Part XII, Chapter A Banach and Hilbert Spaces
The goal of this appendix is to recall fundamental results on Banach and Hilbert spaces. The results collected herein provide a theoretical framework for the mathematical analysis of the finite element method. Some classical results are stated without proof; see Aubin [24], Brezis [97], Lax [321], Rudin [399], Yosida [483], Zeidler [486] for further insight. One important outcome of this appendix is the characterization of bijective operators in Banach spaces. To get started, let us recall the following definition of injective, surjective, and bijective maps.
Definition A.1 (Injection, surjection, bijection). Let E and F be two sets. A function (or map) f : E F is said to be injective if every element of the codomain (i.e., F ) is mapped−→ to by at most one element of the domain (i.e., E). The function is said to be surjective if every element of the codomain is mapped to by at least one element of the domain. Finally, f is bijective if every element of the codomain is mapped to by exactly one element of the domain (i.e., f is both injective and surjective).
A.1 Normed vector spaces
Definition A.2 (Norm). Let V be a vector space over the field K = R or C. A norm on V is a map
: V v v [0, ), (A.1) k·kV ∋ 7−→ k kV ∈ ∞ satisfying the following three properties: (i) Definiteness: v = 0 v = 0. k kV ⇐⇒ (ii) 1-homogeneity: λv V = λ v V , for all λ K and all v V . (iii) Triangle inequality:k k v + |w| k k v + w∈ , for all v,w∈ V . k kV ≤ k kV k kV ∈ 774 Appendix A. Banach and Hilbert Spaces
A seminorm on V is a map from V to [0, ) which satisfies only properties (ii) and (iii). ∞
Definition A.3 (Equivalent norms). Two norms V,1 and V,2 are said to be equivalent on V if there exists a positive numberk·kc such thatk·k
1 c v v c− v , v V. (A.2) k kV,2 ≤ k kV,1 ≤ k kV,2 ∀ ∈ Remark A.4 (Finite dimension). If the vector space V has finite di- mension, all the norms in V are equivalent. This result is false in infinite- dimensional vector spaces. ⊓⊔ Proposition A.5 (Compactness of unit ball). Let V be a normed vector space and let B(0, 1) be the closed unit ball in V . Then, B(0, 1) is compact (for the norm topology) if and only if V is finite-dimensional.
Proof. See Brezis [97, Thm. 6.5], Lax [321, 5.2]. § ⊓⊔ Definition A.6 (Bounded linear maps). Let V and W be two normed vector spaces. (V ; W ) is the vector space of bounded linear maps from V to W . The actionL of A (V ; W ) on an element v V is denoted A(v) or, more simply, Av. Maps∈ inL (V ; W ) are often called∈operators. L Example A.7 (Continuous embedding). Let V and W be two normed vector spaces. Assume that V W and that there is c such that v ⊂ k kW ≤ c v V for all v V . This property means that the embedding of V into W isk continuous.k We∈ say that V is continuously embedded into W and we write . V ֒ W →
A.2 Banach spaces
Definition A.8 (Banach space). A vector space V equipped with a norm V such that every Cauchy sequence (with respect to the metric d(x,y) = k·kx y ) in V has a limit in V is called a Banach space. k − kV A.2.1 Operators in Banach spaces Proposition A.9 (Banach space). Let V be a normed vector space and let W be a Banach space. Equip (V ; W ) with the norm L A(v) W A (V ;W ) = sup k k , A (V ; W ). (A.3) k kL v V v V ∀ ∈L ∈ k k Then, (V ; W ) is a Banach space. L Proof. See Rudin [399, p. 87], Yosida [483, p. 111]. ⊓⊔ Part XII. Appendices 775
Remark A.10 (Notation). In this book, we systematically abuse the no- A(v) W A(v) W tation by writing supv V k v k instead of supv V 0 k v k . ∈ k kV ∈ \{ } k kV ⊓⊔ The Uniform Boundedness Principle (or Banach–Steinhaus Theorem) is a useful tool to study the limit of a sequence of operators in Banach spaces.
