Group Actions On Manifolds: Eckhard Meinrenken

Group actions on manifolds
Eckhard Meinrenken
Lecture Notes, University of Toronto, Spring 2003

Contents
1. Terminology and notation 4

2. Classifying bundles 17
3. Construction of EG by simplicial techniques 27
4. Spectral sequences 37
5. The Chern-Weil construction 43
6. Equivalence between the simplicial and Weil model 53
7. Localization 57
Bibliography 61
3
4 CONTENTS
1. Terminology and notation

1.1. Lie group actions.
Definition 1.1. An action of a Lie group G on a manifold M is a group homomorphism
G → Diff(M ), g 7→ Ag
into the group of diffeomorphisms on M , such that the action map
G × M → M, (g, m) 7→ Ag (m)
is smooth.
We will usually write g.m rather than Ag (m). With this notation, g1 .(g2 .m) = (g1 g2 ).m
and e.m = m.
Remarks 1.2. (a) One has similar definitions of group actions in other categories.
For instance, an action of a topological group G on a topological space X to be a
homomorphism G → Homeo(X) such that the action map G × X → X is continuous.
An action of a (discrete) group G on a set S is simply a homomorphism into the
permutation group of S.
(b) Some people call what we’ve just introduced a left-action, and define a right-action to
be an anti-homomorphism G → Diff(M ). For such a right action g 7→ Bg one would
then write m.g := Bg (m); with this notation
(m.g1 ).g2 = m.(g1 g2 )
Any right action can be turned into a left action by setting Ag = Bg−1 . In this course,
we will avoid working with right actions.
Examples 1.3. 1) An action of the (additive) Lie group G = R is the same thing as a
global flow, while an action of the Lie algebra G = S 1 is sometimes called a periodic flow.
2) Let V be a finite-dimensional vector space. Then V (viewed as an Abelian group)
acts on itself by translation. Also the general linear group GL(V ) acts on V by its defining
representation. The actions fit together to an action of the affine linear group, the semi-direct
product
GL(V ) ⋉ V, (g1 , v1 )(g2 , v2 ) = (g1 g2 , v1 + g1 .v2 ).
(In fact, the formula for the semi-direct product is most easily remembered from this action.)
A group homomorphism G → GL(V ) (i.e. representation) defines a linear action of G on V ,
and more generally a group homomorphism G → GL(V ) ⋉ V is called an affine action.
3) Any Lie group G acts on itself by multiplication from the left, Lg (a) = ga, multiplication
from the right Rg (a) = ag −1 , and also by the adjoint (=conjugation) action
Adg (a) := Lg Rg (a) = gag −1 .
4) Given a G-action on M , and a submanifold N ⊂ M that is G-invariant, one obtains an
action on N by restriction. For example the rotation action on Rn restricts to an action on
the unit sphere. Similarly, given a group homomorphism H → G one obtains an action of H
by composition with G Diff(M ).
5) Suppose H ⊂ G is a closed subgroup, hence (by a theorem of Cartan) a Lie subgroup.
Let G/H by the space of right cosets {aH} with the quotient topology. By a well-known
1. TERMINOLOGY AND NOTATION 5
result from the theory of Lie groups, there is a unique smooth structure on G/H such that the
quotient map G → G/H is smooth. Moreover, the left G-action on G descends to an action
on G/H:
g.(aH) = (ga)H.
For a detailed proof, see e.g. Onishchik-Vinberg, [26, Theorem 3.1].
6) Lie group often arise as transformation groups preserving a certain structure. For in-
stance, the Myers-Steenrod theorem asserts that the group Diff(M, g) of isometries of a Rie-
mannian manifold is a Lie group, compact if M is compact. Similarly, if M is a complex
(or even an almost complex) manifold, the group of diffeomorphisms preserving the (almost)
complex structure is a Lie group, provided M is compact. (By contrast, the group of sym-
plectomorphisms of a symplectic manifold Diff(M, ω) is of course infinite-dimensional!) The
general setting for this type of problem is explained in detail in Kobayashi’s book on trans-
formation groups. Let M be a manifold with a reduction of the structure group of T M to
some subgroups H ⊂ GL(n, R), and let G ⊂ Diff(M ) be the group of automorphisms for this
reduction. Call H elliptic if its Lie algebra does not contain a rank 1 matrix, and finite type if
the kth prolongation
hk = {t : S k+1 Rn → Rn | for all x1 , . . . , xk , t(·, x1 , . . . , xk ) ∈ h ⊂ gl(n, R)}
vanishes for k sufficiently large. For elliptic H, G is always a Lie group, and for finite type H,
G is a Lie group provided M is compact.
7) The group of automorphisms Aut(G) of a Lie group G is itself a Lie group. It contains
the subgroup Int(G) of inner automorphism, i.e. automorphism of the form a 7→ Adg (a). The
left-action of G on itself first together with the action of Aut(G) to an action of the semi-direct
product, Aut(G) ⋉ G, where the product is as follows:
(σ1 , g1 )(σ2 , g2 ) = (σ1 σ2 , g1 σ1 (g2 )).
1.2. Lie algebra actions. Let X(M ) denote the Lie algebra of vector fields on M , with
bracket [X, Y ] = X ◦ Y − Y ◦ X where we view vector fields as derivations on the algebra
C ∞ (M ) of smooth functions.
Definition 1.4. An action of a finite-dimensional Lie algebra g on M is a Lie algebra
homomorphism g → X(M ), ξ 7→ ξM such that the action map
g × M → T M, (ξ, m) 7→ ξM (m)
is smooth.
On the right hand side of this definition, vector fields are viewed as sections of the tangent
bundle T M → M .
Examples 1.5. Any Lie algebra representation g → gl(V ) may be viewed as a Lie algebra
action. If (M, g) is a Riemannian manifold, the Lie algebra X(M, g) = {X| LX (g) = 0} of
Killing vector fields is finite-dimensional (by Myers-Steenrod), and by definition acts on M .
d
If γ : R → M is a smooth curve on M , we denote its tangent vector at γ(0) by dt |t=0 γ(t).
Theorem 1.6. Given an action of a Lie group G on a manifold M , one obtains an action
of the corresponding Lie algebra g, by setting
d
ξM (m) = |t=0 exp(−tξ).m
dt
6 CONTENTS
where exp : g → G is the exponential map for G. The vector field ξM is called the generating
vector field corresponding to ξ.
Proof. Let us first note that if G acts on manifolds M1 , M2 , and if F : M1 → M2 is a
G-equivariant map, i.e.
F (g.m) = g.F (m) ∀m ∈ M1
then the vector fields ξM1 , ξM2 are F -related. since F takes integral curves for ξM1 to integral
curves for ξM2 . Since Lie brackets of pairs of F -related vector fields are again F -related, it
follows that if F is onto, and ξ 7→ ξM1 is a Lie algebra homomorphism, then so is ξ 7→ ξM2 .
Consider the map
F : G × M → M, (a, m) 7→ a−1 .m
This map is onto and has the equivariance property,
F (ag −1 , m) = F (a, g.m)
This reduces the problem to the special case of G acting on itself by the action g 7→ Rg . This
action commutes with left-translations, i.e. each lh : G → G is a G-equivariant map. Thus, it
generating vector fields ξG are left-invariant. Since
d d d
ξG (e) = |t=0 exp(−tξ).e = |t=0 exp(−tξ)−1 = |t=0 exp(tξ) = ξ,
dt dt dt
we conclude ξG = ξ , the left-invariant vector field defined by ξ. But [ξ L , η L ] = [ξ, η]L by
L
definition of the Lie bracket.

