Lecture Notes on Operator Algebras

John M. Erdman
Portland State University

Version March 12, 2011

c 2010 John M. Erdman

E-mail address: [email protected]

 Contents

Chapter 1. LINEAR ALGEBRA AND THE SPECTRAL THEOREM 3

1.1. Vector Spaces and the Decomposition of Diagonalizable Operators 3
1.2. Normal Operators on an Inner Product Space 6

Chapter 2. THE ALGEBRA OF HILBERT SPACE OPERATORS 13

2.1. Hilbert Space Geometry 13
2.2. Operators on Hilbert Spaces 16
2.3. Algebras 19
2.4. Spectrum 21

Chapter 3. BANACH ALGEBRAS 23

3.1. Definition and Elementary Properties 23
3.2. Maximal Ideal Space 26
3.3. Characters 27
3.4. The Gelfand Topology 28
3.5. The Gelfand Transform 31
3.6. The Fourier Transform 32

Chapter 4. INTERLUDE: THE LANGUAGE OF CATEGORIES 35

4.1. Objects and Morphisms 35
4.2. Functors 36
4.3. Natural Transformations 38
4.4. Universal Morphisms 39

Chapter 5. C ∗ -ALGEBRAS 43
5.1. Adjoints of Hilbert Space Operators 43
5.2. Algebras with Involution 44
5.3. C ∗ -Algebras 46
5.4. The Gelfand-Naimark Theorem—Version I 47

Chapter 6. SURVIVAL WITHOUT IDENTITY 49

6.1. Unitization of Banach Algebras 49
6.2. Exact Sequences and Extensions 51
6.3. Unitization of C ∗ -algebras 53
6.4. Quasi-inverses 55
6.5. Positive Elements in C ∗ -algebras 57
6.6. Approximate Identities 59

Chapter 7. SOME IMPORTANT CLASSES OF HILBERT SPACE OPERATORS 61

7.1. Orthonormal Bases in Hilbert Spaces 61
7.2. Projections and Partial Isometries 63
7.3. Finite Rank Operators 65
7.4. Compact Operators 66
iii
iv CONTENTS

Chapter 8. THE GELFAND-NAIMARK-SEGAL CONSTRUCTION 69

8.1. Positive Linear Functionals 69
8.2. Representations 69
8.3. The GNS-Construction and the Third Gelfand-Naimark Theorem 70
Chapter 9. MULTIPLIER ALGEBRAS 73
9.1. Hilbert Modules 73
9.2. Essential Ideals 76
9.3. Compactifications and Unitizations 78
Chapter 10. FREDHOLM THEORY 81
10.1. The Fredholm Alternative 81
10.2. The Fredholm Alternative – continued 82
10.3. Fredholm Operators 83
10.4. The Fredholm Alternative – Concluded 84
Chapter 11. EXTENSIONS 87
11.1. Essentially Normal Operators 87
11.2. Toeplitz Operators 88
11.3. Addition of Extensions 90
11.4. Tensor Products of C ∗ -algebras 94
11.5. Completely Positive Maps 94
Chapter 12. K-THEORY 99
12.1. Projections on Matrix Algebras 99
12.2. The Grothendieck Construction 101
12.3. K0 (A)—the Unital Case 103
12.4. K0 (A)—the Nonunital Case 105
12.5. Exactness and Stability Properties of the K0 Functor 106
12.6. Inductive Limits 107
12.7. Bratteli Diagrams 108
Bibliography 113
Index 115
It is not essential for the value of an education that every idea be understood at the
time of its accession. Any person with a genuine intellectual interest and a wealth
of intellectual content acquires much that he only gradually comes to understand
fully in the light of its correlation with other related ideas. . . . Scholarship is a
progressive process, and it is the art of so connecting and recombining individual
items of learning by the force of one’s whole character and experience that nothing
is left in isolation, and each idea becomes a commentary on many others.

- NORBERT WIENER

1
 CHAPTER 1

LINEAR ALGEBRA AND THE SPECTRAL THEOREM

1.1. Vector Spaces and the Decomposition of Diagonalizable Operators

1.1.1. Convention. In this course, unless the contrary is explicitly stated, all vector spaces will
be assumed to be vector spaces over C. That is, scalar will be taken to mean complex number.
1.1.2. Definition. The triple (V, +, M ) is a (complex) vector space if (V, +) is an Abelian
group and M : C → Hom(V ) is a unital ring homomorphism (where Hom(V ) is the ring of group
homomorphisms on V ).
A function T : V → W between vector spaces is linear if T (u + v) = T u + T v for all u,
v ∈ V and T (αv) = αT v for all α ∈ C and v ∈ V . Linear functions are frequently called linear
transformations or linear maps. When V = W we say that T is an operator on V . The collection
of all linear maps from V to W is denoted by L(V, W ) and the set of operators on V is denoted
by L(V ). Depending on context we denote the identity operator x 7→ x on V by idV or IV or
just I. Recall that if T : V → W is a linear map, then the kernel of T , denoted by ker T , is
T ← ({0}) = {x ∈ V : T x = 0}. Also, the range of T , denoted by ran T , is T → (V ) = {T x : x ∈ V }.
1.1.3. Definition. A linear map T : V → W between vector spaces is invertible (or is an
isomorphism) if there exists a linear map T −1 : W → V such that T −1 T = idV and T T −1 = idW .
Recall that if a linear map is invertible its inverse is unique. Recall also that for a linear operator
T on a finite dimensional vector space the following are equivalent:
(a) T is an isomorphism;
(b) T is injective;
(c) the kernel of T is {0}; and
(d) T is surjective.
1.1.4. Definition. Two operators R and T on a vector space V are similar if there exists an
invertible operator S on V such that R = S −1 T S.
1.1.5. Proposition. If V is a vector space, then similarity is an equivalence relation on L(V ).
1.1.6. Definition. Let V be a finite dimensional vector space and B = {e1 , . . . , en } be a basis
for V . An operator T on V is diagonal if there exist scalars α1 , . . . , αn such that T ek = αk ek for
each k ∈ Nn . Equivalently, T is diagonal if its matrix representation [T ] = [Tij ] has the property
that Tij = 0 whenever i 6= j.
Asking whether a particular operator on some finite dimensional vector space is diagonal is,
strictly speaking, nonsense. As defined the operator property of being diagonal is definitely not a
vector space concept. It makes sense only for a vector space for which a basis has been specified.
This important, if obvious, fact seems to go unnoticed in beginning linear algebra courses, due,
I suppose, to a rather obsessive fixation on Rn in such courses. Here is the relevant vector space
property.
1.1.7. Definition. An operator T on a finite dimensional vector space V is diagonalizable if
there exists a basis for V with respect to which T is diagonal. Equivalently, an operator on a finite
dimensional vector space with basis is diagonalizable if it is similar to a diagonal operator.
3
4 1. LINEAR ALGEBRA AND THE SPECTRAL THEOREM

1.1.8. Definition. Let M and N be subspaces of a vector space V . If M ∩N = {0} and M +N = V ,

then V is the (internal) direct sum of M and N . In this case we write
V =M ⊕N.
We say that M and N are complementary subspaces and that each is a (vector space) comple-
ment of the other. The codimension of the subspace M is the dimension of its complement N .
1.1.9. Example. Let C = C[−1, 1] be the vector space of all continuous real valued functions on
the interval [−1, 1]. A function f in C is even if f (−x) = f (x) for all x ∈ [−1, 1]; it is odd if
f (−x) = −f (x) for all x ∈ [−1, 1]. Let Co = {f ∈ C : f is odd } and Ce = {f ∈ C : f is even }.
Then C = Co ⊕ Ce .
1.1.10. Proposition. If M is a subspace of a vector space V , then there exists a subspace N of V
such that V = M ⊕ N .
1.1.11. Proposition. Let V be a vector space and suppose that V = M ⊕ N . Then for every v ∈ V
there exist unique vectors m ∈ M and n ∈ N such that v = m + n.
1.1.12. Definition. Let V be a vector space and suppose that V = M ⊕ N . We know from 1.1.11
that for each v ∈ V there exist unique vectors m ∈ M and n ∈ N such that v = m + n. Define
a function EM N : V → V by EM N v = n. The function EM N is the projection of V along M
onto N . (Frequently we write E for EM N . But keep in mind that E depends on both M and N .)
1.1.13. Proposition. Let V be a vector space and suppose that V = M ⊕ N . If E is the projection
of V along M onto N , then
(i) E is linear;
(ii) E 2 = E (that is, E is idempotent);
(iii) ran E = N ; and
(iv) ker E = M .
1.1.14. Proposition. Let V be a vector space and suppose that E : V → V is a function which
satisfies
(i) E is linear, and
(ii) E 2 = E.
Then
V = ker E ⊕ ran E
and E is the projection of V along ker E onto ran E.
It is important to note that an obvious consequence of the last two propositions is that a
function T : V → V from a finite dimensional vector space into itself is a projection if and only if
it is linear and idempotent.
1.1.15. Proposition. Let V be a vector space and suppose that V = M ⊕ N . If E is the projection
of V along M onto N , then I − E is the projection of V along N onto M .
As we have just seen, if E is a projection on a vector space V , then the identity operator on
V can be written as the sum of two projections E and I − E whose corresponding ranges form a
direct sum decomposition of the space V = ran E ⊕ ran(I − E). We can generalize this to more
than two projections.
1.1.16. Definition. Suppose that on a vector space V there exist projection operators E1 , . . . ,
En such that
(i) IV = E1 + E2 + · · · + En and
(ii) Ei Ej = 0 whenever i 6= j.
Then we say that IV = E1 + E2 + · · · + En is a resolution of the identity.
 1.1. VECTOR SPACES AND THE DECOMPOSITION OF DIAGONALIZABLE OPERATORS 5

1.1.17. Proposition. If IV = E1 + E2 + · · · + En is a resolution of the identity on a vector space

V , then V = nk=1 ran Ek .
L

1.1.18. Example. Let P be the plane in R3 whose equation is x − z = 0 and L be the line whose
equations are y = 0 and x = −z. Let E be the projection of R3 along L onto P and F be the
projection of R3 along P onto L. Then
1 1
 1
0 − 12
 
2 0 2 2
[E] =  0 1 0  and [F ] =  0 0 0  .
1 1
2 0 2 − 12 0 12
1.1.19. Definition. A complex number λ is an eigenvalue of an operator T on a vector space
V if ker(T − λIV ) contains a nonzero vector. Any such vector is an eigenvector of T associated
with λ and ker(T − λIV ) is the eigenspace of T associated with λ. The set of all eigenvalues of
the operator T is its point spectrum and is denoted by σp (T ).
If M is an n × n matrix, then det(M − λIn ) (where In is the n × n identity matrix) is a
polynomial in λ of degree n. This is the characteristic polynomial of M . A standard way
of computing the eigenvalues of an operator T on a finite dimensional vector space is to find the
zeros of the characteristic polynomial of its matrix representation. It is an easy consequence of
the multiplicative property of the determinant function that the characteristic polynomial of an
operator T on a vector space V is independent of the basis chosen for V and hence of the particular
matrix representation of T that is used.
1.1.20. Example.  The eigenvalues
 of the operator on (the real vector space) R3 whose matrix
0 0 2
representation is 0 2 0 are −2 and +2, the latter having (both algebraic and geometric)
2 0 0
multiplicity 2. The eigenspace associated with the negative eigenvalue is span{(1, 0, −1)} and the
eigenspace associated with the positive eigenvalue is span{(1, 0, 1), (0, 1, 0)}.

The central fact asserted by the finite dimensional vector space version of the spectral theorem
is that every diagonalizable operator on such a space can be written as a linear combination of
projection operators where the coefficients of the linear combination are the eigenvalues of the
operator and the ranges of the projections are the corresponding eigenspaces. Thus if T is a
diagonalizable operator on a finite dimensional vector space V , then V has a basis consisting of
eigenvectors of T .
Here is a formal statement of the theorem.
1.1.21. Theorem (Spectral Theorem: vector space version). Suppose that T is a diagonalizable
operator on a finite dimensional vector space V . Let λ1 , . . . , λn be the (distinct) eigenvalues of T .
Then there exists a resolution of the identity IV = E1 + · · · + En , where for each k the range of the
projection Ek is the eigenspace associated with λk , and furthermore
T = λ1 E1 + · · · + λn En .
Proof. A good proof of this theorem can be found in [17] on page 212. 
1.1.22.
 Example.
 Let T be the operator on (the real vector space) R2 whose matrix representation
−7 8
is .
−16 17
(a) The characteristic polynomial for T is cT (λ) = λ2 − 10λ + 9.
(b) The eigenspace M1 associated with the eigenvalue 1 is span{(1, 1)}.
(c) The eigenspace M2 associated with the eigenvalue 9 is span{(1, 2)}.
6 1. LINEAR ALGEBRA AND THE SPECTRAL THEOREM

(d) We can write T as a linear combination of projection operators. In particular,

   
2 −1 −1 1
T = 1 · E1 + 9 · E2 where [E1 ] = and [E2 ] = .
2 −1 −2 2
(e) Notice that the sum of [E1 ] and [E2 ] is the identity matrix and that their product is the
zero matrix.    
1 1 1 0
(f) The matrix S = diagonalizes [T ]. That is, S −1 [T ] S = .
1 2 0 9
√
 
−1 2
(g) A matrix representing T is .
−4 5

1.2. Normal Operators on an Inner Product Space

1.2.1. Definition. Let V be a vector space. A function which associates to each pair of vectors
x and y in V a complex number hx, yi is an inner product (or a dot product) on V provided
that the following four conditions are satisfied:
(a) If x, y, z ∈ V , then
hx + y, zi = hx, zi + hy, zi.
(b) If x, y ∈ V , then
hαx, yi = αhx, yi.
(c) If x, y ∈ V , then
hx, yi = hy, xi.
(d) For every nonzero x in V we have hx, xi > 0.
Conditions (a) and (b) show that an inner product is linear in its first variable. Conditions (a) and
(b) of proposition 1.2.3 say that an inner product is conjugate linear in its second variable.
When a mapping is linear in one variable and conjugate linear in the other, it is often called
sesquilinear (the prefix “sesqui-” means “one and a half”). Taken together conditions (a)–(d)
say that the inner product is a positive definite conjugate symmetric sesquilinear form.
1.2.2. Notation. If V is a vector space which has been equipped with an inner product and x ∈ V
we introduce the abbreviation p
kxk := hx, xi
which is read the norm of x or the length of x. (This somewhat optimistic terminology is justified
in proposition 1.2.11 below.)
1.2.3. Proposition. If x, y, and z are vectors in an inner product space and α ∈ C, then
(a) hx, y + zi = hx, yi + hx, zi,
(b) hx, αyi = αhx, yi, and
(c) hx, xi = 0 if and only if x = 0.
1.2.4. Example. For vectors x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) belonging to Cn define
n
X
hx, yi = xk yk .
k=1
Then Cn is an inner product space.
1.2.5. Example. Let l2 be the set of all square
P∞summable sequences of complex numbers. (A
∞ 2
sequence x = (xk )k=1 is square summable if k=1 |xk | < ∞.) For vectors x = (x1 , x2 , . . . ) and
y = (y1 , y2 , . . . ) belonging to l2 define
∞
X
hx, yi = xk yk .
k=1
 1.2. NORMAL OPERATORS ON AN INNER PRODUCT SPACE 7

Then l2 is an inner product space. (It must be shown, among other things, that the series in the
preceding definition actually converges.)
1.2.6. Example. For a < b let C([a, b], C) be the family of all continuous complex valued functions
on the interval [a, b]. For every f , g ∈ C([a, b]) define
Z b
hf, gi = f (x)g(x) dx.
a
Then C([a, b]) is an inner product space.
1.2.7. Theorem. In every inner product space the Schwarz inequality
|hx, yi| ≤ kxk kyk.
holds for all vectors x and y.
1.2.8. Proposition. If (xn ) is a sequence in an inner product space V which converges to a vector
a ∈ V , then hxn , yi → ha, yi for every y ∈ V .
1.2.9. Definition. Let V be a vector space. A function k k : V → R : x 7→ kxk is a norm on V if
(i) kx + yk ≤ kxk + kyk for all x, y ∈ V ;
(ii) kαxk = |α| kxk for all x ∈ V and α ∈ R; and
(iii) if kxk = 0, then x = 0.
The expression kxk may be read as “the norm of x” or “the length of x”. If the function k k
satisfies (i) and (ii) above (but perhaps not (iii)) it is a seminorm on V .
A vector space on which a norm has been defined is a normed linear space (or normed
vector space). A vector in a normed linear space which has norm 1 is a unit vector.
1.2.10. Proposition. If k k is norm (or a seminorm) on a vector space V , then kxk ≥ 0 for
every x ∈ V and k0k = 0.
Every inner product space is a normed linear space.
1.2.11. Proposition. Let V be an inner product space. The map x 7→ kxk defined on V in 1.2.2
is a norm on V .
Every normed linear space is a metric space. More precisely, a norm on a vector space induces
a metric d, which is defined by d(x, y) = kx − yk. That is, the distance between two vectors is the
length of their difference.
? _

x x−y

/
y

If no other metric is specified we always regard a normed linear space as a metric space under
this induced metric. Thus every metric (and hence every topological) concept makes sense in a
(semi)normed linear space.
1.2.12. Proposition. Let V be a normed linear space. Define d : V × V → R by d(x, y) = kx − yk.
Then d is a metric on V . If V is only a seminormed space, then d is a pseudometric.
When there is a topology on a vector space, in particular in normed linear spaces, we reserve
the word “operator” for those linear mappings from the space into itself which are continuous. We
are usually not made aware of this conflicting terminology in elementary linear algebra because that
subject focuses primarily on finite dimensional vector and inner product spaces where the question
is moot: on finite dimensional normed linear spaces all linear maps are automatically continuous
(see proposition 1.2.14 below).
8 1. LINEAR ALGEBRA AND THE SPECTRAL THEOREM

1.2.13. Definition. An operator on a normed linear space V is a continuous linear map from
V into itself.
1.2.14. Proposition. If V and W are normed linear spaces and V is finite dimensional, then
every linear map T : V → W is continuous.
Proof. See [5], proposition III.3.4.
1.2.15. Definition. Vectors x and y in an inner product space V are orthogonal (or perpen-
dicular) if hx, yi = 0. In this case we write x ⊥ y. Subsets A and B of V are orthogonal if
a ⊥ b for every a ∈ A and b ∈ B. In this case we write A ⊥ B.
1.2.16. Definition. If M and N are subspaces of an inner product space V we use the notation
V = M ⊕ N to indicate not only that V is the (vector space) direct sum of M and N but also that
M and N are orthogonal. Thus we say that V is the (internal) orthogonal direct sum of M
and N .
1.2.17. Proposition. Let a be a vector in an inner product space V . Then a ⊥ x for every x ∈ V
if and only if a = 0.
1.2.18. Proposition (The Pythagorean theorem). If x ⊥ y in an inner product space, then
kx + yk2 = kxk2 + kyk2 .
1.2.19. Definition. Let V and W be inner product spaces. For (v, w) and (v 0 , w0 ) in V × W and
α ∈ C define
(v, w) + (v 0 , w0 ) = (v + v 0 , w + w0 )
and
α(v, w) = (αv, αw) .
This results in a vector space, which is the (external) direct sum of V and W . To make it into an
inner product space define
h(v, w), (v 0 , w0 )i = hv, v 0 i + hw, w0 i.
This makes the direct sum of V and W into an inner product space. It is the (external orthog-
onal) direct sum of V and W and is denoted by V ⊕ W .
Notice that the same notation ⊕ is used for both internal and external direct sums and for both
vector space direct sums (see definition 1.1.8) and orthogonal direct sums. So when we see the
symbol V ⊕ W it is important to know which category we are in: vector spaces or inner product
spaces, especially as it is common practice to omit the word “orthogonal” as a modifier to “direct
sum” even in cases when it is intended.
1.2.20. Example. In R2 let M be the x-axis and L be the line whose equation is y = x. If we
think of R2 as a (real) vector space, then it is correct to write R2 = M ⊕ L. If, on the other
hand, we regard R2 as a (real) inner product space, then R2 6= M ⊕ L (because M and L are not
perpendicular).
1.2.21. Proposition. Let V be an inner product space. The inner product on V , regarded as a
map from V ⊕ V into C, is continuous. So is the norm, regarded as a map from V into R.
Concerning the proof of the preceding proposition, notice that the maps (v, v 0 ) 7→ kvk + kv 0 k,
(v, v ) 7→ kvk2 + kv 0 k2 , and (v, v 0 ) 7→ max{kvk, kv 0 k} are all norms on V ⊕ V . Which one is
0
p

induced by the inner product on V ⊕ V ? Why does it not matter which one we use in proving that
the inner product is continuous?
1.2.22. Proposition (The parallelogram law). If x and y are vectors in an inner product space,
then
kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 .
 1.2. NORMAL OPERATORS ON AN INNER PRODUCT SPACE 9

1.2.23. Example. Consider the space C([0, 1]) of continuous complex valued functions defined on
[0, 1]. Under the uniform norm
kf ku := sup{|f (x)| : 0 ≤ x ≤ 1}
C([0, 1]) is a normed linear space. There is no inner product on C([0, 1]) which induces this norm.
Hint for proof . Use the preceding proposition.
1.2.24. Proposition (The polarization identity). If x and y are vectors in an inner product space,
then
hx, yi = 14 (kx + yk2 − kx − yk2 + i kx + iyk2 − i kx − iyk2 ) .
1.2.25. Notation. Let V be an inner product space, x ∈ V , and A, B ⊆ V . If x ⊥ a for every
a ∈ A, we write x ⊥ A; and if a ⊥ b for every a ∈ A and b ∈ B, we write A ⊥ B. We define A⊥ ,

⊥
the orthogonal complement of A, to be {x ∈ V : x ⊥ A}. We write A⊥⊥ for A⊥ .
1.2.26. Proposition. If A is a subset of an inner product space V , then A⊥ is a closed linear
subspace of V .
1.2.27. Theorem (Gram-Schmidt Orthogonalization). If {v 1 , . . . , v n } is a linearly independent
subset of an inner product space V , then there exists an orthogonal set {e1 , . . . , en } of vectors such
that span{v 1 , . . . , v n } = span{e1 , . . . , en }.
1.2.28. Corollary. If M is a subspace of a finite dimensional inner product space V then V =
M ⊕ M ⊥.
For a counterexample showing that the preceding result need not hold in an infinite dimensional
space, see example 2.1.6.
1.2.29. Definition. A linear functional on a vector space V is a linear map from V into its
scalar field. The set of all linear functionals on V is the (algebraic) dual space of V . We will
use the notation V # (rather than L(V, C)) for the algebraic dual space.
1.2.30. Theorem (Riesz-Fréchet Theorem). If f ∈ V # where V is a finite dimensional inner
product space, then there exists a unique vector a in V such that
f (x) = hx, ai
for all x in V .
We will prove shortly that every continuous linear functional on an arbitrary inner product
space has the above representation. The finite dimensional version stated here is a special case,
since every linear map on a finite dimensional inner product space is continuous.
1.2.31. Definition. Let T : V → W be a linear transformation between complex inner product
spaces. If there exists a function T ∗ : W → V which satisfies
hT v, wi = hv, T ∗ wi
for all v ∈ V and w ∈ W , then T ∗ is the adjoint (or conjugate transpose, or Hermitian
conjugate) of T .
1.2.32. Proposition. If T : V → W is a linear map between finite dimensional inner product
spaces, then T ∗ exists.
Hint for proof . The functional φ : V × W → C : (v, w) 7→ hT v, wi is sesquilinear. Fix w ∈ W
and define φw : V → C : v 7→ φ(v, w). Then φw ∈ V # . Use the Riesz-Fréchet theorem (1.2.30).
1.2.33. Proposition. If T : V → W is a linear map between finite dimensional inner product
spaces, then the function T ∗ defined above is linear and T ∗∗ = T .
10 1. LINEAR ALGEBRA AND THE SPECTRAL THEOREM

1.2.34. Theorem (The fundamental theorem of linear algebra). If T : V → W is a linear map

between finite dimensional inner product spaces, then
ker T ∗ = (ran T )⊥ and ran T ∗ = (ker T )⊥ .
1.2.35. Definition. An operator U on an inner product space is unitary if U U ∗ = U ∗ U = I,
that is if U ∗ = U −1 .
1.2.36. Definition. Two operators R and T on an inner product space V are unitarily equiv-
alent if there exists a unitary operator U on V such that R = U ∗ T U .
1.2.37. Proposition. If V is an inner product space, then unitary equivalence is in fact an equiv-
alence relation on L(V ).
1.2.38. Definition. An operator T on a finite dimensional inner product space V is unitarily
diagonalizable if there exists an orthonormal basis for V with respect to which T is diagonal.
Equivalently, an operator on a finite dimensional inner product space with basis is diagonalizable
if it is unitarily equivalent to a diagonal operator.
1.2.39. Definition. An operator T on an inner product space is self-adjoint (or Hermitian)
if T ∗ = T .
1.2.40. Definition. A projection P in an inner product space is an orthogonal projection if
it is self-adjoint. If M is the range of an orthogonal projection we will adopt the notation PM for
the projection rather than the more cumbersome E ⊥ .
M M
CAUTION. A projection on a vector space or a normed linear space is linear and idempotent,
while an orthogonal projection on an inner product space is linear, idempotent, and self-adjoint.
This otherwise straightforward situation is somewhat complicated by a common tendency to refer to
orthogonal projections simply as “projections”. In fact, later in these notes we will adopt this very
convention. In inner product spaces ⊕ usually indicates orthogonal direct sum and “projection”
usually means “orthogonal projection”. In many elementary linear algebra texts, where everything
happens in Rn , it can be quite exasperating trying to divine whether on any particular page the
author is treating Rn as a vector space or as an inner product space.
1.2.41. Proposition. If P is an orthogonal projection on an inner product space V , then we have
the orthogonal direct sum decomposition V = ker P ⊕ ran P .
1.2.42. Definition. If IV = P1 + P2 + · · · + Pn is a resolution of the identity in an inner product
space V and each Pk is an orthogonal projection, then we say that I = P1 + P2 + · · · + Pn is an
orthogonal resolution of the identity.
1.2.43. Proposition. If IV = P1L+ P2 + · · · + Pn is an orthogonal resolution of the identity on an
inner product space V , then V = nk=1 ran Pk .
1.2.44. Definition. An operator N on an inner product space is normal if N N ∗ = N ∗ N .
Two great triumphs of linear algebra are the spectral theorem for operators on a (complex) finite
dimensional inner product space (see 1.2.45), which gives a simply stated necessary and sufficient
condition for an operator to be unitarily diagonalizable, and theorem 1.2.46, which gives a complete
classification of those operators.
1.2.45. Theorem (Spectral Theorem for Complex Inner Product Spaces). Let T be an operator
on a finite dimensional inner product space V with (distinct) eigenvalues λ1 , . . . , λn . Then T is
unitarily diagonalizable if and only if it is normal. If T is normal, then there exists an orthogonal
resolution of the identity IV = P1 + · · · + Pn , where for each k the range of the orthogonal projection
Pk is the eigenspace associated with λk , and furthermore
T = λ1 P1 + · · · + λn Pn .
 1.2. NORMAL OPERATORS ON AN INNER PRODUCT SPACE 11

Proof. See [26], page 227. 

1.2.46. Theorem. Two normal operators on a finite dimensional inner product space are unitarily
equivalent if and only if they have the same eigenvalues each with the same multiplicity; that is, if
and only if they have the same characteristic polynomial.
Proof. See [17], page 357. 
Much of the remainder of this course is about the long and difficult adventure of finding ap-
propriate generalizations of the preceding two results to the infinite dimensional setting and the
astonishing landscapes which came into view along the way.
 CHAPTER 2

THE ALGEBRA OF HILBERT SPACE OPERATORS

2.1. Hilbert Space Geometry

Inner products induce norms (see proposition 1.2.11) and norms induce metrics (see 1.2.12), so
the notion of completeness makes sense in both cases. The next definition tells us that a Banach
space is a complete normed linear space and a Hilbert space is a complete inner product space.
2.1.1. Definition. If a normed linear space is complete with respect to the metric induced by its
norm, it is a Banach space. If an inner product space is complete with respect to the metric
induced by the norm induced by the inner product on V , it is a Hilbert space.
2.1.2. Example. The space Cn of n-tuples of complex numbers is a Hilbert space under the inner
product defined in example 1.2.4.
2.1.3. Example. The space l2 of all square summable sequences of complex numbers is a Hilbert
space under the inner product defined in example 1.2.5.
2.1.4. Definition. Let (xn ) be a sequence of elements of a set S and P be some property that
members of S may possess. We say that the sequence (xn ) eventually has property P if there
exists n0 ∈ N such that xn has property P for every n ≥ n0 . (Another way to say the same thing:
xn has property P for all but finitely many n.)
2.1.5. Example. We denote by lc , the vector space of all sequences (an ) of complex numbers which
are eventually zero. (The vector space operations are defined pointwise.) We make the space lc
into an inner product space by defining ha, bi = h (an ), (bn ) i := ∞
P
k=1 n n The resulting space is
a b .
not a Hilbert space.
2.1.6. Example. The vector subspace lc of the Hilbert space l2 shows that corollary 1.2.28 need
not hold for infinite dimensional spaces.
2.1.7. Example. Let l1 be the P family of all sequences (a1 , a2 , a3 , . . . ) which are absolutely
summable; that is, such that ∞ k=1 |ak | < ∞. This is a Banach space under pointwise operations
of addition and scalar multiplication and norm given by
∞
X
kak = |ak | .
k=1
The norm does not arise from an inner product, so it is not a Hilbert space.
2.1.8. Example. Let µ be a measure on a σ-algebra A of subsets of a set S. A complex valued
function f on S is measurable if the inverse image under f of every Borel set (equivalently, every
open set) in C belongs to A. We define an equivalence relation ∼ on the family of measurable
complex valued functions by setting f ∼ g whenever f and g differ on a set of measure zero,
that is, whenever µ({x ∈ S : f (x) 6= g(x)}) = 0. We adopt conventional notation and denote the
equivalence class containing f by f itself (rather than something more logical such as [f ]). We
denote the family of (equivalence classes) of measurable complex valued functions on S by M(S).
A function f ∈ M(S) is square integrable if S |f (x)|2 dµ(x) < ∞. We denote the family of
R
(equivalence classes of) square integrable functions on S by L2 (S). For every f , g ∈ L2 (S) define
Z
hf, gi = f (x)g(x) dµ(x) .
S
13
14 2. THE ALGEBRA OF HILBERT SPACE OPERATORS

With this inner product (and the obvious pointwise vector space operations) L2 (S) is a Hilbert
space.
Here is a standard example of a space which is a Banach space but not a Hilbert space.
2.1.9. Example. As above letR µ be a measure on a σ-algebra A of subsets of a set S. A function
f ∈ M(S) is integrable if S |f (x)| dµ(x) < ∞. We denote the family of (equivalence classes
of) integrable functions on S by L1 (S).
R An attempt to define an inner product on L1 (S) as
R on L2 (S) by setting hf, gi = S f (x)g(x) dµ(x) fails. (Why?) Nevertheless, the function
we did
f 7→ S |f (x)| dµ(x) is a norm on L1 (S). It is denoted by k · k1 . With respect to this norm (and the
obvious pointwise vector space operations), L1 (S) is a Banach space.
2.1.10. Example. If X is a compact Hausdorff space the family C(X) of continuous complex
valued functions on X is a Banach space under the uniform norm, which is defined by
kf ku := sup{|f (x)| : x ∈ X}.
We have seen in example 1.2.23 that this norm does not arise from an inner product.
2.1.11. Example. If X is a locally compact Hausdorff space the uniform norm may not be defined
on the family C(X) of continuous complex valued functions on X. (Why not?) However, it is
defined on the family Cb (X) of bounded continuous complex valued functions on X and on C0 (X)
the family of continuous complex valued functions on X that vanish at infinity (A complex valued
function on X is said to vanish at infinity if for every  > 0 there exists a compact subset K
of X such that |f (x)| <  whenever x ∈
/ K.) Both Cb (X) and C0 (X) are Banach spaces under the
uniform norm (and the obvious pointwise vector space operations).
2.1.12. Example. Let H be the set of all absolutely continuous functions on [0, 1] such that f 0
belongs to L2 ([0, 1]) and f (0) = 0. For f and g in H define
Z 1
hf, gi = f 0 (t)g 0 (t) dt.
0
This is an inner product on H under which H becomes a Hilbert space.
2.1.13. Convention. In the context of Hilbert (and, more generally, Banach) spaces the word
“subspace” will always mean closed vector subspace. To indicate that M is a subspace of H we
write M 4 H. A (not necessarily closed) vector subspace of a Hilbert space is often called by other
names such as linear subspace or linear manifold.
2.1.14. Definition. Let A beWa nonempty subset of a Banach space B. We define the closed
linear span of A (denoted by A) to be the intersection of the family of all subspaces of B which
contain A. This is frequently referred to as the smallest subspace of B containing A.
W
2.1.15. Proposition. The preceding definition makes sense. It is equivalent to defining A to be
the closure of the (linear) span of A.
2.1.16. Definition. Let V be a vector space and a, b ∈ V . Then the segment between a and b,
denoted by Seg[a, b], is defined to be
{(1 − t)a + tb : 0 ≤ t ≤ 1}.
A subset C of V is convex if Seg[a, b] ⊆ C whenever a, b ∈ C.
2.1.17. Proposition. The intersection of a family of convex subsets of a vector space is convex.
2.1.18. Proposition. If T : V → W is a linear map between vector spaces and C is a convex
subset of V , then T → (C) is a convex subset of W .
2.1.19. Definition. Let a be a element of a normed linear space V and r > 0. The open
ball of radius r about a is {x ∈ V : kx − ak < r}. The closed ball of radius r about a is
{x ∈ V : kx − ak ≤ r}. And the sphere of radius r about a is {x ∈ V : kx − ak = r}.
 2.1. HILBERT SPACE GEOMETRY 15

2.1.20. Proposition. Every open ball (and every closed ball) in a normed linear space is convex.
2.1.21. Theorem (Minimizing Vector Theorem). If C is a nonempty closed convex subset of a
Hilbert space H and a ∈ C c , then there exists a unique b ∈ C such that kb − ak ≤ kx − ak for every
x ∈ C.

3x

a /b

2.1.22. Example. The vector space R2 under the uniform metric is a Banach space. To see that
in this space the minimizing vector theorem does not hold take C to be the closed unit ball about
the origin and a to be the point (2, 0).
R 1/2 R1
2.1.23. Example. The sets C1 = {f ∈ C([0, 1], R) : 0 f − 1/2 f = 1} and C2 = {f ∈
R1
L1 ([0, 1], R) : 0 f = 1} are examples that show that neither the existence nor the uniqueness
claims of the minimizing vector theorem necessarily holds in a Banach space.
2.1.24. Theorem (Vector decomposition theorem). Let H be a Hilbert space, M be a subspace
of H, and x ∈ H. Then there exist unique vectors y ∈ M and z ∈ M ⊥ such that x = y + z.
2.1.25. Proposition. Let H be a Hilbert space. Then the following hold:
(a) if M ⊆ H, then M ⊆ M ⊥⊥ ;
(b) if M ⊆ N ⊆ H, then N ⊥ ⊆ M ⊥ ;
(c) if M is a subspace
W of H, then M = M ⊥⊥ ; and
(d) if M ⊆ H, then M = M . ⊥⊥

2.1.26. Proposition. If M is a subspace of a Hilbert space H, then H = M ⊕M ⊥ . (This conclusion

need not follow if M is assumed only to be a linear subspace of H.)
2.1.27. Example. The preceding result says that every subspace of a Hilbert space is a direct
summand. This is not true if M is assumed to be just a linear subspace of the space. For example,
notice that M = l (see example 2.1.5) is a linear subspace of the Hilbert space l2 (see example 2.1.3)
but M ⊥ = {0}.
2.1.28. Proposition. Let M and N be (closed linear) subspaces of a Hilbert space. Then
(a) (M + N )⊥ = (M ∪ N )⊥ = M ⊥ ∩ N ⊥ , and
(b) (M ∩ N )⊥ = M ⊥ + N ⊥ .
2.1.29. Definition. Let V and W be (complex) vector spaces. A function φ : V × W → C is a
sesquilinear functional if it satisfies
(i) φ(u + v, w) = φ(u, w) + φ(v, w),
(ii) φ(αv, w) = αφ(v, w),
(iii) φ(v, w + x) = φ(v, w) + φ(v, x), and
(iv) φ(v, αw) = α φ(v, w)
for all u,v ∈ V , all w, x ∈ W and all α ∈ C. When W = V there are two more terms we will need.
A sesquilinear functional φ : V × V → C is (conjugate) symmetric if
16 2. THE ALGEBRA OF HILBERT SPACE OPERATORS

(v) φ(u, v) = φ(v, u) for all u, v ∈ V .

A sesquilinear functional φ : V × V → C is positive semidefinite if
(vi) φ(v, v) ≥ 0 for all v ∈ V .
If φ : V × V → C is sesquilinear, then the function φ̂ defined on V by φ̂(v) = φ(v, v) is the
quadratic form associated with φ.
2.1.30. Proposition. Let φ be a sesquilinear functional on a vector space V and φ̂ be its associated
quadratic form. Then
(a) 4φ(u, v) = φ̂(u + v) − φ̂(u − v) + iφ̂(u + iv) − iφ̂(u − iv) for all u, v ∈ V ;
(b) φ is uniquely determined by φ̂; and
(c) φ is symmetric if and only if φ̂ is real valued.
Notice that part (a) of the preceding proposition is just a slight generalization of the polarization
identity (proposition 1.2.24).
2.1.31. Definition. A positive symmetric sesquilinear form is a semi-inner product.
2.1.32. Proposition. The Schwarz inequality (see 1.2.7) holds for semi-inner products. That is,
if x and y are elements of a semi-inner product space, then
|hx, yi|2 ≤ hx, xihy, yi .
2.1.33. Proposition. Let V be a semi-inner product space and z ∈ V . Then hz, zi = 0 if and only
if hz, yi = 0 for all y ∈ V .
2.1.34. Proposition. If V is a semi-inner product space, then the set L := {z ∈ V : hz, zi = 0} is
a vector subspace of V and the quotient vector space V /L can be made into an inner product space
by defining
h [x], [y] i := hx, yi
for all [x], [y] ∈ V /L.

2.2. Operators on Hilbert Spaces

2.2.1. Definition. A linear transformation T : V → W between normed linear spaces is bounded
if T (B) is a bounded subset of W whenever B is a bounded subset of V . In other words, a bounded
linear map takes bounded sets to bounded sets. We denote by B(V, W ) the family of all bounded
linear transformations from V into W . A bounded linear map from a space V into itself is an
operator and we denote the family of all operators on V by B(V ).
2.2.2. Proposition. If T : V → W is a linear transformation between normed linear spaces, then
the following are equivalent:
(i) T is continuous at 0.
(ii) T is continuous on V .
(iii) T is uniformly continuous on V .
(iv) The image of the closed unit ball under T is bounded.
(v) T is bounded.
(vi) There exists a number M > 0 such that kT xk ≤ M kxk for all x ∈ V .
2.2.3. Proposition. Let V and W be normed linear spaces. If S, T ∈ B(V, W ) and α ∈ C, then
S + T and αT belong to B(V, W ).
2.2.4. Proposition. Let T : V → W be a bounded linear transformation between normed linear
spaces. Then the following four numbers (exist and) are equal.
(i) sup{kT xk : kxk ≤ 1}
(ii) sup{kT xk : kxk = 1}
 2.2. OPERATORS ON HILBERT SPACES 17

(iii) sup{kT xk kxk−1 : x 6= 0}

(iv) inf{M > 0 : kT xk ≤ M kxk for all x ∈ V }
2.2.5. Definition. If T is a bounded linear map between normed linear spaces, then kT k, called
the norm of T , is defined to be any one of the four expressions in the previous proposition.
2.2.6. Proposition. Let V and W be normed linear spaces. The function
k k : B(V, W ) → R : T 7→ kT k
is a norm. Under this norm B(V, W ) is a Banach space whenever W is.
2.2.7. Proposition. Let U , V , and W be normed linear spaces. If S ∈ B(U, V ) and T ∈ B(V, W ),
then T S ∈ B(U, W ) and kT Sk ≤ kT kkSk.
2.2.8. Example. On any normed linear space V the identity operator
idV = IV = I : V → V : v 7→ v
is bounded and kIV k = 1. The zero operator
0V = 0 : V → V : v 7→ 0
is also bounded and k0V k = 0.
2.2.9. Example. Let T : R2 → R3 : (x, y) 7→ (3x, x + 2y, x − 2y). Then T is bounded and kT k =
√
11.
2.2.10. Example. Let X be a compact Hausdorff space and φ be a continuous real valued function
on X. Define
Mφ : C(X) → C(X) : f 7→ φf .
Then Mφ is an operator on the Banach space C(X) (see example 2.1.10) and kMφ k = kφku . The
operator Mφ is called a multiplication operator.)
2.2.11. Example. Under the uniform norm the family D([0, 1], R) of all continuously differentiable
real valued functions on [0, 1] is a real normed linear space. The differentiation map
D : D([0, 1], R) → C([0, 1], R) : f 7→ f 0
(although linear) is not bounded.
2.2.12. Notation. If V is a normed linear space we denote by V ∗ the family of all continuous
linear functionals on V . We call this the dual space of V .
If V is finite dimensional, then V ∗ = V # (see proposition 1.2.14). By proposition 2.2.6 V ∗ is
always a Banach space (whether or not V itself is complete).
2.2.13. Convention. When we say dual space we always mean V ∗ ; when referring to V # we will
say algebraic dual space.
2.2.14. Example. Let X be a compact topological space. For a ∈ X define
Ea : C(X) → C : f 7→ f (a) .

