Академический Документы
Профессиональный Документы
Культура Документы
Contents
1 Topological Vector Spaces
1.1 What is functional analysis? . . . . . . . . .
1.2 The denition of topological vector spaces .
1.3 Basic properties of topological vector spaces
1.4 Hilbert spaces . . . . . . . . . . . . . . . . .
1.5 Banach spaces . . . . . . . . . . . . . . . . .
1.6 Frechet spaces . . . . . . . . . . . . . . . . .
1.6.1 Seminorms . . . . . . . . . . . . . . .
1.6.2 Frechet spaces . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
2 Linear Functionals
2.1 The Riesz representation theorem . . . . . . .
2.2 The Riesz Representation Theorem for Hilbert
2.3 The Hahn-Banach Theorems . . . . . . . . . .
2.4 A few results about convex sets . . . . . . . .
2.5 The dual of a topological vector space . . . . .
2.5.1 The weak -topology . . . . . . . . . . .
2.5.2 The dual of a normed vector space . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . .
spaces
. . . .
. . . .
. . . .
. . . .
. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
5
5
7
8
16
18
18
20
.
.
.
.
.
.
.
23
23
27
28
30
33
33
35
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
41
42
42
42
46
49
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
55
57
60
62
65
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
5 algebras
5.1 The denition of -algebras . . . . . . .
5.2 Commutative -algebras . . . . . . . .
5.3 -algebras as algebras of operators . . .
5.4 Functional calculus for normal operators
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
67
67
69
70
71
81
A Background results
A.1 Zorns lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
83
Chapter 1
Topological Vector Spaces
1.1
Functional analysis is the study of vector spaces endowed with a topology, and of the maps
between such spaces.
Linear algebra in innite dimensional spaces.
It is a eld of mathematics where linear algebra and geometry/topology meet.
Origins and applications
The study of spaces of functions (continuous, integrable) and of transformations between them (dierential operators, Fourier transform)
The study of dierential and integral equations (understanding the solution set).
Quantum mechanics (the Heisenberg formalism)
1.2
Denition. If and are vector spaces, a map : is called a linear map (or
linear operator) if for every scalars and and every vectors , ,
( + ) = + .
Denition. A topological space is a set together with a collection of subsets of with
the following properties
and are in
The union of arbitrarily many sets from is in
The intersection of nitely many sets from is in .
The sets in are called open.
If and are topological spaces, the is a topological space in a natural way, by
dening the open sets in to be arbitrary unions of sets of the form 1 2 where 1
is open in and 2 is open in .
Denition. A map : is called continuous if for every open set , the set
1 ( ) is open in .
A topological space is called Hausdor if for every , , there are open sets 1
and 2 such that 1 , 2 and 1 2 = .
A neighborhood of is an open set containing . A system of neighborhoods of is a
family of neighborhoods of such that for every neighborhood of there is a member of this
family inside it.
A subset is closed if is open. A subset is called compact if every
covering of by open sets has a nite subcover. The closure of a set is the smallest closed
set that contains it. If is a set then denotes its closure. The interior of a set is the
largest open set contained in it. We denote by () the interior of .
Denition. A topological vector space over the eld (which is either or ) is a vector
space endowed with a topology such that every point is closed and with the property that
both addition and scalar multiplication,
+ : and : ,
are continuous.
Example. is an example of a nite dimensional topological vector space, while ([0, 1])
is an example of an innite dimensional vector space.
A subset of a topological vector space is called bounded if for every neighborhood
of 0 there is a number > 0 such that for every > .
A topological vector space is called locally convex if every point has a system of neighborhoods that are convex.
1.3
1.4
Hilbert spaces
Example. The space ([0, 1]) of continuous functions : [0, 1] with the inner product
, =
()().
0
= , ,
and the distance between two elements is dened to be . Two elements, and , are
called orthogonal if
< , >= 0.
The norm completely determines the inner product by the polarization identity which in
the case of vector spaces over is
1
, = ( + 2 2 )
4
and in the case of vector spaces over is
1
, = ( + 2 2 + + 2 2 ).
4
Note that we also have the parallelogram identity
+ 2 + 2 = 22 + 22 .
Proposition 1.4.1. The norm induced by the inner product has the following properties:
a) = ,
b) (the Cauchy-Schwarz inequality) , ,
c) (the Minkowski inequality aka the triangle inequality) + + .
10
Proof. Part a) follows easily from the denition. For b), choose of absolute value 1 such
that , > 0. Let also be a real parameter. We have
0 2 = ,
= 2 2 (, + < , ) + 2
= 2 2 2 2 , + 2 .
As a quadratic function in this is always nonnegative, so its discriminant is nonpositive.
The discriminant is equal to
4( , 2 2 2 ),
and the fact that this is less than or equal to zero is equivalent to the Cauchy-Schwarz
inequality.
For c) we use the Cauchy-Schwarz inequality and compute
+ 2 = + , + = 2 + , + , + 2
2 + , + , + 2 = 2 + 2 , + 2
2 + 2 + 2 = ( + )2 .
Hence the conclusion.
Proposition 1.4.2. The vector space endowed with the inner product has a natural
topological vector space structure in which the open sets are arbitrary unions of balls of the
form
(, ) = { },
, > 0.
Proof. The continuity of addition follows from the triangle inequality. The continuity of the
scalar multiplication is straightforward.
Denition. A Hilbert space is a vector space endowed with an inner product, which is
complete, in the sense that if is a sequence of points in that satises the condition
0 for , , then there is an element such that 0.
We distinguish two types of convergence in a Hilbert space.
Denition. We say that converges strongly to if 0. We say that
converges weakly to if , , for all .
Using the Cauchy-Schwarz inequality, we see that strong convergence implies weak convergence.
Denition. The dimension of a Hilbert space is the smallest cardinal number of a set of
elements whose nite linear combinations are everywhere dense in the space.
We will only be concerned with Hilbert spaces of either nite or countable dimension.
11
Denition. An orthonormal basis for a Hilbert space is a set of unit vectors that are pairwise
orthogonal and such that the linear combinations of these elements are dense in the Hilbert
space.
Proposition 1.4.3. Every separable Hilbert space has an orthonormal basis.
Proof. Consider a countable dense set in the Hilbert space and apply the Gram-Schmidt
process to it.
From now on we will only be concerned with separable Hilbert spaces.
Theorem 1.4.1. If , 1 is an orthonormal basis of the Hilbert space , then:
a) Every element can be written uniquely as
=
,
, =
,
2 =
2 .
2 =
,
=1
=1
= ,
2
=1
=1
,
2
, +
=1
=1
=1
, +
=1
, 2 .
=1
, =
=1
=1
, 2 2 .
, 2 .
12
Note that
=1
, =
=1
, 2 ,
=1
, .
()().
0
, 2 2
1
[(2 1) ].
2 !
13
Example. The Hardy space on the unit disk 2 (). It consists of the holomorphic functions
on the unit disk for which
( 2
)1/2
1
2
sup
( )
2 0
0<<1
is nite. This quantity is the norm of 2 (); it comes from an inner product. An orthonormal
basis consists of the monomials 1, , 2 , 3 , . . ..
Example. The Segal-Bargmann space
2 (, )
which consists of the holomorphic functions on for which
2
()2 / <
2
1
()() / .
, = ()
,
!
= 0, 1, 2, . . . .
=1
2 < .
We set
< , >=
=1
14
Theorem 1.4.2. Every two Hilbert spaces (over the same eld of scalars) of the same
dimension are isometrically isomorphic.
Proof. Let ( ) and ( ) be orthonormal bases of the rst, respectively second space. The
map
preserves the norm. The uniqueness of writing an element in an orthonormal basis implies
that this map is linear.
Corollary 1.4.1. Every separable Hilbert space over is isometrically isomorphic to either
for some or to 2 (). Every separable Hilbert space over is isometrically isomorphic
to either for some or to 2 ().
Subspaces of a Hilbert space
Proposition 1.4.4. A nite dimensional subspace is closed.
