Research Article
Dietmar Gallistl
Adaptive Nonconforming Finite Element Approximation
of Eigenvalue Clusters
Abstract: This paper analyses an adaptive nonconforming finite element method for eigenvalue clusters of self-adjoint operators and proves optimal convergence rates (with respect to the concept of nonlinear ap-proximation classes) for the apap-proximation of the invariant subspace spanned by the eigenfunctions of the eigenvalue cluster. Applications include eigenvalues of the Laplacian and of the Stokes system.
Keywords: Eigenvalue Problem, Eigenvalue Cluster, Adaptive Finite Element Method, Stokes Operator, Opti-mality
MSC 2010: 65M12, 65M60, 65N25 ||
Dietmar Gallistl:Institut fรผr Mathematik, Humboldt-Universitรคt zu Berlin, Unter den Linden 6, 10099 Berlin, Germany,
e-mail: gallistl@math.hu-berlin.de
1 Introduction
Nonconforming finite element methods (FEMs) are of high interest in computational fluid dynamics where they provide stable low-order discretisations with favourable local mass conservation properties. Especially for eigenvalue problems, the nonconforming discretisation is even more attractive because it allows for a convenient computation of guaranteed lower eigenvalue bounds [16]. In many practical situations the eigen-values of interest form an eigenvalue cluster where all eigenfunctions have to be discretised simultaneously in adaptive algorithms. This paper applies and generalises the technique of the recent work [33] to the non-conformingP1discretisation of the Laplace and Stokes eigenvalue problems and proves optimal convergence rates of the simultaneous adaptive FEM computation for the eigenfunctions in the cluster. Optimal conver-gence rates for adaptive FEMs for eigenvalue problems were established in [15, 26] for simple eigenvalues and in [25] for multiple eigenvalues for conforming finite elements and in [14] for the nonconforming discretisation of the first eigenvalue of the Laplacian. The main difference to the analysis of those results is the additional difficulty that the cluster width should not enter the error estimates as an additive term. Consider an open bounded polyhedral Lipschitz domain ฮฉ โ โ๐for ๐ โฅ 2 and a simplicial triangulation T
โ. Let ๐ be the
in-variant subspace spanned by the eigenfunctions of an eigenvalue cluster and let ๐โdescribe the linear hull
of the corresponding nonconformingP1(Tโ) eigenfunctions. The adaptive algorithm is driven by the explicit residual-based error estimator contributions of all discrete eigenfunctions in the cluster. The main results of this paper state that the error quantities
sup
๐คโ๐ โ๐คโ=1
inf
๐ฃโโ๐โ|||๐ค โ ๐ฃโ|||NC
(in the case of the Laplace eigenproblem โฮ๐ข = ๐๐ข) and sup ๐คโ๐ โ๐คโ=1 inf ๐ฃโโ๐โ(|||๐ค โ ๐ฃโ||| 2 NC+ โ๐(๐ค) โ ๐(๐ฃโ)โ2) 1/2
(in the case of the Stokes eigenproblem โฮ๐ข+(๐ท๐)โค= ๐๐ข; div ๐ข = 0) decay as (card(T
โ) โ card(T0))โ๐, provided
all eigenfunctions belong to the approximation class A๐(resp. AStokes๐ ). Here, โโ โ denotes the ๐ฟ2norm and |||โ |||NC denotes the nonconforming energy norm (i.e., the ๐ฟ2norm of the piecewise derivative). Although one can
square root of the eigenvalue error, this paper merely studies the approximation of the space ๐. An important methodological tool is the higher-order ๐ฟ2control for the eigenfunction approximations which is proven by
means of conforming companion operators. Operators of this kind were introduced in [14, 41] in the two-dimensional case and are generalised in this paper to higher space dimensions ๐ โฅ 2. The resulting ๐ฟ2error
estimates compare the ๐ฟ2error directly with the energy error and therefore do not employ any a priori results of the eigenfunction approximation.
The proofs for optimal convergence rates of adaptive FEMs were initiated in [22, 46] and extended to nonconforming FEMs for the Poisson equation [3, 42] and the Stokes equations [2, 20, 39]. These approaches were recently unified in the axiomatic approach of [12]. The convergence of adaptive FEMs for eigenvalues was proven in [10, 35, 36]. The optimality results [15, 26, 34] concern simple eigenvalues and conforming FEMs while [14] establishes optimality for the nonconforming discretisation of the first Laplace eigenvalue. The first optimality analysis for an adaptive algorithm for multiple eigenvalues [25] based on conforming FEMs introduced a simultaneous bulk criterion for all discrete eigenfunctions of the multiple eigenvalue. In [33] this marking strategy was proven to lead to optimal convergence rates in the case of eigenvalue clusters. The results of this paper establish a corresponding result for the nonconformingP1FEM and the first optimality result for the Stokes eigenproblem.
The remaining parts of this paper are organised as follows. Section 2 describes an abstract framework for the discretisation of eigenvalue clusters. Section 3 introduces the notation on triangulations and presents the conforming companion operators for the nonconformingP1FEM in any space dimension. Section 4 is devoted to the analysis of the adaptive FEM for the eigenvalues of the Laplacian. Section 5 studies the adaptive FEM approximation of the eigenvalues of the Stokes system.
Throughout the paper standard notation on Lebesgue and Sobolev spaces is employed. The integral mean is denoted byโซ. The notation ๐ โฒ ๐ abbreviates ๐ โค ๐ถ๐ for a positive generic constant ๐ถ that may depend onโ
the domain ฮฉ and the initial triangulation T0but not on the mesh-size or the eigenvalue cluster of interest.
The notation ๐ โ ๐ stands for ๐ โฒ ๐ โฒ ๐.
2 Approximation of Eigenvalue Clusters
Let (๐, ๐(โ , โ )) be a separable Hilbert space over โ with induced norm โโ โ๐and let ๐(โ , โ ) be a scalar product on
๐ with induced norm โโ โ๐such that the embedding (๐, โโ โ๐) ๓ณจ โ (๐, โโ โ๐) is compact. This paper is concerned
with eigenvalue problems of the form: Find eigenpairs (๐, ๐ข) โ โ ร ๐ with โ๐ขโ๐= 1 such that
๐(๐ข, ๐ฃ) = ๐๐(๐ข, ๐ฃ) for all ๐ฃ โ ๐. (2.1)
It is well known from the spectral theory of selfadjoint compact operators [23, 40] that the eigenvalue problem (2.1) has countably many eigenvalues, which are real and positive with +โ as only possible accumulation point. Suppose that the eigenvalues are enumerated as
0 < ๐1โค ๐2โค ๐3โค โ โ โ
and let (๐ข1, ๐ข2, ๐ข3, . . .) be some ๐-orthonormal system of corresponding eigenfunctions. For any ๐ โ โ, the
eigenspace corresponding to ๐๐is defined as
๐ธ(๐๐) := {๐ข โ ๐ | (๐๐, ๐ข) satisfies (2.1)} = span{๐ข๐| ๐ โ โ and ๐๐= ๐๐}.
In the present case of an eigenvalue problem of (the inverse of) a compact operator, the spaces ๐ธ(๐๐) have
finite dimension. The discretisation of (2.1) is based on a family (over a countable index set ๐ผ) of separable (not necessarily finite-dimensional) Hilbert spaces ๐โwith scalar products ๐NC(โ , โ ) and ๐NC(โ , โ ) on ๐ + ๐โwith
induced norms โโ โ๐,NCand โโ โ๐,NCsuch that ๐NCand ๐NCcoincide with ๐ and ๐ when restricted to ๐:
The discrete eigenvalue problem seeks eigenpairs (๐โ, ๐ขโ) โ โ ร ๐โwith โ๐ขโโ๐,NC= 1 such that
๐NC(๐ขโ, ๐ฃโ) = ๐โ๐NC(๐ขโ, ๐ฃโ) for all ๐ฃโโ ๐โ.
The discrete eigenvalues can be enumerated
0 < ๐โ,1โค ๐โ,2โค ๐โ,3โค โ โ โ
with corresponding ๐NC-orthonormal eigenfunctions (๐ขโ,1, ๐ขโ,2, ๐ขโ,3, . . .). For a finite cluster of eigenvalues
๐๐+1, . . . , ๐๐+๐of length ๐ โ โ, define the index set ๐ฝ := {๐ + 1, . . . , ๐ + ๐} and the spaces
๐ := span{๐ข๐| ๐ โ ๐ฝ} and ๐โ:= span{๐ขโ,๐| ๐ โ ๐ฝ}.
The eigenspaces ๐ธ(๐๐) may differ for different ๐ โ ๐ฝ.
Assume that the cluster is contained in a compact interval [๐ด, ๐ต] in the sense that {๐๐| ๐ โ ๐ฝ} โช {๐โ,๐| โ โ ๐ผ, ๐ โ ๐ฝ} โ [๐ด, ๐ต]. This implies sup โโ๐ผ(๐,๐)โ๐ฝmax2max{๐ โ1 ๐ ๐โ,๐, ๐โ1โ,๐๐๐} โค ๐ต/๐ด.
Although in the applications in this paper dim(๐โ) will be finite-dimensional, the analysis in this section
admits the case dim(๐โ) โ โ โช {โ}. Let ๐ฝ๐ถ := {1, . . . , dim(๐โ)} \ ๐ฝ denote the complement of ๐ฝ. Assume that
the cluster is separated from the remaining part of the spectrum in the sense that there exists a separation bound ๐๐ฝ:= sup โโ๐ผ sup๐โ๐ฝ๐ถmax๐โ๐ฝ ๐๐ |๐โ,๐โ ๐๐| < โ. (H1)
Given ๐ โ ๐, let ๐ข โ ๐ denote the unique solution to the linear problem ๐(๐ข, ๐ฃ) = ๐(๐, ๐ฃ) for all ๐ฃ โ ๐. The quasi-Ritz projection ๐ โ๐ข โ ๐โis defined as the unique solution to
๐NC(๐ โ๐ข, ๐ฃโ) = ๐NC(๐, ๐ฃโ) for all ๐ฃโโ ๐โ.
Let ๐โdenote the ๐NC-orthogonal projection onto ๐โand define
ฮโ:= ๐โโ ๐ โ.
For any eigenfunction ๐ข โ ๐, the function ฮโ๐ข โ ๐โis regarded as its approximation. This approximation
does not depend on the basis of ๐โ. Notice that ฮโ๐ข is neither computable without knowledge of ๐ข nor
nec-essarily an eigenfunction.
The following result is essentially contained in the textbook [48] and in [10] for a conforming finite ele-ment discretisation of the Laplace eigenvalue problem. The proof presented here extends the arguele-ments of [48] to a more abstract situation.
Proposition 2.1. Any eigenpair (๐, ๐ข) โ โ ร ๐ of (2.1) with โ๐ขโ๐= 1 satisfies
โ๐ โ๐ข โ ฮโ๐ขโ๐,NCโค ๐๐ฝโ๐ข โ ๐ โ๐ขโ๐,NC
and
โ๐ข โ ๐โ๐ขโ๐,NCโค โ๐ข โ ฮโ๐ขโ๐,NCโค (1 + ๐๐ฝ)โ๐ข โ ๐ โ๐ขโ๐,NC.
