Lecture notes

C ∗-algebras and K-theory

Matthias Ludewig
SS 2020

September 20, 2023

If you find any mistakes (even spelling mistakes), please tell me!

1
Contents
1 Some theory of C ∗ -algebras 3
1.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Spectral theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The unitalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 The Gelfand-Naimark theorem . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 The spatial tensor product . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 The K0 -functor 19
2.1 Equivalence of projections . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 The Grothendieck construction . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Definition of K0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Homotopy invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3 The K-theory long exact sequence 34

3.1 Half-Exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Cone and suspension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 The long exact sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Bott Periodicity 42
4.1 The exterior product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 The Töplitz exact sequence and the Bott element . . . . . . . . . . . . . . 44
4.3 The periodicity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 The K1 -functor and the six-term exact sequence 50

5.1 Definition of K1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Identification with Suspension . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3 The index map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4 The exponential map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.5 The six-term exact sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2
1 Some theory of C ∗-algebras
In this section, we give the definition of Banach algebras, C ∗ -algebras and related notions.
Then we review the spectral theory of Banach and C ∗ -algebras. Finally, we prove the
Gelfand-Naimark theorem and the existence of a continuous functional calculus for normal
elements.

1.1 Basic definitions

Definition 1.1 (Banach and C ∗ algebras). All algebras are over the field C.
(a) A Banach algebra is an algebra A together with a norm that turns it into a Banach
space and is submultiplicative, that is
kabk ≤ kakkbk ∀a, b ∈ A.

(b) A ∗-algebra is an algebra A together with an anti-linear involution A → A, a 7→ a∗ ,

the ∗-operation, such that
(ab)∗ = b∗ a∗ ∀a, b ∈ A.

(c) A C ∗ -algebra is a Banach algebra with a ∗-operation satisfying the C ∗ -property

ka∗ ak = kak2 ∀a ∈ A. (1.1)

(d) A Banach algebra A is unital if A has a unit 1 6= 0 such that k1k = 1. (For C ∗ -
algebras, this follows from (1.1).)
The following facts will be used throughout without mentioning.
Lemma 1.2. In any (unital) C ∗ -algebra, we have
kak = ka∗ k and 1∗ = 1. (1.2)

Proof. To see the first identity, calculate kak2 = ka∗ ak ≤ ka∗ kkak, hence kak ≤ ka∗ k;
replacing a by a∗ yields ka∗ k ≤ kak. To see the second identity, observe that 1∗ a =
(a∗ 1)∗ = (a∗ )∗ = 1 for all a ∈ A; therefore 1 = 1∗ 1 = 1∗ . 
Example 1.3 (Continuous functions). Let X be a locally compact Hausdorff space.
The space
C0 (X) = {f : X → C continuous | ∀ε > 0∃K ⊂ X compact : kf |X\K k∞ < ε}
is a commutative Banach algebra with pointwise multiplication and the supremum norm
kf k∞ = supx∈X |f (x)|. It is even a C ∗ -algebra with f ∗ (x) = f (x) (pointwise complex
conjugation). C0 (X) is unital if and only if X is compact. In that case, C0 (X) = C(X),
the algebra of continuous functions on X.

3
Example 1.4 (Algebras of operators). Let H be a complex Hilbert space.
(1) B(H), the algebra of bounded operators on H, is a unital Banach algebra with respect
to the operator norm. It is commutative precisely when dim(H) ≤ 1. It is even a
∗-algebra with respect to taking the adjoint map.

(2) K(H), the algebra of compact operators on H, is a closed subalgebra of B(H). Since
taking adjoints preserves K(H), it is a C ∗ -algebra. It is unital precisely when H is
finite-dimensional and commutative precisely when dim(H) ≤ 1.

Example 1.5 (Direct sums). If A, B are Banach algebras (C ∗ -algebras), their direct
sum A ⊕ B is a Banach algebra (C ∗ -algebra) with the norm k(a, b)k := sup{kak, kbk}.

Example 1.6 (Quotients). If A is a Banach algebra an J ⊂ A is a closed ideal, then

A/J is a Banach algebra with the norm k[a]k = inf{ka + bk | b ∈ J}. If A is a C ∗ -
algebra, things are slightly involved. First of all, it is a fact that any closed ideal J of A is
automatically ∗-closed, i.e. a∗ ∈ J for all a ∈ J [3, Thm. 3.1.3]. Therefore, the ∗-operation
[a]∗ := [a∗ ] is well-defined on A/J. Hence A/J is a ∗-algebra, but showing the C ∗ -identity
k[a]k2 = k[a]∗ [a]k is rather tricky; one way to see this is via approximate units (see e.g.
[3, Thm. 3.1.4] or [2, Prop. 1.8.2]). For another approach, see [4, Exercise 1.B].

Example 1.7 (Matrix algebras). If A is an algebra and n ∈ N, the algebra Mn (A) the
algebra of n × n matrices with entries in A is an algebra with matrix multiplication. If A
is ∗-algebra, there is a ∗-operation on Mn (A) given by
 ∗  
a11 · · · a1n a∗11 · · · a∗n1
 .. .. ..  =  .. .. ..  ,
 . . .   . . . 
∗ ∗
an1 · · · ann a1n · · · anm

turning also Mn (A) into a ∗-algebra. If A is moreover a C ∗ -algebra, Lemma 1.38 shows
that there is a suitable norm on Mn (A) turning it into a C ∗ -algebra.

Definition 1.8 (Homomorphisms). Let A, B be algebras.

(a) A homomorphism Φ : A → B is a linear map which is multiplicative, that is Φ(ab) =
Φ(a)Φ(b) for a, b ∈ A.

(b) If A and B are unital, then we say that Φ is unital if Φ(1A ) = 1B .

(c) If A and B are ∗-algebras, a ∗-homomorphism is a homomorphism Φ : A → B such

that Φ(a∗ ) = Φ(a)∗ for all a ∈ A.

Remark 1.9. If A, B are Banach algebras, we do not require continuity of homomor-

phisms Φ : A → B. Instead, continuity is often automatic (see e.g. Prop. 1.16 and
Lemma 1.22).

4
Example 1.10. If B is another C ∗ -algebra and Φ : A → B a ∗-homomorphism, we get
an induced ∗-homomorphism Mn (Φ) : Mn (A) → Mn (B), which is given by
   
a11 · · · a1n Φ(a11 ) · · · Φ(a1n )
 .. .. ..  7−→  .. .. ..  .
 . . .   . . .  (1.3)
an1 · · · ann Φ(an1 ) · · · Φ(ann )

Clearly, associating matrix algebras assembles to a functor that sends the category of
C ∗ -algebras and ∗-homomorphisms to itself. For convenience, we will usually just write
again Φ instead of Mn (Φ).

1.2 Spectral theory

Definition 1.11. Let A be a unital Banach algebra and let a ∈ A.

(a) ρ(a) = {λ ∈ C | λ − a invertible} is called resolvent set of a.

(b) σ(a) = C \ ρ(a) is called spectrum of a.

If A is moreover a C ∗ -algebra, then

(c) a is called normal if a∗ a = aa∗ .

(d) a is called self-adjoint if a = a∗ .

Example 1.12. If X is a compact Hausdorff space, then for f ∈ C(X), we have σ(f ) =
{f (x) | x ∈ X}. This follows directly from the fact that f ∈ C(X) is invertible precisely
if f (x) 6= 0 for all x ∈ X.

Example 1.13. Let H be a Hilbert space and T ∈ B(H). The essential spectrum σess (T )
of T consists of those λ ∈ C such that λ − T is not a Fredholm operator. Remember
here that T is called Fredholm if it has closed range and finite-dimensional kernel and
cokernel; equivalently (by Atkinsons’s theorem), it is one that admits a parametrix, that
is an operator S such that T S − idH , ST − idH ∈ K(H). Clearly, the latter condition
is equivalent to saying that the class [T ] is invertible in B(H)/K(H). We conclude that
σess (T ) = σ([T ]), the spectrum of [T ] in B(H)/K(H)

Proposition 1.14. Let A be a unital Banach algebra and let a ∈ A.

(a) ρ(a) is open.

(b) σ(a) is compact, more precisely, |λ| ≤ kak for all λ ∈ σ(a).

(c) σ(a) 6= ∅.

If A is moreover a C ∗ -algebra, then

5
(d) if a is normal, then kak2 = ka2 k and kak = sup{|λ| | λ ∈ σ(a)};

(e) if a is self-adjoint, then σ(a) ⊆ R.

(f) σ(a∗ a) ⊂ R≥0 ;

Proof. Except for (f), these results are proven just as the special case A = B(H). Assertion
(f) is non-trivial; a proof can be found in [3, §2.2]. 

Proposition 1.15. Any ∗-algebra has at most one submultiplicative norm satisfying the
C ∗ -property with respect to which it is complete.

Proof. This follows from the fact that the norm is determined by the algebra structure:
For all a ∈ A,
kak2 = ka∗ ak = sup{|λ| | λ ∈ σ(a∗ a)},
where the first equality is the C ∗ -property and the second equality follows from Prop. 1.14(d),
as a∗ a is normal. 

Proposition 1.16. Let A and B be a unital C ∗ -algebras. Then any unital ∗-homomorphism
Φ : A → B is contractive, i.e. kΦ(a)k ≤ kak for all a ∈ A.

Proof. Let a ∈ A and λ ∈ ρ(a). Then λ − a is invertible, say (λ − a)b = 1A . Since Φ is

unital,
1B = Φ(1A ) = Φ((λ − a)b) = (λ − Φ(a))Φ(b).
we conclude that λ − Φ(a) is invertible, with inverse Φ(b), hence λ ∈ ρ(Φ(a)). Hence
ρ(a) ⊆ ρ(Φ(a)) and σ(Φ(a)) ⊆ σ(a). Now by the C ∗ -property and Prop. 1.14(d),

kΦ(a)k2 = kΦ(a)∗ Φ(a)k = sup{|λ| | λ ∈ σ(Φ(a)∗ Φ(a)))}

≤ sup{|λ| | λ ∈ σ(a∗ a))} = ka∗ ak = kak2 .

This finishes the proof. 

Theorem 1.17 (Gelfand-Mazur). Let A be a unital Banach algebra where every ele-
ment 0 6= a ∈ A is invertible. Then A is one-dimensional.

Proof. Let a ∈ A. Since σ(a) 6= ∅ (Prop. 1.14(c)), we can choose λ ∈ C such that λ − a is
not invertible. Hence by assumption on A, a − λ = 0, i.e. a is a multiple of the identity.

6
1.3 The unitalization
Definition 1.18. Let A be a C ∗ -algebra. Its unitalization A+ is the unital ∗-algebra with
underlying vector space A+ = A ⊕ C, product

(a, λ) · (b, µ) := (ab + λb + µa, λµ).

and ∗-operation
(a, λ)∗ := (a∗ , λ).

That the product and ∗-operation given above indeed turn A+ into a ∗-algebra is a
straightforward calculation. The identification a 7→ (a, 0) embeds A into A+ , and we just
write A ⊂ A+ . One easily checks that A is an ideal in A+ . In fact, it is the kernel of the
augmentation map, which is the ∗-homomorphism

εA : A+ −→ C, εA (a, λ) = λ.

Proposition 1.19. For any C ∗ -algebra A, the ∗-algebra A+ is a C ∗ -algebra, that is, there
exists a unique norm of A+ that turns A+ into a C ∗ -algebra.

Proof. We have to prove the existence; uniqueness then follows from Prop. 1.15. If A is
unital, we have A+ ∼
= A ⊕ C as ∗-algebras (via the isomorphism (a, λ) 7→ (a + λ1A , λ))
∗
and A ⊕ C is a C -algebra.
Assume now that A is non-unital. In this case, we define a norm by

k(a, λ)k := sup kab + λbk, (1.4)

kbk≤1

where the supremum is taken over all b ∈ A with kbk ≤ 1 and the norm on the right hand
side is the norm of A. The norm is definite because kab + λbk = 0 for all b ∈ A would
mean that either λ = 0 and hence a = 0 or −λ−1 a is a unit for A. It is submultiplicative
because
k(a, λ) · (b, µ)k = sup kabc + λbc + µac + λµck
kck≤1

ka(bc + µc) + λ(bc + µc)k

= sup kbc + µck
bc+µc6=0 kbc + µck
≤ sup kdk ≤ 1kad + λdk · sup kac + λck = k(a, λ)kk(b, µ)k.
kck≤1

We verify that the norm verifies the C ∗ -identity. To this end, we first observe that in any
C ∗ -algebra, we have the identity

kak = sup kabk = sup kbak, (1.5)

kbk≤1 kbk≤1

7
since on the one hand, if kbk ≤ 1, then kabk ≤ kakkbk ≤ kak by submultiplicativity, and
on the other hand, the C ∗ -identity implies kak = kabk for b = kak−1 a∗ . Now
k(a, λ)∗ (a, λ)k = k(a∗ a + λa + λa∗ , |λ|2 )k
= sup k(a∗ a + λa + λa∗ )b + |λ|2 bk
kbk≤1

= sup sup kc(a∗ a + λa + λa∗ )b + c|λ|2 bk

kck≤1 kbk≤1

≥ sup kb∗ (a∗ a + λa + λa∗ )b + b∗ |λ|2 bk

kbk≤1

= sup k(ab + λb)∗ (ab + λb)k

kbk≤1

= sup kab + λbk2

kbk≤1

= k(a, λ)k2 ,
where we used the C ∗ -identity of the norm of A. The inequality k(a, λ)∗ (a, λ)k ≤ k(a, λ)k2
follows from submultiplicativity.
Completeness follows from the following lemma from functional analysis: If V is a normed
vector space and W ⊂ V is a subspace of codimension one which is complete with respect
to the induced norm, then V is also complete. Indeed, W := A has codimension one in
V := A+ , and the norm that (1.4) induces on W its original norm by (1.5). Hence by the
lemma, V = A+ is complete. 
Remark 1.20. The obvious norm k(a, λ)k1 := kak + |λ| turns A+ into a Banach algebra,
but it does not satisfy the C ∗ -property. However, by (1.4), we have k(a, λ)k ≤ k(a, λ)k1 .
Hence the identity map from A+ with the norm k·k1 to A+ with the norm (1.4) is bounded.
It then follows from the open mapping theorem that also its inverse is bounded, hence
both norms are equivalent.
If A, B are C ∗ -algebras and Φ : A → B is a ∗-homomorphism, we obtain a unital ∗-
homomorphism
Φ+ : A+ −→ B + , Φ+ (a, λ) = (Φ(a), λ).
It is straightforward to check that A 7→ A+ , Φ 7→ Φ+ defines a functor (the unitalization
functor) from the category of C ∗ -algebras and ∗-homomorphisms to the category of unital
C ∗ -algebras and unital ∗-homomorphisms. In fact, since εB ◦ Φ = εA , the target category
of the unitalization functor is the category of augmented unital C ∗ -algebras together with
compatible unital ∗-homomorphisms. This observation motivates the definition of the
K0 -functor below.
Remark 1.21. The unitalization of a C ∗ -algebra A has the following universal property.
For any unital C ∗ -algebra B and any ∗-homomorphism Φ : A → B (not necessarily unital
if A is unital), there exists a unique unital ∗-homomorphism Φ+ : A+ → B that restricts
to Φ on A ⊂ A+ . Hence the unitalization is a left adjoint to the forgetful functor from the
category of unital C ∗ -algebra and unital ∗-homomorphisms to the category of C ∗ -algebras
and ∗-homomorphisms.

8
1.4 The Gelfand-Naimark theorem
Lemma 1.22. Let A be a Banach algebra. Then every homomorphism ϕ : A → C is
continuous with kϕk ≤ 1. If A is unital, we have more precisely either kϕk = 1 or ϕ = 0.

Proof. Suppose that 1 < kϕk ≤ ∞. Then there exists a ∈ A with P∞ |ϕ(a)| > kak. Set
a := ϕ(a) a; then ϕ(a ) = 1, but ka k = |ϕ(a)| kak < 1. Set b = n=1 (a0 )n , where the
0 −1 0 0 −1

series converges absolutely as ka0 k < 1. Then b = a0 + a0 b and therefore

ϕ(b) = ϕ(a0 ) + ϕ(a0 )ϕ(b) = 1 + ϕ(b),

a contradiction. Hence kϕk ≤ 1.

If now ϕ 6= 0, then there exists a ∈ A with ϕ(a) 6= 0. Again setting a0 = ϕ(a)−1 a, we have
ϕ(a0 ) = 1. Therefore, if A has a unit,

k1k = 1 = ϕ(a0 ) = ϕ(a0 1) = ϕ(a0 ) ϕ(1) = ϕ(1),

| {z }
=1

hence kϕk ≥ 1. 

Definition 1.23. For a Banach algebra A, the Gelfand space is the set

ΓA := {ϕ : A → C homomorphism, ϕ 6= 0}.

By Lemma 1.22, each ϕ ∈ ΓA is in fact continuous, i.e. an element of the dual space A0
of the Banach space A. Remember that A0 carries the weak-∗-topology, which can be
characterized as the coarsest topology such that for each a ∈ A, the linear functional
a : A0 → C, b
b a(ϕ) = ϕ(a) is continuous.

Lemma 1.24. Let A be a unital Banach algebra and J ⊂ A a maximal ideal. Then J is
closed.

Proof. Let J be a maximal ideal (in particular proper!). Then, by continuity of the
multiplication, its closure J is again an ideal. Since J ⊂ J and J is maximal, we have
either J = J or J = A. The latter means that J is dense in A; we show that this is not
possible. Namely, any aP∈ A with k1 − ak < 1 is invertible, with inverse given by the
Neumann series a−1 = ∞ n
n=0 (1 − a) . On the other hand, if J is dense, it must have a
non-trivial intersection with the open set {a | k1 − ak < 1}, hence contain an invertible
element. But this would imply J = A, which is impossible since J is a proper ideal. 

Proposition 1.25. Let A be a unital commutative Banach algebra.

(a) ΓA equipped with the weak-∗-topology is a compact Hausdorff space, and for each
a ∈ A, we have b
a ∈ C(ΓA ).

(b) Every maximal ideal J ⊂ A is of the form J = ker(ϕ) for ϕ ∈ ΓA .

9
(c) We have ΓA 6= ∅. More precisely, σ(a) = {ϕ(a) | ϕ ∈ ΓA } for all a ∈ A.

(d) For all a ∈ A, σ(a) = σ(b

a).

Proof. (a) We have

\
ΓA = {ϕ ∈ A0 | ϕ(ab) − ϕ(a) − ϕ(b) = 0} ∩ {ϕ ∈ A0 | ϕ(1) = 1}.
a,b∈A

Since the maps A0 → C, ϕ 7→ ϕ(a) are weak-∗-continuous for each a ∈ A (by the
above characterization of the topology), we see that ΓA is closed. On the other hand
Lemma 1.22, we have kϕk = 1 for each ϕ ∈ ΓA , hence ΓA is a subset of the unit ball of A0 ,
which is compact with respect to the weak-∗-topology, by the Banach-Alaoglu theorem.
We conclude that ΓA is a closed subset of a compact set, hence compact. The b a are
continuous, again by the characterization of the weak-∗-topology.
(b) It is a general fact that for a commutative ring A, the quotient A/J by an ideal J
is a field if and only if J is maximal. Suppose that J is a maximal ideal, so that A/J is
a field. Now, any maximal ideal in a Banach algebra is closed by Lemma 1.24, and the
quotient of a Banach algebra by a closed ideal is again a Banach algebra (see Example 1.6).
From Thm. 1.17, we therefore get A/J ∼ = C. The quotient map ϕ : A → A/J ∼ = C is a
homomorphism with ker(ϕ) = J. Conversely, since for ϕ ∈ ΓA , the quotient A/ ker(ϕ) ∼ =
C is a field, ker(ϕ) must be a maximal ideal.
(c) First ΓA 6= ∅ by (b), as any ideal is contained in a maximal ideal (Zorn’s lemma!).
Suppose that λ ∈ / σ(a), so that λ − a is invertible. Then for every ϕ ∈ ΓA , ϕ(λ −
a) = λ − ϕ(a) ∈ C is non-zero by multiplicativity of ϕ. Hence λ ∈ / {ϕ(a) | ϕ ∈ ΓA }.
This shows that {ϕ(a) | a ∈ A} ⊆ σ(a). Conversely, suppose that λ ∈ σ(a). Then
J = {(λ − a)b | b ∈ A} is an ideal. It is proper since 1 ∈ J would imply 1 = (λ − a)b for
some b ∈ A, hence b = (λ − a)−1 , a contradiction. Therefore, J is contained in a maximal
ideal, which by (b) is of the form ker(ϕ) with ϕ ∈ ΓA . We obtain ϕ((λ − a)b) = 0 for each
b ∈ A; in particular for b = 1, we get ϕ(a) = λ, hence σ(a) ⊆ {ϕ(a) | ϕ ∈ ΓA }.
(d) Since b a ∈ C(ΓA ), Example 1.12 gives

a) = {b
σ(b a(ϕ) | ϕ ∈ ΓA } = {ϕ(a) | ϕ ∈ ΓA }.

