Neumark Theroem
Neumark Theroem
Neumark Theroem
r
X
i
v
:
m
a
t
h
-
p
h
/
9
8
0
7
0
3
0
v
1
2
4
J
u
l
1
9
9
8
Lecture Notes on
C
-Algebras, Hilbert C
-modules,
and Quantum Mechanics
Draft: 8 April 1998
N.P. Landsman
Korteweg-de Vries Institute for Mathematics, University of Amsterdam,
Plantage Muidergracht 24,
1018 TV AMSTERDAM, THE NETHERLANDS
email: [email protected]
homepage: http://turing.wins.uva.nl/
npl/
telephone: 020-5256282
oce: Euclides 218a
2
CONTENTS 3
Contents
1 Historical notes 5
1.1 Origins in functional analysis and quantum mechanics . . . . . . . . . . . . . . . . 5
1.2 Rings of operators (von Neumann algebras) . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Reduction of unitary group representations . . . . . . . . . . . . . . . . . . . . . . 6
1.4 The classication of factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Elementary theory of C
-algebras 12
2.1 Basic denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Banach algebra basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Commutative Banach algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Commutative C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Spectrum and functional calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Positivity in C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7 Ideals in C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.8 States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.9 Representations and the GNS-construction . . . . . . . . . . . . . . . . . . . . . . 38
2.10 The Gelfand-Neumark theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.11 Complete positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12 Pure states and irreducible representations . . . . . . . . . . . . . . . . . . . . . . . 44
2.13 The C
-modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 The C
-algebra of a Hilbert C
-module . . . . . . . . . . . . . . . . . . . . . . . . 61
3.4 Morita equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 Rieel induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.6 The imprimitivity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Group C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.8 C
-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.10 The abstract transitive imprimitivity theorem . . . . . . . . . . . . . . . . . . . . . 78
3.11 Induced group representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.12 Mackeys transitive imprimitivity theorem . . . . . . . . . . . . . . . . . . . . . . . 82
4 Applications to quantum mechanics 83
4.1 The mathematical structure of classical and quantum mechanics . . . . . . . . . . 83
4.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3 Stinesprings theorem and coherent states . . . . . . . . . . . . . . . . . . . . . . . 86
4.4 Covariant localization in conguration space . . . . . . . . . . . . . . . . . . . . . . 87
4.5 Covariant quantization on phase space . . . . . . . . . . . . . . . . . . . . . . . . . 88
Literature 89
4 CONTENTS
5
1 Historical notes
1.1 Origins in functional analysis and quantum mechanics
The emergence of the theory of operator algebras may be traced back to (at least) three develop-
ments.
The work of Hilbert and his pupils in Gottingen on integral equations, spectral theory, and
innite-dimensional quadratic forms (1904-);
The discovery of quantum mechanics by Heisenberg (1925) in Gottingen and (independently)
by Schrodinger in Z urich (1926);
The arrival of John von Neumann in Gottingen (1926) to become Hilberts assistant.
Hilberts memoirs on integral equations appeared between 1904 and 1906. In 1908 his student E.
Schmidt dened the space
2
in the modern sense. F. Riesz studied the space of all continuous linear
maps on
2
(1912), and various examples of L
2
-spaces emerged around the same time. However,
the abstract concept of a Hilbert space was still missing.
Heisenberg discovered a form of quantum mechanics, which at the time was called matrix
mechanics. Schrodinger was led to a dierent formulation of the theory, which he called wave
mechanics. The relationship and possible equivalence between these alternative formulations of
quantum mechanics, which at rst sight looked completely dierent, was much discussed at the
time. It was clear from either approach that the body of work mentioned in the previous paragraph
was relevant to quantum mechanics.
Heisenbergs paper initiating matrix mechanics was followed by the Dreimannerarbeit of Born,
Heisenberg, and Jordan (1926); all three were in Gottingen at that time. Born was one of the
few physicists of his time to be familiar with the concept of a matrix; in previous research he
had even used innite matrices (Heisenbergs fundamental equations could only be satised by
innite-dimensional matrices). Born turned to his former teacher Hilbert for mathematical advice.
Hilbert had been interested in the mathematical structure of physical theories for a long time;
his Sixth Problem (1900) called for the mathematical axiomatization of physics. Aided by his
assistants Nordheim and von Neumann, Hilbert thus ran a seminar on the mathematical structure
of quantum mechanics, and the three wrote a joint paper on the subject (now obsolete).
It was von Neumann alone who, at the age of 23, saw his way through all structures and
mathematical diculties. In a series of papers written between 1927-1932, culminating in his book
Mathematische Grundlagen der Quantenmechanik (1932), he formulated the abstract concept of a
Hilbert space, developed the spectral theory of bounded as well as unbounded normal operators
on a Hilbert space, and proved the mathematical equivalence between matrix mechanics and wave
mechanics. Initiating and largely completing the theory of self-adjoint operators on a Hilbert
space, and introducing notions such as density matrices and quantum entropy, this book remains
the denitive account of the mathematical structure of elementary quantum mechanics. (von
Neumanns book was preceded by Diracs The Principles of Quantum Mechanics (1930), which
contains a heuristic and mathematically unsatisfactory account of quantum mechanics in terms of
linear spaces and operators.)
1.2 Rings of operators (von Neumann algebras)
In one of his papers on Hilbert space theory (1929), von Neumann denes a ring of operators
M (nowadays called a von Neumann algebra) as a
-subalgebra of the algebra B(H) of all
bounded operators on a Hilbert space H (i.e, a subalgebra which is closed under the involution
A A
) that is closed (i.e., sequentially complete) in the weak operator topology. The latter may
be dened by its notion of convergence: a sequence A
n
of bounded operators weakly converges
to A when (, A
n
) (, A) for all H. This type of convergence is partly motivated by
quantum mechanics, in which (, A) is the expectation value of the observable A in the state ,
provided that A is self-adjoint and has unit norm.
6 1 HISTORICAL NOTES
For example, B(H) is itself a von Neumann algebra. (Since the weak topology is weaker than
the uniform (or norm) topology on B(H), a von Neumann algebra is automatically norm-closed
as well, so that, in terminology to be introduced later on, a von Neumann algebra becomes a C
-
algebra when one changes the topology from the weak to the uniform one. However, the natural
topology on a von Neumann algebra is neither the weak nor the uniform one.)
In the same paper, von Neumann proves what is still the basic theorem of the subject: a
-
subalgebra M of B(H), containing the unit operator I, is weakly closed i M
= M. Here the
commutant M
is simply (M
. This theorem is
remarkable, in relating a topological condition to an algebraic one; one is reminded of the much
simpler fact that a linear subspace / of His closed i /
, where /
.
This alternative characterization indicates why von Neumann algebras are important in physics:
the set of bounded operators on H which are invariant under a given group representation U(G)
on H is automatically a von Neumann algebra. (Note that a given group U of unitaries on H may
be regarded as a representation U of U itself, where U is the identity map.)
1.3 Reduction of unitary group representations
The (possible) reduction of U(G) is determined by the von Neumann algebras U(G)
and U(G)
.
For example, U is irreducible i U(G)
U(G)
= CI. When G is
compact, so that U is discretely reducible, this implies that U is a multiple of a xed irreducible
1.3 Reduction of unitary group representations 7
representation U
on a Hilbert space H
, so that H H
/, and U U
I
K
.
When G is not compact, but still assumed to be locally compact, unitary representations may
be reducible without containing any irreducible subrepresentation. This occurs already in the
simplest possible cases, such as the regular representation of G = R on H = L
2
(R); that is, one
puts U(x)(y) = (y x). The irreducible would-be subspaces of H would be spanned by the
vectors
p
(y) := exp(ipy), but these functions do not lie in L
2
(R). The solution to this problem
was given by von Neumann in a paper published in 1949, but written in the thirties (the ideas in
it must have guided von Neumann from at least 1936 on).
Instead of decomposing H as a direct sum, one should decompose it as a direct integral.
(To do so, one needs to assume that H is separable.) This means that rstly one has a measure
space (, ) and a family of Hilbert spaces H
for which () H
with respect to
the measure , one needs a sequence of sections
n
satisfying the two conditions that rstly the
function (
n
(),
m
())
be measurable for all n, m, and secondly that for each xed the
n
span H
are measurable.
For ,
0
it then makes sense to dene
(, ) :=
.
The direct integral
d() H
()
for some (suitably measurable) family of operators A
on H
. We then write
A =
d() A
.
Thus a unitary group representation U(G) on H is diagonal when
U(x)() = U
(x)
d() U
.
Reducing a given representation U on some Hilbert space then amounts to nding a unitary map
V between H and some direct integral Hilbert space, such that each H
carries a representation
U
, and V U(x)V
= U
R
dp H
p
,
8 1 HISTORICAL NOTES
in which each H
p
is C. To reduce the regular representation of R on L
2
(R), one simply performs
a Fourier transform V : L
2
(R) L
2
(R), i.e.,
V (p) =
R
dy e
ipy
(y).
This leads to V U(x)V
e
it
0 z
0 e
it
w
0 0 1
,
where is an irrational real number, t R, and z, w C.
When G is wild, curious phenomena may occur. By denition, a wild group has primary unitary
representations which contain no irreducible subrepresentations. More bizarrely, representations
of the latter type may be decomposed in two alternative ways
U =
G
d
1
() U
G
d
2
() U
,
where the measures
1
and
2
are disjoint (that is, supported by disjoint subsets of
G).
A reducible primary representation U may always be decomposed as U = U
h
U
h
. In case
that U is not equivalent to U
h
, and U is not of type I, it is said to be a representation of type II.
When U is neither of type I nor of type II, it is of type III. In that case U is equivalent to U
h
;
indeed, all (proper) subrepresentations of a primary type III representation are equivalent.
1.4 The classication of factors
Between 1936 and 1953 von Neumann wrote 5 lengthy, dicult, and profound papers (3 of which
were in collaboration with Murray) in which the study of his rings of operators was initiated.
(According to I.E. Segal, these papers form perhaps the most original major work in mathematics
in this century.)
The analysis of Murray and von Neumann is based on the study of the projections in a von
Neumann algebra M (a projection is an operator p for which p
2
= p
V = p
and V V
.
Hence when M = B(H) one has p q i pH and qH have the same dimension, for in that case
one may take any V with the above properties.
An equivalent characterization of arises when we write M = U(G)
d() H
;
M =
d() M
,
where (almost) each M
and II
= O, where
O
consists of all points that are spacelike separated from O. The operation O O
on causally
closed regions in space-time is somewhat analogous to the operation M M
on von Neumann
algebras. Thus Haag proposed that a quantum eld theory should be dened by a net of local
observables; this is a map O M(O) from the set of all causally closed regions in space-time
to the set of all von Neumann algebras on some Hilbert space, such that M(O
1
) M(O
2
) when
O
1
O
2
, and M(O)
= M(O
).
This idea initiated algebraic quantum eld theory, a subject that really got o the ground
with papers by Haags pupil Araki in 1963 and by Haag and Kastler in 1964. From then till
the present day, algebraic quantum eld theory has attracted a small but dedicated group of
mathematical physicists. One of the result has been that in realistic quantum eld theories the
local algebras M(O) must all be isomorphic to the unique hypernite factor of type III
1
discussed
below. (Hence von Neumanns belief that physics should use II
1
factors has not been vindicated.)
A few years later (1967), an extraordinary coincidence took place, which was to play an es-
sential role in the classication of factors of type III. On the mathematics side, Tomita developed
a technique in the study of von Neumann algebras, which nowadays is called modular theory
or Tomita-Takesaki theory (apart from clarifying Tomitas work, Takesaki made essential con-
tributions to this theory). Among other things, this theory leads to a natural time-evolution on
certain factors. On the physics side, Haag, Hugenholtz, and Winnink characterized states of ther-
mal equilibrium of innite quantum systems by an algebraic condition that had previously been
introduced in a heuristic setting by Kubo, Martin, and Schwinger, and is therefore called the KMS
condition. This condition leads to type III factors equipped with a time-evolution which coincided
with the one of the Tomita-Takesaki theory.
In the hands of Connes, the Tomita-Takesaki theory and the examples of type III factors
1.5 C
-algebras 11
provided by physicists (Araki, Woods, Powers, and others) eventually led to the classication of all
hypernite factors of type II and III (the complete classication of all factors of type I is already
given by the list presented earlier). These are factors containing a sequence of nite-dimensional
subalgebras M
1
M
2
. . . M, such that M is the weak closure of
n
M
n
. (Experience shows
that all factors playing a role in physics are hypernite, and many natural examples of factors
constructed by purely mathematical techniques are hypernite as well.) The work of Connes, for
which he was awarded the Fields Medal in 1982, and others, led to the following classication of
hypernite factors of type II and III (up to isomorphism):
There is a unique hypernite factor of type II
1
. (In physics this factor occurs when one
considers KMS-states at innite temperature.)
There is a unique hypernite factor of type II
-algebras
In the midst of the Murray-von Neumann series of papers, Gelfand initiated a separate develop-
ment, combining operator algebras with the theory of Banach spaces. In 1941 he dened the con-
cept of a Banach algebra, in which multiplication is (separately) continuous in the norm-topology.
He proceeded to dene an intrinsic spectral theory, and proved most basic results in the theory of
commutative Banach algebras.
In 1943 Gelfand and Neumark dened what is now called a C
-algebra is isomor-
phic to the norm-closed
-algebra of operators on a Hilbert space. Their paper also contained the
rudiments of what is now called the GNS construction, connecting states to representations. In
its present form, this construction is due to Segal (1947), a great admirer of von Neumann, who
generalized von Neumanns idea of a state as a positive normalized linear functional from B(H) to
arbitrary C
-algebras
and quantum physics. Moreover, the theory of C
-algebra of
the structure.
The rst instance where this led to a deep result which has not been proved in any other
way is the theorem of Gelfand and Raikov (1943), stating that the unitary representations of a
locally compact group separate the points of the group (that is, for each pair x = y there exists a
unitary representation U for which U(x) = U(y). This was proved by constructing a C
-algebra
C
-
algebra are group actions, groupoids, foliations, and complex domains. The same idea lies at the
12 2 ELEMENTARY THEORY OF C
-ALGEBRAS
basis of non-commutative geometry and non-commutative topology. Here the starting
point is another theorem of Gelfand, stating that any commutative C
-algebras.
This strategy has been successful in K-theory, whose non-commutative version is even simpler
than its usual incarnation, and in (de Rham) cohomology theory, whose non-commutative version
is called cyclic cohomology. Finally, homology, cohomology, K-theory, and index theory haven
been unied and made non-commutative in the KK-theory of Kasparov. The basic tool in KK-
theory is the concept of a Hilbert C
-algebras
2.1 Basic denitions
All vector spaces will be dened over C, and all functions will be C-valued, unless we explicitly
state otherwise. The abbreviation i means if and only if, which is the same as the symbol .
An equation of the type a := b means that a is by denition equal to b.
Denition 2.1.1 A norm on a vector space 1 is a map | | : 1 R such that
1. | v | 0 for all v 1;
2. | v |= 0 i v = 0;
3. | v |= [[ | v | for all C and v 1;
4. | v +w | | v | + | w | (triangle inequality).
A norm on 1 denes a metric d on 1 by d(v, w) :=| v w |. A vector space with a norm which
is complete in the associated metric (in the sense that every Cauchy sequence converges) is called
a Banach space. We will denote a generic Banach space by the symbol B.
The two main examples of Banach spaces we will encounter are Hilbert spaces and certain
collections of operators on Hilbert spaces.
Denition 2.1.2 A pre-inner product on a vector space 1 is a map ( , ) : 1 1 C such that
1. (
1
v
1
+
2
v
2
,
1
w
1
+
2
w
2
) =
1
1
(v
1
, w
1
) +
1
2
(v
1
, w
2
) +
2
1
(v
2
, w
1
) +
2
2
(v
2
, w
2
) for
all
1
,
2
,
1
,
2
C and v
1
, v
2
, w
1
, w
2
1;
2. (v, v) 0 for all v 1.
An equivalent set of conditions is
1. (v, w) = (w, v) for all v, w 1;
2. (v,
1
w
1
+
2
w
2
) =
1
(v, w
1
) +
2
(v, w
2
) for all
1
,
2
C and v, w
1
, w
2
1;
3. (v, v) 0 for all v 1.
A pre-inner product for which (v, v) = 0 i v = 0 is called an inner product.
The equivalence between the two denitions of a pre-inner product is elementary; in fact, to
derive the rst axiom of the second characterization from the rst set of conditions, it is enough
to assume that (v, v) R for all v (use this reality with v v + iw). Either way, one derives the
Cauchy-Schwarz inequality
[(v, w)[
2
(v, v)(w, w), (2.1)
2.1 Basic denitions 13
for all v, w 1. Note that this inequality is valid even when ( , ) is not an inner product, but
merely a pre-inner product.
It follows from these properties that an inner product on 1 denes a norm on 1 by | v |:=
-ALGEBRAS
The dual B
of B is the space of all functionals on B. Similarly to the proof of 2.1.5, one shows that
B
is a Banach space. For later use, we quote, without proof, the fundamental Hahn-Banach
theorem.
Theorem 2.1.6 For a functional
0
on a linear subspace B
0
of a Banach space B there exists a
functional on B such that =
0
on B
0
and | |=|
0
|. In other words, each functional dened
on a linear subspace of B has an extension to B with the same norm.
Corollary 2.1.7 When (v) = 0 for all B
then v = 0.
For v = 0 we may dene a functional
0
on Cv by
0
(v) = , and extend it to a functional
on B with norm 1.
Recall that an algebra is a vector space with an associative bilinear operation (multiplication)
: A A A; we usually write AB for A B. It is clear that B(B) is an algebra under operator
multiplication. Moreover, using (2.4) twice, for each v B one has
| ABv | | A | | Bv | | A | | B | | v | .
Hence from (2.2) we obtain | AB | | A | | B |.
Denition 2.1.8 A Banach algebra is a Banach space A which is at the same time an algebra,
in which for all A, B A one has
| AB | | A | | B | . (2.8)
It follows that multiplication in a Banach algebra is separately continuous in each variable.
As we have just seen, for any Banach space B the space B(B) of all bounded operators on B is
a Banach algebra. In what follows, we will restrict ourselves to the case that B is a Hilbert space
H; this leads to the Banach algebra B(H). This algebra has additional structure.
Denition 2.1.9 An involution on an algebra A is a real-linear map A A
= A; (2.9)
(AB)
= B
; (2.10)
(A)
= A
. (2.11)
A
-algebra is an algebra with an involution.
The operator adjoint A A
) := (A, ),
denes an involution on B(H). Hence B(H) is a
-algebra. As in this case, an element A of a C
-
algebra A is called self-adjoint when A
= A. (2.12)
Since one may write
A = A
+iA
:=
A +A
2
+i
AA
2i
, (2.13)
every element of A is a linear combination of two self-adjoint elements.
To see how the norm in B(H) is related to the involution, we pick H, and use the Cauchy-
Schwarz inequality and (2.4) to estimate
| A |
2
= (A, A) = (, A
A) | | | A
A | | A
A | | |
2
.
Using (2.3) and (2.8), we infer that
| A |
2
| A
A | | A
| | A | . (2.14)
This leads to | A | | A
|. Replacing A by A
| | A |, so that
| A
A |=| A |
2
.
This motivates the following denition.
2.2 Banach algebra basics 15
Denition 2.1.10 A C
A | = | A |
2
. (2.16)
In other words, a C
-algebra is a Banach
-algebra in which (2.16) holds.
Here a Banach
-algebra is, of course, a Banach algebra with involution. Combining (2.16) and
(2.15), one derives | A | | A
|=| A | . (2.17)
The same argument proves the following.
Lemma 2.1.11 A Banach
-algebra in which | A |
2
| A
A | is a C
-algebra.
We have just shown that B(H) is a C
-algebra is isomor-
phic to a norm-closed
-algebra in B(H), for some Hilbert space H. Hence the axioms in 2.1.10
characterize norm-closed
-algebras on Hilbert spaces, although the axioms make no reference to
Hilbert spaces at all.
For later use we state some self-evident denitions.
Denition 2.1.12 A morphism between C
) = (A)
(2.19)
for all A, B A. An isomorphism is a bijective morphism. Two C
-algebras is
automatically isometric. For this reason the condition that an isomorphism be isometric is not
included in the denition.
2.2 Banach algebra basics
The material in this section is not included for its own interest, but because of its role in the
theory of C
-algebras. Even in that special context, it is enlightening to see concepts such as the
spectrum in their general and appropriate setting.
Recall Denition 2.1.8. A unit in a Banach algebra A is an element I satisfying IA = AI = A
for all A A, and
| I |= 1. (2.20)
A Banach algebra with unit is called unital. We often write z for zI, where z C. Note that in
a C
, so that I
I = I
;
taking the adjoint, this implies I
-ALGEBRAS
where we have written A+I for (A, ), etc. In other words, the number 1 in C is identied with
I. Furthermore, dene a norm on A
I
by
| A +I |:=| A | +[[. (2.23)
In particular, | I |= 1. Using (2.15) in A, as well as 2.1.1.3, one sees from (2.22) and (2.23) that
| (A +I)(B +I) | | A | | B | +[[ | B | +[[ | A | +[[ [[ =| A +I | | B +I |,
so that A
I
is a Banach algebra with unit. Since by (2.23) the norm of A A in A coincides with
the norm of A + 0I in A
I
, we have shown the following.
Proposition 2.2.1 For every Banach algebra without unit there exists a unital Banach algebra A
I
and an isometric (hence injective) morphism A A
I
, such that A
I
/A C.
As we shall see at the end of section 2.4, the unitization A
I
with the given properties is not
unique.
Denition 2.2.2 Let A be a unital Banach algebra. The resolvent (A) of A A is the set of
all z C for which A zI has a (two-sided) inverse in A.
The spectrum (A) of A A is the complement of (A) in C; in other words, (A) is the set
of all z C for which A zI has no (two-sided) inverse in A.
When A has no unit, the resolvent and the spectrum are dened through the embedding of A in
A
I
= A C.
When A is the algebra of n n matrices, the spectrum of A is just the set of eigenvalues.
For A = B(H), Denition 2.2.2 reproduces the usual notion of the spectrum of an operator on a
Hilbert space.
When A has no unit, the spectrum (A) of A A always contains zero, since it follows from
(2.22) that A never has an inverse in A
I
.
Theorem 2.2.3 The spectrum (A) of any element A of a Banach algebra is
1. contained in the set z C[ [z[ | A |;
2. compact;
3. not empty.
The proof uses two lemmas. We assume that A is unital.
Lemma 2.2.4 When | A | < 1 the sum
n
k=0
A
k
converges to (I A)
1
.
Hence (A zI)
1
always exists when [z[ >| A |.
