Lectures on Cyclic Cohomology

Masoud Khalkhali
Mathematics Department, University of Western Ontario
London ON, Canada

Lectures at IMPAN, Warsaw, September 2016

Simons Semester on Noncommutative Geometry
2
Contents

0.1 Index theory in a noncommutative setting . . . . . . . . . . . . . . 7

0.2 Elliptic operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.3 The Gauss-Bonnet-Chern theorem as an index theorem . . . . . . 8
0.4 The signature theorem . . . . . . . . . . . . . . . . . . . . . . . . . 11
0.5 Riemann-Roch-Hirzebruch theorem . . . . . . . . . . . . . . . . . . 12
0.6 An odd index theorem . . . . . . . . . . . . . . . . . . . . . . . . . 12
0.7 Connes’ noncommutative index theorem . . . . . . . . . . . . . . . 13
0.8 Hochschild cohomology . . . . . . . . . . . . . . . . . . . . . . . . 15
0.9 Hochschild cohomology as a derived functor . . . . . . . . . . . . . 20
0.10 Deformation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
0.11 Topological algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 36
0.12 Hochschild (co)homology computations . . . . . . . . . . . . . . . 38
0.13 Cyclic Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
0.14 Connes’ long exact sequence . . . . . . . . . . . . . . . . . . . . . . 60
0.15 Connes’ spectral sequence . . . . . . . . . . . . . . . . . . . . . . . 64
0.16 Cyclic cohomology computations . . . . . . . . . . . . . . . . . . . 66
0.17 What is the Chern character? . . . . . . . . . . . . . . . . . . . . . 79
0.18 Connes–Chern character in K-theory . . . . . . . . . . . . . . . . . 80
0.19 Connes–Chern character in K-homology . . . . . . . . . . . . . . . 90
0.20 Connes’ index theorem . . . . . . . . . . . . . . . . . . . . . . . . . 101

Appendices 106

A Compact operators, Fredholm operators, and abstract index the-

ory 107

B K-theory 115

Bibliography 119

3
4 Introduction
Introduction

In September 2016 I gave 5 introductory lectures on cyclic cohomology and some

of its applications in IMPAN Warsaw, during the Simons Semester in Noncommu-
tative Geometry. The audience consisted of graduate students and postdocs and
my task was to introduce them to the subject. The following text is an expanded
version of my lectures. In producing these lecture notes I have freely used mate-
rial from my book, Basic Noncommutative Geometry. I would like to thank Piotr
Hajac for the invitation to give these lectures.

5
6 Introduction
Lecture 1: Index theory in a
noncommutative setting

0.1 Index theory in a noncommutative setting

Lecture 1, September 12, 2016. In this first lecture, we try to answer a fun-
damental question: what is index theory and how it relates to noncommutative
geometry? We shall see that the need to extend the Atiyah-Singer index theorem
beyond its original domain had much influence on the development of noncom-
mutative geometry and its basic tools like cyclic cohomology and Connes- Chern
character maps. We shall then formulate the first index theorem in a noncom-
mutative setting due to Connes [19] that exhibits this point in a very clar way .
Most of what we say in these five lectures are geared towards understanding the
statement of this noncommutative index theorem and its proof. In the last lecture
we shall give some applications of this index theorem.
We shall freely use concepts of functional analysis like Fredholm operators,
compact operators and p-summable operators, and notions of differential and al-
gebraic topology, topological K-theory, and differential geometry. The reader can
consult the appendix for abstract Fredholm theory and K-theory.

0.2 Elliptic operators

Elliptic differential operators acting on smooth sections of vector bundles over
closed manifolds define Fredholm operators on the corresponding Sobolev spaces of
sections. Computing the index of such Fredholm operators in terms of topological
data is what the index theorem of Atiyah–Singer achieves. Let M be a smooth
manifold and let E and F be smooth complex vector bundles on M . Let

D : C ∞ (E) → C ∞ (F )

be a liner differential operator. This means that D is a C-linear map which is

locally expressible by a matrix of differential operators. This matrix of course
depends on the choice of local coordinates on M and local frames for E and F .
The principal symbol of D is defined by replacing differentiation by covectors in the

7
8 Introduction

leading order terms of D. The resulting ‘matrix-valued function’ on the cotangent

bundle’
σD ∈ C ∞ (Hom(π ∗ E, π ∗ F ))
can be shown to be invariantly defined. Here π : T ∗ M → M is the natural projec-
tion map of the cotangent bundle.
To compute the principal symbol we can proceed as follows. Let D = [Dij ],
denote the matrix of the differential operators D in some
P local coordinate system
for M , and local frames for E and F . Thus Dij = |k|≤d akij (x)∂k with d =
 
degree of D. Then σD (x, ξ) = σDij (x, ξ) , where
X 1
σDij (x, ξ) = akij (x)( )|k| ξ k
i
|k|=d

is obtained by replacing the partial derivatives by covectors: ∂k → ( 1i )|k| ξ k .

Definition 0.2.1. A differential operator D is called an elliptic operator if for all
x ∈ M and all ξ 6= 0 in Tx∗ M , the matrix σD (x, ξ) is invertible.
Theorem 0.2.1. (Finiteness theorem) Let M be a smooth and closed manifold
and E and F smooth complex vector bundles over M . If D : C ∞ (E) → C ∞ (F ) is
an elliptic operator, then

dim(ker D) < ∞ and dim(coker D) < ∞

In particular, we can define the index of D by

index(D) = dim(ker D) − dim(coker D).

Let W s (E) denote the Sobolev space of sections of E (roughly speaking, it

consists of sections whose ‘derivatives of order s’ are square integrable). The main
results of the theory of linear elliptic PDE’s show that for each s ∈ R, D has a
unique extension to a bounded and Fredholm operator D : W s (E) → W s−d (F ).
Moreover the Fredholm index of D is independent of s and coincides with the
index defined using smooth sections.
We start with a few classic examples of elliptic operators and their indices.

0.3 The Gauss-Bonnet-Chern theorem as an in-

dex theorem
p p+1 p
Let M
∞
Vpbe ∗a Riemannian manifold, and let d : Ω (M ) → Ω (M ), where Ω (M ) :=
C ( T M ) is the space of p-forms on M , denote the de Rham differential. It is
a differential operator but is not elliptic. Let d∗ : Ωp (M ) → Ωp−1 (M ) denote the
adjoint of d. Here d∗ = − ∗ d∗, with ∗ : Ωp (M ) → Ωn−p (M ) the Hodge ∗ operator.
Consider the differential operator

d + d∗ : Ωev (M ) −→ Ωodd (M ),
 0.3 The Gauss-Bonnet-Chern theorem as an index theorem 9

where Ωev (M ) = ⊕Ω2p (M ).This is an elliptic operator. In fact it is easy to find

the symbol of d + d∗ :

σd+d∗ (x, ξ) : ∧ev (Tx∗ M ) → ∧odd (Tx∗ M )

is given by √
σd+d∗ (x, ξ) = −1(ξ ± iξ ),
where iξ is the interior multiplication by ξ. A simple computation then shows
that σd+d∗ (x, ξ) is invertible for ξ 6= 0 and therefore d + d∗ is an elliptic operator.
Consequently, index(d + d∗ ) is defined. Using Hodge theory, one shows that
Proposition 0.3.1. With the notations and assumptions as above
index(d + d∗ ) = χ(M ),
where χ(M ) is the Euler characteristic of M .
In fact, with regard to ker(d + d∗ ) : Ωev → Ωodd , we have (d + d∗ )w = 0 if and
only if dw = 0 and d∗ w = 0 (we can assume ω is homogeneous). Consider the
Laplacian on forms
∆ := (d + d∗ )2 = dd∗ + d∗ d.
We have
ker(d + d∗ )|Ωev = ker ∆ = H2p (M ),
L
p

where H2p (M ) is the space of harmonic 2p forms on M . We need the following

theorem.
Theorem 0.3.1. (Hodge) Under the previous assumptions, the map w 7→ [w]
defines an isomorphism of vector spaces
p
Hp (M ) → HdR (M )

Remark 1. The space Hp (M ) depends on the metric we choose. However, the

dimension of this space is a topological invariant and independent of the choice of
the metric.
Thus we get
2p
dim ker(d + d∗ )|Ωev =
P
p dim HdR (M ).

Similarly we have:
2p+1
dim ker(d + d∗ )|Ωodd =
P
p dim HdR (M ).

This shows that

p
index(d + d∗ ) = p
P
p (−1) dim HdR (M ) = χ(M ).
10 Introduction

Now, we can consider the following question: is there a local formula for χ(M )?
More precisely we want to know if there is a naturally defined cohomology class
whose integral over M gives the Euler characteristic of M. An affirmative answer
to this question is given by the celebrated s Gauss-Bonnet-Chern theorem:
Theorem 0.3.2. Let M be a closed oriented Riemannian manifold of dimension
2n. Its Euler characteristic can be expressed as
Z
−2n
χ(M ) = (2π) P f (Ω).
M

Here, working in a coordinate system, Ω = (wij ) is an skew symmetric matrix

of two forms (the curvature matrix of M), and the Pfaffian P f is an invariant
polynomial on the space of skew symmetric 2n × 2n matrices. The Pfaffian is an
square root of the determinant in the sense that P f (Ω)2 = det(Ω). The first few
are  
0 a
pf = a.
−a 0
 
0 a b c
−a 0 d e
 −b −d 0 f  = af − be + dc.
pf  

−c −e −f 0
 
0 a1 0 0
−a1 0 b1 0 
 
 0
 −b1 0 a2 

 . .. . .. 
pf 
 0 0 −a 2
 = a1 a2 · · · an .

 . .. . ..


 bn−1 

 −bn−1 0 an 
−an 0
In general, if A = (aij ) is an skew symmetric 2n × 2n matrix, its Pfaffian is defined
by
n
1 X Y
pf(A) = n sgn(σ) aσ(2i−1),σ(2i)
2 n! i=1
σ∈S2n
The Gauss-Bonnet-Chern theorem is a prototype of an index theorem. An
invariant of a manifold defined as the index of an elliptic operator, the analytic
index, is expressed in terms of the integral of a differential form, called a topological
index. Thus the above result can be expressed as
analytic index = topological index
Remark 2. In the previous theorem if dim(M ) = 2, we get the original Gauss-
Bonnet theorem which relates the Euler characteristic of a surface to its total
Gaussian curvature: Z
1
χ(M ) = K dvolg .
2π M
 0.4 The signature theorem 11

Here g is the Riemannian metric on M , and K is the Gauss curvature (which is

half of the scalar curvature).

0.4 The signature theorem

A similar approach works for Hirzebruch’s signature theorem. Let M be a closed
oriented manifold of dimension 4n. Consider the intersection pairing on middle
forms Ω2n (M ) :
Z
(α, β) 7→ α ∧ β.
M

It induces a symmetric bilinear form on middle cohomology H 2n (M ). The signa-

ture of M, σ(M ), is defined as the signature of this bilinear form.
Recall the definition of the signature of a bilinear form B on a finite dimensional
vector space V . Let b+ be the dimension of the largest subspace on which B is
positive definite and b− the dimension of the largest subspace on which B is
negative definite. The signature of B is defined as the number b+ − b− .
Using Hodge theory, one shows that σ(M ) is also an index of an operator. Let
ΩpC (M ) denote the complexified p-forms on M. Define an operator γ : ΩpC (M ) →
Ωn−p
C (M ) by
γ(ω) = ip(p−1)+n ? ω, ω ∈ Ωp (M )

Let Ω+ (M ) (resp. Ω− (M )) denote the +1 (resp. -1) eigenspace of γ. Then

using Hodge theory again one can show that the index of elliptic operator

d + d∗ : Ω+ (M ) → Ω− (M )

is equal to the signature of M :

index(d + d∗ ) = σ(M )

The Hirzebruch signature theorem exprsesses this analytic index as a topological

index by the formula
Z
σ(M ) = Ln (p1 , . . . , pn )
M

where pi ’s are Pontryagin classes and L is the Hirzebruch L-polynomial. The first
are

L0 = 1
1
L1 = 3 p1
1 2
L2 = 45 (7p2 − p1 )
L3 = 1
945 (62p3 − 13p1 p2 + 2p31 )
12 Introduction

0.5 Riemann-Roch-Hirzebruch theorem

Let M be a compact complex manifold and E a holomorphic vector bundle on M .
Let Ω(0,p) (M, E) denote the space of (0, p)− forms on M with coefficients in E.
The holomorphic structure on E defines a differential operator

∂¯E : Ω(0,p) (M, E) → Ω(0,p+1) (M, E).

Like before we can assemble the even and odd parts of this Dolbeaux complex to
get and elliptic PDE
M M
∂¯E : Ω(0,2p) (M, E) → Ω(0,2p+1) (M, E).
p p

The holomorphic Euler characteristic of E is defined as the index of this operator

χ(E) = index ∂¯E

Using Hodge theory one shows that this definition coincides with the definition
based on sheaf cohomology.
The Riemann-Roch theorem of Hirzebrch gives a formula for this number in
terms of topological invariants:
Z
χ(E) = ch(E) td(X).
M

0.6 An odd index theorem

The index theorems we have discussed so far are all on even dimensional mani-
folds. In fact the index of any elliptic differential operator on an odd dimensional
manifold is necessarily zero. One way to see this would be to express the index
as the super trace of the heat kernel and use the general form of the asymptotic
expansion of the heat trace. To get examples on odd dimensional manifolds, one
can look at pseudodifferential operators. Here is a the simplest example.
Let H = L2 (S 1 ). Any continuous function f : S 1 → C defines a Toeplitz
operator
Tf = P f P : H + → H + ,
where H + is the Hardy space of functions whose negative Fourier coefficients
vanish and P : H → H + is the corresponding projection. Then it can be shown
that Tf is a Fredholm operator if and only if f is nowhere zero. The following
standard result, known as the Gohberg–Krein index theorem, computes the index
of the Toeplitz operator in terms of the winding number of f :

index(P f P ) = −W (f, 0).

To prove this formula notice that both sides are homotopy invariant. For the left-
hand side this is a consequence of the homotopy invariance of the Fredholm index
 0.7 Connes’ noncommutative index theorem 13

while for the right-hand side it is a standard fact about the winding number. Also,
both sides are additive. Therefore it suffices to show that the two sides coincide
on the generator of π1 (S 1 ), i.e., for f (z) = z. Then P zP is easily seen to be the
forward shift operator given by P zP (en ) = en+1 in the Fourier basis. Clearly then
index(P zP ) = −1 = −W (z, 0).
When f is smooth we have the following well-known formula for the winding
number: Z
1 1
W (f, 0) = f −1 df = ϕ(f −1 , f ),
2πi 2πi
where ϕ is the cyclic 1-cocycle on C ∞ (S 1 ) defined by ϕ(f, g) = f dg. Since this
R

cyclic cocycle is the Connes–Chern character of the Fredholm module (H, F ), the
above equation can be written as
1
h[(H, F )], [f ]i = hChodd (H, F ), Chodd (f )i,
2πi
where the pairing on the right-hand side is between cyclic cohomology and homol-
ogy. As we shall prove next, this is a special case of a very general index formula
of Connes.

0.7 Connes’ noncommutative index theorem

We saw that the gist of an index theorem is the equality of an analytic index,
defined as the index of a Fredholm operator, with a topological index, defined as
the integration of a cohomology class over a homology cycle. To formulate an
index theorem on a noncommutative space one must reformulative the analytic
and topological and index in a new fashion.
Connes’ index formulae that we hsall prove in these lectures (Theorem 0.20.1
and Proposition 0.20.1) acheives this goal as the following commutative diagram.
In fact many aspects of noncommutative geometry are beautifully displayed in this
diagram.
K∗ (A) × K∗ (A)
index /Z

Ch∗ Ch∗ (1)

  
HP ∗ (A) × HP∗ (A) / C.

In this diagram:
1) A is an algebra which may very well be noncommutative.
2) K∗ (A) is the set of even resp. odd finitely summable Fredholm modules over
A. It is closely related to the K-homology of A.
3) K∗ (A) is the algebraic K-theory of A.
14 Introduction

4) Ch∗ is Connes–Chern character in K-homology.

5) Ch∗ is Connes–Chern character in K-theory.
6) HP ∗ (A) is the periodic cyclic cohomology of A and HP∗ (A) is the periodic
cyclic homology of A.
7) The top row is the analytic index map. It computes the Fredholm index of
a Fredholm module twisted by a K-theory class.
8) The bottom row is the natural pairing between cyclic cohomology and ho-
mology. Once composed with vertical arrows it gives the topological index maps

[(H, F, γ)] × [e] 7→ hCh0 (H, F, γ), Ch0 (e)i,

[(H, F )] × [u] 7→ hCh1 (H, F ), Ch1 (u)i.

So the commutativity of the diagram amounts to the following equality:

Topological Index = Analytic Index (2)

Notice that the Atiyah–Singer index theorem amounts to an equality of the above
type, where in this classical case A = C ∞ (M ) is a commutative algebra.
We can summarize early the earliest development of NCG as a way of extending
(2) beyond its classical realm of manifolds and differential operators on them, to
a noncommutative world. The following ingredients were needed:
1. What is a noncommutative space and how to construct one? A major source
of noncommutative spaces is noncommutative quotients, replacing bad quotients
by groupoid algebras.
2. Cohomological apparatus. This includes noncommutative analogues of topo-
logical invariants such as K-theory, K-homology, de Rham cohomology, and Chern
character maps. Cyclic cohomology and the theory of characteristic classes in non-
commutative geometry as we discuss later in these lectures.
The diagram (1) should be seen as the prototype of a series of results in non-
commutative geometry that aims at expressing the analytic index by a topological
formula. In the next step it would be desirable to have a local expression for the
topological index, that is, for the Connes–Chern character Chi . The local index
formula of Connes and Moscovici [25] solves this problem by replacing the char-
acteristic classes Chi (H, F ) by a cohomologous cyclic cocycle Chi (H, D). Here
D is an unbounded operator that defines a refinement of the notion of Fredholm
module to that of a spectral triple, and F is the phase of D. We won’t discuss this
aspect of noncommutative index theory in these lectures.
Lecture 2: Hochschild
cohomology

Roughly speaking, Hochschild homology is the noncommutative analogue of differ-

ential forms on a manifold. This is made precise by Hochschild-Kostant-Rosenberg-
Connes theorem. This theory was introduced by Hochschild to classify square zero
extensions of an associative algebra. Group homology and Lie algebra homology
are examples of Hochschild homology.

0.8 Hochschild cohomology

Let A be a unital algebra over C and M be an A-bimodule. Thus M is a left and
a right A-module and the two actions are compatible in the sense that a(mb) =
(am)b for all a, b in A and m in M . The Hochschild cochain complex of A with
coefficients in M ,

δ δ δ
C 0 (A, M ) −
→ C 1 (A, M ) −
→ C 2 (A, M ) −
→ ··· (3)

denoted (C ∗ (A, M ), δ), is defined by

C 0 (A, M ) = M, C n (A, M ) = Hom(A⊗n , M ), n ≥ 1,

where the differential δ : C n (A, M ) → C n+1 (A, M ) is given by

(δm)(a) = ma − am,
(δf )(a1 , . . . , an+1 ) = a1 f (a2 , . . . , an+1 )
Xn
+ (−1)i+1 f (a1 , . . . , ai ai+1 , . . . , an+1 )
i=1
+ (−1)n+1 f (a1 , . . . , an )an+1 .

Here m ∈ M = C 0 (A, M ), and f ∈ C n (A, M ), n ≥ 1.

One checks that
δ 2 = 0.

15
16 Introduction

The cohomology of the complex (C ∗ (A, M ), δ) is by definition the Hochschild

cohomology of the algebra A with coefficients in the A-bimodule M and will be
denoted by H n (A, M ), n = 0, 1, 2, . . . .
Among all bimodules over an algebra A, the following two bimodules play an
important role.
1) M = A, with bimodule structure a(b)c = abc for all a, b, c in A. In
this case the Hochschild complex C ∗ (A, A) is also known as the deformation or
Gerstenhaber complex of A. It plays an important role in deformation theory of
associative algebras pioneered by Gerstenhaber [37]. For example, it can be shown
that H 1 (A, A) is the space of derivations from A → A modulo inner derivations,
H 2 (A, A) is the space of equivalence classes of infinitesimal deformations of A and
H 3 (A, A) is the space of obstructions for deformations of A (cf. Section 3.3).
2) M = A∗ := Hom(A, C), the linear dual of A, with A-bimodule structure
defined by
(af b)(c) = f (bca)
for all a, b, c in A and f in A∗ . This bimodule is relevant to cyclic cohomology.
Indeed, as we shall see later in this chapter, the Hochschild groups H n (A, A∗ ) and
the cyclic cohomology groups HC n (A) enter into a long exact sequence. Using the
identification

Hom(A⊗n , A∗ ) ' Hom(A⊗(n+1) , C), f 7→ ϕ,

ϕ(a0 , a1 , . . . , an ) = f (a1 , . . . , an )(a0 ),

the Hochschild differential δ is transformed into a differential, denoted b, given by
n
X
(bϕ)(a0 , . . . , an+1 ) = (−1)i ϕ(a0 , . . . , ai ai+1 , . . . , an+1 )
i=0
+ (−1)n+1 ϕ(an+1 a0 , a1 , . . . , an ).

Thus for n = 0, 1, 2 we have the following formulas for b:

(bϕ)(a0 , a1 ) = ϕ(a0 a1 ) − ϕ(a1 a0 ),

(bϕ)(a0 , a1 , a2 ) = ϕ(a0 a1 , a2 ) − ϕ(a0 , a1 a2 ) + ϕ(a2 a0 , a1 ),
(bϕ)(a0 , a1 , a2 , a3 ) = ϕ(a0 a1 , a2 , a3 ) − ϕ(a0 , a1 a2 , a3 )
+ ϕ(a0 , a1 , a2 a3 ) − ϕ(a3 a0 , a1 , a2 ).

From now on the Hochschild complex C ∗ (A, A∗ ) will be simply denoted by

C (A) and the Hochschild cohomology H ∗ (A, A∗ ) by HH ∗ (A).
∗

Example 0.8.1. We give a few examples of Hochschild cohomology, starting in

low dimensions.
1. n = 0. It is clear that

H 0 (A, M ) = {m ∈ M ; ma = am for all a ∈ A}.

 0.8 Hochschild cohomology 17

In particular for M = A∗ ,

H 0 (A, A∗ ) = {f : A → C; f (ab) = f (ba) for all a, b ∈ A}

is the space of traces on A.

2. n = 1. A Hochschild 1-cocycle f ∈ C 1 (A, M ) is simply a derivation, i.e., a
C-linear map f : A → M such that

f (ab) = af (b) + f (a)b

for all a, b in A. A 1-cocycle is a coboundary if and only if the corresponding

derivation is inner, that is there should exists an m in M such that f (a) = ma−am
for all a in A. Therefore
derivations
H 1 (A, M ) = .
inner derivations
Sometimes this is called the space of outer derivations of A with values in the A-
bimodule M . In view of Exercise 0.8.6, for an algebra A, commutative or not, we
can think of Der(A, A) as the Lie algebra of noncommutative vector fields on the
noncommutative space represented by A. Notice that, unless A is commutative,
Der(A, A) need not be an A-module.
3. n = 2. One can show that H 2 (A, M ) classifies abelian extensions of A by
M . Let A be a unital algebra and M be an A-bimodule. By definition, an abelian
extension of A by M is an exact sequence of algebras

0→M →B→A→0

such that B is unital, M has trivial multiplication (i.e., M 2 = 0), and the induced
A-bimodule structure on M coincides with the original bimodule structure. Two
such extensions (M, B, A) and (M, B 0 , A) are called isomorphic if there is a unital
algebra map f : B → B 0 which induces identity maps on M and A. (Notice that if
such an f exists then it is necessarily an isomorphism.) Let E(A, M ) denote the
set of isomorphism classes of such extensions. We define a natural bijection

E(A, M ) ' H 2 (A, M )

as follows. Given an extension as above, let s : A → B be a linear splitting for the

projection B → A, and let f : A ⊗ A → M be its curvature, defined by

f (a, b) = s(ab) − s(a)s(b)

for all a, b in A. One can easily check that f is a Hochschild 2-cocycle and its
class is independent of the choice of the splitting s. In the other direction, given
a 2-cochain f : A ⊗ A → M , we try to define a multiplication on B = A ⊕ M via

(a, m)(a0 , m0 ) = (aa0 , am0 + ma0 + f (a, a0 )).

18 Introduction

It can be checked that this defines an associative multiplication if and only if f is

a 2-cocycle. The extension associated to a 2-cocycle f is the extension

0 → M → A ⊕ M → A → 0.

It can be checked that these two maps are bijective and inverse to each other.
4. A simple computation shows that when A = C is the ground field we have

HH 0 (C) = C and HH n (C) = 0 for n ≥ 1.

Example 0.8.2. Let M be a closed (i.e., compact without boundary), smooth,

oriented, n-dimensional manifold and let A = C ∞ (M ) denote the algebra of com-
plex valued smooth functions on M . For f 0 , . . . , f n ∈ C ∞ (M ), let
Z
0 n
ϕ(f , . . . , f ) = f 0 df 1 . . . df n .
M

The (n + 1)-linear cochain ϕ : A⊗(n+1) → C has three properties: it is continuous

with respect to the natural Fréchet space topology of A (cf. Section 0.11 for more
on this point); it is a Hochschild cocycle; and it is a cyclic cochain (cf. Section 0.12
for more on this). The Hochschild cocycle property that concerns us here, bϕ = 0,
can be checked as follows:
n
X
(bϕ)(f 0 , . . . , f n+1 ) := (−1)i ϕ(f 0 , . . . , f i f i+1 , . . . , f n+1 )
i=0
+ (−1)n+1 ϕ(f n+1 f 0 , . . . , . . . , f n )
n
X Z
i
= (−1) f 0 df 1 . . . d(f i f i+1 ) . . . df n+1
i=0 M
Z
+ (−1)n+1 f n+1 f 0 df 1 . . . df n
M
=0

for all f 0 , . . . , f n+1 ∈ A. Here we used the Leibniz rule for the de Rham differen-
tial d and the graded commutativity of the algebra (Ω∗ M, d) of differential forms
on M .
We have thus associated a Hochschild cocycle to the orientation cycle of the
manifold. This construction admits a vast generalization, as we explain now. Let

Ωp M := Homcont (Ωp M, C) (4)

denote the continuous linear dual of the space of p-forms on M . Here, the (locally
convex) topology of Ωp M is defined by the sequence of seminorms

kωkn = sup |∂ α ωi1 ,...,ip |, |α| ≤ n,

 0.8 Hochschild cohomology 19

where the supremum is over a fixed, finite, coordinate cover for M , and over all
partial derivatives ∂ α of total degree at most n of all components ωi1 ,...,ip of ω.
Elements of Ωp M are called de Rham p-currents on M . For p = 0 we recover the
notion of a distribution on M . Since the de Rham differential d : Ωk M → Ωk+1 M ,
k = 0, 1, . . . , is continuous in the topology of differential forms, by dualizing it we
obtain differentials d∗ : Ωk M → Ωk−1 M , k = 1, 2, . . . and the de Rham complex
of currents on M :
d∗ d∗ d∗
Ω0 M ←−− Ω1 M ←−− Ω2 M ←−− · · · .
The homology of this complex is called the de Rham homology of M and we shall
denote it by HndR (M ), n = 0, 1, . . . .
It is easy to check that for any m-current C, closed or not, the cochain ϕC
defined by
ϕC (f 0 , f 1 , . . . , f m ) := hC, f 0 df 1 . . . df m i
is a Hochschild cocycle on A. As we shall explain in Section 3.4, ϕC is continuous
in the natural topology of A⊗(m+1) and we obtain a canonical map

m
Ωm M → HHcont (C ∞ (M ))

from the space of m-currents on M to the continuous Hochschild cohomology of

C ∞ (M ). By a theorem of Connes [19] this map is an isomorphism. We refer to
Section 3.5 for more details and a dual statement relating differential forms with
Hochschild homology. The corresponding statement for the algebra of regular
functions on a smooth affine variety, the Hochschild–Kostant–Rosenberg theorem,
will be discussed in that section as well.
d
Exercise 0.8.1. Let A1 = C[x, dx ] denote the Weyl algebra of differential opera-
tors with polynomial coefficients, where the product is defined as the composition
of operators. Equivalently, A1 is the unital universal algebra generated by ele-
d d d
ments x and dx with relation dx x − x dx = 1. Show that HH 0 (A1 ) = 0; that is,
A1 carries no nonzero trace.
d
Exercise 0.8.2. Show that any derivation of the Weyl algebra A1 = C [x, dx ] is
inner, i.e., H 1 (A1 , A1 ) = 0.

Exercise 0.8.3. Show that any derivation of the algebra C(X) of continuous
functions on a compact Hausdorff space X is zero. (Hint: If f = g 2 and g(x) = 0
for some x ∈ X then, for any derivation δ, (δf )(x) = 0.)

Exercise 0.8.4. Show that any derivation of the matrix algebra Mn (C) is inner.
(This was proved by Dirac in his first paper on quantum mechanics [34], where
derivations are called quantum differentials).

Exercise 0.8.5. Let Z(A) denote the center of the algebra A. Show that the
Hochschild groups H n (A, M ) are Z(A)-modules.
20 Introduction

Exercise 0.8.6 (Derivations and vector fields). Let U ⊂ Rn be an open set and
let
n
X ∂
X= Xi
i=1
∂x i

be a smooth vector field on U . Define a derivation δX : C ∞ (U ) → C ∞ (U ) by

n
X ∂f
δX (f ) = Xi .
i=1
∂xi

Show that the map X 7→ δX defines a 1-1 correspondence between vector fields
on U and derivations of C ∞ (U ) to itself. Under this isomorphism, the bracket of
vector fields corresponds to commutators of derivations:

δ[X, Y ] = [δX , δY ].

Fix a point m ∈ U and define an A-module structure on C by the map f ⊗ 1 7→

f (m). Show that the set Der(C ∞ (U ), C) of C-valued derivations of C ∞ (U ) is
canonically isomorphic to the (complexified) tangent space of U at m. Extend
these correspondences to arbitrary smooth manifolds. (These considerations form
the beginnings of a purely algebraic approach to some ‘soft’ aspects of differential
geometry including differential forms and tensor analysis, connection and curva-
ture formalism and Chern–Weil theory and is part of ‘differential geometry over
commutative algebras’. It can also be adapted to algebraic geometry.)

0.9 Hochschild cohomology as a derived functor

The original complex (3) that we used to define the Hochschild cohomology is
rarely useful for computations. Instead, the fact that Hochschild cohomology is a
derived functor will allows us, in specific cases, to replace the standard complex (3)
by a much smaller complex and to compute the Hochschild cohomology. In this
section we show that Hochschild cohomology is a derived functor; more precisely
it is an Ext functor. References for the general theory of derived functors and
homological algebra include [14], and [56].
Let Aop denote the opposite algebra of an algebra A. Thus, as a vector space
op
A = A and the new multiplication is defined by a · b := ba. There is a one-to-one
correspondence between A-bimodules and left A ⊗ Aop -modules defined by

(a ⊗ bop )m = amb.

Define a functor from the category of left A ⊗ Aop -modules to the category of
complex vector spaces by

M 7→ HomA⊗Aop (A, M ) = {m ∈ M ; ma = am for all a ∈ A} = H 0 (A, M ).