Proof. Set ๐ฃโ:= ๐ โ๐ข โ ฮโ๐ข and recall dim(๐โ) โ โ โช {โ}. Since the eigenfunctions (๐ขโ,๐| ๐ = 1, . . . , dim(๐โ))
form a ๐NC-orthonormal system of ๐โand ๐ฃโis ๐NC-orthogonal on ๐โ, there exist coefficients (๐ผ๐ | ๐ โ ๐ฝ๐ถ) such
that ๐ฃโ= โ ๐โ๐ฝ๐ถ ๐ผ๐๐ขโ,๐ and โ ๐โ๐ฝ๐ถ ๐ผ2๐ = โ๐ฃโโ2๐,NC.
The definition of ๐ โand the symmetry show that
(๐โ,๐โ ๐)๐NC(๐ โ๐ข, ๐ขโ,๐) = ๐๐NC(๐ข โ ๐ โ๐ข, ๐ขโ,๐).
This and the orthogonality of ๐ฃโand ฮโ๐ข lead to
โ๐ฃโโ2๐,NC= ๐NC(๐ โ๐ข, โ ๐โ๐ฝ๐ถ ๐ผ๐๐ขโ,๐) = ๐NC(๐ข โ ๐ โ๐ข, โ ๐โ๐ฝ๐ถ ๐ผ๐ ๐ ๐โ,๐โ ๐ ๐ขโ,๐).
The Cauchy inequality, the estimate (H1) and the ๐NC-orthogonality of the discrete eigenfunctions therefore
show
โ๐ฃโโ๐,NCโค ๐๐ฝโ๐ข โ ๐ โ๐ขโ๐,NC.
The second claimed chain of inequalities follows from the projection property of ๐โand the triangle inequality.
The following algebraic identity applies frequently in the analysis. It states the important property that, al-though ฮโ๐ข is no eigenfunction in general, ฮโ๐ข satisfies an equation that is similar to an eigenfunction
prop-erty.
Lemma 2.2. Any eigenpair (๐, ๐ข) โ โ ร ๐ of (2.1) satisfies
๐NC(ฮโ๐ข, ๐ฃโ) = ๐๐NC(๐โ๐ข, ๐ฃโ) for all ๐ฃโโ ๐โ.
In other words, ๐ โand ๐โcommute, ๐โโ ๐ โ= ๐ โโ ๐โ.
Proof. The proof is given in [33, Lemma 2.2] and repeated here for convenient reading. The representation of
ฮโ๐ข in terms of the ๐NC-orthonormal basis (๐ขโ,๐)๐โ๐ฝreads as
ฮโ๐ข = โ ๐โ๐ฝ
๐ผ๐๐ขโ,๐ with ๐ผ๐= ๐NC(๐ โ๐ข, ๐ขโ,๐) for all ๐ โ ๐ฝ.
The symmetry of ๐NCand ๐NCproves for any ๐ โ ๐ฝ that
๐ผ๐= ๐NC(๐ โ๐ข, ๐ขโ,๐) = ๐ โ1
โ,๐๐NC(๐ โ๐ข, ๐ขโ,๐) = ๐ โ1
โ,๐๐๐NC(๐ข, ๐ขโ,๐).
Therefore, the discrete eigenvalue problem reveals ๐NC(ฮโ๐ข, ๐ฃโ) = โ
๐โ๐ฝ
๐ผ๐๐โ,๐๐NC(๐ขโ,๐, ๐ฃโ) = ๐ โ ๐โ๐ฝ
๐NC(๐NC(๐ข, ๐ขโ,๐)๐ขโ,๐, ๐ฃโ) = ๐๐NC(๐โ๐ข, ๐ฃโ).
The following result states a comparison of seminorms for the eigenfunctions. The application in the subse-quent sections will be the equivalence of error estimators.
Lemma 2.3. Suppose that
๐ := max
๐โ๐ฝโ๐ข๐โ ฮโ๐ข๐โ๐,NCโค โ1 + 1/(2๐) โ 1 for all โ โ ๐ผ. (H2)
Then, both (๐โ๐ข๐)๐โ๐ฝand (ฮโ๐ข๐)๐โ๐ฝform a basis of ๐โ. For any ๐คโโ ๐โwith โ๐คโโ๐,NC= 1, the coefficients of the
representation ๐คโ= โ๐โ๐ฝ๐ฝ๐๐โ๐ข๐and ๐คโ= โ๐โ๐ฝ๐พ๐ฮโ๐ข๐are controlled as
max { โ ๐โ๐ฝ |๐ฝ๐| 2 , โ ๐โ๐ฝ |๐พ๐| 2 } โค 2 + 4๐ for ๐ = card(๐ฝ).
For any โ โ ๐ผ, any seminorm ๐โon ๐โsatisfies
๐โ1โ ๐โ๐ฝ ๐โ(๐๐๐โ๐ข๐)2โค (๐ต/๐ด)2โ ๐โ๐ฝ ๐โ(๐โ,๐๐ขโ,๐)2โค (๐ต/๐ด)4(2๐ + 4๐2) โ ๐โ๐ฝ ๐โ(๐๐๐โ๐ข๐)2 and ๐โ1โ ๐โ๐ฝ ๐โ(ฮโ๐ข๐)2โค (๐ต/๐ด)2โ ๐โ๐ฝ ๐โ(๐ขโ,๐)2โค (๐ต/๐ด)4(2๐ + 4๐2) โ ๐โ๐ฝ ๐โ(ฮโ๐ข๐)2.
3 The Nonconforming
P
1Finite Element Space
This section introduces the necessary notation on regular simplicial triangulations and recalls some elemen-tary facts on the nonconformingP1finite element space. It furthermore generalises the companion operators from [14] to higher space dimensions.
3.1 Notation on Regular Triangulations
LetT0be a regular simplicial triangulation of ฮฉ in the sense of [47], i.e., โชT0= ฮฉ and any two elements of T0 are either disjoint or share exactly one ๐-dimensional face for ๐ โค ๐ (e.g., a vertex or an edge). Throughout this paper, any regular triangulation of ฮฉ is assumed to be admissible in the sense that it is regular and a refinement ofT0created by the refinement rules of [47] with proper initialisation of the refinement edges [47]. The set of all admissible refinements is denoted by ๐. Given a triangulation Tโ โ ๐, the piecewise constant
mesh-size function โโ:= โTโis defined by โโ|๐:= โ๐:= meas(๐)
1/๐for any simplex ๐ โ T โ.
The set of (๐ โ 1)-dimensional hyper-faces (e.g., edges for ๐ = 2 or faces for ๐ = 3) of Tโis denoted byFโ
while the interior (๐ โ 1)-dimensional hyper-faces are denoted by Fโ(ฮฉ). Let every ๐น โ Fโbe equipped with a
fixed normal vector ๐๐น. Given ๐น โ Fโ(ฮฉ), ๐น = ๐๐+โฉ ๐๐โshared by two simplices (๐+, ๐โ) โ T2โ, and a piecewise
smooth function ๐ฃ, define the jump of ๐ฃ across ๐น by
[๐ฃ]๐น:= ๐ฃ|๐+โ ๐ฃ|๐โ.
For hyper-faces ๐น โ ๐ฮฉ on the boundary, [๐ฃ]๐น := ๐ฃ|๐นdenotes the trace. For a simplex ๐, the set of (๐ โ
1)-dimensional hyper-faces belonging to ๐ is denoted by F(๐).
The set of piecewise polynomial functions of degree โค ๐ with respect to Tโis denoted byP๐(Tโ). The ๐ฟ2
projection ontoP๐(Tโ) is denoted by ฮ ๐ Tโ โก ฮ
๐
โ. The ๐-th order oscillations of a given function ๐ โ ๐ฟ2(ฮฉ) is
defined as
osc๐(๐, Tโ) := โโโ(1 โ ฮ ๐ โ)๐โ๐ฟ2(ฮฉ).
The piecewise action of a differential operator is indicated by the subscript NC, i.e., the piecewise versions of ๐ท and div read as ๐ทNC โก ๐ทNC(โ)and divNC โก divNC(โ)e.g., (๐ทNC๐ฃ)|๐ = ๐ท(๐ฃ|๐) for any ๐ โ Tโ. The dependence
onTโin the notation is dropped whenever there is no risk of confusion.
3.2 Nonconforming Finite Element Space and Companion Operator
The nonconformingP1finite element space, sometimes referred to as CrouzeixโRaviart finite element space [24], reads as
CR10(Tโ) := {๐ฃโโ P1(Tโ) | ๐ฃโis continuous in the interior hyper-facesโ midpoints and vanishes in the midpoints of hyper-faces on the boundary}.
Let, throughout this subsection, ๐โ:= ๐(Tโ) := CR10(Tโ) and ๐ := ๐ป01(ฮฉ). Given an admissible refinement
Tโ+๐โ ๐(Tโ) of Tโ, define the operatorI CR
โ : ๐ + ๐โ+๐โ ๐โby
โซ
๐น
(๐ฃ โ ICRโ ๐ฃ) ๐๐ = 0 for all ๐น โ Fโand all ๐ฃ โ ๐ + ๐โ+๐.
Note thatICRโ is indeed well-defined for functions in CR1
0(Tโ+๐). A (piecewise) integration by parts proves the
projection property ๐ทNCI CR โ = ฮ 0 โ๐ท, i.e., โซ ๐ ๐ทNCI CR โ ๐ฃ ๐๐ฅ = โซ ๐
The proof of the approximation and stability property โโโ1๐ (๐ฃ โ I CR โ ๐ฃ)โ๐ฟ2(๐)+ โ๐ทNC(๐ฃ โ ICR โ ๐ฃ)โ๐ฟ2(๐)โฒ โ(1 โ ฮ 0 โ)๐ทNC๐ฃโ๐ฟ2(๐) (3.2)
for any ๐ฃ โ ๐ + ๐โ+๐and any ๐ โ Tโfollows from the discrete Friedrichs inequality [9, Theorem 10.6.12] and
a scaling argument.
The remaining parts of this subsection present conforming companion operators. The idea behind these operators is to design for a nonconforming finite element function ๐ฃโsome conforming companion ๐ฝ๐+1๐ฃโโ ๐
with certain conservation properties. For ๐ = 2, operators of this kind have been constructed in [14] and independently in [41]. The following result extends [14] to any dimension ๐ โฅ 2.
Proposition 3.1(companion operator in any space dimension). Given any ๐ฃโ โ ๐โthere exists some ๐ฝ๐+1๐ฃโ โ
P๐+1(Tโ) โฉ ๐ such that ๐ฃโโ ๐ฝ๐+1๐ฃโis ๐ฟ2orthogonal onto the spaceP0(Tโ) of piecewise constants, it enjoys the
integral mean property
ฮ 0โ(๐ทNC(๐ฃโโ ๐ฝ๐+1๐ฃโ)) = 0, (3.3)
and it satisfies the approximation and stability property
โโโ1โ (๐ฃโโ ๐ฝ๐+1๐ฃโ)โ๐ฟ2(ฮฉ)+ โ๐ทNC(๐ฃโโ ๐ฝ๐+1๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ2(ฮฉ). (3.4)
Proof. The design follows in three steps.