Example 1.7. The generating vector field for the left action of G on itself is −ξ R , and the
generating vector field for the adjoint action is ξ L − ξ R . The generating vector fields for the
action of an isometry group Iso(M, g) are the Killing vector fields X(M, g).
Remark 1.8. Many people omit the minus sign in the definition of the generating vector
field ξM . But then ξ 7→ ξM is not a Lie algebra homomorphism but an anti-homomorphism.
The minus sign is quite natural if we think of vector fields as derivations: If G acts on M ,
we get an action on the algebra C ∞ (M ) by (g.f )(x) = f (g −1 .x), i.e. a group homomorphism
G → Aut(C ∞ (M )). The generating vector fields is formally (ignoring that Aut(C ∞ (M )) is
infinite-dimensional) the induced map on Lie algebras.
Let us now consider the inverse problem: Try to integrate a given Lie algebra action to an
action of the corresponding group! We will need the following Lemma:
Lemma 1.9. Let G be a connected Lie group, and U ⊂ G an open neighborhood of the group
unit e ∈ G. Then every g ∈ G can be written as a finite product g = g1 · · · gN of elements
gj ∈ U .
Proof. We may assume
S∞ that g −1 ∈ U whenever g ∈ U . For each N , let U N = {g1 · · · gN | gj ∈
U }. WeShave to show N =0 U N = S∞ G. Each U N is open,Shence their union is open as well. If
∞
g ∈ G\ N =0 US, then gU ∈ G\ N =0 U N (for S
N if gh ∈ ∞ N =0 U
N with h ∈ U we would have
∞ ∞
g = (gh)h−1 ∈ N =0 U N .) This showsSthat G\ N =0 U N is also open. Since G is connected,
it follows that the open and closed set ∞ N =0 U
N is all of G.
Corollary 1.10. An action of connected Lie group G on a manifold M is uniquely deter-
mined by its generating vector fields.
Theorem 1.11. Suppose ξ 7→ ξM is a Lie algebra action of g on a manifold M . Then this