But this equals σ(a) by (c). 

Theorem 1.26 (Gelfand representation). Let A be a commutative unital Banach al-

gebra. Then A → C(ΓA ), a 7→ b
a, called Gelfand transform, is a unital homomorphism
with
kb
ak∞ = sup{|λ| | λ ∈ σ(a)} ≤ kak.

Proof. We have

(b
abb)(ϕ) = b
a(ϕ)bb(ϕ) = ϕ(a)ϕ(b) = ϕ(ab) = ab(ϕ),
b eb(ϕ) = ϕ(e) = 1.

10
Hence the Gelfand transform is a unital algebra homomorphism. Moreover,

kb
ak∞ = sup{|b
a(ϕ)| | ϕ ∈ ΓA } = sup{|ϕ(a)| | ϕ ∈ ΓA } = sup{|λ| | λ ∈ σ(a)},

where in the last step, we used Prop. 1.25(c). By Prop. 1.14(b), this is estimated by kak.

Theorem 1.27 (Gelfand-Naimark). Let A be a commutative unital C ∗ -algebra. Then

the Gelfand transform A → C(ΓA ), a 7→ b
a is an isometric ∗-isomorphism.

Proof. We show that the Gelfand transform is a ∗-homomorphism, i.e. ab∗ = b a. First let
a be self-adjoint, for which we have to show that b
a is real. Because a is self-adjoint, we
have σ(a) ⊂ R (Prop. 1.14(e)). But by Prop. 1.25(d) and Example 1.12, we have

R ⊃ σ(a) = σ(b
a) = {b
a(ϕ) | ϕ ∈ ΓA }.

a is real-valued. The general case follows from writing

Hence b
1 1
a = b + ic = (a + a∗ ) + i (a − a∗ ).
2 2i
Then b and c are self-adjoint and by the previous step,

ab∗ = bb∗ − icb∗ = bb − ib

c = bb + ib
c=b
a.

We show that the Gelfand transform is isometric. Since A is commutative, any a ∈ A is

normal, hence
kak = sup{|λ| | λ ∈ σ(a)} (Prop. 1.14(d))
= sup{|λ| | λ ∈ σ(b
a)} (Prop. 1.25(d))
= sup{ba(ϕ) | ϕ ∈ ΓA } (Example 1.12)
= kb
ak ∞ .
b ⊆ C(ΓA ) is dense. To this end,
We use the theorem of Stone-Weierstraß to show that A
we have to show that
b seperates points, i.e. if for ϕ, ψ ∈ ΓA , we have b
(i) A a(ψ) for all a ∈ A, then
a(ϕ) = b
0
ϕ = ψ. But this is clear since ϕ = ψ ∈ A if ϕ(a) = ψ(a) for all a ∈ A.

(ii) No evaluation functional vanishes, i.e. for all ϕ ∈ ΓA , there exists b a ∈ Ab with
a(ϕ) 6= 0. Again, this is clear, because if for some ϕ ∈ Γ, one has b
b a(ϕ) = ϕ(a) = 0
for all a ∈ A, then ϕ = 0, hence ϕ ∈ / ΓA .

(iii) A a ∈ A,
b is closed under complex conjugation. But this follows since for b b ba = ab∗ ∈ A.
b

We conclude that Ab is dense in C(ΓA ). But since a 7→ b

a is isometric, A
b is also closed,
hence A
b = C(ΓA ). 

11
Remark 1.28. It follows from the proof that if A is a C ∗ -algebra any algebra homomor-
phism ϕ : A → C is automatically ∗-preserving. Namely, for any a ∈ A,

ϕ(a∗ ) = ab∗ (ϕ) = b

a(ϕ) = ϕ(a).

Theorem 1.29 (Spectral permanence). Let A be a unital C ∗ -algebra and let B be a

closed subalgebra containing the unit. Then for any a ∈ B, we have σA (a) = σB (a).

Proof. Clearly, σA (a) ⊆ σB (a). Indeed, if λ − a is not invertible in A, then it cannot be

invertible in B. To show the converse, it suffices to show that if a ∈ B is invertible in A,
then a−1 ∈ B (i.e. a is even invertible in B).
Suppose first that a is self-adjoint and let B0 ⊆ A be the unital C ∗ -algebra generated by
a and a−1 and let B00 ⊆ B be the unital C ∗ -algebra generated by a. We want to show that
B0 ⊆ B, and we will do this by establishing that B0 = B00 . To this end, notice that both
B0 and B00 are commutative since a is self-adjoint. Hence by Thm. 1.27, B0 ∼ = C(ΓB0 ) is
−1 0
c ⊆ C(ΓB ) is generated by
generated by the functions b a and ba and the subalgebra B 0 0
the function ba. First observe that for all ϕ ∈ ΓB0 , we have 1 = ϕ(1) = ϕ(a)ϕ(a−1 ), hence
ϕ(a−1 ) = 1/ϕ(a). We apply the theorem of Stone-Weierstraß. To this end, we have to
show
c0 separates points: For ϕ, ψ ∈ ΓB , suppose that f (ϕ) = f (ψ) for all f ∈ B
(i) B c0 . Then
0 0 0
−1
in particular for f = b
a, i.e. ϕ(a) = ψ(a). But then also ϕ(a ) = 1/ϕ(a) = 1/ψ(a) =
ψ(a−1 ), hence ϕ and ψ agree on all Laurent polynomials in a. Since those are dense
in B0 , we must have ϕ = ψ.

(ii) No evaluation functional vanishes: Suppose that there exists ϕ ∈ ΓB0 such that
c0 . Then in particular ϕ(a) = 0, a contradiction to 1 =
f (ϕ) = 0 for all f ∈ B 0
−1
ϕ(a)ϕ(a ).
c0 is closed with respect to complex conjugation: This is clear, since B 0 is a ∗-algebra
(iii) B0 0
and the Gelfand transform is ∗-preserving.

We conclude that B c0 is dense in C(ΓB ). But it is also closed, hence B c0 = C(ΓB ) and
0 0 0 0
0
B0 = B0 , which was to show.
Finally, let a ∈ B be invertible but not necessarily self-adjoint. Then (a−1 )∗ is an inverse
for a∗ , hence the self-adjoint element a∗ a ∈ B is invertible in A with inverse (a∗ a)−1 =
a−1 (a∗ )−1 . By the previous step, (a∗ a)−1 ∈ B, hence also a−1 = (a∗ a)−1 a∗ ∈ B. 

Theorem 1.30 (Continuous functional calculus). Let A be a unital C ∗ -algebra and

let a ∈ A be normal. Then there exists an isometric ∗-homomorphism C(σ(a)) → A,
f 7→ f (a), such that the identity function on σ(a) is mapped to a.

Proof. Let A0 ⊂ A be the C ∗ -algebra generated by a, in other words the closure of the
subalgebra of all polynomials in a and a∗ . Since a is normal, A0 is commutative, hence
Gelfand transform gives an isomorphism A0 ∼ = C(ΓA0 ).

12
We claim that b a : ΓA0 → {b a(ϕ) | ϕ ∈ ΓA0 } is a homeomorphism (here we used Exam-
a is surjective. We claim that b
ple 1.12 and Prop. 1.25(d)). Clearly b a is injective. Suppose
a(ϕ) = b
that b a(ψ), i.e. ϕ(a) = ψ(a). Then also

ϕ(a∗ ) = b
a(ϕ∗ ) = b
a(ϕ) = b a(ψ ∗ ).
a(ψ) = b

Since ϕ and ψ are multiplicative, ϕ and ψ agree on finite sums of an (a∗ )m . Since ϕ is
continuous, ϕ = ψ everywhere so that ba is injective. Now b
a is a bijective continuous map
between two Hausdorff spaces. We have to show that the inverse f = b a−1 is continuous.
To this end, let K ⊂ ΓA0 be closed. Then the preimage of K under f is f −1 (K) = b a(K).
Because ΓA0 is compact, so is K. Since b a is continuous, b
a(K) is compact, hence closed.
This shows that the preimages of closed sets under f are closed, so f is continuous.
Finally, the calculation
{b
a(ϕ) | ϕ ∈ ΓA0 } = σ(b
a) (Example 1.12)
= σA0 (a) (Prop. 1.25(d))
= σA (a) (Thm. 1.29).
finishes the proof. 

Using the continuous functional calculus, one can show the following result on polar
decomposition in a unital C ∗ -algebra A.
Corollary 1.31 (Polar decomposition). Let A be a unital C ∗ -algebra and let a ∈ A
be invertible. Then there exists a unitary u ∈ A such that a = u|a|.

Here a∗ a is self-adjoint and has non-negative spectrum by Prop. 1.14(f), hence its square-
root |a| = (a∗ a)1/2 can be defined using Thm. 1.30.
Proof. Clearly, u is given by u = a|a|−1 . We have to show that it is unitary. To this end,
we calculate

u∗ u = (a∗ a)−1/2 a∗ a(a∗ a)−1/2 = (a∗ a)−1/2 (a∗ a)−1/2 a∗ a = 1,

using that the map C(σ(a∗ a)) → A is an algebra homomorphism. Showing that uu∗ = 1
is more involved.
We first claim that σ(a∗ a) = σ(aa∗ ). To this end, let λ ∈ ρ(a∗ a), in other words λ − a∗ a
is invertible. Since a is invertible, this is equivalent to the invertibility of (λ − a∗ a)a∗ =
a∗ (λ − aa∗ ), which is then equivalent to λ − aa∗ being invertible. Hence ρ(a∗ a) = ρ(aa∗ ),
which implies the result on the spectra.
We now claim that f (a∗ a)a∗ = a∗ f (aa∗ ) for all f ∈ C(σ(a∗ a)) = C(σ(aa∗ )). Because of
the calculation

(a∗ a)k a∗ = (a∗ a)(a∗ a) · · · (a∗ a)a∗ = a∗ (aa∗ )(a · · · a∗ )(aa∗ ) = a∗ (aa∗ )k ,

this is true for any polynomial f . Since polynomials are dense in C(σ(a∗ a)) by the
Weistraß approximation theorem, the claim follows.

13
Finally, we have

uu∗ = a(a∗ a)−1/2 (a∗ a)−1/2 a∗ = a(a∗ a)−1 a∗ = aa∗ (aa∗ )−1 = 1,

where in the second step, we used the identity f (a∗ a)a∗ = a∗ f (aa∗ ) with f (x) = x−1/2 .

Corollary 1.32. Let A, B be unital C ∗ -algebras and let Φ : A → B be an injective unital

∗-homomorphism. Then Φ is isometric.

Proof. Assume the converse. Since by Prop. 1.16, Φ is contractive, there this would mean
that there exists a ∈ A with kak = 1, but kΦ(a)k =: λ0 < 1. By the C ∗ -identity,
also ka∗ ak = 1 and Since Φ is a ∗-homomorphism, also kΦ(a∗ a)k = λ20 < 1. Therefore,
we can choose a continuous function f ∈ C(σ(a∗ a)) such that f (λ) = 0 on [0, λ20 ] and
f (1) = 1. As seen in the proof of Prop. 1.16, σ(Φ(a∗ a)) ⊂ σ(a∗ a). We claim that
Φ(f (a∗ a)) = f (Φ(a∗ a)). Indeed, this holds for f a polynomial since Φ is a homomorphism;
the general case follows since polynomials are dense in C(σ(a∗ a)).
Now ka∗ ak = 1 means that 1 ∈ σ(a∗ a) (Prop. 1.14(d)), hence

kf (a∗ a)k = sup{|f (λ)k | λ ∈ σ(a∗ a)} = 1.

But since kΦ(a∗ a)k = λ20 , we have σ(Φ(a∗ a)) ⊂ [0, λ20 ], hence Φ(f (a∗ a)) = f (Φ(a∗ a)) = 0.
This contradicts the injectivity of Φ. 

Definition 1.33 (Representation). Let A be a C ∗ -algebra.

(a) If H is a Hilbert space, a ∗-homomorphism ρ : A → B(H) is called a ∗-representation.

(b) A ∗-representation is called faithful if it is injective.

(c) A ∗-representation on a Hilbert space H is called irreducible if whenever V ⊆ H is a

closed subspace such that ρ(a)v = 0 for all a ∈ A, v ∈ H, then V = {0}.

Theorem 1.34 (Gelfand-Naimark, non-commutative version). Let A be a C ∗ -algebra.

Then there exists a Hilbert space H together with a faithful and isometric ∗-representation
ρ : A → B(H).

Proof (Sketch). After possibly replacing A by A+ , we may assume that A is unital. The
Gelfand space ΓA is “too small” to characterize A when it is not commutative; indeed,
since for ϕ ∈ ΓA ,
ϕ(ab − ba) = ϕ(a)ϕ(b) − ϕ(b)ϕ(a) = 0,
the Gelfand space ΓA only depends on the commutator subspace A/[A, A].
In the non-commutative case, we therefore instead consider the space

SA := {ϕ : A → C continuous | ∀a ∈ A : ϕ(a∗ a) ≥ 0, kϕk = 1.} ⊂ A0 ,

14
where we give up on the requirement that ϕ is multiplicative. For any such ϕ, the obtain
a (semi-) positive Hermitean form ha, bi := ϕ(a∗ b) on A. The corresponding completion
Hϕ is a Hilbert space that comes with a ∗-representation ρϕ : A → B(H) defined by
ρϕ (a)[b] := [ab]. One then defines
M M
H= Hϕ , ρ= ρϕ (1.6)
ϕ∈SA ϕ∈SA

and shows that the corresponding ρ is faithful. It is isometric by Corollary 1.32. 

Remark 1.35. As obvious from formula (1.6), the representation of A constructed in

the proof above is typically not separable (namely as soon as SA is an uncountable set).
Some algebras in fact to not have a separable representation, e.g. the Calkin algebra
Q(H) := B(H)/K(H), for H a separable Hilbert space [5, Satz IX.3.16].

1.5 The spatial tensor product

In this section, we define the spatial tensor product of C ∗ -algebras, in particular in order
to put C ∗ -norms on matrix algebras Mn (A) over C ∗ -algebras A. General references for
the theory of tensor products on C ∗ -algebras are [4, §T.5] and [1, §3].
For ∗-algebras A, B, the algebraic tensor product A ⊗alg B is an algebra, with product
determined and well-defined (!) by a1 ⊗ b1 · a2 ⊗ b2 = a1 a2 ⊗ b1 b2 for a1 , a2 ∈ A, b1 , b2 ∈ B.
It is moreover a ∗-algebra with the ∗-operation (a ⊗ b)∗ = a∗ ⊗ b∗ .
Definition 1.36 (Spatial tensor product). Let A and B be C ∗ -algebras.
(a) The spatial norm on A ⊗alg B is defined by
n
X
kxkσ := sup k(ρA ⊗alg ρB )(x)k = sup ρA (ai ) ⊗ ρB (bi ) (1.7)
i=1
Pm
for x = n=1 an ⊗bn ∈ A⊗alg B, where the supremum is taken over all ∗-representations
ρA , ρB of A and B on Hilbert spaces H, K.
(b) The spatial tensor product of A and B, denoted by A⊗B, is the completion of A⊗alg B
with respect to the spatial norm.

Some comments on the definition of the spatial norm are in order. First, any pair of
∗-representations ρA , ρB on Hilbert spaces H, K defines a ∗-representation

ρA ⊗alg ρB : A ⊗alg B −→ B(H) ⊗alg B(K) ⊆ B(H ⊗ K).

Here an operator X ⊗ YP∈ B(H) ⊗alg B(K)

P is viewed as operator in B(H ⊗ K); explicitly,
it is given by (X ⊗ Y )( i vi ⊗ wi ) = i X(vi ) ⊗ Y (wi ). The norm is finite, since
n
X n
X n
X
ρA (ai ) ⊗ ρB (bi ) ≤ kρA (ai )kkρB (bi )k ≤ kai kkbi k; (1.8)
i=1 i=1 i=1

15
here we used that ∗-homomorphisms are contractive, by (1.16). It is non-degenerate, as
any C ∗ -algebra has a faithful representation on a Hilbert space (by the Gelfand-Naimark
theorem 1.34) and the induced representation ρA ⊗alg ρB is injective if ρA and ρB are (see
e.g. [4, T.5.1]). Moreover, it is clear from the definition that k · kσ satisfies the C ∗ -identity,
hence A ⊗ B is a C ∗ -algebra.
Remark 1.37. Any pair of representations ρA , ρB of A and B induces a C ∗ -seminorm
on A ⊗alg B by pulling back the operator norm along ρA ⊗alg ρB . Now if ρ0A , ρ0B is any
other pair of representations, then the direct sum representation ρA ⊕ ρ0A , ρB ⊕ ρ0B clearly
induces a larger seminorm this way. This shows that in the supremum in (1.8), it suffices
to only consider faithful representations, because ρA ⊕ ρ0A is faithful as soon as one of
ρA , ρ0A is faithful. It is also easy to see that we may restrict attention to irreducible
representations.

Lemma 1.38. For any C ∗ -algebra A, we have A ⊗ Mn (C) ∼

= Mn (A).
Proof. Clearly, Mn (A) ∼= A ⊗alg Mn (C), so the point is to show that A ⊗alg Mn (C) is
already complete with respect to the spatial norm.
Let ρ be a faithful ∗-representation of A. The representation ρn : Mn (A) → B(H n )
induced by ρ as in (1.3) takes the form
     
a11 · · · a1n v1 Φ(a11 )v1 + · · · + Φ(a1n )vn
ρn  ... .. ..   ..  =  ..
. (1.9)
 
. .   .   .
an1 · · · ann vn Φ(an1 )v1 + · · · + Φ(ann )vn

We check that ρn (Mn (A)) is complete in B(H n ). To this end, we observe that for any
X = (Xij )1≤i,j≤n ∈ B(H n ), one has for each i, j = 1, . . . , n
  2
X11 · · · X1n n
 .. . . X 2
.. ..  = sup Xi1 v1 + · · · + Xin vn ≥ kXij k2 .

 .
Xn1 · · · Xnn i=1 kvk=1

(m)
We conclude that, if a sequence a(m) = (aij )1≤i,j≤n ∈ Mn (A), m ∈ N, is is such that
(m)
ρn (a(m) ) → X ∈ B(H n ), then also ρ(aij ) → Xij for all 1 ≤ i, j ≤ n. Since the image of
ρ is closed, this implies Xij = ρ(aij ) for some aij ∈ A, hence X is in the image of ρn .
Now since the image of ρn is closed, pulling back the norm on B(H n ) to Mn (A) via ρn
gives a complete norm on Mn (A) satisfying the C ∗ -identity. We have to show that this
norm coincides with the spatial norm. By Remark 1.37, it suffices to consider faithful,
irreducible representations in the definition of the spatial norm; for Mn (C), any such
representation is isomorphic to the standard representation idMn (C) : Mn (C) → B(Cn ) =
Mn (C). Also, under the isomorphism Mn (A) ∼ = A ⊗ Mn (C), we have ρn = ρ ⊗alg idMn (C) .
Combining these two observations, we conclude that the spatial norm of a ∈ Mn (A) ∼ =
A ⊗alg Mn (C) is given by
kakσ = sup kρn (a)k,
ρ

16
where the supremum is taken over all faithful representations ρ of A. On the other hand,
we have seen above that each of the norms kakρ := kρn (a)k turns Mn (A) into a C ∗ -algebra,
hence they must all be equal, by Prop. 1.15. This finishes the proof. 