Proof. Let be an -dimensional subspace. Using Gram-Schmidt, produce a basis
1 , 2 , 3 , . . . of such that 1 , 2 , . . . , are a basis for . Then every element in is of
the form
=
=1
and because convergence in norm implies the convergence of coecients, it follows that the
limit of a sequence of elements in is also a linear combination of 1 , 2 , . . . , , hence is in
.
However, if the Hilbert space is innite dimensional, then there are subspaces which are
not closed. For example if 1 , 2 , 3 , . . . is an orthonormal basis, then the linear combinations
of these basis elements dene a subspace which is dense, but not closed because it is not the
whole space.
Denition. We say that an element is orthogonal to a subspace if for every .
The orthogonal complement of a subspace is
= { , = 0 for all } .
Proposition 1.4.5. is a closed subspace of .
Proof. If , and , , then for all ,
+ , = , + , = 0,
which shows that is a subspace. The fact that it is closed follows from the fact that strong
convergence implies weak convergence.
15
Theorem 1.4.3. (The decomposition theorem) If is a closed subspace of the Hilbert space
, then every can be written uniquely as = + , where and .
Proof. (Following Riesz-Nagy) Consider as variable and consider the distances .
Let be their inmum, and a sequence such that
.
Now we use the parallelogram identity to write
( ) + ( )2 + ( ) ( )2 = 2 2 + 2 2 .
Using it we obtain
2 = 2( 2 + 2 ) 4
2( 2 + 2 ) 42 .
+ 2
The last expression converges to 0 when , . This implies that is Cauchy, hence
convergent. Let be its limit. Then = .
Set = . We will show that is orthogonal to . For this, let 0 be an arbitrary
element of . Then for every ,
2 = 2 0 2 = 2 , 0 0 , + 0 , 0 .
Set = , 0 / 0 , 0 to obtain
, 0 2
0.
0 2
(Adapt this proof to prove Cauchy-Schwarz!)
It follows that , 0 = 0, and so = .
If there are other , such that = + , then + = + so
= . This implies = = 0, hence = , = proving
uniqueness.
Corollary 1.4.2. If is a subspace of then ( ) = .
= = ( ) .
Hence cannot be a proper subspace of ( ) .
Exercise. Show that every nonempty closed convex subset of contains a unique element
of minimal norm.
16
1.5
Banach spaces
, (0, ).
Denition. A Banach space is a normed vector space that is complete, namely in which
every Cauchy sequence of elements converges.
Example. Every Hilbert space is a Banach space. In fact, the necessary structure for a
Banach space to have an underlying Hilbert space structure (prove it!) is that the norm
satises
+ 2 + 2 = 2(2 + 2 ).
Example. The space with the norm
(1 , 2 , . . . , ) = sup
is a Banach space.
Example. Let 1 be a real number. The space endowed with the norm
(1 , 2 , . . . , ) = (1 + 2 + + )1/
is a Banach space.
Example. The space ([0, 1]) of continuous functions on [0, 1] is a Banach space with the
norm
= sup ().
[0,1]
17
() = { :
() < },
()
)1/
is a Banach space. It is also separable. In general the space over any measurable space
is a Banach space.
The space () of functions that are bounded almost everywhere is also Banach. Here
two functions are identied if they coincide almost everywhere. The norm is dened by
= inf{ 0 () for almost every }.
The space is not separable.
Example. The Hardy space on the unit disk (). It consists of the holomorphic functions
on the unit disk for which
)1/
( 2
1
( )
sup
2 0
0<<1
is nite. This quantity is the norm of (). The Hardy space () is separable.
Also (), the space of bounded holomorphic functions on the unit disk with the sup
norm is a Banach space.
Example. Let be a domain in . Let also be a positive integer, and 1 < .
The Sobolev space , () is the space of all functions () such that for every
multi-index = (1 , 2 , . . . , ) with 1 + 2 + + , the weak partial derivative
belongs to ().
Here the weak partial derivative of is a function that satises
= (1)
,
, :=
.
The Sobolev spaces with 1 < are separable. However, for = , one denes the
norm to be
max ,
and in this case the Sobolev space is not separable.
18
1.6
Fr
echet spaces
1.6.1
Seminorms
() = inf{ > 0, 1 }.
1.6. FRECHET
SPACES
19
+
=
+
.
+
+
+
It follows that ( + ) + = () + () + 2. Now pass to the limit 0.
b) follows from the denition and c) follows from a) and b).
For d) note that the inclusions show that . For the converse
inequalities, let and choose , such that () < < . Then / so (/) 1
and (/) < 1. Hence / , so () . Vary to obtain .
A family of seminorms on a vector space is called separating if for every = , there
is a seminorm such that ( ) > 0.
Proposition 1.6.3. Suppose has a system of neighborhoods of 0 that are convex and
balanced. Associate so each open set in this system of neighborhoods its Minkowski
functional . Then = { () < 1}, and the family of functionals dened for
all such s is a separating family of continuous functionals.
Proof. If then / is still in for some > 1, so () < 1. If , then /
implies 1 because is balanced and convex. This proves that = { () < 1}.
By Proposition 1.6.2, is a seminorm for all . Applying Proposition 1.6.1 b) we have
that for every > 0 if then
() () ( ) < ,
which proves the continuity of at . Finally, is separating because is Hausdor.
Theorem 1.6.1. Suppose is a separating family of seminorms on a vector space .
Associate to each and to each positive integer the set
1/, = { () < 1/}.
Let be the set of all nite intersections of such sets. Then is a system of convex,
balanced, absorbing neighborhoods of 0, which denes a topology on and turns into a
topological vector space such that every is continuous and a set is bounded if and
only if is bounded for all .
Proof. Proposition 1.6.1 implies that each set 1/, is convex and balanced, and hence so are
the sets in . Consider all translates of sets in , and let the open sets be arbitrary unions
of such translates. We thus obtain a topology on . Because the family is separating,
the topology is Hausdor. We need to check that addition and scalar multiplication are
continuous.
Let be a neighborhood of 0 and let
1/1 ,1 1/2 ,2 1/ , .
Set
= 1/21 ,1 1/22 ,2 1/2 , .
20
1.6.2
Fr
echet spaces
1
.
(, ) =
1 +
2
=1
1.6. FRECHET
SPACES
21
Example. Let be an open subset of the complex plane. There is a sequence of compact
sets 1 2 3 whose union is . Let () be the space of holomorphic
functions on endowed with the seminorms
= sup{ () }.
Then () endowed with these seminorms is a Frechet space.
Theorem 1.6.2. A topological vector space has a norm that induces the topology if and
only if there is a convex bounded neighborhood of the origin.
Proof. If a norm exists, then the open unit ball centered at the origin is convex and bounded.
For the converse, assume is such a neighborhood. By Theorem 1.3.1, contains a
convex balanced neighborhood, which is also bounded. Let be the Minkowski functional
of this neighborhood. By Proposition 1.3.5, the , 0, is a family of neighborhoods of 0.
Moreover, because is bounded, for every there is > 0 such that . Then
so = 0 if and only if = 0. Thus is a norm and the topology is induced by this
norm.
22
Chapter 2
Linear Functionals
In this chapter we will look at linear functionals
: (or ),
where is a vector space.
2.1
This section is based on a series of lectures given by Hari Bercovici in 1990 in Perugia.
Problem: Given a sequence , 0, when does there exist a positive function such
that
()
=
for all 0?
We ask the more general problem, if there is a measure on such that 1 () for
all and
().
=
It is easy to see that not all such sequences are moments. Set
() =
.
=0
Then
() () =
,=0
+ ()
,=0
23
+ .
24
0
1 . . .
.
.
.
,
2
+1
+1 . . . 2
(2.1.1)
Theorem 2.1.1. (M. Riesz) Let be a linear space over and a cone, meaning
that if , and > 0 then + and . Assume moreover that the cone is
proper, meaning that () = {0} and dene the order if and only if .
Let be a subspace and let 0 : be a linear functional such that 0 () 0
for all . Supose that for every there is and > 0 such that
. Then there is a linear functional : such that = 0 and () 0
for all .
