0% found this document useful (0 votes)

69 views48 pages

The Basics of C - Algebras

The document provides an introduction to Banach algebras and C*-algebras. It defines what a Banach algebra is and provides examples including the algebra of continuous functions on a compact Hausdorff space and the algebra of matrices. It introduces key concepts such as the spectrum and resolvent of an element in a Banach algebra. It proves properties of the spectrum including that it is always non-empty and compact. It also introduces the concept of a unital Banach algebra and proves related properties.

Uploaded by

YAGO PEREIRA DOS ANJOS SANTOS

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

69 views48 pages

The Basics of C - Algebras

Uploaded by

YAGO PEREIRA DOS ANJOS SANTOS

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 48

Chapter 1

The Basics of C ∗ -algebras

1.1 Banach algebras

Definition 1.1.1 A normed algebra is a complex algebra A which is a

normed space, and the norm satisfies

kabk ≤ kakkbk for all a, b ∈ A.

If A (with this norm) is complete, then A is called a Banach algebra.

Every closed subalgebra of a Banach algebra is itself a Banach algebra.

Example 1.1.2 Let C be the complex field. Then C is a Banach algebra.

Let X be a compact Hausdorff space and C(X) the set of continuous func-
tions on X. C(X) is a complex algebra with pointwise operations. With
kf k = supx∈X |f (x)|, C(X) is a Banach algebra.

Example 1.1.3 Let Mn be the algebra of n × n complex matrices. By

identifying Mn with B(Cn ), the set of all (bounded) linear maps from the
n-dimensional Hilbert space Cn to Cn , with operator norm, i.e, kxk =
supξ∈Cn ,kξk≤1 kx(ξ)k, we see that Mn is a Banach algebra.

Example 1.1.4 The set of continuous functions A(D) on the closed unit
disk D in the plane which are analytic on the interior is a closed subalgebra
of C(D). Therefore, A(D) is a Banach algebra.

Example 1.1.5 Let X be a Banach space and B(X) be the set of all
bounded linear operators on X. If T, L ∈ B(X), define T L = T ◦ L. Then
B(X) is a complex algebra. With operator norm, B(X) is a Banach algebra.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
1
2 The Basics of C ∗ -algebras

A commutative Banach algebra is a Banach algebra A with the property

that ab = ba for all a, b ∈ A Examples 1.1.2 and 1.1.4 are of commutative
Banach algebras while Example 1.1.3 1.1.5 are not commutative.

Definition 1.1.6 In a unital algebra, an element a ∈ A is called invertible

if there is an element b ∈ A such that ab = ba = 1. In this case b is unique
and written a−1 . The set

GL(A) = {a ∈ A : a is invertible}

is a group under multiplication.

We define the spectrum of an element a to be the set

sp(a) = spA (a) = {λ ∈ C : λ1 − a 6∈ GL(A)}.

Whenever there is no confusion, we will write λ1 simply as λ.

The complement of the spectrum is called the resolvent and R(λ) =
(λ − a)−1 is the resolvent function.

Example 1.1.7 Let A = C(X) be as in 1.1.2. Then sp(f ) = f (X) for all
f ∈ A. In other words, the spectrum of f is the range of f.
Let A = Mn . If a = (aij ) ∈ A, then the reader can check that sp(a) is
the set of eigenvalues of the matrix a.

Proposition 1.1.8 For any a and b in A,

sp(ab) \ {0} = sp(ba) \ {0}.

Proof. If λ 6∈ sp(ab) and λ 6= 0, then there is c ∈ A such that

c(λ − ab) = (λ − ab)c = 1.

Thus c(ab) = λc − 1 = (ab)c. So we compute that

(1 + bca)(λ − ba) = λ − ba + λbca − bcaba

= λ − ba + b(λ − ab)ca = λ − ba + ba = λ,

which shows that λ−1 (1 + bca) is the inverse of λ − ba. Hence λ 6∈ sp(ba)
and sp(ba) \ {0} ⊂ sp(ab) \ {0}.

Definition 1.1.9 A Banach algebra A is said to be unital if it admits a

unit 1 and k1k = 1. Banach algebras in 1.1.2, 1.1.3 and 1.1.4 are unital.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Banach algebras 3

Lemma 1.1.10 Let A be a unital Banach algebra and a be an element of

A such that k1 − ak < 1. Then a ∈ GL(A) and
∞
X
a−1 = (1 − a)n .
n=0

1 k1−ak
Moreover, ka−1 k ≤ 1−k1−ak and k1 − a−1 k ≤ 1−k1−ak .

Proof. Since
∞ ∞
X X 1
k(1 − a)n k ≤ k1 − akn = < ∞,
n=0 n=0
(1 − k1 − ak)
P∞
the series n=0 (1 − a)n is convergent. Let b be its limit in A. Then kbk ≤
1
(1−k1−ak) and
∞
X k1 − ak
k1 − bk ≤ k1 − akn = .
n=1
1 − k1 − ak

One verifies
k
X k
X
a( (1 − a)n ) = (1 − (1 − a))( (1 − a)n ) = 1 − (1 − a)k+1
n=0 n=0

and that it converges to ab = ba = 1 as k → ∞. Hence b is the inverse of a.

Definition 1.1.11 A function f from an open subset Ω ⊂ C to a Banach

algebra is said to be analytic, if for any λ0 ∈ Ω there is an open neigh-
P∞ n
borhood O(λ0 ) such that f (λ) = n=0 an (λ − λ0 ) converges for every
λ ∈ O(λ0 ). To include the case that λ0 = ∞, we say f is analytic at infinity
P∞
if f (λ) = n=0 an λ−n for all λ in a neighborhood of infinity.

Theorem 1.1.12 In any unital Banach algebra A, the spectrum of each

a ∈ A is a non-empty compact subset, and the resolvent function is analytic
on C \ sp(a).

Proof. If |λ| > kak, then kλ−n an k ≤ ( kak n

λ ) . So the series

∞
X an
n=0
λn+1

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
4 The Basics of C ∗ -algebras

converges (in norm). Similarly to 1.1.10,

k
X an ak+1
(λ − a) = 1 − ( )
n=0
λn+1 λk+1

which converges to 1. This shows that supλ∈sp(a) |λ| ≤ kak and R(λ) is
analytic in {λ : |λ| > kak}. Moreover,
|λ|−1
lim kR(λ)k ≤ lim = 0. (e 1.1)
|λ|→∞ |λ|→∞ 1 − ( kak
λ )
1
Similarly, if λ0 − a is invertible and |λ − λ0 | < k(λ0 −a)−1 k , then
∞
X
(λ − a)−1 = (λ0 − λ)n ((λ0 − a)−1 )n+1 .
n=0

This also shows that the resolvent is open. Since sp(a) has been shown to be
bounded, sp(a) is compact. We have also shown that the resolvent function
is analytic on the complement of the spectrum.
In particular, f (R(λ)) is a (scalar) analytic function for every bounded
linear functional f ∈ A∗ . If sp(a) were empty, then f (R(λ)) would be an
entire function for every f ∈ A∗ . However, (e 1.1) shows that f (R(λ)) is
bounded on the plane. Thus Liouville’s theorem implies that f (R(λ)) is a
constant. But (e 1.1) also implies that f (R(λ)) = 0. So, by the Hahn-Banach
theorem, R(λ) = 0. This is a contradiction. Hence sp(a) is not empty.
Corollary 1.1.13 The only simple commutative unital Banach algebra
is C.
Proof. Suppose that A is a unital commutative Banach algebra and a ∈
A is not a scalar. Let λ ∈ sp(a). Set I = (a − λ)A. Then I is clearly a closed
ideal of A. No element of the form (a − λ)b is invertible in the commutative
Banach algebra A. By 1.1.10,
k(a − λ)b − 1k ≥ 1.
So 1 6∈ I and I is proper. Therefore, if A is simple, a must be a scalar,
whence A = C.
Lemma 1.1.14 If p is a polynomial and a is an element of a unital
Banach algebra A, then
sp(p(a)) = p(sp(a)).

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Banach algebras 5

Proof. We may assume that p is not constant. If λ ∈ C, there are

c, β1 , ..., βn ∈ C such that
n
Y
p(z) − λ = c (z − βi ),
i=1

and therefore
n
Y
p(a) − λ = c (a − βi ).
i=1

It is clear that p(a)−λ is invertible if and only if each a−βi is. It follows that
λ ∈ sp(p(a)) if and only if λ = p(α) for some α ∈ sp(a). Thus sp(p(a)) =
p(sp(a)).
Definition 1.1.15 Let A be a unital Banach algebra. If a ∈ A, its spectral
radius is defined to be
r(a) = sup |λ|.
λ∈sp(a)

Theorem 1.1.16 If a is an element in a unital Banach algebra A, then

r(a) = lim kan k1/n .
n→∞

Proof. If λ ∈ sp(a), then λ ∈ sp(an ) by 1.1.14, so |λn | ≤ kan k. There-

fore, r(a) ≤ lim inf n→∞ kan k1/n . Let Ω be the open disk in C with center
0 and radius 1/r(a) (or ∞ if r(a) = 0). If λ ∈ Ω, then 1 − λa ∈ GL(A). If
f ∈ A∗ , then f ((1 − λa)−1 ) is analytic. There are unique complex numbers
zn such that
∞
X
f ((1 − λa)−1 ) = zn λn (λ ∈ Ω).
n=0

However, if |λ| < 1/kak ≤ 1/r(a), then kλak < 1, so (1 − λa)−1 =

P∞ n n −1
P∞
n=0 λ a , and therefore, f ((1 − λa) ) = n=0 λn f (an ). It follows that
zn = f (an ) for all n ≥ 0. Hence the sequence {λn f (an )} converges to zero
for each λ ∈ Ω, and therefore is bounded. Since this is true for every f ∈ A∗ ,
by the principle of uniform boundedness, {λn an } is a bounded sequence.
So we may assume that |λn |kan k ≤ M for all n ≥ 0 and for some positive
number M. Hence
M 1/n
kan k1/n ≤ , n = 0, 1, ....
|λ|

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
6 The Basics of C ∗ -algebras

Consequently,

lim sup kan k1/n ≤ 1/|λ|.

n→∞

1
This implies that if r(a) < |λ| , then lim supn→∞ kan k1/n ≤ 1/|λ|. It fol-
lows that lim supn→∞ kan k1/n ≤ r(a). From what we have shown at the
beginning of this proof, we obtain

lim sup kan k1/n ≤ r(a) ≤ lim inf kan k1/n .

n→∞ n→∞

This implies that r(a) = limn→∞ kan k1/n .

 
1/2 1 0
Example 1.1.17 Let A = M3 and a =  0 1/2 0  . Then kak ≥
0 0 1/3
1 and sp(a) = {1/2, 1/3}. So r(a) = 1/2. It follows that kan k1/n → 1/2.
Let T ∈ B(L2 ([0, 1])) be a bounded linear operator defined by
Z t
T (f ) = f (x)dx.
0
n 1
The reader can compute that kT k ≤ n! . Hence r(T ) = 0. Note that T 6= 0
(see Exercise 1.11.3).
We now establish the holomorphic functional calculus for elements in
Banach algebras (1.1.19). The first application appears in 1.2.9. Later, we
will establish continuous functional calculus for commutative C ∗ -algebras
(1.3.5) and Borel functional calculus for normal elements in von Neumann
algebras (1.8.5).
Definition 1.1.18 Let x be a fixed element in a unital Banach algebra
A. Let f be a holomorphic function in an open neighborhood Of of sp(x),
and C be a smooth simple closed curve in Of enclosing sp(x). We assign
the positive orientation to C as in complex analysis. For each φ ∈ A∗ , we
consider a continuous function which maps λ to f (λ)φ((λ − x)−1 ) on the
curve C. Set
1
Z
L(φ) = f (λ)φ((λ − x)−1 )dλ.
2πi C
The map φ 7→ L(φ) is a linear functional on A∗ and
l
|L(φ)| ≤ kφk sup{|f (λ)|k(λ − x)−1 k : λ ∈ C},
2π

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Banach algebras 7

where l is the length of the curve C. Hence there exists an F ∈ A∗∗ such
that F (φ) = L(φ).
On the other hand, the function λ 7→ f (λ)(λ − x)−1 is a continuous
function from C into A. So the limit of
n
X
f (λi )(λi − x)−1 (λi − λi+1 ) (as maxi |λi − λi+1 | → 0),
i=0

where {λ0 , ..., λn , λn+1 = λ0 } is a partition of the curve C, converges in

norm in A to y. By the continuity of φ, we know that φ(y) = L(φ) = F (φ)
for all φ ∈ A∗ . Hence F ∈ A. By Cauchy’s theorem, F does not depend on
the choice of the curve C, but only on the function f. Therefore we may
denote F by f (x) and write
1
Z
f (x) = f (λ)(λ − x)−1 dλ.
2πi C
We denote by Hol(sp(x)) the algebra of all functions which are holo-
morphic in a neighborhood of sp(x).
Theorem 1.1.19 Fix an element x in a unital Banach algebra A. The
map f 7→ f (x) from Hol(sp(x)) into A is a homomorphism which sends
the constant function 1 to the identity of A and the identity function on
C to the element x. If f (z) = ∞ n
P
P∞ n=0 cn z in a neighborhood of sp(x), then
n
f (x) = n=0 cn x .
Proof. Linearity is clear. Let f and g be functions holomorphic in neigh-
borhoods Of and Og of sp(x), respectively. Set O = Of ∩ Og and let
Ci , i = 1, 2 be smooth simple closed curves in O enclosing sp(x) such that
C1 lies completely inside the curve C2 . Then
1 1
Z Z
f (x)g(x) = ( f (λ)(λ − x)−1 dλ)( g(z)(z − x)−1 dz)
2πi C1 2πi C2
1
Z Z
= − 2 [ f (λ)g(z)(λ − x)−1 (z − x)−1 dz]dλ
4π C1 C2
1 [(z − x)−1 − (λ − x)−1 ]
Z Z
= − 2 f (λ)g(z) dzdλ
4π C1 C2 λ−z
1 1 g(z)
Z Z
= f (λ)(λ − x)−1 ( dz)dλ
2πi C1 2πi C2 z − λ
1 −1 1 f (λ)
Z Z
+ g(z)(z − x) ( dλ)dz
2πi C2 2πi C1 λ − z

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
8 The Basics of C ∗ -algebras

1
Z
= f (λ)g(λ)(λ − x)−1 dλ = (f · g)(x).
2πi C1

1
R g(z)
The second to last equality holds because 2πi C2 z−λ
dz = g(λ) (by the
f (λ)
Cauchy formula) and because is holomorphic inside the curve C2 if
λ−z
R f (λ)
λ ∈ C1 (so that C1 λ−z dλ = 0).
To complete the proof, pick a circle C with center at 0 and large radius.
We note that
∞
X ∞
X
λn−1 (1 − λ−1 x)−1 = λn−1 xk λ−k = xk λn−k−1 ,
n=0 k=0

where the convergence is in norm and is uniform on C. Therefore

Z Z
xk λn−k−1 dλ = ( λn−k−1 dλ)xk = 0 · xk = 0 (e 1.2)
C C

unless k = n, in which case the integral is 2πixn . This implies that

1
Z
n
x = λn (λ − x)−1 dλ. (e 1.3)
2πi C
Now suppose that f (z) = ∞ n
P
n=0 cn z in a neighborhood of sp(x). Then
it converges in an open disk with center 0. Let C be a circle with center at
0 contained in the open disk. Then the series converges uniformly on C, so
from (e 1.3),
∞ ∞
1 1
Z X Z X
−1 n −1
f (x) = f (z)(z−x) dz = cn ( z (z−x) dz) = cn xn .
2πi C n=0
2πi C n=0
Proposition 1.1.20 If I is a closed ideal in a Banach algebra, then A/I
is a Banach algebra with the quotient norm
kāk = ka + Ik = inf ka + bk.
b∈I

Proof. It is well known that A/I, as a normed space is complete. It

remains to show that A/I is a normed algebra. Let ε > 0 and a, b ∈ A.
Then, there are i, j ∈ I such that
ka + ik < ka + Ik + ε and kb + jk < kb + Ik + ε.
Hence, for c = ib + aj + ij ∈ I,
kab + ck ≤ ka + ikkb + jk < (ka + Ik + ε)(kb + Ik + ε).