Lie algebra action integrates to an action of the simply connected Lie group G corresponding
to g, if and only if each ξM is complete.
Proof. Every G-action on M decomposes G × M into submanifolds Lm = {(g, g.m)| g ∈
G}. Note that each Lm projects diffeomorphically onto G. The action may be recovered from
this foliation: Given (g, m) the leaf Lm contains a unique point having g as its first component,
and then the second component is g.m.
The idea of proof, given a g-action, is to construct this foliation from an integrable distri-
bution. Consider the Lie algebra action on G × M , taking ξ to (−ξ R , ξM ) ∈ X(G × M ). The
subbundle of the tangent bundle spanned by the generating vector fields is a distribution of
rank dim G, which is integrable by Frobenius’ theorem. Hence we obtain a foliation of G × M ,
with leaves of dimension dim G.
Let Lm ֒→ G × M be the unique leaf containing the point (e, m). Projection to the first
factor induces a smooth map
πm : Lm → G,
with tangent map taking (−ξ R , ξM ) to −ξ R . Since the tangent map is an isomorphism, the
map Lm → G is a local diffeomorphism (that is, every point in Lm has an open neighborhood
over which the map is a diffeomorphism onto its image). We claim that πm is a diffeomorphism.
Since G is simply connected, it suffices to show that πm is a covering map. Let U0 ⊂ g be a
star-shaped open neighborhood of 0 over which the exponential map is a diffeomorphism, and
U = exp(U0 ). Given (g, m′ ) ∈ Lm and ξ ∈ U0 , the curve t 7→ exp(−tξ)g is an integral curve of
−ξR . Letting Ftξ be the flow of ξM , it follows that t 7→ (exp(−tξ)g, Ftξ (m)) is an integral curve
of (−ξ R , ξM ), so it lies in the leaf Lm . This shows that there exists an open neighborhood
of (g, m′ ) mapping diffeomorphically onto the right translate U g. This proves that πm is a
covering map, ans also that πm is onto.
Using that πm is a diffeomorphism, we can now define the action by
−1
g.m := pr2 (πm (g))
where pr2 denotesQprojection to the second factor. Concretely, the above argument shows that
if we write g = gi with gi = exp(ξi ) then g.m = g1 .(g2 . · · · gN .m) · · · ) where each gi acts
by its time one flow. This description also shows directly that Ag (m) = g.m defines a group
action.
1.3. Terminology.
Definition 1.12. Let G → Diff(M ) be a group action.
(a) For any m ∈ M , the set G.m := {(g, m) g ∈ G} is called the orbit of m. The subgroup
Gm = {g ∈ G| g.m = m} is called the stabilizer of m.
(b) The action is free if all stabilizer groups Gm are trivial.
(c) The action is locally free if all stabilizer groups Gm are discrete.
(d) The action is effective if the kernel of the homomorphism G → Diff(M ) defining the
action is trivial.
(e) The action is transitive if G.m = M for some (hence all) m ∈ M .
The space M/G = {G.m| m ∈ M } is called the orbit space for the given action.
8 CONTENTS
From the definition, it is clear that stabilizer subgroups are closed subgroups of G, hence
Lie subgroups. For any g ∈ G, the stabilizers of a point m and of its translate g.m are related
by the adjoint action:
Gg.m = Adg (Gm ).
Hence, each to each orbit O = G.m there corresponds a conjugacy class of stabilizers. For
subgroups H, H ′ we will write H ∼ H ′ if H is G-conjugate to H ′ . Clearly, this is an equivalence
relation. We denote by (H) the equivalence class of H. We define a partial ordering on
equivalence classes, by writing (H) < (K) if H is G-conjugate to a subgroup of K.
Definition 1.13. For any subgroup H ⊂ G we define,
MH = {m ∈ M | H ⊂ Gm }
MH = {m ∈ M | H = Gm }
M (H) = {m ∈ M | (H) < (Gm )}
M(H) = {m ∈ M | (H) = (Gm )}.
The set M H is the fixed point set of H, and M(H) set of points of orbit type (H).
Notice that the sets M (H) , M(H) are both G-invariant. In fact they are the flow outs of
M H , MH :
M(H) = G.MH , M (H) = G.M H .
1
Examples 1.14. (a) The rotation action of S 1 ⊂ SO(3) on S 2 has fixed point set M S
consisting of the north and south poles. There are two orbit types: The trivial orbit
type H = e) and the orbit type of the fixed points, H = G.
(b) For the G-action on a homogeneous space G/H, there is only one orbit type equal to
H (since there is only one orbit). The action of NG (H)/H, however, has a much more
interesting orbit type decomposition.
(c) Consider the action of SO(3) on itself by conjugation. Let q : SO(3) → [0, π] be the
map that associates to each A ∈ SO(3) the corresponding angle of rotation. Since each
rotation is determined up to conjugacy by its angle, we may view [0, π] as the orbit
space for the conjugation action, with q as the quotient map. There are three different
1
orbit type strata: The fixed point set M SO(3) = q −1 (0), the set M (S ) = q −1 ((0, π))
(where S 1 = SO(2) is the subgroup of rotations about the z-axis) and M (O(2)) = q −1 (π)
consisting of rotations by π. Notice that M (O(2)) is a submanifold diffeomorphic to
RP (2).
(d) Exercise: Study the orbit space decomposition for the conjugation action of the groups
O(2), SU(3), PU(3).
Definition 1.15. Let g → X(M ) be a Lie algebra action.
(a) For m ∈ M , the subalgebra gm = {ξ| ξM (m) = 0} is called the stabilizer algebra of m.
(b) The Lie algebra action is called free if all stabilizer algebras are trivial.
(c) The Lie algebra action is called effective if the kernel of the map g → X(M ) is trivial.
(d) The Lie algebra action is called transitive if the map g → Tm M, ξ → ξM (m) is onto
for all m ∈ M .
Proposition 1.16. Let G → Diff(M ) be a Lie group action, with corresponding Lie algebra
action given by the generating vector fields ξM . Then the Lie algebra of Gm is the stabilizer
algebra gm .
Proof. Let ξ be an element of the Lie algebra of Gm . Thus exp(−tξ).m = m for all t ∈ R.
Taking the derivative at t = 0, we see that ξM (m) = 0. Thus ξ ∈ gm , showing that the Lie
algebra of Gm is contained in gm . For the converse, one has to show that if ξ ∈ gm then
exp(tξ) ∈ Gm (where exp is the exponential map for G). But this follows since the action of
exp(−tξ) is the flow of ξM , which fixes m since ξM (m) = 0.
Note that the stabilizer algebras are slightly special: For a general subalgebra h ⊂ g, the
corresponding connected subgroup H ⊂ G need not be a closed subgroup.
Proposition 1.17. The G-action is locally free if and only if the g-action is free. If the
G-action is transitive then so is the g-action. The converse holds if M is connected. If the
G-action is effective then so is the g-action.
Proof. If the g-action is transitive, it easily follows that the G-orbits must be open. Since
M is connected, it must be a single G-orbit. All other statements are obvious.
1.4. Proper actions.

Definition 1.18. A G-action on a manifold M is called proper if the map
G × M → M × M, (g.m) 7→ (m, g.m)
is proper (pre-images of compact sets are compact).
For instance, the action of a group on itself by left or right multiplication is proper (because
the map G × M → M × M, (g.m) 7→ (m, g.m) is a diffeomorphism in that case). Actions of
compact groups are always proper. Given a proper G-action, the induced action of a closed
subgroup is also proper.
The “irrational flow” on a 2-torus is a non-proper R-action.
Lemma 1.19. The stabilizer groups Gm for a proper group action are all compact.
Proof. The pre-image of the point {m} × {m} ∈ M × M is Gm × {m}.
Thus for example the conjugation action of G on itself is not proper unless G is compact.
Proposition 1.20. The orbits O for a proper G-action on M are embedded, closed sub-
manifolds, with
Tm (O) = {ξM (m)| ξ ∈ g}.
Proof. We need to check the proposition near any given m ∈ O. Consider the map
φ : G → M taking g to g.m. This map is G-equivariant for the left-action of G on itself,
hence it has constant rank. Hence by the constant rank theorem its image is an immersed
submanifold: for each g ∈ G there exists an open neighborhood U ⊂ G such that φ(U ) is an
embedded submanifold of M , with tangent space the image of the tangent map. In particular
this applies to g = e, where the image of the tangent map g 7→ Tm M is clearly spanned by the
generating vector fields. We claim that W ∩ φ(G) = W ∩ φ(U ) for W ⊂ M a sufficiently small
10 CONTENTS
open neighborhood of m. Since φ(U ) = φ(U h) for h ∈ Gm , it is no loss of generality to assume