Theorem 1.39. Let A and B be C ∗ -algebras. Then the supremum (1.7) is in fact a
maximum, which is taken at any pair of faithful representations ρA , ρB . In other words,
for any x ∈ A ⊗alg B and any pair of faithful representations ρA , ρB of A, B, we have

kxkσ = k(ρA ⊗alg ρB )(x)k.

In particular, if A ⊆ B(H), B ⊆ B(K) are C ∗ -subalgebras, then A ⊗ B is isomorphic to

the norm closure of A ⊗alg B ⊆ B(H ⊗ K).

Proof. We first observe that by Lemma 1.38, the theorem is true if A is isomorphic to
Mn (C). In particular, we have

kxkσ = k(idA ⊗ ρB )(x)k

for any faithful representation ρB of B. Namely, for any such representation, the right
hand side gives a complete C ∗ -norm on Mn (C) ⊗ B, which then must be all equal.
In general, let ρA and ρB representations of A and B and let ρ0B be faithful representation
of B. We will show that for all x ∈ A ⊗alg B, we have

k(ρA ⊗alg ρB )(x)k ≤ k(ρA ⊗alg ρ0B )(x)k. (1.10)

By symmetry of the tensor product construction, the same is true when replacing ρA by
a faithful representation ρA0 , and the proposition follows.
To begin with, let V be the directed system of finite-dimensional subspaces of H (see
Example 2.26). For V ∈ V , let PV be the orthogonal projection onto V in H, and let
PV0 := PV ⊗ idK , the orthogonal projection onto V ⊗ K in H ⊗ K. It is then an easy
lemma from functional analysis that

kXk = lim k(PV ⊗ idB(K) )X(PV ⊗ idB(K) )k,

V

for all X ∈ B(H ⊗ K), where the limit is taken in the sense of nets.
We obtain that for any x ∈ A ⊗alg B,

k(ρA ⊗alg ρB )(x)k = lim k(PV ρA PV ⊗alg ρB )(x)k.

V

For each V ∈ V , the map PV ρA PV ⊗alg ρB is the composition of the linear map PV ρA PV ⊗
idB : A ⊗alg B → B(V ) ⊗alg B and the ∗-homomorphism idB(V ) ⊗alg ρB : B(V ) ⊗alg B →
B(V ⊗ K). Therefore, for any x ∈ A ⊗alg B, we have

k(ρA ⊗alg ρB )(x)k = lim k(idB(V ) ⊗ ρB )(PV ρA PV ⊗ idB )(x)k.

V

17
Now since idB(V ) ⊗ ρB is a ∗-homomorphism, hence contractive (Prop. 1.16), we have
k(idB(V ) ⊗ ρB )(PV ρA PV ⊗ idB )(x)k ≤ k(PV ρA PV ⊗ idB )(x)kσ , (1.11)
where the right hand side denotes the spatial norm of B(V ) ⊗alg B = B(V ) ⊗ B. Moreover,
by (the proof of) Lemma 1.38, we have equality in (1.11) if ρ0B is faithful. Combining
these observations, we get that for ρB an arbitrary ∗-representation and ρ0B a faithful
∗-representation, we have
kPV ρA PV ⊗alg ρB )(x)k ≤ k(PV ρA PV ⊗alg ρ0B )(x)k
for all V ∈ V . Taking the limit over V , we obtain (1.10), which finishes the proof. 
If Φ : A → A0 and Ψ : B → B 0 are ∗-homomorphisms, we get an induced ∗-homomorphism
Φ ⊗alg Ψ : A ⊗alg B → A0 ⊗alg B 0 . It is continuous because for ∗-representations ρA0 and
ρB 0 of A0 and B 0 , ρA0 ◦ Φ and ρB 0 ◦ Ψ are ∗-representations of A, respectively B. Therefore
Φ ⊗alg Ψ extends by continuity to a ∗-homomorphism Φ ⊗ Ψ : A ⊗ B → A0 ⊗ B 0 . We
record the following consequence of Thm. 1.39 for later use.
Corollary 1.40. Let A, A0 , B and B 0 be C ∗ -algebras and let Φ : A → A0 , Ψ : B → B 0
be injective ∗-homomorphisms. Then Φ ⊗ Ψ : A ⊗ B → A0 ⊗ B 0 is injective.
Proof. If ρA0 and ρB 0 are faithful representations of A0 , respectively B 0 , then ρA := ρA0 ◦ Φ
and ρB := ρB 0 ◦ Ψ are faithful representations of A, respectively B. Hence for all x ∈
A ⊗alg B,
k(Φ ⊗alg Ψ)(x)kσ = k(ρA0 ⊗alg ρB 0 )(Φ ⊗alg Ψ)(x)k = k(ρA ⊗ ρB )(x)k = kxk.
This shows that Φ ⊗ Ψ is isometric, in particular injective. 
Example 1.41. Let A be a C ∗ -algebra and let X be a compact Hausdorff space. Then
C0 (X) ⊗ A ∼ = C0 (X, A), the C ∗ -algebra of continuous A-valued functions on X vanishing
at infinity.
To see this, observe first there is an obvious injective ∗-homomorphism Φ : C0 (X)⊗alg A →
C0 (X, A), given by Φ(f ⊗ a)(t) = f (t)a. To see that its image is dense, one first observes
that Cc (X, A) (compactly supported functions) is dense in C0 (X, A), hence it suffices to
approximate a given compactly supported function f . This is done using a partition of
unity subordinate to a suitable finite open cover of the support of f . For details, see for
example [4, §T.2, p. 322].
On the other hand, we claim that Φ is continuous with respect to the spatial norm on
C(X)⊗alg A. To this end, for a Hilbert space H, let `2 (X, H)P be Hilbert space of functions
α : X → C with countable support x1 , x2 , · · · ∈ X such that ∞ 2
n=1 |α(xn )| < ∞; we write
`2 (X) if H = C. Now we have a faithful ∗-representation µ : C(X) → `2 (X), given by
µ(f )α = f · α, and given a faithful representation ρA of A on a Hilbert space H, the
∗-representation θ ⊗alg ρA of C(X) ⊗alg A on `2 (X) ⊗ H = `2 (X, H) takes has an obvious
extension to an injective (hence isometric) ∗-representation ρ of C(X, A) on `2 (X, H) such
that ρ ◦ Φ = θ ⊗alg ρA . Now for any x ∈ C(X) ⊗alg A, we have
kΦ(x)k = k(ρ ◦ Φ)(x)k = k(θ ⊗alg ρA )(x)k = kxkσ .
Here in the last step, we used Thm. 1.39. This finishes the proof.

18
2 The K0-functor
In this section, we define the K0 -group K0 (A) of a C ∗ -algebra A. We then state and prove
its main properties.

2.1 Equivalence of projections

Throughout this section, for C ∗ -algebras A, we denote by A
e the C ∗ -algebra given by A if
+
A is unital and A if A is non-unital.
Definition 2.1 (Projections, Isometries, Unitaries). Let A be a C ∗ -algebra.
(a) p ∈ A is called projection if p∗ = p and p2 = p.

(b) v ∈ A is called partial isometry if v ∗ v is a projection.

(c) If A is unital, u ∈ A is called unitary if u∗ u = uu∗ = 1.

Lemma 2.2. Let A be a C ∗ -algebra and let v ∈ A be a partial isometry, so that p = v ∗ v

is a projection. Then also q = vv ∗ is a projection. Moreover,

v = vv ∗ v = vp = qv, v ∗ = v ∗ vv ∗ = pv ∗ = v ∗ q. (2.1)

Proof. By the C ∗ -identity, we have

kv − vv ∗ vk2 = kv(1 − v ∗ v)k2 = k(1 − v ∗ v)v ∗ v(1 − v ∗ v)k = k(1 − p)p(1 − p)k = 0,

hence v = vv ∗ v. Taking the transpose shows v ∗ = v ∗ vv ∗ , so that we have established

(2.1). q is a projection because it is self-adjoint and q 2 = vv ∗ vv ∗ = vv ∗ = q, using (2.1).

Definition 2.3 (Equivalence of projections). Let A be a C ∗ -algebra and let p, q ∈ A

be projections.
(a) p, q are called Murray-von-Neumann equivalent, denoted p ∼ q, if there exists a partial
isometry v ∈ A such that v ∗ v = p and vv ∗ = q.

(b) p, q are called unitarily equivalent, denoted p ∼u q, if there exists a unitary u ∈ A

e
such that q = upu∗ .

(c) p, q are called homotopy equivalent, denoted p ∼h q, if there exists a continuous path
(pt )t∈[0,1] of projections in A, such that p0 = p, p1 = q. Such a path is called homotopy
between p and q.

Lemma 2.4. All the relations in Def. 2.3 are equivalence relations on the set of projec-
tions in A.

19
Proof. The only non-trivial part is the transitivity of Murray-von-Neumann equivalence.
Let p, q, r ∈ A be three projections, and let v, w ∈ A be partial isometries with v ∗ v = p,
vv ∗ = q, w∗ w = q, ww∗ = r. Then

(wv)∗ (wv) = v ∗ w∗ wv = v ∗ qv = v ∗ vv ∗ v = p2 = p,

(wv)(wv)∗ = wvv ∗ w = wqw∗ = ww∗ ww∗ = r2 = r.

Lemma 2.5. Let A be a C ∗ -algebra and let p and q be projections in A with kp−qk < 1/2.
e with q = upu∗ .
Then there exists a unitary u ∈ A

Proof. Set a = qp + (1 − q)(1 − p) ∈ A.

e It satisfies

qa = qp = ap. (2.2)

Because

k1 − ak = kq + p − 2qpk = k(1 − q)(p − q) − q(p − q)k ≤ 2kp − qk < 1,

the element a is invertible, with a−1 = ∞ n

P
n=0 (1 − a) (Neumann series), and from (2.2)
−1
follows q = apa .
To obtain a unitary u with q = upu∗ , we take the unitary u = a|a|−1 from the polar
decomposition of a, see Corollary 1.31 (remember that |a| = (a∗ a)1/2 ).
We claim that p commutes with |a|−1 . First, p commutes with |a|2 because of the calcu-
lation
|a|2 p = a∗ ap = a∗ qa = (qa)∗ a = (ap)∗ a = pa∗ a = p|a|2 ,
where we used (2.2). We conclude that it commutes with all elements of the C ∗ -subalgebra
B ⊆ A generated by |a|2 and 1. Since |a|2 is invertible, B contains |a|−1 , proving the
claim.
Using (2.2) again, we calculate

upu∗ = a|a|−1 pu∗ = ap|a|−1 u∗ = qa|a|−1 u∗ = quu∗ = q. 

Proposition 2.6 (Equivalence of equivalences). Let A be a C ∗ -algebra and let p, q

be projections in A. Then

(a) p ∼h q =⇒ p ∼u q;

(b) p ∼u q =⇒ p ∼ q;
   
p 0 q 0
(c) p ∼ q =⇒ ∼u in M2 (A);
0 0 0 0
   
p 0 q 0
(d) p ∼u q =⇒ ∼h in M2 (A).
0 0 0 0

20
Proof. (a) Let (pt )t∈[0,1] be a homotopy between p and q and choose a subdivision 0 =
t0 < t1 < · · · < tn = 1 such that kpti − pti−1 k < 1/2 for all i = 1, . . . , n. By Lemma 2.5,
there exist unitaries ui in Ae such that pt = ui pt u∗ for each i, hence with u = un · · · u1 ,
i i−1 i
we have p1 = up0 u∗ .
(b) If q = upu∗ for some unitary u ∈ A, e then with v = up, we have v ∗ v = pu∗ up = p2 = p
and vv ∗ = up2 u∗ = upu∗ = q.
(c) Let v ∈ A be a partial isometry with v ∗ v = p, vv ∗ = q. We need to find an element
^ ∗
u∈M 2 (A) with u diag(p, 0)u = diag(q, 0). To this end, define elements of M2 (A) by
e
   
v 1−q q 1−q
w := , s := .
1−p v∗ 1−q q

Clearly, s is unitary. Moroever,

 ∗
v v + (1 − p)2 v ∗ − v ∗ q + v ∗ − pv ∗
  
∗ 1 0
w w= = ,
v − qv + v − vp (1 − q)2 + vv ∗ 0 1

where we used the identities (2.1). Similarly, one calculates ww∗ = diag(1, 1), hence u is
unitary. Also
 ∗
vpv ∗ 0
      ∗   
p 0 ∗ pv 0 vv q 0 q 0
w w =w = = = ,
0 0 0 0 0 0 0 0 0 0

again using (2.1). However, if A is non-unital, we generally have w ∈ /M^ 2 (A). To repair
∗
this, set u = sw and notice that also u diag(p, 0)u = diag(q, 0) and since A ⊆ A e is an
ideal,
qv + (1 − q)(1 − p) (1 − q)v ∗
 
u = sw = ∈M ^2 (A),
(1 − q)v + q(1 − p) 1 − q + qv ∗
(d) Let u ∈ Ae be a unitary such that q = upu∗ . For t ∈ [0, 1], define elements of M2 (A)
e
by
  ∗
cos πt πt


    
− sin u 0 u 0 ∗ p 0
rt := 2
2
, wt := rt r , pt := wt wt∗ .
sin πt
2
cos πt
2
0 1 0 1 t 0 0
e with w0 = diag(1, 1), w1 = diag(u, u∗ ) and
Then (wt )t∈[0,1] is a path of unitaries in M2 (A)
(pt )t∈[0,1] is a path of projections in M2 (A) with p0 = diag(p, 0) and p1 = diag(q, 0). 

2.2 The Grothendieck construction

Remember that a semigroup is a non-empty set S together with a map S ×S → S, (g, h) 7→
g · h, that satisfies (g · h) · k = g · (h · k) for all g, h, k ∈ S (associativity). A semigroup S is
abelian if the multiplication is commutative, g · h = h · g for all g, h ∈ S. A homomorphism
between semigroups S, T is a map ϕ : S → T such that ϕ(gh) = ϕ(g)ϕ(h) for all g, h ∈ S.
Any group is in particular a semigroup.

21
Definition 2.7 (Group completion). Let S be an abelian semigroup. Then a group
completion or Grothendieck group is an abelian group G(S), together with a homomor-
phism ιS : S → G(S) satisfying the following universal property: For every abelian
group H together with a homomorphism ρ : S → H, there is a unique homomorphism
ρe : G(S) → H such that
ιS
S G(S)
∃!
commutes.
ρ

Proposition 2.8 (Grothendieck construction). Let S be an abelian semigroup. Then

a group completion exists and is unique up to unique isomorphism. In fact, there exists
a functor G from abelian semigroups to abelian groups such that for every semigroup S,
G(S) is a group completion of S.

Proof (Sketch). A concrete model is G(S) := S × S/ ≈, where the equivalence relation ≈

is defined by (g, h) ≈ (g 0 , h0 ) if there exists k ∈ S such that g + h0 + k = g 0 + h + k (the
additional k is needed to show transitivity of the relation, in case that S does not have
the cancellation property g + k = h + k ⇒ g = h). One then shows that G(S) is a group
and sets ιS (g) = [g + k, k] for any k ∈ S (any choice of k yields the same element). To
show the universal property, observe that for any g, h ∈ S, [g, h] = ιS (g) − ιS (h), hence
for a given homomorphism ρ : S → H, we must have ρe([g, h]) = ρ(g) − ρ(h); this indeed
gives a well-defined group homomorphism ρe : G(S) → H.
If now ϕ : S → T is a homomorphism of semigroups, then ρ := ιT ◦ ϕ is homomorphism
S to the group G(T ), so by the universal property, there exists a unique group homomor-
phism G(ϕ) := ρe : G(S) → G(T ). One then verifies that for a second homomorphism
ψ : T → U , one has the functoriality G(ψ) ◦ G(ϕ) = G(ψ ◦ ϕ). 

Remark 2.9. By the above, elements of G(S) can be represented by equivalence classes
of pairs of elements in S. We suggestively write g − h := [g, h] for g, h ∈ S.

Remark 2.10. By the universal property of the Grothendieck construction, for any
abelian semigroup S and any abelian group H, the canonical map
HomAb (G(S), H) −→ HomSAb (S, H), ϕ 7−→ ϕ ◦ ιS
is a bijection. Since it is natural in both S and H, this shows that the functor G : SAb →
Ab is left adjoint to the forgetful functor Ab → SAb.

Example 2.11. We have G(N) = Z; in fact the left hand side can be taken as a definition
of Z.

Example 2.12. If we set n + ∞ = ∞ + n = ∞ for n ∈ N and ∞ + ∞ = ∞, then N ∪ {∞}

is a semigroup. Moreover, G(N ∪ {∞}) = {0}, because in groups, the cancellation rule
holds, that is, [n] + [∞] = [∞] implies [n] = [0] for all n ∈ N ∪ {∞}.

22
2.3 Definition of K0
We have canonical embeddings Mn (A) → Mn+1 (A) given by
 
  a11 · · · a1n 0
a11 · · · a1n  .. .. .. .. 
 .. .. ..  7−→  . . . .
 . . .   . (2.3)
an1 · · · ann 0
an1 · · · ann
0 ··· 0 0

We denote by M∞ (A) the union of all the Mn (A), that is, the direct limit in the category
of ∗-algebra. M∞ (A) can be described as the ∗-algebra of infinite matrices with entries in
A, with only finitely many non-zero entries. Since all the inclusions Mn (A) → Mn+1 (A)
are isometric, M∞ (A) inherits a norm which satisfies the C ∗ -property but is not complete.

Definition 2.13. Let A be a C ∗ -algebra.

(a) Two projections p, q ∈ M∞ (A) are equivalent, if for some n ∈ N, p, q ∈ Mn (A) and
p ∼ q in Mn (A). The set of equivalence classes is denoted by V(A).

(b) If p, q ∈ M∞ (A) are projections with p ∈ Mn (A), q ∈ Mm (A), we define

[p] + [q] := [diag(p, q)], where diag(p, q) ∈ Mn+m (A) ⊂ M∞ (A).

To show well-definedness of the addition, notice that for all p ∈ Mn (A), q ∈ Mm (A),
   
p p
 q   0 
   

 0 ∼
  q 
 in M∞ (A).

 0 
 
 0 

.. ..
. .

Remark 2.14. By Prop. 2.6, we can use any of the equivalence relations ∼, ∼u , ∼h to
obtain the same set V(A).

Lemma 2.15. Let A be a C ∗ -algebra. Then V(A) is an abelian semigroup.

Proof. Let p, q, r ∈ M∞ (A) be projections with p ∈ Mn (A), q ∈ Mm (A) and r ∈ Ml (A).

Because of diag(diag(p, q), r) = diag(p, q, r) = diag(p, diag(q, r)) in Mn+m+l (A), the semi-
group operation is associative. With
     
0 q ∗ p 0 ∗ q 0
v= , we have v v= , vv = ,
p 0 0 q 0 p

hence v is a partial isometry and diag(p, q) ∼ diag(q, p) in Mn+m (A). This shows that
the semigroup operation is commutative. 

23
If A and B are C ∗ -algebras and Φ : A → B a ∗-homomorphism, one easily checks that

V(Φ) : V(A) → V(B), [p] 7→ [Φ(p)]

gives a well-defined homomorphism of semigroups. If C is another C ∗ -algebra and Ψ :

B → C is a ∗-homomorphism, one clearly has V(Ψ) ◦ V(Φ) = V(Ψ ◦ Φ), hence V is a
well-defined functor from C ∗ -algebras to abelian semigroups.

Definition 2.16 (The K0 -functor). Let A be a C ∗ -algebra.

(a) The group K0 (A) associated to A is defined by

K0 (A) := ker GV(εA ) : GV(A+ ) → GV(C) .

(b) Given another C ∗ -algebra B with a ∗-homomorphism Φ : A → B, define K0 (Φ) :

K0 (A) → K0 (B) as the unique group homomorphism fitting in the commutative
diagram
GV(εA )
K0 (A) GV(A+ ) GV(C)
K0 (Φ) GV(Φ+ ) (2.4)

K0 (B) GV(B + ) GV(εB )

GV(C).