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
C ∗ -algebras 9

Thus,

kab + Ik ≤ (ka + Ik + ε)(kb + Ik + ε).

Letting ε → 0, we obtain

kab + Ik ≤ ka + Ikkb + Ik.

In other words, A/I is a normed algebra.

1.2 C ∗ -algebras

Definition 1.2.1 An algebra A is called a ∗-algebra if it is a complex

algebra with a conjugate linear involution ∗ which is an anti-isomorphism,
i.e., for any a, b ∈ A and α ∈ C,

(a + b)∗ = a∗ + b∗ , (αa)∗ = ᾱa∗ , a∗∗ = a and (ab)∗ = b∗ a∗ .

If a ∈ A, then a∗ is called the adjoint of a. Let A be a ∗-algebra which is

also a normed algebra. A norm on A that satisfies

ka∗ ak = kak2

for all a ∈ A is called a C ∗ -norm. If, with this norm, A is complete, then A
is called a C ∗ -algebra.
Since kx∗ xk ≤ kx∗ kkxk we have kxk ≤ kx∗ k for all x ∈ A. Thus kx∗ k =
kxk.
A closed ∗-subalgebra of a C ∗ -algebra is also a C ∗ -algebra. Such a ∗-
subalgebra will be called a C ∗ -subalgebra.

Example 1.2.2 (a) Let A be as in 1.1.3 and let a = (aij ) ∈ A. Define

a∗ = (āji ). Then A is a C ∗ -algebra.
(b) Let X be a locally compact Hausdorff space and C0 (X) be the
set of all continuous functions vanishing at infinity. Define f ∗ (t) = f (t)
(for t ∈ X). Then C0 (X) becomes a ∗-algebra. With kf k = supt∈X |f (t)|,
C0 (X) is a C ∗ -algebra. C0 (X) is unital if and only if X is compact.

Example 1.2.3 Let X be a compact Hausdorff space and B(X) be the

set of all bounded Borel functions on X. With kf k = supx∈X |f (x)| and
f ∗ (x) = f (x), B(X) becomes a C ∗ -algebra. For any complex Borel measure

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
C ∗ -algebras 9

Thus,

kab + Ik ≤ (ka + Ik + ε)(kb + Ik + ε).

Letting ε → 0, we obtain

kab + Ik ≤ ka + Ikkb + Ik.

In other words, A/I is a normed algebra.

1.2 C ∗ -algebras

Definition 1.2.1 An algebra A is called a ∗-algebra if it is a complex

algebra with a conjugate linear involution ∗ which is an anti-isomorphism,
i.e., for any a, b ∈ A and α ∈ C,

(a + b)∗ = a∗ + b∗ , (αa)∗ = ᾱa∗ , a∗∗ = a and (ab)∗ = b∗ a∗ .

If a ∈ A, then a∗ is called the adjoint of a. Let A be a ∗-algebra which is

also a normed algebra. A norm on A that satisfies

ka∗ ak = kak2

Example 1.2.2 (a) Let A be as in 1.1.3 and let a = (aij ) ∈ A. Define

Example 1.2.3 Let X be a compact Hausdorff space and B(X) be the

∗
R on X, there is a bounded linear functional Fµ ∈ C(X) such that Fµ (f ) =
µ
X
f dµ. It is clear that (by considering point-evaluation)

sup |Fµ (f )| ≥ kf k
kµk≤1

for any f ∈ B(X). On the other hand, if kµk ≤ 1,

Z
|Fµ (f )| ≤ |f |d|µ| ≤ kf kkµk.
X

Thus the C ∗ -algebra B(X) is the same Banach space when we regard it as
a subspace of the second dual of C(X) as described above.
Definition 1.2.4 Let A be a C ∗ -algebra. An element x ∈ A is normal
if xx∗ = x∗ x and is self-adjoint if x = x∗ . The self-adjoint part of A is
denoted by Asa . For each x ∈ A, the element 12 (x + x∗ ) is in Asa , and is
called the real part of x, and the element − 2i (x − x∗ ) is in Asa and is called
the imaginary part of x. It follows that Asa is a closed real subspace of A
and each element x ∈ A has a unique decomposition as x = y + iz with
y, z ∈ Asa . An element p ∈ Asa is called a projection if p2 = p.
When A admits a unit, we denote the unit (or identity) by 1A , or 1
when there is no confusion. If A has a unit, then A 6= 0. Since 1∗A = 1A ,
k1A k = k1A k2 and k1A k = 1. Thus a C ∗ -algebra with unit is a unital
Banach algebra. We will call it a unital C ∗ -algebra.
An element u in a unital C ∗ -algebra is unitary if u∗ u = uu∗ = 1. Since
1∗ = 1, kuk = 1.
One can always unitize a C ∗ -algebra.
Proposition 1.2.5 For each C ∗ -algebra A there is a C ∗ -algebra Ã with
unit containing A as a closed ideal. If A has no unit, Ã/A = C.
Proof. Let B(A) be the set of all bounded linear operators on A. Consider
the map π : A → B(A) defined by π(a)b = ab for all a, b ∈ A. (This map
is often called the left regular representation of A.) It is clear that π is a
homomorphism. Since kπ(a)bk ≤ kakkbk, kπ(a)k ≤ kak. Also

kak2 = kaa∗ k = kπ(a)a∗ k ≤ kπ(a)kka∗ k = kπ(a)kkak.

Hence π is an isometry. Let 1 denote the identity operator on A and let Ã

be the algebra of operators on A of the form π(a) + λ · 1 with a ∈ A and
λ ∈ C. Since π(A) is complete and Ã/π(A) = C, Ã is also complete. Define

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
C ∗ -algebras 11

an involution on Ã by defining (π(a) + λ1)∗ = π(a∗ ) + λ̄1. For each ε > 0,

there is b ∈ A with kbk = 1 such that

kπ(a) + λ1k2 ≤ ε + k(a + λ)bk2 = ε + kb∗ (a∗ + λ̄)(a + λ)bk

≤ ε + k(a∗ + λ̄)(a + λ)bk ≤ ε + k(π(a∗ ) + λ̄)(π(a) + λ)k.

So Ã becomes a C ∗ -algebra.
Definition 1.2.6 For a non-unital C ∗ -algebra A, the spectrum of x ∈ A,
denoted by sp(x), is defined to be the spectrum of x in Ã.
Lemma 1.2.7 If A is a C ∗ -algebra and x ∈ A is a normal element, then

kxk = r(x).

Proof. If x ∈ Asa , then kx2 k = kxk2 implies that

n n
r(x) = lim kx2 k1/2 = kxk.
n→∞

If in general, x is normal, we obtain

r(x)2 ≤ kxk2 = kx∗ xk = lim k(x∗ x)n k1/n

n→∞

≤ lim (k(x∗ )n kkxn k)1/n ≤ r(x)2 .

n→∞

Hence r(x) = kxk.

Corollary 1.2.8 There is at most one norm on a ∗-algebra making it a
C ∗ -algebra.
Proof. If k · k1 and k · k2 are two norms on a ∗-algebra A making it a
C ∗ -algebra, then

kak2i = ka∗ aki = r(a∗ a) = sup{|λ| : λ ∈ sp(a∗ a)}

(i = 1, 2). Thus kak1 = kak2 .

Lemma 1.2.9 Let A be a C ∗ -algebra and a ∈ Asa . Then sp(a) ⊂ R. If
u ∈ A is a unitary, then sp(u) is a subset of the unit circle.
Proof. Let u be a unitary in a unital C ∗ -algebra A and λ ∈ sp(u).
Since kuk2 = ku∗ uk = k1k = 1, |λ| ≤ 1. However, λ−1 ∈ sp(u−1 ). Since
u−1 = u∗ is also a unitary, we conclude that |λ| = 1. For a ∈ Asa , by
considering Ã if necessary, we may assume that A is unital. The function
P
exp(iz) = ∞ n
n=0 (iz) /n! is an entire function. By applying 1.1.19, we see

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
12 The Basics of C ∗ -algebras

that u = exp(ia) is a unitary (with u∗ = exp(−ia)). If λ ∈ sp(a) and

P
b= ∞ n
n=1 (i) (a − λ)
n−1
/n!, then by 1.1.19,

exp(ia) − eiλ = (exp(i(a − λ)) − 1)eiλ = (a − λ)beiλ .

Since b commutes with a (by 1.1.19), and a−λ is not invertible, exp(ia)−eiλ
is not invertible. Hence |eiλ | = 1, and therefore λ ∈ R. In other words,
sp(a) ⊂ R.

1.3 Commutative C ∗ -algebras

The C ∗ -algebra C0 (X) in 1.2.2 (b) is a commutative C ∗ -algebra. In this

section, we will show that every commutative C ∗ -algebra has this form.

Definition 1.3.1 A multiplicative linear functional on a Banach algebra

A is a nonzero homomorphism of A into C. The set of all multiplicative
linear functionals on A is called the maximal ideal space of A and will be
denoted by Ω(A).
If A is a unital algebra, and I is a proper ideal of A, then an application
of Zorn’s lemma shows that there is an ideal M ⊃ I such that A/M is
simple and nonzero. Such ideals are called maximal ideal. We leave this to
the reader as an exercise.

Theorem 1.3.2 Let A be a unital commutative Banach algebra.

(1) If φ ∈ Ω(A), then kφk = 1.
(2) The space Ω(A) is non-empty, and the map φ 7→ kerφ defines a
bijection from Ω(A) onto the set of all maximal ideals of A.

Proof. Suppose that φ ∈ Ω(A) and a ∈ A such that kak < 1 = φ(a). Let
P∞
b = n=1 an . Then a + ab = b and

φ(b) = φ(a) + φ(a)φ(b) = 1 + φ(b).

This is not possible. So kφk ≤ 1. Since φ(1) = 1, we proved that kφk = 1.

For part (2), let φ ∈ Ω(A). It follows that M = kerφ is a closed ideal
of codimension 1 in A, and thus is maximal. If φ1 , φ2 ∈ Ω(A) and kerφ1 =
kerφ2 , then for each a ∈ A, a − φ2 (a) ∈ kerφ1 . This implies that φ1 (a −
φ2 (a)) = 0 or φ1 (a) = φ2 (a). This shows the map is one-to -one.
Conversely, if M is a maximal ideal, then dist(M, 1) ≥ 1 because the
open unit ball with center at 1 consists of only invertible elements (by

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
12 The Basics of C ∗ -algebras

that u = exp(ia) is a unitary (with u∗ = exp(−ia)). If λ ∈ sp(a) and

P
b= ∞ n
n=1 (i) (a − λ)
n−1
/n!, then by 1.1.19,

exp(ia) − eiλ = (exp(i(a − λ)) − 1)eiλ = (a − λ)beiλ .

Since b commutes with a (by 1.1.19), and a−λ is not invertible, exp(ia)−eiλ
is not invertible. Hence |eiλ | = 1, and therefore λ ∈ R. In other words,
sp(a) ⊂ R.

1.3 Commutative C ∗ -algebras

The C ∗ -algebra C0 (X) in 1.2.2 (b) is a commutative C ∗ -algebra. In this

section, we will show that every commutative C ∗ -algebra has this form.

Definition 1.3.1 A multiplicative linear functional on a Banach algebra

Theorem 1.3.2 Let A be a unital commutative Banach algebra.

(1) If φ ∈ Ω(A), then kφk = 1.
(2) The space Ω(A) is non-empty, and the map φ 7→ kerφ defines a
bijection from Ω(A) onto the set of all maximal ideals of A.

Proof. Suppose that φ ∈ Ω(A) and a ∈ A such that kak < 1 = φ(a). Let
P∞
b = n=1 an . Then a + ab = b and

φ(b) = φ(a) + φ(a)φ(b) = 1 + φ(b).

This is not possible. So kφk ≤ 1. Since φ(1) = 1, we proved that kφk = 1.

1.1.10). It follows that the closure of M still does not contain 1. It is easy
to see that the closure is an ideal. So it is a proper ideal. We conclude that
M itself is closed. So the quotient A/M is a simple commutative Banach
algebra. By Lemma 1.1.13 A/M = C. So this quotient map φ gives a
continuous homomorphism from A → C with kerφ = M. The map is
therefore bijective.
To see that Ω(A) is non-empty, we may assume that A 6= C, otherwise
the identification of A with C gives a nonzero homomorphism. So A is not
simple (by 1.1.13). Let I be a proper ideal of A. Since A has an identity,
there is a maximal proper ideal of A containing I (see 1.3.1).

If A is a unital commutative Banach algebra, it follows from 1.3.2 that

Ω(A) is a subset of the (closed) unit ball of A∗ . So Ω(A) with the relative
weak ∗ -topology becomes a topological space.

Theorem 1.3.3 If A is a unital commutative Banach algebra, then Ω(A)

is a compact Hausdorff space.