Theorem A.11 (Uniform Boundedness Principle). Let V and W be two Banach spaces. Let Ai i I be a family (not necessarily countable) of operators in (V ; W ). Assume{ that} ∈ L
sup Aiv W < , v V. (A.4) i I k k ∞ ∀ ∈ ∈ Then, there is a constant C such that
A v C v , v V, i I. (A.5) k i kW ≤ k kV ∀ ∈ ∀ ∈ Proof. See Brezis [97, p. 32], Lax [321, Chap. 10]. ⊓⊔ Corollary A.12 (Point-wise convergence). Let V and W be two Banach spaces. Let (An)n N be a sequence of operators in (V ; W ) such that, for all ∈ L v V , the sequence (Anv)n N converges as n to a limit in W denoted ∈ ∈ → ∞ Av (this means that the sequence (An)n N converges pointwise to A). Then, the following holds: ∈
(i) supn N An (V ;W ) < ; (ii) A ∈ (Vk; Wk)L; ∞ ∈L (iii) A (V ;W ) lim infn An (V ;W ). k kL ≤ →∞ k kL Proof. Statement (i) is just a consequence of Theorem A.11. Owing to (A.5), we infer that A v C v for all v V and all n N. Letting n , k n kW ≤ k kV ∈ ∈ → ∞ we obtain that Av W C v V , and since A is obviously linear, we infer that statement (ii)k holds.k ≤ Finally,k k statement (iii) results from the fact that Anv W An (V ;W ) v V for all v V and all n N. k k ≤ k kL k k ∈ ∈ ⊓⊔ Remark A.13 (Uniform convergence on compact sets). Note that Corollary A.12 does not claim that (An)n N converges to A in (V ; W ), i.e., uniformly on bounded sets. However, a∈ standard argument showsL that (An)n N converges uniformly to A on compact sets. Indeed, let K V be a ∈ ⊂ compact set. Let ǫ > 0. Set C := supn N An (V ;W ); this quantity is finite owing to statement (i) in Corollary A.12.∈ k K kbeingL compact, we infer that there is a finite set of points xi i I in K such that, for all v K, there is { } ∈1 ∈ i I such that v xi V (6C)− ǫ. Owing to the pointwise convergence of ∈ k − k ≤ 1 (An)n N to A, there is Ni such that, for all n Ni, Anxi Axi W 3 ǫ. Using∈ the triangle inequality and statement (iii)≥ above,k we infer− thatk ≤
A v Av A (v x ) + A x Ax + A(v x ) ǫ, k n − kW ≤ k n − i kW k n i − ikW k − i kW ≤ for all v K and all n maxi I Ni. ∈ ≥ ∈ ⊓⊔ 776 Appendix A. Banach and Hilbert Spaces
Compact operators are encountered in various important situations, e.g., the Peetre–Tartar Lemma A.53 and the spectral theory developed in A.5.1. § Definition A.14 (Compact operator). Let V and W be two Banach spaces. T (V ; W ) is called a compact operator if from every bounded ∈ L sequence (vn)n N in V , one can extract a subsequence (vn )k N such that the ∈ k ∈ sequence (Tvnk )k N converges in W ; equivalently, T maps the unit ball in V into a relatively compact∈ set in W .
Proposition A.15 (Composition with compact operator). Let W , X, Y , Z be four Banach spaces and A (Z; Y ), K (Y ; X), B (X; W ). Assume that K is compact. Then B∈ LK A is compact.∈ L ∈ L ◦ ◦ Example A.16 (Compact injection). A classical example is the case where V and W are two Banach spaces such that the injection of V into W is compact. Then from every bounded sequence (vn)n N in V , one can extract a subsequence that converges in W . ∈ ⊓⊔ A.2.2 Duality We start with real vector spaces and then discuss the extension to complex vector spaces.
Definition A.17 (Dual space, Bounded linear forms). Let V be a normed vector space over R. The dual space of V is defined to be (V ; R) L and is denoted V ′. An element A V ′ is called a bounded linear form. Its action on an element v V is either∈ denoted A(v) (or Av) or by means of duality brackets in the form∈ A, v for all v V . h iV ′,V ∈
Owing to Proposition A.9, V ′ is a Banach space when equipped with the norm
A(v) A, v V ′,V A V ′ = sup | | = sup |h i |, A V ′. (A.6) k k v V v V v V v V ∀ ∈ ∈ k k ∈ k k Note that the absolute value can be omitted from the numerators since A is linear and R-valued, and v can be considered in the supremum. ± Theorem A.18 (Hahn–Banach). Let V be a normed vector space over R and let W be a subspace of V . Let B W ′ = (W ; R) be a bounded linear B(w)∈ L map with norm B W ′ = supw W w . Then, there exists a bounded linear k k ∈ k kV form A V ′ with the following properties: ∈ (i) A is an extension of B, i.e., A(w)= B(w) for all w W . (ii) A = B . ∈ k kV ′ k kW ′ Proof. See Brezis [97, p. 3], Lax [321, Chap. 3], Rudin [399, p. 56], Yosida [483, p. 102]. The above statement is a simplified version of the actual Hahn– Banach Theorem. ⊓⊔ Part XII. Appendices 777
Corollary A.19 (Dual characterization of norm). Let V be a normed vector space over R. Then, the following holds:
v V = sup A(v)= sup A, v V ′V , (A.7) k k A V , A =1 A V , A =1h i ∈ ′ k kV ′ ∈ ′ k kV ′ for all v V , and the supremum is attained. ∈ Proof. Assume v = 0 (the assertion is obvious for v = 0). We first observe that 6 sup A(v) v . Let W = span(v) and let B W ′ be defined A V ′, A V =1 V as B(∈tv)k =kt ′v for all≤ kt k R. Owing to the Hahn–Banach∈ Theorem, there k kV ∈ exists A V ′ such that A = B = 1 and A(v)= B(v)= v . ∈ k kV ′ k kW ′ k kV ⊓⊔ Corollary A.20 (Characterization of density). Let V be a normed space over R and let W be a subspace of V . Assume that any bounded linear form in V ′ vanishing identically on W vanishes identically on V . Then, W = V . Proof. See Brezis [97, p. 8], Rudin [399, Thm. 5.19]. ⊓⊔ Definition A.21 (Adjoint operator). Let V and W be two normed vector spaces over R and let A (V ; W ). The adjoint operator, or dual operator, ∈L A∗ : W ′ V ′ is defined by →
A∗w′,v = w′, Av , (v,w′) V W ′. (A.8) h iV ′,V h iW ′,W ∀ ∈ × Definition A.22 (Double dual). The double dual of a Banach space V over R is the dual of V ′ and is denoted V ′′.