We rst show that the sum is a Cauchy sequence. Indeed, for n > m one has
|
n
k=0
A
k
k=0
A
k
|=|
n
k=m+1
A
k
|
n
k=m+1
| A
k
|
n
k=m+1
| A |
k
.
For n, m this goes to 0 by the theory of the geometric series. Since A is complete, the Cauchy
sequence
n
k=0
A
k
converges for n . Now compute
n
k=0
A
k
(I A) =
n
k=0
(A
k
A
k+1
) = I A
n+1
.
Hence
| I
n
k=0
A
k
(I A) |=| A
n+1
| | A |
n+1
,
2.2 Banach algebra basics 17
which 0 for n , as | A | < 1 by assumption. Thus
lim
n
n
k=0
A
k
(I A) = I.
By a similar argument,
lim
n
(I A)
n
k=0
A
k
= I.
so that, by continuity of multiplication in a Banach algebra, one nally has
lim
n
n
k=0
A
k
= (I A)
1
. (2.24)
The second claim of the lemma follows because (A z)
1
= z
1
(I A/z)
1
, which exists
because | A/z | < 1 when [z[ >| A |.
To prove that (A) is compact, it remains to be shown that it is closed.
Lemma 2.2.5 The set
G(A) := A A[ A
1
exists (2.25)
of invertible elements in A is open in A.
Given A G(A), take a B A for which | B | <| A
1
|
1
. By (2.8) this implies
| A
1
B | | A
1
| | B | < 1. (2.26)
Hence A+B = A(I +A
1
B) has an inverse, namely (I +A
1
B)
1
A
1
, which exists by (2.26) and
Lemma 2.2.4. It follows that all C A for which | AC | < lie in G(A), for | A
1
|
1
.
To resume the proof of Theorem 2.2.3, given A A we now dene a function f : C A by
f(z) := z A. Since | f(z +)f(z) |= , we see that f is continuous (take = in the denition
of continuity). Because G(A) is open in A by Lemma 2.2.5, it follows from the topological denition
of a continuous function that f
1
(G(A)) is open in A. But f
1
(G(A)) is the set of all z C where
z A has an inverse, so that f
1
(G(A)) = (A). This set being open, its complement (A) is
closed.
Finally, dene g : (A) A by g(z) := (z A)
1
. For xed z
0
(A), choose z C such that
[z z
0
[ <| (A z
0
)
1
|
1
. From the proof of Lemma 2.2.5, with A A z
0
and C A z,
we see that z (A), as | Az
0
(A z) |= [z z
0
[. Moreover, the power series
1
z
0
A
n
k=0
z
0
z
z
0
A
k=0
z
0
z
z
0
A
=
1
z
0
A
z
0
z
z
0
A
=
1
z A
= g(z).
Hence
g(z) =
k=0
(z
0
z)
k
(z
0
A)
k1
(2.27)
18 2 ELEMENTARY THEORY OF C
-ALGEBRAS
is a norm-convergent power series in z. For z = 0 we write | g(z) |= [z[
1
| (I A/z)
1
| and
observe that lim
z
IA/z = I, since lim
z
| A/z |= 0 by 2.1.1.3. Hence lim
z
(IA/z)
1
=
I, and
lim
z
| g(z) |= 0. (2.28)
Let A
: z (g(z))
is given by a convergent power series, and (2.28) implies that
lim
z
g
(z) = 0. (2.29)
Now suppose that (A) = , so that (A) = C. The function g, and hence g
, is then dened on
C, where it is analytic and vanishes at innity. In particular, g
-algebras.
Corollary 2.2.6 If every element (except 0) of a unital Banach algebra A is invertible, then A C
as Banach algebras.
Since (A) = , for each A = 0 there is a z
A
C for which A z
A
I is not invertible. Hence
A z
A
I = 0 by assumption, and the map A z
A
is the desired algebra isomorphism. Since
| A |=| zI |= [z[, this isomorphism is isometric.
Dene the spectral radius r(A) of A A by
r(A) := sup[z[, z (A). (2.30)
From Theorem 2.2.3.1 one immediately infers
r(A) | A | . (2.31)
Proposition 2.2.7 For each A in a unital Banach algebra one has
r(A) = lim
n
| A
n
|
1/n
. (2.32)
By Lemma 2.2.4, for [z[ >| A | the function g in the proof of Lemma 2.2.5 has the norm-
convergent power series expansion
g(z) =
1
z
k=0
A
z
k
. (2.33)
On the other hand, we have seen that for any z (A) one may nd a z
0
(A) such that the
power series (2.27) converges. If [z[ > r(A) then z (A), so (2.27) converges for [z[ > r(A). At
this point the proof relies on the theory of analytic functions with values in a Banach space, which
says that, accordingly, (2.33) is norm-convergent for [z[ > r(A), uniformly in z. Comparing with
(2.31), this sharpens what we know from Lemma 2.2.4. The same theory says that (2.33) cannot
norm-converge uniformly in z unless | A
n
| /[z[
n
< 1 for large enough n. This is true for all z for
which [z[ > r(A), so that
lim sup
n
| A |
1/n
r(A). (2.34)
To derive a second inequality we use the following polynomial spectral mapping property.
Lemma 2.2.8 For a polynomial p on C, dene p((A)) as p(z)[ z (A). Then
p((A)) = (p(A)). (2.35)
2.2 Banach algebra basics 19
To prove this equality, choose z, C and compare the factorizations
p(z) = c
n
i=1
(z
i
());
p(A) I = c
n
i=1
(A
i
()I). (2.36)
Here the coecients c and
i
() are determined by p and . When (p(A)) then p(A) I is
invertible, which implies that all A
i
()I must be invertible. Hence (p(A)) implies that
at least one of the A
i
()I is not invertible, so that
i
() (A) for at least one i. Hence
p(
i
()) = 0, i.e., p((A)). This proves the inclusion (p(A)) p((A)).
Conversely, when p((A)) then = p(z) for some z (A), so that for some i one must
have
i
() = z for this particular z. Hence
i
() (A), so that A
i
() is not invertible,
implying that p(A) I is not invertible, so that (p(A)). This shows that p((A)) (p(A)),
and (2.35) follows.
To conclude the proof of Proposition 2.2.7, we note that since (A) is closed there is an (A)
for which [[ = r(A). Since
n
(A
n
) by Lemma 2.2.8, one has [
n
[ | A
n
| by (2.31). Hence
| A
n
|
1/n
[[ = r(A). Combining this with (2.34) yields
lim sup
n
| A |
1/n
r(A) | A
n
|
1/n
.
Hence the limit must exist, and
lim
n
| A |
1/n
= inf
n
| A
n
|
1/n
= r(A).
Denition 2.2.9 An ideal in a Banach algebra A is a closed linear subspace I A such that
A I implies AB I and BA I for all B A.
A left-ideal of A is a closed linear subspace I for which A I implies BA I for all B A.
A right-ideal of A is a closed linear subspace I for which A I implies AB I for all B A.
A maximal ideal is an ideal I = A for which no ideal
I = A,
I = I, exists which contains I.
In particular, an ideal is itself a Banach algebra. An ideal I that contains an invertible element
A must coincide with A, since A
1
A = I must lie in I, so that all B = BI must lie in I. This
shows the need for considering Banach algebras with and without unit; it is usually harmless to
add a unit to a Banach algebra A, but a given proper ideal I = A does not contain I, and one
cannot add I to I without ruining the property that it is a proper ideal.
Proposition 2.2.10 If I is an ideal in a Banach algebra A then the quotient A/I is a Banach
algebra in the norm
| (A) |:= inf
JI
| A+J | (2.37)
and the multiplication
(A)(B) := (AB). (2.38)
Here : A A/I is the canonical projection. If A is unital then A/I is unital, with unit (I).
We omit the standard proof that A/I is a Banach space in the norm (2.37). As far as the
Banach algebra structure is concerned, rst note that (2.38) is well dened: when J
1
, J
2
I one
has
(A +J
1
)(B +J
2
) = (AB +AJ
2
+J
1
B +J
1
J
2
) = (AB) = (A)(B),
since AJ
2
+ J
1
B + J
1
J
2
I by denition of an ideal, and (J) = 0 for all J I. To prove (2.8),
observe that, by denition of the inmum, for given A A, for each > 0 there exists a J I
such that
| (A) | + | A +J | . (2.39)
20 2 ELEMENTARY THEORY OF C
-ALGEBRAS
For if such a J would not exist, the norm in A/I could not be given by (2.37). On the other hand,
for any J I it is clear from (2.37) that
| (A) |=| (A +J) | | A+J | . (2.40)
For A, B A choose > 0 and J
1
, J
2
I such that (2.39) holds for A, B, and estimate
| (A)(B) | = | (A +J
1
)(B +J
2
) |=| ((A + J
1
)(B +J
2
)) |
| (A +J
1
)(B +J
2
) | | A +J
1
| | B +J
2
|
(| (A) | +)(| (B) | +). (2.41)
Letting 0 yields | (A)(B) | | (A) | | (B) |.
When A has a unit, it is obvious from (2.38) that (I) is a unit in A/I. By (2.40) with A = I
one has | (I) | | I |= 1. On the other hand, from (2.8) with B = I one derives | (I) | 1.
Hence | (I) |= 1.
2.3 Commutative Banach algebras
We now assume that the Banach algebra A is commutative (that is, AB = BA for all A, B A).
Denition 2.3.1 The structure space (A) of a commutative Banach algebra A is the set of
all nonzero linear maps : A C for which
(AB) = (A)(B) (2.42)
for all A, B A. We say that such an is multiplicative.
In other words, (A) consists of all nonzero homomorphisms from A to C.
Proposition 2.3.2 Let A have a unit I.
1. Each (A) satises
(I) = 1; (2.43)
2. each (A) is continuous, with norm
| |= 1; (2.44)
hence
[(A)[ | A | (2.45)
for all A A.
The rst claim is obvious, since (IA) = (I)(A) = (A), and there is an A for which
(A) = 0 because is not identically zero.
For the second, we know from Lemma 2.2.4 that A z is invertible when [z[ >| A |, so that
(A z) = (A) z = 0, since is a homomorphism. Hence [(A)[ = [z[ for [z[ > A |, and
(2.45) follows.
Theorem 2.3.3 Let A be a unital commutative Banach algebra. There is a bijective correspon-
dence between (A) and the set of all maximal ideals in A, in that the kernel ker() of each
(A) is a maximal ideal I
1
=
2
because of (2.43).
We now show that every maximal ideal I of A is the kernel of some (A). Since I = A,
there is a nonzero B A which is not in I. Form
I
B
:= BA +J[, A A, J I.
This is clearly a left-ideal; since A is commutative, I
B
is even an ideal. Taking A = 0 we see
I I
B
. Taking A = I and J = 0 we see that B I
B
, so that I
B
= I. Hence I
B
= A, as I is
maximal. In particular, I I
B
, hence I = BA +J for suitable A A, J I. Apply the canonical
projection : A A/I to this equation, giving
(I) = I = (BA) = (B)(A),
because of (2.38) and (J) = 0. hence (A) = (B)
1
in A/I. Since B was arbitrary (though
nonzero), this shows that every nonzero element of A/I is invertible. By Corollary 2.2.6 this yields
A/I C, so that there is a homomorphism : A/I C. Now dene a map : A C by
(A) := ((A)). This map is clearly linear, since and are. Also,
(A)(B) = ((A))((B)) = ((A)(B)) = ((AB)) = (AB),
because of (2.38) and the fact that is a homomorphism.
Therefore, is multiplicative; it is nonzero because (B) = 0, or because (I) = 1. Hence
(A). Finally, I ker() since I = ker(); but if B / I we saw that (B) = 0, so that
actually I = ker().
By 2.3.2.2 we have (A) A
-topology,
on the dual B
-topology.
Proposition 2.3.4 The structure space (A) of a unital commutative Banach algebra A is com-
pact and Hausdor in the Gelfand topology.
The convergence
n
in the w
n
(A)
n
(B) (A)(B) = (
n
(A) (A))
n
(B) +(A)(
n
(B) (B)).
By (2.45) and the triangle inequality, the absolute value of the right-hand side is bounded by
| B | [
n
(A) (A)[+ | A | [
n
(B) (B)[.
All in all, when
n
in the w
-closed.
From (2.44) we have (A) A
1
(the unit ball in A
is w
-ALGEBRAS
We embed A in A
by A
A, where
-
topology on A
A
1
(O) = (A)[ (A) O, (2.47)
where A A and O is an open set in C.
Seen as a map from A to C((A)), the map A
A dened by (2.46) is called the Gelfand
transform.
For any compact Hausdor space X, we regard the space C(X) of all continuous functions on
X as a Banach space in the sup-norm dened by
| f |
:= sup
xX
[f(x)[. (2.48)
A basic fact of topology and analysis is that C(X) is complete in this norm. Convergence in the
sup-norm is the same as uniform convergence. Whats more, it is easily veried that C(X) is even
a commutative Banach algebra under pointwise addition and multiplication, that is,
(f +g)(x) := f(x) +g(x);
(fg)(x) := f(x)g(x). (2.49)
Hence the function 1
X
which is 1 for every x is the unit I. One checks that the spectrum of
f C(X) is simply the set of values of f.
We regard C((A)) as a commutative Banach algebra in the manner explained.
Theorem 2.3.5 Let A be a unital commutative Banach algebra.
1. The Gelfand transform is a homomorphism from A to C((A)).
2. The image of A under the Gelfand transform separates points in (A).
3. The spectrum of A A is the set of values of
A on (A); in other words,
(A) = (
A) =
| A | . (2.51)
The rst property immediately follows from (2.46) and (2.42). When
1
=
2
there is an A A
for which
1
(A) =
2
(A), so that
A(
1
) =
A(
2
). This proves 2.3.5.2.
If A G(A) (i.e., A is invertibe), then (A)(A
1
) = 1, so that (A) = 0 for all (A).
When A / G(A) the ideal I
A
:= AB[, B A does not contain I, so that it is contained in
a maximal ideal I (this conclusion is actually nontrivial, relying on the axiom of choice in the
guise of Hausdors maximality priciple). Hence by Theorem 2.3.3 there is a (A) for which
(A) = 0. All in all, we have showed that A G(A) is equivalent to (A) = 0 for all (A).
Hence A z G(A) i (A) = z for all (A). Thus the resolvent is
(A) = z C[ z = (A) (A). (2.52)
Taking the complement, and using (2.46), we obtain (2.50).
Eq. (2.51) then follows from (2.30), (2.31), (2.46), and (2.48).
2.3 Commutative Banach algebras 23
We now look at an example, which is included for three reasons: rstly it provides a concrete
illustration of the Gelfand transform, secondly it concerns a commutative Banach algebra which is
not a C
-algebra, and thirdly the Banach algebra in question has no unit, so the example illustrates
what happens to the structure theory in the absence of a unit. In this connection, let us note in
general that each (A) has a suitable extension to A
I
, namely
(A +I) := (A) +. (2.53)
The point is that remains multiplicative on A
I
, as can be seen from (2.22) and the denition
(2.42). This extension is clearly unique. Even if one does not actually extend A to A
I
, the existence
of shows that satises (2.45), since this property (which was proved for the unital case) holds
for , and therefore certainly for the restriction of to A.
Consider A = L
1
(R), with the usual linear structure, and norm
| f |
1
:=
R
dx[f(x)[. (2.54)
The associative product dening the Banach algebra structure is convolution, that is,
f g(x) :=
R
dy f(x y)g(y). (2.55)
Strictly speaking, this should rst be dened on the dense subspace C
c
(R), and subsequently be
extended by continuity to L
1
(R), using the inequality below. Indeed, using Fubinis theorem on
product integrals, we estimate
| f g |
1
=
R
dx[
R
dy f(x y)g(y)[
R
dy [g(y)[
R
dx[f(x y)[
=
R
dy [g(y)[
R
dx[f(x)[ =| f |
1
| g |
1
,
which is (2.8).
There is no unit in L
1
(R), since from (2.55) one sees that the unit should be Diracs delta-
function (i.e., the measure on R which assigns 1 to x = 0 and 0 to all other x), which does not lie
in L
1
(R).
We know from the discussion following (2.53) that every multiplicative functional (L
1
(R))
is continuous. Standard Banach space theory says that the dual of L
1
(R) is L
R
dxf(x) (x). (2.56)
The multiplicativity condition (2.42) then implies that (x + y) = (x) (y) for almost all
x, y R. This implies
(x) = exp(ipx) (2.57)
for some p C, and since is bounded (being in L
f(p) =
R
dxf(x)e
ipx
. (2.58)
Hence the Gelfand transform is nothing but the Fourier transform (more generally, many of the
integral transforms of classical analysis may be seen as special cases of the Gelfand transform).
The well-known fact that the Fourier transform maps the convolution product (2.55) into the
24 2 ELEMENTARY THEORY OF C
-ALGEBRAS
pointwise product is then a restatement of Theorem 2.3.5.1. Moreover, we see from 2.3.5.3 that
the spectrum (f) of f in L
1
(R) is just the set of values of its Fourier transform.
Note that the Gelfand transform is strictly a contraction, i.e., there is no equality in the bound
(2.51). Finally, the Riemann-Lebesgue lemma states that f L
1
(R) implies
f C
0
(R), which
is the space of continuous functions on R that go to zero when [x[ . This is an important
function space, whose denition may be generalized as follows.
Denition 2.3.6 Let X be a Hausdor space X which is locally compact (in that each point
has a compact neighbourhood). The space C
0
(X) consists of all continuous functions on X which
vanish at innity in the sense that for each > 0 there is a compact subset K X such that
[f(x)[ < for all x outside K.
So when X is compact one trivially has C
0
(X) = C(X). When X is not compact, the sup-norm
(2.48) can still be dened, and just as for C(X) one easily checks that C
0
(X) is a Banach algebra
in this norm.
We see that in the example A = L
1
(R) the Gelfand transform takes values in C
0
((A)). This
may be generalized to arbitrary commutative non-unital Banach algebras. The non-unital version
of Theorem 2.3.5 is
Theorem 2.3.7 Let A be a non-unital commutative Banach algebra.
1. The structure space (A) is locally compact and Hausdor in the Gelfand topology.
2. The space (A
I
) is the one-point compactication of (A).
3. The Gelfand transform is a homomorphism from A to C
0
((A)).
4. The spectrum of A A is the set of values of
A on (A), with zero added (if 0 is not already
contained in this set).
5. The claims 2 and 4 in Theorem 2.3.5 hold.
Recall that the one-point compactication
X of a non-compact topological space X is the
set X, whose open sets are the open sets in X plus those subsets of X whose complement
is compact in X. If, on the other hand,
X is a compact Hausdor space, the removal of some point
yields a locally compact Hausdor space X =
X` in the relative topology (i.e., the open
sets in X are the open sets in
X minus the point ), whose one-point compactication is, in turn,
X.
To prove 2.3.7 we add a unit to A, and note that
(A
I
) = (A) , (2.59)
where each (A) is seen as a functional on A
I
by (2.53), and the functional is dened by
(A +I) := . (2.60)
There can be no other elements of (A
I
), because the restriction of has a unique multiplicative
extension (2.53) to A
I
, unless it identically vanishes on (A). In the latter case (2.60) is clearly
the only multiplicative possibility.
By Proposition 2.3.4 the space (A
I
) is compact and Hausdor; by (2.60) one has
(A) = (A
I
)` (2.61)
as a set. In view of the paragraph following 2.3.7, in order to prove 2.3.7.1 and 2, we need to show
that the Gelfand topology of (A
I
) restricted to (A) coincides with the Gelfand topology of
(A) itself. Firstly, it is clear from (2.47) that any open set in (A) (in its own Gelfand topology)
is the restriction of some open set in (A
I
), because A A
I
. Secondly, for any A A, C, and
open set O C, from (2.53) we evidently have
(A
I
)[ (A +I) O` = (A)[ (A) O .
2.4 Commutative C
-algebras 25
(When does not lie in the set . . . on the left-hand side, one should here omit the `.)
With (2.47), this shows that the restriction of any open set in (A
I
) to (A) is always open in
the Gelfand topology of (A). This establishes 2.3.7.1 and 2.
It follows from (2.3.5) and (2.60) that
A() = 0 (2.62)
for all A A, which by continuity of
A leads to 2.3.7.3.
The comment preceding Theorem 2.2.3 implies 2.3.7.4. The nal claim follows from the fact
that it holds for A
I
.
2.4 Commutative C
-algebras
The Banach algebra C(X) considered in the previous section is more than a Banach algebra. Recall
Denition 2.1.9. The map f f
, where
f
(B +itI) |=| B
2
+t
2
| | B |
2
+t
2
.
Using (2.65) then yields
2
+ t | B |
2
for all t R. For > 0 this is impossible. For < 0
we repeat the argument with B B, nding the same absurdity. Hence = 0, so that (A)
is real when A = A
=| A | . (2.67)
For general A A we note that A
A |=|
A
A |
=|
A
A |
=|
A |
2
.
26 2 ELEMENTARY THEORY OF C
-ALGEBRAS
In the third equality we used
A
=
A
-algebra, so that (2.16) is satised in it. Hence (2.67) holds for all A A.
It follows that in (2.64) is injective, because if (A) = 0 for some A = 0, then would fail
to be an isometry. (A commutative Banach algebra for which the Gelfand transform is injective
is called semi-simple. Thus commutative C
-
algebra as explained above. A C
-
algebra as explained above. Then (C(X)) (equipped with the Gelfand topology) is homeomorphic
to X.
Each x X denes a linear map
x
: C(X) C by
x
(f) := f(x), which is clearly multiplica-
tive and nonzero. Hence x
x
denes a map E (for Evaluation) from X to (C(X)), given
by
E(x) : f f(x). (2.68)
Since a compact Hausdor space is normal, Urysohns lemma says that C(X) separates points on
X (i.e., for all x = y there is an f C(X) for which f(x) = f(y)). This shows that E is injective.
We now use the compactness of X and Theorem 2.3.3 to prove that E is surjective. The
maximal ideal I
x
:= I
x
in C(X) which corresponds to
x
(C(X)) is obviously
I
x
= f C(X)[ f(x) = 0. (2.69)
Therefore, when E is not surjective there exists a maximal ideal I C(X) which for each x
X contains at a function f
x
for which f
x
(x) = 0 (if not, I would contain an ideal I
x
which
thereby would not be maximal). For each x, the set O
x
where f
x
is nonzero is open, because
f is continuous. This gives a covering O
x
xX
of X. By compactness, there exists a nite
subcovering O
xi
i=1,...,N
. Then form the function g :=
N
i=1
[f
xi
[
2
. This function is strictly
positive by construction, so that it is invertible (note that f C(X) is invertible i f(x) = 0 for
all x X, in which case f
1
(x) = 1/f(x)). But I is an ideal, so that, with all f
xi
I (since all
f
x
I) also g I. But an ideal containing an invertible element must coincide with A (see the
comment after 2.2.9), contradicting the assumption that I is a maximal ideal.
