Abstract
In the present work, we consider weakly-singular integral equations arising from linear second-order elliptic PDE systems with constant coefficients, including, e.g. linear elasticity. We introduce a general framework for optimal convergence of adaptive Galerkin BEM. We identify certain abstract conditions for the underlying meshes, the corresponding mesh-refinement strategy, and the ansatz spaces that guarantee that the weighted-residual error estimator is reliable and converges at optimal algebraic rate if used within an adaptive algorithm. These conditions are satisfied, e.g. for discontinuous piecewise polynomials on simplicial meshes as well as certain ansatz spaces used for isogeometric analysis. Technical contributions include the localization of (non-local) fractional Sobolev norms and local inverse estimates for the (non-local) boundary integral operators associated to the PDE system.
1. Introduction
1.1. State of the art
For the Laplace model problem, adaptive boundary element methods (BEM) using (dis)continuous piecewise polynomials on triangulations have been intensively studied in the literature. In particular, optimal convergence of adaptive mesh-refining algorithms has been proved for polyhedral boundaries [Citation1–3] as well as smooth boundaries [Citation4]. The work [Citation5] allows to transfer these results to piecewise smooth boundaries; see also the discussion in the review article [Citation6]. In [Citation7], these results have been generalized to the Helmholtz problem. In recent years, we have also shown optimal convergence of adaptive isogeometric BEM (IGABEM) using one-dimensional splines for the 2D Laplace problem [Citation8,Citation9]. However, the important case of 3D IGABEM remained open. Moreover, a generalization to other PDE operators is highly non-trivial (see (Equation6(6) (6) ) below), but especially linear elasticity is of great interest in the context of isogeometric analysis.
In [Citation10], we have considered isogeometric finite element methods (IGAFEM). We have derived an abstract framework which guarantees that, first, the classical residual FEM error estimator is reliable, and second, the related adaptive algorithm yields optimal convergence; see [Citation10, Section 2 and 4]. We then showed that, besides standard FEM with piecewise polynomials, this abstract framework covers IGAFEM with hierarchical splines (see [Citation10, Section 3 and 5]) as well as IGAFEM with analysis suitable T-splines (see the recent work [Citation11]).
The aim of the present work is to develop such an abstract framework also for BEM, which is mathematically much more demanding than FEM. In ongoing research [Citation12], we aim to show that this framework covers, besides standard discretizations with piecewise polynomials, also IGABEM with hierarchical splines resp. T-splines.
To this end, the present work focusses on weakly-singular integral equations. For a given Lipschitz domain with compact boundary and right-hand side , we consider (1) (1) Here, the fundamental solution G stems from an elliptic PDE operator (2) (2) where the coefficients are constant for some fixed dimension .
1.2. Outline & contributions
In Section 2, we fix some general notation, recall Sobolev spaces on the boundary, and precisely state the considered problem. Section 3 can be paraphrased as follows. We formulate an adaptive algorithm (Algorithm 3.3) of the form (3) (3) driven by some weighted-residual a posteriori error estimator (see (Equation4(4) (4) ) below) in the frame of conforming Galerkin BEM. The algorithm particularly generates meshes , BEM solutions in associated nested ansatz spaces , and error estimators for all . We formulate five Assumptions (M1)–(M5) on the underlying meshes (Section 3.1), five Assumptions (R1)–(R5) on the mesh refinement (Section 3.2), and six Assumptions (S1)–(S6) on the BEM spaces (Section 3.3). First, these assumptions are sufficient to guarantee that the a posteriori error estimator associated with the BEM solution is reliable, i.e. there exists a constant such that (4) (4) where denotes the local mesh-size function and is the surface gradient. Second, Theorem 3.3 states that Algorithm 3.3 leads to linear convergence at optimal algebraic rate with respect to the number of mesh elements. In Theorem 3.7, we briefly note that the introduced conditions have already been implicitly proved for standard discretizations with piecewise polynomials on conforming triangulations. Moreover, we mention expected applications to adaptive IGABEM on quadrilateral meshes in Remark 3.8.
Section 4 is devoted to the proof of Theorem 3.3. To prove reliability (Equation4(4) (4) ), we use a localization argument (Proposition 4.2), which generalizes earlier works [Citation13,Citation14] for standard discretizations. More precisely, we prove that (5) (5) for all that are -orthogonal onto the ansatz space corresponding to some mesh , where is independent of v and denotes the patch of . Finally, the proof of reliability (Equation4(4) (4) ) requires the derivation of a local Poincaré-type inequality (Proposition 4.9). In Remark 4.10, we note that one obtains at least plain convergence if Algorithm 3.3 is steered by the so-called Faermann estimator which is reliable and efficient. To prove linear convergence at optimal rate for the weighted-residual estimator (Equation4(4) (4) ), we show that the assumptions of Section 3 imply the axioms of adaptivity [Citation6]. The latter are properties for abstract mesh-refinements and abstract error estimators, which automatically yield the desired convergence result. In contrast to [Citation1,Citation3] which (implicitly) verify the axioms of adaptivity only for the Laplace problem, our analysis allows for general PDE operators (Equation2(2) (2) ). The crucial step is the generalization (Proposition 4.13) of the non-trivial local inverse inequality for the non-local boundary integral operator : With the help of a Caccioppoli-type inequality (Lemma 4.12), we prove that there exists a constant such that (6) (6) see [Citation5] for standard BEM discretizations of the Laplacian. Similar estimates hold also for the other boundary integral operators related to (Equation2(2) (2) ), namely the double-layer integral operator , its adjoint , and the hypersingular integral operator . These are stated and proved in Appendix B; again we refer to [Citation5] for standard BEM discretizations of the Laplacian. In each case, ellipticity of the PDE operator is not required for the inverse inequalities.
While the present work focusses on the numerical analysis aspects only, we refer to the literature (see, e.g. [Citation7,Citation15–18]) for numerical experiments for the Laplace problem, the Helmholtz problem, and linear elasticity.
2. Preliminaries
In this section, we fix some general notation, recall Sobolev spaces on the boundary, and precisely state the considered problem. Throughout the work, let for be a bounded Lipschitz domain as in [Citation19, Definition 3.28] and its boundary.
2.1. General notation
Throughout and without any ambiguity, denotes the absolute value of scalars, the Euclidean norm of vectors in , as well as the Hausdorff measure of any d-dimensional set in . Let denote the open ball around x with radius . For , let . Moreover, let , and . We write to abbreviate with some generic constant C>0, which is clear from the context. Moreover, abbreviates . Throughout, mesh-related quantities have the same index, e.g. is the ansatz space corresponding to the mesh . The analogous notation is used for meshes , , , etc.
2.2. Sobolev spaces
For , we define the Hilbert spaces as in [Citation19, page 99] by use of Bessel potentials on and liftings via bi-Lipschitz mappingsFootnote1 that describe Γ. For , it holds that with equivalent norms. We thus may define .
For , any measurable subset , and all , we define the associated Sobolev–Slobodeckij norm (7) (7) It is well known that provides an equivalent norm on ; see, e.g. [Citation20, Lemma 2.19] and [Citation19, Theorem 3.30 and page 99] for and [Citation18, Theorem 2.28] for . Here, denotes the usual (weak) surface gradient which can be defined for almost all as follows. Since Γ is a Lipschitz boundary, there exist an open cover in of Γ such that each can be parametrized by a bi-Lipschitz mapping , where is an open set. By Rademacher's theorem, is almost everywhere differentiable. The corresponding Gram determinant is almost everywhere positive. Moreover, by definition of the space , implies that . With the weak derivative , we can hence define (8) (8) This definition does not depend on the particular choice of the open sets and the corresponding parametrizations ; see, e.g. [Citation18, Theorem 2.28]. With (Equation8(8) (8) ), we immediately obtain the chain rule (9) (9) For , is a realization of the dual space of ; see [Citation19, Theorem 3.30 and page 99]. With the duality bracket , we define an equivalent norm (10) (10) Moreover, we abbreviate (11) (11) [Citation19, page 76] states that the inclusion for is continuous and dense. In particular, form a Gelfand triple in the sense of [Citation21, Section 2.1.2.4] for all , where is interpreted as function in via (12) (12) Here, is the usual complex scalar product on .