Notice that for C ∗ -algebras A, B and ∗-homomorphisms Φ : A → B, there is indeed

a unique group homomorphism K0 (Φ) : K0 (A) → K0 (B) making (2.5) commute: By
injectivity of the inclusions K0 (A) ,→ GV(A+ ) and K0 (B) ,→ GV(B + ), K0 (Φ) must be
the restriction of GV(Φ+ ) to K0 (A), and if x ∈ K0 (A), then

0 = GV(εA )(x) = GV(εB )GV(Φ+ )(x),

hence indeed GV(Φ+ )(x) ∈ ker GV(εB ) = K0 (B).

Remark 2.17. If A is unital, then A+ ∼ = A ⊕ C with εA being the projection onto the
+
second factor. Therefore, GV(A ) = GV(A) ⊕ GV(C) with GV(εA ) being the projection
onto the second factor. If Φ : A → B is a unital ∗-homomorphism, then GV(Φ+ ) =
GV(Φ)⊕GV(idC ) under this identification, hence on the subcategory of unital C ∗ -algebras
and unital ∗-homomorphisms, the functors GV and K0 are naturally isomorphic. We will
therefore often write K0 instead of GV for unital C ∗ -algebras. In particular, for any
C ∗ -algebra A, we have a short exact sequence
K0 (εA )
0 K0 (A) K0 (A+ ) K0 (C) 0

Lemma 2.18. K0 is a functor from C ∗ -algebras to abelian groups.

24
Proof. Let A, B and C be C ∗ -algebras and Φ : A → B, Ψ : B → C be ∗-homomorphisms.
Consider the following diagram.
GV(εA )
K0 (A) GV(A+ ) GV(C)
K0 (Φ) GV(Ψ◦Φ) GV(Φ+ )

K0 (Ψ◦Φ) K0 (B) GV(B + ) GV(C) (2.5)

K0 (εB )

K0 (Φ) GV(Φ+ )

K0 (C) GV(C + ) K0 (εC )

GV(C)

Commutativity of the left-most triangle is equivalent to the desired equality K0 (Ψ) ◦

K0 (Φ) = K0 (Ψ ◦ Φ). All squares commute by definition of the K0 maps, while commuta-
tivity of the other triangle follows from functoriality of GV. In total, we obtain that the
entire diagram commutes. 

Proposition 2.19 (A portrait of K0 ). Let A be a C ∗ -algebra.

(a) Any element x ∈ K0 (A) can be written in the form x = [p] − [1n ] for some n ∈ N,
with a projection p ∈ M∞ (A+ ) and
 
1
 .. 
.
1n :=   ∈ M∞ (A+ ).
 
 1 
0

Moreover, we can arrange p such that p − 1n ∈ M∞ (A).

(b) For projections p, q ∈ M∞ (A+ ), [p] − [q] = 0 in K0 (A) if and only if there exists
m ∈ N such that diag(p, 1m ) ∼ diag(q, 1m ). Here ∼ can be replaced by ∼u or ∼h .

(c) If B is another C ∗ -algebra and Φ : A → B is a homomorphism, then for all projections

p, q ∈ Mn (A), we have K0 (Φ)([p] − [q]) = [Φ+ (p)] − [Φ+ (q)].

Proof. (a), first part. By definition of the Grothendieck group, any element x ∈ K0 (A)
can be written as x = [p] − [q] with projections p, q ∈ M∞ (A+ ) for some n ∈ N. Since q
is a projection, so is 1n − q, and
 
q 0
[q] + [1n − q] = = [1n ],
0 1n − q
as
     
q 0 ∗ 1n 0 q 1n − q
u u = for the unitary u= .
0 1n − q 0 0 1n − q q

25
Therefore, if p ∈ Mm (A+ ) and q ∈ Mn (A+ ),
 

p 0
[p] − [q] = [p] + [1n − q] − [q] + [1n − q] = − [1n ],
0 1n − q

as claimed.
(b) Suppose that [p]−[q] = 0 in K0 (A) for projections p, q ∈ Mn (A+ ). By the definition of
the Grothendieck group, this implies that there exists a projection r ∈ Mn (A+ ) ⊂ M∞ (A)
such that diag(p, r) ∼ diag(q, r). We then also have diag(p, r, 1n − r) ∼ diag(q, r, 1n − r).
But by a similar calculation to the one just above, we have diag(p, r, 1n − r) ∼ diag(p, 1n )
and diag(q, r, 1n − r) ∼ diag(q, 1n ). Conversely, if diag(p, r) ∼ diag(q, r), then
   
p 0 q 0
[p] − [q] = ([p] + [r]) − ([q] + [r]) = − = 0.
0 r 0 r

(a), second part. Let x = [p] − [1n ] ∈ K0 (A), where p ∈ M∞ (A+ ) is a projection. Since
x ∈ ker K0 (εA ), we have [εA (p)] − [1n ] = 0. By (b), there exists m ∈ N with p ∈ Mm (A+ )
and a unitary u ∈ Mm (C) ⊆ Mm (A+ ) such that uεA (p)u∗ = 1n . Then with the projection
p0 = upu∗ , we still have x = [p0 ] − [1n ], but εA (p0 ) − 1n = 0, hence p0 − 1n ∈ Mm (A).
(c) This follows from the universal property of the Grothendieck group, compare the proof
sketch of Prop. 2.8. 

Example 2.20 (K0 of C). Two projections p, q ∈ Mn (C) are equivalent if and only if
they have the same rank, and the rank is additive under taking direct sums. Therefore
V(C) = N0 and K0 (C) = Z (see Example 2.11). We conclude that the map

τ : K0 (C) −→ Z, [p] − [q] 7−→ tr(p) − tr(q) (2.6)

is a well-defined group isomorphism.

Example 2.21 (K0 of B(H)). Let H be a Hilbert space and A = B(H). Two projec-
tions in Mn (A) ∼= B(H n ) are equivalent if and only if they have the same rank, and the
rank is additive under taking direct sums. Here the rank can be any number in N0 ∪ {∞},
and the Grothendieck group of this semigroup is zero (Example 2.12).

2.4 Homotopy invariance

Definition 2.22 (Homotopy). Let A and B be C ∗ -algebras.

(a) A homotopy between ∗-homomorphisms Φ, Ψ : A → B is a family (Φt )t∈[0,1] of ∗-

homomorphisms such that Φ0 = Φ, Φ1 = Ψ and such that t →
7 Φt (a) is continuous
for every a ∈ A.

(b) Two ∗-homomorphisms Φ, Ψ : A → B are called homotopic if there exists a homotopy

between them.

26
(c) A ∗-homomorphism Φ : A → B is a homotopy equivalence if there exists a ∗-
homomorphism Φ0 : B → A such that both Φ ◦ Φ0 and Φ0 ◦ Φ are homotopic to
the identity.

Example 2.23. If X, Y are compact topological spaces, then a continuous map ϕ :

X → Y induces a ∗-homomorphism ϕ∗ : C(Y ) → C(X), f 7→ ϕ∗ f by pullback. If ϕ
and ψ are two such maps, then a homotopy (of continuous maps) induces a homotopy of
∗-homomorphisms between ϕ∗ and ψ ∗ , and if ϕ is a homotopy equivalence (in the sense
of topology), then ϕ∗ is a homotopy equivalence in the sense of Def. 2.22.

Theorem 2.24 (Homotopy invariance). Let A and B be C ∗ -algebras.

(a) If Φ, Ψ : A → B are homotopic ∗-homomorphisms, then K0 (Φ) = K0 (Ψ).
(b) If Φ : A → B is a homotopy equivalence, then K0 (Φ) : K0 (A) → K0 (B) is an
isomorphism.

Proof. (a) Let (Φt )t∈[0,1] be a homotopy between Φ and Ψ. Let x = [p] − [q] ∈ K0 (A)
with p, q ∈ Mn (A). Then (Φ+ +
t (p))t∈[0,1] and (Φt (q))t∈[0,1] are homotopies of projections
between Φ (p) and Ψ (p), respectively Φ (q), Ψ+ (q). Hence by Prop. 2.19(c),
+ + +

K0 (Φ)(x) = [Φ+ (p)] − [Φ+ (q)] = [Ψ+ (p)] − [Ψ+ (q)] = K0 (Ψ)(x).

(b) If Φ : A → B is a homotopy equivalence with homotopy inverse Φ0 , then by functori-

ality of K0 and the results of (a),

idK0 (A) = K0 (idA ) = K0 (Φ0 ) ◦ K0 (Φ) and idK0 (B) = K0 (idB ) = K0 (Φ) ◦ K0 (Φ0 ).

Hence K0 (Φ) and K0 (Φ0 ) must be isomorphisms. 

2.5 Continuity
Definition 2.25 (Directed set). A directed set is a set I with a partial order ≤ such
that any two elements have a common upper bound. In other words, for all objects
i, j ∈ I, there exists k ∈ I such that i ≤ k and j ≤ k.

Example 2.26. Any subset S ⊆ R gives rise to a directed set with the usual order
relation. The same statement is true for S replaced by any totally ordered set. An
example for a directed set which does not come from a total order is the set of finite-
dimensional subspaces of a Hilbert space H, ordered by inclusion.

Definition 2.27 (Direct limits). Let C be a category and let I be a directed set.
(a) A direct system in C is a collection of objects ci , i ∈ I, together with a collection of
morphisms ϕji : ci → cj for all i, j ∈ I with i ≤ j, such that ϕkj ◦ ϕji = ϕki whenever
i ≤ j ≤ k.

27
(b) A cocone to a direct system {ci , ϕji }I in C is an object c of C together with a collection
of morphisms ψi : ci → c for each i ∈ I, such that whenever i ≤ j, the diagram

ci
ψi

ϕji c commutes.

ψj
cj

(c) A cocone {c, ψi }I to a direct system {ci , ϕji }I is called a direct limit or colimit, if it
satisfies the following universal property: For every other cocone c0 , there exists a
unique morphism c → c0 in C such that whenever i ≤ j, the diagram

ψi0
ci
ψi

ϕji c ∃!
c0 (2.7)

ψj
cj ψj0

commutes. The direct limit is denoted by lim ci or colim ci .

−→

Remark 2.28. A directed set I gives rise to a category with objects the elements of I
and precisely one morphism i → j if i ≤ j. From this point of view, a direct system in C
is just a functor I → C (i.e. a diagram in C), and the cocone and colimit coincide with
the corresponding notions in category theory.
Any directed set is in particular a filtered category. The latter is slightly more general
in that one drops the assumption that there is at most one morphism between any two
objects; instead one requires the existence of equalizers for each pair of parallel morphisms
α, α0 : i → j, i.e. a morphism β : j → k such that β ◦ α = β ◦ α0 . All results below are
true with general filtered diagrams instead of directed sets, but it is usual in this context
(and slightly more convenient) to restrict to directed sets.

Example 2.29 (Direct limits of sets). In the category of sets, direct limits always
exist. If {ci , ϕji }I is a direct system of sets, a direct limit c can be constructed explicitly
by a .
c= ci ∼ (2.8)
i∈I

where for x ∈ ci , y ∈ cj , we declare x ∼ y if and only if there exist k ∈ I such that

ϕki (x) = ϕkj (x). The maps ψi : ci → c are just the obvious maps ψi (x) = [x].

28
Example 2.30 (Filtered colimits of algebraic structures). If C is a category of al-
gebraic structures such as the category of semigroups, groups, algebras or ∗-algebras,
direct limits always exist. To realize the direct limit of a direct system {ci , ϕji }I , use
the construction in Example 2.29 of the colimit c as a set and observe that one obtains
induced algebraic structures on it in a (semi-)obvious way. For example, if each of the
objects ci is a semigroup, we obtain a well-defined multiplication on the set c defined in
(2.8) by defining [x] · [y] = [ϕki (x) · ϕkj (y)] for x ∈ ci , y ∈ cj , where k ∈ I is such that
i ≤ k and j ≤ k.
Example 2.31 (Filtered colimits of C ∗ -algebras). In the category of C ∗ -algebras,
direct limits exist. To realize the direct limit of a direct system {Ai , Φji }I , start with the
colimit A in the category of ∗-algebras (see Example 2.30), and define a seminorm on A
as follows. For [a] ∈ A represented by a ∈ Ai , set
k[a]k := inf{kΦji (a)k | i ≤ j}, (2.9)
where we take the infimum over all j ∈ I with j ≥ i. One checks that this seminorm
is independent of the choice of a. Since ∗-homomorphisms are always contractive by
Prop. 1.16, the seminorm is finite (for non-unital algebras, the same result is true, after
passing to the unitalization). It satisfies the C ∗ -identity, as induced from that of the Ai .
The completion of A with respect to this norm is the required direct limit (note that the
canonical map A → A is not necessarily injective as the seminorm may be degenerate).
Example 2.32. Let {Ai , Φji }I be a direct system of C ∗ -algebras with direct limit {A, Ψi }I .
Then {Mn (A), Mn (Ψi )}I is the direct limit of the direct system {Mn (Ai ), Mn (Φji )}I of
C ∗ -algebras.
Lemma 2.33. Let A be a C ∗ -algebra and a ∈ A be self-adjoint with ka2 − ak ≤ ε for
some ε < 41 . Then there exists a projection p ∈ A such that ka − pk ≤ 2ε.
Proof. Let B ⊂ A be the C ∗ -subalgebra generated by a. Since a is self-adjoint, B is
commutative and by Thm. 1.30, B ∼= C(σ(a)). Let f (λ) = λ2 − λ. Then
ε ≥ ka2 − ak = kf (a)k = sup |f (λ)|.
λ∈σ(a)

Some analysis shows that (provided ε ≤ 41 ), we have |λ2 − λ| ≤ ε if and only if

1 √ 1 √ 
λ ∈ [−δ 0 , δ] ∪ [1 − δ, 1+ δ 0 ], 1 − 1 − 4ε , δ 0 =


with δ= 1 + 4ε − 1 .
2 2
Notice that as functions of ε, we have δ, δ 0 ≤ 2ε whenever ε ≤ 14 . We conclude that
σ(a) ⊂ [−δ 0 , δ] ∪ [1 − δ, 1 + δ 0 ], and that if ε < 41 , then 12 ∈
/ σ(a). Therefore, the function
(
0 if λ ≤ 12
H(λ) = satisfies sup |H(λ) − λ| ≤ 2ε,
1 if λ > 21 λ∈σ(a)

and is continuous on σ(a), that is H ∈ C(σ(a)). Therefore the self-adjoint element

p := H(a) ∈ A satisfies kp − ak ≤ 2ε and because H 2 = H, p is a projection. 

29
Corollary 2.34. Let {Ai , Φji }I be a direct system of C ∗ -algebras with direct limit {A, Ψi }I .
For each projection p ∈ A and any ε > 0, there exists i ∈ I and a projection pi ∈ Ai such
that kp − Ψi (pi )k ≤ ε.

Proof. By the explicit description of A (see Example 2.31), there exists a sequence an ∈
Ain , n ∈ N such that Ψin (an ) → p in A. Since p is self-adjoint (after possibly replacing an
by 12 (an +a∗n )) we may assume that an is self-adjoint. As multiplication in A is continuous,
the sequence Ψin (an )2 converges to p2 in A. Therefore, given any ε > 0, we can choose
n ∈ N large enough so that both
ε ε
kΨin (an )2 − p2 k < , and kΨin (an ) − pk < .
5 5
Using that p = p2 is a projection, we therefore get
2ε
kΨin (a2n − an )k ≤ kΨin (an )2 − p2 k + kΨin (an ) − pk <
5
By the definition of the norm of A, there exists j ≥ in such that a := Φjin (an ) ∈ Aj also
satisfies ka2 − ak < 2ε
5
. Therefore by Lemma 2.33, there exists a projection pj ∈ Aj such
that ka − pj k ≤ 4ε
5
. With this projection,

kp − Ψj (pj )k ≤ kp − Ψj Φjin (an )k + kΨj (a) − Ψj (pj )k

ε 4ε
≤ kp − Ψin (an )k + ka − pj k ≤ + = ε.
5 5
where we used that ∗-homomorphisms between C ∗ -algebras are contractive (Prop. 1.16).

Proposition 2.35. If {Ai , Φji }I is a direct system of C ∗ -algebras with direct limit {A, Ψi }I ,
then the collection {V(Ai ), V(Φji )}I is a direct system of semigroups, and

V(A) ∼
= lim V(Ai ).
−→

Proof. We verify the universal property. To this end, let {S, ψi }I be a cocone to the
direct system of semigroups {V(Ai ), V(Φji )}I . We have to show that there exists a unique
semigroup homomorphism χ : V(A) → S such that χ ◦ V(Ψi ) = ψi for all i ∈ I.
On elements x ∈ V(A) of the form x = [Ψi (pi )] for some projection pi ∈ Mn (Ai ), this
homomorphism must be given by

χ(x) = χ([Ψi (pi )]) = χ ◦ V(Ψi )([pi ]) = ψi ([pi ]). (2.10)

But by Corollary 2.34, any element x ∈ V(A) is of this form: Indeed, if x = [p] for some
projection p ∈ Mn (A), then (since Mn (A) = lim Mn (Ai ) by Example 2.32), Corollary 2.34
−→
provides the existence of a projection pi ∈ Mn (Ai ) such that kp − Ψi (pi )k < 14 . By
Lemma 2.5 and Prop. 2.6, we therefore have p ∼ Ψi (pi ), in other words [p] = [Ψi (pi )].

30
This shows that χ is uniquely determined by (2.10), and it is compatible with the maps
V(Ψi ) and ψi by construction. It is also easy to see that it is additive. It therefore
only remains show that χ is indeed unambiguously defined by (2.10). In other words,
we have to show that if pi ∈ Mn (Ai ) and pj ∈ Mn (Aj ) are two projections such that
[Ψi (pi )] = [Ψj (pj )], then ψi ([pi ]) = ψj ([pj ]).
Assume first that kΨi (pi ) − Ψj (pj )k < 12 . Then by the definition of the norm of the direct
limit Mn (A), there exists k ≥ i, j such that also kΦki (pi ) − Φkj (pj )k < 12 , which (again by
Lemma 2.5 and Prop. 2.6) implies that Φki (pi ) ∼ Φkj (pj ), hence
ψi ([pi ]) = ψk ◦ V(Φki ([pi ]) = ψk ([Φki (pi )]) = ψk ([Φkj (pj )]) = · · · = ψj ([pj ]).
In general, (after possibly increasing matrix dimensions), let (qt )t∈[0,1] a homotopy of
projections with q0 = Ψi (pi ) and q1 = Ψj (pj ). Choose a partition 0 = t0 < t1 < · · · <
tn = 1 with kqtk − qtk−1 k < 14 and projections pik ∈ Aik with kΨik (pik ) − qtk k < 14 ; here we
let pi0 = pi and pin = pj . Now
1
kΨik (pik ) − Ψik−1 (pik−1 )k ≤ kΨik (pik ) − qtk k + kqtk − qtk−1 k + kqtk−1 − Ψik−1 (pik−1 )k < .
2
Hence by the previous step, ψik (pik ) = ψik−1 (pik−1 ) for all k = 1, . . . , n, which finishes the
proof. 
Theorem 2.36 (Continuity). K-theory commutes with direct limits. In other words, if
{Ai , Φji }I is a direct system of C ∗ -algebras with direct limit {A, Ψi }I , then the collection
{K0 (Ai ), K0 (Φji )}I is a direct system of groups, and
K0 (A) = K0 lim Ai ∼


−→ = lim K (A ).
−→ 0 i

Proof. By functoriality of K0 , {K0 (Ai ), K0 (Φji )}I and {GV(A+ ), GV(Φ+ ji )}I are direct
systems of abelian groups. Moreover, we have the constant direct system {GV(C), id}I
(of course, GV(C) ∼= Z, but we don’t need this fact). Putting these together, we obtain
a short exact sequence of direct systems of abelian groups, i.e. for each i ≤ j, we have a
commutative diagram
GV(εAi )
0 K0 (Ai ) GV(A+
i ) GV(C) 0

K0 (Φji ) GV(Φ+
ji )

0 K0 (Aj ) GV(A+
j ) GV(εAj )
GV(C) 0.

This is just the definition of K0 ; the compatibility of these diagrams for three indices
i ≤ j ≤ k is just its functoriality, see the proof of Lemma 2.18. It is now well-known that
such a short exact sequence of direct systems yields a short exact sequence of the direct
limits, that is, we get a short exact sequence

0 lim K0 (Ai ) lim GV(A+

i ) lim GV(C) 0 (2.11)
−→ −→ −→

31
The term on the right is just isomorphic GV(C), as the corresponding direct system is
constant. To identify the middle term, we use that both the unitalization functor and the
Grothendieck functor are left adjoints, as noted in Remark 1.21 and Remark 2.10, and
it is a standard fact from category theory that left adjoints commute with direct limits.
Moreover, the functor V commutes with direct limits by Prop. 2.35. We conclude that
lim GV(A+
i ) = GV(lim A+ ) = GV(A+ ).
−→ −→ i
Put together, we obtain the commutative diagram

0 lim K0 (Ai ) lim GV(A+

i ) lim GV(C) 0
−→ −→ −→
(2.12)

0 K0 (A) GV(A+ ) GV(C) 0

with exact rows. By the five lemma, the canonical map lim K0 (Ai ) → K0 (A) is an
−→
isomorphism. 