Step 1. The operator ๐ฝ1: ๐โโ ๐1(Tโ) โฉ ๐ acts on any function ๐ฃโโ ๐โby averaging the function values at
each interior vertex ๐ง, i.e.,
๐ฝ1๐ฃโ(๐ง) = card(Tโ(๐ง))โ1 โ ๐โTโ(๐ง)
๐ฃโ|๐(๐ง) for all ๐ง โ Nโ(ฮฉ)
whereTโ(๐ง) := {๐ โ Tโ | ๐ง โ ๐} is the set of simplices that contain the vertex ๐ง. This operator is also known as enriching operator in the context of fast solvers [8]. The proof of the approximation property
โโโ1โ (๐ฃโโ ๐ฝ1๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ2(ฮฉ) (3.5)
is included in [11, Theorem 5.1] for ๐ = 2. A generalisation to higher dimensions is outlined in the proof of [13, Theorem 4.9]. This and an inverse estimate [9] imply the stability property
โ๐ทNC(๐ฃโโ ๐ฝ1๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ
2(ฮฉ). (3.6)
Step 2. Given any hyper-face ๐น = conv{๐ง1, . . . , ๐ง๐} with nodal P1conforming basis functions ๐1, . . . , ๐๐ โ
P1(Tโ) โฉ ๐, the quadratic edge-bubble function
โญ๐น:= (2๐ โ 1)! (๐ โ 1)! ๐ โ ๐=1 ๐๐
is supported on the patch of ๐น (that is the union of simplices which ๐น belongs to) and satisfiesโซโ
๐นโญ๐น๐๐ = 1.
For any function ๐ฃโโ ๐โthe operator ๐ฝ๐: ๐โโ P๐(Tโ) โฉ ๐ acts as
๐ฝ๐๐ฃโ:= ๐ฝ1๐ฃโ+ โ ๐นโFโ(ฮฉ)
(โโซ
๐น
(๐ฃโโ ๐ฝ1๐ฃโ) ๐๐ )โญ๐น.
An immediate consequence of this choice reads as โ โซ ๐น ๐ฝ๐๐ฃโ๐๐ = โโซ ๐น ๐ฃโ๐๐ for all ๐น โ Fโ.
An integration by parts shows the integral mean property of the gradients ฮ 0
โ๐ท๐ฝ๐= ๐ทNC, i.e., โซ ๐ ๐ท๐ฝ๐๐ฃโ๐๐ฅ = โซ ๐ ๐ทNC๐ฃโ๐๐ฅ for all ๐ โ Tโ.
Let ๐ โ Tโwith ๐น โ F(๐). The scaling โโญ๐นโ๐ฟ2(ฮฉ)โฒ โ๐/2
๐ and the Hรถlder and trace inequalities [30] show
โ๐โ1๓ตฉ๓ตฉ๓ตฉ๓ตฉ๓ตฉ๓ตฉ ๓ตฉ๐นโF(๐)โ (โโซ ๐น (๐ฃโโ ๐ฝ1๐ฃโ) ๐๐ )โญ๐น๓ตฉ๓ตฉ๓ตฉ๓ตฉ๓ตฉ๓ตฉ ๓ตฉ๐ฟ2(๐)โฒ โ (๐โ2)/2 ๐ โ ๐นโF(๐) ๓ตจ๓ตจ๓ตจ ๓ตจ๓ตจ๓ตจ ๓ตจโโซ ๐น (๐ฃโโ ๐ฝ1๐ฃโ) ๐๐ ๓ตจ๓ตจ๓ตจ๓ตจ๓ตจ๓ตจ ๓ตจ โฒ โโ1/2๐ โ ๐นโF(๐) โ๐ฃโโ ๐ฝ1๐ฃโโ๐ฟ2(๐น) โฒ โโ1๐ โ๐ฃโโ ๐ฝ1๐ฃโโ๐ฟ2(๐)+ โ๐ทNC(๐ฃโโ ๐ฝ1๐ฃโ)โ๐ฟ2(๐).
This, the triangle inequality and the properties (3.5)โ(3.6) yield โโโ1โ (๐ฃโโ ๐ฝ๐๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ
2(ฮฉ). (3.7)
The stability property of ๐ฝ๐follows with an inverse estimate [9]
โ๐ทNC(๐ฃโโ ๐ฝ๐๐ฃโ)โ๐ฟ2(ฮฉ)โฒ โโโ1
โ (๐ฃโโ ๐ฝ๐๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ2(ฮฉ).
Step 3. On any simplex ๐ = conv{๐ง1, . . . , ๐ง๐+1} with nodal basis functions ๐1, . . . , ๐๐+1, the volume bubble
function is defined by โญ๐:= (2๐ + 1)! ๐! ๐+1 โ ๐=1 ๐๐โ ๐ป 1 0(int(๐)) and satisfiesโซโ ๐โญ๐๐๐ฅ = 1. Define ๐ฝ๐+1๐ฃโ:= ๐ฝ๐๐ฃโ+ โ ๐โTโ (โโซ ๐ (๐ฃโโ ๐ฝ๐๐ฃโ) ๐๐ฅ)โญ๐.
The difference ๐ฃโโ ๐ฝ๐+1๐ฃโis ๐ฟ2-orthogonal to all piecewise constant functions. Since โญ๐vanishes on all ๐น โ Fโ,
๐ฝ๐+1enjoys the integral mean property ฮ 0โ๐ท๐ฝ๐+1= ๐ทNC. The Hรถlder inequality and (3.7) imply
๓ตจ๓ตจ๓ตจ ๓ตจ๓ตจ๓ตจ ๓ตจโซโ ๐ (๐ฃโโ ๐ฝ๐๐ฃโ) ๐๐ฅ๓ตจ๓ตจ๓ตจ๓ตจ๓ตจ๓ตจ ๓ตจโฒ โ โ๐/2 ๐ โ๐ฃโโ ๐ฝ๐๐ฃโโ๐ฟ2(๐)โฒ โโ(๐โ2)/2 ๐ min๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ2(ฮฉ). The scaling โ๐ทโญ๐โ๐ฟ2(ฮฉ)โ โ(๐โ2)/2
๐ and the triangle inequality prove the stability property
โ๐ทNC(๐ฃโโ ๐ฝ๐+1๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ
2(ฮฉ).
A piecewise Poincarรฉ inequality proves the approximation property โโโ1โ (๐ฃโโ ๐ฝ๐+1๐ฃโ)โ๐ฟ2(ฮฉ)โฒ min
๐ฃโ๐โ๐ทNC(๐ฃโโ ๐ฃ)โ๐ฟ2(ฮฉ).
4 Eigenvalues of the Laplacian
This section studies the adaptive nonconforming FEM approximation of the Laplace eigenproblem. Section 4.1 presents ๐ฟ2and best-approximation estimates for the linear Poisson problem. Section 4.2 introduces the
dis-cretisation of the eigenvalue problem. A โtheoreticalโ (i.e., non-computable) error estimator and its discrete reliability are analysed in Section 4.3. Sections 4.4 and 4.5 present the practical AFEM and prove contraction and optimal convergence rates.
4.1 Nonconforming FEM for the Poisson Model Problem
This subsection revisits the nonconformingP1discretisation of the linear Poisson equation. Let ๐ := ๐ป1 0(ฮฉ)
be equipped with the scalar products
and induced norms |||๐ฃ||| := ๐(๐ฃ, ๐ฃ)1/2and โ๐ฃโ := ๐(๐ฃ, ๐ฃ)1/2. Given ๐ โ ๐ฟ2(ฮฉ), the weak formulation of the Poisson problem โฮ๐ข = ๐ under homogeneous Dirichlet boundary conditions reads as
๐(๐ข, ๐ฃ) = ๐(๐, ๐ฃ) for all ๐ฃ โ ๐. (4.1)
The nonconforming finite element discretisation is based on the space ๐โ:= CR10(Tโ) and the scalar product
๐NC(๐ฃโ, ๐คโ) := (๐ทNC๐ฃโ, ๐ทNC๐คโ)๐ฟ2(ฮฉ) for all (๐ฃโ, ๐คโ) โ ๐โ2
with norm |||โ |||NC := ๐NC(โ , โ ) and seeks ๐ขโโก ๐ โ๐ข โ ๐โsuch that
๐NC(๐ขโ, ๐ฃโ) = ๐(๐, ๐ฃโ) for all ๐ฃโโ ๐โ. (4.2)
A posteriori and a priori error estimates as well as best-approximation properties for this problem are well-studied in the literature [6, 21, 28, 37]. Error estimates in the ๐ฟ2 norm require a modification of the usual
duality argument for conforming finite element methods. The following proposition establishes an ๐ฟ2error estimate. The main ingredient is the use of the companion operator ๐ฝ๐+1. For ๐ = 2, this result was first
ob-tained in [14] and [18]. A similar approach has independently been developed in [41] for ๐ = 2. The result presented here compares the ๐ฟ2error directly with the energy error and therefore uses no a priori results of
the eigenfunction approximation. This is important as the ๐ฟ2control will usually lead to higher-order terms which can be absorbed for โโ0โโโช 1.
Let 0 < ๐ โค 1 indicate the elliptic regularity index of the Poisson problem โฮ๐ข = ๐ with homogeneous Dirichlet boundary conditions in the sense that โ๐ขโ๐ป1+๐ (ฮฉ)โค ๐ถ(๐ )โ๐โ๐ฟ2(ฮฉ).
Proposition 4.1(๐ฟ2error estimate for the linear problem). The exact solution ๐ข to (4.1) and the discrete
solu-tion ๐ขโto (4.2) satisfy
โ๐ข โ ๐ขโโ โฒ โโ0โ ๐
โ|||๐ข โ ๐ขโ|||NC.
Proof. Let ๐ := ๐ข โ ๐ขโand let ๐ง โ ๐ denote the solution of
๐(๐ง, ๐ฃ) = ๐(๐, ๐ฃ) for all ๐ฃ โ ๐.
Recall the companion operator ๐ฝ๐+1from Proposition 3.1. Since ฮ 0โ(๐ขโโ ๐ฝ๐+1๐ขโ) = 0, it holds that
โ๐โ2= ๐(๐ฝ๐+1๐ขโโ ๐ขโ, ๐) + ๐(๐, ๐ข โ ๐ฝ๐+1๐ขโ)
= ๐(๐ฝ๐+1๐ขโโ ๐ขโ, (1 โ ฮ 0โ)๐) + ๐(๐ง, ๐ข โ ๐ฝ๐+1๐ขโ).
Piecewise Poincarรฉ inequalities and (3.4) lead to ๐(๐ฝ๐+1๐ขโโ ๐ขโ, (1 โ ฮ 0 โ)๐) โฒ โโ0โ 2 โ|||๐||| 2 NC.
Since ๐ is perpendicular to the conforming finite element functions in P1(T)โฉ๐ and since ฮ 0โ๐ทNC(๐ขโโ๐ฝ๐+1๐ขโ) =
0, the ScottโZhang quasi-interpolation ๐ง๐ถโ P1(T) โฉ ๐ of ๐ง [45] satisfies
๐(๐ง, ๐ข โ ๐ฝ๐+1๐ขโ) = ๐NC(๐, ๐ง) + ๐NC(๐ขโโ ๐ฝ๐+1๐ขโ, ๐ง)
= ๐NC(๐, ๐ง โ ๐ง๐ถ) + ๐NC(๐ขโโ ๐ฝ๐+1๐ขโ, ๐ง โ ๐ง๐ถ).
The Cauchy inequality and (3.4) imply
๐NC(๐, ๐ง โ ๐ง๐ถ) + ๐NC(๐ขโโ ๐ฝ๐+1๐ขโ, ๐ง โ ๐ง๐ถ) โฒ |||๐|||NC|||๐ง โ ๐ง๐ถ|||NC.