Hence E is surjective; since we already found it is injective, E must be a bijection. It remains
to be shown that E is a homeomorphism. Let X
o
denote X with its originally given topology, and
write X
G
for X with the topology induced by E
1
. Since
f E = f by (2.68) and (2.46), and the
Gelfand topology on (C(X)) is the weakest topology for which all functions
f are continuous,
we infer that X
G
is weaker than X
o
(since f, lying in C(X
o
), is continuous). Here a topology T
1
is called weaker than a topology T
2
on the same set if any open set of T
1
contains an open set of
T
2
. This includes the possibility T
1
= T
2
.
Without proof we now state a result from topology.
Lemma 2.4.4 Let a set X be Hausdor in some topology T
1
and compact in a topology T
2
. If T
1
is weaker than T
2
then T
1
= T
2
.
2.4 Commutative C
-algebras 27
Since X
o
and X
G
are both compact and Hausdor (the former by assumption, and the lat-
ter by Proposition 2.3.4), we conclude from this lemma that X
0
= X
G
; in other words, E is a
homeomorphism. This concludes the proof of 2.4.3.
Proposition 2.4.3 shows that X as a topological space may be extracted from the Banach-
algebraic structure of C(X), up to homeomorphism. Hence if C(X) C(Y ) as a C
-algebra,
where Y is a second compact Hausdor space, then X Y as topological spaces. Given the
isomorphism A C(X) constructed above, a second isomorphism A C(Y ) is therefore only
possible if X Y . This proves the nal claim of Theorem 2.4.1.
The condition that a compact topological space be Hausdor is sucient, but not necessary
for the completeness of C(X) in the sup-norm. However, when X is not Hausdor yet C(X) is
complete, the map E may fail to be injective since in that case C(X) may fail to separate points
on X.
On the other hand, suppose X is locally compact but not compact, and consider A = C
b
(X);
this is the space of all continuous bounded functions on X. Equipped with the operations (2.48),
(2.49), and (2.63) this is a commutative C
:= A
+I. (2.70)
However, the straightforward norm (2.23) cannot be used, since it is not a C
-algebra.
1. The map : A B(A) given by
(A)B := AB (2.71)
establishes an isomorphism between A and (A) B(A).
2. When A has no unit, dene a norm on A
I
by
| A+I |:=| (A) +I |, (2.72)
where the norm on the right-hand side is the operator norm (2.2) in B(A), and I on the
right-hand side is the unit operator in B(A). With the operations (2.22) and (2.70), the
norm (2.72) turns A
I
into a C
| / | A |=| (A)
A
| A |
| (A) |;
in the last step we used (2.4) and | (A
/ | A |) |= 1. Hence
| (A) |=| A | . (2.73)
Being isometric, the map must be injective; it is clearly a homomorphism, so that we have proved
2.4.5.1.
It is clear from (2.22) and (2.70) that the map A+I (A) +I (where the symbol I on the
left-hand side is dened below (2.22), and the I on the right-hand side is the unit in B(A)) is a
28 2 ELEMENTARY THEORY OF C
-ALGEBRAS
morphism. Hence the norm (2.72) satises (2.15), because (2.8) is satised in B(A). Moreover, in
order to prove that the norm (2.72) satises (2.16), by Lemma 2.1.11 it suces to prove that
| (A) +I |
2
| ((A) +I)
(AB +B) | .
Here we used (2.16) in A. Using (2.71), the right-hand side may be rearranged as
| (B
)(A
+I)(A +I)B | | (B
) | | ((A) +I)
((A) +I) | | B | .
Since | (B
) |=| B
|=| B |= 1 by (2.73) and (2.17), and | B |= 1 also in the last term, the
inequality (2.74) follows by letting 0.
Hence the C
-algebra A
I
and an isometric (hence injective) morphism A A
I
, such that A
I
/A C.
The uniqueness of A
I
follows from Corollary 2.5.3 below. On the other hand, in view of the
fact that both (2.23) and (2.72) dene a norm on A
I
satisfying the claims of Proposition 2.2.1, we
conclude that the unital Banach algebra A
I
called for in that proposition is not, in general, unique.
In any case, having established the existence of the unitization of an arbitrary non-unital C
-
algebra, we see that, in particular, a commutative non-unital C
-algebraic setting;
the only nontrivial point compared to the situation for Banach algebras is the generalization of
Lemma 2.4.2. This now reads
Lemma 2.4.7 Let X be a locally compact Hausdor space, and regard C
0
(X) as a commutative
C
-subalgebra A of C
0
(X) which separates points on X, and is such that for each x X
there is an f A such that f(x) = 0, coincides with C
0
(X).
At the end of the day we then nd
Theorem 2.4.8 Let A be a commutative C
-subalgebra C
(A, I) is
commutative. In particular, when A is self-adjoint, C
-algebra.
2.5 Spectrum and functional calculus 29
1. The spectrum
A
(A) of A in A coincides with the spectrum
C
(A,I)
(A) of A in C
(A, I) (so
that we may unambiguously speak of the spectrum (A)).
2. The spectrum (A) is a subset of R.
3. The structure space (C
(A, I) is isomorphic
to C((A)). Under this isomorphism the Gelfand transform
A : (A) R is the identity
function id
(A)
: t t.
Recall (2.25). Let A G(A) be normal in A, and consider the C
-algebra C
(A, A
1
, I)
generated by A, A
1
, and I. One has (A
1
)
= (A
)
1
, and A, A
, A
1
, (A
)
1
and I all commute
with each other. Hence C
(A, A
1
, I) is commutative; it is the closure of the space of all polynomials
in A, A
, A
1
, (A
)
1
, and I. By Theorem 2.4.1 we have C
(A, A
1
, I) C(X) for some compact
Hausdor space X. Since A is invertible and the Gelfand transform (2.46) is an isomorphism,
A
is invertible in C(X) (i.e.,
A(x) = 0x for all x X). However, for any f C(X) that is nonzero
throughout X we have 0 <| f |
2
ff
1 pointwise, so that 0 1
X
| f |
2
ff < 1 pointwise,
hence
| 1
X
ff / | f |
2
< 1.
Here f
| f |
2
k=0
I
ff
| f |
2
k
. (2.75)
Hence
A
1
is a norm-convergent limit of a sequence of polynomials in
A and
A
. Gelfand trans-
forming this result back to C
(A, A
1
, I), we infer that A
1
is a norm-convergent limit of a sequence
of polynomials in A and A
. Hence A
1
lies in C
(A, A
1
, I) = C
(A, I).
Now replace A by Az, where z C. When A is normal Az is normal. So if we assume that
A z G(A) the argument above applies, leading to the conclusion that the resolvent
A
(A) in
A coincides with the resolvent
C
(A,I)
(A) in C
A
(A) =
C
(A,I)
(A).
According to Theorem 2.4.1, the function
A is real-valued when A = A
(A,I)
(A) is real, so that by the previous result (A) is real.
Finally, given the isomorphism C
(A, I))),
according to 2.3.5.3 the function
A is a surjective map from X to (A). We now prove injectivity.
When
1
,
2
X and
1
(A) =
2
(A), then, for all n N, we have
1
(A
n
) =
1
(A)
n
=
2
(A)
n
=
2
(A
n
)
by iterating (2.42) with B = A. Since also
1
(I) =
2
(I) = 1 by (2.43), we conclude by linearity
that
1
=
2
on all polynomials in A. By continuity (cf. 2.3.2.2) this implies that
1
=
2
on
C
. (2.77)
In particular, the norm of f(A) in C
-ALGEBRAS
Theorem 2.5.1.3 yields an isomorphism C((A)) C
-algebra A there is no
other norm in which A is a C
-algebra).
First assume A = A
). (2.78)
Since A
r(A
A). (2.79)
Since the spectrum is determined by the algebraic structure alone, (2.79) shows that the norm is
determined by the algebraic structure as well.
Note that Corollary 2.5.3 does not imply that a given
-algebra can be normed only in one way
so as to be completed into a C
-algebras satises
| (A) | | A |, (2.80)
and is therefore automatically continuous.
When z (A), so that (A z)
1
exists in A, then (A z) is certainly invertible in B, for
(2.18) implies that ((A z))
1
= ((A z)
1
). Hence (A) ((A)), so that
((A)) (A). (2.81)
Hence r((A)) r(A), so that (2.80) follows from (2.79).
For later use we note
Lemma 2.5.5 When : A B is a morphism and A = A
then
f((A)) = (f(A)) (2.82)
for all f C((A)) (here f(A) is dened by the continuous functional calculus, and so is f((A))
in view of (2.81)).
The property is true for polynomials by (2.18), since for those f has its naive meaning. For
general f the result then follows by continuity.
2.6 Positivity in C
-algebras
A bounded operator A B(H) on a Hilbert space H is called positive when (, A) 0 for
all H; this property is equivalent to A
= A and (A) R
+
, and clearly also applies to
closed subalgebras of B(H). In quantum mechanics this means that the expectation value of the
observable A is always positive.
Classically, a function f on some space X is positive simply when f(x) 0 for all x X. This
applies, in particular, to elements of the commutative C
-algebra C
0
(X) (where X is a locally
compact Hausdor space). Hence we have a notion of positivity for certain concrete C
-algebras,
which we would like to generalize to arbitrary abstract C
-algebra; it will, for example, play a central role in the proof of the
Gelfand Neumark theorem. In particular, one is interested in nding a number of equivalent
characterizations of positivity.
2.6 Positivity in C
-algebras 31
Denition 2.6.1 An element A of a C
A[ c by 2.3.5.3 and 2.5.1.3, so that | c1
(A)
A |
c. Gelfand transform-
ing back to C
(A, I), this implies | cI A | c for all c | A | by 2.5.2. Inverting this argument,
one sees that if | cI A | c for some c | A |, then (A) R
+
.
Use this with A A +B and c =| A | + | B |; clearly c | A +B | by 2.1.1.4. Then
| cI (A +B) | | (| A | A) | + | (| B | B) | c,
where in the last step we used the previous paragraph for A and for B separately. As we have
seen, this inequality implies A +B A
+
.
Finally, when A A
+
and A A
+
it must be that (A) = 0, hence A = 0 by (2.78) and
(2.30).
This is important, because a convex cone in a real vector space is equivalent to a linear partial
ordering, i.e., a partial ordering in which A B implies A+C B+C for all C and A B
for all R
+
. The real vector space in question is the space A
R
of all self-adjoint elements of A.
The equivalence between these two structures is as follows: given A
+
R
:= A
+
one denes A B if
B A A
+
R
, and given one puts A
+
R
= A A
R
[ 0 A.
For example, when A = A
B[ B A. (2.87)
32 2 ELEMENTARY THEORY OF C
-ALGEBRAS
When (A) R
+
and A = A
then
A A
R
is dened by the continuous functional calculus
for f =
, and satises
A
2
= A. Hence A
+
A
2
[ A A
R
. The opposite inclusion follows
from (2.76) and 2.5.1.2. This proves (2.86).
The inclusion A
+
B
, where A
+
, A
A
+
and A
+
A
= 0. Moreover, | A
| | A |.
Apply the continuous functional calculus with f = id
(A)
= f
+
f
, where id
(A)
(t), f
+
(t) =
maxt, 0, and f
.
We use this lemma to prove that B
B[ B A A
+
. Apply the lemma to A = B
B (noting
that A = A
). Then
(A
)
3
= A
(A
+
A
)A
= A
AA
= A
BA
= (BA
BA
.
Since (A
) R
+
because A
)
3
0.
Hence (BA
BA
0.
Lemma 2.6.6 If C
C A
+
for some C A then C = 0.
By (2.13) we can write C = D +iE, D, E A
R
, so that
C
C = 2D
2
+ 2E
2
CC
. (2.88)
Now for any A, B A one has
(AB) 0 = (BA) 0. (2.89)
This is because for z = 0 the invertibility of ABz implies the invertibility of BAz. Namely, one
computes that (BAz)
1
= B(AB z)
1
Az
1
I. Applying this with A C and B C
we
see that the assumption (C
C) R
implies (CC
) R
, hence (CC
) R
+
. By (2.88),
(2.86), and 2.6.2.2 we see that C
C 0, i.e., (C
C) R
+
, so that the assumption C
C A
+
now yields (C
C) = 0. Hence C = 0 by 2.6.2.3.
The last claim before the lemma therefore implies BA
= 0. As (A
)
3
= (BA
BA
= 0
we see that (A
)
3
= 0, and nally A
B = A
+
, which lies in A
+
.
An important consequence of (2.87) is the fact that inequalities of the type A
1
A
2
for
A
1
, A
2
A
R
are stable under conjugation by arbitrary elements B A, so that A
1
A
2
implies
B
A
1
B B
A
2
B. This is because A
1
A
2
is the same as A
2
A
1
0; by (2.87) there is an A
3
A
such that A
2
A
1
= A
3
A
3
. But clearly (A
3
B)
A
3
B 0, and this is nothing but B
AB B
A
2
B.
For example, replace A in (2.84) by A
A | A |
2
I. Applying the
above principle gives
B
AB | A |
2
B
B (2.90)
for all A, B A.
2.7 Ideals in C
-algebras
An ideal I in a C
-algebra A is a family I
, where
is some directed set (i.e., a set with a partial order and a sense in which ), with the
following properties:
2.7 Ideals in C
-algebras 33
1.
I
= I
(2.91)
and (I
| 1; (2.92)
2.
lim
| I
A A |= lim
| AI
A |= 0 (2.93)
for all A A.
For example, the C
-algebra C
0
(R) has no unit (the unit would be 1
R
, which does not vanish
at innity because it is constant), but an approximate unit may be constructed as follows: take
= N, and take I
n
to be a continuous function which is 1 on [n, n] and vanishes for [x[ > n +1.
One checks the axioms, and notes that one certainly does not have I
n
1
R
in the sup-norm.
Proposition 2.7.2 Every non-unital C
:=
i
A
i
A
i
. Clearly B
is
self-adjoint, and according to 2.6.4 and 2.6.2.2 one has (B) R
+
, so that n
1
I +B
is invertible
in A
I
. Hence we may form
I
:= B
(n
1
I +B
)
1
. (2.94)
Since B
is self-adjoint and B
)
1
), one has
I
= I
. Although (n
1
I+B
)
1
is computed in A
I
, so that it is of the form C+I for some C A
and C, one has I
= B
C +B
that (I
) [0, 1].
Putting C
i
:= I
A
i
A
i
, a simple computation shows that
i
C
i
C
i
= n
2
B
(n
1
I +B
)
2
. (2.95)
We now apply (2.77) with A B
and f(t) = n
2
t(n
1
+ t)
2
. Since f 0 and f assumes its
maximum at t = 1/n, one has sup
tR
+ [f(t)[ = 1/4n. As (B) R
+
, it follows that | f |
1/4n,
hence | n
2
B
(n
1
I + B
)
2
| 1/4n by (2.77), so that |
i
C
i
C
i
| 1/4n by (2.95). Lemma
2.6.3 then shows that | C
i
C
i
| 1/4n for each i = 1, . . . , n. Since any A A sits in some directed
subset of with n , it follows from (2.16) that
lim
| I
A A |
2
= lim
| (I
A A)
A A |= lim
| C
i
C
i
|= 0.
The other equality in (2.93) follows analogously.
Finally, when A is separable one may draw all A
i
occurring as elements of from a
countable dense subset, so that is countable.
The main properties of ideals in C
-algebra A.
1. If A I then A
-algebra is self-adjoint.
2. The quotient A/I is a C
:= (A
). (2.96)
34 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Note that (2.96) is well dened because of 2.7.3.1.
Put I
:= A
J I I
because I
is a C
. Take J I. Using
(2.16) and (2.91), we estimate
| J
|
2
=| (J I
J)(J
) |=| (JJ
JJ
) I
(JJ
JJ
) |
| (JJ
JJ
) | + | I
(JJ
JJ
) | | (J
J J
JI
) | + | I
| | (JJ
JJ
) | .
As we have seen, J
J I I
| J
|= 0. But I
lies in I I
, so certainly I
| A AI
| . (2.97)
It is obvious from (2.37) that
| A AI
| | (A) | . (2.98)
To derive the opposite inequality, add a unit I to A if necessary, pick any J I, and write
| AAI
|=| (A +J)(I I
) +J(I
I) | | A+J | | I I
| + | JI
J | .
Note that
| I I
| 1 (2.99)
by 2.7.1.1 and the proof of 2.6.2. The second term on the right-hand side goes to zero for ,
since J I. Hence
lim
| A AI
| | A +J | . (2.100)
For each > 0 we can choose J I so that (2.39) holds. For this specic J we combine (2.98),
(2.100), and (2.39) to nd
lim
| A AI
| | (A) | | A AI
| .
Letting 0 proves (2.97).
We now prove (2.16) in A/I. Successively using (2.97), (2.16) in A
I
, (2.99), (2.97), (2.38), and
(2.96), we nd
| (A) |
2
= lim
| A AI
|
2
= lim
| (A AI
(A AI
) |
= lim
| (I I
)A
A(I I
) | lim
| I I
| | A
A(I I
) | lim
| A
A(I I
) |
=| (A
A) |=| (A)(A
) |=| (A)(A)
| .
Lemma 2.1.11 then implies (2.16).
This seemingly technical result is very important.
Corollary 2.7.5 The kernel of a morphism between two C
-algebra is the kernel of some morphism. Hence every morphism has norm 1.
2.8 States 35
The rst claim is almost trivial, since (A) = 0 implies (AB) = (BA) = 0 for all B by
(2.18). Also, since is continuous (see 2.5.4) its kernel is closed.
The converse follows from Theorem 2.7.3, since I is the kernel of the canonical projection
: A A/I, where A/I is a C
B) | =| B
B |. Put A := B
B, noting that A
-algebras is closed. In
particular, (A) is a C
-subalgebra of B.
Dene : A/ ker() B by ([A]) = (A), where [A] is the equivalence class in A/ ker()
of A A. By the theory of vector spaces, is a vector space isomorphism between A/ ker()
and (A), and = . In particular, is injective. According to 2.7.5 and 2.7.3.2, the space
A/ ker() is a C
-algebra.
2.8 States
The notion of a state in the sense dened below comes from quantum mechanics, but states play
a central role in the general theory of abstract C
-algebras also.
Denition 2.8.1 A state on a unital C
B) =| B |
2
0, and normalization is obvious
from (I) = (, ) = 1.
Theorem 2.8.2 The state space of A = C(X) consists of all probability measures on X.
By the Riesz theorem of measure theory, each positive linear map : C(X) C is given by a
regular positive measure
(X) = 1, so
that
is a probability measure.
The positivity of with 2.6.4 implies that (A, B)
:= (A
B)[
2
(A
A)(B
B), (2.103)
which will often be used. Moreover, for all A A one has
(A
) = (A), (2.104)
as (A
) = (A
I) = (A, I)
= (I, A)
= (A).
Partly in order to extend the denition of a state to non-unital C
-algebras, we have
36 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Proposition 2.8.3 A linear map : A C on a unital C
B)(I) (I)
2
| B
B |= (I)
2
| B |
2
.
Hence | | (I). Since the upper bound is reached by B = I, we have (2.105).
To prove the converse claim, we rst note that the argument around (2.65) may be copied,
showing that is real on A
R
. Next, we show that A 0 implies (A) 0. Choose s > 0 small
enough, so that | I sA | 1. Then (assuming = 0)
1 | I sA |=
| |
(I)
| I sA |
[(I sA)[
(I)
.
Hence [(I) s(A)[ (I), which is only possible when (A) 0.
We now pass to states on C
+
A
+iA
+
iA
, (2.106)
where A
+
etc. are positive. Since | A
| | A | and | A
| | A | by (2.13), we have | B | | A |
for B = A
+
, A
, A
+
, or A
+
) | + | O(A
) | + | O(A
+
) | + | O(A
) | 4C | A | .
Now assume that O is not bounded; by the previous argument it is not bounded on A
+
, so that
for each n N there is an A
n
A
+
1
so that | O(A
n
) | n
3
(here A
+
1
consists of all A A
+
with
| A | 1). The series
n=0
n
2
A
n
obviously converges to some A A
+
. Since O is positive, we
have O(A) n
2
O(A
n
) 0 for each n. Hence by (2.85)
| O(A) | n
2
| O(A
n
) | n
for all n N, which is impossible since | O(A) | is some nite number. Thus O is bounded on
A
+
, and therefore on A by the previous paragraph.
Choosing B = C, we see that a state on a unital C
-algebras; Proposition 2.8.5 then provides an alternative proof of 2.8.3.1. Hence in the
non-unital case we may replace the normalization condition in 2.8.1 as follows.
Denition 2.8.6 A state on a C
I
(A +I) := (A) +. (2.107)
This obviously satises (2.101); it remains to prove positivity.
Since a state on A is bounded by 2.8.5, we have [(AAI
I
((A +I)
(A +I)) [(A) +[
2
0.
Hence is positive by (2.87).
There are lots of states:
Lemma 2.8.8 For every A A and a (A) there is a state
a
on A for which (A) = a. When
A = A
a
of to A of norm 1. By 2.8.3.2
a
is a state, which clearly satises
a
(A) =
a
(A) = a.
Since (A) is closed by 2.2.3.2, there is an a (A) for which r(A) = [a[. For this a one has
[(A)[ = [a[ = r(A) =| A | by (2.78).
An important feature of a state space o(A) is that it is a convex set. A convex set C in a
vector space ( is a subset of 1 such that the convex sum v + (1 )w belongs to ( whenever
v, w ( and [0, 1]. Repeating this process, it follows that
i
p
i
v
i
belongs to ( when all
p
i
0 and
i
p
i
= 1, and all v
i
(. In the unital case it is clear that o(A) is convex, since both
positivity and normalization are clearly preserved under convex sums. In the non-unital case one
arrives at this conclusion most simply via 2.8.7.
We return to the unital case. Let o(A) be the state space of a unital C
-algebra A. We saw
in 2.8.3 that each element of o(A) is continuous, so that o(A) A
. Since w
-limits obviously
preserve positivity and normalization, we see that o(A) is closed in A
by 2.8.3.1, so that
o(A) is compact in the (relative) w
is
of the form (
+) = c
1
1
+ c
2
2
. Positive elements of C C are fo the form
+ with 0
and 0, so that a positive functional must have c
1
0 and c
2
0. Finally, since I = 1
+1,
normalization yields c
1
+ c
2
= 1. We conclude that o(C C) may be identied with the interval
[0, 1].
Now consider A = M
2
(C). We identify M
2
(C) with its dual through the pairing (A) =
Tr A. It follows that o(A) consists of all positive 2 2 matrices with Tr = 1; these are the
density matrices of quantum mechanics. To identify o(A) with a familiar compact convex set, we
parametrize
=
1
2
1 +x y +iz
y iz 1 x
, (2.108)
where x, y, z R. The positivity of this matrix then corresponds to the constraint x
2
+y
2
+z
2
1.
Hence o(M
2
(C)) is the unit ball in R
3
.
