The constants and h0h_0 may depend on the phasewise Poincaré bounds but are independent of the position of Γ\Gamma relative to the mesh. They are also independent of βΓ\beta_\Gamma and the contrast κ2/κ1\kappa_2/\kappa_1, provided that the Poincaré bounds are uniform in these parameters.

Remark 6.2 (What contrast independence means here). Since the norms on both sides of (57) contain coefficient weights, contrast independence refers to the constants in these norms, subject to the stated Poincaré bounds. For perfect bonding, Burman, Guzmán, Sánchez and Sarkis [8, Thm. 4.3] prove a stronger flux estimate whose constant is independent of the coefficients. Their proof uses a discrete extension from one phase to the whole domain. We have not examined whether boundary value correction preserves that estimate.

Sketch of proof. We follow the coercivity proof of Theorem 5.1, using the weights (6.3) to control the coefficient factors [15, 11]. The last two lines of (6.7) are treated as in the boundary problem, with [v][v] in place of vv. The second line gives the additional misalignment term −⟨[κ∂nhv]−Π0(v),{v}⟩=−⟨[κ∇v]⋅(nh−n∗),{v}⟩-\langle[\kappa\partial_{n_h}v]-\Pi_0(v),\{v\}\rangle=-\langle[\kappa\nabla v]\cdot(\boldsymbol{n}_h-\boldsymbol{n}^*),\{v\}\rangle, which is bounded in Lemma 6.1 below. Consistency follows from Theorem 6.1, and we bound the first two terms of (6.12) as in Corollary 5.1. For the flux-jump residual, we retain the coefficient weights before applying the Cauchy–Schwarz inequality. A Taylor remainder from phase ii paired with the trace from phase jj has coefficient κiw3−jκj−1/2\kappa_i w_{3-j}\kappa_j^{-1/2}. Extracting the solution weight κi1/2\kappa_i^{1/2} leaves the factor

κi1/2w3−jκj−1/2≤1,i,j∈{1,2}.\kappa_i^{1/2}w_{3-j}\kappa_j^{-1/2}\le1,\qquad i,j\in\{1,2\}.

The calculation is the same as in Lemma 6.1 below, so the continuous trace estimate gives an O(δhm+1)O(\delta_h^{m+1}) bound for this residual in the weighted norms. We treat the average-flux residual coefficient by coefficient as well, and use the boundary penalty for the trace residual. Together, these estimates imply the stated uniform O(h−1/2δhm+1)O(h^{-1/2}\delta_h^{m+1}) bound. The perfect-bonding case follows by taking the scaled-data limit, while the interpolation argument is the phasewise version of Theorem 5.3. □

Lemma 6.1 (The interface misalignment remainder). Let the weights be those of (6.3) and assume, as in (5.4), that κi∥vi∥h2≤Λ2∥∣v∣∥h2\kappa_i\|v_i\|_{h}^{2}\le\Lambda^2\|\lvert v\rvert\|_{h}^{2} for i=1,2i=1,2. Then

∣⟨[κ∇v]⋅(nh−n∗),{v}⟩Γh∣≤CcΓh1/2∥∣v∣∥h2,\left|\langle[\kappa\nabla v]\cdot(\boldsymbol{n}_h-\boldsymbol{n}^*),\{v\}\rangle_{\Gamma_h}\right|\le C c_\Gamma h^{1/2}\|\lvert v\rvert\|_{h}^{2},

where CC depends on Λ\Lambda, but, for a fixed bound on Λ\Lambda, is independent of hh, βΓ\beta_\Gamma and the contrast κ2/κ1\kappa_2/\kappa_1.