2.6 Stability
Proposition 2.37. Let {Ai , Φji }I be a direct system of C ∗ -algebras with direct limit
{A, Ψi }I . Suppose that each of the structure maps Φji : Ai → Aj is injective. Then
for any C ∗ -algebras B, {Ai ⊗ B, Φji ⊗ idB }I is a direct system of C ∗ -algebras with direct
limit {A ⊗ B, Ψi ⊗ idB }I .

Proof. To begin with, let A◦ = i∈I Ψi (Ai ) be the direct limit of {Ai , Φji }I in the category
S
of ∗-algebras. We claim that A◦ ⊗alg B is dense in A ⊗ B. Indeed, if a ∈ A is the limit of
a sequence (an )n∈N in A◦ , then for any b ∈ B, an ⊗ b converges to a ⊗ b in A ⊗ B, as by
(1.8),
kan ⊗ b − a ⊗ bk = k(an − a) ⊗ bk ≤ kan − akkbk.
Hence the closure of A◦ ⊗alg B in A ⊗ B contains A ⊗alg B, which is dense by the definition
of A ⊗ B.
Clearly {A ⊗ B, Φi ⊗ idB }I is a cocone. To verify the universal property, let {C, Ψ0i }I be
another cocone; we have to define a ∗-homomorphism Ξ : A◦ → B such that Ξ ◦ Ψi = Ψ0i
for all i ∈ I. Clearly, on the subset A◦ ⊗alg B, Ξ must be given by
Ξ(Ψi (a) ⊗ b) = Ψ0i (a ⊗ b), for a ∈ Ai , b ∈ B.
One easily verifies that this gives a well-defined ∗-homomorphism Ξ : A◦ ⊗alg B → C,
which satisfies Ξ ◦ Ψi = Ψ0i by construction. We havePto verify that Ξ is continuous with
respect to the spatial norm. To this end, let x = m ◦
n=1 Ψin (an ) ⊗ bn ∈ A ⊗alg B, for
an ∈ Ain and bn ∈ B. Then there exists i ∈ I with i ≥ in for all n = 1, . . . , m, hence
m
X m
X 
x= Φi (Φiin (an )) ⊗ bn = (Ψi ⊗ idB ) Φiin (an ) ⊗ bn . (2.13)
n=1 n=1

32
The fact about the spatial tensor product we use now is that since Ψi is assumed to be
injective, so is Φi ⊗ idB (Corollary 1.40). Therefore,
m
X  m
X
kΞ(x)k = Ψi Φiin (an ) ⊗ bn ≤ Φiin (an ) ⊗ bn = kxk
n=1 n=1

Here we used that ∗-homomorphisms are contractive (Prop. 1.16) together with (2.13)
and the fact that Φi ⊗ idB is isometric (Corollary 1.32). 

Corollary 2.38. Let A be a C ∗ -algebra. Then the completion of the infinite matrix al-
gebra M∞ (A) with respect to the C ∗ -norm induced by the inclusions Mn (A) → M∞ (A)
is isomorphic to the spatial tensor product A ⊗ K, where K is the algebra of compact
operators on an (infinite-dimensional) separable Hilbert space.

Proof. By Lemma 1.38, we have Mn (A) = A ⊗ Mn (C), and the embeddings Mn (A) →
Mm (A) for m ≥ n take the form idA ⊗ Jmn , where Jmn : Mn (C) → Mm (C) is the canonical
inclusion.
The limit of the direct system {Mn (C), Jmn }N in the category of ∗-algebras is M∞ (C),
which can be identified with a dense subalgebra of F(H), the algebra of finite rank op-
erators on H = `2 (N), and the induced norm is just the operator norm. Hence the
C ∗ -algebraic direct limit is its closure, the space of compact operators,

lim Mn (C) = K(H).

−→
The result now follows from Prop. 2.37. 

Theorem 2.39 (Stability). Let A be a C ∗ -algebra. Then the inclusion C → K = K(H)

as rank one operators induces an isomorphism

K0 (A) ∼
= K0 (A ⊗ K).

Proof. Consider the direct system {Mn (A), Jmn }N , with direct limit {A ⊗ K(H), Jn }N
(Corollary 2.38). Therefore, by Continuity of K0 , Thm. 2.36,

lim K0 (Mn (A)) = K0 (A ⊗ K(H)).

−→
On the other hand, by construction of K0 , each of the maps K0 (Jmn ) : K0 (Mn (A)) →
K0 (Mm (A)) are isomorphisms. Hence {Mn (A), Jmn }N is the constant direct system, with
each turn isomorphic to K0 (A) and the connecting maps the identity under this identifi-
cation. The result follows. 

33
3 The K-theory long exact sequence
Let A be a C ∗ -algebra and J ⊂ A a closed ideal, leading to the short exact sequence
ι π
0 J A A/J 0. (3.1)

In this section, we construct the long exact sequence of K-theory corresponding to this
short exact sequence. Throughout, we denote the projection map on the quotient by
π : A → A/J and the inclusion map of the ideal by ι : J → A.
Notice that associated to the short exact sequence (3.1), we also have the short exact
sequences
0 Mn (J) ι Mn (A) π Mn (A/J) 0
(3.2)
ι π+
0 J A+ (A/J)+ 0
There are several further short exact sequences derived from this one, see §3.2.

3.1 Half-Exactness
Definition 3.1 (Homotopy of unitaries). Let A be a unital C ∗ -algebra. Two uni-
taries u, v ∈ A are homotopic, denoted by u ∼h v, if there exists a continuous path of
unitaries (ut )t∈[0,1] such that u0 = u and u1 = v.

Lemma 3.2 (Lifting unitaries). Let A be a unital C ∗ -algebra and J ⊂ A a closed ideal.
Then for any unitary u ∈ A/J with u ∼h 1, there exists a unitary u
e ∈ A with π(eu) = u
e ∼h 1 in A.
and u

Proof. First assume that ku − 1k < 2. Then σ(u) is contained in {λ ∈ C | |λ − 1| < 2}.
In particular, −1 ∈
/ σ(u). On the other hand σ(u) is contained in the unit circle, as u is
unitary. Therefore the complex logarithm (defined such that Log eiθ = iθ for θ ∈ (−π, π))
is a continuous function on σ(u). Hence we may define z := Log(u) ∈ A/J. z is skew-
adjoint, since
z ∗ = Log(u)∗ = Log(u∗ ) = Log(u−1 ) = − Log(u) = −z.
Let ze ∈ A, be a lift of z (which exists by surjectivity of π). We may arrange ze to be
z − ze∗ )/2). Then u
skew-adjoint (by possibly replacing ze by (e e := exp(e
z ) is the required lift
of u. It is connected to 1 by the continuous path (ut )t∈[0,1] of unitaries given by
ut = exp(t Log(u)) (3.3)
For a general unitary u with u ∼h 1, let (ut )t∈[0,1] be a continuous path of unitaries with
u1 = u and u0 = 1n . Choose a partition 0 = t0 < t1 < · · · < tn = 1 of [0, 1] such that
kuti − uti−1 k < 2 for each i = 1, . . . , n. Then ku∗ti−1 uti − 1n k < 2, hence there exist lifts
ei ∈ A+ of u∗ti−1 uti . But then w
w e1 · · · w
en is a lift of u. Concatenating the paths (3.3) gives
a continuous path of unitaries from u to 1. 

34
Remark 3.3. Conversely, the above proof shows that any unitary u ∈ A with ku−1k < 2
automatically satisfies u ∼h 1, where the homotopy is implemented by the path (3.3)

Corollary 3.4. For any unitary u ∈ A/J, the unitary diag(u, u∗ ) ∈ M2 (A/J) has a
e ∈ M2 (A) with w
unitary lift w e ∼h 12n .

Proof. As seen in part (d) of the proof of Prop. 2.6, we have diag(u, u∗ ) ∼h 12n , hence
the statement follows from Lemma 3.2. 

Theorem 3.5 (Half-exactness). Let A be a C ∗ -algebra and J ⊂ A a closed ideal. Then

the sequence of groups
K0 (ι) K0 (π)
K0 (J) K0 (A) K0 (A/J)

is exact.

Proof. Clearly, if x ∈ K0 (J), then by functoriality, K0 (π) ◦ K0 (ι)(x) = K0 (π ◦ ι)(x) = 0.

Hence im K0 (ι) ⊆ ker K0 (π).
Let now x ∈ ker K0 (π) with K0 (π)(x) = 0. We have to show that x = K0 (ι)(y) for some
y ∈ K0 (J). According to Prop. 2.19(a), there exist a projection p ∈ M∞ (A+ ) and n ∈ N
such that x = [p] − [1n ] and p − 1n ∈ M∞ (A). Since K0 (π)(x) = 0, by Prop. 2.19(b), we
have diag(π + (p), 1k ) ∼u 1n+k in M∞ ((A/J)+ ) for some k ∈ N. Denote p0 := diag(p, 1k ),
and for some m ∈ N large enough, let u ∈ Mm ((A/J)+ ) be a unitary such that
 + 
+ 0 ∗ π (p)
u π (p ) u = u u∗ = 1n+k ∈ Mm ((A/J)+ )
1k

Let w ∈ M2m (A+ ) be a unitary lift of diag(u, u∗ ), which exists by Corollary 3.4, and set
 0 
p
q := w w∗ ∈ M2m (A+ ).
0

Then by construction, [q] − [1n+k ] is another representative for x. On the other hand, we
have   0  ∗   0 ∗   
+ u p u up u 1n+k
π (q) = = = ,
u∗ 0 u 0 0
so q − 1n+k ∈ M2m (J) and q ∈ M2m (J + ). Therefore y := [q] − [1n+k ] ∈ K0 (J) is an
element with K0 (ι)(y) = x. 

3.2 Cone and suspension

Definition 3.6 (Cone and suspension). Let A be a C ∗ -algebra.
(a) The cone of A is the C ∗ -algebra CA defined by

CA := {f ∈ C([0, 1], A) | f (0) = 0}.

35
(b) The suspension of A is the C ∗ -subalgebra SA ⊆ CA defined by

SA := {f ∈ CA | f (1) = 0.}

For ∗-homomorphisms Φ : A → B, we define CΦ : CA → CB by Φ(f )(t) = Φ(f (t));

SΦ : SA → SB is defined by the same formula. It is then clear that both C and S are
functors sending C ∗ -algebras to C ∗ -algebras. It is straightforward to verify that both are
exact functors, that is, applying them to the short exact sequence (3.1), one obtains short
exact sequences
Cι Cπ
0 CJ CA C(A/J) 0
(3.4)
Sι Sπ
0 SJ SA S(A/J) 0.

Definition 3.7 (Mapping cone and cylinder). Let A, B be C ∗ -algebras and let Φ :
A → B be a ∗-homomorphism.

(a) The mapping cone CΦ of Φ is defined as

CΦ = {(a, f ) ∈ A ⊕ CB | f (1) = Φ(a)}.

(b) The mapping cylinder ZΦ of Φ is defined as

ZΦ = {(a, f ) ∈ A ⊕ C([0, 1], B) | f (0) = Φ(a)}.

The mapping cone and mapping cylinders extend to functors from the category whose ob-
jects are ∗-homomorphisms Φ : A → B and whose morphisms are commutative diagrams
Φ
A B
ΨA ΨB (3.5)
A0 Φ0
B0

to the category of C ∗ -algebras. Namely, given such a commutative diagram, one obtains a
∗-homomorphisms Ψ : ZΦ → ZΦ0 and Ψ : CΦ → CΦ0 by setting Ψ(a, f ) = (ΦA (a), ΨB ◦f ) ∈
A0 ⊕C([0, 1], B 0 ). It is easy to check functoriality with respect to concatenation of diagrams
(3.5).

Lemma 3.8. Let A be a C ∗ -algebra.

(a) The mapping cone CA is contractible, that is, the inclusion Φ : {0} → CA of the
trivial C ∗ -algebra is a homotopy equivalence.

(b) For any ∗-homomorphism Φ : A → B, projection onto the first component p1 : ZΦ →

A is a homotopy equivalence.

36
Proof. (a) Let Φ0 : CA → {0} be the trivial map (a ∗-homomorphism) and for t ∈ [0, 1],
consider the ∗-homomorphism Φt : CA → CA, f 7→ ft , where ft (s) = f (ts). Then for
any f ∈ CA, t 7→ Φt (f ) is a continuous map (by compactness of [0, 1]) hence (Φt )t∈[0,1]
is a homotopy with Φ1 = id and Φ0 = Φ ◦ Φ0 . It follows now from Thm. 2.24(b) that
K0 (CA) = {0}.
(b) We claim that a homotopy inverse is given by c : A → ZΦ , a 7→ (a, ca ), where
ca ∈ C([0, 1], B) is the function with ca (t) ≡ Φ(a) for all t ∈ [0, 1]: First, p1 ◦ c = idA .
On the other hand, consider the family of ∗-automorphisms Ψs , s ∈ [0, 1], of ZΦ given by
Ψs (a, f ) := (a, fs ), where for f ∈ C([0, 1], B), the function fs ∈ C([0, 1], B) is given by
(
f (t − s) s ≤ t
fs (t) :=
f (1) s≥t
Then Ψ1 = c ◦ p1 , while Ψ0 = idA . Hence also c ◦ p1 is homotopic to the identity. 

3.3 The long exact sequence

Lemma 3.9. Let A be a C ∗ -algebra, J ⊆ A be a closed ideal and suppose that A/J is
contractible. Then K0 (ι) : K0 (J) → K0 (A) is an isomorphism.
Proof. First, notice that K0 (ι) is surjective by half-exactness of K0 (Thm. 3.5), because
K0 (A/J) = {0} by contractibility of A/J and homotopy invariance of K0 (Thm. 2.24).
To show injectivity of K0 (ι), we use the mapping cylinder Zι . Notice here that by
Lemma 3.8(b), the projection map p1 : Zι → J, (a, f ) 7→ a is a homotopy equivalence
and hence K0 (p1 ) : K0 (Zι ) → K0 (J) is an isomorphism (Thm. 2.24).
On the other hand, Zι admits a surjective ∗-homomorphism ζ : Zι → Cπ to the mapping
cone of π, given by ζ(a, f ) = (f (1), π ◦ f ), with ker(ζ) = CJ. Consider the following
commutative diagram with exact rows and colums, where i and p are the obvious inclusion
and projection maps, in view of Cπ ⊆ A ⊕ C(A/J).

ι π
0 J A A/J 0
p1 ' p

0 CJ Zι ζ
Cπ 0
i

S(A/J)

0
Since CJ is contractible by Lemma 3.8(a) and S(A/J) is contractible by assumption, the
homomorphisms K0 (ζ) and K0 (p) are injective, again by half-exactness and homotopy

37
invariance of K0 (Thms 3.5 & 2.24). This shows that K0 (ι) = K0 (p)K0 (ζ)K0 (p1 )−1 is
injective as well. 

Theorem 3.10 (Long exact sequence). Let A be a C ∗ -algebra and let J ⊂ A be a

closed ideal. Then there exists a boundary map δ : K0 (S(A/J)) → K0 (J) such that we
have an exact sequence of groups

K0 (S(A/J)) K0 (SA) K0 (SJ)

δ (3.6)

K0 (J) K0 (A) K0 (A/J).

Moreover, the map is functorial with respect to the short exact sequence, i.e. if Φ : A → A0
is a ∗-homomorphism taking J to a closed ideal J 0 ⊆ A0 , then we obtain a commutative
diagram
δ
K0 (S(A/J)) K0 (J)

K0 (S(A0 /J 0 )) δ0
K0 (J 0 ).

Proof. We will use the short exact sequences

i p
0 S(A/J) Cπ A 0
(3.7)
j
0 J Cπ C(A/J) 0

involving the mapping cone Cπ of π : A → A/J. Again, all maps involved are just the
obvious inclusion and projection maps coming from viewing Cπ ⊂ A ⊕ C(A/J).
Definition of the boundary map: This is done with the diagram
K0 (Sπ) −δ K0 (ι)
K0 (SA) K0 (S(A/J)) K0 (J) K0 (A).
∼
= (3.8)
K0 (i) K0 (j)
K0 (Cπ )

Since C(A/J) is contractible by Lemma 3.8(a), Lemma 3.9 implies that the map K0 (j)
is an isomorphism. Therefore, we can define δ = −K0 (j)−1 K0 (i). Naturality of δ follows
from the functoriality of the cone construction. It is left to verify exactness of the top
row of this diagram.

38
Exactness at K0 (J): We have the commutative diagram
K0 (i) K0 (p)
K0 (S(A/J)) K0 (Cπ ) K0 (A)
∼
= K0 (j)

K0 (S(A/J)) −δ
K0 (J) K0 (ι)
K0 (A).

The top row is exact (in the middle) by half-exactness of K0 , Thm. 3.5, applied to the
first short exact sequence in (3.7). Since K0 (j) is an isomorphism, this implies that also
the bottom row is exact.
Exactness at K0 (S(A/J)): At this point we know that for any C ∗ -algebra A1 with a closed
ideal J1 ⊆ A1 , the sequence
δ1 K0 (ι1 ) K0 (π1 )
K0 (S(A1 /J1 ) K0 (J1 ) K0 (A1 ) K0 (A1 /J1 ) (3.9)

is exact. The trick is to apply this to the first exact sequence in (3.7), that is, we set
J1 := S(A/J) and A1 := Cπ , with maps ι1 = i, π1 = p, which then gives A1 /J1 ∼ = A.
Substituting these definitions in (3.9), we obtain that the top row of the diagram
δ1 K0 (i) K0 (p)
K0 (SA) K0 (S(A/J)) K0 (Cπ ) K0 (A)
∼
= K0 (j) (3.10)
K0 (SA) K0 (Sπ)
K0 (S(A/J)) −δ
K0 (J) K0 (ι)
K0 (A)

is exact. We have seen that the two squares on the right hand side commute; since
both K0 (j) and K0 (σ) are isomorphisms, the exactness of the top row implies that of the
bottom row, provided that we can verify that the left-most square commutes as well, that
is, δ1 = K0 (Sπ).
To see this, we need to look more closely at the derived short exact sequences (3.7) for
our new exact sequence (3.9). These are
i1 p1
0 SA Cπ1 Cπ 0
(3.11)
j1
and 0 S(A/J) Cπ1 CA 0,

involving the mapping cone Cπ1 of π1 : Cπ → A. Since π1 is just the projection onto the
first factor of Cπ ⊆ A ⊕ C(A/J), upon going through the definitions, one finds that the
mapping cone can be identified with
Cπ1 = {(g, f ) ∈ C(A/J) ⊕ CA | g(1) = π(f (1))},
in such a way that the maps i1 and j1 in (3.11) are just the obvious inclusion maps under
this identification. Since δ1 = −K0 (j1 )−1 K0 (i1 ), we have
δ1 = K0 (Sπ) ⇐⇒ −K0 (i1 ) = K0 (j1 ◦ Sπ).