Standard a priori error estimates [9] and the elliptic regularity imply |||๐ง โ ๐ง๐ถ||| โฒ โโ0โ๐ โโ๐งโ๐ป1+๐ (ฮฉ)โฒ โโ0โ๐
โโ๐โ.
The combination of the above estimates proves
The next result states a best-approximation property in any space dimension. It generalises some recent re-sults of the medius analysis [7, 21, 37] to arbitrary space dimensions. The result is stated with a refined oscil-lation term osc1(๐, Tโ). This will be important for the analysis of eigenvalue problems.
Proposition 4.2(best-approximation property). The solution ๐ข โ ๐ to (4.1) with right-hand side ๐ โ ๐ฟ2(ฮฉ) and
the discrete solution ๐ขโโ ๐โto (4.2) satisfy
|||๐ข โ ๐ขโ|||NC โฒ โ(1 โ ฮ 0
โ)๐ท๐ขโ + osc1(๐, Tโ).
Proof. The projection property (3.1) of the nonconforming interpolation operatorICRโ and the Pythagoras theorem show that
|||๐ข โ ๐ขโ|||2NC= |||๐ขโโ I CR โ ๐ข||| 2 NC+ |||๐ข โ I CR โ ๐ข||| 2 NC.
Since |||๐ข โ ICRโ ๐ข|||NC = โ(1 โ ฮ 0โ)๐ท๐ขโ, it remains to estimate the first term on the right-hand side. Set ๐โ :=
๐ขโโ I CR
โ ๐ข. The properties of the companion operator from Proposition 3.1 show that
|||๐ขโโ I CR โ ๐ข|||
2
NC= ๐NC(๐ขโโ ๐ข, ๐โ) = ๐(๐, ๐โโ ๐ฝ๐+1๐โ) + ((1 โ ฮ 0โ)๐ท๐ข, ๐ทNC(๐ฝ๐+1โ 1)๐โ)๐ฟ2(ฮฉ).
The approximation and stability properties (3.4) show that this is bounded by (โโโ๐โ + โ(1 โ ฮ
0
โ)๐ท๐ขโ)|||๐โ|||NC.
The efficiency โโโ๐โ โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ + osc1(๐, Tโ) in the spirit of [49] follows from arguments similar to those
of [33, Proposition 3.1]. This concludes the proof.
4.2 Discretisation of the Laplace Eigenvalue Problem
The Laplace eigenvalue problem seeks eigenpairs (๐, ๐ข) โ โ ร ๐ with โ๐ขโ = 1 such that
๐(๐ข, ๐ฃ) = ๐๐(๐ข, ๐ฃ) for all ๐ฃ โ ๐. (4.3)
The finite element discretisation based on a regular triangulationTโseeks discrete eigenpairs (๐โ, ๐ขโ) โ โร๐โ with โ๐ขโโ = 1 and
๐NC(๐ขโ, ๐ฃโ) = ๐โ๐(๐ขโ, ๐ฃโ) for all ๐ฃโโ ๐โ. (4.4)
Adopt the notation of Section 2 with exact and discrete eigenvalues
0 < ๐1โค ๐2โค โ โ โ and 0 < ๐โ,1โค โ โ โ โค ๐โ,dim(๐โ)
and their corresponding ๐-orthonormal systems of eigenfunctions
(๐ข1, ๐ข2, ๐ข3, . . .) and (๐ขโ,1, ๐ขโ,2, . . . , ๐ขโ,dim(๐โ)).
Recall the definitions of Section 2: The set ๐ฝ = {๐ + 1, . . . , ๐ + ๐} describes the eigenvalue cluster of interest and ๐ := span{๐ข๐ | ๐ โ ๐ฝ} and ๐โ := span{๐ขโ,๐ | ๐ โ ๐ฝ} are the exact and discrete invariant subspaces (not
necessarily eigenspaces) related to the cluster. In the present situation, the quasi-Ritz projection ๐ โmaps the
solution ๐ข โ ๐ of the linear problem (4.1) to the solution ๐ โ๐ข of the discrete linear problem (4.2). With the ๐ฟ2
projection ๐Tโ := ๐โonto ๐โlet ฮTโ := ฮโ:= ๐โโ ๐ โ.
The remaining parts of this subsection prove an ๐ฟ2error estimate as well as a best-approximation result. Proposition 4.3(๐ฟ2error control). Provided โโ
0โโ โช 1, any eigenpair (๐, ๐ข) โ โ ร ๐ with โ๐ขโ = 1 satisfies
โ๐ข โ ๐โ๐ขโ โค โ๐ข โ ฮโ๐ขโ โฒ (1 + ๐๐ฝ)โ๐ข โ ๐ โ๐ขโ โค ๐ถ๐ฟ2(1 + ๐๐ฝ)โโ0โ๐
โ|||๐ข โ ฮโ๐ข|||NC
Proof. Note that ๐ โ๐ข solves (4.2) with right-hand side ๐ := ๐๐ข. The combination of Proposition 2.1 with
Propo-sition 4.1 and PropoPropo-sition 4.2 yields
โ๐ข โ ๐โ๐ขโ โค โ๐ข โ ฮโ๐ขโ โฒ (1 + ๐๐ฝ)โโ0โโ๐ (|||๐ข โ ฮโ๐ข|||NC+ osc1(๐๐ข, Tโ)).
Provided โโ0โโ โช 1, the oscillation term can be absorbed.
Proposition 4.4(best-approximation property). Provided โโ0โโ โช 1, any eigenpair (๐, ๐ข) โ โ ร ๐ of (4.3)
with โ๐ขโ = 1 satisfies
|||๐ข โ ฮโ๐ข|||NC โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ.
Proof. The triangle inequality proves for the quasi-Ritz projection ๐ โ๐ข that
|||๐ข โ ฮโ๐ข|||NCโค |||๐ข โ ๐ โ๐ข|||NC+ |||๐ โ๐ข โ ฮโ๐ข|||NC.
Set ๐โ:= ๐ โ๐ข โ ฮโ๐ข. The definition of ๐ โand the discrete problem (cf. Lemma 2.2) prove that
|||๐ โ๐ข โ ฮโ๐ข|||2NC = ๐NC(๐ โ๐ข โ ฮโ๐ข, ๐โ) = ๐๐(๐ข โ ๐โ๐ข, ๐โ).
Hence, the Cauchy and discrete Friedrichs inequalities [9, Theorem 10.6.12] and the ๐ฟ2control from
Proposi-tion 4.3 prove that
|||๐ โ๐ข โ ฮโ๐ข|||NCโฒ ๐(1 + ๐๐ฝ)โโ0โ ๐
โ|||๐ข โ ฮโ๐ข|||NC.
The combination of the foregoing estimates with Proposition 4.2 results in |||๐ข โ ฮโ๐ข|||NC โฒ โ(1 โ ฮ
0
โ)๐ท๐ขโ + ๐(1 + ๐๐ฝ)โโ0โ ๐
โ|||๐ข โ ฮโ๐ข|||NC+ osc1(๐๐ข, Tโ).
If โโ0โโ โช 1 is sufficiently small, the higher-order terms on the right-hand side can be absorbed.
4.3 Theoretical Error Estimator and Discrete Reliability
The analysis relies on a theoretical, non-computable error estimator that does not depend on the choice of the discrete eigenfunctions. This idea was first presented in [25]. Given an eigenpair (๐, ๐ข), the error estimator includes the elementwise residuals in terms of ๐โ๐ข and ฮโ๐ข. More precisely, define, for any ๐ โ Tโ,
๐โ2(๐, ๐, ๐ข) := โ 2 ๐โ๐๐โ๐ขโ 2 ๐ฟ2(๐)+ โ ๐นโF(๐) โโ1๐โ[ฮโ๐ข]๐นโ 2 ๐ฟ2(๐น)
and, for any subsetK โ Tโ,
๐2โ(K, ๐๐, ๐ข๐) := โ ๐โK ๐2โ(๐, ๐๐, ๐ข๐) and ๐ 2 โ(K) := โ ๐โ๐ฝ ๐โ2(K, ๐๐, ๐ข๐).
The following shorthand notation for higher-order terms will be frequently used in the remaining parts of this section. For (โ, ๐) โ โ20define (with the constant ๐ถ๐ฟ2from Proposition 4.3)
๐โ,๐ := โโ0โ๐ โ๐(1 + ๐๐ฝ)๐ถ๐ฟ2โ|||๐ข โ ฮโ๐ข|||2+ |||๐ข โ ฮโ+๐๐ข|||2. (4.5)
The theoretical error estimator satisfies the following discrete reliability.
Proposition 4.5(discrete reliability). There exists a constant ๐ถdrelโ 1 solely dependent on T0with โโ0โโ โช 1
such that any eigenpair (๐, ๐ข) โ โ ร ๐ of (4.3) with โ๐ขโ = 1 satisfies
2|||ฮโ+๐๐ข โ ฮโ๐ข|||2โค ๐ถ2drel(๐ 2
โ(Tโ\ Tโ+๐, ๐, ๐ข) + ๐2โ,๐).
Proof. Let ๐ฃโ+๐denote the best-approximation (with respect to the norm |||โ |||NC) of ฮโ๐ข in ๐โ+๐. The
Pythago-ras theorem reads as
|||(ฮโ+๐โ ฮโ)๐ข||| 2 NC= |||ฮโ+๐๐ข โ ๐ฃโ+๐||| 2 NC+๐ค min โ+๐โ๐โ+๐ |||๐คโ+๐โ ฮโ๐ข||| 2 NC.
The second term has been estimated in [13, Theorem 3.1] by means of the jumps of ฮโ๐ข. For the analysis of
the first term, let ๐โ+๐:= ฮโ+๐๐ข โ ๐ฃโ+๐. The projection property (3.1) of the nonconforming interpolation and
the discrete eigenvalue problems (cf. Lemma 2.2) reveal that |||ฮโ+๐๐ข โ ๐ฃโ+๐||| 2 NC= ๐NC((ฮโ+๐โ ฮโ)๐ข, ๐โ+๐) = ๐๐((๐โ+๐โ ๐โ)๐ข, ๐โ+๐) + ๐๐(๐โ๐ข, (1 โ I CR โ )๐โ+๐).
The ๐ฟ2error estimate from Proposition 4.3 and the approximation and stability property (3.2) conclude the
proof.
The reliability of the error estimator is an immediate consequence.