∗
Then Ea ∈ C ∗ (X) (where C ∗ (X) is a shortened notation for C(X) ) and kEa k = 1.
2.2.15. Example. Recall from example 2.1.3 that the family l2 of all square summable sequences
of complex numbers is a Hilbert space. Let
S : l2 → l2 : (x1 , x2 , x3 , . . . ) 7→ (0, x1 , x2 , . . . ) .
Then S is an operator on l2 ,called the unilateral shift operator, and kSk = 1.
18 2. THE ALGEBRA OF HILBERT SPACE OPERATORS

2.2.16. Example. Let a = (a1 , a2 , a3 , . . . ) be a bounded sequence of complex numbers. Define a

mapping
Da : l2 → l2 : (x1 , x2 , x3 , . . . ) 7→ (a1 x1 , a2 x2 , a3 x3 , . . . ) .
Then Da is an operator on l2 whose norm is kaku . This is a diagonal operator.
2.2.17. Example. Let (S, A, µ) be a sigma-finite measure space and L2 (S) be the Hilbert space of
all (equivalence classes of) complex valued functions on S which are square integrable with respect
to µ (see example 2.1.8). Let k : S × S → C be square integrable with respect to the product
measure µ × µ on S × S. Define int k on L2 (S) by
Z
int k(f )(x) := k(x, y)f (y) dµ(y)
S
for every x ∈ S. Then int k is an operator on L2 (S). If K = int k for some k ∈ L2 (µ × µ), then
K is an integral operator and k is its kernel. (This is another use of the word “kernel”;
it has nothing whatever to do with the more common use of the word: ker K = K ← ({0})—see
definition 1.1.2 ).
2.2.18. Example. Let H = L2 ([0, 1]) be the real Hilbert space of all (equivalence classes of) real
valued functions on [0, 1] which are square integrable with respect to Lebesgue measure. Define V
on H by Z x
V f (x) = f (t) dt . (2.1)
0
Then V is an operator on H. This is a Volterra operator and is an example of an integral
operator. (What is its kernel k ?)
There is nothing important about the choice of the space L2 ([0, 1]) in this example. Many
spaces admit Volterra operators. For instance, we could just as well have used equation (2.1) to
define an operator V on the Banach space C([0, 1]).
2.2.19. Proposition. The kernel of a bounded linear map A : H → K between Hilbert spaces is a
closed linear subspace of H.
2.2.20. Proposition. Let A, B ∈ B(H, K) where H and K are Hilbert spaces. Then A = B if
and only if hAx, yi = hBx, yi for all x ∈ H and y ∈ K.
2.2.21. Proposition. If H is a complex Hilbert space and T ∈ B(H) satisfies hT z, zi = 0 for all
z ∈ H, then T = 0. The corresponding result fails in real Hilbert spaces.
Hint for proof . In the hypothesis replace z first by x + y and then by x + iy.
Despite our convention that in these notes all vector spaces are complex, the word “complex”
was added to the hypotheses of the preceding proposition to draw attention to the fact that it is
one of the few facts which holds only for complex spaces. While proposition 2.2.20 holds for both
real and complex Hilbert spaces 2.2.21 does not. (Consider the operator which rotates the plane
R2 by 90 degrees.)
2.2.22. Example. Let H be a Hilbert space and a ∈ H. Define ψa : H → C : x 7→ hx, ai. Then
ψa ∈ H ∗ .
Now we generalize theorem 1.2.30 to the infinite dimensional setting. This result is sometimes
called the Riesz representation theorem (which invites confusion with the more substantial results
about representing certain linear functionals as measures) or the little Riesz representation theorem.
2.2.23. Theorem (Riesz-Fréchet Theorem). If f ∈ H ∗ where H is a Hilbert space, then there
exists a unique vector a in H such that f = ψa . Furthermore, kak = kψa k.
Proof. See [5], I.3.4.
 2.3. ALGEBRAS 19

2.3. Algebras
2.3.1. Definition. A (complex) algebra is a (complex) vector space A together with a binary
operation (x, y) 7→ xy, called multiplication, which satisfy
(i) (ab)c = a(bc),
(ii) (a + b)c = ac + bc,
(iii) a(b + c) = ab + ac, and
(iv) α(ab) = (αa)b = a(αb)
for all a, b, c ∈ A and α ∈ C. In other words, an algebra is a vector space which is also a ring and
satisfies (iv). If an algebra A has a multiplicative identity (or unit), that is, a nonzero element 1
(or 1A ) which satisfies
(v) 1 a = a 1 = a
for every a ∈ A, then the algebra is unital. An algebra A for which ab = ba for all a, b ∈ A is a
commutative algebra.
A map φ : A → B between algebras is an algebra homomorphism if it is linear and multi-
plicative (meaning that φ(aa0 ) = φ(a)φ(a0 ) for all a, a0 ∈ A). If the algebras A and B are unital a
homomorphism φ : A → B is unital if φ(1A ) = 1B .
A subset B of an algebra A is a subalgebra of A if it is an algebra under the operations it
inherits from A. A subalgebra B of a unital algebra A is a unital subalgebra if it contains the
multiplicative identity of A. CAUTION. To be a unital subalgebra it is not enough for B to have
a multiplicative identity of its own; it must contain the identity of A. Thus, an algebra can be both
unital and a subalgebra of A without being a unital subalgebra of A. (An example is given later
in 3.1.4.)

2.3.2. Convention. As with vector spaces, all algebras in the sequel will be assumed to be complex
algebras unless the contrary is explicitly stated.
2.3.3. Proposition. An algebra can have at most one multiplicative identity.
2.3.4. Definition. An element a of a unital algebra A is left invertible if there exists an element
al in A (called a left inverse of a) such that al a = 1 and is right invertible if there exists
an element ar in A (called a right inverse of a)such that aar = 1. The element is invertible
if it is both left invertible and right invertible. The set of all invertible elements of A is denoted
by inv A.
An element of a unital algebra can have at most one multiplicative inverse. In fact, more is
true.
2.3.5. Proposition. If an element of a unital algebra has both a left inverse and a right inverse,
then these two inverses are equal (and so the element is invertible).
When an element a of a unital algebra is invertible its (unique) inverse is denoted by a−1 .
2.3.6. Proposition. If a is an invertible element of a unital algebra, then a−1 is also invertible
and
−1
a−1 = a.
2.3.7. Proposition. If a and b are invertible elements of a unital algebra, then their product ab is
also invertible and
(ab)−1 = b−1 a−1 .
2.3.8. Proposition. If a and b are invertible elements of a unital algebra, then
a−1 − b−1 = a−1 (b − a)b−1 .
2.3.9. Proposition. Let a and b be elements of a unital algebra. If both ab and ba are invertible,
then so are a and b.
20 2. THE ALGEBRA OF HILBERT SPACE OPERATORS

Hint for proof . Use proposition 2.3.5.

2.3.10. Proposition. Let a and b be elements of a unital algebra. Then 1 − ab is invertible if and
only if 1 − ba is.
Proof. If 1 − ab is invertible, then 1 + b(1 − ab)−1 a is the inverse of 1 − ba. 
2.3.11. Example. The simplest example of a (complex) algebra is the family C of complex num-
bers. It is both unital and commutative.
2.3.12. Example. If X is a nonempty topological space, then the family C(X) of all continuous
complex valued functions on X is an algebra under the usual pointwise operations of addition,
scalar multiplication, and multiplication. It is both unital and commutative.
2.3.13. Example. If X is a locally compact Hausdorff space which is not compact, then (under
pointwise operations) the family C0 (X) of all continuous complex valued functions on X which
vanish at infinity (see example 2.1.11) is a commutative algebra. However, it is not a unital
algebra.
2.3.14. Example. The family Mn of n × n matrices of complex numbers is a unital algebra
under the usual matrix operations of addition, scalar multiplication, and multiplication. It is not
commutative when n > 1.
We will be making considerable use of a generalization of the preceding example.
2.3.15. Example. Let A be an algebra. Make the family Mn (A) of n × n matrices of elements of
A into an algebra by using the same rules for matrix operations that are used for Mn . Thus Mn
is just Mn (C). The algebra Mn (A) is unital if and only if A is.
2.3.16. Example. If V is a normed linear space, then B(V ) is a unital algebra. If dim V > 1,
the algebra is not commutative. (See propositions 2.2.3 and 2.2.7.)
2.3.17. Definition. An left ideal in an algebra A is a vector subspace J of A such that AJ ⊆ J.
(For right ideals, of course, we require JA ⊆ J.) We say that J is an ideal if it is a two-sided
ideal, that is, both a left and a right ideal. A proper ideal is an ideal which is a proper subset
of A.
The ideals {0} and A are often referred to as the trivial ideals of A. The algebra A is simple
if it has no nontrivial ideals.
A maximal ideal is a proper ideal that is properly contained in no other proper ideal. We
denote the family of all maximal ideals in an algebra A by Max A. A minimal ideal is a nonzero
ideal that properly contains no other nonzero ideal.

2.3.18. Convention. Whenever we refer to an ideal in an algebra we understand it to be a two-

sided ideal (unless the contrary is stated).
2.3.19. Proposition. No invertible element in a unital algebra can belong to a proper ideal.
2.3.20. Proposition. Every proper ideal in a unital algebra A is contained in a maximal ideal.
Thus, in particular, Max A is nonempty whenever A is a unital algebra.
Hint for proof . Zorn’s lemma.
2.3.21. Proposition. Let a be an element of a commutative algebra A. If A is unital and a is not
invertible, then aA is a proper ideal in A.
2.3.22. Definition. The ideal aA in the preceding proposition is the principal ideal generated
by a.
 2.4. SPECTRUM 21

2.3.23. Definition. Let J be a proper ideal in an algebra A. Define an equivalence relation ∼ on

A by
a∼b if and only if b − a ∈ J.
For each a ∈ A let [a] be the equivalence class containing a. Let A/J be the set of all equivalence
classes of elements of A. For [a] and [b] in A/J define
[a] + [b] := [a + b] and [a][b] := [ab]
and for α ∈ C and [a] ∈ A/J define
α[a] := [αa] .
Under these operations A/J becomes an algebra. It is the quotient algebra of A by J. The
notation A/J is usually read “A mod J”. The surjective algebra homomorphism
π : A → A/J : a 7→ [a]
is called the quotient map.
2.3.24. Proposition. The preceding definition makes sense and the claims made therein are cor-
rect. Furthermore, the quotient algebra A/J is unital if A is.
Hint for proof . You will need to show that:
(a) ∼ is an equivalence relation.
(b) Addition and multiplication of equivalence classes is well defined.
(c) Multiplication of an equivalence class by a scalar is well defined.
(d) A/J is an algebra.
(e) The “quotient map” π really is a surjective algebra homomorphism.
(At what point is it necessary that we factor out an ideal and not just a subalgebra?)

2.4. Spectrum
2.4.1. Definition. Let a be an element of a unital algebra A. The spectrum of a, denoted by
σA (a) or just σ(a), is the set of all complex numbers λ such that a − λ1 is not invertible.
2.4.2. Example. If z is an element of the algebra C of complex numbers, then σ(z) = {z}.
2.4.3. Example. Let X be a compact Hausdorff space. If f is an element of the algebra C(X) of
continuous complex valued functions on X, then the spectrum of f is its range.
2.4.4. Example. Let X be an arbitrary topological space. If f is an element of the algebra Cb (X)
of bounded continuous complex valued functions on X, then the spectrum of f is the closure of its
range.
2.4.5. Example. Let S be a positive measure space and [f ] ∈ L∞ (S) be an (equivalence class of)
essentially bounded function(s) on S. Then the spectrum of [f ] is its essential range.
2.4.6. Example. The family M3 of 3 × 3 matrices of complex
 numbersis a unital algebra under
5 −6 −6
the usual matrix operations. The spectrum of the matrix −1 4 2  is {1, 2}.
3 −6 −4
2.4.7. Proposition. Let a be an element of a unital algebra such that a2 = 1. Then either
(i) a = 1, in which case σ(a) = {1}, or
(ii) a = −1, in which case σ(a) = {−1}, or
(iii) σ(a) = {−1, 1}.
1
Hint for proof . In (iii) to prove σ(a) ⊆ {−1, 1}, consider (a + λ1).
1 − λ2
22 2. THE ALGEBRA OF HILBERT SPACE OPERATORS

2.4.8. Proposition. An element a of an algebra is idempotent if a2 = a. Let a be an idempotent

element of a unital algebra. Then either
(i) a = 1, in which case σ(a) = {1}, or
(ii) a = 0, in which case σ(a) = {0}, or
(iii) σ(a) = {0, 1}.
1

Hint for proof . In (iii) to prove σ(a) ⊆ {0, 1}, consider 2
a + (λ − 1)1 .
λ−λ
2.4.9. Proposition. Let a be an invertible element of a unital algebra. Then λ ∈ σ(a) if and only
if λ1 ∈ σ(a−1 ).
The next proposition is a simple corollary of proposition 2.3.10.
2.4.10. Proposition. If a and b are elements of a unital algebra, then, except possibly for 0, the
spectra of ab and ba are the same.
 CHAPTER 3

BANACH ALGEBRAS

3.1. Definition and Elementary Properties

3.1.1. Definition. If an algebra A is equipped with a norm k · k satisfying
kxyk ≤ kxk kyk
for all x, y ∈ A, then we say that it is a normed algebra and that k · k is an algebra norm.
We also require that if A is unital, then to be a normed algebra it must also satisfy
k1A k = 1
where 1A is the multiplicative identity of A. A complete normed algebra is a Banach algebra.
A map f : A → B between Banach algebras is a Banach algebra homomorphism if it is
both an algebraic homomorphism and a bounded linear map between A and B (regarded as Banach
spaces).
3.1.2. Convention. In the context of Banach algebras, when we speak of a homomorphism, or
just a morphism, we mean a Banach algebra homomorphism.
3.1.3. Proposition. Let A and B be Banach algebras. The direct sum of A and B, denoted by
A ⊕ B, is defined to be the vector space direct sum of A and B on which a multiplication has been
defined by
(a, b) (a0 , b0 ) = (aa0 , bb0 )
(for a, a0 ∈ A and b, b0 ∈ B) and a norm defined by
k(a, b)k = max{kak, kbk} .
This makes A ⊕ B into a Banach algebra.
3.1.4. Example. The algebra C ∼
= C ⊕ {0} is unital and is a subalgebra of C ⊕ C, but it is not a
unital subalgebra of C ⊕ C.
3.1.5. Proposition. In a Banach algebra the operations of addition and multiplication (regarded
as maps from A ⊕ A to A) are continuous and scalar multiplication (regarded as a map from C ⊕ A
to A) is also continuous.
3.1.6. Example. Let X be a compact Hausdorff space. The family C(X) of all continuous complex
valued functions on X under pointwise operations of addition, multiplication, and scalar multipli-
cation is a unital commutative Banach algebra under the uniform norm
kf ku := sup{|f (x)| : x ∈ X} .
(See examples 2.1.10 and 2.3.12.) Similarly if X is locally compact and Hausdorff, then C0 (X) and
Cb (X) are commutative Banach algebras (see example 2.1.11).
3.1.7. Example. If V is a Banach space, then with composition as multiplication B(V ) is a unital
Banach algebra (see example 2.3.16 and proposition 2.2.6).
3.1.8. Example. Let l1 (Z) be the family of all bilateral sequences
(. . . , a−2 , a−1 , a0 , a1 , a2 , . . . )
23
24 3. BANACH ALGEBRAS

which are absolutely summable; that is, such that ∞

P
k=−∞ |ak | < ∞. This is a Banach space
under pointwise operations of addition and scalar multiplication and norm given by
X∞
kak = |ak | .
k=−∞

For a, b ∈ l1 (Z) define a ∗ b to be the sequence whose nth entry is given by

∞
X
(a ∗ b)n = an−k bk .
k=−∞

The operation ∗ P
is called convolution. (To see where the definition comes from try multiplying
the power series ∞ k and
P∞ k .) With this additional operation l (Z) becomes a unital
−∞ ka z b
−∞ k z 1
commutative Banach algebra.
3.1.9. Example. The Banach space L1 (R) (see example 2.1.9) can be made into a commutative
Banach algebra. For f , g ∈ L1 (R) define
Z ∞
h(x) = f (x − y)g(y) dy (1)
−∞
whenever the function y 7→ f (x − y)g(y) belongs to L1 (R). Then h(x) is defined and finite for
almost all x ∈ R. Set h(x) = 0 whenever (1) is undefined. Furthermore h belongs to L1 (R) and
khk1 ≤ kf k1 kgk1 . The function h is usually denoted by f ∗ g; this is the convolution of f and g.
(In this definition does any problem arise from the fact that members of L1 are in fact equivalence
classes of functions?)
3.1.10. Definition. Let (ak ) be a sequence of vectors in a normed linear space V . We say that the
sequence (ak ) isPsummable, or that the series ∞
P
n
a P∞if there exists an element b ∈ V
k=1 k converges
such that kb − k=1 ak k → 0 as n → ∞. In this case we write k=1 ak = b.
3.1.11. Proposition (The Neumann series). Let P a bekan element of a unital Banach algebra A.
If kak < 1, then 1 − a ∈ inv A and (1 − a)−1 = ∞
k=0 a .

Hint for proof . In a unital algebra we take a0 to mean 1A . Start by proving that the sequence
(1, a, a2 , . . . ) is summable by showing that the sequence of partial sums nk=0 ak is Cauchy.
P

◦
3.1.12. Proposition. If A is a unital Banach algebra, then inv A ⊆ A.
Hint for proof . Let a ∈ inv A. Show, for sufficiently small h, that 1 − a−1 h is invertible.
3.1.13. Proposition. If a belongs to a unital Banach algebra and kak < 1, then
kak
(1 − a)−1 − 1 ≤ .
1 − kak
3.1.14. Proposition. Let A be a unital Banach algebra. The map a 7→ a−1 from inv A into itself
is a homeomorphism.
3.1.15. Notation. Let f be a complex valued function on some set S. Denote by Zf the set of all
points x in S such that f (x) = 0. This is the zero set of f .
3.1.16. Proposition. The invertible elements in the Banach algebra C(X) of all continuous com-
plex valued functions on a compact Hausdorff space X are the functions which vanish nowhere.
That is,
inv C(X) = {f ∈ C(X) : Zf = ∅} .
3.1.17. Proposition. Let A be a unital Banach algebra. The map r : a 7→ a−1 from inv A into
itself is differentiable and at each invertible element a, we have dra (h) = −a−1 ha−1 for all h ∈ A.
 3.1. DEFINITION AND ELEMENTARY PROPERTIES 25

3.1.18. Proposition. Let a be an element of a unital Banach algebra A. Then the spectrum of a
is compact and |λ| ≤ kak for every λ ∈ σ(a).
Hint for proof . Use the Heine-Borel theorem. To prove that the spectrum is closed notice that
(σ(a))c = f ← (inv A) where f (λ) = a − λ1 for every complex number λ. Also show that if |λ| > kak,
then 1 − λ−1 a is invertible.
3.1.19. Definition. Let a be an element of a unital Banach algebra. The resolvent mapping
for a is defined by
Ra : C \ σ(a) → A : λ 7→ (a − λ1)−1 .
◦
3.1.20. Definition. Let U ⊆ C and A be a unital Banach algebra. A function f : U → A is
analytic on U if
f (z) − f (a)
f 0 (a) := lim
z→a z−a
exists for every a ∈ U . A complex valued function which is analytic on all of C is an entire
function.
3.1.21. Proposition. For a an element of a unital Banach algebra A and φ a bounded linear
functional on A let f := φ ◦ Ra : C \ σ(a) → C. Then
(i) f is analytic on its domain, and
(ii) f (λ) → 0 as |λ| → ∞.
Hint for proof . For (i) use proposition 2.3.8.
In order to prove our next major result, that the spectrum of an element is never empty (see
theorem 3.1.25), we need two theorems: Liouville’s theorem from complex variables and the Hahn-
Banach theorem from functional analysis.
3.1.22. Theorem (Liouville’s theorem). Every bounded entire function on C is constant.
A proof of this theorem can be found in nearly any text on complex variables.
What is known as the Hahn-Banach theorem is really a family of related theorems that guarantee
the existence of a generous supply of linear functionals. Some authors refer to the version given
below, which says that linear functionals on subspaces can be extended without increasing their
norm, as the Bohnenblust-Sobczyk-Suhomlinov theorem.
3.1.23. Theorem (Hahn-Banach theorem). If M is a linear subspace of a normed linear space V
and f ∈ M ∗ , then there exists an extension fˆ of f to all of V such that kfˆk = kf k.
Proof. See [15], theorem 14.12.
3.1.24. Corollary. Let M be a linear subspace of a normed linear space V . If z is a vector in M c
such that the distance d(z, M ) between z and M is strictly greater than zero, then there exists a
linear functional g ∈ V ∗ such that g → (M ) = {0}, g(z) = d(z, M ), and kgk = 1.
Proof. See [15], corollary 14.13.
3.1.25. Theorem. The spectrum of every element of a unital Banach algebra is nonempty.
Hint for proof . Argue by contradiction. Use Liouville’s theorem to show that φ◦Ra is constant
for every bounded linear functional φ on A. Then use the (corollary to the) Hahn-Banach theorem
to prove that Ra is constant. Why must this constant be 0?
It is important to keep in mind that we are working only with complex  algebras.
 This result
0 1
is false for real Banach algebras. An easy counterexample is the matrix regarded as an
−1 0
element of the (real) Banach algebra of all 2 × 2 matrices of real numbers.
The next result says that essentially the only (complex) Banach division algebra is the field of
complex numbers.
26 3. BANACH ALGEBRAS

3.1.26. Theorem (Gelfand-Mazur theorem). If A is a unital Banach algebra in which every

nonzero element is invertible, then A is isometrically isomorphic to C.
Hint for proof . Let B = {λ1 : λ ∈ C}. Use the preceding result (theorem 3.1.25) to show that
B = A.
3.1.27. Proposition. Let a be an element of a unital algebra. Then σ(an ) = [σ(a)]n for every
n ∈ N. (The notation [σ(a)]n means {λn : λ ∈ σ(a)}.)
3.1.28. Definition. Let a be an element of a unital algebra. The spectral radius of a, denoted
by ρ(a), is defined to be sup{|λ| : λ ∈ σ(a)}.
n

n Proposition. If a is an element of a unital Banach algebra, then ρ(a) ≤ kak and ρ(a ) =
3.1.29.
ρ(a) .
3.1.30. Theorem (Spectral radius formula). If a is an element of a unital Banach algebra, then
ρ(a) = inf kan k1/n : n ∈ N = lim kan k1/n .

n→∞
Proof. The proof while not terribly deep is a bit complicated. A good version can be found
in [1], pages 19–20.
3.1.31. Exercise. The Volterra operator V (defined in example 2.2.18) is an integral operator on
the Banach space C([0, 1]).
(a) Compute the spectrum of the operator V . (Hint. Show that kV n k ≤ (n!)−1 for every
n ∈ N and use the spectral radius formula.)
(b) What does (a) say about the possibility of finding a continuous function f which satisfies
the integral equation Z x
f (x) − µ f (t) dt = h(x),
0
where µ is a scalar and h is a given continuous function on [0, 1]?
(c) Use the idea in (b) and the Neumann series expansion for 1 − µV (see proposition 3.1.11)
to calculate explicitly (and in terms of elementary functions) a solution to the integral
equation Z x
f (x) − 1
2 f (t) dt = ex .
0
3.1.32. Proposition. Let A be a unital Banach algebra and B a closed subalgebra containing 1A .
Then
(a) inv B is both open and closed in B ∩ inv A;
(b) σA (b) ⊆ σB (b) for every b ∈ B; and
(c) if b ∈ B and σA (b) has no holes (that is, if its complement in C is connected), then
σA (b) = σB (b).
Hint for proof . For part (c) consider (for b ∈ B) the function
fb : (σA (b))c → B ∩ inv A : λ 7→ b − λ1 .

3.2. Maximal Ideal Space

3.2.1. Proposition. If J is a proper ideal in a unital Banach algebra, then so is its closure.
3.2.2. Corollary. Every maximal ideal in a unital Banach algebra is closed.
3.2.3. Proposition. Let J be a proper closed ideal in a Banach algebra A. On the quotient algebra
A/J (see definition 2.3.23) define a function
k k : A/J → R : α 7→ inf{kuk : u ∈ α} .
This function is a norm on A/J, under this norm A/J is a Banach algebra, and the quotient map
is continuous with kπk ≤ 1. The Banach algebra A/J is the quotient algebra of A by J.
 3.3. CHARACTERS 27

3.2.4. Proposition. Let I be a proper closed ideal in a unital commutative Banach algebra A.
Then I is maximal if and only if A/I is a field.
The following is an immediate consequence of the preceding proposition and the Gelfand-Mazur
theorem (3.1.26).
3.2.5. Corollary. If I is a maximal ideal in a commutative unital Banach algebra, then A/I is
isometrically isomorphic to C.
3.2.6. Example. For every subset C of a topological space X the set
JC := f ∈ C(X) : f → (C) = {0}


is an ideal in C(X). Furthermore, JC ⊇ JD whenever C ⊆ D ⊆ X. (In the following we will

write Jx for the ideal J{x} .)
3.2.7. Proposition. Let X be a compact topological space and I be a proper ideal in C(X). Then
there exists x ∈ X such that I ⊆ Jx .
3.2.8. Proposition. Let x and y be points in a compact Hausdorff space. If Jx ⊆ Jy , then x = y.
3.2.9. Proposition. Let X be a compact Hausdorff space. A subset I of C(X) is a maximal ideal
in C(X) if and only if I = Jx for some x ∈ X.
3.2.10. Corollary. If X is a compact Hausdorff space, then the map x 7→ Jx from X to Max C(X)
is bijective.
Compactness is an important ingredient in proposition 3.2.9.


3.2.11. Example. In the Banach algebra Cb (0, 1) of bounded continuous functions on the in-
terval (0, 1) there exists a maximal ideal I such that for no point x ∈
(0, 1) is I = Jx . Let I be
a maximal ideal containing the ideal S of all functions f in Cb (0, 1) for which there exists a
neighborhood Uf of 0 in R such that f (x) = 0 for all x ∈ Uf ∩ (0, 1).

3.3. Characters
3.3.1. Definition. A character (or nonzero multiplicative linear functional) on an
algebra A is a nonzero homomorphism from A into C. The set of all characters on A is denoted
by ∆A.
3.3.2. Proposition. Let A be a unital algebra and φ ∈ ∆A. Then
(a) φ(1) = 1;
(b) if a ∈ inv A, then φ(a) 6= 0;
(c) if a is nilpotent (that is, if an = 0 for some n ∈ N), then φ(a) = 0;
(d) if a is idempotent (that is, if a2 = a), then φ(a) is 0 or 1; and
(e) φ(a) ∈ σ(a) for every a ∈ A.
We note in passing that part (e) of the preceding proposition does not give us an easy way of
showing that the spectrum σ(a) of an algebra element is nonempty. This would depend on knowing
that ∆(A) is nonempty.
3.3.3. Example. The identity map is the only character on the algebra C.
 
3.3.4. Example. Let A be the algebra of 2 × 2 matrices a = aij such that a12 = 0. This algebra
has exactly two characters φ(a) = a11 and ψ(a) = a22 . Hint. Use proposition 3.3.2.
3.3.5. Example. The algebra of all 2 × 2 matrices of complex numbers has no characters.
3.3.6. Proposition. Let A be a unital algebra and φ be a linear functional on A. Then φ ∈ ∆A
if and only if ker φ is closed under multiplication and φ(1) = 1.
Hint for proof . For the converse apply φ to the product of a − φ(a)1 and b − φ(b)1 for a, b ∈ A.
28 3. BANACH ALGEBRAS

3.3.7. Proposition. Every multiplicative linear functional on a unital Banach algebra A is con-
tinuous. In fact, if φ ∈ ∆(A), then φ is contractive and kφk = 1.
3.3.8. Example. Let X be a topological space and x ∈ X. We define the evaluation func-
tional at x, denoted by EX x, by
EX x : C(X) → C : f 7→ f (x) .
This functional is a character on C(X) and its kernel is Jx . When there is only one topological
space under discussion we simplify the notation from EX x to Ex . Thus, in particular, for f ∈ C(X)
we often write Ex (f ) for the more cumbersome EX x(f ).
Proposition 3.3.7 turns out to be very important: it says that characters on a unital Banach
algebra A all live on the unit sphere of the dual space A∗ . The trouble with the unit sphere in the
dual space is that, while it is closed and bounded, it is not compact in the usual (norm) topology
on A∗ . We need a new topology on A∗ , one that is weak enough to make the closed unit ball (and
hence the unit sphere) compact and yet strong enough to be Hausdorff.

3.4. The Gelfand Topology

3.4.1. Proposition. Let B be a Banach space. For each x ∈ B let
x∗∗ : B ∗ → C : f 7→ f (x) .
Then x∗∗ ∈ B ∗∗ for every x ∈ B and the map
JB : B → B ∗∗ : x 7→ x∗∗
is a linear isometry.
3.4.2. Definition. The map JB in the preceding proposition is the natural injection of a
Banach space B into its second dual B ∗∗ . (The rationale for the use of the word “natural” will
appear shortly—see example 4.3.2.) In situations where there is a single Banach space under
discussion we usually write J for JB . A Banach space B is reflexive if the map JB is surjective.
CAUTION. To show that a Banach space is reflexive it is not enough to show that there exists an
isometric isomorphism from the space to its second dual; what must be proved is that the natural
injection J is an isometric isomorphism.
3.4.3. Definition. Let B be a Banach space and J : B → B ∗∗ be the natural injection of B into
its second dual. The w∗ -topology (pronounced weak star topology) is the weak topology on B ∗
determined by ran J. That is, it is the weakest topology on B ∗ which makes every functional on
B ∗ of the form x∗∗ continuous. Thus we take as a subbase for the w∗ -topology on B ∗ all sets of
◦
the form (x∗∗ )← (U ) = {f ∈ B ∗ : f (x) ∈ U } where x ∈ B and U ⊆ C.
3.4.4. Proposition. The w∗ -topology on the dual of a Banach space is Hausdorff and is weaker
than the norm topology.
3.4.5. Notation. In the following proposition we use the notation Fin(S) to denote the family of
finite subsets of a set S.
3.4.6. Proposition. Let B be a Banach space. The family of all subsets of B ∗ of the form
V (f ; A; ) := {g ∈ B ∗ : |f (x) − g(x)| <  for all x ∈ A} ,
where f ∈ B ∗ , A ∈ Fin(B), and  > 0, is a base for the w∗ -topology on B ∗ .
3.4.7. Notation. Let (fλ ) be a net in the dual B ∗ of a Banach space and let g ∈ B ∗ . To indicate
w∗ / g.
that the net (fλ ) converges to g in the w∗ -topology we write fλ
The following result is a great joy. It tells us that the w∗ -topology on the dual of a Banach
space is nothing more than the topology of pointwise convergence.
 3.4. THE GELFAND TOPOLOGY 29

3.4.8. Proposition. Let (fλ ) be a net in the dual B ∗ of a Banach space B and g ∈ B ∗ . Then
w∗ / g if and only if fλ (x) → g(x) in C for every x ∈ B.
fλ
3.4.9. Definition. In proposition 3.3.7 we discovered that every character on a unital Banach
algebra A lives on the unit sphere of the dual A∗ . Thus we may give the set ∆(A) of characters on
A the relative w∗ -topology it inherits from A∗ . This is the Gelfand topology on ∆(A) and the
resulting topological space we call the character space (or the structure space) of A.
In order to show that the character space is compact we need an important theorem from
functional analysis.
3.4.10. Theorem (Alaoglu’s theorem). If V is a normed linear space, then the closed unit ball of
its dual V ∗ is compact in the w∗ -topology.
Proof. See [5], theorem V.3.1.
3.4.11. Proposition. The character space of a unital Banach algebra is a compact Hausdorff
space.
3.4.12. Example. The maximal ideal space of the unital Banach algebra l1 (Z) (see example 3.1.8)
is (homeomorphic to) the unit circle T.
Hint for proof . For each z ∈ T define
∞
X
ψz : l1 (Z) → C : a 7→ ak z k .
k=−∞

Show that ψz ∈ ∆l1 (Z). Then show that the map

ψ : T → ∆l1 (Z) : z 7→ ψz
is a homeomorphism.
3.4.13. Proposition. If φ ∈ ∆A where A is a unital algebra, then ker φ is a maximal ideal in A.
Hint for proof . To show maximality, suppose I is an ideal

in A which properly contains ker φ.
Choose z ∈ I \ ker φ. Consider the element x − φ(x)/φ(z) z where x is an arbitrary element of A.
3.4.14. Proposition. A character on a unital algebra is completely determined by its kernel.
Hint for proof . Let a be an element of the algebra and φ be a character. For how many complex
numbers λ can a2 − λa belong to the kernel of φ?
3.4.15. Corollary. If A is a unital algebra, then the map φ 7→ ker φ from ∆A to Max A is injective.
3.4.16. Proposition. Let I be a maximal ideal in a unital commutative Banach algebra A. Then
there exists a character on A whose kernel is I.
Hint for proof . Use corollary 3.2.5 Why can we think of the quotient map as a character?
3.4.17. Corollary. If A is a unital commutative Banach algebra, then the map φ 7→ ker φ is a
bijection from ∆A onto Max A.
3.4.18. Definition. Let A be a unital commutative Banach algebra. In light of the preceding
corollary we can give Max A a topology under which it is homeomorphic to the character space ∆A.
This is the maximal ideal space of A. Since ∆A and Max A are homeomorphic it is common
practice to identify them and so ∆A is often called the maximal ideal space of A.
3.4.19. Definition. Let X be a compact Hausdorff space and x ∈ X. Recall that in example 3.3.8
we defined EX x, the evaluation functional at x by
EX x(f ) := f (x)
30 3. BANACH ALGEBRAS

for every f ∈ C(X). The map

EX : X → ∆C(X) : x 7→ EX x
is the evaluation map on X. As was mentioned earlier when only one topological space is being
considered we usually shorten EX to E and EX x to Ex .
3.4.20. Notation. To indicate that two topological spaces X and Y are homeomorphic we write
X ≈Y.
3.4.21. Proposition. Let X be a compact Hausdorff space. Then the evaluation map on X
EX : X → ∆C(X) : x 7→ EX x
is a homeomorphism. Thus we have
X ≈ ∆C(X) ≈ Max C(X) .
More is true: not only is each EX a homeomorphism between compact Hausdorff spaces,
but E itself is a natural equivalence between functors—the identity functor and the ∆C functor.
Discussion of this will have to wait until we discuss the language of categories. (See example 4.3.3.)
It suffices for the moment to say that the identification between a compact Hausdorff space X and
its character space and its maximal ideal space is so strong that many people speak of them as
if they were actually equal. It is very common to hear, for example, that “the maximal ideals in
C(X) are just the points of X”. Although not literally true, it does sound a bit less intimidating
than “the maximal ideals of C(X) are the kernels of the evaluation functionals at points of X”.
3.4.22. Proposition. Let X and Y be compact Hausdorff spaces and F : X → Y be continuous.
Define C(F ) on C(Y ) by
C(F )(g) = g ◦ F
for all g ∈ C(Y ). Then
(a) C(F ) maps C(Y ) into C(X).
(b) The map C(F ) is a contractive unital Banach algebra homomorphism.
(c) C(F ) is injective if and only if F is surjective.
(d) C(F ) is surjective if and only if F is injective.
(e) If X is homeomorphic to Y , then C(X) is isometrically isomorphic to C(Y ).
3.4.23. Proposition. Let A and B be unital commutative Banach algebras and T : A → B be a
unital algebra homomorphism. Define ∆T on ∆B by
∆T (ψ) = ψ ◦ T
for all ψ ∈ ∆B. Then
(a) ∆T maps ∆B into ∆A.
(b) The map ∆T : ∆B → ∆A is continuous.
(c) If T is surjective, then ∆T is injective.
(d) If T is an (algebra) isomorphism, then ∆T is a homeomorphism.
(e) If A and B are (algebraically) isomorphic, then ∆A and ∆B are homeomorphic.
3.4.24. Corollary. Let X and Y be compact Hausdorff spaces. If C(X) and C(Y ) are algebraically
isomorphic, then X and Y are homeomorphic.
3.4.25. Corollary. Two compact Hausdorff spaces are homeomorphic if and only if their algebras
of continuous complex valued functions are (algebraically) isomorphic.
3.4.26. Corollary. Let X and Y be compact Hausdorff spaces. If C(X) and C(Y ) are algebraically
isomorphic, then they are isometrically isomorphic.
 3.5. THE GELFAND TRANSFORM 31

3.5. The Gelfand Transform

3.5.1. Definition. Let A be a commutative Banach algebra and a ∈ A. Define
ΓA a : ∆A → C : φ 7→ φ(a)
for every φ ∈ ∆(A). (Alternative notations: when no confusion seems likely we frequently write Γa
or b
a for ΓA a.) The map ΓA is the Gelfand transform on A.
Since ∆A ⊆ A∗ it is clear that ΓA a is just the restriction of a∗∗ to the character space of A.
Furthermore the Gelfand topology on ∆A is the relative w∗ -topology, the weakest topology under
which a∗∗ is continuous on ∆A for each a ∈ A, so ΓA a is a continuous function on ∆A. Thus
ΓA : A → C(∆A).
As a matter of brevity and convenience the element ΓA a = Γa = b a is usually called just the
Gelfand transform of a—because the phrase the Gelfand transform on A evaluated at a is awkward.
3.5.2. Definition. We say that a family F of functions on a set S separates points of S (or is
a separating family of functions on S) if for every pair of distinct elements x and y of S there
exists f ∈ F such that f (x) 6= f (y).
3.5.3. Proposition. Let X be a compact topological space. Then C(X) is separating if and only if
X is Hausdorff.
3.5.4. Proposition. If A is a unital commutative Banach algebra, then ΓA : A → C(∆A) is a
unital contractive algebra homomorphism having norm one. Furthermore the range of ΓA is a
separating subalgebra of C(∆A).
3.5.5. Proposition. Let a be an element of a unital commutative Banach algebra A. Then a is
invertible in A if and only if â is invertible in C(∆A).
3.5.6. Proposition. Let A be a unital commutative Banach algebra and a be an element of A.
Then ran â = σ(a) and kâku = ρ(a).
3.5.7. Definition. An element a of a Banach algebra is quasinilpotent if lim kan k1/n = 0.
n→∞

3.5.8. Proposition. Let a be an element of a unital commutative Banach algebra A. Then the
following are equivalent:
(a) a is quasinilpotent;
(b) ρ(a) = 0;
(c) σ(a) = {0};
(d) Γa = 0;
(e) φ(a)T= 0 for every φ ∈ ∆A;
(f) a ∈ Max A.
3.5.9. Definition. A Banach algebra is semisimple if it has no nonzero quasinilpotent elements.
3.5.10. Proposition. Let A be a unital commutative Banach algebra. Then the following are
equivalent:
(a) A is semisimple;
(b) if ρ(a) = 0, then a = 0;
(c) if σ(a) = {0}, then a = 0;
(d) the Gelfand transform ΓA is a monomorphism (that is, an injective homomorphism);
T φ(a) = 0 for every φ ∈ ∆A, then a = 0;
(e) if
(f) Max A = {0}.
3.5.11. Proposition. Let A be a unital commutative Banach algebra. Then the following are
equivalent:
(a) ka2 k = kak2 for all a ∈ A;
32 3. BANACH ALGEBRAS

(b) ρ(a) = kak for all a ∈ A; and

(c) the Gelfand transform is an isometry; that is, kb
aku = kak for all a ∈ A.
3.5.12. Example. The Gelfand transform on l1 (Z) is not an isometry.
Recall from proposition 3.4.21 that when X is a compact Hausdorff space the evaluation map-
ping EX identifies the space X with the maximal ideal space of the Banach algebra C(X). Thus
according to proposition 3.4.22 the mapping CEX identifies the algebra C(∆(C(X))) of continuous
functions on this maximal ideal space with the algebra C(X) itself. It turns out that the Gelfand
transform on the algebra C(X) is just the inverse of this identification map.
3.5.13. Example. Let X be a compact Hausdorff space. Then the Gelfand transform on the
Banach algebra C(X) is an isometric isomorphism. In fact, on C(X) the Gelfand transform ΓX is
(CEX )−1 .