Proof. As in Lemma 5.2, we apply the discrete trace inequality to the gradients and the continuous trace inequality (5.3) to the functions. This gives

∥[κ∇v]∥Γh≤∑iκi∥∇vi∥Γh≲h−1/2∑iκi1/2Ei,\|[\kappa\nabla v]\|_{\Gamma_h}\le\sum_i\kappa_i\|\nabla v_i\|_{\Gamma_h}\lesssim h^{-1/2}\sum_i\kappa_i^{1/2}E_i,
∥{v}′∥Γh≤∑iw3−i∥vi∥L2(Γh)≲∑iw3−iκi−1/2Λi,\|\{v\}'\|_{\Gamma_h}\le\sum_i w_{3-i}\|v_i\|_{L^2(\Gamma_h)}\lesssim\sum_i w_{3-i}\kappa_i^{-1/2}\Lambda_i,

where Ei:=(κi∥∇vi∥h2+jh,i(vi,vi))1/2≤∥∣v∣∥hE_i:=(\kappa_i\|\nabla v_i\|_h^2+j_{h,i}(v_i,v_i))^{1/2}\le\|\lvert v\rvert\|_h and Λi:=(1+Λ2)1/2∥∣v∣∥h\Lambda_i:=(1+\Lambda^2)^{1/2}\|\lvert v\rvert\|_h. The product contains the four coefficients κi1/2w3−jκj−1/2\kappa_i^{1/2}w_{3-j}\kappa_j^{-1/2}. The two diagonal ones are w2w_2 and w1w_1, both at most 1. The two off-diagonal ones are both equal to

(κ1κ2)1/2κ2κ1+κ2=κ1κ2κ1+κ2≤12\left(\frac{\kappa_1}{\kappa_2}\right)^{1/2}\frac{\kappa_2}{\kappa_1+\kappa_2} =\frac{\sqrt{\kappa_1\kappa_2}}{\kappa_1+\kappa_2}\le\frac{1}{2}

by the arithmetic–geometric mean inequality. Thus all four coefficients are bounded by 1 independently of the contrast, and multiplication by θh≤2cΓh\theta_h\le2c_\Gamma h gives the result. The harmonic weights [11] are essential here, since arithmetic weights would introduce the off-diagonal factors 12(κ1/κ2)1/2\frac{1}{2}(\kappa_1/\kappa_2)^{1/2} and its reciprocal. □

Remark 6.3 (The debonded limit). At βΓ=0\beta_{\Gamma}=0, the enclosed phase solves a pure Neumann problem, whose solution is determined only up to a constant. We therefore fix its mean before applying Poincaré’s inequality, as in the related normalisation used in [8 Thm. 4.3]. For βΓ>0\beta_{\Gamma}>0, the bonding term controls this constant, but that control weakens as βΓ\beta_{\Gamma} tends to zero. Thus Theorem 6.2 does not assert uniformity in this limit unless the phasewise Poincaré constants remain bounded.

Imperfect bonding as the interface Robin condition

A. Hansbo and P. Hansbo [16] considered elasticity problems with strong and weak discontinuities, coupling the two phases through a cohesive law that includes the limit of perfect bonding. The third equation in (45) is the scalar analogue of this law. Our formulation (6.7) uses the same doubled unknowns and weighted averages, with Taylor transports and a flux-jump correction to account for the surrogate interface.

Boundary value correction also has a direct interpretation as an interface compliance. To see this, take m=1m=1 and assume continuous flux, so that gN=0g_N=0. Then, on Γh\Gamma_h,

[T1(u)]=[u]+ϱh[∂nhu]=[u]+Kκ1∂nhu1+O((θh+δh)δh),[\mathcal{T}_1(u)] = [u] + \varrho_h [\partial_{n_h}u] = [u] + \mathcal{K}\kappa_1\partial_{n_h}u_1 + O((\theta_h+\delta_h)\delta_h),

where K:=ϱh(κ1−1−κ2−1)\mathcal{K}:=\varrho_h(\kappa_1^{-1}-\kappa_2^{-1}). Thus the transported condition [T1(u)]=gD∘ph[\mathcal{T}_1(u)]=g_D\circ p_h gives, to this order,

[u]=gD∘ph−Kκ1∂nhu1.(58)[u] = g_D\circ p_h-\mathcal{K}\kappa_1\partial_{n_h}u_1. \tag*{(58)}

This has the algebraic form of an imperfect bonding law with compliance K\mathcal{K}, although the geometric compliance can have either sign. The geometric offset thus acts as an artificial compliance, weighted by the jump of the inverse diffusivity. Since ϱh=O(h2)\varrho_h=O(h^2), this contribution has order h2h^2 and, at this order, adds to the physical compliance 1/βΓ1/\beta_{\Gamma}.