Proposition 4.6(reliability and efficiency). Provided โโ0โโ โช 1, any eigenpair (๐, ๐ข) โ โ ร ๐ of (4.3) with โ๐ขโ = 1 satisfies |||๐ข โ ฮโ๐ข|||2NC โค ๐ถ 2 drel๐ 2 โ(Tโ, ๐, ๐ข). (4.6)
For some constant ๐ถeff โ 1, it holds that
๐โ(Tโ, ๐, ๐ข) 2
โค ๐ถ2eff|||๐ข โ ฮโ๐ข||| 2
NC. (4.7)
Proof. The reliability
2|||๐ข โ ฮโ๐ข||| 2 NCโค ๐ถ 2 drel(๐ 2 โ(Tโ, ๐, ๐ข) + โโ0โ 2๐ โ๐ 2 (1 + ๐๐ฝ) 2 |||๐ข โ ฮโ๐ข||| 2 NC)
follows from the discrete reliability on a sequence of meshesTโ+๐with โโโ+๐โโ โ 0 and the a priori con-vergence result of Proposition 4.4. Provided the initial mesh is sufficiently fine, the higher-order terms on the right-hand side can be absorbed. The efficiency
2๐2โ(Tโ, ๐, ๐ข) โค ๐ถ 2
eff(1 + ๐โโ0โ 1+๐
โ (1 + ๐๐ฝ)๐ถ๐ฟ2)2|||๐ข โ ฮโ๐ข|||2NC
follows from the triangle inequality and the ๐ฟ2error control from Proposition 4.3 combined with the standard arguments of [49]. The assumption โโ0โโ โช 1 implies
๐2โ(Tโ, ๐, ๐ข) โค ๐ถ2eff|||๐ข โ ฮโ๐ข|||2NC.
4.4 Adaptive Algorithm and Contraction Property
This subsection presents the adaptive algorithm and proves the contraction property.
For any simplex ๐ โ Tโ, the explicit residual-based error estimator consists of the sum of the residuals of
the computed discrete eigenfunctions (๐ขโ,๐)๐โ๐ฝ,
๐2โ(๐) := โ ๐โ๐ฝ
(โ2๐โ๐โ,๐๐ขโ,๐โ2๐ฟ2(๐)+ โ
๐นโF(๐)
โโ1๐ โ[๐ขโ,๐]๐นโ2๐ฟ2(๐น)).
Let, for any subsetK โ T,
๐2โ(K) := โ ๐โK
๐2โ(๐).
For simple eigenvalues this type of error estimator was introduced in [29]. The adaptive algorithm is driven by this computable error estimator and runs the following loop.
Algorithm 4.7(nonconforming AFEM for the Laplace eigenproblem). Input: Initial triangulationT0, bulk parameter 0 < ๐ โค 1.
for โ = 0, 1, 2, . . .
Solve. Compute discrete eigenpairs (๐โ,๐, ๐ขโ,๐)๐โ๐ฝof (4.4) with respect toTโ.
Mark. Choose a minimal subsetMโโ Tโsuch that ๐๐2โ(Tโ) โค ๐โ2(Mโ). Refine. GenerateTโ+1:= refine(Tโ, Mโ) with the refinement rules of [47].
end for
Output: Triangulations (Tโ)โand discrete solutions ((๐โ,๐, ๐ขโ,๐)๐โ๐ฝ)โ.
The first important observation is that, by Lemma 2.3, the non-computable error estimator ๐โ(Mโ) satisfies
the bulk criterion
ฬ
๐๐โ(Tโ) โค ๐โ(Mโ)
for the modified bulk parameter
ฬ
๐ := ((๐ต/๐ด)4(2๐2+ 4๐3))โ1๐ < 1. (4.8) The following proposition states the error estimator reduction property.
Proposition 4.8(error estimator reduction for ๐โ). Provided the assumptions (H1) and (H2) (see Lemma 2.3)
hold, there exist constants 0 < ๐1< 1 and 0 < ๐พ < โ such that Tโand its one-level refinementTโ+1generated
by Algorithm 4.7 and any eigenfunction ๐ข โ ๐ with โ๐ขโ = 1 and eigenvalue ๐ satisfy (with ๐โ,1from (4.5))
๐2โ+1(Tโ+1, ๐, ๐ข) โค ๐1๐2โ(Tโ, ๐, ๐ข) + ๐พ(|||ฮโ+1๐ข โ ฮโ๐ข|||2NC+ โโ0โ2โ๐ 2 โ,1).
Proof. The standard techniques of [22, 46] and the bulk criterion (4.8) lead to a constant ฬ๐พ such that
๐โ+12 (Tโ+1, ๐, ๐ข) โค ๐1๐2โ(Tโ, ๐, ๐ข) + ฬ๐พ(|||ฮโ+1๐ข โ ฮโ๐ข|||2NC+ โโโ+1๐(๐โ+1โ ๐โ)๐ขโ2).
The triangle inequality for the term โโโ+1๐(๐โ+1โ ๐โ)๐ขโ and the ๐ฟ2error control from Proposition 4.3 prove the
result.
The next technical result is needed for the reduction of the volume contribution of the error estimator. In-equalities of this type were previously utilised in [42] for ๐ = 2 for the linear Poisson problem and in [13] for boundary value problems for ๐ โฅ 2.
Lemma 4.9(control of the volume contribution). Provided โโ0โโ โช 1, any triangulation Tโ โ ๐ and any
ad-missible refinementTโ+๐ofTโsatisfy for any 0 < ๐ฟ < โ and any eigenpair (๐, ๐ข) โ โ ร ๐ of (4.3) with โ๐ขโ = 1 that โโโ+๐๐๐โ+๐๐ขโ 2 ๐ฟ2(ฮฉ)+ (1 + ๐ฟโ1)(1 โ 2โ2/๐)โโโ๐๐โ๐ขโ2๐ฟ2(โช(T โ\Tโ+๐))โค 2(1 + ๐ฟ)โโ0โ 2 โ๐ 2 โ,๐+ (1 + ๐ฟ โ1 )โโโ๐๐โ๐ขโ 2 ๐ฟ2(ฮฉ).
Proof. The triangle and Young inequalities prove for any 0 < ๐ฟ < โ that
โโโ+๐๐๐โ+๐๐ขโ2๐ฟ2(ฮฉ)โค (1 + ๐ฟ)โโโ+๐๐(๐โ+๐๐ข โ ๐โ๐ข)โ2๐ฟ2(ฮฉ)+ (1 + ๐ฟโ1)โโโ+๐๐๐โ๐ขโ2๐ฟ2(ฮฉ). The relation โ๐โ+๐โค โ ๐ โ/2 on Tโ\ Tโ+๐proves โโโ๐๐โ๐ขโ 2 ๐ฟ2(โช(T โ\Tโ+๐))โค (1 โ 2 โ2/๐ )โ1(โโโ๐๐โ๐ขโ 2 ๐ฟ2(ฮฉ)โ โโโ+๐๐๐โ๐ขโ2๐ฟ2(ฮฉ)).
The preceding two displayed formulas together with Proposition 4.3 prove the result.
In the case of nonconforming discretisations of eigenvalue problems, the Galerkin orthogonality is violated at two points. First, the nonlinearity leads to a perturbation of the right-hand side. Furthermore, the non-conforming finite element functions are not admissible test functions in the continuous problem and, thus, additional techniques enter the analysis. The notion of โquasi-orthogonalityโ traces back to [17].
Proposition 4.10(quasi-orthogonality). Under the hypothesis โโ0โโ โช 1 there exists a constant ๐ถqosuch that
any eigenpair (๐, ๐ข) โ โ ร ๐ of (4.3) with โ๐ขโ = 1, any Tโโ ๐, and any admissible refinement Tโ+๐ofTโsatisfy
|2๐NC(๐ข โ ฮโ+๐๐ข, ฮโ+๐๐ข โ ฮโ๐ข)| โค ๐ถqo(โโโ๐๐โ๐ขโ๐ฟ2(โชT
Proof. Some algebraic manipulations with the projection property (3.1) of the nonconforming interpolation
and the discrete eigenvalue problems (cf. Lemma 2.2) reveal ๐NC((1 โ ฮโ+๐)๐ข, (ฮโ+๐โ ฮโ)๐ข) = ๐NC(ฮโ+๐๐ข, I CR โ+๐(1 โ ฮโ+๐)๐ข) โ ๐NC(ฮโ๐ข, I CR โ (1 โ ฮโ+๐)๐ข) = ๐๐(๐โ+๐๐ข, I CR โ+๐(1 โ ฮโ+๐)๐ข) โ ๐๐(๐โ๐ข, I CR โ (1 โ ฮโ+๐)๐ข) = ๐๐(๐โ๐ข, (I CR โ+๐โ I CR โ )(1 โ ฮโ+๐)๐ข) + ๐๐((๐โ+๐โ ๐โ)๐ข, I CR โ+๐(1 โ ฮโ+๐)๐ข).
SinceICRโ+๐๐ฃ|๐ = ICRโ ๐ฃ|๐for all ๐ โ Tโโฉ Tโ+๐, the first term of the right-hand side can be controlled with (3.2) as ๐๐(๐โ๐ข, (I CR โ+๐โ I CR โ )(1 โ ฮโ+๐)๐ข) โฒ โโโ๐๐โ๐ขโ๐ฟ2(โชT โ\Tโ+๐)โ๐ทNC(1 โ ฮโ+๐)๐ขโ๐ฟ2(โชTโ\Tโ+๐).
For the second term, the discrete Friedrichs inequality [9, Theorem 10.6.12] and the stability ofICRโ reveal ๐๐((๐โ+๐โ ๐โ)๐ข, I
CR
โ+๐(1 โ ฮโ+๐)๐ข) โฒ ๐โ(๐โ+๐โ ๐โ)๐ขโ|||๐ข โ ฮโ+๐๐ข|||NC.
The triangle inequality and Proposition 4.3 control the term ๐โ(๐โ+๐โ ๐โ)๐ขโ by ๐โ,๐from (4.5). This concludes
the proof.
The following contraction property implies the convergence of the adaptive algorithm.
Proposition 4.11(contraction property). Under the condition โโ0โโ โช 1, there exist 0 < ๐2 < 1 and 0 < ๐ฝ, ๐พ < โ such that, for any eigenpair (๐, ๐ข) โ โร๐ with โ๐ขโ = 1, the term ๐2โ:= ๐2โ(Tโ, ๐, ๐ข)+๐ฝ|||๐ขโฮโ๐ข|||2NC+๐พโโโ๐โ๐ขโ2
satisfies
๐2โ+1โค ๐2๐ 2
โ for all โ โ โ0.
Proof. Throughout the proof, the following shorthand notation applies:
๐โ:= |||๐ข โ ฮโ๐ข|||NC, ๐โ+1:= |||๐ข โ ฮโ+1๐ข|||NC, ๐ 2 โ := ๐ 2 โ(Tโ, ๐, ๐ข), ๐ 2 โ+1:= ๐ 2 โ+1(Tโ+1, ๐, ๐ข).
The error estimator reduction from Proposition 4.8 and elementary algebraic manipulations plus the quasi-orthogonality (Proposition 4.10) lead to
๐2โ+1+ ๐พ๐ 2 โ+1โค ๐1๐ 2 โ+ ๐พ(๐ 2 โ+ 2๐(๐ข โ ฮโ+1๐ข, (ฮโโ ฮโ+1)๐ข) + โโ0โ 2 โ๐ 2 โ,1) โค ๐1๐ 2 โ+ ๐พ(๐ 2 โ+ ๐ถqo(โโโ๐๐โ๐ขโ๐ฟ2(โชT โ\Tโ+1)+ ๐โ,1)๐โ+1+ โโ0โ 2 โ๐ 2 โ,1).
This and the Young inequality for any 0 < ๐ < 1 lead to ๐2โ+1+ ๐พ(1 โ ๐ถqo๐/2)๐2โ+1โค ๐1๐โ2+ ๐พ(๐ 2 โ+ ๐ถqo/๐(โโโ๐๐โ๐ขโ2๐ฟ2(โชT โ\Tโ+๐)+ ๐ 2 โ,1) + โโ0โ2โ๐ 2 โ,1).