3.6. The Fourier Transform

3.6.1. Definition. For each f in the commutative Banach algebra L1 (R) (see example 3.1.9) define
a function fe by
Z ∞
fe(x) = f (t)e−itx dt .
−∞

The function fe is the Fourier transform of f .

An important fact from real analysis, the so-called Riemann-Lebesgue lemma, is very useful in
studying the Fourier transform.
3.6.2. Proposition (Riemann-Lebesgue Lemma). If f ∈ L1 (R), then fe ∈ C0 (R).
Proof. See any good real analysis text, for example [15](21.39).
3.6.3. Proposition. The Fourier transform
F : L1 (R) → C0 (R) : f 7→ fe
is a homomorphism of Banach algebras. (See example 3.1.9.)
What particular property of the homomorphism in the preceding proposition would you guess
is particularly useful in applications?
3.6.4. Proposition. The Banach algebra L1 (R) is not unital.
3.6.5. Example. The maximal ideal space of the Banach algebra L1 (R) is R itself and the Gelfand
transform on L1 (R) is the Fourier transform.
Proof. See [2], pages 169–171.

The function t 7→ eit is a bijection from the interval [−π, π) to the unit circle T in the complex
plane. One consequence of this is that we need not distinguish between
(i) 2π-periodic functions on R,
(ii) all functions on [−π, π),
(iii) functions f on [−π, π] such that f (−π) = f (π), and
(iv) functions on T.
In the sequel we will frequently without further explanation identify these classes of functions.
Another convenient identification is the one between the unit circle T in C and the maximal
ideal space of the algebra l1 (Z). The homeomorphism ψ between these two compact Hausdorff
space was defined in example 3.4.12. It is often technically more convenient in working with the
 3.6. THE FOURIER TRANSFORM 33

Gelfand transform Γa of an element a ∈ l1 (Z) to treat it as a function, let’s call it Ga , whose domain
is T as the following diagram suggests.
T
Ga
ψ

 
∆l1 (Z) /C
Γa

Thus for a ∈ l1 (Z) and z ∈ T we have

∞
X
Ga (z) = Γa (ψz ) = ψz (a) = an z n .
k=−∞

3.6.6. Definition. If f ∈ L1 ([−π, π)), the Fourier series for f is the series
∞
X
fe(n) exp(int) −π ≤t≤π
n=−∞

where the sequence fe is defined by

Z π
1
fe(n) = 2π f (t) exp(−int) dt
−π

for all n ∈ Z. The doubly infinite sequence fe is the Fourier transform of f , and the number fe(n)
is the nth Fourier coefficient of f . If fe ∈ l1 (Z) we say that f has an absolutely convergent
Fourier series. The set of all continuous functions on T with absolutely convergent Fourier series
is denoted by A C(T).
3.6.7. Proposition. If f is a continuous 2π-periodic function on R whose Fourier transform is
zero, then f = 0.
3.6.8. Corollary. The Fourier transform on C(T) is injective.
3.6.9. Proposition. The Fourier transform on C(T) is a left inverse of the Gelfand transform
on l1 (Z).
3.6.10. Proposition. The range of the Gelfand transform on l1 (Z) is the unital commutative
Banach algebra A C(T).
3.6.11. Proposition. There are continuous functions whose Fourier series diverge at 0.
Proof. See, for example, [15], exercise 18.45.)
What does the preceding result say about the Gelfand transform Γ : l1 (Z) → C(T)?
Suppose a function f belongs to AC(T) and is never zero. Then 1/f is certainly continuous
on T, but does it have an absolutely convergent Fourier series? One of the first triumphs of the
abstract study of Banach algebras was a very simple proof of the answer to this question given
originally by Norbert Wiener. Wiener’s original proof by comparison was quite difficult.
3.6.12. Theorem (Wiener’s theorem). Let f be a continuous function on T which is never zero.
If f has an absolutely convergent Fourier series, then so does its reciprocal 1/f .
3.6.13. Example. The Laplace transform can also be viewed as a special case of the Gelfand
transform. For details see [2], pages 173–175.
 CHAPTER 4

INTERLUDE: THE LANGUAGE OF CATEGORIES

4.1. Objects and Morphisms

4.1.1. Definition. Let A be a class, whose members we call objects. For every pair (S, T ) of
objects we associate a set Mor(S, T ), whose members we call morphisms from S to T . We assume
that Mor(S, T ) and Mor(U, V ) are disjoint unless S = U and T = V .
We suppose further that there is an operation ◦ (called composition) that associates with
every α ∈ Mor(S, T ) and every β ∈ Mor(T, U ) a morphism β ◦ α ∈ Mor(S, U ) in such a way that:
(1) γ ◦ (β ◦ α) = (γ ◦ β) ◦ α whenever α ∈ Mor(S, T ), β ∈ Mor(T, U ), and γ ∈ Mor(U, V );
(2) for every object S there is a morphism IS ∈ Mor(S, S) satisfying α ◦ IS = α whenever
α ∈ Mor(S, T ) and IS ◦ β = β whenever β ∈ Mor(R, S).
Under these circumstances the class A, together with the associated families of morphisms, is a
category.
4.1.2. Example. The category SET comprises the class of all sets (as objects) and all functions
between these sets (as morphisms).
4.1.3. Example. The category CpH comprises the class of all compact Hausdorff spaces (as
objects) and all continuous functions between these spaces (as morphisms).
4.1.4. Example. We will denote by BAN∞ the category of Banach spaces and continuous linear
maps between these spaces.
4.1.5. Example. The category BA∞ comprises the class of all Banach algebras (as objects) and all
continuous algebra homomorphisms between these algebras (as morphisms). We will further denote
by CBA∞ the category of commutative Banach algebras and continuous algebra homomorphisms,
and by UCBA∞ the category of unital commutative Banach algebras and unital continuous algebra
homomorphisms. Following Palmer [21] we will refer to these categories as topological categories
of Banach algebras.
4.1.6. Example. We will denote by BAN1 the category of Banach spaces and contractive linear
maps between these spaces.
4.1.7. Example. The category BA1 comprises the class of all Banach algebras (as objects) and all
contractive algebra homomorphisms between these algebras (as morphisms). We will further denote
by CBA1 the category of commutative Banach algebras and contractive algebra homomorphisms,
and by UCBA1 the category of unital commutative Banach algebras and unital contractive algebra
homomorphisms. Again following Palmer [21] we will refer to these categories as geometric
categories of Banach algebras.
4.1.8. Example. The category HIL comprises the class of all Hilbert spaces (as objects) and all
bounded linear maps between these spaces (as morphisms).
4.1.9. Example. We may regard the set N of natural numbers as a category where for m, n ∈ N
the morphisms in Mor(m, n) are the m×n matrices and composition is taken to be ordinary matrix
multiplication.
4.1.10. Example. A monoid is a semigroup with identity. We may regard a monoid (M, ∗ ) as a
category with a single object M (or any other object you choose). Then there is a single class of
35
36 4. INTERLUDE: THE LANGUAGE OF CATEGORIES

morphisms Mor(M, M ) and the morphisms in this class are the elements of the monoid M . The
composite of two morphisms a and b is their product a ∗ b in M .
4.1.11. Notation. If C is a category we denote by C0 the class of objects in C and by C1 the
class of morphism in C. Thus, for example, the notation X ∈ CpH0 would indicate that X is
a compact Hausdorff space and f ∈ CpH1 would mean that f is a continuous function between
compact Hausdorff spaces.
φ
4.1.12. Notation. In these notes we restrict the notation A / B to morphisms. When this
notation appears it should be clear from context what category is being discussed. We then infer
that A and B are objects in that category and φ : A → B is a morphism.
4.1.13. Definition. The terminology for inverses of morphisms in categories is essentially the same
α / β
as for functions. Let S T and T / S be morphisms in a category. If β ◦ α = IS , then β is a
left inverse of α and, equivalently, α is a right inverse of β. We say that the morphism α is
β
an isomorphism (or is invertible) if there exists a morphism T / S which is both a left and a
−1
right inverse for α. Such a function is denoted by α and is called the inverse of α. To indicate
that objects A and B in some category are isomorphic we will, in general, use the notation A ∼ = B.
There is one exception however: in various categories of topological spaces X ≈ Y means that the
spaces X and Y are homeomorphic (topologically isomorphic).

4.2. Functors
4.2.1. Definition. If A and B are categories a (covariant) functor F from A to B (written
A
F / B) is a pair of maps: an object map F which associates with each object S in A an
object F (S) in B and a morphism map (also denoted by F ) which associates with each morphism
f ∈ Mor(S, T ) in A a morphism F (f ) ∈ Mor(F (S), F (T )) in B, in such a way that
(1) F (g ◦ f ) = F (g) ◦ F (f ) whenever g ◦ f is defined in A; and
(2) F (idS ) = idF (S) for every object S in A.

The definition of a contravariant functor (or cofunctor) A / B differs from the

F

preceding definition only in that, first, the morphism map associates with each morphism f ∈
Mor(S, T ) in A a morphism F (f ) ∈ Mor(F (T ), F (S)) in B and, second, condition (1) above is
replaced by
(10 ) F (g ◦ f ) = F (f ) ◦ F (g) whenever g ◦ f is defined in A.
4.2.2. Definition. A lattice is a partially ordered set in which every pair of elements has a
supremum and an infimum. A lattice L is order complete if the sup A and inf A exist (in L) for
every nonempty subset A of L.
4.2.3. Example. Let S be a nonempty set.
(a) The power set P(S) of S partially ordered by ⊆ is an order complete lattice.
(b) The class of order complete lattices and order preserving maps is a category.
(c) For each function f between sets let P(f ) = f → . Then P is a covariant functor from
the category of sets and functions to the category of order complete lattices and order
preserving maps.
(d) For each function f between sets let P(f ) = f ← . Then P is a contravariant functor from
the category of sets and functions to the category of order complete lattices and order
preserving maps.
4.2.4. Example. Let T : B → C be a bounded linear map between Banach spaces. Define
T ∗ : C ∗ → B ∗ : g 7→ g T .
The map T ∗ is the adjoint of T .
 4.2. FUNCTORS 37

(a) The composite g T is in B ∗ and the pair of maps V 7→ V ∗ and T 7→ T ∗ is a contravariant

functor from the category BAN∞ of Banach spaces and bounded linear maps into itself.
The pair of maps B 7→ B ∗∗ and T 7→ T ∗∗ is a covariant functor. We will call this the
second dual functor.
(b) If T is contractive, then g ◦ T ∈ B ∗ and the pair of maps B 7→ B ∗ and T 7→ T ∗ is a
contravariant functor from the category BAN1 of Banach spaces and contractive linear
maps into itself.
(c) We may regard the functor C (see example 4.2.5) as a contravariant functor from the
category CpH of compact Hausdorff spaces and continuous maps into the category BAN1
of Banach spaces and contractive linear maps (because every Banach algebra is a Banach
space). The composite of this functor with the one in part (b) is a covariant functor
from
∗
CpH to BAN1 . We will denote this composite by C ∗ so that C ∗ (X) means C(X) .
4.2.5. Example. Let X and Y be compact Hausdorff spaces and F : X → Y be continuous. As
in proposition 3.4.22 define C(F ) on C(Y ) by
C(F )(g) = g ◦ F
for all g ∈ C(Y ). Then the pair of maps X 7→ C(X) and F 7→ C(F ) is a contravariant functor from
the category CpH of compact Hausdorff spaces and continuous maps to the category UCBA1
of unital commutative Banach algebras and contractive unital algebra homomorphisms. Since
every contractive homomorphism is continuous we may, if we choose, regard C as a functor from
the category CpH to the category UCBA∞ of unital commutative Banach algebras and unital
continuous algebra homomorphisms.
4.2.6. Example. Let A and B be unital commutative Banach algebras and T : A → B be a unital
algebra homomorphism. As in proposition 3.4.23 define ∆T on ∆B by
∆T (ψ) = ψ ◦ T
for all ψ ∈ ∆B. Then the pair of maps A 7→ ∆A and T 7→ ∆T is a contravariant functor from
the category UCBA of unital commutative Banach algebras and unital algebra homomorphisms
to the category CpH of compact Hausdorff spaces and continuous maps.
4.2.7. Example. All the categories that are of interest in this course are concrete categories. A
concrete category is, roughly speaking, one in which the objects are sets usually with additional
structure (algebraic operations, inner products, norms, topologies, and the like) and the morphisms
are maps (functions) which preserve, in some sense, the additional structure. If A is an object in
f
some concrete category C, we denote by A its underlying set. And if A / B is a morphism in
C we denote by f the map from A to B regarded simply as a function between sets. It is easy
to see that  , which takes objects in C to objects in SET (the category of sets and maps) and
morphisms in C to morphisms in SET, is a functor. It is referred to as the forgetful functor.
(The definite article here is inaccurate but nearly universal.) In the category VEC of vector spaces
and linear maps, for example,  causes a vector space V to “forget” about its addition and scalar
multiplication; V is just a set. And if T : V → W is a linear transformation, then T : V → W
is just a map between sets—it has “forgotten” about preserving operations (which are no longer
there to preserve). For more precise definitions of concrete categories and forgetful functors consult
any text on category theory.
4.2.8. Convention. In these notes all categories are concrete.
4.2.9. Definition. In light of the preceding convention it makes sense to define in any (concrete)
category a monomorphism to be an injective morphism and an epimorphism to be a surjective
morphism. This terminology, which makes no sense in arbitrary categories, deviates from the usage
in category theory proper where monomorphism (or monic) means left-cancellable and epimorphism
(or epic) means right-cancellable. In most of the concrete categories which we consider morphisms
38 4. INTERLUDE: THE LANGUAGE OF CATEGORIES

are injective if and only if they are left-cancellable. The agreement between surjective and right-
cancellable is less general. In the category of topological spaces and continuous maps, for example,
morphisms are right-cancellable if and only if they have dense range.

4.2.10. Example. If C is a category let C2 be the category whose objects are ordered pairs of
f
objects in C and whose morphisms are ordered pairs of morphisms in C. Thus if A / C and
g (f,g)
/ D are morphisms in C, then (A, B) / (C, D) (where (f, g)(a, b) = f (a), g(b) for all


B
a ∈ A and b ∈ B) is a morphism in C2 . Composition of morphism is defined in the obvious way:
(f, g) ◦ (h, j) = (f ◦ h, g ◦ j). We define the diagonal functor C
D / C2 by D(A) := (A, A).
This is a covariant functor.

4.3. Natural Transformations

F //
4.3.1. Definition. Let A and B be categories and A B be covariant functors. A natural
G
transformation from F to G is a map τ which assigns to each object A in category A a morphism
τA f
F (A) / G(A) in category B in such a way that for every morphism A / A0 in A the following
diagram commutes.
F (f )
F (A) / F (A0 )

τA τA0

 
G(A) / G(A0 )
G(f )

We denote such a transformation by F / G. (The definition of a natural transformation between

τ

two contravariant functors should be obvious: just reverse the horizontal arrows in the preceding
diagram.)
τ /
A natural transformation F G is a natural equivalence (or a natural isomorphism)
if each morphism τA is an isomorphism in B. Two functors are naturally equivalent if there
exists a natural equivalence between them.

4.3.2. Example. On the category BAN∞ of Banach spaces and bounded linear transformations
let I be the identity functor and ( · )∗∗ be the second dual functor (see example 4.2.4(a) ). Show
that the map J which takes each Banach space B to its natural injection JB (see definition 3.4.2) is
a natural transformation from I to ( · )∗∗ . In the category of reflexive Banach spaces and bounded
linear maps this natural transformation is a natural equivalence.

4.3.3. Example. The mapping E : X 7→ EX , which takes compact Hausdorff spaces to their
corresponding evaluation maps is a natural equivalence between the identity functor and the functor
∆C in the category of compact Hausdorff spaces and continuous maps. (See example 3.3.8 and
proposition 3.4.21.)

Thus X and ∆C(X) are not only homeomorphic, they are naturally homeomorphic. This is the
justification for the very common informal assertion that the maximal ideals of C(X) “are” just the
points of X.

4.3.4. Example. The Gelfand transform Γ is a natural transformation from the identity functor to
the C∆ functor on the category UCBA of unital commutative Banach algebras and unital algebra
homomorphisms.
 4.4. UNIVERSAL MORPHISMS 39

4.4. Universal Morphisms

Much of mathematics involves the construction of new objects from old ones—things such as
products, coproducts, quotients, completions, compactifications, and unitizations. More often than
not it is possible to characterize such a construction by the existence of a unique morphism having
some particular property. Because this morphism and its corresponding property characterize the
construction in question, they are referred to as a universal morphism and a universal property,
respectively. Here is one very common way in which such morphisms arise.

4.4.1. Definition. Let A / B be a functor between categories A and B and B be an object

F

in B. A pair (A, u) with A an object in A and u a B-morphism from B to F(A) is a universal

morphism for B (with respect to the functor F) if for every object A0 in A and every B-morphism
f ˜
B / F(A0 ) there exists a unique A-morphism A f / A0 such that the following diagram commutes.

B
u / F(A) A

F(f˜) f˜
f
  
F(A0 ) A0 (4.1)

In this context the object A is often referred to as a universal object in A.

4.4.2. Example. Let S be a set and VEC / SET be the forgetful functor from the category

ι /
VEC to the category SET. If there exists a vector space V and an injection S V which
constitute a universal morphism for S (with respect to ), then V is the free vector space
over S. Of course merely defining an object does not guarantee its existence. In fact, free vector
spaces exist over arbitrary sets. Given the set S let V be the set of all complex valued functions
on S which have finite support. Define addition and scalar multiplication pointwise. The map
ι : s 7→ χ{s} of each element s ∈ S to the characteristic function of {s} is the desired injection. To
verify that V is free over S it must be shown that for every vector space W and every function
f fe
S / W there exists a unique linear map V / W which makes the following diagram commute.

S
ι / V V

f˜ f˜
f
  
W W

4.4.3. Example. Let S be a nonempty set and let S 0 = S∪{1} where 1 is any element not belonging
to S. A word in the language S is a sequence s of elements of S 0 which is eventually 1 and satisfies:
if sk = 1, then sk+1 = 1. The constant sequence (1, 1, 1, . . . ) is called the empty word. Let F
be the set of all words of members in the language S. Suppose that s = (s1 , . . . , sm , 1, 1, . . . ) and
t = (t1 , . . . , tn , 1, 1, . . . ) are words (where s1 , . . . , sm , t1 , . . . , tn ∈ S). Define

x ∗ y := (s1 , . . . , sm , t1 , . . . , tn , 1, 1, . . . ).

This operation is called concatenation. It is not difficult to see that the set F under concatenation
is a monoid (a semigroup with identity) where the empty word is the identity element. This is the
free monoid generated by S. If we exclude the empty word we have the free semigroup
40 4. INTERLUDE: THE LANGUAGE OF CATEGORIES

generated by S. The associated diagram is

S
ι / F F

f˜ f˜
f
  
G G
where ι is the obvious injection s 7→ (s, 1, 1, . . . ) (usually treated as an inclusion mapping), G is
an arbitrary semigroup, f : S → G is an arbitrary function, and  is the forgetful functor from the
category of monoids and homomorphisms (or the category of semigroups and homomorphisms) to
the category SET.
4.4.4. Example. Here is the usual presentation of the coproduct of two objects in a category. Let
A1 and A2 be two objects in a category C. A coproduct of A1 and A2 is a triple (Q, ι1 , ι2 ) with
ιk
Q an object in C and Ak / Q (k = 1, 2) morphisms in C which satisfies the following condition:
fk
if B is an arbitrary object in C and Ak / B (k = 1, 2) are arbitrary morphisms in C, then there
f
exists a unique C-morphism Q / B which makes the following diagram commute.

? BO _
f1 f2
f

A1 /Qo A2 (4.2)
ι1 ι2

It may not be obvious at first glance that this construction is universal in the sense of defini-
tion 4.4.1. To see that it in fact is, let D be the diagonal functor from a category C to the category
of pairs C2 (see example 4.2.10). Suppose that (Q, ι1 , ι2 ) is a coproduct of the C-objects A1 and A2
ι /
in the sense defined above. Then A = (A1 , A2 ) is an object in C2 , A D(Q) is a C2 -morphism,
and the pair (A, ι) is universal in the sense of 4.4.1. The diagram corresponding to diagram (4.1) is

A
ι / D(Q) Q

D(f˜) f˜
f
  
D(B) B (4.3)

fk
where B is an arbitrary object in C and (for k = 1, 2) Ak / B are arbitrary C-morphisms so
that f = (f1 , f2 ) is a C2 -morphism.
4.4.5. Example. The coproduct of two objects H and K in the category HIL of Hilbert spaces
and bounded linear maps (and more generally in the category of inner product spaces and linear
maps) is their (external orthogonal) direct sum H ⊕ K (see 1.2.19).
4.4.6. Example. The coproduct of two objects S and T in the category SET is their disjoint
union S ] T .
4.4.7. Example. Let A and B be Banach spaces. On the Cartesian product A × B define addition
and scalar multiplication pointwise. For every (a, b) ∈ A × B let k(a, b)k = max{kak, kbk}. This
makes A × B into a Banach space, which is denoted by A ⊕ B and is called the direct sum of A
and B. The direct sum is a coproduct in the topological category BAN∞ of Banach spaces but
not in the corresponding geometrical category BAN1 .
 4.4. UNIVERSAL MORPHISMS 41

4.4.8. Example. To construct a coproduct on the geometrical category BAN1 of Banach spaces
define the vector space operations pointwise on A × B but as a norm use k(a, b)k1 = kak + kbk for
all (a, b) ∈ A × B.
Virtually everything in category theory has a dual concept—one that is obtained by reversing
all the arrows. We can, for example, reverse all the arrows in diagram (4.1).

4.4.9. Definition. Let A / B be a functor between categories A and B and B be an object

F

in B. A pair (A, u) with A an object in A and u a B-morphism from F(A) to B is a co-universal

morphism for B (with respect to F) if for every object A0 in A and every B-morphism
f ˜
F(A0 ) / B there exists a unique A-morphism A0 f / A such that the following diagram commutes.

B_o
u
F(A) AO
O

F(f˜) f˜
f

F(A0 ) A0 (4.4)
Some authors reverse the convention and call the morphism in 4.4.1 co-universal and the one here
universal. Other authors, this one included, call both universal morphisms.
4.4.10. Convention. Morphisms of both the types defined in 4.4.1 and 4.4.9 will be referred to
as universal morphisms.
4.4.11. Example. The usual categorical approach to products is as follows. Let A1 and A2 be
two objects in a category C. A product of A1 and A2 is a triple (P, π1 , π2 ) with P an object in
ιk
C and Q / Ak (k = 1, 2) morphisms in C which satisfies the following condition: if B is an
fk
arbitrary object in C and B / Ak (k = 1, 2) are arbitrary morphisms in C, then there exists a
f
unique C-morphism B / P which makes the following diagram commute.

B
f1 f2
f

  
A1 o π1 P π2
/ A2 (4.5)

This is (co)-universal in the sense of definition 4.4.9.

4.4.12. Example. The product of two objects H and K in the category HIL of Hilbert spaces
and bounded linear maps (and more generally in the category of inner product spaces and linear
maps) is their (external orthogonal) direct sum H ⊕ K (see 1.2.19).
4.4.13. Example. The product of two objects S and T in the category SET is their Cartesian
product S × T .
4.4.14. Example. If A and B are Banach algebras their direct sum A ⊕ B is a product in both
the topological and geometric categories, BA∞ and BA1 , of Banach algebras. Compare this to
the situation discussed in examples 4.4.7 and 4.4.8.
4.4.15. Proposition. Universal objects in a category are essentially unique.
4.4.16. Definition. Let M and N be metric spaces. We say that N is a completion of M if N
is complete and M is isometric to a dense subset of N .
4.4.17. Proposition. Every metric space has a completion.
42 4. INTERLUDE: THE LANGUAGE OF CATEGORIES

Hint for proof . Let (M, d) be a metric space and fix a ∈ M . For each x ∈ M define φx : M → R
by φx (u) = d(x, u) − d(u, a). Show first that for every x ∈ M the function φx belongs to the space
Cb (M, R) of bounded real valued continuous functions on M . Then show that φ : M → Cb (M, R)
is an isometry. (To verify that du (φx , φy ) ≥ d(x, y) notice that kφx − φy ku ≥ |φx (y) − φy (y)|. )
Explain why the closure of ran φ in Cb (M ) is a completion of M .
4.4.18. Example. Let MS1 be the category of metric spaces and contractive maps and CMS1
be the category of complete metric spaces and contractive maps. The map M 7→ M c from MS1 to
CMS1 which takes a metric space to its completion is universal in the sense of definition 4.4.1.
The following consequence of proposition 4.4.15 allows us to speak of the completion of a metric
space.
4.4.19. Corollary. Metric space completions are unique (up to isometry).
Every inner product space is a metric space by propositions 1.2.11 and 1.2.12. It is an important
fact that completing an inner product space as a metric space produces a Hilbert space.
4.4.20. Proposition. Let V be an inner product space and H be the metric space completion
of V . Then the inner product on V can be extended to an inner product on H and the metric on
H induced by this inner product is the same as the original metric on H.
4.4.21. Proposition. Let V be an inner product space and H be its completion (see proposi-
tion 4.4.20). Then every operator on V extends to an operator on H.
Next is a closely related result which we will need in the sequel.
4.4.22. Proposition. Let D be a dense subset of a Hilbert space H and suppose that S and T are
operators on H such that hSu, vi = hu, T vi for all u, v ∈ D. Then hSx, yi = hx, T yi for all x,
y ∈ H.
4.4.23. Definition. A category A is a subcategory of category B if A0 ⊆ B0 and A1 ⊆ B1 .
It is a full subcategory of B if, additionally, for all objects A1 and A2 in A every B-morphism
τ /
A1 A2 is also an A-morphism.
4.4.24. Example. The category of unital algebras and unital algebra homomorphism is a subcat-
egory of the category of algebras and algebra homomorphisms, but not a full subcategory.
4.4.25. Example. The category of complete metric spaces and continuous maps is a full subcat-
egory of the category of metric spaces and continuous maps.
 CHAPTER 5

C ∗ -ALGEBRAS

5.1. Adjoints of Hilbert Space Operators

5.1.1. Definition. Let H and K be Hilbert spaces. A sesquilinear functional φ : H × K → C on
H × K is bounded if there exists a constant M > 0 such that
|φ(x, y)| ≤ M kxkkyk
for all x ∈ H and y ∈ K.
5.1.2. Proposition. If φ : H × K → C is a bounded sesquilinear functional on the product of two
Hilbert spaces, then the following numbers (exist and) are equal:
• sup{|φ(x, y)| : kxk ≤ 1, kyk ≤ 1}
• sup{|φ(x, y)| : kxk = kyk = 1}
 
|φ(x, y)|
• sup : x, y 6= 0
kxk kyk
• inf{M > 0 : |φ(x, y)| ≤ M kxk kyk for all x ∈ H, y ∈ K}.
The proof of the preceding proposition is virtually identical to the one for linear maps (see
proposition 2.2.4).
5.1.3. Definition. Let φ : H × K → C be a bounded sesquilinear functional on the product of two
Hilbert spaces. We define kφk, the norm of φ, to be any of the (equal) expressions in the preceding
result.
5.1.4. Proposition. Let A : H → K be a bounded linear map between Hilbert spaces. Then
φ : H × K → C : (x, y) 7→ hAx, yi is a bounded sesquilinear functional on H × K and kφk = kAk.
5.1.5. Proposition. Let φ : H × K → C be a bounded sesquilinear functional on the product of
two Hilbert spaces. Then there exist unique bounded linear maps B ∈ B(H, K) and C ∈ B(K, H)
such that
φ(x, y) = hBx, yi = hx, Cyi
for all x ∈ H and y ∈ K. Also, kBk = kCk = kφk.

Hint for proof . Show that for every x ∈ H the map y 7→ φ(x, y) is a bounded linear functional
on K. Use the Riesz-Fréchet theorem (2.2.23).

The next proposition provides an entirely satisfactory extension of proposition 1.2.32 to the
infinite dimensional setting: adjoints of Hilbert space operators always exist.
5.1.6. Proposition. Let A : H → K be a bounded linear map between Hilbert spaces. The mapping
(x, y) 7→ hAx, yi from H × K into C is a bounded sesquilinear functional. Then there exists an
unique bounded linear map A∗ : K → H such that
hAx, yi = hx, A∗ yi
for all x ∈ H and y ∈ K. This is the adjoint of A (see definition 1.2.31). Also kA∗ k = kAk.
43
44 5. C ∗ -ALGEBRAS

5.1.7. Example. If S is the unilateral shift operator on l2 (see example 2.2.15)

S : l2 → l2 : (x1 , x2 , x3 , . . . ) 7→ (0, x1 , x2 , . . . ),
then its adjoint is given by
S ∗ : l2 → l2 : (x1 , x2 , x3 , . . . ) 7→ (x2 , x3 , x4 , . . . ).
Example 2.2.10 dealt with multiplication operators on C(X) where X is a compact Hausdorff
space. Multiplication operators can be defined on many spaces; there is nothing special in this
respect about C(X). For example, we can just as well consider multiplication operators on spaces
of square integrable functions.
5.1.8. Example. Let (S, A, µ) be a sigma-finite measure space and L2 (S) be the Hilbert space
of all (equivalence classes of) complex valued functions on S which are square integrable with
respect to µ. Let φ be an essentially bounded complex valued µ-measurable function on S. Define
Mφ on L2 (S) by Mφ (f ) := φf. Then Mφ is an operator on L2 (S); it is called a multiplication
operator. Its norm is given by kMφ k = kφk∞ and its adjoint by Mφ ∗ = Mφ .
5.1.9. Example. Let (S, A, µ) be a sigma-finite measure space and L2 (X) be the Hilbert space of
all (equivalence classes of) complex valued functions on S which are square integrable with respect
to µ. Let K be the integral operator whose kernel is the function k : S × S → C which is square
integrable with respect to the product measure µ × µ on S × S (see example 2.2.17). The adjoint
K ∗ of K is also an integral operator and its kernel is the function k ∗ defined by k ∗ (x, y) = k(y, x).
5.1.10. Proposition. Let A and B be operators on a Hilbert space H and α ∈ C. Then
(i) (A + B)∗ = A∗ + B ∗ ;
(ii) (αA)∗ = αA∗ ;
(iii) A∗∗ = A; and
(iv) (AB)∗ = B ∗ A∗ .
5.1.11. Proposition. Let A be an operator on a Hilbert space. Then
kA∗ Ak = kAk2 .
We now give an infinite dimensional generalization of the fundamental theorem of linear alge-
bra 1.2.34. We cannot expect 1.2.34 to hold exactly in the infinite dimensional case because the
range of an operator is not necessarily closed, while orthogonal complements of subspaces always
are. (An example of an operator with non-closed range is given in 10.2.5.)
5.1.12. Theorem. If A is an operator on a Hilbert space, then
(i) ker A∗ = (ran A)⊥ , and
(ii) ran A∗ = (ker A)⊥ .
5.1.13. Proposition. An operator A on a Hilbert space H is self-adjoint if and only if hAx, xi ∈ R
for every x ∈ H.
5.1.14. Proposition. The pair of maps H 7→ H and A 7→ A∗ taking every Hilbert space to itself
and every bounded linear map between Hilbert spaces to its adjoint is a contravariant functor from
the category HIL to itself.

5.2. Algebras with Involution

5.2.1. Definition. An involution on an algebra A is a map x 7→ x∗ from A into A which satisfies
(i) (x + y)∗ = x∗ + y ∗ ,
(ii) (αx)∗ = α x∗ ,
(iii) x∗∗ = x, and
(iv) (xy)∗ = y ∗ x∗
 5.2. ALGEBRAS WITH INVOLUTION 45

for all x, y ∈ A and α ∈ C. An algebra on which an involution has been defined is a ∗ -algebra
(pronounced “star algebra”). An algebra homomorphism φ between ∗ -algebras which preserves
involution (that is, such that φ(a∗ ) = (φ(a))∗ ) is a ∗ -homomorphism (pronounced “star ho-
momorphism”. A ∗ -homomorphism φ : A → B between unital algebras is said to be unital if
φ(1A ) = 1B . In the category of ∗ -algebras and ∗ -homomorphisms, the isomorphisms (called for
emphasis ∗ -isomorphisms) are the bijective ∗ -homomorphisms.
5.2.2. Example. In the algebra C of complex numbers the map z 7→ z of a number to its complex
conjugate is an involution.
5.2.3. Example. The map of an n × n matrix to its conjugate transpose is an involution on the
unital algebra Mn (see example 2.3.14).
5.2.4. Example. Let X be a compact Hausdorff space. The map f 7→ f of a function to its
complex conjugate is an involution in the algebra C(X).
5.2.5. Example. The map T 7→ T ∗ of a Hilbert space operator to its adjoint is an involution in
the algebra B(H) (see proposition 5.1.10).
5.2.6. Proposition. Let a and b be elements of a ∗ -algebra. Then a commutes with b if and only
if a∗ commutes with b∗ .
5.2.7. Proposition. In a unital ∗ -algebra 1∗ = 1.
5.2.8. Proposition. If a ∗ -algebra A has a left multiplicative identity e, then A is unital and
e = 1A .
5.2.9. Proposition. Let a be an element of a unital ∗ -algebra. Then a∗ is invertible if and only
if a is. And when a is invertible we have

∗
(a∗ )−1 = a−1 .
5.2.10. Proposition. Let a be an element of a unital ∗ -algebra. Then λ ∈ σ(a) if and only if
λ ∈ σ(a∗ ).
5.2.11. Corollary. For every element a of a ∗ -algebra ρ(a∗ ) = ρ(a).
5.2.12. Definition. An element a of a ∗ -algebra A is self-adjoint (or Hermitian) if a∗ = a. It
is normal if a∗ a = aa∗ . And it is unitary if a∗ a = aa∗ = 1. The set of all self-adjoint elements of
A is denoted by H(A), the set of all normal elements by N(A), and the set of all unitary elements
by U(A).
5.2.13. Proposition. Let a be an element of a ∗ -algebra. Then there exist unique self-adjoint
elements u and v such that a = u + iv.
Hint for proof . Think of the special case of writing a complex number in terms of its real and
imaginary parts.
5.2.14. Definition. Let S be a subset of a ∗ -algebra A. Then S ∗ = {s∗ : s ∈ S}. The subset
S is self-adjoint if S ∗ = S.
A nonempty self-adjoint subalgebra of A is a ∗ -subalgebra (or a sub-∗ -algebra).
CAUTION. The preceding definition does not say that the elements of a self-adjoint subset of a
∗ -algebra are themselves self-adjoint.
5.2.15. Definition. In an algebra A with involution a ∗ -ideal is a self-adjoint ideal in A.
5.2.16. Proposition. Let φ : A → B be a ∗ -homomorphism between ∗ -algebras. Then the kernel
of φ is a ∗ -ideal in A and the range of φ is a ∗ -subalgebra of B.
5.2.17. Proposition. If J is a ∗ -ideal in a ∗ -algebra A, then defining [a]∗ = [a∗ ] for each a ∈ A
makes the quotient A/J into a ∗ -algebra and the quotient map a 7→ [a] is a ∗ -homomorphism. The
∗ -algebra A/J is,of course, the quotient of A by J.
46 5. C ∗ -ALGEBRAS

5.3. C ∗ -Algebras
5.3.1. Definition. A C ∗ -algebra is a Banach algebra A with involution which satisfies
ka∗ ak = kak2
for every a ∈ A. This property of the norm is usually referred to as the C ∗ -condition. An algebra
norm satisfying this condition is a C ∗ -norm. A C ∗ -subalgebra of a C ∗ -algebra A is a closed
∗ -subalgebra of A.
5.3.2. Example. The vector space C of complex numbers with the usual multiplication of complex
numbers and complex conjugation z 7→ z as involution is a unital commutative C ∗ -algebra.
5.3.3. Example. If X is a compact Hausdorff space, the algebra C(X) of continuous complex
valued functions on X is a unital commutative C ∗ -algebra when involution is taken to be complex
conjugation.
5.3.4. Example. If X is a locally compact Hausdorff space, the Banach algebra C0 (X) = C0 (X, C)
of continuous complex valued functions on X which vanish at infinity is a (not necessarily unital)
commutative C ∗ -algebra when involution is taken to be complex conjugation.
5.3.5. Example. If (X, µ) is a measure space, the algebra L∞ (X, µ) of essentially bounded measur-
able complex valued functions on X (again with complex conjugation as involution) is a C ∗ -algebra.
(Technically, of course, the members of L∞ (X, µ) are equivalence classes of functions which differ
on sets of measure zero.)
5.3.6. Example. The algebra B(H) of bounded linear operators on a Hilbert space H is a unital
C ∗ -algebra when addition and scalar multiplication of operators are defined pointwise, composition
is taken as multiplication, the map T 7→ T ∗ , which takes an operator to its adjoint, is the involution,
and the norm is the usual operator norm. (See proposition 5.1.11 for the crucial C ∗ -property
kT ∗ T k = kT k2 .)
5.3.7. Example. A special case of the preceding example is the set Mn of n × n matrices of
complex numbers. We saw in example 5.2.3 that Mn is a unital algebra with involution. To make
it into a C ∗ -algebra simply identify each matrix with the (necessarily bounded) linear operator in
B(Cn ) which it represents.
5.3.8. Proposition. In every C ∗ -algebra involution is an isometry. That is, ka∗ k = kak for every
element a in the algebra.
In definition 3.1.1 of normed algebra we made the special requirement that the identity element
of a unital normed algebra have norm one. In C ∗ -algebras this requirement is redundant.
5.3.9. Corollary. In a unital C ∗ -algebra k1k = 1.
5.3.10. Corollary. Every unitary element in a unital C ∗ -algebra has norm one.
5.3.11. Corollary. If a is an element of a C ∗ -algebra A such that ab = 0 for every b ∈ A, then
a = 0.
5.3.12. Proposition. Let a be a normal element of a unital C ∗ -algebra. Then ka2 k = kak2 and
therefore ρ(a) = kak.
5.3.13. Corollary. Let A be a unital commutative C ∗ -algebra. Then ka2 k = kak2 and ρ(a) = kak
for every a ∈ A.
5.3.14. Corollary. On a unital commutative C ∗ -algebra A the Gelfand transform Γ is an isometry;
that is, kΓa ku = kb
aku = kak for every a ∈ A.
5.3.15. Corollary. The norm of a unital C ∗ -algebra is unique in the sense that given a unital
algebra A with involution there is at most one norm which makes A into a C ∗ -algebra.
 5.4. THE GELFAND-NAIMARK THEOREM—VERSION I 47

5.3.16. Proposition. If h is a self-adjoint element of a unital C ∗ -algebra, then σ(h) ⊆ R.

5.3.17. Proposition. If u is a unitary element of a unital C ∗ -algebra, then σ(u) ⊆ T.
5.3.18. Proposition. If a is an element in a C ∗ -algebra, then
kak = sup{kxak : kxk ≤ 1}
= sup{kaxk : kxk ≤ 1}.
5.3.19. Corollary. If a is an element of a C ∗ -algebra A, the operator La , called left multipli-
cation by a and defined by
La : A → A : x 7→ ax
is a (bounded linear) operator on A. Furthermore, the map
L : A → B(A) : a 7→ La
is both an isometry and an algebra homomorphism.
5.3.20. Proposition. The closure of a ∗-subalgebra of a C ∗ -algebra A is a C ∗ -subalgebra of A.
5.3.21. Definition. Let S be a nonempty subset of a C ∗ -algebra A. The intersection of the family
of all C ∗ -subalgebras of A which contain S is the C ∗ -subalgebra generated by S. We denote
it by C ∗ (S). (It is easy to see that the intersection of a family of C ∗ -subalgebras really is a C ∗ -
algebra.) In some cases we shorten the notation slightly: for example, if a ∈ A we write C ∗ (a)
for C ∗ ({a}).
5.3.22. Proposition. Let S be a nonempty subset of a C ∗ -algebra A. For each natural number n
elements a of A for which there exist x1 , x2 , . . . , xn in S ∪ S ∗
define the set Wn to be the set of all S
such that a = x1 x2 · · · xn . Let W = ∞ n=1 Wn . Then
C ∗ (S) = span W .