So far, we have only dealt with scalar-valued functions. For , , , we define . If , and is an arbitrary measurable set, we define and analogously. With the definition (13) (13) it holds that . Note that with can be identified with the dual space of , where we set (14) (14) As before, we abbreviate (15) (15) and set (16) (16) The spaces can also be defined as trace spaces or via interpolation, where the resulting norms are always equivalent with constants depending only on the dimension d and the boundary Γ. More details and proofs are found, e.g. in the monographs [Citation19–21].
2.3. Continuous problem
We consider a general second-order linear system of PDEs (17) (17) where the coefficients are constant for some fixed dimension . We suppose that . Moreover, we assume that is elliptic on in the sense of the Lax–Milgram lemma, i.e. the sesquilinear form (18) (18) satisfies that (19) (19) of the matrices in the sense of [Citation19, page 119]. Here, the standard complex scalar product on is denoted by .
Let be a corresponding (matrix-valued) fundamental solution in the sense of [Citation19, page 198], i.e. a distributional solution of , where δ denotes the Dirac delta distribution. For , we define the single-layer operator as (20) (20) According to [Citation19, page 209 and 219–220] and [Citation22, Corollary 3.38], this operator can be extended for arbitrary ] to a bounded linear operator (21) (21) [Citation19, Theorem 7.6] states that is always coercive, i.e. elliptic up to some compact perturbation. We assume that it is elliptic even without perturbation, i.e. (22) (22) This is particularly satisfied for the Laplace problem or for the Lamé problem, where the case d = 2 requires an additional scaling of the geometry Ω; see, e.g. [Citation20, Chapter 6]. Moreover, the sesquilinear form is continuous due to (Equation21(21) (21) ), i.e. it holds with that (23) (23) Given a right-hand side , we consider the boundary integral equation (24) (24) Such equations arise from (and are even equivalent to) the solution of Dirichlet problems of the form in Ω with u = g on Γ for some ; see, e.g. [Citation19, page 226–229] for more details. The Lax–Milgram lemma provides existence and uniqueness of the solution of the equivalent variational formulation of (Equation24(24) (24) ) (25) (25) In particular, we see that is an isomorphism. In the Galerkin boundary element method, the test space is replaced by some discrete subspace . Again, the Lax–Milgram lemma guarantees existence and uniqueness of the solution of the discrete variational formulation (26) (26) Moreover, can in fact be computed by solving a linear system of equations. Note that (Equation21(21) (21) ) implies that for arbitrary . The additional regularity instead of is only needed to define the residual error estimator (Equation37(37a) (37a) ) below. For a more detailed introduction to boundary integral equations, the reader is referred to the monographs [Citation19–21].
3. Axioms of adaptivity (revisited)
The aim of this section is to formulate an adaptive algorithm (Algorithm 3.3) for conforming BEM discretizations of our model problem (Equation24(24) (24) ), where adaptivity is driven by the residual a posteriori error estimator (see (Equation37(37a) (37a) ) below). We identify conditions for the underlying meshes, the mesh-refinement, as well as the boundary element spaces which ensure that the residual error estimator is reliable and fits into the general framework of [Citation6] and which hence guarantee optimal convergence behavior of the adaptive algorithm. We mention that we have already identified similar (but not identical) conditions for the finite element method in [Citation10, Section 3]. The main result of this work is Theorem 3.3 which is proved in Section 4.
3.1. Meshes
Throughout, is a mesh of the boundary of the bounded Lipschitz domain in the following sense:
is a finite set of compact Lipschitz domains on Γ, i.e. each element T has the form , where is a compactFootnote2 Lipschitz domain in and is bi-Lipschitz.
covers Γ, i.e. .
For all with , the intersection has -dimensional Hausdorff measure zero.
We suppose that there is a countably infinite set of admissible meshes. In order to ease notation, we introduce for the corresponding mesh-width function (27) (27) For , we define the patches of order inductively by (28) (28) The corresponding set of elements is (29) (29) If for some , we simply write and . For , we define and . To abbreviate notation, the index q = 1 is omitted, e.g. and .
We assume the existence of constants such that the following assumptions are satisfied for all :
(M1) | Bounded element patch: The number of elements in a patch is uniformly bounded, i.e. | ||||
(M2) | Local quasi-uniformity: Neighboring elements have comparable diameter, i.e. | ||||
(M3) | Shape-regularity: It holds that | ||||
(M4) | Elements lie in the center of their patches: It holdsFootnote3 that | ||||
(M5) | Local seminorm estimate: For all , it holds that |
The following proposition shows that (M5) is actually always satisfied. However, in general the multiplicative constant depends on the shape of the point patches. The proof is inspired by [Citation23, Proposition 2.2], where an analogous assertion for norms instead of seminorms is found. For and d = 2, we have already shown the assertion in the recent own work [Citation16, Lemma 4.5]. For polyhedral domains Ω with triangular meshes, it is proved in [Citation24, Proposition 3.3] via interpolation techniques. A detailed proof for our setting is found in [Citation17, Proposition 5.2.2], where we essentially follow the proof of [Citation16, Lemma 4.5].
Proposition 3.1
Let be a bounded and connected Lipschitz domain and be bi-Lipschitz, i.e. there exists a constant such that (30) (30) Then, for arbitrary there exists a constant such that (31) (31) The constant depends only on the dimension d, σ, the set and .
3.2. Mesh refinement
For and an arbitrary set of marked elements , we associate a corresponding refinement with , i.e. at least the marked elements are refined. Moreover, we suppose for the cardinalities that if and else. Let be the set of all such that there exist meshes and marked elements with and . We assume that there exists a fixed initial mesh with .
We suppose that there exist and such that all meshes satisfy for arbitrary marked elements with corresponding refinement , the following elementary properties (R1)–(R3):
(R1) | Son estimate: One step of refinement leads to a bounded increase of elements, i.e. | ||||
(R2) | Father is union of sons: Each element is the union of its successors, i.e. | ||||
(R3) | Reduction of sons: Successors are uniformly smaller than their father, i.e. |
By induction and the definition of , one easily sees that (R2)–(R3) remain valid if is an arbitrary mesh in . In particular, (R2)–(R3) imply that each refined element is split into at least two sons, wherefore (32) (32) Besides (R1)–(R3), we suppose the following less trivial requirements (R2)–(R3) with generic constants :
(R4) | Closure estimate: Let be a sequence in such that with some for all . Then, it holds that | ||||
(R5) | Overlay property: For all , there exists a common refinement which satisfies the overlay estimate |
3.3. Boundary element space
With each , we associate a finite dimensional space of vector valued functions (33) (33) Let be the corresponding Galerkin approximation of from (Equation24(24) (24) ), i.e. (34) (34) We note the Galerkin orthogonality (35) (35) as well as the resulting Céa type quasi-optimality (36) (36) We assume the existence of constants , , and such that the following properties (S1)–(S4) hold for all :
(S1) | Inverse inequality: For all , it holds that | ||||
(S2) | Nestedness: For all , it holds that | ||||
(S3) | Local domain of definition: For all , , and , it holds that | ||||
(S4) | Componentwise local approximation of unity: For all and all , there exists some with such that only the jth component does not vanish, i.e. and |
Remark 3.2
Clearly, (S4) is in particular satisfied if is a product space, i.e. , and each component satisfies (S4).
Besides (S1)–(S4), we suppose that there exist constants as well as such that for all and , there exists a linear Scott–Zhang-type operator with the following properties (S5)–(S6):
(S5) | Local projection property. Let from (S3). For all and with , it holds that | ||||
(S6) | Local -stability. For all and , it holds that |
3.4. Error estimator
Let . Due to the regularity assumption , the mapping property (Equation21(21) (21) ), and , it holds that for all . This allows to employ the weighted-residual a posteriori error estimator (37a) (37a) where the local refinement indicators read (37b) (37b) The latter estimator goes back to the works [Citation25,Citation26], where reliability (Equation42(42) (42) ) is proved for standard 2D BEM with piecewise polynomials on polygonal geometries, while the corresponding result for 3D BEM is found in [Citation15].