Two features distinguish this correction from a physical compliance. Because K\mathcal{K} may change sign with ϱh\varrho_h, we treat it as a perturbation in the stability analysis. Moreover, K=0\mathcal{K}=0 when κ1=κ2\kappa_1=\kappa_2, so the first-order correction vanishes under the assumed flux continuity. At a boundary, the weights θ\theta and 1−θ1-\theta give a corresponding interpretation. The Dirichlet limit corrects the value and the Neumann limit corrects the flux, while both corrections contribute for a Robin condition.

Numerical examples

We use structured triangulations of the unit square and continuous PpP_p elements, with one chord per cut element for the surrogate boundary. The ghost penalty (18) uses γg=5⋅10−2\gamma_g=5\cdot10^{-2}, and the Nitsche parameter is set to γ=20\gamma=20. The first geometry is a circle of radius 0.3137, centred at (0.5017, 0.4983). The second is a four-petal flower with level set Φ=r2−R2+ccos⁡(4φ)\Phi=r^2-R^2+c\cos(4\varphi), where R=0.31R=0.31 and c=0.04c=0.04, about the same centre. Neither boundary is aligned with the mesh, as shown in Figure 1.

The circle and flower geometries

Figure 1. The two geometries at N=16N = 16: the background mesh, the active elements (shaded), the surrogate boundary Γh\Gamma_h (thick) and the exact boundary Γ\Gamma (thin).

For the boundary value problems, we take u=cos⁡(2x)sin⁡(2y)u=\cos(2x)\sin(2y) with κ=1\kappa=1 and c=1c=1, compute ff and gg from this solution, and evaluate boundary data at ph(x)∈Γp_h(x)\in\Gamma. Cut cells are integrated using a centroid fan with signed triangle areas. With sufficiently accurate triangle quadrature, this gives exact integration of polynomials on a simple polygon. On each element, we obtain the nodal basis and the derivatives required by the Taylor operators by inverting the Vandermonde matrix of a scaled monomial basis.

We first apply a patch test on a curved boundary. For a polynomial solution of degree pp and m=pm=p, the operators Tm\mathcal{T}_m, Λm\Lambda_m and Πm\Pi_m are exact, so the residuals (32) and (55) vanish. This algebraic test uses m=pm=p, whereas the error analysis assumes m≤p−1m\leq p-1. The relative H1H^1 error is below 2⋅10−52\cdot10^{-5} for p=1p=1, 4⋅10−134\cdot10^{-13} for p=2p=2 and 6⋅10−106\cdot10^{-10} for p=3p=3. These bounds hold for β∈{0,1,103,∞}\beta\in\{0,1,10^3,\infty\} in both the boundary and interface tests, with the larger error for p=3p = 3 consistent with the conditioning of the monomial Vandermonde matrix. The patch test thus checks the geometry maps, Taylor operators, Nitsche terms, ghost penalty and quadrature together.

We also measure the geometric errors for N=12,…,192N = 12,\ldots,192. The ratio δh/h2\delta_h/h^2 stays below 0.26 on the circle and 1.02 on the flower, while the corresponding bounds for θh/h\theta_h/h are 1.18 and 5.78. These measurements are consistent with the orders in Lemma 3.1.

The Robin family: convergence and the Taylor order.