5.4. The Gelfand-Naimark Theorem—Version I

5.4.1. Notation. The category CSA comprises C ∗ -algebras as objects and ∗-homomorphisms as
morphisms. Closely related, of course, is the category UCSA of unital C ∗ -algebras and unital
∗-homomorphisms.
In the categories CSA and UCSA every bijective morphism is an isomorphism. We will write
∗
A∼
= B to indicate that C ∗ -algebras A and B are ∗ -isomorphic.
5.4.2. Proposition. A ∗ -homomorphism φ : A → B between C ∗ -algebras is a ∗ -isomorphism if
and only if it is bijective.
5.4.3. Proposition. Let A be a unital C ∗ -algebra. If a ∈ A is self-adjoint, then its Gelfand
transform b
a is real valued.
5.4.4. Proposition. On a unital C ∗ -algebra A every character preserves involution, thus the
Gelfand transform ΓA is a ∗-homomorphism.
5.4.5. Theorem (Stone-Weierstrass Theorem). Let X be a compact Hausdorff space. Every unital
separating ∗ -subalgebra of C(X) is dense.
Proof. See [9], theorem 2.40.
The first version of the Gelfand-Naimark theorem says that any unital commutative C ∗ -algebra
is the algebra of all continuous functions on some compact Hausdorff space. Of course the word
is in the preceding sentence means is isometrically ∗ -isomorphic to. The compact Hausdorff space
referred to is the character space of the algebra.
5.4.6. Theorem (Gelfand-Naimark Theorem I). Let A be a unital commutative C ∗ -algebra. Then
the Gelfand transform ΓA : a 7→ b
a is an isometric ∗ -isomorphism of A onto C(∆A).
48 5. C ∗ -ALGEBRAS

5.4.7. Theorem (Abstract Spectral Theorem). If a is a normal element of a unital C ∗ -algebra A,

then the C ∗ -algebra C(σ(a)) is isometrically ∗ -isomorphic to C ∗ (1A , a).
Hint for proof . Use the Gelfand transform of a to identify the maximal ideal space of C ∗ (1A , a)
with the spectrum of a. Apply the functor C. Compose the resulting map with Γ−1 where Γ is the
Gelfand transform on the C ∗ -algebra C ∗ (1A , a).
5.4.8. Example. Suppose that in the preceding theorem ψ : C(σ(a)) → C ∗ (1, a) implements the
isometric ∗ -isomorphism. Then the image under ψ of the constant function 1 on the spectrum of a
is 1A and the image under ψ of the identity function λ 7→ λ on the spectrum of a is a.
5.4.9. Example. Let T be a normal operator on a Hilbert space H whose spectrum is contained
in [0, ∞). Suppose that ψ : C(σ(T )) → C ∗ (I, T ) implements
√ the isometric ∗ -isomorphism between
these two C ∗ -algebras. Then there is at least one operator T whose square is T . Indeed, whenever
f is a continuous function on the spectrum of a normal operator T , we may meaningfully speak of
the operator f (T ).
5.4.10. Proposition (Spectral mapping theorem). Let a be a self-adjoint element in a unital
C ∗ -algebra A and f be a continuous complex valued function on the spectrum of a. Then
σ(f (a)) = f → (σ(a)) .
 CHAPTER 6

SURVIVAL WITHOUT IDENTITY

6.1. Unitization of Banach Algebras

6.1.1. Definition. Let A be an algebra. Define A ./ C to be the set A × C on which addition and
scalar multiplication are defined pointwise and multiplication is defined by
(a, λ) · (b, µ) = (ab + µa + λb, λµ).
If the algebra A is equipped with an involution, define an involution on A ./ C pointwise; that is,
(a, λ)∗ := (a∗ , λ). (The notation A ./ C is not standard.)
6.1.2. Proposition. If A is an algebra, then A ./ C is a unital algebra in which A is embedded as
(that is, isomorphic to) an ideal such that (A ./ C)/A ∼ = C. The identity of A ./ C is (0, 1). If A
is a ∗ -algebra, then A ./ C is a unital ∗ -algebra in which A is a ∗ -ideal such that (A ./ C)/A ∼
= C.
6.1.3. Definition. The algebra A ./ C is the unitization of the algebra (or ∗ -algebra) A.
6.1.4. Notation. In the preceding construction elements of the algebra A ./ C are technically
ordered pairs (a, λ). They are usually written differently. Let ι : A → A ./ C : a 7→ (a, 0) and
π : A ./ C → C : (a, λ) 7→ λ. It follows, since (0, 1) is the identity in A ./ C, that
(a, λ) = (a, 0) + (0, λ)
= ι(a) + λ1A./C
It is conventional to treat ι as an inclusion mapping. Thus it is reasonable to write (a, λ) as
a + λ1A./C or simply as a + λ1. No ambiguity seems to follow from omitting reference to the
multiplicative identity, so a standard notation for the pair (a, λ) is a + λ.
6.1.5. Definition. Let A be a nonunital algebra. Define the spectrum of an element a ∈ A to
be the spectrum of a regarded as an element of A ./ C; that is, σA (a) := σA./C (a).
6.1.6. Definition. Let A be a normed algebra (with or without involution). On the unitization
A ./ C of A define k(a, λ)k := kak + |λ|.
6.1.7. Proposition. Let A be a normed algebra. The mapping (a, λ) 7→ kak + |λ| defined above is
a norm under which A ./ C is a normed algebra. If A is a Banach algebra (respectively, a Banach
∗ -algebra), then A ./ C is a Banach algebra (respectively, a Banach ∗ -algebra). The resulting
Banach algebra (or Banach ∗ -algebra) is the unitization of A and will be denoted by Ae .
With this expanded definition of spectrum many of the earlier facts for unital Banach algebras
remain true in the more general setting. In particular, for future reference we restate items 3.1.18,
3.1.25, 3.1.29, and 3.1.30.
6.1.8. Proposition. Let a be an element of a Banach algebra A. Then the spectrum of a is compact
and |λ| ≤ kak for every λ ∈ σ(a).
6.1.9. Proposition. The spectrum of every element of a Banach algebra is nonempty.

n
6.1.10. Proposition. For every element a of a Banach algebra ρ(a) ≤ kak and ρ(an ) = ρ(a) .
6.1.11. Theorem (Spectral radius formula). If a is an element of a Banach algebra, then
ρ(a) = inf kan k1/n : n ∈ N = lim kan k1/n .

n→∞

49
50 6. SURVIVAL WITHOUT IDENTITY

6.1.12. Definition. Let A be an algebra. A left ideal J in A is a modular left ideal if there
exists an element u in A such that au − a ∈ J for every a ∈ A. Such an element u is called a
right identity with respect to J. Similarly, a right ideal J in A is a modular right ideal
if there exists an element v in A such that va − a ∈ J for every a ∈ A. Such an element v is called
a left identity with respect to J. A two-sided ideal J is a modular ideal if there exists
an element e which is both a left and a right identity with respect to J.
6.1.13. Proposition. An ideal J in an algebra is modular if and only if it is both left modular and
right modular.
Hint for proof . Show that if u is a right identity with respect to J and v is a left identity with
respect to J, then vu is both a right and left identity with respect to J.
6.1.14. Proposition. An ideal J in an algebra A is modular if and only if the quotient algebra
A/J is unital.
6.1.15. Example. Let X be a locally compact Hausdorff space. For every x ∈ X the ideal Jx is
a maximal modular ideal in the C ∗ -algebra C0 (X) of continuous complex valued functions on X.
Proof. By the locally compact Hausdorff space version of Urysohn’s lemma (see, for example,
[11], theorem 17.2.10) there exists a function g ∈ C0 (X) such that g(x) = 1. Thus Jx is mod-
ular because g is an identity with respect to Jx . Since C0 (X) = Jx ⊕ span{g} the ideal Jx has
codimension 1 and is therefore maximal. 
6.1.16. Proposition. If J is a proper modular ideal in a Banach algebra, then so is its closure.
6.1.17. Corollary. Every maximal modular ideal in a Banach algebra is closed.
6.1.18. Proposition. Every proper modular ideal in a Banach algebra is contained in a maximal
modular ideal.
6.1.19. Proposition. Let A be a commutative Banach algebra and φ ∈ ∆A. Then ker φ is a
maximal modular ideal in A and A/ ker φ is a field. Furthermore, every maximal modular ideal is
the kernel of exactly one character in ∆A.
6.1.20. Proposition. Every multiplicative linear functional on a commutative Banach algebra A
is continuous. In fact, every character is contractive.
6.1.21. Example. Let Ae be the unitization of a commutative Banach algebra A (see proposi-
tion 6.1.7). Define
φ∞ : Ae → C : (a, λ) 7→ λ .
Then φ∞ is a character on Ae .
6.1.22. Proposition. Every character φ on a commutative Banach algebra A has a unique exten-
sion to a character φe on the unitization Ae of A. And the restriction to A of any character on
Ae , with the obvious exception of φ∞ , is a character on A.
6.1.23. Proposition. If A is a commutative Banach algebra, then
(a) ∆A is a locally compact Hausdorff space,
(b) ∆Ae = ∆A ∪ {φ∞ },
(c) ∆Ae is the one-point compactification of ∆A, and
(d) the map φ 7→ φe is a homeomorphism from ∆A onto ∆Ae \ {φ∞ }.
If A is unital (so that ∆A is compact), then φ∞ is an isolated point of ∆Ae .
6.1.24. Theorem. If A is a commutative Banach algebra with a nonempty character space, then
the Gelfand transform
Γ = ΓA : C0 (∆A) : a 7→ ba = Γa
is a contractive algebra homomorphism and ρ(a) = kb aku . Furthermore, if A is not unital, then
a ∪ {0} for every a ∈ A.
σ(a) = ran b
 6.2. EXACT SEQUENCES AND EXTENSIONS 51

6.2. Exact Sequences and Extensions

The rather simple procedure for the unitization of Banach ∗ -algebras (see 6.1.1 and 6.1.6)
does not carry over to C ∗ -algebras. The norm defined in 6.1.6 does not satisfy the C ∗ -condition
(definition 5.3.1). It turns out that the unitization of C ∗ -algebras is a bit more complicated. Before
examining the details we look at some preparatory material on exact sequences and extensions of
C ∗ -algebras.
6.2.1. Definition. A sequence of C ∗ -algebras and ∗ -homomorphisms
φn+1
··· / An−1 φn / An / An+1 / ···

is said to be exact at An if ran φn = ker φn+1 . A sequence is exact if it is exact at each of its
constituent C ∗ -algebras. A sequence of C ∗ -algebras and ∗ -homomorphisms of the form
φ ψ
0 /A /E /B /0 (6.1)
is a short exact sequence. (Here 0 denotes the trivial 0-dimensional C ∗ -algebra, and the
unlabeled arrows are the obvious ∗-homomorphisms.) The short exact sequence of C ∗ -algebras (6.1)
is split exact if there exists a ∗ -homomorphism ξ : B → E such that ψ ◦ ξ = idB .
The preceding definitions were for the category CSA of C ∗ -algebras and ∗ -homomorphisms.
Of course there is nothing special about this particular category. Exact sequences make sense in
many situations, in, for example, various categories of Banach spaces, Banach algebras, Hilbert
spaces, vector spaces, Abelian groups, modules, and so on.
Often in the context of C ∗ -algebras the exact sequence (6.1) is referred to as an extension.
Some authors refer to it as an extension of A by B (for example, [29] and [7]) while others say it
is an extension of B by A ([20], [12], and [3]). In [12] and [3] the extension is defined to be the
sequence 6.1; in [29] it is defined to be the ordered triple (φ, E, ψ); and in [20] and [7] it is defined
to be the C ∗ -algebra E itself. Regardless of the formal definitions it is common to say that E is
an extension of A by B (or of B by A).
6.2.2. Convention. In a C ∗ -algebra the word ideal will always mean a closed two-sided ∗ -ideal
(unless, of course, the contrary is explicitly stated). We will show shortly (in proposition 6.6.6)
that requiring an ideal in a C ∗ -algebra to be self-adjoint is redundant. A two-sided algebra ideal
of a C ∗ -algebra which is not necessarily closed will be called an algebraic ideal. A self-adjoint
algebraic ideal of a C ∗ -algebra will be called an algebraic ∗-ideal.
6.2.3. Proposition. The kernel of a ∗ -homomorphism φ : A → B between C ∗ -algebras is an ideal
in A and its range is a ∗ -subalgebra of B.
6.2.4. Proposition. Consider the following diagram in the category of C ∗ -algebras and ∗ -homomorphisms
j
0 /A /B k /C /0

f g h
  
0 / A0 / B0 / C0 /0
j0 k0

If the rows are exact and the left square commutes, then there exists a unique ∗ -homomorphism
h : C → C 0 which makes the right square commute.
6.2.5. Proposition (The Five Lemma). Suppose that in the following diagram of C ∗ -algebras and
∗ -homomorphisms
j
0 /A /B k /C /0

f g h
  
0 / A0 / B0 / C0 /0
j0 k0
52 6. SURVIVAL WITHOUT IDENTITY

the rows are exact and the squares commute. Prove the following.
(1) If g is surjective, so is h.
(2) If f is surjective and g is injective, then h is injective.
(3) If f and h are surjective, so is g.
(4) If f and h are injective, so is g.
6.2.6. Definition. Let A and B be C ∗ -algebras. We define the (external) direct sum of A
and B, denoted by A ⊕ B, to be the Cartesian product A × B with pointwise defined algebraic
operations and norm given by
k(a, b)k = max{kak, kbk}
for all a ∈ A and b ∈ B. An alternative notation for the element (a, b) in A ⊕ B is a ⊕ b.
6.2.7. Example. Let A and B be C ∗ -algebras. Then the direct sum of A and B is a C ∗ -algebra
and the following sequence is split short exact:
ι1 π2
0 /A / A⊕B /B /0

The indicated maps in the preceding are the obvious ones:

ι1 : A → A ⊕ B : a 7→ (a, 0) and π2 : A ⊕ B → B : (a, b) 7→ b .
This is the direct sum extension.
6.2.8. Proposition. If A and B are nonzero C ∗ -algebras, then their direct sum A ⊕ B is a product
in the category CSA of C ∗ -algebras and ∗ -homomorphisms (see example 4.4.11). The direct sum
is unital if and only if both A and B are.
6.2.9. Definition. Let A and B be C ∗ -algebras and E and E 0 be extensions of A by B. These
extensions are strongly equivalent if there exists a ∗ -isomorphism θ : E → E 0 that makes the
diagram
0 /A /E /B /0

θ

0 /A / E0 /B /0

commute.
6.2.10. Proposition. In the preceding definition it is enough to require θ to be a ∗ -homomorphism.
6.2.11. Proposition. Let A and B be C ∗ -algebras. An extension
φ ψ
0 /A /E /B /0

is strongly equivalent to the direct sum extension A⊕B if and only if there exists a ∗ -homomorphism
ν : E → A such that ν ◦ φ = idA .
6.2.12. Proposition. If the sequences of C ∗ -algebras
φ ψ
0 /A /E /B /0 (6.2)
and
φ0 ψ0
0 /A / E0 /B /0 (6.3)
are strongly equivalent and (6.2) splits, then so does (6.3).
 6.3. UNITIZATION OF C ∗ -ALGEBRAS 53

6.3. Unitization of C ∗ -algebras

6.3.1. Proposition. Let A be a C ∗ -algebra. Then there exists a unital C ∗ -algebra A
e in which A
is embedded as an ideal such that the sequence
0 /A /A
e /C /0 (6.4)
is split exact. If A is unital then the sequence (6.4) is strongly equivalent to the direct sum extension,
so that A e∼= A ⊕ C. If A is not unital, then A e is not isomorphic to A ⊕ C.

Hint for proof . The proof this result is a little complicated. Everyone should go through all
the details at least once in his/her life. What follows is an outline of a proof.
Notice that we speak of the unitization of C ∗ -algebra A whether or not A already has a unit
(multiplicative identity). We divide the argument into two cases.
Case 1: the algebra A is unital.
(1) On the algebra A ./ C define
k(a, λ)k := max{ka + λ1A k, |λ|}
and let A
e := A ./ C together with this norm.
(2) Prove that the map (a, λ) 7→ k(a, λ)k is a norm on A ./ C.
(3) Prove that this norm is an algebra norm.
(4) Show that it is, in fact, a C∗-norm on A ./ C.
(5) Observe that it is an extension of the norm on A.
(6) Prove that A ./ C is a C ∗ -algebra by verifying completeness of the metric space induced
by the preceding norm.
(7) Prove that the sequence
Q
/A ι /A
eo / /0
0 C
ψ

is split exact (where ι : a 7→ (a, 0), Q : (a, λ) 7→ λ, and ψ : λ 7→ (0, λ)).

∗
(8) Prove that A e∼= A ⊕ C.

Case 2: the algebra A is not unital.

(9) Prove the following simple fact.
6.3.2. Lemma. Let A be an algebra and B be a normed algebra. If φ : A → B is an algebra
homomorphism, the function a 7→ kak := kφ(a)k is a seminorm on A. The function is a
norm if φ is injective.
(10) Recall that we defined the operator La , left multiplication by a, in 5.3.19. Now let
A] := {La + λIA ∈ B(A) : a ∈ A and λ ∈ C}
and show that A] is a normed algebra.
(11) Make A] into a ∗ -algebra by defining
(La + λIA )∗ := La∗ + λIA
for all a ∈ A and λ ∈ C.
(12) Define
φ : A ./ C → A] : (a, λ) 7→ La + λIA
and verify that φ is a ∗ -homomorphism.
(13) Prove that φ is injective.
(14) Use (9) to endow A ./ C with a norm which makes it into a unital normed algebra. Let
Ae := A ./ C with the norm pulled back by φ from A]
54 6. SURVIVAL WITHOUT IDENTITY

(15) Verify the following facts.

(a) The map φ : A e → A] is an isometric isomorphism.
(b) ran L is a closed subalgebra of ran φ = A] ⊆ B(A).
(c) IA ∈/ ran L.
(16) Prove that the norm on A e satisfies the C ∗ -condition.
(17) Prove that A e is a unital C ∗ -algebra. (To show that A e is complete we need only show

∞
that A is complete. To this end let φ(an , λn ) n=1 be a Cauchy sequence in A] . To
]

show that this sequence converges it suffices to show that it has a convergent subsequence.
Showing that the
sequence (λn ) is bounded allows us to extract from it a convergent
subsequence λnk . Prove that (Lank ) converges.)
(18) Prove that the sequence

0 /A /A
e /C /0 (6.5)
is split exact.
(19) The C ∗ -algebra A
e is not equivalent to A ⊕ C.

6.3.3. Definition. The C ∗ -algebra A

e constructed in the preceding proposition is the unitization
of A.
Note that the expanded definition of spectrum given in 6.1.5 applies to C ∗ -algebras since the
added identity is purely an algebraic matter and is the same for C ∗ -algebras as it is for general
Banach algebras. Thus many of the earlier facts stated for unital C ∗ -algebras remain true. In
particular, for future reference we restate items 5.3.12, 5.3.13, 5.3.14, 5.3.15, 5.3.16, 5.4.3, and 5.4.4.
6.3.4. Proposition. Let a be a normal element of a C ∗ -algebra. Then ka2 k = kak2 and therefore
ρ(a) = kak.
6.3.5. Corollary. If A is a commutative C ∗ -algebra, then ka2 k = kak2 and ρ(a) = kak for every
a ∈ A.
6.3.6. Corollary. On a commutative C ∗ -algebra A the Gelfand transform Γ is an isometry; that
is, kΓa ku = kb
aku = kak for every a ∈ A.
6.3.7. Corollary. The norm of a C ∗ -algebra is unique in the sense that given a algebra A with
involution there is at most one norm which makes A into a C ∗ -algebra.
6.3.8. Proposition. If h is a self-adjoint element of a C ∗ -algebra, then σ(h) ⊆ R.
6.3.9. Proposition. If a is a self-adjoint element in a C ∗ -algebra, then its Gelfand transform b
a
is real valued.
6.3.10. Proposition. Every character on a C ∗ -algebra A preserves involution, thus the Gelfand
transform ΓA is a ∗-homomorphism.
An immediate result of the preceding results is the second version of the Gelfand-Naimark
theorem, which says that any commutative C ∗ -algebra is (isometrically unitally ∗ -isomorphic to)
the algebra of all those continuous functions on some locally compact Hausdorff space which vanish
at infinity. As was the case with the first version of this theorem (see 5.4.6) the locally compact
Hausdorff space referred to is the character space of the algebra.
6.3.11. Theorem (Gelfand-Naimark Theorem II). Let A be a commutative C ∗ -algebra. Then the
Gelfand transform ΓA : a 7→ b
a is an isometric unital ∗ -isomorphism of A onto C0 (∆A).
It follows from the next proposition that the unitization process is functorial.
6.3.12. Proposition. Every ∗-homomorphism φ : A → B between C ∗ -algebras has a unique exten-
sion to a unital ∗-homomorphism φe : A
e→B
e between their unitizations.
 6.4. QUASI-INVERSES 55

In contrast to the situation in general Banach algebras there is no distinction between topo-
logical and geometric categories of C ∗ -algebras. One of the most remarkable aspects of C ∗ -algebra
theory is that ∗-homomorphisms between such algebras are automatically continuous—in fact, con-
tractive. It follows that if two C ∗ -algebras are algebraically ∗-isomorphic, then they are isometrically
isomorphic.
6.3.13. Proposition. Every ∗-homomorphism between C ∗ -algebras is contractive.
6.3.14. Proposition. Every injective ∗-homomorphism between C ∗ -algebras is an isometry.
e = X ∪ {∞} be its one-
6.3.15. Proposition. Let X be a locally compact Hausdorff space and X
point compactification. Define
ι : C0 (X) → C(X)
e : f 7→ fe
where 
f (x), if x ∈ X;
fe(x) =
0, if x = ∞.
Also let E∞ be defined on C(X)
e by E∞ (g) = g(∞). Then the sequence
E∞
0 / C0 (X) ι / C(X)
e /C /0

is exact.
In the preceding proposition we refer to X e as the one-point compactification of X even in the
case that X is compact to begin with. Most definitions of compactification require a space to be
dense in any compactification. (See my remarks in the beginning of section 17.3 of [11].) We have
previously adopted the convention that the unitization of a unital algebra gets a new multiplicative
identity. In the spirit of consistency with this choice we will in the sequel subscribe to the convention
that the one-point compactification of a compact space gets an additional (isolated) point. (See
also the terminology introduced in 9.3.1.)
From the point of view of the Gelfand-Naimark theorem (6.3.11) the fundamental insight
prompted by the next proposition is that the unitization of a commutative C ∗ -algebra is, in some
sense, the “same thing” as the one-point compactification of a locally compact Hausdorff space.
6.3.16.
Proposition. If X is a locally compact Hausdorff space, then the unital C ∗ -algebras
∼
C0 (X) and C(X)e are isometrically ∗ -isomorphic.

Proof. Define
∼
θ : C0 (X) → C(X) e : (f, λ) 7→ fe + λ1 e
X
(where 1Xe is the constant function 1 on X). Then consider the diagram
e

/ C0 (X) / C0 (X) ∼ /C /0


0

θ

0 / C0 (X) ι e E∞
/ C(X) /C /0

The top row is exact by proposition 6.3.1, the bottom row is exact by proposition 6.3.15, and the
diagram obviously commutes. It is routine to check that θ is a ∗ -homomorphism. Therefore θ is
an isometric ∗ -isomorphism by proposition 6.2.10 and corollary 6.3.14. 

6.4. Quasi-inverses
6.4.1. Definition. An element b of an algebra A is a left quasi-inverse for a ∈ A if ba = a + b.
It is a right quasi-inverse for a if ab = a + b. If b is both a left and a right quasi-inverse for a
it is a quasi-inverse for a. When a has a quasi-inverse denote it by a0 .
56 6. SURVIVAL WITHOUT IDENTITY

6.4.2. Proposition. If b is a left quasi-inverse for a in an algebra A and c is a right quasi-inverse

for a in A, then b = c.
6.4.3. Proposition. Let A be a unital algebra and let a, b ∈ A. Then
(a) a0 exists if and only if (1 − a)−1 exists; and
(b) b−1 exists if and only if (1 − b)0 exists.
Hint for proof . For the reverse direction in (a) consider a + c − ac where c = 1 − (1 − a)−1 .
6.4.4. Proposition. An element a of a Banach algebra A is quasi-invertible if and only if it is not
an identity with respect to any maximal modular ideal in A.
Proposition. Let A be a Banach algebra and a ∈ A. If ρ(a) < 1, then a0 exists and
6.4.5. P
a =− ∞
0 k
k=1 a .

6.4.6. Proposition. Let A be a Banach algebra and a ∈ A. If kak < 1, then a0 exists and
kak kak
≤ ka0 k ≤ .
1 + kak 1 − kak
6.4.7. Proposition. Let A be a unital Banach algebra and a ∈ A. If ρ(1 − a) < 1, then a is
invertible in A.
6.4.8. Proposition. Let A be a Banach algebra and a, b ∈ A. If b0 exists and kak < (1 + kb0 k)−1 ,
then (a + b)0 exists and
kak (1 + kb0 k)2
k(a + b)0 − b0 k ≤ .
1 − kak (1 + kb0 k)
Hint for proof . Show first that u = (a − b0 a)0 exists and that u + b0 − ub0 is a left quasi-inverse
for a + b.

Compare the next result with propositions 3.1.12 and 3.1.14

6.4.9. Proposition. The set QA of quasi-invertible elements of a Banach algebra A is open in A,
and the map a 7→ a0 is a homeomorphism of QA onto itself.
6.4.10. Notation. If a and b are elements in an algebra we define a ◦ b := a + b − ab.
6.4.11. Proposition. If A is an algebra, then QA is a group under ◦.
6.4.12. Proposition. If A is a unital Banach algebra and inv A is the set of its invertible elements,
then
Ψ : QA → inv A : a 7→ 1 − a
is an isomorphism.
6.4.13. Definition. If A is a algebra and a ∈ A, we define the q-spectrum of a in A by
a
σ̆A (a) := {λ ∈ C : λ 6= 0 and ∈/ QA } ∪ {0}.
λ
In a unital algebra, the preceding definition is “almost” the usual one.
6.4.14. Proposition. Let A be a unital algebra and a ∈ A. Then for all λ 6= 0, we have λ ∈ σ(a)
if and only if λ ∈ σ̆(a).

6.4.15. Proposition. If an algebra A is not unital and a ∈ A, then σ̆A (a) = σ̆Ae(a), where A
e is
the unitization of A.
 6.5. POSITIVE ELEMENTS IN C ∗ -ALGEBRAS 57

6.5. Positive Elements in C ∗ -algebras

6.5.1. Definition. A self-adjoint element a of a C ∗ -algebra A is positive if σ(a) ⊆ [0, ∞). In this
case we write a ≥ 0. We denote the set of all positive elements of A by A+ . This is the positive
cone of A. For any subset B of A let B + = B ∩ A+ . We will use the positive cone to induce a
partial ordering on A: we write a ≤ b when b − a ∈ A+ .
6.5.2. Definition. Let ≤ be a relation on a nonempty set S. If the relation ≤ is reflexive and
transitive, it is a preordering. If ≤ is a preordering and is also antisymmetric, it is a partial
ordering.
A partial ordering ≤ on a real vector space V is compatible with (or respects) the operations
(addition and scalar multiplication) on V if for all x, y, z ∈ V
(a) x ≤ y implies x + z ≤ y + z, and
(b) x ≤ y, α ≥ 0 imply αx ≤ αy.
A real vector space equipped with a partial ordering which is compatible with the vector space
operations is an ordered vector space.
6.5.3. Definition. Let V be a vector space. A subset C of V is a cone in V if αC ⊆ C for every
α ≥ 0. A cone C in V is proper if C ∩ (−C) = {0}.
6.5.4. Proposition. A cone C in a vector space is convex if and only if C + C ⊆ C.
6.5.5. Example. If V is an ordered vector space, then the set
V + := {x ∈ V : x ≥ 0}
is a proper convex cone in V . This is the positive cone of V and its members are the positive
elements of V .
6.5.6. Proposition. Let V be a real vector space and C be a proper convex cone in V . Define
x ≤ y if y − x ∈ C. Then the relation ≤ is a partial ordering on V and is compatible with the
vector space operations on V . This relation is the partial ordering induced by the cone C.
The positive cone V + of the resulting ordered vector space is just C itself.
6.5.7. Proposition. If a is a self-adjoint element of a unital C ∗ -algebra and t ∈ R, then
(i) a ≥ 0 whenever ka − t1k ≤ t; and
(ii) ka − t1k ≤ t whenever kak ≤ t and a ≥ 0.
6.5.8. Example. The positive cone of a C ∗ -algebra A is a closed proper convex cone in the real
vector space H(A).
6.5.9. Proposition. If a and b are positive elements of a C ∗ -algebra and ab = ba, then ab is
positive.
6.5.10. Proposition. Every positive element of a C ∗ -algebra A has a unique positive nth root
(n ∈ N). That is, if a ∈ A+ , then there exists a unique b ∈ A+ such that bn = a.
Proof. Hint. The existence part is a simple application of the C ∗ -functional calculus (that
is, the abstract spectral theorem 5.4.7). The element b given by the functional calculus is positive
in the algebra C ∗ (1, a). Explain why it is also positive in A. The uniqueness argument deserves
considerable care.
6.5.11. Theorem (Jordan Decomposition). If c is a self-adjoint element of a C ∗ -algebra A, then
there exist positive elements c+ and c− of A such that c = c+ − c− and c+ c− = 0.
6.5.12. Lemma. If c is an element of a C ∗ -algebra such that −c∗ c ≥ 0, then c = 0.
6.5.13. Proposition. If a is an element of a C ∗ -algebra, then a∗ a ≥ 0.
6.5.14. Proposition. If c is an element of a C ∗ -algebra, then the following are equivalent:
58 6. SURVIVAL WITHOUT IDENTITY

(i) c ≥ 0;
(ii) there exists b ≥ 0 such that c = b2 ; and
(iii) there exists a ∈ A such that c = a∗ a,
6.5.15. Example. If T is an operator on a Hilbert space H, then T is a positive member of the
C ∗ -algebra B(H) if and only if hT x, xi ≥ 0 for all x ∈ H.
Hint for proof . Showing that if T ≥ 0 in B(H), then hT x, xi ≥ 0 for all x ∈ H is easy: use
proposition 6.5.14 to write T as S ∗ S for some operator S.
For the converse suppose that hT x, xi ≥ 0 for all x ∈ H. It is easy to see that this implies
that T is self-adjoint. Use the Jordan decomposition theorem6.5.11 to write T as T + − T − . For
arbitrary u ∈ H let x = T − u and verify that 0 ≤ hT x, xi = −h(T − )3 u, ui. Now (T − )3 is a positive
element of B(H). (Why?) Conclude that (T − )3 = 0 and therefore T − = 0. (For additional detail
see [8], page 37.)
√
6.5.16. Definition. For an arbitrary element a of a C ∗ -algebra we define |a| to be a∗ a.
6.5.17. Proposition. If a is a self-adjoint element of a C ∗ -algebra, then
|a| = a+ + a− .
6.5.18. Example. The absolute value in a C ∗ -algebra need notbe subadditive;  that is, |a + b|
1 1 0 0
need not be less than |a| + |b|. For example, in M2 (C) take a = and b = . Then
1 1 0 −2
  √ 
0 0 2 √0
|a| = a, |b| = , and |a + b| = . If |a + b| − |a| − |b| were positive, then, according
0 2 0 2
to example 6.5.15, h (|a + b| − |a| − |b|) x , x i would be positive for every vector x ∈ C2 . But this is
not true for x = (1, 0).
6.5.19. Proposition. If φ : A → B is a ∗ -homomorphism between C ∗ -algebras, then φ(a) ∈ B +
whenever a ∈ A+ . If φ is a ∗ -isomorphism, then φ(a) ∈ B + if and only if a ∈ A+ .
6.5.20. Proposition. Let a be a self-adjoint element of a C ∗ -algebra A and f a continuous complex
valued function on the spectrum of a. Then f ≥ 0 in C(σ(a)) if and only if f (a) ≥ 0 in A.
6.5.21. Proposition. If a is a self-adjoint element of a C ∗ -algebra A, then kak1A ± a ≥ 0.
6.5.22. Proposition. If a and b are self-adjoint elements of a C ∗ -algebra A and a ≤ b, then
x∗ ax ≤ x∗ bx for every x ∈ A.
6.5.23. Proposition. If a and b are elements of a C ∗ -algebra with 0 ≤ a ≤ b, then kak ≤ kbk.
6.5.24. Proposition. Let A be a unital C ∗ -algebra and c ∈ A+ . Then c is invertible if and only
if c ≥ 1 for some  > 0.
6.5.25. Proposition. Let A be a unital C ∗ -algebra and c ∈ A. If c ≥ 1, then c is invertible and
0 ≤ c−1 ≤ 1.
6.5.26. Proposition. If a is a positive invertible element in a unital C ∗ -algebra, then a−1 is
positive.
1
Next we show that the notation a− 2 is unambiguous.
1
6.5.27. Proposition. Let a ∈ A+ where A is a unital C ∗ -algebra. If a is invertible, so is a 2 and
1
−1
1
a2 = a−1 2 .
6.5.28. Proposition. Let a and b be elements of a C ∗ -algebra. If 0 ≤ a ≤ b and a is invertible,
then b is invertible and b−1 ≤ a−1 .
√ √
6.5.29. Proposition. If a and b are elements of a C ∗ -algebra and 0 ≤ a ≤ b, then a ≤ b.
 6.6. APPROXIMATE IDENTITIES 59

6.5.30. Example. Let a and b be elements of a C ∗ -algebra with 0 ≤ a ≤ b. It is not necessarily

the case that a2 ≤ b2 .
   
∗ 1 0 1 1 1
Hint for proof . In the C -algebra M2 let a = and b = a + 2 .
0 0 1 1

6.6. Approximate Identities

6.6.1. Definition. An approximate identity (or approximate unit) in a C ∗ -algebra A is an
increasing net (eλ )λ∈Λ of positive elements of A such that keλ k ≤ 1 and aeλ → a (equivalently,
eλ a → a) for every a ∈ A. If such a net is in fact a sequence, we have a sequential approximate
identity. In the literature be careful of varying definitions: many authors omit the requirements
that the net be increasing and/or that it be bounded.
6.6.2. Example. Let A = C0 (R). For each n ∈ N let Un = (−n, n) and let en : R → [0, 1] be a
function in A whose support is contained in Un+1 and such that en (x) = 1 for every x ∈ Un . Then
(en ) is a (sequential) approximate identity for A.
6.6.3. Proposition. If A is a C ∗ -algebra, then the set
Λ := {a ∈ A+ : kak < 1}
is a directed set (under the ordering it inherits from H(A)).
Hint for proof . Verify that the function
t
f : [0, 1) → R+ : t 7→ (1 − t)−1 − 1 =
1−t
+
induces an order isomorphism f : Λ → A . (An order isomorphism between partially ordered
sets is an order preserving bijection whose inverse also preserves order.) A careful proof of this
result involves checking a rather large number of details.
6.6.4. Proposition. If A is a C ∗ -algebra, then the set
Λ := {a ∈ A+ : kak < 1}
is an approximate identity for A.
6.6.5. Corollary. Every C ∗ -algebra A has an approximate identity. If A is separable then it has
a sequential approximate identity.
6.6.6. Proposition. Every closed (two-sided) algebraic ideal in a C ∗ -algebra is self-adjoint.
6.6.7. Proposition. If J is an ideal in a C ∗ -algebra A, then A/J is a C ∗ -algebra.
Proof. See [16], theorem 1.7.4, or [7], pages 13–14, or [12], theorem 2.5.4.
6.6.8. Example. If J is an ideal in a C ∗ -algebra A, then the sequence
0 /J /A π / A/J /0

is short exact.
6.6.9. Proposition. The range of a ∗-homomorphism between C ∗ -algebras is closed (and therefore
itself a C ∗ -algebra).
6.6.10. Definition. A C ∗ -subalgebra B of a C ∗ -algebra A is hereditary if a ∈ B whenever
a ∈ A, b ∈ B, and 0 ≤ a ≤ b.
6.6.11. Proposition. Suppose x∗ x ≤ a in a C ∗ -algebra A. Then there exists b ∈ A such that
1 1
x = ba 4 and kbk ≤ kak 4 .
Proof. See [7], page 13.
60 6. SURVIVAL WITHOUT IDENTITY

6.6.12. Proposition. Suppose J is an ideal in a C ∗ -algebra A, j ∈ J + , and a∗ a ≤ j. Then a ∈ J.

Thus ideals in C ∗ -algebras are hereditary.
6.6.13. Theorem. Let A and B be C ∗ -algebras and J be an ideal in A. If φ is a ∗ -homomorphism
from A to B and ker φ ⊇ J, then there exists a unique ∗ -homomorphism φe : A/J → B which makes
the following diagram commute.
A

φ
π


A/J /B
φe

Furthermore, φe is injective if and only if ker φ = J; and φe is surjective if and only if φ is.

6.6.14. Corollary. If 0 /A φ /E /B / 0 is a short exact sequence of C ∗ -algebras, then

E/ ran φ and B are isometrically ∗ -isomorphic.
6.6.15. Corollary. Every C ∗ -algebra A has codimension one in its unitization A;
e that is, dim A/A
e =
1.
6.6.16. Proposition. Let A be a C ∗ -algebra, B be a C ∗ -subalgebra of A, and J be an ideal in A.
Then
B/(B ∩ J) ∼ = (B + J)/J .
Let B be a unital subalgebra of an arbitrary algebra A. It is clear that if an element b ∈ B is
invertible in B, then it is also invertible in A. The converse turns out to be true in C ∗ -algebras: if
b is invertible in A, then its inverse lies in B. This is usually expressed by saying that every unital
C ∗ -subalgebra of a C ∗ -algebra is inverse closed.
6.6.17. Proposition. Let B be a unital C ∗ -subalgebra of a C ∗ -algebra A. If b ∈ inv(A), then
b−1 ∈ B.
6.6.18. Corollary. Let B be a unital C ∗ -subalgebra of a C ∗ -algebra A and b ∈ B. Then
σB (b) = σA (b) .
6.6.19. Corollary. Let φ : A → B be a unital C ∗ -monomorphism of a C ∗ -algebra A and a ∈ A.
Then
σ(a) = σ(φ(a)) .
 CHAPTER 7

SOME IMPORTANT CLASSES OF HILBERT SPACE

OPERATORS

7.1. Orthonormal Bases in Hilbert Spaces

7.1.1. Definition. A basis for a Hilbert space H is a maximal orthonormal set in H.
CAUTION. This notion of a basis for a Hilbert space (sometimes, for emphasis, called an or-
thonormal basis) should not be confused with the usual vector space (Hamel) basis for the un-
derlying vector space. For a Hilbert space the basis vectors are required to be of length one and
to be mutually perpendicular; but it is not necessary that every vector in the space be a linear
combination of basis vectors (but see proposition 7.1.10(d) below).
7.1.2. Proposition. Every orthonormal set in a Hilbert space H can be extended to a basis for H.
7.1.3. Corollary. Every nonzero Hilbert space has a basis.