3.5. Adaptive algorithm
We consider the following concrete realization of the abstract algorithm from [Citation6, Algorithm 2.2].
3.6. Optimal convergence
Define (39) (39) and for all s>0 (40) (40) We say that the solution lies in the approximation class s with respect to the estimator if (41) (41) By definition, implies that the error estimator on the optimal meshes decays at least with rate . The following main theorem states that each possible rate s>0 is in fact realized by Algorithm 3.1. We stress that the range of possible s>0 depends, in particular, on the chosen admissible meshes . The proof is given in Section 4. It essentially follows by verifying the axioms of adaptivity from [Citation6]. Such an optimality result was first proved in [Citation3] for the Laplace operator on a polyhedral domain Ω. As ansatz space, they considered piecewise constants on shape-regular triangulations. [Citation1] in combination with [Citation5] extends the assertion to piecewise polynomials on shape-regular curvilinear triangulations of some piecewise smooth boundary Γ. Independently, [Citation4] proved the same result for globally smooth Γ and general self-adjoint and elliptic boundary integral operators.
Theorem 3.4
Let be the sequence of meshes generated by Algorithm 3.3. Then, there hold the following assertions (i)–(iii):
Suppose (M1)–(M5), and (S4). Then, the residual error estimator satisfies reliability, i.e. there exists a constant such that (42) (42)
Suppose (M1)–(M5), (R2)–(R3), (S1)–(S2) and (S4). Then, for arbitrary and , the estimator converges linearly, i.e. there exist constants and such that (43) (43)
Suppose (M1)–(M5), (R1)–(R5), and (S1)–(S6). Then, there exists a constant such that for all and the estimator converges at optimal rate, i.e. for all s>0 there exist constants such that (44) (44) where the lower bound requires only (R1) to hold.
Remark 3.5
If the sesquilinear form is Hermitian, then , and are independent of ; see Remark 4.14 below.
Remark 3.6
Let be an open subset of and let denote the operator that extends a function defined on to a function on Γ by zero. We define the space of restrictions endowed with the quotient norm and its dual space . According to [Citation5, Section 2.1], can be extended to an isometric operator . Then, one can consider the integral equation (45) (45) where . In the literature, such problems are known as screen problems; see, e.g. [Citation21, Section 3.5.3]. Theorem 3.3 holds analogously for the screen problem (Equation45(45) (45) ). Indeed, the works [Citation1,Citation3–5] cover this case as well. However, to ease the presentation, we focus on closed boundaries .
Remark 3.7
(a) Let us additionally assume that contains all componentwise constant functions, i.e. (46) (46) Then, under the assumption that as , one can show that . To see this, recall that is continuously and densely embedded in which is itself continuously and densely embedded in . For and arbitrary , let with . We abbreviate the projection operator for all . For all , the projection property (S5) in combination with our additional assumption (Equation46(46) (46) ), the triangle inequality, and the local -stability (S6) show that With this, the Poincaré-type inequality from Lemma 4.6 below, and (M1)–(M3), we see that Summing over all elements, we obtain that With (M1)–(M4), Proposition 4.1 and Lemma 4.8 from below prove that ≲. Overall, this shows that Since and ϵ was arbitrary, this concludes the proof.
(b) The latter observation allows to follow the ideas of [Citation27] and to show that the adaptive algorithm yields convergence provided that the sesquilinear forms as well as are only coercive, i.e. elliptic up to some compact perturbation and that the continuous problem is well-posed; see also the introductory text of Section 4.4. This includes, e.g. adaptive BEM for the Helmholtz equation; see [Citation20, Section 6.9]. For details, the reader is referred to [Citation7,Citation27].
3.7. Application to BEM with piecewise polynomials on triangulations
For d = 2, 3, we fix the reference simplex as the closed convex hull of the d vertices . The convex hull of any d−1 vertices is called facet. A set of subsets of Γ is called κ-shape regular triangulation if the following properties (i)–(v) are satisfied:
is a finite set of elements T of the form , where is a bi-Lipschitz mapping whose Lipschitz constants are bounded from above by κ.
covers Γ, i.e. .
There are no hanging nodes in the sense that the intersection of any with is either empty or a common facet, i.e. for some facets f and of .
The parametrizations of neighboring elements are compatible in the sense that for all nodes z (i.e. images of the under an element map ), there exists an interval for d = 2 and a convex polygonal for d = 3 respectively as well as a bijective and bi-Lipschitz continuous mapping such that is affine for all .
If d = 2, is locally-quasi uniform in the sense that for all with .
Up to the fact that we allow to be bi-Lipschitz instead of , this definition is slightly stronger than [Citation5, Definition 2.4]. The property (iii) stems from [Citation21, Assumption 4.3.25] and is stronger than the corresponding assumption [Citation5, Definition 2.4 (iii)]. Further, (i) implies [Citation5, Definition 2.4 (v)], i.e. for all , there holds with the extremal eigenvalues and that (47) (47) see, e.g. [Citation24, (3.26)–(3.27)] or [Citation17, Lemma 5.2.1].
Let be a -shape regular triangulation. For d = 2, we define as in [Citation28] via 1D-bisection in the parameter domain. For d = 3, we define as in [Citation29] via newest vertex bisection in the parameter domain. In particular, all corresponding refinements are again κ-shape regular triangulations with some fixed κ depending on . We also note that the number of different in (iv) is uniformly bounded, i.e. there exist only finitely many reference node patches. Finally, let be a fixed polynomial order. For each , we associate the space of (transformed) piecewise polynomials (48) (48) For this concrete setting, we already pointed out that [Citation1] in combination with [Citation5] proved linear convergence (Equation43(43) (43) ) at optimal rate (Equation44(44) (44) ) if is the Laplace operator. The following theorem generalizes this result to arbitrary as in Section 2.3.
Theorem 3.8
Piecewise polynomials on κ-shape regular triangulations satisfy the abstract properties (M1)–(M5), (R1)–(R5), and (S1)–(S6), where the constants depend only on the dimension D, the regularity constant κ, the initial mesh and the polynomial order p. By Theorem 3.4, this implies reliability (Equation42(42) (42) ) of the error estimator and linear convergence (Equation43(43) (43) ) at optimal rate (Equation44(44) (44) ) for the adaptive strategy from Algorithm 3.3.
Proof.
The elementary mesh properties (M1)–(M3) are verified in [Citation5, Section 2.3]. (M4) is stated in [Citation5, Section 4.1]. (M5) follows from Proposition 3.1 together with the fact that there are only finitely many reference node patches.
For d = 2, the son estimate (R1) is clearly satisfied with . For d = 3, it is well known that NVB satisfies (R1) with . Further, (R1) holds true by definition. Reduction of sons (R3) is obviously satisfied in the parameter domain, i.e. for all with . Since is bi-Lipschitz, this property transfers to the physical domain, i.e. , where depends only on κ; see, e.g. [Citation17, Section 4.5.3] for details. For d = 2, (R4)–(R5) are found in [Citation28, Theorem 2.3]. For d = 3, the closure estimate (R4) is proved in [Citation30, Theorem 2.4], [Citation31, Theorem 6.1], or [Citation32, Theorem 2], where the latter result avoids any additional assumption on . The overlay property is proved in [Citation31, Proof of Lemma 5.2] or [Citation33, Section 2.2].
The inverse inequality (S1) for piecewise polynomials on the boundary is proved, e.g. in [Citation5, Lemma A.1]. Nestedness (S2) is trivially satisfied. Also (S3) is trivially satisfied with . Clearly, (S4) holds with for and , where denotes the indicator function on T. Finally, for , we define with the elementwise -orthogonal projection (49) (49) This definition immediately yields (S5)–(S6) with .