Proof. First, let us assume = + with . Suppose we are able to set () = .
If such that + then we should have () + () 0 so 0 () + 0. All
we need to show is that inf{0 () + } is not . Write = with
and . Then , so 0 () 0 (). We set
() = inf(0 () + }.
For the general case, apply Zorns Lemma. Consider the set of functionals :
such that , positive, and = 0 . Order it by
if and only if and = .
If ( ) is totally ordered, then = is a subspace, and = on for all is
a functional that is larger that all . Hence the conditions of Zorns Lemma are satised.
If : is a maximal functional, then = , for if but not in , then we can
extend to + as seen above.
Theorem 2.1.2. (F. Riesz) Let : ([0, 1]) be a positive linear functional. Then
there is a unique positive measure on [0, 1] such that
1
( ) =
()().
(2.1.2)
0
Proof. We use the theorem of M. Riesz. Let ([0, 1]) be the space of bounded functions on
[0, 1]. We have Set = ([0, 1]) and = ([0, 1]). The conditions of Theorem 2.1.1 are
satised, because every bounded function is the dierence between a continuous bounded
function and a positive function. Hence there is a positive linear functional : such
that ([0, 1]) = . Dene the monotone increasing function : [0, 1] such that
() = ([0,] ).
25
{0} +
( ,+1 ]) {0} +
( + )( ,+1 ]) .
({0} ) +
(( ,+1 ]) )) ( ) ({0} ) +
(( ,+1 ]) )) + (1).
(0) +
( (+1 ) ( )) ( ) (0) +
( (+1 ( )) + (1).
For those with more experience in measure theory, here is the general statement of this
result.
Theorem 2.1.3. Let be a compact space, in which the Borel sets are the -algebra
generated by open sets. Let : () be a positive linear functional. Then there is a
unique regular (positive) measure on such that
( ) =
.
(2.1.3)
Proof. Denote by [] the real valued polynomial functions on and by () the continuous functions with compact support. Consider
= [] + (),
= [],
and
= { () 0 for all }.
For a polynomial () =
=0
, let
0 () =
=0
26
The conditions of Theorem 2.1.1 are veried. Then there is a linear positive functional
: such that = 0 . By Theorem 2.1.2, on every interval [, ], 1
there
is a measure such that if is continuous with the support in [, ], then ( ) =
() (). Uniqueness implies that for 1 > 2 , 1 [2 , 2 ] = 2 . Hence we can
The fact that is a nite measure is proved as follows. Given an interval [, ], let
be compactly supported, such that
[,] 1.
Then
() =
()().
If is an even degree polynomial with positive dominant coecient, then it can be approximated from below by compactly supported continuous functions, and so using the positivity
of we conclude that for every such function
()().
() ( ) =
()
()().
Let be a polynomial of even degree with dominant coecient positive, whose degree is less
than . Then for every > 0,
( )()().
( )
Said dierently
()
This can only happen if
()() ()
() =
()() .
()().
Varying and we conclude that this is true for every with even degree and with positive
dominant coecient. Since every polynomial can be written as the dierence between two
such polynomials, the property is true for all polynomials.
2.2
27
Theorem 2.2.1. (The Riesz Representation Theorem) Let be a Hilbert space and let
: be a continuous linear functional. Then there is such that
() = , , for all .
Proof. Let us assume that is not identically equal to zero, for otherwise we can choose
= 0.
Because is continuous, Ker = 1 (0) is closed. Let = Ker . Then is one
dimensional, because if 1 , 2 were linearly independent in , then ((2 )1 (1 )2 ) = 0,
but (2 )1 (1 )2 is a nonzero vector orthogonal to the kernel of . Let be a nonzero
vector in , so that () = 0. Replace by = /(). Let = / 2 . Then
() = 1/ 2 = , .
Every vector can be written uniquely as = + with Ker and a scalar.
Then
() = ( + ) = () = ,
= + , = , .
Example. If : 2 () 7 is a continuous linear functional, then there is an 2 function
such that
()(),
( ) =
Example. Consider the Hardy space 2 (), and the linear functionals ( ) = (), .
Then for all , is continuous, and so there is a function () 2 () such that
() = , .
The function (, ) 7 () is called the reproducting kernel of the Hardy space.
() =
() () .
28
2.3
() = ()
29
30
Theorem 2.3.4. (Hahn-Banach) Suppose and are disjoint, nonempty, convex sets in
a locally convex topological vector space .
a) If is open then there is a continuous linear functional on and such that
Re() < Re() for all , .
b) If is compact and is closed then there is a continuous linear functional and 1 , 2
such that
Re() 1 < 2 Re()
for all and .
Proof. We just consider the real case. Because is open, every point of is internal. So
there is and such that { () } and { () }. Let 0 be a point in
. Then ( 0 ) = () (0 ) (0 ). So there is an open neighborhood of 0
such that () if , where = (0 ).. Choosing to be balanced, we conclude
that () for all . But this is the condition that is continuous.
We claim that because is open () < for all . If not, let be such that
() = . Given , there is a small (0, 1) such that ( )/(1 ) (because of
the continuity of addition and multiplication). Call this point . Then = + (1 ), and
because () and () are both less than or equal to , () = () = () = . Hence
is constant in a neighborhood of . Consequently is constant in a neighborhood of 0, and
because the neighborhoods of 0 are absorbing, it is constant everywhere. This is impossible.
Hence a) is true.
For b) we use Proposition 1.3.3 to conclude that there is an open set such that +
+ = . By shrinking, we can make balanced and convex. Let 0 {0} such
that (0 ) = = 0. Then
sup () + sup (),
2.4
For a subset of a vector space , we denote by Ext() the extremal points of , namely
the points which cannot be written as = + (1 ) with , {}, (0, 1).
This denition can be extended from points to sets by saying that a subset of is extermal
if for , and (0, 1) such that +(1) , it automatically follows that , .
Note that a point is extemal if an only if {} is an extremal set.
Also, for a subset of the vector space , we denote by co() the convex hull of ,
namely the convex set consisting of all points of the form + (1 ) where , and
[0, 1].
Theorem 2.4.1. (Krein-Milman) Suppose is a locally convex topological vector space,
and let be a subset of that is compact and convex. Then
= co(Ext()).
31
It is not hard to see that 0 is also extremal. Hence we are in the conditions of Zorns
Lemma. We deduce that has minimal elements.
Let be a minimal element; we claim it is a singleton. Arguing by contradiction,
let us assume that has two distinct points. By the Hahn-Banach Theorem there is a
continuous linear functional : such that () = (). Let
= max ().
Dene
1 = { 0 () = }.
Then 1 is a nonempty extremal subset of and consequently an extremal subset of .
It is also compact and convex, which contradicts the minimality of . Hence contains
only one point. This proves
Ext() = .
Let = co(Ext()). Note that is compact. Assume = . Then there is .
By the Hahn-Banach theorem, there is a continuous linear functional : such that
max () < (). Set
2 = { () = max ()}.
32
To prove this, let = {(1 , 2 , . . . , ) [0, 1] = 1}, and consider the function
: 1 2 . . . ,
(1 , . . . , , 1 , . . . , ) =
.
33
Proof. Let
= { : nonempty, compact, convex , ( ) for all }.
Note that , so this family is nonempty. Order by inclusion and note that if is a
subfamily that is totally ordered, then because is compact,
0 = = .
Clearly (0 ) 0 , so the conditions of Zorns lemma are satised. It follows that has
minimal elements. Let 0 be such a minimal element. We claim that it is a singleton.
Assume, to the contrary, that 0 contains , with = . Let be a neighborhood of
0 such that , and let be the neighborhood of 0 associated to by the denition
of equicontinuity. Then for every , () () , for else, because 1 ,
= 1 ( ()) 1 ( ()) .
Set = 12 ( + ). Then 0 . Let
() = { () }.
the () is -invariant, hence so is its closure 1 = (). Consequently, co(1 ) is a
-invariant, compact convex subset of 0 . The minimality of 0 implies
0 = co(1 ).