39
The idea is therefore to construct a homotopy between the ∗-homomorphisms i1 and
j1 ◦ Sπ from SA to Cπ1 . Going through the definition, one finds
i1 (f ) = (0, f ), (j1 ◦ Sπ)(f ) = (π ◦ f, 0).
Consider now the collection (Φs )s∈[0,1] of ∗-homomorphisms Φs : SA → Cπ1 given by
(
0 t ∈ [0, 1 − s]
fsC(A/J) (t) =
π(f (2 − t − s)) t ∈ [1 − s, 1]
Φs (f ) = (fsC(A/J) , fsCA ), with (
0 t ∈ [0, s]
fsCA (t) =
f (t − s) t ∈ [s, 1].
It is a homotopy with Φ0 = i1 and Φ1 = j1 ◦ Sπ ◦ σ, where σ : SA → SA is the ∗-
automorphism defined by σ(f )(t) = f (1 − t). By homotopy invariance (Thm. 2.24), we
therefore have K0 (i1 ) = K0 (j1 ◦ Sπ) ◦ K0 (σ), or equivalently δ1 = −K0 (Sπ) ◦ K0 (σ).
To finish the proof, one could now show that K0 (σ) = −idK0 (SA) , which is not too hard.
However this is not necessary: Instead, one can observe that so far, we have shown
that the diagram (3.10) commutes if one replaces the left-most identity arrow by the
automorphism −K0 (σ). But also for this new diagram, exactness of the top row implies
that of the bottom row. 
Corollary 3.11 (Split-exactness). Let A be a C ∗ -algebra with a closed ideal J ⊆ A
such that the short exact sequence (3.1) splits, that is, there exists a ∗-homomorphism
s : A/J → A such that π ◦ s = idA/J . Then
K0 (A) ∼
= K0 (J) ⊕ K0 (A/J).

Proof. The splitting map s provides a group homomorphism K0 (s) : K0 (J) → K0 (A)
such that K0 (π)K0 (s) = idK0 (A/J) . This shows that K0 (π) must be surjective. Taking
suspensions, we obtain that also K0 (Sπ) is surjective, hence by exactness, δ = 0. Therefore
K0 (ι) is injective. Therefore, we obtain a split exact short exact sequence
K0 (ι) K0 (π)
0 K0 (J) K0 (A) K0 (A/J) 0.
K0 (s)

As we are in the category of abelian groups, this implies that K0 (ι) ⊕ K0 (s) : K0 (J) ⊕
K0 (A/J) → K0 (A) is an isomorphism. 
Remark 3.12. Of course, if A is isomorphic to J ⊕ A/J as a C ∗ -algebra, such that ι
and π are just the inclusion respectively the projection map under this identification,
then the result follows directly from the definition of K0 . However, the existence of a
splitting s : A/J → A does not imply A ∼ = J ⊕ A/J as C ∗ -algebras. For example, if
A is non-unital, then usually A+ is not isomorphic to the direct sum A ⊕ C. However,
Corollary 3.11 implies that we always have
K0 (A+ ) ∼
= K0 (A) ⊕ Z. (3.12)

40
To obtain a formula for the boundary map, we need the following lemma.

Lemma 3.13. Let A be a C ∗ -algebra and let f ∈ Mm (CA+ ) be a projection with f (0) =
1n , where n ≤ m. Then there exists a unitary u ∈ Mm ((CA)+ ) with u(0) = 1m such that
f (t) = u(t)1n u(t)∗ .

Proof. We have f ∼h 1n , as the path fs defined by fs (t) = f (st) is a homotopy.

By Prop. 2.6(a), this implies f ∼u 1n . This implies the existence of a unitary u0 ∈
Mm ((CA)+ ) such that f (t) = u0 (t)1n u0 (t)∗ for t ∈ [0, 1]. For t = 0, this implies
1n = u0 (0)1n u(0), hence u0 (0) = diag(v, w) for unitaries v ∈ Mn (A), w ∈ Mm−n (A)
(see Lemma 5.8 below). Therefore u(t) = u0 (t)u0 (0)∗ also satisfies f (t) = u(t)1n u(t)∗ , and
u(0) = 1m . 

Proposition 3.14. Let A be a C ∗ -algebra and let J ⊂ A be a closed ideal. Then the
boundary map in the long exact sequence (3.6) has the following explicit formula: Rep-
resent x ∈ K0 (S(A/J)) as x = [f ] − [1n ], where f ∈ Mm (S(A/J)+ ), m ≥ n such that
f − 1n ∈ Mm (S(A/J)), and choose a projection fe ∈ Mm ((CA)+ ), fe(0) = 1n such that
π + (fe(t)) = f (t) for all t ∈ [0, 1]. Then

δ(x) = [fe(1)] − [fe(0)]. (3.13)

Proof. To see the existence of a lift, consider f as a projection in the larger space
M∞ (C(A/J)+ ). Then by Lemma 3.13, there exists a unitary u ∈ Mm (C(A/J)+ ) with
u(0) = 1m such that f = u1n u∗ . We have u ∼h 1m since the family of paths (us )s∈[0,1]
defined by us (t) = u(st) provides a homotopy; hence by Lemma 3.2, there exists a uni-
e ∈ Mm (CA+ ) of u, which automatically satisfies u
tary lift u e(0) = 1m . Then fe defined by
fe(t) := u e(t)∗ is the desired lift of f .
e(t)1n u
We have fe(1) ∈ Mm (J + ), as

π + (fe(1)) = π + (e u(1))∗ = u(1)1n u(1)∗ = f (1) = 1n .

u(1))1n π + (e

In particular, since fe(0) = 1n , this implies fe(1) − fe(0) ∈ Mn (J), hence [fe(1)] − [fe(0)] is
a well-defined element of K0 (J).
Since δ is defined by δ = −K0 (j)−1 ◦ K0 (i), in order to see the formula (3.13), we compare
K0 (j)([fe(1)] − [fe(0)]) with K0 (i)(x) in K0 (Cπ ). Remember that the maps j : J → Cπ
and i : S(A/J) → Cπ are given by j(a) = (a, 0), respectively i(f ) = (0, f ). Therefore,

j + (fe(1)) = j(fe − 1n ) + j + (1n ) = (fe(1) − 1n , 0) + (1n , 1n ) = (fe(1), 1n ),

where we use that (1, 1) is the unit of (CA)+ . We therefore have

K0 (j)([fe(1)] − [fe(0)]) = [(fe(1), 1n )] − [(fe(0), 1n )].

41
On the other hand,

K0 (i)(x) = [i+ (f )] − [i+ (1n )] = [(1n , f )] − [(1n , 1n )]

We have to show that these elements are add to zero in K0 (Cπ ). This follows from the
calculation    
fe(1) 0 1 n 0
[(fe(1), 1n )] + [(1n , f )] = ,
0 1n 0 f
   
(∗) fe(1) 0 f 0
= ,
0 1n 0 1n
   
(†) 1n 0 1n 0
= ,
0 1n 0 1n
= 2[(1n , 1n )],
where we have to justify the equalities (∗) and (†). For (∗), we use the homotopy (qs )s∈[0,1]
of projections in M2m (Cπ+ ), defined by

cos πs πs
     


fe(1) 0 1n 0 ∗ 2
− sin 2
qs = , rs r , where rs =
;
sin πs cos πs


0 1n 0 f s 2 2

observe here that due to the fact that f (1) = 1n , each qs indeed defines a matrix with
values in the mapping cone Cπ .
For the equality (†), consider the homotopy (ps )s∈[0,1] of projections in Mm (Cπ+ ) given by

ps = (fes (1), π ◦ fes ), with fes (t) = fe(st).

Then p0 = (1n , 1n ) and p1 = (fe(1), f ). Stabilising this, we obtain a homotopy implement-

ing (†). 

4 Bott Periodicity
In this section, we prove the main result of operator K-theory, Bott periodicity. Through-
out this section, we identify S n C ⊗ A ∼
= S n A, in view of Example 1.41.

4.1 The exterior product

Let A, B be two C ∗ -algebras. If p ∈ Mn (A), q ∈ Mm (B) are projections, then their tensor
product p ⊗ q ∈ Mn (A) ⊗ Mm (B) ∼ = Mnm (A ⊗ B) is again a projection. Applying the V
functor, we obtain a well-defined map

× : V(A) × V(B) −→ V(A ⊗ B), ([p], [q]) 7−→ [p ⊗ q], (4.1)

which is N-bilinear. Applying the Grothendieck construction, we obtain a Z-bilinear

product on the associated Grothendieck groups. On the category of unital C ∗ -algebras,

42
where we can identify K0 = GV (see Remark 2.17), this gives a product map × : K0 (A) ×
K0 (B) → K0 (A ⊗ B). By construction, the map is natural in the sense that for any pair
of unital ∗-homomorphisms Φ : A → A0 , Ψ : B → B 0 , the diagram
×
K0 (A) × K0 (B) K0 (A ⊗ B)
K0 (Φ)×K0 (Ψ) K0 (Φ⊗Ψ) (4.2)

K0 (A0 ) × K0 (B 0 ) ×
K0 (A0 ⊗ B 0 )

commutes.
To extend this construction to the non-unital case, we need the following lemma.
Lemma 4.1. Let A, B be C ∗ -algebras. Then we naturally have

K0 (A+ ⊗ B + ) ∼
= K0 (A ⊗ B) ⊕ K0 (A) ⊕ K0 (B) ⊕ Z.

Moreover, under this identification, we have

K0 (A ⊗ B) = ker(K0 (εA ⊗ idB + )) ∩ ker(K0 (idA+ ⊗ εB )) ⊂ K0 (A+ ⊗ B + ). (4.3)

Proof. This follows easily from split-exactness, Corollary 3.11, as we have the split exact
sequences
εA ⊗idB +
0 A ⊗ B+ A+ ⊗ B + B+ 0,

idA ⊗εB
0 A⊗B A ⊗ B+ A 0,
as well as K0 (A+ ) = K0 (A) ⊕ Z, K0 (B + ) = K0 (B) ⊕ Z, see (3.12). 

Corollary 4.2. The product map defined above sends K0 (A)×K0 (B) ⊂ K0 (A+ )×K0 (B + )
to K0 (A ⊗ B) ⊂ K0 (A+ ⊗ B + ).

Proof. If x ∈ K0 (A) ⊂ K0 (A+ ) and y ⊂ K0 (B) ⊂ K0 (B + ), then by the naturality

property (4.2),
K0 (εA ⊗ idB + )(x × y) = K0 (εA )(x) × y = 0,
K0 (idA+ ⊗ εB )(x × y) = x × K0 (εB )(y) = 0
From (4.3), it then follows that x × y ∈ K0 (A ⊗ B). 

We can now make the following definition.

Definition 4.3 (Exterior product). Let A and B be C ∗ -algebras. The product

× : K0 (A) × K0 (B) −→ K0 (A ⊗ B), (x, y) 7→ x × y, (4.4)

defined above is called the exterior product.

43
Lemma 4.4 (Properties of the exterior product). Let A, B be C ∗ -algebras.
(a) The exterior product is natural, in the sense that for any pair of ∗-homomorphisms
Φ : A → A0 , Ψ : B → B 0 , the diagram (4.2) commutes.

(b) The class 1 ∈ Z ∼= K0 (C) is a two-sided unit for the exterior product, meaning that
under the canonical isomorphisms K0 (C ⊗ A) ∼ = K0 (A) and K0 (A ⊗ C) ∼= K0 (A), the
elements 1 × x, respectively x × 1 are identified with x, for any x ∈ K0 (A).

(c) The exterior product is commutative, in the sense that

x × y = K0 (σ)(y × x), (4.5)

for all x ∈ K0 (A) and y ∈ K0 (B), where σ : A ⊗ B → B ⊗ A is the symmetry

isomorphism of the tensor product.

Proof. All of these properties are induced by the analogous properties of the product
(4.1), for which they are easily verified. 

4.2 The Töplitz exact sequence and the Bott element

Throughout this section, we denote K = K(`2 (N)), B = B(`2 (N)).
Definition 4.5 (Toeplitz algebra). The Toeplitz algebra T ⊂ B is the subalgebra gen-
erated by the shift operator, explicitly
(
αn−1 n ≥ 2
(Sα)n = α ∈ `2 (N).
0 n = 1,

Lemma 4.6. We have K ⊂ T .

Proof. Let e1 , e2 , . . . be the canonical basis of `2 (N). Then id − SS ∗ = e1 ⊗ e∗1 , the

projection onto the one-dimensional subspace spanned by e1 . More generally, we have
em ⊗ e∗n = S m (id − SS ∗ )(S ∗ )n for any m, n ∈ N. Taking the linear span of these operators,
we see that T contains all finite rank operators. But since by definition, T is norm-closed,
it must contain the closure of the finite rank operators, which is K. 

Proposition 4.7. We have σ(S) = D := {λ ∈ C | |λ| ≤ 1} and σess (S) = T.

Remember here that the essential spectrum is the set of number λ ∈ C such that λ − S
is not a Fredholm operator (see Example 1.13). We use the following criterion.
Lemma 4.8. Let H be a Hilbert space and T ∈ B(H). Given λ ∈ C, assume that there
exists a sequence (vn )n∈N in H without accumulation point such that kvn k = 1 for each
n ∈ N and kT vn − λvn k → 0. Then λ ∈ σess (T ).

44
Proof. Let (vn )n∈N be a sequence with kvn k = 1 for each n ∈ N and kT vn − λvn k → 0.
Suppose that λ − T is a Fredholm operator. Then there exists S ∈ B(H) such that
S(λ − T ) = idH + K, with K ∈ K(H). Since K is compact and kvn k = 1 for each n ∈ N,
the sequence (Kvn )n∈N has an accumulation point w ∈ H. On the other hand, we have
vn = S(λ − T )vn − Kvn .
Since S(λ − T )vn converges to zero, after passing the a subsequence, the right hand side
converges to w. Therefore (vn )n∈N has an accumulation point. 

Proof (of Prop. 4.7). First of all, observe that since kSk = 1, we have σ(S) ⊆ {λ ∈ C |
|λ| ≤ 1}.
For each λ ∈ C with |λ| < 1, the sequence α with αn = λn is contained in `2 (N) and
satisfies S ∗ α = λα. Hence
P∞ λn∈ σ(S
∗
) and λ ∈ σ(S). On the other hand, since S ∗ S = id,
the operator T := − n=0 λ (S ∗ )n+1 satisfies (λ − S)T = id and
∞
X ∞
X ∞
X
T (λ − S) = − λn+1 (S ∗ )n+1 + λn SS ∗ (S ∗ )n = id + (SS ∗ − id) λn (S ∗ )n .
n=0 n=0 n=0

Since id − SS ∗ ∈ K, we see that T is a parametrix for S, so that S is Fredholm. We

conclude that λ ∈
/ σess (T).
(m) √
Let now λ ∈ C with |λ| = 1. For m ∈ N, define a sequence α(m) ∈ `2 (N) by αn = λn / m
(m)
if n ≤ m and αn = 0 for n > m. Then kα(m) k = 1 and
(
0 if n < m or n > m
(λ − S)α(m) = m+1
√
λ / m if n = m.

We obtain that k(λ − S)α(m) k2 = 1/m, which converges to zero as m → ∞. Moreover,

since the sequence α(m) converges pointwise to zero, the only possible accumulation point
is zero; but this is impossible since kα(m) k = 1 for all m ∈ N. Hence α(m) has no
accumulation point. We conclude from Lemma 4.8 that λ ∈ σess (S). 

Proposition 4.9. There exists a unique surjective ∗-homomorphism π : T → C(T) such

that π(S) = z, the identity function on T. Moreover, ker(π) = K, hence we have a short
exact sequence
ι π
0 K T C(T) 0. (4.6)

Proof. Clearly, K is an ideal in T . Consider the C ∗ -algebra A := T /K. Since T is

generated by S, A is generated by [S]. As S ∗ S = id and id − SS ∗ ∈ K, [S] ∈ T /K is
unitary, so A is commutative, and by Thm. 1.30, we have an isomorphism A ∼ = C(σ([S]))
such that [S] 7→ idσ([S]) .
It remains to show that σ([S]) = T. By Prop. 1.29, the spectrum of [S] in A is the same as
the spectrum of [S] in the Calkin algebra B/K. Therefore, by Prop. 4.7 σ([S]) = σess (S) =
T (see Example 1.13). 

45
Indentify SC ⊂ C([0, 1]) with {f ∈ C(T) | f (1) = 0} ⊂ C(T) by sending e2πit to the
function z ∈ C(T). Let T0 = ker(q), where q = ev1 ◦ π : T → C. Then the diagram
ι π
0 K T0 SC 0
(4.7)
ι π
0 K T C(T) 0

has exact rows.

Definition 4.10 (Bott element). A Bott element is an element b ∈ K0 (S 2 C) such that

δ(b) ∈ K0 (K) is the class defined by a rank one projection, where δ is the boundary map
to the upper row in (4.7).

Theorem 4.11 (Existence of Bott element). A Bott element exists.

Proof. We identify SC with {f ∈ C(T) | f (1) = 0} ⊂ C(T) by sending the function

f (s) = e2πis to the identity function z on C(T). Define a unitary uBott ∈ M2 (S 2 C+ ) ⊂
M2 (SC(T)+ ) by

cos πt − sin πt cos πt sin πt






2 2
z 2 2
uBott (t) = (4.8)
sin πt πt πt πt





2
z cos 2
− sin 2
cos 2

It satisfies uBott (0) = 12 , uBott (1) = diag(z, z), therefore

 
1 0 ∗
pBott = uBott u (4.9)
0 0 Bott

is a projection in M2 (S 2 C+ ) with pBott − 11 ∈ M2 (S 2 C). Hence b := [pBott ] − [11 ] defines

an element of K0 (S 2 C).
To calculate δ(b), we use (3.13). The unitary U ∈ M2 (ST0+ ) = M2 (ST ) defined by

∗
cos πt πt πt πt




2

− sin

∗ 2 S cos 2

sin 2

U (t) = ∗
sin πt2
S cos πt
2
SS + (1 − SS ) − sin πt
2
cos πt
2

is a lift of uBott with U (0) = 12 . Hence

         
1 0 ∗ 1 0 1 0 1 0
δ(b) = U (1) U (1) − = − = [1 − SS ∗ ]
0 0 0 0 0 1 − SS ∗ 0 0

Since 1 − SS ∗ is a rank one projection in K, the result follows. 

We finish this section with the following lemma. which is needed in the next section.

46
Lemma 4.12. For any C ∗ -algebra A, the rows of the commutative diagram
ι⊗idA π⊗idA
0 K⊗A T0 ⊗ A SA 0

ι⊗idA π⊗idA
0 K⊗A T ⊗A C(T) ⊗ A 0

are exact.

Proof. It suffices to consider the second sequence. By Corollary 1.40, the map ι ⊗ idA is
injective, hence we naturally have K ⊗ A ⊂ T ⊗ A. It is also straightforward to see that
K ⊗ A is an ideal in T ⊗ A. Because the relation (π ⊗ idA ) ◦ (ι ⊗ idA ) = 0 is still true
(since it holds on the dense subset K ⊗alg A ⊂ K ⊗ A), we obtain a ∗-homomorphism

Φ : (T ⊗ A)/(K ⊗ A) −→ C(T) ⊗ A.

It is surjective, because the dense inclusion C(T) ⊗alg A ⊂ C(T) ⊗ A factors through Φ:

C(T) ⊗alg A ∼
= (T ⊗alg A)/(K ⊗alg A)
Ψ
(T ⊗ A)/(K ⊗ A) Φ
C(T) ⊗ A.