Remark 3.8
We mention that Theorem 3.8 is also valid if d = 2 and is chosen as set of (transformed) splines which are piecewise polynomials with certain differentiability conditions at the break points. The required properties are (implicitly) verified in [Citation16]. As in [Citation10] (resp. [Citation11]), where we have verified the abstract FEM framework of [Citation10] for IGAFEM with hierarchical splines [Citation34] and the mesh-refinement from [Citation10] (resp. T-splines with the mesh-refinement from [Citation35]), the verification of the present abstract BEM framework for 3D IGABEM will be addressed in the future work [Citation12].
4. Proof of Theorem 3.4
In the following sections, we prove Theorem 3.4. Reliability (Equation42(42) (42) ) is treated explicitly in Section 4.2. It follows immediately from an auxiliary result on the localization of the Sobolev–Slobodeckij norm which is investigated in Section 4.1. To prove Theorem 3.4 (ii)–(iii), we verify the following abstract properties (E1)–(E4) for the error estimator. Together with (R1), the closure estimate (R4), and the overlay property (R5), these already imply linear convergence of the estimator at optimal algebraic rate; see [Citation6].
There exist , , , , , , and such that there hold:
(E1) | Stability on non-refined elements: For all and , it holds that | ||||
(E2) | Reduction on refined elements: For all and , it holds that | ||||
(E3) | General quasi-orthogonality: It holds that and for all , the sequence from Algorithm 3.3 satisfies that | ||||
(E4) | Discrete reliability: For all and all , there exists with such that |
4.1. Localization of the Sobolev–Slobodeckij norm
Let . In contrast to the integer case, for , the norm is not additive in the sense that Although the upper bound ‘≲’ is in general false (see [Citation36, Section 3]), the lower bound ‘⪆’ can be proved elementarily for arbitrary .
Proposition 4.1
Let and . Then, (M1) implies the existence of a constant such that for any there holds that (50) (50) The constant depends only on the constant from (M1)
Proof.
With the abbreviation (51) (51) (M1) shows that This concludes the proof.
However, if one replaces the elements T by some overlapping patches, then also the converse inequality is satisfied for functions which are -orthogonal to the ansatz space .
Proposition 4.2
Let and . Then, (M1)–(M4) and (S4) imply the existence of a constant such that for any which satisfies that for all and all where are the functions from (S4), it holds that (52) (52) The constant depends only on the dimension , Γ, and the constants from (M1)–(M4) and (S4).
With this result, one can immediately construct a reliable and efficient error estimator, namely the so-called Faermann estimator; see Remark 4.10. For d = 2, the result of the proposition goes back to [Citation13], where is chosen as space of splines transformed via the arclength parametrization onto the one-dimensional boundary. In the recent own works [Citation37], we generalized the assertion to rational splines, where we could also drop the restriction that γ is the arclength parametrization. For d = 3, [Citation14] proved the result for discrete spaces which contain certain (transformed) polynomials of degree on a curvilinear triangulation of Γ. Our proof of Proposition 4.2 is inspired by [Citation14]. The key ingredient is the assumption (S4) which is exploited in Lemma 4.7. Before proving Proposition 4.2, we provide an easy corollary which is the key ingredient for the proof of reliability (Equation42(42) (42) ).
Corollary 4.3
Let . Then, (M1)–(M5) and (S4) imply the existence of a constant such that for any which satisfies that for all and all where are the functions from (S4), it holds that (53) (53) The constant depends only on the dimension d, Γ, as well as the constants from (M1)–(M5) and (S4).
To prove Proposition 4.2, we start with the following basic estimate, which is proved in [Citation38, Lemma 8.2.4] or in [Citation17, Lemma 5.3.1].
Lemma 4.4
For all there is a constant such that for all and all there holds that (54) (54) The constant depends only on the parameter λ, the dimension d, and Γ.
The following lemma is the first step towards the localization of the norm for certain functions . In [Citation14, Lemma 3.1], this result is stated for triangular meshes. The elementary proof extends to our situation; see also [Citation17, Lemma 5.3.2] for details.
Lemma 4.5
Let and . Then, (M4) implies the existence of a constant C>0 such that for all it holds that (55) (55) The constant C depends only on the dimension d, and the constant from (M4).
It remains to control the second summand in (Equation55(55) (55) ). To this end, we need the following elementary Poincaré-type inequality of [Citation13, Lemma 2.5].
Lemma 4.6
For any and any measurable there holds for all that (56) (56)
We start to estimate the second summand in (Equation55(55) (55) ).
Lemma 4.7
Let and . Then, (M1)–(M3) and (S4) imply the existence of a constant C>0 such that for all with for all where are the functions from (S4), it holds that (57) (57) where is the constant from (S4). The constant C depends only on the dimension d, and the constants from (M1)–(M3) and (S4).
Proof.
We prove (Equation57(57) (57) ) for each component of v, where . Then, squaring and summing up all components, we conclude the proof. (S4) and Lemma 4.6 show that (58) (58) Now, we apply the orthogonality and (S4) to get for the second summand that Inserting this in (Equation58(58) (58) ) gives that (59) (59) With (S4) and (M1)–(M3), we see that . Further, (S4) implies that . Inserting this in (Equation59(59) (59) ) and using again (S4), we derive that Altogether, this concludes the proof.
The following lemma allows us to further estimate the term of (Equation57(57) (57) ).
Lemma 4.8
Let and . Then, (M1)–(M4) imply the existence of a constant such that for all and all there holds that (60) (60) The constant depends only on the dimension and the constants from (M1)–(M4).
Proof.
Without loss of generality, we may assume that D = 1. We prove the assertion in two steps.
Step 1: Let be a chain of elements in with for and if , where . We set for and prove by induction on m that there exists a constant which depends only on and (M2)–(M4), such that (61) (61) For m = 1, (Equation61(61) (61) ) with even holds with equality. Thus the induction hypothesis reads: for all and for any chain of elements in , it holds that (62) (62) Let with for and for i = m−1. For all , we abbreviate . The definition (Equation7(7) (7) ) of the Sobolev–Slobodeckij seminorm shows that With the induction hypothesis (Equation62(62) (62) ), it remains to estimate . First, we note that for , it holds that (63) (63) Moreover, (M4) shows that . Since , (M2) shows . Hence, we can proceed the estimate of (Equation63(63) (63) ) This implies that Note that by (M2)–(M3). Together with the induction hypothesis (Equation62(62) (62) ), this concludes the induction step.
Step 2: We come to the assertion itself. By definition, we have that Let . First, we suppose that . Then, there exists a chain as in Step 1 with and . Step 1 proves that If , the same estimate holds true. Since the number of is uniformly bounded by a constant, which depends only on the constant of (M1) and q, this estimate concludes the proof.
With the property (M5), one immediately derives the following Poincaré inequality.
Proposition 4.9
Let and . Then, (M1)–(M5) and (S4) imply the existence of a constant such that for all which satisfy that for all , where are the functions from (S4), it holds that (64) (64) where is the constant from (S4). The constant depends only on the dimension and the constants from (M1)–(M5) and (S4).
Proof.
We apply Lemma 4.7 and Lemma 4.8 to see that For with , we fix some point . With (M5), we continue our estimate (M1)–(M3) imply that on , and that the last term of the latter estimate can be bounded from above (up to a multiplicative constant) by . This concludes the proof.
With all the preparations, we can finally prove the main result of this section.
Proof
Proof of Proposition 4.2
Together with (M3), Lemma 4.5 proves that It remains to estimate the second sum. With Lemma 4.7 and Lemma 4.8, we see that (65) (65) If and with , then and . Plugging this into (Equation65(65) (65) ) shows that and (see (M1)) concludes the proof.