By the Krein-Milman Theorem (Theorem 2.4.1), 0 has extremal points. Applying Milmans
Theorem (Theorem 2.4.2), we deduce that every extremal point of 0 lies in 1 . Let be
such a point.
Consider the set
= {( , , ) } 0 0 0 .
Since 1 = (), and 0 0 is compact, there is a point (1 , 1 ) 0 0
such that (0 , 1 , 1 ) . Indeed, if this were not true, then every (1 , 1 ) 0 0
had a neighborhood (1 ,1 ) and there would exist a neighborhood (1 ,1 ) of such that
(1 ,1 ) (1 ,1 ) = . Then 0 0 is covered by nitely many of the (1 ,1 ) and the
intersection of the corresponding (1 , 1 )s is a neighborhood of that does not intersect
1 .
Because 2 = + for all , we get 20 = 1 + 1 , hence 0 = 1 = 1 , because 0
is an extremal point. But for all , hence 1 1 , and so 1 = 1 .
This is a contradiction, which proves our initial assumption was false, and the conclusion
follows.
2.5
2.5.1
34
= 1, 2, . . . , },
= (
) ,
35
The set
:
, () = (()) .
We will show
(1) () is closed.
(2) : () is a homeomorphism.
For (1) assume that ( ) is in (). Dene () = 1 for such that .
Approximating ( ) with linear functionals (), = 1, 2, . . . , . We have
( + ) = () + ().
For large enough, ( + ) approximates well (), while () and () approximate
() and (). By passing to the limit we obtain that is linear.
Also for , () 1, and again by passing to the limit, () 1. This implies
the continuity of , as well as the fact that it lies in (). This proves (1).
For (2), note that is one-to-one, hence it is an inclusion. Moreover, the weak topology
was chosen so that it coincides with the topology induced by the product topology. Hence
(2).
2.5.2
If is a normed vector space, then is also a normed space with the norm
= sup{() 1}.
Proposition 2.5.2. The dual of a normed space is a Banach space.
Proof. The only dicult part is to show that is Banach. Let , 1 be a Cauchy
sequence in . Dene () = lim (). It is not hard to check that is linear. On
the other hand, because is Cauchy, is Cauchy as well, by the triangle inequality.
For a given , if we choose large enough then
() () < ,
so
() < ( + 1).
Because , 1 is a bounded sequence (being Cauchy), it follows that is a bounded
linear functional, and we are done.
36
So has two topologies the one induced by the norm, and the weak topology. It is
not hard to check that the second is weaker than the rst.
Here is an example.
Theorem 2.5.3. Let [1, ). Then ( ([0, 1])) = ([0, 1]), where satises 1/+1/ =
1. A function ([0, 1]) denes a functional by
( ) =
()().
0
Proof. Note that every ([0, 1]) denes a continuous linear functional by the above
formula because of Holders inequality:
1 .
Moreover, it is not hard to see that if 1 and 2 dene the same functional then 1 = 2
almost everywhere. This follows from the fact that if
1
()[1 () 2 ()] = 0
0
Approximating the functions in ([0, 1]) by step functions, and using the continuity of both
the left-hand side on ([0, 1]) and of the right-hand side on ([0, 1]) ([0, 1]), we
deduce that
1
()() for all ([0, 1]).
( ) =
0
37
1
Case 2. > 1. We want to show that if 0 ()() is nite for all ([0, 1]), then
([0, 1]). By multiplying by /, we can make be positive, so let us consider just
this case.
Let be a step function that approximates from below, 0. Consider
1
1
1
1/
1/
1/
1/
( ) =
() ()
() () =
().
0
() 1/ .
(2.5.1)
We also have
1/
()
0
]1/
1
0
()
()
0
)1/
()
0
)1/
( ) =
.
Moreover = .
Theorem 2.5.5. (([0, 1])) is the set of nite complex valued measures on [0, 1].
Proof. Each nite measure denes a continuous linear functional by
1
().
( ) =
0
Let us prove conversely, that every linear functional is of this form. For every complex
continuous linear functional , we have = Re + Im where the real and the imaginary
part are themselves continuous. So we reduce the problem to real functionals. We show that
38
each such functional is the dierence between two positive functionals, and then apply the
Riesz Representation Theorem.
For 0, set
+ ( ) = sup{() ([0, 1]), 0 }.
Because is continuous, hence bounded, + takes nite values. Since = 0 , and
(0) = 0, we have that + is positive.
It is clear that + ( ) = + ( ), for 0, 0. Also, since = 0 , and (0) = 0,
we have that + is positivie. It is also clear that + (1 + 2 ) + (1 ) + + (2 ) because we
can use for 1 + 2 the 1 + 2 .
On the other hand, if = 1 + 2 , and , set 1 = sup( 2 , 0) and 2 = inf(, 2 ).
Then 1 + 2 = , and 0 , = 1, 2. Hence
() = (1 ) + (2 ) + (1 ) + + (2 ).
Consequently + ( ) = + (1 + 2 ) + (1 ) + + (2 ). Therefore we must have equality.
This holds of course for 1 , 2 0.
For arbitrary , write = 1 2 , where 1 , 2 0, and dene + ( ) = + (1 ) + (2 ).
It is not hard to see that + is well dened, linear, and positive. Also + is a linear
positive functional. We have
= + (+ ),
and the claim is proved. We can therefore write every continuous complex linear functional
as
= 1 2 + (3 4 ),
where , = 1, 2, 3, 4 are positive. Each of these is given by a positive measure , by the
Riesz representation Theorem, so is given by the complex measure
= 1 2 + (3 4 ).
Remark 2.5.1. Using the general form of the Riesz Representation Theorem, we see that
[0, 1] can be replaced by any compact space.
Theorem 2.5.6. (Banach-Alaoglu) Let be a normed vector space. Then the closed unit
ball in is weak -compact.
Here are some applications.
Proposition 2.5.3. Place a number in [0, 1] at each node of the lattice 2 such that the
number at each node is the average of the four numbers at the closest nodes. Then all
numbers are equal.
39
Proof. Consider the Banach space (2 ). Let be the set of elements in (2 ) satisfying
the condition from the statement. Then is a weak -closed subset of the unit ball; by
applying the Banach-Alaoglu theorem we deduce that it is weak -compact. It is also convex.
By the Krein-Milman Theorem, = co(Ext()). Let : 2 [0, 1] be an extremal point in
. Let , be the operators that shift up and left. Then , 1 , , 1 are functions
with the same property, and
1
= ( + 1 + + 1 ).
4
Because is extremal, = = 1 = = 1 , meaning that is constant. In
fact = 0 or = 1. The convex hull of the two extremal constant functions is the set of
all constant functions with values in [0, 1], this set is closed, so consists only of constant
functions. Done.
Theorem 2.5.7. The space 1 () is not the dual of any normed space.
Proof. If 1 () were the dual of a normed space, then the Banach-Alaoglu Theorem implies
that the closed unit ball in 1 () is weak -compact. By the Krein-Milman Theorem it has
extreme points. But this is not true, since every function in the closed unit ball can be
written as the convex combination of two functions in the unit ball.
Consider ([0, 1]), the Banach space of continuous functions on [0, 1], with the norm
= sup[0,1] ().
Theorem 2.5.8. (Stone-Weierstrass) Let ([0, 1]) be a subalgebra with the following
properties
(1) if then ,
(2) the function identically equal to 1 is in ,
(3) separates the points of [0, 1].
Then is dense in ([0, 1]),
Proof. (de Brange) We argue by contradiction. Let
= { ([0, 1]) 1, = 0}.
By the Banach-Alaoglu Theorem, it is compact in the weak topology. is also convex, so
by the Krein-Milman Theorem it has extremal points. Moreover, the Hahn-Banach Theorem
implies that = {0}, so then Krein-Milman implies the existence of at least two extremal
points. This means that there is an extremal functional that is not identically equal
to zero.