that U is invariant under right-translation by Gm :
U Gm = U.
Now if the claim was false, we could find a sequence of points gk .m ∈ O, converging to m,
but with gk 6∈ U . The sequence (m, gk .m) ∈ M × M is contained in a compact set, hence
by properness the sequence (gk , m) ∈ G × M must be contained in a compact set. Hence, by
passing to a subsequence we may assume that the sequence gk converges: gk → g∞ . Passing
to the limit in gk .m → m, we see g∞ ∈ Gm . Since U is open, this means gk ∈ U for large k, a
contradiction.
In immediate consequence of the fact that the orbits for a proper action are closed, is that
the orbit space M/G is Hausdorff. Our main goal in this Section is the cross-section theorem,
giving a local normal form for the action near any G-orbit. A slightly weaker version of the
theorem may be stated as follows. Let O = G.m be an orbit, T O ⊂ T M |O its tangent bundle,
and
νO = T M |O /T O
the normal bundle. Both T M and T O carry actions of G by vector bundle automorphisms,
hence νO carries an induced G-action. The tubular neighborhood theorem says that there
exists a diffeomorphism ψ : νO → U ⊂ M onto an open neighborhood of O, such that ψ
restricts to the identity map on O. The main point of the cross-section theorem is that one
may choose the map ψ to be G-equivariant. Since G acts transitively on O, this then reduces
the problem of studying the G-action near O to the study of the linear action of H = Gm on
the fiber V = (νO )|m , called the slice representation.
We will begin our discussion of the slice theorem by considering two extreme cases: (i) the
action being free, and (ii) m being a fixed point for the G-action.
Theorem 1.21. If G → Diff(M ) is a proper, free action, the orbit space M/G admits a
unique smooth structure such that the quotient map π : M → M/G is a submersion. It makes
M into a principal G-bundle over M/G.
We recall that a principal H-bundle is a H-manifold P together with a smooth map π :
P → B onto another manifold, having the following local triviality property: For each x ∈ B,
there exists an open neighborhood U of x and a H-equivariant diffeomorphism
π −1 (U ) → U × H
where the H-action on the right hand side is h.(x, h1 ) = (x, h1 h−1 ). One calls P the total
space and B the base of the principal bundle. The maps π −1 (U ) → U × H are called local
trivializations.
Proof. Since the action is free, the infinitesimal action map gives an isomorphism Tm (G.m) =
g for all m ∈ M . Choose a submanifold S ⊂ M with m ∈ S and g ⊕ Tm S = Tm M . The action
map restricts to a smooth map G × S → M , with tangent map at (e, m) equal to the given
splitting. By continuity, the tangent map stays invertible at (e, m′ ) for m′ ∈ S sufficiently
close to m. Thus, choosing S sufficiently small we may assume this is true for all points in
S. By equivariance, it then follows that G × S → M has invertible tangent map everywhere,
thus it is a local diffeomorphism onto its image. In fact, choosing S smaller if necessary it be-
comes a diffeomorphism onto its image: Otherwise, we could choose a non-convergent sequence
(gk , mk ) ∈ G × S with mk → m and gk .mk = m. Thus gk−1 .m = mk → m. The sequence of
points (m, mk ) = (gk .mk , m) is contained in a compact set, hence by properness the sequence gk
is contained in a compact set. The equation gk−1 .m = mk → m shows that for any convergent
subsequence of gk , the limit stabilizes m, hence is e. Thus gk → e, contradicting the choice of
sequence. The diffeomorphism G × S → V ⊂ M identifies V /G with S, hence gives a manifold
structure on V /G. This defines a smooth structure on M/G such that the quotient map is a
submersion. 1
It is important to note that the Theorem becomes false if we consider principal bundles in
the category of topological spaces. In the definition of a topological principal H-bundle, one
replaces manifolds by topological spaces, smooth maps by continuous maps, and Lie groups by
topological groups.
Exercise 1.22. Give an example of a compact topological space with a free action of Z2
that does not define a principal bundle. (There is an easy such example, using the indiscrete
topology on a set. Give an example where the space has a more reasonable topology, e.g. with
P a subspace of R3 .)
Let M be a manifold with a proper free G-action. Given an action of a second group
G1 on M , such that the actions of G and G1 commute, the quotient space inherits a natural
G1 -action. (Smoothness is automatic by properties of quotient maps). For instance, let H
be a closed subgroup of G. The restriction of the right action to H is proper, and commutes
with the G-action from the left. Hence the induced action on G/H makes G → G/H into a
G-equivariant principal H-bundle.
At another extreme, we now consider the case that a point m ∈ M is fixed under the group
action. For proper actions, this can only happen if the group is compact.
Theorem 1.23. Let M be a G-manifold, with G compact, and m ∈ M G a fixed point for
the action. Then there exists a G-invariant open neighborhood U of m and a G-equivariant
diffeomorphism Tm M ∼= U taking 0 to m.
Proof. Choose a G-invariant Riemannian metric on M . The exponential map exp0 :
Tm M → M for this metric is G-equivariant, and its differential at the identity is invertible.
Hence it defines a G-equivariant diffeomorphism Bǫ (0) → U ⊂ M for ǫ > 0 sufficiently small.
But Bǫ (0) is diffeomorphic to all of V , choosing a diffeomorphism preserving radial directions.

Proposition 1.24. Let G × M → M be a proper G-action, and H ⊂ G a subgroup.
Then the each component of the set M H of H-fixed points is a (topologically) closed embedded
submanifold of M .
Proof. We first observe that the fixed point set of H coincides with the fixed point set of
its closure H. Indeed, if m ∈ M is H-fixed, and hi ∈ H a sequence converging to h ∈ H, then
1Compatibility of the local manifold structures is automatic: For a given map F : M → N , from a manifold
M onto a set N , there is at most one manifold structure on N such that F is a submersion: A function on N is
smooth if and only if its pull-back to M is smooth.
12 CONTENTS
h.m = limi hi .m = m, showing H ⊂ Gm . Since Gm is compact, H is compact also. Thus we