 0.9 Hochschild cohomology as a derived functor 21

We show that Hochschild cohomology is the left derived functor of the functor
M ; H 0 (A, M ). We assume that A is unital. Since A is naturally a left A ⊗ Aop -
module, we can consider its bar resolution. It is defined by
b0 b0 b0
0 ← A ←−− B1 (A) ←−− B2 (A) ←−− · · · , (5)
where Bn (A) = A ⊗ Aop ⊗ A⊗n is the free left A ⊗ Aop -module generated by A⊗n .
The differential b0 is defined by
b0 (a ⊗ b ⊗ a1 ⊗ · · · ⊗ an ) = aa1 ⊗ b ⊗ a2 ⊗ · · · ⊗ an
n−1
X
+ (−1)i (a ⊗ b ⊗ a1 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an )
i=1
+ (−1)n (a ⊗ an b ⊗ a1 ⊗ · · · ⊗ an−1 ).
Define the operators s : Bn (A) → Bn+1 (A), n ≥ 0, by
s(a ⊗ b ⊗ a1 ⊗ · · · ⊗ an ) = 1 ⊗ b ⊗ a ⊗ a1 ⊗ · · · ⊗ an .
One checks that
b0 s + sb0 = id,
which shows that (B(A), b0 ) is acyclic and hence is a free resolution of A as a left
A ⊗ Aop -module. Now, for any A-bimodule M we have an isomorphism of cochain
complexes
HomA⊗Aop (B(A), M ) ' (C ∗ (A, M ), δ),
which shows that Hochschild cohomology is the left derived functor of the Hom
functor:
H n (A, M ) ' ExtnA⊗Aop (A, M ) for all n ≥ 0.
One can therefore use any projective resolution of A, or any injective resolution of
M , as a left A ⊗ Aop -module to compute the Hochschild cohomology groups.
Before proceeding further let us recall the definition of the Hochschild homology
of an algebra A with coefficients in a bimodule M . The Hochschild homology
complex of A with coefficients in M is the complex
δ δ δ
C0 (A, M ) ←− C1 (A, M ) ←− C2 (A, M ) ←− · · · (6)
denoted by (C∗ (A, M ), δ), where
C0 (A, M ) = M and Cn (A, M ) = M ⊗ A⊗n , n = 1, 2, . . . ,
and the Hochschild boundary δ : Cn (A, M ) → Cn−1 (A, M ) is defined by
δ(m ⊗ a1 ⊗ · · · ⊗ an ) = ma1 ⊗ a2 ⊗ · · · ⊗ an
n−1
X
+ (−1)i m ⊗ a1 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an
i=1
+ (−1)n an m ⊗ a1 ⊗ · · · ⊗ an .
22 Introduction

The Hochschild homology of A with coefficients in M is, by definition, the homol-

ogy of the complex (C∗ (A, M ), δ). We denote this homology by Hn (A, M ), n =
0, 1, . . . . It is clear that
H0 (A, M ) = M/[A, M ],
where [A, M ] is the C-linear subspace of M spanned by commutators am − ma
for a in A and m in M .
The following facts are easily established:
1) Hochschild homology, H∗ (A, M ), is the right derived functor of the functor
M ; A ⊗A⊗Aop M = H0 (A, M ) from the category of left A ⊗ Aop -modules to the
category of complex vector spaces, i.e.,
op
Hn (A, M ) ' TorA⊗A
n (A, M ).

For the proof one can simply use the bar resolution (5) as we did for cohomology.
2) (Duality) Let M ∗ = Hom(M, C). It is an A-bimodule via (af b)(m) =
f (bma). One checks that the natural isomorphism

Hom(A⊗n , M ∗ ) ' Hom(M ⊗ A⊗n , C), n = 0, 1, . . .

is compatible with differentials. Thus, since we are over a field of characteristic 0,

we have natural isomorphisms

H n (A, M ∗ ) ' (Hn (A, M ))∗ , n = 0, 1, . . .

From now on the Hochschild homology groups H∗ (A, A) will be denoted by HH∗ (A).
In view of the above duality, we have the isomorphisms

HH n (A) ' HHn (A)∗ , n ≥ 0,

where by our earlier convention HH n (A) stands for H n (A, A∗ ).

Example 0.9.1. Let A = C[x] be the algebra of polynomials in one variable. It is
easy to check that the following complex is a resolution of A as a left A⊗A-module:
ε d
0←
− C[x] ←− C[x] ⊗ C[x] ←− C[x] ⊗ C[x] ⊗ C ←
− 0, (7)

where the differentials are the unique A ⊗ A-linear extensions of the maps

ε(1 ⊗ 1) = 1, d(1 ⊗ 1 ⊗ 1) = x ⊗ 1 − 1 ⊗ x. (8)

To check its acyclicity, notice that it is isomorphic to the complex

ε d
0←
− C[x] ←− C[x, y] ←− C[x, y] ←
− 0,

where now
ε(f (x, y)) = f (x, x), d(f (x, y)) = (x − y)f (x, y).
 0.9 Hochschild cohomology as a derived functor 23

By tensoring this resolution with the right A ⊗ A-module A, we obtain a com-

plex with zero differentials
0
0←
− C[x] ←− C[x] ←
−0

and hence (
C[x] if i = 0, 1,
HHi (C[x]) '
0 if i ≥ 2.
The complex (7) is a simple example of a Koszul resolution. In the next example
we generalize it to polynomials in several variables.
Example 0.9.2. Let A = C[x1 , . . . , xn ] be the algebra of polynomials in n vari-
ables. Let V be an n-dimensional complex vector space over. The Koszul resolu-
tion of A, as a left A ⊗ A-module, is defined by
ε d
0 ← A ←− A⊗A ←− A⊗A⊗Ω1 ← · · · ← A⊗A⊗Ωi ← · · · ← A⊗A⊗Ωn ← 0, (9)
Vi
where Ωi = V is the i-th exterior power of V . The differentials ε and d are
defined in (8). d has a unique extensionVto a graded derivation of degree −1 on the
graded commutative algebra A ⊗ A ⊗ V . Notice that A ' S(V ), the symmetric
algebra of the vector space V .
Let K(S(V )) denote the Koszul resolution (9). To show that it is exact we
notice that
K(S(V ⊕ W )) ' K(S(V ) ⊗ K(S(W ))).
Since the tensor product of two exact complexes is again exact (notice that we
are over a field of characteristic zero), the exactness of K(S(V )) can be reduced
to the case where V is 1-dimensional, which was treated in the last example. See
Exercise 0.9.6 for an explicit description of the resolution (9).
As in the one dimensional case, the differentials in the complex A ⊗A⊗A
K(S(V )) are all zero and we obtain
S(V )⊗S(V )
HHi (S(V )) = Tori (S(V ), S(V )
Vi
= S(V ) ⊗ V.

The right-hand side is isomorphic to the module of algebraic differential forms on

S(V ). So we can write this result as

HHi (S(V )) = Ωi (S(V )),

which is a special case of the Hochschild–Kostant–Rosenberg theorem mentioned

before. More generally, if M is a symmetric A-bimodule, the differentials of
M ⊗A⊗A K(S(V )) vanish and we obtain
Vi
Hi (S(V ), M ) ' M ⊗ V, i = 0, 1, . . . , n,

and 0 otherwise.
24 Introduction

Example 0.9.3. In Section 0.11 we shall define the continuous analogues of

Hochschild and cyclic (co)homology as well as Tor and Ext functors. Here is a
simple example. The continuous analogue of the resolution (7) for the topological
algebra A = C ∞ (S 1 ) is the topological Koszul resolution
ε d
− C ∞ (S 1 ) ←− C ∞ (S 1 ) ⊗
0← ˆ C ∞ (S 1 ) ←− C ∞ (S 1 ) ⊗
ˆ C ∞ (S 1 ) ⊗ C ←
− 0, (10)

with differentials given by (8). Here ⊗ ˆ denotes the projective tensor product of
locally convex spaces (cf. Section 0.11 for definitions). To verify the exactness, the
only non-trivial step is to check that ker ε ⊂ im d. To this end, notice that if we
identify
C ∞ (S 1 ) ⊗
ˆ C ∞ (S 1 ) ' C ∞ (S 1 × S 1 ),
the differentials are given by

(εf )(x) = f (x, x), (d1 f )(x, y) = (x − y)f (x, y).

Now the homotopy formula

Z 1
∂
f (x, y) = f (x, x) − (x − y) f (x, y + t(x − y)) dt
0 ∂y
shows that ker ε ⊂ im d. Alternatively, one can use Fourier series to establish the
exactness (cf. Exercise 0.9.5).
To compute the continuous Tor functor, we apply the functor − ⊗ ˆ A⊗A
ˆ to the
above complex. We obtain
0
− C ∞ (S 1 ) ←− C ∞ (S 1 ) ←
0← −0

and hence (
Ωi S 1 if i = 0, 1,
HHicont (C ∞ (S 1 )) =
0 if i ≥ 2,
where Ωi S 1 ' C ∞ (S 1 )dxi is the space of differential forms of degree i on S 1 .
A similar computation, using a continuous version of Ext by applying the
functor Homcontˆ (−, A) gives
A⊗A
(
i ∞ 1 Ωi S 1 if i = 0, 1,
HHcont (C (S )) =
0 if i ≥ 2.

Here Ωi S 1 = (Ωi S 1 )∗ , the continuous dual of i-forms, is the space of i-currents

on S 1 .
Notice how the identification C ∞ (S 1 ) ⊗
ˆ C ∞ (S 1 ) ' C ∞ (S 1 × S 1 ) played an
important role in the above proof. The algebraic tensor product C ∞ (S 1 )⊗C ∞ (S 1 ),
on the other hand, is only dense in C ∞ (S 1 × S 1 ) and this makes it very difficult
to write a resolution to compute the algebraic Hochschild groups of C ∞ (S 1 ). In
fact these groups are not known so far!
 0.9 Hochschild cohomology as a derived functor 25

Example 0.9.4 (Cup product). Let A and B be unital algebras. What is the
relation between the Hochschild homology groups of A ⊗ B and those of A and
B? One can construct (cf. [14], [56] for details) chain maps

C∗ (A ⊗ B) → C∗ (A) ⊗ C∗ (B),
C∗ (A) ⊗ C∗ (B) → C∗ (A ⊗ B)

inducing inverse isomorphisms. We obtain

L
HHn (A ⊗ B) ' HHp (A) ⊗ HHq (B) for all n ≥ 0.
p+q=n

Now, if A is commutative, the multiplication m : A ⊗ A → A is an algebra map

and, in combination with the above map, induces an associative and graded com-
mutative product on HH∗ (A).
Exercise 0.9.1. Let A and B be unital algebras. Give a direct proof of the
isomorphism
HH0 (A ⊗ B) ' HH0 (A) ⊗ HH0 (B).
Dually, show that there is a natural map

HH 0 (A) ⊗ HH 0 (B) → HH 0 (A ⊗ B),

but it need not be surjective in general.

Exercise 0.9.2. Let

A = T (V ) = C ⊕ V ⊕ V ⊗2 ⊕ · · · ,

be the tensor algebra of a vector space V . Show that the complex

ε d
− A ←− A ⊗ Aop ←− A ⊗ Aop ⊗ V ←
0← − 0,

with differentials induced by

ε(1 ⊗ 1) = 1, d(1 ⊗ 1 ⊗ v) = v ⊗ 1 − 1 ⊗ v, v ∈ V,

is a free resolution of A as a left A ⊗ Aop -module. Conclude that A has Hochschild

homological dimension 1 in the sense that Hn (A, M ) = 0 for all A-bimodules M
and all n ≥ 2. Compute H0 (A, M ) and H1 (A, M ).
Exercise 0.9.3 (Normalization). Let M be an A-bimodule. A cochain f : A⊗n →
M is called normalized if f (a1 , . . . , an ) = 0 whenever ai = 1 for some i. Show that
∗
normalized cochains Cnorm (A, M ) form a subcomplex of the Hochschild complex
∗
C (A, M ) and that the inclusion
∗
Cnorm (A, M ) ,→ C ∗ (A, M )

is a quasi-isomorphism. (Hint: Introduce a normalized version of the bar resolu-

tion.)
26 Introduction

Exercise 0.9.4. Let A = C[x]/(x2 ) denote the algebra of dual numbers. Use the
normalized Hochschild complex to compute HH∗ (A).
Exercise 0.9.5. Use Fourier series to show that the sequence (10) is exact.
Exercise 0.9.6. Let V be an n-dimensional vector space. Show that the following
complex is a free resolution of S(V ), the symmetric algebra of V , as a left S(V ) ⊗
S(V )-module
ε i i i i
S(V ) ←− S(V 2 ) ←−X−− S(V 2 )⊗E1 ←−X−− S(V 2 )⊗E2 ←−X−− · · · ←−X−− S(V 2 )⊗En ← 0,
Vk
where Ek = V , and iX is the interior multiplication (contraction) with respect
to the vector field
n
X ∂
X= (xi − yi )
i=1
∂y i

on V 2 = V × V . (Hint: Use the Cartan homotopy formula diX + iX d = LX to

find a contracting homotopy for iX .)
Exercise 0.9.7 (A resolution for the algebraic noncommutative torus [19]). Let

A = ChU1 , U2 i/(U1 U2 − λU2 U1 )

be the universal unital algebra generated by invertible elements U1 and U2 with

Vi U1 U2 = λU2 U1 . We assume that λ ∈ C is not a root of unity. Let
relation
Ωi = V , where V is a 2-dimensional vector space with basis e1 and e2 . Consider
the complex of left A ⊗ Aop -modules
ε d d
− A ←− A ⊗ Aop ←−−
0← 0
A ⊗ Aop ⊗ Ω1 ←−−
1
A ⊗ Aop ⊗ Ω2 ←
− 0, (11)

where ε is the multiplication map and the other differentials are defined by

d0 (1 ⊗ 1 ⊗ ej ) = 1 ⊗ Uj − Uj ⊗ 1, j = 1, 2,

d1 (1 ⊗ 1 ⊗ e1 ∧ e2 ) = (U2 ⊗ 1 − λ ⊗ U2 ) ⊗ e1 − (λU1 ⊗ 1 − 1 ⊗ U1 ) ⊗ e2 .
Show that (11) is a resolution of A as an A ⊗ Aop -module and use it to compute
HH∗ (A).
Exercise 0.9.8. The Weyl algebra A1 is defined in Exercise 0.8.1. By giving a
‘small’ resolution of length two for A1 as a left A1 ⊗ Aop
1 -module show that
(
C if i = 2,
HHi (A1 ) '
0 if i 6= 2.

Show that HH2 (A1 ) is generated by the class of the 2-cycle

1 ⊗ p ⊗ q − 1 ⊗ q ⊗ p + 1 ⊗ 1 ⊗ 1,
 0.10 Deformation theory 27

d
where q = x and p = dx . Extend this result to higher order Weyl algebras
⊗n
An = A1 and show that
(
C if i = 2n,
HHi (An ) '
0 if i 6= 2n.

Can you give an explicit formula for the generator of HH2n (An )?

0.10 Deformation theory

Let A be a unital complex algebra. An increasing filtration on A is an increasing
sequence
S i of subspaces of A, F i (A) ⊂ F i+1 (A), i = 0, 1, 2, . . . , with 1 ∈ F 0 (A),
i F (A) = A, and

F i (A)F j (A) ⊂ F i+j (A) for all i, j.

Let F −1 (A) = 0. The associated graded algebra of a filtered algebra is the graded
algebra
L F i (A)
Gr(A) = F i−1 (A) .
i≥0

Definition 0.10.1. An almost commutative algebra is a filtered algebra whose

associated graded algebra Gr(A) is commutative.
Being almost commutative is equivalent to the commutator condition

[F i (A), F j (A)] ⊂ F i+j−1 (A) (12)

for all i, j. As we shall see, Weyl algebras and, more generally, algebras of dif-
ferential operators on a smooth manifold, and universal enveloping algebras are
examples of almost commutative algebras.
Let A be an almost commutative algebra. The original Lie algebra bracket
[x, y] = xy − yx on A induces a Lie algebra bracket { } on Gr(A) via the formula

{x + F i , y + F j } := [x, y] + F i+j−2 .

Notice that by the almost commutativity assumption (12), [x, y] is in F i+j−1 (A)
and Gr(A), with its grading shifted by one, is indeed a graded Lie algebra. The
induced Lie bracket on Gr(A) is compatible with its multiplication in the sense
that for all a ∈ Gr(A), the map b 7→ {a, b} is a derivation. The algebra Gr(A)
is called the semiclassical limit of the almost commutative algebra A. It is an
example of a Poisson algebra as we recall later in this section.
Notice that as vector spaces, Gr(A) and A are linearly isomorphic, but their
algebra structures are different as Gr(A) is always commutative but A need not
be commutative. A linear isomorphism

q : Gr(A) → A
28 Introduction

can be regarded as a ‘naive quantization map’. Of course, linear isomorphisms

always exist but they are hardly interesting. One usually demands more. For
example one wants q to be a Lie algebra map in the sense that

q{a, b} = [q(a), q(b)] (13)

for all a, b in Gr(A). This is one form of Dirac’s quantization rule, going back to
Dirac’s paper [34]. One normally thinks of A as the algebra of quantum observables
of a system acting as operators on a Hilbert space, and of Gr(A) as the algebra
of classical observables of functions on the phase space. No-go theorems, e.g. the
celebrated Groenewold–Van Hove Theorem (cf. [40] for discussions and precise
statements; see also Exercise 0.10.2), states that, under reasonable irreducibility
conditions, this is almost never possible. The remedy is to have q defined only for
a special class of elements of Gr(A), or satisfy (13) only in an asymptotic sense
as Planck’s constant h goes to zero. As we shall discuss later in this section, this
can be done in different ways, for example in the context of formal deformation
quantization [53].
The notion of a Poisson algebra captures the structure of semiclassical limits.

Definition 0.10.2. Let P be a commutative algebra. A Poisson structure on P

is a Lie algebra bracket (a, b) 7→ {a, b} on A such that for any a ∈ A, the map
b 7→ {a, b}: A → A is a derivation of A. That is, for all b, c in A we have

{a, bc} = {a, b}c + b{a, c}.

In geometric examples (see below) the vector field defined by the derivation
b 7→ {a, b} is called the Hamiltonian vector field of the Hamiltonian function a.

Definition 0.10.3. A Poisson algebra is a pair (P, { , }) where P is a commutative

algebra and { , } is a Poisson structure on P .

We saw that the semiclassical limit P = Gr(A) of any almost commutative

algebra A is a Poisson algebra. Conversely, given a Poisson algebra P one may
ask if it is the semiclassical limit of an almost commutative algebra. This is one
form of the problem of quantization of Poisson algebras, the answer to which for
general Poisson algebras is negative.

Example 0.10.1. A Poisson manifold is a manifold M whose algebra of smooth

functions A = C ∞ (M ) is a Poisson algebra (we should also assume that the bracket
{ , } is continuous in the Fréchet topology of A, or, equivalently, is a bidifferential
operator). It is not difficult to see that all Poisson structures on A are of the form

{f, g} := hdf ∧ dg, πi,

V2
where π ∈ C ∞ ( (T M )) is a smooth 2-vector field on M . This bracket clearly
satisfies the Leibniz rule in each variable and one checks that it satisfies the Jacobi
 0.10 Deformation theory 29

V3
identity if and only if [π, π] = 0, where the Schouten bracket [π, π] ∈ C ∞ ( (T M ))
is defined in local coordinates by
n  
X ∂πik ∂πkj ∂πji
[π, π]ijk = πlj + πli + πlk .
∂xl ∂xl ∂xl
l=1

The Poisson bracket in local coordinates is given by

X ∂f ∂g
{f, g} = πij .
ij
∂xi ∂xj

Symplectic manifolds are the simplest examples of Poisson manifolds. They cor-
respond to non-degenerate Poisson structures. Recall that a symplectic form on a
manifold is a non-degenerate closed 2-form on the manifold. Given a symplectic
form ω, the associated Poisson bracket is given by

{f, g} = ω(Xf , Xg ),

where the vector field Xf is the symplectic dual of df and is defined by requiring
that the equation df (Y ) = ω(Xf , Y ) holds for all smooth vector fields Y on M .
∞
Let Cpoly (T ∗ M ) be the algebra of smooth functions on T ∗ M which are polyno-
mial in the cotangent direction. It is a Poisson algebra under the natural symplectic
structure of T ∗ M . This Poisson algebra is the semiclassical limit of the algebra of
differential operators on M , as we shall see in the next example.
Example 0.10.2 (Differential operators on commutative algebras). Let A be a
commutative unital algebra. We define an algebra D(A) ⊂ EndC (A) inductively
as follows. Let
D0 (A) = A = EndA (A) ⊂ EndC (A)
denote the set of differential operators of order zero on A, i.e., A-linear maps from
A → A. Assuming Dk (A) has been defined for 0 ≤ k < n, we let Dn (A) be the set
of all operators D in EndC (A) such that for any a ∈ A, [D, a] ∈ Dn−1 (A). The
set [
D(A) = Dn (A)
n≥0

is a subalgebra of EndC (A), called the algebra of differential operators on A. It is an

almost commutative algebra under the filtration given by subspaces Dn (A), n ≥ 0.
Elements of Dn (A) are called differential operators of order n. For example, a
linear map D : A → A is a differential operator of order one if and only if it is of
the form D = δ + a, where δ is a derivation on A and a ∈ A.
For general A, the semiclassical limit Gr(D(A)) and its Poisson structure are
not easily identified except for coordinate rings of smooth affine varieties or alge-
bras of smooth functions on a manifold. In this case a differential operator D of
order k is locally given by X
D= aI (x)∂ I ,
|I|≤k
30 Introduction

where I = (i1 , . . . , in ) is a multi-index, ∂ I = ∂i1 ∂i2 . . . ∂in is a mixed partial

derivative, and n is the dimension of the manifold. This expression depends on
the local coordinates but its leading terms of total degree n have an invariant
meaning provided that we replace ∂i with ξi ∈ T ∗ M . For ξ ∈ Tx∗ M , let
X
σp (D)(x, ξ) := aI (x)ξ I .
|I|=k

Then the function σp (D) : T ∗ M → C, called the principal symbol of D, is invari-

∞
antly defined and belongs to Cpoly (T ∗ M ). The algebra Cpoly
∞
(T ∗ M ) inherits a
canonical Poisson structure as a subalgebra of the Poisson algebra C ∞ (T ∗ M ) and
we have the following
Proposition 0.10.1. The principal symbol map induces an isomorphism of Pois-
son algebras
σp : Gr D(C ∞ (M )) −−
∼ ∞
→ Cpoly (T ∗ M ).
Example 0.10.3 (Weyl algebra). Let A1 := D C[X] be the Weyl algebra of dif-
ferential operators on the line. Alternatively, A1 can be described as the unital
complex algebra defined by generators x and p with

px − xp = 1.
d
The map x 7→ x, p 7→ dx defines the isomorphism. Physicists prefer to write the
h
defining relation as the canonical commutation relation pq − qp = 2πi 1, where h is
Planck’s constant and p and q represent momentum and position operators. This
is not without merit because we can then let h → 0 and obtain the commutative
algebra of polynomials in p and q as the semiclassical limit. Also, i is necessary
if we want to consider p and q as selfadjoint operators (why?). Then one can use
h d
the representation q 7→ x, p 7→ 2πi dx .
Any element of A1 has a unique expression as a differential operator with
di
P
polynomial coefficients ai (x) dx i where the standard filtration is by degree of

the differential operator. The principal symbol map

n
X di 
σp ai (x) = an (x)y n .
i=0
dxi

defines an algebra isomorphism Gr(A1 ) ' C[x, y]. The induced Poisson bracket
on C[x, y] is the classical Poisson bracket
∂f ∂g ∂f ∂g
{f, g} = − .
∂x ∂y ∂y ∂x
In principle, the Weyl algebra An is the algebra of differential operators on
C[x1 , . . . , xn ]. Alternatively, it can be defined as the universal algebra defined by
2n generators x1 , . . . , xn , p1 , . . . , pn with

[pi , xi ] = δij and [pi , pj ] = [xi , xj ] = 0

 0.10 Deformation theory 31

for all i, j. Notice that An ' A1 ⊗ · · · ⊗ A1 (n factors). A lot is known about

Weyl algebras and a lot remains to be known, including the Dixmier conjecture
about the automorphisms of An . The Hochschild and cyclic cohomology of An are
computed in [36] (cf. also [56]).
Example 0.10.4 (Universal enveloping algebras). Let U (g) denote the enveloping
algebra of a Lie algebra g. By definition, U (g) is the quotient of the tensor algebra
T (g) by the two-sided ideal generated by x ⊗ y − y ⊗ x − [x, y] for all x, y ∈ g. For
p ≥ 0, let F p (U (g)) be the subspace generated by tensors of degree at most p.
This turns U (g) into a filtered algebra and the Poincaré–Birkhoff–Witt theorem
asserts that its associated graded algebra is canonically isomorphic to the sym-
metric algebra S(g). The algebra isomorphism is induced by the symmetrization
map s : S(g) → Gr(U (g)), defined by
1 X
s(X1 X2 . . . Xp ) = Xσ(1) . . . Xσ(p) .
p!
σ∈Sp

Note that S(g) is the algebra of polynomial functions on the dual space g∗ , which
is a Poisson manifold under the bracket

{f, g}(X) = [Df (X), Dg(X)]

for all f, g ∈ C ∞ (g∗ ) and X ∈ g∗ . Here we have used the canonical isomorphism
g ' g∗∗ , to regard the differential Df (X) ∈ g∗∗ as an element of g. The induced
Poisson structure on polynomial functions coincides with the Poisson structure in
Gr(U (g)).
Example 0.10.5 (Algebra of formal pseudodifferential operators on the circle).
d
This algebra is obtained by formally inverting the differentiation operator ∂ := dx
and then completing the resulting algebra.
Pn A formal pseudodifferential operator on
the circle is an expression of the form −∞ ai (x)∂ i , where each ai (x) is a Laurent
polynomial. The multiplication is uniquely defined by the rules ∂x − x∂ = 1 and
∂∂ −1 = ∂ −1 ∂ = 1. We denote the resulting algebra by Ψ1 . The Adler–Manin
trace on Ψ1 [58], also called the noncommutative residue, is defined by
n Z
X
i
 1
Tr ai (x)∂ = Res(a−1 (x); 0) = a−1 (x).
−∞
2πi S 1

This is a trace on Ψ1 . In fact one can show that Ψ1 /[Ψ1 , Ψ1 ] is 1-dimensional

which means that any trace on Ψ1 is a multiple of Tr. Notice that for the Weyl
algebra A1 we have [A1 , A1 ] = A1 .
Another interesting difference between Ψ1 and A1 is that Ψ1 admits non-inner
derivations (see exercise below). The algebra Ψ1 has a nice generalization to
algebras of pseudodifferential operators in higher dimensions. The appropriate
extension of the above trace is the noncommutative residue of Wodzicki (cf. [68];
see also [21] for relations with the Dixmier trace and its role in noncommutative
Riemannian geometry).
32 Introduction

So far in this section we saw at least one way to formalize the idea of quanti-
zation through the notion of an almost commutative algebra and its semiclassical
limit which is a Poisson algebra. A closely related notion is formal deformation
quantization, or star products, going back to [5], [62]. It is also closely related to
the theory of deformations of associative algebras developed originally by Gersten-
haber, as we recall now.
Let A be an algebra, which may be noncommutative, over C, and let A[[h]]
be the algebra of formal power series over A. A formal deformation of A is an
associative C[[h]]-linear multiplication

∗h : A[[h]] ⊗ A[[h]] → A[[h]]

such that ∗0 is the original multiplication. Writing

a ∗h b = B0 (a, b) + hB1 (a, b) + h2 B2 (a, b) + · · · ,

where Bi : A ⊗ A → A are Hochschild 2-cochains on A with values in A, we see

that the initial value condition on ∗h is equivalent to B0 (a, b) = ab for all a, b ∈ A.
Let us define a bracket { , } on A by

{a, b} = B1 (a, b) − B1 (b, a)

or, equivalently, but more suggestively, by

a ∗h b − b ∗h a
{a, b} := Lim as h → 0.
h
Using the associativity of the star product a ∗h (b ∗h c) = (a ∗h b) ∗h c, it is easy to
check that B1 is a Hochschild 2-cocycle for the Hochschild cohomology of A with
coefficients in A, i.e., it satisfies the relation

aB1 (b, c) − B1 (ab, c) + B1 (a, bc) − B1 (a, b)c = 0

for all a, b, c in A. Clearly the bracket { , } satisfies the Jacobi identity. In short,
(A, { , }) is an example of what is sometimes called a noncommutative Poisson
algebra. If A is a commutative algebra, then it is easy to see that it is indeed a
Poisson algebra in the sense of Definition 0.10.3.
The bracket { , } can be regarded as the infinitesimal direction of the deforma-
tion, and the deformation problem for a commutative Poisson algebra amounts to
finding higher order terms Bi , i ≥ 2, given B0 and B1 .
The associativity condition on ∗h is equivalent to an infinite system of equations
involving the cochains Bi that we derive now. They are given by

B0 ◦ Bn + B1 ◦ Bn−1 + · · · + Bn ◦ B0 = 0 for all n ≥ 0,

or, equivalently,
n−1
X
Bi ◦ Bn−i = δBn . (14)
i=1
 0.10 Deformation theory 33

Here, the Gerstenhaber ◦-product of 2-cochains f, g : A ⊗ A → A is defined as the

3-cochain
f ◦ g(a, b, c) = f (g(a, b), c) − f (a, g(b, c)).
Notice that a 2-cochain f defines an associative product if and only if f ◦ f = 0.
Also notice that the Hochschild coboundary δf can be written as

δf = −m ◦ f − f ◦ m,

where m : A ⊗ A → A is the multiplication of A. These observations lead to the

associativity equations (14).
To solve these equations starting with B0 = m, by antisymmetrizing we can
always assume that B1 is antisymmetric and hence we can assume B1 = 21 { , }.
AssumePnB0 , B1 , . . . , Bn have been found so that (14) holds. Then one can show
that i=1 Bi ◦ Bn−i is a cocycle. Thus we can find a Bn+1 satisfying (14) if
and only if this cocycle is a coboundary, i.e., its class in H 3 (A, A) should vanish.
The upshot is that the third Hochschild cohomology H 3 (A, A) is the space of
obstructions for the deformation quantization problem. In particular if H 3 (A, A) =
0 then any Poisson bracket on A can be deformed. Notice, however, that this is
only a sufficient condition and is by no means necessary, as will be shown below.
In the most interesting examples, e.g. for A = C ∞ (M ), H 3 (A, A) 6= 0. To see
this consider the differential graded Lie algebra (C ∗ (A, A), [ , ], δ) of continuous
HochschildVcochains on A, and the differential graded Lie algebra, with zero dif-
ferential, ( (T M ), [ , ], 0) of polyvector fields on M . The bracket in the first is
the Gerstenhaber bracket and in the second is the Schouten bracket of polyvector
fields. By a theorem of Connes (see the resolution in Lemma 44 in [19]), we know
that the antisymmetrization map

α : C∞ T M ), 0 → (C ∗ (A, A), δ
V

sending a polyvector field X1 ∧ · · · ∧ Xk to the functional ϕ defined by

ϕ(f 1 , . . . , f k ) = df 1 (X1 )df 2 (X2 ) . . . df k (Xk )

is a quasi-isomorphism of differential graded algebras. In particular, it induces an

isomorphism of graded commutative algebras
L k L ∞ Vk

H (A, A) ' C TM .
k k

The map α, however, is not a morphism of Lie algebras and that is where
the real difficulty of deforming a Poisson structure is hidden. The formality
theorem of M. Kontsevich [53] states that as a differential graded Lie algebra,
(C ∗ (A, A), δ, [ , ]) is formal in the sense that it is quasi-isomorphic to its cohomol-
ogy. Equivalently, it means that one can perturb the map α, by adding an infinite
number of terms, to a morphism of L∞ -algebras. This shows that the V original
deformation problem of Poisson structures can be transferred to (C ∞ ( T M ), 0)
34 Introduction

where it is unobstructed since the differential in the latter DGL is zero. Later in
this section we shall give a couple of simple examples where deformations can be
explicitly constructed.
There is a much deeper structure hidden in the deformation complex of an
associative (C ∗ (A, A), δ) than first meets the eye and we can only barely scratch
the surface here. The first piece of structure is the cup product. Let C ∗ =
C ∗ (A, A). The cup product ^ : C p × C q → C p+q is defined by

(f ^ g)(a1 , . . . , ap+q ) = f (a1 , . . . , ap )g(ap+1 , . . . , ap+q ).