Table 1 reports least-squares convergence rates in the H1H^1 seminorm and in L2L^2 over the four meshes N=12,24,48,96N = 12,24,48,96, for p=1,2,3p = 1,2,3, m=0,1,2m = 0,1,2 and the three representative values β=0\beta= 0 (Neumann), β=1\beta= 1 (Robin) and β=∞\beta= \infty (Dirichlet).

ppmmβ=0\beta= 0β=1\beta= 1β=∞\beta= \infty
H1H^1L2L^2H1H^1L2L^2H1H^1L2L^2
100.981.790.981.920.992.01
110.981.790.981.920.992.02
202.151.772.201.821.541.87
212.222.962.223.162.223.29
222.222.962.223.162.223.29
301.811.781.851.821.331.90
313.404.073.404.243.414.06
323.404.003.404.193.414.36

Table 1. Least-squares convergence rates on the circle over N=12,…,96N = 12,\ldots,96. Optimal is pp in H1H^1 and p+1p + 1 in L2L^2.

The Dirichlet results agree with the rule (4.5). For m=0m = 0 and p≥2p \ge2, the predicted energy rate is min⁡(p,2m+32)=32\min(p,2m+\frac{3}{2}) = \frac{3}{2}, compared with measured rates of 1.54 for p=2p = 2 and 1.33 for p=3p = 3. Adding one Taylor term raises these rates to 2.22 and 3.41.

At fixed finite β\beta, the improved estimate in Theorem 5.3 predicts an energy rate of min⁡(p,2m+2)\min(p,2m+2). For p=2p = 2 and m=0m = 0, the predicted order 2 agrees with the measured rates of 2.15 at β=0\beta= 0 and 2.20 at β=1\beta= 1. The optimal energy convergence in these cases therefore follows from the analysis. For p=3p = 3 and m=0m = 0, the predicted rate is again 2, while the measured rates are 1.81 and 1.85 on the tested meshes. The distinction between finite β\beta and the Dirichlet limit comes from the complementary weights in Remark 4.1, which confine the half-order loss in the uniform estimate to the β\beta-weighted part of the residual.

Figure 2. Energy-norm convergence on the circle for p=1,2,3p=1,2,3. Dashed curves use m=0m=0 and solid curves use m=1m=1. The reference slopes are hph^p and h3/2h^{3/2}.

The improved energy estimate does not imply optimal L2L^2 convergence. For p=2p=2 and m=0m=0, the measured L2L^2 rates range from 1.77 to 1.87, increasing with m=1m=1 to between 2.96 and 3.29. This agrees with the need for a correction discussed in Remark 5.2. At β=0\beta=0, where only the flux transport Λm\Lambda_m is active, increasing mm from 0 to 1 raises the L2L^2 rate from 1.77 to 2.96.

We also vary β\beta over ten orders of magnitude at N=48N=48, with the results shown in Figure 3, left. For m=0m=0, the error grows monotonically by a factor 2.69 between the Neumann and Dirichlet limits. The transition occurs near β≃ρ=γκ/h\beta\simeq\rho=\gamma\kappa/h, which is of order 10310^3 on this mesh. For m=1m=1, the variation is at most 0.1 per cent over the tested range, so the correction removes most of the observed dependence on β\beta in this example.

Figure 3. Left: the H1H^1 error at N=48N = 48 as β\beta varies from the Neumann to the Dirichlet limit. The curves are nearly flat for m=1m = 1. Right: the interface problem with (solid) and without (dashed) the flux-jump term of (6.7), for three contrasts.

On the non-convex flower, where ϱh\varrho_h changes sign along Γh\Gamma_h, the energy rates for p=2p=2 and m=1m=1 are 2.24, 2.24 and 2.23 for β=0,1,∞\beta=0,1,\infty. For m=0m=0, the Dirichlet rate falls to 1.22. The corrected method therefore retains its convergence rate despite the changing sign of ϱh\varrho_h.

The interface problem and the flux-jump correction

We use the same circular interface with κ1=1\kappa_1=1 and κ2=10\kappa_2=10, and choose the exact solutions u1=sin⁡(1.7x)cos⁡(1.7y)u_1=\sin(1.7x)\cos(1.7y) and u2=0.3+x2−12y2+0.4xyu_2=0.3+x^2-\frac{1}{2}y^2+0.4xy. Both the solution jump and the flux jump are non-zero, as is the tangential flux jump in Remark 6.1.

