Functional Analysis

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Appendix A

Functional Analysis

Remark A.1. Motivation. The study of the existence and uniqueness of so-
lutions of the Navier–Stokes equations as well as the finite element error
analysis requires tools from functional analysis, in particular the use of func-
tion spaces, certain inequalities, and imbedding theorems. There will be no
difference in the notation for functions spaces for scalar, vector-valued, and
tensor-valued functions.
Let Ω ⊂ Rd , d ∈ {2, 3}, be a domain, i.e., Ω is an open set. ✷

A.1 Metric Spaces, Banach Spaces, and Hilbert Spaces

Definition A.2. Metric space. Let X 6= ∅ be a set. A map d : X ×X → R


is called a metric on X if for all x, y, z ∈ X it is
i) d(x, y) = 0 ⇐⇒ x = y,
ii) symmetry: d(x, y) = d(y, x),
iii) triangle inequality: d(x, y) ≤ d(x, z) + d(z, y).
Then (X, d) is called a metric space. ✷

Definition A.3. Isometric metric space. Two metric spaces (X1 , d1 ) and
X2 , d2 ) are called isometric, if there is a surjective map g : X1 → X2 such
that for all x, y ∈ X1 it is d1 (x, y) = d2 (g(x), g(y)). ✷

Definition A.4. Cauchy sequence, convergent sequence. Let {xn }∞ n=1


be a sequence in a metric space (X, d). It is called a Cauchy sequence if for
each ε > 0 there is a N ∈ N such that

d(xk , xl ) < ε ∀ k, l ≥ N.

The sequence {xn }∞


n=1 converges to x ∈ X, denoted by xn → x, if

lim d(xn , x) = 0.
n→∞

167
168 A Functional Analysis

Definition A.5. Complete metric space. A metric space (X, d) is called


complete, if each Cauchy sequence converges in X. That means, for each
Cauchy sequence {xn }∞
n=1 there exists an element x ∈ X such that xn → x.

Definition A.6. Norm, triangle inequality, seminorm, normed space.


Let X be a linear space over R (or C). A mapping k·kX : X → R is called
a norm on X if
i) definiteness: kxkX = 0 if and only if x = 0,
ii) homogeneity: kαxkX = |α| kxkX for all x ∈ X, α ∈ R,
iii) the triangle inequality holds: kx + ykX ≤ kxkX + kykX for all x, y ∈ X.
A mapping from X to R that satisfies only ii) and iii) is called a seminorm
on X.
The space (X, k·kX ) is called normed space. ✷

Definition A.7. Equivalent norms. Two norms k·kX,1 , k·kX,2 of a normed


space X are called equivalent, if there are two positive constants C1 and C2
such that
C1 kxkX,1 ≤ kxkX,2 ≤ C2 kxkX,1 ∀ x ∈ X.

Remark A.8. On norms.


• All norms in finite-dimensional spaces are equivalent.
• A normed space (X, k·kX ) becomes a metric space with the induced metric

d(x1 , x2 ) = kx1 − x2 kX , x1 , x2 ∈ X.

Definition A.9. Banach space. A normed space is called complete if it is


a complete metric space with the induced metric. A complete normed space
is called Banach space. ✷

Remark A.10. Compact set, precompact set. A subset Y of a normed space


X is called compact if every sequence of elements in Y has a subsequence
that converges in the norm of X to an element of Y . The set Y is called
precompact if its closure Y is compact.
Compact sets are closed and bounded. The reverse statement is only true
for finite-dimensional spaces. ✷

Definition A.11. Inner product, scalar product. Let X be a linear space


over R. A map (·, ·)X : X × X → R is called symmetric sesquilinear form if
for all x, y, z ∈ X and all α ∈ R it holds that
i) symmetry: (x, y)X = (y, x)X ,
ii) (αx, y)X = α(x, y),
A.1 Metric Spaces, Banach Spaces, and Hilbert Spaces 169

iii) (x, y + z) = (x, y) + (x, z).


The symmetric sesquilinear form (·, ·)X is called positive semi-definite if for
all x ∈ X it is (x, x)X ≥ 0. A positive semi-definite symmetric sesquilinear
form with
(x, x)X = 0 ⇐⇒ x = 0
is called inner product or scalar product on X. ✷

Definition A.12. Induced norm, inner product space, Hilbert space.


Let (·, ·)X be an inner product on X, then (X, (·, ·)X ) is called pre Hilbert
space. The inner product induces the norm
1/2
kxkX = (x, x)X

in X. A complete inner product space is called Hilbert space.


For simplicity of notation, the subscript at the inner product symbol will
be neglected if the inner product is clear from the context. ✷

Lemma A.13. Cauchy–Schwarz inequality. Let (X, (·, ·)) be an inner


product space, then it holds the so-called Cauchy–Schwarz inequality

|(x, y)| ≤ kxkX kykX ∀ x, y ∈ X. (A.1)

Example A.14. Cauchy–Schwarz inequality for sums. Consider X = Rn with


the standard inner product for vectors, then one obtains with the triangle
inequality and the Cauchy–Schwarz inequality (A.1)

n

n n
!1/2 n
!1/2
X X X X
x i yi ≤ |xi | |yi | ≤ x2i yi2 , (A.2)



i=1 i=1 i=1 i=1

for all x = (x1 , . . . , xn )T , y = (y1 , . . . , yn )T ∈ Rn . ✷

Example A.15. Hölder inequality for sums. The Cauchy–Schwarz inequality


(A.2) is a special case of the Hölder inequality

n n
!1/p n
!1/q
X X p
X q 1 1
|ai bi | ≤ |ai | |bi | + = 1.
, 1 < p, q < ∞,
i=1 i=1 i=1
p q
(A.3)
The following inequalities for sums of non-negative real numbers hold:
n n
!p n
X X 1/p
X 1 1
ai ≤ ai ≤ np/q ai , ai ≥ 0, p ∈ (1, ∞), + = 1. (A.4)
i=1 i=1 i=1
p q

The right inequality of (A.4) is just a consequence of (A.3). ✷


170 A Functional Analysis

Definition A.16. Orthogonal elements, orthogonal complement of a


subspace. Let X be a normed space endowed with an inner product (·, ·).
Two elements x, y ∈ X are said to be orthogonal if (x, y) = 0.
Let Y ⊂ X be a subspace of X, then Y ⊥ = {x ∈ X : (x, y) = 0 for all y ∈
Y } is the orthogonal complement of Y . ✷
Lemma A.17. Orthogonal complement is closed subspace. Let W ⊂
V be a subspace of a Hilbert space V . Then, W ⊥ is a closed subspace of V .
Lemma A.18. Young’s inequality. Let a, b ∈ R, then the following in-
equality is called Young’s inequality:

t p t−q/p q 1 1
ab ≤ a + b , + = 1, 1 < p, q < ∞, t > 0. (A.5)
p q p q
Proof. The proof is based on the strict convexity of the exponential, which follows from
the strict positivity of the second derivative. This property reads for α, β ∈ R and p, q as
in (A.5)  
α β 1 1
exp + ≤ exp(α) + exp(β).
p q p q
Choosing α = ln (tap ) and β = ln t−q/p bq gives (A.5).


m×n
Lemma A.19. Estimate for a Rayleigh quotient. Let A ∈ R be a
matrix, then it is

xT AT Ax
= λmin AT A ,

inf
x∈R n ,x6=0 xT x

where λmin AT A is the smallest eigenvalue of AT A. The infimum is taken,




i.e., it is even a minimum. The quotient on the left-hand side is called


Rayleigh quotient.
Proof. The matrix AT A is symmetric and positive semi-definite. Hence, all eigenvalues
are non-negative, the (normalized eigenvectors) {φi }n n
i=1 form a basis of R , and they are
mutually orthonormal. Let the eigenvalues be ordered such that

0 ≤ λmin AT A = λ1 ≤ λ2 ≤ . . . ≤ λn .


Pn
Each vector x ∈ Rn can be written in the form x = i=1 xi φi . Using that the eigenvectors
are orthonormal, it follows that xT x = n 2
P
i=1 xi and

n
X X n
n X n
X n
X
xT AT Ax = xT xi AT Aφi = λ i x j xi φ j φ i = λi x2i ≥ λmin AT A x2i .
i=1 j=1 i=1 i=1 i=1

Hence, one gets


xT AT Ax
≥ λmin AT A .

inf
x∈Rn ,x6=0 xT x
Choosing x = x1 φ1 , x1 6= 0, leads to

xT AT Ax λ1 x21
= λ1 = λmin AT A ,

=
T
x x x21
A.2 Function Spaces 171

such that the equal sign holds. 

A.2 Function Spaces

Definition A.20. Derivatives and multi-index. A multi-index α is a vec-


tor α = (α1 , . . . , αn ) with αi ∈ N ∪ {0}, i = 1, . . . , n. Derivatives are denoted
by
n
∂ |α| X
Dα = , with |α| = αi .
∂xα αn
1 . . . ∂xn
1
i=1

Low order derivatives are also denoted by subscripts, e.g.,


∂u
∂x u = .
∂x

Definition A.21. Spaces of continuously differentiable functions
C m (Ω), C m (Ω), and CB m
(Ω). Let m ∈ N ∪ {0}, then the space of m-times
continuously differentiable functions in Ω is denoted by

C m (Ω) = f : f and all its derivatives up to order m




are continuous in Ω .

It is

\
C ∞ (Ω) = C m (Ω).
m=0
m
The space C (Ω) for m < ∞ is defined by

C m (Ω) = f : f ∈ C m (Ω) and all derivatives can be extended




continuously to Ω .

One defines

\
C ∞ (Ω) = C m (Ω).
m=0

Finally, the following space is introduced


m
(Ω) = f : f ∈ C m (Ω) and f is bounded .

CB (A.6)


Remark A.22. Spaces of continuously differentiable functions C m (Ω), C m (Ω),
m
and CB (Ω).
• If Ω is bounded, then C m (Ω), equipped with the norm
172 A Functional Analysis
X
kf kC m (Ω) = max |Dα f (x)| ,
x∈Ω
0≤|α|≤m

is a Banach space.
m
• The space CB (Ω) becomes a Banach space with the norm

kf kC m (Ω) = max sup |Dα f (x)| .


B 0≤|α|≤m x∈Ω

• It is
m
C m (Ω) ⊂ CB (Ω) ⊂ C m (Ω).
Consider, e.g., Ω = (0, 1) and f (x) = sin(1/x), then f ∈ CB (Ω) but
f 6∈ C(Ω).

Definition A.23. Support. Let f ∈ C(Ω), then

supp(f ) = {x : f (x) 6= 0}

is the support of f (x). The closure is taken with respect to Rd . A function


f ∈ C(Ω) is said to have a compact support, if the support of f (x) is bounded
in Rd and if supp(f ) ⊂ Ω. ✷

Definition A.24. The space C0m (Ω). The space C0m (Ω) is given by

C0m (Ω) = {f : f ∈ C m (Ω) and supp(f ) is compact in Ω}.

In the literature, the space C0∞ (Ω) is often denoted by D(Ω).


An important space for the study of the Navier–Stokes equations is

C0,div (Ω) = {f : f ∈ C0∞ (Ω), ∇ · f = 0}. (A.7)

Definition A.25. The spaces C m,α (Ω), spaces of Hölder continuous


functions. Let M ∈ Rd , d ∈ {2, 3}, be a set and let α ∈ (0, 1]. Then, the
constant  
|f (x) − f (y)|
|f |C 0,α (M ) = sup α
x6=y∈M |x − y|
is called Hölder coefficient or Hölder constant. For α = 1, it is usually called
Lipschitz constant.
Let Ω be bounded. For m ∈ N ∪ {0}, the following spaces are defined

C m,α (Ω) = f ∈ C m (Ω) : Dβ f C 0,α (Ω) < ∞, |β| = m .




For m = 0, these spaces are called spaces of Hölder continuous functions and
for α = 1, space of Lipschitz continuous functions. ✷
A.2 Function Spaces 173

Remark A.26. The spaces C m,α (Ω). The spaces C m,α (Ω) are Banach spaces
if they are equipped with the norm
X
kf kC m,α (Ω) = kf kC m (Ω) + [Dβ f ]C 0,α (Ω) .
|β|=m

Definition A.27. Spaces of (Lebesgue) integrable functions Lp (Ω).


The Lebesgue spaces are defined by
 Z 
p p
L (Ω) = f : |f (x)| dx < ∞ , p ∈ [1, ∞),

where the integral is to be understood in the sense of Lebesgue. The space


L∞ (Ω) is the space of all functions that are bounded for almost all x ∈ Ω

L∞ (Ω) = {f : |f (x)| < ∞ for almost all x ∈ Ω} .


Remark A.28. Lebesgue spaces.
• The space Lp (Ω) is a normed vector space with norm
Z 1/p
kf kLp (Ω) = |f (x)|p dx , p ∈ [1, ∞).

• An important special case is L2 (Ω) since this space is a Hilbert space.


The inner product (f, g)L2 (Ω) of L2 (Ω) and the induced norm are given
by Z
1/2
(f, g)L2 (Ω) = f (x)g(x) dx, kf kL2 (Ω) = (f, f )L2 (Ω) .

• The space L∞ (Ω) becomes a Banach space if it is equipped with the norm

kf kL∞ (Ω) = ess sup |f (x)|,


x∈Ω

where ess supx∈Ω is the essential supremum.


• Let |Ω| < ∞ and 1 ≤ p ≤ q ≤ ∞. If u ∈ Lq (Ω), then u ∈ Lp (Ω) and
Z 1/p−1/q
kukLp (Ω) ≤ dx kukLq (Ω) , (A.8)

see (Adams, 1975, Theorem 2.8).



Example A.29. Cauchy–Schwarz inequality and Hölder’s inequality. Let f ∈
Lp (Ω) and g ∈ Lq (Ω) with p, q ∈ [1, ∞] and 1/p + 1/q = 1. Then it is
174 A Functional Analysis

f g ∈ L1 (Ω) and the Hölder inequality holds

kf gkL1 (Ω) ≤ kf kLp (Ω) kgkLq (Ω) . (A.9)

For p = q = 2, this inequality is called Cauchy–Schwarz inequality

kf gkL1 (Ω) ≤ kf kL2 (Ω) kgkL2 (Ω) . (A.10)

Definition A.30. Sobolev spaces W k,p (Ω). Let k ∈ N and p ∈ [1, ∞]. The
Sobolev space W k,p (Ω) consists of all integrable functions f : Ω → R such
that for each multi-index α with |α| ≤ k, the derivative Dα f exists in the
weak sense and it belongs to Lp (Ω). ✷
Remark A.31. Sobolev spaces.
• It is Lp (Ω) = W 0,p (Ω).
• A norm in Sobolev spaces is defined by
 1/p
p
kDα f kLp (Ω)
 P
|α|≤k if p ∈ [1, ∞),
kf kW k,p (Ω) = P
α
|α|≤k ess supx∈Ω |D f | if p = ∞.

Sobolev spaces equipped with this norm are Banach spaces, e.g., see
(Evans, 2010, p. 262).
• The Sobolev spaces for p = 2 are Hilbert spaces. They are often denoted
by W m,2 (Ω) = H m (Ω) and they are equipped with the inner product
X
(f, g)H k (Ω) = (Dα f, Dα g)L2 (Ω) .
|α|≤k

• In particular, the Sobolev spaces of first order are important for the study
of the Navier–Stokes equations
 Z 
W 1,p (Ω) = f : |f (x)|p + |∇f (x)|p dx < ∞ , p ∈ [1, ∞),

which are equipped with the norm


Z 1/p
kf kW 1,p (Ω) = |f (x)|p + |∇f (x)|p dx , p ∈ [1, ∞).

• The definition of Sobolev spaces can be extended to k ∈ R, e.g., see Adams


(1975).

Definition A.32. Sobolev spaces W0k,p (Ω). The Sobolev spaces W0k,p (Ω)
are defined by the closure of C0∞ (Ω) in the norm of W k,p (Ω). ✷
A.2 Function Spaces 175

Remark A.33. On the smoothness of the boundary. The Sobolev imbedding


theorem requires that Ω has the so-called cone property or the strong local
Lipschitz property. In the case that Ω is bounded, these assumptions reduce
to the requirement that Ω has a locally Lipschitz boundary, (Adams, 1975, p.
67). That means, each point x on the boundary ∂Ω of Ω has a neighborhood
Ux such the ∂Ω ∩ Ux is the graph of a Lipschitz continuous function. ✷

Theorem A.34. Trace theorem, (Lions & Magenes, 1972, Theorem 9.4),
(Galdi, 2011, Theorem II.4.1 for m = 1). Let Ω be a bounded domain with
locally Lipschitz boundary ∂Ω. Then, there is a bounded linear operator T :
W 1,q (Ω) → Lr (∂Ω), q ∈ [1, ∞), such that
i) r ∈ [1, q(d − 1)/(d − q)] if q < d and r ∈ [1, ∞) else,
ii) T f = f |∂Ω if f ∈ W 1,q (Ω) ∩ C Ω ,
iii) kT f kLr (∂Ω) ≤ C kf kW 1,q (Ω) for each f ∈ W 1,q (Ω), with the constant
C depending only on q and Ω.
The mapping
s0
∂j f
Y  
s s−j−1/2
H (Ω) → H (∂Ω), f 7→ , j = 0, 1, . . . , s0 (A.11)
j=0
∂nj

is continuous, where s0 is the greatest integer such that s0 < s − 1/2, and
n is the outward pointing unit normal vector. The mapping is surjective and
there exists a continuous right inverse.

Theorem A.35. Functions with vanishing trace, (Galdi, 2011, Theo-


rem II.4.2), (Evans, 2010, p. 273). Let the assumptions of Theorem A.34 be
given. Then f ∈ W01,p (Ω) if and only if T f = 0 on ∂Ω.

Theorem A.36. Poincaré’s inequality, Poincaré–Friedrichs’ inequal-


ity, (Galdi, 2011, Theorem II.5.1), (Gilbarg & Trudinger, 1983, p. 164). Let
f ∈ W01,p (Ω), then
 1/d
|Ω|
kf kLp (Ω) ≤ k∇f kLp (Ω) = CPF k∇f kLp (Ω) p ∈ [1, ∞), (A.12)
ωd

where ωd is the volume of the unit ball in Rd .

Remark A.37. Poincaré’s inequality. Poincaré’s inequality (A.12) holds also


for functions v ∈ H 1 (Ω) with v = 0 on Γ0 ⊂ Γ with |Γ0 | > 0.
Poincaré’s inequality stays valid for vector-valued functions v if Ω is
bounded with a locally Lipschitz boundary, v ∈ W 1,q (Ω), 1 ≤ q < ∞,
and v · n = 0 on ∂Ω, see (Galdi, 1994, Section II.4). ✷

Theorem A.38. Density of continuous functions in Sobolev spaces,


(Gilbarg & Trudinger, 1983, p. 154). The subspace C ∞ (Ω)∩W k,p (Ω) is dense
in W k,p (Ω).
176 A Functional Analysis

Remark A.39. Density of continuous functions in Sobolev spaces. For C ∞ (Ω)


to be dense in W k,p (Ω), one needs some smoothness assumptions on the
boundary ∂Ω, like ∂Ω is C 1 or the so-called segment property, e.g., see
(Gilbarg & Trudinger, 1983, p. 155). This segment property follows from
the strong local Lipschitz property, see (Adams, 1975, p. 67). ✷

Theorem A.40. Interpolation theorem for Sobolev spaces, (Adams,


1975, Theorem 4.17). Let Ω ⊂ Rd be a bounded domain with a locally Lip-
schitz boundary and let p ∈ [1, ∞). Then there exists a constant C(m, p, Ω)
such that for 0 ≤ j ≤ m and any u ∈ W m,p (Ω)
j/m (m−j)/m
kukW j,p (Ω) ≤ C(m, p, Ω) kukW m,p (Ω) kukLp (Ω) . (A.13)

In addition, (A.13) is valid for all u ∈ W0m,p (Ω) with a constant C(m, p, d)
independent of Ω.

Remark A.41. Imbedding theorems. Imbedding theorems for Sobolev spaces


are used frequently in the analysis of partial differential equations. The imbed-
ding theorems state that all functions belonging to a certain space do belong
also to another space and that the norm of the functions in the larger space
can be estimated by the norm in the smaller space. Let V be a Banach space
such that an imbedding W m,p (Ω) → V holds. Then, there is a constant C
depending on Ω such that

kvkV ≤ C kvkW m,p (Ω)

for all functions v ∈ W m,p (Ω). The validity of imbeddings depends on the
dimension d of the domain Ω. The larger the dimension, the less imbeddings
are valid, compare Example A.44. ✷

Theorem A.42. The Sobolev imbedding theorem, (Adams, 1975, The-


orem 5.4, Remark 5.5. (6), Theorem 6.2). Let Ω ⊂ Rd be a bounded domain
with a locally Lipschitz boundary. Let j and m be non-negative integers and
let p satisfy 1 ≤ p < ∞.
i) Let mp < d, then the imbedding

dp
W j+m,p (Ω) → W j,q (Ω), 1≤q≤ (A.14)
d − mp
holds. In particular, it is
dp
W m,p (Ω) → Lq (Ω), 1≤q≤ . (A.15)
d − mp

ii) Suppose mp = d. Then the imbedding

W m,p (Ω) → Lq (Ω), 1≤q<∞ (A.16)


A.2 Function Spaces 177

is valid. If in addition p = 1, then this imbedding holds also for q = ∞

W d,1 (Ω) → L∞ (Ω). (A.17)

and even
W d,1 (Ω) → CB (Ω),
see (A.6) for the definition of latter space.
iii) Suppose that mp > d, then the imbedding

W m,p (Ω) → CB (Ω) (A.18)

holds.
iv) Suppose mp > d > (m − 1)p, then

d
W j+m,p (Ω) → C j,λ Ω

for 0<λ≤m− . (A.19)
p

v) Suppose d = (m − 1)p, then

W j+m,p (Ω) → C j,λ Ω



for 0 < λ < 1.

This imbedding holds for λ = 1 if p = 1 and d = m − 1.


vi) All imbeddings are true for arbitrary domains provided the W spaces
undergoing the imbedding are replaced with the corresponding W0
spaces.
vii) Rellich–Kondrachov theorem: The imbeddings (A.14) – (A.16) are
compact with the conditions on Ω stated at the beginning of the theo-
rem, i.e., the imbedding operator is compact, see Definition A.63.

Remark A.43. Spaces of continuous functions in Ω. Since the compact imbed-


ding
C j,λ Ω → C j Ω
 
j ≥ 0, 0 < λ ≤ 1,
holds for bounded domains, (Adams, 1975, Theorem 1.31), one can derive
from Theorem A.42, cases iv) and v), also imbeddings for C j Ω : if mp >
d ≥ (m − 1)p, p ∈ [1, ∞), then

W j+m,p (Ω) → C j Ω .

(A.20)

Example A.44. Important Sobolev imbeddings. Let d = 2. Then, it follows


from (A.16) that

H 1 (Ω) = W 1,2 (Ω) → Lq (Ω), q ∈ [1, ∞). (A.21)

For d = 3, one gets with (A.15) that


178 A Functional Analysis

H 1 (Ω) = W 1,2 (Ω) → Lq (Ω), q ∈ [1, 6]. (A.22)

Remark A.45. Spaces of functions defined in space-time domains. Let X be


any normed space introduced above that is equipped with the norm k · kX
and let (t0 , t1 ) be a time interval. Then, the following function space on the
space-time domain can be defined
 Z t1 
Lp (t0 , t1 ; X) = f (t, x) : kf kpX (τ ) dτ < ∞ , p ∈ [1, ∞).
t0

The norm of Lp (t0 , t1 ; X) is


Z t1 1/p
kf kLp (t0 ,t1 ;X) = kf kpX (τ ) dτ , p ∈ [1, ∞).
t0

The modifications for p = ∞ are the same as for the Lebesgue spaces. ✷

A.3 Some Definitions, Statements, and Theorems

Remark A.46. Convolution. The convolution of two scalar functions f and g


is defined by
Z Z
(f ∗ g) (y) = f (y − x) g (x) dx = f (x) g (y − x) dx = (g ∗ f ) (y) ,
R R

provided that the integrals exist for almost all y ∈ R. ✷

Remark A.47. Fourier transform: definition and some properties. The Fourier
transform of a scalar function f is defined by
Z
F (f ) (y) = f (x) e−ixy dx (A.23)
R

and the inverse Fourier transform of F (y) by

1
Z
−1
F (F ) (x) = F (y) eixy dy. (A.24)
2π R

It holds
F (f ∗ g) = F (f ) F (g) , F (f g) = F (f ) ∗ F (g) . (A.25)
If f is differentiable and lim|x|→∞ f (x) = 0, integration by parts yields

yF (f ) (y) = −iF (f ′ ) (y) .


A.3 Some Definitions, Statements, and Theorems 179

This formula implies the relations


2
kyk2 F (f ) = −F (∆f ) , (A.26)
1 −1

2 F (f ) = −F ∆ (f ) , (A.27)
kyk2
1 
−1

2 F (f ) = F (I − c∆) (f ) . (A.28)
1 + c kyk2

The Lr (Ω) norm of f ∗ g, 1 ≤ r ≤ ∞, can be estimated by Young’s in-


equality for convolutions (sometimes also called Hölder’s inequality for con-
 1 ≤ p, q ≤ ∞,
volutions), e.g., see (Hörmander, 1990, Section IV.4.5). Let
p−1 + q −1 ≥ 1, and r−1 = p−1 + q −1 − 1. For f ∈ Lp Rd and g ∈ Lq Rd ,


it is f ∗ g ∈ Lr Rd and Young’s inequality for convolution

kf ∗ gkLr (Ω) ≤ kf kLp (Ω) kgkLq (Ω) (A.29)

holds. If p−1 + q −1 − 1 = 0, then f ∗ g ∈ Lr Rd is continuous and bounded.




Definition A.48. Absolutely continuous function. Let I ⊂ R be an


interval. A function f : I → R is called absolutely continuous if for every ε >
0 there is a δ > 0 such that whenever
P a finite sequenceP
of pairwise disjoint sub-
intervals (xk , yk ) of I satisfies k (yk − xk ) < δ, then k |f (yk ) − f (xk )| < ε.

Remark A.49. Absolutely continuous functions. Absolute continuity of a func-


tion is a stronger condition than continuity and even uniform continuity. On
a compact interval I = [a, b], absolute continuity of a function f is equivalent
to the property that this function has a derivative f ′ almost everywhere, the
derivative is Lebesgue integrable, and it holds
Z t
f (t) = f (a) + f ′ (τ ) dτ ∀ t ∈ [a, b]
a

(fundamental theorem of calculus). ✷

Theorem A.50. Local existence and uniqueness theorem of Cara-


théodory, (Carathéodory, 1918, Kap. 11), (Kamke, 1944, p. 34), (Filippov,
1988, Section 1.1). Let fm (t, y1 , . . . , yn ), m = 1, . . . , n, be defined in

ΩT = (t0 , t0 + T ) × Ωy

with Ωy = {y = (y1 , . . . , yn )T : ky − y 0 k2 ≤ b} for some b > 0, let the


functions f1 , . . . , fn for each fixed system y1 , . . . , yn be measurable with respect
to t, let for each fixed t the functions f1 , . . . , fn be continuous with respect to
y1 , . . . , yn , and let
180 A Functional Analysis

|fm (t, y1 , . . . , yn )| ≤ F (t), m = 1, . . . , n,

where F (t) is a Lebesgue integrable function in (t0 , t0 + T ). Then there exists


a system of absolutely continuous functions y1 (t), . . . , yn (t) that satisfies for
all t in some interval [t0 , t0 + a], 0 < a ≤ T ,
Z t
ym (t) = y0m + fm (s, y1 (s), . . . , yn (s)) ds, m = 1, . . . , n. (A.30)
t0

At each point where the term in the integral is continuous, the functions
satisfy the ordinary differential equation
d
ym (t) = fm (t, y1 , . . . , yn ), m = 1, . . . , n. (A.31)
dt
If in addition for any two points (t, y 1 , . . . , y n ), (t, ŷ1 , . . . , ŷn ) ∈ ΩT the
Lipschitz condition
n
X
|fm (t, y 1 , . . . , y n ) − fm (t, ŷ1 (t), . . . , ŷn (t))| ≤ G(t) |y l − ŷl (t)| ,
l=1

m = 1, . . . , n, with a Lebesgue integrable function G(t) is satisfied, then there


exists exactly one solution of (A.30) in [t0 , t0 + a].
Remark A.51. On Carathéodory’s theorem. The theorem of Carathéodory is
an extension of the theorem of Peano to ordinary differential equations of
type (A.31) with discontinuous right-hand side. ✷
Remark A.52. On Gronwall’s lemma. Gronwall’s lemma is an important tool
for the analysis and finite element analysis of time-dependent problems. Two
versions of this lemma in the continuous setting will be given below, see
Emmrich (1999) for complete proofs and a discussion of the differences of
these versions. ✷
Lemma A.53. Gronwall’s lemma in integral form. Let T ∈ R+ ∪ ∞,
f, g, ∈ L∞ (0, T ), and λ ∈ L1 (0, T ), λ (t) ≥ 0 for almost all t ∈ [0, T ]. Then
Z t
f (t) ≤ g (t) + λ (s) f (s) ds a.e. in [0, T ] (A.32)
0

implies for almost all t ∈ [0, T ] that


Z t Z t 
f (t) ≤ g (t) + exp λ (τ ) dτ λ (s) g (s) ds. (A.33)
0 s

If g ∈ W 1,1 (0, T ), it follows that


Z t  Z t  Z s 
f (t) ≤ exp λ (τ ) dτ g (0) + exp − λ (τ ) dτ g ′ (s) ds.
0 0 0
A.3 Some Definitions, Statements, and Theorems 181

Moreover, if g(t) is a monotonically increasing continuous function, it holds


Z t 
f (t) ≤ exp λ (τ ) dτ g (t) . (A.34)
0

Proof. For illustration, the derivation of (A.33) and (A.34) will be presented.
(A.33). Let  Z t Z t
f˜(t) = exp − λ (τ ) dτ λ(s)f (s) ds,
0 0
then one obtains for almost all t ∈ [0, T ] with the product rule, the Leibniz integral rule,
(A.32), and λ(t) ≥ 0
 Z t  Z t 
f˜′ (t) = exp − λ (τ ) dτ −λ(t) λ(s)f (s) ds + λ(t)f (t)
0 0
 Z t 
≤ exp − λ (τ ) dτ (λ(t) (g(t) − f (t)) + λ(t)f (t))
0
 Z t 
= exp − λ (τ ) dτ λ(t)g(t).
0

Integration yields, using f˜(0) = 0,


Z t  Z s 
f˜(t) ≤ exp − λ (τ ) dτ λ(s)g(s) ds.
0 0

With (A.32), one obtains


 Z t   Z t Z t
exp − λ (τ ) dτ (f (t) − g(t)) ≤ exp − λ (τ ) dτ λ (s) f (s) ds
0 0 0
Z t  Z s 
= f˜(t) ≤ exp − λ (τ ) dτ λ(s)g(s) ds.
0 0
R 
t
Multiplying this inequality with exp 0
λ (τ ) dτ and bringing g(t) to the right-hand side
gives (A.33).
(A.34). If g(t) is a monotonically increasing continuous function, one gets from (A.33),
using that g(t) takes its largest value at the final time and that λ(t) ≥ 0, and applying the
fundamental theorem of calculus
 Z t Z t  
f (t) ≤ g(t) 1 + exp λ (τ ) dτ λ (s) ds
0 s
 Z t Z t  Z s  
= g(t) 1 + exp λ (τ ) dτ exp − λ (τ ) dτ λ (s) ds
0 0 0
 Z t Z t   Z s  
d
= g(t) 1 + exp λ (τ ) dτ − exp − λ (τ ) dτ ds
0 0 ds 0
 Z t   Z t  
= g(t) 1 + exp λ (τ ) dτ − exp − λ (τ ) dτ +1
0 0
Z t 
= exp λ (τ ) dτ g(t).
0


182 A Functional Analysis

Lemma A.54. Gronwall’s lemma in differential form. Let T ∈ R+ ∪∞,


f ∈ W 1,1 (0, T ), and g, λ ∈ L1 (0, T ). Then

f ′ (t) ≤ g (t) + λ (t) f (t) a.e. in [0, T ] (A.35)

implies for almost all t ∈ [0, T ]


Z t  Z t Z t 
f (t) ≤ exp λ (τ ) dτ f (0) + exp λ (τ ) dτ g (s) ds. (A.36)
0 0 s

Proof. Defining
 Z t 
f˜(t) = exp − λ (τ ) dτ f (t) = exp(−Λ(t))f (t), (A.37)
0

applying the chain rule, the Leibniz integral rule, and (A.35) gives

f˜′ (t) = −Λ′ (t) exp(−Λ(t))f (t) + exp(−Λ(t))f ′ (t) = exp(−Λ(t)) f ′ (t) − λ(t)f (t)


≤ exp(−Λ(t))g(t).

Integration in (0, t) and using (A.37) yields


 Z t  Z t
f˜(t) − f˜(0) = exp − λ (τ ) dτ f (t) − f (0) ≤ exp(−Λ(s))g(s) ds.
0 0
R 
t
Multiplication with exp 0
λ (τ ) dτ gives (A.36). 

Lemma A.55. Variation of Gronwall’s lemma in differential form.


Let T ∈ R+ ∪ ∞, f ∈ W 1,1 (0, T ), h, g, λ ∈ L1 (0, T ), and h (t) , λ (t) ≥ 0 a.e.
in (0, T ). Then,

f ′ (t) + h (t) ≤ g (t) + λ (t) f (t) a.e. in [0, T ] (A.38)

implies for almost all t ∈ [0, T ]


Z t
f (t) + h (s) ds (A.39)
0
Z t  Z t Z t 
≤ exp λ (τ ) dτ f (0) + exp λ (τ ) dτ g (s) ds.
0 0 s

Moreover, if g (t) ≥ 0 a.e. in (0, T ), it holds


Z t Z t  Z t 
f (t) + h (s) ds ≤ exp λ (τ ) dτ f (0) + g (s) ds . (A.40)
0 0 0

Proof. From (A.38), it follows a.e. in [0, T ] that


f ′ (s) − λ (s) f (s) + h (s) ≤ g (s) .

The positivity of the exponential implies


A.3 Some Definitions, Statements, and Theorems 183
 Z s   Z s 
f ′ (s) − λ (s) f (s) + h (s) ≤ exp −

exp − λ (τ ) dτ λ (τ ) dτ g (s) .
0 0

Integration on (0, t) ⊂ [0, T ] gives


 Z t  Z t  Z s 
exp − λ (τ ) dτ f (t) − f (0) + exp − λ (τ ) dτ h (s) ds
0 0 0
Z t  Z s 
≤ exp − λ (τ ) dτ g (s) ds. (A.41)
0 0

Using the monotonicity of the exponential yields


 Z t Z t Z t  Z s 
exp − λ (τ ) dτ h (s) ds ≤ exp − λ (τ ) dτ h (s) ds.
0 0 0 0

Applying this inequality to bound the left-hand


R side of
 (A.41) from below and multiplica-
t
tion of the resulting inequality with exp 0 λ (τ ) dτ proves (A.39).
If g is non-negative, one obtains
Z t Z t  Z t Z t
exp λ (τ ) dτ g (s) ds ≤ exp λ (τ ) dτ g (s) ds,
0 s 0 0

from which (A.40) follows. 

Lemma A.56. Discrete Gronwall’s lemma, (Heywood & Rannacher,


1990, Lemma 5.1). Let k, B, an , bn , cn , αn be non-negative numbers for in-
tegers n ≥ 1 and let the inequality
N
X +1 N
X +1 N
X +1
aN +1 + k bn ≤ B + k cn + k αn an for N ≥ 0 (A.42)
n=1 n=1 n=1

hold. If kαn < 1 for all n = 1, . . . , N + 1, then


N +1 N +1 N +1
! !
X X αn X
aN +1 + k bn ≤ exp k B+k cn for N ≥ 0.
n=1 n=1
1 − kαn n=1
(A.43)
If the inequality
N
X +1 N
X +1 N
X
aN +1 + k bn ≤ B + k cn + k αn an for N ≥ 0 (A.44)
n=1 n=1 n=1

is given, then it holds


N +1 N N +1
! !
X X X
aN +1 + k bn ≤ exp k αn B+k cn for N ≥ 0. (A.45)
n=1 n=1 n=1

Definition A.57. Weak convergence and weak∗ convergence, (Yosida,


1995, p. 120, p. 125). A sequence {xn }∞n=1 in a normed linear space X is said
to be weakly convergent if a finite limit limn→∞ f (xn ) exists for each f ∈ X ′ ,
184 A Functional Analysis

where X ′ is the (strong) dual space of X. If

lim f (xn ) = f (x) ∀ f ∈ X ′,


n→∞

then {xn }∞
n=1 is called to be weakly convergent to x, in notation xn −⇀ x.
A sequence {fn }∞ ′
n=1 in the (strong) dual space X of a linear space X is
said to be weakly∗ convergent if a finite limit limn→∞ fn (x) exists for every
x ∈ X. The sequence is said to converge weakly∗ to an element f ∈ X ′ if

lim fn (x) = f (x) ∀ x ∈ X,


n→∞


in notation fn −
⇀ f. ✷

Remark A.58. On the weak and weak∗ convergence.


• If the limit x or f exist, then the limit is unique, (Yosida, 1995, p. 120).
• If X is a reflexive Banach space and if {xn }∞ n=1 ⊂ X is bounded, then
there exists a subsequence {xnl }∞ ∞
l=1 ⊂ {xn }n=1 and an element x ∈ X
such that xnl −⇀ x, see (Evans, 2010, p. 723).
• If {fn }∞ ′ ′
n=1 ⊂ X is bounded in the dual X of X and X is a separable
Banach space, then there exists a weakly∗ convergent subsequence.

Definition A.59. Linear operator, range, kernel. Let X and Y be real


Banach spaces. A mapping A : X → Y is a linear operator if

A (αx1 + βx2 ) = αAx1 + βAx2 ∀ x1 , x2 ∈ X, α, β ∈ R.

The range or image of A is given by

range(A) = {y ∈ Y : y = Ax for some x ∈ X} .

The kernel or the null space of A is defined by

ker(A) = {x ∈ X : Ax = 0} .

Definition A.60. Bounded operator, continuous operator. An opera-


tor A : X → Y , X, Y Banach spaces, is bounded if

kAxkY
kAk = sup = sup kAxkY = sup kAxkY < ∞.
x∈X kxkX x∈X,kxkX ≤1 x∈X,kxkX =1
(A.46)
The operator A is continuous in x0 ∈ X if for each ε > 0 there is a δ > 0 such
that for all x ∈ X with kx − x0 kX < δ, it follows that kAx − Ax0 kY < ε. The
operator A is called a continuous operator if A is continuous for all x ∈ X. ✷
A.3 Some Definitions, Statements, and Theorems 185

Remark A.61. Equivalent definition of a continuous operator. The operator


A : X → Y , X, Y Banach spaces, is continuous in x0 ∈ X if and only if
for all sequences {xn }∞
n=1 , xn ∈ X, with xn → x0 it holds that Axn → Ax0
in Y . ✷

Lemma A.62. Properties of bounded linear operators, (Kolmogorov


& Fomı̄n, 1975, §4.5.2, §4.5.3), (Yosida, 1995, p.43). Let X, Y be Banach
spaces.
i) A bounded linear operator A : X → Y is continuous.
ii) A continuous linear operator A : X → Y is bounded.
iii) The set

L(X, Y ) = {A : A is a bounded linear operator from X to Y }

is a Banach space endowed with the norm (A.46).

Definition A.63. Compact operator. An operator A : X → Y is com-


pact, if A(x) is precompact in Y for every bounded set X̃ ⊂ X. ✷

Theorem A.64. Rank-nullity theorem. Let A : V → W be a linear map


between two finite-dimensional linear spaces V and W , then it holds

dim V = dim (ker (A)) + dim (range(A)) .

Definition A.65. Linear functional. A (real) linear functional f defined


on a Banach space X is a linear operator with range(f ) ⊂ R. ✷

Definition A.66. Bounded bilinear form, coercive bilinear form, V -


elliptic bilinear form. Let b(·, ·) : V × V → R be a bilinear form on the
Banach space V . Then, it is bounded if

|b(u, v)| ≤ M kukV kvkV ∀ u, v ∈ V, M > 0, (A.47)

where the constant M is independent of u and v. The bilinear form is coercive


or V -elliptic if
2
b(u, u) ≥ m kukV ∀ u ∈ V, m > 0, (A.48)
where the constant m is independent of u. ✷

Remark A.67. Application to an inner product. Let V be a Hilbert space.


Then the inner product a(·, ·) is a bounded and coercive bilinear form, since
by the Cauchy–Schwarz inequality

|a(u, v)| ≤ kukV kvkV ∀ u, v ∈ V,


2
and obviously a(u, u) = kukV . Hence, the constants can be chosen to be
M = 1 and m = 1. ✷
186 A Functional Analysis

Theorem A.68. Banach’s fixed point theorem, (Gilbarg & Trudinger,


1983, Theorem 5.1). A contraction mapping T in a Banach space B has a
unique fixed point, that is there exists a unique solution x ∈ B of the equation
T x = x.
The statement holds true if B is replaced by any closed subset, see (Gilbarg
& Trudinger, 1983, p. 74).

Theorem A.69. Brouwer’s fixed point theorem, (Gilbarg & Trudinger,


1983, Theorem 11.1). Let S be a compact convex set in a Banach space B
and let T be a continuous mapping of S into itself. Then T has a fixed point,
that is, T x = x for some x ∈ S.

Theorem A.70. Theorem of Banach on the inverse operator, (Kol-


mogorov & Fomı̄n, 1975, p. 225). Let A : X → Y be a bounded linear
operator that defines a one-to-one mapping between the Banach spaces X
and Y . Then, the inverse operator A−1 is bounded.

Theorem A.71. Closed Range Theorem of Banach, (Yosida, 1995,


p. 205). Let X, Y be Banach spaces, let A : X → Y be a bounded linear
operator, and let A′ : Y ′ → X ′ be its dual. Then, the following statements
are equivalent:
i) range(A) is closed in Y ,
ii) range(A) = {y ∈ Y : hy ′ , yiY ′ ,Y = 0 for all y ′ ∈ ker(A′ )},
iii) range(A′ ) is closed in X,
iv) range(A′ ) = {x′ ∈ X ′ : hx′ , xiX ′ ,X = 0 for all x ∈ ker(A)}.

Theorem A.72. Hahn–Banach Theorem, (Yosida, 1995, Section IV.1),


(Triebel, 1972, p. 67). Let X be a Banach space, let Y be a subspace of X,
and let f be a bounded linear functional defined on Y . Then, there exists an
extension g of f to X, where g is a linear functional with the same norm
as f .

You might also like