Since Ψ : C(T) ⊗alg A → (T ⊗ A)/(K ⊗ A) is an injective ∗-homomorphism with dense

image (also by the diagram above), it induces a C ∗ -norm k · kα on C(T) ⊗alg A such that
the corresponding completion C(T) ⊗α A ∼ = (T ⊗ A)/(K ⊗ A). As seen, it comes with
surjective ∗-homomorphism C(T) ⊗α A → C(T) ⊗ A, therefore k · kσ ≤ k · kα .
There are several ways to see that k · kα ≤ k · kσ . For example, it is a fact that C(T) is
nuclear, meaning that all C ∗ -norms on C(T) ⊗alg A coincide. Another approach uses the
group C ∗ -algebra C ∗ (Z), which is the C ∗ -subalgebra of B(`2 (Z)) generated by the unilat-
eral translation U , defined by (U α)n = αn+1 for α = (αn )n∈Z ∈ `2 (Z). It is commutative,
and since σ(U ) = T, the Gelfand transform provides a canonical isomorphism to C(T)
(this is just the inverse discrete Fourier transform). Now there is a contractive linear map

s : C(T) ∼
= C ∗ (Z) −→ T , f 7−→ Tf := V fˇV ∗ ,

where fˇ is the inverse Gelfand transform of f and V : `2 (Z) → `2 (N) is the orthogonal
projection. s is a section of π, that is (π ◦ s) = idC(T) .
The map s is not a ∗-homomorphism (so the associated K-theory sequence does not
split), but is a completely positive map, meaning that for all n ∈ N, the induced map on
matrices Mn (s) maps positive elements to positive elements. In particular, tensoring with
idA provides a contraction s ⊗ idA : C(T) ⊗ A → T ⊗ A (a general fact about completely
positive maps, which can also easily be seen from the concrete form of s). We therefore
obtain a contractive linear map
s⊗idA
C(T) ⊗ A T ⊗A (T ⊗ A)/(K ⊗ A) ∼
= C(T) ⊗α A.

Hence k · kα ≤ k · kσ and C(T) ⊗α A ∼

= C(T) ⊗ A. 

47
Remark 4.13. Above, we have essentially proved the following general result for general
C ∗ -algebras A, B and a closed ideal J ⊆ A: Assume that there exists a completely positive
map s : A/J → A such that π ◦ s = idA/J or that A/J is nuclear. Then the short sequence
ι⊗idB π⊗idB
0 J ⊗B A⊗B A/J ⊗ B 0

is exact.

4.3 The periodicity theorem

Throughout, let b ∈ K0 (S 2 C) be the Bott element constructed in the proof of Thm. 4.11.
In fact, it will follows from Bott periodicity, Thm. 4.15 below, that the Bott element is
in fact unique; this is irrelevant for the proof of Bott periodicity, but justifies to refer to
“the” Bott element henceforth.
Definition 4.14 (Bott map). For any C ∗ -algebra A the Bott map of A is the map

βA : K0 (A) −→ K0 (S 2 C ⊗ A) = K0 (S 2 A), x 7→ b × x,

given by taking the exterior product with the Bott element.

It is clear from naturality of the exterior product, Lemma 4.4(a), that the Bott map is
natural, that is, for each ∗-homomorphism Φ : A → B between C ∗ -algebras A, B, we
have a commutative diagram
K0 (Φ)
K0 (A) K0 (B)

βA βB (4.10)

K0 (S 2 A) K0 (S 2 B).
K0 (S 2 Φ)

In other words, the Bott maps assemble to a natural transformation β : K0 ⇒ K0 S 2 .

Theorem 4.15 (Bott periodicity). For each C ∗ -algebra A, the Bott map βA is an iso-
morphism. In other words, the functors K0 and K0 S 2 are naturally isomorphic.

The proof of Thm. 4.15 goes by constructing an inverse transformation α : K0 S 2 → K0 .

This is done as follows: Tensoring the upper sequence of (4.7) with a given C ∗ -algebra A,
we obtain the sequence

0 K⊗A T0 ⊗ A SA 0. (4.11)

which is exact by Lemma 4.12. By Thm. 3.10, it therefore gives rise to a long exact
sequence in K-theory, the relevant part of which is
δA
··· K0 (S(T0 ⊗ A)) K0 (S 2 A) K0 (K ⊗ A) K0 (T0 ⊗ A) ···

48
with δA the corresponding differential. It is natural in A, as the differential depends
naturally on the exact sequence.
Let λ : C → K be the inclusion as rank one operators, so that K0 (λ ⊗ idA ) : K0 (A) →
K0 (K ⊗ A) is an isomorphism by Thm. 2.39. We now define

αA : K0 (S 2 A) → K0 (A), by αA := K0 (λ ⊗ idA )−1 ◦ δA .

It is then clear that the maps αA assemble to a natural transformation of functors α :

K0 S 2 ⇒ K0 . In other words, for any ∗-homomorphism Φ : A → B between C ∗ -algebras
A, B, the diagram
K0 (S 2 Φ)
K0 (S 2 A) K0 (S 2 B)
αA αB

K0 (A) K0 (Φ)
K0 (B).

commutes.
Lemma 4.16. For any other C ∗ -algebra B and x ∈ K0 (S 2 A), y ∈ K0 (B), we have

αA⊗B (x ⊗ y) = αA (x) × y.

Proof. Observe that by the definition (3.8) we have δA = K0 (jA )−1 ◦ K0 (iA ), where jA :
K ⊗ A → Cπ⊗idA and iA : S(SC ⊗ A) → Cπ⊗idA are the inclusion maps into the mapping
cone. Under the canonical isomorphisms Cπ⊗idA ∼ = Cπ ⊗ A and S(SC ⊗ A) ∼ = S 2 C ⊗ A,
these maps take the form jA = jC ⊗ idA and iA = iC ⊗ idA . We obtain that

αA = K0 (λ ⊗ idA )−1 K0 (jC ⊗ idA )−1 K0 (iC ⊗ idA ).

The statement now follows from the naturality (4.2) of the product. 

Proof. We show that both α ◦ β and β ◦ α are the identity transformation. First, the
identity αA ◦ βA = id follows from the calculation

(αA ◦ βA )(x) = αA (b × x) = αC (b) × x = 1 × x = x,

for x ∈ K0 (A), where we used property (2) and then property (1) of α.
Showing the identity βA ◦ αA = id is more involved. For any C ∗ -algebra A, denote by

σA : S 2 C ⊗ A → A ⊗ S 2 C, f ⊗ a 7−→ a ⊗ f

the “flip map”. Observe that for these maps, we have the identity

(idS 2 C ⊗ σA ) ◦ (σS 2 C ⊗ idA ) = σS 2 C⊗A : S 2 C ⊗ S 2 C ⊗ A −→ S 2 C ⊗ A ⊗ S 2 C. (4.12)

The important fact is now that

K0 (σS 2 C ⊗ idA ) = id. (4.13)

49
on K0 (S 2 C ⊗ S 2 C ⊗ A). To see this, identify S 2 C ∼= C0 ((0, 1)2 ) ∼
= C0 (R2 ) and notice that
under this identification, σS 2 C (f ) = Q∗ f , where Q is the linear map given by the matrix
 
1
 1
Q =b  1
.

1

Since this is a determinant one orthogonal matrix, it can be connected to the identity
matrix by a continuous path (Qt )t∈[0,1] in SO(4); then Φt (f ) := Q∗t f , t ∈ [0, 1], is a
continuous family of ∗-homomorphisms with Φ1 = σS 2 C , Φ0 = id. The claim now follows
from homotopy invariance, Thm. 2.24.
With these preparations, we calculate using that

x × b = K0 (σS 2 C⊗A )(b × x) Lemma 4.4(c)

= K0 (idS 2 C ⊗ σA )K0 (σS 2 C ⊗ idA )(b × x) (4.12) (4.14)
= K0 (S 2 σA )(b × x). (4.13)

Here we used that idS 2 C ⊗σA ∼

= S 2 σA under the identification S 2 C⊗A ∼
= S 2 A. Calculating
further, we get for any x ∈ K0 (S C ⊗ A) ∼
2 2
= K0 (S A) that

(K0 (σA ) ◦ βA ◦ αA )(x) = K0 (σA )(b × αA (x))

= αA (x) × b
= αA⊗S 2 C (x × b) Lemma 4.16(b)
= (αA⊗S 2 C ◦ K0 (S 2 σA ))(b × x) (4.14)
= (K0 (σA ) ◦ αS 2 C⊗A )(b × x) naturality of α
= K0 (σA )(x) α left inverse to β

Because σA and hence K0 (σA ) is an isomorphism, the result follows. 

Corollary 4.17. The Bott element is unique.

Proof. By Bott periodicity, the map βC = K0 (λ)−1 ◦ δC : K0 (C) → K0 (S 2 C) is an isomor-

phism, that is, K0 (S 2 C) = K0 (C) = Z. Since K0 (λ) is an isomorphism, δC : K0 (S 2 C) →
K0 (K) is an isomorphism as well. Hence there exists a unique element b such that δ(b)
corresponds to the element 1 ∈ Z ∼ = K0 (K). 

5 The K1-functor and the six-term exact sequence

In this chapter, we finish our exposition of the K-theory of C ∗ -algebras by introducing the
K1 -functor, which gives important interpretation for the boundary maps in the K-theory
six-term sequence.

50
5.1 Definition of K1
Definition 5.1 (Unitary groups). Let A be a C ∗ -algebra. For any n ∈ N, write
Un+ (A) := {u ∈ Mn (A)+ | u unitary and u = 1n + a, a ∈ Mn (A)}.
Denote by Un+ (A)0 ⊂ Un+ (A) the normal subgroup of those unitaries homotopic to 1n .
+
There exist the obvious inclusion maps Un+ (A) → Un+1 (A) given by
 
  a11 · · · a1n 0
a11 · · · a1n  . ... .. .. 
 .. . . ..  7−→  .. . .
 . . .   .
an1 · · · ann 0 
an1 · · · ann
0 ··· 0 1
+
By U∞ (A), we denote the union of all the Un+ (A), that is, the direct limit with respect
+
to the above inclusion maps. We have U∞ (A) ⊂ M∞ (A)+ , which induces a topology on
+
U∞ (A).
Definition 5.2 (The K1 -functor). Let A be a C ∗ -algebra.
(a) We define
+ +
K1 (A) := U∞ (A)/U∞ (A)0 .
(b) If B is another C ∗ -algebra and Φ : A → B is a ∗-homomorphism, we define
K1 (Φ) : K1 (A) → K1 (B), [u] 7−→ [Φ+ (u)].

In total, K1 is a functor from the category of C ∗ -algebras to the category of groups.

Remark 5.3. If A is unital, then A+ = A ⊕ C and the map u 7→ (u − 1n , 1) provides
an isomorphism from the unitary group Un (A) to Un+ (A). However, even in the unital
case, we need the groups Un+ (A) to deal with non-unital ∗-homomorphisms Φ : A → B.
Namely, for u ∈ Mn (A) unitary, Φ(u) is in general only a partial isometry, but Φ+ maps
+ +
U∞ (A) to U∞ (B).
Lemma 5.4. Let A be a C ∗ -algebra. Elements x, y ∈ K1 (A) coincide if and only if there
exists n ∈ N and a homotopy (ut )t∈[0,1] of unitaries ut ∈ Un+ (A) with x = [u0 ] and y = [u1 ].
+
Proof. First observe that two unitaries u0 , u1 ∈ U∞ (A) represent the same class in K1 (A)
if and only if they are homotopic: If they are homotopic, they are clearly in the same
+
connected component, that is, in the same coset of U∞ (A)0 . Conversely, elements in same
connected component can be joined by a path of unitaries.
We now prove that one can restrict to homotopies that lie in some Un+ (A) throughout.
Clearly, homotopies in Un+ (A) give rise to homotopies in U∞ +
(A). Conversely, let (ut )t∈[0,1]
+
be a homotopy in U∞ (A). Choose a partition 0 = t0 < t1 < · · · < tn = 1 such that
+
kuti − uti−1 k < 2. Then the unitaries uti all lie in some Um (A), for some m ∈ N. But by
+
remark Remark 3.3, since kuti − uti−1 k < 2 (this also holds within Um (A)), there exists a
+
homotopy in Um between uti−1 and uti ; concatenating these homotopies gives a homotopy
+
in Um (A) between u0 and u1 . 

51
Lemma 5.5. For any C ∗ -algebra A, K1 (A) is abelian. Moreover, if u ∈ Un+ (A) and
+
v ∈ Um (A), then [uv] = [diag(u, v)] in K1 (A).

Proof. Let x, y ∈ K1 (A) and write x = [u], y = [w] with u, w ∈ Un+ (A). Define elements
+ +
in U2n (A) by [This choice of rt is not contained in U2n (A), just in U2n (A+ )!]

cos πt 1n − sin πt


    
1n u 0 w 0
rt := 2 2 , wt := rt rt∗ .
sin πt πt



2
1 n cos 2
1 n 0 1 n 0 1 n

Then (wt )t∈[0,1] is a continuous path with w0 = diag(uw, 1n ) and w1 = diag(u, w). Define
a continuous path (wt0 )t∈[0,1] by swapping the roles of u and w in the formula above, so
that w00 = diag(wu, 1n ) and w10 = diag(w, u). Finally, define a path (vt )t∈[0,1] by
 
u 0 ∗
vt = rt r .
0 w t
+
Then (vt )t∈[0,1] is a continuous path in U2n (A) with v0 = diag(u, w) and v1 = diag(w, u).
+
Concatenating these paths appropriately gives a continuous path of unitaries in U2n (A)
from diag(uv, 1n ) to diag(vu, 1n ). This proves the claim. 

5.2 Identification with Suspension

Theorem 5.6. For any C ∗ -algebra A, there exists a canonical isomorphism

ηA : K1 (A) −→ K0 (SA)

such that for each ∗-homomorphism Φ : A → B, the diagram

K1 (Φ)
K1 (A) K1 (B)

ηA ηB (5.1)

K0 (SA) K0 (SΦ)
K0 (SB).

commutes. In other words, the maps ηA assemble to a natural isomorphism of functors

η : K1 ⇒ K0 S.

We will need the following lemma.

Lemma 5.7. Let A be a C ∗ -algebra. Let m ≥ n and let w ∈ M2n (A) be unitary such that
   
1n 0 1n 0
w ∗
w = ∈ M2 (Mn (A)) ∼
= M2n (A).
0 0 0 0

Then there exist unitaries u, v ∈ Mn (A), such that w = diag(u, v).

52
Proof. Let  
u a
w=
b v
with u, a, b, v ∈ Mn (A). We have to show that a = b = 0. Since w is unitary,
(
1 n = u ∗ u + a∗ a
     ∗ ∗
1n 0 ∗ u a u b
= ww = ∗ ∗ in particular
0 1n b v a v 1n = bb∗ + vv ∗ .

But (
= u∗ u
    ∗ ∗  
u a 1n 0 u b 1n 0 1n
= implies
b v 0 0 a∗ v ∗ 0 0 0 = bb∗ .
Putting together, we get a∗ a = b∗ b = 0, hence a = b = 0 (this follows from the C ∗ -
property, as kak2 = ka∗ ak = 0 and similarly for b). 

Proof (of Thm. 5.7). We will start with the definition of ηA , then show injectivity and
surjectivity of ηA and then verify that the square (5.1) commutes.
Definition of ηA : For A a C ∗ -algebra the map ηA : K1 (A) → K0 (SA) is defined as follows.
Given x ∈ K1 (A), write x = [u] with u ∈ Un+ (A) and let (wt )t∈[0,1] be a homotopy in
+
U2n (A) with w1 = diag(u, u∗ ) and w0 = 12n (such a homotopy exists by Corollary 3.4).
Then set
 
1n 0
wt∗ .
 
ηA (x) := f − [1n ] ∈ K0 (SA) with f (t) = wt
0 0

Notice that indeed, f (t) is a projection for every t ∈ [0, 1] and f (0) = f (1) = 1n , so
f ∈ M2n (SA+ ).
We have to check that ηA is independent from the choice of representative in Un+ (A) and
the choice of homotopy (wt )t∈[0,1] , as well as the choice of n ∈ N.

(1) Independence of n ∈ N: Write

 ∗
at at at c∗t
  
at b t
wt = so that f (t) = . (5.2)
ct d t ct a∗t dt c∗t

If we set u0 := diag(u, 1m ) for m ∈ N, then (wt0 )t∈[0,1] with

 
at 0 b t 0
 0 1m 0
wt0 := 
 ct 0 d t 0 
 (5.3)
0 0 0 1m

53
 is a homotopy of unitaries from diag(u0 , (u0 )∗ ) to 12n+2m , and the corresponding path
of projections is
 ∗
at at 0 at c∗t 0
 ∗
at at at c∗t 0 0
   
1n
0 0 0
0
1m 0  ∗ ∗
f 0 (t) := wt0   (wt0 )∗ =  0 ∗ 1m 0 ∗ 0 = σ  ct at dt ct 0 0 σ ∗
   
00 0 0  ct at 0 dt ct 0  0 0 1m 0
0
0 0 0 0 0 0 0 0 0 0 0
 ∗ ∗
at at at c∗t 0 0
   
1n 0 0 0 1n 0 0 0
0 0 1m 0   ct a∗t dt c∗t 0 0  0 0 1m 0 
= 0 1n
   .
0 0  0 0 1m 0  0 1n 0 0
0 0 0 1m 0 0 0 0 0 0 0 1m
(5.4)
0
With a view on (5.2), this shows that f ∼u diag(f, 1m ) for all t ∈ [0, 1], hence
 
0 f 0
[f ] − [1n+m ] = − [1n+m ] = [f ] − [1n ],
0 1m
as desired.
(2) Independence of representative and homotopy: Let u0 ∈ Un+ (A) with u0 ∼h u and let
+
(wt0 )t∈[0,1] be a homotopy in U2n (A) with w10 = diag(u0 , (u0 )∗ ) and w0 = 12n . We will
show that the path of projections f 0 (t) = wt0 diag(1n , 0)(wt0 )∗ is unitary equivalent to
the path f .
To this end, let (ut )t∈[0,1] be a homotopy in Un+ (A) with u0 = u and u1 = u0 (here
we need to possibly increase n before). Set now v(t) = wt diag(u∗ ut , uu∗t )(wt0 )∗ . Then
v(t) is unitary for each t ∈ [0, 1], with v(0) = 12n and
 ∗    ∗ 0  0 ∗   
u u1 0 0 ∗ u 0 uu 0 (u ) 0 1n 0
v(1) = w1 (w1 ) = = .
0 uu∗1 0 u∗ 0 u(u0 )∗ 0 u0 0 1n
Hence v is a unitary element in M2n (SA+ ). Moreover,
 ∗   ∗   
0 ∗ u ut 0 1n 0 ut u 0 ∗ 1n 0
v(t)f (t)v(t) = wt wt = wt wt∗ = f (t).
0 uu∗t 0 0 0 ut u∗ 0 0
This shows that f 0 ∼u f in M2n (SA+ ).
Homomorphism property: If x, x0 ∈ K1 (A), represent them by unitaries u, u0 ∈ Un+ (A).
By Lemma 5.5, we have x + x0 = diag(u, u0 ). Let (wt )t∈[0,1] and (wt0 )t∈[0,1] be homotopies
of unitaries in M2n (A+ ) with w0 = w00 and w1 = diag(u, u∗ ), w10 = diag(u0 , (u0 )∗ ) and let
f, f 0 ∈ M2n (A+ ) be the corresponding projections so that ηA (x) = [f ] − [1n ], ηA (x0 ) =
[f 0 ] − [1n ]
We define vt = s diag(wt , wt0 )s∗ , where s ∈ M4n (C) ⊂ M4n (A+ ) is the permutation matrix
that previously appeared in (5.4). This gives a homotopy (vt )t∈[0,1] with v0 = 14n and
v1 = (diag(u, u0 , u∗ , (u0 )∗ )). Hence
 
12n 0 ∗
ηA (x + y) = [g] − [12n ], where g(t) = vt v .
0 0 t

54
But
 
 1n
 wt∗
      
12n 0 ∗ wt 0  0 0 ∗ f 0 ∗
vt vt = s 0 

 0 (wt0 )∗ s = s 0 f 0 s ,

0 0 0 wt 1n
0

hence, since the loop constant equal to s defines an element of M4n (C) ⊂ M4n ((SA)+ ),
we have

ηA (x + x0 ) = [g] − [12n ] = [f ] − [1n ] + [f 0 ] − [1n ] = ηA (x) + ηA (x0 ),

as desired.
Injectivity: Let x ∈ K1 (A) with ηA (x) = 0. Represent x = [u] with u ∈ Un+ (A) and let
f (t) = wt diag(1n , 0)wt∗ ∈ M2n (A+ ), where f is a homotopy from diag(u, u∗ ) to 12n in
M2n (A). Then
0 = ηA (x) = [f ] − [1n ]
in K0 (SA).
We first treat the special case that f ∼u diag(1n , 0) in M2n ((SA)+ ). This means that
there exists a homotopy (vt )t∈[0,1] of unitaries in M2n (A+ ) with v0 = v1 = 12n and for all
t ∈ [0, 1],    
1n 0 ∗ 1n 0
= vt f (t)vt = vt wt (wt vt )∗ .
0 0 0 0
By Lemma 5.8, vt wt has the form vt wt = diag(ut , u0t ) for homotopies of unitaries (ut )t∈[0,1] ,
(u0t )t∈[0,1] . By construction, u0 = 1n and u1 = u, so that (ut )t∈[0,1] implements u ∼ 1n .
Therefore x = [u] = 0.
We finish by showing that the general case can be reduced to the special case just treated.
In general, [f ] − [1n ] = 0 only means that diag(f, 1m ) ∼u 1n+m for some m ∈ N and
all t ∈ [0, 1]. Write u0 = diag(u, 1m ) ∈ Mn+m (A+ ) (which is also a representative for x)
and let (wt0 )t∈[0,1] be the homotopy from diag(u0 , (u0 )∗ ) to 12n+2m given in (5.3). Then as
calculated in (5.4),
 
1n 0 0 0
 0 1m 0
diag(f (t), 1m ) ∼u wt0   0 ∗
 0 0 0 0 (wt ) =: f (t).
0

0 0 0 0

Thus
− [1n+m ] = [f 0 ] − [1n+m ],


ηA (x) = [f ] − [1n ] = diag(f, 1m )
where by the choice of f 0 , we have f 0 ∼u 1n+m in M2n+2m ((SA)+ ). This reduces to the
special case above.
Surjectivity: Let y ∈ K0 (SA). By Prop. 2.19(a), we can represent y = [f ] − [1n ] for some
n ∈ N and some projection f ∈ M∞ ((SA)+ ) with f − 1n ∈ M∞ (SA). As discussed in

55
Prop. 3.14 there exists m ≥ n and a homotopy (wt )t∈[0,1] of unitaries in Mm (A+ ) such
that f (t) = wt diag(1n , 0)wt∗ for all t ∈ [0, 1]. Moreover, we may assume that m = 2n
(otherwise represent y = [(diag(f, 1k ))] − [1n+k ] for some suitable k instead). We now
have    
1n 0 1n 0
= f (1) = w1 w1∗ ,
0 0 0 0
so by Lemma 5.8, w1 = diag(u, v) for unitaries u, v ∈ Un+ (A). If now (wt0 )t∈[0,1] is a
homotopy of unitaries with w00 = 12n and w10 = diag(u, u∗ ), then by definition, we have
 
0 0 0 1n 0
ηA ([u]) = [f ] − [1n ], where f (t) = wt (wt0 )∗ .
0 0

To see that y is in the image of ηA , we will show that [f 0 ] − [1n ] = y.