Proof. It is easy to check that Ω(A) is weak∗ closed in the closed unit
ball of A∗ . Since the unit ball of A∗ is weak∗ compact (Banach-Alaoglu
theorem), Ω(A) is weak∗ compact.

Definition 1.3.4 Suppose that A is a unital commutative Banach alge-

bra and Ω(A) is its maximal ideal space. If a ∈ A, we define a function ǎ
by

ǎ(t) = t(a) ( t ∈ Ω(A) ).

The set {φ ∈ Ω(A) : |φ(a)| ≥ α} is weak∗ closed (for every α > 0) in the
closed unit ball of A∗ . Thus it is weak∗ compact by the Banach-Alaoglu
theorem. Hence ǎ ∈ C(Ω(A)). Thus we define a map Γ : a → ǎ from A into
C(Ω(A)). This map is called the Gelfand transform. It is clear that Γ is a
homomorphism.
Assume A is a non-unital C ∗ -algebra and Ã is its unitization. If φ :
A → C is a nonzero homomorphism, then φ̃(a + λ) = φ(a) + λ (a ∈ A
and λ ∈ C) is a homomorphism from Ã into C. Thus Ω(A) is a subset of
Ω(Ã). In fact Ω(Ã) = Ω(A) ∪ {π}, where π is determined by Ã/A = C,
i.e., kerπ = A. Since Ω(Ã) is a compact Hausdorff space, we conclude that
Ω(A) is a locally compact Hausdorff space. The restriction of the Gelfand
transform of Ã on A maps A into C0 (Ω(A)).

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
14 The Basics of C ∗ -algebras

The following is the Gelfand theorem for commutative C ∗ -algebras. Note

that if A is unital, Ω(A) is compact and C0 (Ω(A)) = C(Ω(A)).
Theorem 1.3.5 Suppose that A is a commutative C ∗ -algebra. Then the
Gelfand transform a 7→ ǎ is a ∗-preserving isometry from A onto C0 (Ω(A)).
Proof. Suppose first that A is unital. Let t ∈ Ω(A) and a ∈ A then
t(a) − a ∈ ker t. Since kert is a maximal ideal, t(a) ∈ sp(a). Therefore, if a ∈
Asa , then t(a) ∈ R by 1.2.9. By writing a = (1/2)(a+a∗ )+i[(1/2i)(a−a∗ )],
we conclude that t(a∗ ) = t(a) for each a ∈ A. This shows that Γ(a) = ǎ is
a ∗-preserving homomorphism from A into C(Ω).
If a is invertible, then Γ(a−1 ) = Γ(a)−1 , since Γ is a homomorphism.
Conversely, if a is not invertible, then the ideal I = aA is proper and thus
is contained in a maximal ideal M (we still assume that A is unital). If
φ ∈ Ω(A) and kerφ = M, then φ(a) = 0. Thus ǎ is not invertible in C(Ω(A)).
This implies that ǎ and a have the same spectrum and this coincides with
the range of ǎ (see 1.1.7). Since the norm in C(Ω(A)) is the supremum norm
and the range of ǎ is sp(a), we conclude that kǎk∞ = r(a) = kak, since a is
normal (see 1.2.7). This says that Γ is a ∗-preserving isometry from A into
C(Ω(A)). Since points in Ω(A) are multiplicative linear functionals which
can be only distinguished by elements in A, {ǎ : a ∈ A} separates points in
Ω(A). It follows from the Stone-Weierstrass Theorem that Γ is surjective.
For the non-unital case, from the above, Γ : Ã → C(Ω(Ã)) is a surjective
∗-preserving isometry. Since Ã/A = C, Γ(A) is a maximal ideal. Note as
in 1.3.4, we may write Ω(Ã) = Ω(A) ∪ {π}, where π is determined by
the quotient map Ã → Ã/A = C. Then C0 (Ω(A)) is the (maximal) ideal
(of C(Ω(Ã))) of continuous functions vanishing at π, which is identified
naturally with C0 (Ω(A)).
Let S be a subset of a C ∗ -algebra A. Then the C ∗ -subalgebra of A
generated by S, denoted by C ∗ (S), is the smallest C ∗ -subalgebra of A
containing S.
Corollary 1.3.6 If a is a normal element in a unital C ∗ -algebra A, then
there is an isometric ∗-isomorphism from C ∗ (a) to C0 (sp(a) \ {0}) which
sends a to the identity function on sp(a).
Proof. Since a is normal, C ∗ (a) is commutative. Note that φ ∈ Ω(C ∗ (a))
is determined by φ(a) = λ. Thus the map from Ω(C ∗ (a)) into C taking φ
to φ(a) is a homeomorphism onto ǎ(Ω(A)). From the Gelfand transform,
ǎ(Ω(A)) = sp(a). This map identifies ǎ with the identity function z on sp(a)

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Commutative C ∗ -algebras 15

as desired. It follows from 1.3.5 that this map is an isometric ∗-isomorphism

.
Definition 1.3.7 If a is a normal element of A and f ∈ C0 (sp(a) \ {0})
we denote by f (a) the element of A corresponding to f via the isomorphism
given in 1.3.6. Corollary 1.3.6 is known as the continuous functional calculus
for normal elements.
The following corollary is often called the spectral mapping theorem.
Corollary 1.3.8 Let a be a normal element in a unital C ∗ -algebra A
and f ∈ C(sp(a)). Then

f (sp(a)) = sp(f (a)).

Proof. This follows immediately from the above corollary.

Theorem 1.3.9 Let A = C(X) and B = C(Y ), where X and Y are
two compact Hausdorff spaces. Then A ∼
= B if and only if X and Y are
homeomorphic.
Proof. Let φ : A → B be an isomorphism. Define τ : Y → X by
τ (y)(f ) = y(φ(f )). Here we identify X with the maximal ideal space (mul-
tiplicative states) of A and Y with the maximal ideal space of B. The
continuity of φ implies that τ is continuous. Since φ is onto, there is f ∈ A
such that τ (y1 ) 6= τ (y2 ) if y1 6= y2 . Thus τ is injective. Since Y is a compact
Hausdorff space, F = τ (Y ) is a closed subset of X. Let

I = {f ∈ C(X) : f |F = 0}.

Then kerφ ⊃ I. Thus F = X. In other words τ is onto, whence τ is a

homeomorphism.
Conversely if τ : Y → X is a homeomorphism, define φ(f )(y) = f (τ (y)).
To verify that φ is an isomorphism one can simply reverse the above argu-
ment, and we leave this to the reader.
Remark 1.3.10 The Gelfand representation theorem for commutative
C ∗ -algebras is fundamentally important. Even in a non-commutative C ∗ -
algebra A, we often obtain useful information of A via the study of certain
commutative C ∗ -subalgebras of A. So Theorem 1.3.5 certainly plays an
important role in studying non-commutative C ∗ -algebras as well. Theorem
1.3.9 shows that to study commutative C ∗ -algebras, it is equivalent to study
their maximal ideal spaces. Therefore the commutative C ∗ -algebra theory is

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
16 The Basics of C ∗ -algebras

the theory of topology. Much of general C ∗ -algebra theory can be described

as non-commutative topology.

1.4 Positive cones

Example 1.4.1 Let H be a Hilbert space and let B(H) be the space of all
bounded operators from H to H. If T1 , T2 ∈ B(H), we define (T1 T2 )(v) =
T1 (T2 (v)) for v ∈ H. With the operator norm kT k = supkvk=1 kT (v)k,
B(H) is a Banach algebra. If T ∈ B(H), define T ∗ by hT x, yi = hx, T ∗ yi
(x, y ∈ H), where h·, ·i is the inner product on H. Then one easily checks

kT ∗ T k = sup |hT ∗ T x, yi| = sup |hT x, T yi| = kT k2 .

kxk=1=kyk kxk=1=kyk

Thus B(H) is a C ∗ -algebra.

Every closed ∗-subalgebra of B(H) is a C ∗ -algebra. Later in this chapter
we will show that every C ∗ -algebra is a closed ∗-subalgebra of B(H) for
some Hilbert space.
Example 1.4.2 Let H be a Hilbert space. An operator T ∈ B(H) is said
to be compact if it maps bounded sets to precompact subsets. If the range
of T is finite dimensional, then T is compact. Denote by K(H) the set of
all compact operators in B(H). K(H) is clearly a closed subspace of B(H).
If T ∈ K(H), L ∈ B(H), then T L, LT ∈ K(H). Therefore K(H) is a closed
ideal of B(H). Moreover if T ∈ K(H), then T ∗ ∈ K(H). Therefore K is a
C ∗ -subalgebra of B(H). When H is separable infinite dimensional Hilbert
space, we will use K for K(H). K is a very important example of a (simple)
C ∗ -algebra.
Example 1.4.3 To show that any C ∗ -algebra is a C ∗ -subalgebra of
B(H) for some Hilbert space H, we need to study the order structure of C ∗ -
algebras. Recall that an operator T ∈ B(H) is called positive if hT v, vi ≥ 0
for all v ∈ H. In the case that H is finite dimensional, T is positive if and
only if it is self-adjoint and all eigenvalues are nonnegative.
If A = C(X), an element f ∈ A is positive if f (t) ≥ 0 for all t ∈ X.
We now introduce the following definition.
Definition 1.4.4 An element a in a C ∗ -algebra A is positive if a ∈ Asa
and sp(a) ⊂ R+ . We write a ≥ 0 if a is positive. The set of all positive
elements in A will be denoted by A+ .

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
16 The Basics of C ∗ -algebras

the theory of topology. Much of general C ∗ -algebra theory can be described

as non-commutative topology.

1.4 Positive cones

kT ∗ T k = sup |hT ∗ T x, yi| = sup |hT x, T yi| = kT k2 .

kxk=1=kyk kxk=1=kyk

Thus B(H) is a C ∗ -algebra.

A projection p is always positive, since p2 = p and p ∈ Asa implies that

sp(p) = {0, 1}, by 1.3.6.
Lemma 1.4.5 Let A be a C ∗ -algebra and let a ∈ A. Then the following
are equivalent:
(i) a ≥ 0;
(ii) a = b2 for some b ∈ Asa ;
(iii) a = a∗ and kt − ak ≤ t for any t ≥ kak;
(iv) a = a∗ and kt − ak ≤ t for some t ≥ kak.
Proof. (i) ⇒ (ii): In C ∗ (a), by 1.3.6, we set b = a1/2 .
(ii) ⇒ (i): Consider the C ∗ -subalgebra C ∗ (b). Then a ∈ C ∗ (b) and, by
1.3.6, C ∗ (b) = C0 (sp(b)). We see that a = a∗ since b ∈ Asa .
(i) ⇒ (iii): Since t − a is normal, from 1.3.5 we have (with t ≥ kak),

kt − ak = sup{|t − λ| : λ ∈ sp(a)} ≤ t.

(iii) ⇒ (iv) is immediate.

(iv) ⇒ (i): If λ ∈ sp(a) then t − λ ∈ sp(t − a). Thus

|t − λ| ≤ kt − ak ≤ t.

Therefore λ > 0 since λ ≤ t.

Definition 1.4.6 Let a ∈ Asa . Then a2 ∈ A+ . Set |a| = (a2 )1/2 (by
1.3.6). Then a+ = (1/2)(|a| + a) and a− = (1/2)(|a| − a). By the Gelfand
transform, |a|, a+ and a− are positive. Moreover, by 1.3.6, a+ a− = 0 and
a = a+ − a− .
Corollary 1.4.7 If A is a unital C ∗ -algebra, then A is the linear span of
unitaries.
Proof. Let a ∈ Asa and kak ≤ 1. Then 1 − a2 ∈ A+ . Put

u = a + i(1 − a2 )1/2 and v = a − i(1 − a2 )1/2 .

Then u∗ = v and u∗ u = uu∗ = 1. So u and v are unitaries. On the other

hand, we have a = (1/2)(u + v).
Theorem 1.4.8 The set A+ is a closed cone (a + b ∈ A+ , if a, b ∈ A+
and A+ ∩ A− = {0} ) and a ∈ A+ if and only if a = x∗ x for some x ∈ A.
Proof. It follows from (iii) of 1.4.5 that A+ is closed and, if a ∈ A+ , then
λa ∈ A+ for any λ ∈ R+ . To see that A+ is a cone, take a and b in A+ . By

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
18 The Basics of C ∗ -algebras

(iii) of 1.4.5,

k(kak + kbk) − (a + b)k = k(kak − a) + (kbk − b)k

≤ kkak − ak + kkbk − bk ≤ kak + kbk.

Since kak + kbk ≥ ka + bk, from the above and (iv) of 1.4.5, a + b ∈ A+ .
Now suppose that a = x∗ x for some x ∈ A. Then a∗ = a. Set a = a+ −a−
as in 1.4.6. We have
1/2 1/2 1/2 1/2 1/2 1/2
−(xa− )∗ (xa− ) = −a− x∗ xa− = −a− (a+ − a− )a− = a2− ∈ A+ .
1/2
Put xa− = b + ic with b, c ∈ Asa . Then
1/2 1/2 1/2 1/2
(xa− )(xa− )∗ = 2(b2 + c2 ) + [−(xa− )∗ (xa− )] ∈ A+

since A+ is a cone. It follows from 1.1.8 that sp(a2− ) ⊂ R+ ∩ R− = {0}.

Therefore a− = 0 and a ≥ 0.
Definition 1.4.9 In Asa , we write b ≤ a if a−b ≥ 0. Since Asa = A+ −A+
and A+ ∩ (−A+ ) = {0}, Asa becomes a partially ordered real vector space.
Theorem 1.4.10 Let A be a C ∗ -algebra.
(1) A+ = {a∗ a : a ∈ A}.
(2) If a, b ∈ Asa and c ∈ A, then a ≤ b implies that c∗ ac ≤ c∗ bc.
(3) If 0 ≤ a ≤ b, kak ≤ kbk.
(4) If A is unital and a, b ∈ A+ are invertible, then a ≤ b implies that
0 ≤ b−1 ≤ a−1 .
Proof. Condition (1) follows from 1.4.8 and the fact that positive ele-
ments have positive square roots. To see (2), we note that c∗ bc − c∗ ac =
c∗ (b − a)c. So (2) follows from (1). For (3), we may assume that A is unital.
The inequality b ≤ kbk · 1 follows from 1.3.5 by considering C ∗ (b, 1). Hence
a ≤ kbk · 1. We then apply 1.3.5 to C ∗ (a, 1) to obtain kak ≤ kbk.
To see (4), we note that if b ≥ 0 and invertible, then b−1 ≥ 0 (by 1.3.6). It
follows from (2) that b−1/2 ab−1/2 ≤ b−1/2 bb−1/2 = 1. Hence ka1/2 b−1/2 k ≤
1 (by (3)). Thus ka1/2 b−1 a1/2 k ≤ 1 which implies that a1/2 b−1 a1/2 ≤ 1. By
(2), we have

b−1 = (a−1/2 )(a1/2 b−1 a1/2 )(a−1/2 ) ≤ (a−1/2 )1(a−1/2 ) = a−1 .