may assume H is a compact embedded Lie subgroup of G. The Lemma (with G = H) shows
that a neighborhood of m in M is H-equivariantly modeled by a neighborhood of 0 in Tm M .
In particular M H corresponds to (Tm M )H , which is a linear subspace of Tm M .
To formulate the slice theorem, we need some more notation. Suppose P → B is a principal
H-bundle. For any linear representation of H on a vector space V , the quotient (P × V )/H is
naturally a vector bundle over P/H = B: Indeed, any local trivialization π −1 (U ) = U × H of
P gives rise to a fiberwise linear local trivialization
(π −1 (U ) × V )/H = (U × H × V )/H ∼
=U ×V
of (P × V )/H. One calls this the associated vector bundle with fiber V , and writes
P ×H V := (P × V )/H → B
If another group G acts on P by principal bundle automorphisms (i.e. the action of G commutes
with the action of H), the associated bundle becomes a G-equivariant vector bundle.
A special case of this construction is the case P = G, with H a closed subgroup of G acting
by the right action. In this case G×H V is a G-equivariant vector bundle over the homogeneous
space G/H. It is easy to see that any G-equivariant vector bundle over a homogeneous space
is of this form. The following result is due to Koszul and Palais.
Theorem 1.25 (Slice theorem for proper actions). Let M be a manifold with proper action
of G. Let m ∈ M , with stabilizer H = Gm , and denote by V = Tm M/Tm (G.m) the slice
representation. Then there exists a G-equivariant diffeomorphism G ×H V → M taking [(e, 0)]
to m.
Proof. As above, we can choose an H-equivariant diffeomorphism Tm M → U ⊂ M ,
taking 0 to m. Identifying V with a complement of Tm (G.m) in Tm M (e.g. the orthogonal
complement for an invariant inner product), we obtain an H-equivariant embedding V → M , as
a submanifold transversal to the orbit G.m. It extends to a G-equivariant map G ×H V → M .
As before, we see that the map G ×H V → M is in fact a diffeomorphism over G ×H Bǫ (0) for
ǫ sufficiently small. Choosing an H-equivariant diffeomorphism V ∼= Bǫ (0) the Slice Theorem
is proved.
Corollary
S 1.26 (Partitions of unity). If G × M → M is a proper group action, and
M = α Uα a locally finite cover by invariant open sets, there P
exists an invariant partition of
unity: I.e functions χα supported in Uα , with 0 ≤ χα ≤ 1, and α χα = 1.
Proof. Exercise.
Corollary 1.27 (Invariant Riemannian metrics). If G×M → M is a proper group action,
there exists a G-invariant Riemannian metric on M .
Proof. Given an H-invariant inner product on V and an H-invariant inner product on
g, one naturally constructs an invariant Riemannian metric on G ×H V . This constructs the
desired metric near any orbit, and globally by a partition of unity.
Corollary 1.28 (Equivariant tubular neighborhood theorem). If G × M → M is a proper
group action, and N ⊂ M a G-invariant embedded submanifold with normal bundle νN , there
exists a G-equivariant diffeomorphism from νN to a neighborhood of N in M , restricting to the

identity map from the zero section N ⊂ νN to N ⊂ M .
Proof. This follows from the standard proof of the tubular neighborhood theorem, by
using a G-invariant Riemannian metric on M .
1.5. The orbit type decomposition. We will now describe the orbit type stratification
of a manifold
S with a proper action. 2 We begin with some general definitions. A decomposition
X = i Xi of a topological space is called locally finite if each compact set in X meets only
finitely many Xi . It satisfies the frontier condition if
Xi ∩ Xj 6= ∅ ⇒ Xi ⊂ Xj .
In this case define a partial ordering of the pieces Xi , and in fact of the indexing set, by setting
i ≤ j ⇔ Xi ⊂ Xj . The depth of a piece Xi is defined to be the largest k for which there exist
pieces Xij with ik > ik−1 > i1 ≥ i. The depth of a (finitely) decomposed space is the largest
depth of any of its pieces.
If X is a decomposed space, then also the open cone over X,
cone(X) := [0, ∞) × X/({0} × X)
is a decomposed space, with pieces the tip of the cone together with all cone(X)i = (0, 1) × Xi .
Clearly,
depth(cone(X)) = 1 + depth(X)
Following Sjamaar-Lerman we define the notion of a stratified singular space3 as follows:
Definition 1.29.
S A depth k stratification of a topological space X is a locally finite de-
composition X = i Xi satisfying the frontier condition, with each Xi a smooth manifold
(called the stratum), with the following property: For each m ∈ Xi ⊂ X there exists an open
neighborhood U ⊂ Xi around m, and a stratified space L of depth at most k − 1, together with
a homeomorphism
U × cone(L) → V ⊂ X
preserving the decompositions, and restricting to diffeomorphisms between strata.
Now let M be a manifold with a proper G-action, and X = M/G S the orbit space. (Recall
that this is a singular space, in general). The decomposition M = (H) M(H) into orbit types
S
induces a decomposition X = (H) X(H) where X(H) = M(H) /G. Decompose X further as
[
X= Xi
i
where each Xi is a component of some X(Hi ) , and
[
M= Mi
2A good reference for this material is the paper by Sjamaar-Lerman [28]. For stratified spaces in general,
see e.g. the book [15] by Goresky-MacPherson.
3There are other notions of a stratified space, the most common definition being due to Whitney. See
Goresky-MacPherson [15] for this definition and Duistermaat-Kolk [13] for its application in the context of
group actions.
14 CONTENTS
be the corresponding decomposition of M , where Mi is the pre-image of Xi . (Of course, if G