38 2 ELEMENTARY THEORY OF C
-ALGEBRAS
2.9 Representations and the GNS-construction
The material of this section explains how the usual Hilbert space framework of quantum mechanics
emerges from the C
-algebraic setting.
Denition 2.9.1 A representation of A on a Hilbert space H is a (complex) linear map :
A B(H) satisfying
(A B) = (A)(B);
(A
) = (A)
(2.109)
for all A, B A.
A representation is automatically continuous, satisfying the bound
| (A) | | A | . (2.110)
This is because is a morphism; cf. (2.80). In particular, | (A) |=| A | when is faithful by
Lemma 2.7.6.
There is a natural equivalence relation in the set of all representations of A: two representa-
tions
1
,
2
on Hilbert spaces H
1
, H
2
, respectively, are called equivalent if there exists a unitary
isomorphism U : H
1
H
2
such that U
1
(A)U
=
2
(A) for all A A.
The map (A) = 0 for all A A is a representation; more generally, such trivial may occur
as a summand. To exclude this possibility, one says that a representation is non-degenerate if 0
is the only vector annihilated by all representatives of A.
A representation is called cyclic if its carrier space H contains a cyclic vector for ; this
means that the closure of (A) (which in any case is a closed subspace of H) coincides with H.
Proposition 2.9.2 Any non-degenerate representation is a direct sum of cyclic representations.
The proof uses a lemma which appears in many other proofs as well.
Lemma 2.9.3 Let M be a
-algebra in B(H), take a nonzero vector H, and let p be the
projection onto the closure of M. Then p M
Ap = 0 with p
= I p; this reads
Ap = pAp. When A = A
then
(Ap)
= pA = (pAp)
= pAp = Ap,
so that [A, p] = 0. By (2.13) this is true for all A M.
Apply this lemma with M = (A); the assumption of non-degeneracy guarantees that p is
nonzero, and the conclusion implies that A p(A) denes a subrepresentation of A on pH. This
subrepresentation is clearly cyclic, with cyclic vector . This process may be repeated on p
H,
etc.
If is a non-degenerate representation of A on H, then any unit vector H denes a state
o(A), referred to as a vector state relative to , by means of (2.102). Conversely, from any
state o(A) one can construct a cyclic representation
on a Hilbert space H
with cyclic
vector
in the following way. We restrict ourselves to the unital case; the general case follows
by adding a unit to A and using 2.8.7.
Construction 2.9.4 1. Given o(A), dene the sesquilinear form ( , )
0
on A by
(A, B)
0
:= (A
B). (2.111)
Since is a state, hence a positive functional, this form is positive semi-denite (this means
that (A, A)
0
0 for all A). Its null space
^
= A A[ (A
A) = 0 (2.112)
is a closed left-ideal in A.
2.9 Representations and the GNS-construction 39
2. The form ( , )
0
projects to an inner product ( , )
. If V : A A/^
:= (A, B)
0
. (2.113)
The Hilbert space H
by
is continuous. Hence
by continuous
extension of (2.114), where it satises (2.109).
4. The cyclic vector is dened by
= V I, so that
(
(A)
) = (A) A A. (2.115)
We now prove the various claims made here. First note that the null space ^
of ( , )
0
can be
dened in two equivalent ways;
^
:= A A[ (A, A)
0
= 0 = A A[ (A, B)
0
= 0 B A. (2.116)
The equivalence follows from the Cauchy-Schwarz inequality (2.103). The equality (2.116) implies
that ^
to
A/^
of H
.
To prove that
is continuous on A/^
, we compute |
(A) |
2
for = V B, where
A, B A. By (2.113) and step 2 above, one has |
(A) |
2
= (B
AA B) | A |
2
(B
B). But (B
B) =| |
2
, so that
|
(A) | | A | (2.117)
follows from (2.3).
For later use we mention that the GNS-construction yields
(
(A)
(B)
) = (A
B). (2.118)
Putting B = A yields
|
(A)
|
2
= (A
A), (2.119)
which may alternatively be derived from (2.115) and the fact that
is a representation.
Proposition 2.9.5 If ((A), H) is cyclic then the GNS-representation (
(A), H
) dened by any
vector state (corresponding to a cyclic unit vector H) is unitarily equivalent to ((A), H).
This is very simple to prove: the operator U : H
(A)
by U
(A)
(A)
to UH
and .
Corollary 2.9.6 If the Hilbert spaces H
1
, H
2
of two cyclic representations
1
,
2
each contain a
cyclic vector
1
H
1
,
2
H
2
, and
1
(A) := (
1
,
1
(A)
1
) = (
2
,
2
(A)
2
) =:
2
(A)
for all A A, then
1
(A) and
2
(A) are equivalent.
By 2.9.5 the representation
1
is equivalent to the GNS-representation
1
, and
2
is equivalent
to
2
. On the other hand,
1
and
2
are induced by the same state, so they must coincide.
40 2 ELEMENTARY THEORY OF C
-ALGEBRAS
2.10 The Gelfand-Neumark theorem
One of the main results in the theory of C
-algebras is
Theorem 2.10.1 A C
.
Theorem 2.10.1 then follows by taking H = H
u
; the desired isomorphism is
u
. To prove that
u
is injective, suppose that
u
(A) = 0 for some A A. By denition of a direct sum, this implies
(A)
= 0, hence |
(A)
|
2
= 0; by (2.119) this means
(A
-algebra M
n
(A) for a given C
k
M
ik
N
kj
, with the dierence that one now multiplies elements of A rather than
complex numbers. In particular, the order has to be taken into account. The involution in M
n
(A)
is, of course, given by (M
)
ij
= M
ji
, in which the involution in A replaces the usual complex
conjugation in C. One may identify M
n
(A) with A M
n
(C) in the obvious way.
When is a faithful representation of A (which exists by Theorem 2.10.1), one obtains a faithful
realization
n
of M
n
(A) on HC
n
, dened by linear extension of
n
(M)v
i
:= (M
ij
)v
j
; we here
look at elements of H C
n
as n-tuples (v
1
, . . . , v
n
), where each v
i
H. The norm | M | of
M M
n
(A) is then simply dened to be the norm of
n
(M). Since
n
(M
n
(A)) is a closed
-
algebra in B(H C
n
) (because n < ), it is obvious that M
n
(A) is a C
B in M
n
(A), then
(A) = (B)
(B) B(H
.
There exists a Hilbert space H
, a representation
of A on H
(with W
(O(A)) = W
(A)W A A. (2.120)
Equivalently, with p := WW
),
H
:= pH
, and U :
H
to H
to
H
, so that U
is unitary, one has
U
(O(A))U
1
= p
(A)p. (2.121)
The proof consists of a modication of the GNS-construction. It uses the notion of a partial
isometry. This is a linear map W : H
1
H
2
between two Hilbert spaces, with the property that
H
1
contains a closed subspace /
1
such that (W, W)
2
= (, )
1
for all , /
1
, and W = 0
on /
1
. Hence W is unitary from /
1
to W/
1
. It follows that WW
= [/
2
] and W
W = [/
1
] are
projections onto the image and the kernel of W, respectively.
We denote elements of H
.
Construction 2.11.3 1. Dene the sesquilinear form ( , )
0
on AH
0
:= (v,
(O(A
B))w)
. (2.122)
Since O is completely positive, this form is positive semi-denite; denote its null space by
^
.
2. The form ( , )
0
projects to an inner product ( , )
on AH
/^
. If V
: AH
AH
/^
(A v), V
(B w))
:= (A v, B w)
0
. (2.123)
The Hilbert space H
is the closure of A H
/^
/^
by linear extension of
(A)V
(B w) := V
(A)^
(A) | | A |, (2.125)
so that
(A
) =
(A)
.
4. The map W : H
, dened by
Wv := V
I v (2.126)
is a partial isometry. Its adjoint W
: H
A v =
(O(A))v, (2.127)
from which the properties W
i,j
(A
i
v
i
, A
j
v
j
)
0
=
i,j
(v
i
,
(O(A
i
A
j
))v
j
)
. (2.128)
42 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Now consider the element A of M
n
(A) with matrix elements A
ij
= A
i
A
j
. Looking in a faithful
representation
n
as explained above, one sees that
(z, Az) =
i,j
(z
i
, (A
i
A
j
)z
j
) =
i,j
((A
i
)z
i
, (A
j
)z
j
) =| Az |
2
0
where Az =
i
A
i
z
i
. Hence A 0. Since O is completely positive, it must be that B, dened by
its matrix elements B
ij
:= O(A
i
A
j
), is positive in M
n
(B). Repeating the above argument with
A and replaced by B and
AI
n
| A |
2
I
n
with the matrix B, whose rst row is (B
1
, . . . , B
n
),
and which has zeros everywhere else; the adjoint B
1
, . . . , B
n
)
T
, and all other entries zero. This leads to 0 B
AB | A |
2
B
B. Since O is
completely positive, one has O
n
(B
AB) | A |
2
O
n
(B
(B)
and any vector (v
1
, . . . v
n
) H
C
n
one has
i,j
(v
i
,
(O(B
i
A
AB
j
))v
j
) | A |
2
i,j
(v
i
,
(O(B
i
B
j
))v
j
). (2.129)
With =
i
V
B
i
v
i
, from (2.122), (2.124), and (2.129) one then has
|
(A) |
2
=
i,j
(AB
i
v
i
, AB
j
v
j
)
0
=
i,j
(v
i
,
(O(B
i
A
AB
j
))v
j
)
| A |
2
i,j
(v
i
,
(O(B
i
B
j
))v
j
)
=| A |
2
i,j
(B
i
v
i
, B
j
v
j
)
0
=| A |
2
(V
i
B
i
v
i
, V
j
B
j
v
j
)
=| A |
2
| |
2
.
To show that W is a partial isometry, use the denition to compute
(Wv, Ww)
= (V
I v, V
I w)
= (I v, I w)
0
= (v, w)
,
where we used (2.122) and O(I) = I.
To check (2.127), one merely uses the denition of the adjoint, viz. (w, W
= (Ww, )
for
all w H
and H
(A)Wv = W
(A)V
(I v) = W
(A v) =
(O(A))v.
Being a partial isometry, one has p = WW
(O(A))U
1
= W
(O(A))W
= WW
(A)WW
= p
(A)p.
When O fails to preserve the unit, the above construction still applies, but W is no longer a
partial isometry; one rather has | W |
2
=| O(I) |. Thus it is no longer possible to regard H
as a
subspace of H
.
If A and perhaps B are non-unital the theorem holds if O can be extended (as a positive
map) to the unitization of A, such that the extension preserves the unit I (perhaps relative to the
unitization of B). When the extension exists but does not preserve the unit, one is in the situation
of the previous paragraph.
The relevance of Stinesprings theorem for quantum mechanics stems from the following result.
2.11 Complete positivity 43
Proposition 2.11.4 Let A be a commutative unital C
i
f
i
(x)M
i
for f
i
C(X) and M
i
M
n
(C), and the
sum is nite, are dense in C(X, M
n
(C)).
2. Such F is positive i all f
i
and M
i
are positive.
3. Positive elements G of C(X, M
n
(C)) can be norm-approximated by positive Fs, i.e., when
G 0 there is a sequence F
k
0 such that lim
k
F
k
= G.
4. O
n
(F) is positive when F is positive.
5. O
n
is continuous.
6. If F
k
G 0 in C(X, M
n
(C)) then O(G) = lim
k
O(F
k
) is a norm-limit of positive elements,
hence is positive.
We now prove each of these claims.
1. Take G C(X, M
n
(C)) and pick > 0. Since G is continuous, the set
O
x
:= y X, | G(x) G(y) | <
is open for each x X. This gives an open cover of X, which by the compactness of X
has a nite subcover O
x1
, . . . O
x
l
. A partition of unity subordinate to the given cover
is a collection of continuous positive functions
i
C(X), where i = 1, . . . , l, such that the
support of
i
lies in O
xi
and
l
i=1
i
(x) = 1 for all x X. Such a partition of unity exists.
Now dene F
l
C(X, M
n
(C)) by
F
l
(x) :=
l
i=1
i
(x)G(x
i
). (2.130)
Since | G(x
i
) G(x) | < for all x O
xi
, one has
| F
l
(x) G(x) |=|
l
i=1
i
(x)(G(x
i
) G(x)) |
l
i=1
i
(x) | G(x
i
) G(x) | <
l
i=1
i
(x) = .
Here the norm is the matrix norm in M
n
(C). Hence
| F
l
G |= sup
xX
| F
l
(x) G(x) | < .
2. An element F C(X, M
n
(C)) is positive i F(x) is positive in M
n
(C) for each x X. In
particular, when F(x) = f(x)M for some f C(X) and M M
n
(C) then F is positive i
f is positive in C(X) and M is positive in M
n
(C). By 2.6.2.2 we infer that F dened by
F(x) =
i
f
i
(x)M
i
is positive when all f
i
and M
i
are positive.
3. When G in item 1 is positive then each G(x
i
) is positive, as we have just seen.
4. On F as specied in 2.11.4.1 one has O
n
(F) =
i
O(f
i
) M
i
. Now each operator B
i
M
is positive in M
n
(B) when B
i
and M are positive (as can be checked in a faithful repre-
sentation). Since O is positive, it follows that O
n
maps each positive element of the form
F =
i
f
i
M
i
into a positive member of M
n
(B).
5. We know from 2.8.5 that O is continuous; the continuity of O
n
follows because n < .
6. A norm-limit A = lim
n
A
n
of positive elements in a C
n
B
n
, and limB
n
= B exist because of (2.16). Finally, A = B
B by continuity
of multiplication, i.e., by (2.15).
44 2 ELEMENTARY THEORY OF C
-ALGEBRAS
2.12 Pure states and irreducible representations
We return to the discussion at the end of 2.8. One sees that the compact convex sets in the
examples have a natural boundary. The intrinsic denition of this boundary is as follows.
Denition 2.12.1 An extreme point in a convex set K (in some vector space) is a member
of K which can only be decomposed as
=
1
+ (1 )
2
, (2.131)
(0, 1), if
1
=
2
= . The collection
e
K of extreme points in K is called the extreme
boundary of K. An extreme point in the state space K = o(A) of a C
-algebra A is called a
pure state. A state that is not pure is called a mixed state.
When K = o(A) is a state space of a C
-algebras that the qualier closed may be omitted from this denition, but we will not prove
this. Clearly, the dening representation
d
of the matrix algebra M
N
on C
N
is irreducible. In
the innite-dimensional case, the dening representations
d
of B(H) on H is irreducible as well.
Proposition 2.12.3 Each of the following conditions is equivalent to the irreducibility of (A) on
H:
1. The commutant of (A) in B(H) is I [ C; in other words, (A)
= B(H) (Schurs
lemma);
2. Every vector in H is cyclic for (A) (recall that this means that (A) is dense in H).
The commutant (A)
is a
-algebra in B(H), so when it is nontrivial it must contain a self-
adjoint element A which is not a multiple of I. Using Theorem 2.14.3 below and the spectral
theorem, it can be shown that the projections in the spectral resolution of A lie in (A)
if A does.
Hence when (A)
= CI.
Conversely, when (A)
= CI
irreducible.
When there exists a vector Hfor which (A) is not dense in H, we can form the projection
onto the closure of (A). By Lemma 2.9.3, with M = (A), this projection lies in (A)
, so that
cannot be irreducible by Schurs lemma. Hence irreducible every vector cyclic. The
converse is trivial.
We are now in a position to answer the question posed before 2.12.2.
2.12 Pure states and irreducible representations 45
Theorem 2.12.4 The GNS-representation
(A)
by Schurs
lemma. Let
. If p
= 0 then Ap
= pA
= 0 for all A A,
so that p = 0 as
is cyclic. Similarly, p
, where and
/ | p
|,
:= p
/ | p
|, and =| p
|
2
. Hence cannot be pure. This proves pure
irreducible.
In the opposite direction, suppose
A) (A
B)[
2
2
1
(A
A)
1
(B
B) (A
A)(B
B) (2.132)
for all A, B. This allows us to dene a quadratic form (i.e., a sesquilinear map)
Q on
(A)
by
Q(
(A)
(B)
) :=
1
(A
B). (2.133)
This is well dened: when
(A
1
)
(A
2
)
then ((A
1
A
2
)
(A
1
A
2
)) = 0 by (2.119),
so that
[
Q(
(A
1
)
(B)
)
Q(
(A
2
)
(B)
)[
2
= [
1
((A
1
A
2
)
B)[
2
0
by (2.132); in other words,
Q(
(A
1
)
(B)
) =
Q(
(A
2
)
(B)
). Similarly for B.
Furthermore, (2.132) and (2.119) imply that
Q is bounded in that
[
Q(, )[ C | | | |, (2.134)
for all ,
(A)
by continuity.
Moreover, one has
Q(, ) =
Q(, ) (2.135)
by (2.104) with A A
B and
1
.
Lemma 2.12.5 Let a quadratic form
Q on a Hilbert space H be bounded, in that (2.134) holds
for all , H, and some constant C 0. There is a bounded operator Q on H such that
such
that
(
(A)
, Q
(B)
) =
1
(A
B). (2.136)
It is the immediate from (2.109) that [Q,
(A)
; since
is
irreducible one must have Q = tI for some t R; hence (2.136), (2.133), and (2.118) show that
1
is proportional to , and therefore equal to by normalization, so that is pure.
From 2.9.5 we have the
Corollary 2.12.6 If ((A), H) is irreducible then the GNS-representation (
(A), H
) dened by
any vector state (corresponding to a unit vector H) is unitarily equivalent to ((A), H).
Combining this with 2.12.4 yields
46 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Corollary 2.12.7 Every irreducible representation of a C
-algebra C
0
(X) (equipped with the
relative w
-topology) is homeomorphic to X.
In view of Proposition 2.4.3 and Theorems 2.4.1 and 2.4.8, we merely need to establish a
bijective correspondence between the pure states and the multiplicative functionals on C
0
(X).
The case that X is not compact may be reduced to the compact case by passing from A = C
0
(X)
to A
I
= C(
X); cf. 2.4.6 and 2.3.7 etc. This is possible because the unique extension of a pure
state on C
0
(X) to a state on C(
X) guaranteed by 2.8.7 remains pure. Moreover, the extension of
a multiplicative functional dened in (2.53) coincides with the extension
I
of a state dened in
(2.107), and the functional in (2.60) clearly denes a pure state.
Thus we put A = C(X). Let
x
(C(X)) (cf. the proof of 2.4.3), and suppose a functional
satises 0
x
. Then ker(
x
) ker(), and ker() is an ideal. But ker(
x
) is a maximal
ideal, so when = 0 it must be that ker(
x
) = ker(). Since two functionals on any vector space
are proportional when they have the same kernel, it follows from 2.12.8 that
x
is pure.
Conversely, let be a pure state, and pick a g C(X) with 0 g 1
X
. Dene a functional
g
on C(X) by
g
(f) := (fg). Since (f)
g
(f) = (f(1 g)), and 0 1 g 1
X
, one
has 0
g
. Hence
g
= t for some t R
+
by 2.12.8. In particular, ker(
g
) = ker(). It
follows that when f ker(), then fg ker() for all g C(X), since any function is a linear
combination of functions for which 0 g 1
X
. Hence ker() is an ideal, which is maximal because
the kernel of a functional on any vector space has codimension 1. Hence is multiplicative by
Theorem 2.3.3.
It could be that no pure states exist in o(A); think of an open convex cone. It would follow that
such a C
-algebra has lots of pure states. For example, we may now rene
Lemma 2.8.8 as follows
2.13 The C
a
(A) = a. There exists a pure state such that [(A)[ =| A |.
We extend the state in the proof of 2.8.8 to C
. We
may further restrict this direct sum by dening two states to be equivalent if the corresponding
GNS-representations are equivalent, and taking only one pure state in each equivalence class. Let
us refer to the ensuing set of pure states as [{(A)]. We then have
A
r
(A) :=
[P(A)]
(A). (2.138)
It is obvious that the proof of 2.10.1 still goes through.
The simplest application of this renement is
Proposition 2.12.12 Every nite-dimensional C
-algebra M
n
(C) of n n matrices
to innite-dimensional Hilbert spaces H is the C
a
(A), which excludes the possibility that (A) and
a
(A) are equivalent (as the spectrum is
invariant under unitary transformations).
Another argument against B(H) is that it is non-separable in the nom-topology even when H
is separable. The appropriate generalization of M
n
(C) to an innite-dimensional Hilbert space H
turns out to be the C
-algebra B
0
(H) of compact operator on H. In non-commutative geometry
elements of this C
-
algebra.
48 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Denition 2.13.1 Let H be a Hilbert space. The
-algebra B
f
(H) of nite-rank operators on
H is the (nite) linear span of all nite-dimensional projections on H. In other words, an operator
A B(H) lies in B
f
(H) when AH := A[ H is nite-dimensional.
The C
-algebra B
0
(H) of compact operators on H is the norm-closure of B
f
(H) in B(H);
in other words, it is the smallest C
-algebra B
0
(H) is an ideal in B(H).
3. If A B
0
(H) then AB
1
is compact in H (with the norm-topology). Here B
1
is the unit ball
in H, i.e., the set of all H with | | 1.
Firstly, for any sequence (or net) A
n
B
f
(H) we may choose a unit vector
n
(A
n
H)
.
Then (A
n
I) = , so that | (A
n
I) |= 1. Hence sup
=1
| (A
n
I) | 1, hence
| A
n
I | 0 is impossible by denition of the norm (2.3) in B(H) (hence in B
0
(H)).
Secondly, when A B
f
(H) and B B(H) then AB B
f
(H), since ABH = AH. But since
BA = (A
, and B
f
(H) is a
-algebra, one has A
B
f
(H) and hence BA B
f
(H).
Hence B
f
(H) is an ideal in B(H), save for the fact that it is not norm-closed (unless H has nite
dimension). Now if A
n
A then A
n
B AB and BA
n
BA by continuity of multiplication in
B(H). Hence B
0
(H) is an ideal by virtue of its denition.
Thirdly, note that the weak topology on H(in which
n
i (,
n
) (, ) for all H)
is actually the w
-topology under the duality of H with itself given by the Riesz-Fischer theorem.
Hence the unit ball B
1
is compact in the weak topology by the Banach-Alaoglu theorem. So if we
can show that A B
0
(H) maps weakly convergent sequences to norm-convergent sequences, then
A is continuous from H with the weak topology to H with the norm-topology; since compactness
is preserved under continuous maps, it follows that AB
1
is compact.
Indeed, let
n
in the weak topology, with |
n
|= 1 for all n. Since
| |
2
= (, ) = lim
n
(,
n
) | | |
n
|= ,
one has | | 1. Given > 0, choose A
f
B
f
(H) such that | A A
f
| < /3, and put
p := [A
f
H], the nite-dimensional projection onto the image of A
f
. Then
| A
n
A |=| (A A
f
)
n
+ (A A
f
) +A
f
(
n
) |
1
3
+
1
3
+ | A
f
| | p(
n
) | .