Theorem 1.4.11 Let A be a C ∗ -algebra. If a, b ∈ A+ such that a ≤ b,
then aα ≤ bα for any 0 ≤ α ≤ 1.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive cones 19

Proof. By considering Ã, we may assume that A is unital. Fix 0 ≤ a ≤ b

in A+ . Let C = {α ∈ R+ : aα ≤ bα }. Then 0, 1 ∈ C. Since the function
fn (t) = tαn → tα if αn → α for αn ≥ 0 uniformly on [0, K] for any K > 0,
xαn → xα for any positive element x ∈ A+ if αn → α (and αn ≥ 0). Since
A+ is closed, if αn ∈ C and αn → α, then bα − aα ∈ A+ . This shows that
C is closed. To prove the theorem, it suffices to show that C is convex. We
first consider the case that both a and b are invertible. Let α, β ∈ C. Then

b−α/2 aα b−α/2 ≤ 1 and b−β/2 aβ b−β/2 ≤ 1.

Hence, by 1.4.10, kb−α/2 aα b−α/2 k ≤ 1 and kb−β/2aβ b−β/2 k ≤ 1. So

kaα/2 b−α/2 k2 ≤ 1 and kaβ/2 b−β/2 k2 ≤ 1. Therefore (using the fact that
sp(xy) = sp(yx))

1 ≥ k(b−β/2 aβ/2 )(aα/2 b−α/2 )k = kb−β/2 [a(α+β)/2 b−α/2 ]k

≥ r(b−β/2 [a(α+β)/2 b−α/2 ]) = r(a(α+β)/2 b−(α+β)/2 )
= r([a(α+β)/2 b−(α+β)/4 ]b−(α+β)/4 ) = r(b−(α+β)/4 a(α+β)/2 b−(α+β)/4 )
= kb−(α+β)/4 a(α+β)/2 b−(α+β)/4 k

Therefore we have b−(α+β)/4 a(α+β)/2 b−(α+β)/4 ≤ 1. Hence a(α+β)/2 ≤

b(α+β)/2 . Therefore C contains (1/2)(α + β). Consequently, C ⊃ [0, 1]. In
general, for any ε > 0, we have

(b + ε)α − (a + ε)α ≥ 0 (0 ≤ α ≤ 1).

Since A+ is closed, letting ε → 0, we obtain aα ≤ bα .

Example 1.4.12 It is not true that 0 ≤ a ≤ b implies a2 ≤ b2 in general

non-commutative C ∗ -algebras. For example, let A = M2 and let

1 0 1 1
p= and q = 1/2 .
0 0 1 1

Then both p and q are projections and p ≤ p + q. But p2 = p 6≤ (p + q)2 =

p + q + pq + qp, since

3 2
q + pq + qp = 1/2
2 1

is not positive.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
20 The Basics of C ∗ -algebras

1.5 Approximate identities, hereditary C ∗ -subalgebras and

quotients

To deal with non-unital C ∗ -algebras, and avoid the troubles caused by the
absence of unit, one can often embed A into Ã. However, more often, one
has to work in the original non-unital C ∗ -algebra. Therefore the notion of
approximate identity is essential.
Example 1.5.1 Let H be a Hilbert space with an orthonormal basis
{vn }∞
n=1 . The C -algebra K, the set of all compact operators on H, is a non-
∗

unital C -algebra. Let pn be the projection (from H) onto span{v1 , ...., vn }.

∗

Then {pn } is an increasing sequence of projections in K. The reader easily

checks that, for any x ∈ K,
kpn x − xk → 0 and kx − xpn k → 0 as n → ∞.
The sequence {pn } plays a role which is similar to that of identity in a
unital C ∗ -algebra.
Let X be a non-compact but σ-compact and locally compact Hausdorff
space X. Then C0 (X) = A is non-unital. Moreover, X = ∪∞ n=1 Xn , where
each Xn is compact and Xn+1 contains a neighborhood of Xn . It is easy to
produce an increasing sequence of positive functions fn ∈ C0 (X) such that
0 ≤ fn ≤ 1, fn (t) = 1 on Xn and fn (t) = 0 if t 6∈ Xn+1 . One checks that,
for any g ∈ C0 (X),
kgfn − gk → 0 as n → ∞.
Definition 1.5.2 An approximate identity for a C ∗ -algebra A is an in-
creasing net {eλ}λ∈Λ of positive elements in the closed unit ball of A such
that

a = lim aeλ .
λ

Equivalently, a = limλ eλ a for all a ∈ A.

Lemma 1.5.3 Let A be a C ∗ -algebra and denote by Λ the set of all
elements a ∈ A+ with kak < 1. Then Λ is upwards-directed; i.e., if a, b ∈ Λ,
then there exists some c ∈ Λ such that a, b ≤ c.
Proof. Suppose that a ∈ A+ . Then 1 + a ∈ GL(Ã), and a(1 + a)−1 =
1 − (1 + a)−1 . We claim that
a, b ∈ A+ and a ≤ b → a(1 + a)−1 ≤ b(1 + b)−1 . (e 5.4)

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Approximate identities, hereditary C ∗ -subalgebras and quotients 21

In fact, if a ≤ b, then 1+a ≤ 1+b which implies that (1+b)−1 ≤ (1+a)−1

by 1.4.10 (4). Consequently 1 − (1 + a)−1 ≤ 1 − (1 + b)−1 . This is the same
as

a(1 + a)−1 ≤ b(1 + b)−1 .

So the claim is proved.

We note also (by 1.3.6) that if a ∈ A+ , then a(1 + a)−1 ∈ Λ. Now
suppose that a, b ∈ Λ. Put

x = a(1 − a)−1 , y = b(1 − b)−1 and c = (x + y)(1 + x + y)−1 .

Since x+y ∈ A+ , c ∈ Λ. We note that (1+x)−1 = 1−a and x(1+x)−1 = a.

So, by e 5.4, a = x(1 + x)−1 ≤ c, since x ≤ x + y. Similarly, b ≤ c. This
proves the lemma.

The following positive continuous functions will be used throughout this

book.

Definition 1.5.4 Let

if ε ≤ t

 1
fε (t) = linear if ε/2 ≤ t < ε (e 5.5)
0 if 0 ≤ t ≤ ε/2


so that fε ∈ C0 ((0, K]) for any K > 0, and 0 ≤ fε ≤ 1. Note that if

α ≥ β > 0, then fβ ≥ fα . Also f 2n
1 f1 = f1 .
n n
Let a ∈ A+ . Then it follows from 1.3.6 that fε (a)a → a if ε → 0.

Theorem 1.5.5 Every C ∗ -algebra A admits an approximate identity.

Indeed, if Λ is the upwards-directed set of all a ∈ A+ with kak < 1 and
eλ = λ for all λ ∈ Λ, then {eλ }λ∈Λ forms an approximate identity for A.

Proof. From 1.5.3 {eλ } is an increasing net in the closed unit ball of A.
We need to show that limλ aeλ = a for all a ∈ A. Since Λ spans A, it suffices
to assume that a ∈ A+ . Since {ka(1 − eλ )ak} is decreasing, it suffices to
show that there are un ∈ {eλ} such that ka(1 − un )ak → 0 (as n → ∞).
Note that un = (1 − n12 )f n1 (a) ∈ Λ. We have t(1 − (1 − n12 )f n1 (t))t → 0 as
n → ∞ uniformly on [0, 1]. It follows from 1.3.6 that

ka(1 − un )ak → 0 as n → ∞.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
22 The Basics of C ∗ -algebras

Corollary 1.5.6 If A is separable, then A admits a countable approxi-

mate identity.

Proof. Fix a dense sequence {xn } of A. Let {eλ} be an approximate

identity for A. Choose ek ∈ {eλ } such that

kxi ek − xi k < 1/k (1 ≤ i ≤ k)

and ek+1 ≥ ek , k = 1, 2, ..., . Thus {ek } forms an approximate identity for

Definition 1.5.7 A C ∗ -algebra is said to be σ-unital if A admits a count-

able approximate identity.
The above corollary (1.5.6) shows that every separable C ∗ -algebra is
σ-unital.

Definition 1.5.8 A C ∗ -subalgebra B of A is said to be hereditary if for

any a ∈ A+ and b ∈ B+ the inequality a ≤ b implies that a ∈ B. Obviously,
0 and A are hereditary C ∗ -subalgebras of A. Any intersection of hereditary
C ∗ -subalgebras is hereditary. Let S ⊂ A. The hereditary C ∗ -subalgebra
generated by S is the smallest hereditary C ∗ -subalgebra of A containing S.
Such a hereditary C ∗ -subalgebra is denoted by Her(S).

Lemma 1.5.9 Let A be a C ∗ -algebra and a ∈ A+ . Then aAa is the

hereditary C ∗ -subalgebra generated by a.

Proof. Let B = aAa. It is easy to see that aAa is a ∗-subalgebra. It

follows that B is a C ∗ -subalgebra.
Claim (1): If c ∈ B+ then cAc ⊂ B. In fact, if axn a → c, then, for any
y ∈ A, axn ayaxn a → cyc. Therefore cAc ⊂ B. This proves the claim.
Claim (2): Let en = f1/n (a). Then {en } forms an approximate identity
for B. In fact by 1.5.4, ken a − ak → 0 as n → ∞. Suppose x = aba for some
b ∈ A. Then we have ken x − xk → 0 and kxen − xk → 0 as n → ∞. Since
aAa is dense in B, we conclude that {en } is an approximate identity for B.
Since aen a → a2 , a2 ∈ B. Hence a ∈ B and C ∗ (a) ⊂ B. In particular,
from Claim (1), f1/n (a)Af1/n (a) ⊂ B.
Now suppose that 0 ≤ c ≤ b with b ∈ B+ . We will show that c ∈ B. By
1.4.11, c1/2 ≤ b1/2 . Thus

k(1 − en )c(1 − en )k ≤ k(1 − en )b(1 − en )k ≤ k(1 − en )bk → 0.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Approximate identities, hereditary C ∗ -subalgebras and quotients 23

Therefore k(1−en )c1/2 k → 0, or en c1/2 → c1/2 . Hence en cen → c. By Claim

(1), en cen ∈ B. Therefore c ∈ B. So B is hereditary. We have shown that
a ∈ aAa. It follows that Her(a) = aAa.

Theorem 1.5.10 A C ∗ -algebra A is σ-unital if and only if there is a ∈

A+ such that Her(a) = A.

Proof. We have shown (from Claim (2) in the proof of 1.5.9) that A is
σ-unital if A = Her(a) for some a ∈ A+ .
For the converse, let {en } be an approximate identity for A. Put a =
P∞ n n
n=1 (1/2 )en and B = Her(a). Since en ≤ 2 a, by 1.5.9, en ∈ B for all n.
As in the proof of 1.5.9, en ben ∈ B for any b ∈ A+ . Since en b1/2 → b1/2 ,
en ben → b. Therefore b ∈ B. It follows that A = B.

Corollary 1.5.11 If A is a σ-unital C ∗ -algebra, then A admits a count-

able approximate identity {en } satisfying

en en+1 = en = en+1 en n = 1, 2, ....

Proof. Let A = aAa for some a ∈ A+ . Set en = f1/2n (a). Then {en } is
an approximate identity. By 1.5.4, f1/2n f1/2(n+1) = f1/2n for all n.

Theorem 1.5.12 If I is an ideal in a C ∗ -algebra A, then I is a hereditary

C ∗ -subalgebra of A. If {eλ} is an approximate identity for I, then for each
a ∈ A,

kāk = inf ka + bk = lim ka − eλ ak = lim ka − aeλ k.

b∈I λ λ

Proof. Let B = I ∗ ∩ I. Then B is a C ∗ -subalgebra of A. Let {eλ} be an

approximate identity for B. Note that eλ ∈ B ⊂ I. If a ∈ I,
limλ a∗ a(1 − eλ ) = 0. Hence

lim ka − aeλ k2 = lim k(1 − eλ )a∗ a(1 − eλ )k ≤ lim ka∗ a(1 − eλ )k = 0.

λ λ λ

Therefore a = limλ aeλ . Hence a∗ = limλ eλ a∗ . Since eλ a∗ ∈ I, we conclude

that a∗ ∈ I. Therefore I is C ∗ -subalgebra.
To see it is hereditary, let 0 ≤ c ≤ a, where c ∈ A+ and a ∈ I+ . By
1.5.9 c ∈ Her(a) = aAa. But aAa ⊂ I. Hence I is hereditary.
Now suppose that {eλ} is an approximate identity for I.
By definition, kāk = inf b∈I ka + bk for all a ∈ A. Fix a ∈ A. Put
α = inf b∈I ka + bk and β = limλ ka(1 − eλ )a∗ k (it exists since the net is

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
24 The Basics of C ∗ -algebras

decreasing). We have

α2 ≤ ka(1 − eλ )k2 = ka(1 − eλ )2 a∗ k ≤ ka(1 − eλ )a∗ k → β.

Hence α2 ≤ β. Thus it suffices to show that α2 ≥ β. To do this, let ε > 0

and take b ∈ I such that α + ε ≥ ka + bk. Then

(α + ε)2 ≥ ka + bkk1 − eλ kka∗ + b∗ k ≥ k(a + b)(1 − eλ )(a∗ + b∗ )k → β

since both kb(1 − eλ )k and kb∗ (1 − eλ )k tend to zero. Thus we have

(α + ε)2 ≥ β

for every ε > 0. Hence α2 ≥ β.

Corollary 1.5.13 If I is a closed ideal of A, then A/I equipped with its
natural operation is a C ∗ -algebra. (In particular, x̄∗ = x∗ .)
Proof. It follows from 1.1.20 that it suffices to show that kx̄∗ x̄k = kx̄k2 .
Let {eλ} be an approximate identity for I. Then,

kx̄∗ x̄k = lim kx∗ x(1 − eλ )k ≥ lim k(1 − eλ )x∗ x(1 − eλ )k

λ λ
= lim kx(1 − eλ )k = kx̄k2 .2
λ
Definition 1.5.14 From now on, an ideal of a C ∗ -algebra is always
closed, and a homomorphism from a C ∗ -algebra to another is always a
∗-homomorphism.
Theorem 1.5.15 Each homomorphism h : A → B is norm decreasing,
and h(A) is always a C ∗ -subalgebra of B. If h is injective, then it is an
isometry.
Proof. If a ∈ Asa , then sp(h(a)) \ {0} ⊂ sp(a) \ {0}. Since kak = r(a) by
1.2.7, we conclude that for each a ∈ A,

kh(a)k2 = kh(a∗ a)k ≤ ka∗ ak = kak2 .