For perfect bonding, the H1H^1 rate is 1.00 for p=1p=1. For p=2p=2, it increases from 1.40 with m=0m=0 to 2.17 with m=1m=1, while for p=3p=3 the corresponding rates are 1.36 and 3.32. These results agree with the rule (21), and the case βΓ=10\beta_\Gamma=10 gives similar results on these meshes. At N=48N=48 and p=2p=2, we also vary βΓ\beta_\Gamma over nine orders of magnitude. With m=1m=1, the error varies by 0.3 per cent, indicating weak dependence on the bonding stiffness in the tested range.

To isolate the flux-jump correction, we run the method with and without the second line of (51). In the latter case, the flux-jump datum is imposed directly on Γh\Gamma_h. Both runs use p=2p=2, m=1m=1 and perfect bonding, with the results given in Table 2.

contrast κ2/κ1\kappa_2/\kappa_1withwithout
H1H^1 rateL2L^2 rateH1H^1 rateL2L^2 rate
12.163.241.821.88
102.173.281.401.88
1002.173.271.311.88

Table 2. With and without the flux-jump term of (6.7), p=2p = 2, m=1m = 1, perfect bonding.

Omitting the flux-jump correction reduces the rate at each contrast. At contrast 1, the H1H^1 rate falls from 2.16 to 1.82, while at contrasts 10 and 100 it falls from 2.17 to 1.40 and 1.31, respectively. The tangential flux jump is (κ1−κ2)∂tu1+κ2∂t[u](\kappa_1-\kappa_2)\partial_t u_1+\kappa_2\partial_t[u], which need not vanish at unit contrast. In this example, it remains non-zero because the jump varies along the interface. Equal coefficients and a constant jump would make the term vanish. The corrected rates are almost unchanged over the three contrasts, supporting the weighted estimate in Theorem 6.2 for these cases. This correction adds two terms on the cut elements.

The symmetric preconditioner.

Corollary 5.2 gives a contraction factor of order h1/2h^{1/2} when Ah0A_h^0 preconditions the corrected form, a prediction we test with m=1m = 1 in Table 3. The observed factor is obtained from the iteration e↦e−(Ah0)−1Ahee \mapsto e - (A_h^0)^{-1}A_h e on the homogeneous problem, starting from a random vector. We also report iteration counts for Richardson iteration and GMRES, both using the same factorisation of Ah0A_h^0 and a stopping tolerance of 10−810^{-8} for the relative residual.

caseqq observed N=12N = 12N=96N = 96iterations (max) RichardsonGMRES∥uit−udir∥/∥udir∥\lVert u_{\mathrm{it}} - u_{\mathrm{dir}}\rVert/\lVert u_{\mathrm{dir}}\rVert
β=0\beta= 0, p=2p = 20.02060.0041541.26⋅10−91.26 \cdot10^{-9}
β=0\beta= 0, p=3p = 30.02260.0069553.91⋅10−103.91 \cdot10^{-10}
β=1\beta= 1, p=2p = 20.01810.0037542.44⋅10−102.44 \cdot10^{-10}
β=1\beta= 1, p=3p = 30.02220.0073547.44⋅10−117.44 \cdot10^{-11}
β=∞\beta= \infty, p=2p = 20.01860.0041443.13⋅10−93.13 \cdot10^{-9}
β=∞\beta= \infty, p=3p = 30.02380.0044443.32⋅10−93.32 \cdot10^{-9}

Table 3. The uncorrected symmetric form as a preconditioner, m=1m = 1: observed contraction factor on the coarsest and finest meshes, the largest iteration count over N=12,…,96N = 12, \ldots, 96 to a relative residual 10−810^{-8}, and the largest discrepancy with the direct solution of the non-symmetric system.