Since the weak and the norm topology on a nite-dimensional Hilbert space coincide, they coincide
on pH, so that we can nd N such that | p(
n
) | < /3 for all n > N. Hence | A
n
A |
< .
Corollary 2.13.3 A self-adjoint operator A B
0
(H) has an eigenvector
a
with eigenvalue a
such that [a[ =| A |.
Dene f
A
: B
1
R by f
A
() :=| A |
2
. When
n
weakly with |
n
|= 1, then
[f
A
(
n
)f
A
()[ = [(
n
, A
A(
n
))(
n
, A
A)[ | A
A(
n
) | +[(
n
, A
A)[.
The rst term goes to zero by the proof of 2.13.2.3 (noting that A
A B
0
(H)), and the second goes
to zero by denition of weak convergence. Hence f
A
is continuous. Since B
1
is weakly compact,
f
A
assumes its maximum at some
a
. This maximum is | A |
2
by (2.3). Now the Cauchy-
Schwarz inequality with = 1 gives | A |
2
= (, A
A) | A
A |, with equality i A
A is
proportional to . Hence when A
= A the property | A |
2
=| A
a
|
2
with |
a
|= 1 implies
A
2
a
= a
2
a
, where a
2
=| A |
2
. The spectral theorem or the continuous functional calculus with
f(A
2
) =
A
2
= A implies A
a
= a
a
. Clearly [a[ =| A |.
2.13 The C
i
a
i
[
i
] (norm-convergent
sum), where each eigenvalue a
i
has nite multiplicity. Ordering the eigenvalues so that a
i
a
j
when i > j, one has lim
i
[a
i
[ = 0. In other words, the set of eigenvalues is discrete, and can
only have 0 as a possible accumulation point.
This ordering is possible because by 2.13.3 there is a largest eigenvalue.
Let A B
0
(H) be self-adjoint, and let p be the projection onto the closure of the linear
span of all eigenvectors of A. As in Lemma 2.9.3 one sees that [A, p] = 0, so that (pA)
= pA.
Hence p
A |= 0 2.13.3, which
implies that A restricted to p
H are eigenvectors
with eigenvalue zero. This contradicts the denition of p
H unless p
H = 0. This proves A
compact and self-adjoint A diagonalizable.
Let A be compact and self-adjoint, hence diagonalizable. Normalize the eigenvectors
i
:=
ai
to unit length. Then lim
i
(,
i
) = 0 for all H, since the
i
form a basis, so that
(, ) =
i
[(,
i
)[
2
, (2.139)
which clearly converges. Hence
i
0 weakly, so | A
i
|= [a
i
[ 0 by (the proof of)
2.13.2.3. Hence lim
i
[a
i
[ = 0. This proves A compact and self-adjoint A diagonalizable
with lim
i
[a
i
[ = 0.
Let now A be self-adjoint and diagonalizable, with lim
i
[a
i
[ = 0. For N < and H
one then has
| (A
N
i=1
a
i
[
i
]) |
2
=|
i=N+1
a
i
(
i
, )
i
|
2
i=N+1
[a
i
[
2
[(,
i
)[
2
[a
N
[
2
i=N+1
[(,
i
)[
2
.
Using (2.139), this is [a
N
[
2
(, ), so that lim
N
| A
N
i=1
a
i
[
i
] |= 0, because lim
N
[a
N
[ =
0. Since the operator
N
i=1
a
i
[
i
] is clearly of nite rank, this proves that A is compact. Hence
A self-adjoint and diagonalizable with lim
i
[a
i
[ = 0 A compact.
Finally, when A is compact its restriction to any closed subspace of H is compact, which by
2.13.2.1 proves the claim about the multiplicity of the eigenvalues.
We now wish to compute the state space of B
0
(H). This involves the study of a number of
subspaces of B(H) which are not C
-algebras, but which are ideals of B(H), except for the fact
that they are not closed.
Denition 2.13.5 The Hilbert-Schmidt norm | A |
2
of A B(H) is dened by
| A |
2
2
:=
i
| Ae
i
|
2
, (2.140)
where e
i
i
is an arbitrary basis of H; the right-hand side is independent of the choice of the basis.
The Hilbert-Schmidt class B
2
(H) consists of all A B(H) for which | A |
2
< .
The trace norm | A |
1
of A B(H) is dened by
| A |
1
:=| (A
A)
1
4 |
2
2
, (2.141)
where (A
A)
1
4
is dened by the continuous functional calculus. The trace class B
1
(H) consists
of all A B(H) for which | A |
2
< .
To show that (2.140) is independent of the basis, we take a second basis u
i
i
, with corre-
sponding resolution of the identity I =
i
[u
i
] (weakly). Aince I =
i
[e
i
] we then have
| A |
2
2
:=
i,j
(e
j
, u
i
)(u
i
, A
Ae
j
) =
i,j
(A
Au
i
, e
j
)(e
j
, u
i
) =
i
| Au
i
|
2
.
50 2 ELEMENTARY THEORY OF C
-ALGEBRAS
If A B
1
(H) then
Tr A :=
i
(e
i
, Ae
i
) (2.142)
is nite and independent of the basis (when A / B
1
(H), it may happen that Tr A depends on
the basis; it may even be nite in one basis and innite in another). Conversely, it can be shown
that A B
1
(H) when Tr
+
A < , where Tr
+
is dened in terms of the decomposition (2.106) by
Tr
+
A := Tr A
+
Tr A
+ iTr A
+
iTr A
. For A B
1
(H) one has Tr
+
A = Tr A. One always
has the equalities
| A |
1
= Tr [A[; (2.143)
| A |
2
= Tr [A[
2
= Tr A
A, (2.144)
where
[A[ :=
A. (2.145)
In particular, when A 0 one simply has | A |
1
= Tr A, which does not depend on the basis,
whether or not A B
1
(H). The properties
Tr A
A = Tr AA
(2.146)
for all A B(H), and
Tr UAU
= Tr A (2.147)
for all positive A B(H) and all unitaries U, follow from (2.142) by manipulations similar to
those establishing the basis-independence of (2.140). Also, the linearity property
Tr (A +B) = Tr A+ Tr B (2.148)
for all A, B B
1
(H) is immediate from (2.142).
It is easy to see that the Hilbert-Schmidt norm is indeed a norm, and that B
2
(H) is complete
in this norm. The corresponding properties for the trace norm are nontrivial (but true), and will
not be needed. In any case, for all A B(H) one has
| A | | A |
1
; (2.149)
| A | | A |
2
. (2.150)
To prove this, we use our old trick: although | B | | B | for all unit vectors , for every > 0
there is a
|
2
+. Put B = (A
A)
1
4
, and note that
| (A
A)
1
4
|
2
=| A | by (2.16). Completing
to a basis e
i
i
, we have
| A |=| (A
A)
1
4
|
2
| (A
A)
1
4
|
2
+
i
| (A
A)
1
4
e
i
|
2
+ =| A |
1
+.
Letting 0 then proves (2.149). The same trick with | A || A
| + establishes (2.150).
The following decomposition will often be used.
Lemma 2.13.6 Every operator A B(H) has a polar decomposition
A = U[A[, (2.151)
where [A[ =
A) = (, [A[
2
) = ([A[, [A[).
Hence U is an isometry on ran([A[). In particular, U is well dened, for this property implies that
if [A[
1
= [A[
2
then U[A[
1
= U[A[
2
. Then extend U to the closure of ran([A[) by continuity,
and put U = 0 on ran([A[)
A, (2.152)
and that U
i
(e
i
, [A[e
i
) < , for every > 0
we can nd N() such that
i>N()
(e
i
, [A[e
i
) < . Let p
N()
be the projection onto the linear
span of all e
i
, i > N(). Using (2.16) and (2.149), we have
| [A[
1
2
p
N()
|
2
=| p
N()
[A[p
N()
| | p
N()
[A[p
N()
|
1
< ,
so that [A[
1
2
p
N()
[A[
1
2
in the operator-norm topology. Since the star is norm-continuous by
(2.17), this implies p
N()
[A[
1
2
[A[
1
2
. Now p
N()
[A[
1
2
obviously has nite rank for every > 0, so
that [A[
1
2
is compact by Denition 2.13.1. Since A = U[A[
1
2
[A[
1
2
by (2.151), Proposition 2.13.2.2
implies that A B
0
(H).
The proof that B
2
(H) B
0
(H) is similar: this time we have
| [A[p
N()
|
2
=| p
N()
[A[
2
p
N()
| | p
N()
[A[
2
p
N()
|
2
< ,
so that [A[p
N()
[A[, with the same conclusion.
Finally, we use Theorem 2.13.4 to rewrite (2.143) and (2.140) as
| A |
1
=
i
a
i
;
| A |
2
=
i
a
2
i
, (2.154)
where the a
i
are the eigenvalues of [A[. This immediately gives
| A |
2
| A |
1
, (2.155)
implying B
1
(H) B
2
(H).
Finally, the claim about proper inclusions is trivially established by producing examples on the
basis of 2.13.4 and (2.154).
The chain of inclusions (2.153) is sometimes seen as the non-commutative analogue of
c
(X)
1
(X)
2
(X)
0
(X)
(X),
where X is an innite discrete set. Since
1
(X) =
0
(X)
and
(X) =
1
(X)
=
0
(X)
, this
analogy is strengthened by the following result.
Theorem 2.13.8 One has B
0
(H)
= B
1
(H) and B
1
(H)
= B
0
(H)
is identied with B
1
(H), and
A B
1
(H)
B. (2.157)
To show that this is well dened, use (2.1) and (2.144).
Lemma 2.13.9 For B
1
(H) and A B(H) one has
[Tr A[ | A | | |
1
. (2.158)
52 2 ELEMENTARY THEORY OF C
-ALGEBRAS
Using (2.151) for and (2.1) for the inner product (2.157), as well as (2.146) and (2.141), we
estimate
[Tr A[
2
= [Tr AU[[
1
2
[[
1
2
[ = [((AU[[
1
2
)
, [[
1
2
)[
| [[
1
2
|
2
2
| (AU[[
1
2
)
|
2
2
=| |
1
Tr ([[
1
2
U
AU[[
1
2
).
Now observe that if 0 A
1
A
2
then Tr A
2
Tr A
2
for all A
1
, A
2
B
0
(H), since on account
of 2.13.4 one has A
1
A
2
i all eigenvalues of A
1
are all eigenvalues of A
2
. Then use (2.154).
From (2.90) we have [[
1
2
U
AU[[
1
2
| AU |
2
, so from the above insight we arrive at
Tr ([[
1
2
U
AU[[
1
2
) | |
1
| AU |
2
| A |
2
,
since U is a partial isometry. Hence we have (2.158).
We now prove B
0
(H)
= B
1
(H). It is clear from 2.13.9 that
B
1
(H) B
0
(H)
, (2.159)
with
| | | |
1
. (2.160)
To prove that B
0
(H)
B
1
(H), we use (2.150). For B
0
(H)
and A B
2
(H) B
0
(H) we
therefore have
[ (A)[ | | | A | | | | A |
2
.
Hence B
2
(H)
; since B
2
(H) is a Hilbert space, by Riesz-Fischer there is an operator B
2
(H)
such that (A) = Tr A for all A B
2
(H). In view of (2.153), we need to sharpen B
2
(H)
to B
1
(H). To do so, choose a nite-dimensional projection p, and note that p[[ B
f
(H)
B
1
(H); the presence of p even causes the sum in (2.142) to be nite in a suitable basis. Now use
the polar decomposition = U[[ with (2.152) to write
Tr p[[ = Tr pU
= Tr pU
= (pU
);
changing the order inside the trace is justied by naive arguments, since the sum in (2.142) is
nite. Using the original assumption B
0
(H)
, we have
[Tr p[[ [ | | | pU
| | | | p |=| | (2.161)
since U is a partial isometry, whereas | p |= 1 in view of (2.16) and p = p
2
= p
. Now choose
a basis of H, and take p to be the projection onto the subspace spanned by the rst N elements;
from (2.142) and (2.161) we then have
[Tr p[[ [ = [
N
i=1
(e
i
, [[e
i
)[ | | .
It follows that the sequence s
N
:= [
N
i=1
(e
i
, [[e
i
)[ is bounded, and since it is positive it must
have a limit. By (2.161) and (2.143) this means that | |
1
| |, so that B
1
(H), hence
B
0
(H)
B
1
(H). Combining this with (2.159) and (2.160), we conclude that B
0
(H)
= B
1
(H)
and | |
1
=| |.
We turn to the proof of B
1
(H)
, with
|
A | | A | . (2.162)
To establish the converse, pick
A B
1
(H)
= [ >< [, so that
[ [ >< [ [ =
([ >< [)
[ >< [ =
(, )[ >< [ =| | | | [].
Since, for any projection p, the number Tr p is the dimension of pH (take a basis whose elements
lie either in pH or in p
-algebra B
0
(H) of all compact operators on
some Hilbert space H consists of all density matrices, where a density matrix is an element
B
1
(H) which is positive ( 0) and has unit trace (Tr = 1).
2. The pure state space of B
0
(H) consists of all one-dimensional projections.
3. The C
-algebra B
0
(H) possesses only one irreducible representation, up to unitary equiva-
lence, namely the dening one.
Diagonalize =
i
p
i
[
i
]; cf. 2.13.4 and 2.153. Using A = [
i
], which is positive, the condition
(A) 0 yields p
i
0. Conversely, when all p
i
0 the operator is positive. The normalization
condition | |=| |
1
=
p
i
= 1 (see 2.8.6) and (2.154)) yields 2.13.10.1.
The next item 2.13.10.2 is then obvious from 2.12.1.
Finally, 2.13.10.3 follows from 2.13.10.2 and Corollaries 2.12.7 and 2.12.6.
Corollary 2.13.10.3 is one of the most important results in the theory of C
-algebras. Applied
to the nite-dimensional case, it shows that the C
-algebra M
n
(C) of n n matrices has only one
irreducible representation.
The opposite extreme to a pure state on B
0
(H) is a faithful state , for which by denition
the left-ideal ^
(B
0
(H)) on the Hilbert space B
2
(H) of Hilbert-Schmidt
operators given by left-multiplication, i.e.,
-ALGEBRAS
It is obvious from (2.140) that for A B(H) and B B
2
(H) one has
| AB |
2
| A | | B |
2
, (2.169)
so that the representation (2.168) is well-dened (even for A B(H) rather than merely A
B
0
(H)). Moreover, when A, B B
2
(H) one has
Tr AB = Tr BA. (2.170)
This follows from (2.146) and the identity
AB =
1
4
3
n=0
i
n
(B +i
n
A
(B +i
n
A
). (2.171)
When B
1
(H) and 0 then
1/2
B
2
(H); see (2.143) and (2.144). It is easily seen that
1/2
is cyclic for
(B
0
(H)) when is faithful. Using (2.157) and (2.170) we compute
(
1/2
,
(A)
1/2
) = Tr
1/2
(A)
1/2
= Tr A = (A).
The equivalence between
and
of a collection M of bounded operators consists of all bounded operators which commute with
all elements of M; the bicommutant M
is (M
.
We rst give the nite-dimensional version of the theorem; this is already nontrivial, and its
proof contains the main idea of the proof of the innite-dimensional case as well.
Proposition 2.14.1 Let M be a
-algebra (and hence a C
-algebra) in M
n
(C) containing I (here
n < ). Then M
= M.
The idea of the proof is to take n arbitrary vectors
1
, . . . ,
n
in C
n
, and, given A M
,
construct a matrix A
0
M such that A
i
= A
0
i
for all i = 1, . . . , n. Hence A = A
0
M. We
will write H for C
n
.
Choose some =
1
H, and form the linear subspace M of H. Since H is nite-
dimensional, this subspace is closed, and we may consider the projection p = [M] onto this
subspace. By Lemma 2.9.3 one has p M
. Hence A M
n
(where I
n
is the unit in M
n
(B(H))); this is the
diagonal matrix in M
n
(B(H)) in which all diagonal entries are A.
Now use the rst part of the proof, with the substitutions H H
n
, M (M), A A := (A),
and
1
+. . .
+
n
. Hence given
1
+. . .
+
n
and (A) (M) there exists A
0
(M)
such
that
(A)(
1
+. . .
+
n
) = A
0
(
1
+. . .
+
n
). (2.172)
For arbitrary B M
n
(B(H)), compute ([B, (A)])
ij
= [B
ij
, A]. Hence (M)
= M
n
(M
). It is
easy to see that M
n
(M
= M
n
(M
), so that
(M)
= (M
). (2.173)
2.14 The double commutant theorem 55
Therefore, A
0
= (A)
0
for some A
0
M. Hence (2.172) reads A
i
= A
0
i
for all i = 1, . . . , n.
As it stands, Proposition 2.14.1 is not valid when M
n
(C) is replaced by B(H), where dim(H) =
. To describe the appropriate renement, we dene two topologies on B(H) which are weaker
than the norm-topology we have used so far (and whose denition we repeat for convenience).
Denition 2.14.2 The norm-topology on B(H) is dened by the criterion for convergence
A
A i | A
A i | (A
A) |0
for all H. A basis for the strong topology is given by all sets of the form
O
s
(A,
1
, . . . ,
n
) := B B(H)[ | (B A)
i
| < i = 1, . . . , n, (2.175)
where A B(H),
1
, . . . ,
n
H, and > 0.
The weak topology on B(H) is dened by the convergence A
A i [(, (A
A))[ 0
for all H. A basis for the weak topology is given by all sets of the form
O
w
(A,
1
, . . . ,
n
,
1
, . . . ,
n
) := B B(H)[ [(
i
, (A
A)
i
)[ < i = 1, . . . , n,
(2.176)
where A B(H),
1
, . . . ,
n
,
1
, . . . ,
n
H, and > 0.
These topologies should all be seen in the light of the general theory of locally convex topological
vector spaces. These are vector spaces whose topology is dened by a family p
of semi-norms;
recall that a semi-norm on a vector space 1 is a function p : 1 R satisfying 2.1.1.1, 3, and 4. A
net v
(v
v) 0 for all .
The norm-topology is dened by a single semi-norm, namely the operator norm, which is even
a norm. Its open sets are generated by -balls in the operator norm, whereas the strong and the
weak topologies are generated by nite intersections of -balls dened by semi-norms of the form
p
s
= M;
2. M is closed in the weak operator topology;
3. M is closed in the strong operator topology.
It is easily veried from the denition of weak convergence that the commutant N
of a
-algebra
N is always weakly closed: for if A
N, and B N, then
(, [A, B]) = (, AB) (B
, A) = lim
(, A
B) (B
, A
) = lim
(, [A
, B]) = 0.
If M
= M then M = N
for N = M
56 3 HILBERT C
) | < .
For H
n
this means that
| (A A
)(
1
+. . .
+
n
) |
2
=
n
i=1
| (A A
)
i
|
2
<
2
.
Noting the inclusion
i=1
| (A B)
i
|
2
<
2
O
s
(A,
1
, . . . ,
n
)
(cf. (2.175)), it follows that A
= M.
3 Hilbert C
:
1
(^
) ^
F
X F for which =
X
(where
X
: X F X is the projection onto the rst factor).
The maps
= (,
F
), so that
F
restricted
to
1
(x) provides a homeomorphism between the latter and the typical ber F. Each subset
1
(x) is called a ber of B. One may think of B as X with a copy of F attached at each point.
The simplest example of a bundle over a base X with typical ber F is the trivial bundle
B = X F, with (x, f) := x. According to the denition, any bundle is locally trivial in the
specied sense.
Denition 3.1.2 A vector bundle is a bundle in which
1. each ber is a nite-dimensional vector space, such that the relative topology of each ber
coincides with its topology as a vector space;
2. each local trivialization
F
:
1
(x) F (where x ^
) is linear.
A complex vector bundle is a vector bundle with typical ber C
m
, for some m N.
We will generically denote vector bundles by the letter V, with typical ber F = V . The
simplest vector bundle over X with ber V = C
n
is the trivial bundle V = X C
n
. This
bundle leads to possibly nontrivial sub-bundles, as follows. Recall the denition of M
n
(A) in
2.11, specialized to A = C(X) in the proof of 2.11.4. If X is a compact Hausdor space, then
M
n
(C(X)) C(X, M
n
(C)) is a C
1
(x) := p(x)C
n
. The space V
p
inherits a topology and a projection (onto the rst co-ordinate)
from XC
n
, relative to which all axioms for a vector bundle are satised. Note that the dimension
of p(x) is independent of x, because p is continuous and X is connected.
The converse is also true.
Proposition 3.1.3 Let V be a complex vector bundle over a connected compact Hausdor space
X, with typical ber C
m
. There is an integer n m and an idempotent p C(X, M
n
(C)) such
that V X C
n
, with
1
(x) = p(x)C
n
.
3.1 Vector bundles 57
The essence of the proof is the construction of a complex vector bundle V
such that V V
is
trivial (where the direct sum is dened berwise); this is the bundle X C
n
.
Following the philosophy of non-commutative geometry, we now try to describe vector bundles
in terms of C
R
(f)(x) := f(x)(x). (3.1)
Since C(X) is commutative, this is, of course, a left-action as well.
For example, in the trivial case one has the obvious isomorphisms
(X C
m
) C(X, C
m
) C(X) C
m
m
C(X). (3.2)
A fancy way of saying this is that (X C
m
) is a nitely generated free module for C(X).
Here a free (right-) module c for an algebra A is a direct sum c =
n
A of a number of copies of
A itself, on which A acts by right-multiplication, i.e.,
R
(B)A
1
. . . A
n
:= A
1
B . . . A
n
B. (3.3)
If this number is nite one says that the free module is nitely generated.
When V is non-trivial, one obtains (V) as a certain modication of a nitely generated free
module for C(X). For any algebra A, and idempotent p M
n
(A), the action of p on
n
A commutes
with the action by A given by right-multiplication on each component. Hence the vector space
p
m
A is a right- A-module, called projective. When m < , one calls p
m
A a nitely
generated projective module for A.
In particular, when V = X C
n
and V
p
is the vector bundle described prior to 3.1.3, we see
that
(V
p
) = p
n
C(X) (3.4)
under the obvious (right-) action of C(X).
This lead to the Serre-Swan theorem.
Theorem 3.1.4 Let X be a connected compact Hausdor space. There is a bijective corre-
spondence between complex vector bundles V over X and nitely generated projective modules
c(V) = (V) for C(X).
This is an immediate consequence of Proposition 3.1.3: any vector bundle is of the form V
p
,
leading to (V
p
) as a nitely generated projective C(X)-module by (3.4). Conversely, given such a
module p
n
C(X), one has p C(X, M
n
(C)), and thereby a vector bundle V
p
as described prior
to 3.1.3.
Thus we have achieved our goal of describing vector bundles over X purely in terms of concepts
pertinent to the C
= p in addition to p
2
= p). Any other Hermitian structure on V
p
may be shown to be
equivalent to this canonical one.