is connected then the Mi will be connected.) We will call this the orbit type decomposition of
M and of the orbit space X, respectively.
S S
Theorem 1.30 (Orbit type stratification). The decompositions M = Mi and X = Xi
are locally finite and satisfy the frontier condition. Each Mi is a smooth embedded submanifold
of M , and Xi = Mi /G inherits a unique manifold structure for which the quotient map is a
submersion. With these manifold structures, the above decompositions are in fact stratifications.
Proof. Near any orbit G.m ⊂ M(H) , with Gm = H, M is modeled by the associated
bundle G ×H V , with action
g1 .[(g, v)] = [(g1 g, v)]
By definition of the associated bundle, we have [(g1 g, v)] = [g, v] if and only there exists h ∈ H
with (g1 g, v) = (gh−1 , h.v). This means h ∈ Hv and g1 = gh−1 g −1 . That is, the stabilizer
group of [(g, v)] is
G[(g,v)] = Adg (Hv ),
where Hv ⊂ H is the stabilizer of v. In particular, all stabilizer groups of points in the model
are subconjugate to H. The stabilizer group is conjugate to H if and only if Hv = H. This
shows
(G ×H V )(H) = G ×H V H = V H × G/H
which is a vector subbundle, in particular a submanifold. This shows that all Mi are embedded
submanifolds. Furthermore,
(G ×H V )(H) /G = V H
showing that the Xi are smooth manifolds also.
We next analyze how the strata fit together. Note that G[(g,v)] = G[(g,tv)] for all t 6= 0. It
follows that each orbit type stratum (G ×H V )(H1 ) for (H1 ) < (H) is invariant under scaling
(g, v) 7→ (g, tv). In particular, the zero section G/H is in the closure of each orbit type stratum.
It follows that a component Mi of M(H) can meet the closure of some M(H ′ ) only if (H ′ ) < (H),
and in that case it is in fact contained in the closure.
Choose an H-invariant inner product on V and let W be the orthogonal complement of
H
V in V . We have,
G ×H V = V H × (G ×H W ) = V H × (G ×H cone S(W ))
where S(W ) ⊂ W is the unit sphere bundle. Notice that all stabilizer groups for the action
of H on S(W ) are proper subgroups of H. Thus, by induction, we may assumeSthat the orbit
type decomposition for the H-action on S(W ) gives a stratification: S(W ) = j S(W )j . By
the above discussion, the orbit type decomposition for G ×H V is given by
(G ×H V )j = V H × (0, ∞) × (G ×H S(W )j ),
together with the stratum V H × G/H. This shows that the orbit type decomposition is a
stratification. The orbit space X = M/G is locally modeled by
(G ×H V )/G = V /H = V H × W/H = V H × cone(S(W )/H)
and induction shows that this is a stratified singular space.
Each M(H) is naturally a fiber bundle over X(H) , with fiber G/H. One can be more precise:
Theorem 1.31. For any H ⊂ G there is a natural principal K-bundle over P(H) → X(H) ,
with K = NG (H)/H, such that
M(H) = P(H) ×K (G/H).
Proof. Recall that MH = {m ∈ M | Gm = H}. Clearly G.MH = M(H) . In the local model
G ×H V ,
(G ×H V )H = NG (H)/H × V H .
Thus MH is a submanifold. If m ∈ MH , then g.m ∈ MH if and only if g ∈ NG (H). The
stabilizer for the action of NG (H) on MH is exactly H everywhere. Thus K = NG (H)/H acts
freely, with quotient X(H) . Let K = NG (H)/H act on G/H by the action induced from the
right multiplication, and let G act by left multiplication. Then
MH ×K G/H → M(H) , [(m, gH)] 7→ g.m
is well-defined, and is a diffeomorphism.
1.6. Principal orbit type theorem.

Theorem 1.32 (Principal orbit type theorem). Let G × M → M be a proper group action,
with connected orbit space M/G. Among the conjugacy classes of stabilizer groups, there is a
unique conjugacy class (Hprin ) with the property that (Hprin ) < (H) for any other stabilizer
group H = Gm . The corresponding orbit type stratum Mprin := M(Hprin ) is open and dense in
M , and its quotient Xprin = M(Hprin ) /G ⊂ X is open, dense and connected.
On calls Hprin (or any subgroup conjugate to it) a principal stabilizer, and Mprin , Xprin the
principal stratum of M, X. Note that if G is connected, then Mprin itself is connected.
Proof. Note first that by definition of “depth”,