Because ([0, 1]) is the space of nite complex measures (Theorem 2.5.5), is given by
a measure . We claim that every function in is constant on the support of . If this is so
then because the functions in separate points, the support of consists of just one point,
so = 0 for 0 [0, 1] and . Because = 0, and 1 , we get that = 0, a
contradiction. Hence the conclusion.
40
Let us prove the claim. Suppose there is not constant on the support of . We
have = 1 + 2 , so 1 = ( + )/2 and 2 = ( )/2, and because , 1 , 2 as
well. One of these is nonconstant, so we may assume that is real valued. Replacing by
( + )/ we may assume 0 1. Dene the measures 1 and 2 by
1 = ,
2 = (1 ).
Then = 1 + 2 . Note
1both zero on .
1 that 1 , 2 are
We have 1 = 0 , 2 = 0 (1 ). And also = = 1 + 2 .
Then
)
)
(
(
2
1
1 +
2 .
=
1
2
Note that
1 ,
2
1
2
Because is an extreme point, either 1 or 2 is zero. So must be identically equal to 1,
a contradiction. The claim is proved, and so it the theorem.
Remark 2.5.2. Using the general form of the Riesz Representation Theorem, we see that
[0, 1] can be replaced by any compact space.
Chapter 3
Fundamental Results about Bounded
Linear Operators
3.1
3.1.1
We now start looking at continuous linear operators between topological vector spaces:
: .
Proposition 3.1.1. Let : be a linear operator between topological vector spaces
that is continuous at 0. Then is continuous everywhere, moreover, for every open neighborhood of 0 there is an open neighborhood of 0 such that if then .
Denition. A linear operator is called bounded if it maps bounded sets to bounded sets.
Proposition 3.1.2. A continuous linear operator is bounded.
Proof. Let : be a continuous linear operator. Consider a bounded set . Let
also be a neighborhood of 0 in . Because is continous, there is a neighborhood of 0 in
such that ( ) . Choose {0} such that . Then () = () .
It follows that () is bounded.
Denition. Let : be a linear operator. The kernel of is
ker( ) = { = 0}.
The range or image of is
im( ) = { there is with = }.
Both ker( ) and im( ) are vector spaces. If is a continuous linear operator between
topological vector spaces, then ker( ) is closed. This is not necessarily true about im( ).
41
3.2
3.2.1
3.2.2
From now on we will focus just on continuous linear operators between Banach spaces.
Theorem 3.2.2. Let : be a linear operator. Then is continuous if and only if
it is bounded.
Proof.
Denition. Let be a bounded linear operator. The norm of is
= sup{ 1}.
Theorem 3.2.3. (Banach-Steinhaus) Let be a Banach space, a normed space, and
let be a family of continuous operators from to . Suppose that for all ,
sup < . Then sup < .
Proof. Let
= { for all }.
These sets are convex and balanced. They are also closed, so by the Baire Category Theorem
is such that the interior of is nonempty. Because is convex and balanced, its interior
contains the origin. Hence there is a ball 0, centered at origin such that for
all and with . We have / for all , and the theorem is
proved.
Here is an application that I learned from Hari Bercovici. We have
1
=
(1)1 1 .
+ 1 =1
43
The left-hand side takes the value 1/2 when = 1, so it is natural to impose that the
right-hand side converges to 1/2. A way to do this is to consider the sequence =
1 1
(3.2.1)
converges to the same limit as when the latter converges (Cesaro), but moreover for = 1
(3.2.1) converges to 1/2.
Denition. A summation method associates to each convergent sequence , 1 another
convergent sequence , 1 such that
(1) lim = lim ;
(2) =
=1 for = 1, 2, . . ., where is an array of complex numbers that does
not depend on and denes the summation method.
An example of a summation method, introduced by Cesaro, is = 1/, = 1, 2, . . .,
1 , and = 0 otherwise.
Theorem 3.2.4. (Toeplitz) The array , , 1, denes a summation method if and
only if it satises the following three conditions
(1) lim = 0, = 1, 2, . . .;
(2) lim
=1 = 1;
(3) sup
=1 < .
Proof. Let us prove that the three conditions are necessary. If = for some , then
= . The fact that 0implies lim = 0, hence (1).
If = 1, 1, then =
=1 . Because 1, it follows that lim
=
1, hence (2).
For (3) we apply the Banach-Steinhaus Theorem. Denote by 0 the Banach space of
convergent sequences with the sup norm (i.e. continuous functions on {} with the sup
norm, where {} is given the topology such that the map () = 1/
from it to is
a homeomorphism onto the image). Let , 1 be a sequence
such that
=1 converges
0
and
=
sup
=
.
=1
{ }0 , 1
The fact that the left-hand side does not exceed the right-hand side follows from the triangle
inequality. On the other hand, if = / , 1 and zero otherwise makes
=
=1 . Taking we obtain that the right-hand side is less than or
equal to the left-hand side. Hence the two are equal.
Dene
: 0 ,
({ }) =
=1
Then (0 ) and
=
=1
The sequence ({ }), 1 is bounded for all convergent sequences { }. Hence by the
Banach-Steinhaus Theorem, , 1 is bounded, which is (3).
Now let us check that the conditions are sucient. Let = sup
=1 . Consider
a sequence converging to . We want to show that converges to as well. We compute
(
)
+
1
=1
=1
=1
+
1
=1
=1
+ sup +
1 .
+1
=1
=1
We obtain lim = 0.
45
1 (, ) =
and
2 : ,
2 (, ) = .
Both these operators are linear and continuous. The operator 1 is invertible and bijective.
By the Inverse Mapping Theorem (Theorem 3.2.6) its inverse is also continuous. We have
= 2 11 , and hence is invertible.
Here is an application.
3.3
47
Corollary 3.3.1. ker( ) is weak closed, im( ) is dense if and only if is injective, and
is injective if and only if im( ) is weak dense.
Theorem 3.3.2. Let : be a bounded linear operator between Banach spaces.
The following conditions are equivalent:
(a) im( ) is closed in ;
(b) im( ) is weak closed in ;
(c) im( ) is norm closed in .
49
Hence 1. We have
< (0 ) 0 0 0 .
We deduce that if then necessarily ( ).
Now let us show that moreover ( ). Rescaling we may assume = 1. Then
( ), and hence for every and every > 0 there is such that
and < . Choose 1 . Let = 31 (1 1 ). Dene the sequences and
inductively as follows. Assume is already picked, and let be such that
and
< . Set +1
.
=
If =
, the =
= ( +1 ) = 1 . Hence 1 ( ). This proves
our claim. The conclusion follows.
Using the lemma we conclude that im() = im(), and so the image of is closed. But
im() = im( ), and so the theorem is proved.
Theorem 3.3.3. Let : be a bounded linear operator between Banach spaces.
Then im( ) = if and only if is one-to-one and im( ) is norm closed.
Proof. By Proposition 3.3.1 is one-to-one. By the Open Mapping Theorem there is
> 0 such that
{ } { () 1}.
Then for a functional ,
= sup{( )() 1} = sup{( ) 1}
sup{() } = .
We claim that given this inequality, im( ) is closed. Let im( ), and let be such
that . Then is Cauchy, so the above inequality implies is Cauchy. If is
its limit, then = . The implication is proved.
By Theorem 3.3.2, im( ) is closed, and by Proposition 3.3.1 it is dense. Hence
im( ) = .
3.4
Let be a Hilbert space over and let : be a bounded linear operator. There is
a dierent construction of based on the Riesz representation theorem. Recall that there is
an antilinear isometry between and which associates to each functional the element
such that () = , .
The linear operator 7 induces a linear operator 7 . Moreover, the two
operators have the same norm. We will use the notation for the second. A direct way to
dene this operator is by the equality
, = , .
(3.4.1)
Because the adjoint is dened using the inner product, we will use the following lemma
several times.
51
2 3
+
1!
2!
3!
is well dened because the series is absolutely convergent hence convergent. We have
exp() = exp(),
and because and commute,
exp() exp() = exp() exp() = .
It follows that exp() is unitary.
Corollary 3.4.1. If is a bounded operator, then exp[( + )] and exp( ) are
unitary.