Notice that ^ is associative and one checks that this product is compatible with
the differential δ and hence induces an associative graded product on H ∗ (A, A).
What is not so obvious however is that this product is graded commutative for
any algebra A [37].
The second piece of structure on (C ∗ (A, A), δ) is a graded Lie bracket. It is
based on the Gerstenhaber circle product ◦ : C p × C q → C p+q−1 defined by

(f ◦ g)(a1 , . . . , ap+q−1 )
p−1
X
= (−1)|g|(|f |+i−1) f (a1 , . . . , g(ai , . . . , ai+p ), . . . , ap+q−1 ).
i=1

Notice that ◦ is not an associative product. Nevertheless one can show that [37]
the corresponding graded bracket [ , ] : C p × C q → C p+q−1

[f, g] = f ◦ g − (−1)(p−1)(q−1) g ◦ f

defines a graded Lie algebra structure on the deformation cohomology H ∗ (A, A).
Notice that the Lie algebra grading is now shifted by one.
What is most interesting is that the cup product and the Lie algebra structure
are compatible in the sense that [ , ] is a graded derivation for the cup product; or
in short, (H ∗ (A, A), ^, [ , ]) is a graded Poisson algebra.
We give a couple of examples where deformations can be explicitly constructed.
Example 0.10.6. The simplest non-trivial Poisson manifold is the dual g∗ of
a finite dimensional Lie algebra g. Let Uh (g) = T (g)/I, where the ideal I is
generated by
x ⊗ y − y ⊗ x − h[x, y], x, y ∈ g.
This is simply the enveloping algebra of the rescaled bracket h[−, −]. By the Poin-
caré–Birkhoff–Witt theorem, the antisymmetrization map αh : S(g) → Uh (g) is a
linear isomorphism. We can define a ∗-product on S(g) by
∞
X
f ∗h g = αh−1 (αh (f )αh (g)) = hn Bn (f, g).
n=0

With some work one can show that the Bn are bidifferential operators and hence
the formula extends to a ∗-product on C ∞ (g∗ ).
 0.10 Deformation theory 35

Example 0.10.7 (Weyl–Moyal quantization). Consider the algebra generated by

x and y with relation xy − yx = hi 1. Let f , g be polynomials in x and y. Iterated
application of the Leibniz rule gives the formula for the product
∞  n
X 1 −ih
f ∗h g = Bn (f, g),
n=0
n! 2

where B0 (f, g) = f g, B1 (f, g) = {f, g} is the standard Poisson bracket, and for
n ≥ 2,
n  
X n
Bn (f, g) = (−1)n (−1)k (∂xk ∂yn−k f )(∂xn−k ∂yk g).
k
k=0

Notice that this formula makes sense for f, g ∈ C ∞ (R2 ) and defines a deformation
of this algebra with its standard Poisson structure. This can be extended to
arbitrary constant Poisson structures on R2 ,
X
{f, g} = π ij ∂i f ∂j g.

The Weyl–Moyal ∗ product is then given by

 
h X ij
f ∗h g = exp − i π ∂i ∧ ∂j (f, g).
2

Exercise 0.10.1. Show that the Weyl algebra A1 is a simple algebra, i.e., it has
no non-trivial two-sided ideals; prove the same for An . In a previous exercise we
asked to show that any derivation of A1 is inner. Is it true that any automorphism
of A1 is inner?

Exercise 0.10.2. In Example 0.10.3 show that there is no linear map q : C[x, y] →
A1 such that q(1) = 1 and q{f, g} = [q(f ), q(g)] for all f and g. This is an
important special case of the Groenewold–van Hove no-go theorem ([40]).

Exercise 0.10.3. Let A = C[x]/(x2 ) be the algebra of dual numbers. It is a

non-smooth algebra. Describe its algebra of differential operators.

Exercise 0.10.4. Unlike the algebra of differential operators, the algebra of

pseudodifferential operators Ψ1 admits non-inner derivations. Clearly log ∂ :=
P∞ n
− 1 (1−∂) n ∈/ Ψ1 , but show that for any a ∈ Ψ1 , we have [ log ∂, a] ∈ Ψ1 and
therefore the map
a 7→ δ(a) := [ log ∂, a]
defines a non-inner derivation of Ψ1 . The corresponding Lie algebra 2-cocycle

ϕ(a, b) = Tr(a[ log ∂, b])

is the Radul cocycle [54].

36 Introduction

0.11 Topological algebras

For applications of Hochschild and cyclic cohomology to noncommutative geome-
try, it is crucial to consider topological algebras, topological bimodules, topological
resolutions, and continuous cochains and chains. For example, while the alge-
braic Hochschild groups of the algebra of smooth functions on a smooth manifold
are not known, and perhaps are hopeless to compute, its continuous Hochschild
(co)homology as a topological algebra can be computed as we recall in Exam-
ple 0.9.3 below. We shall give only a brief outline of the definitions and refer the
reader to [19], [21] for more details. A good reference for locally convex topological
vector spaces and topological tensor products is the book of Treves on the subject.
There is no difficulty in defining continuous analogues of Hochschild and cyclic
cohomology groups for Banach algebras. One simply replaces bimodules by Ba-
nach bimodules, that is a bimodule which is also a Banach space where the left and
right module actions are bounded operators, and cochains by continuous cochains.
Since the multiplication of a Banach algebra is a continuous operation, all oper-
ators including the Hochschild boundary and the cyclic operators extend to this
continuous setting. The resulting Hochschild and cyclic theory for C ∗ -algebras,
however, is hardly useful and tends to vanish in many interesting examples. This
is hardly surprising since the definition of any Hochschild or cyclic cocycle of di-
mension bigger than zero involves differentiating the elements of the algebra in
one way or another. (See Exercise 0.8.3 or, more generally, the Remark below.)
This is in sharp contrast with topological K-theory where the right setting, e.g.
for Bott periodicity to hold, is the setting of Banach or C ∗ -algebras.

Remark 3. By combining results of Connes [15] and Haagerup [41], we know

that a C ∗ -algebra is amenable if and only if it is nuclear. Amenability refers to
the property that for all n ≥ 1,
n
Hcont (A, M ∗ ) = 0

for an arbitrary Banach dual bimodule M ∗ . In particular, for any nuclear C ∗ -

algebra
n
HHcont n
(A) = Hcont (A, A∗ ) = 0 for all n ≥ 1. Using Connes’ long exact sequence
(see Section 0.14), we obtain, for any nuclear C ∗ -algebra A, the vanishing results
2n
HCcont (A) = A∗ 2n+1
and HCcont (A) = 0

for all n ≥ 0. Nuclear C ∗ -algebras form a large class which includes commutative
algebras, the algebra of compact operators, and reduced group C ∗ -algebras of
amenable groups [8].

The right class of topological algebras for Hochschild and cyclic cohomology
turns out to be the class of locally convex algebras [19]. An algebra A which is
simultaneously a locally convex topological vector space is called a locally convex
algebra if its multiplication map A ⊗ A → A is (jointly) continuous. That is, for
 0.11 Topological algebras 37

any continuous seminorm p on A there is a continuous seminorm p0 on A such that

p(ab) ≤ p0 (ab) for all a, b in A.
We should mention that there are topological algebras with a locally convex
topology for which the multiplication map is only separately continuous. But
we do not dwell on this more general class in this book as they appear rarely
in applications. This distinction between separate and joint continuity of the
multiplication map disappears for the class of Fréchet algebras. By definition, a
locally convex algebra is called a Fréchet algebra if its topology is metrizable and
complete. Many examples of ‘smooth noncommutative spaces’ that one encounters
in noncommutative geometry are in fact Fréchet algebras.
Example 0.11.1. Basic examples of Fréchet algebras include the algebra A =
C ∞ (M ) of smooth functions on a closed smooth manifold and the smooth non-
commutative tori Aθ and their higher dimensional analogues. We start with a
simple down to earth example where A = C ∞ (S 1 ). We consider the elements of
A as smooth periodic functions of period one on the line. Its topology is defined
by the sequence of norms

pn (f ) = sup kf (k) k∞ , 0 ≤ k ≤ n,

where f (k) is the k-th derivative of f and k k∞ is the sup norm. We can equivalently
use the sequence of norms
n
X 1 (k)
qn (f ) = kf k∞ .
i=0
k!

Notice that qn ’s are submultiplicative, that is qn (f g) ≤ qn (f )qn (g). Locally convex

algebras whose topology is induced by a family of submultiplicative seminorms are
known to be projective limits of Banach algebras. This is the case in all examples
in this section.
In general, the topology of C ∞ (M ) is defined by the sequence of seminorms

kf kn = sup |∂ α f |, |α| ≤ n,

where the supremum is over a fixed, finite, coordinate cover for M . The Leibniz
rule for derivatives of products shows that the multiplication map is indeed jointly
continuous. See the Exercise 0.11.1 for the topology of Aθ .
Given locally convex topological vector spaces V1 and V2 , their projective ten-
sor product is a locally convex space V1 ⊗ ˆ V2 together with a universal jointly
ˆ
continuous bilinear map V1 ⊗ V2 → V1 ⊗ V2 (cf. [39]). That is, for any locally con-
vex space W , we have a natural isomorphism between jointly continuous bilinear
maps V1 × V2 → W and continuous linear maps V1 ⊗ ˆ V2 → W . Explicitly, the
ˆ
topology of V1 ⊗ V2 is defined by the family of seminorms p ⊗ ˆ q, where p, q are
continuous seminoms on V1 and V2 respectively, and
 Pn Pn
p⊗ˆ q(t) := inf i p(ai )q(bi ); t = i ai ⊗ bi , ai ∈ V1 , bi ∈ V2 .
38 Introduction

Then V1 ⊗ ˆ V2 is defined as the completion of V1 ⊗ V2 under the above family of

seminorms.
One of the nice properties of the topology of C ∞ (M ) is that it is nuclear (see
[39]). In particular for any other smooth compact manifold N , the natural map

C ∞ (M ) ⊗
ˆ C ∞ (N ) → C ∞ (M × N )

is an isomorphism of topological algebras. This plays an important role in com-

puting the continuous Hochschild cohomology of C ∞ (M ).
Let A be a locally convex topological algebra. A topological left A-module is
a locally convex vector space M endowed with a continuous left A-module action
A⊗ ˆ M → M . A topological free left A-module is a module of the type M = A ⊗ ˆ V
where V is a locally convex space. A topological projective module is a topological
module which is a direct summand in a free topological module.
Given a locally convex algebra A, let
ˆ
n
Ccont (A) = Homcont (A⊗n , C)

be the space of continuous (n + 1)-linear functionals on A. All the algebraic defi-

nitions and results of this chapter extend to this topological setting. In particular
one defines the continuous Hochschild and cyclic cohomology groups of a locally
convex algebra. One must be careful, however, in dealing with resolutions. The
right class of topological projective (in particular free) resolutions are those res-
olutions that admit a continuous linear splitting. This extra condition is needed
when one wants to prove comparison theorems for resolutions and, eventually, in-
dependence of cohomology from resolutions. We shall not go into details here since
this is very well explained in [19].
Exercise 0.11.1. The sequence of norms

pk (a) = Sup {(1 + |n| + |m|)k |amn |}

m,n

defines a locally convex topology on the smooth noncommutative torus Aθ . Show

that the multiplication of Aθ is continuous in this topology.

0.12 Hochschild (co)homology computations

We give a few examples of Hochschild (co)homology computations. In particular
we shall see that group (co)homology and Lie algebra (co)homology are instances of
Hochschild (co)homology. We start by recalling the classical results of Hochschild–
Kostant–Rosenberg [47] and Connes [19] which identifies the Hochschild homol-
ogy of smooth commutative algebras with the algebra of differential forms. By a
smooth commutative algebra we mean either the topological algebra A = C ∞ (M )
of smooth complex-valued functions on a closed smooth manifold M , or the al-
gebra A = O[X] of regular function on a smooth affine algebraic variety X. We
start with the latter case.
 0.12 Hochschild (co)homology computations 39

Example 0.12.1 (Smooth commutative algebras). Algebras of regular functions

on a smooth affine variety can be characterized abstractly through various equiv-
alent conditions (cf. Proposition 3.4.2 in [56]). For example, one knows that a
finitely generated commutative algebra A is smooth if and only if it has the lifting
property with respect to nilpotent extensions. More precisely, A is smooth if and
only if for any pair (C, I), where C is a commutative algebra and I is an ideal such
that I 2 = 0, the map

Homalg (A, C) → Homalg (A, C/I)

is surjective. Examples of smooth algebras include polynomial algebras C[x1 , . . . , xn ],

algebras of Laurent polynomials C[x1 , x−1 −1
1 , . . . , xn , xn ], and coordinate rings of
affine algebraic groups. The algebra C[x, y]/(xy) is not smooth.
We recall the definition of the algebraic de Rham complex of a commutative, not
necessarily smooth, algebra A. The module of 1-forms, or Kähler differentials, over
A, denoted by Ω1A , is by definition a left A-module Ω1A endowed with a universal
derivation
d : A → Ω1A .
This means that any other derivation δ : A → M into a left A-module M uniquely
factorizes through d. One usually defines Ω1A := I/I 2 where I is the kernel of the
multiplication map A ⊗ A → A. Note that since A is commutative this map is
an algebra homomorphism and I is an ideal. The left multiplication defines a left
A-module structure on Ω1A . The derivation d is defined by

d(a) = a ⊗ 1 − 1 ⊗ a mod (I 2 ).

Checking its universal property is straightforward. One defines the space of n-

forms on A as the n-th exterior power of the A-module Ω1A :
Vn
ΩnA := A Ω1A ,

where the exterior product is over A. There is a unique extension of d to a graded

derivation of degree one,
d : Ω∗A → Ω∗+1
A .

It satisfies d2 = 0. The algebraic de Rham cohomology of A is the cohomology of

the complex (Ω∗A , d). For some examples of this construction see exercises at the
end of this section.
Let us compare the complex of differential forms with trivial differential (Ω∗A , 0),
with the Hochschild complex of A with coefficients in A, (C∗ (A), b). Consider the
antisymmetrization map

εn : ΩnA → A⊗(n+1) , n = 0, 1, 2, . . . ,
X
εn (a0 da1 ∧ · · · ∧ dan ) = sgn(σ)a0 ⊗ aσ(1) ⊗ · · · ⊗ aσ(n) ,
σ∈Sn
40 Introduction

where Sn denotes the symmetric group on n letters. We also have maps

µn : A⊗(n+1) → ΩnA , n = 0, 1, . . . ,

µn (a0 ⊗ a1 ⊗ · · · ⊗ an ) = a0 da1 ∧ · · · ∧ dan .

One checks that both maps are morphisms of complexes, i.e.,

b ◦ εn = 0 and µn ◦ b = 0.

Moreover, one has

µn ◦ εn = n! id.
Assuming the ground field has characteristic zero, it follows that the antisym-
metrization map induces an inclusion

ΩnA ,→ HHn (A), n = 0, 1, 2, . . . . (15)

In particular, for any commutative algebra A over a field of characteristic zero, the
space of differential n-forms on A is always a direct summand of the Hochschild
homology group HHn (A).
This map, however, need not be surjective in general (cf. Exercises below).
This has to do with the singularity of the underlying geometric space represented
by A. The Hochschild–Kostant–Rosenberg theorem [47] states that if A is the
algebra of regular functions on a smooth affine variety, then the above map is an
isomorphism. Notice that we have verified this fact for polynomial algebras in
Example 0.9.2.

Example 0.12.2 (Algebras of smooth functions). This is a continuation of Ex-

ample 0.8.2. Let M be a smooth closed manifold and A = C ∞ (M ) the algebra of
smooth complex-valued functions on M . It is a locally convex (in fact, Fréchet)
topological algebra as we explained in Section 0.11. Let Ωn M (resp. Ωn M ) denote
the space of n-forms (resp. n-currents) on M . Consider the map
n
Ωn M → Ccont (C ∞ (M )), C 7→ ϕC ,

where
ϕC (f0 , f1 , . . . , fn ) := hC, f0 df1 . . . dfn i.
It is easily checked that this map defines a morphism of complexes
∗
(Ω∗ M, 0) → (Ccont (C ∞ (M )), b).

In [19], using an explicit resolution, Connes shows that the induced map on coho-
mologies is an isomorphism. Thus we have a natural isomorphism between space
of de Rham currents on M and (continuous) Hochschild cohomology of C ∞ (M ):

i
Ωi M ' HHcont (C ∞ (M )), i = 0, 1, . . . (16)
 0.12 Hochschild (co)homology computations 41

Without Vgoing into details, we shall briefly indicate the resolution introduced
k ∗
in [19]. Let TC M denote the bundle of complexified k-forms on M and Ek be
its pullback under the projection pr2 : M × M → M . Let X be a vector field on
M 2 = M × M such that in a neighborhood of the diagonal ∆(M ) ⊂ M × M and
in a local geodesic coordinate system (x1 , . . . , xn , y1 , . . . , yn ) it looks like
n
X ∂
X= (xi − yi ) .
i=1
∂y i

We assume that away from the diagonal X is nowhere zero. Such an X can always
be found, provided that M admits a nowhere zero vector field. The latter condition
is clearly equivalent to vanishing of the Euler characteristic of M . By replacing
M by M × S 1 the general case can be reduced to this special case.
The following is then shown to be a continuous projective resolution of C ∞ (M )
as a C ∞ (M × M )-module [19]:
ε i
C ∞ (M ) ←− C ∞ (M 2 ) ←−X−− C ∞ (M 2 , E1 )
i i i
←−X−− C ∞ (M 2 , E2 ) ←−X−− · · · ←−X−− C ∞ (M 2 , En ) ←
− 0,
where iX is interior multiplication by X. By applying the Hom functor HomA⊗A (−, A∗ )
we obtain a complex with zero differentials
0 0 0 0
Ω0 M −−→ Ω1 M −−→ · · · −−→ Ωn M −−→ 0,
and hence the isomorphism (16).
The analogous result for Hochschild homology uses the map
C ∞ (M ) ⊗ ˆ C ∞ (M ) → Ωn M
ˆ ··· ⊗

defined by
f0 ⊗ · · · ⊗ fn 7→ f0 df1 . . . dfn .
It is easy to check that this defines a morphism of complexes
C∗cont (C ∞ (M ), b) → (Ω∗ M, 0).
Using the above resolution and by essentially the same argument, one shows that
the induced map on homologies is an isomorphism between continuous Hochschild
homology of C ∞ (M ) and differential forms on M :

HHicont (C ∞ (M )) ' Ωi M, i = 0, 1, . . .

Example 0.12.3 (Group algebras). It is clear from the original definitions that
group (co)homology is an example of Hochschild (co)homology. Let G be a group
and M be a left G-module. The standard complex for computing the cohomology
of G with coefficients in M is the complex (cf. [14], [56])
δ δ δ
M −−→ C 1 (G, M ) −−→ C 2 (G, M ) −−→ · · · ,
42 Introduction

where
C n (G, M ) = {f : Gn → M },
and the differential δ is defined by

(δm)(g) = gm − m,

n
X
δf (g1 , . . . , gn+1 ) = g1 f (g2 , . . . , gn+1 ) + (−1)i f (g1 , . . . , gi gi+1 , . . . , gn+1 )
i=1
+ (−1)n+1 f (g1 , g2 , . . . , gn ).

Let A = CG denote the group algebra of G over the complex numbers. Then
M is a CG-bimodule via the actions

g · m = g(m), m·g =m

for all g in G and m in M . It is clear that for all n,

C n (CG, M ) ' C n (G, M ),

and the isomorphism preserves the differentials. It follows that the group coho-
mology of G with coefficients in M coincides with the Hochschild cohomology of
CG with coefficients in M :

H n (CG, M ) ' H n (G, M ).

Conversely, it is easy to see that the Hochschild cohomology of CG with coef-

ficients in a bimodule M reduces to group cohomology. Let M ad = M as a vector
space and define a left G-action on M ad by

g · m = gmg −1 .

Define a linear isomorphism i : C n (G, M ad ) → C n (CG, M ) by

(if )(g1 , . . . , gn ) = f (g1 , . . . , gn )g1 g2 . . . gn .

It can be checked that i commutes with differentials and hence is an isomorphism

of complexes (MacLane isomorphism)

C ∗ (G, M ad ) −−
∼
→ C ∗ (CG, M ).

Of course, there is a similar result for homology.

Of particular importance is an understanding of HH∗ (CG) = H∗ (CG, CG) =
H∗ (G, CG), i.e., when M = CG and G is acting by conjugation. By a theorem of
Burghelea [11], the Hochschild and cyclic homology groups of CG decompose over
the set of conjugacy classes of G where each summand is the group homology (with
trivial coefficients) of a group associated to the conjugacy class. We recall this
 0.12 Hochschild (co)homology computations 43

result for Hochschild homology here and later we shall discuss the corresponding
result for cyclic homology.
The whole idea can be traced back to the following simple observation. Let
τ : CG → C be a trace on the group algebra. It is clear that τ is constant on each
conjugacy class of G and, conversely, the characteristic function of each conjugacy
class defines a trace on CG. Thus we have
Y
HH 0 (CG) = C,
hGi

where hGi denotes the set of conjugacy classes of G. Dually, for homology we have
L
HH0 (CG) = C.
hGi

We focus on homology and shall extend the above observation to higher dimen-
sions. Dual cohomological versions are straightforward.
Clearly we have
(CG)⊗(n+1) = CGn+1 .
For each conjugacy class c ∈ hGi, let Bn (G, c) be the linear span of all (n + 1)-
tuples (g0 , g1 , . . . , gn ) ∈ Gn+1 such that

g0 g1 . . . gn ∈ c.

It is clear that B∗ (G, c) is invariant under the Hochschild differential b. We there-

fore have a decomposition of the Hochschild complex of CG into subcomplexes
indexed by conjugacy classes:
L
C∗ (CG, CG) = B∗ (G, c).
c∈hGi

Identifying the homology of the component corresponding to the conjugacy

class of the identity is rather easy. For other components one must work harder.
Let c = {e} denote the conjugacy class of the identity element of G. The map
(g0 , g1 , . . . , gn ) 7→ (g1 , g2 , . . . , gn ) is easily seen to define an isomorphism of vector
spaces
Bn (G, {e}) ' CGn .
Moreover, under this map the Hochschild differential b goes over to the differential
for the group homology of G with trivial coefficients. It follows that

H∗ (B(G, {e})) ' H∗ (G, C).

Next we describe the Hochschild homology of other components. For an ele-

ment g ∈ G, let
Cg = {h ∈ G; hg = gh}
44 Introduction

denote the centralizer of g in G. Notice that the isomorphism class of this

group depends only on the conjugacy class of g. One checks that the inclusion
i : Cn (Cg , C) → Bn (G, c) defined by

i(g1 , . . . , gn ) = ((g1 g2 . . . gn )−1 g, g1 , . . . , gn−1 )

is a chain map. One can in fact show, by an explicit chain homotopy, that i is a
quasi-isomorphism. It therefore follows that, for each conjugacy class c and each
g ∈ c, we have
H∗ (B(G, c)) = H∗ (Cg , C).
Putting everything together this shows that the Hochschild homology of CG
decomposes as a direct sum of group homologies of centralizers of conjugacy classes
of G, a result due to Burghelea [11] (cf. also [56], for purely algebraic proofs):
L
HH∗ (CG) ' H∗ (Cg ) (17)
hGi

The corresponding dual statement for Hochschild cohomology reads as

Y
HH ∗ (CG) ' H ∗ (Cg ).
hGi

Example 0.12.4 (Enveloping algebras). We show that Lie algebra (co)homology

is an example of Hochschild (co)homology, a result which goes back to Cartan–
Eilenberg [14]. Let g be a Lie algebra and M be a (left) g-module. This simply
means that we have a Lie algebra morphism

g → EndC (M ).

The Lie algebra homology of g with coefficients in M is, by definition, the homology
of the Chevalley–Eilenberg complex
δ V1 δ V2 δ V3 δ
M ←− M ⊗ g ←− M ⊗ g ←− M ⊗ g ←− · · · ,

where the differential δ is defined by

δ(m ⊗ X) = X(m),

δ(m ⊗ X1 ∧ X2 ∧ · · · ∧ Xn ) =
X
(−1)i+j m ⊗ [Xi , Xj ] ∧ X1 ∧ · · · ∧ X̂i ∧ · · · ∧ X̂j ∧ · · · ∧ Xn
i<j
X
+ (−1)i Xi (m) ⊗ X1 ∧ · · · ∧ X̂i ∧ · · · ∧ Xn .
i

(One checks that δ 2 = 0.)

 0.12 Hochschild (co)homology computations 45

Let U (g) denote the enveloping algebra of g. Given a U (g)-bimodule M , we

define a left g-module M ad , where M ad = M and
X · m := Xm − mX
for all X ∈ g and m ∈ M . Define a map
Vn
ε : M ad ⊗ g → M ⊗ U (g)⊗n
from the Lie algebra complex to the Hochschild complex by
X
ε(m ⊗ X1 ∧ · · · ∧ Xn ) = sgn(σ)m ⊗ Xσ(1) ⊗ · · · ⊗ Xσ(n) .
σ∈Sn

One checks that ε : C Lie (g, M ad ) → C(U (g), M ) is a chain map (prove this!).
We claim that it is indeed a quasi-isomorphism, i.e., it induces an isomorphism
between the corresponding homology groups:
H∗ (g, M ad ) ' H∗ (U (g), M ).
We refer to [14], [56] for its standard proof.
Example 0.12.5 (Morita invariance of Hochschild homology). Let A and B be
unital Morita equivalent algebras. Let X be an equivalence A–B-bimodule and Y
be an inverse B–A-bimodule. Let M be an A-bimodule and N = Y ⊗A M ⊗A X
the corresponding B-bimodule. Morita invariance of Hochschild homology states
that there is a natural isomorphism
Hn (A, M ) ' Hn (B, N )
for all n ≥ 0. A proof of this can be found in [56]. There is a similar result, with a
similar proof, for cohomology. We sketch a proof of this result for the special case
where B = Mk (A) is the algebra of k by k matrices over A. The main idea is to
introduce the generalized trace map.
Let M be an A-bimodule and Mk (M ) be the space of k by k matrices with
coefficients in M . It is a bimodule over Mk (A) in an obvious way. The generalized
trace map is defined by
Tr : Cn (Mk (A), Mk (M )) → Cn (A, M ),

Tr(α0 ⊗ m0 ⊗ α1 ⊗ a1 ⊗ · · · ⊗ αn ⊗ an ) = tr(α0 α1 . . . αn )m0 ⊗ a1 ⊗ · · · ⊗ an ,

where αi ∈ Mk (C), ai ∈ A, m0 ∈ M , and tr : Mk (C) → C is the standard trace of
matrices.
As an exercise the reader should show that Tr is a chain map. Let i : A →
Mk (A) be the map that sends a in A to the matrix with only one nonzero com-
ponent in the upper left corner equal to a. There is a similar map M → Mk (M ).
These induce a map
I : Cn (A, M ) → Cn (Mk (A), Mk (M )),
46 Introduction

I(m ⊗ a1 ⊗ · · · ⊗ an ) = i(m) ⊗ i(a1 ) ⊗ · · · ⊗ i(an ).

We have Tr ◦ I = id, which is easily checked. It is however not true that
I ◦ Tr = id. There is instead a homotopy between I ◦ Tr and id (cf. [56]). It follows
that Tr and I induce inverse isomorphisms between homologies.
As a special case of Morita invariance, by choosing M = A, we obtain an
isomorphism of Hochschild homology groups

HH n (A) ' HH n (Mk (A))

for all n and k.

Example 0.12.6 (Inner derivations and inner automorphisms). We need to know,

for example when defining the noncommutative Chern character later in this chap-
ter, that inner automorphisms act by the identity on Hochschild homology and
inner derivations act by zero. Let A be an algebra, u ∈ A be an invertible element
and let a ∈ A be any element. They induce the chain maps Θ : Cn (A) → Cn (A)
and La : Cn (A) → Cn (A) defined by

Θ(a0 ⊗ · · · ⊗ an ) = ua0 u−1 ⊗ · · · ⊗ uan u−1 ,

and
n
X
La (a0 ⊗ · · · ⊗ an ) = a0 ⊗ · · · ⊗ [a, ai ] ⊗ · · · ⊗ an .
i=0

Lemma 0.12.1. Θ induces the identity map on Hochschild homology and La

induces the zero map.

Proof. The maps hi : A⊗n+1 → A⊗n+2 , i = 0, . . . , n,

hi (a0 ⊗ · · · ⊗ an ) = (a0 u−1 ⊗ ua1 u−1 ⊗ · · · ⊗ u ⊗ ai+1 ⊗ · · · ⊗ an )

define a homotopy
n
X
h= (−1)i hi
i=0

between id and Θ.
For the second part one checks that the maps h0i : A⊗n+1 → A⊗n+2 , i =
0, . . . , n,
hi (a0 ⊗ · · · ⊗ an ) = (a0 ⊗ · · · ⊗ ai ⊗ a ⊗ · · · ⊗ an ),
define a homotopy
n
X
0
h = (−1)i h0i
i=0

between La and 0.
 0.12 Hochschild (co)homology computations 47

Exercise 0.12.1. Let A = S(V ) be the symmetric algebra of a vector space V .

Show that its module of Kähler differentials Ω1S (V ) is isomorphic to S(V ) ⊗ V , the
free left S(V )-module generated by V , where the universal differential is given by
n
X
d(v1 v2 . . . vn ) = (v1 . . . vˆi . . . vn ) ⊗ vi .
i=1

Exercise 0.12.2 (Additivity of HH∗ ). Show that for unital algebras A and B,
there is a natural isomorphism

HHn (A ⊕ B) ' HHn (A) ⊕ HHn (B)

for all n ≥ 0.
Exercise 0.12.3. Show that non-inner automorphisms need not act by the iden-
tity on HH∗ .
48 Introduction
Lecture 3: Cyclic
cohomology

0.13 Cyclic Cohomology

To define a noncommutative de Rham type theory for noncommutative algebras is
a highly non-trivial matter. Note that the usual algebraic formulation of de Rham
theory is based on the module of Kähler differentials and its exterior algebra, which
has no analogue for noncommutative algebras. This is in sharp contrast with the
situation in K-theory where extending the topological K-theory to noncommuta-
tive Banach algebras is straightforward.
Instead, the noncommutative analogue of de Rham homology, called cyclic
cohomology by Connes, was found by a careful analysis of the algebraic structures
deeply hidden in (super )traces of products of commutators. These expressions are
directly defined in terms of an elliptic operator and its parametrix and were shown,
via an index formula, to give the index of the operator when paired with a K-
theory class. This connection with elliptic theory, K-homology, and K-theory, will
be explored later in these lectures.
Cyclic cohomology is defined in [19] through a remarkable subcomplex of the
Hochschild complex. We recall this definition in this section. Later in this chapter
we give two other definitions. While these three definitions are equivalent to each
other, as we shall see each has its own merits and strengths.
Let A be an algebra over the complex numbers and (C ∗ (A), b) denote the
Hochschild complex of A with coefficients in the A-bimodule A∗ . We have, from
Section 3.1,

C n (A) = Hom(A⊗(n+1) , C), n = 0, 1, . . . ,

and
n
X
(bf )(a0 , . . . ,an+1 ) = (−1)i f (a0 , . . . , ai ai+1 , . . . , an+1 )
i=0
+ (−1)n+1 f (an+1 a0 , . . . , an )

for all f ∈ C n (A).