The contraction factor never exceeds 0.028, and its smallest value on the finest mesh is 0.0037. Over the four meshes, the least-squares rates range from 0.58 to 0.88 across the six cases, consistent with the O(h1/2)O(h^{1/2}) bound of Corollary 5.2. At the Dirichlet limit, ρs→0\rho_s \to0, so the term responsible for the half-order bound in Proposition 5.1 vanishes from (5.20), leaving an O(h)O(h) perturbation.

No more than five iterations are needed in these tests, and the counts decrease under refinement. The last column compares the iterative and direct solutions of the non-symmetric system, with small discrepancies consistent with the stopping tolerance and finite precision arithmetic. The method therefore uses one symmetric factorisation and applies the geometric correction at each iteration.

Conclusions

We have presented a cut finite element method with boundary value correction for Robin and Neumann conditions, including an extension to interfaces with imperfect bonding. By transporting the solution and its gradient to the exact boundary and evaluating the flux with the exact normal, the method retains a piecewise affine geometry for all integration.

The energy analysis gives a geometric error of O(h−1/2δhm+1)O(h^{-1/2}\delta_h^{m+1}) uniformly through the Dirichlet limit, but the continuous trace inequality sharpens this to O(δhm+1)O(\delta_h^{m+1}) for fixed finite β\beta. The distinction comes from the trace residual, which carries the Dirichlet weight, and leads to the sufficient conditions 2m+32≥p2m + \frac{3}{2} \ge p and 2m+2≥p2m + 2 \ge p when δh=O(h2)\delta_h = O(h^2). In particular, p=2p = 2 and m=0m = 0 give second-order energy convergence for fixed finite β\beta, explaining the optimal energy rates in Table 1 under the stated assumptions. The analysis does not establish the larger slopes measured on these meshes, and the corresponding L2L^2 estimate still requires the adjoint-consistency argument discussed in Remark 5.2.

For interface problems, the additional flux-jump correction accounts for the different normals used by integration by parts and by the prescribed jump condition. The numerical examples show a substantial loss of accuracy when this term is omitted, whereas the corrected method gives energy rates close to two for quadratic elements at all three tested contrasts. Uniform stability remains subject to the Poincaré bounds stated in the analysis, with a normalization for any uncoupled constant mode at zero bonding stiffness.

The perturbation estimate also suggests a simple preconditioner based on the symmetric form assembled on the affine geometry. The preconditioned solvers require at most five iterations in the numerical examples, with the geometric correction applied without a separate factorisation.

Funding. EB was supported in part by EPSRC grants EP/X042650/1 and EP/V050400/1. PH was supported by the Swedish Research Council under grant 2022-03908. MGL was supported in part by the Swedish Research Council under grants 2021-04925 and 2025-05562, the Knut and Alice Wallenberg Foundation under grant KAW 2025.0277, and the Swedish Research Programme Essence.