∞
7.1.4. Example. For each n ∈ N let en = δnk k=1 be the constant zero sequence except for the nth
entry whose value is one. Then {en : n ∈ N} is a basis for the Hilbert space l2 (see example 2.1.3).
7.1.5. Example. For each integer n define the function en on [0, 2π] by
en (t) = √1
2π
eint
for 0 ≤ t ≤ 2π. Then {en : n ∈ Z} is a basis for the Hilbert space L2 ([0, 2π]) (see example 2.1.8).
7.1.6. Definition. Let V be a normed linear space and A ⊆ V . For every F ∈ Fin A define
X
sF = F.


Then s = sF F ∈Fin A is a net in V . If this net converges, the set A is said to be summable; the
P
limit of the net is the sum of A and is denoted by A. Indexed sets require a slightly different
notation. Suppose, for example, that A = {xλ : λ ∈ Λ} where Λ is an arbitrary index set. Then for
each F ∈ Fin Λ X X

sF = xλ = {xλ : λ ∈ F } .
λ∈F
As
P above s is a P net in H. If it converges {xλ : λ ∈ Λ} is summable and its sum is denoted by
λ∈Λ x λ (or by {x
Pλ : λ ∈ Λ}). An alternative way of saying
P that {xλ : λ ∈ Λ} is summable is to
say that the series λ∈Λ xλ converges or that the sum λ∈Λ xλ exists.
7.1.7. Proposition (Bessel’s inequality). If E is an orthonormal subset of a Hilbert space H and
x ∈ H, then X
|hx, ei|2 ≤ kxk2 .
e∈E

7.1.8. Corollary. If E is an orthonormal subset of a Hilbert space H and x ∈ H, then {e ∈

E : hx, ei =
6 0} is countable.
7.1.9. Proposition. If E is an orthonormal subset of a Hilbert space H and x ∈ H, then {hx, eie :
e ∈ E} is summable.
61
62 7. SOME IMPORTANT CLASSES OF HILBERT SPACE OPERATORS

7.1.10. Proposition. Let E be an orthonormal set in a Hilbert space H. Then the following are
equivalent.
(a) E is maximal (that is, E is a basis).
(b) E is total (that is, if x ⊥
W E, then x = 0).
(c) E is X
complete (that is, E = H).
(d) x = hx, eie for all x ∈ H. (Fourier expansion)
e∈E X
(e) hx, yi = hx, eihe, yi for all x, y ∈ H. (Parseval’s identity)
e∈E
X
(f) kxk2 = |hx, ei|2 for all x ∈ H. (Parseval’s identity)
e∈E

The coefficients hx, ei in the Fourier expansion of the vector x (in part (d) of proposition 7.1.10)
are the Fourier coefficients of x with respect to the basis E. It is easy to see that these
coefficients are unique.
P
7.1.11. Proposition. Let E be a basis for a Hilbert space H and x ∈ H. If x = e∈E αe e, then
αe = hx, ei for each e ∈ E.
7.1.12. Example. By writing out the Fourier expansion of the identity function f : x 7→ x in the
Hilbert space L2 ([0, 2π]) with
P∞ respect to the basis given in example 7.1.5, we demonstrate that the
1 1 2
sum of the infinite series k=1 k2 is 6 π .
7.1.13. Definition. A mapping T : V → W between vector spaces is conjugate linear if T (u +
v) = T u + T v and T (αv) = α T v for all u, v ∈ V and all α ∈ C. A bijective conjugate linear map
between vector spaces is an anti-isomorphism.
7.1.14. Example. Every inner product is conjugate linear in its second variable.
7.1.15. Example. For a Hilbert space H, the mapping ψ : H → H ∗ : a 7→ ψa defined in exam-
ple 2.2.22 is an isometric anti-isomorphism between H and its dual space.
7.1.16. Definition. A conjugate linear mapping C : V → V from a vector space into itself which
satisfies C 2 = idV is called a conjugation on V .
7.1.17. Example. Complex conjugation z 7→ z is an example of a conjugation in the vector
space C.
7.1.18.
P Example. Let H be a Hilbert space and B be a basis for H. Then the map x 7→
e∈B he, xie is a conjugation and an isometry on H.
The term “anti-isomorphism” is a bit misleading. It suggests something entirely different from
an isomorphism. In fact, an anti-isomorphism is nearly as good as an isomorphism. The next
proposition says in essence that a Hilbert space and its dual are isomorphic because they are anti-
isomorphic.
7.1.19. Proposition. Let H be a Hilbert space, ψ : H → H ∗ be the anti-isomorphism defined
in 2.2.22 (see 7.1.15), and C be the conjugation defined in 7.1.18. Then the composite Cψ −1 is an
isometric isomorphism from H ∗ onto H.
7.1.20. Corollary. Every Hilbert space is isometrically isomorphic to its dual.
7.1.21. Proposition. Let H be a Hilbert space and ψ : H → H ∗ be the anti-isomorphism defined
in 2.2.22 (see 7.1.15). If we define hf, gi := hψ −1 g, ψ −1 f i for all f , g ∈ H ∗ , then H ∗ becomes a
Hilbert space isometrically isomorphic to H. The resulting norm on H ∗ is its usual norm.
7.1.22. Corollary. Every Hilbert space is reflexive.
 7.2. PROJECTIONS AND PARTIAL ISOMETRIES 63

In example 4.2.4 we defined the adjoint of a linear operator between Banach spaces and in
proposition 5.1.6 we defined the adjoint of a Hilbert space operator. In the case of an operator on a
Hilbert space (which is also a Banach space) what is the relationship between these two “adjoints”?
They certainly are not equal since the former acts between the dual spaces and the latter between
the original spaces. In the next proposition we make use of the anti-isomorphism ψ defined in 2.2.22
(see 7.1.15) to demonstrate that the two adjoints are “essentially” the same.
7.1.23. Proposition. Let T be an operator on a Hilbert space H and ψ be the anti-isomorphism
defined in 2.2.22. If we denote (temporarily) the Banach space dual of T by T 0 : H ∗ → H ∗ , then
T 0 = ψ T ∗ ψ −1 . That is, the following diagram commutes.
T0 / H∗
HO ∗ O

ψ ψ

H /H
T∗

7.1.24. Definition. Let A be a subset of a Banach space B. Then the annihilator of A, denoted
by A⊥ , is {f ∈ B ∗ : f (a) = 0 for every a ∈ A}.
It is possible for the conflict in notations between 1.2.25 and 7.1.24 to cause confusion. To see
that the annihilator of a subspace and its orthogonal complement are “essentially” the same thing,
use the isometric anti-isomorphism ψ between H and H ∗ discussed in 7.1.15 to identify them.
7.1.25. Proposition. Let M be a subspace of
a Hilbert space H. If we (temporarily) denote the
annihilator of M by M a , then M a = ψ → M ⊥ .

7.2. Projections and Partial Isometries

7.2.1. Convention. In this and subsequent sections our attention is focused primarily on operators
on Hilbert spaces. In this context the term projection is always taken to mean orthogonal projection
(see definition 1.2.40). Thus a Hilbert space operator is called a projection if P 2 = P and P ∗ = P .
We generalize the definition of “(orthogonal) projection” from Hilbert spaces to ∗ -algebras.
7.2.2. Definition. A projection in a ∗ -algebra A is an element p of the algebra which is idem-
potent (p2 = p) and self-adjoint (p∗ = p). The set of all projections in A is denoted by P(A). In the
case of B(H), the bounded operators on a Hilbert space, we write P(H), or, if H is understood,
just P, for P(B(H)).
7.2.3. Proposition. Every operator on a Hilbert space that is an isometry on the orthogonal
complement of its kernel has closed range.
7.2.4. Proposition. Let P be a projection on a Hilbert space H. Then
(i) P x = x if and only if x ∈ ran P ;
(ii) ker P = (ran P )⊥ ; and
(iii) H = ker P ⊕ ran P .
In part (iii) of the preceding proposition the symbol ⊕ stands of course for orthogonal direct
sum (see the paragraph following 1.2.40).
7.2.5. Proposition. Let M be a subspace of a Hilbert space H. If P ∈ B(H), if P x = x for every
x ∈ M , and if P x = 0 for every x ∈ M ⊥ , then P is the projection of H onto M .
7.2.6. Proposition. Let p and q be projections in a ∗ -algebra. Then the following are equivalent:
(i) pq = 0;
(ii) qp = 0;
64 7. SOME IMPORTANT CLASSES OF HILBERT SPACE OPERATORS

(iii) qp = −pq;
(iv) p + q is a projection.
7.2.7. Definition. Let p and q be projections in a ∗ -algebra. If any of the conditions in the
preceding result holds, then we say that p and q are orthogonal and write p ⊥ q. (Thus for
operators on a Hilbert space we would correctly speak of orthogonal orthogonal projections!)
7.2.8. Proposition. Let P and Q be projections on a Hilbert space H. Then P ⊥ Q if and only
if ran P ⊥ ran Q. In this case P + Q is an orthogonal projection whose kernel is ker P ∩ ker Q and
whose range is ran P + ran Q.
7.2.9. Example. On a Hilbert space (orthogonal) projections need not commute. For example let
P be the projection of the (real) Hilbert space R2 onto the line y = x and Q be the projection of
R2 onto the x-axis. Then P Q 6= QP .
7.2.10. Proposition. Let p and q be projections in a ∗ -algebra. Then pq is a projection if and
only if pq = qp.
7.2.11. Proposition. Let P and Q be projections on a Hilbert space H. If P Q = QP , then P Q
is a projection whose kernel is ker P + ker Q and whose range is ran P ∩ ran Q.
7.2.12. Proposition. Let p and q be projections in a ∗ -algebra. Then the following are equivalent:
(i) pq = p;
(ii) qp = p;
(iii) q − p is a projection.
7.2.13. Definition. Let p and q be projections in a ∗ -algebra. If any of the conditions in the
preceding result holds, then we write p  q.
7.2.14. Proposition. Let P and Q be projections on a Hilbert space H. Then the following are
equivalent:
(i) P  Q;
(ii) kP xk ≤ kQxk for all x ∈ H; and
(iii) ran P ⊆ ran Q.
In this case Q − P is a projection whose kernel is ran P + ker Q and whose range is ran Q ran P .
Notation: Let H, M , and N be subspaces of a Hilbert space. The assertion H = M ⊕ N , may be
rewritten as M = H N (or N = H M ).
7.2.15. Proposition. The operation  defined in 7.2.13 for projections on a ∗ -algebra A is a
partial ordering on P(A). If p is a projection in A, then 0  p  1.
7.2.16. Proposition. Suppose p and q are projections on a ∗ -algebra A. If pq = qp, then the
infimum of p and q, which we denote by p f q, exists with respect to the partial ordering  and
p f q = pq. The infimum p f q may exist even when p and q do not commute. A necessary and
sufficient condition that p ⊥ q hold is that both p f q = 0 and pq = qp hold.
7.2.17. Proposition. Suppose p and q are projections on a ∗ -algebra A. If p ⊥ q, then the
supremum of p and q, which we denote by p g q, exists with respect to the partial ordering  and
p g q = p + q. The supremum p g q may exist even when p and q are not orthogonal.
7.2.18. Proposition. Let A be C ∗ -algebra, a ∈ A+ , and 0 <  ≤ 21 . If ka2 − ak < /2, then there
exists a projection p in A such that kp − ak < .
7.2.19. Definition. An element v of a C ∗ -algebra A is a partial isometry if v ∗ v is a projection
in A. (Since v ∗ v is always self-adjoint, it is enough to require that v ∗ v be idempotent.) The element
v ∗ v is the initial (or support) projection of v and vv ∗ is the final (or range) projection
of v. (It is an obvious consequence of the next proposition that if v is a partial isometry, then vv ∗
is in fact a projection.)
 7.3. FINITE RANK OPERATORS 65

7.2.20. Proposition. If v is a partial isometry in a C ∗ -algebra, then

(i) vv ∗ v = v ; and
(ii) v ∗ is a partial isometry.
Hint for proof . Let z = v − vv ∗ v and consider z ∗ z.
7.2.21. Proposition. Let v be a partial isometry in a C ∗ -algebra A. Then its initial projection p
is the smallest projection (with respect to the partial ordering  on P(A) ) such that vp = v and its
final projection q is the smallest projection such that qv = v.
7.2.22. Proposition. If V is a partial isometry on a Hilbert space H (that is, if V is a partial
isometry in the C ∗ -algebra B(H) ), then the initial projection V ∗ V is the projection of H onto
(ker V )⊥ and the final projection V V ∗ is the projection of H onto ran V .
Because of the preceding result (ker V )⊥ is called the initial (or support) space of V and
ran V is sometimes called the final space of V .)
7.2.23. Proposition. An operator on a Hilbert space is a partial isometry if and only if it is an
isometry on the orthogonal complement of its kernel.
7.2.24. Theorem (Polar Decomposition). If T is an operator on a Hilbert space H, then there
exists a partial isometry V on H such that
(i) the initial space of V is (ker T )⊥ ,
(ii) the final space of V is ran T , and
(iii) T = V |T |.
This decomposition of T is unique in the sense that if T = V0 P where V0 is a partial isometry and
P is a positive operator such that ker V0 = ker P , then P = |T | and V0 = V .
Proof. See [5], VIII.3.11; [7], theorem I.8.1; [18], theorem 6.1.2; [20], theorem 2.3.4; or [23],
theorem 3.2.17.
7.2.25. Corollary. If T is an invertible operator on a Hilbert space, then the partial isometry in
the polar decomposition (see 7.2.24) is unitary.
7.2.26. Proposition. If T is a normal operator on a Hilbert space H, then there exists a unitary
operator U on H which commutes with T and satisfies T = U |T |.
Proof. See [23], proposition 3.2.20.
7.2.27. Exercise. Let (S, A, µ) be a σ-finite measure space and φ ∈ L∞ (µ). Find the polar
decomposition of the multiplication operator Mφ (see example 5.1.8).

7.3. Finite Rank Operators

7.3.1. Definition. The rank of an operator is the dimension of its range. Thus an operator of
finite rank is one which has finite dimensional range. We denote by FR(V ) the collection of
finite rank operators on a vector space V .
7.3.2. Proposition. Let H be a Hilbert space. The family of all finite rank operators on H is a
minimal (algebraic) ∗ -ideal in the ∗ -algebra B(H).
Proof. See [9], proposition 5.5.
The next definition is the usual one for a positive operator on a Hilbert space. But recall
example 6.5.15.
7.3.3. Definition. A self-adjoint operator T on a Hilbert space H is positive if hT x, xi ≥ 0 for
every x ∈ H. In this case we may write T ≥ 0. For self-adjoint operators S and T on H (as for
elements of any C ∗ -algebra), we write S ≤ T (or T ≥ S) if T − S ≥ 0.
66 7. SOME IMPORTANT CLASSES OF HILBERT SPACE OPERATORS

7.3.4. Proposition. For a self-adjoint operator T on a Hilbert space the following are equivalent:
(i) σ(T ) ⊆ [0, ∞);
(ii) there exists an operator S on H such that T = S ∗ S; and
(iii) T is positive.
7.3.5. Notation. For vectors x and y in a Hilbert space H let
x ⊗ y : H → H : z 7→ hz, yix .
7.3.6. Proposition. If x and y are nonzero vectors in a Hilbert space H, then x ⊗ y is a rank 1
operator in B(H).
Let us extend the terminology of definition 2.1.29 from complex valued functions to functions
whose codomain is a ∗ -algebra.
7.3.7. Definition. Suppose V is a vector space and A is a ∗ -algebra. A function φ : V × V → A is
sesquilinear if it is linear

∗ in its first variable and conjugate linear in its second. It is conjugate
symmetric if φ(x, y) = φ(y, x) for all x, y ∈ V .
If A is a C ∗ -algebra we say that the mapping φ is positive semidefinite if φ(x, x) ≥ 0 for
every x ∈ V .
7.3.8. Proposition. If H is a Hilbert space, then
φ : H × H → B(H) : (x, y) 7→ x ⊗ y
is a positive semidefinite, conjugate symmetric, sesquilinear mapping.
7.3.9. Proposition. If H is a Hilbert space, x and y are elements of H, and T ∈ B(H), then
T (x ⊗ y) = (T x) ⊗ y
and therefore
(x ⊗ y)T = x ⊗ (T ∗ y) .
7.3.10. Lemma. If x is a vector in a Hilbert space H, then x ⊗ x is a rank 1 projection if and
only if x is a unit vector.
7.3.11. Proposition. Let M = span{e1 , . . . , en } where {e1 , . . . , en } is an orthonormal subset of a
Hilbert space H. Then
Xn
PM = ek ⊗ ek .
k=1
7.3.12. Proposition. Every finite rank Hilbert space operator is a linear combination of rank one
projections.
Proof. See [20], theorem 2.4.6.

7.4. Compact Operators

7.4.1. Definition. An operator K on a Hilbert space H is compact if the image of the closed
unit ball in H under K is a compact (equivalently, totally bounded) subset of H. We denote by
K(H) the family of all compact operators on H. (More generally, a bounded linear map K between
normed linear spaces is compact if the image under K of the closed unit ball is relatively compact.)
7.4.2. Example. The integral operator K = int k defined in example 2.2.17 on the Hilbert space
L2 (S) (where (S, A, µ) is a sigma-finite measure space) is a compact Hilbert space operator.
7.4.3. Example. The Volterra operator defined in example 2.2.18 on L2 ([0, 1]) is a compact Hilbert
space operator.
7.4.4. Example. The identity operator on a Banach space is compact if and only if the space is
finite dimensional.
 7.4. COMPACT OPERATORS 67

Proof. See [4], theorem 4.4.6.

7.4.5. Proposition. Let H be an infinite dimensional Hilbert space. Then the family K(H) of
compact operators on H is the closure of the set of all finite rank operators on H.
Proof. See [5] theorem II.4.4 or [9], theorem 5.9.
7.4.6. Proposition. Let H be an infinite dimensional Hilbert space. Then the family K(H) of
compact operators on H is a minimal ideal in the C ∗ -algebra B(H).
Proof. See [5], theorem VI.3.4 or [9], corollary 5.11.
 CHAPTER 8

THE GELFAND-NAIMARK-SEGAL CONSTRUCTION

8.1. Positive Linear Functionals

8.1.1. Definition. Let A be an algebra with involution. For each linear functional τ on A and
each a ∈ A define τ ? (a) = τ (a∗ ). We say that τ is Hermitian if τ ? = τ . Notice that a linear
functional τ : A → C is Hermitian if and only if it preserves involution; that is, τ ? = τ if and only
if τ (a∗ ) = τ (a) for all a ∈ A.
CAUTION. The τ ? defined above should not be confused with the usual adjoint mapping τ ∗ : C∗ →
A∗ . Context (or use of a magnifying glass) should make it clear which is intended.
8.1.2. Proposition. A linear functional τ on a C ∗ -algebra A is Hermitian if and only if τ (a) ∈ R
whenever a is self-adjoint.
As is always the case with maps between ordered vector spaces, positive maps are the ones that
take positive elements to positive elements.
8.1.3. Definition. A linear functional τ on a C ∗ -algebra A is positive if τ (a) ≥ 0 whenever a ≥ 0
in A for all a ∈ A.
8.1.4. Proposition. Every positive linear functional on a C ∗ -algebra is Hermitian.
8.1.5. Proposition. The family of positive linear functionals is a proper convex cone in the real
vector space of all Hermitian linear functionals on a C ∗ -algebra. The cone induces a partial ordering
on the vector space: τ1 ≤ τ2 whenever τ2 − τ1 is positive.
8.1.6. Definition. A state of a C ∗ -algebra A is a positive linear functional τ on A such that
τ (1) = 1.
8.1.7. Example. Let x be a vector in a Hilbert space H. Define
ωx : B(H) → C : T 7→ hT x, xi .
Then ωx is a positive linear functional on B(H). If x is a unit vector, then ωx is a state of B(H).
A state τ is a vector state if τ = ωx for some unit vector x.
8.1.8. Proposition (Schwarz inequality). If τ is a positive linear functional on a C ∗ -algebra A,
then
|τ (b∗ a)|2 ≤ τ (a∗ a)τ (b∗ b)
for all a, b ∈ A.
8.1.9. Proposition. A linear functional τ on a C ∗ -algebra A is positive if and only if it is bounded
and kτ k = τ (1A ).
Proof. See [18], pages 256–257.

8.2. Representations
8.2.1. Definition. Let A be a C ∗ -algebra.
A representation of A is a pair (π, H) where H
is a Hilbert space and π : A → B(H) is a ∗ -homomorphism. Usually one says simply that π is a
representation of A. When we wish to emphasize the role of the particular Hilbert space we say
that π is a representation of A on H. Depending on context we may write either πa or π(a) for
69
70 8. THE GELFAND-NAIMARK-SEGAL CONSTRUCTION

the Hilbert space operator which is the image of the algebra element a under π. A representation
π of A on H is nondegenerate if π(A)H is dense in H.

8.2.2. Convention. We add to the preceding definition the following requirement: if the C ∗ -
algebra A is unital, then a representation of A must be a unital ∗ -homomorphism.
8.2.3. Definition. A representation π of a C ∗ -algebra A on a Hilbert space H is faithful if it is
injective. If there exists a vector x ∈ H such that π → (A)x = {πa (x) : a ∈ A} is dense in H, then
we say that the representation π is cyclic and that x is a cyclic vector for π.
8.2.4. Example. Let (S, A, µ) be a σ-finite measure space and L∞ = L∞ (S, A, µ) be the C ∗ -algebra
of essentially bounded µ-measurable functions on S. As we saw in example 5.1.8 for each φ ∈ L∞
the corresponding multiplication operator Mφ is an operator on the Hilbert space L2 = L2 (S, A, µ).
The mapping M : L∞ → B(L2 ) : φ 7→ Mφ is a faithful representation of the C ∗ -algebra L∞ on the
Hilbert space L2 .
8.2.5. Example. Let C([0, 1]) be the C ∗ -algebra of continuous functions on the interval [0, 1]. For
each φ ∈ C([0, 1]) the corresponding multiplication operator Mφ is an operator on the Hilbert space
L2 = L2 ([0, 1]) of functions on [0, 1] which are square-integrable with respect to Lebesgue measure.
The mapping M : C([0, 1]) → B(L2 ) : φ 7→ Mφ is a faithful representation of the C ∗ -algebra C([0, 1])
on the Hilbert space L2 .
8.2.6. Example. Suppose that π is a representation of a unital C ∗ -algebra A on a Hilbert space H
and x is a unit vector in H. If ωx is the corresponding vector state of B(H), then ωx ◦ π is a state
of A.
8.2.7. Exercise. Let X be a locally compact Hausdorff space. Find an isometric (therefore faithful)
representation (π, H) of the C ∗ -algebra C0 (X) on some Hilbert space H..
8.2.8. Definition. Let ρ be a state of a C ∗ -algebra A. Then
Lρ := {a ∈ A : ρ(a∗ a) = 0}
is called the left kernel of ρ.
Recall that as part of the proof of Schwarz inequality 8.1.8 for positive linear functionals we
verified the following result.
8.2.9. Proposition. If ρ is a state of a C ∗ -algebra A, then ha, bi0 := ρ(b∗ a) defines a semi-inner
product on A.
8.2.10. Corollary. If ρ is a state of a C ∗ -algebra A, then its left kernel Lρ is a vector subspace of
A and h [a], [b] i := ha, bi0 defines an inner product on the quotient algebra A/Lρ .
8.2.11. Proposition. Let ρ be a state of a C ∗ -algebra A and a ∈ Lρ . Then ρ(b∗ a) = 0 for every
b ∈ A.
8.2.12. Proposition. If ρ is a state of a C ∗ -algebra A, then its left kernel Lρ is a closed left ideal
in A.

8.3. The GNS-Construction and the Third Gelfand-Naimark Theorem

The following theorem is known as the Gelfand-Naimark-Segal construction (the GNS-construction).
8.3.1. Theorem (GNS-construction). Let ρ be a state of a C ∗ -algebra A. Then there exists a
cyclic representation πρ of A on a Hilbert space Hρ and a unit cyclic vector xρ for πρ such that
ρ = ωxρ ◦ πρ .
8.3.2. Notation. In the following material πρ , Hρ , and xρ are the cyclic representation, the Hilbert
space, and the unit cyclic vector guaranteed by the GNS-construction starting with a given state ρ
of a C ∗ -algebra
 8.3. THE GNS-CONSTRUCTION AND THE THIRD GELFAND-NAIMARK THEOREM 71

8.3.3. Proposition. Let ρ be a state of a C ∗ -algebra A and π be a cyclic representation of A on

a Hilbert space H such that ρ = ωx ◦ π for some unit cyclic vector x for π. Then there exists a
unitary map U from Hρ to H such that x = U xρ and π(a) = U πρ (a)U ∗ for all a ∈ A.
L
8.3.4. Definition. Let {Hλ : λ ∈ Λ} be a family of Hilbert spaces. Denote by Hλ the set of all
X λ∈Λ
kxλ k2 < ∞. On
S L
functions x : Λ → Hλ : λ 7→ xλ such that xλ ∈ Hλ for each λ ∈ Λ and Hλ
λ∈Λ λ∈Λ λ∈Λ
define addition and scalar multiplication pointwise; thatPis. (x + y)λ = xλ + yλ and (αx)λ = αxλ
for all λ ∈ Λ, and define
L an inner product by hx, yi = λ∈Λ hxλ , yλ i. These operations (are well
defined and) make Hλ into a Hilbert space. It is the direct sum of the Hilbert spaces Hλ .
λ∈Λ
Various notations for elements of this direct sum occur in the literature: x, (xλ ), (xλ )λ∈Λ , and
⊕λ xλ are common.
Now suppose that {Tλ : λ ∈ Λ} is a family of Hilbert space operators where Tλ ∈ B(Hλ ) for
each λ ∈ Λ. Suppose further that
L sup{kTλ k : λ ∈ Λ} < ∞. Then T (xλ )λ∈Λ = (T Lλ xλ )λ∈Λ defines an
operator on the Hilbert space λ Hλ . The operator T is usually denoted by λ Tλ and is called
the direct sum of the operators Tλ .
L
8.3.5.
L Proposition. The claimsL made in the preceding definition that λ∈Λ Hλ is a Hilbert space
and λ Tλ is an operator on Hλ are correct.
λ∈Λ

8.3.6. Example. Let A be a C ∗ -algebra and {πλ : λ ∈ Λ} be a family of representations of A on

Hilbert spaces Hλ so that πλ (a) ∈ B(Hλ ) for each λ ∈ Λ and each a ∈ A. Then
L L
L
π = λ πλ : A → B λ Hλ : a 7→ λ πλ (a)
L
is a representation of A on the Hilbert space λ Hλ . It is the direct sum of the representations πλ .
8.3.7. Theorem. Every C ∗ -algebra has a faithful representation.
Proof. See [18], page 281 or [5], page 253.
An obvious restatement of the preceding theorem is a third version of the Gelfand-Naimark
theorem, which says that every C ∗ -algebra is (essentially) an algebra of Hilbert space operators.
8.3.8. Corollary (Gelfand-Naimark Theorem III). Every C ∗ -algebra is isometrically ∗ -isomorphic
to a C ∗ -subalgebra of B(H) for some Hilbert space H.
 CHAPTER 9

MULTIPLIER ALGEBRAS

9.1. Hilbert Modules

9.1.1. Notation. The inner products that occur previously in these notes and that one encounters
in standard textbooks and monographs on Hilbert spaces, functional analysis, and so on, are linear
in the first variable and conjugate linear in the second. Most contemporary operator algebraists
have chosen to work with objects called right Hilbert A-modules (A being a C ∗ -algebra). For such
modules it turns out to be more convenient to have “inner products” that are linear in the second
variable and conjugate linear in the first. While this switch in conventions may provoke some slight
irritation, it is, mathematically speaking, of little consequence. Of course, we would like Hilbert
spaces to be examples of Hilbert C-modules. To make this possible we equip a Hilbert space, whose
inner product is denoted by h , i with a new “inner product” defined by hx | yi := hy, xi. This
“inner product” is linear in the second variable and conjugate linear in the first. I will try to be
consistent in using h , i for the standard inner product and h | i for the one which is linear
in its second variable.
Another (very common) solution to this problem is to insist that an inner product is always
linear in its second variable and “correct” standard texts, monographs, and papers accordingly.
9.1.2. Convention. In light of the preceding remarks we will in the sequel use the word “sesquilin-
ear” to mean either linear in the first variable and conjugate linear in the second or linear in the
second variable and conjugate linear in the first.
9.1.3. Definition. Let A be a nonzero C ∗ -algebra. A vector space V is an A-module if there is
a bilinear map
B : V × A → V : (x, a) 7→ xa
such that x(ab) = (xa)b holds for all x ∈ V and a, b ∈ A. We also require that x1A = x for every
x ∈ V if A is unital. (A function of two variables is bilinear if it is linear in both of its variables.)
9.1.4. Definition. For clarity in some of the succeeding material it will be convenient to have
the following formal definition. A (complex) vector space is a triple (V, +, M ) where (V, +) is
an Abelian group and M : C → Hom(V, +) is a unital ring homomorphism. (As you would guess,
Hom(V, +) denotes the unital ring of endomorphisms of the group (V, +).) Thus an algebra is
an ordered quadruple (A, +, M, · ) where (A, +, M ) is a vector space and · : A × A → A : (a, b) 7→
a · b = ab is a binary operation on A satisfying the four conditions of 2.3.1.
9.1.5. Exercise. Make sure the preceding definition of vector space is equivalent to the one you
are accustomed to.
9.1.6. Definition. Let (A, +, M, · ) be a (complex) algebra. Then Aop is the algebra (A, +, M, ∗)
where a ∗ b = b · a for all a, b ∈ A. This is the opposite algebra of A. It is just A with the order
of multiplication reversed. If A and B are algebras, then any function f : A → B induces in an
obvious fashion a function from A into B op (or from Aop into B, or from Aop into B op ). We will
denote all these functions simply by f .
9.1.7. Definition. Let A and B be algebras. A function φ : A → B is an antihomomorphism
if the function f : A → B op is a homomorphism. A bijective antihomomorphism is an anti-
isomorphism. In example 7.1.15 we encountered anti-isomorphisms of Hilbert spaces. These
73
74 9. MULTIPLIER ALGEBRAS

were a bit different since instead of reversing multiplication (which Hilbert spaces don’t have) they
conjugate scalars. In either case an anti-isomorphism is not something terribly different from an
isomorphism but actually something quite similar.
The notion of an A-module (where A is an algebra) was defined in 9.1.3. You may prefer
the following alternative definition, which is more in line with the definition of vector space given
in 9.1.4.
9.1.8. Definition. Let A be an algebra. An A-module is an ordered quadruple (V, +, M, Φ)
where (V, +, M ) is a vector space and Φ : A → L(V ) is an algebra homomorphism. If A is unital
we require also that Φ be unital.
9.1.9. Exercise. Check that the definitions of A-module given in 9.1.3 and 9.1.8 are equivalent.
We now say precisely what it means,when A is a C ∗ -algebra, to give an A-module an A-valued
inner product.
9.1.10. Definition. Let A be a C ∗ -algebra. A semi-inner product A-module is an A-module
V together with a mapping
β : V × V → A : (x, y) 7→ hx| yi
which is linear in its second variable and satisfies
(i) hx| yai = hx| yia,
(ii) hx| yi = hy| xi∗ , and
(iii) hx| xi ≥ 0
for all x, y ∈ V and a ∈ A. It is an inner product A-module (or a pre-Hilbert A-module)
if additionally
(iv) hx| xi = 0 implies that x = 0
when x ∈ V . We will refer to the mapping β as an A-valued (semi-)inner product on V .
9.1.11. Example. Every inner product space is an inner product C-module.
9.1.12. Proposition. Let A be a C ∗ -algebra and V be a semi-inner product A-module. The semi-
inner product h | i is conjugate linear in its first variable both literally and in the sense that
hva| wi = a∗ hv| wi for all v, w ∈ V and a ∈ A.
9.1.13. Proposition (Schwarz inequality—for inner product A-modules). Let V be an inner prod-
uct A-module where A is a C ∗ -algebra. Then
hx| yi∗ hx| yi ≤ khx| xik hy| yi
for all x, y ∈ V .
Hint for proof . Show that no generality is lost in assuming that khx| xik = 1. Consider the
positive element hxa − y| xa − yi where a = hx| yi. Use propositions 6.5.21 and 6.5.22.
9.1.14. Definition. For every element v of an inner product A-module (where A is a C ∗ -algebra)
define
kvk := khv| vik1/2 .
9.1.15. Proposition (Yet another Schwarz inequality). Let A be a C ∗ -algebra and V be an inner
product A-module. Then for all v, w ∈ V
khv| wik ≤ kvk kwk.
9.1.16. Corollary. If v and w are elements of an inner product A-module (where A is a C ∗ -
algebra), then kv + wk ≤ kvk + kwk and the map x 7→ kxk is a norm on V .
 9.1. HILBERT MODULES 75

9.1.17. Proposition. If A is a C ∗ -algebra and V is an inner product A-module, then

kvak ≤ kvk kak
for all v ∈ V and a ∈ A.
9.1.18. Definition. Let A be a C ∗ -algebra and V be an inner product A-module. If V is complete
with respect to (the metric induced by) the norm defined in 9.1.14, then V is a Hilbert A-module.
9.1.19. Example. For a and b in a C ∗ -algebra A define
ha| bi := a∗ b .
Then A is itself a Hilbert A-module. Any closed right ideal in A is also a Hilbert A-module.
9.1.20. Definition. Let V and W be Hilbert A-modules where A is a C ∗ -algebra. A mapping
T : V → W is A-linear if it is linear and if T (va) = T (v)a holds for all v ∈ V and a ∈ A. The
mapping T is a Hilbert A-module morphism if it is bounded and A-linear.
Recall from proposition 5.1.6 that every Hilbert space operator has an adjoint. This is not true
for Hilbert A-modules
9.1.21. Definition. Let V and W be Hilbert A-modules where A is a C ∗ -algebra. A function
T : V → W is adjointable if there exists a function T ∗ : W → V satisfying
hT v| wi = hv| T ∗ wi
for all v ∈ V and w ∈ W . The function T ∗ , if it exists, is the adjoint of T . Denote by L(V, W )
the family of adjointable maps from V to W . We shorten L(V, V ) to L(V ).
9.1.22. Proposition. Let V and W be Hilbert A-modules where A is a C ∗ -algebra. If a function
T : V → W is adjointable, then it is a Hilbert A-module morphism. Furthermore if T is adjointable,
then so is its adjoint and T ∗∗ = T .
9.1.23. Example. Let X be the unit interval [0, 1] and let Y = {0}. With its usual topology X is
a compact Hausdorff space and Y is a subspace of X. Let A be the C ∗ =algebra C(X) and J0 be the
ideal {f ∈ A : f (0) = 0} (see proposition 3.2.9). Regard V = A and W = J0 as Hilbert A-modules
(see example 9.1.19). Then the inclusion map ι : V → W is a Hilbert A-module morphism which
is not adjointable.
9.1.24. Proposition. Let A be a C ∗ -algebra. The pair of maps V 7→ V , T 7→ T ∗ is a contravariant
functor from the category of Hilbert A-modules and adjointable maps into itself.
9.1.25. Proposition. Let A be a C ∗ -algebra and V be a Hilbert A-module. Then L(V ) is a unital
C ∗ -algebra.
9.1.26. Notation. Let V and W be Hilbert A-modules where A is a C ∗ -algebra. For v ∈ V and
w ∈ W let
Θv,w : W → V : x 7→ vhw| xi .
(Compare this with 7.3.5.)
9.1.27. Proposition. The map Θ defined above is sesquilinear.
9.1.28. Proposition. Let V and W be Hilbert A-modules where A is a C ∗ -algebra. For every
v ∈ V and w ∈ W the map Θv,w is adjointable and (Θv,w )∗ = Θw,v .
The next proposition generalizes proposition 7.3.9.
9.1.29. Proposition. Let U , V , W , and Z be Hilbert A-modules where A is a C ∗ -algebra. If
S ∈ L(Z, W ) and T ∈ L(V, U ), then
T Θv,w = ΘT v,w and Θv,w S = Θv,S ∗ w
for all v ∈ V and w ∈ W .
Θv,w
Z
S /W /V T /U
76 9. MULTIPLIER ALGEBRAS

9.1.30. Proposition. Let U , V , and W be Hilbert A-modules where A is a C ∗ -algebra. Suppose

u ∈ U ; v, v 0 ∈ V ; and w ∈ W . Then
Θu,v Θv0 ,w = Θuhv| v0 i,w = Θu,whv0 | vi .
9.1.31. Notation. Let A be a C ∗ -algebra and V and W be Hilbert A-modules. We denote by
K(W, V ) the closed linear span of {Θv,w : v ∈ V and w ∈ W }. As usual we shorten K(V, V ) to K(V ).
9.1.32. Proposition. If A is a C ∗ -algebra and V is a Hilbert A-module, then K(V ) is an ideal in
the C ∗ -algebra L(V ).
The next example is intended as justification for the standard practice of identifying a C ∗ -
algebra A with K(A).
∗
9.1.33. Example. If we regard a C ∗ -algebra A as an A-module (see example 9.1.19), then K(A) ∼
=
A.
Hint for proof . As in corollary 5.3.19 define for each a ∈ A the left multiplication operator
La : A → A : x 7→ ax. Show that each such operator is adjointable and that the map L : A →
L(A) : a 7→ La is a ∗ -isomorphism onto a C ∗ -subalgebra of L(A). Then verify that K(A) is the
closure of the image under L of the span of products of elements of A.
9.1.34. Example. Let H be a Hilbert space regarded as a C-module. Then K(H) (as defined
in 9.1.31) is the ideal of compact operators on H (see proposition 7.4.6).
Proof. See [25], example 2.27.
The preceding example has lead many researchers, when dealing with an arbitrary Hilbert
module V , to refer to members of K(V ) as compact operators. This is dubious terminology since
such operators certainly need not be compact. (For example, if we regard an infinite dimensional
unital C ∗ -algebra A as an A-module, then Θ1,1 = IA , but the identity operator on A is not
compact—see example 7.4.4.)
The fact that in these notes rather limited use is made of Hilbert C ∗ -modules should not lead
you to think that their study is specialized and/or of marginal interest. To the contrary, it is
currently an important and vigorous research area having applications to fields as diverse as K-
theory, graph C ∗ -algebras, quantum groups, quantum probability, vector bundles, non-commutative
geometry, algebraic and geometric topology, operator spaces and algebras, and wavelets. Take a
look at Michael Frank’s webpage [13], Hilbert C*-modules and related subjects—a guided reference
overview, where he lists 1531 references (as of his 11.09.10 update) to books, papers, and theses
dealing with such modules and categorizes them by application. There is an interesting graphic (on
page 9) illustrating the growth of this field of mathematics. It covers material from the pioneering
efforts in the 50’s and early 60’s (0–2 papers per year) to the time of this writing (about 100 papers
per year).

