Fa Eng 2017

Functional analysis and partial differential
equations
Winter term 2017/2018
Prof. Dr. Stefan Müller
http://hcm.uni-bonn.de/people/mueller/
What is functional analysis?

Functional analysis is, roughly speaking, analysis on infinite dimensional
spaces. The functional analytic point of view has meanwhile become essen-
tial in many areas of mathematics, e.g. in
• Geometry and topology (index theorems, loop spaces, the diffeomor-

phism group of a manifold as an infinite dimensional Lie group, ...)
• Probability
• Numerical analysis (seemingly very different approximation of a prob-

lem often correspond simply to different choices of bases in an infinite
dimensional space; robust behaviour of very large finite dimensional
approximation is often related to good estimates for the infinite di-
mensional problem)
• Mathematical physics (self adjoint operators, spectral theory, ...)
• Partial differential equations (see below)
This course will give an introduction to functional analysis and covers in

particular the central theorems of linear functional analysis. The course
is also suitable for students, who have not followed the course
’Einführung in die partiellen Differentialgleichungen’ (Introduc-
tion to partial differential equations). Partial differential equations
will be used as examples of applications of methods from functional analy-
sis, but no previous knowledge of partial differential equations is assumed.
A typical problem is the following. Let X and Y be infinite (normed,
metric, topological, ...) vector spaces and let A : X → Y be a linear and
continuous map. Under which conditions does A have a continuous inverse?
An important theorem of linear algebra states that for X = Y = Rd we
have A injective ⇔ A surjective ⇔ A−1 exists and is continuous. This is no
longer true for infinite dimensional spaces.
1
Here is an example from the theory of partial differential equations. Let
X denote the space C 2 (U ) of twice differentiable functions from a bounded
and open set U ⊂ Rn to R with bounded second derivatives, which converge
to 0 on ∂U . Let Y be the corresponding space of continuous maps P and
∂2
bounded maps from U to R and let A be the Laplaceoperator ∆ = j ∂x 2.
j
The question of the solvability of the PDG ∆u = f is thus translated into
the question of the invertibility of a certain operator. In this example it
happens that A is injective, but not surjective. One obtains a much better
theory if one replaces the classical spaces C 2 and C 0 by Sobolev spaces (see
below).
The course will cover four big themes.
1 Spaces and topologies/ notions of convergence

An important theorem of analysis states that a subset E of Rn is compact
if and only if it is bounded and closed. As a result, every bounded sequence
in Rn has a convergent subsequence.
This assertion does not hold in infinite dimensional spaces. Compact-
ness is nonetheless a key property. We will thus introduce new topologies
(the weak topology and the weak* topology), for which good compactness
properties hold.
Another important aspect is that the classical function spaces such as
the space C k of k times differentiable functions are not well adapted to
other natural mathematical structures. On C 1 , for example, one can define
a scalar product Z
hu, vi := uv + ∇u · ∇v
Nonetheless the space C 1 is not complete under the corresponding norm

||u|| := hu, ui1/2 . The completion leads to the Sobolev space W 1,2 and we
will study the central properties of Sobolev spaces (imbedding theorems into
classical spaces, Poincaré inequality, approximation properties, ...).
2 Linear operators
Key theorems are the Hahn-Banach theorem (a continuous linear functional
on a subspace can be extended to the whole space), Baire’s theorem and the
closed graph theorem. Baire’s theorem has a number of strong consequences,
e.g. the principle of uniform boundedness: A family of linear and continuous
2
maps Tj from a Banach space X to a Banach space Y which is pointwise
bounded is already uniformly bounded, i.e.,
sup ||T j x|| < ∞ ∀x ⇒ sup sup ||T j x|| < ∞.

j j x:||x||≤1
3 Hilbert spaces
A Hilbert space H is a vector spaces with a scalar product which is closed
under the norm induced by the scalar product. A key property is that for
every closed subspace U and every x ∈ H there exists a unique best approx-
imation of y ∈ U of x. Moreover y is given by the orthogonal projection of
x onto U .
This simple geometric fact has strong consequence, e.g. the Riesz rep-
resentation theorem (every linear and continuous functional T : H → R is
of the form T (x) = hz, xi for some z ∈ H) and the Lax-Milgram theorem.
This leads to a new approach for the solution of the equation Ax = y. One
considers the bilinear form a(x, z) := hAx, zi and look for solutions of the
equation a(x, z) = (y, z) ∀z. In this way the PDE −∆u = f in U , u = 0 on
∂U can be reformulated as
Z Z
∇u∇ϕ = f ϕ ∀ϕ
U U
(to see this, multiply the equation by ϕ and integrate by parts). The left
hand side of the equation defines a scalar product on the space W01,2 of
Sobolev functions with zero boundary values and for fixed f ∈ L2 (U ), the
right hand side defines a continuous linear map from W01,2 to R. Thus, the
Riesz representation theorem yields the existence and uniqueness of a solu-
tion u ∈ W01,2 . The Lax-Milgram theorem yields existence and uniqueness
for general bilinear forms which are continuous and coercive (i.e. a(x, x) ≥
c||x||2 , c > 0). In this way, one immediately obtains existence
P and unique-
ness of solutions of general partial differential equations − i,j ∂i (aij (x)∂j u(x)) =
fPas long as the coefficients are bounded (and measurable) and elliptic (i.e.,
aij ξi ξj ≥ c|x|2 ∀ξ ∈ Rn , with c > 0).
This Hilbert space approach does not only lead to a very flexible and
powerful notion of solutions but also creates a rigorous link to the natural
minimization principles in physics. At the same time the Hilbert space
approach leads to a very natural understanding and analysis of numerical
approximation schemes, which often corresponds to an approximation in a
well chosen finite dimensional subspace.
3
4 Eigenvalues and spectral theory
An important theorem of linear algebra states that a symmetric matrix has
an orthonormal basis of eigenvectors. We will prove a generalization of this
result to compact and symmetric operators on a Hilbert space. For natural
choices of the operator (usually the inverse of a differential operator) the
corresponding orthonormal systems of eigenfunctions yield a unification and
generalization of a number of classical orthonormal systems (Fourier series,
Legendre polynomials, spherical harmonics, . . . ).
For noncompact bounded or unbounded symmetric (or self adjoint) op-
erators A one needs a generalization of the concept of an eigenvalue. One
says that λ ∈ C belongs to the spectrum of A if the operator λ − A does not
have a continuous inverse. The concept of spectrum is a central notion in
many areas of mathematics ranging from mathematical physics to number
theory (Selberg trace formula).
Literature
The course will be mostly based on the following two books:
H.W. Alt, Linear functional analysis, Springer, 2016
H. Brezis, Functional analysis, Sobolev spaces and partial differential equa-
tions, Springer
Further relevant literature will be announced in the course as needed.
Prerequisites
Analysis III (i.e. measure and integration, in particular the Lebesgue inte-
gral)

Fa Eng 2017

Uploaded by

Copyright:

Available Formats

Fa Eng 2017

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fa Eng 2017

Uploaded by

Copyright:

Available Formats

Functional analysis and partial differential

What is functional analysis?

• Geometry and topology (index theorems, loop spaces, the diffeomor-

• Numerical analysis (seemingly very different approximation of a prob-

• Mathematical physics (self adjoint operators, spectral theory, ...)

• Partial differential equations (see below)

This course will give an introduction to functional analysis and covers in

1 Spaces and topologies/ notions of convergence

Nonetheless the space C 1 is not complete under the corresponding norm

sup ||T j x|| < ∞ ∀x ⇒ sup sup ||T j x|| < ∞.

You might also like