depth(Mi ) > 0 ⇔ ∃j 6= i : Mi ⊂ Mj
⇔ ∃j 6= i : Mi ∩ Mj 6= ∅
6 i : Mi ∩ Mj ∩ Mj 6= ∅
⇔ ∃j =
⇔ Mi is not closed.
Thus the orbit type strata of depth 0 are all open, and all other orbit type strata Mj are
embedded submanifolds of positive codimension. It follows that the union of depth 0 orbit
type strata is open and dense. We have to show that there exists a unique orbit type stratum
of depth 0 (recall that Xj = Mj /G are all connected by definition).
We use induction on the depth of the stratification. For any point m ∈ Mj with depth(Mj ) >
0, consider the local model G×H V = V H ×G×H W . where H = Gm and V = Tm M/Tm (G.m).
By induction on the dimension of the depth of the stratification, the theorem applies for the
H-action on S(W ). In particular, there exists a principal stabilizer Hprin ⊂ H for this action,
with S(W )(Hprin ) /H connected. Then Hprin is also a principal stabilizer for the G-action on
G ×H V , and
(G ×H V )(Hprin ) /G = V(Hprin ) /H = V H × (0, ∞) × S(W )(Hprin ) /H
16 CONTENTS
is connected, open and dense. This shows that if Mi , Mi′ have depth 0, and their closures
intersect in m ∈ Mj , then the two must be equal. 4
In general, Mprin need not be connected even if M is connected: A counterexample is the
action of Z2 on R generated by t 7→ −t.
Besides the orbit type decomposition, one can also consider the decomposition into infini-
tesimal orbit types, by partitioning M into
M(h) = {m ∈ M | gm ∼ h}
(where ∼ denotes G-conjugacy of subalgebras of g). and then decomposing each M(h) further
into connected components. Thus M(h) is the union of all M(H) where H ranges over subgroups
having h as its Lie algebra. Again, the local model shows that each M(h) is a submanifold, in
fact
(G ×H V )(h) = G ×H (V h)
where V h is the subspace fixed by h. If we decompose M into the M(h) ’s and then decompose
further into connected components Mi , we again obtain a stratification. We call these the
infinitesimal orbit type strata. The same inductive argument as before shows that if M is
connected, then there exists a unique open stratum, which we denote Mreg and call regular
elements, following Duistermaat-Kolk [13]5
Theorem 1.33. For any proper G-action on a connected manifold M , the set Mreg of
regular elements is open, dense and connected.
Proof. The complement is the union of all M(h) with h a non-minimal infinitesimal sta-
bilizer. In the local model, (G ×H V )(h) = G ×H V h. But V h has codimension at least two
in V , since h is the Lie algebra of a compact group, and non-trivial representations of com-
pact groups are at least 2-dimensional. Hence M(h) has codimension at least 2, so removing it
doesn’t disconnect M .
1.7. Example: The adjoint action of G on its Lie algebra. Let G be a compact,
connected Lie group acting on its Lie algebra g by the adjoint action. We would like to describe
the orbit type decomposition for this action, as well as the orbit space. Let T be a maximal
torus (i.e. a maximal connected abelian subgroup) in G, and t its Lie algebra. We will need
the following two facts from the theory of Lie groups:
a) Any two maximal tori are conjugate in G. (The standard proof of this is to show that
at all regular points, the map G/T × T → G, (gT, t) 7→ gtg −1 is orientation-preserving. Hence
it has positive mapping degree, which implies that it must by onto.)
b) If H is any torus in G, and g commutes with all elements of H, there exists a torus
containing H ∪ {g}. (To prove this, consider the closed subgroup B generated by H and g. It is
easy to see that B is the direct product of a torus and a cyclic group. Hence there exists x ∈ B
such that the subgroup generated by x is dense in B. Choose a maximal torus T containing x,
then also B ⊂ T and we are done.)
4Strictly speaking, the inductive argument only worked for finite depth stratifications. However, if
depth(M ) = ∞ the same proof shows how to go from finite to infinite depth: Note that the H-actions on
S(W ) always have finite depth, by compactness.
5The terminology “regular” is not entirely standard, in contrast to “principal”.
2. CLASSIFYING BUNDLES 17
Part b) shows in particular that maximal tori are maximal abelian. (The converse does not
hold in general.) It also shows that all stabilizers Gξ for the adjoint action are connected, and
in fact that Gξ is the union of all maximal tori for which the Lie algebra contains ξ. Indeed, if
g ∈ Gξ , then the torus H given as the closure of the 1-parameter subgroup exp(tξ) fixes g, so
there is a maximal torus containing H ∪ {g}. (Conversely, any such torus lies in Gξ .) Since all
Gξ are connected, orbit types and infinitesimal orbit types coincide.
For any maximal torus T , one can choose ξ such that the closure of exp(tξ) equals T . Then
Gξ = T , since any other element stabilizing ξ would commute with T . Thus, the principal
stabilizer for the adjoint action is (T ). To determine the corresponding principal orbit type
stratum, choose an invariant inner product on g, and let
g = gT ⊕ m = t ⊕ m
denote the orthogonal decomposition. 6 The subspace m decomposes into a direct sum of irre-
ducible representations of T . By representation theory of tori, any irreducible representation of
T is equivalent to a representation of the form exp(ξ) 7→ R2πhα,ξi where Rφ is the 2-dimensional
rotation defined by φ. The element α ∈ t∗ is a root and the corresponding 2-dimensional sub-
space mα is called a root space.7 We thus find that given elements ξ ∈ t, ζ ∈ g, commute if
and only if ζ has no component in any root space mα with hα, ξi = 0. We therefore find:
Theorem 1.34. For any ξ ∈ t, the infinitesimal stabilizer gξ is the direct sum of t together
with all root space mα such that hα, ξi = 0. The stabilizer group Gξ is the connected subgroup
of G with Lie algebra gξ .
The equations hα, ξi = 0 subdivide t into chambers, and each wall corresponds to a given
orbit type. We know that the interior of each chamber corresponds to the principal orbit type,
i.e their union is MT ⊂ M(T ) . Since the principal orbit type stratum is connected this easily
implies that the Weyl group NG (T )/T acts transitively on the set of chambers.
Of course, there is much more to be said about this example, see e.g. Broecker-tom Dieck
[7] or Duistermaat-Kolk [13].
2. Classifying bundles
2.1. Principal bundles. We recall the definition of principal bundles in the topological
category.
Definition 2.1. Let G be a topological group. A principal G-bundles is a topological
space P , together with a continuous action of G satisfying the following local triviality condi-
tion: For any x ∈ B = P/G there exists an open neighborhood U os x and a G-equivariant
homeomorphism
π −1 (U ) → U × G.
Here π : P → B is the quotient map and the action of G on U × G is given by g.(y, h) =
(y, hg −1 ). One calls P the total space and B the base of the principal bundle.
6In fact, m is the unique such complement, but this need not concern us here.
7The sign of α changes if one changes the orientation of R2 .
18 CONTENTS
Thus, a principal G-bundle is the special case of a fiber bundle, where all fibers are principal
homogeneous G-spaces (spaces with free transitive G-actions), and the local trivializations take
the fibers to the “standard” principal homogeneous G-spaces, G itself.
A morphism of principal bundles π : P → B and π ′ : P ′ → B ′ . is a commutative diagram
P / P′

B / B′
where the upper horizontal map is G-equivariant.
Examples 2.2. (a) The trivial bundle P = B × G, with action g.(y, h) = (y, hg −1 ).
(It is standard to write principal bundle actions as multiplications from the right.)
In general, a principal bundle is isomorphic to the trivial bundle if and only if it
admits a section σ : B → P, π ◦ σ = idB ; in this case the isomorphism is given by
B × G → P, (x, g) 7→ g −1 .σ(x).
(b) We had seen that if G is a Lie group acting freely and properly on a manifold M , then
P = M is a principal G-bundle over B = M/G. Important cases include: S n as a
principal Z2 -bundle over RP n , and S 2n+1 as a principal U(1)-bundle over CP n .
(c) More generally, for k ≤ n one has the Stiefel manifold StR (k, n) of orthonormal k-
frames in Rn ,
StR (k, n) = {(v1 , . . . , vk ) ∈ Rkn | vi · vj = δij }
as a principal O(k)-bundle over the Grassmann manifold GrR (k, n) of k-planes in Rn .