4.2. Reliability (42)
Let . Recall that is an isomorphism. Due to Galerkin orthogonality (Equation35(35) (35) ), Corollary 4.3 leads to (66) (66)
Remark 4.10
Proposition 4.1 and Proposition 4.2 show that (67) (67) This is even true for arbitrary without the additional restriction . In particular, (68) (68) provides a local error indicator. The corresponding error estimator is often referred to as Faermann estimator. In BEM, it is the only known estimator which is reliable and efficient (without further assumptions as, e.g. the saturation assumption [Citation39, Section 1]). Obviously, one could replace the residual estimator in Algorithm 3.3 by . However, due to the lack of an h-weighting factor, it is unclear whether the reduction property (E2) of Section 4.2 is satisfied. [Citation24, Theorem 7] proves at least plain convergence of even for if one uses piecewise constants on affine triangulations of Γ as ansatz space. The proof immediately extends to our current situation, where the assumptions (M1)–(M5), (R2)–(R3), and (S1)–(S2) are employed. The key ingredient is the construction of an equivalent mesh-size function which is contractive on each element which touches a refined element, i.e. there exists a uniform constant such that (69) (69) The existence of such a mesh-size function is proved in [Citation6, Section 8.7] for shape-regular triangular meshes. The proof works verbatim for the present setting.
4.3. Convergence of
Nestedness (S2) ensures that is a closed subspace of and hence admits a unique Galerkin solution . Note that is also a Galerkin approximation of . Hence, the Céa lemma (Equation36(36) (36) ) with φ replaced by proves that as . In particular, we obtain that .
4.4. An inverse inequality for
In Proposition 4.13, we establish an inverse inequality for the single-layer operator . Throughout this section, neither the ellipticity of nor the ellipticity of are exploited (and we can drop these assumption here). Indeed, it is sufficient to assume that is only coercive. Then, the definitions and properties presented in Section 2.3 remain valid; see [Citation19, page 119]. For the Laplace operator , an inverse estimate as in Proposition 4.13 was already proved in [Citation3, Theorem 3.1] for shape-regular triangulations of a polyhedral boundary Γ. Independently, [Citation4] derived a similar result for globally smooth Γ and arbitrary self-adjoint and elliptic boundary integral operators. In [Citation5, Theorem 3.1], [Citation3, Theorem 3.1] is generalized to piecewise polynomial ansatz functions on shape-regular curvilinear triangulations. In particular, our Proposition 4.13 does not only extend these results to arbitrary general meshes as in Section 3.1, but is also completely novel for, e.g. linear elasticity. The proof follows the lines of [Citation5, Section 4]. We start with the following lemma, which was proved in [Citation40, Theorem 4.1] on shape-regular triangulations. With Lemma 4.6, the proof immediately extends to our situation; see also [Citation17, Lemma 5.3.11].
Lemma 4.11
For , let be the set of all functions whose D components are -piecewise constant functions on Γ. Let be the corresponding -projection. Then, (M1) and (M3) imply for arbitrary the existence of a constant C>0 such that (70) (70) The constant C depends only on the dimension D, the boundary and the constants from (M3).
In contrast to [Citation5], we cannot use the Caccioppoli-type inequality from [Citation41, Lemma 5.7.1] which is only shown for the Poisson problem there. Therefore, we prove the following generalization. For an open set and an arbitrary , we abbreviate and .
Lemma 4.12
Let , and be a weak solution of . Then, and there exists a constant C>0 such that (71) (71) The constant C depends only on the dimensions and the coefficients of the partial differential operator .
Proof.
By [Citation19, Theorem 4.16], there holds that for all , and the Sobolev embedding theorem proves that . In particular, u is a strong solution of on . To prove (Equation71(71) (71) ), let be an arbitrary constant vector, and define with the affine bijection , for . Since the coefficients of are constant and u is a strong solution, there holds for all with that (72) (72) We define the right-hand side as , i.e. This shows that is a strong (and thus weak) solution of a coercive system of second-order PDEs with smooth coefficients and smooth right-hand side. The application of [Citation19, Theorem 4.16] yields the existence of a constant , which depends only on and the coefficients of the matrices , such that (73) (73) Standard scaling arguments prove that Plugging this into (Equation73(73) (73) ), we obtain that (74) (74) We choose λ as the integral mean . The Cauchy–Schwarz inequality implies that Using this and the Poincaré inequality in (Equation74(74) (74) ), we see that Together with the fact that , this concludes the proof.
For the proof of the next proposition, we need the linear and continuous single-layer potential from [Citation19, Theorem 6.11] (75) (75) where U is an arbitrary bounded domain with . The single-layer operator is just the trace of , i.e. (76) (76) see [Citation19, page 219–220]. Indeed, for , [Citation19, page 201–202] states the following integral representation: (77) (77)
Proposition 4.13
Suppose (M1)–(M5). For let be a weight function which satisfies for some and all that (78) (78) Then, there exists a constant such that for all it holds that (79) (79) The constant depends only on (M1)–(M5), Γ, the coefficients of , and the admissibility constant α. The particular choice shows that (80) (80)
Proof.
The proof works essentially as in [Citation5, Section 4]. Therefore, we mainly emphasize the differences and refer to [Citation17, Proposition 5.3.15] for further details.
By (M4) and with the abbreviation and , there holds for all that . This provides us with an open covering of . We show that this is even locally finite in the sense that there exists a constant C>0 with for all : Let . Clearly, we may assume that . Choose with such that is minimal, and let with . If with , the triangle inequality yields that . By choice of , (M4) hence yields that . Thus, , and (M1) implies that (81) (81) We fix (independently of ) a bounded domain with for all . We define for the near-field and the far-field of by (82) (82) In the following five steps, the near-field and the far-field are estimated separately. The first two steps deal with the near-field, whereas the last three steps deal with the far-field.
Step 1: As in [Citation5, Lemma 4.1], one shows that for all , all -piecewise (componentwise) constant functions with satisfy that (83) (83) The proof of [Citation5] uses only (Equation81(81) (81) ) and the fact that with for the Laplacian fundamental solution G. However, according to [Citation19, Theorem 6.3 and Corollary 6.5] this fact is also valid for general coercive second-order PDEs with coefficients. Moreover, [Citation5] bounds only the -seminorm in (Equation83(83) (83) ), but the -norm can be bounded similarly due to (see again [Citation19, Theorem 6.3 and Corollary 6.5]).
Step 2: With Step 1, one shows as in [Citation5, Proposition 4.2] that and with (84) (84) In the proof, one applies the stability of (see (Equation21(21) (21) )) and (see (Equation75(75) (75) )). Moreover, the approximation property (Equation70(70) (70) ) is exploited by splitting and choosing . Note that [Citation5] only proves (Equation84(84) (84) ) with instead of .
Step 3: We consider the far-field. We set . According to [Citation19, Theorem 6.11], for all , the potential is a solution of the transmission problem where and denote the jump of the traces and the conormal derivatives respectively (see [Citation19, page 117] for a precise definition) across the boundary Γ. Twofold integration by parts that uses these jump conditions shows that weakly on . Since for all , Lemma 4.12 shows that with (85) (85) Note that [Citation5] proves (Equation85(85) (85) ) even without the term . Indeed, since the kernel of the Laplace operator contains all constants, [Citation5] employs a Poincaré inequality to bound by .
Step 4: With inequality (Equation85(85) (85) ) at hand, one can prove the following local far-field bound for the single-layer potential (86) (86) The proof works as in [Citation5, Lemma 4.4] and relies on a standard trace inequality on Ω, the Caccioppoli inequality (Equation85(85) (85) ) as well as the Besicovitch covering theorem. Note that in [Citation5], the estimates (Equation85(85) (85) ) and thus (Equation86(86) (86) ) even hold without the -norm of , since the Laplace problem is considered.
Step 5: Finally, (Equation81(81) (81) ), (Equation82(82) (82) ), (Equation84(84) (84) ), (Equation86(86) (86) ), and the stability of (see (Equation75(75) (75) )) easily lead to the far-field bound for (87) (87) For the simple proof, we refer to [Citation5, Proposition 4.5]. By definition (Equation82(82) (82) ), (Equation87(87) (87) ) together with (Equation84(84) (84) ) from Step 2 concludes the proof.