Proposition A.23 (Isometric embedding into V ′′). Let V be a Banach space over R. Then, V ′′ is a Banach space, and the linear map JV : V V ′′ defined by →
J v,w′ = w′,v , (v,w′) V V ′, (A.9) h V iV ′′,V ′ h iV ′,V ∀ ∈ × is an isometry.
Proof. That V ′′ is a Banach space results from Proposition A.9. That JV is an isometry results from
JV v V ′′ = sup JV v,w′ V ′′,V ′ = sup w′,v V ′,V = v V , k k w V |h i | w V |h i | k k ′∈ ′ ′∈ ′ w′ =1 w′ =1 k kV ′ k kV ′ where the last equality results from Corollary A.19. ⊓⊔
Remark A.24 (Map JV ). Since the map JV is an isometry, it is injective. As a result, V can be identified with the subspace J (V ) V ′′. It may V ⊂ happen that the map JV is not surjective. In this case, the space V is a 1 1 proper subspace of V ′′. For instance, L∞(D)= L (D)′ but L (D) ( L∞(D)′ with strict inclusion; see B.4 or Brezis [97, 4.3]. § § ⊓⊔ 778 Appendix A. Banach and Hilbert Spaces
Definition A.25 (Reflexive Banach spaces). Let V be a Banach space over R. V is said to be reflexive if JV is an isomorphism.
Let now V be a normed vector space over C. The notion of dual space of V can be defined as in Definition A.17 by setting V ′ = (V ; C). However, in the context of weak formulations of PDEs with complex-valueL d functions, it is more convenient to work with maps A : V C that are antilinear; this means that A(v + w)= Av + Aw for all v,w V→(as usual), but A(λv)= λv for all λ C and all v V , where λ denotes∈ the complex conjugate of λ (instead ∈ ∈ of A(λv)= λv, in which case the map is linear). We denote by V ′ the vector space of antilinear maps that are bounded with respect to the norm (A.6) (note that we are now using the modulus in the numerators). Our aim is to extend the result of Corollary A.19 to measure the norm of the elements of V by the action of the elements of V ′. To this purpose, it is useful to consider V also as a vector space over R by restricting the scaling λv to λ R and v V . The corresponding vector space is denoted VR to ∈ ∈ distinguish it from V (thus, V and VR are the same sets, but equipped with different structures). For instance, if V = Cm so that dim(V ) = m, then dim(VR) = 2m; a basis of V is the set ek 1 k m with ek,l = δkl (the Kro- { } ≤ ≤ necker symbol) for all l 1:m , while a basis of VR is the set ek,iek 1 k m with i2 = 1. Another∈ example { } is V = L2(0, 2π; C) for which{ an Hilbertian} ≤ ≤ − basis is the set cos(nx), sin((n + 1)x) n N, while an Hilbertian basis of VR is { } ∈ the set cos(nx),i cos(nx), sin((n + 1)x),i sin((n + 1)x) n N. { } ∈ Let VR′ be the dual space of VR, i.e., spanned by bounded R-linear maps from V to R.
