Abstract
In this paper, we present a novel approach to the approximate solution of elliptic partial differential equations on compact submanifolds of \(\mathbb {R}^{d}\), particularly compact surfaces and the surface equation \({\Delta }_{\mathbb {M}} u - \lambda u=f\). In the course of this, we reconsider differential operators on such submanifolds to deduce suitable penalty based functionals. These functionals are based on the residual of the equation in an integral representation, extended by a penalty on the first-order normal derivative. The general framework we develop is accompanied by error analysis and exemplified by numerical examples employing tensor product B-splines.
Similar content being viewed by others
References
Adalsteinsson, D., Sethian, J.A.: Transport and diffusion of material quantities on propagating interfaces via level set methods. J. Comput. Phys. 185 (1), 271–288 (2003)
Adams, R.: Sobolev Spaces. Academic Press (1975)
Adams, R.A., Fournier, J.J.: Sobolev Spaces. Academic Press (2003)
Agranovich, M.S.: Sobolev Spaces, Their Generalizations and Elliptic Problems in Smooth and Lipschitz Domains. Springer (2016)
Aubin, T.: Some Nonlinear Problems in Riemannian Geometry. Springer Science & Business Media (2013)
Bartezzaghi, A., Dedè, L., Quarteroni, A.: Isogeometric analysis of high order partial differential equations on surfaces. Comput. Methods Appl. Mech. Eng. 295, 446–469 (2015)
Bazilevs, Y., Calo, V.M., Cottrell, J.A., Evans, J.A., Hughes, T.J.R., Lipton, S., Scott, M.A., Sederberg, T.W.: Isogeometric analysis using t-splines. Comput. Methods Appl. Mech. Eng. 199(5–8), 229–263 (2010)
Berger, M.: A Panoramic View of Riemannian Geometry. Springer Science & Business Media (2012)
Bergh, J., Löfström, J.: Interpolation Spaces: an Introduction. Springer (1976)
Bertalmıo, M., Cheng, L.T., Osher, S., Sapiro, G.: Variational problems and partial differential equations on implicit surfaces. J. Comput. Phys. 174(2), 759–780 (2001)
Burman, E., Hansbo, P., Larson, M.G.: A stabilized cut finite element method for partial differential equations on surfaces: the Laplace–Beltrami operator. Comput. Methods Appl. Mech. Eng. 285, 188–207 (2015)
Burman, E., Hansbo, P., Larson, M.G., Massing, A.: Cut finite element methods for partial differential equations on embedded manifolds of arbitrary codimensions. ArXiv e-prints (2016)
Burman, E., Hansbo, P., Larson, M.G., Massing, A.: A cut discontinuous Galerkin method for the Laplace–Beltrami operator. IMA J. Numer. Anal. 37(1), 138–169 (2017)
Burman, E., Hansbo, P., Larson, M.G., Massing, A., Zahedi, S.: Full gradient stabilized cut finite element methods for surface partial differential equations. Comput. Methods Appl. Mech. Eng. 310, 278–296 (2016)
Canzani, Y.: Analysis on manifolds via the Laplacian. Lecture Notes available at: http://www.math.harvard.edu/canzani/docs/Laplacian.pdf (2013)
Chernyshenko, A.Y., Olshanskii, M.A.: An adaptive octree finite element method for PDEs posed on surfaces. Comput. Methods Appl. Mech. Eng. 291, 146–172 (2015)
Deckelnick, K., Dziuk, G., Elliott, C.M., Heine, C.J.: An h-narrow band finite-element method for elliptic equations on implicit surfaces. IMA J. Numer. Anal. 30(2), 351–376 (2009)
Dedè, L., Quarteroni, A.: Isogeometric analysis for second order partial differential equations on surfaces. Comput. Methods Appl. Mech. Eng. 284, 807–834 (2015)
Delfour, M., Zolésio, J. P.: Shapes and geometries: analysis, differential calculus, and optimization. Society for Industrial and Applied Mathematics (2001)
Dziuk, G.: Finite elements for the Beltrami operator on arbitrary surfaces. Partial Differential Equations and Calculus of Variations, pp. 142–155. Springer (1988)
Dziuk, G., Elliott, C.M.: Finite element methods for surface PDEs. Acta Numerica 22, 289–396 (2013)
Eason, E.D.: A review of least-squares methods for solving partial differential equations. Int. J. Numer. Methods Eng. 10(5), 1021–1046 (1976)
Foote, R.L.: Regularity of the distance function. Proc. Amer. Math. Soc. 92(1), 153–155 (1984)
Fuselier, E.J., Wright, G.B.: A high-order kernel method for diffusion and reaction-diffusion equations on surfaces. J. Sci. Comput. 56(3), 535–565 (2013)
Gallot, S., Hulin, D., Lafontaine, J.: Riemannian Geometry. Springer (1990)