49
50 Introduction

The following definition is fundamental and marks our departure from Hochschild
cohomology:
Definition 0.13.1. An n-cochain f ∈ C n (A) is called cyclic if
f (an , a0 , . . . , an−1 ) = (−1)n f (a0 , a1 , . . . , an )
for all a0 , . . . , an in A. We denote the space of cyclic n-cochains on A by Cλn (A).
Lemma 0.13.1. The space of cyclic cochains is invariant under the action of b,
i.e., for all n we have
b Cλn (A) ⊂ Cλn+1 (A).
Proof. Define the operators λ : C n (A) → C n (A) and b0 : C n (A) → C n+1 (A) by
(λf )(a0 , . . . , an ) = (−1)n f (an , a0 , . . . , an−1 ),
Xn
(b0 f )(a0 , . . . , an+1 ) = (−1)i f (a0 , . . . , ai ai+1 , . . . , an+1 ).
i=0

One checks that

(1 − λ)b = b0 (1 − λ).
Since
Cλn (A) = Ker(1 − λ),
the lemma is proved.
We therefore have a subcomplex of the Hochschild complex, called the cyclic
complex of A:
b b b
Cλ0 (A) −−→ Cλ1 (A) −−→ Cλ2 (A) −−→ · · · (18)

Definition 0.13.2. The cohomology of the cyclic complex is called the cyclic
cohomology of A and will be denoted by HC n (A), n = 0, 1, 2, . . . .
A cocycle for the cyclic cohomology group HC n (A) is called a cyclic n-cocycle
on A. It is an (n + 1)-linear functional f on A which satisfies the two conditions
(1 − λ)f = 0 and bf = 0.
The inclusion of complexes
(Cλ∗ (A), b) ,→ (C ∗ (A), b)
induces a map I from the cyclic cohomology of A to the Hochschild cohomology
of A with coefficients in the A-bimodule A∗ :
I : HC n (A) → HH n (A), n = 0, 1, 2, . . . .
We shall see that this map is part of a long exact sequence relating Hochschild
and cyclic cohomology. For the moment we mention that I need not be injective
or surjective (see example below).
 0.13 Cyclic Cohomology 51

Example 0.13.1. Let A = C, the ground field. We have

Cλ2n (C) ' C, Cλ2n+1 (C) = 0,

so the cyclic complex reduces to

0 → C → 0 → C → ··· .

It follows that for all n ≥ 0,

HC 2n (C) = C, HC 2n+1 (C) = 0.

Since HH n (C) = 0 for n ≥ 1, we conclude that the map I need not be injective
and the cyclic complex is not a retraction of the Hochschild complex.
Example 0.13.2. It is clear that, for any algebra A, HC 0 (A) = HH 0 (A) is the
space of traces on A.
Example 0.13.3. Let A = C ∞ (M ) be the algebra of smooth complex valued
functions on a closed smooth oriented manifold M of dimension n. We check that
Z
ϕ(f 0 , f 1 , . . . , f n ) := f 0 df 1 . . . df n ,
M

is a cyclic n-cocycle on A. We have already checked the cocycle property of ϕ,

bϕ = 0, in Example 0.8.2. The cyclic property of ϕ

ϕ(f n , f 0 , . . . , f n−1 ) = (−1)n ϕ(f 0 , . . . , f n )

is more interesting and is related to Stokes’ formula. In fact since

Z Z
(f n df 0 . . . df n−1 − (−1)n f 0 df 1 . . . df n ) = d(f n f 0 df 1 . . . df n−1 ),
M M

we see that the cyclic property of ϕ follows from a special case of Stokes’ formula:
Z
dω = 0,
M

valid for any (n − 1)-form ω on a closed n-manifold M .

The last example can be generalized in several directions. For example, let V be
an m-dimensional closed singular chain (a cycle) on M , e.g. V can be a closed m-
dimensional submanifold of M . Then integration on V defines an m-dimensional
cyclic cocycle on A:
Z
ϕ(f 0 , f 1 , . . . , f m ) = f 0 df 1 . . . df m .
V

We obtain a map

Hm (M, C) → HC m (C ∞ (M )), m = 0, 1, . . . ,
52 Introduction

from the singular homology of M (or its equivalents) to the cyclic cohomology of
C ∞ (M ).
Let
Ωp M := Homcont (Ωp M, C)
denote the continuous dual of the space of p-forms on M . Elements of Ωp M are
de Rham p-currents on M as defined in Section 3.1. A p-current is called closed if
for any (p − 1)-form ω we have hC, dωi = 0.
It is easy to check that for any m-current C, closed or not, the cochain

ϕC (f 0 , f 1 , . . . , f m ) := hC, f 0 df 1 . . . df m i,

is a Hochschild cocycle on C ∞ (M ). Now if C is closed, then one can easily check

that ϕC is a cyclic m-cocycle on C ∞ (M ). We thus obtain natural maps

Ωm M → HH m (C ∞ (M )) and Zm M → HC m (C ∞ (M )),

where Zm (M ) ⊂ Ωm M is the space of closed m-currents on M .

A noncommutative generalization of this procedure involves the notion of a
cycle on an algebra due to Connes [19] that we recall now. It gives a geometric and
intuitively appealing presentation of cyclic cocycles. It also leads to a definition of
cup product in cyclic cohomology and the S operator, as we shall indicate later.
Let (Ω, d) be a differential graded algebra. Thus

Ω = Ω0 ⊕ Ω1 ⊕ Ω2 ⊕ · · ·

is a graded algebra and d : Ω∗ → Ω∗+1 is a square zero graded derivation in the

sense that

d(ω1 ω2 ) = d(ω1 )ω2 + (−1)deg(ω1 ) ω1 d(ω2 ) and d2 = 0

for all homogenous elements ω1 and ω2 of Ω.

Definition 0.13.3. A closed graded trace of dimension n on a differential graded

algebra (Ω, d) is a linear map Z
: Ωn → C

such that
Z Z
dω = 0 and (ω1 ω2 − (−1)deg(ω1 )deg(ω2 ) ω2 ω1 ) = 0

for all ω in Ωn−1 , ω1 in Ωi , ω2 in Ωj and i + j = n.

R R
Definition 0.13.4. An n-cycle over an algebra A is a triple (Ω, , ρ) where
is an n-dimensional closed graded trace on (Ω, d) and ρ : A → Ω0 is an algebra
homomorphism.
 0.13 Cyclic Cohomology 53

R
Given an n-cycle (Ω, , ρ) over A its character is a cyclic n-cocycle on A defined
by
Z
ϕ(a0 , a1 , . . . , an ) = ρ(a0 )dρ(a1 ) . . . dρ(an ). (19)

Checking the cyclic cocycle conditions bϕ = 0 and (1 − λ)ϕ = 0 is straightforward

but instructive. To simplify the notation we drop the homomorphism ρ and write
ϕ as
Z
ϕ(a0 , a1 , . . . , an ) = a0 da1 . . . dan .
R
We have, using the Leibniz rule for d and the graded trace property of ,

(bϕ)(a0 , . . . , an+1 )
X n Z
= (−1)i a0 da1 . . . d(ai ai+1 ) . . . dan+1
i=0
Z
+ (−1)n+1 an+1 a0 da1 . . . dan
Z Z
n 0 1 n n+1 n+1
= (−1) a da . . . da · a + (−1) an+1 a0 da1 . . . dan

= 0.
R
Notice that we did not need to use the ‘closedness’ of so far. This will be needed
however to check the cyclic property of ϕ:
Z Z
0 n 0 1 n n
(1 − λ)ϕ(a , . . . , a ) = a da . . . da − (−1) an da0 . . . dan
Z
= (−1)n−1 d(an a0 da1 . . . dan−1 )

= 0.

Conversely, one can show that any cyclic cocycle on A is obtained from a cycle
over A via (19). To this end, we introduce the algebra (ΩA, d),

ΩA = Ω0 A ⊕ Ω1 A ⊕ Ω2 A ⊕ · · · ,

called the algebra of noncommutative differential forms on A as follows. ΩA is

the universal (non-unital) differential graded algebra generated by A as a subal-
gebra. We put Ω0 A = A, and let Ωn A be linearly generated over C by expressions
a0 da1 . . . dan and da1 . . . dan for ai ∈ A (cf. [19] for details). Notice that even if
A is unital, ΩA is not a unital algebra and in particular the unit of A is only an
idempotent in ΩA. The differential d is defined by

d(a0 da1 . . . dan ) = da0 da1 . . . dan and d(da1 . . . dan ) = 0.

54 Introduction

The universal property of (ΩA, d) is the fact that for any (not necessarily unital)
differential graded algebra (Ω, d) and any algebra map ρ : A → Ω0 there is a unique
extension of ρ to a morphism of differential graded algebras,
ρ̂ : ΩA → Ω. (20)
Now given a cyclic n-cocycle ϕ on A, define a linear map ϕ : Ωn A → C by
R

Z
(a0 + λ1) da1 . . . dan = ϕ(a0 , . . . , an ).
ϕ
R
It is easy to check that ϕ is a closed graded trace on ΩA whose character is ϕ.
Summarizing, we have shown that the relation
Z
(a0 + λ1) da1 . . . dan = ϕ(a0 , a1 , . . . , an )
ϕ

defines a one-to-one correspondence:

{cyclic n-cocycles on A} ' {closed graded traces on Ωn A} (21)

Notice that for n = 0 we recover the relation in Example 0.13.2 between cyclic
0-cocycles on A and traces on A.
Example 0.13.4 (A 2-cycle on the noncommutative torus). Let δ1 , δ2 : Aθ → Aθ
denote the canonical derivations of the noncommutative torus and τ : Aθ → C its
canonical trace (cf. Example ??). It can be shown by a direct computation that
the 2-cochain ϕ defined on Aθ by
ϕ(a0 , a1 , a2 ) = (2πi)−1 τ (a0 (δ1 (a1 )δ2 (a2 ) − δ2 (a1 )δ1 (a2 )))
is a cyclic
R 2-cocycle. It can also be realizedVas the character of the following 2-cycle
∗
(Ω, d, ) on Aθ as follows. Let Ω = Aθ ⊗ C2 be the tensor product of Aθ with
the exterior algebra of the vector space C2 = Ce1 ⊕ Ce2 . The differential d is
defined by
da = δ1 (a)e1 + δ2 (a)e2 , d(a ⊗ e1 ) = −δ2 (a)e1 ∧ e2 , d(a ⊗ e2 ) = δ1 (a)e1 ∧ e2 .
: Ω2 → C is defined by
R
The closed graded trace
Z
a ⊗ e1 ∧ e2 = (2πi)−1 τ (a).
R
The graded trace property of is a consequence of the trace property of τ and its
closedness follows from the invariance of τ under the infinitesimal automorphisms
δ1 and δ2 , that is, the property τ (δi (a)) = 0 for all a and i = 1, 2. Now it is clear
that the character of this cycle is the cyclic 2-cocycle ϕ defined above:
Z
a0 da1 da2 = ϕ(a0 , a1 , a2 ).
 0.13 Cyclic Cohomology 55

In the remainder of this section we indicate a variety of different sources of

cyclic cocycles, e.g. from group cocycles or Lie algebra cycles.

Example 0.13.5 (From group cocycles to cyclic cocycles). Let G be a discrete

group and A = CG be its group algebra. Let c(g1 , . . . , gn ) be a group n-cocycle
on G. Thus c : Gn → C satisfies the cocycle condition

g1 c(g2 , . . . , gn+1 ) − c(g1 g2 , . . . , gn+1 ) + · · · + (−1)n+1 c(g1 , . . . , gn ) = 0

for all g1 , . . . , gn+1 in G. Assume c is normalized in the sense that

c(g1 , . . . , gn ) = 0

if gi = e for some i, or if g1 g2 . . . gn = e. (It can be shown that any cocycle is

cohomologous to a normalized one). One checks that
(
c(g1 , . . . , gn ), if g0 g1 . . . gn = e,
ϕc (g0 , . . . , gn ) =
0 otherwise,

is a cyclic n-cocycle on the group algebra CG (cf. [18], [21], or exercises below).
In this way one obtains a map from the group cohomology of G to the cyclic
cohomology of CG,

H n (G, C) → HC n (CG), c 7→ ϕc .

By a theorem of Burghelea [11], the cyclic cohomology group HC n (CG) de-

composes over the conjugacy classes of G and the component corresponding to
the conjugacy class of the identity contains the group cohomology H n (G, C) as a
summand. (See Example 0.16.3 in this chapter.)

Example 0.13.6 (From Lie algebra homology to cyclic cohomology). We start

with a simple special case. Let A be an algebra, τ : A → C be a trace, and let
δ : A → A be a derivation on A. We assume that the trace is invariant under the
action of the derivation in the sense that

τ (δ(a)) = 0

for all a ∈ A. Then one checks that

ϕ(a0 , a1 ) := τ (a0 δ(a1 ))

is a cyclic 1-cocycle on A. A simple commutative example of this is when A =

C ∞ (S 1 ), τ corresponds to the Haar measure, and δ = dx
d
. Then one obtains the
fundamental class of the circle
Z
ϕ(f0 , f1 ) = f0 df1 .
56 Introduction

See below for a noncommutative example, with A = Aθ , the smooth noncommu-

tative torus.
This construction can be generalized. Let δ1 , . . . , δn be a commuting family of
derivations on A, and let τ be a trace on A which is invariant under the action of
the δi , i = 1, . . . , n. Then one can check that
X
ϕ(a0 , . . . , an ) := sgn(σ)τ (a0 δσ(1) (a1 ) . . . δσ(n) (an )) (22)
σ∈Sn

is a cyclic n-cocycle on A. Again we give a commutative example and postpone

a noncommutative example to below. Let A = Cc∞ (RnR) be the algebra of smooth
∂
compactly supported functions on Rn . Let τ (f ) = Rn f and δi = ∂x i
. The
corresponding cyclic cocycle is given, using the wedge product, by the formula
Z
ϕ(f0 , . . . , fn ) = f0 df1 ∧ df2 ∧ · · · ∧ dfn ,
Rn
∂f
dxi .
P
where df = i ∂x i
Everything we did so far in this example lends itself to a grand generalization
as follows. Let g be a Lie algebra acting by derivations on an algebra A. This
means that we have a Lie algebra map

g → Der(A, A)

from g to the Lie algebra of derivations of A. Let τ : A → C be a trace which is

invariant under the action of g, i.e.,

τ (X(a)) = 0 for all X ∈ g, a ∈ A.

For each n ≥ 0, define a linear map

Vn
g → C n (A), c 7→ ϕc ,

where X
ϕc (a0 , a1 , . . . , an ) = sgn(σ)τ (a0 Xσ(1) (a1 ) . . . Xσ(n) (an )) (23)
σ∈Sn

if c = X1 ∧ · · · ∧ Xn and extended linearly.

It can be shown that ϕc is a Hochschild cocycle for any c, and that it is a cyclic
cocycle if c is a Lie algebra cycle. (See Exercise 0.13.5.)
We therefore obtain, for each n ≥ 0, a map

χτ : HnLie (g, C) → HC n (A), c 7→ ϕc ,

from the Lie algebra homology of g with trivial coefficients to the cyclic cohomology
of A [18]. Vn
In particular if g is abelian then of course HnLie (g) = (g) and we recover our
previously defined map (22):
Vn
(g) → HC n (A), n = 0, 1, . . . .
 0.13 Cyclic Cohomology 57

Here is an example of this construction which first appeared in [16]. Let A =

Aθ denote the “algebra of smooth functions” on the noncommutative torus. Let
X1 = (1, 0), X2 = (0, 1). There is an action of the abelian Lie algebra R2 on Aθ
defined on generators of Aθ by

X1 (U ) = U, X1 (V ) = 0,
X2 (U ) = 0, X2 (V ) = V.

The induced derivations on Aθ are given by

X  X
X1 am,n U m V n = mam,n U m V n ,
X  X
X2 am,n U m V n = nam,n U m V n .

It is easily checked that the trace τ on Aθ defined by

X 
τ am,n U m V n = a0,0

is invariant under the above action of R2 . The generators of H∗Lie (R2 , C) are: 1,
X1 , X2 , X1 ∧ X2 .
We therefore obtain the following 0-dimensional, 1-dimensional and 2-dimensional
cyclic cocycles on Aθ :

ϕ0 (a0 ) = τ (a0 ), ϕ1 (a0 , a1 ) = τ (a0 X1 (a1 )), ϕ01 (a0 , a1 ) = τ (a0 X2 (a1 )),

ϕ2 (a0 , a1 , a2 ) = τ (a0 (X1 (a1 )X2 (a2 ) − X2 (a1 )X1 (a2 ))).

It is shown in [19] that these classes form a basis for the continuous periodic
cyclic cohomology of Aθ .
Example 0.13.7 (Cup product in cyclic cohomology). As we indicated before,
the notion of cycle over an algebra can be used to give a natural definition of a cup
product for cyclic cohomology. By specializing one of the variables to the ground
field, we obtain the S-operation. R0
Let (Ω, , ρ) be an m-dimensional cycle on an algebra A and (Ω0 , , ρ0 ) an
R

n-dimensional cycle on an algebra B. Let Ω ⊗ Ω0 denote the (graded) tensor

product of the differential graded algebras Ω and Ω0 . By definition, we have

(Ω ⊗ Ω0 )k = Ωi ⊗ Ω0j ,
L
i+j=k

d(ω ⊗ ω 0 ) = (dω) ⊗ ω 0 + (−1)deg(ω) ω ⊗ (dω 0 ).

Let Z 00 Z Z 0
0
ω⊗ω = ω ω0 if deg(ω) = m, deg(ω 0 ) = n.
R 00
It is easily checked that is a closed graded trace of dimension m + n on Ω ⊗ Ω0 .
58 Introduction

Using the universal property (20) of noncommutative differential forms, applied

to the map ρ ⊗ ρ0 : A ⊗ B → Ω0 ⊗ Ω00 , one obtains a morphism of differential graded
algebras
(Ω(A ⊗ B), d) → (Ω ⊗ Ω0 , d).
We therefore obtain a closed graded trace of dimension m + n on (Ω(A ⊗ B), d).
In [19] it is shown that the resulting cup product map in cyclic cohomology,

# : HC m (A) ⊗ HC n (B) → HC m+n (A ⊗ B)

is well defined.
We give a couple of simple examples of cup product computations.
Example 0.13.8 (The generalized trace map). Let ψ be a trace on B. Then
ϕ 7→ ϕ # ψ defines a map

HC m (A) → HC m (A ⊗ B).

Explicitly we have

(ϕ # ψ)(a0 ⊗ b0 , . . . , am ⊗ bm ) = ϕ(a0 , . . . , am )ψ(b0 b1 . . . bm ).

A special case of this construction plays a very important role in cyclic cohomology
and noncommutative geometry. Let ψ = tr : Mn (C) → C be the standard trace.
Then ‘cupping with trace’ defines a map

HC m (A) → HC m (Mk (A)).

Example 0.13.9 (The periodicity operator S). Another important special case
of the cup product is when we choose B = C and ψ to be the fundamental cyclic
2-cocycle on C defined by
ψ(1, 1, 1) = 1.
This leads to an operation of degree 2 on cyclic cohomology:

S : HC n (A) → HC n+2 (A), ϕ 7→ ϕ # ψ.

The formula simplifies to

Z
(Sϕ)(a0 , . . . , an+2 ) = a0 a1 a2 da3 . . . dan+2
ϕ
Z
+ a0 da1 (a2 a3 ) da4 . . . dan+2 + · · ·
ϕ
Z
+ a0 da1 . . . dai−1 (ai ai+1 )dai+2 . . . dan+2 + · · ·
ϕ
Z
+ a0 da1 . . . dan (an+1 an+2 ).
ϕ

In the next section we give a different approach to S via the cyclic bicomplex.
 0.13 Cyclic Cohomology 59

So far we have studied the cyclic cohomology of algebras. There is a ‘dual’

theory called cyclic homology which we introduce now. Let A be an algebra and
for n ≥ 0 let
Cn (A) = A⊗(n+1) .
For each n ≥ 0, define the operators
b : Cn (A) → Cn−1 (A), b0 : Cn (A) → Cn−1 (A), λ : Cn (A) → Cn (A)
by
n−1
X
b(a0 ⊗ · · · ⊗ an ) = (−1)i (a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an )
i=0
+ (−1)n (an a0 ⊗ a1 ⊗ · · · ⊗ an−1 ),
n−1
X
b0 (a0 ⊗ · · · ⊗ an ) = (−1)i (a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an ),
i=0
λ(a0 ⊗ · · · ⊗ an ) = (−1)n (an ⊗ a0 ⊗ · · · ⊗ an−1 ).
The relation
(1 − λ)b0 = b(1 − λ)
can be easily established. Clearly (C∗ (A), b) is the Hochschild complex of A with
coefficients in the A-bimodule A. Let
Cnλ (A) := Cn (A)/ Im(1 − λ).
The relation (1 − λ)b0 = b(1 − λ) shows that the operator b is well defined on the
quotient complex C∗λ (A). The complex
(C∗λ (A), b)
is called the cyclic complex of A and its homology, denoted by HCn (A), n =
0, 1, . . . , is called the cyclic homology of A.
Example 0.13.10. For n = 0,
HC0 (A) ' HH0 (A) ' A/[A, A]
is the commutator quotient of A. Here [A, A] denotes the subspace of A generated
by the commutators ab − ba, for a and b in A. In particular if A is commutative
then HC0 (A) = A.
Exercise 0.13.1. Give a description of Hochschild cocycles on A in terms of linear
functionals on ΩA similar to (21).
Exercise 0.13.2. Let ϕ ∈ HC 0 (A) be a trace on A. Show that
(Sϕ)(a0 , a1 , a2 ) = ϕ(a0 a1 a2 ).
Find an explicit formula for S n ϕ for all n. Let ϕ ∈ HC 1 (A). Express Sϕ in terms
of ϕ.
60 Introduction

Exercise 0.13.3 (Area as a cyclic cocycle). Let f, g : S 1 → R be smooth functions.

The map u 7→ (f (u),
R g(u)) defines a smooth closedR curve in the plane. Its signed
area is given by f dg. Notice that ϕ(f, g) = f dg is a cyclic 1-cocycle on
C ∞ (S 1 ).

Exercise 0.13.4. Let c : Z2 × Z2 → C be the map

c((a, b), (c, d)) = ad − bc.

Show that c is a normalized group 2-cocycle in the sense of Example 0.13.5. Show
that the associated cyclic 2-cocycle on the group algebra CZ2 extends to its smooth
completion and coincides, up to scale, with the volume form on the two torus.

Exercise 0.13.5. Check that 1) for any c, the cochain ϕc defined in (23) is a
Hochschild cocycle, i.e., bϕc = 0; 2) if c is a Lie algebra cycle, i.e., if δ(c) = 0, then
ϕc is a cyclic cocycle.

0.14 Connes’ long exact sequence

Our goal in this section is to establish the long exact sequence of Connes relating
Hochschild and cyclic cohomology groups. There is a similar sequence relating
Hochschild and cyclic homology. Connes’ sequence is the long exact sequence
of a short exact sequence and the main difficulty in the proof is to identify the
cohomology of the quotient as cyclic cohomology, with a shift in dimension, and
to identify the maps.
Let A be an algebra and let Cλ and C denote its cyclic and Hochschild cochain
complexes, respectively. Consider the short exact sequence of complexes
π
0 → Cλ → C −→ C/Cλ → 0. (24)

Its associated long exact sequence is

· · · → HC n (A) → HH n (A) → H n (C/Cλ ) → HC n+1 (A) → · · · . (25)

We need to identify the cohomology groups H n (C/Cλ ). To this end, consider the
short exact sequence
1−λ N
0 → C/Cλ −−−→ (C, b0 ) −−→ Cλ → 0, (26)

where the operator N is defined by

N = 1 + λ + λ2 + · · · + λn : C n → C n .

The relations
N (1 − λ) = (1 − λ)N = 0 and bN = N b0
 0.14 Connes’ long exact sequence 61

can be verified and they show that 1 − λ and N are morphisms of complexes
in (26). As for the exactness of (26), the only non-trivial part is to show that
ker(N ) ⊂ im(1 − λ). But this follows from the relation

(1 − λ)(1 + 2λ + 3λ2 + · · · + (n + 1)λn ) = N − (n + 1) idC n .

Now, assuming A is unital, the middle complex (C, b0 ) in (26) can be shown
to be exact. In fact we have a contracting homotopy s : C n → C n−1 defined by

(sϕ)(a0 , . . . , an−1 ) = (−1)n−1 ϕ(a0 , . . . , an−1 , 1),

which satisfies
b0 s + sb0 = id .
The long exact sequence associated to (26) looks like

· · · → H n (C/Cλ ) → Hbn0 (C) → HC n (A) → H n+1 (C/Cλ ) → Hbn+1

0 (C) → · · · .
(27)
Since Hbn0 (C) = 0 for all n, it follows that the connecting homomorphism
∼
δ : HC n−1 (A) −−→ H n (C/Cλ ) (28)

is an isomorphism for all n ≥ 0. Using this in (25), we obtain Connes’ long exact
sequence relating Hochschild and cyclic cohomology:

I B S
· · · → HC n (A) −−→ HH n (A) −−→ HC n−1 (A) −−→ →HC n+1 (A) → · · · (29)

The operators B and S can be made more explicit by finding the connecting ho-
momorphisms in the above long exact sequences. Notice that B is the composition
of maps from (25) and (28):
π δ −1
B : HH n (A) −−→ H n (C/Cλ ) −−→ HC n−1 (A).

We have, on the level of cohomology, B = (1 − λ)−1 b0 N −1 . Remarkably this can

be expressed, on the level of cochains, by Connes’ operator B:

B = N s(1 − λ).

In fact, we have

N s(1 − λ)(1 − λ)−1 b0 N −1 [ϕ] = N sb0 N −1 [ϕ]

= N (1 − b0 s)N −1 [ϕ]
= (1 − bN sN −1 )[ϕ]
= [ϕ].

We can also write B as

B = N s(1 − λ) = N B0 ,
62 Introduction

where B0 : C n → C n−1 is defined by

B0 ϕ(a0 , . . . , an−1 ) = ϕ(1, a0 , . . . , an−1 ) − (−1)n ϕ(a0 , . . . , an−1 , 1).

Using the relations (1 − λ)b = b0 (1 − λ), (1 − λ)N = N (1 − λ) = 0, bN = N b0 ,

and sb0 + b0 s = 1, it is easy to show that

bB + Bb = 0 and B 2 = 0.

Let
S 0 : HC n−1 (A) → HC n+1 (A)
be the composition of connecting homomorphisms in (28) and (25) associated to
the short exact sequences (24) and (26):

S 0 : HC n−1 (A) −−
∼
→ H n (C/Cλ ) → HC n+1 (A)

Therefore we have
S 0 [ϕ] = [b(1 − λ)−1 b0 N −1 ϕ].
Any cochain ψ ∈ (1 − λ)−1 b0 N −1 ϕ has the property that bψ is cyclic, as can be
easily checked, and B[ψ] = [ϕ]. For the latter notice that

B(1 − λ)−1 b0 N −1 ϕ = N s(1 − λ)(1 − λ)−1 b0 N −1 ϕ

= N (1 − b0 s)N −1 ϕ
= ϕ − bN sϕ.

This gives us the formula

S 0 [ϕ] = [bψ] = [bB −1 ϕ].

So far we have a long exact sequence

I B S0
· · · → HC n (A) −−→ HH n (A) −−→ HC n−1 (A) −−→ HC n+1 (A) → · · · . (30)

At this point an important remark is in order. The operator S 0 as defined above,

coincides, up to scale, with the periodicity operator S defined in Example 0.13.9.
In fact, using the explicit formulae for both S and S 0 , one shows (cf. also [18],
Lemma 4.34) that for any cyclic (n − 1)-cocycle [ϕ] ∈ HC n−1 (A),

S[ϕ] = n(n + 1)S 0 [ϕ].

Thus in the exact sequence (30) we can replace S 0 with its scalar multiple S and
this of course will give Connes’ exact sequence (29). For future use we record the
new formula for S : HC n−1 (A) → HC n+1 (A),

S[ϕ] = n(n + 1) bB −1 [ϕ] = n(n + 1)[b(1 − λ)−1 b0 N −1 ϕ]. (31)

 0.14 Connes’ long exact sequence 63

Using the periodicity operator S, the periodic cyclic cohomology of an algebra

A is defined as the direct limit under the operator S of cyclic cohomology groups:
HP i (A) := Lim HC 2n+i (A), i = 0, 1.
−→

Notice that since S has degree 2 there are only two periodic groups.
Typical applications of Connes’ IBS long exact sequence are to extract infor-
mation on cyclic cohomology from Hochschild cohomology. We list some of them:
1) Let f : A → B be an algebra homomorphism and suppose that the induced
maps on Hochschild groups
f ∗ : HH n (B) → HH n (A)
are isomorphisms for all n ≥ 0. Then
f ∗ : HC n (B) → HC n (A)
is an isomorphism for all n ≥ 0 as well. This simply follows by comparing the
IBS sequences for A and B and applying the Five Lemma. For example, using
Lemma 0.12.1, it follows that inner automorphisms act as identity on (periodic)
cyclic cohomology.
Maps between cohomology groups need not be induced by algebra maps. For
example if f : (C ∗ (B), b) → (C ∗ (A), b) is a morphism of Hochschild complexes and
if f commutes with the cyclic operator λ, then it induces a map (Cλ∗ (B), b) →
(Cλ∗ (A), b) between cyclic complexes. Using the IBS sequence, we conclude that if
the induced maps between Hochschild cohomology groups are isomorphisms then
the induced maps between cyclic groups are isomorphisms as well. For example
using Lemma 0.12.1 we conclude that derivations act trivially on (periodic) cyclic
cohomology groups.
2) (Morita invariance of cyclic cohomology) Let A and B be Morita equivalent
unital algebras. The Morita invariance property of cyclic cohomology states that
there is a natural isomorphism
HC n (A) ' HC n (B), n = 0, 1, . . . .
For a proof of this fact in general see [56]. In the special case where B = Mk (A) a
simple proof can be given as follows. Indeed, by Morita invariance of Hochschild
cohomology, we know that the inclusion i : A → Mk (A) induces isomorphisms on
Hochschild groups and therefore on cyclic groups by 1) above.
3) (Normalization) A cochain f : A⊗(n+1) → C is called normalized if
f (a0 , a1 , . . . , an ) = 0
whenever ai = 1 for some i ≥ 1. It is clear that normalized cyclic cochains form
∗
a subcomplex (Cλ,norm (A), b) of the cyclic complex of A. Since the corresponding
inclusion for Hochschild complexes is a quasi-isomorphism (Exercise 0.9.3), using
the IBS sequence we conclude that the inclusion of cyclic complexes is a quasi-
isomorphism as well.
64 Introduction

Exercise 0.14.1. Show that (26) is exact (the interesting part is to show that
Ker N ⊂ Im(1 − λ)).