Thus h is norm decreasing.

Now assume that h is injective. To show that h is an isometry, we need
to show that kh(a)k2 = kak2 , i.e., kh(a∗ a)k = ka∗ ak. Thus it suffices to
show that kh(x)k = kxk for all x ∈ A+ . Fix such an x and by restricting
to C ∗ (x), we may assume that A is commutative. Also, by considering Ã
and extending h to h̃ : Ã → B̃ if necessary, we may assume that A and B

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive linear functionals and a Gelfand-Naimark theorem 25

are unital. Thus, by 1.3.5, we may assume that A = C(S), where S is a

compact Hausdorff space. Since h(A) is commutative, by taking its closure,
we may also assume that B = C(T ) for some compact Hausdorff space
T. Define h∗ : T → S by h∗ (t) = t ◦ h. As in 1.3.9, since h is injective,
h∗ (T ) = S. Therefore, for each g ∈ C(S),

kgk = sup |g(s)| = sup |g(h∗ (t))| = kh(g)k.

s∈S t∈T

Thus h is an isometry.
To complete the proof, we note the I = kerh is an ideal of A (closed). So
h induces an isomorphism from A/I into B. Therefore A/I is isometrically
∗-isomorphic to h(A). It follows that h(A) is a C ∗ -subalgebra.

Definition 1.5.16 From now on, an isomorphism h from a C ∗ -algebra

A to another C ∗ -algebra B is always a ∗-isomorphism. Furthermore, it is
an isometric ∗-isomorphism. If A is isomorphic to B, we write A ∼
= B.

Corollary 1.5.17 Let I be a closed ideal of a C ∗ -algebra A and let B

be a C ∗ -subalgebra of A. Then B + I is a C ∗ -subalgebra of A. Moreover,
B/(B ∩ I) ∼ = (B + I)/I.

Proof. Clearly B + I is a ∗-subalgebra of A containing I and B. Let

π : A → A/I be the quotient map. By 1.5.15, π(B) is closed in A/I,
whence the preimage of π(B) is closed, i.e., B + I is closed.
Note that the map b + B ∩ I 7→ b + I is a ∗-isomorphism from B/(B ∩ I)
to (B + I)/I. Thus, by 1.5.15, it is a C ∗ -isomorphism.

1.6 Positive linear functionals and a Gelfand-Naimark

theorem

Definition 1.6.1 A linear map φ : A → B between C ∗ -algebras is said

to be self-adjoint if φ(Asa ) ⊂ Bsa , and positive if φ(A+ ) ⊂ B+ . It follows
that if φ is positive then φ is self-adjoint.
Every homomorphism h : A → B is positive.
If B = C, then a positive linear map φ : A → C is called a positive
linear functional. If, in addition, φ is bounded and kφk = 1, then φ is called
a state on A. If φ is a linear functional, we write φ ≥ 0 if it is positive.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive linear functionals and a Gelfand-Naimark theorem 25

are unital. Thus, by 1.3.5, we may assume that A = C(S), where S is a

kgk = sup |g(s)| = sup |g(h∗ (t))| = kh(g)k.

s∈S t∈T

Definition 1.5.16 From now on, an isomorphism h from a C ∗ -algebra

A to another C ∗ -algebra B is always a ∗-isomorphism. Furthermore, it is
an isometric ∗-isomorphism. If A is isomorphic to B, we write A ∼
= B.

Corollary 1.5.17 Let I be a closed ideal of a C ∗ -algebra A and let B

be a C ∗ -subalgebra of A. Then B + I is a C ∗ -subalgebra of A. Moreover,
B/(B ∩ I) ∼ = (B + I)/I.

Proof. Clearly B + I is a ∗-subalgebra of A containing I and B. Let

1.6 Positive linear functionals and a Gelfand-Naimark

theorem

Definition 1.6.1 A linear map φ : A → B between C ∗ -algebras is said

A positive linear functional τ : A → C is called a trace if φ(u∗ au) = φ(a)

for all a ∈ A and all unitaries u ∈ Ã. A trace which is also a state is called
a tracial state.

Example 1.6.2 Let A = C(X), where X is a compact Hausdorff space,

and let µ be a positive Borel finite measure. Then the linear functional τ
defined by
Z
τ (f ) = f dµ

is positive. Since A is commutative, τ is also a trace.

Example 1.6.3 Let A = Mn . Define a linear functional T r on A by

X
n
T r((aij )) = aii .
i=1

This is a trace. It is called the standard trace on Mn . The normalized trace

tr on A is defined to be tr = (1/n)T r. It is a tracial state.

Example 1.6.4 Let A be a C ∗ -subalgebra of B(H), where H is a Hilbert

space. Let v ∈ H be a nonzero vector. Define f (a) = ha(v), vi for a ∈ A.
Then f is positive. To see this, note that for a ≥ 0, there is x ∈ A such
that a = x∗ x. Thus f (a) = hx∗ x(v), vi = hx(v), x(v)i = kx(v)k ≥ 0. This
functional is, in general, not a trace.

Lemma 1.6.5 Every positive linear functional on a C ∗ -algebra A is

bounded.

Proof. Let φ be a positive linear functional on A. If φ is not bounded,

then there is a sequence {xn } ⊂ A with kxk k ≤ 1 such that |φ(xk )| → ∞.
Since φ(Asa ) ⊂ R, by considering the real part and imaginary part of xk ,
without loss of generality, we may assume that xk ∈ Asa . By considering
xk = (xk )+ − (xk )− , we may further assume that xk ∈ A+ . By passing to
a subsequence if necessary, we may assume that φ(xk ) ≥ 2k , k = 1, 2, ..., .
P
Set x = ∞ xk
k=1 2k . Then x ∈ A+ . For any n,

X
∞ X∞
xk
−k
n≤ 2 φ(xk ) = φ( ) ≤ φ(x).
2k
k=1 k=1

This is impossible. Thus φ must be bounded.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive linear functionals and a Gelfand-Naimark theorem 27

The following is the Cauchy-Schwarz inequality.

Lemma 1.6.6 If φ is a positive linear functional on a C ∗ -algebra A, then

for all a, b ∈ A

|φ(b∗ a)|2 ≤ φ(b∗ b)φ(a∗ a). (e 6.6)

Proof. For each complex number λ, we have φ((λa + b)∗ (λa + b)) ≥ 0.
∗
We may assume that φ(b∗ a) 6= 0. With λ = t|φ(a b)|
φ(b∗ a) and t ∈ R, this gives

t2 φ(a∗ a) + 2t|φ(b∗ a)| + φ(b∗ b) ≥ 0

(φ(b∗ a) = φ(a∗ b)). If φ(a∗ a) = 0, the above becomes

2t|φ(b∗ a)| + φ(b∗ b) ≥ 0

for all t ∈ R. By taking negative t with large |t|, we conclude that φ(b∗ a) =
0. The inequality (e 6.6) holds in this case.
∗
If φ(a∗ a) 6= 0, choose t = − |φ(b a)|
φ(a∗ a) . We obtain

|φ(b∗ a)|2 |φ(b∗ a)|2

− 2 + φ(b∗ b) ≥ 0.
φ(a∗ a) φ(a∗ a)

Thus (e 6.6) follows.

Theorem 1.6.7 An element φ in A∗ is positive if and only if limλ φ(eλ ) =

kφk for some approximate identity {eλ} in A.

Proof. If φ is a positive linear functional on A and {eλ } is any approx-

imate identity for A, then φ(eλ ) is increasing. Hence it has a limit, say l.
Then l ≤ kφk. For each a ∈ A with kak ≤ 1, by 1.6.6,

|φ(eλ a)|2 ≤ φ(e2λ )φ(a∗ a) ≤ φ(eλ )kφk ≤ lkφk.

Since φ is continuous (1.6.5), we obtain |φ(a)|2 ≤ lkφk, whence kφk2 ≤ lkφk

and l = kφk.
To prove the converse, suppose that {φ(eλ )} converges to kφk. We first
show that φ(Asa ) ⊂ R. Let a ∈ Asa with kak ≤ 1 and write φ(a) = α + iβ
with α, β ∈ R. By multiplying by −1 if necessary, we may assume that
β ≥ 0. Choose eλ so that keλa − aeλ k < 1/n. Then

kneλ − iak2 = kn2 eλ + a2 − in(aeλ − eλ a)k ≤ n2 + 2.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
28 The Basics of C ∗ -algebras

On the other hand

lim |φ(neλ − ia)|2 = (nkφk + β)2 + α2 .

Combining these two inequalities, we obtain

(nkφk + β)2 + α2 ≤ (n2 + 2)kφk2

for all n. Thus

2nkφkβ + β 2 + α2 ≤ 2kφk2 .

for all n. Hence β = 0. So φ(Asa ) ⊂ R.

Now if a ∈ A+ with kak ≤ 1, then eλ − a ∈ Asa and eλ − a ≤ eλ ≤ 1.
Therefore

φ(eλ − a) ≤ kφk.

Taking the limit, we obtain kφk − φ(a) ≤ kφk. It follows that φ(a) ≥ 0
which implies that φ ≥ 0.

Remark 1.6.8 We actually proved that limλ φ(eλ ) = kφk for any ap-
proximate identity {eλ} of A. If A is unital then Theorem 1.6.7 also implies
that φ ≥ 0 if and only if φ(1) = kφk.

Lemma 1.6.9 Let Ã be the C ∗ -algebra obtained by adjoining a unit to

the C ∗ -algebra A. For each positive linear functional φ on A define an
extension φ̃ on Ã by setting φ̃(1) = kφk. Then φ̃ is positive on Ã and
kφ̃k = kφk.

Proof. Let {eλ } be an approximate identity for A. For each a ∈ A and

α ∈ C, by 1.4.10 (3),

lim sup kαeλ + ak2 = lim sup k|α|2 e2λ + ᾱeλ a + αa∗ eλ + a∗ ak
λ λ
≤ lim sup k|α|2 1 + ᾱeλ a + αa∗ eλ + a∗ ak = kα1 + ak2 .
λ

It follows from 1.6.7 that

|φ̃(α + a)| = lim |φ(αeλ + a)| ≤ kα1 + akkφk.

Hence kφ̃k = kφk. Since φ̃(1) = kφ̃k, by 1.6.7, φ̃ ≥ 0.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive linear functionals and a Gelfand-Naimark theorem 29

Proposition 1.6.10 Let B be a C ∗ -subalgebra of a C ∗ -algebra A. For

each positive linear functional φ on B there is a positive linear functional
φ̃ on A such that φ̃|B = φ and kφ̃k = kφk. If furthermore B is hereditary,
then the extension is unique.
Proof. We may assume that A has a unit by replacing A by Ã if necessary.
Extend φ to C ∗ (B, 1A ) ∼
= B̃ by setting φ(1) = kφk. It follows from 1.6.9
that so extended φ is positive and preserves the norm. Therefore, without
loss of generality, we may assume that A has a unit 1 and 1 ∈ B .
By the Hahn-Banach theorem, there is a linear functional φ̃ such that
φ̃|B = φ and kφ̃k = kφk. Since φ̃(1) = kφ̃k, by 1.6.7, φ̃ ≥ 0.
To prove the second part, assume that B is hereditary and let {eλ } be
an approximate identity for B. Suppose that ψ is a positive linear functional
on A with ψ|B = φ and kψk = kφk. Then kψk = limλ φ(eλ ) by 1.6.7. So
limλ ψ(1A − eλ ) = 0. It follows that ψ((1A − eλ )2 ) ≤ ψ(1A − eλ ) → 0. By
1.6.6, for any c ∈ A,

|ψ((1 − eλ )ac)|2 ≤ ψ((1A − eλ )2 )ψ(c∗ a∗ ac) → 0.

Therefore

ψ(a) = lim ψ(eλ aeλ ).

Since eλ Ãeλ ⊂ B, we have

ψ(a) = lim ψ(eλ aeλ ) = lim φ(eλ aeλ )

λ λ

for every a ∈ A. Thus ψ = φ̃.

Corollary 1.6.11 Let a ∈ A be a normal element. Then there is a state
φ on A such that |φ(a)| = kak.
Proof. Let B = C ∗ (a). Since r(a) = kak, by 1.2.7, there is λ ∈ sp(a)
such that |λ| = kak. Identify B with C0 (sp(a)), by 1.3.6, we define a state
φ1 on B by φ1 (g) = g(λ) for g ∈ C0 (sp(a)). In particular, |φ1 (a)| = kak.
By 1.6.10, there is a state φ on A such that φ|B = φ1 .
Definition 1.6.12 Let A be a C ∗ -algebra and f ∈ A∗ . Define f ∗ (a) =
f (a∗ ). Denote by fsa and fim the self-adjoint linear functionals (1/2)(f +f ∗ )
and (1/2i)(f − f ∗ ), respectively. We also use Ref for the real part of f, i.e.,
Ref (a) = (1/2)[f (a) + f (a)] (for a ∈ A). Ref is a real linear functional on
A (regard A as a real Banach space). One should note that fsa and Ref

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
30 The Basics of C ∗ -algebras

are different. It is standard that kf k = kRef k. This is usually used in the

proof of the Hahn-Banach theorem (to pass from the real version to the
complex version). It will be used later in these notes.
We have the following non-commutative Jordan decomposition theorem.
Proposition 1.6.13 Let A be a C ∗ -algebra and f ∈ A∗ . Then f is a
linear combination of states. More precisely, we have the following decom-
position:

f = fsa + ifim , fsa = (fsa )+ − (fsa )− and fim = (fim )+ − (fim )− ,

where kfsa k ≤ kf k, kfimk ≤ kf k, (fsa )+ , (fsa )− , (fim )+ and (fim )− are

positive and

kfsa k = k(fsa )+ k + k(fsa )− k and kfim k = k(fim )+ k + k(fim )− k.