Greer, J.B.: An improvement of a recent Eulerian method for solving PDEs on general geometries. J. Sci. Comput. 29(3), 321–352 (2006)
Güneysu, B., Pigola, S.: The Calderón-Zygmund inequality and Sobolev spaces on noncompact Riemannian manifolds. Adv. Math. 281, 353–393 (2015)
Hughes, T.J., Cottrell, J.A., Bazilevs, Y.: Isogeometric analysis: Cad, finite elements, nurbs, exact geometry and mesh refinement. Comput. Methods Appl. Mech. Eng. 194(39–41), 4135–4195 (2005)
Laugwitz, D.: Differential and Riemannian Geometry. Academic Press (1965)
Le Gia, Q.T., Sloan, I.H., Wendland, H.: Multiscale RBF collocation for solving PDEs on spheres. Numer. Math. 121(1), 99–125 (2012)
Lee, J.M.: Introduction to Smooth Manifolds. Springer (2003)
Lehmann, N., Maier, L.B., Odathuparambil, S., Reif, U.: Ambient approximation on hypersurfaces. Constr. Approx. 49(1), 175–190 (2019)
Lunardi, A.: Interpolation theory. Scuola Normale Superiore (1998)
Maier, L.B.: Ambient Approximation of Functions and Functionals on Embedded Submanifolds. Ph.D. thesis, Technische Universität Darmstadt (2018)
Maier, L.B.: Ambient approximation on embedded submanifolds. Constr. Approx., 1–29 (2020)
Odathuparambil, S.: Ambient Spline Approximation on Manifolds. Ph.D. thesis, Technische Universität Darmstadt (2016)
Olshanskii, M.A., Reusken, A.: Error analysis of a space-time finite element method for solving pdes on evolving surfaces. SIAM J. Numer. Anal. 52(4), 2092–2120 (2014)
Olshanskii, M.A., Reusken, A., Grande, J.: A finite element method for elliptic equations on surfaces. SIAM J. Numer. Anal. 47(5), 3339–3358 (2009)
Olshanskii, M.A., Reusken, A., Xu, X.: An Eulerian space-time finite element method for diffusion problems on evolving surfaces. SIAM J. Numer. Anal. 52(3), 1354–1377 (2014)
Olshanskii, M.A., Reusken, A., Xu, X.: A stabilized finite element method for advection–diffusion equations on surfaces. IMA J. Numer. Anal. 34 (2), 732–758 (2014)
Petersen, P., Axler, S., Ribet, K.: Riemannian Geometry. Springer (2006)
Piret, C.: The orthogonal gradients method: a radial basis functions method for solving partial differential equations on arbitrary surfaces. J. Comput. Phys. 231(14), 4662–4675 (2012)
Schumaker, L.: Spline Functions: Basic Theory. Cambridge University Press (1981)
Scott, M.A., Simpson, R.N., Evans, J.A., Lipton, S., Bordas, S.P., Hughes, T.J., Sederberg, T.W.: Isogeometric boundary element analysis using unstructured t-splines. Comput. Methods Appl. Mech. Eng. 254, 197–221 (2013)
Sethian, J.A.: Level Set Methods and Fast Marching Methods: Evolving Interfaces in Computational Geometry, Fluid Mechanics, Computer Vision, and Materials Science, vol. 3. Cambridge University Press (1999)
Triebel, H.: Theory of Function Spaces. Birkhäuser, Basel (1983)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by: Francesca Rapetti
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 A.1 Parameterising compact submanifolds
For technical reasons it is convenient to give some of the arguments in the following proofs using an inverse atlas for the submanifold. Therefore, we will briefly recall a suitable inverse atlas, using [34, Sect. 9.1] and almost copying the corresponding statements in [35, Sect. 6]:
As our submanifolds are required to be compact, we can assume to be equipped with some finite inverse atlas \(\{(\varphi _{i},{\omega ^{0}_{i}})\}_{i\in I}\) where each \({\omega ^{0}_{i}}\subseteq {\mathbb R}^{k}\) is a ball of fixed radius δ0 > 0. On each of these balls, φi is a so-called exponential map\(\exp _{\xi _{i},{\mathbb M}}\) of \({\mathbb M}\) around \(\xi _{i}\in {\mathbb M}\) with ξi = φi(0). For any arbitrary \(\xi \in {\mathbb M}\), the exponential map is defined as
where γξ, x is the uniquely defined arc-length parameterisation for the geodesic that contains ξ and has direction vx for some suitable identification vx in \({\mathrm T}_{\xi }{\mathbb M}\) of x in \({\mathbb R}^{k}\). The validity of this construction is a consequence of the compactness of \({\mathbb M}\) and the fact that the so-called injectivity radiusδ(y) in which the exponential map is an injective map to \(\mathbb {M}\) is itself a smooth function of \(y\in {\mathbb M}\) (cf. [5, Paragr. 3], [8, Prop. 88]). In particular, we can thereby determine a minimax injectivity radius \(\delta _{{\mathbb M}}\) that is valid over all of \({\mathbb M}\).