9.2. Essential Ideals

9.2.1. Example. If A and B are C ∗ -algebras, then A (more precisely, A ⊕ {0}) is an ideal in
A ⊕ B.
9.2.2. Convention. As the preceding example suggests, it is conventional to regard A as a subset
of A ⊕ B. In the sequel we will do this without further mention.
9.2.3. Notation. For an element c of an algebra A let
 Xp 
Ic := a0 c + cb0 + ak cbk : p ∈ N, a0 , . . . , ap , b0 , . . . , bp ∈ A
k=1

9.2.4. Proposition. If c is an element of an algebra A, then Ic is an (algebraic) ideal in A.

 9.2. ESSENTIAL IDEALS 77

Notice that in the preceding proposition no claim is made that the algebraic ideal Ic must be
proper. It may well be the case that Ic = A (as, for example, when c is an invertible element of a
unital algebra).
9.2.5. Definition. Let c be an element of a C ∗ -algebra A. Define Jc , the principal ideal
containing c, to be the intersection of the family of all (closed) ideals of A which contain c. Clearly,
Jc is the smallest ideal containing c.
9.2.6. Proposition. In a C ∗ -algebra the closure of an algebraic ideal is an ideal.
9.2.7. Example. The closure of a proper algebraic ideal in a C ∗ -algebra need not be a proper
ideal. For example, lc , the set of sequences of complex numbers which are eventually zero, is dense
in the C ∗ -algebra l0 = C0 (N). (But recall proposition 3.2.1.)
9.2.8. Proposition. If c is an element of a C ∗ -algebra, then Jc = Ic .
9.2.9. Notation. We adopt a standard notational convention. If A and B are nonempty subsets
of an algebra. By AB we mean the linear span of products of elements in A and elements in B;
that is, AB = span{ab : a ∈ A and b ∈ B}. (Note that in definition 2.3.17 it makes no difference
whether we take AJ to mean the set of products of elements in A with elements in J or the span
of that set.)
9.2.10. Proposition. If I and J are ideals in a C ∗ -algebra, then IJ = I ∩ J.

A nonunital C ∗ -algebra A can be embedded as an ideal in a unital C ∗ -algebra in different

ways. The smallest unital C ∗ -algebra containing A is its unitization A e (see proposition 6.3.1). Of
∗
course there is no largest unital C -algebra in which A can be embedded as an ideal because if A is
embedded as an ideal in a unital C ∗ -algebra B and C is any unital C ∗ -algebra, then A is an ideal
in the still larger unital C ∗ -algebra B ⊕ C. The reason this larger unitization is not of much interest
is that the intersection of the ideal C with A is {0}. This motivates the following definition.
9.2.11. Definition. An ideal J in a C ∗ -algebra A is essential if and only if I ∩ J 6= 0 for every
nonzero ideal I in A.
9.2.12. Example. A C ∗ -algebra A is an essential ideal in its unitization A
e if and only if A is not
unital.
9.2.13. Example. If H is a Hilbert space the ideal of compact operators K(H) is an essential ideal
in the C ∗ -algebra B(H).
9.2.14. Definition.
 Let J be an ideal in a C ∗ -algebra A. Then we define J ⊥ , the annihilator
of J, to be a ∈ A : Ja = {0} .
9.2.15. Proposition. If J is an ideal in a C ∗ -algebra, then so is J ⊥ .
9.2.16. Proposition. An ideal J in a C ∗ -algebra A is essential if and only if J ⊥ = {0}.

⊥
9.2.17. Proposition. If J is an ideal in a C ∗ -algebra, then J ⊕ J ⊥ = {0}.
9.2.18. Notation. Let f be a (real or) complex valued function on a set S. Then
Zf := {s ∈ S : f (s) = 0}.
This is the zero set of f .

Suppose that A is a nonunital commutative C ∗ -algebra. By the second Gelfand-Naimark the-

orem 6.3.11 there exists a noncompact locally compact Hausdorff space X such that A = C0 (X).
(Here, of course, we are permitting ourselves a conventional abuse of language: for literal cor-
rectness the indicated equality should be an isometric ∗ -isomorphism.) Now let Y be a compact
78 9. MULTIPLIER ALGEBRAS

Hausdorff space in which X is an open subset and let B = C(Y ). Then B is a unital commutative
C ∗ -algebra. Regard A as embedded as an ideal in B by means of the map ι : A → B : f 7→ fe where

f (y), if y ∈ X;
f (y) =
e
0, otherwise.
Notice that the closed set X c is {Zfe : f ∈ C0 (X)}.
T

9.2.19. Proposition. Let the notation be as in the preceding paragraph. Then the ideal A is
essential in B if and only if the open subset X is dense in Y .
Thus the property of an ideal being essential in the context of unitizations of nonunital com-
mutative C ∗ -algebras corresponds exactly with the property of an open subspace being dense in
the context of compactifications of noncompact locally compact Hausdorff spaces.

9.3. Compactifications and Unitizations

In definition 6.3.3 we called the object whose existence was proved in the preceding propo-
sition 6.3.1 “the” unitization of a C ∗ -algebra. The definite article there is definitely misleading.
Just as a topological space may have many different compactifications, a C ∗ -algebra may have
many unitizations. If A is a nonunital commutative C ∗ -algebra, it is clear from corollary 6.6.15
that the unitization A e is the smallest possible unitization of A. Similarly, in topology, if X is a
noncompact locally compact Hausdorff space, then its one-point compactification is obviously the
smallest possible compactification of X. Recall that in proposition 6.3.16 we established the fact
that constructing the smallest unitization of A is “essentially” the same thing as constructing the
smallest compactification of X.
It is sometimes convenient (as, for example, in the preceding paragraph) to take a “unitization”
of an algebra that is already unital and sometimes convenient to take a “compactification” of a
space that is already compact. Since there appears to be no universally accepted terminology, I
introduce the following (definitely nonstandard, but I hope helpful) language.
9.3.1. Definition. Let A and B be C ∗ -algebras and X and Y be Hausdorff topological spaces.
We will say that
(1) B is a unitization of A if B is unital and A is (∗ -isomorphic to) a C ∗ -subalgebra of B;
(2) B is an essential unitization of A if B is unital and A is (∗ -isomorphic to) an essential
ideal of B;
(3) Y is a compactification of X if it is compact and X is (homeomorphic to) a subspace
of Y ; and
(4) Y is an essential compactification of X if it is compact and X is (homeomorphic to)
a dense subspace of Y
Perhaps a few words are in order concerning the bits in parentheses in the preceding definition.
It is seldom the case that a topological space X is literally a subset of a particular compactification
of X or that a C ∗ -algebra A is a subset of a particular unitization of A. While certainly true that
it is frequently convenient to regard one C ∗ -algebra as a subset of another when in fact the first is
merely ∗ -isomorphic to a subset of the second, there are also occasions when it clarifies matters to
specify the actual embeddings involved. If the details of these distinctions are not entirely familiar,
the next two definitions are intended to help.
9.3.2. Definition. Let A and B be C ∗ -algebras. We say that A is embedded in B if there exists
an injective ∗ -homomorphism ι : A → B; that is, if A is ∗ -isomorphic to a C ∗ -subalgebra of B (see
propositions 6.6.9 and 6.3.14). The injective ∗ -homomorphism ι is an embedding of A into B. In
this situation it is common practice to treat A and the range of ι as identical C ∗ -algebras. The
pair (B, ι) is a unitization of A if B is a unital C ∗ -algebra and ι : A → B is an embedding. The
unitization (B, ι) is essential if the range of ι is an essential ideal in B.
 9.3. COMPACTIFICATIONS AND UNITIZATIONS 79

9.3.3. Definition. Let X and Y be Hausdorff topological spaces. We say that X is embedded
in Y if there exists a homeomorphism j from X to a subspace of Y . The homeomorphism j is a
embedding of X into Y . As in C ∗ -algebras it is common practice to identify the range of j with
the space X. The pair (Y, j) is a compactification of X if Y is a compact Hausdorff space and
j : X → Y is an embedding. The compactification (Y, j) is essential if the range of j is dense
in Y .
We have discussed the smallest unitization of a C ∗ -algebra and the smallest compactification
of a locally compact Hausdorff space. Now what about a largest, or even maximal, unital algebra
containing A? Clearly there is no such thing, for if B is a unital algebra containing A, then so is
B ⊕ C where C is any unital C ∗ -algebra. Similarly, there is no largest compact space containing X:
if Y is a compact space containing X, then so is the topological disjoint union Y ] K where K
is any nonempty compact space. However, it does make sense to ask whether there is a maximal
essential unitization of a C ∗ -algebra or a maximal essential compactification of a locally compact
Hausdorff space. The answer is yes in both cases. The well-known Stone-Čech compactification
β(X) is maximal among essential compactifications of a noncompact locally compact Hausdorff
space X. Details can be found in any good topology text. One readable standard treatment is [30],
items 19.3–19.12. More sophisticated approaches make use of some functional analysis—see, for
example, [5], chapter V, section 6. There turns out also to be a maximal essential unitization of a
nonunital C ∗ -algebra A—it is called the multiplier algebra of A.
We say that an essential unitization M of a C ∗ -algebra A is maximal if any C ∗ -algebra that
contains A as an essential ideal embeds in M . Here is a more formal statement.
9.3.4. Definition. An essential unitization (M, j) of a C ∗ -algebra A is said to be maximal if for
every embedding ι : A → B whose range is an essential ideal in B there exists a ∗ -homomorphism
φ : B → M such that φ ◦ ι = j.
9.3.5. Proposition. In the preceding definition the ∗ -homomorphism φ, if it exists must be injec-
tive.
9.3.6. Proposition. In the preceding definition the ∗ -homomorphism φ, if it exists must be unique.
Compare the following definition with 8.2.1.
9.3.7. Definition. Let A and B be C ∗ -algebras and V be a Hilbert A-module. A ∗ -homomorphism
φ : B → L(V ) is nondegenerate if φ→ (B) V is dense in V .
9.3.8. Proposition. Let A, B, and J be C ∗ -algebras, V be a Hilbert B-module, and ι : J → A be
an injective ∗ -homomorphism whose range is an ideal in A. If φ : J → L(V ) is a nondegenerate
∗ -homomorphism, then there exists a unique extension of φ to a ∗ -homomorphism φ : A → L(V )
which satisfies φ ◦ ι = φ.
9.3.9. Proposition. If A is a nonzero C ∗ -algebra, then (L(A), L) is a maximal essential unitization
of A. It is unique in the sense that if (M, j) is another maximal essential unitization of A, then
there exists a ∗ -isomorphism φ : M → L(A) such that φ ◦ j = L.
9.3.10. Definition. Let A be a C ∗ -algebra. We define the multiplier algebra of A, to be the
family L(A) of adjointable operators on A. From now on we denote this family by M (A).
 CHAPTER 10

FREDHOLM THEORY

10.1. The Fredholm Alternative

In 1903 Erik Ivar Fredholm published a seminal paper on integral equations in the journal Acta
Mathematica. Among many important results was the theorem we know today as the Fredholm
alternative. We state a version of this result in the language available to Fredholm at the time.
10.1.1. Proposition (Fredholm Alternative I). Let k be a continuous complex valued function on
the unit square [0, 1] × [0, 1]. Either the nonhomogeneous equations
Z 1
λf (s) − k(s, t)f (t) dt = g(s) and (1)
0
Z 1
λh(s) − k(t, s)h(t) dt = j(s) (2)
0
have solutions f and h for every given g and j, respectively, the solutions being unique, in which
case the corresponding homogeneous equations
Z 1
λf (s) − k(s, t)f (t) dt = 0 and (3)
0
Z 1
λh(s) − k(t, s)h(t) dt = 0 (4)
0
have only the trivial solution; —or else—
the homogeneous equations (3) and (4) have the same (nonzero) finite number of linearly indepen-
dent solutions f1 , . . . , fn and h1 , . . . , hn , respectively, in which case the nonhomogeneous equations
(1) and (2) have a solution if and only if g and j satisfy
Z 1
hk (t)g(t) dt = 0 and (5)
0
Z 1
j(t)fk (t) dt = 0 (6)
0
for k = 1, . . . , n.
By 1906 David Hilbert had noticed that integration as such had very little to do with the
correctness of this result. What was important, he discovered, was the compactness of the resulting
integral operator. (In the early 20th century compactness was called complete continuity. The term
Hilbert space was used as early as 1911 in connections with the sequence space l2 . It was not until
1929 that John von Neumann introduced the notion of—and axioms defining—abstract Hilbert
spaces.) So here is a somewhat updated version of Fredholm alternative.
10.1.2. Proposition (Fredholm Alternative II). If K is a compact Hilbert space operator, λ ∈ C,
and T = λI − K, then either the nonhomogeneous equations
T f = g and (1’)
∗
T h=j (2’)

81
82 10. FREDHOLM THEORY

have solutions f and h for every given g and j, respectively, the solutions being unique, in which
case the corresponding homogeneous equations
T f = 0 and (3’)
∗
T h=0 (4’)
have only the trivial solution; —or else—
the homogeneous equations (3’) and (4’) have the same (nonzero) finite number of linearly indepen-
dent solutions f1 , . . . , fn and h1 , . . . , hn , respectively, in which case the nonhomogeneous equations
(1’) and (2’) have a solution if and only if g and j satisfy
hk ⊥ g and (5’)
j ⊥ fk (6’)
for k = 1, . . . , n.
Notice that by making use of a few elementary facts concerning kernels and ranges of operators
and orthogonality in Hilbert spaces, we can compress the statement of 10.1.2 quite a bit.
10.1.3. Proposition (Fredholm Alternative IIIa). If T = λI − K where K is a compact Hilbert
space operator and λ ∈ C, then
(1) T is injective if and only if it is surjective,
(2) ran T ∗ = (ker T )⊥ , and
(3) dim ker T = dim ker T ∗ .
Also, conditions (1) and (2) hold for T ∗ as well as T .

10.2. The Fredholm Alternative – continued

10.2.1. Definition. An operator T on a Banach space is a Riesz-Schauder operator if it can be
written in the form T = S + K where S is invertible, K is compact, and SK = KS.
The material in section 7.1 and the preceding definition make it possible to generalize the
version of the Fredholm alternative given in 10.1.3 to Banach spaces.
10.2.2. Proposition (Fredholm Alternative IIIb). If T is a Riesz-Schauder operator on a Banach
space, then
(1) T is injective if and only if it is surjective,
(2) ran T ∗ = (ker T )⊥ , and
(3) dim ker T = dim ker T ∗ < ∞.
Also, conditions (1) and (2) hold for T ∗ as well as T .
10.2.3. Proposition. If M is a closed subspace of a Banach space, then M ⊥ ∼
= (B/M )∗ .
Proof. See [5], page 89.
10.2.4. Definition. If T : V → W is a linear map between vector spaces, then its cokernel is
defined by
coker T = W/ ran T .
Recall that the codimension of a subspace U of a vector space V is the dimension of V /U . Thus
when T is a linear map dim coker T = codim ran T .
In the category HIL of Hilbert spaces and continuous linear maps the range of a morphism
need not be an object of the category. Specifically the range of an operator need not be closed.
10.2.5. Example. The Hilbert space operator
T : l2 → l2 : (x1 , x2 , x3 , . . . ) 7→ (x1 , 12 x2 , 13 x3 , . . . )
is injective, self-adjoint, compact, and contractive, but its range, while dense in l2 , is not all of l2 .
 10.3. FREDHOLM OPERATORS 83

Although it is incidental to our present purposes this is a convenient place to note the fact that
the sum of two subspaces of a Hilbert space need not be a subspace.
10.2.6. Example. Let T be the operator on the Hilbert space l2 defined in example 10.2.5, M =
l2 ⊕ {0}, and N be the graph of T . Then M and N are both (closed) subspaces of the Hilbert
space l2 ⊕ l2 but M + N is not.
Proof. Verification of example 10.2.6 follows easily from the following result and example 10.2.5.
10.2.7. Proposition. Let H be a Hilbert space, T ∈ B(H), M = H ⊕ {0}, and N be the graph
of T . Then
(a) The set N is a subspace of H ⊕ H.
(b) The operator T is injective if and only if M ∩ N = { (0, 0) }.
(c) The range of T is dense in H if and only if M + N is dense in H ⊕ H.
(d) The operator T is surjective if and only if M + N = H ⊕ H.
The good news for the theory of Fredholm operators is that operators with finite dimensional
cokernels automatically have closed range.
10.2.8. Proposition. If a bounded linear map A : H → K between Hilbert spaces has finite di-
mensional cokernel, then ran A is closed in K.
We observe in Fredholm alternative IIIb 10.2.2 that condition (1) is redundant and also that (2)
holds for any Banach space operator with closed range. This enables us to rephrase 10.2.2 more
economically.
10.2.9. Proposition (Fredholm alternative IV). If T is a Riesz-Schauder operator on a Banach
space, then
(1) T has closed range and
(2) dim ker T = dim ker T ∗ < ∞.

10.3. Fredholm Operators

10.3.1. Definition. Let H be a Hilbert space and K(H) be the ideal of compact operators in the
C ∗ -algebra B(H). Then the quotient algebra (see proposition 6.6.7) Q(H) := B(H)/K(H) is the
Calkin algebra. As usual the quotient map taking B(H) onto B(H)/K(H) is denoted by π so
that if T ∈ B(H) then π(T ) = [T ] = T + K(H) is its corresponding element in the Calkin algebra.
An element T ∈ B(H) is a Fredholm operator if π(T ) is invertible in Q(H). We denote the
family of all Fredholm operators on H by F(H).
10.3.2. Proposition. If H is a Hilbert space, then F(H) is a self-adjoint open subset of B(H)
which is closed under compact perturbations (that is, if T is Fredholm and K is compact, then T +K
is Fredholm).
10.3.3. Theorem (Atkinson’s theorem). A Hilbert space operator is Fredholm if and only if it has
finite dimensional kernel and cokernel.
Proof. See [1], theorem 3.3.2; [3], theorem I.8.3.6; [9], theorem 5.17; [16], theorem 2.1.4;
or [29], theorem 14.1.1.
Interestingly the dimensions of the kernel and cokernel of a Fredholm operator are not very
important quantities. Their difference however turns out to be very important.
10.3.4. Definition. If T is a Fredholm operator on a Hilbert space then its Fredholm index (or
just index) is defined by
ind T := dim ker T − dim coker T .
10.3.5. Example. Every invertible Hilbert space operator is Fredholm with index zero.
84 10. FREDHOLM THEORY

10.3.6. Example. The Fredholm index of the unilateral shift operator is −1.
10.3.7. Example. Every linear map T : V → W between finite dimensional vector spaces is
Fredholm and ind T = dim V − dim W .
10.3.8. Example. If T is a Fredholm operator on a Hilbert space, then ind T ∗ = − ind T .
10.3.9. Example. The index of any normal Fredholm operator is 0.
10.3.10. Lemma. Let S : U → V and T : V → W be linear transformations between vector spaces.
Then (there exist linear mappings such that) the following sequence is exact.
0 / ker S / ker T S / ker T / coker S / coker T S / coker T / 0.

10.3.11. Lemma. If V0 , V1 , . . . , Vn are finite dimensional vector spaces and the sequence
0 / V0 / V1 / ... / Vn /0

is exact, then
n
X
(−1)k dim Vk = 0 .
k=0

10.3.12. Proposition. Let H be a Hilbert space. Then the set F(H) of Fredholm operators on H
is a semigroup under composition and the index function ind is an epimorphism from F(H) onto
the additive semigroup Z of integers.
Proof. Hint. Use 10.3.10, 10.3.11, 10.3.5, 10.3.6, and 10.3.8.

10.4. The Fredholm Alternative – Concluded

The next result implies that every Fredholm operator of index zero is of the form invertible
plus compact.
10.4.1. Proposition. If T is a Fredholm operator of index zero on a Hilbert space, then there
exists a finite rank partial isometry V such that T − V is invertible.
Proof. See [23], lemma 3.3.14 or [29], proposition 14.1.3.
10.4.2. Lemma. If F is a finite rank operator on a Hilbert space, then I + F is Fredholm with
index zero.
Proof. See [23], lemma 3.3.13 or [29], lemma 14.1.4.
10.4.3. Notation. Let H be a Hilbert space. For each integer n we denote the family of all
Fredholm operators of index n on H by Fn (H) or just Fn .
10.4.4. Proposition. In every Hilbert space F0 + K = F0 .
Proof. See [9], lemma 5.20 or [29], 14.1.5.
10.4.5. Corollary (Fredholm alternative V). Every Riesz-Schauder operator on a Hilbert space is
Fredholm of index zero.
Proof. This follows immediately from the preceding proposition 10.4.4 and example 10.3.5.

We have actually proved a stronger result: our final (quite general) version of the Fredholm
alternative.
10.4.6. Corollary (Fredholm alternative VI). A Hilbert space operator is Riesz-Schauder if and
only if it is Fredholm of index zero.
Proof. Proposition 10.4.1 and corollary 10.4.5. 
 10.4. THE FREDHOLM ALTERNATIVE – CONCLUDED 85

10.4.7. Proposition. If T is a Fredholm operator of index n ∈ Z on a Hilbert space H and K is

a compact operator on H, then T + K is also Fredholm of index n; that is
Fn + K = Fn .
Proof. See [16], proposition 2.1.6; [23], theorem 3.3.17; or [29], proposition 14.1.6.
10.4.8. Proposition. Let H be a Hilbert space. Then Fn (H) is an open subset of B(H) for each
integer n.
Proof. See [29], proposition 14.1.8.
10.4.9. Definition. A path in a topological space X is a continuous map from the interval [0, 1]
into X. Two points p and q in X are said to be connected by a path (or homotopic in X)
if there exists a path f : [0, 1] → X in X such that f (0) = p and f (1) = q. In this case we write
p ∼h q.
10.4.10. Proposition. The relation ∼h of homotopy equivalence defined above is an equivalence
relation on the set of points of a topological space.
10.4.11. Definition. If X is a topological space and ∼h is the relation of homotopy equivalence,
then the resulting equivalence classes are the path components of X.
The next proposition identifies the path components of the set of Fredholm operators as the
sets Fn of operators with index n.
10.4.12. Proposition. Operators S and T in the space F(H) of Fredholm operators on a Hilbert
space H are homotopic in F(H) if and only if they have the same index.
Proof. See [29], corollary 14.1.9.
 CHAPTER 11

EXTENSIONS

11.1. Essentially Normal Operators

11.1.1. Definition. Let T be an operator on a Hilbert space H. The essential spectrum of T ,
denoted by σe (T ), is the spectrum of the image of T in the Calkin algebra Q(H); that is,

σe (T ) = σQ(H) (π(T )) .

11.1.2. Proposition. If T is an operator on a Hilbert space H, then

σe (T ) = {λ ∈ C : T − λI ∈
/ F(H)} .

11.1.3. Proposition. The essential spectrum of a self-adjoint Hilbert space operator T is the union
of the accumulation points of the spectrum of T with the eigenvalues of T having infinite multiplicity.
The members of σ(T ) \ σe (T ) are the isolated eigenvalues of finite multiplicity.

Proof. See [16], proposition 2.2.2.

Here are two theorems from classical functional analysis.

11.1.4. Theorem (Weyl). If S and T are operators on a Hilbert space whose difference is compact,
then their spectra agree except perhaps for eigenvalues.

11.1.5. Theorem (Weyl-von Neumann). Let T be a self-adjoint operator on a separable Hilbert

space H. For every  > 0 there exists a diagonalizable operator D such that T − D is compact and
kT − Dk < .

Proof. See [6], 38.1; [7], corollary II.4.2; or [16], 2.2.5.

11.1.6. Definition. Let H and K be Hilbert spaces. Operators S ∈ B(H) and T ∈ B(K) are
essentially unitarily equivalent (or compalent) if there exists a unitary map U : H → K
such that S − U T U ∗ is a compact operator on H. (We extend definitions 1.2.35 and 1.2.36 in the
obvious way: U ∈ B(H, K) is unitary if U ∗ U = IH and U U ∗ = IK ; and S and T are unitarily
equivalent if there exists a unitary map U : H → K such that S = U T U ∗ .)

11.1.7. Proposition. Self-adjoint operators S and T on separable Hilbert spaces are essentially
unitarily equivalent if and only if they have the same essential spectrum.

Proof. See [16], proposition 2.2.4.

11.1.8. Definition. A Hilbert space operator T is essentially normal if its commutator

[T, T ∗ ] := T T ∗ − T ∗ T is compact. That is to say: T is essentially normal if its image π(T ) is
a normal element of the Calkin algebra. The operator T is essentially self-adjoint if T − T ∗
is compact; that is, if π(T ) is self-adjoint in the Calkin algebra.

11.1.9. Example. The unilateral shift operator S (see example 2.2.15) is essentially normal (but
not normal).
87
88 11. EXTENSIONS

11.2. Toeplitz Operators

11.2.1. Definition. Let ζ : T → T : z 7→ z be the identity function on the unit circle. Then
{z n : n ∈ Z} is an orthonormal basis for the Hilbert space L2 (T) of (equivalence classes of) functions
square integrable on T with respect to (suitably normalized) arc length measure. We denote by H 2
the subspace of L2 (T) which is the closed linear span of {ζ n : n ≥ 0} and by P+ the (orthogonal)
projection on L2 (T) whose range (and codomain) is H 2 . The space H 2 is an example of a Hardy
space. For every φ ∈ L∞ (T) we define a mapping Tφ from the Hilbert space H 2 into itself by
Tφ = P+ Mφ (where Mφ is the multiplication operator defined in example 5.1.8). Such an operator
is a Toeplitz operator and φ is its symbol. Clearly Tφ is an operator on H 2 and kTφ k ≤ kφk∞ .
11.2.2. Example. The Toeplitz operator Tζ acts on the basis vectors ζ n of H 2 by Tζ (ζ n ) = ζ n+1 ,
so it is unitarily equivalent to S the unilateral shift.
11.2.3. Proposition. The map T : L∞ (T) → B(H 2 ) : φ 7→ Tφ is positive, linear, and involution
preserving.
11.2.4. Example. The Toeplitz operators Tζ and Tζ show that the map T in the preceding
proposition is not a representation of L∞ (T) on B(H 2 ). (Compare this with example 8.2.4.)
11.2.5. Notation. Let H ∞ := H 2 ∩ L∞ (T). This is another example of a Hardy space.
11.2.6. Proposition. An essentially bounded function φ on the unit circle belongs to H ∞ if and
only if the multiplication operator Mφ maps H 2 into H 2 .
Although (as we saw in example 11.2.4) the map T defined in proposition 11.2.3 is not in general
multiplicative, it is multiplicative for a large class of functions.
11.2.7. Proposition. If φ ∈ L∞ (T) and ψ ∈ H ∞ , then Tφψ = Tφ Tψ and Tψφ = Tψ Tφ .
11.2.8. Proposition. If the Toeplitz operator Tφ with symbol φ ∈ L∞ (T) is invertible then the
function φ is invertible.
Proof. See [9], proposition 7.6.
11.2.9. Theorem (Hartman-Wintner Spectral Inclusion Theorem). If φ ∈ L∞ (T), then σ(φ) ⊆
σ(Tφ ) and ρ(Tφ ) = kTφ k = kφk∞ .
Proof. See [9], corollary 7.7 or [20], theorem 3.5.7.
11.2.10. Corollary. The mapping T : L∞ (T) → B(H 2 ) defined in proposition 11.2.3 is an isom-
etry.
11.2.11. Proposition. A Toeplitz operator Tφ with symbol φ ∈ L∞ (T) is compact if and only if
φ = 0.
Proof. See [20], theorem 3.5.8.
11.2.12. Proposition. If φ ∈ C(T) and ψ ∈ L∞ (T), then the semi-commutators Tφ Tψ − Tφψ and
Tψ Tφ − Tψφ are compact.
Proof. See [1], proposition 4.3.1; [7], corollary V.1.4; or [20], lemma 3.5.9.
11.2.13. Corollary. Every Toeplitz operator with continuous symbol is essentially normal.
11.2.14. Definition. Suppose that H is a separable Hilbert space with basis {e0 , e1 , e2 , . . . } and
that T is an operator on H whose (infinite) matrix representation is [tij ]. If the entries in this
matrix depend only on the difference of the indices i − j (that is, if each diagonal parallel to the
main diagonal is a constant sequence), then the matrix is a Toeplitz matrix.
11.2.15. Proposition. Let T be an operator on the Hilbert space H 2 . The matrix representation
of T with respect to the usual basis {ζ n : n ≥ 0} is a Toeplitz matrix if and only if S ∗ T S = T (where
S is the unilateral shift).
 11.2. TOEPLITZ OPERATORS 89

Proof. See [1], proposition 4.2.3.

11.2.16. Proposition. If Tφ is a Toeplitz operator with symbol φ ∈ L∞ (T), then S ∗ Tφ S = Tφ .
Conversely, if R is an operator on the Hilbert space H 2 such that S ∗ RS = R, then there exists a
unique φ ∈ L∞ (T) such that R = Tφ .
Proof. See [1], theorem 4.2.4.
11.2.17. Definition. The Toeplitz algebra T is the C ∗ -subalgebra of B(H 2 ) generated by the
unilateral shift operator; that is, T = C ∗ (S).
11.2.18. Proposition. The set K(H 2 ) of compact operators on H 2 is an ideal in the Toeplitz
algebra T.
11.2.19. Proposition. The Toeplitz algebra comprises all compact perturbations of Toeplitz oper-
ators with continuous symbol. That is,
T = {Tφ + K : φ ∈ C(T) and K ∈ K(H 2 )} .
Furthermore, if Tφ + K = Tψ + L where φ, ψ ∈ C(T) and K, L ∈ K(H 2 ), then φ = ψ and K = L.
Proof. See [1], theorem 4.3.2 or [7], theorem V.1.5.
11.2.20. Proposition. The map π ◦ T : C(T) → Q(H 2 ) : φ 7→ π(Tφ ) is a unital ∗ -monomorphism.
So the map α : C(T) → T/K(H 2 ) : φ 7→ π(Tφ ) establishes and isomorphism between the C ∗ -algebras
C(T) and T/K(H 2 ).
Proof. See [16], proposition 2.3.3 or [20], theorem 3.5.11.
11.2.21. Corollary. If φ ∈ C(T), then σe (Tφ ) = ran φ.
11.2.22. Proposition. The sequence
β
0 / K(H 2 ) /T / C(T) /0

is exact. It does not split.

Proof. The map β is defined by β(R) := α−1 π(R) for every R ∈ T (where α is the isomor-

phism defined in the preceding proposition 11.2.20). See [1], remark 4.3.3; [7], theorem V.1.5; or
[16], page 35.
11.2.23. Definition. The short exact sequence in the preceding proposition 11.2.22 is the Toeplitz
extension of C(T) by K(H 2 ).
11.2.24. Remark. A version of the diagram for the Toeplitz extension which appears frequently
looks something like
β
/ K(H 2 ) /To / /0
0 C(T) (1)
T
where β is as in 11.2.22 and T is the mapping φ 7→ Tφ . It is possible to misinterpret this diagram.
It may suggest to the unwary that this is a split extension especially in as much as it is certainly
true that β ◦ T = IC(T) . The trouble, of course, is that this is not a diagram in the category CSA
of C ∗ -algebras and ∗ -homomorphisms. We have already seen in example 11.2.4 that the mapping
T is not a ∗ -homomorphism since it does not always preserve multiplication. Invertible elements
in C(T) need not lift to invertible elements in the Toeplitz algebra T; so the ∗ -epimorphism β does
not have a right inverse in the category CSA.
Some authors deal with the problem by saying that the sequence (1) is semisplit (see, for
example, Arveson [1], page 112). Others borrow a term from category theory where the word
“section” means “right invertible”. Davidson [7], for example, on page 134, refers to the mapping
β as a “continuous section” and Douglas [9], on page 179, says it is an “isometrical cross section”.
While it is surely true that β has T as a right inverse in the category SET of sets and maps and
even in the category of Banach spaces and bounded linear maps, it has no right inverse in CSA.
90 11. EXTENSIONS

11.2.25. Proposition. If φ is a function in C(T), then Tφ is Fredholm if and only if it is never

zero.
Proof. See [1], page 112, corollary 1; [7], theorem V.1.6; [9], theorem 7.2 6; or [20], corollary
3.5.12.
11.2.26. Proposition. If φ is an invertible element of C(T), then there exists a unique integer n
such that φ = ζ n exp ψ for some ψ ∈ C(T).
Proof. See [1], propositions 4.4.1 and 4.4.2 or [20], lemma 3.5.14.
11.2.27. Definition. The integer whose existence is asserted in the preceding proposition 11.2.26
is the winding number of the invertible function φ ∈ C(T). It is denoted by w(φ).
11.2.28. Theorem (Toeplitz index theorem). If Tφ is a Toeplitz operator with a nowhere vanishing
continuous symbol, then it is a Fredholm operator and
ind(Tφ ) = −w(φ) .
Proof. Elementary proofs can be found in [1], theorem 4.4.3 and [20]. For those with a
little background in homotopy of curves proposition 11.2.26, which leads to the definition above of
winding number, can be bypassed. It is an elementary fact in homotopy theory that the fundamental
group π 1 (C \ 0) of the punctured plane is infinite cyclic. There is an isomorphism τ from π 1 (C \ 0)
to Z that associates the integer +1 with (the equivalence class containing) the function ζ. This
allows us to associate with each invertible member φ of C(T) the integer corresponding under τ
to its equivalence class. We call this integer the winding number of φ. Such a definition makes
it possible to give a shorter more elegant proof of the Toeplitz index theorem. For a treatment in
this vein see [9], theorem 7.26 or [16], theorem 2.3.2. For background concerning the fundamental
group see [19], chapter two; [28], appendix A; or [30], sections 32 and 33.
11.2.29. Theorem (Wold decomposition). Every proper (that is, non-unitary) isometry on a
Hilbert space is a direct sum of copies of the unilateral shift or else a direct sum of a unitary
operator together with copies of the shift.
Proof. See [7], theorem V.2.1; [14], problem (and solution) 149; or [20], theorem 3.5.17.
11.2.30. Theorem (Coburn). Let v be an isometry in a unital C ∗ -algebra A. Then there ex-
ists a unique unital ∗ -homomorphism τ from the Toeplitz algebra T to A such that τ (Tζ ) = v.
Furthermore, if vv ∗ 6= 1, then τ is an isometry.
Proof. See [20], theorem 3.5.18 or [7], theorem V.2.2.

11.3. Addition of Extensions

From now on all Hilbert spaces will be separable and infinite dimensional.
11.3.1. Proposition. A Hilbert space operator T is essentially self-adjoint if and only if it is a
compact perturbation of a self-adjoint operator.
11.3.2. Proposition. Two essentially self-adjoint Hilbert space operators T1 and T2 are essentially
unitarily equivalent if and only if they have the same essential spectrum.
The analogous result does not hold for essentially normal operators.
11.3.3. Example. The Toeplitz operators Tζ and Tζ 2 are essentially normal with the same essential
spectrum, but they are not essentially unitarily equivalent.
11.3.4. Definition. We now restrict our attention to a special class of extensions. If H is a Hilbert
space and A is a C ∗ -algebra, we say that (E, φ) is an extension of K = K(H) by A if E is a unital
C ∗ -subalgebra of B(H) containing K and φ is a unital ∗ -homomorphism such that the sequence
φ
0 /K ι /E /A /0 (11.1)
 11.3. ADDITION OF EXTENSIONS 91

(where ι is the inclusion mapping) is exact. In fact, we will be concerned almost exclusively with
the case where A = C(X) for some compact metric space X. We will say that two extensions (E, φ)
and (E 0 , φ 0 ) are equivalent if there exists an isomorphism ψ : E → E 0 which makes the following
diagram commute.
φ
0 /K ι /E /A /0 (11.2)
ψ|K) ψ
  φ0
0 /K ι / E0 /A /0

Notice that this differs slightly from the definition of strong equivalence given in 6.2.9. We denote
the family of all equivalence classes of such extensions by Ext A. When A = C(X) we write Ext X
rather than Ext C(X).

11.3.5. Definition. If U : H1 → H2 is a unitary mapping between Hilbert spaces, then the mapping

AdU : B(H1 ) → B(H2 ) : T 7→ U T U ∗

is called conjugation by U .

It is clear that AdU is an isomorphism between the C ∗ -algebras B(H1 ) and B(H2 ). In partic-
ular, if U is a unitary operator on H1 , then AdU is an automorphism of both K(H1 ) and B(H1 ).
Furthermore, conjugations are the only automorphisms of the C ∗ -algebra K(H).

11.3.6. Proposition. If H is a Hilbert space and φ : K(H) → K(H) is an automorphism, then

φ = AdU for some unitary operator U on H

Proof. See [7], lemma V.6.1.

11.3.7. Proposition. If H is a Hilbert space and A is a C ∗ -algebra, then extensions (E, φ) and
(E 0 , φ 0 ) in Ext A are equivalent if and only if there exists a unitary operator U in B(H) such that
E 0 = U EU ∗ and φ = φ 0 AdU .

11.3.8. Example. Suppose that T is an essentially normal operator on a Hilbert space H. Let ET
be the unital C ∗ -algebra generated by T and K(H). Since π(T ) is a normal element of the Calkin
algebra, the unital C ∗ -algebra ET /K(H) that it generates is commutative. Thus the abstract spectral
theorem 5.4.7 gives us a C ∗ -algebra isomorphism Ψ : C(σe (T )) → ET /K(H). Let φT = Ψ−1 ◦ π E .
T
Then the sequence
φ
0 / K(H) ι / ET T / C(σe (T )) /0

is exact. This is the extension of K(H) determined by T .

11.3.9. Proposition. Let T and T 0 be essentially normal operators on a Hilbert space H. These
operators are essentially unitarily equivalent if and only if the extensions they determine are equiv-
alent.

11.3.10. Proposition. If E is a C ∗ -algebra such that K(H) ⊆ E ⊆ B(H) for some Hilbert space H,
X is a nonempty compact subset of C, and (E, φ) is an extension of K(H) by C(X), then every
element of E is essentially normal.