The reliability (4.6) proves for any 0 < ๐ < โ that this is bounded by (๐1+ ๐พ๐๐ถ 2 drel)๐ 2 โ+ ๐พ((1 โ ๐)๐ 2 โ+ ๐ถqo/๐(โโโ๐๐โ๐ขโ 2 ๐ฟ2(โชT โ\Tโ+๐)+ ๐ 2 โ,1) + โโ0โ 2 โ๐ 2 โ,1).
Lemma 4.9 states for any 0 < ๐ฟ < โ and ๐๐:= (1 โ 2โ2/๐) that
โโโ๐๐โ๐ขโ2๐ฟ2(โช(T โ\Tโ+1))โค 2๐ฟโโ0โ2โ๐2โ,1 ๐๐ + โโโ๐๐โ๐ขโ 2 ๐๐ โ โโโ+1๐๐โ+1๐ขโ 2 (1 + ๐ฟโ1)๐ ๐ . Altogether, ๐2โ+1+ ๐พ((1 โ ๐ถqo๐/2)๐2โ+1+ ๐ถqoโโโ+1๐๐โ+1๐ขโ2 ๐(1 + ๐ฟโ1)๐ ๐ ) โค (๐1+ ๐พ๐๐ถ2drel)๐ 2 โ+ ๐พ((1 โ ๐)๐ 2 โ+ (๐ โ1 ๐ถqo(1 + 2๐ฟโโ0โ2โ/๐๐) + โโ0โ2โ)๐ 2 โ,1+ ๐ถqoโโโ๐๐โ๐ขโ2 ๐๐๐ ).
Define ๐ก(โ0, ๐, ๐ฟ) := ๐ถ2drelโโ0โ2๐ โ๐ 2 (1 + ๐๐ฝ)2๐ถ2๐ฟ2๐พ(๐โ1๐ถqo(1 + 2๐ฟโโ0โ2โ ๐๐ ) + โโ0โ2โ).
Recall the definition (4.5) of ๐โ,1. The reliability (4.6) implies ๐พ(๐โ1๐ถqo(1 + 2๐ฟโโ0โโ2 /๐๐) + โโ0โ2โ)๐ 2
โ,1โค ๐ก(โ0, ๐, ๐ฟ)(๐โ2+ ๐ 2 โ+1).
This and the fact that โโโ๐๐โ๐ขโ2โค ๐2โtogether with the foregoing estimates prove
(1 โ ๐ก(โ0, ๐, ๐ฟ))๐2โ+1+ ๐พ((1 โ ๐ถqo๐/2)๐2โ+1+ ๐ถqoโโโ+1๐๐โ+1๐ขโ2 ๐(1 + ๐ฟโ1)๐ ๐ ) โค (๐1+ ๐พ๐๐ถ2drel+ ๐ก(โ0, ๐, ๐ฟ) + ๐พ๐)๐2โ+ ๐พ((1 โ ๐)๐ 2 โ+ ( ๐ถqo ๐๐๐ โ ๐)โโโ๐๐โ๐ขโ2). Hence, for ๐ฝ := ๐พ(1 โ ๐ถqo๐/2) 1 โ ๐ก(โ0, ๐, ๐ฟ) , ๐พ := ๐พ๐ถqo ๐(1 + ๐ฟโ1)๐ ๐(1 โ ๐ก(โ0, ๐, ๐ฟ)) , and ๐2:= max{ ๐1+ ๐พ๐๐ถ2drel+ ๐ก(โ0, ๐, ๐ฟ) + ๐พ๐ 1 โ ๐ก(โ0, ๐, ๐ฟ) , 1 โ ๐ 1 โ ๐ถqo๐/2 , (1 + ๐ฟ โ1 )(๐ถqoโ ๐2๐๐) ๐ถqo }, it follows that ๐โ+1+ ๐ฝ๐ 2 โ+1+ ๐พโโโ+1๐๐โ+1๐ขโ 2 โค ๐2(๐โ+ ๐ฝ๐ 2 โ+ ๐พโโโ๐๐โ๐ขโ 2 ).
Choose ๐ฟ := ๐ถqo/(๐2๐๐) and ๐ < 2๐๐ถโ1qo. The choice of sufficiently small ๐, ๐ and โโ0โโyields ๐2< 1.
4.5 Optimal Convergence Rates
Let, for any ๐ โ โ, the set of triangulations in ๐ whose cardinality differs from that of T0by ๐ or less be
denoted by
๐(๐) := {T โ ๐ | card(T) โ card(T0) โค ๐}.
Define the seminorm
|๐ข|A๐ := sup ๐โโ๐ ๐ inf Tโ๐(๐)โ(1 โ ฮ 0 T)๐ท๐ขโ
and the approximation class
A๐:= {๐ฃ โ ๐ | |๐ฃ|A๐ < โ}. Define the following alternative set, also referred to as approximation class
ANC,ฮ๐ := {๐ข โ ๐ | |๐ข|ANC,ฮ ๐ < โ} for |๐ข|ANC,ฮ๐ := sup ๐โโ๐ ๐ inf Tโ๐(๐)|||๐ข โ ฮT๐ข|||NC
for the eigenfunction approximation ฮT๐ข with respect to a triangulation T. Proposition 4.4 proves that these
two approximation classes are equivalent in the sense that any eigenfunction ๐ข โ ๐ belongs to A๐if and
only if it belongs to ANC,ฮ๐ . The following theorem states optimality of Algorithm 4.7. The proof follows in the remaining parts of this section.
Theorem 4.12(optimal convergence rates). Provided the bulk parameter ๐ โช 1 and the initial mesh-size โโ0โโ โช 1 are sufficiently small, Algorithm 4.7 computes sequences of triangulations (Tโ)โand discrete
eigen-pairs ((๐โ,๐, ๐ขโ,๐)๐โ๐ฝ)โwith optimal rate of convergence in the sense that, for some constant ๐ถopt,
sup โโโ (card(Tโ) โ card(T0)) 2๐ โ ๐โ๐ฝ |||๐ข๐โ ฮโ๐ข๐|||2NCโค ๐ถoptโ ๐โ๐ฝ |๐ข๐|2ANC,ฮ๐ .
Proposition 4.4 implies the following immediate consequence.
Corollary 4.13. Provided the bulk parameter ๐ โช 1 and the initial mesh-size โโ0โโ โช 1 are sufficiently small,
Algorithm 4.7 computes triangulations (Tโ)โand discrete eigenpairs ((๐โ,๐, ๐ขโ,๐)๐โ๐ฝ)โwith optimal rate of
conver-gence in the sense that
sup โโโ(card(Tโ) โ card(T0)) ๐ sup ๐คโ๐ โ๐คโ=1 inf ๐ฃโโ๐โ|||๐ค โ ๐ฃโ|||NCโฒ ( โ ๐โ๐ฝ |๐ข๐|2A๐) 1/2 .
The remaining part of this subsection is devoted to the proof of Theorem 4.12 which follows the methodol-ogy of [22, 46] as in [33]. The optimality proof of this section is concerned with the simultaneous error of all eigenfunction approximations. Consider
ฮ2โ:= ๐ 2 โ(Tโ) + ๐ฝ โ ๐โ๐ฝ |||๐ข๐โ ฮโ๐ข๐|||2NC+ ๐พ โ ๐โ๐ฝ โโโ๐๐๐โ๐ข๐โ2 for all โ โ โ0
for the parameters ๐ฝ and ๐พ from Proposition 4.11. The proof excludes the pathological case ฮ0 = 0. Choose
0 < ๐ โค โ๐โ๐ฝ|๐ข๐|2ANC,ฮ๐ /ฮ
2
0, and set ๐(โ) := โ๐ ฮโ. Let ๐(โ) โ โ be minimal with the property
โ ๐โ๐ฝ |๐ข๐|2ANC,ฮ ๐ โค ๐(โ) 2 ๐(โ)2๐.
Let for a fixed โ โ โ, ฬTโโ ๐ denote the optimal triangulation of cardinality
card(ฬTโ) โค card(T0) + ๐(โ)
in the sense that the projection ฬฮ := ฮฬT
โwith respect to ฬTโsatisfies
โ ๐โ๐ฝ |||๐ข๐โ ฬฮ๐ข๐||| 2 NCโค ๐(โ) โ2๐ โ ๐โ๐ฝ |๐ข๐| 2 ANC,ฮ๐ โค ๐(โ) 2 (4.9)
and define ฬTโ:= Tโโ ฬTโas the overlay [22], that is, the smallest common refinement ofTโand ฬTโ. The argu-ments of [22, 33] lead to card(Tโ\ ฬTโ) โค ๐(โ) โค 2( โ ๐โ๐ฝ |๐ข๐|2ANC,ฮ๐ ) 1/(2๐) ๐(โ)โ1/๐. Let ฬฮ := ฮฬT
โdenote the projection with respect to ฬTโ.
Lemma 4.14. Provided โโ0โโโช 1, it holds that
โ ๐โ๐ฝ |||๐ข๐โ ฬฮ๐ข๐||| 2 NC โฒ ๐(โ) 2 .
Proof. Recall that by definition of the overlay [22] the triangulations ฬTโand ฬTโare nested. Hence, the best-approximation result of Proposition 4.4 and (4.9) prove
โ ๐โ๐ฝ |||๐ข๐โ ฬฮ๐ข๐|||2NCโฒ โ ๐โ๐ฝ |||๐ข๐โ ฬฮ๐ข๐|||2NCโค ๐(โ) 2 . Lemma 4.15(key argument). Provided โโ0โโโช 1, there exists ๐ถ2โ 1 such that
๐โ2(Tโ) โค ๐ถ2๐2โ(Tโ\ ฬTโ).
Proof. The triangle inequality and the Young inequality imply for any ๐ โ ๐ฝ that
|||๐ข๐โ ฮโ๐ข๐|||2NCโค 2|||๐ข๐โ ฬฮ๐ข๐|||2NC+ 2|||ฬฮ๐ข๐โ ฮโ๐ข๐|||2NC.
Hence, the discrete reliability from Proposition 4.5 leads to |||๐ข๐โ ฮโ๐ข๐|||2NCโค (2 + ๐ถ 2 drel๐ 2 ๐โโ0โ2๐ โ(1 + ๐๐ฝ)2๐ถ๐ฟ22)|||๐ข๐โ ฬฮ๐ข๐|||2NC + ๐ถ2drel๐ 2 ๐โโ0โ2๐ โ(1 + ๐๐ฝ)2๐ถ2๐ฟ2|||๐ข๐โ ฮโ๐ข๐|||2NC+ ๐ถ2drel๐โ2(Tโ\ ฬTโ, ๐๐, ๐ข๐).
The term with |||๐ข๐โฮโ๐ข๐|||2NCcan be absorbed for sufficiently small โโ0โโโช 1. Therefore, Lemma 4.14 implies
for constants ๐ถ3โ 1 โ ๐ถ4and โโ0โโโช 1 that
โ
๐โ๐ฝ
|||๐ข๐โ ฮโ๐ข๐|||2NC โค ๐ถ3๐(โ)2+ ๐ถ4๐โ2(Tโ\ ฬTโ).