Moreover, by the same arguments we can demand that δ0 > 0 is chosen such that there are \(\delta _{1}<\delta _{0}<\delta _{2}<\delta _{{\mathbb M}}\) and the balls with radii δ1, δ2 yield an inverse atlas as well. For future reference, we term those balls as parameter spaces by \({\omega ^{1}_{i}}{\subset \omega ^{0}_{i}}{\subset \omega ^{2}_{i}}\).
Furthermore, suitably shrinking δ1, δ0, δ2 further if necessary, we can also demand that there are smooth maps Ti and Ni that assign orthonormal frames of the tangent and normal space in \(\varphi _{i}(x)\in {\mathbb M}\) to \(x{\in \omega _{i}^{j}}\), respectively, for j = 0, 1, 2.
By another compactness argument, we can then deduce that all three maps φi, Ti, Ni are C-bounded on the balls of radius δ2. That is, for any multi-index α and all i ∈ I there is a constant cα > 0 such that it holds |Dαg(x)| < cα for g = φi, Ti, Ni and any \(x{\in \omega _{i}^{2}}\) (cf. [32]).
1.2 A.2 Proof of Lemma 2
First of all, we recall right from above that there is a locally smooth map \(N:\varphi (\omega )\subseteq {\mathbb M}\to {\mathbb R}^{d\times (d-k)}\) that assigns an orthonormal frame ν1(y), ... , νd−k(y) of the normal space \({\mathrm N}_{y}{\mathbb M}\) to each y ∈ φ(ω). We conclude immediately by the definition of the tangential Hessian that locally for any \(v,w\in {\mathrm T}{\mathbb M}\)
Looking closer at the remaining D(N ⋅∇NF), we find by the product rule
Multiplication of this by cotangent vt from the left and by tangent w from the right will annihilate the first sum. This in mind, we can reinsert and obtain
As the bilinear form induced by Dνi(x) can be assumed to be bounded for any i = 1, ... , d − k, because N can be chosen to be C-bounded, we directly deduce that
Hence, we obtain the desired result.