Proof. It is clear that fsa and fim are self-adjoint and kfsa k, kfim k ≤
kf k are obvious. For the rest of the proof, we may assume that f is self-
adjoint. Let Ω be the set of all positive linear functionals g with kgk ≤ 1.
With the weak∗ -topology, Ω is compact. View A as a closed subspace of
C(Ω). Note that A+ ⊂ C(Ω). Let f ∈ A∗ . Then f can be extended to a
bounded linear functional f˜ on C(Ω) with the same norm. Therefore there
is a complex Radon measure µ on Ω such that
Z
˜
f (x) = xdµ (x ∈ C(Ω)).
Ω

Let µ = ν1 + iν2 , where ν1 and ν2 are signed measures. So νj (j = 1, 2)

gives self-adjoint bounded linear functionals on A which will be denoted by
fj (j = 1, 2). Since f is self-adjoint, (f2 )|A = 0. So f1 extends f (necessarily
they have the same norm). Therefore we may assume that f˜ is self-adjoint.
So we may assume that µ is a signed measure. Then by Jordan decom-
position, we have positive measures µj (j = 1, 2) such that µ = µ1 − µ2
and kµk = kµ1 k + kµ2 k. Each µj gives a positive linear functional fj on A
(j = 1, 2). We have the following
˜ A k ≤ k(f1 )|A k + k(f2 )|A k ≤ kf1 k + kf2 k = kfk
kf k = kf| ˜ = kf k.

Thus kf k = k(f1 )|A k + k(f2 )|A k.

Definition 1.6.14 A representation of a C ∗ -algebra A is a pair (H, π),
where H is a Hilbert space and π : A → B(H) is a homomorphism. We say

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Positive linear functionals and a Gelfand-Naimark theorem 31

(H, π) is faithful if π is injective. A cyclic representation is a representation

(H, π) with a vector v ∈ H such that π(A)v is dense in H; and the vector
v is called cyclic.

Theorem 1.6.15 For each positive linear functional φ on a C ∗ -algebra

A there is a cyclic representation (Hφ , πφ ) of A with a cyclic vector vφ such
that hπφ (a)vφ , vφ i = φ(a) for all a ∈ A.

Proof. Define the left kernel of φ as the set

Nφ = {a ∈ A : φ(a∗ a) = 0}.

Since φ is continuous, Nφ is a closed subspace. It follows from the Cauchy-

Schwarz inequality (1.6.6) that Nφ is a closed left ideal. On A/Nφ define

hā, b̄i = φ(b∗ a). (e 6.7)

It is easy to see that the above is well defined on A/Nφ × A/Nφ . By the
Cauchy-Schwarz inequality (1.6.6) again, A/Nφ becomes an inner product
space. Let Hφ be the completion. Define πφ (a)(x̄) = ax for all a, x ∈ A. It
is well-defined. In fact, if y ∈ A such that ȳ = x̄, then a(y − x) ∈ Nφ , since
Nφ is a left ideal. Thus πφ (a) is a linear map on A/Nφ . Furthermore,

kπφ (a)x̄k2 = kaxk2 = φ(x∗ a∗ ax) ≤ kak2 φ(x∗ x) = kak2 kx̄k2

(the inequality holds because a∗ a ≤ kak2 and φ is positive). Thus πφ can

be uniquely extended to an operator (still denoted by πφ (a)) on Hφ such
that kπφ (a)k ≤ kak. Also,

hπφ (a)(x̄), ȳi = φ(y ∗ ax) = hx̄, πφ (a∗ )(ȳ)i

which shows that πφ (a)∗ = πφ (a∗ ). So πφ : A → B(Hφ ) gives a representa-

tion.
To complete the proof, let {eλ } be an approximate identity for A. Then
for λ < µ,

kēµ − ēλ k2 = φ((eµ − eλ )2 ) ≤ φ(eµ − eλ ).

Since φ(eλ ) → kφk, the net {ēλ } is convergent with a limit vφ ∈ Hφ . For
each a ∈ A we have

πφ (a)(vφ ) = lim πφ (a)(ēλ ) = lim aeλ = ā,

λ λ

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
32 The Basics of C ∗ -algebras

since a → ā is continuous. Hence vφ is cyclic. Moreover, since

hπφ (a∗ a)(vφ ), vφ i = hā, āi = φ(a∗ a),

we have hπφ (a)(vφ ), vφ i = φ(a) for all a ∈ A+ (by 1.4.8) and by linearity
for all a ∈ A.

Definition 1.6.16 Let (Hλ , πλ )λ∈Λ be a family of representations. Let

H = ⊕λ Hλ be the Hilbert space direct sum. Define π(a)({vλ }) =
{πλ (a)(vλ )}. One verifies that π : A → B(H) gives a representation of
A. This representation is called the direct sum of {πλ }.
Let S be the state space of A. By 1.6.15, for each t ∈ S, there is a
representation πt of A. The direct sum πU of {πt }t∈S is called the universal
representation of A.

Theorem 1.6.17 (Gelfand-Naimark) If A is a C ∗ -algebra, then it has

a faithful representation. In other words, every C ∗ -algebra is isometrically
∗-isomorphic to a C ∗ -subalgebra of B(H) for some Hilbert space H.

Proof. We will show that the universal representation πU is faithful. Let

a ∈ A be nonzero with kak ≤ 1. It follows from 1.6.11 that there is τ ∈ S
such that |τ ((a∗ a)2 )| = k(a∗ a)2 k = kak4 . Then

kπU (a)k2 ≥ kπτ (a)k2 ≥ kπτ (a)((a∗ a)1/2 )k2

= τ ((a∗ a)1/2 (a∗ a)(a∗ a)1/2 ) = τ ((a∗ a)2 ) = kak4 .

Thus πU is injective.

1.7 Von Neumann algebras

Definition 1.7.1 Let H be a Hilbert space and B(H) be the C ∗ -algebra

of all bounded operators on H. The strong (operator) topology on B(H) is
the locally convex space topology associated with the family of semi-norms
of the form x 7→ kx(ξ)k, x ∈ B(H) and ξ ∈ H. In other words, a net {xλ }
converges strongly to x and only if {xλ (ξ)} converges to x(ξ) for all ξ ∈ H.
The weak (operator) topology on B(H) is the locally convex space topol-
ogy associated with the family of semi-norms of the form x → 7 |hx(ξ), ηi|,
x ∈ B(H) and ξ, η ∈ H. In other words, a net {xλ } converges weakly to
x ∈ B(H) if and only if h(xλ − x)(ξ), ηi → 0 for all ξ, η ∈ H.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
32 The Basics of C ∗ -algebras

since a → ā is continuous. Hence vφ is cyclic. Moreover, since

hπφ (a∗ a)(vφ ), vφ i = hā, āi = φ(a∗ a),

we have hπφ (a)(vφ ), vφ i = φ(a) for all a ∈ A+ (by 1.4.8) and by linearity
for all a ∈ A.

Definition 1.6.16 Let (Hλ , πλ )λ∈Λ be a family of representations. Let

Theorem 1.6.17 (Gelfand-Naimark) If A is a C ∗ -algebra, then it has

a faithful representation. In other words, every C ∗ -algebra is isometrically
∗-isomorphic to a C ∗ -subalgebra of B(H) for some Hilbert space H.

Proof. We will show that the universal representation πU is faithful. Let

a ∈ A be nonzero with kak ≤ 1. It follows from 1.6.11 that there is τ ∈ S
such that |τ ((a∗ a)2 )| = k(a∗ a)2 k = kak4 . Then

kπU (a)k2 ≥ kπτ (a)k2 ≥ kπτ (a)((a∗ a)1/2 )k2

= τ ((a∗ a)1/2 (a∗ a)(a∗ a)1/2 ) = τ ((a∗ a)2 ) = kak4 .

Thus πU is injective.

1.7 Von Neumann algebras

Definition 1.7.1 Let H be a Hilbert space and B(H) be the C ∗ -algebra

Example 1.7.2 Let H be an infinite dimensional Hilbert space with an

orthonormal basis {en }∞ n=1 . Define an (ξ) = hξ, en ie1 for ξ ∈ H (n = 1, 2, ...).
Then an ∈ B(H). We have
limn→∞ kan (ξ)k = limn→∞ |hξ, en i| = 0. Thus an strongly converges to
zero. However, kan k = 1 since kan (en )k = 1 for all n. Thus the strong
topology is weaker than the norm topology.
Define bn (ξ) = hξ, e1 ien (n = 1, ...). Then, for any ξ, η ∈ H,
|hbn (ξ), ηi| = |hξ, e1 ihen , ηi| → 0, as n → ∞. So bn converges weakly to
zero. However, kbn (e1 )k = ken k = 1. So bn does not converge to zero in the
strong topology.
The reader may want to take a look at exercises (1.11.18-1.11.22) for
some additional information about the weak and strong (operator) topolo-
gies.
Proposition 1.7.3 Let {aλ } be an increasing net of positive operators
in B(H) which is also bounded above. Then {aλ } converges strongly to a
positive operator a ∈ B(H).
Proof. Suppose that there is M > 0 such that kaλ k ≤ M. For any ξ ∈ H,
{haλ (ξ), ξi} is a bounded increasing sequence. So it is convergent. Using the
polarization identity
3
X
haλ (ξ), ηi = (1/4) ik haλ (ξ + ik η), ξ + ik ηi,
k=0

we see that {haλ )(ξ), ηi} is convergent for all ξ, η ∈ H. Denote by L(ξ, η) the
limit. The map (ξ, η) 7→ L(ξ, η) is linear in the first variable and conjugate
linear in the second. We also have

|L(ξ, η)| = lim |haλ (ξ), ηi| ≤ M kξkkηk

for all ξ, η ∈ H. By the Riesz representation theorem, there is a ∈ B(H)

such that ha(ξ), ηi = L(ξ, η) for all ξ, η ∈ H. Clearly kak ≤ M and aλ ≤ a.
Moreover,

ka(ξ) − aλ (ξ)k2 = k(a − aλ )1/2 (a − aλ )1/2 (ξ)k2

≤ ka − aλ kk(a − aλ )1/2 (ξ)k2 ≤ 2M h(a − aλ )(ξ), ξi,

and h(a − aλ )(ξ), ξi → 0, so a(ξ) = limλ aλ (ξ). Thus aλ converges strongly

to a.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
34 The Basics of C ∗ -algebras

Definition 1.7.4 If H is a Hilbert space, we write H (n) for the orthogonal

sum of n copies of H. If a ∈ Mn (B(H)), we define φ(a) ∈ B(H (n) ) by setting

Xn n
X
φ(a)(x1 , ..., xn ) = ( a1j (xj ), ..., anj (xj ))
j=1 j=1

for all (x1 , ..., xn ) ∈ H (n) .

It is easy to verify that the map

φ : Mn (B(H)) → B(H (n) ), a 7→ φ(a),

is a ∗-isomorphism. We call φ the canonical ∗-isomorphism of Mn (B(H))

onto B(H (n) ), and use it to identify these two algebras. We define a norm on
Mn (B(H)) making it a C ∗ -algebra by setting kak = kφ(a)k. The following
inequalities for a ∈ Mn (B(H)) are easily verified:
n
X
kaij k ≤ kak ≤ kakl k (i, j = 1, ..., n) (e 7.8)
k,l=1

For each i ≤ n let Pi be the projection of H (n) onto the ith copy
of H. Each element x ∈ B(H (n) ) has a representation (aij )1≤i,j≤n with
aij ∈ B(H). We define an amplification ρ (of multiplicity n) from B(H) →
B(H (n) ) by setting ρ(a) = diag(a, ..., a) (where a repeats n times).

Lemma 1.7.5 Let φ be a linear functional on B(H). Then the following

are equivalent:
Pn
(i) φ(a) = k=1 ha(ξk ), ηk i for some ξ1 , ..., ξn , η1 , ..., ηn ∈ H and for all
a ∈ B(H);
(ii) φ is weakly continuous;
(iii) φ is strongly continuous.

Proof. It is obvious that (i) ⇒ (ii) ⇒ (iii). To prove (iii) ⇒ (i), suppose
that φ is strongly continuous. Therefore, there exist vectors ξ1 , ..., ξn and
δ > 0 such that |φ(a)| ≤ 1, whenever maxk {ka(ξk )k} ≤ δ for all a ∈ B(H).
δa
For any a ∈ B(H), put b = (Pn kaξ 2 1/2 . Then |φ(b)| ≤ 1. Thus
kk ) k=1

Xn
|φ(a)| ≤ (1/δ)( kaξk k2 )1/2 . (e 7.9)
k=1

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Von Neumann algebras 35

for all a ∈ B(H). With notation as in 1.7.4, we define ξ = ξ1 ⊕ · · · ⊕ ξn in

H (n) . On the vector subspace V = {ρ(a)ξ : a ∈ B(H)}, define

ψ(ρ(a)ξ) = φ(a).

From the definition (in 1.7.4), ψ is a linear functional (on the span) and
|ψ(ρ(a)ξ)| ≤ (1/δ)kρ(a)ξk. So it is a bounded linear functional. It extends
to a bounded linear functional on H0 , the closure of V. By the Riesz rep-
resentation theorem, there is a vector η = η1 ⊕ · · · ⊕ ηn ∈ H0 ⊂ H (n) such
that
n
X
φ(a) = hρ(a)(ξ), ηi = haξk , ηk i.
k=1

Corollary 1.7.6 Each strongly closed convex set in B(H) is weakly

closed.

We leave this to the reader for an exercise (1.11.23).

Definition 1.7.7 For each subset M ⊂ B(H), let M ′ denote the com-
mutant of M, i.e.,

M ′ = {a ∈ B(H) : ab = ba for all b ∈ M }.

It is easy to verify that M ′ is weakly closed. If M is self-adjoint, then M ′

is a C ∗ -algebra. We will write M ′′ for (M ′ )′ .

The following is von Neumann’s double commutant theorem.

Theorem 1.7.8 Let M be a C ∗ -subalgebra of B(H) containing the iden-

tity. The following are equivalent:
(i) M = M ′′ .
(ii) M is weakly closed.
(iii) M is strongly closed.

Proof. The implication (i) ⇒ (ii) ⇔ (iii) follows from 1.7.7 and 1.7.6.
We will show (iii) ⇒ (i). Fix ξ ∈ H let P be the projection on the closure
of {aξ : a ∈ M }. Note that P ξ = ξ since 1 ∈ M. Since P aP = aP for all
a ∈ M, P a∗ = P a∗ P for all a ∈ M. Therefore P ∈ M ′ . Let x ∈ M ′′ . Then
P x = xP. Hence xξ ∈ P H. Thus for any ε > 0 there is an a ∈ M with
k(x − a)ξk < ε.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
36 The Basics of C ∗ -algebras

To show that x is in the strong closure of M, take finitely many vectors

ξ1 , ..., ξn ∈ H. Set ξ = ξ1 ⊕ · · · ⊕ ξn ∈ H (n) . It is easily checked that

ρ(M )′ = {a ∈ B(H (n) ) : aij ∈ M ′ }.