Exercise 0.14.2. Prove the relations bB + Bb = 0 and B 2 = 0. (They will be

used later, together with b2 = 0, to define the (b, B)-bicomplex).

Exercise 0.14.3. Let A = C be the ground field. Compute the operators

B : C n (C) → C n+1 (C) and S : HC n (C) → HC n+2 (C). Conclude that HP 2n (C) =
C and HP 2n+1 (C) = 0.

Exercise 0.14.4. Let A = Mn (C). Show that the cochains ϕ2n : A⊗(2n+1) → C
defined by
ϕ2n (a0 , . . . , a2n ) = Tr(a0 a1 . . . a2n )
are cyclic cocycles on A. We have S[ϕ2n ] = λ2n [ϕ2n+2 ]. Compute the constants
λ2n .

Exercise 0.14.5. Give examples of algebras whose Hochschild groups are iso-
morphic in all dimensions but whose cyclic groups are not isomorphic. In other
words, an ‘accidental’ isomorphism of Hochschild groups does not imply cyclic
cohomologies are isomorphic (despite the long exact sequence).

0.15 Connes’ spectral sequence

The cyclic complex (18) and the long exact sequence (29), useful as they are,
are not powerful enough for computations. A much deeper relation between
Hochschild and cyclic cohomology groups is encoded in Connes’ (b, B)-bicomplex
and the associated spectral sequence that we shall briefly recall now, following
closely the original paper [19].
Let A be a unital algebra. The (b, B)-bicomplex of A, denoted by B(A), is the
bicomplex
.. .. ..
. . .

C 2 (A)
B / C 1 (A) B / C 0 (A)
O O
b b

C 1 (A)
B / C 0 (A)
O
b

C 0 (A).

Of the three relations

b2 = 0, bB + Bb = 0, B 2 = 0,
 0.15 Connes’ spectral sequence 65

only the middle relation is not obvious. But this follows from the relations b0 s +
sb0 = 1, (1 − λ)b = b0 (1 − λ) and N b0 = bN , already used in the previous section.
The total complex of a bicomplex (C ∗,∗ , d1 , d2 ) is defined as the complex
(Tot C, d), where (Tot C)n = p+q=n C p,q and d = d1 + d2 . The following result
L

is fundamental. It shows that the resulting Connes’ spectral sequence obtained by

filtration by rows which has Hochschild cohomology for its E 1 terms, converges to
cyclic cohomology.
Theorem 0.15.1 ([19]). The map ϕ 7→ (0, . . . , 0, ϕ) is a quasi-isomorphism of
complexes
(Cλ∗ (A), b) → (Tot B(A), b + B).
This is a consequence of the vanishing of the E 2 term of the second spectral
sequence (filtration by columns) of B(A). To prove this consider the short exact
sequence of b-complexes
0 → Im B → Ker B → Ker B/ Im B → 0
By a hard lemma of Connes ([19], Lemma 41), the induced map
Hb (Im B) → Hb (Ker B)
is an isomorphism. It follows that Hb (Ker B/ Im B) vanishes. To take care of the
first column one appeals to the fact that
Im B ' Ker(1 − λ)
is the space of cyclic cochains.
We give an alternative proof of Theorem 0.15.1 above. To this end, consider
the cyclic bicomplex C(A) defined by
.. .. ..
. . .

C 2 (A)
1−λ
/ C 2 (A) N / C 2 (A) 1−λ
/ ···
O O O
b −b0 b

C 1 (A)
1−λ
/ C 1 (A) N / C 1 (A)(A) 1−λ
/ ···
O O O
b −b0 b

C 0 (A)
1−λ
/ C 0 (A) N / C 0 (A)(A) 1−λ
/ ···.

The total cohomology of C(A) is isomorphic to cyclic cohomology:

H n (Tot C(A)) ' HC n (A), n ≥ 0.
This is a consequence of the fact that the rows of C(A) are exact except in degree
zero. To see this, define the homotopy operator
1
H= (1 + 2λ + 3λ2 + · · · + (n + 1)λn ) : C n (A) → C n (A). (32)
n+1
66 Introduction

1
We have (1 − λ)H = n+1 N − id, which of course implies the exactness of rows in
positive degrees and for the first column we are left with the cyclic complex:
.. .. ..
. . .

Cλ2 (A) /0 /0 / ···

O O O
b

Cλ1 (A) /0 /0 / ···

O O O
b

Cλ0 (A) /0 /0 / ···.

We are therefore done with the proof of Theorem 0.15.1 provided we can prove
that Tot B(A) and Tot C(A) are quasi-isomorphic. The next proposition proves
this by an explicit formula:
Proposition 0.15.1. The complexes Tot B(A) and Tot C(A) are homotopy equiv-
alent.
Proof. We define explicit chain maps between these complexes and show that they
are chain homotopic via explicit homotopies. Define

I : Tot B(A) → Tot C(A), I = id +N s,

J : Tot C(A) → Tot B(A), J = id +sN.

One checks that I and J are chain maps.

Now consider the operators

g : Tot B(A) → Tot B(A), g = N s2 B0 ,

h : Tot C(A) → Tot C(A), h = s,

where B0 = s(1 − λ).

We have, by direct computation:

I ◦ J = id +hδ + δh,
J ◦ I = id +gδ 0 + δ 0 g,

where δ (resp. δ 0 ) denotes the differential of Tot C(A) (resp. Tot B(A)).
There is a similar result for cyclic homology.

0.16 Cyclic cohomology computations

Cyclic cohomology has been computed for many algebras, most notably algebras of
smooth functions, group algebras and crossed product algebras, groupoid algebras,
 0.16 Cyclic cohomology computations 67

noncommutative tori, universal enveloping algebras, and almost commutative al-

gebras. Equipped with these core examples, one can then use general results like
additivity, Morita invariance, homotopy invariance, Künneth formulae, and exci-
sion [31], to compute the (periodic) cyclic cohomology of even larger classes of
algebras. The main technique to deal with core examples is to find a suitable
resolution for the Hochschild complex to compute the Hochschild cohomology first
and then find the action of the operator B on the Hochschild complex. In good
cases, the E 2 -term of the spectral sequence associated with the (b, B)-bicomplex
vanishes and one ends up with a computation of cyclic cohomology. To illustrate
this idea, we recall some of these computations in this section.
Example 0.16.1 (Algebras of smooth functions). Let A = C ∞ (M ) denote the
algebra of smooth complex-valued functions on a closed smooth manifold M with
its natural Fréchet algebra topology, and let (ΩM, d) denote the de Rham complex
ˆ
of M . Let Cn (A) = A⊗(n+1) denote the space of continuous n-chains on A. We
saw in Example 0.12.2 that the map µ : Cn (A) → Ωn M defined by
1
µ(f0 ⊗ · · · ⊗ fn ) = f0 df1 ∧ · · · ∧ dfn ,
n!
induces an isomorphism between the continuous Hochschild homology of A and
differential forms on M :
HHncont (A) ' Ωn M.
To compute the continuous cyclic homology of A, we first show that under the
map µ the operator B corresponds to the de Rham differential d. More precisely,
for each integer n ≥ 0 we have a commutative diagram:

Cn (A)
µ
/ Ωn M

B d
 
Cn+1 (A)
µ
/ Ωn+1 M .

We have

µB(f0 ⊗ · · · ⊗ fn )
X n
=µ (−1)ni (1 ⊗ fi ⊗ · · · ⊗ fi−1 − (−1)n fi ⊗ · · · ⊗ fi−1 ⊗ 1)
i=0
n
1 X
= (−1)ni dfi . . . dfi−1
(n + 1)! i=0
1
= (n + 1)df0 . . . dfn = dµ(f0 ⊗ · · · ⊗ fn ).
(n + 1)!
It follows that µ defines a morphism of bicomplexes

B(A) → Ω(A),
68 Introduction

where Ω(A) is the bicomplex

.. .. ..
. . .

Ω2 M o Ω1 M o Ω0 M
d d
0 0
 
Ω1 M o Ω0 M
d
0

Ω0 M .

Since µ induces isomorphisms on row homologies, it induces isomorphisms on total

homologies as well. Thus we have [19]:

HCncont (C ∞ (M )) ' Ωn M/ im d ⊕ HdR

n−2 k
(M ) ⊕ · · · ⊕ HdR (M ),

where k = 0 if n is even and k = 1 if n is odd. Notice that the top part, for
n ≤ dim(M ), consists of the so called co-closed differential n-forms on M .
Using the corresponding periodic complexes, one concludes that the continuous
periodic cyclic homology of C ∞ (M ) is given by

HPkcont (C ∞ (M )) '
L 2i+k
HdR (M ), k = 0, 1.
i

There are of course dual results relating continuous cyclic cohomology of C ∞ (M )

and de Rham homology of M . Let (Ω∗ M, d) denote the complex of de Rham cur-
n
rents on M . Recall from Example 0.12.2 that the map Ωn M → Ccont (A) defined
by sending a current C ∈ Ωn M to the cochain ϕC , where

ϕC (f 0 , f 1 , . . . , f n ) = hC, f 0 df 1 ∧ · · · ∧ df n i

is a quasi-isomorphism. By basically following the same route as above we obtain

the following theorem of Connes [19]:

n
HCcont (C ∞ (M )) ' Z n (M ) ⊕ Hn−2
dR
(M ) ⊕ · · · ⊕ HkdR (M ) (33)

where Z n (M ) is the space of closed de Rham n-currents on M and k = 0 if n is

even and k = 1 if n is odd. Finally, for the continuous periodic cyclic cohomology
we obtain:
k
(C ∞ (M )) '
L dR
HPcont H2i+k (M ), k = 0, 1 (34)
i

Now (33) shows that cyclic cohomology is not homotopy invariant. In fact while
the de Rham cohomology components are homotopy invariant the top component
Z n (M ) cannot be homotopy invariant. Formula (34) on the other hand shows
that the periodic cyclic cohomology of C ∞ (M ) is homotopy invariant. This is a
 0.16 Cyclic cohomology computations 69

special case of the homotopy invariance of periodic cyclic cohomology [19], [56].
More precisely, for any algebras A and B and smoothly homotopic algebra maps
f0 , f1 : A → B, we have f0∗ = f1∗ : HP ∗ (B) → HP ∗ (A). In particular the algebras
A and A[x] have isomorphic periodic cyclic cohomologies.

Example 0.16.2 (Smooth commutative algebras). Let A = O(X) be the coordi-

nate ring of an affine smooth variety over C and let (Ω∗A , d) denote the de Rham
complex of A. As we saw in Example 0.12.1, by Hochschild–Kostant–Rosenberg’s
theorem, the map
µ : Cn (A) → ΩnA
induces an isomorphism between Hochschild homology of A and the differential
forms on X. By the same method as in the previous example one then arrives at
the isomorphisms
n−2
HCn (O(X)) ' ΩnA / im d ⊕ HdR k
(X) ⊕ · · · ⊕ HdR (X),
L dR
HPk (O(X)) ' H2i+k (X), k = 0, 1.
i

n
Notice that the de Rham cohomology HdR (X) appearing on the right side is
n
isomorphic to the singular cohomology H (Xtop , C) of the underlying topological
space of X.
When X is singular the relations between the cyclic homology of O(X) and
the topology of Xtop can be quite complicated. The situation for periodic cyclic
homology however is quite straightforward, as the following theorem of Feigin and
Tsygan [36] indicates:

H 2i+k (Xtop , C).

L
HPk (O(X)) ' (35)
i

Notice that X need not be smooth and the cohomology on the right-hand side is
the singular cohomology.

Example 0.16.3 (Group algebras). Let CG denote the group algebra of a discrete
group G. Here, to be concrete, we work over C, but results hold over any field
of characteristic zero. As we saw in Example 0.12.3, the Hochschild complex of
CG decomposes over the set hGi of conjugacy classes of G and the homology of
each summand is isomorphic to the group homology of a group associated to the
conjugacy class:
L
HHn (CG) ' Hn (Cg ),
hGi

where Cg is the centralizer of a representative g of a conjugacy class of G [11],

[21], [56]. Recall the decomposition
L
C∗ (CG, b) = B(G, c)
c∈hGi
70 Introduction

of the Hochschild complex of CG from Example 0.12.3, where for each conjugacy
class c ∈ hGi, Bn (G, c) is the linear span of all (n+1)-tuples (g0 , g1 , . . . , gn ) ∈ Gn+1
such that
g0 g1 . . . gn ∈ c.
It is clear that Bn (G, c), n = 0, 1, 2, . . . , are invariant not only under the Hochschild
differential b, but also under the cyclic operator λ. Let

Bnλ (G, c) = Bn (G, c)/ im(1 − λ).

We then have a decomposition of the cyclic complex of CG into subcomplexes

indexed by conjugacy classes:

C∗λ (CG, b) =
L λ
B∗ (G, c).
c∈Ĝ

The homology of B∗λ (G, {e}) was first computed by Karoubi [49], [56] in terms
of the group homology of G. The result is

Hn (B∗λ (G, {e}) =

L
Hn−2i (G).
i

Burghelea’s computation of the cyclic homology of CG [11] (cf. also [?], [?] for
a purely algebraic proof) can be described as follows. Let hGifin and hGi∞ denote
the set of conjugacy classes of elements of finite, and infinite orders, respectively.
For an element g ∈ G, let Ng = Cg /hgi, where hgi is the group generated by g and
Cg is the centralizer of g. Notice that the isomorphism type of Ng only depends
on the conjugacy class of g. In each conjugacy class c we pick a representative
g ∈ c once and for all. Now if g is an element of finite order we have

Hn (B∗λ (G, c)) =

L
Hn−2i (Cg ).
i≥0

On the other hand, if g is of infinite order we have

Hn (B∗λ (G, c)) = Hn (Ng ).

Putting these results together we obtain:

L L
L
HCn (CG) ' Hn−2i (Cg ) Hn (Ng ) (36)
hGifin i≥0 hGi∞

In particular, the Hochschild

L group has Hn (G) as a direct summand, while the
cyclic homology group has i Hn−2i (G) as a direct summand (corresponding to
the conjugacy class of the identity of G).
Example 0.16.4 (Noncommutative torus). We shall briefly recall Connes’ compu-
tation of the Hochschild and cyclic cohomology groups of smooth noncommutative
tori [19]. In Example ?? we showed that when θ is rational the smooth noncommu-
tative torus Aθ is Morita equivalent to C ∞ (T 2 ), the algebra of smooth functions
 0.16 Cyclic cohomology computations 71

on the torus. One can then use the Morita invariance of Hochschild and cyclic
cohomology to reduce the computation of these groups to those for the algebra
C ∞ (T 2 ). This takes care of the computation for rational θ. So, through the rest
of this example we assume that θ is irrational and we denote the generators of Aθ
by U1 and U2 with the relation U2 U1 = λU1 U2 , where λ = e2πiθ .
Let B = Aθ ⊗ ˆ Aopθ . There is a topological free resolution of Aθ as a left
B-module
ε b1 b2
Aθ ←− B ⊗ Ω0 ←−− B ⊗ Ω1 ←−− B ⊗ Ω2 ←− 0,
Vi 2
where Ωi = C , i = 0, 1, 2 is the i-th exterior power of C2 . The differentials are
given by

b1 (1 ⊗ ej ) = 1 ⊗ Uj − Uj ⊗ 1, j = 1, 2,
b2 (1 ⊗ (e1 ∧ e2 )) = (U2 ⊗ 1 − λ ⊗ U2 ) ⊗ e1 − (λU1 ⊗ 1 − 1 ⊗ U1 ) ⊗ e2 ,
ε(a ⊗ b) = ab.

The following result completely settles the question of continuous Hochschild

cohomology of Aθ when θ is irrational. Recall that an irrational number θ is said
to satisfy a Diophantine condition if |1 − λn |−1 = O(nk ) for some positive integer
k.

Proposition 0.16.1 ([19]). Let θ ∈

/ Q. Then the following holds.

a) One has HH 0 (Aθ ) = C.

b) If θ satisfies a Diophantine condition then HH i (Aθ ) is 2-dimensional for

i = 1 and is 1-dimensional for i = 2.

c) If θ does not satisfy a Diophantine condition, then HH i (Aθ ) are infinite

dimensional non-Hausdorff spaces for i = 1, 2.

Remarkably, for all values of θ, the periodic cyclic cohomology is finite dimen-
sional and is given by

HP 0 (Aθ ) = C2 , HP 1 (Aθ ) = C2 .

An explicit basis for these groups is given by cyclic 1-cocycles

ϕ1 (a0 , a1 ) = τ (a0 δ1 (a1 )) and ϕ1 (a0 , a1 ) = τ (a0 δ2 (a1 )),

and by cyclic 2-cocycles

ϕ(a0 , a1 , a2 ) = τ (a0 (δ1 (a1 )δ2 (a2 ) − δ2 (a1 )δ1 (a2 ))) and Sτ,

where δ1 , δ2 : Aθ → Aθ are the canonical derivations defined by

X  X X
δ1 amn U1m U2n = mamn U1m U2n and δ2 (U1m U2n ) = namn U1m U2n ,
72 Introduction

and τ : Aθ → C is the canonical trace (cf. Example ??). Note that Sτ (a0 , a1 , a2 ) =
τ (a0 a1 a2 ).
Let O(Tθ2 ) denote the (dense) subalgebra of Aθ generated by U1 and U2 . In
Exercise 0.9.7 we ask the reader to show that the (algebraic) Hochschild groups of
O(Tθ2 ) are finite dimensional for all values of θ.
Exercise 0.16.1. Since CZ = C[z, z −1 ] is both a group algebra and a smooth al-
gebra, we have two descriptions of its Hochschild and cyclic homologies. Compare
the two descriptions and show that they are the same.
Exercise 0.16.2. Let G be a finite group. Use Burghelea’s theorem in Exam-
ple 0.16.3 to compute the Hochschild and cyclic homology of CG. Alternatively,
one knows that CG is a direct sum of matrix algebras and one can use the Morita
invariance of Hochschild and cyclic theory. Compare the two approaches.
Exercise 0.16.3. Let D∞ = Z o Z2 be the infinite dihedral group. Use (36) to
compute the cyclic homology of the group algebra CD∞ .

Exercise 0.16.4. Let X = {(x, y) ∈ C2 ; xy = 0}, and let A = O(X) be the

coordinate ring of X. Verify (35) for A.
Exercise 0.16.5. Prove directly, without using Proposition 0.16.1, that when θ is
irrational Aθ has a unique trace and therefore HH 0 (Aθ ) = C. Describe the traces
on Aθ for rational θ.
Lecture 4: Cyclic modules

The original motivation of [18] was to define cyclic cohomology of algebras as a

derived functor. Since the category of algebras and algebra homomorphisms is not
even an additive category (for the simple reason that the sum of two algebra ho-
momorphisms is not an algebra homomorphism in general), the standard (abelian)
homological algebra is not applicable. Let k be a unital commutative ring. In [18],
the category Λk of cyclic k-modules appears as an ‘abelianization’ of the category
of k-algebras. Cyclic cohomology is then shown to be the derived functor of the
functor of traces, as we explain in this section.
Recall that the simplicial category ∆ is a small category whose objects are the
totally ordered sets ([56])
[n] = {0 < 1 < · · · < n}, n = 0, 1, 2, . . . .
A morphism f : [n] → [m] of ∆ is an order preserving, i.e., monotone non-
decreasing, map f : {0, 1, . . . , n} → {0, 1, . . . , m}. Of particular interest among
the morphisms of ∆ are faces δi and degeneracies σj ,
δi : [n − 1] → [n], i = 0, 1, . . . , n,
σj : [n + 1] → [n], j = 0, 1, . . . , n.
By definition δi is the unique injective morphism missing i and σj is the unique
surjective morphism identifying j with j + 1. It can be checked that they satisfy
the following simplicial identities:
δj δi = δi δj−1 if i < j,
σj σi = σi σj+1 if i ≤ j,

δi σj−1 if i < j,

σj δi = id[n] if i = j or i = j + 1,

δi−1 σj if i > j + 1.


Every morphism of ∆ can be uniquely decomposed as a product of faces followed

by a product of degeneracies.
The cyclic category Λ has the same set of objects as ∆ and in fact contains
∆ as a subcategory. Morphisms of Λ are generated by simplicial morphisms and

73
74 Introduction

new morphisms τn : [n] → [n], n ≥ 0, defined by τn (i) = i + 1 for 0 ≤ i < n and

τn (n) = 0. We have the following extra relations:
τn δi = δi−1 τn−1 , 1 ≤ i ≤ n,
τn δ0 = δn ,
τn σi = σi−1 τn+1 , 1 ≤ i ≤ n,
2
τn σ0 = σn τn+1 ,
τnn+1 = id .
It can be shown that the classifying space BΛ of the small category Λ is homotopy
equivalent to the classifying space BS 1 = CP ∞ [18].
A cyclic object in a category C is a functor Λop → C. A cocyclic object in C is a
functor Λ → C. For any commutative unital ring k, we denote the category of cyclic
k-modules by Λk . A morphism of cyclic k-modules is a natural transformation
between the corresponding functors. Equivalently, a morphism f : X → Y consists
of a sequence of k-linear maps fn : Xn → Yn compatible with the face, degeneracy,
and cyclic operators. It is clear that Λk is an abelian category. The kernel and
cokernel of a morphism f are defined pointwise: (Ker f )n = Ker fn : Xn → Yn and
(Coker f )n = Coker fn : Xn → Yn . More generally, if A is any abelian category
then the category ΛA of cyclic objects in A is itself an abelian category.
Let Algk denote the category of unital k-algebras and unital algebra homomor-
phisms. There is a functor
\ : Algk → Λk
defined as follows. To an algebra A we associate the cyclic module A\ defined by
A\n = A⊗(n+1) , n ≥ 0,
with face, degeneracy and cyclic operators given by
δi (a0 ⊗ a1 ⊗ · · · ⊗ an ) = a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an ,
δn (a0 ⊗ a1 ⊗ · · · ⊗ an ) = an a0 ⊗ a1 ⊗ · · · ⊗ an−1 ,
σi (a0 ⊗ a1 ⊗ · · · ⊗ an ) = a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗ · · · ⊗ an ,
τn (a0 ⊗ a1 ⊗ · · · ⊗ an ) = an ⊗ a0 ⊗ · · · ⊗ an−1 .

A unital algebra map f : A → B induces a morphism of cyclic modules f \ : A\ →

B \ by f \ (a0 ⊗ · · · ⊗ an ) = f (a0 ) ⊗ · · · ⊗ f (an ), and this defines the functor \.
Example 0.16.5. We have
HomΛk (A\ , k \ ) ' T (A),
where T (A) is the space of traces from A → k. To a trace τ we associate the cyclic
map (fn )n≥0 , where
fn (a0 ⊗ a1 ⊗ · · · ⊗ an ) = τ (a0 a1 . . . an ), n ≥ 0.
It can be easily shown that this defines a one-to-one correspondence.
 0.16 Cyclic cohomology computations 75

Now we can state the following fundamental theorem of Connes [18] which
greatly extends the above example:

Theorem 0.16.1. Let k be a field of characteristic zero. For any unital k-algebra
A there is a canonical isomorphism

HC n (A) ' ExtnΛk (A\ , k \ ) for all n ≥ 0.

Before sketching its proof, we mention that combined with the above example
the theorem implies that cyclic cohomology is, in some sense, the non-abelian
derived functor of the functor of traces

A ; T (A)

from the category of k-algebras to the category of k-modules.

Sketch of proof. The main step in the proof of Theorem 0.16.1 is to find an injective
resolution of k \ in Λk . The required injective cyclic modules will be the dual of
some projective cyclic modules that we define first. For each integer m ≥ 0, let us
define a cyclic module C m where

(C m )n = k HomΛ ([m], [n])

is the free k-module generated by the set of all cyclic maps from [m] → [n].
Composition in Λ defines a natural cyclic module structure on each C k . For any
cyclic module M we clearly have HomΛk (C m , M ) = Mm . This of course implies
that each C m is a projective cyclic module. (Recall that an object P of an abelian
category is called projective if the functor M 7→ Hom(P, M ) is exact in the sense
that it sends any short exact sequence in the category into a short exact sequence
of abelian groups.) The corresponding projective resolution of k \ is defined as the
total complex of the following double complex:
.. .. ..
. . .

CO 2
1−λ
/ C2 N / C2 1−λ
/ ···
O O
b b0 b
(37)
CO 1 1−λ
/ C1 N / C1 1−λ
/ ···
O O
b b0 b

C0
1−λ
/ C0 N / C0 1−λ
/ ···,

where the cyclic module maps b, b0 , λ and N are defined by

k
X k−1
X
b(f ) = (−1)i f ◦ δi , b0 (f ) = (−1)i f ◦ δi ,
i=0 i=0
76 Introduction

λ(f ) = (−1)k f ◦ τk , and N = 1 + λ + · · · + λk

Now by a direct argument one shows that the row homologies of the above bicom-
plex (37) are trivial in positive dimensions. Thus to compute its total homology
it remains to compute the homology of the complex of complexes:
b b
C 0 /(1 − λ) ←−− C 1 /(1 − λ) ←−− · · · .

It can be shown that for each fixed m, the complex of vector spaces
b b
C 0m /(1 − λ) ←−− C 1m /(1 − λ) ←−− · · ·

coincides with the complex that computes the simplicial homology of the simplicial
set ∆m . The simplicial set ∆m is defined by ∆m n = Hom∆ ([m], [n]) for all n ≥ 0.
The geometric realization of ∆m is the closed unit ball in Rm which is of course
contractible. It follows that the total homology of the above bicomplex is the
cyclic module k \ . We note that for this argument K need not be a field. Now
if k is a field of characteristic zero the cyclic modules C m , m ≥ 0, defined by
(C m )n = Homk ((C m )n , k) are injective cyclic modules. Dualizing the bicomplex
(37), finally we obtain an injective resolution of k \ as a cyclic module. To compute
the Ext∗Λk (A\ , k \ ) groups, we apply the functor HomΛk (A\ , −) to this resolution.
We obtain the bicomplex
.. .. ..
. . .

C 2 (A)
1−λ
/ C 2 (A) N / C 2 (A) 1−λ
/ ···
O O O
b −b0 b
(38)
1
C (A)
1−λ
/ C 1 (A) N / C1 1−λ
/ ···
O O O
b −b0 b

C 0 (A)
1−λ
/ C 0 (A) N / C 0 (A) 1−λ
/ ···.

We are done with the proof of Theorem 0.16.1, provided we can show that the
cohomology of (38) is isomorphic to the cyclic cohomology of A. But this we have
already shown in the last section. This finishes the proof of the theorem.
A remarkable property of the cyclic category Λ, not shared by the simplicial
category, is its self-duality in the sense that there is a natural isomorphism of
categories Λ ' Λop [18]. Roughly speaking, the duality functor Λop → Λ acts as
identity on objects of Λ and exchanges face and degeneracy operators while sending
the cyclic operator to its inverse. Thus to a cyclic (resp. cocyclic) module one can
associate a cocyclic (resp. cyclic) module by applying the duality isomorphism.
Example 0.16.6 (Hopf cyclic cohomology). We give a very non-trivial example
of a cocyclic module. Let H be a Hopf algebra. A character of H is a unital
algebra map δ : H → C. A group-like element is a nonzero element σ ∈ H such
 0.16 Cyclic cohomology computations 77

that ∆σ = σ ⊗ σ. Following [26], [27], we say (δ, σ) is a modular pair if δ(σ) = 1,

and a modular pair in involution if

S̃δ2 (h) = σhσ −1

for all h in H. Here the δ-twisted antipode S̃δ : H → H is defined by

X
S̃δ (h) = δ(h(1) )S(h(2) ).

Now let (H, δ, σ) be a Hopf algebra endowed with a modular pair in involution.
\
In [26] Connes and Moscovici attach a cocyclic module H(δ,σ) to this data as
follows. Let

\,0 \,n
H(δ,σ) =C and H(δ,σ) = H ⊗n for n ≥ 1.

Its face, degeneracy and cyclic operators δi , σi , and τn are defined by

δ0 (h1 ⊗ · · · ⊗ hn ) = 1 ⊗ h1 ⊗ · · · ⊗ hn ,
δi (h1 ⊗ · · · ⊗ hn ) = h1 ⊗ · · · ⊗ ∆(hi ) ⊗ · · · ⊗ hn for 1 ≤ i ≤ n,
78 Introduction

δn+1 (h1 ⊗ · · · ⊗ hn ) = h1 ⊗ · · · ⊗ hn ⊗ σ,
σi (h1 ⊗ · · · ⊗ hn ) = h1 ⊗ · · · ⊗ hi ε(hi+1 ) ⊗ · · · ⊗ hn for 0 ≤ i < n,
n−1
τn (h1 ⊗ · · · ⊗ hn ) = ∆ S̃δ (h1 ) · (h2 ⊗ · · · ⊗ hn ⊗ σ).

Checking the cyclic property of τn , i.e., τnn+1 = 1 is a highly non-trivial task. The
\
cyclic cohomology of the cocyclic module H(δ,σ) is the Hopf cyclic cohomology of
the triple (H, δ, σ). (cf. also [1], [42], [43] for more examples of cyclic modules
arising from actions and coactions of Hopf algebras on algebras and coalgebras.)
Lecture 5: Connes–Chern
character

Classicaly, Chern character relates the K-theory of a space to its ordinary coho-
mology theory. In noncommutative geometry, in addition to K-theory there is also
a very important dual K-homology theory built out of abstract elliptic operators
on the noncommutative space. In this lecture we look at the noncommutative ana-
logues of Chern character maps for both K-theory and K-homology, with values
in cyclic homology and cyclic cohomology, respectively. As we mentioned before,
it was the search for a noncommutative analogue of the Chern character in K-
homology that eventually led Alain Connes to the discovery of cyclic cohomology.
K-theory, K-homology, cyclic homology and cohomology, via their allied Chern
character maps, enter into a beautiful index formula of Connes which plays an
important role in applications of noncommutative geometry.
We mentione that of all characteristic classes, the Chern character is the only
one that admits an extension to the noncommutative world. In particular there are
no analogues of Chern classes or Pontryagin classes for noncommutative algebras.

0.17 What is the Chern character?

The classical Chern character is a natural transformation from K-theory to ordi-
nary cohomology theory with rational coefficients [61]. More precisely, for each
compact Hausdorff space X we have a natural ring homomorphism

Ch : K 0 (X) → H 2i (X, Q),

L
i≥0

where K 0 (resp. H) denotes the K-theory (resp. Čech cohomology with rational
coefficients). Moreover, thanks to result of Atiyah and Hirzebruch, it is known
that Ch is a rational isomorphism. It satisfies certain axioms and these axioms
completely characterize Ch. We shall not recall these axioms here since they are
not very useful for finding the noncommutative analogue of Ch. What turned
out to be most useful in this regard was the Chern–Weil definition of the Chern

79
80 Introduction

character for smooth manifolds,

Ch : K 0 (X) → 2i
L
HdR (X),
i≥0

using differential geometric notions of connection and curvature on vector bundles

over smooth manifolds. (cf. [61], and Example 0.18.4 in this section). Now let us
describe the situation in the noncommutative case.
In [16], [19], [21], Connes shows that Chern–Weil theory admits a vast general-
ization to a noncommutative setting. For example, for a not necessarily commuta-
tive algebra A and each integer n ≥ 0 there are natural maps, called Connes–Chern
character maps,
Ch02n : K0 (A) → HC2n (A),
Ch2n+1
1 : K1 (A) → HC2n+1 (A)
from the K-theory of A to its cyclic homology.
Alternatively, these maps can be defined as a pairing between cyclic cohomology
and K-theory:

HC 2n (A) ⊗ K0 (A) → C, HC 2n+1 (A) ⊗ K1 (A) → C. (39)

These pairings are compatible with the periodicity operator S in cyclic cohomology
in the sense that
h[ϕ], [e]i = hS[ϕ], [e]i
for all cyclic cocycles ϕ and K-theory classes [e], and thus induce a pairing

HP i (A) ⊗ Ki (A) → C, i = 0, 1,

between periodic cyclic cohomology and K-theory.