[Citation5] does not only treat the single-layer operator but also derives similar inverse estimates as in (Equation79(79) (79) ) for the double-layer operator, the adjoint double-layer operator, and the hyper-singular operator. With similar techniques as in Proposition 4.13, we will also generalize this result in Appendix B. However, we will indeed only need the inverse estimate (Equation80(80) (80) ) for the single-layer operator in the remainder of the paper.
4.5. Stability on non-refined elements (E1)
We show that the assumptions (M1)–(M5) and (S1)–(S2) imply stability (E1), i.e. the existence of such that for all , and all , it holds that (88) (88) In Section 4.6, we will fix the constant for defined in (E1) such that . The reverse triangle inequality and the fact that on prove that (S2) shows that . Therefore, the inverse inequalities from (S1) and (Equation80(80) (80) ) are applicable, which implies (Equation88(88) (88) ). The constant depends only on d, D, Γ, the coefficients of , and the constants from (M1)–(M5) and (S1).
4.6. Reduction on refined elements (E2)
We show that the assumptions (M1)–(M5), (R2)–(R3), and (S1)–(S2) imply reduction on refined elements (E2), i.e. the existence of and such that for all and all , it holds that (89) (89) With this, we can fix the constant for defined in (E1) as (90) (90) Let . First, we apply the triangle inequality Clearly, (R2)–(R3) show that and on ω. Thus we can proceed the estimate as follows: Since according to (S2), we can apply the inverse estimates (S1) and (Equation80(80) (80) ). Together with the Young inequality, we derive for arbitrary that Choosing sufficiently small, we obtain (Equation89(89) (89) ). The constant depends only on d, D, Γ, the coefficients of , and the constants from (M1)–(M5), (R2)–(R3), and (S1).
4.7. General quasi-orthogonality (E3)
According to [Citation6, Section 4.3], Section 4.3, Section 4.5, and Section 4.6 already imply estimator convergence . Therefore, reliability (Equation42(42) (42) ) implies error convergence . In particular, we obtain that . Recall that we have already fixed the constant in (Equation90(90) (90) ). We introduce the principal part of as the corresponding partial differential operator without lower-order terms (91) (91) According to [Citation19, Lemma 4.5], the principal part is coercive on . We denote its corresponding single-layer operator which can be defined as in Section 2.3 by . Our assumption easily implies that is self-adjoint; see, e.g. [Citation19, page 218]. With (Equation76(76) (76) ) and (EquationA14(B.14) (B.14) ) below, we particularly see the stability of for all . Thus the Rellich compactness theorem [Citation19, Theorem 3.27] implies that is compact. This yields that is an elliptic operator which can be written as the sum of a self-adjoint operator plus a compact operator . From [Citation27,Citation42], we thus derive the general quasi-orthogonality (E3) (see also [Citation17, Section 4.4.3] for all details), i.e. the existence of (92) (92) and such that (93) (93)
Remark 4.14
If the sesquilinear form is Hermitian, (Equation93(93) (93) ) follows from the Pythagoras theorem and norm equivalence Together with reliability (Equation42(42) (42) ), this proves (Equation93(93) (93) ) even for , and is independent of the sequence .
4.8. Discrete reliability (E4)
The proof of (E4) is inspired by [Citation3, Proposition 5.3] which considers piecewise constants on shape-regular triangulations as ansatz space. Under the assumptions (M1)–(M5), (Equation32(32) (32) ), and (S1)–(S6), we show that there exist such that for all and all , the subset (94) (94) satisfies that The last two properties are obvious with by validity of (M1) and (Equation32(32) (32) ). The first estimate is proved in three steps:
Step 1: For , let be the corresponding projection operator from (S5)–(S6). Ellipticity (Equation22(22) (22) ), nestedness (S2) of the ansatz spaces, and the definition (Equation34(34) (34) ) of the Galerkin approximations yield that (S3) shows that for any . Moreover, one easily sees that (95) (95) Hence, the local projection property (S5) of is applicable and proves that on . With , Lemma A.1 provides a smooth cut-off function with such that on , on , and , where the hidden constant depends only on d and (M1)–(M4). This leads to (96) (96) We bound the two terms and separately. Since is the dual space of , it holds that (97) (97) The Cauchy–Schwarz inequality shows that Since , it holds that . This together with the fact that on , the local -stability (S6) and (M1)–(M3) implies that With the inverse inequality (S1) applied to (see (S2)), the latter estimate implies that (98) (98) Plugging (Equation97(97) (97) )–(Equation98(98) (98) ) into (Equation96(96) (96) ) shows that (99) (99)
Step 2: Next, we deal with the first summand of (Equation99(99) (99) ). With and , this implies that (100) (100) By Galerkin orthogonality (Equation35(35) (35) ), we see that is -orthogonal to all functions of which includes in particular the functions from (S4). Hence, we can apply Proposition 4.9. Together with (M1)–(M3) and recalling (Equation94(94) (94) ), (Equation100(100) (100) ) proves that Step 3: It remains to consider the second summand of (Equation99(99) (99) ). Lemma 4.5 in conjunction with shape-regularity (M3) implies that We have already dealt with the second summand in Step 2 (see (Equation100(100) (100) )). For the first one, we fix again some for any . (M1)–(M3) and (M5) show that With the product rule and (EquationA2(A2) (A2) ), we continue our estimate Note that we have already dealt with the first summand in Step 2 (see (Equation100(100) (100) )). Finally, and (see (Equation94(94) (94) )) prove for the second one that With this, we conclude the proof of discrete reliability (E4). The constant depends only on , Γ, and the constants from (M1)–(M5) and (S1)–(S6).
Acknowledgments
The authors acknowledge support through the Austrian Science Fund (FWF) under grant J4379-N,P29096,W1245.
Disclosure statement
No potential conflict of interest was reported by the authors.
Additional information
Funding
Notes
1 For and , a mapping is bi-Lipschitz if it is bijective and γ as well as its inverse are Lipschitz continuous.
2 A compact Lipschitz domain is the closure of a bounded Lipschitz domain. For d = 2, it is the finite union of compact intervals with non-empty interior.
3 We use the convention .
References
- Feischl M, Führer T, Karkulik M, et al. Quasi-optimal convergence rates for adaptive boundary element methods with data approximation, part I: weakly-singular integral equation. Calcolo. 2014;51(4):531–562.
- Feischl M, Führer T, Karkulik M, et al. Quasi-optimal convergence rates for adaptive boundary element methods with data approximation, part II: hyper-singular integral equation. Electron Trans Numer Anal. 2015;44:153–176.
- Feischl M, Karkulik M, Markus Melenk J, et al. Quasi-optimal convergence rate for an adaptive boundary element method. SIAM J Numer Anal. 2013;51(2):1327–1348.
- Gantumur T. Adaptive boundary element methods with convergence rates. Numer Math. 2013;124(3):471–516.
- Aurada M, Feischl M, Führer T, et al. Local inverse estimates for non-local boundary integral operators. Math Comp. 2017;86(308):2651–2686.
- Carstensen C, Feischl M, Page M, et al. Axioms of adaptivity. Comput Math Appl. 2014;67(6):1195–1253.
- Bespalov A, Betcke T, Haberl A, et al. Adaptive BEM with optimal convergence rates for the Helmholtz equation. Comput Methods Appl Mech Engrg. 2019;346:260–287.
- Feischl M, Gantner G, Haberl A, et al. Optimal convergence for adaptive IGA boundary element methods for weakly-singular integral equations. Numer Math. 2017;136(1):147–182.
- Gantner G, Praetorius D, Schimanko S. Adaptive isogeometric boundary element methods with local smoothness control. Math Models Methods Appl Sci. 2020;30:261–307.
- Gantner G, Haberlik D, Praetorius D. Adaptive IGAFEM with optimal convergence rates: hierarchical B-splines. Math Models Methods Appl Sci. 2017;27(14):2631–2674.
- Gantner G, Praetorius D. Adaptive IGAFEM with optimal convergence rates: T-splines. Comput Aided Geom Design. 2020;81:101906.
- Gantner G, Praetorius D. Adaptive BEM for elliptic PDE systems, part II: isogeometric analysis for weakly-singular integral equations. In Preparation. 2020.