1.3 A.3 Proof of \({\mathrm H}^{1}({\mathbb M})={\mathrm H}^{1}_{\textsc {t}}({\mathbb M}),{\mathrm H}^{2}({\mathbb M})={\mathrm H}^{2}_{\textsc {t}}({\mathbb M})\)
Since we have a finite inverse atlas, we restrict ourselves here to one pair (φ, ω) from the inverse atlas, and we understand the norms \(\left |\!\!{}\left | f\right |\!\!{}\right |_{{\mathrm H}_{\textsc {t}}^{1}({\mathbb M})}\) and \(\left |\!\!{}\left | f\right |\!\!{}\right |_{{\mathrm H}_{\textsc {t}}^{2}({\mathbb M})}\) in the Sobolev case as metric closures of the smooth functions. Then by the transformation law
By considering additionally the chain rule and the fact that the jacobian Jφ(x) maps \(x\in {\mathbb M}\) onto a basis frame for the tangent space, and this map is C-bounded, we can further deduce that
So we need to take care about the second-order terms only. Recalling the notation \({\mathrm {E\!\!{}x}}_{{\mathrm N}} f\), we have for arbitrary coordinate directions xi, xj with respect to ω that by the chain rule
By definition, the set
forms a pointwise basis of the tangent space, and the corresponding basis transform has a bounded spectrum due to C-boundedness by conception of the inverse atlas. So we can deduce by standard linear algebra arguments on basis changes that for bases τ1, ... , τk (making up the frame Ti(x)) and ∂1φ, ... , ∂kφ of \(\mathrm {T}_{x}(\mathbb {M})\)
where we used further that \(\nabla _{{\mathbb M}} f=\nabla {{\mathrm {E\!\!{}x}}_{\mathrm N}} f\) by conception. Thereby, we can deduce in turn that
which gives one part of the required inequality. For the other, we argue by contradiction: We assume that we have a sequence \((f_{n})_{n\in \mathbb {N}}\) of functions with corresponding \( \mathrm {E\!\!{}x}_{\mathrm {N}} f_{n}=\mathrm {E\!\!{}x}_{\mathrm {N}} f_{n}\) such that \(\left |\!\!{}\left | f_{n}\right |\!\!{}\right |_{{\mathrm H}_{\textsc {t}}^{2}({\mathbb M})}^{2}\ge 1\) but
Without restriction, we demand these functions to be smooth. Then, we can directly deduce from the known equivalence of \(\left |\!\!{}\left | f\right |\!\!{}\right |_{\mathrm {H}^{1}(\mathbb {M})}\) and \( \left |\!\!{}\left | f\right |\!\!{}\right |_{\mathrm {H}^{1}_{\textsc {t}}(\mathbb {M})}\), and from Sobolev embeddings, that also
so again it boils down to the second-order term only. There we see now that in particular
On the other hand, we can deduce with \(\partial _{i,j}\varphi := \left (\frac {\partial ^{2}\varphi _{1}}{\partial x_{i}\partial x_{j}}, ... , \frac {\partial ^{2}\varphi _{d}}{\partial x_{i}\partial x_{j}}\right )^{t}\) and the chain rule
Integrating over both after summing over all i, j gives
On the other hand, Hölder’s inequality and (A.3.1) gives that
Inserting this in (A.3.2) and rearranging terms gives that
whereby we have deduced a contradiction.
1.4 A.4 Proving spline approximation of normal derivatives
1.4.1 A.4.1 Extended parameter spaces and parameterisations
With the assertions of Appendix A.1, we can use the inverse atlas for \({\mathbb M}\) to deduce a set of corresponding parameterisations {Φi}i∈I and parameter spaces \(\{{{\Omega }_{i}^{h}}\}_{i\in I}\) that parameterise \(\mathbb {U}_{h}({\mathbb M})\) in a very convenient way: We choose \({{\Omega }_{i}^{h}}:=\omega _{i}\times {\mathrm B}^{(d-k)}_{h}(0)\) for open ball \({\mathrm B}^{(d-k)}_{h}(0)\subseteq {\mathbb R}^{(d-k)}\) of radius h and νj the j th basis vector of the normal frame according to Ni by
Moreover, we can in particular deduce the following lemma according to [35, Lem. 6.1], [34, Sect. 2.2], [36, Lem. 3]:
Lemma 6
Let\(m\in {\mathbb N}\), let\({\mathbb M}\subseteq {\mathbb R}^{d}\)be a smooth compact submanifold, contained in a family\(\mathbb {U}_{h}({\mathbb M})\)of tubular neighbourhoods with 0 < h < h0. Let (φi, ωi)i∈Iand\(({\Phi }_{i},{{\Omega }_{i}^{h}})_{i\in I}\)be the inverse atlases as introduced above. Then there are fixed a1, a2 > 0 independent of h such that for any\(f\in {\mathrm H}^{m}({\mathbb M})\)and fixed b1, b2 > 0 independent of h
whereby in particular for suitable c1, c2 > 0
1.4.2 A.4.2 The integer order case in Theorem 3
We will first concentrate on the integer order case, and deduce the fractional afterwards via some function space interpolation arguments. The proofs are quite similar to those provided in [35] for the standard norm approximation. We can reuse much of the lemmas and argumentation from there, with just a couple of suitable adaptions.