11.3.11. Definition. If φ1 : A1 → B and φ2 : A2 → B are unital ∗ -homomorphisms between

C ∗ -algebras, then a pullback of A1 and A2 along φ1 and φ2 , denoted by (P, π1 , π2 ), is a
C ∗ -algebra P together with a pair of unital ∗ -homomorphisms π1 : P → A1 and π2 : P → A2 which
satisfy the following two conditions:
92 11. EXTENSIONS

(i) the diagram

π2
P / A2

π1 φ2

 
A1 /B
φ1

commutes and
(ii) if ρ1 : Q → A1 and ρ2 : Q → A2 are unital ∗ -homomorphisms of C ∗ -algebras such that the
diagram
ρ2
Q / A2

ρ1 φ2

 
A1 /B
φ1

commutes, then there exists a unique unital ∗ -homomorphism Ψ : Q → P which makes

the diagram
Q
ρ2
Ψ

 )/
P π2 A2
ρ1

π1 φ2

 
A1 /B
φ1
commute.
11.3.12. Proposition. Let H be a Hilbert space, A be a unital C ∗ -algebra, and τ : A → Q(H) be
a unital ∗ -monomorphism. Then there exists (uniquely up to isomorphism) a pullback (E, π1 , π2 )
of A and B(H) along τ and π such that (E, π2 ) is an extension of K(H) by A which makes the
following diagram commute.
π2
0 / K(H) ι /E /A /0 (11.3)
π1 τ
 
0 / K(H) / B(H) π / Q(H) /0

Proof. (Sketch.) Let E = {T ⊕ a ∈ B(H) ⊕ A : τ (a) = π(T )}, π1 : E → B(H) : T ⊕ a 7→ T ,

and π2 : E → A : T ⊕ a 7→ a. The uniqueness (of any pullback) is proved using the usual “abstract
nonsense”. 
11.3.13. Proposition. Let H be a Hilbert space, A be a unital C ∗ -algebra, and (E, φ) be an
extension of K(H) by A. Then there exists a unique ∗ -monomorphism τ : A → Q(H) which makes
the diagram (11.3) commute.
11.3.14. Definition. Let H be a Hilbert space and A be a unital C ∗ -algebra. Two unital ∗ -
monomorphisms τ1 and τ2 from A into Q(H) are unitarily equivalent if there exists a unitary
 11.3. ADDITION OF EXTENSIONS 93

operator on H such that τ2 = AdU τ1 . Unitary equivalence of unital ∗ -monomorphisms is of course

an equivalence relation. The equivalence class containing τ1 is denoted by [τ1 ].
11.3.15. Proposition. Let H be a Hilbert space and A be a unital C ∗ -algebra. Two extensions of
K(H) by A are equivalent if and only if their corresponding unital ∗ -monomorphisms (see 11.3.13)
are unitarily equivalent.
11.3.16. Corollary. If H is a Hilbert space and A a unital C ∗ -algebra there is a one-to-one
correspondence between equivalence classes of extensions of K(H) by A and unitary equivalence
classes of unital ∗ -monomorphisms from A into the Calkin algebra Q(H).
11.3.17. Convention. In light of the preceding corollary we will regard members of Ext A (or
Ext X) as either equivalence classes of extensions or unitary equivalence classes of ∗ -monomorphisms,
whichever seems the most convenient at the moment.
11.3.18. Proposition. Every (separable infinite dimensional) Hilbert space H is isometrically
isomorphic to H ⊕ H. Thus the C ∗ -algebras B(H) and B(H ⊕ H) are isomorphic.
11.3.19. Definition. Let τ1 , τ2 : A → Q(H) be unital ∗ -monomorphisms (where A is a C ∗ -algebra
and H is a Hilbert space). We define a unital ∗ -monomorphism τ1 ⊕ τ2 : A → Q(H) by


(τ1 ⊕ τ2 )(a) := ρ τ1 (a) ⊕ τ2 (a)
for all a ∈ A where (as in Douglas[10]) ν is the isomorphism established in 11.3.18 and ρ is the
map which makes the following diagram commute.
B(H) ⊕ B(H) / B(H ⊕ H) ν / B(H) (11.4)

π⊕π π

 
Q(H) ⊕ Q(H) / Q(H)
ρ

We then define the obvious operation of addition on Ext A:

[τ1 ] + [τ2 ] := [τ1 ⊕ τ2 ]
for [τ1 ], [τ2 ] ∈ Ext A.
11.3.20. Proposition. The operation of addition (given in 11.3.19) on Ext A is well defined and
under this operation Ext A becomes a commutative semigroup.
11.3.21. Definition. Let A be a unital C ∗ -algebra and r : A → B(H) be a nondegenerate repre-
sentation of A on some Hilbert space H. Let P be a Hilbert space projection and M be the range
of P . Suppose that P r(a) − r(a)P ∈ K(H) for every a ∈ A. Denote by PM the projection P
with its codomain set equal to its range; that is, PM : H → M : x 7→ P x. Then for each a ∈ A we
define the abstract Toeplitz operator Ta ∈ B(M ) with symbol a associated with the
pair (r, P ) by Ta = PM r(a) M .
11.3.22. Definition. Let notation be as in the preceding definition 11.3.21. Then we define the
abstract Toeplitz extension τP associated with the pair (r, P ) by
τP : A → Q(M ) : a 7→ π(Ta ) .
Notice that the (unital ∗ -monomorphism associated with the concrete) Toeplitz extension de-
fined in 11.2.23 is an example of an abstract Toeplitz extension, and also that, in general, abstract
Toeplitz extensions need not be injective.
11.3.23. Definition. Let A be a unital C ∗ -algebra and H a Hilbert space. In the spirit of [16],
definition 2.7.6, we will say that a unital ∗ -monomorphism τ : A → Q(H) is semisplit if there
exists a unital ∗ -monomorphism τ 0 : A → Q(H) such that τ ⊕ τ 0 splits.
94 11. EXTENSIONS

11.3.24. Proposition. Suppose that A is a unital C ∗ -algebra and H is a Hilbert space. Then a
unital ∗ -monomorphism τ : A → Q(H) is semisplit if and only if it is unitarily equivalent to an
abstract Toeplitz extension.
Proof. See [16], proposition 2.7.10.

11.4. Tensor Products of C ∗ -algebras

Class Project for Winter Term
Presentation based on [29], Appendix T.

11.5. Completely Positive Maps

In example 5.3.7 it is shown how the set Mn of n × n matrices of complex numbers can be
made into a C ∗ -algebra. We now generalize that example to the algebra Mn (A) of n × n matrices
of elements of a C ∗ -algebra A.
11.5.1. Example. In example 2.3.15 it is asserted that if A is a C ∗ -algebra, then under the usual
algebraic operations the set Mn (A) of n × n matrices of elements of A is an algebra. This algebra
can be made into a ∗ -algebra by taking conjugate transposition as involution. That is, define
[aij ]∗ := aj i ∗
 

where aij ∈ A for 1 ≤ i, j ≤ n.

Let H be a Hilbert space. For the moment it need not be infinite dimensional or separable.
Denote by H n its n-fold direct sum. That is,
n
M
H n := Hk
k=1

where Hk = H for k = 1, . . . , n. Let [Tjk ] ∈ Mn (B(H)). For x = x1 ⊕ · · · ⊕ xn ∈ H n define

n
X n
X
T : H n → H n : x 7→ [Tjk ]x = T1k xk ⊕ · · · ⊕ Tnk xk .
k=1 k=1

Then T ∈ B(H n ). Furthermore, the map

Ψ : Mn B(H) → B(H n ) : [Tjk ] 7→ T

is an isomorphism of ∗ -algebras. Use Ψ to transfer the operator norm on B(H n ) to Mn B(H) .

That is, define
[Tjk ] := kTk .
This makes Mn B(H) into a C ∗ -algebra isomorphic to B(H n ).


Now suppose that A is an arbitrary C ∗ -algebra. Version III of the Gelfand-Naimark theorem
(see 8.3.8) allows us to identify A with a C ∗ -subalgebra
of B(H) for some Hilbert space H and,
consequently, Mn (A) with a C ∗ -subalgebra of Mn B(H) . With this identification Mn (A) becomes
a C ∗ -algebra. (Notice that by corollary 5.3.15 norms on C ∗ -algebras are unique; so it is clear that the
norm on Mn (A) is independent of the particular way in which A is represented as a C ∗ -subalgebra
of operators on some Hilbert space.)
11.5.2. Example. Let k ∈ N and A = Mk . Then for every n ∈ N
Mn (A) = Mn (Mk ) ∼
= Mnk .
 11.5. COMPLETELY POSITIVE MAPS 95

The isomorphism is the obvious one: just delete the inner brackets. For example, the isomorphism
from M2 (M2 ) to M4 is given by
 
   
 a11 a12 b11 b12   
a11 a12 b11 b12
 
 a21 a22 b21 b22 

 a21 a22 b21 b22 
 7→ 
  

     c11 c12 d11 d12 
 c11 c12 d d
11 12 c21 c22 d21 d22

 
 c21 c22 d21 d22 

11.5.3. Definition. A mapping φ : A → B between C ∗ -algebras is positive if it takes positive

elements to positive elements; that is, if φ(a) ∈ B + whenever a ∈ A+ .
11.5.4. Example. Every ∗ -homomorphism between C ∗ -algebras is positive. (See proposition 6.5.19.)
11.5.5. Proposition. Every positive linear map between C ∗ -algebras is bounded.
11.5.6. Notation. Let A and B be C ∗ -algebras and n ∈ N. A linear map φ : A → B induces a linear
map φ( n) := φ ⊗ idMn : A ⊗ Mn → B ⊗ Mn . As we have seen A ⊗ Mn can be identified
withMn (A).
Thus we may think of φ( n) as the map from Mn (A) to Mn (B) defined by φ( n) [ajk ] = φ(ajk ) .


It is easy to see that if φ preserves multiplication, then so does each φ( n) and if φ preserves
involution, then so does each φ( n) . Positivity, however, is not always preserved as the next exam-
ple 11.5.8 shows.
11.5.7. Definition. In Mn the standard matrix units are the matrices ejk (1 ≤ j, k ≤ n)
whose entry in the j th row and k th column is 1 and whose other entries are all 0.
11.5.8. Example. Let A = M2 . Then φ : A → A : a 7→ at (where at is the transpose of a) is a
positive mapping. The map φ(2) : M2 (A) → M2 (A) is not positive. To see this let e11 , e12 , e21 ,
and e22 be the standard matrix units for A. Then
 
 11 12   11 1 0 0 0
φ(e ) φ(e12 ) e11 e21
  
e e 0 0 1 0
φ(2) 21 22 = 21 22 = 12 22 =  0 1 0 0

e e φ(e ) φ(e ) e e
0 0 0 1
which is not positive.
11.5.9. Definition. A linear mapping φ : A → B between unital C ∗ -algebras is n-positive (for
n ∈ N) if φ( n) is positive. It is completely positive if it is n-positive for every n ∈ N.
11.5.10. Example. Every ∗ -homomorphism between C ∗ -algebras is completely positive.
∗
 Proposition. Let A be a unital C -algebra and a ∈ A. Then kak ≤ 1 if and only if
11.5.11.

1 a
is positive in M2 (A).
a∗ 1
∗ ∗
 Proposition. Let A be a unital C -algebra and a, b ∈ A. Then a a ≤ b if and only if
11.5.12.

1 a
is positive in M2 (A).
a∗ b
11.5.13. Proposition. Every unital 2-positive map between unital C ∗ -algebras is contractive.
11.5.14. Proposition (Kadison’s inequality). If φ : A → B is a unital 2-positive map between
unital C ∗ -algebras, then
∗
φ(a) φ(a) ≤ φ(a∗ a)
for every a ∈ A.
96 11. EXTENSIONS

11.5.15. Definition. It is easy to see that if φ : A → B is a bounded linear map between C ∗ -

algebras, then φ( n) is bounded for each n ∈ N. If kφkcb := sup{kφ( n) k : n ∈ N} < ∞, then φ is
completely bounded.
11.5.16. Proposition. If φ : A → B is a completely positive unital linear map between unital
C ∗ -algebras, then φ is completely bounded and
kφ(1)k = kφk = kφkcb .
Proof. See [22].
11.5.17. Theorem (Stinespring’s dilation theorem). Let A be a unital C ∗ -algebra, H be a Hilbert
space, and φ : A → B(H) be a unital linear map. Then φ is completely positive if and only if
there exists a Hilbert space H0 , an isometry V : H → H0 , and a nondegenerate representation
r : A → B(H0 ) such that φ(a) = V ∗ r(a)V for every a ∈ A.
Proof. See [16], theorem 3.1.3 or [22], theorem 4.1.
11.5.18. Notation. Let A be a C ∗ -algebra, f : A → B(H0 ), and g : A → B(H), where H0
and H are Hilbert spaces. We write g . f if there exists an isometry V : H → H0 such that
g(a) − V ∗ f (a)V ∈ K(H) for every a ∈ A. The relation . is a preordering but not a partial
ordering; that is, it is reflexive and transitive but not antisymmetric.
11.5.19. Theorem (Voiculescu). Let A be a separable unital C ∗ -algebra, r : A → B(H0 ) be a
nondegenerate representation, and φ : A → B(H) (where H and H0 are Hilbert spaces). If φ(a) = 0
whenever r(a) ∈ K(H0 ), then φ . r.
Proof. The proof is long and complicated but “elementary”. See [16], sections 3.5–3.6.
11.5.20. Proposition. Let A be a separable unital C ∗ -algebra, H be a Hilbert space, and τ1 ,
τ2 : A → Q(H) be unital ∗ -monomorphisms. If τ2 splits, then τ1 + τ2 is unitarily equivalent to τ1 .
Proof. See [16], theorem 3.4.7.
11.5.21. Corollary. Suppose that A is a separable unital C ∗ -algebra, H is a Hilbert space, and
τ : A → Q(H) is a split unital ∗ -monomorphism. Then [τ ] is an additive identity in Ext A.
Note that a unital ∗ -monomorphism τ : A → Q(H) from a separable unital C ∗ -algebra to
the Calkin algebra of a Hilbert space H is semisplit if and only if [τ ] has an additive inverse in
Ext A. The next proposition says that this happens if and only if τ has a completely positive lifting
to B(H).
11.5.22. Proposition. Let A be a separable unital C ∗ -algebra and H be a Hilbert space. A ∗ -
monomorphism τ : A → Q(H) is semisplit if and only if there exists a unital ∗ -homomorphism
τe : A → B(H) such that the diagram
B(H)
?
τe
π


A / Q(H)
τ
commutes.
Proof. See [16], theorem 3.1.5.
11.5.23. Definition. A C ∗ -algebra A is nuclear if for every C ∗ -algebra B there is a unique
C ∗ -norm on the algebraic tensor product A B.
11.5.24. Example. Every finite dimensional C ∗ -algebra is nuclear.
 11.5. COMPLETELY POSITIVE MAPS 97

Proof. See [20], theorem 6.3.9.

11.5.25. Example. The C ∗ -algebra Mn is nuclear.
Proof. See [3], II.9.4.2.
11.5.26. Example. If H is a Hilbert space, the C ∗ -algebra K(H) of compact operators is nuclear.
Proof. See [20], example 6.3.2.
11.5.27. Example. If X is a compact Hausdorff space, then the C ∗ -algebra C(X) is nuclear.
Proof. See [22], proposition 12.9.
11.5.28. Example. Every commutative C ∗ -algebra is nuclear.
Proof. See [12], theorem 7.4.1 or [29], theorem T.6.20.
The conclusion of the next result is known as the completely positive lifting property.
11.5.29. Proposition. Let A be a nuclear separable unital C ∗ -algebra and J be a separable ideal
in a unital C ∗ -algebra B. Then for every completely positive map φ : A → B/J there exists a
completely positive map φe : A → B which makes the diagram
?B
φ
e


A / B/J
φ

commute.
Proof. See [16], theorem 3.3.6.
11.5.30. Corollary. If A is a nuclear separable unital C ∗ -algebra, then Ext A is an Abelian group.
 CHAPTER 12

K-THEORY

12.1. Projections on Matrix Algebras

Recall from definition 10.4.9 that points p and q in a topological space X are homotopic in
X if there exists a continuous function f : [0, 1] → X : t 7→ ft such that f0 = p and f1 = q. In this
case we write p ∼h q in X (or just p ∼h q when the space X is clear from context).
12.1.1. Proposition. The relation ∼h between points in a topological space is an equivalence
relation. The equivalence classes are the path components of X.
12.1.2. Notation. In a unital C ∗ -algebra A the path component of the space of unitary operators
containing the identity is denoted by U0 (A). That is, U0 (A) = {u ∈ U(A) : u ∼h 1 in U(A)}.
12.1.3. Example. If h is a self-adjoint element of a unital C ∗ -algebra A, then exp(ih) ∈ U0 (A).
12.1.4. Proposition. If u is a unitary element in a unital C ∗ -algebra A whose spectrum is not all
of T, then u ∈ U0 (A).
12.1.5. Proposition. Let u1 , u2 , u3 , and u4 be unitary elements in a unital C ∗ -algebra A. If
u1 ∼h u2 and u3 ∼h u4 , then u1 u3 ∼h u2 u4 .
12.1.6. Proposition. If u1 and u2 are unitary elements in a unital C ∗ -algebra A such that ku1 −
u2 k < 2, then u1 ∼h u2 .
12.1.7. Example. If u1 and u2 are unitary elements in a unital C ∗ -algebra A, then
   
u1 0 u u 0
∼h 1 2
0 u2 0 1


in U M2 (A) .
12.1.8. Proposition. If A is a unital C ∗ -algebra, then U0 (A) is a normal subgroup of U(A) which
is both open and closed in U(A). Furthermore, u ∈ U0 (A) if and only if u = exp(ih1 ) · · · exp(ihn )
for some finite collection of self-adjoint elements h1 , . . . , hn in A.
Proof. See [27], proposition 2.1.6.
12.1.9. Proposition (Polar decomposition). If a is an invertible element of a unital C ∗ -algebra,
then u := a|a|−1 (exists and) is unitary. Clearly then, a = u|a|.
Proof. See [27], proposition 2.1.8.
12.1.10. Proposition. If p and q are projections in a unital C ∗ -algebra, then k2p − 1k = 1 and
kp − qk ≤ 1.
12.1.11. Notation. If p and q are projections in a C ∗ -algebra A, we write p ∼u q (p is unitarily
equivalent to q) if there exists a unitary element u ∈ A e such that q = upu∗ .
12.1.12. Proposition. The relation ∼u is an equivalence relation on the family P(A) of projections
in a C ∗ -algebra A.
12.1.13. Notation. If p and q are projections in a C ∗ -algebra A, we write p ∼ q (p is Murray-
von Neumann equivalent to q) is there exists an element v ∈ A such that v ∗ v = p and vv ∗ = q.
(Note that such a v is automatically a partial isometry.)
99
100 12. K-THEORY

12.1.14. Proposition. The relation ∼ of Murray-von Neumann equivalence is an equivalence

relation on the family P(A) of projections in a C ∗ -algebra A.
12.1.15. Proposition. If p and q are projections in a C ∗ -algebra, then
p ∼h q =⇒ p ∼u q =⇒ p ∼ q .
12.1.16. Proposition. If p and q are projections in a unital C ∗ -algebra A, then p ∼u q if and
only if there exists an element u ∈ U(A) such that q = upu∗ .
12.1.17. Definition. An element s of a unital C ∗ -algebra is an isometry if s∗ s = 1.
12.1.18. Example. If p and q are projections in a C ∗ -algebra, then
p ∼ q =⇒
6 p ∼u q .
For example, if s is a nonunitary isometry, then s∗ s ∼ ss∗ , but s∗ s u ss∗ .
12.1.19. Notation. If a1 , a2 , . . . an are elements of a C ∗ -algebra A, then diag(a1 , a2 , . . . , an ) is
the diagonal matrix in Mn (A) whose main diagonal consists of the elements a1 , . . . , an . We also
use the notations for block matrices. For example if a is an m × m matrix and b is an n × n matrix,
a 0
then diag(a, b) is the (m + n) × (m + n) matrix .
0 b
12.1.20. Proposition. Let p and q be projections in a C ∗ -algebra A. Then
p ∼ q =⇒ diag(p, 0) ∼u diag(q, 0) in M2 (A).
12.1.21. Example. If p and q are projections in a C ∗ -algebra, then
p ∼u q =⇒
6 p ∼h q .
An example of this phenomenon is not easy to come by. To see what is involved look at [27],
examples 2.2.9 and 11.3.4.
12.1.22. Proposition. Let p and q be projections in a C ∗ -algebra A. Then
p ∼u q =⇒ diag(p, 0) ∼h diag(q, 0) in M2 (A).
12.1.23. Notation. When A is a C ∗ -algebra and n ∈ N we let
Pn (A) = P(Mn (A)) and
∞
[
P∞ (A) = Pn (A).
n=1

We now extend Murray-von Neumann equivalence to matrices of different sizes.

12.1.24. Definition. If A is a C ∗ -algebra, p ∈ Pm (A), and q ∈ Pn (A), we set p ∼◦ q if there exists
v ∈ Mn,m (A) such that
v∗v = p and vv ∗ = q.
Note: if m = n, then ∼◦ is just Murray-von Neumann equivalence ∼. (Here, Mn,m (A) is the
set of n × m matrices with entries belonging to A.)
12.1.25. Proposition. The relation ∼◦ defined above is an equivalence relation on P∞ (A).
12.1.26. Definition. For each C ∗ -algebra A we define a binary operation ⊕ on P∞ (A) by
p ⊕ q = diag(p, q) .
Thus if p ∈ Pm (A) and q ∈ Pn (A), then p ⊕ q ∈ Pm+n (A).
In the next proposition 0n is the additive identity in Pn (A).
12.1.27. Proposition. Let A be a C ∗ -algebra and p ∈ P∞ (A). Then p = p ⊕ 0n for every n ∈ N.
 12.2. THE GROTHENDIECK CONSTRUCTION 101

12.1.28. Proposition. Let A be a C ∗ -algebra and p, p0 , q, q 0 ∈ P∞ (A). If p ∼◦ p0 and q ∼◦ q 0 ,

then p ⊕ q ∼◦ p0 ⊕ q 0 .
12.1.29. Proposition. Let A be a C ∗ -algebra and p, q ∈ P∞ (A). Then p ⊕ q ∼◦ q ⊕ p.
12.1.30. Proposition. Let A be a C ∗ -algebra and p, q ∈ Pn (A) for some n. If p ⊥ q, then p + q
is a projection in Mn (A) and p + q ∼◦ p ⊕ q.
12.1.31. Proposition. If A is a C ∗ -algebra, then P∞ (A) is a commutative semigroup under the
operation ⊕.
12.1.32. Notation. If A is a C ∗ -algebra and p ∈ P∞ (A), let [p ]D be the equivalence class con-
taining p determined by the equivalence relation ∼◦ . Also let D(A) := { [p ]D : p ∈ P∞ (A) }.
12.1.33. Definition. Let A be a C ∗ -algebra. Define a binary operation + on D(A) by
[p ]D + [q ]D := [p ⊕ q ]D
where p, q ∈ P∞ (A).
12.1.34. Proposition. The operation + defined in 12.1.33 is well defined and makes D(A) into a
commutative semigroup.
12.1.35. Example. Making use of properties of the trace function on Mn it is not difficult to see
that
D(C) = Z+ = {0, 1, 2, . . . } .
12.1.36. Example. If H is a Hilbert space, then
D(B(H)) = Z+ ∪ {∞}.
(Use the usual addition on Z+ and let n + ∞ = ∞ + n = ∞ whenever n ∈ Z+ ∪ {∞}.)
12.1.37. Example. For the C ∗ -algebra C ⊕ C we have
D(C ⊕ C) = Z+ ⊕ Z+ .

12.2. The Grothendieck Construction

12.2.1. Definition. Let (S, +) be a commutative semigroup. Define a relation ∼ on S × S by
(a, b) ∼ (c, d) if there exists k ∈ S such that a + d + k = b + c + k.
12.2.2. Proposition. The relation ∼ defined above is an equivalence relation.
12.2.3. Notation. For the equivalence relation ∼ defined in 12.2.1 the equivalence class containing
the pair (a, b) will be denote by ha, bi rather than by [(a, b)].
12.2.4. Definition. Let (S, +) be a commutative semigroup. On G(S) define a binary operation
(also denoted by +) by:
ha, bi + hc, di := ha + c, b + di .
12.2.5. Proposition. The operation + defined above is well defined and under this operation G(S)
becomes and Abelian group.
The Abelian group (G(S), +) is called the Grothendieck group of S.
12.2.6. Definition. An injective additive map φ : S → G from a semigroup S into a group G is
an embedding of S into G.
12.2.7. Proposition. For a semigroup S and an arbitrary a ∈ S define a mapping
γS : S → G(S) : s 7→ hs + a, ai .
The mapping γS , called the Grothendieck map, is well defined and is a semigroup homomor-
phism. It is an embedding if and only if the semigroup S has the cancellation property.
102 12. K-THEORY

Saying that the Grothendieck map is well defined means that its definition is independent of
the choice of a. We frequently write just γ for γS .
12.2.8. Example. Both N = {1, 2, 3, . . . } and Z+ = {0, 1, 2, . . . } are commutative semigroups
under addition. They have the same Grothendieck group
G(N) = G(Z+ ) = Z .
12.2.9. Example. Let S be the commutative additive semigroup Z+ ∪ {∞} (see example 12.1.36).
Then G(S) = {0}.
12.2.10. Example. Let Z0 be the (commutative) multiplicative semigroup of nonzero integers.
Then G(Z0 ) = Q0 , the Abelian multiplicative group of nonzero rational numbers.
12.2.11. Proposition. If S is a commutative semigroup, then
G(S) = {γ(x) − γ(y) : x, y ∈ S} .
12.2.12. Proposition (Universality of the Grothendieck map). Let S be a commutative (additive)
semigroup and G(S) be its Grothendieck group. If H is an Abelian group and φ : S → H is
an additive map, then there exists a unique group homomorphism ψ : G(S) → H such that the
following diagram commutes.
γ
S / G(S) G(S)

ψ ψ
φ
  
H H (12.1)
In the preceding diagram G and H are just G and H regarded as semigroups and ψ is the
corresponding semigroup homomorphism. In other words, the forgetful functor  “forgets” only
about identities and inverses but not about the operation of addition. Thus the triangle on the left
is a commutative diagram in the category of semigroups and semigroup homomorphisms.
12.2.13. Proposition. Let φ : S → T be a homomorphism of commutative semigroups. Then the
map γT ◦ φ : S → G(T ) is additive. By proposition 12.2.12 there exists a unique group homomor-
phism G(φ) : G(S) → G(T ) such that the following diagram commutes.
φ
S /T

γ γ
S T

 
G(S) / G(T )
G(φ)

12.2.14. Proposition (Functorial property of the Grothendieck construction). The pair of maps
S 7→ G(S), which takes semigroups to their corresponding Grothendieck groups, and φ 7→ G(φ),
which takes semigroup homomorphisms to group homomorphism (as defined in 12.2.13) is a co-
variant functor from the category of commutative semigroups and semigroup homomorphisms to
the category of Abelian groups and group homomorphisms.
One slight advantage of the rather pedantic inclusion of a forgetful functor in diagram (12.1) is
that it makes it possible to regard the Grothendieck map γ : S 7→ γS as a natural transformation
of functors.
12.2.15. Corollary (Naturality of the Grothendieck map). Let  be the forgetful functor on
Abelian groups which “forgets” about identities and inverses but not the group operation as in 12.2.12.
Then G is a covariant functor from the category of semigroups and semigroup homomorphisms to
 12.3. K0 (A)—THE UNITAL CASE 103

itself. Furthermore, the Grothendieck map γ : S 7→ γS is a natural transformation from the identity
functor to the functor G.
φ
S /T

γ γ
S T

 
G(S) / G(T )
G(φ)

12.3. K0 (A)—the Unital Case

12.3.1. Definition. Let A be a unital C ∗ -algebra. Let K0 (A) := G(D(A)), the Grothendieck
group of the semigroup D(A) defined in 12.1.32 and 12.1.33.
12.3.2. Definition. For a unital C ∗ -algebra A define


[ ]0 : P∞ (A) → K0 (A) : p 7→ γD(A) [p ]D .
12.3.3. Definition. Let A be a unital C ∗ -algebra and p, q ∈ P∞ (A). We say that P is stably
equivalent to q and write p ∼s q if there exists a projection r ∈ P∞ (A) such that p ⊕ r ∼◦ q ⊕ r.
12.3.4. Proposition. Stable equivalence ∼s is an equivalence relation on P∞ (A).
12.3.5. Proposition. Let A be a unital C ∗ -algebra and p, q ∈ P∞ (A). Then p ∼s q if and only if
p ⊕ 1n ∼◦ q ⊕ 1n for some n ∈ N.
12.3.6. Proposition (Standard Picture of K0 (A) when A is unital). If A is a unital C ∗ -algebra,
then
K0 (A) = {[p ]0 − [q ]0 : p, q ∈ P∞ (A)}
= {[p ]0 − [q ]0 : n ∈ N and p, q ∈ Pn (A)} .
12.3.7. Proposition. Let A be a unital C ∗ -algebra and p, q ∈ P∞ (A). Then [p ⊕ q ]0 = [p ]0 + [q ]0 .
12.3.8. Proposition. Let A be a unital C ∗ -algebra and p, q ∈ Pn (A). If p ∼h q in Pn (A), then
[p ]0 = [q ]0 .
12.3.9. Proposition. Let A be a unital C ∗ -algebra and p, q ∈ Pn (A). If p ⊥ q in Pn (A), then
p + q ∈ Pn (A) and [p + q ]0 = [p ]0 + [q ]0 .
12.3.10. Proposition. Let A be a unital C ∗ -algebra and p, q ∈ P∞ (A). Then [p ]0 = [q ]0 if and
only if p ∼s q.
Proof. See [27], proposition 3.1.7.
In the next proposition the forgetful functor  is the one described in proposition 12.2.12.
12.3.11. Proposition (Universal Property of K0 —Unital Case). Let A be a unital C ∗ -algebra, G
be an Abelian group, and ν : P∞ (A) → G be a semigroup homomorphism that satisfies
(i) ν(0A ) = 0G and
(ii) if p ∼h q in Pn (A) for some n, then ν(p) = ν(q).
Then there exists a unique group homomorphism νe : K0 (A) → G such that the following diagram
commutes.
[ ]0
P∞ (A) / K0 (A) K0 (A)

ν ν
e νe

$  
G G
104 12. K-THEORY

Proof. See [27], proposition 3.1.8.

12.3.12. Definition. A ∗ -homomorphism φ : A → B between C ∗ -algebras extends, for every n ∈
N, to a ∗ -homomorphism φ : Mn (A) → Mn (B) and also (since ∗ -homomorphisms take projections
to projections) to a ∗ -homomorphism φ from P∞ (A) to P∞ (B). For such a ∗ -homomorphism φ
define
ν : P∞ (A) → K0 (B) : p 7→ [φ(p)]0 .
Then ν is a semigroup homomorphism satisfying conditions (i) and (ii) of proposition 12.3.11
according to
which there exists a unique group homomorphism K0 (φ) : K0 (A) → K0 (B) such that
K0 (φ) [p ]0 = ν(p) for every p ∈ P∞ (A).
12.3.13. Proposition. The pair of maps A 7→ K0 (A), φ 7→ K0 (φ) is a covariant functor from the
category of unital C ∗ -algebras and ∗ -homomorphisms to the category of Abelian groups and group
homomorphisms. Furthermore, for all ∗ -homomorphisms φ : A → B between unital C ∗ -algebras
the following diagram commutes.
φ
P∞ (A) / P∞ (B)

[ ]0 [ ]0

 
K0 (A) / K0 (B)
K0 (φ)

12.3.14. Notation. For C ∗ -algebras A and B let Hom(A, B) be the family of all ∗ -homomorphisms
from A to B.
12.3.15. Definition. Let A and B be C ∗ -algebras and a ∈ A. For φ, ψ ∈ Hom(A, B) let
da (φ, ψ) = kφ(a) − ψ(a)k .
Then da is a pseudometric on Hom(A, B). The weak topology generated by the family {da : a ∈ A}
is the point-norm topology on Hom(A, B).
12.3.16. Definition. Let A and B be C ∗ -algebras and φ0 , φ1 ∈ Hom(A, B). A homotopy from
φ0 to φ1 is a ∗ -homomorphism from A to C([0, 1], B) such that φ0 = E0 ◦ φ and φ1 = E1 ◦ φ. (Here,
Et is the evaluation functional at t ∈ [0, 1].) We say that φ0 and φ1 are homotopic if there exists
a homotopy from φ0 to φ1 , in which case we write φ0 ∼h φ1 .
12.3.17. Proposition. Two ∗ -homomorphisms φ0 , φ1 : A → B between C ∗ -algebras are homotopic
if and only if there exists a point-norm continuous path from φ0 to φ1 in Hom(A, B).
12.3.18. Definition. We say that C ∗ -algebras are homotopically equivalent if there exist ∗ -
homomorphisms φ : A → B and ψ : B → A such that ψ ◦ φ ∼h idA and φ ◦ ψ ∼h idB . A C ∗ -algebra
is contractible if it is homotopically equivalent to {0}.
12.3.19. Proposition. Let A and B be unital C ∗ -algebras. If φ ∼h ψ in Hom(A, B), then K0 (φ) =
K0 (ψ).
Proof. See [27], proposition 3.2.6.
12.3.20. Proposition. If unital C ∗ -algebras A and B are homotopically equivalent, then K0 (A) ∼
=
K0 (B).
Proof. See [27], proposition 3.2.6.
12.3.21. Proposition. If A is a unital C ∗ -algebra, then the split exact sequence
/A ι /A
eo
π / /0
0 C (12.2)
λ
 12.4. K0 (A)—THE NONUNITAL CASE 105

induces another split exact sequence

K0 (π)
/ K0 (A)
K0 (ι)
/ K0 (A)
e o / / 0.
0 K0 (C) (12.3)
K0 (λ)

Proof. See [27], lemma 3.2.8.

12.3.22. Example. For every n ∈ N, K0 (Mn ) ∼
= Z.
12.3.23. Example. If H is a Hilbert space, then K0 (B(H)) ∼
= 0.
12.3.24. Definition. Recall that a topological space X is contractible if there is a point a in
the space and a continuous function f : [0, 1] × X → X such that f (1, x) = x and f (0, x) = a for
every x ∈ X.
12.3.25. Example. If X is a contractible compact Hausdorff space, then K0 (C(X)) ∼
= Z.

12.4. K0 (A)—the Nonunital Case

12.4.1. Definition. Let A be a nonunital C ∗ -algebra. Recall that the split exact sequence (12.2)
for the unitization of A induces a split exact sequence (12.3) between the corresponding K0 groups.
Define
K0 (A) = ker(K0 (π)) .
12.4.2. Proposition. For a nonunital C ∗ -algebra A the mapping [ ]0 : P∞ (A) → K0 (A)
e may be
regarded as a mapping from P∞ (A) into K0 (A).
12.4.3. Proposition. For both unital and nonunital C ∗ -algebras the sequence
0 / K0 (A) / K0 (A)
e / K0 (C) /0

is exact.
12.4.4. Proposition. For both unital and nonunital C ∗ -algebras the group K0 (A) is (isomorphic
to) ker(K0 (π)).
12.4.5. Proposition. If φ : A → B is a ∗ -homomorphism between C ∗ -algebras, then there exists
a unique ∗ -homomorphism K0 (φ) which makes the following diagram commute.
K0 (πA )
K0 (A) / K0 (A)
e / K0 (C)

K0 (φ) K0 (φ)
e

 
K0 (B) / K0 (B)
e / K0 (C)
K0 (πB )

12.4.6. Proposition. The pair of maps A 7→ K0 (A), φ 7→ K0 (φ) is a covariant functor from the
category CSA of C ∗ -algebras and ∗ -homomorphisms to the category of Abelian groups and group
homomorphisms.
In propositions 12.3.19 and 12.3.20 we asserted the homotopy invariance of the functor K0 for
unital C ∗ -algebras. We now extend the result to arbitrary C ∗ -algebras.
12.4.7. Proposition. Let A and B be C ∗ -algebras. If φ ∼h ψ in Hom(A, B), then K0 (φ) = K0 (ψ).
Proof. See [27], proposition 4.1.4.
12.4.8. Proposition. If C ∗ -algebras A and B are homotopically equivalent, then K0 (A) ∼
= K0 (B).
Proof. See [27], proposition 4.1.4.
106 12. K-THEORY

12.4.9. Definition. Let π and λ be the ∗ -homomorphisms in the split exact sequence (12.2) for
the unitization of as C ∗ -algebra A. Define the scalar mapping s : Ae→A e by s := λ ◦ π.
e for A
Every member of A e can be written in the form a + α1 e for some a ∈ A and α ∈ C. Notice that
A
s(a + α1Ae) = α1Ae and that x − s(x) ∈ A for every x ∈ A.e For each natural number n the scalar
mapping induces a corresponding map s = sn : Mn (A) e → Mn (A).
e An element x ∈ Mn (A) e is a
scalar element of Mn (A) e if s(x) = x.

12.4.10. Proposition (Standard Picture of K0 (A) for arbitrary A). If A is a C ∗ -algebra, then
K0 (A) = { [p ]0 − [s(p)]0 : p, q ∈ P∞ (A)}
e .

12.5. Exactness and Stability Properties of the K0 Functor

12.5.1. Definition. A covariant functor F from a category A to a category B is exact if it takes
exact sequences in A to exact sequences in B. It is split exact if it takes split exact sequences
to split exact sequences. And it is half exact provided that whenever the sequence
j
0 / A1 / A2 k / A3 /0

is exact in A, then
F (j) F (k)
F (A1 ) / F (A2 ) / F (A3 )

is exact in B.
12.5.2. Proposition. The functor K0 is half exact.
Proof. See [27], proposition 4.3.2.
12.5.3. Proposition. The functor K0 is split exact.
Proof. See [27], proposition 4.3.3.
12.5.4. Proposition. The functor K0 preserves direct sums. That is, if A and B are C ∗ -algebras,
then K0 (A ⊕ B) = K0 (A) ⊕ K0 (B).

12.5.5. Example. If A is a C ∗ -algebra, then K0 (A)

e = K0 (A) ⊕ Z.

Despite being both split exact and half exact the functor K0 is not exact. Each of the next two
examples is sufficient to demonstrate this.
12.5.6. Example. The sequence
ψ
/ C0 (0, 1) ι / C [0, 1] /C⊕C /0



0


where ψ(f ) = f (0), f (1) , is clearly exact; but K0 (ψ) is not surjective.
12.5.7. Example. If H is a Hilbert space the exact sequence

0 / K(H) ι / B(H) π / Q(H) /0

associated with the Calkin algebra Q(H) is exact but K0 (ι) is not injective. (This example requires
a fact we have not yet derived: K0 (K(H)) ∼
= Z.)
Next is an important stability property of the functor K0 .
12.5.8. Proposition. If A is a C ∗ -algebra, then K0 (A) ∼
= K0 (Mn (A)).
Proof. See [27], proposition 4.3.8.
 12.6. INDUCTIVE LIMITS 107

12.6. Inductive Limits

12.6.1. Definition. In any category an inductive sequence is a pair (A, φ) where A = (Aj ) is
a sequence of objects and φ = (φj ) is a sequence of morphisms such that φj : Aj → Aj+1 for each j.
An inductive limit (or direct limit) of the sequence (A, φ) is a pair (L, µ) where L is an object
and µ = (µn ) is a sequence of morphisms µj : Aj → L which satisfy
(i) µj = µj+1 ◦ φj for each j ∈ N, and
(ii) if (M, λ) is a pair where M is an object and λ = (λj ) is a sequence of morphisms λj : Aj →
M satisfying λj = λj+1 ◦ φj , then there exists a unique morphism ψ : L → M such that
λj = ψ ◦ µj for each j ∈ N.
Abusing language we usually say that L is the inductive limit of the sequence (Aj ) and write
L=− lim
→ Aj .

58 L

µj µj+1

φj
··· / Aj / Aj+1 / ··· ψ

λj λj+1

&) 
M
12.6.2. Proposition. Inductive limits (if they exist in a category) are unique (up to isomorphism).
12.6.3. Proposition. Every inductive sequence (A, φ) of C ∗ -algebras has an inductive limit. (And
so does every inductive sequence of Abelian groups.)
Proof. See [3], II.8.2.1; [27], proposition 6.2.4; or [29], appendix L.
12.6.4. Proposition. If (L, µ) is the inductive limit of an inductive sequence (A, φ) of C ∗ -algebras,
then L = µn → (An ) and kµm (a)k = limn→∞ kφn,m (a)k = inf n≥m kφn,m (a)k for all m ∈ N and
S
a ∈ A.
Proof. See [27], proposition 6.2.4.
12.6.5. Definition. An approximately finite dimensional C ∗ -algebra (an AF-algebra for
short) is the inductive limit of a sequence of finite dimensional C ∗ -algebras.
12.6.6. Example. If (An ) is an increasing sequence of C ∗ -subalgebras of a C ∗ -algebra D (with
ιn : An → An+1 being the inclusion map for each n), ∗
S∞then (A, ι) is an inductive sequence of C -
algebras whose inductive limit is (B, j) where B = n=1 An and jn : An → B is the inclusion map
for each n.
φ1 φ2 φ3
12.6.7. Example. The sequence M1 = C / M2 / M3 / . . . (where φn (a) = diag(a, 0)
∗
for each n ∈ N and a ∈ Mn ) is an inductive sequence of C -algebras whose inductive limit is the
C ∗ -algebra K(H) of compact operators on a Hilbert space H.
108 12. K-THEORY

Proof. See [27], section 6.4.

12.6.8. Example. The sequence Z /Z 1 /Z 2 /Z / . . . (where n : Z → Z satisfies n(1) = n)

3 4

is an inductive sequence of Abelian groups whose inductive limit is the set Q of rational numbers.
2 / 2 / 2 / 2 /
12.6.9. Example. The sequence Z Z Z Z . . . is an inductive sequence of Abelian
groups whose inductive limit is the set of dyadic rational numbers.
The next result is referred to the continuity property of K0 .
12.6.10. Proposition. If (A, φ) is an inductive sequence of C ∗ -algebras, then


K0 lim
−→ An = − lim
→ K0 (An ) .
Proof. See [27], theorem 6.3.2.
12.6.11. Example. If H is a Hilbert space, then K0 (K(H)) ∼
= Z.
12.7. Bratteli Diagrams
12.7.1. Proposition. Nonzero algebra homomorphisms from Mk into Mn exist only if n ≥ k, in
which case they are precisely the mappings of the form
a 7→ u diag(a, a, . . . , a, 0) u∗
where u is a unitary matrix. Here there are m copies of a and 0 is the r × r zero matrix where
n = mk + r. The number m is the multiplicity of φ.
Proof. See [24], corollary 1.3.
12.7.2. Example. An example of a homomorphism from M2 into M7 is
 
a b 0 0 0 0 0
c d 0 0 0 0 0
 
  0 0 a b 0 0 0
a b  
A= 7→ 
0 0 c d 0 0 0 = diag(A, A, 0)
c d 0 0 0 0 0
 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
where m = 2 and r = 3.