- Faermann B. Localization of the Aronszajn–Slobodeckij norm and application to adaptive boundary elements methods. Part I. The two-dimensional case. IMA J Numer Anal. 2000;20(2):203–234.
- Faermann B. Localization of the Aronszajn–Slobodeckij norm and application to adaptive boundary element methods. Part II. The three-dimensional case. Numer Math. 2002;92(3):467–499.
- Carstensen C, Maischak M, Stephan EP. A posteriori error estimate and h-adaptive algorithm on surfaces for Symm's integral equation. Numer Math. 2001;90(2):197–213.
- Feischl M, Gantner G, Haberl A, et al. Adaptive 2D IGA boundary element methods. Eng Anal Bound Elem. 2016;62:141–153.
- Gantner G. Optimal adaptivity for splines in finite and boundary element methods. PhD thesis, Institute for Analysis and Scientific Computing, TU Wien, 2017.
- Mitscha-Eibl G. Adaptive BEM und FEM-BEM-Kopplung für die Lamé-Gleichung. Master's thesis, Institute for Analysis and Scientific Computing, TU Wien, 2014.
- McLean W. Strongly elliptic systems and boundary integral equations. Cambridge: Cambridge University Press; 2000.
- Steinbach O. Numerical approximation methods for elliptic boundary value problems. New York: Springer; 2008.
- Sauter SA, Schwab C. Boundary element methods. Berlin: Springer; 2011.
- Hofmann S, Mitrea M, Taylor M. Singular integrals and elliptic boundary problems on regular Semmes–Kenig– Toro domains. Int Math Res Not. 2009;2010(14):2567–2865.
- Di Nezza E, Palatucci G, Valdinoci E. Hitchhiker' s guide to the fractional Sobolev spaces. Bull Sci Math. 2012;136(5):521–573.
- Feischl M, Führer T, Mitscha-Eibl G, Praetorius D, et al. Convergence of adaptive BEM and adaptive FEM-BEM coupling for estimators without h-weighting factor. Comput Meth Appl Math. 2014;14(4):485–508.
- Carstensen C. An a posteriori error estimate for a first-kind integral equation. Math Comp. 1997;66(217):139–156.
- Carstensen C, Stephan EP. Adaptive boundary element methods for some first kind integral equations. SIAM J Numer Anal. 1996;33(6):2166–2183.
- Bespalov A, Haberl A, Praetorius D. Adaptive FEM with coarse initial mesh guarantees optimal convergence rates for compactly perturbed elliptic problems. Comput Methods Appl Mech Engrg. 2017;317:318–340.
- Aurada M, Feischl M, Führer T, et al. Efficiency and optimality of some weighted-residual error estimator for adaptive 2D boundary element methods. Comput Methods Appl Math. 2013;13(3):305–332.
- Stevenson R. The completion of locally refined simplicial partitions created by bisection. Math Comp. 2008;77(261):227–241.
- Binev P, Dahmen W, DeVore R. Adaptive finite element methods with convergence rates. Numer Math. 2004;97(2):219–268.
- Stevenson R. Optimality of a standard adaptive finite element method. Found Comput Math. 2007;7(2):245–269.
- Karkulik M, Pavlicek D, Praetorius D. On 2D newest vertex bisection: optimality of mesh-closure and H1-stability of L2-projection. Constr Approx. 2013;38(2):213–234.
- Manuel Cascon J, Kreuzer C, Nochetto RH, et al. Quasi-optimal convergence rate for an adaptive finite element method. SIAM J Numer Anal. 2008;46(5):2524–2550.
- Vuong A-V, Giannelli C, Jüttler B, et al. A hierarchical approach to adaptive local refinement in isogeometric analysis. Comput Methods Appl Mech Engrg. 2011;200(49):3554–3567.
- Morgenstern P, Peterseim D. Analysis-suitable adaptive T-mesh refinement with linear complexity. Comput Aided Geom Design. 2015;34:50–66.
- Carstensen C, Faermann B. Mathematical foundation of a posteriori error estimates and adaptive mesh-refining algorithms for boundary integral equations of the first kind. Eng Anal Bound Elem. 2001;25(7):497–509.
- Feischl M, Gantner G, Praetorius D. Reliable and efficient a posteriori error estimation for adaptive IGA boundary element methods for weakly-singular integral equations. Comput Methods Appl Mech Engrg. 2015;290:362–386.
- Hackbusch W. Integral equations: theory and numerical treatment. Basel: Birkhäuser; 1995.
- Ferraz-Leite S, Praetorius Dirk. Simple a posteriori error estimators for the h-version of the boundary element method. Computing. 2008;83(4):135–162.
- Carstensen C, Praetorius D. Averaging techniques for the effective numerical solution of symm's integral equation of the first kind. SIAM J Sci Comput. 2006;27(4):1226–1260.
- Morrey Jr CB. Multiple integrals in the calculus of variations. Springer, Berlin, reprint of the 1966 original edition, 2008.
- Feischl M, Führer T, Praetorius D. Adaptive FEM with optimal convergence rates for a certain class of nonsymmetric and possibly nonlinear problems. SIAM J Numer Anal. 2014;52(2):601–625.
Appendices
Appendix 1. Smooth characteristic functions
The following lemma provides a smooth cut-off function as used in the proof of discrete reliability (E4); see Section 4.8. For regular simplicial meshes as in Section 3.7, such functions can easily be constructed by means of hat functions; see [Citation5, Section 5.3]. For the present abstract setting, the construction is more technical.
Lemma A.1
Let and . Suppose (M1)–(M4). Then there exists a function such that for almost all (A.1a) (A.1a) (A.1b) (A.1b) (A.1c) (A.1c) Further, there exists a constant C>0 such that (A2) (A2) The constant C depends only on the dimension d and the constants from (M1)–(M4).
Proof.
In the following three steps, we will even prove the existence of a function with an open superset such that the restriction to Γ has the desired properties. With the constants from (M1)–(M2) and (M4), we define for all , (A3) (A3) Step 1: First, we construct an equivalent smooth mesh-size function with uniformly bounded gradient on Γ. Let be a standard mollifier with on , on , and . For s>0, we set . By convolution, we define (A4) (A4) Note that for . Thus (M4) and the choice (EquationA3(A3) (A3) ) of yield that . Together with (M1)–(M2) and , this implies for the interior of any that Note that the restriction to the interior is indeed necessary since the second term is discontinuous across faces of . However, by continuity of , this estimate is also satisfied if is replaced by , i.e. . The fact that on shows that also the converse estimate is valid. This leads to (A5) (A5) In particular, this proves the existence of an open set such that on O. Finally, we consider the gradient of for . Recall that . Together with the Hölder inequality, , and (M1)–(M2), this proves that (A6) (A6)
Step 2: In this step, we construct and prove (EquationA.1a(A.1a) (A.1a) )–(EquationA.1c(A.1c) (A.1c) ). For , we define the quasi-convolution Since on O, the chain rule shows that any derivative with respect to x of the term exists and is continuous at all . This yields that and ; see, e.g. [Citation19, page 98–99]. Since , it holds that (A7) (A7) If and thus , the properties of show that . Due to for all (which follows from (EquationA5(A5) (A5) )), this is particularly satisfied if . This proves (EquationA.1a(A.1a) (A.1a) ). Moreover, (EquationA7(A7) (A7) ) shows that for all (and hence verifies (EquationA.1b(A.1b) (A.1b) )), and if . For (EquationA.1c(A.1c) (A.1c) ), it thus remains to prove that implies that . We prove the contraposition. Let and suppose that . Then, there exists and such that and . The triangle inequality yields that (A8) (A8) Now, we differ two different cases. If , then we have that . The choice (EquationA3(A3) (A3) ) of together with (M4) shows that . If , then we have that . Let with and with . Together with (EquationA5(A5) (A5) ) and (EquationA8(A8) (A8) ), this yields that Hence, (M4) implies that and thus . Any that is sufficiently close to z also satisfies that and thus . Since , can be particularly chosen in . Hence, we see that This is equivalent to , which concludes that .