0.18 Connes–Chern character in K-theory

We start by defining the pairings (39). Let ϕ be a cyclic 2n-cocycle on an algebra
A. For each integer k ≥ 1, the formula

ϕ̃(m0 ⊗ a0 , . . . , m2n ⊗ a2n ) = tr(m0 . . . m2n )ϕ(a0 , . . . , a2n ) (40)

defines a cyclic 2n-cocycle ϕ̃ ∈ Zλ2n (Mk (A)). Let e ∈ Mk (A) be an idempotent

representing a class in K0 (A). Define a bilinear map

HC 2n (A) ⊗ K0 (A) → C (41)

by the following formula:

h[ϕ], [e]i = (n!)−1 ϕ̃(e, . . . , e) (42)

 0.18 Connes–Chern character in K-theory 81

Let us first check that the value of the pairing depends only on the cyclic co-
homology class of ϕ in HC 2n (A). It suffices to assume k = 1 (why?). Let ϕ = bψ
with ψ ∈ Cλ2n−1 (A), be a coboundary. Then we have

ϕ(e, . . . , e) = bψ(e, . . . , e)
= ψ(ee, e, . . . , e) − ψ(e, ee, . . . , e) + · · · + (−1)2n ψ(ee, e, . . . , e)
= ψ(e, . . . , e)
= 0,

where the last relation follows from the cyclic property of ψ.

To verify that the value of h[ϕ], [e]i, for fixed ϕ, only depends on the class
of [e] ∈ K0 (A) we have to check that for u ∈ GLk (A) an invertible matrix, we
have h[ϕ], [e]i = h[ϕ], [ueu−1 ]i. It again suffices to show this for k = 1. But this
is exactly the fact, proved in Section 3.7, that inner automorphisms act by the
identity on cyclic cohomology. Formula (42) can be easily seen to be additive in
[e] under the direct sum e ⊕ f of idempotents. This shows that the pairing (41) is
well defined.
Proposition 0.18.1. For any cyclic cocycle ϕ ∈ Zλ2n (A) and idempotent e ∈
Mk (A) we have
h[ϕ], [e]i = hS[ϕ], [e]i.
Proof. Without loss of generality we can assume that k = 1. Using our explicit
formula (31) for the S-operator, we have

S[ϕ] = (2n + 1)(2n + 2)[b(1 − λ)−1 b0 N −1 ϕ],

where N −1 ϕ = 1
2n+1 ϕ (since ϕ is cyclic), and

−1
(1 − λ)−1 b0 ϕ = (1 + 2λ + 3λ2 + · · · + (2n + 2)λ2n+1 )b0 ϕ.
2n + 2
Thus we have

Sϕ(e, . . . , e) = −b(1 + 2λ + 3λ2 + · · · + (2n + 2)λ2n+1 )b0 ϕ(e, . . . , e)

= (n + 1)b0 ϕ(e, . . . , e)
= (n + 1)ϕ(e, . . . , e).

Now we have
1 1
hS[ϕ], [e]i = (Sϕ)(e, . . . , e) = ϕ(e, . . . , e) = h[ϕ], [e]i.
(n + 1)! n!

Example 0.18.1 (n = 0). HC 0 (A) is the space of traces on A. Therefore the

Connes–Chern pairing for n = 0 reduces to a map

{traces on A} × K0 (A) → C,
82 Introduction

k
X
hτ, [e]i = τ (eii ),
i=1

where e = [eij ] ∈ Mk (A) is an idempotent. The induced function on K0 (A)

is called the dimension function and denoted by dimτ . This terminology is sug-
gested by the commutative case. In fact if X is a compact connected topological
space, then τ (f ) = f (x0 ), x0 ∈ X, defines a trace on C(X), and for a vector
bundle E on X, dimτ (E) is the rank of the vector bundle E and is an integer.
One of the striking features of noncommutative geometry is the existence of non-
commutative vector bundles with non-integral dimensions. A beautiful example
of this phenomenon is shown is Example ?? through the Powers–Rieffel projection
e ∈ Aθ with τ (e) = θ, where τ is the canonical trace on the noncommutative torus
(cf. also [21]).
Here is a slightly different approach to this dimension function. Let E be a finite
projective right A-module. A trace τ on A induces a trace on the endomorphism
algebra of E,
Tr : EndA (E) → C
as follows. First assume that E = An is a free module. Then EndA (E) ' Mn (A)
and our trace map is defined by
X
Tr(ai,j ) = aii .
i

It is easy to check that the above map is a trace. In general, there is an A-module
F such that E ⊕F ' An is a free module and EndA (E) embeds in Mn (A). One can
check that the induced trace on EndA (E) is independent of the choice of splitting.
Now, from our description of Tr in terms of τ , it is clear that
hτ, [E]i = dimτ (E) = Tr(idE )
for any finite projective A-module E.
The topological information hidden in an idempotent is much more subtle than
just its ‘rank’, as two idempotents, say vector bundles, can have the same rank
but still be non-isomorphic. In fact traces can only capture the 0-dimensional
information. To know more about idempotents and K-theory we need the higher
dimensional analogues of traces, which are cyclic cocycles, and the pairing (42).
As we saw in Section 0.15 cyclic cocycles can also be realized in the (b, B)-
bicomplex picture of cyclic cohomology. Given an even cocycle
ϕ = (ϕ0 , ϕ2 , . . . , ϕ2n )
in the (b, B)-bicomplex, its pairing with an idempotent e ∈ Mk (A) can be shown
to be given by
n  
X k! 1
h[ϕ], [e]i = (−1)k ϕ2k e − , e, . . . , e (43)
(2k)! 2
k=1
 0.18 Connes–Chern character in K-theory 83

(cf. Exercise 0.18.2).

When A is a Banach (or at least a suitable topological) algebra, to verify that
the pairing (42) is well defined, it suffices to check that for a smooth family of
idempotents et , 0 ≤ t ≤ 1, ϕ(et , . . . , et ) is constant in t. There is an alternative
“infinitesimal proof” of this fact which is worth recording [21]:

Lemma 0.18.1. Let et , 0 ≤ t ≤ 1, be a smooth family of idempotents in a Banach

algebra A. There exists a smooth family xt , 0 ≤ t ≤ 1, of elements of A such that

. d
et := (et ) = [xt , et ] for 0 ≤ t ≤ 1.
dt

Proof. Let
. . .
xt = [et , et ] = et et − et et .

Differentiating the idempotent condition e2t = et with respect to t we obtain

d 2 . . .
(e ) = et et + et et = et .
dt t
Multiplying this last relation on the left by et yields
.
et et et = 0.

Now we have
. . . . .
[xt , et ] = [et et − et et , et ] = et et + et et = et .

It follows that if τ : A → C is a trace (= a cyclic zero cocycle), then

d d .
hτ, et i = τ (et ) = τ (et ) = τ ([xt , et ]) = 0.
dt dt
Hence the value of the pairing, for a fixed τ , depends only on the homotopy class
of the idempotent. This shows that the pairing

{traces on A} × K0 (A) → C

is well defined.
This is generalized in

Lemma 0.18.2. Let ϕ(a0 , . . . , a2n ) be a cyclic 2n-cocycle on A and let et be a

smooth family of idempotents in A. Then the number

h[ϕ], [et ]i = ϕ(et , . . . , et )

is constant in t.
84 Introduction

Proof. Differentiating with respect to t and using the above lemma, we obtain
d . .
ϕ(et , . . . , et ) = ϕ(et , . . . , et ) + ϕ(et , et , . . . , et ) + · · ·
dt
.
· · · + ϕ(et , . . . , et , et )
2n
X
= ϕ(et , . . . , [xt , et ], . . . , et )
i=0
= Lxt ϕ(et , . . . , et ).
We saw in Section 0.14 that inner derivations act trivially on Hochschild and cyclic
cohomology. This means that for each t there is a cyclic (2n − 1)-cochain ψt such
that the Lie derivative Lxt ϕ = bψt . We then have
d
ϕ(et , . . . , et ) = (bψt )(et , . . . , et ) = 0.
dt
The formulas in the odd case are as follows. Given an invertible matrix u ∈
Mk (A), representing a class in K1alg (A), and an odd cyclic (2n − 1)-cocycle ϕ on
A, we define
2−(2n+1)
h[ϕ], [u]i := ϕ̃(u−1 − 1, u − 1, . . . , u−1 − 1, u − 1), (44)
(n − 12 ) . . . 12
where the cyclic cocycle ϕ̃ is defined in (40). As we saw in Section 0.14, any cyclic
cocycle can be represented by a normalized cocycle for which ϕ(a0 , . . . , a1 ) = 0 if
ai = 1 for some i. When ϕ is normalized, formula (44) reduces to

2−(2n+1)
h[ϕ], [u]i = ϕ̃(u−1 , u, . . . , u−1 , u) (45)
(n − 12 ) . . . 12

As in the even case, the induced pairing HC 2n+1 (A) ⊗ K1alg (A) → C is com-
patible with the periodicity operator: for any odd cyclic cocycle ϕ ∈ Zλ2n+1 (A)
and an invertible u ∈ GLk (A), we have
h[ϕ], [u]i = hS[ϕ], [u]i.
These pairings are just manifestations of perhaps more fundamental maps that
define the even and odd Connes–Chern characters
Ch02n : K0 (A) → HC2n (A),
Ch2n+1
1 : K1 (A) → HC2n+1 (A),
as we describe them now. In the even case, given an idempotent e = (eij ) ∈ Mk (A),
we define for each n ≥ 0,
−1
Ch2n
0 (e) = (n!) Tr(e ⊗ e ⊗ · · · ⊗ e)
| {z }
2n+1
X (46)
= ei0 i1 ⊗ ei1 i2 ⊗ · · · ⊗ ei2n i0 ,
i0 ,i1 ,...,i2n
 0.18 Connes–Chern character in K-theory 85

where on the right-hand side the class of the tensor in A⊗(2n+1) / Im(1 − λ) is
understood. In low dimensions we have
k
X
Ch00 (e) = eii ,
i=1
k X
X k X
k
Ch20 (e) = ei0 i1 ⊗ ei1 i2 ⊗ ei2 i0 ,
i0 =1 i1 =1 i2 =1

etc. To check that Ch2n

0 (e) is actually a cycle, notice that

1
b(Ch2n
0 (e)) = (1 − λ) Tr(e ⊗ · · · ⊗ e),
2 | {z }
2n

which shows its class is zero in the quotient.

In the odd case, given an invertible matrix u ∈ Mk (A), we define

Ch2n+1
1 ([u]) = Tr((u−1 − 1) ⊗ (u − 1) ⊗ · · · ⊗ (u−1 − 1) ⊗ (u − 1)).
| {z }
2n+2

Example 0.18.2. Let A = C ∞ (S 1 ) denote the algebra of smooth complex-valued

functions on the circle. One knows that K1 (A) ' K 1 (S 1 ) ' Z and u(z) = z is a
generator of this group. Let
Z
ϕ(f0 , f1 ) = f0 df1
S1

denote the cyclic cocycle on A representing the fundamental class of S 1 in de Rham

homology. Notice that this is a normalized cocycle since ϕ(1, f ) = ϕ(f, 1) = 0 for
all f ∈ A. We have
Z
h[ϕ], [u]i = ϕ(u, u−1 ) = u du−1 = −2πi.
S1

Alternatively, the Connes–Chern character

Ch11 ([u]) = u ⊗ u−1 ∈ HC1 (A) ' HdR

1
(S 1 )

is the class of the differential form ω = z −1 dz, representing the fundamental class
of S 1 in de Rham cohomology.

Example 0.18.3. Let A = C ∞ (S 2 ) and let e ∈ M2 (A) denote the idempotent

representing the Hopf line bundle on S 2 :
 
1 1 + x3 x1 + ix2
e= .
2 x1 − ix2 1 − x3
86 Introduction

Let us check that under the map

HC2 (A) → Ω2 S 2 , a0 ⊗ a1 ⊗ a2 7→ a0 da1 da2 ,

the Connes–Chern character of e corresponds to the fundamental class of S 2 . We

have

Ch20 (e) = Tr(e ⊗ e ⊗ e) 7→ Tr(edede)

  
1 1 + x3 x1 + ix2 dx3 dx1 + idx2
= Tr
8 x1 − ix2 1 − x3 dx1 − idx2 −dx3
 
dx3 dx1 + idx2
· .
dx1 − idx2 −dx3

Performing the computation one obtains

−i
Ch20 (e) 7→ (x1 dx2 dx3 − x2 dx1 dx3 + x3 dx1 dx2 ).
2
One can then integrate this 2-form on the two-sphere S 2 . The result is −2πi.
Notice that for the unit of the algebra 1 ∈ A, representing the trivial rank one line
bundle on S 2 , we have Ch00 (1) = 1 and Ch2n0 (1) = 0 for all n > 0. Thus e and
1 represent different K-theory classes in K0 (A). A fact which cannot be proved
using just Ch00 (e) = Tr(e) = 1.
Example 0.18.4. For smooth commutative algebras, the noncommutative Chern
character reduces to the classical Chern character. We verify this only in the
even case. The verification hinges on two things: the Chern–Weil approach to
characteristic classes via connections and curvatures, and the general fact, valid
even in the noncommutative case, that an idempotent e ∈ Mn (A) is more than
just a (noncommutative) vector bundle as it carries with it a god-given connection:

idempotent = noncommutative vector bundle + connection

Let X be a smooth closed manifold, A = C ∞ (X), and let Ω• X denote the de Rham
complex of X. The alternative definition of the classical Chern character Ch, called
the Chern–Weil theory, uses the differential geometric notions of connection and
curvature on vector bundles as we briefly recall now [61]. Let E be a complex
vector bundle on X and let ∇ be a connection on E. Thus by definition,

∇ : C ∞ (E) → C ∞ (E) ⊗A Ω1 X

is a C-linear map satisfying the Leibniz rule

∇(f ξ) = f ∇(ξ) + ξ ⊗ df

for all smooth sections ξ of E and smooth functions f on X. Let

ˆ : C ∞ (E) ⊗A Ω• X → C ∞ (E) ⊗A Ω•+1 X
∇
 0.18 Connes–Chern character in K-theory 87

denote the natural extension of ∇ to E-valued differential forms. It is uniquely

defined by virtue of the graded Leibniz rule
ˆ
∇(ξω) ˆ
= ∇(ξ)ω + (−1)deg ξ ξdω

for all ξ ∈ C ∞ (E) ⊗A Ω• X and ω ∈ Ω• X. The curvature of ∇ is the operator

ˆ 2 ∈ EndΩ• X (C ∞ (E) ⊗A Ω• X) = C ∞ (End(E)) ⊗ Ω• X,
∇

which can be easily checked to be ΩX-linear. Thus it is completely determined by

its restriction to C ∞ (E). This gives us the curvature form of ∇ as a ‘matrix-valued
2-form’
R ∈ C ∞ (End(E)) ⊗ Ω2 X.
Let
Tr : C ∞ (End(E)) ⊗A Ωev X → Ωev X
denote the canonical trace. The Chern character of E is then defined to be the
class of the non-homogeneous even form

Ch(E) = Tr(eR ).
1
(We have omitted the normalization factor of 2πi to be multiplied by R.) One
shows that Ch(E) is a closed form and that its cohomology class is independent
of the choice of connection.
Now let e ∈ Mn (C ∞ (X)) be an idempotent representing the smooth vector
bundle E on X. Smooth sections of E are in one-to-one correspondence with
smooth map ξ : X → Cn such that eξ = ξ. One can check that the following
formula defines a connection on E, called the Levi-Civita or Grassmannian con-
nection:
∇(ξ) = edξ ∈ C ∞ (E) ⊗A Ω1 X.
Computing the curvature form, we obtain
ˆ 2 (ξ) = ed(edξ) = ededξ.
R(ξ) = ∇

Differentiating the relation ξ = eξ, we obtain dξ = (de)ξ + edξ. Also, by differen-

tiating the relation e2 = e, we obtain ede · e = 0. If we use these two relations in
the above formula for R, we obtain

R(ξ) = ededeξ,

and hence the following formula for the matrix valued curvature 2-form R:

R = edede.

Using ede · e = 0, we can easily compute powers of R. They are given by

Rn = (edede)n = e |dede .{z

. . dede} .
2n
88 Introduction

The classical Chern–Weil formula for the Chern character Ch(E) is

 X Rn 
Ch(E) = Tr(eR ) = Tr ∈ Ωeven (X),
n!
n≥0

so that its n-th component is given by

Rn 1 1
Tr = Tr((edede)n ) = Tr(ede . . . de) ∈ Ω2n X.
n! n! n!
The Connes–Chern character of e defined in (46) is
−1
Ch2n
0 (e) := (n!) Tr(e ⊗ · · · ⊗ e).

We see that under the canonical map

2n
HC2n (A) → HdR (M ), a0 ⊗ · · · ⊗ a2n 7→ a0 da1 . . . da2n ,

Ch2n
0 (e) is mapped to the component of Ch(E) of degree 2n.

Example 0.18.5 (Noncommutative Chern–Weil theory). It may happen that an

element of K0 (A) is represented by a finite projective module, rather than by an
explicit idempotent. It is then important to have a formalism that would give the
value of its pairing with cyclic cocycles. This is in fact possible and is based on a
noncommutative version of Chern–Weil theory developed by Connes in [16], [19]
that we sketch next.
Let A be an algebra. By a noncommutative differential calculus on A we mean
a triple (Ω, d, ρ) such that (Ω, d) is a differential graded algebra and ρ : A → Ω0
is an algebra homomorphism. Thus

Ω = Ω0 ⊕ Ω1 ⊕ Ω1 ⊕ Ω2 ⊕ · · ·

is a graded algebra, and we assume that the differential d : Ωi → Ωi+1 increases

the degree, and d is a graded derivation in the sense that

d(ω1 ω2 ) = d(ω1 )ω2 + (−1)deg(ω1 ) ω1 d(ω2 ) and d2 = 0.

Given a differential calculus on A and a right A-module E, a connection on E

is a C-linear map
∇ : E → E ⊗A Ω1
satisfying the Leibniz rule

∇(ξa) = ∇(ξ)a + ξ ⊗ da

for all ξ ∈ E and a ∈ A.

Let
ˆ : E ⊗A Ω• → E ⊗A Ω•+1
∇
 0.18 Connes–Chern character in K-theory 89

be the (necessarily unique) extension of ∇ which satisfies the graded Leibniz rule
ˆ
∇(ξω) ˆ
= ∇(ξ)ω + (−1)deg ξ ξdω
with respect to the right Ω-module structure on E ⊗A Ω. It is defined by
ˆ ⊗ ω) = ∇(ξ)ω + (−1)deg ω ξ ⊗ dω.
∇(ξ
ˆ 2 : E ⊗A Ω• → E ⊗A Ω• , which can be
The curvature of ∇ is the operator ∇
easily checked to be Ω-linear:
ˆ 2 ∈ EndΩ (E ⊗A Ω) = EndA (E) ⊗ Ω.
∇
Let : Ω2n → C be a closed graded trace representing a cyclic 2n-cocycle ϕ
R

on A (cf. Definition 0.13.4). Now since E is finite projectiveRover A it follows that

E ⊗A Ω is finite projective Rover Ω and therefore the trace : Ω → C extends to
a trace, denoted again by , on EndA (E) ⊗ Ω. The following result of Connes
relates the value of the pairing as defined above to its value computed through the
Chern–Weil formalism: Z
1 ˆ 2n
h[ϕ], [E]i = ∇
n!
The next example is a concrete illustration of this method.
Example 0.18.6. Let E = S(R) denote the Schwartz space of rapidly decreasing
functions on the real line. The operators
(ξ · U )(x) = ξ(x + θ), (ξ · V )(x) = e2πix ξ(x)
turn S(R) into a right Aθ -module for all ξ ∈ S(R). It is the simplest of a series
of modules Ep,q on the noncommutative torus, defined by Connes in [16]. It turns
out that E is finite projective, and for the canonical trace τ on Aθ we have
hτ, Ei = −θ.
In Example 0.13.4 we defined a differential calculus, in fact a 2-cycle, on Aθ . It
is easy to see that a connection on ∇ : E → E ⊗A Ω1 with respect to this calculus
is simply given by a pair of operators ∇1 , ∇2 : E → E (‘covariant derivatives’ with
respect to noncommutative vector fields δ1 and δ2 ) satisfying
∇j (ξa) = (∇j ξ)a + ξδj (a), j = 0, 1,
for all ξ ∈ E and a ∈ Aθ .
One can now check that the following formulae define a connection on E [16],
[21]:
s dξ
∇1 (ξ)(s) = − ξ(s), ∇2 (ξ)(s) = (s).
θ ds
The curvature of this connection is constant and is given by
1
∇2 = [∇1 , ∇2 ] = id ∈ EndAθ (E).
θ
90 Introduction

Exercise 0.18.1. Let E be a finite projective right A-module. Show that EndA (E) '
E ⊗A E ∗ where A∗ = HomA (E, A). The canonical pairing E ⊗A E ∗ → A defined
by ξ ⊗ f 7→ f (ξ) induces a map EndA (E) → A/[A, A] = HC0 (A). In particular if
τ : A → C is a trace on A, the induced trace on EndA (E) is simply obtained by
composing τ with the canonical pairing between E and E ∗ .
Exercise 0.18.2. Verify that under the natural quasi-isomorphism between (b, B)
and cyclic complexes, formula (43) corresponds to (42).
Exercise 0.18.3. Show that a right A-module E admits a connection with respect
to the universal differential calculus (ΩA, d), if and only if E is projective.

0.19 Connes–Chern character in K-homology

By K-homology for spaces we mean the theory which is dual to topological K-
theory. While such a theory can be constructed using general techniques of al-
gebraic topology, a beautiful and novel idea of Atiyah [2] (in the even case), and
Brown–Douglas–Fillmore [10] (in the odd case) was to use techniques of index
theory, functional analysis and operator algebras to define a K-homology theory
(cf. also [4]). What is even more interesting is that the resulting theory can be
extended to noncommutative algebras and pairs with K-theory. This extension,
in full generality, then paved the way for Kasparov’s bivariant KK-theory which
unifies both K-theory and K-homology into a single bivariant theory (cf. [50] and
[7]). Unfortunately the name K-homology is used even when one is dealing with
algebras, despite the fact that the resulting functor is in fact contravariant for
algebras, while K-theory for algebras is covariant. We hope this will cause no
confusion for the reader.
To motivate the discussions, we start this section by recalling the notion of an
abstract elliptic operator over a compact space [2]. This will then be extended to
the noncommutative setting by introducing the notion of an, even or odd, Fredholm
module over an algebra [19]. The Connes–Chern character of a Fredholm module is
introduced next. We shall then define the index pairing between K-theory and K-
homology, which indicates the sense in which these theories are dual to each other.
The final result of this section is an index formula of Connes which computes the
index pairing in terms of Connes–Chern characters for K-theory and K-homology.
Let X be a compact Hausdorff space. The even cycles for Atiyah’s theory in [2]
are abstract elliptic operators (H, F ) over C(X). This means that H = H + ⊕ H −
is a Z2 -graded Hilbert space, π : C(X) → L(H) with
 + 
π (a) 0
π(a) =
0 π − (a)

is an even representation of C(X) in the algebra of bounded operators on H, and

F : H → H with  
0 Q
F = (47)
P 0
 0.19 Connes–Chern character in K-homology 91

is an odd bounded operator with F 2 − I ∈ K(H) being compact. This data must
satisfy the crucial condition

[F, π(a)] ∈ K(H)

for all a ∈ C(X). We shall make no attempt at turning these cycles into a homology
theory. Suffice it to say that the homology theory is defined as the quotient of the
set of these cycles by a homotopy equivalence relation (cf. [44], [32] for a recent
account).
When X is a smooth closed manifold, the main examples of abstract elliptic
operators in the above sense are given by elliptic pseudodifferential operators of
order 0, D : C ∞ (E + ) → C ∞ (E − ) acting between sections of vector bundles E +
and E − on X. Let P : H + → H − denote the natural extension of D to a bounded
operator where H + = L2 (E + ) and H − = L2 (E − ), and let Q : H − → H + denote
a parametrix of P . Define F by (47). Then with C(X) acting as multiplication
operators on H + and H − , basic elliptic theory shows that (H, F ) is an elliptic
operator in the above sense on C(X).
If e ∈ Mn (C(X)) is an idempotent representing a vector bundle on X, then
the formula
h(H, F ) , [e]i := index Fe+
with the Fredholm operator Fe+ := eF e : eH + → eH − can be shown to define a
pairing between the K-theory of X and abstract elliptic operators on X. This is
the duality between K-homology and K-theory.
A modification of the above notion of abstract elliptic operator, which makes
sense over noncommutative algebras, both in the even and odd case, is the following
notion of Fredholm module in [19] which is an important variation of a related
notion from [2], [10], [51] (cf. the remark below).

Definition 0.19.1. An odd Fredholm module over an algebra A is a pair (H, F )

where
1) H is a Hilbert space endowed with a representation

π : A → L(H);

2) F ∈ L(H) is a bounded selfadjoint operator with F 2 = I;

3) for all a ∈ A we have

[F, π(a)] = F π(a) − π(a)F ∈ K(H). (48)

For 1 ≤ p < ∞, let Lp (H) denote the Schatten ideal of p-summable operators.
A Fredholm module (H, F ) is called p-summable if, instead of (48), we have the
stronger condition:
[F, π(a)] ∈ Lp (H) (49)
for all a ∈ A. Since Lp (H) ⊂ Lq (H) for p ≤ q, a p-summable Fredholm module is
clearly q-summable for all q ≥ p.
92 Introduction

Definition 0.19.2. An even Fredholm module over an algebra A is a triple

(H, F, γ) such that (H, F ) is a Fredholm module over A in the sense of the above
definition and γ : H → H is a bounded selfadjoint operator with γ 2 = I and such
that
F γ = −γF, π(a)γ = γπ(a) (50)
for all a ∈ A.

Let H + and H − denote the +1 and −1 eigenspaces of γ. They define an

orthogonal decomposition H = H + ⊕ H − . With respect to this decomposition,
equations (50) are equivalent to saying that π is an even representation and F is
an odd operator, so that we can write
 +   
π (a) 0 0 Q
π(a) = and F = ,
0 π − (a) P 0

where π + and π − are representations of A on H + and H − , respectively. The notion

of a p-summable even Fredholm module is defined as in the odd case above.

Remark 4. Notice that in the preceding example with F defined by (47), and in
general in [2], [10], [51], the condition F 2 = I for a Fredholm module only holds
modulo compact operators. Similarly for the two equalities in equation (50). It
is shown in [21] that with simple modifications one can replace such (H, F ) by an
equivalent Fredholm module in which these equations, except (48), hold exactly.
The point is that as far as pairing with K-theory is concerned the set up in [2],
[10], [51] is enough. It is for the definition of the Chern character and pairing with
cyclic cohomology that one needs the exact equalities F 2 = I and (50), as well as
the finite summability assumption (49).

Let (H, F ) be an odd p-summable Fredholm module over an algebra A and let
n be an integer such that 2n ≥ p. To simplify the notation, from now on in our
formulae the operator π(a) will be denoted by a. Thus an expression like a0 [F, a1 ]
stands for the operator π(a0 )[F, π(a1 )], etc. We define a (2n − 1)-cochain on A by

ϕ2n−1 (a0 , a1 , . . . , a2n−1 ) = Tr(F [F, a0 ][F, a1 ] . . . [F, a2n−1 ]) (51)

where Tr denotes the operator trace. Notice that by our p-summability assump-
tion, each commutator is in Lp (H) and hence, by Hölder inequality for Schatten
class operators (cf. Appendix B), their product is in fact a trace class operator as
soon as 2n ≥ p.

Proposition 0.19.1. ϕ2n−1 is a cyclic (2n − 1)-cocycle on A.

Proof. For a ∈ A, let da := [F, a]. The following relations are easily established:
for all a, b ∈ A we have

d(ab) = ad(b) + da · b and F da = −da · F (52)

 0.19 Connes–Chern character in K-homology 93

Notice that for the second relation the assumption F 2 = 1 is essential. Now ϕ2n−1
can be written as

ϕ2n−1 (a0 , a1 , . . . , a2n−1 ) = Tr(F da0 da1 . . . da2n−1 ),

and therefore
2n
X 
(bϕ2n−1 )(a0 , . . . , a2n ) = Tr (−1)i F da0 . . . d(ai dai+1 ) . . . da2n
i=0
+ (−1)2n+1 Tr(F d(a2n a0 )da1 . . . da2n−1 ).

Using the derivation property of d, we see that most of the terms cancel and we
are left with just four terms

= Tr(F a0 da1 . . . da2n ) − Tr(F da0 . . . da2n−1 a2n )

+ Tr(F a2n da0 . . . da2n−1 ) + Tr(F da2n a0 da1 . . . da2n−1 ).

Using the relation F da = −da · F and the trace property of Tr we see that the
second and third terms cancel. By the same argument the first and last terms
cancel as well. This shows that ϕ2n−1 is a Hochschild cocycle.
To check the cyclic property of ϕ2n−1 , again using the relation F da = −da · F ,
and the trace property of Tr, we have

ϕ2n−1 (a2n−1 , a0 , . . . , a2n−2 ) = Tr(F da2n−1 da0 . . . da2n−2 )

= − Tr(da2n−1 F da0 . . . da2n−2 )
= − Tr(F da0 . . . da2n−2 da2n−1 )
= −ϕ2n−1 (a0 , . . . , a2n−2 , a2n−1 ). 2

Notice that if 2n ≥ p − 1, then the cyclic cocycle (51) can be written as

ϕ2n−1 (a0 , a1 , . . . , a2n−1 ) = 2 Tr(a0 [F, a1 ] . . . [F, a2n−1 ])

= 2 Tr(a0 da1 . . . da2n−1 )

which looks remarkably like a noncommutative analogue of the integral

Z
f0 df1 . . . df2n−1 .
M

Now the products [F, a0 ][F, a1 ] . . . [F, a2m−1 ] are trace class for all m ≥ n.
Therefore we obtain a sequence of odd cyclic cocycles

ϕ2m−1 (a0 , a1 , . . . , a2m−1 ) = Tr(F [F, a0 ][F, a1 ] . . . [F, a2m−1 ]), m ≥ n.