Proof. We have ( + ) = + , and [( )/] = ( )/.
53
, () = (),
Chapter 4
Banach Algebra Techniques in
Operator Theory
4.1
Banach algebras
This section and the next follow closely R.G. Douglas, Banach Algebra Techniques in Operator Theory, Academic Press 1972 with some input from Rudins Functional Analysis.
Denition. A Banach algebra is an associative algebra with unit 1 over the complex (or
real) numbers that is at the same time a Banach space, and so that the norm satises
, and 1 = 1.
Example. The Banach algebra () of bounded linear operators on a Banach space .
Example. The Banach algebra of continuous functions ([0, 1]).
We will almost always be concerned with Banach algebras over the complex numbers.
Denition. A series
=0
=0
<
1
.
1 1
55
56
= 1 lim = 1.
Hence
1 = (1 )1 = 1 + + 2 + 3 + .
By the triangle inequality
1 1 + + 2 + 3 + =
1
1
=
.
1
1 1
Denition. For a Banach algebra , let , , and be respectively the sets of invertible
elements, right invertible elements that are not invertible, and left invertible elements that
are not invertible.
Proposition 4.1.2. If is a Banach algebra, then each of the sets , , and is open.
Proof. If is invertible, and
1
,
1
then
1 1 1 < 1.
Hence 1 1 is invertible, and so is (1 1 ) = . This proves that for every
there is a ball of radius 1/1 centered at and contained in . Hence is open.
By the same argument, if and is such that = 1, then if is such that
< 1/ then is invertible. We have (()1 ) = 1, showing that is left invertible.
Note that itself cannot be invertible, or else , being close to it would be invertible too.
This proves open. The proof that is open is similar.
Proposition 4.1.3. If is a Banach algebra and is the subgroup of invertible elements,
then the map
7 1
is continuous.
57
Proof. Fix . We want to show that for every > 0, there is > 0 such that if
< then 1 1 < . We have
1 1 = 1 ( )1 1 1 .
So we have to choose so that 1 < /1 for all such that < . If
< 1/(21 ), then 1 1 < 1/2 and so by Theorem 4.1.1
1 1 1 = (1 )1 1
1
1 = 21 .
1 12
,
1
2 21 2
4.2
58
() = (( )1 ).
Hence
lim () = lim
limsup
1
.
1 /
1
(/ 1)1
1
1
1
(/ 1)1 limsup
1 /
where for the last step we used Theorem 4.1.1. This last limit is zero. Hence is a bounded
holomorphic function. By Liouvilles Theorem it is constant.
Using the Hahn-Banach Theorem we deduce that 7 ( )1 is constant, and since
the inverse is unique, it follows that 7 is constant. But this is clearly not true.
Hence our assumption was false, and the spectrum is nonempty.
Since the map ( ) is continuous, and (the set of invertible elements) is open,
the inverse image of through this map is open. But the inverse image of is the resolvent.
Hence the resolvent is open, and therefore the spectrum is closed. Being bounded (as it lies
inside the disk of radius ), it is compact.
In view of this theorem we dene the spectral radius to be
() = sup{ ()}.
59
Proof. Fix an element and let > . Then using Theorem 4.1.1 we can write
( )1 = 1 + 2 + 3 2 + .
The series converges absolutely on every circle (0, ) centered at the origin and radius
> . We can therefore integrate term by term and write
=
( )1 , = 1, 2, 3, . . .
(4.2.1)
2 (0,)
Let be a continuous linear functional. Then as we saw before (( )1 ) is holomorphic.
From (4.2.1) we deduce
(( )1 ).
( ) =
2 (0,)
The right-hand side is an integral of a holomorphic function, and so by Cauchys theorem
the equality also holds true for all circles for which (( )1 ) is dened. Thus the equality
holds for (). Because the Hahn-Banach theorem we can conclude that
( )1 , for 0, ().
=
2 (0,)
Let () be the maximum of ()1 on (0, ), (which is nite because 7 ()1
is continuous). Then
+1 (),
0, ().
().
Hence
() lim sup 1/ .
60
4.3
Let . Then () is a compact subset of the plane. Consider a domain that contains
(), and let be a smooth oriented contour (maybe made out of several curves) that does
not cross itself such that () is surrounded by and such that travels around () in the
counterclockwise direction.
For a holomorphic function in , we have the Cauchy formula
1
(0 ) =
()( 0 )1 .
2
Now let us replace 0 by . Then on , the element ( )1 is dened. With Cauchys
formula in mind, we can dene
1
() =
()( )1 .
(4.3.1)
2
Lemma 4.3.1. The operator () is well dened and does not depend on the contour .
1
1
1
1
()(( ) ) =
()( )
2
2
does not depend on . So () itself does not depend on .
() =
,
since again the series converges. The integral formula (4.3.1) would be meaningful only if in
this particular situation the two versions coincide. And indeed, we have the following result.
61
()( )1
=
2
Proof. Choose large enough so that > and sup > are as small as
we wish. Then we can ignore these sums and consider just the case where () =
=0 .
To prove the result in this case, it suces to check it for a power of . Thus let us show
that
=
( )1 .
2
Now we can rely on Cauchys theorem about the integral of a holomorphic function, to to
make a circle of radius greater than . Because on > , we can expand
( )1 =
/ +1 .
0
The series on the right is absolutely convergent, so we can integrate term-by-term to write
(
)
1
1
1
1
( ) =
.
2
2
=0
All of the integrals are zero, except for the one where = , which is equal to 2. Hence
the result is , as desired.
A slight modication of the proof yields the following more general result.
62
Theorem 4.3.2. Let be a domain in that contains (). Endow the space of holomorphic functions on , (), with the topology of uniform convergence on compact subsets.
Then the map () , 7 () is a continuous algebra homomorphism.
Proof. The only dicult step is multiplicativity. But we have multiplicativity for polynomials, and hence for rational functions. By Runges theorem, every function in () is the
limit of rational functions. By passing to the limit in () () = ( )(), we conclude
that multiplicativity holds in general.
Theorem 4.3.3. (The Spectral Mapping Theorem for Holomorphic Functions) Let be a
holomorphic function in a neighborhood of the spectrum of . Then
( ()) = (()).
Proof. Let (). Then as before () () = ( )() with a holomorphic function
with the same domain as . By the previous theorem
() () = ( )(),
so () () is not invertible. Hence (()) ( ()).
For the opposite inclusion, let ( ()). If () is nowhere zero on the spectrum
of , then () = ( () )1 is dened on the spectrum of , and then
( () )( )1 () = 1
which cannot happen. So () is zero for some (), that is (()).
4.4
In this section we will construct a Banach algebra which is not the algebra of bounded linear
operators on a Banach space. For this we introduce the notion of a compact operator.
Denition. Let be a Banach space. An operator () is called compact if the
closure of the image of the unit ball is compact.
Example. If is such that im() is nite dimensional, then is compact. Such an operator
is said to be of nite rank.
Theorem 4.4.1. The set () of compact linear operators on is a closed two-sided ideal
of ().
Proof. Let 1 and 2 be compact operators. Then 1 (0,1 ) and 2 (0,1 ) are compact.
Then
(1 + 2 )(0,1 ) 1 (0,1 ) + 2 (0,1 )
and the latter is compact because is the image through the continuous map (, ) 7 +
of the compact set 1 (0,1 2 (0,1 ) . This proves that 1 + 2 is compact.
63
Also for every , if is compact then is compact, because the image of the set
1 (0,1 ) through the continuous map 7 is compact.
Finally, if () and () then ((0,1 )) is the image of a compact set
through a continuous map, so it is compact. It follows that (0,1 ) lies inside a compact
set, so its closure is compact. So is compact.
On the other hand, (0,1 ) is a subset of 0, for some , so (0,1 ) is a closed subset
of the compact set (0, ), hence is compact. This proves that is compact.
We thus showed that () is an ideal. Let us prove that it is closed. Let , 1,
be a sequence of compact operators that is norm convergent to an operator . We want
to prove that is compact. For this we use the characterization of compactness in metric
spaces: Every sequence contains a convergent subsequence.