Suppose first that v ∼h u∗ in Un+ (A). Then there exists a homotopy (st )t∈[0,1] of unitaries
with s0 = 1n and s1 = v ∗ u∗ . Therefore
     ∗
1n 0 ∗ 1n 0 1n 0 1n 0
f (t) = wt wt = wt wt∗
0 0 0 st 0 0 0 st
   ∗
1n 0 0 ∗ 0 0 1n 0
= wt (wt ) f (t) wt wt∗
0 st 0 st

for all t ∈ [0, 1]. Now one easily checks that homotopy (wt diag(1n , st )(wt0 )∗ ) starts and
ends at 12n , hence defines a unitary element in M2n ((SA)+ ) and implements f ∼u f 0 .
In general, diag(u, v) ∼ 12n only implies that diag(u, 1m ) ∼h diag(v, 1m ) for some m ∈ N,
but this case can be reduced to the previous one by stabilising appropriately, as before.
Commutativity of (5.1): Let x ∈ K1 (A) and represent x = [u] with u ∈ Un+ (A). Let
moreover (wt )t∈[0,1] be a homotopy with w1 = diag(u, u∗ ) and w0 = 12n , so that ηA (x) =
[f ] − [1n ] with f (t) := wt diag(1n , 0)wt∗ . Then K1 (Φ)(x) = [Φ+ (u)] and (Φ+ (wt ))t∈[0,1] is
a homotopy between diag(Φ+ (u), Φ+ (u)∗ ) and 12n . Therefore
   
+ + 1n 0 ∗
Φ (wt ) − [1n ] = Φ+ (f ) − [1n ]
+


ηB ([Φ (u)]) = Φ (wt )
0 0

hence
ηB ◦ K1 (Φ)(x) = ηB ([Φ+ (u)])
= Φ+ (f ) − [1n ]



= K0 (SΦ)([f ] − [1n ])
= K0 (SΦ) ◦ ηA (x),
as desired. 

Example 5.8 (The Bott element, again). The function z ∈ SC+ = C(T) is unitary
and hence defines an element of K1 (SC). We calculate the image of the class [z] under

56
ηSC : K1 (SC) → K0 (S 2 C). To this end, we need a path w of unitaries with w(0) = 12 ,
w(1) = diag(z, z). The standard construction, exhibited in the proof of Prop. 2.6(d), is

cos πt − sin πt cos πt sin πt

 

  


z 0 2 2
z 0 2 2
w(t) =
,
sin πt πt πt πt




0 1 2
cos 2
0 1 − sin 2
cos 2

which is just uBott , defined in (4.8). Hence

     
1 0 ∗ 1 0
η([z]) = uBott u − = b,
0 0 Bott 0 0

the Bott element.

5.3 The index map

Definition 5.9 (Index map). Let A be a C ∗ -algebra and J ⊂ A a closed ideal. Let
δ be the boundary map to the short exact sequence (3.1). The index map is the group
homomorphism Ind : K1 (A/J) −→ K0 (J) making the diagram
ηA/J
K1 (A/J) K0 (S(A/J))

δ
Ind
K0 (J)

commutative.

Proposition 5.10 (Formula for the index map). Let A be a C ∗ -algebra and J ⊂ A
+
a closed ideal. Given u ∈ Un+ (A/J), choose w
e ∈ U2n e = diag(u, u∗ ). Then
(A) with π + (w)
     
1n 0 ∗ 1n 0
Ind([u]) = w we − ∈ K0 (J). (5.5)
0 0 0 0
e

Proof. We remark first of all that such a lift we exists by Corollary 3.4). Observe that
Ind([u]) does not depend on the choice of lift. Namely, if we0 is another left, then we set
0 ∗
ve = w
ew e and observe that π (v) = 12n , hence ve ∈ K0 (J ). Therefore, in V(J + ), we have
+ +

           
0 1n 0 0 ∗ 1n 0 ∗ ∗ 1n 0
w (we ) = vew w
e ve = w w∗
0 0 0 0 0 0
e e e

We may therefore assume that w e ∼h 12n (such a lift exists by Corollary 3.4) . In this
+
case, let (w
et )t∈[0,1] be a continuous family of unitaries w et ∈ U2n (A) such that w
e0 = 12n ,
+
w
e1 = w.e Set moreover wt = π (w et ) for t ∈ [0, 1]. Then by the definition of η,
 
1n 0
ηA/J ([u]) = [f ] − [1n ], f (t) = wt wt∗ .
0 0

57
Moreover, the path fe of projections in M2n (A+ ) given by fe(t) = w et∗ is a lift
et diag(1n , 0)w
of f , hence by the formula (3.13), we have

δ([f ]) = [fe(1)] − [fe(0)].

But this is precisely (5.5). 

Proposition 5.11. Let A be a unital C ∗ -algebra and let J ⊂ A be a closed ideal. Let
v ∈ A be a partial isometry such that 1 − v ∗ v ∈ J and 1 − vv ∗ ∈ J. Then π(v) ∈ U(A/J)
and
Ind([π(v)]) = [1 − v ∗ v] − [1 − vv ∗ ] ∈ K0 (J).

Here we use Remark 5.3 to identify U(A/J) with U + (A/J), in order to see how π(v)
defines an element of K1 (A/J).
Proof. Since 1 − v ∗ v, 1 − vv ∗ ∈ J, we have

π(v)∗ π(v) = π(v ∗ v) = 1 and π(v)π(v)∗ = 1.

Hence π(v) is indeed unitary in A/J. Moreover, one easily checks that

1 − vv ∗
 
v
u := ∈ M2 (A)
1 − v∗v v∗

is a unitary lift of diag(π(v), π(v)∗ ) ∈ M2 (A/J). Now by (5.5), we have

     
1 0 ∗ 1 0
Ind([π(v)]) = u u −
0 0 0 0
 ∗   
vv 0 1 0
= −
0 1 − v∗v 0 0
= [vv ∗ ] + [1 − v ∗ v] − [1]
= [1 − v ∗ v] − [1 − vv ∗ ],

where we used that [1] = [vv ∗ ] + [1 − vv ∗ ] (see e.g. the proof of Prop. 2.19(a)). 

Example 5.12 (The Bott element, once more). Because the adjoint shift operator
S ∗ ∈ T0+ = T is a partial isometry lifting the unitary z ∈ SC+ , we have by Prop. 5.12

Ind([z]) = [1 − SS ∗ ] − [1 − S ∗ S] = [1 − SS ∗ ].

This gives another proof that ηSC ([z]) is a Bott element.

Remember that an operator T ∈ B is called Fredholm if π(T ) ∈ B/K is invertible. In this

case, its index is defined as

ind(T ) := dim ker(T ) − dim coker(T ) = dim ker(T ) − dim ker(T ∗ ).

58
Remark 5.13. Let T ∈ B be a Fredholm operator. That π(T ) is invertible means that
there exists a parametrix S ∈ B with T S − 1 =: K ∈ K and ST − 1 =: L ∈ K. Hence
ker(T ) ⊆ ker(ST ) = ker(1 + L), which is the eigenspace to eigenvalue −1 of L. But since
L is compact, this is finite-dimensional. Similarly, ker(T ∗ ) ⊆ ker(S ∗ T ∗ ) = ker(1 + K ∗ ) is
finite-dimensional. Thus ind(T ) is well-defined.

Proposition 5.14. Let V ∈ B be partial isometry which is Fredholm. Then

Ind([π(V )]) = [Pker(V ) ] − [Pker(V ∗ ) ] (5.6)

where Pker(V ) and Pker(V ∗ ) are the orthogonal projections onto ker(V ), respectively ker(V ∗ )
and Ind is the index map for the pair K ⊂ B. In particular,

τ (Ind(π(V ))) = ind(V ), (5.7)

where τ : K0 (K) → Z is the isomorphism that sends a rank one projection to 1 ∈ Z.

Proof. We claim that

1 − V ∗ V = Pker(V ) , and 1 − V V ∗ = Pker(V ∗ ) . (5.8)

Indeed, if ϕ ∈ ker(V ), then (1 − V ∗ V )ϕ = ϕ, while if ϕ ∈ ker(V )⊥ = im(V ∗ ), we have

ϕ = V ∗ ψ for some ψ ∈ H. Therefore using (2.1)

(1 − V ∗ V )ϕ = ϕ − V ∗ V V ∗ ψ = ϕ − V ∗ ψ = ϕ − ϕ = 0.

This shows the first identity in (5.8); the second follows from replacing V by V ∗ . Formula
(5.6) is now a consequence of Lemma 5.12.
Formula (5.7) follows from observing that τ ([P ]−[Q]) = tr(P )−tr(Q) (see Example 2.20)
and that the trace of a projection is equal to its rank. 

5.4 The exponential map

We start by deriving a more explicit formula for the Bott map in terms of the K1 group.

Proposition 5.15. Let A be a C ∗ -algebra. For a projection p ∈ Mn (A+ ), define the

projection loop fp ∈ Mn (C([0, 1], A+ )) by

fp (t) = e−2πit p + 1n − p. (5.9)

−1
Then the composition βA0 := ηSA ◦ βA : K0 (A) → K1 (SA) is given by the formula

βA0 ([p] − [q]) 7−→ [fp fq∗ ].

for projections p, q ∈ Mn (A+ ) such that p − q ∈ Mn (A).

59
Remark 5.16. We generally do not have fp ∈ Un+ (SA). However, if [p] − [q] ∈ K0 (A)
such that p − q ∈ Mn (A), then εA (p) = εA (q). Since εA (fp (t)) = fεA (p) (t) ∈ Mn (C), we
therefore obtain εA (fq (t)fq (t)∗ ) = 1n in Mn (C), hence fp fq∗ ∈ Un+ (SA).
−1
Proof. It suffices to verify this for unital algebras A, since ηSA ◦ βA is the restriction of
−1 −1
ηSA+ ◦ βA+ . As ηSA ◦ βA is homomorphism, it moreover suffices to verify that
  
1 0
ηSA ([fp ]) = b × [p] = [pBott ⊗ p] − ⊗p .
0 0
+
To calculate the left hand side, we need to choose a path w in U2n (SA) connecting 12n
∗ 2
to diag(fp , fp ); in other words, an element w ∈ C([0, 1] , A) with w(t, 0) = w(0, s) =
w(1, s) = 12n and w(t, 1) = diag(fp , fp∗ ). Then
     
1n 0 ∗ 1n 0
ηSA ([fp ]) = w w −
0 0 0 0

A possible choice is

  ∗
cos πs − sin πs cos πs sin πs
 



fp 0 2 2
fp 0 2 2
w=
sin πs πs πs πs





0 1n 2
cos 0 1 n − sin cos

2
 2  2
cos πs πs πs πs



2
p − sin 2
zp cos 2
sin 2
1 n − p 0
=
+
sin πs cos πs − sin πs cos πs




2
zp 2
p 2 2
0 1n − p
 
1 0
= uBott ⊗ p + ⊗ (1n − p),
0 1

where we wrote z = e2πit and uBott is the unitary (4.8) used in the definition (4.9) of the
Bott projection. Therefore
        
1 0 ∗ 1 0 1n 0
ηSA ([fp ]) = uBott u ⊗p + ⊗ (1n − p) −
0 0 Bott 0 0 0 0
  
1 0
= [pBott ⊗ p] − ⊗p ,
0 0

which is what needed to be shown. 

Definition 5.17 (Exponential map). Let A be a C ∗ -algebra and J ⊂ A be a closed

ideal. Then the corresponding exponential map is the map Exp : K0 (A/J) → K1 (J) such
that the square
βA/J
K0 (A/J) ∼
=
K0 (S 2 (A/J))
Exp Sδ (5.10)
ηJ
K1 (J) ∼
=
K0 (SJ),
commutes, where Sδ is the boundary map to the suspended ideal SJ ⊆ SA.

60
Proposition 5.18 (Formula for the exponential map). Let A be a C ∗ -algebra and
let J ⊂ A be a closed ideal. The exponential map can be described as follows. Given
x ∈ K0 (A/J), represent x = [p] − [1k ] with a projection p ∈ Mn ((A/J)+ ), n ≥ k such that
p − 1k ∈ Mn (A/J). Then  
Exp(x) = exp(2πie p) , (5.11)
where pe ∈ Mn (A+ ) is some self-adjoint lift of p.

Remark 5.19. We emphasize that pe is not required to be a projection as well. In fact, if

pe ∈ Mn (A+ ) is also a projection, then it has spectrum σ(e p) ⊆ {0, 1}, hence exp(2πie
p) =
1n , which represents the zero element of K1 (J). So in this sense, the exponential map
provides a measure of the failure of p to lift to a projection.
+
Proof. Observe first that indeed exp(−2πie p) ∈ Um (J), as it is a unitary in Mm (J + ) and

π + exp(2πie


p) = exp(2πip) = 1m ,

hence exp(2πiep) − 1m ∈ Mm (J).

As before, we may assume that A is unitary. Let p ∈ Mn (A/J) be a projection. To use
the previous results, extend the Diagram (5.10) as follows:

0
βA/J K1 (S(A/J))
ηS(A/J)
βA/J
K0 (A/J) ∼
=
K0 (S 2 (A/J))
S Ind
Exp Sδ
ηJ
K1 (J) ∼
=
K0 (SJ),

where S Ind : K1 (S(A/J)) → K0 (SJ) is the index map corresponding to the suspended
ideal SJ ⊂ SA. Then by Prop. 5.16 and Def. 5.10, we can write

(Sδ ◦ βA/J )([p]) = (Sδ ◦ ηS(A/J) )([fp ]) = S Ind([fp ]),

We then want to verify

S Ind([fp ]) = ηJ ([p]).
e ∈ M2n (SA) be a unitary lift of diag(fp , fp∗ ), that
To calculate the left hand side, let w
is w(0)
e = w(1)
e = 12n and π + (w(t))
e = diag(fp (t), fp∗ (t)) for all t ∈ [0, 1]. Then by the
formula for the index map, Prop. 5.11, we have
     
1n 0 ∗ 1n 0
S Ind([fp ]) = w we − .
0 0 0 0
e

On the other hand, let pe ∈ Mn (A) be a self-adjoint lift of p and define u

e(t) := exp(2πite
p).
Then
u(t)) = exp(2πitp) = e2πit p + 1n − p = fp (t)∗ .
π + (e

61
Therefore, if we set  
ue(t) 0
ve(t) := w(t) ,
0 u e(t)∗
e

then π + (e
v (t)) = 12n for all t ∈ [0, 1], hence we obtain a continuous path of unitaries
in M2n (J + ) with ve(0) = 12n and ve(1) = diag(exp(2πie p), exp(−2πiep)). Therefore, by
definition of ηJ ,
     
1n 0 ∗ 1n 0
ηJ ([exp(2πie
p)]) = ve ve −
0 0 0 0
     ∗    
u
e 0 1n 0 u
e 0 ∗ 1n 0
= w we −
0 u e∗ 0 0 0 u 0 0
e
e
     
1 0 1n 0
= we n we∗ − = S Ind([p])
0 0 0 0

This finishes the proof. 

5.5 The six-term exact sequence

Theorem 5.20 (The six-term sequence). Let A be a C ∗ -algebra and let J ⊂ A be a
closed ideal. Then the six-term sequence
K0 (ι) K0 (π)
K0 (J) K0 (A) K0 (A/J)

Ind Exp

K1 (A/J) K1 (π)
K1 (A) K1 (ι)
K1 (J)

is exact.

Proof. So far, from Thm. 3.10, we know the exactness of the (non-dashed) spiral sequence

K0 (S 2 ι) K0 (S 2 π)
K0 (S 2 J) K0 (S 2 A) K0 (S 2 (A/J))

βJ ∼
= βA ∼
= βA/J ∼
=

K0 (ι) K0 (π)
K0 (J) K0 (A) K0 (A/J)
Sδ

K0 (S(A/J)) K0 (Sπ)
K0 (SA) K0 (Sι)
K0 (SJ).

62
However, the Bott periodicity isomorphisms (dashed) provide an exact K0 -K0 S-six-term
sequence, where the right boundary map K0 (A/J) → K0 (SJ) is Sδ ◦ βA/J . Here the
commutativity of the above diagram follows from naturality of β. The commutativity of
the diagram
K0 (ι) K0 (π)
K0 (J) K0 (A) K0 (A/J)

δ Sδ◦βA/J

Ind
K0 (S(A/J)) K0 (Sπ)
K0 (SA) K0 (Sι)
K0 (SJ) Exp

ηA/J ηA ηJ

K1 (A/J) K1 (π)
K1 (A) K1 (ι)
K1 (J),

which follows from naturality of η, then implies the exactness of the corresponding K0 -
K1 -sequence. 

References
[1] N. Brown and N. Ozawa. C ∗ -algebras and finite-dimensional approximations. 2008.

[2] J. Dixmier. C ∗ -algebras. North-Holland Publishing Co., Amsterdam-New York-

Oxford, 1977. Translated from the French by Francis Jellett, North-Holland Mathe-
matical Library, Vol. 15.

[3] G. Murphy. C ∗ -algebras and operator theory. Academic Press Inc., Boston-San Diego-
New York, 1990.

[4] N. E. Wegge-Olsen. K-theory and C ∗ -algebras. Oxford Science Publications. The

Clarendon Press, Oxford University Press, New York, 1993. A friendly approach.

[5] D. Werner. Funktionalanalysis.

63