References

  1. [1]N. Atallah, C. Canuto and G. Scovazzi, The second-generation shifted boundary method and its numerical analysis, Comput. Methods Appl. Mech. Engrg. 372 (2020), 113341.DOI
  2. [2]S. Bertoluzza, M. Montardini, M. Pennacchio and D. Prada, The virtual element method on polygonal pixel-based tessellations, J. Comput. Phys. 518 (2024), Paper No. 113334.arxiv.org/abs/2206.03449
  3. [3]S. Bertoluzza, M. Montardini and D. Prada, The high order virtual element method on approximate domains: Neumann boundary conditions, arXiv:2608.07100 [math.NA] (2026).arxiv.org/abs/2608.07100
  4. [4]S. Bertoluzza, M. Pennacchio and D. Prada, Weakly imposed Dirichlet boundary conditions for 2D and 3D virtual elements, Comput. Methods Appl. Mech. Engrg. 400 (2022), Paper No. 115454.DOI
  5. [5]J. Boiveau, E. Burman, S. Claus and M. G. Larson, Fictitious domain method with boundary value correction using penalty-free Nitsche method, J. Numer. Math. 26 (2018), no. 2, 64–68.arxiv.org/abs/1610.04482
  6. [6]J. H. Bramble, T. Dupont and V. Thomée, Projection methods for Dirichlet’s problem in approximating polygonal domains with boundary-value corrections, Math. Comp. 26 (1972), 869–879.DOI
  7. [7]E. Burman, Ghost penalty, C. R. Math. Acad. Sci. Paris 348 (2010), no. 21–22, 1217–1220.DOI
  8. [8]E. Burman, J. Guzmán, M. A. Sánchez and M. Sarkis, Robust flux error estimation of an unfitted Nitsche method for high-contrast interface problems, IMA J. Numer. Anal. 38 (2018), no. 2, 646–668.arxiv.org/abs/1602.00603
  9. [9]E. Burman, P. Hansbo and M. G. Larson, A cut finite element method with boundary value correction, Math. Comp. 87 (2018), no. 310, 633–657.
  10. [10]E. Burman, P. Hansbo and M. G. Larson, Dirichlet boundary value correction using Lagrange multipliers, BIT Numer. Math. 60 (2020), no. 1, 235–260.arxiv.org/abs/1903.07104
  11. [11]E. Burman and P. Zunino, A domain decomposition method based on weighted interior penalties for advection-diffusion-reaction problems, SIAM J. Numer. Anal. 44 (2006), no. 4, 1612–1638.DOI
  12. [12]J. Cheung, M. Perego, P. Bochev and M. Gunzburger, Optimally accurate higher-order finite element methods for polytopial approximations of domains with smooth boundaries, Math. Comp. 88 (2019), no. 319, 2187–2219.arxiv.org/abs/1710.05628
  13. [13]M. Dryja, On discontinuous Galerkin methods for elliptic problems with discontinuous coefficients, Comput. Methods Appl. Math. 3 (2003), no. 1, 76–85.DOI
  14. [14]T. Dupont, J. Guzmán and L. R. Scott, Obtaining higher-order Galerkin accuracy when the boundary is polygonally approximated, J. Numer. Math. 33 (2025), no. 1, 55–86.arxiv.org/abs/2001.03082
  15. [15]A. Hansbo and P. Hansbo, An unfitted finite element method, based on Nitsche’s method, for elliptic interface problems, Comput. Methods Appl. Mech. Engrg. 191 (2002), no. 47–48, 5537–5552.DOI
  16. [16]A. Hansbo and P. Hansbo, A finite element method for the simulation of strong and weak discontinuities in solid mechanics, Comput. Methods Appl. Mech. Engrg. 193 (2004), no. 33–35, 3523–3540.DOI
  17. [17]Y. Hou, Y. Liu and T. Zhao, A high-order mixed finite element method for second order elliptic equations on curved domain with boundary value corrections, Int. J. Numer. Anal. Model. 23 (2026), no. 3, 442–459.DOI
  18. [18]M. Juntunen and R. Stenberg, Nitsche’s method for general boundary conditions, Math. Comp. 78 (2009), no. 267, 1353–1374.
  19. [19]C. Lehrenfeld, High order unfitted finite element methods on level set domains using isoparametric mappings, Comput. Methods Appl. Mech. Engrg. 300 (2016), 716–733.DOI
  20. [20]A. Main and G. Scovazzi, The shifted boundary method for embedded domain computations. Part I: Poisson and Stokes problems, J. Comput. Phys. 372 (2018), 972–995.DOI
  21. [21]A. Main and G. Scovazzi, The shifted boundary method for embedded domain computations. Part II: Linear advection-diffusion and incompressible Navier-Stokes equations, J. Comput. Phys. 372 (2018), 996–1026.DOI
  22. [22]Z. Zhong and S. A. Meguid, On the elastic field of a spherical inhomogeneity with an imperfectly bonded interface, J. Elasticity 55 (1999), no. 2, 101–140.

Paper details

vlnc:W_yerzbx4f
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents