Function Spaces for Differential Equations

Thomas Götz
goetz@uni-koblenz.de
 0

Organization of the Course

pdf–slides with the content (download them)

Audio–podcast in addition (download it)
Listen to the podcast and follow the slides in parallel.
You may stop the audio podcast at any time.
Make your own notes, sketches, try proofs on your own.
Discuss with your course mates on skype, whatsapp, ...
Weekly chat to discuss problems
(Thu 08:30–10:00, starts Apr 16)
Exercise Sheets for the tutorials
Weekly video–conference during the tutorial session
(Wed 10:15, starts on Apr 22)
Details to be found in OLAT
Th. Götz (UKO) Function Spaces 2 / 64
 0

Why Function Spaces?

Linear models are often used in applied math as first approximation.
Their solution, typically a function, lives in function spaces.
Consider e.g.
population growth x 1 “ µx,
pendulum oscillation ϕ2 ` ω 2 ϕ “ 0,
slow and viscous flow ∇ ¨ u “ 0 and ∇p “ ∆ u.
All these problems are linear in following the sense:

If x and x̃ are solutions, then x ` x̃ and λx for λ P R are also

solutions.

But take care

This does not imply, that a solution x is a linear function itself!
Th. Götz (UKO) Function Spaces 3 / 64
 0

Example (Population Model x 1 “ µx)

Solution is given by xptq “ ce µt for c P R arbitrary
Why? Check this yourself!
This is not a linear function.
However, if xptq and x̃ptq are two solutions, i.e.
xptq “ ce µt and x̃ptq “ c̃e µt , then
1 1 1
1 px ` x̃q “ x ` x̃ “ µpx ` x̃q, so x ` x̃ is also a solution.
1 1
2 pλxq “ λx “ λµx “ µpλxq, so λx is also a solution.

Th. Götz (UKO) Function Spaces 4 / 64

 1 Linear Spaces

Function Spaces for Differential Equations

General Concepts of Linear Spaces

Section 1 Linear Spaces

Th. Götz (UKO) Function Spaces 5 / 64

 1 Linear Spaces

Definition 1.1 (Linear Space)

A (real) linear space is a set V with an addition ` : V ˆ V Ñ V and
a scalar multiplication ¨ : R ˆ V Ñ V such that
1 v `w “ w `v for all v , w P V .
2 u ` pv ` w q “ pu ` v q ` w for all u, v , w P V .
3 There exists a unique element 0 P V , called zero vector, such
that v ` 0 “ v .
4 For each v P V exists a unique element ´v P V such that
v ` p´v q “ v ´ v “ 0.
5 α ¨ pv ` w q “ α ¨ v ` α ¨ w for all v , w P V and all α P R.
6 α ¨ pβ ¨ v q “ pαβq ¨ v for all v P V and all α, β P R.
7 1¨v “ v for all v P V .
8 0¨v “ 0 P V for all v P V .
We write αv for α ¨ v .
Th. Götz (UKO) Function Spaces 6 / 64
 1 Linear Spaces

Remark
Linear spaces are also called vector spaces, their elements are
called vectors.
The real field R can be replaced by the complex field C mutatis
mutandis (with the necessary changes).
In the sequel we will just consider the real case and point out
structural differences between the real and complex case if
needed.
Note, that the symbol 0 denotes several different objects, the
scalar number zero 0 P R and the zero vector 0 P V .
We have not defined the multiplication of vectors!

Th. Götz (UKO) Function Spaces 7 / 64

 1 Linear Spaces

Example
For n P N and v1 , . . . , vn P R, the n–tupels

v “ pv1 , . . . , vn q,

are the vectors in the usual Euclidean vector space V “ Rn .

The addition and scalar multiplication are defined by
u ` v “ pu1 ` v1 , . . . , un ` vn q and
αv “ pαv1 , . . . , αvn q
The zero vector equals p0, 0, . . . , 0q.

The vectors in R2 or R3 are often used to describe points in the

two–dimensional plane or three–dimensional space.

Th. Götz (UKO) Function Spaces 8 / 64

 1 Linear Spaces

Example (Functions as vectors)

Let Ω “ ra, bs denote an interval. The set

CpΩq :“ tf : Ω Ñ R : f is continuous on Ωu

of scalar, real valued and continuous functions on Ω is a linear space.

For two functions f , g P CpΩq and α P R we define
the sum f ` g by pf ` g qpxq :“ f pxq ` g pxq
the scalar multiple αf by pαf qpxq :“ α ¨ f pxq for all x P Ω.
(Pointwise sum and scalar multiple).
To show the property f ` g P CpΩq, let f , g P CpΩq. Then we have
to show, that their sum f ` g is also continuous on Ω. Let x P Ω be
arbitrary and let xn Ñ x be any sequence. Then

pf ` g qpxn q “ f pxn q ` g pxn q Ñ f pxq ` g pxq “ pf ` g qpxq

Th. Götz (UKO) Function Spaces 9 / 64

 1 Linear Spaces

Example (Sum & scalar multiple of functions)

2.5
f
g
f+g
2 1/2 g

1.5

0.5

0
0 0.5 1 1.5 2

Th. Götz (UKO) Function Spaces 10 / 64

 1 Linear Spaces

Remark
Recall, that a function f : Ω Ñ R is called (sequentially) continuous
in a point x P Ω, if for all sequences pxn qnPN Ă Ω with lim xn “ x it
holds that lim f pxn q “ f pxq “ f plim xn q.
A function f is called continuous on Ω, if it is continuous for all
x P Ω.
We also write xn Ñ x instead of lim xn “ limnÑ8 xn “ x.

Th. Götz (UKO) Function Spaces 11 / 64

 1 Linear Spaces

Definition 1.2 (Spaces of continuous functions)

Let Ω Ă R be an (unbounded) interval. By

CpΩq “ C 0 pΩq “ C 0 pΩ; Rq :“ tf : Ω Ñ R : f is continuous on Ωu

we denote the set of all continuous functions on Ω. By

" *
k k pkq dk
C pΩq “ C pΩ; Rq :“ f : Ω Ñ R : f “ k f P CpΩq
dx

we denote the set of all k–times continuously differentiable functions

on Ω. By
č
C 8 pΩq “ C 8 pΩ; Rq :“ C k pΩ; Rq
kPN

we denote the set of all 8–often cont. diff’able functions on Ω.

Th. Götz (UKO) Function Spaces 12 / 64
 1 Linear Spaces

Theorem 1.1
The sets C k pΩ; Rq for k P N arbitrary and C 8 pΩ; Rq are linear spaces.
Their zero vector is the zero function 0 : Ω Ñ R defined by 0pxq “ 0
for all x P Ω and the negative vector ´f is defined by
p´f qpxq “ ´f pxq for x P Ω.

To show that C k pΩq is a linear space, we have to show, that the sum
of two k–times cont. diff’able functions is again k–times cont.
diff’able. And that the same holds true for scalar multiples of a
k–times cont. diff’able function.
This should be clear from your calculus courses.

Th. Götz (UKO) Function Spaces 13 / 64

 1 Linear Spaces

Definition 1.3 (Linear independence, Dimension)

Let V be a vector space. Pairwise different vectors v1 , . . . , vn P V ,
vi ‰ vj are called linearly independent, iff

α1 v1 ` . . . αn vn “ 0 always implies α1 “ 0 “ ¨ ¨ ¨ “ αn .

Only the trivial linear combination yields the zero vector.

We write dim V “ N P N, if the maximal number of linearly
independent vectors in V equals N and call N the dimension of V .
We write dim V “ 8, if for each n P N there exist n linearly
independent vectors in V . In this case we call V an
infinite–dimensional space.
The space V is called finite–dimensional, if 0 ď dim V ă 8.
For V “ t0u we define dim V “ 0 (zero–dimensional space).

The zeros vector is always linearly dependent.

Th. Götz (UKO) Function Spaces 14 / 64
 1 Linear Spaces

Example
dim Rn “ n.
How to prove this?

Th. Götz (UKO) Function Spaces 15 / 64

 1 Linear Spaces

Example
dim Rn “ n.

Proof.
We consider the vectors e1 “ p1, 0, . . . , 0q, e2 “ p0, 1, 0, . . . , 0q, . . . ,
en “ p0, . . . , 0, 1q and their linear combination α1 e1 ` ¨ ¨ ¨ ` αn en . If
the linear combination equals 0 we get

0 “ α1 e1 ` ¨ ¨ ¨ ` αn en
“ pα1 , 0, . . . , 0q ` p0, α2 , 0, . . . , 0q ` ¨ ¨ ¨ ` p0, . . . , 0, αn q
“ pα1 , α2 , . . . , αn q

Hence α1 “ 0 “ ¨ ¨ ¨ “ αn . Therefore the so–called standard basis

vectors e1 , . . . en are linearly independent.
To show that there are not more than n linearly independent vectors
assume there would and show a contradiction.
Th. Götz (UKO) Function Spaces 15 / 64
 1 Linear Spaces

Example
dim C 0 pra, bsq “ 8.

How to prove this?

Th. Götz (UKO) Function Spaces 16 / 64

 1 Linear Spaces

Example
dim C 0 pra, bsq “ 8.

Proof.
Set fk pxq :“ x k for k P N. Let α0 , . . . αk P R.
We consider the linear combination α0 f0 ` . . . αk fk “ 0 in Cpra, bsq
and get α0 ` α1 x ` ¨ ¨ ¨ ` αn x n “ 0 for all x P ra, bs.
Hence by the fundamental theorem of algebraa we get

α0 “ 0 “ ¨ ¨ ¨ “ αn .

Therefore the functions (vectors) f0 , . . . fn are linearly independent for

any n P N.
a
A polynomial of degree n, which is not the zero polynomial, has at
most n distinct zeros

Th. Götz (UKO) Function Spaces 16 / 64

 1 Linear Spaces

Definition 1.4
Let V be a linear space and U, W Ă V be subsets and α P R.
We call U a subspace, if U is itself a linear space.
We define

U ` W :“ tu ` w : u P U and w P W u ,
αU :“ tαu : u P Uu ,
U ˆ W :“ tpu, w q : u P U and w P W u ,

Th. Götz (UKO) Function Spaces 17 / 64

 2 Normed spaces

Function Spaces for Differential Equations

General Concepts of Linear Spaces

Section 2 Normed spaces

Th. Götz (UKO) Function Spaces 18 / 64

 2 Normed spaces

One often needs to define the distance of two elements (vectors) in a

linear space.
The general concept of convergence also relies on the concept of
distance. An element x is called the limit of a sequence pxn qnPN , if
the distance between individual sequence elements xn and the
assumed limit element x get smaller as the index n increases and
finally gets arbitrarily small, i.e. tends to zero.
So the concept of distance is essential for defining limits.

Th. Götz (UKO) Function Spaces 19 / 64

 2 Normed spaces

Definition 2.1 (Norm)

Let V be a linear space (over the real or complex field). A map
}¨} : V Ñ R (always R) is called norm (on V ), if it has the following
properties
1 }v } ě 0 for all v P V and }v } “ 0 iff v “ 0.

2 }αv } “ |α| ¨ }v } f5or all v P V , α P R, C.

3 }v ` w } ď }v } ` }w } for all v , w P V . (triangle inequality)

A linear space V together with a norm }¨} is called normed space.
A map d : V ˆ V Ñ R defined by dpu, v q “ }u ´ v } is called a
metric or distance of u and v .
Try to make a small sketch to visualize the triangle inequality. Here

Th. Götz (UKO) Function Spaces 19 / 64

 2 Normed spaces

Definition 2.1 (Norm)

Let V be a linear space (over the real or complex field). A map
}¨} : V Ñ R (always R) is called norm (on V ), if it has the following
properties
1 }v } ě 0 for all v P V and }v } “ 0 iff v “ 0.

2 }αv } “ |α| ¨ }v } f5or all v P V , α P R, C.

3 }v ` w } ď }v } ` }w } for all v , w P V . (triangle inequality)

A linear space V together with a norm }¨} is called normed space.
A map d : V ˆ V Ñ R defined by dpu, v q “ }u ´ v } is called a
metric or distance of u and v .
}v }

}u}
}u ` v }

Th. Götz (UKO) Function Spaces 19 / 64

 2 Normed spaces

Example
Consider the vector space Rn and
}v }1 :“ ni“1 |vi |
ř
bř
n 2
}v }2 :“ i“1 |vi |

}v }8 :“ maxni“1 |vi |.
The three above are norms on Rn .
How to show this?

Th. Götz (UKO) Function Spaces 20 / 64

 2 Normed spaces

Example
Consider the vector space Rn and
}v }1 :“ ni“1 |vi |
ř
bř
n 2
}v }2 :“ i“1 |vi |

}v }8 :“ maxni“1 |vi |.
The three above are norms on Rn .

Proof.
We just show triangle inequality for }¨}1 . Let u, v P Rn . Then
n
ÿ n
ÿ
}u ` v }1 “ |ui ` vi | ď |ui | ` |vi | triangle inequality in R
i“1 i“1
ÿn n
ÿ
“ |ui | ` |vi | “ }u}1 ` }v }1
i“1 i“1

Th. Götz (UKO) Function Spaces 20 / 64

 2 Normed spaces

To show the triangle inequality for the 2–norm, we need

Lemma 2.1 (Schwarz inequality)

Let x “ px1 , . . . , xn q and y “ py1 , . . . , yn q. Then
˜ ¸2
n
ÿ n
ÿ n
ÿ
xi yi ď xi2 ¨ yi2
i“1 i“1 i“1

Th. Götz (UKO) Function Spaces 21 / 64

 2 Normed spaces

Proof.
Since 0 ď pa ˘ bq2 “ a2 ˘ 2ab ` b 2 we get for all a, b P R

˘2ab ď a2 ` b 2
dÿ dÿ
2
We define a :“ xi { xj and b :“ yi { yj2 and sum over i.
j j
Then
¨ ˛2 ¨ ˛2
ř
2xi yi
i ˝ b xi ‚ ` ˝ b yi ‚
ÿ
˘2ab “ ˘ bř bř ď ř 2 ř 2
2 2
j xj ¨ j yj i j xj j yj
ř 2 ř 2
x y
ď ři i2 ` ři i2 “ 1 ` 1 “ 2
j xj j yj

Th. Götz (UKO) Function Spaces 22 / 64

 2 Normed spaces

Proof continued.
Therefore ˜ ¸1{2 ˜ ¸1{2
n
ÿ n
ÿ n
ÿ
˘ xi yi ď xj2 ¨ yj2
i“1 j“1 j“1

If we square this, we get the Schwarz inequality.

Now we can proceed to the triangle inequality for the 2–norm.

 2 Normed spaces

Proof: Triangle inequality for 2–norm.

Let u, v P Rn n
ÿ ÿ
}u ` v }22 “ pui ` vi q2 “ pui2 ` 2ui vi ` vi2 q
i“1 i
˜ ¸1{2 ˜ ¸1{2
ÿ ÿ ÿ ÿ
ď ui2 ` 2 ui2 vi2 ` vi2
i i i i
2 2
“ }u} ` 2 }u} ¨ }v } ` }v } “ p}u} ` }v }q2
 2 Normed spaces

Example
Let V “ Cpra, bs; Rq for a, b P R, a ă b. Then

}f }8 :“ max |f pxq|
xPra,bs

defines a norm, the so–called maximum or infinity norm.

Also ˆż b ˙1{2
2
}f }2 :“ |f pxq| dx
a

defines a norm, the so–called 2– or L2 –norm.

Th. Götz (UKO) Function Spaces 25 / 64

 2 Normed spaces

Example
Let V “ C 1 pra, bs; Rq for a ă b. Then

}f } :“ }f }8 ` }f 1 }8

defines a norm.

Th. Götz (UKO) Function Spaces 26 / 64

 2 Normed spaces

Definition 2.2 (Convergence)

Let pun qnPN be a sequence in a normed space pV , }¨}q. We write
limnÑ8 un “ u, iff
lim }un ´ u} “ 0
nÑ8

and call the sequence pun qnPN convergent with limit u.

If a sequence pun qnPN converges to u we write also un Ñ u for n Ñ 8.

How is the term limnÑ8 }un ´ u} defined? Ñ Calculus

Th. Götz (UKO) Function Spaces 27 / 64

 2 Normed spaces

Definition 2.2 (Convergence)

Let pun qnPN be a sequence in a normed space pV , }¨}q. We write
limnÑ8 un “ u, iff
lim }un ´ u} “ 0
nÑ8

and call the sequence pun qnPN convergent with limit u.

If a sequence pun qnPN converges to u we write also un Ñ u for n Ñ 8.

Remark
Formally limnÑ8 }un ´ u} “ 0 is defined as:
For all ε ą 0, there exists an index N P N such that }un ´ u} ď ε for
all indices n ě N.
With the mathematical quantors, this reads as

@ε ą 0 DN P N : }un ´ u} ď ε @n ě N.

Th. Götz (UKO) Function Spaces 27 / 64

 2 Normed spaces

Example
Let un pxq :“ x n be a sequence in Cpr0, 1sq and consider the
maximum–norm.
# Then the sequence does not converge to
0 x ă1,
ϕpxq :“
1 x “ 1.

Why that?
1
x4
0.9
x8
0.8 x 50

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1

Th. Götz (UKO) Function Spaces 28 / 64

 2 Normed spaces

Example
Let un pxq :“ x n be a sequence in Cpr0, 1sq and consider the
maximum–norm.
# Then the sequence does not converge to
0 x ă1,
ϕpxq :“
1 x “ 1.

Proof.
}un ´ ϕ}8 “ maxxPr0,1s |x n ´ ϕpxq| “ max0ăxă1 x n .
Choosing n P N fixed, we can pick a point x close to 1 such that
x n ą ε for any given ε. Therefore the sequence does not
converge.

Th. Götz (UKO) Function Spaces 28 / 64

 2 Normed spaces

Example
Consider again the same space Cpr0, 1sq but now endowed with the
ş1
norm }u}1 “ 0 |upxq| dx. Then, for the same sequence un pxq “ x n as
before we get
un pxq “ x n Ñ 0

Why that? How to see this?

Th. Götz (UKO) Function Spaces 29 / 64

 2 Normed spaces

Example
Consider again the same space Cpr0, 1sq but now endowed with the
ş1
norm }u}1 “ 0 |upxq| dx. Then, for the same sequence un pxq “ x n as
before we get
un pxq “ x n Ñ 0

Proof.
ż1 ż1
1
}un ´ 0}1 “ |un pxq| dx “ x n dx “ Ñ0
0 0 n`1

What do we learn from this and the previous example?

Th. Götz (UKO) Function Spaces 29 / 64

 2 Normed spaces

Example
Consider again the same space Cpr0, 1sq but now endowed with the
ş1
norm }u}1 “ 0 |upxq| dx. Then, for the same sequence un pxq “ x n as
before we get
un pxq “ x n Ñ 0

Proof.
ż1 ż1
1
}un ´ 0}1 “ |un pxq| dx “ x n dx “ Ñ0
0 0 n`1

Remark
Convergence of a sequence can depend on the chosen norm!

Th. Götz (UKO) Function Spaces 29 / 64

 2 Normed spaces

Theorem 2.2
Let pV , }¨}q be a normed space and let un , vn , u, v P V and αn , α P R.
Assume un Ñ u and vn Ñ v . Then
1 The limit is unique, i.e. if lim un “ u and lim un “ ũ, then u “ ũ.

2 The sequence pun qnPN is bounded, i.e. there exists r ě 0 such

that }un } ď r for all n P N.
3 }un } Ñ }u}, un ` vn Ñ u ` v and αn un Ñ αu, if αn Ñ α.

Proof. ad 1.
Assume two limits u and ũ of the same sequence pun qnPN . Then

}u ´ ũ} “ }pu ´ un q ` pun ´ ũq}

ď }u ´ un } ` }un ´ ũ} Ñ 0 ` 0

Hence u “ ũ.
Th. Götz (UKO) Function Spaces 30 / 64
 2 Normed spaces

Theorem 2.2
Let pV , }¨}q be a normed space and let un , vn , u, v P V and αn , α P R.
Assume un Ñ u and vn Ñ v . Then
1 The limit is unique, i.e. if lim un “ u and lim un “ ũ, then u “ ũ.

2 The sequence pun qnPN is bounded, i.e. there exists r ě 0 such

that }un } ď r for all n P N.
3 }un } Ñ }u}, un ` vn Ñ u ` v and αn un Ñ αu, if αn Ñ α.

Proof. ad 2 & 3.
(2) Since }un ´ u} Ñ 0, the real sequence }un ´ u} is bounded, let’s
say by R, i.e. }un ´ u} ď R for all n P N. Then for all n P N

}un } “ }pun ´ uq ` u} ď }un ´ u} ` }u} ď R ` }u}

(3) Do it on your own.

Th. Götz (UKO) Function Spaces 30 / 64
 2 Normed spaces

Definition 2.3 (Cauchy sequence)

A sequence pun qnPN in a normed space is called a Cauchy sequence, if
for all ε ą 0 there exists an index N “ Npεq P N such that

}un ´ um } ă ε @n, m ě N .

Th. Götz (UKO) Function Spaces 31 / 64

 2 Normed spaces

Theorem 2.3
In a normed space every convergent sequence is a Cauchy sequence.

but ...

Th. Götz (UKO) Function Spaces 32 / 64

 2 Normed spaces

Theorem 2.3
In a normed space every convergent sequence is a Cauchy sequence.

Remark
Not every Cauchy sequence is also convergent.

This leads to ...

Th. Götz (UKO) Function Spaces 32 / 64

 2 Normed spaces

Theorem 2.3
In a normed space every convergent sequence is a Cauchy sequence.

Remark
Not every Cauchy sequence is also convergent.

Definition 2.4 (Banach Space)

A normed space pB, }¨}q is called a Banach space, if every Cauchy
sequence is convergent.
In this case, the space is also called complete.

Th. Götz (UKO) Function Spaces 32 / 64

 2 Normed spaces

Example
The spaces pR, |¨|q, pC, |¨|q (both equipped with the usual absolute
value as norms) and pRn , }¨}8 q, pCn , }¨}8 q are Banach spaces.

For R and C this follows from their axiomatic definition

(completeness axiom).
For Rn : Let xk “ px1,k , . . . , xn,k q denote a sequence in Rn .
Then limkÑ8 }xk ´ x}8 “ 0, iff each component converges, i.e.
limkÑ8 |xi,k ´ xi | “ 0 for all i “ 1, . . . n.
That is, the convergence in Rn is equivalent to the
component–wise convergence in R.
Given a Cauchy sequence in Rn , then each component is a
Cauchy sequence in R. Now, each component converges in R,
hence the entire sequence in Rn .

Th. Götz (UKO) Function Spaces 33 / 64

 2 Normed spaces

Example
Let Ω “ ra, bs be a compact (closed and bounded) interval. Then
pCpΩq, }¨}8 q is a Banach space.
The convergence fn Ñ f as n Ñ 8 in CpΩq means
}fn ´ f }8 “ max |fn pxq ´ f pxq| Ñ 0 as n Ñ 8.
aďxďb

What do we have to do to prove this? Take a Cauchy sequence in

C pΩq and show that is converges to a function, that is also an
element in C pΩq.

Th. Götz (UKO) Function Spaces 34 / 64

 2 Normed spaces

Example
Let Ω “ ra, bs be a compact (closed and bounded) interval. Then
pCpΩq, }¨}8 q is a Banach space.
The convergence fn Ñ f as n Ñ 8 in CpΩq means
}fn ´ f }8 “ max |fn pxq ´ f pxq| Ñ 0 as n Ñ 8.
aďxďb

Proof.
Let pfn qnPN Ă CpΩq be a Cauchy sequence, i.e.
}fn ´ fm }8 “ max |fn pxq ´ fm pxq| ă ε for all n, m ě N
aďxďb

Now, choose x P Ω fixed. Then the sequence of function values

pfn pxqqnPN is a Cauchy sequence in R and hence it converges to some
value f pxq.
It remains to show that the function x ÞÑ f pxq is continuous.

Th. Götz (UKO) Function Spaces 34 / 64

 2 Normed spaces

Proof ctd.
Therefore let ε ą 0 Then there exists some δ ą 0 such that
|fn pxq ´ fn py q| ă ε{3 for all |x ´ y | ă δ and all n P N since all
sequence elements fn are continuous functions. Choose a y such that
|x ´ y | ă δ. Then there exists an index Ñ P N such that
ε ε
|fn pxq ´ f pxq| ă and |fn py q ´ f py q| ă
3 3

for n ě Ñ since both real sequences pfn pxqq and pfn py qq converge.
Therefore

|f pxq ´ f py q| ď |f pxq ´ fn pxq| ` |fn pxq ´ fn py q| ` |fn py q ´ f py q|

ε ε ε
ă ` ` “ε
3 3 3

Th. Götz (UKO) Function Spaces 35 / 64

 2 Normed spaces

Theorem 2.4
Let B be a Banach space and let pun qnPN Ă B be a Cauchy sequence
which has a convergent subsequence punk qkPN , i.e. unk Ñ u for
k Ñ 8. Then the entire sequence converges to u, i.e. un Ñ u as
n Ñ 8.
Think about a proof.

Th. Götz (UKO) Function Spaces 36 / 64

 2 Normed spaces

Theorem 2.4
Let B be a Banach space and let pun qnPN Ă B be a Cauchy sequence
which has a convergent subsequence punk qkPN , i.e. unk Ñ u for
k Ñ 8. Then the entire sequence converges to u, i.e. un Ñ u as
n Ñ 8.

Proof.
Let ε ą 0. Since pun q is a Cauchy sequence, there exists N such that
}un ´ um } ă ε{2 for all n, m ě N. Since the subsequence punk q
converges to u, there exists an index k large enough, such that
nk ě N and }unk ´ u} ă ε{2. Hence

}un ´ u} ď }un ´ unk } ` }unk ´ u} ă ε{2 ` ε{2 “ ε

Th. Götz (UKO) Function Spaces 36 / 64

 2 Normed spaces

Example
Consider pCpΩq, }¨}8 q for Ω “ r0, 1s and the sequence un pxq “ x n .
Then un is no Cauchy sequence, since
› › ˇ ˇ
}un ´ u2n }8 “ ›x n ´ x 2n ›8 “ max ˇx n ´ x 2n ˇ
0ďxď1

which is taken at x “ 2´1{n and max0ďxď1 |x n ´ x 2n | “ 1{4

independent of n.

Th. Götz (UKO) Function Spaces 37 / 64

 3 Hilbert Spaces

Function Spaces for Differential Equations

General Concepts of Linear Spaces

Section 3 Hilbert Spaces

Th. Götz (UKO) Function Spaces 38 / 64

 3 Hilbert Spaces

In this section we will generalize the concept of orthogonality to

general vector spaces. For that purpose we need again some
definitions.

Th. Götz (UKO) Function Spaces 39 / 64

 3 Hilbert Spaces

Definition 3.1 (Inner / Scalar product)

Let V be a linear space. A map x¨, ¨y : V ˆ V Ñ R, C is called an
inner product or scalar product on V , if
1 xu, uy ě 0 for all u P V and xu, uy “ 0 iff u “ 0.

2 xu, αv ` βw y “ αxu, v y ` βxu, w y

3 xu, v y “ xv , uy in the real case

xu, v y “ xv , uy in the complex case, where z denotes the

complex conjugate.
Two vectors u, v P V are called orthgonal, iff xu, v y “ 0.

Th. Götz (UKO) Function Spaces 39 / 64

 3 Hilbert Spaces

Example
On the spaces Rn and Cn the Euclidean (standard) scalar product
ÿ
xx, y y “ xi yi for Rn
ÿ
xx, y y “ xi yi for Cn

are inner products according to the above definition.

In particular we have

xei , ej y “ 0

and for x “ px1 , . . . , xn q we get

xx, ei y “ xi .

Check each of the properties on your own.

Th. Götz (UKO) Function Spaces 40 / 64
 3 Hilbert Spaces

Example
On Cpra, bsq we consider
żb
xf , g y :“ f pxq g pxq dx
a

Then this defines a inner product.

Why? We consider xf , f y “ 0 iff f “ 0. The rest is on your own.

Th. Götz (UKO) Function Spaces 41 / 64

 3 Hilbert Spaces

Example
On Cpra, bsq we consider
żb
xf , g y :“ f pxq g pxq dx
a

Then this defines a inner product.

Proof.
Assume f ‰ 0, then there exists a ξ P ra, bs with f pξq ą 0 (wlog).
Since f is continuous, there exists an entire interval rξ ´ δ, ξ ` δs,
where f ą 0. What, if ξ “ a or ξ “ b?
Therefore ż b ż ξ`δ
xf , f y “ f pxq2 dx ě f pxq2 dx ą 0.
a ξ´δ

Th. Götz (UKO) Function Spaces 41 / 64

 3 Hilbert Spaces

Theorem 3.1 (Schwarz inequality; inner product version)

Let V be a linear space with inner product. Then for all u, v P V

|xu, v y| ď xu, uy1{2 ¨ xv , v y1{2

Proof.
Let v ‰ 0. Then
0 ď xu ´ αv , u ´ αv y “ xu, uy ´ αxu, v y ´ α rxv , uy ´ αxv , v ys

and now choose α “ xv , uy{xv , v y to get

xu, v y xv , uy
0 ď xu, uy ´
xv , v y

and hence |xu, v y|2 ď xu, uy ¨ xv , v y

Th. Götz (UKO) Function Spaces 42 / 64

 3 Hilbert Spaces

Lemma 3.2
Each inner product x¨, ¨y defines a norm via
a
}v } “ xv , v y

We have to prove the three properties of a norm. Here just triangle

inequality. Rest on your own.

Th. Götz (UKO) Function Spaces 43 / 64

 3 Hilbert Spaces

Lemma 3.2
Each inner product x¨, ¨y defines a norm via
a
}v } “ xv , v y

Proof.
Let u, v P V . Then

}u ` v }2 “ xu ` v , u ` v y “ xu, uy ` xu, v y ` xu, v y ` xv , v y

“ }u}2 ` 2 ¨ Re xu, v y ` }v , v }2
ď }u}2 ` 2 }u} ¨ }v } ` }v }2 “ p}u} ` }v }q2

So the triangle inequality holds.

Th. Götz (UKO) Function Spaces 43 / 64

 3 Hilbert Spaces

Definition 3.2 (Hilbert Space)

A Banach space where the norm is induced by an inner product is
called a Hilbert space.

In other words: A linear space with inner productaand where each

Cauchy sequence converges wrt the norm }v } “ xv , v y is a Hilbert
space.

Th. Götz (UKO) Function Spaces 44 / 64

 3 Hilbert Spaces

Example
The finite–dimensional spaces Rn and Cn with their standard scalar
products and induced norms
ÿ ÿ
Rn : xx, y y “ xi yi }x}22 “ xi2
ÿ ÿ 2
Cn : xx, y y “ xi yi }x}22 “ |xi |

are Hilbert spaces.

Th. Götz (UKO) Function Spaces 45 / 64

 3 Hilbert Spaces

Example
The linear space Cpra, bsq with the inner product and norm
żb
xf , g y “ f pxq g pxq dx
a
ˆż b ˙1{2
2
}f }2 “ f pxq dx
a

is NOT a Hilbert space.

We will see later on, that it is not complete, i.e. there are Cauchy
sequences that do not converge.

Th. Götz (UKO) Function Spaces 46 / 64

 3 Hilbert Spaces

Definition 3.3 (Bilinear Form)

Let V be a normed space. A map a : V ˆ V Ñ R, C is called a
bounded bilinear form, if it has the following properties
1 apαu ` βv , w q “ αapu, v q ` βapv , w q

apu, αv ` βw q “ αapu, v q ` βapu, w q

2 There exists a constant d ą 0, such that

|apu, v q| ď d ¨ }u} ¨ }v } .

We call ap¨, ¨q
symmetric, iff apu, v q “ apv , uq
positive, iff 0 ď apu, uq
strongly positive, if there exists a constant c ą 0 such that

c ¨ }u}2 ď apu, uq .
Th. Götz (UKO) Function Spaces 47 / 64
 3 Hilbert Spaces

Theorem 3.3
Let V be a normed space with inner product x¨, ¨y. Then the inner
product is continuous, i.e. un Ñ u and vn Ñ v imply
xun , vn y Ñ xu, v y.

How to prove it?

Th. Götz (UKO) Function Spaces 48 / 64

 3 Hilbert Spaces

Theorem 3.3
Let V be a normed space with inner product x¨, ¨y. Then the inner
product is continuous, i.e. un Ñ u and vn Ñ v imply
xun , vn y Ñ xu, v y.

Proof.
Since pvn qnPN is bounded, it follows from Schwarz inequality that

|xun , vn y ´ xu, v y| “ |xun ´ u, vn y ` xu, vn ´ v y|

ď |xun ´ u, vn y| ` |xu, vn ´ v y|
ď }un ´ u} ¨ }v } ` }u} ¨ }vn ´ v } Ñ 0

Th. Götz (UKO) Function Spaces 48 / 64

 3 Hilbert Spaces

Remark
The inner product of a Hilbert space is continuous.

But this is not only true for inner products ...

Th. Götz (UKO) Function Spaces 49 / 64

 3 Hilbert Spaces

Remark
The inner product of a Hilbert space is continuous.

Theorem 3.4
Let ap¨, ¨q be a bounded bilinear form on a normed linear space. Then
ap¨, ¨q is continuous.

The proof is left as an exercise.

Th. Götz (UKO) Function Spaces 49 / 64

 3 Hilbert Spaces

Theorem 3.5
Let H be a Hilbert space with inner product x¨, ¨y. Let v1 , ¨ ¨ ¨ P Hz t0u
be pairwise orthogonal, i.e. xvi , vj y “ 0 for i ‰ j and vi ‰ 0. Then
the vectors (or any finite subset of them) are linearly independent.

Let’s prove it...

Th. Götz (UKO) Function Spaces 50 / 64

 3 Hilbert Spaces

Theorem 3.5
Let H be a Hilbert space with inner product x¨, ¨y. Let v1 , ¨ ¨ ¨ P Hz t0u
be pairwise orthogonal, i.e. xvi , vj y “ 0 for i ‰ j and vi ‰ 0. Then
the vectors (or any finite subset of them) are linearly independent.

Proof.
Consider a linear combination α1 v1 ` ¨ ¨ ¨ ` αn vn “ 0. Choose one vj
with j arbitrary and we get

0 “ xvj , 0y “ xvj , α1 v1 ` ¨ ¨ ¨ ` αn vn y
“ α1 xvj , v1 y ` ¨ ¨ ¨ ` αn xvj , vn y “ αj xvj , vj y
“ αj }vj }2

Since vj ‰ 0 we obtain αj “ 0 for j arbitrary.

Th. Götz (UKO) Function Spaces 50 / 64

 3 Hilbert Spaces

Theorem 3.6 (Parallelogram identity)

Let H be a Hilbert space. Then
´ ¯
}u ` v }2 ` }u ´ v }2 “ 2 }u}2 ` }v }2

Why is this called Parallelogram identity? Make a sketch.

Prove it.

Th. Götz (UKO) Function Spaces 51 / 64

 4 Some Topology

Function Spaces for Differential Equations

General Concepts of Linear Spaces

Section 4 Some Topology

Th. Götz (UKO) Function Spaces 52 / 64

 4 Some Topology

With the concept of norms and distances, we can also define

neighborhoods of points. This happens next

Th. Götz (UKO) Function Spaces 53 / 64

 4 Some Topology

Definition 4.1
Let V be a normed space. For u0 P V and ε ą 0 we define

Bε pu0 q :“ tu P V : }u ´ u0 } ă εu

called the ε–neighborhood of (or the ε–ball around) the point u0 P V .

A set M Ă V is called
open, iff for each u0 P M there exists some ε–neighborhood
Bε pu0 q Ă M entirely contained in M.
closed, iff its complement V zM is open.
An open subset Npu0 q Ă V containing a point u0 P V is called an
open neighborhood of u0 .

ε
u
Bε puq
Th. Götz (UKO) Function Spaces 53 / 64
 4 Some Topology

Remark
The empty set H and the entire space V are open and closed at the
same time!
There exists subsets, that are neither open nor closed.

Remark
In some literature, the notation Br puq refers to the closed ball of
radius r around u, i.e.

Br puq “ tv P V : }u ´ v } ď r u .

Th. Götz (UKO) Function Spaces 54 / 64

 4 Some Topology

Theorem 4.1
Let V be a normed space and M Ă V . The following are equivalent
1 M is closed.

2 If a sequence pun qNPN Ă M is convergent to u, then u P M.

(the limit of a convergent sequence in M is also in M)

Lets proof the individual directions...

lim un

Th. Götz (UKO) Function Spaces 55 / 64

 4 Some Topology

Theorem 4.1
Let V be a normed space and M Ă V . The following are equivalent
1 M is closed.

2 If a sequence pun qNPN Ă M is convergent to u, then u P M.

(the limit of a convergent sequence in M is also in M)

Proof p1q ùñ p2q.

Let pun qnPN Ă M and un Ñ u. Assume u R M, i.e. u P V zM. Since
M is closed, we have V zM open and hence there exists ε ą 0 s.t.
Bε puq Ă V zM. From the convergence }Un ´ u} Ñ 0 we get, that
there exists an index N s.t. }un ´ u} ă ε for all n ě N and hence
un P Bε puq Ă V zM for all n ě N. This contradicts pun qnPN Ă M.

Th. Götz (UKO) Function Spaces 55 / 64

 4 Some Topology

Theorem 4.1
Let V be a normed space and M Ă V . The following are equivalent
1 M is closed.

2 If a sequence pun qNPN Ă M is convergent to u, then u P M.

(the limit of a convergent sequence in M is also in M)

Proof p2q ùñ p1q.

Suppose M is not closed, i.e. V zM is not open. Then there exits
u P V zM s.t. no ε–neighborhood Bε puq is contained in V zM.
Choosing ε “ 1{n we construct a sequence Un P B1{n puq and un P M
for all n P N. Hence }un ´ u} ă 1{n Ñ 0 and due to (2) we have
u P M. This contradicts u P V zM.

Th. Götz (UKO) Function Spaces 55 / 64

 4 Some Topology

Definition 4.2
Let M Ă V be a subset of a normed space V . We define
span M = smallest linear subspace of V containing M
M = smallest closed set in V containing M
M̊ = largest open set in V also contained in M
BM “ MzM̊ as the boundary of M
A point u P V is called an interior, boundary or exterior point of M,
iff u P M̊, u P BM or u P pV ˚zMq.

Th. Götz (UKO) Function Spaces 56 / 64

 4 Some Topology

Remark
č
M“ tC closed and M Ă C u
C ĂV
ď
M̊ “ tO open and O Ă Mu
OĂV

Th. Götz (UKO) Function Spaces 57 / 64

 4 Some Topology

Definition 4.3 (Convexity)

A subset M Ă V of a linear space V is called convex, iff for all
u, v P M
αu ` p1 ´ αqv P M @0 ď α ď 1 .
A function f : M Ñ R is called convex, iff M is convex and for all
u, v P M

f pαu ` p1 ´ αqv q ď αf puq ` p1 ´ αqf pv q @0 ď α ď 1 .

A set is convex if for any two points also the line joining those two
points belongs to M.

M not convex!

Th. Götz (UKO) Function Spaces 58 / 64

 4 Some Topology

Lemma 4.2
Let (V , }¨}q be a normed space. Then
1 The ε–ball Bε pU0 q is convex for any u0 P V and ε ą 0.

2 The norm }¨} : V Ñ R is continuous and convex.

That’s not too difficult...

Th. Götz (UKO) Function Spaces 59 / 64

 4 Some Topology

Lemma 4.2
Let (V , }¨}q be a normed space. Then
1 The ε–ball Bε pU0 q is convex for any u0 P V and ε ą 0.

2 The norm }¨} : V Ñ R is continuous and convex.

Proof.
1 Let v , w P Bε pu0 q and 0 ď α ď 1. Then

}αv ` p1 ´ αqw } “ }αpv ´ u0 q ` p1 ´ αqpw ´ u0 q}

ď α }v ´ u0 } ` p1 ´ αq }w ´ u0 }
ă αε ` p1 ´ αqε “ ε

2 Exercise.

Th. Götz (UKO) Function Spaces 59 / 64

 4 Some Topology

Definition 4.4
Let V be a normed space.
A subspace U Ă V is called a closed subspace, if U vis a
subspace and a closed set.
For a subset M Ă V we define
§ co M = smallest convex subset in V containing M, the
convex hull.
§ co M = smallest closed convex set in V containing M.

Th. Götz (UKO) Function Spaces 60 / 64

 4 Some Topology

Definition 4.5 (Compactness)

Let V be a normed space and M Ă V . M is called
bounded, iff there exists r ě 0 such that }u} ď r for all u P M.
relatively (sequentially) compact, iff each sequence pun qnPN Ă M
kÑ8
has a convergent subsequence unk ÝÑ u P V .
(sequentially) compact, iff each sequence pun qnPN Ă M has a
kÑ8
convergent subsequence unk ÝÑ u and u P M.

Th. Götz (UKO) Function Spaces 61 / 64

 4 Some Topology

Theorem 4.3
M is compact, iff it is relatively compact and closed.

Proof as an Exercise.

Th. Götz (UKO) Function Spaces 62 / 64

 4 Some Topology

Example
Let M be a subset of Rn , Cn with the norm |¨| in R, C or }¨}8 in
Rn , Cn for n ě 2.
Then M is relatively compact, iff it is bounded.
In 1d the proof is a classical result from calculus.

Th. Götz (UKO) Function Spaces 63 / 64

 4 Some Topology

Example
Let M be a subset of Rn , Cn with the norm |¨| in R, C or }¨}8 in
Rn , Cn for n ě 2.
Then M is relatively compact, iff it is bounded.

Proof for R.
Let M be bounded and let pun qnPN Ă M. Then, due to the
Bolzano–Weierstraß Theorema , there exists a convergent
subsequence.
Now assume M relatively compact. Suppose M is not bounded.
Then there exists a sequence pun qnPN Ă M s.t. }un } ě n. Since M is
relatively compact, this sequence should have a convergent
subsequence which has to be bounded. Contradiction.
a
Any bounded sequence in R has a convergent subsequence. Proof
uses nested intervals.

Th. Götz (UKO) Function Spaces 63 / 64

 4 Some Topology

Remark
Finite–dimensional normed spaces are Banach spaces!
Let V be a finite–dimensional normed space.
1 Let U Ă V be a subspace. Then U is closed.

2 Let M Ă V be a subset. Then M is compact, iff it is closed and

bounded.
The above is only true in finite dimensions. For infinite dimensional
spaces this does not hold true in general.

Th. Götz (UKO) Function Spaces 64 / 64