First of all, we recall a spline quasi-interpolation operator Qh can be chosen such that it meets all the requirements presented in [32, 34, 35]. Revising this, we start with the set \(\{B_{i,h}^{m}\}_{i\in {\mathbb Z}^{d}}\) of tensor product B-splines of order \(m\in {\mathbb N}\) over the uniform grid \(h{\mathbb Z}^{d}\), Then we choose a set of corresponding functionals \(\{{\Lambda }_{i,h}\}_{i\in {\mathbb Z}^{d}}\) bounded on \({\mathrm L}_{2}({\mathbb R}^{d})\) such that Λi, hg = 0 for any g with support in the complement of the open ball \({\mathrm B}_{mdh}(ih):=\{x:\left |\!\!{}\left | x-ih\right |\!\!{}\right |_{2}<mdh\}\) and define
For this kind of operator, the following lemma holds (cf. [35, Lem. 6.4/6.8], [34, Lem. 3.17/3.28]):
Lemma 7
Let \(0<n \le m\in {\mathbb N}\) and 0 ≤ μ < n. Let Bh[ζ] := ζ + (−h, h)d be the box centred at \(\zeta \in {\mathbb R}^{d}\). Let \({{\mathrm S}_{h}^{m}}({\mathbb R}^{d})\) be the tensor product spline space of order m with grid width h. Then there is a spline quasi-interpolation operator Qh such that for a constant c independent of h or \(f\in {\mathrm H}^{n}({\mathrm B}_{bh}[\zeta ])\) and some b ≥ a ≥ 1
where pζ, f is the polynomial that coincides with Qhf on one spline cell that contains ζ.
This operator is subsequently applied to \({\mathrm {E\!\!{}x}}_{{\mathrm N}} f\) in order to accomplish the desired proof of Theorem 3. To this end, we let \(\{{\textsc c}_{h}(\zeta )\}_{\zeta \in {\mathrm {Z}_{a,h}}}\) be the set of cells of length h that belong to basis functions whose support intersects \(\mathbb {U}_{h}({\mathbb M})\). Let therein the set Za, h contain the centres of these cells, so that again each cell has the form ζ + (−h/2, h/2)d.
Let then \(S_{h} ={\mathrm Q}_{h} {\mathrm {E\!\!{}x}}_{{\mathrm N}} f\) on these cells and let sh be its trace on \({\mathbb M}\). Let ch be an arbitrary cell. Then \((S_{h})_{|{\textsc c}_{h}}=P_{{\textsc c}_{h}}\) for some polynomial \(P_{{\textsc c}_{h}}\) of order m, and we call by \(p_{{\textsc c}_{h}}={\mathrm {T\!\!{}r}}_{{\mathbb M}} P_{{\textsc c}_{h}}\) the trace of \(P_{{\textsc c}_{h}}\), seen as a polynomial on \({\mathbb R}^{d}\), on all of \({\mathbb M}\). Because any cell is compactly contained in the respective box Bh[ζ] := ζ + (−h, h)d, we have that
By [35, Sect. 3.2] and [34, Sect. 9.1.2] it is ensured that there is a suitable pair \(({\Phi }_{i},{\Omega }_{i}^{ah}=\omega _{i}\times {\mathrm B}_{ah}^{(d-k)}(0))\) corresponding to some (φi, ωi) from the inverse atlas and a suitable, globally valid a > 0 such that for \(\xi ={\Phi }_{i}^{-1}(\zeta )\) it holds \({\Phi }_{i}^{-1}({\mathrm B}_{h}[\zeta ])\subseteq {\mathrm B}_{c_{1} h}[\xi ]\subset {\Omega }_{i}^{ah}\). With this choice, we proceed by omitting the index i ∈ I and the index ch. To obtain the desired relation, we now have to look at the map N = (ν1, ... , ν(d−k)) that gives us a normal frame on the image of ω that is smooth and C-bounded on ω. We decompose then \(\left |\!\!{}\left |(\nabla _{{\mathrm N}} P)\right |\!\!{}\right |_{2}^{2}\) as follows in a finite sum of normal directional derivatives: It holds in any φ(x) for x ∈ ω that
Consequently, we have with the triangle inequality, suitably increased c and ξω as the orthogonal projection of ξ onto ω that
Now we have to notice that on each Ω we have by the definition of Φ via (A) and of N with its component column ν = νj that (DνP) ∘Φ = ∂j(P ∘Φ) for some standard coordinate direction j. We can thus obtain because of \(\mathrm D_{\nu } {{\mathrm {E\!\!{}x}}_{\mathrm N}} f =0\) and Lemma 6 that