Lemma A.26 (Isometry for V ′). The map I : V ′ A I(A) VR′ such that I(A)(v)= (A(v)) for all v V , is a bijective isometry.∋ 7→ ∈ ℜ ∈ Proof. The operator I(A) maps onto R and is linear since I(A)(tv) = (A(tv)) = (tA(v)) = t (A(v)) = tI(A)(v) for all t R and all v V . ℜMoreover, I(ℜA) is boundedℜ since ∈ ∈
I(A)(v)= (A(v)) A(v) A v , ℜ ≤| | ≤ k kV ′ k kV for all v V , so that I(A) A . Furthermore, the map I is injective VR′ V ′ because ∈(A(v)) = 0k for allkv ≤V kimpliesk (A(iv)) = 0, i.e., (A(v))=0 so ℜ ∈ ℜ ℑ that A(v) = 0. Let us now prove that I is surjective. Let ψ VR′ and consider the map A : V C so that ∈ → A(v)= ψ(v)+ iψ(iv), v V. ∀ ∈ (Recall that ψ is only R-linear.) By construction, I(A) = ψ, and the map A : V C is antilinear; indeed, for all λ C, writing λ = µ + iν with µ,ν R→, we infer that ∈ ∈ Part XII. Appendices 779
A(λw)= ψ(µw + iνw)+ iψ(iµw νw) − = µψ(w)+ νψ(iw)+ iµψ(iw) iνψ(w) − = µ (ψ(w)+ iψ(iw)) iν (ψ(w)+ iψ(iw)) = λA(v), − for all w V , where we have used the R-linearity of ψ. Let us finally show ∈ A(v) that A V ψ V . Let v V be such that A(v) = 0 and set λ = C. k k ′ ≤ k k R′ ∈ 6 A(v) ∈ Then, | | 1 1 1 1 A(v) = λ− A(v)= A(λ− v)= ψ(λ− v)+ iψ(iλ− v), | | 1 but since ψ takes values in R, we infer that ψ(iλ− v) = 0, so that A(v) = 1 | | ψ(λ− v). As a result,
1 A(v) ψ V λ− v V = ψ V v V , | | ≤ k k R′ k k k k R′ k k since λ = 1. This concludes the proof. | | ⊓⊔ Corollary A.27 (Dual characterization of norm). Let V be a normed vector space over C. Then, the following holds:
v V = max (A(v)). (A.10) k k A V , A =1 ℜ ∈ ′ k kV ′ for all v V . ∈ Proof. Combine the result of Corollary A.19 with Lemma A.26. ⊓⊔ Remark A.28 (Use of modulus). Note that it is possible to replace (A.10) by v V = maxA V , A =1 A(v) since it is always possible to multiply A ′ V ′ in thek k supremum∈ by ak unitaryk | complex| number so that A(v) is real and non- negative. ⊓⊔ Remark A.29 (Hahn–Banach). A version of the Hahn–Banach Theo- rem A.18 in complex vector spaces can be derived similarly to the above construction; see Lax [321, p. 27]. ⊓⊔ The rest of the material is adapted straightforwardly. The adjoint of an operator A (V ; W ) is still defined by (A.8), and one can verify that it ∈ L maps (linearly) bounded antilinear maps in W ′ to bounded antilinear maps in V ′. Moreover, the bidual is defined by considering bounded antilinear forms on V ′, and the linear isometry extending that from Proposition A.23 is such that J v,w′ = w ,v . h V iV ′′,V ′ h ′ iV ′,V A.2.3 Interpolation between Banach spaces Interpolating between Banach spaces is a useful tool to bridge between known results so as to derive new results that could difficult to obtain directly. An important application is the derivation of interpolation error estimates in fractional-order Sobolev spaces. There are many interpolation methods; see, 780 Appendix A. Banach and Hilbert Spaces e.g., Bergh and L¨ofstr¨om [47], Tartar [443] and references therein. For sim- plicity, we focus here on the real interpolation K-method; see [47, 3.1] and [443, Chap. 22]. § Let V0 and V1 be two normed vector spaces, continuously embedded into a common topological vector space . Then, V0 + V1 is a normed vector space with the (canonical) norm v V= inf ( v + v ). Moreover, k kV0+V1 v=v0+v1 k 0kV0 k 1kV1 if V0 and V1 are Banach spaces, then V0 + V1 is also a Banach space; see [47, Lem. 2.3.1]. For all v V + V and all t> 0, define ∈ 0 1
K(t,v)= inf ( v0 V0 + t v1 V1 ). (A.11) v=v0+v1 k k k k
For all t> 0, v K(t,v) defines a norm on V0 +V1 equivalent to the canonical norm. One can7→ also verify that t K(t,v) is nondecreasing and concave (and therefore continuous) and that t7→ 1 K(t,v) is increasing. 7→ t Definition A.30 (Interpolated space). Let θ (0, 1) and let p [1, ]. ∈ ∈ ∞ The interpolated space [V0,V1]θ,p is defined to be
θ [V0,V1]θ,p = v V0 + V1 t− K(t,v) Lp(R ; dt ) < , (A.12) { ∈ | k k + t ∞}
1 p dt p where ϕ Lp(R ; dt ) = ∞ ϕ(t) for p [1, ) and ϕ L (R ; dt ) = k k + t 0 | | t ∈ ∞ k k ∞ + t sup0