There is no reason to restrict the dimension of the bers so as to be nite-dimensional. A
Hilbert bundle is dened by replacing nite-dimensional vector space in 3.1.2.1 by Hilbert
space, still requiring that all bers have the same dimension (which may be innite). A Hilbert
bundle with nite-dimensional bers is evidently the same as a Hermitian vector bundle. The
simplest example of a Hilbert bundle is a Hilbert space, seen as a bundle over the base space
consisting of a single point.
The following class of Hilbert bundles will play a central role in the theory of induced group
representations.
Proposition 3.1.6 Let H be a closed subgroup of a locally compact group G, and take a uni-
tary representation U
of H on a Hilbert space H
. Then H acts on G H
by h : (x, v)
(xh
1
, U
:= G
H
H
= (GH
)/H (3.5)
by this action is a Hilbert bundle over X = G/H, with projection
([x, v]
H
) := [x]
H
(3.6)
and typical ber H
.
Here [x, v]
H
is the equivalence class in G
H
H
of (x, v) G H
, and [x]
H
= xH is the
equivalence class in G/H of x G. Note that the projection
is well dened.
The proof relies on the fact that G is a bundle over G/H with projection
(x) = [x]
H
(3.7)
and typical ber H. This fact, whose proof we omit, implies that every q G/H has a neighbour-
hood ^
, so that
= (,
H
) :
1
(^
) ^
(xh) =
H
(x)h. (3.8)
This leads to a map
:
1
(^
) ^
, given by
([x, v]
H
) := ([x]
H
, U
(
H
(x))v). This
map is well dened because of (3.8), and is a local trivialization of G
H
H
-modules
What follows generalizes the notion of a Hilbert bundle in such a way that the commutative C
-
algebra C(X) is replaced by an arbitrary C
-module over a C
-algebra B consists of
A complex linear space c.
A right-action
R
of B on c (i.e.,
R
maps B linearly into the space of all linear operators
on c, and satises
R
(AB) =
R
(B)
R
(A)), for which we shall write B :=
R
(B), where
c and B B.
A sesquilinear map ' , `
B
: c c B, linear in the second and anti-linear in the rst entry,
satisfying
', `
B
= ', `
B
; (3.9)
', B`
B
= ', `
B
B; (3.10)
', `
B
0; (3.11)
', `
B
= 0 = 0, (3.12)
for all , c and B B.
3.2 Hilbert C
-modules 59
The space c is complete in the norm
| |:=| ', `
B
|
1
2
. (3.13)
We say that c is a Hilbert B-module, and write c B.
One checks that (3.13) is indeed a norm: | |
2
equals sup(', `
B
), where the supremum
is taken over all states on B. Since each map
(', `
B
) is a semi-norm (i.e., a norm
except for positive deniteness) by (3.11), the supremum is a semi-norm, which is actually positive
denite because of Lemma 2.8.8 and (3.12).
The B-action on c is automatically non-degenerate: the property B = 0 for all B B implies
that ', `
B
B = 0 for all B, hence ', `
B
= 0 (when B is unital this is follows by taking B = I;
otherwise one uses an approximate unit in B), so that = 0 by (3.12).
When all conditions in 3.2.1 are met except (3.12), so that | | dened by (3.13) is only a
semi-norm, one simply takes the quotient of c by its subspace of all null vectors and completes,
obtaining a Hilbert C
', `
B
. (3.14)
Example 3.2.2 1. Any C
B.
Note that the norm (3.13) coincides with the C
-norm by (2.16).
2. Any Hilbert space H is a Hilbert C-module H C in its inner product.
3. Let H be a Hilbert bundle H over a compact Hausdor space X. The space of continuous
sections c = (H) of H is a Hilbert C
0
(H) the function ', `
C(X)
is dened by
', `
C(X)
: x ((x), (x))
x
, (3.15)
where the inner product is the one in the ber
1
(x). The right-action of C(X) on (H) is
dened by (3.1).
In the third example the norm in (H) is | | = sup
x
| (x) |, where | (x) |= ((x), (x))
1
2
x
,
so that it is easily seen that c is complete.
Many Hilbert C
-algebra is a
-algebra satisfying all properties of a C
-algebra
B, dene a pre-Hilbert
B-module
c
B as in Denition 3.2.1, except
that the nal completeness condition is omitted.
Proposition 3.2.3 In a pre-Hilbert
B-module (and hence in a Hilbert B-module) one has the
inequalities
| B | | | | B |; (3.16)
', `
B
', `
B
| |
2
', `
B
; (3.17)
| ', `
B
| | | | | . (3.18)
To prove (3.16) one uses (3.14), (2.90), (2.85), and (2.16). For (3.17) we substitute ', `
B
-algebra C
: c c
such that
', A`
B
= 'A
, `
B
(3.19)
for all , c is called adjointable.
Theorem 3.2.5 An adjointable map is automatically C-linear, B-linear (that is, (A)B = A(B)
for all c and B B), and bounded. The adjoint of an adjointable map is unique, and the
map A A
(c, B) is a C
-algebra.
Each element A C
(c, B)
c B.
The property of C-linearity is immediate. To establish B-linearity one uses (3.14); this also
shows that A
(c, B) when A C
(c, B).
To prove boundedness of a given adjointable map A, x c and dene T
: c B by
T
:= 'A
A, `
B
. It is clear from (3.18) that | T
| | A
A |, so that T
is bounded. On
the other hand, since A is adjointable, one has T
= ', A
A`
B
, so that, using (3.18) once
again, one has | T
| | A
A | | |. Hence sup| T
(c, B) is norm-closed, as one easily veries from (3.19) and (3.13) that if A
n
A
then A
n
converges to some element, which is precisely A
(c, B) is a Banach algebra, so that its satises (2.15). To check (2.16) one
infers from (3.13) and the denition (3.19) of the adjoint that | A |
2
| A
(c, B). When the right-hand side vanishes for all , it must be that ', Z`
B
= 0 for all ,
hence for = Z, so that Z = 0. Here we used the fact that B = 0 for all and B in the linear
span of 'c, c`
B
implies B = 0, for by (3.10) it implies that ', `
B
B = 0.
Under a further assumption (which is by no means always met in our examples) one can
completely characterize C
-module c B the C
-algebra C
(c, B) coincides
with the space L(c)
B
of all bounded C-linear and B-linear maps on c.
3.3 The C
-algebra of a Hilbert C
-module 61
In view of Theorem 3.2.5 we only need to show that a given map A L(c)
B
is adjointable.
Indeed, for xed c dene
A,
: c B by
A,
(Z) := ', AZ`
B
. By self-duality this must
equal ', Z`
B
for some , which by denition is A
.
In the context of Example 3.2.2.1, one may wonder what C
(A, A).
When A has a unit, one therefore has C
(A, A) is isomorphic to the idealizer of (A) in B(H) (this is the set of all B B(H) for which
B(A) (A) for all A A). One thus obtains
C
(C
0
(X), C
0
(X)) = C
b
(X); (3.21)
C
(B
0
(H), B
0
(H)) = B(H). (3.22)
Eq. (3.21) follows by taking (C
0
(X)) to be the representation on L
2
(X) by multiplication opera-
tors (where L
2
is dened by a measure with support X), and (3.22) is obtained by taking (B
0
(H))
to be the dening representation; see the paragraph following 2.13.1.
In Example 3.2.2.2 the C
-algebra C
-modules as well.
3.3 The C
-algebra of a Hilbert C
-module
In preparation for the imprimitivity theorem, and also as a matter of independent interest, we
introduce the analogue for Hilbert C
-modules of the C
-algebra B
0
(H) of compact operators on
a Hilbert space. This is the C
-module.
Denition 3.3.1 The C
-algebra C
0
(c, B) of compactoperators on a Hilbert C
-module c B
is the C
-subalgebra of C
0
(c, B) c B, and call this a dual pair.
The word compact appears between quotation marks because in general elements of C
0
(c, B)
need not be compact operators. The signicance of the notation introduced at the end of the
denition will emerge from Theorem 3.3.3 below. Using the (trivially proved) properties
(T
B
,
)
= T
B
,
; (3.24)
AT
B
,
= T
B
A,
; (3.25)
T
B
,
A = T
B
,A
, (3.26)
where A C
0
(c, B) is a (closed 2-sided) ideal in
C
-algebra by Theorem 3.2.5. From (3.16) and (3.18) one nds the bound
| T
B
,
| | | | | . (3.27)
One sees from the nal part of the proof of Theorem 3.2.5 that C
0
(c, B) acts non-degenerately on
c.
When C
0
(c, B) has a unit it must coincide with C
(c, B).
Proposition 3.3.2 1. When c = B = A (see Example 3.2.2.1) one has
C
0
(A, A) A. (3.28)
This leads to the dual pair A A A.
62 3 HILBERT C
0
(H, C) = B
0
(H), (3.29)
whence the dual pair B
0
(H) H C.
One has T
A
,
= (
, is an
isometric morphism as well. It is, in particular, injective. When A has a unit it is obvious that
is surjective; in the non-unital case the existence of an approximate unit implies that the linear
span of all
is dense in A. Extending to C
0
(A, A) by continuity, one sees from Corollary
2.7.7 that (C
0
(A, A)) = A.
Eq. (3.29) follows from Denition 2.13.1 and the fact that the linear span of all T
C
,
is
B
f
(H).
A Hilbert C
-modules.
Given a complex linear space c, the conjugate space c is equal to c as a real vector space, but
has the conjugate action of complex scalars.
Theorem 3.3.3 Let c be a full Hilbert B-module. The expression
', `
C
0
(E,B)
:= T
B
,
(3.30)
in combination with the right-action
R
(A) := A
, where A C
0
(c, B), denes c as a full
Hilbert C
-module over C
0
(c, B). In other words, from c B one obtains c C
0
(c, B). The
left-action
L
(B) := B
0
(c, C
0
(c, B)) B. (3.31)
We call A := C
0
(c, B); in the references to (3.9) etc. below one should substitute A for B when
appropriate. The properties (3.9), (3.10), and (3.11) follow from (3.24), (3.26), and Lemma 3.5.2,
respectively.
To prove (3.12), we use (3.30) with = , (3.23) with Z = , (3.10), (3.14), and (3.13) to
show that ', `
A
= 0 implies | ', `
3
B
|= 0. Since ', `
B
is positive by (3.11), this implies
', `
B
= 0, hence = 0 by (3.12).
It follows from (3.14) and (3.26) that each
L
(B) is adjointable with respect to ' , `
A
. Moreover,
applying (3.13), (3.30), (3.27), and (3.16) one nds that
L
(B) is a bounded operator on c with
respect to | |
A
, whose norm is majorized by the norm of B in B. The map
L
is injective because
c is non-degenerate as a right-B-module.
Let c
c
be the completion of c in | |
A
; we will shortly prove that c
c
= c. It follows from
the previous paragraph that
L
(B) extends to an operator on c
c
(denoted by the same symbol),
and that
L
maps B into C
(c
c
, A). It is trivial from its denition that
L
is a morphism. Now
observe that
L
(', `
B
) = T
A
,
, (3.32)
for the denitions in question imply that
T
A
,
Z = ', Z`
A
= T
B
Z,
= Z', `
B
. (3.33)
The fullness of c B and the denition of C
0
(c
c
, A) imply that
L
: B C
0
(c
c
, A) is an
isomorphism. In particular, it is norm-preserving by Lemma 2.7.6.
The space c is equipped with two norms by applying (3.13) with B or with A; we write | |
B
and | |
A
. From (3.30) and (3.27) one derives
| |
A
| |
B
. (3.34)
3.4 Morita equivalence 63
For c we now use (3.13), the isometric nature of
L
, and (3.32) to nd that
| |
B
=| T
A
,
|
1
2
. (3.35)
From (3.27) with B A one then derives the converse inequality to (3.34), so that | |
A
=| |
B
.
Hence c
c
= c, as c is complete in | |
B
by assumption. The completeness of c as a Hilbert B-
module is equivalent to the completeness of c as a Hilbert A-module.
We have now proved (3.31). Finally noticing that as a Hilbert C
0
(c, B), the proof of Theorem 3.3.3 is ready.
For later reference we record the remarkable identity
', `
C
0
(E,B)
Z = ', Z`
B
, (3.36)
which is a restatement of (3.33).
3.4 Morita equivalence
The imprimitivity theorem establishes an isomorphism between the respective representation the-
ories of two C
0
(c, B). We write A
M
B and A c B.
Proposition 3.4.2 Morita equivalence is an equivalence relation in the class of all C
-algebras.
The reexivity property B
M
B follows from (3.28), which establishes the dual pair B B
B. Symmetry is implied by (3.31), proving that A c B implies B c A.
The proof of transitivity is more involved. When A
M
B and B
M
C we have the chain of dual
pairs
A c
1
B c
2
C.
We then form the linear space c
1
B
c
2
(which is the quotient of c
1
c
2
by the ideal 1
B
generated
by all vectors of the form
1
B
2
1
B
2
), which carries a right-action
R
(C) given by
R
(C)(
1
B
2
) :=
1
B
(
2
C). (3.37)
Moreover, we can dene a sesquilinear map ' , `
C
on c
1
B
c
2
by
'
1
B
2
,
1
B
2
`
C
:= '
2
, '
1
,
1
`
B
2
`
C
. (3.38)
With (3.37) this satises (3.9) and (3.10); as explained prior to (3.14), one may therefore construct
a Hilbert C
-module, denoted by c
B
c
2
.)
Apart from the right-action
R
(C), the space c
carries a left-action
L
(A): the operator
L
(A)(
1
B
2
) := (A
1
)
B
2
(3.39)
is bounded on c
1
B
c
2
and extends to c
0
(c
, C) =
L
(A). (3.40)
Using (3.23), the denition of
B
, and (3.10), it is easily shown that
L
(T
B
12,2
B
,1
)
1
B
2
=
1
B
'
2
,
2
'
1
,
1
`
B
`
B
2
. (3.41)
64 3 HILBERT C
0
(c
2
, C) = B; as in (3.30), with B C, and c c
2
, this yields
', `
B
= T
C
,
. Substituting this in the right-hand side of (3.41), and using (3.23) with B C,
the right-hand side of (3.41) becomes
1
B
2
'
2
'
1
,
1
`
B
,
2
`
C
. Using B
=
L
(B) (see
3.3.3), (3.19) with B C, (3.38), and (3.23) with B C, we eventually obtain
T
C
1
B
2,1
B
2
=
L
(T
B
12,2
B
,1
). (3.42)
This leads to the inclusion C
0
(c
, C)
L
(A). To prove the opposite inclusion, one picks a
double sequence
i
2
,
i
2
such that
N
i
T
C
i
2
,
i
2
is an approximate unit in B = C
0
(c
2
, C). One
has lim
N
N
i
i
2
'
i
2
, Z`
C
= Z from (3.23), and a short computation using (3.23) with (3.38) then
yields
lim
N
N
i
T
C
1
B
i
2
,1
B
i
2
=
L
(T
B
1,1
).
Hence
L
(A) C
0
(c
C, implying that A
M
C. This proves transitiv-
ity.
Here is a simple example of this concept.
Proposition 3.4.3 The C
-algebra B
0
(H) of compact operators is Morita-equivalent to C, with
dual pair B
0
(H) H C. In particular, the matrix algebra M
n
(C) is Morita-equivalent to C.
This is immediate from (3.29). In the nite-dimensional case one has M
n
(C) C
n
C, where
M
n
(C) and C act on C
n
in the usual way. The double Hilbert C
-algebras
A and
B;
a full pre-Hilbert
B-module
c;
a left-action of
A on
c, such that
c can be made into a full pre-Hilbert
A-module with respect
to the right-action
R
(A) := A
;
the identity
', `
A
Z = ', Z`
B
(3.44)
(for all , , Z
c) relating the two Hilbert C
-module structures;
the bounds
'B, B`
A
| B |
2
', `
A
; (3.45)
'A, A`
B
| A |
2
', `
B
(3.46)
for all A
A and B
B.
3.5 Rieel induction 65
Then A
M
B, with dual pair A c B, where c is the completion of
c as a Hilbert B-module.
Using Corollary 3.2.4 we rst complete
c to a Hilbert B-module c. By (3.46), which implies
| A | | A | | | for all A
A and
c, the action of
A on
c extends to an action of A on
c. Similarly, we complete
c to a Hilbert A-module c
c
; by (3.45) the left-action
L
(B) := B
extends to an action of B on c
c
. As in the proof of Theorem 3.3.3, one derives (3.34) and its
converse for
c, so that the B-completion c of
c coincides with the A-completion c
c
of
c; that
is, c
c
= c.
Since c is a full pre-Hilbert
A-module, the A-action on c is injective, hence faithful. It follows
from (3.44), Theorem 3.3.3, and (once again) the fullness of c, that A C
0
(c, B). In particular,
each A A automatically satises (3.19).
Clearly, (3.44) is inspired by (3.36), into which it is turned after use of this proposition. We
will repeatedly use Proposition 3.4.4 in what follows; see 3.9.3 and 3.10.1.
3.5 Rieel induction
To formulate and prove the imprimitivity theorem we need a basic technique, which is of interest
also in a more general context. Given a Hilbert B-module c, the goal of the Rieel induction
procedure described in this section is to construct a representation
of C
.
Construction 3.5.1 Suppose one has a Hilbert C
-module c B.
1. Given a state
0
on c by
(, )
0
:=
(', `
B
). (3.47)
Since
and ' , `
B
are positive (cf. (3.11)), this form is positive semi-denite. Its null space
is
= c [
(, )
0
= 0. (3.48)
2. The form
( , )
0
projects to an inner product
( , )
on the quotient c/
^
. If
V
: c c/
^
,
V
:=
(, )
0
. (3.49)
The Hilbert space
H
is the closure of c/
^
(C
by
(A)
:=
V
A; (3.50)
it follows that
is continuous. Since c/
^
is dense in
H
, the operator
(A) may be
dened on all of
H
(A)
|
2
=
(A, A)
0
. Using (3.47), (3.20), and (3.18) in succes-
sion, one nds that
|
(A) | | A | . (3.51)
66 3 HILBERT C
(A) |,
o(B). (3.53)
A similar argument combined with Corollary 2.5.3 shows that
is. As a corollary, one infers a useful property, which will be used, e.g., in the proof of
Theorem 3.3.3.
Lemma 3.5.2 Let A C
(A) 0.
When one starts from a representation
-module c B.
1. Given a representation
, the sesquilinear
form ( , )
0
is dened on c H
0
:= (v,
(', `
B
)w)
, (3.54)
where v, w H
and ' , `
B
are. The null
space is
^
c H
[ (
,
)
0
= 0. (3.55)
As in (2.116), we may equally well write
^
c H
[ (
,
)
0
= 0
c H
. (3.56)
2. The form ( , )
0
projects to an inner product ( , )
on the quotient c H
/^
, dened by
(V
, V
:= (
,
)
0
, (3.57)
where V
: c H
c H
/^
is the
closure of c H
/^
(C
by continuous extension of
(A)V
:= V
(A I
), (3.58)
where I
(A) | | A | . (3.59)
To prove that the form dened in (3.54) is positive semi-denite, we assume that
(B) is cyclic
(if not, the argument below is repeated for each cyclic summand; see 2.9.2). With
=
i
i
v
i
and v
i
=
(B
i
) (where is a cyclic vector for
,
)
0
= (v,
(', `
B
)v)
with :=
i
i
B
i
. Hence (
,
)
0
0 by (3.10) and the
positivity of
: B B(H
).
Similarly, one computes |
(A)V
|
2
= (v,
('A, A`
B
)v)
(', `
B
)v)
|
2
, this proves (3.59).
Paraphrasing the comment after the rst version of the construction,
is faithful when
is.
Also, it is not dicult to verify that
is non-degenerate when
is.
3.6 The imprimitivity theorem 67
A c B H
?
-
H
H
H
H
H
H
H
H
Hj
-
induction
.
Then dene a linear map
U : c c H
by
U :=
. (3.60)
According to (3.47), (3.54), and (2.115), this map has the property
(
U,
U)
0
=
(
,
)
0
. (3.61)
By (3.49) and (3.57) the map
U therefore quotients to a unitary isomorphism U :
H
, which
by (3.50) and (3.58) duly intertwines
and
.
Of course, any subspace of C
. This
particularly applies when one has a given (pre-) C
-algebra A and a
-homomorphism : A
C
(A) on H
(A)
for
such that
is equivalent to the
Rieel-induced representation
(A) =
(B)
(B) =
(A)
(A).
See Figure 2. Starting with
(B), we construct
(B) from
Rieel induction with respect to the dual pair B c A. We then construct a unitary map
U : H
which intertwines
and
.
We rst dene
U : c c H
by linear extension of
U v :=
(', `
B
)v. (3.62)
Note that
U is indeed C-linear. Using (3.62), the properties (2.19) and (2.18) with
, (3.54),
and (3.23), one obtains
(
U
1
1
v
1
,
U
2
2
v
2
)
= (
1
v
1
, T
B
1,2
2
v
2
)
0
. (3.63)
68 3 HILBERT C
B c A H
:= H
-
-
@
@
@
@
@R
B H
@
@
@
@
@R
-
Figure 2: Quantum imprimitivity theorem: H
and
0
(c, B) to use (3.36), and subsequently (3.57) and (3.58), all
read from right to left. The right-hand side of (3.63) is then seen to be equal to (V
1
v
1
,
('
1
,
2
`
A
)V
2
v
2
)
. Now put
and H
= H
U
1
1
v
1
,
U
2
2
v
2
)
= (V
(
1
V
1
v
1
), V
(
2
V
2
v
2
))
. (3.64)
In particular,
U annihilates
, where
c H
, whenever
^
or V
. Hence
we see from the construction rstly of H
= H
from c H
, and secondly of H
from c H
(V
v) :=
U v =
(', `
B
)v. (3.65)
Using the assumptions that the Hilbert C
, so that U is
unitary.
To verify that U intertwines
and
(B)V
( V
v) =
('
L
(B), `
B
)v, (3.66)
where the left-action of B B on c is as dened in 3.3.3. Thus writing
L
(B) = B
,
using (3.14), (2.18) with
, and (3.65) from right to left, the right-hand side of (3.66) is seen
to be
(B)UV
(V
v). Hence U
(B) =
2 leads to
2
. This also proves that the
bijective correspondence
(B)
is irreducible and
as above, decomposes
(B) as
2
, and thus arrives at a contradiction, since
.
Combined with Proposition 3.4.3, this theorem leads to a new proof of Corollary 2.13.10.
3.7 Group C
-algebras
In many interesting applications, and also in the theory of induced representation as originally
formulated for groups by Frobenius and Mackey, the C
-algebra.