We invoke now the approximation order for tp-spline quasi-interpolation as presented in Lemma 7. The first summand can then be bounded by using the evident relation \(|\mathrm D_{\nu } (P-{{\mathrm {E\!\!{}x}}_{\mathrm N}} f)|\le \left |\!\!{}\left | \nabla (P-{{\mathrm {E\!\!{}x}}_{\mathrm N}} f)\right |\!\!{}\right |_{2}\) in the form
Therein, b ≥ a > 0 are suitable fixed constants that can be chosen valid for any i ∈ I. For the second summand, we proceed via a version of Friedrichs’ inequality for arbitrary codimensions (cf. [35, Lem. 6.9] or [34, Thm. 9.24]) to obtain for \(R := (\mathrm D_{\nu } P)\circ {\Phi }-({\mathrm {E\!\!{}x}}_{{\mathrm N}}(\mathrm D_{\nu } P))\circ {\Phi }\) the relation
To achieve that, we notice that it holds \(({\mathrm {E\!\!{}x}}_{{\mathrm N}}(\mathrm D_{\nu } P))({\Phi }(x,z))=({\mathrm {E\!\!{}x}}_{{\mathrm N}}(\mathrm D_{\nu } P))({\Phi }(x,0))\), whereby in particular \(\partial _{z}^{\beta } (({\mathrm {E\!\!{}x}}_{{\mathrm N}}(\mathrm D_{\nu } P))\circ {\Phi })=0=\partial _{z}^{\beta }\partial _{j}({{\mathrm {E\!\!{}x}}_{\mathrm N}} f\circ {\Phi })\). Consequently we can deduce by virtue of Friedrichs’ inequality for functions that finite dimensional polynomials in certain coordinates [35, Lemma 6.9] that it holds
Now we apply the approximation power of splines to (A.4.4), and deduce
Consequently, we can bound in (A.4.2) by (A.4.3) and (A.4.5) that
It remains to sum over all cells. To accomplish that, we have to choose b large enough to work out for each cell in at least one extended parameter space; this will be no problem if h0 is small enough. And we have to choose h0 small enough to achieve that \(\mathbb {U}_{dbh}(\mathbb {M})\) has the closest point property. Then because b was fixed over a fixed extended parameter space and the number of these is finite, we can choose b maximal among them to become independent of the extended parameter spaces. Because clearly any \(z\in \mathbb {U}_{dbh}(\mathbb {M})\) can be in at most n0 balls Bbh[ζ] for some fixed n0, we deduce
and obtain thereby with (A.4.6) and Lemma 6 that
1.4.3 A.4.3 The fractional order case in Theorem 3
To deduce the fracional convergence orders for normal derivatives, we restrict ourselves now to an arbitrary \((\varphi _{i},{\omega _{i}^{0}})\) with corresponding \((\varphi _{i},{\omega _{i}^{2}})\). This is feasible, as all the orders provided for the integer case are valid there as well: The integer proof essentially just took place on one such parameter space. The only modification to make affects the cells considered: We would no longer use all cells that intersect \({\mathbb M}\), but only those that intersect \(\varphi ({\omega _{i}^{0}})\). And we would further assume h small enough that \({\mathrm B}_{b h}[\zeta ]\cap \mathbb {U}_{db h}({\mathbb M})={\mathrm B}_{b h}[\zeta ]\cap \mathbb {U}_{db h}(\varphi _{i}({\omega _{i}^{2}}))\) for relevant cell centres ζ whose corresponding cells intersect \(\varphi _{i}({\omega _{i}^{0}})\). Thereby we circumvent the need to use function space interpolation on submanifolds and just apply it on the parameter space, a Euclidean domain. Then, the required validity of the convergence order for integers gives that it holds in particular:
To see this, we note that the quasi-interpolation operator Qh is local in the following sense: For sufficiently small h we have that \(\| (g - f)\circ \varphi _{i}\|_{{\mathrm H}^{\ell }({\omega _{i}^{2}})}=0\) implies \(\| \nabla _{{\mathrm N}}({\mathrm Q}_{h}({\mathrm {E\!\!{}x}}_{{\mathrm N}} g - {\mathrm {E\!\!{}x}}_{{\mathrm N}} f))\circ \varphi _{i}\|_{{\mathrm L}_{2}({\omega _{i}^{0}})}=0\) for any \(g\in {\mathrm H}^{\ell }({\mathbb M})\). This is a consequence of the required locality of quasi-interpolation operator Qh and the choice of the \({\omega _{i}^{0}},{\omega _{i}^{2}}\).