Therefore ρ(x) ∈ ρ(M )′′ . From what we have proved in the first part of the
proof, we obtain a ∈ M such that
n
X
k(x − a)ξk k2 = k(ρ(x) − ρ(a))ξk2 < ε2 .
k=1

It follows that x is in the strong closure of M. Thus x ∈ M. In other words,

M ′′ ⊂ M. Since clearly M ⊂ M ′′ , M = M ′′ .

Definition 1.7.9 A weakly closed C ∗ -subalgebra M ⊂ B(H) is called a

von Neumann algebra. In other words, a C ∗ -subalgebra M ⊂ B(H) is a
von Neumann algebra if M = M ′′ and 1 ∈ M.

Definition 1.7.10 Let A be a C ∗ -algebra and πU : A → B(HU ) be

the universal representation. Then (πU (A))′′ is called the enveloping von
Neumann algebra (or universal weak closure) of A and will be denoted
by A′′ . A representation π : A → B(H) is said to be non-degenerate if
{π(a)H : a ∈ A} is dense in H.

Proposition 1.7.11 Every non-degenerate representation is a direct sum

of cyclic representations

The proof is an exercise (see 1.11.14).

Lemma 1.7.12 Let f ∈ A∗ be self-adjoint. Then there are vectors ξ, η ∈

HU with kξk2 , kηk2 ≤ kf k such that f (a) = hπU (a)(ξ), ηi for all a ∈ A.

Proof. To save notation, we may assume that kf k = 1. By 1.6.13, there

are positive linear functionals fj , j = 1, 2, such that f = f1 − f2 and
kf k = kf1 k + kf2 k. Each fj is a positive scalar multiple of a state on A.
Therefore there are mutually orthogonal vectors ξj ∈ HU (j = 1, 2) such
that kξj k2 = kfj k and fj (a) = hπU (a)(ξj ), ξj i, j = 1, 2. Set ξ = ξ1 ⊕ ξ2 and
η = ξ1 ⊕ (−ξ2 ). Then f (a) = hπU (a)ξ, ηi for all a ∈ A. Furthermore,

kηk2 = kξk2 = kξ1 k2 + kξ2 k2 ≤ kf k.

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Von Neumann algebras 37

Theorem 1.7.13 The enveloping von Neumann algebra A′′ of a C ∗ -

algebra A is isometrically isomorphic, as a Banach space, to the second
dual A∗∗ of A which is the identity map on A.

Proof. Let j : A → A∗∗ be the usual embedding. Let (HU , πU ) be the

universal representation of A. For each f ∈ A∗ , by 1.6.13, there are ξf , ηf ∈
HU such that f (a) = hπU (a)ξf , ηf i. Define J : A′′ → A∗∗ by J(x)(f ) =
hx(ξf ), ηf i for each state f. It follows from 1.7.12 that J(x) defines an
element in A∗∗ . In fact (using 1.7.12), we have

kJ(x)k = sup |f (x)| ≤ sup |hx(ξ), ηi| ≤ 4kxk.

kf k≤1,f ∈A∗ kξk,kηk≤2

On the other hand, if kxk ≤ 1, and ξ ∈ HU with kξk ≤ 1, hπU (a)ξ, x(ξ)i for
a ∈ A defines a linear functional φξ,x(ξ) on A with kφξ,x(ξ) k ≤ kx(ξ)k. So

|J(x)(φξ,x(ξ) )| = |hx(ξ), x(ξ)i| = kx(ξ)kkx(ξ)k.

Hence kJ(x)k ≥ kx(ξ)k for all ξ ∈ HU with kξk ≤ 1. This implies that
kJ(x)k ≥ kxk.
Suppose that x ∈ A′′sa and {aλ } ⊂ A such that πU (aλ ) converges to x
weakly in A′′ . By replacing aλ by (1/2)(aλ +a∗λ ) (see Exercise (1.11.18)), we
may assume that aλ are self-adjoint. This implies that, for any self-adjoint
φ ∈ A∗ , J(x)(φ) is real. Fix f ∈ A∗ with kf k ≤ 1, x ∈ A′′sa , and assume
that |f (x)| = eiθ f (x). Define F = eiθ f. Note that 1 ≥ kF k = kReF k. We
have

|f (x)| = F (x) = ReF (x) = Fsa (x) = |hx(ξ), ηi|

for some ξ, η ∈ HU with kξk, kηk ≤ 1 (by 1.7.12). Thus

|f (x)| = |hx(ξ), ηi| ≤ kxk.

This implies that kJ(x)k = kxk for all x ∈ A′′sa .

In general, note that M2 (A′′ ) = (M2 (A))′′ . Let J2 : (M2 (A))′′ →
M2 (A)∗∗ be as above. Fix f ∈ A∗ and x ∈ A′′ . There are ξf , ηf ∈ HU
such that J(x)(f ) = hx(ξf ), ηf i.
In H (2) , set ξ ′ = 0 ⊕ ξf and η ′ = ηf ⊕ 0. Define a linear functional φ on
Q, where

Q = {(aij ) ∈ M2 (A) : a11 = a22 = 0},

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
38 The Basics of C ∗ -algebras

by φ(z) = hz(ξ ′ ), η ′ i for z ∈ Q. So kφk = kf k. Extend φ further to M2 (A)

such that kφk = kf k. By 1.6.13 and 1.7.12, there are ξ1 , η1 ∈ HU such
that J2 (z)(φ) = hz(ξ1 ), η1 i for all z ∈ M2 (A)′′ . Suppose that z = (zij ) ∈
(M2 (A))′′ such that z11 = z22 = 0. Since z is in the weak (operator)
closure
0 x
of πU (Q), J2 (z)(φ) = hz(ξ ′ ), η ′ i = f (z12 ). Put b = . Then b ∈
x∗ 0
′′
M2 (A )sa . Therefore, from what we have shown, kJ2 (b)k = kbk = kxk. So

|J(x)(f )| = |J2 (b)(φ)| ≤ kbkkφk = kxkkf k.

This implies that kJ(x)k ≤ kxk. Therefore J is an isometry. Since A is

weakly dense in A′′ , J(A) is dense in J(A′′ ) in the weak∗ -topology as a
subset of (A∗ )∗ . However, j(A) is dense in (A∗ )∗ in the weak∗ -topology.
Therefore, since J(a) = j(a) for a ∈ A, J(A′′ ) = (A∗ )∗ = A∗∗ .
The following corollary follows from the above theorem and the uniform
boundedness theorem.
Corollary 1.7.14 Let {aλ } be a net in a C ∗ -algebra A′′ . If {aλ } con-
verges in the weak operator topology in A′′ , then {kaλ k} is bounded.
Remark 1.7.15 The weak operator topology for A as a subalgebra of A′′
acting on HU is called the σ-weak topology. So the weak operator topology
for A′′ is sometimes called σ-weak topology.

1.8 Enveloping von Neumann algebras and the spectral

theorem

Lemma 1.8.1 Let (π1 , H1 ) and (π2 , H2 ) be two cyclic representations of

a C ∗ -algebra A with cyclic vectors ξ1 and ξ2 (kξ1 k = kξ2 k). Then there
exists an isometry u : H1 → H2 such that π1 = u∗ π2 u with u(ξ1 ) = ξ2 if
and only if hπ1 (a)ξ1 , ξ1 i = hπ2 (a)ξ2 , ξ2 i for all a ∈ A.
Proof. If uξ1 = ξ2 then

hπ1 (a)ξ1 , ξ1 i = hu∗ π2 (a)uξ1 , ξ1 i = hπ2 (a)ξ2 , ξ2 i

for all a ∈ A.
Conversely, define a linear map u from π1 (A)ξ1 onto π2 (A)ξ2 by
u(π1 (a)ξ1 ) = π2 (a)ξ2 . Since

ku(π1 (a)ξ1 )k2 = hπ2 (a∗ a)ξ2 , ξ2 i = hπ1 (a∗ a)ξ1 , ξ1 i = kπ1 (a)ξ1 k2 ,

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
38 The Basics of C ∗ -algebras

by φ(z) = hz(ξ ′ ), η ′ i for z ∈ Q. So kφk = kf k. Extend φ further to M2 (A)

|J(x)(f )| = |J2 (b)(φ)| ≤ kbkkφk = kxkkf k.

This implies that kJ(x)k ≤ kxk. Therefore J is an isometry. Since A is

1.8 Enveloping von Neumann algebras and the spectral

theorem

Lemma 1.8.1 Let (π1 , H1 ) and (π2 , H2 ) be two cyclic representations of

hπ1 (a)ξ1 , ξ1 i = hu∗ π2 (a)uξ1 , ξ1 i = hπ2 (a)ξ2 , ξ2 i

for all a ∈ A.
Conversely, define a linear map u from π1 (A)ξ1 onto π2 (A)ξ2 by
u(π1 (a)ξ1 ) = π2 (a)ξ2 . Since

ku(π1 (a)ξ1 )k2 = hπ2 (a∗ a)ξ2 , ξ2 i = hπ1 (a∗ a)ξ1 , ξ1 i = kπ1 (a)ξ1 k2 ,

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Enveloping von Neumann algebras and the spectral theorem 39

we see that u extends to an isometry from the closure of π1 (A)ξ1 onto the
closure of π2 (A)ξ2 . Since ξ1 and ξ2 are cyclic vectors, u is an isometry from
H1 onto H2 . We have

uπ1 (a)π1 (b)ξ1 = π2 (ab)ξ2 = π2 (a)uπ1 (b)ξ1

for all a, b ∈ A. Thus uπ1 (a) = π2 (a)u since {π1 (b)ξ1 : b ∈ A} is dense in
H1 . The lemma follows.

Theorem 1.8.2 Let A be a C ∗ -algebra and π : A → B(H) be a non-

degenerate representation. Then there is a unique π ′′ : A′′ → π(A)′′ such
that π ′′ |A = π and it is (σ-) weak-weak continuous, i.e., if {xλ } is a weak
convergent net in the von Neumann algebra A′′ then {π ′′ (xλ )} is a weak
convergent net in B(H).

Proof. First assume that (π, H) is cyclic with cyclic vector ξ1 and kξ1 k =
1. Then φ(a) = hπ1 (a)ξ1 , ξ1 i is a state. It follows from 1.8.1 that we may
assume that (π, H) = (πφ , Hφ ). Let pφ be the projection of HU onto Hφ .
Then pφ πU (a) = πU (a)pφ = πφ (a) for all a ∈ A. Thus pφ ∈ A′ . It is clear
that x 7→ pφ x is weak-weak continuous from A′′ into π(A)′′ which extends
π (from A to π(A)).
By 1.7.11, in general, π = ⊕λ πλ , where each πλ is a cyclic representa-
tion. Therefore, there are states φλ of A and an isometry U : H → ⊕λ Hφλ
such that U ∗ πU = ⊕λ πφλ . So we may assume that π = ⊕λ πφλ . Let p be
the projection from HU onto ⊕λ Hφλ . Then the map x 7→ px is weak-weak
continuous and extends π.

Definition 1.8.3 Let X be a compact Hausdorff space. Let B(X) be the

set of all bounded Borel functions on X. B(X) is a C ∗ -algebra containing
C(X) as a C ∗ -subalgebra. It follows from 1.2.3 that B(X) is a subspace
in C(X)∗∗ . By 1.7.13, C(X)′′ = C(X)∗∗ . So B(X) is a C ∗ -subalgebra of
C(X)′′ . We say {fλ} converges weakly to f in B(X), if they do so as ele-
ments of the von Neumann algebra C(X)′′ . If {fλ } is bounded and converges
pointwise everywhere on X, then, by Lebesgue’s Dominated Convergence
Theorem, {fλ } converges weakly.

Corollary 1.8.4 Let φ : C(X) → B(H) be a unital homomorphism. Then

there is a unique homomorphism φ̃ : B(X) → B(H) such that φ̃|C(X) = φ
and {fλ } converges weakly to f in B(X) implies that {φ̃(fλ )} converges
weakly to φ̃(f ) in B(H).

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
40 The Basics of C ∗ -algebras

Definition 1.8.5 Let a ∈ B(H) be a normal element. Then A = C ∗ (1, a)

is a C ∗ -subalgebra. Denote by j : A → B(H) the embedding. Then by
1.8.4, let j̃ : B(X) → B(H) be the extension, where X = sp(a). We will use
f (a) for j̃(f ), f ∈ B(X). Thus the following is called the Borel functional
calculus.

Corollary 1.8.6 Let M be a von Neumann algebra and a ∈ M a nor-

mal element. Then there is a weak-weak continuous homomorphism from
B(sp(a)) → M which maps z → a.

Definition 1.8.7 Let X be a compact Hausdorff space and H be a

Hilbert space. A spectral measure E relative to (X, H) is a map from the
Borel sets of X to the set of projections in B(H) such that
(1) E(∅) = 0, E(X) = 1;
(2) E(S1 ∩ S2 ) = E(S1 )E(S2 ) for all Borel sets S1 , S2 of X;
(3) for all ξ, η ∈ H, the function Eξ,η : S 7→ hE(S)ξ, ηi is a regular Borel
complex measure.
Pn
For any simple function g = k=1 αk χSk ,
Z n
X
g(λ)dEλ = αk E(Sk )
X k=1

is an operator in B(H) (Sk are Borel sets). Let f ∈ B(X). There is a

sequence {gn } of simple functions such that kgn − f k → 0 as Rn → ∞.
Since µ(S) = hE(S)ξ, ηi defines a regular Borel complex measure,
R X gn dEλ
converges weakly to an element (in B(H)). Denote by X f dEλ the weak
limit. Clearly this limit does not depend on the choice of gn .
We have the following corollary otherwise known as the spectral theorem.

Corollary 1.8.8 Let a be a normal operator on a Hilbert space H. Then

there is a unique spectral measure E relative to (sp(a), H) such that
Z
a= λdEλ .
sp(a)

Proof. By the Borel functional calculus, the embedding j : C ∗ (1, a) →

B(H) extends to a weak-weak continuous homomorphism j̃ : B(X) →
B(H). For each Borel subset of sp(a), j̃(χS ) = E(S) is a projection. It
is easy to verify that {E(S) : S Borel} forms a spectral measure. Since
the identity function on sp(a) is Borel, it follows from the Borel functional

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
Enveloping von Neumann algebras and the spectral theorem 41

calculus that
Z
a= λdEλ .
sp(a)

We leave it to the reader to check the uniqueness of E.