Let , 1 be a sequence of points in the unit ball of . Let us examine the sequence
, 1. For every > 0, there is () such that for (),
. For (),
+ + 2 + .
The sequence has a convergent subsequence, and so we can nd a subsequence
such that < 3 for all , . Do this for = 1, then choose the rst term of a
sequence to be 1 . Inductively let = 1/, and choose from the previous sequence
a subseqence such that < 3 and let be the rst term of this subsequence.
The result is a Cauchy sequence , which therefore converges. We conclude that is
compact.
Theorem 4.4.2. Let () be a compact operator. Then
(a) If im() is closed, then dimim() < .
(b) If = 0, then dimker( ) < .
(c) If dim = , then ().
Proof. (a) If im() is closed, then it is a Banach space. The Open Mapping Theorem
implies that the image of the unit ball is a neighborhood of the origin. This neighborhood
is compact, and this only happens if im() is nite dimensional.
(b) The operator ker( ) is a multiple of the identity operator. This operator is
also compact. By (a) this can only happen if we are in a nite dimensional situation.
(c) The operator cannot be onto.
Theorem 4.4.3. Let be a Banach algebra and let be a two-sided closed ideal. Then
/ is a Banach algebra with the norm
[] = inf{ + }.
Here we denote by [] the image of under the quotient map.
64
Also
[] + [] = [ + ] = inf{ + + } = inf{ + + + , }
inf{ + } + inf{ + } = [] + [].
Thus is a norm.
Next, let us show that the norm satises the requirements from the denition of a Banach
algebra. First,
[1] = inf{1 } = 1,
where the equality is attained for = 0, and one cannot have 1 < 1 for in that case
must be invertible and hence cannot be an element of an ideal.
Secondly, for , , we have
[][] = [] = inf{ } inf{( 1 )( 2 ) 1 , 2 }
inf{ 1 1 } inf{ 2 2 } = [][].
Finally, let us show that / is complete. Showing that every Cauchy sequence is
convergent is equivalent to showing that every absolutely convergent series is convergent. It
is clear that the fact that every Cauchy sequence is convergent implies that every absolutely
convergent series is convergent. For the converse, let , 1 be a Cauchy sequence. By
choosing a subsequence,
we may assume that 1/2 whenever , . Set
= +1
. Then
is absolutelyconvergent, and its sum is the limit of .
So let [ ] be a series such that [
] = < . Then for each there is
Denition. An operator with nite dimensional kernel and with closed image of nite
codimension is called Fredholm.
Theorem 4.4.4. (Atkinson) Let be a Hilbert space. Then the Fredholm operators form
the preimage through the quotient map of the invertible elements of ()/().
4.5
65
66
not in the closure of . Thus the closure of is an ideal, and because of maximality, this
ideal must be . So is closed.
The quotient algebra / is a division algebra, because is maximal. So by the
Gelfand-Mazur Theorem it is . The quotient map is the desired multiplicative functional.
Recall that for every , the function
: ( )1 given by
() = () is continuous,
Chapter 5
algebras
5.1
(5.1.1)
(5.1.2)
The two conditions (5.1.1) and (5.1.2) are equivalent, though it is hard to show that
(5.1.1) implies (5.1.2). Thus our working denition will be the one with (5.1.2), what is
usually called a -algebra. This condition implies (5.1.1) as follows:
2 = .
Hence and ( ) = . So = . Then = 2 =
. From these calculations we conclude that in a algebra the involution is an
isometry.
Example. The algebra () of bounded linear operators on a Hilbert space with the involution dened by taking the adjoint.
67
CHAPTER 5. ALGEBRAS
68
Theorem 5.1.1. In a -algebra the spectrum of a unitary element is contained in the unit
circle, and the spectrum of a self-adjoint element is contained in the real axis.
Proof. If is unitary, then 1 = 1 = = 2 , so = = 1 = 1. Then
if > 1, then is invertible. Also, if < 1, then 1 1 is invertible, and so is
(1 1 ) = . Hence
() { < 1}.
Let be the -algebra. If is self-adjoint, then = exp() is unitary. Indeed,
= exp() = exp(), and
= exp() exp() = exp( ) = 1 = .
Because () is a subset of the unit disk, and, by the Spectral Mapping Theorem, () =
exp(()), the spectrum of must be real.
5.2
69
Commutative -algebras
() = ( )().
CHAPTER 5. ALGEBRAS
70
Hence and coincide on , so they are equal.
Finally, let us show that is continuous. Let
0 , = { ( ) 0 < }.
Set 0 = 1 (0 ). Then
1 (0 , ) = { ( ) 0 ( ) < },
which is open in the weak topology. Hence is continuous.
Because and ( ) are compact Hausdor spaces, is a homeomorphism.
5.3
for all .
71
Then / is an inner product space. Consider the completion of this space, which
is therefore a Hilbert space. We have
2 = (2 ) = ,
where
= = (2 )1/2 = ( )1/2 .
The element can by dened because the function () = ( )1/2 is continuous on
( ), so we can use Theorem 5.2.2. So, because is positive,
( ) (2 ) = 2 ( ).
It follows that
( + ) 2 + ,
so () is continuous. This implies that () can be extended to the entire Hilbert space
. This representation is cyclic, with cyclic vector 1 + . Also, ()1, 1 = (1 1) =
().
The set of states is a weak closed convex subset of the unit ball of . The extremal
points are called pure states.
Theorem 5.3.2. (Gelfand-Naimark) Every -algebra admits an isometric -representation.
If the -algebra is separable, then the Hilbert space can be chosen to be separable as well.
Proof. Consider the set of pure states and dene
: ( ),
where the sum is taken over all pure states. It suces to show that is faithful, namely
one-to-one, because the fact that it is an isometric -homomorphsm then follows from Theorem 5.4.3 proved in next section. For this we use the theorem of M. Riesz about extension
of positive functionals. Let be a nonzero element of . Then there is a state such that
( ) > 0.
Consider the GNS representation associated to , and let be its cyclic vector. Then
()2 = (), () = ( ), = ( ) > 0.
By the Krein-Milman theorem, there is a pure state that satises ( ) > 0. In this case
= 0, hence the representation is faithful. The theorem is proved.
5.4
CHAPTER 5. ALGEBRAS
72
Denition. Let be a Hilbert space. The weak operator topology is the topology dened
by the open sets
(0 ; 1 , 2 , . . . , ; 1 , 2 , . . . , ; ) = { () ( 0 ) , < , = 1, 2, . . . , }.
The strong operator topology is the topology dened by the open sets
(0 ; 1 , 2 , . . . , ; ) = { () ( 0 ) < , = 1, 2, . . . , }.
Denition. A von Neumann algebra is a -subalgebra of () that is weakly closed.
Remark 5.4.1. If is a self-adjoint subalgebra of (), then its weak closure is a von
Neumann algebra. If is commutative, then its closure is commutative.
Proposition 5.4.1. If is a normal operator on , then the von Neumann algebra
generated by is commutative. If is the set of multiplicative functionals on , then
the Gelfand transform is a -isometrical isomorphism of onto ( ).
We want to show that there is a unique space structure on and a unique isometrical isomorphism : (( )) which extends the functional calculus with
continuous functions dened by the Spectral Theorem (Theorem 5.2.2).
Assume we have a nite positive regular Borel measure on . We can assume that
the measure of the entire space is 1, so that we have a probability measure. For the moment,
we work in this hypothesis.
The map 7 , where : 2 ( ) 2 ( ), = identies ( )
with a maximal commutative von Neumann subalgebra of the algebra of operators on
2 ( ).
Proposition 5.4.2. The weak operator topology and the weak topology on coincide.
Proof. = 1 , and recall that every function in 1 is the product of two 2 functions.
Thus an element of the form
( ) =
with and 1 , can also be represented as
1 2 = 1 , 2
where = 1 2 . Hence the conclusion.
Proposition 5.4.3. The space (( )) is weak -dense in (( )).