For the actual proof of the fractional case, we can now proceed via the so-called interpolation property for fractional Sobolev spaces (cf. [33, Th. 1.1.6], [43, Th. 6.30], [9, Def. 2.4.1], [46, Sect. 2.4.1]):
Lemma 8
Let \({\Omega }_{1},{\Omega }_{2}\subseteq {\mathbb R}^{d}\) be sets with a smooth or without a boundary. Let r = 𝜃r1 + (1 − 𝜃)r2 for 𝜃 ∈ (0, 1) and reals \(0\le r_{1}\le r_{2}<\infty \) and let ϱ = 𝜃ϱ1 + (1 − 𝜃)ϱ2 for reals \(0\le {\varrho }_{1}\le {\varrho }_{2}<\infty \). If \({\mathrm {\Lambda }}:{\mathrm H}^{r_{i}}({\Omega }_{1})\mapsto {\mathrm H}^{{\varrho }_{i}}({\Omega }_{2})\) is bounded for i = 1, 2, then \({\mathrm {\Lambda }}:{\mathrm H}^{r}({\Omega }_{1})\mapsto {\mathrm H}^{{\varrho }}({\Omega }_{2})\) is bounded as well and
Now we apply the interpolation property to \({\mathrm {\Lambda }}={\mathrm {T\!\!{}r}}_{{\mathbb M}}\nabla _{{\mathrm N}}{\mathrm Q}_{h} {\mathrm {E\!\!{}x}}_{{\mathrm N}} \). First, we choose suitable integers r1 = ⌊r⌋, r2 = ⌈r⌉. Then we choose 0 < 𝜃 < 1 such that r = 𝜃r1 + (1 − 𝜃)r2. We obtain for ϱ1 = ϱ2 = 0 in the interpolation property for the operator \({\mathrm {T\!\!{}r}}_{{\mathbb M}}\nabla _{{\mathrm N}}{\mathrm Q}_{h} {\mathrm {E\!\!{}x}}_{{\mathrm N}} \) the relation
The presence of the operation “\(\circ {}\varphi _{i}\)” therein does no harm, as we can always demand that an arbitrary \(g\in {\mathrm H}^{ r }({\omega _{i}^{2}})\) has the form \(g=\tilde g\circ \varphi _{i}\). This gives the desired fractional relation.
1.4.4 A.4.4 Corollary 1
This corollary deals with the fact that it is also important to have approximation results just for the approximation of the normal derivatives in the ambient space. Effectively, we have already provided most of the necessary steps in the proof of Appendix 6.4.2, but we recapitulate the relevant arguments here briefly: For any tubular neighbourhood \(\mathbb {U}_{h}({\mathbb M})\) that is sufficiently narrow and \(r\in {\mathbb N}\) we have the relation
Therein, \({\mathrm C}_{h}(\mathbb {U}_{h}({\mathbb M}))\) is the set of all cells that intersect \(\mathbb {U}_{h}({\mathbb M})\), clearly also a superset of the cells \({\mathrm C}_{h}({\mathbb M})\) that intersect \({\mathbb M}\) itself. The first inequality in the second line is due to Lemma 7 and the same summation argument as in (A.4.7), while the second inequality is due to Lemma 6. To deduce the fractional relation with \(f\in {\mathrm H}^{r}({\mathbb M})\) for \(r\notin {\mathbb N}\), we just have to apply the interpolation property from Lemma 8 once more.
Rights and permissions
About this article
Cite this article
Maier, LB. Ambient residual penalty approximation of partial differential equations on embedded submanifolds. Adv Comput Math 46, 62 (2020). https://doi.org/10.1007/s10444-020-09802-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10444-020-09802-1
Keywords
- Spline approximation
- Partial differential equation
- Collocation
- Numerical solution of PDE
- Elliptic equations
- Penalty method
- Manifold
- Closest point representation
- Multivariate spline