Let ๐ถeqdenote the constant of ๐ถ3ฮ2โโค ๐ถeq๐2โ(Tโ) (which exists by reliability). The efficiency (4.7), the definition
of ๐(โ) and the preceding estimates prove ๐ถeffโ2๐ 2 โ(Tโ) โค ๐ถ3๐(โ) 2 + ๐ถ4๐ 2 โ(Tโ\ ฬTโ) โค ๐๐ถeq๐ 2 โ(Tโ) + ๐ถ4๐ 2 โ(Tโ\ ฬTโ).
For a sufficiently small choice of ๐, the constant ๐ถ2:= (๐ถโ2effโ ๐๐ถeq)โ1๐ถ4is positive.
The finish of the optimality proof follows the arguments of [22, 46]. The proof is identical to that of [33, Lemma 7.3] and therefore omitted.
Lemma 4.16(finish of the optimality proof). The choice
0 < ๐ โค 1/(๐ถ2(๐ต/๐ด)4(2๐2+ 4๐3))
implies the existence of a constant ๐ถ(๐) such that
(card(Tโ) โ card(T0)) ๐ ( โ ๐โ๐ฝ |||๐ข๐โ ฮโ๐ข๐|||2NC) 1/2 โค ๐ถ(๐)( โ ๐โ๐ฝ |๐ข๐|2ANC,ฮ๐ ) 1/2 .
5 Eigenvalues of the Stokes System
This section studies the adaptive nonconforming FEM approximation of the Stokes eigenproblem. Section 5.1 presents new ๐ฟ2and best-approximation estimates for the linear Stokes equations. Section 5.2 introduces the
discretisation of the eigenvalue problem. A theoretical error estimator and its discrete reliability are analysed in Section 5.3. Sections 5.4 and 5.5 present the practical AFEM and prove contraction and optimal convergence rates. Whenever there is no significant modification compared to the case of the eigenvalues of the Laplacian, the arguments are merely sketched.
5.1 Nonconforming Discretisation of the Stokes Equations
One important advantage of the nonconformingP1finite element method is that it provides a stable low-order discretisation of the Stokes equations [24]. The strong form of the linear Stokes equations for a given force ๐ seeks the velocity field ๐ข and the pressure ๐ such that
โฮ๐ข + (๐ท๐)โค= ๐ and div ๐ข = 0 in ฮฉ, ๐ข|๐ฮฉ= 0.
Conforming finite elements satisfying the constraint div ๐ข = 0 pointwise a.e. are rather complicated, see [38, 44]. The nonconformingP1finite element satisfies the favourable local mass-conservation property for the piecewise divergence.
Let ๐ := [๐ป1
0(ฮฉ)]๐and ๐ := ๐ฟ20(ฮฉ) := {๐ โ ๐ฟ2(ฮฉ) | โซฮฉ๐ ๐๐ฅ = 0} and define the bilinear form
๐(๐ฃ, ๐ค) := (๐ท๐ฃ, ๐ท๐ค)๐ฟ2(ฮฉ) for all (๐ฃ, ๐ค) โ ๐2
with induced norm |||โ |||. Furthermore define
๐(๐ฃ, ๐) := โ(div ๐ฃ, ๐)๐ฟ2(ฮฉ) for all (๐ฃ, ๐) โ ๐ ร ๐
Given ๐ โ [๐ฟ2(ฮฉ)]๐, the linear Stokes problem seeks (๐ข, ๐) โ ๐ ร ๐ such that ๐(๐ข, ๐ฃ) + ๐(๐ฃ, ๐) = ๐(๐, ๐ฃ) for all ๐ฃ โ ๐,
๐(๐ข, ๐) = 0 for all ๐ โ ๐. (5.1)
This mixed system can be reformulated as an elliptic problem. Let ๐ := {๐ฃ โ ๐ | div ๐ฃ = 0} denote the space of divergence-free vector fields. Problem (5.1) is equivalent to seeking ๐ข โ ๐ such that
๐(๐ข, ๐ฃ) = ๐(๐, ๐ฃ) for all ๐ฃ โ ๐ (5.2)
and the pressure variable ๐ plays the role of a Lagrange multiplier. The equivalence with (5.1) follows from the Ladyzhenskaya lemma [1, 9] which states that the divergence operator div : ๐ โ ๐ has a continuous right-inverse. Note that (5.1) carries more information than (5.2) in the sense that the pressure variable ๐ extracts information from ๐ โ [๐ฟ2(ฮฉ)]๐even if ๐ is zero as an element of the dual space ๐โ.
The nonconformingP1finite element discretisation of the linear Stokes equations is based on the non-conforming finite element space ๐โ:= [CR10(Tโ)]๐and ๐โ:= P0(Tโ) โฉ ๐ฟ20(ฮฉ) and the bilinear forms
๐NC(๐ฃโ, ๐คโ) := (๐ทNC๐ฃโ, ๐ทNC๐คโ)๐ฟ2(ฮฉ) for all (๐ฃโ, ๐คโ) โ ๐2
โ
with induced norm |||โ |||NCand
๐NC(๐ฃโ, ๐โ) := โ(divNC๐ฃโ, ๐โ)๐ฟ2(ฮฉ) for all (๐ฃโ, ๐โ) โ ๐โร ๐โ.
The nonconforming FEM seeks (๐ขโ, ๐โ) โ ๐โร ๐โsuch that
๐NC(๐ขโ, ๐ฃโ) + ๐NC(๐ฃโ, ๐โ) = ๐(๐, ๐ฃโ) for all ๐ฃโโ ๐โ,
๐NC(๐ขโ, ๐โ) = 0 for all ๐โโ ๐โ.
(5.3)
The well-posedness follows from the discrete inf-sup condition [4] 0 < ๐ฝ โค inf ๐โโ๐โ\{0} sup ๐ฃโโ๐โ\{0} ๐NC(๐ฃโ, ๐โ) |||๐ฃโ|||NCโ๐โโ . (5.4)
Obviously, the discrete solution ๐ขโof (5.3) is piecewise divergence-free, divNC๐ขโ = 0. The equivalent
formu-lation based on the space ๐โ:= {๐ฃโโ ๐โ| divNC๐ฃโ= 0} reads as
๐NC(๐ขโ, ๐ฃโ) = ๐(๐, ๐ฃโ) for all ๐ฃโโ ๐โ. (5.5)
Note that the nonconforming interpolation operatorICRโ maps the space ๐ onto ๐โ. This follows from the projection property (3.1). It is well-established in the literature [27] and follows from the discrete inf-sup condition (5.4) of the system (5.3) that the error in the pressure variable can be controlled as
โ๐ โ ๐โโ โฒ โโโ๐โ + |||๐ข โ ๐ขโ|||NC. (5.6)
The main difference with respect to the analysis of the Laplace operator is that the pressure variable enters the analysis even if one considers the elliptic formulations (5.2) and (5.5). One reason is that the companion operator ๐ฝ๐+1from Proposition 3.1 does not map the space ๐โon ๐ only. Also the efficiency error estimate of
the volume term โโโ๐โ leads to a pressure term on the right-hand side.
The following best-approximation result has been proved in [19] with techniques from the medius anal-ysis [37] for the case ๐ = 2:
โ๐ โ ๐โโ + |||๐ข โ ๐ขโ|||NCโฒ โ(1 โ ฮ 0
โ)๐โ + โ(1 โ ฮ 0
โ)๐ท๐ขโ + osc0(๐, Tโ).
The following result gives a generalisation to ๐ โฅ 2 space dimensions with a refined oscillation term. Proposition 5.1(best-approximation result). Let ๐ โ [๐ฟ2(ฮฉ)]๐. Then, the solution (๐ข, ๐) โ ๐ ร ๐ of (5.1) and
the discrete solution (๐ขโ, ๐โ) โ ๐โร ๐โof (5.3) satisfy
|||๐ข โ ๐ขโ|||NC+ โ๐ โ ๐โโ โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ + โ(1 โ ฮ 0
Proof. The projection property (3.1) of the nonconforming interpolation operatorICRโ and the Pythagoras theorem show that
|||๐ข โ ๐ขโ|||2NC = |||๐ขโโ I CR โ ๐ข||| 2 NC+ |||๐ข โ I CR โ ๐ข||| 2 NC.
Since |||๐ข โ ICRโ ๐ข|||NC = โ(1 โ ฮ 0โ)๐ท๐ขโ, it remains to estimate the first term on the right-hand side. Set ๐โ :=
๐ขโโ I CR
โ ๐ข. The properties of the companion operator from Proposition 3.1 and divNC๐ขโ = 0 = divNCI CR โ ๐ข show that |||๐ขโโ I CR โ ๐ข||| 2 NC= ๐NC(๐ขโโ ๐ข, ๐โ) = ๐(๐, ๐โโ ๐ฝ๐+1๐โ) โ ๐NC(๐โโ ๐ฝ๐+1๐โ, (1 โ ฮ 0โ)๐) + ((1 โ ฮ 0 โ)๐ท๐ข, ๐ทNC(๐ฝ๐+1โ 1)๐โ)๐ฟ2(ฮฉ).
The approximation and stability properties (3.4) show that this is bounded by (โโโ๐โ + โ(1 โ ฮ 0 โ)๐โ + |||๐ขโโ I CR โ ๐ข|||NC)|||๐โ|||NC. The efficiency โโโ๐โ โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ + โ(1 โ ฮ 0
โ)๐โ + osc1(๐, Tโ) in the sense of [49] follows from arguments
similar to those of [33, Proposition 3.1]. This and (5.6) conclude the proof. Remark 5.2. One may ask whether possibly an estimate of the type
|||๐ข โ ๐ขโ|||NCโฒ โ(1 โ ฮ 0
โ)๐ท๐ขโ + oscillations
may be valid. To see that the estimate is indeed untrue, consider the case of a simply-connected domain ฮฉ for ๐ = 2 and the constant right-hand side ๐ = (1, 1). Clearly, ๐ is an irrotational vector field which implies that there is a function ๐ โ ๐ป1(ฮฉ) such that ๐ = ๐ท๐. The integration by parts therefore shows that
๐(๐, ๐ฃ) = 0 for all ๐ฃ โ ๐.
Hence, ๐ข = 0 and the right-hand side of the estimate equals zero, while the left-hand side equals |||๐ขโ|||NC. The
latter, however, is not zero because ๐ does not represent the zero functional in the dual space ๐โโ, although
it is zero in ๐โ. This is due to the fact that the integration by parts with functions ๐ฃ
โโ ๐โleads to additional
jump terms.
The next result is an ๐ฟ2error estimate for arbitrary regularity of the solution. Let 0 < ๐ โค 1 indicate the elliptic regularity of the problem (5.1) in the sense that [31, 43]
โ๐ขโ๐ป1+๐ (ฮฉ)+ โ๐โ๐ป๐ (ฮฉ)โค ๐ถ(๐ )โ๐โ๐ฟ2(ฮฉ). (5.7)
Proposition 5.3(๐ฟ2error control for the linear Stokes problem). The exact solution (๐ข, ๐) โ ๐ร๐ of the linear
problem (5.1) and its nonconforming finite element approximation (๐ขโ, ๐โ) โ ๐โร ๐โfrom (5.3) satisfy
โ๐ข โ ๐ขโโ โฒ โโโโ ๐
โ(|||๐ข โ ๐ขโ|||NC+ โ๐ โ ๐โโ + osc1,1(๐, Tโ)).