3.7 Group C
-algebras 69
We start with the denition of the group algebra C
y,zG|yz=x
f(y)g(z) (3.67)
and the involution
f
(x) := f(x
1
). (3.68)
It is easy to check that the multiplication is associative as a consequence of the associativity of
the product in G. In similar vein, the operation dened by (3.68) is an involution because of the
properties (x
1
)
1
= x and (xy)
1
= y
1
x
1
at the group level.
A representation of C(G) on a Hilbert space H is dened as a morphism : C(G) B(H).
Proposition 3.7.1 There is a bijective correspondence between non-degenerate representations
of the
-algebra C(G) and unitary representations U of G, which preserves unitary equivalence and
direct sums (and therefore preserves irreducibility). This correspondence is given in one direction
by
(f) :=
xG
f(x)U(x), (3.69)
and in the other by
U(x) := (
x
), (3.70)
where
x
(y) := (xy).
It is elementary to verify that is indeed a representation of C(G) when U is a unitary
representation of G, and vice versa. Putting x = e in (3.70) yields (
e
) = I, so that cannot be
degenerate.
When U
1
(x) = V U
2
(x)V
G
dx = 1. The
Banach space L
1
(G) and the Hilbert space L
2
(G) are dened with respect to the Haar measure.
The convolution product is dened, initially on C
c
(G), by
f g(x) :=
G
dy f(xy
1
)g(y); (3.71)
it is evident that for a nite group this expression specializes to (3.67). The involution is given by
(3.68). As in the nite case, one checks that these operations make C
c
(G) a
-algebra; this time
one needs the invariance of the Haar measure at various steps of the proof.
Proposition 3.7.2 The operations (3.71) and (3.68) are continuous in the L
1
-norm; one has
| f g |
1
| f |
1
| g |
1
; (3.72)
| f
|
1
= | f |
1
. (3.73)
Hence L
1
(G) is a Banach
-algebra under the continuous extensions of (3.71) and (3.68) from
C
c
(G) to L
1
(G).
70 3 HILBERT C
|
1
=| f |
1
, so that
the involution is certainly continuous. The proof of (3.72) is a straightforward generalization of
the case G = R; cf. (2.55). This time we have
| f g |
1
=
G
dx[
G
dy f(xy
1
)g(y)[
G
dy [g(y)[
G
dx[f(xy
1
)[
=
G
dy [g(y)[
G
dx[f(x)[ =| f |
1
| g |
1
,
which is (3.72).
In order to equip L
1
(G) with a C
L
(f) := f . (3.74)
is bounded, satisfying |
L
(f) | | f |
1
. The linear map
L
: L
1
(G) B(L
2
(G)) is a faithful
representation of L
1
(G), seen as a Banach
-algebra as in 3.7.2.
Introducing the left-regular representation U
L
of G on L
2
(G) by
U
L
(y)(x) := (y
1
x), (3.75)
it follows that
L
(f) =
G
dxf(x)U
L
(x). (3.76)
The boundedness of
L
(f) then follows from Lemma 3.7.7 below. One easily veries that
L
(fg) =
L
(f)
L
(g) and
L
(f
) =
L
(f)
.
To prove that
L
is faithful, we rst show that L
1
(G) possesses the analogue of an approximate
unit (see 2.7.1 for C
of e, so that each ^
is invariant under x x
1
; this basis is
partially ordered by inclusion. Take I
= N
times
a normalization factor ensuring that | I
|
1
= 1. Eq. (2.91) then holds by virtue of (3.68) and the
invariance of ^
f(x) = N
dy f(y
1
x) and f I
(x) = N
dy f(xy
1
. For f C
c
(G) one therefore
has lim
f = f and lim
f I
. Hence | f |
1
= 0 by Lemma 3.7.4, so that f = 0 and
L
is injective.
Denition 3.7.5 The reduced group C
-algebra C
r
(G) is the smallest C
-algebra in B(L
2
(G))
containing
L
(C
c
(G)). In other words, C
r
(G) is the closure of the latter in the norm
| f |
r
:=|
L
(f) | . (3.77)
3.7 Group C
-algebras 71
Perhaps the simplest example of a reduced group algebra is obtained by taking G = R.
Proposition 3.7.6 One has the isomorphism
C
r
(R) C
0
(R). (3.78)
It follows from the discussion preceding 2.3.6 that the Fourier transform (2.58) maps L
1
(G)
into a subspace of C
0
(R) which separates points on R. It is clear that for every p R there is an
f L
1
(R) for which
f(p) = 0. In order to apply Lemma 2.4.7, we need to verify that
| f |
r
=|
f |
. (3.79)
Since the Fourier transform turns convolution into pointwise multiplication, the left-regular repre-
sentation
L
on L
2
(R) is Fourier-transformed into the action on L
2
(R) by multiplication operators.
Hence (3.78) follows from Lemma 2.4.7.
This example generalizes to arbitrary locally compact abelian groups. Let
G be the set of all
irreducible unitary representations U
f() :=
G
dxf(x)U
(x). (3.80)
By the same arguments as for G = R, one obtains
C
r
(G) C
0
(
G). (3.81)
We return to the general case, where G is not necessarily abelian. We have now found a C
-
algebra which may play the role of C(G) for locally compact groups. Unfortunately, the analogue
of Proposition 3.7.1 only holds for a limited class of groups. Hence we need a dierent construction.
Let us agree that here and in what follows, a unitary representation of a topological group is always
meant to be continuous.
Lemma 3.7.7 Let U be an arbitrary unitary representation of G on a Hilbert space H. Then
(f), dened by
(f) :=
G
dxf(x)U(x) (3.82)
is bounded, with
| (f) | | f |
1
. (3.83)
The integral (3.82) is most simply dened weakly, that is, by its matrix elements
(, (f)) :=
G
dxf(x)(, U(x)).
Since U is unitary, we have [(, (f))[ (F, F)
L
2
(G)
for all H, where F(x) :=| |
[f(x)[.
The Cauchy-Schwarz inequality then leads to [(, (f))[ | f |
1
| |
2
. Lemma 2.12.5 then
leads to (3.83).
Alternatively, one may dene (3.82) as a Bochner integral. We explain this notion in a more
general context.
Denition 3.7.8 Let X be a measure space and let B be a Banach space. A function f : X B
is Bochner-integrable with respect to a measure on X i
f is weakly measurable (that is, for each functional B
X
d(x)f(x) can be manipulated as if it were an
ordinary (Lebesgue) integral. For example, one has
|
X
d(x) f(x) |
X
d(x) | f(x) | . (3.84)
Thus reading (3.82) as a Bochner integral, (3.83) is immediate from (3.84).
The following result generalizes the correspondence between U
L
in (3.75) and
L
in (3.76) to
arbitrary representations.
Theorem 3.7.9 There is a bijective correspondence between non-degenerate representations of
the Banach
-algebra L
1
(G) which satisfy (3.83), and unitary representations U of G. This corre-
spondence is given in one direction by (3.82), and in the other by
U(x)(f) := (f
x
), (3.85)
where f
x
(y) := f(x
1
y). This bijection preserves direct sums, and therefore irreducibility.
Recall from 2.9.2 that any non-degenerate representation of a C
(I
(I
x
(I
x
) strongly on a dense domain. The property U(x)U(y) = U(xy) then follows from (3.85)
and (3.71). The unitarity of each U(x) follows by direct calculation, or from the following argument.
Since | (I
x
) | | I
x
|
1
= 1, we infer that | U(x) | 1 for all x. Hence also | U(x
1
) | 1, which
is the same as | U(x)
1
| 1 We see that U(x) and U(x)
1
are both contractions; this is only
possible when U(x) is unitary.
Finally, if U is reducible there is a projection E such that [E, U(x)] = 0 for all x G. It follows
from (3.82) that [(f), E] = 0 for all f, hence is reducible. Conversely, if is reducible then
[E, (I
x
)] = 0 for all x G; by the previous paragraph this implies [E, U(x)] = 0 for all x. The
nal claim then follows from Schurs lemma 2.12.3.1. 2
This theorem suggests looking at a dierent object from C
r
(G). Inspired by 2.10.2 one puts
Denition 3.7.10 The group C
-algebra C
| (f) |, (3.87)
where the sum is over all representations (L
1
(G)) of the form (3.82), in which U is an irre-
ducible unitary representation of G, and only one representative of each equivalence class of such
representations is included.
3.8 C
-algebra C
(G)) and f L
1
(G) one
has
| (f) | | f | | f |
1
. (3.88)
Hence the restriction (L
1
(G)) satises (3.83), and therefore corresponds to U(G) by Theorem
3.7.9. Conversely, given U(G) one nds (L
1
(G)) satisfying (3.83) by 3.7.9; it then follows from
(3.88) that one may extend to a representation of C
(G) by continuity.
In conjunction with (3.79), the second denition of C
r
(G). The reason is that for
G one has
(f) =
f() C, so that the norms (3.87) and (3.79) coincide. In particular, one has
C
(R
n
) C
0
(R
n
). (3.89)
For general locally compact groups, looking at 3.7.5 we see that
C
r
(G) =
L
(C
(G)) C
(G)/ ker(
L
). (3.90)
A Lie group group is said to be amenable when the equality C
r
(G) = C
(G)) is faithful i G is amenable. We have just seen that all locally compact abelian
groups are amenable. It follows from the Peter-Weyl theorem that all compact groups are amenable
as well. However, non-compact semi-simple Lie groups are not amenable.
3.8 C
I
(A +I) := (A) +I. (3.93)
Denition 3.8.1 An automorphic action of a group G on a C
x
y
(A) =
xy
(A); (3.94)
x
(AB) =
x
(A)
x
(A); (3.95)
(A
) = (A)
(3.96)
for all x, y G and A, B A.
A C
-algebra A, and
an automorphic action of G on A such that for each A A the function from G to A, dened by
x |
x
(A) |, is continuous.
74 3 HILBERT C
G
dx | f(x) | (3.97)
is nite. The operations
f g(x) :=
G
dy f(y)
y
(g(y
1
x)); (3.98)
f
(x) :=
x
(f(x
1
)
) (3.99)
turn L
1
(G, A, ) into a Banach
-algebra.
As usual, we have assumed that G is unimodular; with a slight modication one may extend
these formulae to the non-unimodular case. The integral (3.98) is dened as a Bochner integral;
the assumptions in Denition 3.7.8 are satised as a consequence of the continuity assumption in
the denition of a C
-dynamical system. To verify the properties (3.72) and (3.73) one follows the
same derivation as for L
1
(G), using (3.84) and (3.91). The completeness of L
1
(G, A, ) is proved
as in the case A = C, for which L
1
(G, A, ) = L
1
(G).
In order to generalize Theorem 3.7.9, we need
Denition 3.8.3 A covariant representation of a C
= (
x
(A)). (3.100)
Here is an elegant and useful method to construct covariant representations.
Proposition 3.8.4 Let (G, A, ) be a C
with
cyclic vector
(A)
of H
by
U(x)
(A)
:=
(
x
(A))
. (3.102)
This operator is well dened, and denes a unitary representation of G on H
.
If
(A)
(B)
then ((AB)
x
(A
B)) = 0 by (3.101), so that |
(
x
(A B))
|
2
= 0 by (2.119). Hence
(
x
(A))
(
x
(B)), so that U(x)
(A)
= U(x)
(B)
.
Furthermore, (3.94) implies that U(x)U(y) = U(xy), whereas (3.102) and (3.101) imply that
(U(x)
(A)
, U(x)
(B)
) = (
(A)
(B)
).
This shows rstly that U(x) is bounded on
(A)
by
continuity. Secondly, U(x) is a partial isometry, which is unitary from H
to the closure of
U(x)H
. Taking A =
x
1(B) in (3.102), one sees that U(x)H
(A)
, whose closure is H
because
. (3.103)
Proposition 3.8.4 describes the way unitary representations of the Poincare group are con-
structed in algebraic quantum eld theory, in which is then taken to be the vacuum state on
the algebra of local observables of the system in question. Note, however, that not all covariant
representations of a C
G
dx (f(x))U(x); (3.104)
in the other direction one denes Af : x Af(x) and
x
(f) : y
x
(f(x
1
y)), and puts
U(x)(f) = (
x
(f)); (3.105)
(A)(f) = (Af), (3.106)
where is a cyclic vector for a cyclic summand of (C
(G,
A)).
This bijection preserves direct sums, and therefore irreducibility.
The proof of this theorem is analogous to that of 3.7.9. The approximate unit in L
1
(G, A, )
is constructed by taking the tensor product of an approximate unit in L
1
(G) and an approximate
unit in A. The rest of the proof may then essentially be read o from 3.7.9.
Generalizing 3.7.10, we put
Denition 3.8.6 Let (G, A, ) be a C
(G, A, ) of
G and A is the closure of the Banach
-algebra algebra L
1
(G, A, ) in the norm
| f |:=|
u
(f) |, (3.107)
where
u
is the direct sum of all non-degenerate representations of L
1
(G, A, ) which are bounded
as in (3.83).
Equivalently, C
| (f) |, (3.108)
where the sum is over all representations (L
1
(G, A, )) of the form (3.104), in which (U, ) is an
irreducible covariant representation of (G, A, ), and only one representative of each equivalence
class of such representations is included.
Here we simply say that a covariant representation (U, ) is irreducible when the only bounded
operator commuting with all U(x) and (A) is a multiple of the unit. The equivalence between
the two denitions follows from (2.138) and Theorem 3.8.5.
Theorem 3.8.7 Let (G, A, ) be a C
-algebras
We now come to an important class of crossed products, in which A = C
0
(Q), where Q is a locally
compact Hausdor space, and
x
is dened as follows.
Denition 3.9.1 A (left-) action L of a group G on a space Q is a map L : GQ Q, satisfying
L(e, q) = q and L(x, L(y, q)) = L(xy, q) for all q Q and x, y G. If G and Q are locally compact
we assume that L is continuous. If G is a Lie group and Q is a manifold we assume that L is
smooth. We write L
x
(q) = xq := L(x, q).
We assume the reader is familiar with this concept, at least at a heuristic level. The main
example we shall consider is the canonical action of G on the coset space G/H (where H is a
closed subgroup of G). This action is given by
x[y]
H
:= [xy]
H
, (3.109)
where [x]
H
:= xH; cf. 3.1.6 etc. For example, when G = SO(3) and H = SO(2) is the subgroup
of rotations around the z-axis, one may identify G/H with the unit two-sphere S
2
in R
3
. The
SO(3)-action (3.109) is then simply the usual action on R
3
, restricted to S
2
.
Assume that Q is a locally compact Hausdor space, so that one may form the commutative
C
-algebra C
0
(Q); cf. 2.4. A G-action on Q leads to an automorphic action of G on C
0
(Q), given
by
x
(
f) : q
f(x
1
q). (3.110)
Using the fact that G is locally compact, so that e has a basis of compact neighbourhoods, it is
easy to prove that the continuity of the G-action on Q implies that
lim
xe
|
x
(
f)
f |= 0 (3.111)
for all
f C
c
(Q). Since C
c
(Q) is dense in C
0
(Q) in the sup-norm, the same is true for
f C
0
(Q).
Hence the function x
x
(
f) from G to C
0
(Q) is continuous at e (as
e
(
f) =
f). Using (3.94) and
(3.91), one sees that this function is continuous on all of G. Hence (G, C
0
(Q), ) is a C
-dynamical
system.
It is quite instructive to look at covariant representations (U, ) of (G, C
0
(Q), ) in the special
case that G is a Lie group and Q is a manifold. Firstly, given a unitary representation U of a Lie
group G on a Hilbert space H one can construct a representation of the Lie algebra g by
dU(X) :=
d
dt
U(Exp(tX))
|t=0
. (3.112)
When H is innite-dimensional this denes an unbounded operator, which is not dened on all
of H. Eq. (3.112) makes sense when is a smooth vector for a U; this is an element H
for which the map x U(x) from G to H is smooth. It can be shown that the set H
U
of
smooth vectors for U is a dense linear subspace of H, and that the operator idU(X) is essentially
self-adjoint on H
U
. Moreover, on H
U
one has
[dU(X), dU(Y )] = dU([X, Y ]). (3.113)
Secondly, given a Lie group action one denes a linear map X
X
from g to the space of all
vector elds on Q by
X
f(q) :=
d
dt
f(Exp(tX)q)
|t=0
, (3.114)
where Exp : g G is the usual exponential map.
The meaning of the covariance condition (3.100) on the pair (U, ) may now be claried by
re-expressing it in innitesimal form. For X g,
f C
c
(Q), and R`0 we put
O
(
X) := idU(X); (3.115)
O
(
f) := (
f). (3.116)
3.9 Transformation group C
-algebras 77
From the commutativity of C
0
(Q), (3.113), and (3.100), respectively, we then obtain
i
[O
(
f), O
( g)] = 0; (3.117)
i
[O
(
X), O
Y )] = O
[O
(
X), O
(
f)] = O
(
X
f). (3.119)
These equations hold on the domain H
U
, and may be seen as a generalization of the canonical
commutation relations of quantum mechanics. To see this, consider the case G = Q = R
n
,
where the G-action is given by L(x, q) := q + x. If X = T
k
is the kth generator of R
n
one has
k
:=
T
k
= /q
k
. Taking f = q
l
, the lth co-ordinate function on R
n
, one therefore obtains
k
q
l
=
l
k
. The relations (3.117) - (3.119) then become
i
[O
(q
k
), O
(q
l
)] = 0; (3.120)
i
[O
T
k
), O
T
l
)] = 0; (3.121)
i
[O
(
T
k
), O
(q
l
)] =
l
k
. (3.122)
Hence one may identify O
(q
k
) and O
(
T
k
) with the quantum position and momentum observables,
respectively. (It should be remarked that O
(q
k
) is an unbounded operator, but one may show
from the representation theory of the Heisenberg group that O
(q
k
) and O
(
T
k
) always possess a
common dense domain on which (3.120) - (3.122) are valid.)
Denition 3.9.2 Let L be a continuous action of a locally compact group on a locally compact
space Q. The transformation group C
-algebra C
(G, C
0
(Q), )
dened by the automorphic action (3.110).
Conventionally, the G-action L on Q is not indicated in the notation C
G
dx sup
qQ
[f(x, q)[ (3.123)
is nite; cf. (3.97). In this realization, the operations (3.98) and (3.99) read
f g(x, q) =
G
dy f(y, q)g(y
1
x, y
1
q); (3.124)
f
(x, q) = f(x
1
, x
1
q). (3.125)
As always, G is here assumed to be unimodular. Here is a simple example.
Proposition 3.9.3 Let a locally compact group G act on Q = G by L(x, y) := xy. Then
C
(G, G) B
0
(L
2
(G)) as C
-algebras.
We start from C
c
(GG), regarded as a dense subalgebra of C
G
dy f(xy
1
, x)(y). (3.126)
78 3 HILBERT C
) = (f)
, so that is
a representation of the
-algebra C
c
(G G). It is easily veried that the Hilbert-Schmidt-norm
(2.140) of (f) is
| (f) |
2
2
=
G
dxdy [f(xy
1
, x)[
2
. (3.127)
Since this is clearly nite for f C
c
(G G), we conclude from (2.153) that (C
c
(G G))
B
0
(L
2
(G)). Since (C
c
(G G)) is dense in B
2
(L
2
(G)) in the Hilbert-Schmidt-norm (which is a
standard fact of Hilbert space theory), and B
2
(L
2
(G)) is dense in B
0
(L
2
(G)) in the usual operator
norm (since by Denition 2.13.1 even B
f
(L
2
(G)) is dense in B
0
(L
2
(G))), we conclude that the
closure of (C
c
(GG)) in the operator norm coincides with B
0
(L
2
(G)).
Since is evidently faithful, the equality (C
(G, G)) = B
0
(L
2
(G)), and therefore the iso-
morphism C
(G, G) B
0
(L
2
(G)), follows from the previous paragraph if we can show that the
norm dened by (3.108) coincides with the operator norm of (). This, in turn, is the case if all
irreducible representations of the
-algebra C
c
(GG) are unitarily equivalent to .
To prove this, we proceed as in Proposition 3.4.4, in which we take
A = C
c
(G G),
B = C,
and
c = C
c
(G). The pre-Hilbert C
-module C
c
(G) C is dened by the obvious C-action on
C
c
(G), and the inner product
', `
C
:= (, )
L
2
(G)
. (3.128)
The left-action of
A on
c is as dened in (3.126), whereas the C
c
(G G)-valued inner product
on
C
c
(G) is given by
', `
Cc(GG)
:= (y)(x
1
y). (3.129)
It is not necessary to consider the bounds (3.45) and (3.46). Following the proof of Theorem
3.6.1, one shows directly that there is a bijective correspondence between the representations of
C
c
(GG) and of C.
3.10 The abstract transitive imprimitivity theorem
We specialize to the case where Q = G/H, where H is a closed subgroup of G, and the G-action
on G/H is given by (3.109). This leads to the transformation group C
-algebra C
(G, G/H).
Theorem 3.10.1 The transformation group C
-algebra C
(H).
We need to construct a full Hilbert C
-module c C
0
(c, C
(H)) is isomor-
phic to C
(G, G/H). This will be done on the basis of Proposition 3.4.4. For simplicity we assume
that both G and H are unimodular. In 3.4.4 we take
A = C
c
(G, G/H), seen as a dense subalgebra of A = C
B = C
c
(H), seen as a dense subalgebra of B = C
(H);
c = C
c
(G).
We make a pre-Hilbert C
c
(H)-module C
c
(G) C
c
(H) by means of the right-action
R
(f) = f : x
H
dh(xh
1
)f(h). (3.130)
Here f C
c
(H) and C
c
(G). The C
c
(H)-valued inner product on C
c
(G) is dened by
', `
Cc(H)
: h
G
dx(x)(xh). (3.131)
Interestingly, both formulae may be written in terms of the right-regular representation U
R
of H
on L
2
(G), given by
U
R
(h)(x) := (xh). (3.132)
3.11 Induced group representations 79
Namely, one has
R
(f) =
H
dhf(h)U(h
1
), (3.133)
which should be compared with (3.82), and
', `
Cc(H)
: h (, U(h))
L
2
(G)
. (3.134)
The properties (3.9) and (3.10) are easily veried from (3.68) and (3.71), respectively. To prove
(3.11), we take a vector state
on C
. Hence for
f C
c
(H) L
1
(H) one has
(f) = (
(f)
) =
H
dhf(h)(
, U
(h)
), (3.135)
where U
(C
G
dxf(x) =
G/H
d(q)
H
dhf(s(q)h) (3.136)
for any f C
c
(G), and any measurable map s : G/H G for which s = id (where : G G/H
is the canonical projection (x) := [x]
H
= xH). Combining (3.135), (3.131), and (3.136), we nd
(', `
Cc(H)
) =
G/H
d(q) |
H
dh(s(q)h)U
(h)
|
2
. (3.137)
Since this is positive, this proves that
(', `
Cc(H)
) is positive for all representations
of
C
L
(f)(x) =
G
dy f(xy
1
, [x]
H
)(y); (3.138)
', `
Cc(G,G/H)
: (x, [y]
H
)
H
dh(yh)(x
1
yh). (3.139)
Using (3.124) and (3.125), one may check that
L
is indeed a left-action, and that C
c
(G)
C
c
(G, G/H) is a pre-Hilbert C
R
(f) :=
L
(f
); cf. 3.4.4. Also, using (3.139), (3.138), (3.130), and (3.131), it is easy to verify
the crucial condition (3.44).