Recall the range projection p of an operator a ∈ B(H) is the projection

on the closure of {a(ξ) : ξ ∈ H}.

Proposition 1.8.9 If M is a von Neumann algebra, then it contains the

range projection of every element in M.

Proof. Let a ∈ M. It is clear that we may assume that kak ≤ 1. Since

(aa∗ )1/2 and a have the same range projection, we may assume that a ≥ 0.
Note that {a1/n } is increasing and bounded. It follows from 1.7.3 that
{a1/n } converges strongly to a positive element, say q ∈ M. Let f = χ(0,kak] .
Then, by 1.8.6, a1/n converges weakly to f (a), which is a projection. There-
fore q = f (a). Since a1/n ∈ C ∗ (a) and the polynomials are dense in
C(sp(a)), q(H) ⊂ a(H). One the other hand, qa = f (a)a = a. Hence
a(H) ⊂ q(H). Therefore q is the range projection of a.

Definition 1.8.10 An operator u on H is called a partial isometry if u∗ u

is a projection. Since

(uu∗ )3 = u(u∗ u)(u∗ u)u∗ = (uu∗ )2 .

sp(uu∗ ) = {0, 1} by the spectral mapping theorem. Therefore, uu∗ is also

a projection.

Proposition 1.8.11 Each element x in a von Neumann algebra M has

a polar decomposition: there is a unique partial isometry u ∈ M such that
u∗ u is the range projection of |x| and x = u|x|.

Proof. Set un = x((1/n) + |x|)−1 and denote by p the range projection

of |x|. Since x = xp, we have un = un p. We compute that

(un −um )∗ (un −um )

1 1 1 1
= [( + |x|)−1 −( + |x|)−1 )]x∗ x[( + |x|)−1 −( + |x|)−1 ]
n m n m
1 1
= [( + |x|)−1 − ( + |x|)−1 )]2 |x|2 . (e 8.10)
n m

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−
42 The Basics of C ∗ -algebras

This converges weakly to zero (as n, m → ∞) by the Borel functional

calculus. So, for any ξ ∈ H,

k(un − um )(ξ)k2 = |h(un − um )∗ (un − um )(ξ), ξi| → 0

as n, m → ∞. This implies that {un } converges strongly to an element

u ∈ M with up = u. As in (e 8.10),
1 1
(un |x| − um |x|)∗ (un |x| − um |x|) = [( + |x|)−1 − ( + |x|)−1 )]2 |x|4 .
n m
By the continuous functional calculus (see 1.3.6), the above converges in
norm (as n, m → ∞). On the other hand, by the Borel functional calculus,
((1/n)+|x|)−1 |x| converges weakly to p. Thus {un |x|} converges in norm to
x. Hence x = u|x|. Since pu∗ up = (u∗ )(u) = u∗ u, p(u∗ u) = (u∗ u)p = u∗ u.
On the other hand, x∗ x = |x|u∗ u|x|. For any ξ, η ∈ H,

h(u∗ u − p)(|x|(ξ)), |x|ηi = h|x|(u∗ u − p)(|x|(ξ)), ηi = 0.

Regarding u∗ u − p as an operator on p(H), the above implies that u∗ u = p.

To see the decomposition is unique, let x = v|x| and v ∗ v = p. Then
v|x| = u|x|, or (v − u)|x| = 0. Therefore, (v − u)p = 0. This implies that
v = u.

1.9 Examples of C ∗ -algebras

In this section we give some examples of C ∗ -algebras. More will be presented

later.
We first give more information about the C ∗ -algebras K and B(H). If
H is a Hilbert space, an operator x ∈ B(H) is said to have finite-rank, if
the range of x is a finite dimensional subspace. Denote by F (H) the set of
finite-rank operators on H. It is easy to check that F (H) is a ∗-subalgebra
of B(H) and is an ideal (not necessary closed) of B(H).
Clearly every operator in F (H) is compact. It is also easy to see that
F (H) is a linear span of rank-one projections.

Lemma 1.9.1 If H is a Hilbert space and K(H) is the C ∗ -algebra of all

compact operators on H, then F (H) is dense in K(H).
Proof. Since F (H) and K(H) are both self-adjoint, it suffices to show
that every self-adjoint element x ∈ K(H) is in F (H). We may even further

!∀#∃%& ∋!∋ ∋
%()∗∗+++∋+∃#,∃∋−∗−.%−.∃∗/012∋%−

(Studies in Advanced Mathematics) Kehe Zhu - An Introduction To Operator Algebras-CRC Press (1993)
No ratings yet
(Studies in Advanced Mathematics) Kehe Zhu - An Introduction To Operator Algebras-CRC Press (1993)
169 pages
Zhu K An - Introduction - To - Operator - Algebras (-CRC 1991)
100% (1)
Zhu K An - Introduction - To - Operator - Algebras (-CRC 1991)
168 pages
Functional AnFunctional Analysis Lecture Notes Jeff Schenkeralysis Jeff Schenker
0% (1)
Functional AnFunctional Analysis Lecture Notes Jeff Schenkeralysis Jeff Schenker
169 pages
C Algebras and Operator Theory
No ratings yet
C Algebras and Operator Theory
297 pages
208 C - Algebras: Marc Rieffel Notes by Qiaochu Yuan Spring 2013
No ratings yet
208 C - Algebras: Marc Rieffel Notes by Qiaochu Yuan Spring 2013
55 pages
Pub Function-Algebras307
No ratings yet
Pub Function-Algebras307
276 pages
Lecture Notes On Banach Algebras and Operator Theory
No ratings yet
Lecture Notes On Banach Algebras and Operator Theory
39 pages
Banach Algebras An Introduction Pure and Applied Mathematics
100% (1)
Banach Algebras An Introduction Pure and Applied Mathematics
358 pages
CStar LectureNotes
No ratings yet
CStar LectureNotes
64 pages
Von Neumann
No ratings yet
Von Neumann
127 pages
Course Wilde
No ratings yet
Course Wilde
76 pages
Functional Analysis Notes
No ratings yet
Functional Analysis Notes
84 pages
Cstar Notes
No ratings yet
Cstar Notes
67 pages
Lecture No Two Banach Algebra, Notations and Basic Definitions
No ratings yet
Lecture No Two Banach Algebra, Notations and Basic Definitions
8 pages
Cstar
No ratings yet
Cstar
157 pages
Operator Theory of Banach Algebra
No ratings yet
Operator Theory of Banach Algebra
93 pages
Rieffel, Math 208, C-Algebras
No ratings yet
Rieffel, Math 208, C-Algebras
55 pages
CStar Notes
No ratings yet
CStar Notes
73 pages
Oa 2
No ratings yet
Oa 2
12 pages
Stable Range in Banach Algebra JPAA 32 (1984), 289-300
No ratings yet
Stable Range in Banach Algebra JPAA 32 (1984), 289-300
12 pages
Banach Spaces: K K Is Called A Seminorm
No ratings yet
Banach Spaces: K K Is Called A Seminorm
84 pages
Dual Spaces of Banach Algebras
No ratings yet
Dual Spaces of Banach Algebras
24 pages
Algebra 2
No ratings yet
Algebra 2
14 pages
Pub Operator-Theory
No ratings yet
Pub Operator-Theory
66 pages
Operator Algebras An Introduction
No ratings yet
Operator Algebras An Introduction
30 pages
Inverse Closed Algebra Convex-Cones in Ordered Banach Algebras Mangat Ram
No ratings yet
Inverse Closed Algebra Convex-Cones in Ordered Banach Algebras Mangat Ram
5 pages
Roe J. - Notes On Operator Algebras (2000)
No ratings yet
Roe J. - Notes On Operator Algebras (2000)
110 pages
Banalg
No ratings yet
Banalg
55 pages
Banach Algebra PDF
No ratings yet
Banach Algebra PDF
26 pages
Linear Algebra Via Complex Analysis: Alexander P. Campbell Daniel Daners Corrected Version January 24, 2014
No ratings yet
Linear Algebra Via Complex Analysis: Alexander P. Campbell Daniel Daners Corrected Version January 24, 2014
17 pages
K-Theory For C: - Algebras, and Beyond
No ratings yet
K-Theory For C: - Algebras, and Beyond
110 pages
Linear Algebra Via Complex Analysis: Alexander P. Campbell and Daniel Daners Revised Version September 13, 2012
No ratings yet
Linear Algebra Via Complex Analysis: Alexander P. Campbell and Daniel Daners Revised Version September 13, 2012
19 pages
Banachalgebras
No ratings yet
Banachalgebras
22 pages
Operatortheory TKSM 2
No ratings yet
Operatortheory TKSM 2
18 pages
Cstar Algebras JAE
No ratings yet
Cstar Algebras JAE
51 pages
Banach Algebras
No ratings yet
Banach Algebras
17 pages
Operatortheory TKSM 1
No ratings yet
Operatortheory TKSM 1
20 pages
Operator Algebras and Their Application in Physics: Lecture Notes On
No ratings yet
Operator Algebras and Their Application in Physics: Lecture Notes On
70 pages
Chapter Vi Banach Algebras 1988
No ratings yet
Chapter Vi Banach Algebras 1988
18 pages
Chapter II Elementary Spectral Theory 1988
No ratings yet
Chapter II Elementary Spectral Theory 1988
21 pages
Spectral Mapping Theorem For Polynomials
No ratings yet
Spectral Mapping Theorem For Polynomials
28 pages
Long Chen: Min H H H
No ratings yet
Long Chen: Min H H H
11 pages
Continuity of Approximate Point Spectrum On The Algebra
No ratings yet
Continuity of Approximate Point Spectrum On The Algebra
14 pages
Banach Algebras and Spectral Theory (Lectures) - Coleman
No ratings yet
Banach Algebras and Spectral Theory (Lectures) - Coleman
96 pages
Wermuth RemarkCommutingOperator 1997
No ratings yet
Wermuth RemarkCommutingOperator 1997
5 pages
442C Banach Algebras 2009-10
No ratings yet
442C Banach Algebras 2009-10
47 pages
Corrected
No ratings yet
Corrected
7 pages
Kernels of Hilbert Module Maps: A Counterexample: Jens Kaad and Michael Skeide January 2021
No ratings yet
Kernels of Hilbert Module Maps: A Counterexample: Jens Kaad and Michael Skeide January 2021
6 pages
4751 Chap1 9
No ratings yet
4751 Chap1 9
10 pages
Sema Paper10
No ratings yet
Sema Paper10
14 pages
Tips - C Algebras and Operator Theory
No ratings yet
Tips - C Algebras and Operator Theory
297 pages
8 The Spectral Analysis of Selfadjoint Operators
No ratings yet
8 The Spectral Analysis of Selfadjoint Operators
9 pages
Banalg 1
No ratings yet
Banalg 1
16 pages
Spectral Theory For Unbounded Operators
No ratings yet
Spectral Theory For Unbounded Operators
12 pages
Operator Algebra S
No ratings yet
Operator Algebra S
54 pages
Bse 2
No ratings yet
Bse 2
8 pages
NCG 1
No ratings yet
NCG 1
10 pages
II. The Exponential Function: II.1. Smooth Functions Defined by Power Series
No ratings yet
II. The Exponential Function: II.1. Smooth Functions Defined by Power Series
24 pages
Gaur 2
No ratings yet
Gaur 2
3 pages
Chapter 1 4 Latest 123
No ratings yet
Chapter 1 4 Latest 123
93 pages
PM IMG Configuration Tracker KDS Key Data Structure
No ratings yet
PM IMG Configuration Tracker KDS Key Data Structure
8 pages
Stability & Determinacy of Trusses PDF
No ratings yet
Stability & Determinacy of Trusses PDF
5 pages
MYSYSTEM2
No ratings yet
MYSYSTEM2
9 pages
The Manning Equation
No ratings yet
The Manning Equation
10 pages
2022-2023 ASVAB Arithmetic Reasoning and Mathematics
No ratings yet
2022-2023 ASVAB Arithmetic Reasoning and Mathematics
4 pages
(Tutorial) Dynamic Analysis For High Speed Two - Final - Blue
No ratings yet
(Tutorial) Dynamic Analysis For High Speed Two - Final - Blue
28 pages
AF12 Chapter 5 Solutions
No ratings yet
AF12 Chapter 5 Solutions
111 pages
Norton
No ratings yet
Norton
14 pages
Complete Deep Learning Interview Question
No ratings yet
Complete Deep Learning Interview Question
46 pages
Course - Outline - Math 2030B - F2019 PDF
No ratings yet
Course - Outline - Math 2030B - F2019 PDF
5 pages
New Microsoft Office PowerPoint Presentation
No ratings yet
New Microsoft Office PowerPoint Presentation
27 pages
QA 27 Geometry - 2
No ratings yet
QA 27 Geometry - 2
33 pages
Synthetic Image Transformation
No ratings yet
Synthetic Image Transformation
13 pages
UT 1 Science Class 10
No ratings yet
UT 1 Science Class 10
6 pages
Similarity Class 9 and 10
No ratings yet
Similarity Class 9 and 10
9 pages
Red e CVC
No ratings yet
Red e CVC
19 pages
Input Output Models
No ratings yet
Input Output Models
4 pages
Gray Code
No ratings yet
Gray Code
6 pages
Presentation On Research Paper "The Peak Pool Boiling Heat Flux On Horizontal Cylinders"
No ratings yet
Presentation On Research Paper "The Peak Pool Boiling Heat Flux On Horizontal Cylinders"
19 pages
Terbilang
No ratings yet
Terbilang
3 pages
BOA Spot Tool Application Guide v1.0
No ratings yet
BOA Spot Tool Application Guide v1.0
48 pages
Computer Engineering and Systems Group Orientation 2013
No ratings yet
Computer Engineering and Systems Group Orientation 2013
29 pages
GATE 2024 Mining Engineering MN Solutions
No ratings yet
GATE 2024 Mining Engineering MN Solutions
8 pages
Frege and Names
No ratings yet
Frege and Names
23 pages
Fundamentals of Computer Programming: Arrays (CLO3)
No ratings yet
Fundamentals of Computer Programming: Arrays (CLO3)
17 pages
A3 - 1bm15me039 - Nyquist Plot Using Matlab
No ratings yet
A3 - 1bm15me039 - Nyquist Plot Using Matlab
12 pages
Restricting and Sorting Data Questions
No ratings yet
Restricting and Sorting Data Questions
22 pages
Inmo 2000
No ratings yet
Inmo 2000
1 page