Step 3: Finally, we prove (EquationA2(A2) (A2) ). We recall that on O; see Step 1. With the identity matrix and the matrix , elementary calculations prove for all and all that Considering the norm, we see that Together with , this shows for all that Thus (EquationA5(A5) (A5) )–(EquationA6(A6) (A6) ) and (M3) prove that for almost all . Moreover, for smooth functions, the surface gradient is the orthogonal projection of the gradient ∇ onto the tangent plane; see, e.g. [Citation18, Lemma 2.22]). With the outer normal vector ν, this implies that almost everywhere on Γ and concludes the proof with the previous estimate.
Appendix 2
Inverse inequalities for other integral operators
In Proposition 4.13, we have generalized an inverse estimate from [Citation5] for the single-layer operator . [Citation5] additionally derived similar estimates for the double-layer operator , the adjoint double-layer operator , and the hyper-singular operator ; see, e.g. [Citation19, page 218] for a precise definition (where these operators are denoted by and R, respectively). Although [Citation5] considered only the Laplace problem, the techniques of the proof of Proposition 4.13 extend the result at least to partial differential operators without lowest-order term cu. With some further effort, one can even prove it for arbitrary PDE operators with constant coefficients as in Section 2.3, where, as in Section 4.4, ellipticity of can be replaced by ellipticity up to some compact perturbation. To this end, one requires additional regularity of the trace operator , which is satisfied for piecewise smooth boundaries Γ; see, e.g. [Citation21, Remark 3.1.18].
For the proof, we will frequently use [Citation19, Theorem 4.24], which reads as follows: Let be arbitrary with in the weak sense and . Then, and (B.1) (B.1)
Proposition B.1
Suppose (M1)–(M5). For let be a weight function which satisfies for some and all that (B.2) (B.2) Then, there exists a constant such that for all and (B.3) (B.3) If we additionally suppose that the trace operator satisfies the stability there exists a constant such that for all and (B.4) (B.4) The constants and depend only on (M1)–(M5), Γ, the coefficients of and the admissibility constant α.
Proof of (EquationB3(A3) (A3) ).
Proof of (EquationB.3(A3) (A3) )
The bound for is just the assertion of Proposition 4.13. With the results from the proof of Proposition 4.13, one can easily estimate the second summand as in [Citation5, Section 6]. In particular, the stability of is exploited, which is stated in [Citation19, page 209]. Further, one uses the fact from [Citation19, page 218] that , where denotes the interior modified conormal derivative from [Citation19, page 117–118].
Proof of (EquationB4(A4) (A4) ).
Proof of (EquationB.4(A4) (A4) )
As in the proof of Proposition 4.13, we abbreviate and for all . Let . Further, let be the smooth quasi-indicator function of T from Lemma A.1. With the Poincaré inequality (Equation56(56) (56) ), the localization properties (Equation55(55) (55) ) as well as (Equation60(60) (60) ), and (M5), one can easily verify that (B.5) (B.5) We fix (independently of ) a bounded domain with for all . Let denote the double-layer potential from [Citation19, Theorem 6.11]. With these preparations and if has no lower-order terms, i.e. for all as well as c = 0, the proof of (EquationA4(A4) (A4) ) follows as in [Citation5, Section 5 and Section 6]. In particular, one exploits the fact that (B.6) (B.6) which follows from the representation formula [Citation19, Theorem 6.10] together with the assumption that has no lower-order terms. The proof of (EquationB.4(B.4) (B.4) ) then employs the Poincaré inequality (EquationB.5(B.5) (B.5) ), the property (EquationB.6(B.6) (B.6) ), the stability of and of from [Citation22, Corollary 3.38] or [Citation19, page 209], and the Caccioppoli inequality (Equation71(71) (71) ) in combination with the transmission property [Citation19, Theorem 6.11].
To prove (EquationA12(B.4) (B.4) ) for general with lower-order terms, where (EquationB.6(B.6) (B.6) ) is in general false, we recall the principal part from (Equation91(91) (91) ). In the following five steps, we show for the corresponding double-layer operator and the hyper-singular operator that (B.7) (B.7) are continuous. Since (EquationB.4(B.4) (B.4) ) is satisfied for the operators corresponding to , this and the trivial estimate will conclude the proof.
Step 1: Let be the Newton potentials from [Citation19, Theorem 6.1] corresponding to . According to [Citation19, Theorem 6.1], they satisfy the mapping property for all . In the proof of the latter stability, the fundamental solution is defined in terms of the Fourier transformation. The definition involves a multivariate polynomial resp. associated to resp. (which is obtained from the differential operator by replacing the derivatives with variables) such that for and ; see [Citation19, Equation (6.7)]. Indeed, the latter inequality is the key of the proof. As elementary analysis even shows that ) with the identity matrix , one sees along the lines of [Citation19, Theorem 6.1] the additional regularity (B.8) (B.8) Since multiplication with a fixed compactly supported smooth function is stable (see, e.g. [Citation19, Theorem 3.20]), we also see for arbitrary the continuity of (B.9) (B.9) Step 2: For sufficiently large , the sesquilinear forms and from (Equation18(18) (18) ) are both even elliptic on . Let be the unique weak solution of and . Similarly, let be the solution of and . We extend both functions by zero outside of Ω. Since and , the representation formula [Citation19, Theorem 6.10] yields that (B.10) (B.10) We note that and with the conormal derivative . To see (EquationB.7(B.7) (B.7) ) and hence to conclude the proof, we thus only have to bound the trace as well as the conormal derivative of each summand in (EquationB.10(B.10) (B.10) ) separately, which will be done in the following three steps.
Step 3: By definition, the trace of the first summand in (EquationB.10(B.10) (B.10) ) vanishes, i.e. . According to (EquationA9(B.1) (B.1) ) (where the differential operator can also be chosen as instead of ), the normal derivative satisfies that Since the first summand vanishes and due to , the stability of the solution mapping and gives that (B.11) (B.11) Step 4: Due to our assumption that the trace operator is well-defined and continuous, the stability of the Newton potentials, the stability of the solution mapping and , the trace of the second summand in (EquationB.10(B.10) (B.10) ) satisfies that Note that and are indeed potentials, i.e. weakly and weakly. With (EquationB.1(B.1) (B.1) ) (applied for and ), the stability of the Newton potentials, and the estimates for the trace of and , we thus see that Again, the stability of the solution mappings allows to bound the last terms by .
Step 5: To deal with the third summand in (EquationB.10(B.10) (B.10) ), we first rewrite it as follows: (B.12) (B.12) Due to the stability of (see (Equation21(21) (21) ) and (Equation76(76) (76) )) and (EquationB.11(B.11) (B.11) ), we have for the first summand in (EquationB.12(B.12) (B.12) ) that (B.13) (B.13) To deal with the conormal derivative, we apply again (EquationB.1(B.1) (B.1) ) together with the fact that is a potential, i.e. weakly. This leads to The first summand can be bounded as in (EquationB.13(B.13) (B.13) ). For the second summand, we use the stability (Equation75(75) (75) ) in combination with and (EquationB.11(B.11) (B.11) ).
Finally, it only remains to bound the trace as well as the conormal derivative of the second summand in (EquationB.12(B.12) (B.12) ). Choosing in the additional regularity (EquationB.9(B.9) (B.9) ), one sees as in the proof of [Citation19, page 203] (which proves stability of ) that (B.14) (B.14) With the assumption , this implies that (B.15) (B.15) Recall that with . Thus [Citation19, Theorem 4.25] (which states boundedness of the Dirichlet to Neumann mapping) gives that (B.16) (B.16) Moreover, (EquationA9(B.1) (B.1) ), the fact that , the stability (Equation75(75) (75) ), and (EquationB.16(B.16) (B.16) ) show for the conormal derivative that Overall, we have estimated the trace as well as the conormal derivative of all terms in (EquationB.10(B.10) (B.10) ). Since and , this verifies the stability (EquationB.7(A7) (A7) ) and thus concludes the proof.