The next proposition shows that these cyclic cocycles are related to each other via
the periodicity S-operator of cyclic cohomology:
94 Introduction

Proposition 0.19.2. For all m ≥ n we have

 
1
Sϕ2m−1 = − m + ϕ2m+1 .
2
Proof. Let us define a 2m-cochain ψ2m on A by the formula
ψ2m (a0 , a1 , . . . , a2m ) = Tr(F a0 da1 . . . da2m ).
We claim that
1
Bψ2m = (2m) ϕ2m−1 and bψ2m = − ϕ2m+1 .
2
In fact, we have Bψ2m = N B0 ψ2m , where
B0 ψ2m (a0 , . . . , a2m−1 ) = ψ2m (1, a0 , . . . , a2m−1 )
− (−1)2m−1 ψ2m (a0 , . . . , a2m−1 , 1)
= Tr(F da0 . . . da2m−1 ),
and hence
(N B0 )ψ2m (a0 , . . . , a2m−1 ) = Tr(F da0 . . . da2m−1 ) − Tr(F da2m−1 da0 . . . da2m−2 )
+ · · · − Tr(F da1 . . . da0 )
= (2m) Tr(F da0 . . . da2m−1 )
= (2m)ϕ2m−1 (a0 , . . . , a2m−1 )
and
(bψ2m )(a0 , . . . , a2m+1 ) = Tr(F a0 a1 da2 . . . da2m+1 ) − Tr(F a0 d(a1 a2 ) . . . da2m+1 )
+ · · · + Tr(F a0 da1 . . . d(a2m a2m+1 ))
− Tr(F a2m+1 a0 da1 . . . da2m ).
After cancelations, only two terms remain which can be collected into a single
term:
= Tr(F a0 da1 . . . da2m · a2m+1 ) − Tr(F a2m+1 a0 da1 . . . da2m )
1
= − Tr(F da0 da1 . . . da2m+1 )
2
1
= − ϕ2m+1 (a0 , . . . , a2m+1 ).
2
The above computation shows that
1
bB −1 ϕ2m−1 = − ϕ2m+1 .
2(2m)
Now using formula (31) for the operator S, we have
 
−1 1
Sϕ2m−1 = (2m)(2m + 1)bB ϕ2m−1 =− m+ ϕ2m+1 .
2
 0.19 Connes–Chern character in K-homology 95

The odd Connes–Chern characters Ch2m−1 = Ch2m−1 (H, F ), are defined by

rescaling the cocycles ϕ2m−1 appropriately. Let

Ch2m−1 (a0 , . . . , a2m−1 )

 
m 1 1
:= (−1) 2 m − . . . Tr(F [F, a0 ][F, a1 ] . . . [F, a2m−1 ]).
2 2
The following is an immediate corollary of the above proposition:
Corollary 0.19.1. We have

S(Ch2m−1 ) = Ch2m+1 for all m ≥ n.

Definition 0.19.3. The Connes–Chern character of an odd p-summable Fred-

holm module (H, F ) over an algebra A is the class of the cyclic cocycle Ch2m−1
in the odd periodic cyclic cohomology group HP odd (A).
By the above corollary, the class of Ch2m−1 in HP odd (A) is independent of the
choice of m.
Example 0.19.1. Let A = C(S 1 ) (S 1 = R/Z) act on H = L2 (S 1 ) as multipli-
cation operators. Let F (en ) = en if n ≥ 0 and F (en ) = −en for n < 0, where
en (x) = e2πinx , n ∈ Z, denotes the standard basis of H. Clearly F is selfadjoint
and F 2 = I. ToP show that [F, π(f )] is a compact operator for all f ∈ C(S 1 ), no-
tice that if f = |n|≤N an en is a finite trigonometric sum then [F, π(f )] is a finite
rank operator and hence is compact. In general we can uniformly approximate
a continuous function by a trigonometric sum and show that the commutator is
compact for any continuous f . This shows that (H, F ) is an odd Fredholm module
over C(S 1 ).
This Fredholm module is not p-summable for any 1 ≤ p < ∞. If we restrict it
to the subalgebra C ∞ (S 1 ) of smooth functions, then it can be checked that (H, F )
is in fact p-summable for all p > 1, but is not 1-summable even in this case.
Let us compute the Chern character of this Fredholm module with A = C ∞ (S 1 ).
By the above definition, Ch1 (H, F ) = [ϕ1 ] is the class of the following cyclic 1-
cocycle in HP odd (A):

ϕ1 (f0 , f1 ) = Tr(F [F, f0 ][F, f1 ]),

and the question is if we can identify this cocycle with some local formula. We
claim that
Z
4
Tr(F [F, f0 ][F, f1 ]) = f0 df1 for all f0 , f1 ∈ A.
2πi
To verify the claim, it suffices to check it for the basis elements f0 = em , f1 = en
for all m, n ∈ Z. The right-hand side is easily computed:
(
if m + n 6= 0,
Z
4 0
em den =
2πi 4n if m + n = 0.
96 Introduction

To compute the left-hand side, notice that



 0 if k ≥ 0, n+k ≥ 0,

−2e
n+k if k ≥ 0, n+k < 0,
[F, en ](ek ) =


 2e n+k if k < 0, n+k ≥ 0,
0 if k < 0, n+k < 0.


From this we conclude that F em [F, en ] = 0 if m + n P6= 0. To compute the operator

trace for m + n = 0, we use the formula Tr(T ) = k hT (ek ), ek i for the trace of a
trace class operator T . Using the above information we have


 0 if k ≥ 0, n + k ≥ 0,

4e if k ≥ 0, n + k < 0,
k
F [F, e−n ][F, en ](ek ) =


 4e k if k < 0, n + k ≥ 0,
0 if k < 0, n + k < 0,


from which we readily obtain

Tr(F [F, e−n ][F, en ]) = 4n for all n ∈ Z.

This finishes the proof.

Next we turn to the even case. Let (H, F, γ) be an even p-summable Fredholm
module over an algebra A and let n be an integer such that 2n + 1 ≥ p. Define a
2n-cochain on A by the formula

ϕ2n (a0 , a1 , . . . , a2n ) = Tr(γF [F, a0 ][F, a1 ] . . . [F, a2n ]). (53)

Proposition 0.19.3. ϕ2n is a cyclic 2n-cocycle on A.

Proof. The proof is similar to the odd case and is left to the reader. Apart from
relations (52), one needs the auxiliary relation γ da = −da γ for the proof as
well.
Notice that if we have the stronger condition 2n ≥ p, then the cyclic cocycle
(53) can be written as

ϕ2n (a0 , a1 , . . . , a2n ) = Tr(γa0 [F, a1 ] . . . [F, a2n ])

= Tr(γa0 da1 . . . da2n ).

As in the odd case, we obtain a sequence of even cyclic cocycles ϕ2m , defined
by
ϕ2m (a0 , a1 , . . . , a2m ) = Tr(γF [F, a0 ][F, a1 ] . . . [F, a2m ]), m ≥ n.
Proposition 0.19.4. For all m ≥ n we have

Sϕ2m = −(m + 1)ϕ2m+2 .

 0.19 Connes–Chern character in K-homology 97

Proof. Define a (2m + 1)-cochain ψ2m+1 on A by the formula

ψ2m+1 (a0 , a1 , . . . , a2m+1 ) = Tr(γ F a0 da1 . . . da2m+1 ).

The following relations can be proved as in the odd case:

1
Bψ2m+1 = (2m + 1)ϕ2m and bψ2m+1 = − ϕ2m+2 .
2
It shows that
1
bB −1 ϕ2m = − ϕ2m+2 ,
2(2m + 1)
so that using formula (31) for the operator S we obtain

Sϕ2m = (2m + 1)(2m + 2)bB −1 ϕ2m = −(m + 1)ϕ2m+2 .

The even Connes–Chern characters Ch2m = Ch2m (H, F, γ) are now defined
by rescaling the even cyclic cocycles ϕ2m :

(−1)m m!
Ch2m (a0 , a1 , . . . , a2m ) := 2 Tr(γF [F, a0 ][F, a1 ] . . . [F, a2m ]) (54)

The following is an immediate corollary of the above proposition:

Corollary 0.19.2. We have

S(Ch2m ) = Ch2m+2 for all m ≥ n.

Definition 0.19.4. The Connes–Chern character of an even p-summable Fred-

holm module (H, F, γ) over an algebra A is the class of the cyclic cocycle Ch2m (H, F, γ)
in the even periodic cyclic cohomology group HP even (A).
By the above corollary, the class of Ch2m in HP even (A) is independent of m.
Example 0.19.2 (A noncommutative example). Following [19], we construct an
even Fredholm module over A = Cr∗ (F2 ), the reduced group C ∗ -algebra of the free
group on two generators. This Fredholm module is not p-summable for any p, but
by restricting it to a properly defined dense subalgebra of A (which plays the role
of ‘smooth functions’ on the underlying noncommutative space), we shall obtain
a 1-summable Fredholm. We shall also identify the character of this 1-summable
module. It is known that a group is free if and only if it has a free action on a
tree. Let then T be a tree with a free action of F2 , and let T 0 and T 1 denote the
set of vertices and 1-simplices of T , respectively. Let

H + = `2 (T 0 ) and H − = `2 (T 1 ) ⊕ C,

and let the canonical basis of `2 (T 0 ) (resp. `2 (T 1 )) be denoted by εq , q ∈ T 0 (resp.

q ∈ T 1 ). Fixing a vertex p ∈ T 0 , we can define a one-to-one correspondence

ϕ : T 0 − {p} → T 1
98 Introduction

by sending q ∈ T 0 − {p} to the unique 1-simplex containing q and lying between

p and q. This defines a unitary operator P : H + → H − by

P (εq ) = εϕ(q) if q 6= p, and P (εp ) = (0, 1).

The action of F2 on T 0 and T 1 induces representations of Cr∗ (F2 ) on `2 (T 0 ) and

`2 (T 1 ) and on H − = `2 (T 1 ) ⊕ C by the formula a(ξ, λ) = (aξ, 0). Let H =
H + ⊕ H − and
0 P −1
   
1 0
F = , γ= .
P 0 0 −1
To check that (H, F, γ) is a Fredholm module over A we need to verify that
[F, a] ∈ K(H) for all a ∈ Cr∗ (F2 ). Since the group algebra CF2 is dense in Cr∗ (F2 ),
it suffices to check that for all a = g ∈ F2 , the commutator [F, g] is a finite rank
operator. This in turn is a consequence of the easily established fact that for all
g ∈ F2 ,
ϕ(gq) = gϕ(q) for all q 6= g −1 p.
In fact, for q ∈ T 0 we have

[F, g](εq ) = F (εgq ) − gF (εq ) = εϕ(gq) − εgϕ(q) = 0

6 g −1 p, and [F, g](εg−1 p ) = (0, 1) − gεϕ(g−1 p) . A similar argument works for

if q =
the basis elements εq , q ∈ T 1 . This shows that [F, g] is a rank one operator.
Let
A = {a ∈ A; [F, a] ∈ L1 (H)}.
Using the relation [F, ab] = a[F, b] + [F, a]b, it is clear that A is a subalgebra of A.
It is also dense in A as it contains the group algebra CF2 . Though we do not need
it now, it can also be shown that A is stable under holomorphic functional calculus
and in particular the inclusion A ⊂ A induces an isomorphism K0 (A) → K0 (A)
in K-theory (cf. Section 4.3 for more on this). By its very definition, we now have
an even 1-summable Fredholm module (H, F, γ) over A and it remains to compute
its character. Let τ : A → C denote the canonical trace on A. We claim that
1
Tr(γF [F, a]) = τ (a) for all a ∈ A, (55)
2
so that
Ch(H, F, γ)(a) = τ (a).
To verify the claim, notice that

a − P −1 aP
 
0
γF [F, a] = ,
0 −a + P aP −1

so that
1
Tr(γF [F, a]) = Tr(a − P −1 aP ).
2
 0.19 Connes–Chern character in K-homology 99

Now for the operator a − P −1 aP : H + → H + we have

haεq , εq i = τ (a) for all q ∈ T 0 ,

and
h(P −1 aP )(εp ), εp i = 0 and h(P −1 aP )(εq ), εq i = τ (a)
for all q 6= p, from which (55) follows.
Our next goal is to define the index pairing between Fredholm modules over A
and the K-theory of A. Notice that for this we do not need to assume that the
Fredholm module is finitely summable. We start with the even case. Let (H, F, γ)
be an even Fredholm module over an algebra A and let e ∈ A be an idempotent.
Let
Fe+ : eH0 → eH1
denote the restriction of the operator eF e to the subspace eH0 . It is a Fredholm
operator. To see this, let Fe− : eH1 → eH0 denote the restriction of eF e to the
subspace eH1 . We claim that the operators Fe+ Fe− −1 and Fe− Fe+ −1 are compact.
Atkinson’s theorem then shows that Fe+ is Fredholm. The claim follows from the
following computation:

eF eeF e = e(F e − eF + eF )F e = e[F, e]F e + e,

and the fact that [F, e] is a compact operator.

For an idempotent e ∈ A let us define a pairing:

h(H, F, γ), [e]i := index Fe+

More generally, if e ∈ Mn (A) is an idempotent in the algebra of n by n matrices

over A, we define
h(H, F, γ), [e]i := h(H n , F n , γ), [e]i,
where (H n , F n ) is the n-fold inflation of (H, F ) defined by H n = H ⊗ Cn , F n =
F ⊗ ICn . It is easily seen that (H n , F n ) is a Fredholm module over Mn (A) and if
(H, F ) is p-summable then so is (H n , F n ). It is easily checked that the resulting
map is additive with respect to direct sum of idempotents and is conjugation
invariant. This shows that each even Fredholm module, which need not be finitely
summable, induces an additive map on K-theory:

h(H, F, γ), −i : K0 (A) → C.

There is a similar index pairing between odd Fredholm modules over A and
the algebraic K-theory group K1alg (A). Let (H, F ) be an odd Fredholm module
over A and let U ∈ A× be an invertible element in A. Let P = F 2+1 : H → H be
the projection operator defined by F . Let us check that the operator

PUP : PH → PH
100 Introduction

is a Fredholm operator. Again the proof hinges on Atkinson’s theorem and noticing
that P U −1 P is an inverse for P U P modulo compact operators. We have
P U P P U −1 P − IP H = P U P U −1 P − IP H = P (U P − P U + P U )U −1 P − IP H
1
= P [U, P ]U −1 P + P − IP H = P [U, F ]U −1 P.
2
But [F, U ] is a compact operator by our definition of Fredholm modules and hence
the last term is compact too. Similarly one checks that P U −1 P P U P − IP H is a
compact operator as well. We can thus define the index pairing:

h(H, F ), [U ]i := index (P U P )

If the invertible U happens to be in Mn (A) we can apply the above definition to

the n-fold iteration of (H, F ), as in the even case above, to define the pairing.
The resulting map can be shown to induce a well-defined additive map
h(H, F ), −i : K1alg (A) → C.
Example 0.19.3. Let (H, F ) be the Fredholm module of Example ?? and let
f ∈ C(S 1 ) be a nowhere zero continuous function on S 1 representing an ele-
ment of K1alg (C(S 1 )). We want to compute the index pairing h[(H, F )], [f ]i =
index(P f P ). The operator P f P : H + → H + is called a Toeplitz operator. The
following standard result, known as the Gohberg–Krein index theorem, computes
the index of a Toeplitz operator in terms of the winding number of f :
h[(H, F )], [f ]i = index(P f P ) = −W (f, 0).
To prove this formula notice that both sides are homotopy invariant. For the left-
hand side this is a consequence of the homotopy invariance of the Fredholm index
while for the right-hand side it is a standard fact about the winding number. Also,
both sides are additive. Therefore it suffices to show that the two sides coincide
on the generator of π1 (S 1 ), i.e., for f (z) = z. Then P zP is easily seen to be the
forward shift operator given by P zP (en ) = en+1 in the given basis. Clearly then
index(P zP ) = −1 = −W (z, 0).
When f is smooth we have the following well-known formula for the winding
number: Z
1 1
W (f, 0) = f −1 df = ϕ(f −1 , f ),
2πi 2πi
where ϕ is the cyclic 1-cocycle on C ∞ (S 1 ) defined by ϕ(f, g) = f dg. Since this
R

cyclic cocycle is the Connes–Chern character of the Fredholm module (H, F ), the
above equation can be written as
1
h[(H, F )], [f ]i = hChodd (H, F ), Chodd (f )i,
2πi
where the pairing on the right-hand side is between cyclic cohomology and homol-
ogy. As we shall prove next, this is a special case of a very general index formula
of Connes.
 0.20 Connes’ index theorem 101

0.20 Connes’ index theorem

Now what makes the Connes–Chern character in K-homology useful is the fact
that it can capture the analytic index by giving a topological formula for the index.
More precisely we have the following index formula due to Connes [19]:

Theorem 0.20.1. Let (H, F, γ) be an even p-summable Fredholm module over A

and n be an integer such that 2n + 1 ≥ p. If e is an idempotent in A then

(−1)n
index(Fe+ ) = ϕ2n (e, e, . . . , e),
2
where the cyclic 2n-cocycle ϕ2n is defined by

ϕ2n (a0 , a1 , . . . , a2n ) = Tr(γF [F, a0 ][F, a1 ] . . . [F, a2n ]).

Proof. We use the following fact from the theory of Fredholm operators (cf. Propo-
sition A.2 for a proof): let P 0 : H 0 → H 00 be a Fredholm operator and let Q0 : H 00 →
H 0 be such that for an integer n ≥ 0, 1 − P 0 Q0 ∈ Ln+1 (H 00 ) and 1 − Q0 P 0 ∈
Ln+1 (H 0 ). Then

index (P 0 ) = Tr(1 − Q0 P 0 )n+1 − Tr(1 − P 0 Q0 )n+1 .

We can also write the above formula as a supertrace

index(P 0 ) = Tr(γ 0 (1 − F 02 )n+1 ), (56)

where the operators F 0 and γ 0 acting on H 0 ⊕ H 0 are defined by

0 Q0
   
1 0
F0 = and γ 0
= .
P0 0 0 −1

We apply this result to P 0 = Fe+ , Q0 = Fe− , H 0 = e0 H0 and H 00 = e1 H1 . By

our summability assumption, both operators 1 − P 0 Q0 and 1 − Q0 P 0 are in Ln+1
and we can apply (56). We have

index(Fe+ ) = Tr(γ 0 (1 − F 02 )n+1 ) = Tr(γ (e − (eF e)2 )n+1 ).

As in the proof of Proposition 0.19.1 let de := [F, e]. Using the relations e2 = e,
ede · e = 0, and edede = de · de · e, we have

e − (eF e)2 = eF (F e − eF )e = (eF − F e + F e)(F e − eF )e = −edede

and hence (e − (eF e)2 )n+1 = (−1)n+1 (edede)n+1 = (−1)n+1 e(de)2n+2 . Thus the
index can be written as

index(Fe+ ) = (−1)n+1 Tr(γe(de)2n+2 ).

102 Introduction

On the other hand, using de = ede + de · e, we have

ϕ2n (e, e, . . . , e) = Tr(γF (de)2n+1 )

= Tr(γ(F e)e(de)2n+1 + γF de · ee(de)2n
= Tr(F e − eF + eF )e(de)2n+1 ) + γ(eF − F e + F e)de · e(de)2n
= Tr(γde · e(de)2n+1 − Tr(γdede · e(de)2n )
= − Tr(γdede · e(de)2n ) − Tr(γdede · e(de)2n )
= −2 Tr(γe(de)2n+2 )

which of course proves the theorem. In the above computation we used the fact
that Tr(γeF e(de)2n+1 ) = Tr(γF ede · e(de)2n ) = 0.

Using the pairing HC 2n (A) ⊗ K 0 (A) → C between cyclic cohomology and

K-theory defined in (42), and the definition of the Connes–Chern character of
(H, F, γ) in (54), the above index formula can be written as

index(Fe+ ) = hCh2n (H, F, γ), [e]i,

or in its stable form

index(Fe+ ) = hCheven (H, F, γ), [e]i.

There is yet another way to interpret the index formula as

h(H, F, γ), [e]i = hCh2n (H, F, γ), Ch2n [e]i,

where on the left-hand side we have the pairing between K-homology and K-theory
and on the right-hand side the pairing between cyclic cohomology and homology.
The corresponding index formula in the odd case is as follows:

Proposition 0.20.1. Let (H, F ) be an odd p-summable Fredholm module over A

and let n be an integer such that 2n ≥ p. If u is an invertible element in A, then

(−1)n
index(P uP ) = ϕ2n−1 (u−1 , u, . . . , u−1 , u),
22n

where the cyclic cocycle ϕ2n−1 is defined by

ϕ2n−1 (a0 , a1 , . . . , a2n−1 ) = Tr(F [F, a0 ][F, a1 ] . . . [F, a2n−1 ]).

Proof. Let P = 1+F

2 , H 0 = P H, P 0 = P uP : H 0 → H 0 , Q0 = P u−1 P : H 0 → H 0 ,
 0.20 Connes’ index theorem 103

and du = [F, u]. We have

1 − Q0 P 0 = 1 − P u−1 P P uP
= 1 − pu−1 (P u − uP + uP )P
= 1 − P u−1 [P, u]P − P
1
= − P u−1 du · P
2
1 1
= P du−1 · uP = P du−1 (uP − P u + P u)
2 2
1
= − P du du + P du−1 P u
−1
4
1
= − P du−1 du,
4
where in the last step the relation P du−1 P = 0 was used. This relation follows
from
[P, u−1 ] = [P 2 , u−1 ] = P [P, u−1 ] + [P, u−1 ]P.
Since, by our summability assumption, du = [F, u] ∈ L2n (H) and similarly du−1 ∈
L2n (H), we have 1 − Q0 P 0 ∈ Ln (H 0 ).
A similar computation shows that

1
1 − P 0 Q0 = − P dudu−1 ,
4
and hence 1 − P 0 Q0 ∈ Ln (H 0 ).
Using formula (56) for the index, we obtain

index(P uP ) = Tr((1 − Q0 P 0 )n ) − Tr((1 − P 0 Q0 )n )

(−1)n −1 (−1)n
= Tr((P du du) n
) − Tr((P dudu−1 )n )
22n 22n
(−1)n −1 (−1)n
= Tr(P (du du) n
) − Tr(P (dudu−1 )n )
22n 2 2n

(−1)n (−1)n
   
1+F −1 n 1+F −1 n
= Tr (du du) − Tr (dudu )
22n 2 22n 2
n
(−1)
= Tr(F (du−1 du)n )
22n
(−1)n
= ϕ2n−1 (u−1 , u, . . . , u−1 , u),
22n
where in the last step we used the relations

Tr((du−1 du)n ) = Tr((dudu−1 )n ) and Tr(F (du−1 du)n ) = − Tr(F (dudu−1 )n ).

104 Introduction

Using the pairing HC 2n−1 (A)⊗K1alg (A) → C and the definition of Ch2n−1 (H, F ),
the above index formula can be written as

index(P uP ) = hCh2n−1 (H, F ), [u]i,

or in its stable form

index(P uP ) = hChodd (H, F ), [u]i.

There is yet another way to interpret the index formula as

h(H, F ), [u]i = hCh2n−1 (H, F ), Ch2n−1 [u]i,

where on the left-hand side we have the pairing between K-homology and K-theory
and on the right-hand side the pairing between cyclic cohomology and homology.
Example 0.20.1 (A noncommutative connected space). A projection in a ∗-
algebra is an element e satisfying e2 = e = e∗ . It is called a trivial projection
if e = 0 or e = 1. It is clear that a compact space X is connected if and only if the
algebra C(X) has no non-trivial projections. Let us agree to call a noncommutative
space represented by a C ∗ -algebra A connected if A has no non-trivial projections.
The Kadison conjecture states that the reduced group C ∗ -algebra of a torsion-free
discrete group is connected. This conjecture, in its full generality, is still open
although it has now been verified for various classes of groups [?]. Methods of
noncommutative geometry play an important role in these proofs. The validity of
the conjecture for free groups was first established by Pimsner and Voiculescu [?]
using techniques of K-theory. Here we reproduce Connes’ proof of this conjecture
for free groups. Note that the conjecture is obviously true for the finitely generated
free abelian groups Zn , since by Fourier theory, or the Gelfand–Naimark theorem,
C ∗ (Zn ) ' C(Tn ), and the n-torus T n is of course connected.
Let τ : Cr∗ (F2 ) → C be the canonical normalized trace. It is positive and
faithful in the sense that for all a ∈ A, τ (aa∗ ) ≥ 0 and τ (aa∗ ) = 0 if and only if
a = 0. Thus if we can show that for a projection e, τ (e) is an integer then we can
deduce that e = 0 or e = 1. In fact since e is a projection we have 0 ≤ e ≤ 1 and
therefore 0 ≤ τ (e) ≤ 1, and by integrality we have τ (e) = 1 or τ (e) = 0. Since τ
is faithful from 0 = τ (e) = τ (ee∗ ) we have e = 0. A similar argument works for
τ (e) = 1.
Now the proof of the integrality of τ (e) is based on Connes’ index formula in
Theorem 0.20.1 and is remarkably similar to proofs of classical integrality theorems
for characteristic numbers in topology using an index theorem: to show that a
number τ (e) is an integer it suffices to shows that it is the index of a Fredholm
operator. Let (H, F, γ) be the even 1-summable Fredholm module over the dense
subalgebra A ⊂ Cr∗ (F2 ) defined in Example 0.19.2. The index formula combined
with (55), shows that if e ∈ A is a projection then
1
τ (e) = Tr(γF e[F, e]) = index(Fe+ ) ∈ Z
2
 0.20 Connes’ index theorem 105

is an integer and we are done. To prove the integrality result for idempotents in
A which are not necessarily in A, we make use of the fact that A is stable under
holomorphic functional calculus. Let e ∈ A be an idempotent. For any  > 0
there is an idempotent e0 ∈ A such that ke − e0 k < . In fact, since A is dense in
A we can first approximate it by an element g ∈ A. Since sp(e) ⊂ {0, 1}, sp(g)
is concentrated around 0 and 1. Let f be a holomorphic function defined on an
open neighborhood of sp(g) which is identically equal to 0 around 0 and identically
equal to 1 around 1. Then
Z
1
e0 = f (g) = f (z)(z1 − e)−1 dz,
2πi γ

is an idempotent in A which is close to e. As we showed before (cf. formula (B.1)),

close idempotents are equivalent in the sense that e = ue0 u−1 for an appropriate
u ∈ A. In particular we conclude that τ (e) = τ (ue0 u−1 ) = τ (e0 ) is an integer.

In connection with Exercise 0.20.6 it is appropriate to mention that there is a

refinement of the notion of Fredholm module to that of a spectral triple that plays a
very important role in further developments of noncommutative geometry. Broadly
speaking, going from Fredholm modules to spectral triples is like passing from
the conformal class of a metric to the Riemannian metric itself. Spectral triples
simultaneously provide a notion of Dirac operator in noncommutative geometry,
as well as a Riemannian type distance function for noncommutative spaces.
To motivate the definition of a spectral triple, we recall that the Dirac operator
D/ on a compact Riemannian spinc manifold acts as an unbounded selfadjoint
operator on the Hilbert space L2 (M, S) of L2 -spinors on the manifold M . If we let
C ∞ (M ) act on L2 (M, S) by multiplication operators, then one can check that for
any smooth function f , the commutator [D, f ] = Df − f D extends to a bounded
operator in L2 (M, S). Now the geodesic distance d on M can be recovered from
the following distance formula of Connes [21]:

d(p, q) = sup{|f (p) − f (q)|; k [D, f ] k≤ 1} for all p, q ∈ M. (57)

The triple (C ∞ (M ), L2 (M, S), D

/ ) is a commutative example of a spectral triple.
In general, in the odd case, a spectral triple is a triple (A, H, D), where A is a
∗-algebra represented by bounded operators on a Hilbert space H, and D, encoding
the Dirac operator and metric, is an unbounded selfadjoint operator on H. It is
required that D interacts with the algebra in a bounded fashion, i.e., that for all
a ∈ A the commutators [D, a] = Da − aD are well defined on the domain of D and
extend to bounded operators on H. It is further postulated that the operator D
should have compact resolvent in the sense that (D + λ)−1 ∈ K(H) for all λ ∈ / R.
This last condition implies that the spectrum of D consists of a discrete set of
eigenvalues with finite multiplicity. A spectral triple is called finitely summable if
(D2 + 1)−1 ∈ Lp (H) for some 1 ≤ p < ∞.
Given a spectral triple as above and assuming that D is invertible, one checks
that with F := D|D|−1 , the phase of D, (A, H, F ) is a Fredholm module. By
106 Introduction

passing from the spectral triple to the corresponding Fredholm module, we lose
the metric structure, but still retain the topological information, in particular the
index pairing, encoded by the triple. For examples of spectral triples arising in
physics and number theory the reader should consult [23].

Exercise 0.20.1. The Fredholm module of Example 0.19.2 can be defined over any
free group. For Γ = Z one obtains an even Fredholm module over C ∗ (Z) ' C(S 1 ).
Identify this Fredholm module and its character.
Exercise 0.20.2. Give an example of a discrete group Γ and a projection e ∈ CΓ
such that τ (e) is not an integer (τ is the canonical trace).

Exercise 0.20.3. Let (H, F ) be an odd p-summable Fredholm module over an

algebra A. What happens if in the cochain (51) we replace (2n − 1) by an even
integer. Similarly for even Fredholm modules.
Exercise 0.20.4. Show that the Fredholm module in Example 0.19.1 is p-summable
for any p > 1 but is not 1-summable. If we consider it as a Fredholm module over
the algebra C k (S 1 ) of k-times continuously differentiable functions then (H, F ) is
p-summable for some p > 1. Find a relation between k and p.
Exercise 0.20.5. Show that the Fredholm module over Cr∗ (F2 ) in Example 0.19.2
is not 1-summable.

Exercise 0.20.6. Let D = −i dx d

: C ∞ (S 1 ) → C ∞ (S 1 ). It has an extension to a
selfadjoint unbounded operator D : Dom(D) ⊂ L2 (S 1 ) → L2 (S 1 ). Show that the
arc distance on S 1 can be recovered from D via the formula

dist(p, q) = sup{|f (p) − f (q)|; k[D, π(f )k ≤ 1}, (58)

where π(f ) is the multiplication by f operator. The triple (C ∞ (S 1 ), L2 (S 1 ), D) is

an example of a spectral triple and (58) is a prototype of a very general distance
formula of Connes that recovers the distance on a Riemannian spinc manifold from
its Dirac operator (cf. formula (57) in this section and the last chapter of [21]).
Appendix A

Compact operators,
Fredholm operators, and
abstract index theory

The theory of operators on Hilbert space is essential for noncommutative geome-

try. Operator theory is the backbone of von Neumann and C ∗ -algebras and these
are natural playgrounds for noncommutative measure theory and topology. We
saw, for example, that K-homology has a natural formulation in operator theoretic
terms using compact and Fredholm operators and it is this formulation that lends
itself to generalization to the noncommutative setup. Similarly, the more refined
aspects of noncommutative geometry, like noncommutative metric and Rieman-
nian geometry, can only be formulated through spectral invariants of operators on
Hilbert space.
We assume that the reader is familiar with concepts of Hilbert space, bounded
operators on Hilbert space, and basic spectral theory as can be found in the first
chapters of, e.g. [55], [63], [35]. A good reference for ideals of compact operators
is [65]. For basic Fredholm theory and abstract index theory we recommend [35],
[55]. In this section H will always stand for a Hilbert space over the complex
numbers and L(H) for the algebra of bounded linear operators on H. The adjoint
of an operator T shall be denoted by T ∗ .
Our first task in this section is to introduce several classes of ideals in L(H),
most notably ideals of compact operators and the Schatten ideals. Let F(H)
denote the set of finite rank operators on H, i.e., operators whose range is finite
dimensional. F(H) is clearly a two-sided ∗-ideal in L(H) and in fact it is easy to
show that it is the smallest proper two-sided ideal in L(H).
Let
K(H) := F(H)
be the norm closure of F(H). It is clearly a norm closed two-sided ∗-ideal in L(H).
An operator T is called compact if T ∈ K(H). Let H1 denote the closed unit ball

107
108 A Compact operators, Fredholm operators, and abstract index theory

of H. It can be shown that an operator T ∈ L(H) is compact if and only if the

norm closure T (H1 ) is a compact subset of H in norm topology. It follows that
the range of a compact operator can never contain a closed infinite dimensional
subspace. The spectrum of a compact operator is a countable subset of C with 0
as its only possible limit point. Any nonzero point in the spectrum is an eigenvalue
whose corresponding eigenspace is finite dimensional. For a compact operator T ,
let
µ1 (T ) ≥ µ2 (T ) ≥ µ3 (T ) ≥ · · ·

denote the sequence of singular values of T . By definition, µn (T ) is the n-th

1
eigenvalue of |T | := (T ∗ T ) 2 , the absolute value of T ,
It can be shown that if H is separable and infinite dimensional, which is the
case in almost all examples, then K(H) is the unique proper and closed two-sided
ideal of L(H). In this case it is also the largest proper two-sided ideal of L(H).
Thus for any other two sided operator ideal J we have

F(H) ⊂ J ⊂ K(H).