Proof. Let (๐ง, ๐) โ ๐ ร ๐ denote the solution of problem (5.1) with right-hand side ๐ := ๐ข โ ๐ขโand set ๐ฃ :=
๐ข โ ๐ฝ๐+1๐ขโfor the companion operator ๐ฝ๐+1from Proposition 3.1. Since ฮ 0โ(๐ขโโ ๐ฝ๐+1๐ขโ) = 0, it holds that
โ๐โ2= ๐(๐ฝ๐+1๐ขโโ ๐ขโ, ๐) + ๐(๐, ๐ฃ) = (๐ฝ๐+1๐ขโโ ๐ขโ, (1 โ ฮ 0โ)๐)๐ฟ2(ฮฉ)+ ๐(๐ง, ๐ฃ) + ๐(๐ฃ, ๐).
Piecewise Poincarรฉ inequalities and (3.4) lead to
(๐ฝ๐+1๐ขโโ ๐ขโ, (1 โ ฮ 0โ)๐)๐ฟ2(ฮฉ)โฒ โโ0โ2โ|||๐|||2NC.
The definition of ๐ฃ and div ๐ข = 0 = divNC๐ขโprove
๐(๐ง, ๐ฃ) + ๐(๐ฃ, ๐) = ๐NC(๐, ๐ง) + ๐NC((1 โ ๐ฝ๐+1)๐ขโ, ๐ง) + ๐NC(๐ขโโ ๐ฝ๐+1๐ขโ, ๐). (5.8)
The projection property (3.1) ofICRโ and the continuous and discrete problems (5.1) and (5.3) followed by the approximation and stability properties (3.2) ofICRโ show for the first term on the right-hand side of (5.8) that
๐NC(๐, ๐ง) = ๐(๐ข, ๐ง) โ ๐NC(๐ขโ, I CR โ ๐ง) = (๐, ๐ง โ I CR โ ๐ง)๐ฟ2(ฮฉ)โฒ โโโ๐โโ(1 โ ฮ 0 โ)๐ท๐ง)โ.
Recall that divNCI CR
โ ๐ง = div ๐ง = 0. The projection property (3.3) and the stability (3.4) of ๐ฝ๐+1show for the
second term on the right-hand side of (5.8) that
๐NC((1 โ ๐ฝ๐+1)๐ขโ, ๐ง) = (๐ทNC(1 โ ๐ฝ๐+1)๐ขโ, (1 โ ฮ 0โ)๐ท๐ง)๐ฟ2(ฮฉ)โค |||๐ข โ ๐ขโ|||NCโ(1 โ ฮ 0โ)๐ท๐งโ.
Since ฮ 0โdiv(๐ขโโ ๐ฝ๐+1๐ขโ) = 0, the third contribution of (5.8) satisfies
๐NC((๐ขโโ ๐ฝ๐+1๐ขโ), ๐) = ๐NC(๐ขโโ ๐ฝ๐+1๐ขโ, (1 โ ฮ 0
โ)๐) โค |||๐ขโโ ๐ฝ๐+1๐ขโ|||NCโ(1 โ ฮ 0 โ)๐โ.
The best-approximation property (3.4) of ๐ฝ๐+1proves that |||๐ขโโ ๐ฝ๐+1๐ขโ|||NCโฒ |||๐|||NC. Altogether,
โ๐โ2โฒ โโ0โ2โ|||๐||| 2
NC+ โโโ๐โโ(1 โ ฮ 0โ)๐ท๐ง)โ + |||๐|||NC(โ(1 โ ฮ 0โ)๐โ + โ(1 โ ฮ 0 โ)๐ท๐งโ).
Standard a priori estimates [9] and the elliptic regularity (5.7) imply โ(1 โ ฮ 0โ)๐ท๐ง)โ + โ(1 โ ฮ
0
โ)๐โ โฒ โโ0โ ๐ โโ๐โ.
The combination of the above estimates proves
โ๐โ โฒ โโ0โ๐ โ(|||๐|||NC+ โโโ๐โ).
An efficiency estimate similar to that of [33, Proposition 3.1] proves โโโ๐โ โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ + โ(1 โ ฮ
0
โ)๐โ + osc1,1(๐, Tโ).
This concludes the proof.
Remark 5.4. The right-hand side in Proposition 5.3 is also an upper bound for ๐โ๐โin the ๐ปโ1norm. Although
the proof is not difficult, it is not given here because the ๐ปโ1error control is not required in the analysis of this paper.
5.2 Discretisation of the Stokes Eigenvalue Problem
The Stokes eigenvalue problem seeks (๐, ๐ข, ๐) โ โ ร ๐ ร ๐ with โ๐ขโ = 1 such that ๐(๐ข, ๐ฃ) + ๐(๐ฃ, ๐) = ๐ ๐(๐ข, ๐ฃ) for all ๐ฃ โ ๐,
๐(๐ข, ๐) = 0 for all ๐ โ ๐. (5.9)
Although (๐, ๐ข, ๐) is rather a triple than a pair, it is referred to as eigenpair and identified with the pair (๐, (๐ข, ๐)). As in the foregoing section, an equivalent formulation reads as
๐(๐ข, ๐ฃ) = ๐ ๐(๐ข, ๐ฃ) for all ๐ฃ โ ๐. The nonconforming FEM seeks (๐ขโ, ๐โ) โ ๐โร ๐โwith โ๐ขโโ = 1 such that
๐NC(๐ขโ, ๐ฃโ) + ๐NC(๐ฃโ, ๐โ) = ๐โ๐(๐ขโ, ๐ฃโ) for all ๐ฃโโ ๐โ,
๐NC(๐ขโ, ๐โ) = 0 for all ๐โโ ๐โ.
(5.10)
An equivalent formulation reads as
๐NC(๐ขโ, ๐ฃโ) = ๐โ๐(๐ขโ, ๐ฃโ) for all ๐ฃโโ ๐โ. (5.11)
The elliptic formulation on the spaces ๐ and ๐โshows that this problem fits in the framework of Section 2
(where ๐ from Section 2 is replaced by ๐) with exact and discrete eigenvalues 0 < ๐1โค ๐2โค โ โ โ and 0 < ๐โ,1โค โ โ โ โค ๐โ,dim(๐โ)
and their corresponding ๐-orthonormal systems of eigenfunctions
(๐ข1, ๐ข2, ๐ข3, . . .) โ ๐โ and (๐ขโ,1, ๐ขโ,2, . . . , ๐ขโ,dim(๐โ)) โ ๐
dim(๐โ)
โ .
The corresponding pressures are denoted by ๐1, ๐2, . . . and ๐โ,1, . . . , ๐โ,dim(๐โ), respectively. Recall the
defini-tions of Section 2: The set ๐ฝ = {๐ + 1, . . . , ๐ + ๐} describes the eigenvalue cluster of interest and ๐ := span{๐ข๐ |
๐ โ ๐ฝ} โ ๐ and ๐โ := span{๐ขโ,๐ | ๐ โ ๐ฝ} โ ๐โare the exact and discrete invariant subspaces (not
nec-essarily eigenspaces) related to the cluster. In the present situation, the quasi-Ritz projection ๐ โmaps the
solution ๐ข โ ๐ of the linear problem (5.2) to the solution ๐ โ๐ข โ ๐โof the discrete linear problem (5.5) with
discrete pressure ๐(๐ โ๐ข) โ ๐โfrom (5.3). The ๐ฟ2projection onto ๐โis denoted by ๐Tโ := ๐โ. Furthermore
ฮTโ := ฮโ := ๐โโ ๐ โ. In view of Lemma 2.2, the discrete pressure ๐(ฮโ๐ข) โ ๐โcorresponding to ฮโ๐ข is
defined via
๐NC(ฮโ๐ข, ๐ฃโ) + ๐NC(๐ฃโ, ๐(ฮโ๐ข)) = ๐๐(๐โ๐ข, ๐ฃโ) for all ๐ฃโโ ๐โ. (5.12)
It is not difficult to see that ๐(ฮโ๐ข) is well-defined: Lemma 2.2 shows that ฮโ๐ข solves the discrete source
problem (5.5) with right-hand side ๐ = ๐โ๐ข. Hence, ๐(ฮโ๐ข) is the discrete pressure (or Lagrange multiplier) of
(5.3).
The following result gives an ๐ฟ2error estimate for the eigenfunctions. Proposition 5.5(๐ฟ2error estimate). Provided โโ
0โโ โช 1, there exists a constant ๐ถ๐ฟ2such that any eigenpair
(๐, ๐ข, ๐) โ โ ร ๐ ร ๐ of (5.9) with โ๐ขโ = 1 satisfies โ๐ข โ ๐โ๐ขโ โค โ๐ข โ ฮโ๐ขโ โค ๐ถ๐ฟ2(1 + ๐๐ฝ)โโ0โ๐ โ(โ(1 โ ฮ 0 โ)๐ท๐ขโ + โ(1 โ ฮ 0 โ)๐โ).
Proof. Proposition 2.1 and the ๐ฟ2error estimate from Proposition 5.3 result in the following inequality for the
solution (๐ โ๐ข, ๐(๐ โ๐ข)) of (5.3) to the right-hand side ๐ := ๐๐ข,
โ๐ข โ ๐โ๐ขโ โค โ๐ข โ ฮโ๐ขโ โฒ (1 + ๐๐ฝ)โโโโ๐ โ(|||๐ข โ ๐ โ๐ข|||NC+ โ๐ โ ๐(๐ โ๐ข)โ + osc1,1(๐๐ข, Tโ)).
The best-approximation result for the linear Stokes problem (Proposition 5.1) therefore yields โ๐ข โ ฮโ๐ขโ โฒ (1 + ๐๐ฝ)โโโโ ๐ โ(โ(1 โ ฮ 0 โ)๐ท๐ขโ + โ(1 โ ฮ 0 โ)๐โ + osc1(๐๐ข, Tโ)).
If the initial mesh-size is sufficiently small, the discrete Friedrichs inequality [9, Theorem 10.6.12] allows to absorb the oscillation terms on the right-hand side.
The ๐ฟ2error control and the best-approximation of the quasi-Ritz projection from Proposition 5.1 result in the
following best-approximation property for the eigenfunction approximation.
Proposition 5.6(best-approximation property). Provided the initial mesh-size is sufficiently fine โโ0โโ โช 1,
any eigenpair (๐, ๐ข, ๐) โ โ ร ๐ ร ๐ of (5.10) with โ๐ขโ = 1 satisfies
|||๐ข โ ฮโ๐ข|||NC+ โ๐ โ ๐(ฮโ๐ข)โ โฒ โ(1 โ ฮ 0โ)๐ท๐ขโ + โ(1 โ ฮ 0 โ)๐โ.
Proof. The ๐ฟ2control of Proposition 5.5 and the best-approximation result for the linear case of Proposition 5.1 enable the arguments from the proof of Proposition 4.4. The details are omitted for brevity.
5.3 Theoretical Error Estimator and Discrete Reliability
The analysis relies on a theoretical, non-computable error estimator that does not depend on the choice of the discrete eigenfunctions. Given an eigenpair (๐, ๐ข), the theoretical error estimator includes the elementwise residuals in terms of ๐โ๐ข and ฮโ๐ข. More precisely, define, for any ๐ โ Tโ,
๐โ2(๐, ๐, ๐ข) := โ 2
๐โ๐๐โ๐ขโ2๐ฟ2(๐)+ โ
๐นโF(๐)