To complete the proof, one needs to show that the Hilbert C
-modules C
c
(G) C
c
(H) and
C
c
(G) C
c
(G, G/H) are full, and that the bounds (3.45) and (3.46) are satised. This is indeed
the case, but an argument that is suciently elementary for inclusion in these notes does not
seem to exist. Enthusiastic readers may nd the proof in M.A. Rieel, Induced representations of
C
dened by (3.5) by
means of
U
(x) : [y, v]
H
[xy, v]
H
. (3.140)
Since the left-action x : y xy of G on itself commutes with the right-action h : y yh of H on
G, the action (3.140) is clearly well dened.
The G-action U
) of H
, dened on
()
(H
) by
U
()
(x)
()
(q) := U
(x)
()
(x
1
q)). (3.141)
One should check that U
()
(x)
()
is again a section, in that
(U
()
(x)
()
(q)) = q; see (3.6).
This section is evidently continuous, since the G-action on G/H is continuous.
There is a natural inner product on the space of sections (H
), given by
(
()
,
()
) :=
G/H
d(q) (
()
(q),
()
(q))
, (3.142)
where is the measure on G/H dened by (3.136), and ( , )
(q) H
) of continuous sections
of H
with compact support (in the norm derived from this inner product).
When the measure is G-invariant (which is the case, for example, when G and H are uni-
modular), the operator U
()
(x) dened by (3.141) satises
(U
()
(x)
()
, U
()
(x)
()
) = (
()
,
()
). (3.143)
When fails to be G-invariant, it can be shown that it is still quasi-invariant in the sense that ()
and (x
1
) have the same null sets for all x G. Consequently, the Radon-Nikodym derivative
q d(x
1
(q))/d(q) exists as a measurable function on G/H. One then modies (3.141) to
U
()
(x)
()
(q) :=
d(x
1
(q))
d(q)
U
(x)
()
(x
1
q)). (3.144)
Proposition 3.11.1 Let G be a locally compact group with closed subgroup H, and let U
be a
unitary representation of H on a Hilbert space H
) of L
2
-sections
of the Hilbert bundle H
as the completion of
c
(H
), so that it is bounded,
and can be extended to L
2
(H
) by continuity. Since U
()
(x) is invertible, with inverse U
()
(x
1
),
it is therefore a unitary operator. The property U
()
(x)U
()
(y) = U
()
(xy) is easily checked.
The representation U
()
(G) is said to be induced by U
(H).
Proposition 3.11.2 In the context of 3.11.1, dene a representation
()
(C
0
(G/H)) on L
2
(H
)
by
()
(
f)
()
(q) :=
f(q)
()
(q). (3.145)
The pair (U
()
(G),
()
(C
0
(G/H))) is a covariant representation of the C
-dynamical system
(G, C
0
(G/H), ), where is given by (3.110).
3.11 Induced group representations 81
Given 3.11.1, this follows from a simple computation.
Note that the representation (3.145) is nothing but the right-action (3.1) of (C
0
(G/H)) on
L
2
(H
of the bundle H
may alternatively be
represented as a map
: G H
(xh
1
) = U
(h)
(x). (3.146)
Such a map denes a section
()
by
()
((x)) = [x,
(x)]
H
, (3.147)
where : G G/H is given by (3.7). The section
()
thus dened is independent of the choice
of x
1
((x)) because of (3.146).
For
()
to lie in
c
(H
) is given by
(
) :=
G/H
d((x)) (
(x),
(x))
; (3.148)
the integrand indeed only depends on x through (x) because of (3.146).
Denition 3.11.3 The Hilbert space H
: G H
on
by
[y, U
(x)
(y)]
H
:= U
()
(x)
()
((y)). (3.149)
Using (3.141), (3.147), and (3.140), as well as the denition x(y) = x[y]
H
= [xy]
H
= (xy) of the
G-action on G/H (cf. (3.7)), we obtain
U
()
(x)
()
((y)) = U
(x)
()
(x
1
(y))) = U
(x)[x
1
y,
(x
1
y)]
H
= [y,
(x
1
y)]
H
.
Hence we infer from (3.149) that
U
(y)
(x) =
(y
1
x). (3.150)
Replacing (3.141) by (3.144) in the above derivation yields
U
(y)
(x) =
d((y
1
x))
d((x))
(y
1
x). (3.151)
Similarly, in the realization H
(
f)
(x) :=
f([x]
H
)
(x). (3.152)
Analogous to 3.11.2, we then have
Proposition 3.11.4 In the context of 3.11.1, dene a representation
(C
0
(G/H)) on H
(cf.
3.11.3) by (3.152). The pair (U
(G),
(C
0
(G/H))), where U
H
()
given by
V
((x)) := [x,
(x)]
H
, (3.153)
in the sense that
V U
(y)V
1
= U
()
(y) (3.154)
for all y G, and
V
(
f)V
1
=
()
(
f) (3.155)
for all
f C
0
(G/H).
82 3 HILBERT C
(G),
(C
0
(G/H)). The original imprimitivity theorem of Mackey, which historically preceded
Theorems 3.6.1 and 3.10.1, states that all covariant pairs (U(G), (C
0
(G/H)) arise in this way.
Theorem 3.12.1 Let G be a locally compact group with closed subgroup H, and consider the
C
(G),
(C
0
(G/H)) of
(G, C
0
(G/H), ), given by 3.11.3, (3.151) and (3.152). Conversely, any covariant representation
(U, ) of (G, C
0
(G/H), ) is unitarily equivalent to a pair of this form.
This leads to a bijective correspondence between the space of equivalence classes of unitary
representations of H and the space of equivalence classes of covariant representations (U, ) of
the C
(H) and C
(G, G/H))
to (U(G), (C
0
(G/H)), respectively.
The explicit form of the correspondence remains to be established. Let us start with a technical
point concerning Rieel induction in general. Using (3.57), (3.47), and (3.13), one shows that
|
V | | |, where the norm on the left-hand side is in
H
obtained by Rieel-inducing
from a pre-Hilbert C
-module is the same as the induced space constructed from its completion.
The same comment, of course, applies to H
.
We will use a gerenal technique that is often useful in problems involving Rieel induction.
Lemma 3.12.2 Suppose one has a Hilbert space H
) and a
linear map
U : c H
satisfying
(
U
,
U
= (
,
)
0
(3.157)
for all
,
c H
.
Then
U quotients to an isometric map between c H
/^
. When
the image is dense this map extends to a unitary isomorphism U : H
. Otherwise, U is
unitary between H
(C
(C
(c, B)),
dened by continuous extension of
(A)
U
:=
U(A I
). (3.158)
It is obvious that ^
= ker(
U.
We use this lemma in the following way. To avoid notational confusion, we continue to de-
note the Hilbert space H
-module
83
C
c
(G) C
c
(H) dened in the proof of 3.10.1, by H
Uv(x) :=
H
dh(xh)U
(h)v. (3.159)
Note that the equivariance condition (3.146) is indeed satised by the left-hand side, as follows
from the invariance of the Haar measure.
Using (3.54), (3.131), and (3.82), with G H, one obtains
(v, w)
0
=
H
dh(, U
R
(h))
L
2
(G)
(v, U
(h)w)
H
dh
G
dx(x)(xh)(v, U
(h)w)
;
(3.160)
cf. (3.132). On the other hand, from (3.159) and (3.148) one has
(
Uv,
U w)
H
H
dh
G/H
d((x))
H
dk (xk)(xh)(U
(k)v, U
(h)w)
. (3.161)
Shifting h kh, using the invariance of the Haar measure on H, and using (3.136), one veries
(3.157). It is clear that
U(C
c
(G) H
) is dense in H
.
Using (3.158) and (3.138), one nds that the induced representation of C
(G, G/H) on H
is
given by
(f)
(x) =
G
dy f(xy
1
, [x]
H
)
(y); (3.162)
this looks just like (3.138), with the dierence that in (3.138) lies in C
c
(G), whereas
in
(3.162) lies in H
(f)
(C
(G),
(C
0
(G/H)) by the correspondence
(3.104) - (3.106) of Theorem 3.8.5.
4 Applications to quantum mechanics
4.1 The mathematical structure of classical and quantum mechanics
In classical mechanics one starts from a phase space S, whose points are interpreted as the pure
states of the system. More generally, mixed states are identied with probability measures on S.
The observables of the theory are functions on S; one could consider smooth, continuous, bounded,
measurable, or some other other class of real-vaued functions. Hence the space A
R
of observables
may be taken to be C
(S, R), C
0
(S, R), C
b
(S, R), or L
S
d() f(). (4.1)
The physical interpretation of this pairing is that in a state the observable f has expectation
value ', f`. In general, this expectation value will be unsharp, in that ', f`
2
= ', f
2
`. However,
in a pure state (seen as the Dirac measure
-algebra B
0
(H) of all compact operators on H;
see Corollary 2.13.10.1. On the other hand, with the same choice of the state space, if we take A
R
to be the space B(H)
R
of all bounded self-adjoint operators on H, then the space of observables
is the dual of the (linear space spanned by the) state space, rather then vice versa; see Theorem
2.13.8.
In the C
(S, R), or C
0
(S, R), etc, and L(H) denotes some
space of self-adjoint operators on H, such as B
0
(H)
R
or B(H)
R
. Given the physical meaning of
a classical observable f, one then ascribes the same physical interpretation to the corresponding
quantum observable O(f). This provides the physical meaning of al least all operators in the
image of O. It is desirable (though not strictly necessary) that O preserves positivity, as well as
the (approximate) unit.
It is quite convenient to assume that A
0
R
= C
0
(S, R), which choice discards what happens at
innity on S. We are thus led to the following
Denition 4.2.1 Let X be a locally compact Hausdor space. A quantization of X consists of
a Hilbert space H and a positive map O : C
0
(X) B(H). When X is compact it is required that
O(1
X
) = I, and when X is non-compact one demands that O can be extended to the unitization
C
0
(X)
I
by a unit-preserving positive map.
4.2 Quantization 85
Here C
0
(X) and B(H) are, of course, regarded as C
(4.5)
for all f C
0
(X); this implies that f C
0
(X, R) is mapped into a self-adjoint operator.
There is an interesting reformulation of the notion of a quantization in the above sense.
Denition 4.2.2 Let X be a set with a -algebra of subsets of X. A positive-operator-
valued measure or POVM on X in a Hilbert space H is a map A() from to B(H)
+
(the set of positive operators on H), satisfying A() = 0, A(X) = I, and A(
i
i
) =
i
A(
i
) for
any countable collection of disjoint
i
(where the innite sum is taken in the weak operator
topology).
A projection-valued measure or PVM is a POVM which in addition satises A(
1
2
) =
A(
1
)A(
2
) for all
1
,
2
.
Note that the above conditions force 0 A() I. A PVM is usually written as E();
it follows that each E() is a projection (take
1
=
2
in the denition). This notion is familiar
from the spectral theorem.
Proposition 4.2.3 Let X be a locally compact Hausdor space, with Borel structure . There is
a bijective correspondence between quantizations O : C
0
(X) B(H), and POVMs A() on
S in H, given by
O(f) =
S
dA(x) f(x). (4.6)
The map O is a representation of C
0
(X) i A() is a PVM.
The precise meaning of (4.6) will emerge shortly. Given the assumptions, in view of 2.3.7 and
2.4.6 we may as well assume that X is compact.
Given O, for arbitrary H one constructs a functional
,
on C(X) by
,
(f) :=
(, O(f)) Since O is linear and positive, this functional has the same properties. Hence the
Riesz representation theorem yields a probability measure
,
on X. For one then puts
(, A()) :=
,
(), dening an operator A() by polarization. The ensuing map A()
is easily checked to have the properties required of a POVM.
Conversely, for each pair , H a POVM A() in H denes a signed measure
,
on X by means of
,
() := (, A()). This yields a positive map O : C(X) B(H) by
(, O(f)) :=
X
d
,
(x) f(x); the meaning of (4.6) is expressed by this equation.
Approximating f, g C(X) by step functions, one veries that the property E()
2
= E() is
equivalent to O(fg) = O(f)O(g).
Corollary 4.2.4 Let A() be a POVM on a locally compact Hausdor space X in a Hilbert
space H
, a projection p on H
, a unitary map U : H
pH
,
and a PVM E() on H
() := Tr A() (4.7)
is the probability that, in a state , the system in question is localized in S.
When X is a conguration space Q, it is usually sucient to take the positive map O to be
a representation of C
0
(Q) on H. By Proposition 4.2.3, the situation is therefore described by a
PVM E() on Q in H. The probability that, in a state , the system is localized in Q
is
p
() := Tr E(). (4.8)
86 4 APPLICATIONS TO QUANTUM MECHANICS
4.3 Stinesprings theorem and coherent states
By Proposition 2.11.4, a quantization O : C
0
(X) B(H) is a completely positive map, and
Denition 4.2.1 implies that the conditions for Stinesprings Theorem 2.11.2 are satised. We
will now construct a class of examples of quantization in which one can construct an illuminating
explicit realization of the Hilbert space H
S
d()(
1
,
)(
,
2
) = (
1
,
2
). (4.9)
for all
1
,
2
H. The
d() [
], (4.10)
where [] is the projection onto the one-dimensional subspace spanned by (in Diracs notation
one would have [] = [ >< [).
The positive map O corresponding to the POVM A() by Proposition 4.2.3 is given by
O(f) =
S
d() f()[
]. (4.11)
In particular, one has O(1
S
) = I.
For example, when S = T
R
3
= R
6
, so that = (p, q), one may take
(p,q)
(x) = ()
n/4
e
1
2
ipq+ipx
e
(xq)
2
/2
(4.12)
in H = L
2
(R
3
). Eq. (4.9) then holds with d(p, q) = d
3
pd
3
q/(2)
3
. Extending the map O from
C
0
(S) to C
= L
2
(S, d). (4.13)
The map W : H H
is then given by
W() := (
, ). (4.14)
It follows from (4.9) that W is a partial isometry. The representation (C
0
(S)) is given by
(f)() = f()(). (4.15)
Finally, for (2.121) one has the simple expression
O(f) = UO(f)U
1
= pfp. (4.16)
Eqs. (4.13) and (4.16) form the core of the realization of quantum mechanics on phase space.
One realizes the state space as a closed subspace of L
2
(S) (dened with respect to a suitable
measure), and denes the quantization of a classical observable f C
0
(S) as multiplication by f,
sandwiched between the projection onto the subspace in question. This should be contrasted with
the usual way of doing quantum mechanics on L
2
(Q), where Q is the conguration space of the
system.
In specic cases the projection p = WW
R
3
considered above one may pass to complex variables by putting z = (q ip)/
2.
4.4 Covariant localization in conguration space 87
We then map L
2
(T
R
3
, d
3
pd
3
q/(2)
3
) into / := L
2
(C
3
, d
3
zd
3
z exp(zz)/(2i)
3
) by the unitary
operator V , given by
V (z, z) := e
1
2
zz
(p = (z z)/
2, q = (z +z)/
2). (4.17)
One may then verify from (4.14) and (4.12) that V pV
1
is the projection onto the space of entire
functions in /.
4.4 Covariant localization in conguration space
In elementary quantum mechanics a particle moving on R
3
with spin j N is described by the
Hilbert space
H
j
QM
= L
2
(R
3
) H
j
, (4.18)
where H
j
= C
2j+1
carries the irreducible representation U
j
(SO(3)) (usually called T
j
). The basic
physical observables are represented by unbounded operators Q
S
k
(position), P
S
k
(momentum), and
J
S
k
(angular momentum), where k = 1, 2, 3. These operators satisfy the commutation relations
(say, on the domain o(R
3
) H
j
)
[Q
S
k
, Q
S
l
] = 0; (4.19)
[P
S
k
, Q
S
l
] = i
kl
; (4.20)
[J
S
k
, Q
S
l
] = i
klm
Q
S
m
; (4.21)
[P
S
k
, P
S
l
] = 0; (4.22)
[J
S
k
, J
S
l
] = i
klm
J
S
m
; (4.23)
[J
S
k
, P
S
l
] = i
klm
P
S
m
, (4.24)
justifying their physical interpretation.
The momentum and angular momentum operators are most conveniently dened in terms of a
unitary representation U
j
QM
of the Euclidean group E(3) = SO(3) R
3
on H
j
QM
, given by
U
j
QM
(R, a)(q) = U
j
(R)(R
1
(q a)). (4.25)
In terms of the standard generators P
k
and T
k
of R
3
and SO(3), respectively, one then has P
S
k
=
idU
j
QM
(P
k
) and J
S
k
= idU
j
QM
(T
k
); see (3.112). The commutation relations (4.22) - (4.24) follow
from (3.113) and the commutation relations in the Lie algebra of E(3).
Moreover, we dene a representation
j
QM
of C
0
(R
3
) on H
j
QM
by
j
QM
(
f) =
f I
j
, (4.26)
where
f is seen as a multiplication operator on L
2
(R
3
). The associated PVM E() on R
3
in H
j
QM
(see 4.2.3) is E() =
I
j
, in terms of which the position operators are given by
Q
S
k
=
R
3
dE(x)x
k
; cf. the spectral theorem for unbounded operators. Eq. (4.19) then reects the
commutativity of C
0
(R
3
), as well as the fact that
j
QM
is a representation.
Identifying Q = R
3
with G/H = E(3)/SO(3) in the obvious way, one checks that the canonical
left-action of E(3) on E(3)/SO(3) is identied with its dening action on R
3
. It is then not hard to
verify from (4.25) that the pair (U
j
QM
(E(3)),
j
QM
(C
0
(R
3
))) is a covariant representation of the C
-
dynamical system (E(3), C
0
(R
3
), ), with given by (3.110). The commutation relations (4.20),
(4.21) are a consequence of the covariance relation (3.156).
Rather than using the unbounded operators Q
S
k
, P
S
k
, and J
S
k
, and their commutation rela-
tions, we therefore state the situation in terms of the pair (U
j
QM
(E(3)),
j
QM
(C
0
(R
3
))). Such a
pair, or, equivalently, a non-degenerate representation
j
QM
of the transformation group C
-algebra
C
(E(3), R
3
) (cf. 3.8.7, then by denition describes a quantum system which is localizable in R
3
.
and covariant under the dening action of E(3). It is natural to require that
j
QM
be irreducible,
in which case the quantum system itself is said to be irreducible.
88 4 APPLICATIONS TO QUANTUM MECHANICS
Proposition 4.4.1 An irreducible quantum system which is localizable in R
3
and covariant under
E(3) is completely characterized by its spin j N. The corresponding covariant representation
(U
j
(E(3)),
j
(C
0
(R
3
))), given by 3.11.3, (3.151), and (3.152), is equivalent to the one described
by (4.18), (4.25), and (4.26).
This follows from Theorem 3.12.1. The representation U
j
QM
(E(3)) dened in (4.25) is unitarily
equivalent to the induced representation U
j
. To see this, check that the unitary map V : H
j
H
j
QM
dened by V
j
(q) :=
j
(e, q) intertwines U
j
and U
j
QM
. In addition, it intertwines the
representation (3.152) with
j
QM
as dened in (4.26).
This is a neat explanation of spin in quantum mechanics.
Generalizing this approach to an arbitrary homogeneous conguration space Q = G/H, a non-
degenerate representation of C
(G),
j
(C
0
(G/H))) is equivalent to the one described by (3.151) and
(3.152).
This is immediate from Theorem 3.12.1.
For example, writing the two-sphere S
2
as SO(3)/SO(2), one infers that SO(3)-covariant quan-
tum particles on S
2
are characterized by an integer n Z. For each unitary irreducible represen-
tation U of SO(2) is labeled by such an n, and given by U
n
() = exp(in).
4.5 Covariant quantization on phase space
Let us return to quantization theory, and ask what happens in the presence of a symmetry group.
The following notion, which generalizes Denition 3.8.3, is natural in this context.
Denition 4.5.1 A generalized covariant representation of a C
-dynamical system
(G, C
0
(X), ), where arises from a continuous G-action on X by means of (3.110), consists of a
pair (U, O), where U is a unitary representation of G on a Hilbert space H, and O : C
0
(X) B(H)
is a quantization of C
0
(X) (in the sense of Denition 4.2.1), which for all x G and
f C
0
(X)
satises the covariance condition
U(x)O(
f)U(x)
= O(
x
(
f)). (4.28)
This condition may be equivalently stated in terms of the POVM A() associated to O
(cf. 4.2.3) by
U(x)A()U(x)
1
= A(x). (4.29)
Every (ordinary) covariant representation is evidently a generalized one as well, since a rep-
resentation is a particular example of a quantization. A class of examples of truly generalized
covariant representations arises as follows. Let (U(G), (C
0
(G/H)) be a covariant representation
on a Hilbert space /, and suppose that U(G) is reducible. Pick a projection p in the commu-
tant of U(G); then (pU(G), p p) is a generalized covariant representation on H = p/. Of course,
(U, ) is described by Theorem 3.12.1, and must be of the form (U
-
dynamical system (G, C
0
(G/H), ), dened with respect to the canonical G-action on G/H.
There exists a unitary representation U
)
of (G, C
0
(G/H), ) on the Hilbert space H
in the commutant of U
(G), p
in Construction 2.11.3 by
H
and
and
,
H
, H
). We identify B
in 2.11.2 and 2.11.3 with B(H), where H is specied in 4.5.2; we therefore omit the representation
= H.
For x G we dene a linear map
U(x) on C
0
(G/H) H by linear extension of
U(x)f :=
x
(f) U(x). (4.30)
Since
x
y
=
xy
, and U is a representation,
U is clearly a G-action. Using the covariance
condition (4.28) and the unitarity of U(x), one veries that
(
U(x)f ,
U(x)g )
0
= (f , g )
0
, (4.31)
where ( , )
0
is dened in (2.122). Hence
U(G) quotients to a representation
U
(G) on
H
. Com-
puting on C
0
(G/H) H and then passing to the quotient, one checks that (
) is a covariant
representation on
H
) (up to unitary
equivalence).
Finally, the projection p dened in 2.11.2 commutes with all
U
- and W
-algebras:
1943-1993. Cont. Math. 167, pp. 21-53. Amer. Math. Soc., Providence (RI).
Kadison, R.V. and J.R. Ringrose [1983] Fundamentals of the Theory of Operator Algebras I. Aca-
demic Press, New York.
Kadison, R.V. and J.R. Ringrose [1986] Fundamentals of the Theory of Operator Algebras II.
Academic Press, New York.
90 4 APPLICATIONS TO QUANTUM MECHANICS
Lance, E.C. [1995] Hilbert C