An interesting point of view, advocated by Connes and of fundamental impor-

tance for noncommutative geometry [21], is that compact operators are the true
counterparts of infinitesimals in noncommutative geometry. If we regard L(H) as
a replacement for C in noncommutative geometry (as in going from c-numbers to
q-numbers in quantum mechanics), then compact operators should be regarded as
infinitesimals. Classically, an infinitesimal is a ‘number’ whose absolute value is
less than any positive number! The following lemma shows that the norm of a
compact operator can be made as small as we wish, provided we stay away from
a finite dimensional subspace:

Lemma A.1. Let T be a compact operator. For any ε > 0 there is a finite
dimensional subspace V ⊂ H such that kP T P k < ε, where P is the orthogonal
projection onto the orthogonal complement of V .

The first thorough study of the ideal structure of L(H) was done by Calkin
[13]. Among the two-sided ideals of L(H), and perhaps the most important ones
for noncommutative geometry, are the Schatten ideals, and ideals related to the
Dixmier trace [21]. Let us recall the definition of the former class of ideals next.
A compact operator T ∈ K(H) is called a trace class operator if
∞
X
µn (T ) < ∞.
n=1

Let en , n ≥ 1, be an orthonormal basis of H. It is easy to see that if T is trace

class then X
Tr(T ) := hT ei , ei i
i
 A Compact operators, Fredholm operators, and abstract index theory 109

is finite and is independent of the choice of basis. We denote the set of trace class
operators by L1 (H). It is a two sided ∗-ideal in L(H). Using the definition of the
trace Tr, it is easy to check that if A and B are both trace class, then

Tr(AB) = Tr(BA). (A.1)

What is much less obvious though, and that is what we actually used in Chapter
4, is that if both AB and BA are trace class then (A.1) still holds. A proof of this
can be given using Lidski’s theorem. This theorem is one of the hardest facts to
establish about trace class operators (cf. [65] for a proof).

Theorem A.1 (Lidski’s theorem). If A is a trace class operator then

∞
X
Tr(A) = λi ,
i

where the summation is over the set of eigenvalues of A.

Now since for any two operators A and B, AB and BA have the same spectrum
(and spectral multiplicity) except for 0 (cf. Exercise ??) we obtain the

Corollary A.1. Assume A and B are bounded operators such that AB and BA
are both trace class. Then (A.1) holds.

Next we define the class of Schatten-p ideals for p ∈ [1, ∞) by

Lp (H) := {T ∈ L(H); |T |p ∈ L1 (H)}.

Thus T ∈ Lp (H) if and only if

∞
X
µn (T )p < ∞.
n=1

It is clear that if p ≤ q then Lp (H) ⊂ Lq (H). The Schatten p-norm is defined by

∞
X
kT kpp = µn (T )p .
n=1

Proposition A.1. 1) Lp (H) is a two-sided ideal of L(H).

1 1
2) (Hölder inequality) Let p, q, r ∈ [1, ∞] with r = p + 1q . For any S ∈ Lp (H)
and T ∈ Lq (H), we have ST ∈ Lr (H) and

kST kr ≤ kSkp kT kq .

In particular if Ai ∈ Ln (H) for i = 1, 2, . . . , n, then their product A1 A2 . . . An

is in L1 (H).
110 A Compact operators, Fredholm operators, and abstract index theory

Example A.1. 1. Let us fix an orthonormal basis en , n = 0, 1, 2, . . . , in H. The

diagonal operator defined by T en = λn enP, n ≥ 0 is compact if and only if λn → 0
as n → ∞. It is in Lp (H) if and only if i |λi |p < ∞. By the spectral theorem
for compact operators, every selfadjoint compact operator is unitarily equivalent
to a diagonal operator as above.
2. Integral operators with L2 kernels provide typical examples of operators in
2
L (H), the class of Hilbert–Schmidt operators. Let K be a complex-valued square
integrable function on X × X where (X, µ) is a measure space. Then the operator
TK on L2 (X, µ) defined by
Z
(TK f )(x) = K(x, y)f (y) dµ
X

is a Hilbert–Schmidt (in particular compact) operator, with

Z Z
2 2
kTK k2 = kKk2 = |K(x, y)|2 dxdy.
X X

Under suitable conditions, e.g. when X is compact Hausdorff and the kernel
K is continuous, TK is a trace class operator and
Z
Tr(TK ) = K(x, x) dx.
X

In the remainder of this section we shall recall some basic definitions and facts
about Fredholm operators and index. A bounded linear operator T : H1 → H2
between two Hilbert spaces is called a Fredholm operator if its kernel and cokernel
are both finite dimensional:

dim ker(T ) < ∞, dim coker(T ) < ∞.

The index of a Fredholm operator is the integer

index(T ) := dim ker(T ) − dim coker(T )

= dim ker(T ) − dim ker(T ∗ ).

We list some of the standard properties of Fredholm operators and the index
that are frequently used in noncommutative geometry:
1. (Atkinson’s theorem) A bounded operator T : H1 → H2 is Fredholm if
and only if it is invertible modulo compact operators, that is, if there exists an
operator S : H2 → H1 such that 1 − ST and 1 − T S are compact operators on H1
and H2 respectively. S is called a parametrix for T . It can also be shown that T
is Fredholm if and only if it is invertible module finite rank operators.
Let C := L(H)/K(H) denote the Calkin algebra and π : L(H) → C be the
quotient map. (By general C ∗ -algebra theory, a quotient of a C ∗ -algebra by a
closed two sided ∗-ideal is a C ∗ -algebra in a natural way). Thus Atkinson’s theorem
 A Compact operators, Fredholm operators, and abstract index theory 111

can be reformulated as saying that an operator T is Fredholm if and only if π(T )

is invertible in the Calkin algebra. This for example immediately implies that
Fredholm operators form an open subset of L(H) which is invariant under compact
perturbations.
2. If T1 and T2 are Fredholm operators then T1 T2 is also a Fredholm operator
and
index(T1 T2 ) = index(T1 ) + index(T2 ).
3. The Fredholm index is stable under compact perturbations: if K is a com-
pact operator and T is Fredholm, then T + K is Fredholm and
index(T + K) = index(T ).
4. The Fredholm index is a homotopy invariant: if Tt , t ∈ [0, 1] is a norm
continuous family of Fredholm operators then
index(T0 ) = index(T1 ).
It is this homotopy invariance, or continuity, of the index that makes it computable
and extremely useful. Note that dim ker Tt can have jump discontinuities.
5. Let Fred(H) denote the set of Fredholm operators on a separable infinite
dimensional Hilbert space. It is an open subset of L(H) and the index map
index : Fred(H) → Z
induces a one-to-one correspondence between the connected components of Fred(H)
and Z.
6. Fred(H) is a classifying space for K-theory. More precisely, by a theorem
of Atiyah and Jänich, for any compact Hausdorff space X, we have a canonical
isomorphism of abelian groups
K 0 (X) ' [X, Fred(H)],
where [X, Fred(H)] is the set of homotopy classes of norm continuous maps from
X → Fred(H). Thus continuous families of Fredholm operators on X, up to
homotopy, gives the K-theory of X.
7. (Calderón’s formula [12]) Let P : H1 → H2 be a Fredholm operator and let
Q : H2 → H1 be a parametrix for P . Assume that for some positive integer n,
(1 − P Q)n and (1 − QP )n are both trace class operators. Then we have
index(P ) = Tr(1 − QP )n − Tr(1 − P Q)n . (A.2)
Here is an alternative formulation of the above result. Let H = H1 ⊕H2 . It is a

super Hilbert space with even and odd parts given by H1 and H2 . Let F = P0 Q 0
and γ be the corresponding grading operator. Then we have
index(P ) = Trs (1 − F 2 )n , (A.3)
where Trs is the supertrace of trace class operators, defined by Trs (X) = Tr(γX).
112 A Compact operators, Fredholm operators, and abstract index theory

Example A.2. We give a few examples of Fredholm operators and Fredholm

index.
1. Any operator T : H1 → H2 where both H1 and H2 are finite dimensional is
Fredholm. Its index is independent of T and is given by

index (T ) = dim(H1 ) − dim(H2 ).

If only one of H1 or H2 is finite dimensional then no T can be Fredholm. This

shows that the class of Fredholm operators and index is a purely infinite dimen-
sional phenomenon and are only interesting when both H1 and H2 are infinite
dimensional.
2. Let us fix an orthonormal basis en , n = 0, 1, 2, . . . , for H. The unilateral
shift operator is defined by

T (ei ) = ei+1 , i ≥ 0.

It is easy to see that T is injective and its range is the closed subspace spanned by
ei , i ≥ 1. Thus T is a Fredholm operator with index(T ) = −1. Its adjoint T ∗ (the
backward shift) has index +1. Their powers T m and T ∗m are m-step forward and
backward shifts, respectively, with index(T m ) = m and index(T ∗m ) = −m.
3. We saw in Section 0.19 that for any odd Fredholm modules (H, F ) over
an algebra A and an invertible element U ∈ A the operator P U P : P H → P H is
a Fredholm operator, where P = 1+F 2 is the projection onto the 1-eigenspace of
F . Similarly for an even Fredholm module (H, F, γ) over A and an idempotent
e ∈ A, the operator Fe+ : (eF e)+ : e+ H+ → e− H− is Fredholm.
4. Elliptic differential operators acting on smooth sections of vector bundles
over closed manifolds define Fredholm operators on the corresponding Sobolev
spaces of sections. Computing the index of such Fredholm operators is what the
index theorem of Atiyah–Singer achieves. Let M be a smooth manifold and let E
and F be smooth complex vector bundles on M . Let

D : C ∞ (E) → C ∞ (F )

be a liner differential operator. This means that D is a C-linear map which is

locally expressible by an m × n matrix of differential operators. This matrix of
course depends on the choice of local coordinates on M and local frames for E and
F . The principal symbol of D is defined by replacing differentiation by covectors in
the leading order terms D. The resulting ‘matrix-valued function on the cotangent
bundle’
σD ∈ C ∞ (Hom(π ∗ E, π ∗ F ))
can be shown to be invariantly defined. Here π : T ∗ M → M is the natural projec-
tion map of the cotangent bundle. A differential operator D is called elliptic if for
all x ∈ M and all nonzero ξ ∈ Tx∗ M , the principal symbol σD (x, ξ) is an invertible
matrix.
 A Compact operators, Fredholm operators, and abstract index theory 113

Let W s (E) denote the Sobolev space of sections of E (roughly speaking, it

consists of sections whose ‘derivatives of order s’ are square integrable). The main
results of the theory of linear elliptic PDE’s show that for each s ∈ R, D has a
unique extension to a bounded and Fredholm operator D : W s (E) → W s−n (F )
between Sobolev spaces (n is the order of the differential operator D). Moreover
the Fredholm index of D is independent of s and coincides with the index defined
using smooth sections.
Remark 5. Even in a purely algebraic context, the notion of Fredholm operator
makes sense and defines an interesting class of linear operators. A linear map : T :
V1 → V2 is called Fredholm if its kernel and cokernel are both finite dimensional.
In that case we can define the index of T as the difference

index(T ) = dim(kerT ) − dim(cokerT ).

It is easy to see that this concept is interesting only if V1 and V2 are both infinite
dimensional. The stongest results are obtained when T is a bounded linear operator
between Hilbert spaces.
Exercise A.1. Any invertible operator is clearly Fredholm and its index is zero.
Thus any compact perturbation of a an invertible operator is Fredholm and its
index is zero. Is it true that any Fredholm operator with zero index is a compact
perturbation of an invertible operator?

Exercise A.2. Prove Calderón’s formula (A.2) for n = 1.

Exercise A.3. Formula (A.3) relates the Fredholm index with the operator trace.
Here is a similar formula. Let H be a Z2 -graded Hilbert space and let D be an
2
unbounded odd selfadjoint operator on H such that e−tD is a trace class operator
for all t > 0. Show that index(D) := dim ker(D ) − dim ker(D− ) is well defined
+

and is given by the McKean–Singer formula

2
index(D) = Trs (e−tD ) for all t > 0.
114 A Compact operators, Fredholm operators, and abstract index theory
Appendix B

K-theory

We start by briefly recalling the definitions of the functors K0 and K1 . Let A

be a unital algebra and let P(A) denote the set of isomorphism classes of finitely
generated projective right A-modules. Under the operation of direct sum, P(A)
is an abelian monoid. The group K0 (A) is, by definition, the Grothendieck group
of the monoid P(A) in the sense that there is a universal additive map P(A) →
K0 (A). Thus elements of K0 (A) can be written as [P ]−[Q] for P, Q ∈ P(A), with
[P ] − [Q] = [P 0 ] − [Q0 ] if and only if there is an R ∈ P(A) such that P ⊕ Q0 ⊕ R '
P 0 ⊕ Q ⊕ R.
There is an alternative description of K0 (A) in terms of idempotents in matrix
algebras over A that is often convenient. An idempotent e ∈ Mn (A) defines a
right A-module map
e : An → An
by left multiplication by e. Let Pe = eAn be the image of e. The relation

An = eAn ⊕ (1 − e)An

shows that Pe is a finite projective right A-module. Different idempotents may

define isomorphic modules. This happens, for example, if e and f are equivalent
idempotents (sometimes called similar) in the sense that

e = uf u−1
S
for some invertible u ∈ GL(n, A). Let M (A) = Mn (A) be the direct limit of
the matrix algebras Mn (A) under the embeddings Mn (A) → Mn+1 (A) defined by
a 7→ ( a0 00 ). Similarly let GL(A) be the direct limit of the groups GL(n, A). It acts
on M (A) by conjugation.

Definition B.0.1. Two idempotents e ∈ Mk (A) and f ∈ Ml (A) are called stably
equivalent if their images in M (A) are equivalent under the action of GL(A).

The following is easy to prove and answers our original question.

115
116 B K-theory

Lemma B.0.1. The projective modules Pe and Pf are isomorphic if and only if
the idempotents e and f are stably equivalent.
Let Idem(M (A))/ GL(A) denote the set of stable equivalence classes of idem-
potents over A. This is an abelian monoid under the operation
 
e 0
(e, f ) 7→ e ⊕ f := .
0 f
It is clear that any finite projective module is of the type Pe for some idempo-
tent e. In fact writing P ⊕ Q ' An , one can let e be the idempotent corresponding
to the projection map (p, q) 7→ (p, 0). These observations prove the following
lemma.
Lemma B.0.2. For any unital ring A, the map e 7→ Pe defines an isomorphism
of monoids
Idem(M (A))/ GL(A) ' P(A).
Given an idempotent e = (eij ) ∈ Mn (A), its image under a homomorphism
f : A → B is the idempotent f∗ (e) = (f (eij )). This is our formula for f∗ : K0 (A) →
K0 (B) in the idempotent picture of K-theory. It turns A → K0 (A) into a functor
from unital algebras to abelian groups.
For a unital Banach algebra A, K0 (A) can be described in terms of connected
components of the space of idempotents of M (A) under its inductive limit topology
(a subset V ⊂ M (A) is open in the inductive limit topology if and only if V ∩Mn (A)
is open for all n). It is based on the following important observation: Let e and f
be idempotents in a unital Banach algebra A and assume ke − f k < 1/k2e − 1k.
Then e ∼ f . In fact with
v = (2e − 1)(2f − 1) + 1 (B.1)
and u = 21 v, we have ueu−1 = f . To see that u is invertible note that ku − 1k < 1.
One consequence of this fact is that if e and f are in the same path component of
the space of idempotents in A, then they are equivalent. As a result we have, for
any Banach algebra A, an isomorphism of monoids
P(A) ' π0 (Idem(M (A))),
where π0 is the functor of path components.
For C ∗ -algebras, instead of idempotents it suffices to consider only the pro-
jections. A projection is a self-adjoint idempotent (p2 = p = p∗ ). The reason is
that every idempotent in a C ∗ -algebra is similar to a projection [7]: let e be an
idempotent and set z = 1 + (e − e∗ )(e∗ − e). Then z is invertible and positive
and one shows that p = ee∗ z −1 is a projection and is similar to e. In fact, it can
be shown that the set of projections of a C ∗ -algebra is a retraction of its set of
idempotents. Let Proj(M (A)) denote the space of projections in M (A). We have
established isomorphisms of monoids
P(A) ' π0 (Idem(M (A))) ' π0 (Proj(M (A)))
 117

which reflects the coincidence of stable equivalence, Murray–von Neumann equiv-

alence, and homotopy equivalence in Proj (M (A)).
Starting with K1 , algebraic and topological K-theory begin to differ from each
other. We shall briefly indicate the definition of algebraic K1 , and the necessary
modification needed for topological K 1 . Let A be a unital algebra. The algebraic
K1 of A is defined as the abelianization of the group GL(A):

K1alg (A) := GL(A)/[GL(A), GL(A)],

where [ · , · ] denotes the commutator subgroup. Applied to A = C(X), this defini-

tion does not reproduce the topological K 1 (X). For example for A = C = C(pt)
we have K1alg (C) ' C× where the isomorphism is induced by the determinant map

det : GL(C) → C× ,

while K 1 (pt) = 0. It turns out that, to obtain the right result, one should divide
GL(A) by a bigger subgroup, i.e., by the closure of its commutator subgroup. This
works for all Banach algebras and will give the right definition of topological K1 .
A better approach however is to define the higher K groups in terms of K0 and
the suspension functor [7].
Bibliography

[1] R. Akbarpour and M. Khalkhali, Hopf algebra equivariant cyclic homology and
cyclic homology of crossed product algebras. J. Reine Angew. Math. 559 (2003),
137–152. 78
[2] M. F. Atiyah, Global theory of elliptic operators. In Proc. Internat. Conf. on Func-
tional Analysis and Related Topics (Tokyo, 1969), University of Tokyo Press, Tokyo
1970, 21–30. 90, 91, 92
[3] H. Bass, Algebraic K-theory. W. A. Benjamin, Inc., New York 1968.
[4] P. Baum and R. Douglas, Index theory, bordism, and K-homology. Contemp. Math.
10 (1982), 1–31. 90
[5] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Sternheimer, Deformation
theory and quantization. I. Deformations of symplectic structures. Ann. Physics
111 (1978), no. 1, 61–110; II. Physical applications. Ann. Physics 111 (1978), no.
1, 111–151. 32
[6] J. Bellisard, A. van Elst and H. Schultz-Baldes, The noncommutative geometry of
the quantum Hall effect. J. Math. Phys. 35 (1994), no. 10, 5373–5451.
[7] B. Blackadar, K-theory for operator algebras. Math. Sci. Res. Inst. Publ. 5,
Springer-Verlag, New York 1986. 90, 116, 117
[8] B. Blackadar, Operator algebras. In Theory of C ∗ -algebras and von Neumann al-
gebras, Encyclopaedia Math. Sci. 122, Operator Algebras and Non-commutative
Geometry III, Springer-Verlag, Berlin 2006. 36
[9] J.-B. Bost, Principe d’Oka, K-théorie et systèmes dynamiques non commutatifs.
Invent. Math. 101 (1990), no. 2, 261–333.
[10] L. Brown, R. Douglas, and P. Fillmore, Extensions of C ∗ -algebras and K-homology.
Ann. of Math. (2) 105 (1977), no. 2, 265–324. 90, 91, 92
[11] D. Burghelea, The cyclic homology of the group rings. Comment. Math. Helv. 60
(1985), no. 3, 354–365. 42, 44, 55, 69, 70
[12] A. P. Calderón, The analytic calculation of the index of elliptic equations. Proc.
Nat. Acad. Sci. U.S.A. 57 (1967), 1193–1194. 111
[13] J. W. Calkin, Two-sided ideals and congruences in the ring of bounded operators
in Hilbert space. Ann. of Math. (2) 42 (1941), 839–873. 108
[14] H. Cartan and S. Eilenberg, Homological algebra. Princeton University Press,
Princeton, N.J., 1999. 20, 25, 41, 44, 45

119
120 Bibliography

[15] A. Connes, On the cohomology of operator algebras. J. Funct. Anal. 28 (1978),

no. 2, 248–253. 36
[16] A. Connes, C ∗ algébres et géométrie différentielle. C. R. Acad. Sci. Paris Sér. A-B
290 (1980), no. 13, A599–A604. 57, 80, 88, 89
[17] A. Connes, A survey of foliations and operator algebras. In Operator algebras and
applications, Part I (Kingston, Ont., 1980), Proc. Sympos. Pure Math. 38, Amer.
Math. Soc., Providence, R.I., 1982, 521–628.
[18] A. Connes, Cohomologie cyclique et foncteurs Extn . C. R. Acad. Sci. Paris Sr. I
Math. 296 (1983), no. 23, 953–958. 55, 56, 62, 73, 74, 75, 76
[19] A. Connes, Noncommutative differential geometry. Chapter I: The Chern character
in K-homology, Preprint IHES octobre 1982; Chapter II: de Rham homology and
noncommutative algebra, Preprint IHES février 1983; Inst. Hautes tudes Sci. Publ.
Math. 62 (1985), 41–144. 7, 19, 26, 33, 36, 38, 40, 41, 49, 52, 53, 57, 58, 64, 65, 68,
69, 70, 71, 80, 88, 90, 91, 97, 101
[20] A. Connes, Cyclic cohomology and the transverse fundamental class of a foliation.
In Geometric methods in operator algebras (Kyoto, 1983), Pitman Res. Notes Math.
Ser. 123, Longman Sci. Tech., Harlow 1986, 52–144.
[21] A. Connes, Noncommutative geometry. Academic Press, San Diego, Ca., 1994. 31,
36, 55, 69, 80, 82, 83, 89, 92, 105, 106, 108
[22] A. Connes and M. Marcolli, A walk in the noncommutative garden. In An invita-
tion to noncommutative geometry (ed. by M. Khalkhali and M. Marcolli), World
Scientific Publishing, Hackensack, N.J., 2008, 1–128.
[23] A. Connes and M. Marcolli, Noncommutative geometry, quantum fields and mo-
tives. Amer. Math. Soc. Colloq. Publ. 55, Amer. Math. Soc., Providence, R.I.,
2008. 106
[24] A. Connes and H. Moscovici, Cyclic cohomology, the Novikov conjecture and hy-
perbolic groups. Topology 29 (1990), no. 3, 345–388.
[25] A. Connes and H. Moscovici, The local index formula in noncommutative geometry.
Geom. Funct. Anal. 5 (1995), no. 2, 174–243. 14
[26] A. Connes and H. Moscovici, Hopf algebras, cyclic cohomology and the transverse
index theorem, Comm. Math. Phys. 198 (1998), no. 1, 199–246. 77
[27] A. Connes and H. Moscovici, Cyclic cohomology and Hopf algebra symmetry. Con-
ference Moshé Flato 1999 (Dijon), Lett. Math. Phys. 52 (2000), no. 1, 1–28. 77
[28] A. Connes and M. Rieffel, Yang–Mills for noncommutative two-tori. In Operator
algebras and mathematical physics (Iowa City, Iowa, 1985), Contemp. Math. 62,
Amer. Math. Soc., Providence, R.I., 1987, 237–266.
[29] J. Cuntz, Cyclic theory, bivariant K-theory and the bivariant Chern-Connes char-
acter. In Cyclic homology in non-commutative geometry, Encyclopaedia Math.
Sci. 121, Operator Algebras and Non-commutative Geometry, II, Springer-Verlag,
Berlin 2004, 1–71.
[30] J. Cuntz and D. Quillen, Cyclic homology and nonsingularity. J. Amer. Math. Soc.
8 (1995), no. 2, 373–442.
[31] J. Cuntz, and D. Quillen, Excision in bivariant periodic cyclic cohomology. Invent.
Math. 127 (1997), no. 1, 67–98. 67
 Bibliography 121

[32] J. Cuntz, R. Meyer and J. Rosenberg, Topological and bivariant K-theory. Ober-
wolfach Semin. 36, Birkhäuser, Basel 2007. 91
[33] J. Cuntz, G. Skandalis and B. Tsygan, Cyclic homology in non-commutative ge-
ometry. Encyclopaedia Math. Sci. 121, Operator algebras and non-commutative
geometry, II, Springer-Verlag, Berlin 2004.
[34] P. Dirac, The fundamental equations of quantum mechanics. Proc. Royal Soc. Lon-
don A 109 (1926), 642–653. 19, 28
[35] R. Douglas, Banach algebra techniques in operator theory. Second edition, Grad.
Texts in Math. 179. Springer-Verlag, New York 1998. 107
[36] B. Feı̆gin and B. L. Tsygan, Additive K-theory. In K-theory, arithmetic and geom-
etry (Moscow, 1984–1986), Lecture Notes in Math. 1289, Springer-Verlag, Berlin
1987, 67–209. 31, 69
[37] M. Gerstenhaber, The cohomology structure of an associative ring. Ann. of Math.
(2) 78 (1963), 267–288. 16, 34
[38] J. M. Gracia-Bondia, J. C. Várilly, and H. Figueroa, Elements of noncommutative
geometry. Birkhuser Adv. Texts, Birkhuser, Boston 2000.
[39] A. Grothendieck, Produits tensoriels topologiques et espaces nucléaires. Mem.
Amer. Math. Soc. (1955), no. 16. 37, 38
[40] V. Guillemin and S. Sternberg, Variations on a theme by Kepler. Amer. Math. Soc.
Colloq. Publ. 42. Amer. Math. Soc., Providence, R.I., 1990. 28, 35
[41] U. Haagerup, All nuclear C ∗ -algebras are amenable. Invent. Math. 74 (1983), no.
2, 305–319. 36
[42] P. Hajac, M. Khalkhali, B. Rangipour, and Y. Sommerhäuser, Stable anti-Yetter-
Drinfeld modules. C. R. Math. Acad. Sci. Paris 338 (2004), no. 8, 587–590. 78
[43] P. Hajac, M. Khalkhali, B. Rangipour, and Y. Sommerhäuser, Hopf-cyclic homol-
ogy and cohomology with coefficients. C. R. Math. Acad. Sci. Paris 338 (2004),
no. 9, 667–672. 78
[44] N. Higson and J. Roe, Analytic K-homology. Oxford Math. Monogr., Oxford Uni-
versity Press, Oxford 2000. 91
[45] N. Higson and J. Roe (eds.), Surveys in noncommutative geometry, Clay Math.
Proc. 6, Amer. Math. Soc., Providence, R.I., 2006.
[46] G. Hochschild, On the cohomology groups of an associative algebra. Ann. of Math.
(2) 46, (1945), 58–67.
[47] G. Hochschild, B. Kostant, and A. Rosenberg, Differential forms on regular affine
algebras. Trans. Amer. Math. Soc. 102 (1962), 383–408. 38, 40
[48] M. Karoubi, K-theory. An introduction. Grundlehren Math. Wiss. 226, Springer-
Verlag, Berlin 1978.
[49] M. Karoubi, Homologie cyclique et K-théorie. Astrisque 149 (1987). 70
[50] G. G. Kasparov, The operator K-functor and extensions of C ∗ -algebras. Izv. Akad.
Nauk SSSR Ser. Mat. 44 (1980), no. 3, 571–636 (in Russian). 90
[51] G. G. Kasparov, Equivariant KK-theory and the Novikov conjecture. Invent. Math.
91 (1988), no. 1, 147–201. 91, 92
122 Bibliography

[52] M. Khalkhali and M. Marcolli (eds.), An invitation to noncommutative geometry,

World Scientific Publishing, Hackensack, N.J., 2008.
[53] M. Kontsevich, Deformation quantization of Poisson manifolds. Lett. Math. Phys.
66 (2003), no. 3, 157–216. 28, 33
[54] O. Kravchenko and B. Khesin, A central extension of the algebra of pseudodif-
ferential symbols. Funktsional. Anal. i Prilozhen. 25 (1991), no. 2, 83–85; English
transl. Funct. Anal. Appl. 25 (1991), no. 2, 152–154. 35
[55] P. Lax, Functional analysis. Pure Appl. Math., Wiley-Interscience, John Wiley and
Sons, New York 2002. 107
[56] J. L. Loday, Cyclic Homology. Second edition, Grundlehren Math. Wiss. 301,
Springer-Verlag, Berlin 1998. 20, 25, 31, 39, 41, 44, 45, 46, 63, 69, 70, 73
[57] J.-L. Loday and D. Quillen, Cyclic homology and the Lie algebra homology of
matrices. Comment. Math. Helv. 59 (1984), no. 4, 569–591.
[58] Yu. I. Manin, Algebraic aspects of nonlinear differential equations. In Current prob-
lems in mathematics, Vol. 11, Akad. Nauk SSSR Vsesojuz. Inst. Naucn. i Tehn.
Informacii, Moscow 1978, 5–152 (in Russian). 31
[59] M. Marcolli and V. Mathai, Towards the fractional quantum Hall effect: a non-
commutative geometry perspective.Noncommutative geometry and number theory:
where arithmetic meets geometry and physics, Aspects Math. E37, Vieweg, Wies-
baden 2006, 235–261.
[60] J. Milnor and J. Moore, On the structure of Hopf algebras. Ann. of Math. (2) 81
(1965), 211–264.
[61] J. Milnor and J. Stasheff, Characteristic classes. Ann. of Math. Stud. 76, Princeton
University Press, Princeton, N.J., 1974. 79, 80, 86
[62] J. E. Moyal, Quantum mechanics as a statistical theory. Proc. Cambridge Philos.
Soc. 45 (1949), 99–124. 32
[63] M. Reed and B. Simon, Methods of modern mathematical physics. I. Functional
analysis. Second edition. Academic Press, New York 1980. 107
[64] J. Rosenberg, Algebraic K-theory and its applications. Grad. Texts in Math. 147,
Springer-Verlag, New York 1994.
[65] B. Simon, Trace ideals and their applications. London Math. Soc. Lecture Note Ser.
35, Cambridge University Press, Cambridge 1979. 107, 109
[66] B. Tsygan, Homology of matrix Lie algebras over rings and the Hochschild homol-
ogy. Uspekhi Mat. Nauk 38 (1983), no. 2, 217–218 (in Russian).
[67] J. Várilly, An introduction to noncommutative geometry. EMS Ser. Lect. Math.,
EMS Publishing House, Zürich 2006.
[68] M. Wodzicki, Noncommutative residue. I. Fundamentals. In K-theory, arithmetic
and geometry (Moscow, 1984–1986), Lecture Notes in Math. 1289, Springer-Verlag,
Berlin 1987, 320–399. 31