Proof. We will show that the unit ball in (( )) is weak -dense
in the unit ball in
73
=1
=1
Because the measure is regular, we can choose such that the integrals are as small as
desired.
Recall that a vector is cyclic for an algebra () if is dense in and separating
if = 0 implies = 0. If is commutative, then cyclic implies separating, because
= 0 implies ker( ), hence = 0.
Theorem 5.4.1. If is a normal operator on such that has a cyclic vector, then
there is a positive regular Borel measure supported on ( ) = and an isometrical
isomorphism from onto 2 (( ), ) such that the map
: (2 (( ), ),
( ) = 1
= ( ), for (( )).
( )
If the support of were not the entire spectrum, then, by Urysohns lemma, we could nd
a continuous function that is 1 somewhere on the spectrum and is zero on the support of
CHAPTER 5. ALGEBRAS
74
2
0 = ( ) = ( ), ( ) = ( ), =
2 = 0,
( )
impossible. So supp() = ( ).
Dene
0 : 2 (( ), ),
0 ( ( )) = .
The computation
22
( )
2 = 2 ( ), = ( )2
( ) = 1
75
However, not all operators have cyclic vectors. Instead we will use non-separating vectors
and replace by the smallest invariant subspace containing a non-separating vector. We
proceed to show that every normal has a separating vector.
An easy application of Zorns lemma yields the following result.
Proposition 5.4.4. Every commutative -algebra is contained in a maximal commutative
von Neumann algebra.
Denition. If (), then the commutant of , denoted , is the set of operators in
() which commute with every operator in .
Proposition 5.4.5. A -algebra in () is a maximal commutative von Neumann algebra
if and only if it is equal to its own commutant.
Proof. Let be the -algebra. If is commutative, then . If is maximal
commutative then necessarily we have equality.
Conversely, if equality holds, then is a von Neumann algebra. It must be maximal
commutative, for if commutes with everything in , then = .
Lemma 5.4.1. Let (), is a closed subspace of , and the orthogonal projection
onto . Then = if and only if is an invariant subspace for both and .
Moreover, in this case both and are invariant for .
Proof. is invariant for if and only if = . So is invariant for both and
if and only if = and = . The latter is equivalent, by conjugating
to = . So is invariant for both and if and only if = (in which
case the equality to is superuous). Note also that invariant for implies
invariant for (by the equality , = , ).
Denition. A subspace of is a reducing subspace for if it satises any of the equivalent
conditions from the statement of the above lemma.
Lemma 5.4.2. If is a -algebra contained in () and , then the projection onto
is in .
Proof. By Lemma 5.4.1, it suces to show that is invariant for both and for every
. Note that so both and leave invariant.
Theorem 5.4.2. If is a maximal commutative von Neumann algebra on a separable
Hilbert space , then has a cyclic vector.
Proof. Let be the set of all collections of projections { } in such that
For each there is {0} so that is the projection onto
= = 0 for = .
CHAPTER 5. ALGEBRAS
76
Clearly is not empty, since we can build an element in starting with one vector, via
Lemma 5.4.2. Order by inclusion. The hypothesis of Zorns Lemma is satised. Pick a
maximal element { } .
Let be the collection of all nite subsets of the index set partially ordered by inclusion
and let { } be the net of the orthgonal projections dened by
=
.
( )2 = ( )2 , = ( ), = , , .
77
Let be a separable Hilbert space and normal on . By Corollary 5.4.1, the commutative von Neumann algebra has a separating vector . If we set = , then
both and are invariant under . We can therefore dene a map : ( )
by ( ) = .
Lemma 5.4.3. The map dened above is a -isometrical isomorphism. Moreover () ( ) =
( ) ( ) for all .
Proof. In view of the previous theorem, let us show that is one-to-one. And indeed, if
( ) = 0 then = 0, because = . So = 0, because is separating of .
The equality of spectra is proved as follows.
First,
() ( ) = ( ).
Indeed, () ( ) ( ) because the inverse of might or might not be in .
Moreover, because the resolvent is open both for () and for , ( ) is obtained
from () ( ) by adding to it some bounded components of its complement. So if is
invertible in (), then ( )( ) is self-adjoint, so its spectrum is real and hence
necessarily the same in () and . So this operator must be invertible in , and hence
so is . Next
( ) = ( ) ( )
because is a -isometrical isomorphism onto the image. Repeating the above argument we
also have
( ) ( ) = ( ) ( )
and we are done.
Theorem 5.4.4. (Functional Calculus for Normal Operators - Version I) Let be a normal
operator on the separable Hilbert space let the : (( )) be the Gelfand
transform. Then there is a positive regular Borel measure having support ( ) and a
-isometrical isomorphism from onto (( ), ) which extends . Moreover is
unique up to mutual absolute continuity while (( ), ) and are unique.
Proof. Let be a separating vector for , = , and
: ( ),
( ) = .
Let be the von Neumann algebra generated by . The map is continuous in the
weak operator topology (because it is obtained by restricting the domain). Hence ( )
. Moreover, if
0 : (( ))
is the Gelfand transform, then = 0 .
CHAPTER 5. ALGEBRAS
78
Because is normal and has the cyclic vector , by Theorem 5.4.1 there is a positive
regular Borel measure with support ( ) = ( ) (here we use the previous lemma),
and a -isometrical (onto) isomorphism
0 : (, ( ), ), such that 0 = .
Moreover, 0 is continuous form the weak operator topology of to the weak -topology on
(( ), ). Hence = 0 is a -isometrical isomorphism from into (( ), ),
continuous in the weak/weak topologies, and which extends the Gelfand transform.
The only thing that remains to show is that takes onto (( ), ). For this we
need the following result.
Lemma 5.4.4. Let be a Hilbert space. Then the unit ball of () is compact in the
weak operator topology.
Proof. The proof is from the book of Kadison and Ringrose, Fundamental of the theory of
operator algebras. For two vectors , , let , be the closed disk of radius
in the complex plane. The mapping which assigns to each (())1 the point
{ , , }
,
,
79
The spectral measure of a normal operator is dened as follows. For each Borel set
( ), let
() = 1 ( ).
Because
2 = =
() is an orthogonal projection. Moreover, if = , then ()( ) = ( )() =
0. Hence
(=1 ) =
( ).
=1
( ), =
, .
( )
It turns out that ( ) is the functional calculus dened by the Gelfand transform. In fact
muc more is true.
Let : ( ) be a measurable function. There is a countable collection of open
disks, , 1, that form a basis for the topology on ( ). Let be the union of those
disks for which ( 1 ( )) = 0. Then ( 1 ()) = 0. The complement of is the
essential range of . We say that is essentially bounded if its essential range is bounded.
Theorem 5.4.5. (Functional Calculus for Normal Operators - Version II) There is a isometrical isomorphism : (( ), ) which is onto, dened by the formula
( ), =
, .
( )
Moreover, = 1 .
Proof. Check on step functions, then use density.
This justies the notation
( ) =
.
( )
In particular,
=
.
( )
80
CHAPTER 5. ALGEBRAS
Theorem 5.4.6. (The spectral mapping theorem) The spectrum of ( ) is the essential
range of .
Proposition 5.4.6. If is normal and has spectral measure , and if ( ), then
( ) is also normal and the spectral measure of ( ) is dened by ( ) () = ( 1 ()).
Example. If is self-adjoint, then the spectral measure of is supported on a compact
subset of . If is unitary, then the spectral measure of is supported on a compact subset
of the unit circle.
Example. Let be a normal operator on . Let 1 , . . . , be the eigenvalues of . The
spectral measure associates to each eigenvalue the projection onto its eigenspace.
Chapter 6
Topics presented by the students
The abstract index
Volterra operators; the Fredholm alternative
Multiplication operators
The Fourier transform
Applications of the spectral theorem for normal operators
Distributions
81
82
Appendix A
Background results
A.1
Zorns lemma
Theorem A.1.1. Suppose a partially ordered set has the property that every totally
ordered subset has an upper bound in . Then the set contains at least one maximal
element.
Remark A.1.1. This result is proved using the Axiom of Choice.
83