12.7.3. Proposition. Every finite dimensional C ∗ -algebra A is isomorphic to a direct sum of

matrix algebras
A ' Mk 1 ⊕ · · · ⊕ Mk r .
Suppose A and B are finite dimensional C ∗ -algebras, so that
A ' Mk1 ⊕ · · · ⊕ Mkr and B ' Mn1 ⊕ · · · ⊕ Mns
and suppose that φ : A → B is a unital ∗ -homomorphism.
  Then φ is determined (up to unitary
equivalence in B) by an s × r matrix m = mij of positive integers such that
m k = n. (12.4)
Here k = (k1 , . . . , kr ), n = (n1 , . . . , ns ), and the number mij is the multiplicity of the map
φ
Mkj /A /B / Mn
i

Proof. See [7], theorem III.1.1.

12.7.4. Definition. Let φ be as in the preceding proposition 12.7.3. A Bratteli diagram for φ
consists of two rows (or columns) of vertices labeled by the kj and the ni together with mij edges
connecting kj to ni (for 1 ≤ j ≤ r and 1 ≤ i ≤ s).
 12.7. BRATTELI DIAGRAMS 109

12.7.5. Example. Suppose φ : C ⊕ C → M4 ⊕ M3 is given by

  
λ 0 0 0  
 0 λ 0 0  λ 0 0 
(λ, µ) 7→  
 0 0 λ 0 
 , 0 λ 0 
.
0 0 µ
0 0 0 µ

Then m11 = 3, m12 = 1, m21 = 2, and m22 = 1. Notice that

    
3 1 1 4
mk = = = n.
2 1 1 3
A Bratteli diagram for φ is
1 /// 4
?

/
1 3
12.7.6. Example. Suppose φ : C ⊕ M2 → M3 ⊕ M5 ⊕ M2 is given by
   
  λ 0 0
λ 0
(λ, b) 7→  ,  0 b 0 , b  .
0 b
0 0 b

Then m11 = 1, m12 = 1, m21 = 1, m22 = 2, m31 = 0, and m32 = 1. Notice that
   
1 1   3
1
m k = 1 2 = 5 = n.
2
0 1 2

A Bratteli diagram for φ is

4= 3

4*4 5

*2

12.7.7. Example. Suppose φ : M3 ⊕ M2 ⊕ M2 → M9 ⊕ M7 is given by

   
a 0 0 a 0 0
(a, b, c) 7→ 0 a 0 , 0 b 0  .
0 0 0 0 0 c
110 12. K-THEORY

Then m11 = 2, m12 = 0, m13 = 0, m21 = 1, m22 = 1, and m23 = 1. Notice that this time something
peculiar happens: we have m k 6= n:
 
  3    
2 0 0   6 9
mk = 2 = ≤ = n.
1 1 1 7 7
2

What’s the problem here? Well, the result stated earlier was for unital ∗ -homomorphisms, and this
φ is not unital. In general, this will be the best we can expect: m k ≤ n.
Here is the resulting Bratteli diagram for φ:

++ 9

2
"+
37

12.7.8. Example. If K is the Cantor set, then C(K) is an AF -algebra. To see this write K as
the intersection of a decreasing family of closed subsets Kj each of which consists of 2j disjoint
closed subintervals of [0, 1]. For each j ≥ 0 let Aj be the subalgebra of functions in C(K) which are
j
constant on each of the intervals making up Kj . Thus Aj ' C2 for each j ≥ 0. The imbedding

φj : Aj → Aj+1 : (a1 , a2 , . . . , a2j ) 7→ (a1 , a1 , a2 , a2 , . . . , a2j , a2j )

splits each minimal projection into the sum of two minimal projections. For φ0 thecorresponding

  1 0
1 1 0
matrix m0 of “partial multiplicities” is ; the matrix m1 corresponding to φ1 is 
0 1; and so

1
0 1
on. Thus the Bratteli diagram for the inductive limit C(K) = − lim
→ jA is:

51
:1
)1

D1
51
$
1
)1
1 ···
51
:1
)1

1
51
$
1
)1
 12.7. BRATTELI DIAGRAMS 111

12.7.9. Example. Here is an example of a so-called CAR-algebra (CAR = Canonical Anticom-

mutation Relations). For j ≥ 0 let Aj = M2j and
 
φ j : Aj / Aj+1 : a 7→ a 0 .
0 a
The multiplicity “matrix” m for each φj is just the 1 × 1 matrix [2]. We see that for each j
m k(j) = [2] [2j ] = [2j+1 ] = n(j).
This leads to the following Bratteli diagram

1 // 2 // 4 //
8 // 16 //
32 //
···

for the inductive limit C = −

lim
→ Aj .
However the range of φj is contained in a subalgebra Bj+1 ' M2j ⊕ M2j of Aj+1 . (Take
B0 = A0 = C.) Thus for j ∈ N we may regard φj as a mapping from Bj to Bj+1 :
   
b 0 b 0
φj : (b, c) 7→ , .
0 c 0 c
 
1 1
Now the multiplicity matrix m for each φj is and we see that for each j
1 1

1 1 2j−1
    j
2
m k(j) = = j = n(j).
1 1 2j−1 2
Then, since C = −
lim
→ Aj = −
lim
→Bj , we have a second (quite different) Bratteli diagram for the
AF-algebra C.

81 /2 /4 /8 / 16 / ···
B B B A @

1
&     
1 /2 /4 /8 / 16 / ···



12.7.10. Example. This is the Fibonacci algebra. For j ∈ N define sequences pj and qj
by the familiar recursion relations:
p1 =q1 = 1,
pj+1 =pj + qj , and
qj+1 = pj .
For all j ∈ N let Aj = Mpj ⊕ Mqj and

  
a 0
φj : Aj → Aj+1 : (a, b) 7→ ,a .
0 b
 
1 1
The multiplicity matrix m for each φj is and for each j
1 0
      
1 1 pj p j + qj pj+1
m k(j) = = = = n(j).
1 0 qj pj qj+1
112 12. K-THEORY

The resulting Bratteli diagram for F = −lim

→Aj is
81 /2 /3 /5 /8 / ···
B B B B @

1
&     
1 1 2 3 5 ···
 Bibliography

1. William Arveson, A Short Course on Spectral Theory, Springer-Verlag, New York, 2002, [A2]. 26, 83, 88, 89, 90
2. Larry Baggett and Watson Fulks, Fourier Analysis, Anjou Press, Boulder, Colorado, 1979, [BF]. 32, 33
3. Bruce Blackadar, Operator Algebras: Theory of C ∗ -algebras and von Neumann Algebras, Springer-Verlag, Berlin,
2006, [B2]. 51, 83, 97, 107
4. A. L. Brown and A. Page, Elements of Functional Analysis, Van Nostrand Reinhold, London, 1970, [BP]. 67
5. John B. Conway, A Course in Functional Analysis, second ed., Springer-Verlag, New York, 1990, [Cw1]. 8, 18,
29, 65, 67, 71, 79, 82
6. , A Course in Operator Theory, American Mathematical Society, Providence, Rhode Island, 2000, [Cw2].
87
7. Kenneth R. Davidson, C*-Algebras by Example, American Mathematical Society, Providence, R.I., 1996, [Da].
51, 59, 65, 87, 88, 89, 90, 91, 108
8. Robert S. Doran and Victor A. Belfi, Characterizations of C*-Algebras, Marcel Dekker, New York, 1986, [DB].
58
9. Ronald G. Douglas, Banach Algebra Techniques in Operator Theory, Academic Press, New York, 1972, [Do1]. 47,
65, 67, 83, 84, 88, 89, 90
10. , C*-Algebra Extensions and K-Homology, Princeton University Press, Princeton, 1980, [Do2]. 93
11. John M. Erdman, A Companion to Real Analysis, 2007,
http://web.pdx.edu/~erdman/CRA/CRAlicensepage.html, [E]. 50, 55
12. Peter A. Fillmore, A User’s Guide to Operator Algebras, John Wiley and Sons, New York, 1996, [Fi]. 51, 59, 97
13. Michael Frank, Hilbert C*-modules and related subjects—a guided reference overview, 2010,
http://www.imn.htwk-leipzig.de/~mfrank/mlit.pdf, [Fr]. 76
14. Paul R. Halmos, A Hilbert Space Problem Book, second ed., Springer-Verlag, New York, 1982, [H]. 90
15. Edwin Hewitt and Karl Stromberg, Real and Abstract Analysis, Springer-Verlag, New York, 1965, [HS]. 25, 32,
33
16. Nigel Higson and John Roe, Analytic K-Homology, Oxford University Press, Oxford, 2000, [HR]. 59, 83, 85, 87,
89, 90, 93, 94, 96, 97
17. Kenneth Hoffman and Ray Kunze, Linear Algebra, second ed., Prentice Hall, Englewood Cliffs,N.J., 1971, [HK].
5, 11
18. Richard V. Kadison and John R. Ringrose, Fundamentals of the Theory of Operator Algebras, volumes I–IV,
Academic Press, New York, 1983, [KR]. 65, 69, 71
19. William S. Massey, Algebraic Topology: An Introduction, Harcourt, Brace, and World, New York, 1967, [Ma]. 90
20. Gerard J. Murphy, C*-Algebras and Operator Theory, Academic Press, San Diego, 1990, [Mu]. 51, 65, 66, 88, 89,
90, 97
21. Theodore W. Palmer, Banach Algebras and the General Theory of ∗ -Algebras I–II, Cambridge University Press,
Cambridge, 1994/2001, [Pm]. 35
22. Vern Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge University Press, Cambridge, 2002,
[Pl]. 96, 97
23. Gert K. Pedersen, Analysis Now: Revised Printing, Springer-Verlag, New York, 1995, [Pe1]. 65, 84, 85
24. Stephen C. Power, Limit Algebras: an Introduction to Subalgebras of C*-algebras, Longman Scientific and Tech-
nical, Harlow, Essex, 1992, [Po]. 108
25. Iain Raeburn and Dana P. Williams, Morita Equivalence and Continuous-Trace C*-Algebras, American Mathe-
matical Society, Providence, R.I., 1998, [RW]. 76
26. Steven Roman, Advanced Linear Algebra, second ed., Springer, New York, 2005, [R]. 11
27. M. Rørdam, F. Larsen, and N. J. Laustsen, An Introduction to K-Theory for C*-Algebras, Cambridge University
Press, Cambridge, 2000, [RLL]. 99, 100, 103, 104, 105, 106, 107, 108
28. Andrew H. Wallace, Algebraic Topology: Homology and Cohomology, W. A. Benjamin, New York, 1970, [Wa]. 90
29. N. E. Wegge-Olsen, K-Theory and C*-Algebras, Oxford University Press, Oxford, 1993, [W-O]. 51, 83, 84, 85,
94, 97, 107
30. Stephen Willard, General Topology, Addison-Wesley, Reading, Mass., 1968, [Wi]. 79, 90

113
 Index

A
φ
/ B (φ is a morphism from A to B), 36 p ∼u q (unitary equivalence of projections), 99
−1
α (inverse of a morphism α), 36 A∗ (adjoint of a Hilbert space morphism), 43
C0 (the objects in category C), 36 S ∗ (set of adjoints of elements of S), 45
C1 (the morphisms in category C), 36 T ∗ (adjoint of a Banach space morphism), 36
Aop (opposite algebra of A), 73 V ∗ (dual of a normed linear space), 17
AB (in algebras the span of products of elements), 77 V # (algebraic dual space), 9
[p ]D + [q ]D (addition in D(A)), 101 τ ? (conjugate of τ (a∗ )), 69
p g q (supremum of projections), 64 x∗∗ (image of x under the natural injection), 28
p f q (infimum of projections), 64 4 (closed linear subspace of), 14
p ⊥ q (orthogonality of projections), 64 H N , 64
a ◦ b (operation in an algebra), 56 p ⊕ q (binary operation in P∞ (A)), 100
x ⊗ y, 66 J ⊥ (annihilator of J), 77
A ⊕ B (direct sum of C ∗ -algebras), 52 A
We (unitization of a Banach algebra), 49
a ⊕ b (an element of A ⊕ B), 52 A (closed linear span of A), 14
A ./ C (unitization of an algebra), 49
A∼ absolute value
= B (A and B are isomorphic), 36
in a C ∗ -algebra, 58
X ≈ Y (X and Y are homeomorphic), 30
∗ A C(T) (absolutely convergent Fourier series, 33
A∼ = B (A and B are ∗ -isomorphic), 47 absolutely convergent, 33
x ⊥ y (orthogonal vectors), 8 absolutely summable
[T, T ∗ ] (commutator of T ), 87 bilateral sequence, 24
[p ]0 (image under the Grothendieck map of [p ]D ), 103 sequence, 13
[p]D (∼◦ equivalence class of p), 101 abstract
ha, bi (pair in Grothendieck construction), 101 spectral theorem, 48
hx, yi (inner product), 6 Topelitz
hx | yi (inner product), 73 extension, 93
V + (positive cone in an ordered vector space), 57 operator, 93
|a| (absolute value in a C ∗ -algebra), 58 addition
kxk (norm of x), 7 on the Grothendieck group, 101
k kcb (norm of completely bounded map), 96 adjoint, 75
k ku (uniform norm), 14 as a functor, 44
A⊥ (annihilator of a set), 63 index of, 84
A⊥ (orthogonal complement of a set), 9 of a bounded linear Banach space map, 36
A+ (positive cone), 57 of a bounded linear Hilbert space map, 43
c+ (positive part of c), 57 of a multiplication operator, 44
c− (negative part of c), 57 of an integral operator, 44
≤ (notation for a partial ordering), 57 of an operator on an inner product space, 9
a ≤ b (order relation in a C ∗ -algebra), 57 existence in HS, 43
g . f (relation between operator valued maps), 96 of the unilateral shift, 44
p  q (ordering of projections in a ∗ -algebra), 64 adjointable, 75
(a, b) ∼ (c, d) (equivalence in Grothendieck AdU (conjugation by U ), 91
construction), 101 AF -algebra
p ∼ q (Murray-von Neumann equivalence), 99 C(K) is an (K is the Cantor set), 110
φ0 ∼h φ1 (homotopy between ∗ -homomorphisms), AF-algebra, 107
104 Alaoglu’s theorem, 29
p ∼h q (homotopy equivalence of points), 85, 99 algebra, 19, 73
p ∼◦ q (Murray-von Neumann equivalence), 100 C ∗ -, 46
p ∼s q (stable equivalence), 103 Banach, 23

115
116 INDEX

Calkin, 83 space, 13
commutative, 19 L1 (S) as a, 14
homomorphism, 19 C(X) as a, 14
multiplier, 79 C0 (X) as a, 14
norm, 23 Cb (X) as a, 14
normed, 23 l1 as a, 13
opposite, 73 basis
quotient, 21 for a Hilbert space, 61
simple, 20 orthonormal, 61
Toeplitz, 89 Bessel’s inequality, 61
unital, 19 bilinear, 73
algebraic bounded
∗-ideal, 51 completely, 96
dual space, 9 linear map, 16
ideal, 51 unitary, 87
A-linear, 75 sesquilinear functional, 43
A-module, 73, 74 B(V )
Hilbert, 75 as a C ∗ -algebra, 46
A as a, 75 as a Banach space, 17
inner product, 74 as a unital algebra, 20
morphism, 75 as a unital Banach algebra, 23
pre-Hilbert, 74 operators on a normed linear space, 16
semi-inner product, 74 B(V, W )
analytic, 25 bounded linear maps between normed spaces, 16
annihilator, 63, 77 Bratteli diagram, 108
anti-isomorphism, 62, 73
antihomomorphism, 73 Calkin algebra, 83
approximate CAR-algebras, 111
identity, 59 category, 35
existence of an, 59 N, 35
sequential, 59 BAN1 , 35
unit, 59 BAN∞ , 35
approximately finite dimensional C ∗ -algebra, 107 BA1 , 35
Atkinson’s theorem, 83 BA∞ , 35
A-valued (semi-)inner product, 74 CBA1 , 35
CBA∞ , 35
BA1 CMS1 , 42
Banach algebras and contractive homomorphisms, CSA, 47
35 CpH, 35
products in, 41 HIL, 35
BA∞ MS1 , 42
Banach algebras and continuous homomorphisms, SET, 35
35 UCBA1 , 35
products in, 41 UCBA∞ , 35
ball UCSA, 47
closed, 14 concrete, 37
open, 14 geometric, 35
BAN1 monoid, 35
Banach spaces and contractive linear maps, 35 topological, 35
coproducts in, 41 C2 (pairs of objects and morphisms), 38
BAN∞ CBA1
Banach spaces and continuous linear maps, 35 commutative Banach algebras and contractive
coproducts in, 40 homomorphisms, 35
Banach CBA∞
algebra, 23 commutative Banach algebras and continuous
C(X) as a, 23 homomorphisms, 35
C0 (X) as a, 23 character, 27
Cb (X) as a, 23 space, 29
B(V ) as a, 23 of C(X), 30
direct sum, 23 of l1 (Z), 29
homomorphism of, 23 closed
 INDEX 117

ball, 14 as a C ∗ -algebra, 46
inverse, 60 as a Banach algebra, 23
linear span, 14 as a Banach space, 14
CMS1 as a nuclear C ∗ -algebra, 97
complete metric spaces and contractive maps, 42 as a unital algebra, 20
co-universal continuous functions on X, 14
morphism, 41 Cb (X)
Coburn’s theorem, 90 as a Banach algebra, 23
codimension, 4, 82 as a Banach space, 14
cofunctor, 36 bounded continuous function on X, 14
cokernel, 82 C0 (X)
commutative, 19 as a C ∗ -algebra, 46
commutator, 87 as a Banach algebra, 23
compact as a Banach space, 14
operator, 66 as a nonunital algebra, 20
compactification, 78, 79 continuous functions vanishing at infinity, 14
essential, 78, 79 contractible
compalent, 87 C ∗ -algebras, 104
compatibility topological space, 105
of orderings with operations, 57 contravariant functor, 36
complement conventions
of a vector subspace, 4 A is a subset of A ⊕ B, 76
orthogonal, 9 about dual space, 17
complete about homomorphisms
order, 36 between Banach algebras, 23
orthonormal set, 62 about morphisms
completely between Banach algebras, 23
bounded, 96 about sesquilinear, 73
positive, 95 about universality, 41
lifting property, 97 algebras are complex, 19
completion categories are concrete, 37
of a metric space, 41 Ext A consists of extensions or morphisms, 93
of an inner product space, 42 Hilbert spaces are separable and infinite
universality of, 42 dimensional (after section 9.2), 90
C ideals are closed, 51
as a C ∗ -algebra, 46 ideals are self-adjoint, 51
Cn ideals are two-sided, 20
as a Hilbert space, 13 in an algebra AB denotes the span of products, 77
components projections in Hilbert spaces are orthogonal, 63
path, 85, 99 representations of unital C∗-algebras are unital, 70
composition subspaces are closed, 14
of morphisms, 35 vector spaces are complex, 3
concatenation, 39 convergence
concrete category, 37 in the w∗ -topology, 28
cone, 57 of a series, 61
positive, 57 convex, 14
proper, 57 convolution
conjugate in L1 (R), 24
linear, 6, 62 in l1 (Z), 24
symmetric, 15, 66 coproduct, 40
conjugation, 62, 91 in BAN1 , 41
connected in BAN∞ , 40
by a path, 85 in HIL, 40
construction in SET, 40
Gelfand-Naimark-Segal, 70 covariant functor, 36
continuity CpH
functor, 37 compact Hausdorff spaces and continuous maps, 35
of K0 , 108 CSA
C (the functor), 30, 37 C ∗ -algebras and ∗-homomorphisms, 47
C(X) products in, 52
118 INDEX

C ∗ -algebra, 46 ET (C ∗ -algebra generated by T and K(H)), 91

L∞ (S) as a, 46 eigenspace, 5
C as a, 46 eigenvalue, 5
C(X) as a, 46 eigenvector, 5
C0 (X) as a, 46 embedding, 101
B(H) as a, 46 C ∗ -algebraic, 78
L(V ) as a, 75 topological, 79
direct sum of, 52 EM N (projection along M onto N ), 4
generated by a set, 47 empty word, 39
Mn as a, 46 entire, 25
C ∗ -condition, 46 epimorphism, 37
C ∗ -embedding, 78 equivalence
C ∗ -norm, 46 essential unitary, 87
C ∗ -subalgebra, 46 homotopy
C ∗ (S) (C ∗ -subalgebra generated by S), 47 of ∗ -homomorphisms, 104
C ∗ -algebra of points, 99
AF-algebras, 107 Murray-von Neumann, 99, 100
CAR-algebras, 111 natural, 38
Fibonacci algebra, 111 of extensions, 91
inductive limit, 107 stable, 103
Mn (A) as a, 94 strong
nuclear, 96 of extensions, 52
cyclic unitary, 10, 87, 99
representation, 70 of ∗ -monomorphisms, 92
vector, 70 essential
compactification, 78, 79
D(A) (set of equivalence classes of projections), 101 ideal, 77
decomposition spectrum, 87
Jordan, 57 unitization, 78
polar, 65, 99 maximal, 79
Wold, 90 essentially
normal, 87
∆ (the functor), 30
self-adjoint, 87
∆(A)
unitarily equivalent, 87
character space of A, 27
evaluation
diagonal
functional, 28
functor, 38
EX
operator, 3, 18
evaluation map, 30
diag(a1 , . . . , an ) (diagonal matrix in Mn (A)), 100
EX x or Ex (evaluation functional at x), 28
diagonalizable
evaluation map, 30
operator, 3
eventually, 13
unitarily, 10
exact
diagram
functor, 106
Bratteli, 108
sequence, 51
dilation
split, 51
theorem of Stinespring, 96
Ext A, Ext X
direct
equivalence classes of extensions, 91
limit, 107 is an Abelian group, 97
direct sum extension, 51
extension, 52 determined by an essentially normal operator, 91
external, 8 direct sum, 52
internal, 4 equivalence of, 91
of C ∗ -algebras, 52 of K(H) by A, 90
of Banach algebras, 23 strong equivalence of, 52
of Banach spaces, 40 Toeplitz, 89
of Hilbert space operators, 71 abstract, 93
of Hilbert spaces, 71 external
of representations, 71 direct sum, 8, 52
orthogonal, 8
dual space, 17 F(H) (Fredholm operators on H), 83
algebraic, 9 Fn (H), Fn (Fredholm operators of index n), 84
 INDEX 119

faithful representation, 70 Gelfand transform

Fibonacci algebra, 111 of an element, 31
final on a Banach algebra, 31
projection, 64 Gelfand-Mazur theorem, 26
space, 65 Gelfand-Naimark
finite Theorem I, 47
dimensional Theorem II, 54
approximately, 107 Theorem III, 71
rank, 65 Gelfand-Naimark-Segal construction, 70
Fin(S) (finite subsets of S), 28 geometric category, 35
five lemma, 51 Grothendieck
forgetful functor, 37 construction
Fourier functorial property of, 102
coefficient, 33, 62 group, 101
expansion, 62 map, 101
series, 33 naturality of, 103
absolutely convergent, 33 universal property of, 102
transform group
on L1 ([−π, π]), 33 fundamental, 90
on L1 (R), 32 Grothendieck, 101
FR(V ) (operators of finite rank on V ), 65
Fredholm H 2 , 88
index (see index), 83 H ∞ , 88
operator, 83 H(A) (self-adjoint elements of a ∗ -algebra), 45
Fredholm alternative Hahn-Banach theorem, 25
version I, 81 half exact functor, 106
version II, 81 Hardy spaces
version IIIa, 82 H 2 , 88
version IIIb, 82 H ∞ , 88
version IV, 83 Hartman-Wintner spectral inclusion theorem, 88
version V, 84 hereditary, 59
version VI, 84 Hermitian, 45
free linear functional, 69
monoid, 39 operator, 10
semigroup, 39 HIL
vector space, 39 coproducts in, 40
full subcategory, 42 Hilbert spaces and bounded linear maps, 35
functional products in, 41
linear, 9 Hilbert
functor A-module, 75
K0 , 105 A as a, 75
∆ as a, 30 morphism, 75
C as a, 30 space, 13
adjoint, 44 L2 (S) as a, 13
continuity, 37 Cn as a, 13
contravariant, 36 l2 as a, 13
covariant, 36 Hom(A, B) (set of ∗ -homomorphisms between
diagonal, 38 C ∗ -algebras), 104
exact, 106 homomorphism
forgetful, 37 algebra, 19
Grothendieck, 102 of Banach algebras, 23
half exact, 106 unital, 19
second dual, 37 homotopic
split exact, 106 ∗ -homomorphisms, 104
fundamental equivalence
group, 90 of ∗ -homomorphisms, 104
theorem of linear algebra, 10 points in a topological space, 85, 99
homotopy
G(S) (Grothendieck group of S), 101 invariance of K0 , 104
γS (Grothendieck map), 101
Gelfand topology, 29 Ic , 76
120 INDEX

ideal left, 36
∗ -, 45 of a morphism, 36
algebraic, 51 right, 36
algebraic ∗-, 51 invertible, 36
essential, 77 in an algebra, 19
in C(X), 27 left
in a C ∗ -algebra, 51 in an algebra, 19
in an algebra, 20 linear map
left, 20 between vector spaces, 3
left modular, 50 operator
maximal, 20 index of, 83
minimal, 20 polar decomposition of, 65
modular, 50 right
principal, 20, 77 in an algebra, 19
proper, 20 inv A (invertible elements in an algebra), 19
right, 20 involution, 44
right modular, 50 is an isometry, 46
trivial, 20 isometry
idempotent, 22, 27 in a unital C ∗ -algebra, 100
identity proper, 90
approximate, 59 isomorphism
left in a category, 36
with respect to an ideal, 50 natural, 38
operator, 17 of vector spaces, 3
idV or IV or I (identity operator), 3 order, 59
index
of a Fredholm operator, 83 JB (natural injection), 28
of a normal Fredholm operator, 84 Jc (principal (closed) ideal containing c), 77
JC (continuous functions vanishing on C), 27
of an invertible operator, 83
Jordan decomposition, 57
of the adjoint of an operator, 84
of the unilateral shift, 84
K(H)
induced
as a nuclear C ∗ -algebra, 97
partial ordering, 57 as minimal ideal in B(H), 67
inductive compact operators on a Hilbert space, 66
limit, 107 K(V, W ), K(V ), 76
sequence, 107 K0
inequality as a covariant functor, 105
Bessel, 61 continuity of, 108
Kadison, 95 homotopy invariance of, 104
Schwarz, 7, 16, 69 universal property of, 103
infimum K0 (A)
of projections in a ∗ -algebra, 64 for nonunital A, 105
initial for unital A, 103
projection, 64 is 0 when A = B(H), 105
space, 65 is Z when A = C(X), X contractible, 105
inner product, 6 is Z when A = Mn , 105
A-module, 74 is Z when A = K(H), 108
A-valued, 74 K0 (φ)
semi-, 16 nonunital case, 105
space unital case, 104
l as a, 13 Kadison’s inequality, 95
integrable, 14 kernel
integral operator, 18 left, 70
adjoint of an, 44 of a linear map, 3
compactness of, 66 of an integral operator, 18
internal
direct sum, 4 L(V )
orthogonal direct sum, 8 as a C ∗ -algebra, 75
inverse L(V, W ), L(V ) (adjointable maps), 75
closed, 60 l1
 INDEX 121

as a Banach space, 13 space, 29

l2 , 6 for L1 (R), 32
basis for, 61 of C(X), 30
as a Hilbert space, 13 of l1 (Z), 29
lc measurable, 13
as an inner product space, 13 metric
Λ (members of A+ in the open unit ball), 59 induced by a norm, 7
lattice, 36 metric space
order complete, 36 completion of, 41
L2 (S) minimal ideal, 20
as a Hilbert space, 13 minimizing vector theorem, 15
basis for, 61 modular
L1 (S) ideal, 50
as a Banach space, 14 left ideal, 50
L∞ (S) right ideal, 50
as a C ∗ -algebra, 46 module, 73, 74
left inner product A-, 74
ideal, 20 morphism, 75
inverse pre-Hilbert A-, 74
of a morphism, 36 semi-inner product A-, 74
kernel, 70 monoid, 35, 39
multiplication operator, 47 free, 39
length, 7 monomorphism, 31, 37
limit morphism, 35
direct, 107 co-universal, 41
inductive, 107 Hilbert A-module, 75
limAj (inductive limit), 107 universal, 39
−→
linear, 3 Mor(S, T ) (morphisms between objects, 35
A-, 75 MS1
conjugate, 6, 62 metric spaces and contractive maps, 42
functional, 9 multiplication operator
Hermitian, 69 adjoint of a, 44
multiplicative, 27 on C(X), 17
positive, 69 on L2 (S), 44
sesqui-, 6 polar decomposition of, 65
span multiplicative, 19
closed, 14 identity, 19
L(V ), 3 linear functional, 27
Liouville’s theorem, 25 multiplicity
of an algebra homomorphism, 108
Mφ (multiplication operator), 44 multiplier algebra, 79
Mn (A) Murray-von Neumann equivalence, 99, 100
n × n matrices of elements of an algebra, 20
as a C ∗ -algebra, 94 natural
as an algebra, 20 equivalence, 38
Mn injection, 28
n × n matrices of complex numbers, 20 is a linear isometry, 28
as a C ∗ -algebra, 46 isomorphism, 38
as a nuclear C ∗ -algebra, 97 transformation, 38
as a unital algebra, 20 Grothendieck map is a, 103
Mn,m (A) (matrices with entries from A), 100 negative
M (A) (multiplier algebra of A), 79 part
matrix of a self-adjoint element, 57
Toeplitz, 88 Neumann series, 24
units, 95 nilpotent, 27
maximal nondegenerate
essential unitization, 79 ∗ -homomorphism, 79
Max A representation, 70
homeomorphic to ∆(A), 29 norm, 7
set of maximal ideals in an algebra, 20 C ∗ -, 46
maximal ideal, 20 algebra, 23
122 INDEX

of a continuous linear map, 17 abstract, 93

uniform, 9, 14 unilateral shift, 17
normal adjoint of, 44
element, 45 index of, 84
essentially, 87 unitary, 10
Fredholm operator Volterra, 18
index of, 84 and integral equations, 26
operator, 10 compactness of, 66
conditions for unitary equivalence, 11 with non-closed range, 82
polar decomposition of, 65 zero, 17
N(A) (normal elements of a ∗ -algebra), 45 opposite algebra, 73
normed order
algebra, 23 complete lattice, 36
linear space, 7 isomorphism, 59
vector space, see normed linear space ordered
n-positive, 95 vector space, 57
nuclear, 96 ordering
C(X) is, 97 partial, 57
Mn is, 97 of projections, 64
K(H) is, 97 pre-, 57
commutative C ∗ -algebras are, 97 orthogonal, 8
finite dimensional algebras are, 96 complement, 9
Hilbert space projections, 64
objects, 35 projection
open in an inner product space, 10
ball, 14 projections
operator, 16 in a ∗ -algebra, 64
adjoint, 36, 43 resolution of the identity, 10
index of, 84 orthonormal
compact, 66 basis, 61
diagonal, 3, 18
diagonalizable, 3 Pn (A), P∞ (A) (projections in matrix algebras), 100
essentially normal, 87 parallelogram law, 8
essentially self-adjoint, 87 Parseval’s identity, 62
Fredholm, 83 partial
Hermitian, 10 isometry, 64
identity, 17 final projection of a, 64
integral, 18 initial projection of a, 64
adjoint of an, 44 range projection of a, 64
compactness of, 66 support projection of a, 64
invertible ordering, 57
index of, 83 induced by a proper convex cone, 57
polar decomposition of, 65 of projections, 64
left multiplication, 47 path, 85
multiplication components, 85
adjoint of a, 44 points connected by a, 85
on C(X), 17 path components, 99
on L2 (S), 44 perpendicular, 8
polar decomposition of, 65 perturbation, 83
normal, 10 φ∞ (a character on the unitization of an algebra), 50
polar decomposition of, 65 φe (extension of a character φ to the unitization of an
normal Fredholm algebra), 50
index of, 84 π (quotient map from B(H) to Q(H)), 83
of finite rank, 65 π(T ) (element of the Calkin algebra containing T ), 83
on a normed linear space, 8 picture (see standard picture), 103, 106
on a vector space, 3 PM (orthogonal projection onto M ), 10
orthogonal projection, 10 point spectrum, 5
positive, 65 point-norm topology, 104
Riesz-Schauder, 82 polar decomposition, 65
self-adjoint, 10 of a multiplication operator, 65
Toeplitz, 88 of an invertible C ∗ -algebra element, 99
 INDEX 123

or a normal operator, 65 ∗ -algebra, 45

or an invertible operator, 65 C ∗ -algebra, 59
polarization identity, 9, 16 algebra, 21
positive Banach algebra, 26
n-, 95 quotient map, the, 21
completely, 95
cone, 57 range
in an ordered vector space, 57 of a linear map, 3
element (of an ordered vector space), 57 projection, 64
element of a C ∗ -algebra, 57 rank, 65
linear functional, 69 reflexive, 28
operator, 65 representation, 69
part cyclic, 70
of a self-adjoint element, 57 cyclic vector for a, 70
semidefinite, 16, 66 faithful, 70
pre-Hilbert nondegenerate, 70
A-module, 74 resolution of the identity
preordering, 57 in vector spaces, 4
principal orthogonal, 10
ideal, 77 resolvent
principal ideal, 20 mapping, 25
product, 41 Ra (resolvent mapping), 25
in BA1 , 41 respects operations, 57
in BA∞ , 41 ρA (a) or ρ(a) (spectral radius), 26
in CSA, 52 Riesz-Fréchet theorem, 18
in HIL, 41 finite dimensional version, 9
in SET, 41 Riesz-Schauder operator, 82
inner, 6 right
projection ideal, 20
final, 64 inverse
in a vector space, 4 of a morphism, 36
initial, 64
on a ∗ -algebra, 63 S (unilateral shift operator), 17
orthogonal, 10, 64 scalar
range, 64 element, 106
support, 64 mapping, 106
projections Schwarz inequality, 7, 16, 69
difference of, 64 second dual functor, 37
infimum of, 64 segment, 14
ordering of, 64 self-adjoint
orthogonality of, 64 element, 45
product of, 64 spectrum of, 47, 54
sum of, 64 essentially, 87
supremum of, 64 operator, 10
P(A) (projections on an algebra), 63 subset of a ∗ -algebra, 45
P(H), P(B(H)) (orthogonal projections on Hilbert semi-inner product, 16
space), 63 A-module, 74
proper semidefinite
cone, 57 positive, 16, 66
ideal, 20 semigroup
isometry, 90 free, 39
pullback, 91 seminorm, 7
Pythagorean theorem, 8 semisimple, 31
semisplit, 93
Q(H) (the Calkin algebra), 83 separating family, 31
QA (quasi-invertible elements, 56 separation
q-spectrum, 56 of points, 31
quadratic form, 16 sequence
quasi-inverse, 55 exact, 51
quasinilpotent, 31 inductive, 107
quotient short exact, 51
124 INDEX

split exact, 51 ∗ -homomorphism, 45

sequential unital, 45
approxiamte identity, 59 ∗ -ideal, 45
sesquilinear, 6, 15, 66, 73 ∗ -isomorphism, 45
SET ∗ -subalgebra (self-adjoint subalgebra), 45
coproducts in, 40 state, 69
products in, 41 vector, 69
sets and functions, 35 Stinespring
short exact sequence, 51 dilation theorem, 96
σA (a) or σ(a) (spectrum of a), 21 Stone-Weierstrass theorem, 47
σe (T ) (essential spectrum of T ), 87 structure
similar, 3 space, 29
simple subalgebra, 19
algebra, 20 ∗ -, 45
space C ∗ -, 46
algebraic dual, 9 unital, 19
Banach, 13 subcategory, 42
character, 29 full, 42
final, 65 subspace
Hardy, 88 of a Banach space, 14
Hilbert, 13 sum of subspaces need not be a, 83
initial, 65 summable
maximal ideal, 29 absolutely, 13
normed linear, 7 sequence, 24
structure, 29 set of vectors, 61
support, 65 square, 6
spectral support
mapping theorem, 48 projection, 64
radius, 26 space, 65
formula for, 26, 49 supremum
spectral theorem of projections in a ∗ -algebra, 64
abstract, 48 symbol
for normal operators, 10 of a Toeplitz operator, 88
of an abstract Toeplitz operator, 93
vector space version, 5
symmetric
spectrum, 21
conjugate, 15, 66
essential, 87
in nonunital algebras, 49 Tφ (Toeplitz operator with symbol φ), 88
of a bounded continuous function, 21 Ta (abstract Toeplitz operator), 93
of a complex number, 21 T (unit circle in the complex plane), 32
of a continuous function, 21 T (Toeplitz algebra), 89
of a self-adjoint element, 47, 54 Θv,w , 75
of a unitary element, 47 Toeplitz
of an element whose square is 1, 21 algebra, 89
of an essentially bounded function, 21 extension, 89
of an idempotent element, 22 abstract, 93
point, 5 index theorem, 90
q-, 56 matrix, 88
sphere, 14 operator, 88
split exact abstract, 93
functor, 106 symbol of, 88
sequence, 51 unilateral shift as a, 88
square integrable, 13 topological
square summable, 6 categories, 35
stable equivalence, 103 embedding, 79
standard topology
matrix units, 95 w∗ -, 28
picture of K0 (A) Gelfand, 29
for arbitrary A, 106 point-norm, 104
when A is unital, 103 total, 62
∗ -algebra, 45 transformation
 INDEX 125

natural, 38 vector
trivial space, 3, 73
ideal, 20 free, 39
normed, 7
U0 (A) (component of unitaries containing 1), 99 ordered, 57
UCBA1 state, 69
unital commutative Banach algebras and unit, 7
contractive homomorphisms, 35 VEC
UCBA∞ vector spaces and linear maps, 37
unital commutative Banach algebras and vector decomposition theorem, 15
continuous homomorphisms, 35 Voiculescu’s theorem, 96
UCSA Volterra operator, 18
unital C ∗ -algebras and unital ∗-homomorphisms, and integral equations, 26
47 compactness of, 66
uniform
norm, 9, 14 w(φ) (winding number of φ), 90
unilateral shift, 17 Weyl’s theorem, 87
adjoint of, 44 Weyl-von Neumann theorem, 87
essential normality of, 87 Wiener’s theorem, 33
index of, 84 winding number, 90
is a Toeplitz operator, 88 Wold decomposition, 90
unit, 19 word, 39
approximate, 59 empty, 39
vector, 7 w∗ -star topology
unital is Hausdorff, 28
∗ -homomorphism, 45 is weaker than the norm topology, 28
algebra, 19 w∗ -topology, 28
homomorphism, 19 base for, 28
subalgebra, 19
zero
unitarily
operator, 17
diagonalizable, 10
Zf (zero set of a function), 24
unitary
zero set, 24, 77
bounded linear map, 87
ζ (identity function on the unit circle), 88
element, 45
Zf (zero set of f ), 77
spectrum of, 47
equivalence
essential, 87
of ∗ -monomorphisms, 92
of operators, 10, 87
of projections, 99
operator, 10
U(A) (unitary elements of a ∗ -algebra), 45
unitization, 78
essential, 78
maximal essential, 79
of a ∗ -algebra, 49
of a C ∗ -algebra, 54
of a Banach ∗ -algebra, 49
of a normed algebra, 49
units
matrix, 95
universal
morphism, 39
object, 39
uniqueness of, 41
property, 39
of K0 (unital case), 103
of completions, 42
of the Grothendieck map, 102

vanish at infinity, 14