Indeed, the Stiefel manifold is a homogeneous space StR (k, n) = O(n)/O(n − k), where
we think of O(n − k) as the subgroup of O(n) fixing Rk ⊂ Rn . (Alternatively, StR (k, n)
may be thought of as the space of linear injections Rk → Rn preserving inner products.)
The Grassmann manifold is a homogeneous space GrR (k, n) = O(n)/(O(n−k)×O(k)).
The quotient map takes v1 , . . . , vk to the k-plane they span.
(d) Similarly we have a complex Stiefel manifold of unitary k-frames in Cn , which is a
principal U(k)-bundle over the complex Grassmannian GrC (k, n).
(e) If V is a real vector space of dimension k, the space of isomorphisms V → Rk is a
principal homogeneous space for GL(k, R). Thus is ρ : E → B is a real vector bundle
of rank k, we obtain a principal GL(k, R)-bundle Fr(E) → B (called the frame bundle)
with fibers π −1 (x) the space of isomorphisms ρ−1 (x) → Rk . (One can think of this
isomorphism as introducing a basis (frame) in ρ−1 (x), hence the name.) Similarly, if E
carries a fiberwise inner product, one has a bundle of orthonormal frames FrO (E) → B
which is a principal O(k) bundle.
(f) If E → B is a complex vector bundle of dimension k, one similarly defines the (complex)
frame bundle Fr(E) → B with structure group GL(k, C), and given fiberwise Hermitian
inner products, one defines a unitary frame bundle FrU (E) → B.
We will need two basic constructions with principal bundles π : P → B.

2.2. Pull-backs. Let π : P → B be a principal G-bundle. If f : X → B is a continuous

map, one defines a new principal bundle f ∗ B → X with fibers
(f ∗ P )x = Pf (x) .
(A local trivialization of P over U gives a local trivialization of f ∗ P over f −1 (U ).) One has a
commutative diagram,
f ∗P /P

X f
/B
which in fact defines f ∗P up to isomorphism.

Lemma 2.3. Let X be a paracompact Hausdorff space (e.g. CW complex or a manifold).
If f0 , f1 : X → B are homotopic maps, the pull-back bundles f0∗ P and f1∗ P are isomorphic.
The Lemma is essentially equivalent to the statement that for any principal bundle P over
X × I, the two pull-backs to X × {0} and X × {1} are isomorphic. We will omit the proof,
which can be found e.g. in Husemoller’s book.
Corollary 2.4. If f : X → Y is a homotopy equivalence between paracompact Hausdorff
spaces, the pull-back map sets up a bijections between PrinG (X) and PrinG (Y ). In particu-
lar, if X is a contractible paracompact Hausdorff space, then PrinG (X) has only one element
consisting of the trivial bundle.
Exercise 2.5. Show that if π : P → B is any principal bundle, π ∗ P ∼
= P × G.
2.3. Associated bundles. Let F be a topological space with a continuous G-action.
Then the associated bundle
P ×G F := (P × F )/G
is a fiber bundle over B = P/G with fiber F . Indeed, local trivializations π −1 (U ) → U × G
of P give rise to local trivializations ρ−1 (U ) → U × F of the bundle ρ : P ×G F → B. If F
is a vector space on which G acts linearly, the associated bundle is a vector bundle. If F is a
principal homogeneous H-space on which G acts by morphisms of such spaces, the associated
bundle is a principal H-bundle.
Examples 2.6.
Given a vector bundle E → B, the associated bundle Fr(E) ×GL(k,R) Rk for the defining action
of GL(k, R) recovers E.
Similarly, if P → B is any principal G-bundle, P ×G G = P for the left-action of G on itself.
If Pi → Xi are two principal G-bundles, one may view P1 ×G P2 as a fiber bundle over B1 with
fiber P2 or as a fiber bundle over B2 with fiber P1 .
If ρ : P ×G F → B is an associated bundle, then ρ∗ P ∼
= P × F since the following diagram
commutes:
P ×F /P

P ×G F ρ
/B
20 CONTENTS
2.4. Classifying G-bundles. We recall the following notions from topology: Given a
sequence of spaces and inclusions X 0 ⊂ X 1 ⊂ X 2 ⊂ · · · , the colimit X ∞ = colimn→∞ X n is
the union of the spaces X n , with the weak topology where a set is open in X if its intersection
with all X n is open. This implies that a function on X is continuous if and only if its restriction
to all X n is continuous. A particular case of this construction is a CW-complex: Here X 0 is
a discrete set, while X n is obtained from X n−1 by attaching n-cells, by a family of attaching
maps φλ : ∂Dn → X n−1 .
From now on, we will mostly deal with CW-complexes, or paracompact Hausdorff spaces
that are homotopy equivalent to a CW-complex. For example, smooth manifolds have CW-
complex structures. For compact topological manifolds, the existence of CW-complex struc-
tures is known, except in dimension 4 where this question is open. On the other hand, compact
topological manifolds are homotopy equivalent to CW-complexes. See Hatcher [18, p. 529].
Definition 2.7. A classifying principal G-bundle is a principal G-bundle EG → BG, where
the total space EG is contractible, and where BG is a paracompact Hausdorff space homotopy
equivalent to a CW-complex.
Examples 2.8. (a) Let S ∞ = colimn→∞ S n be the “infinite dimensional sphere”. S ∞
is an example of an infinite-dimensional CW-complex, with two cells in each dimension
(the upper and lower hemispheres). Each S n is a principal Z2 -bundle over RP n , and
this makes S ∞ into a principal Z2 -bundle over RP ∞ = colimn→∞ RP n . RP ∞ inherits
a CW-structure from S ∞ , with one cell in each dimension.
Lemma 2.9. The infinite sphere S ∞ is contractible.
Proof. (cf. Stöcker-Zieschang, “Algebraische Topologie, p.58) View S ∞ as the
P
“unit sphere” ||x|| = ( x2i )1/2 = 1 inside R∞ = colimn→∞ Rn . We denote by f :
R∞ → R∞ the shift operator
f (x1 , x2 , . . .) = (0, x1 , x2 , . . .)
and by x⋆ = (1, 0, . . .) our base point. Define a homotopy ht : S ∞ → S ∞ as follows:
(1 − t)x + t f (x)
ht (x) =
||(1 − t)x + t f (x)||
This is well-defined: the enumerator is never 0 since x, f (x) are always linearly inde-
pendent. (To see that this is a homotopy, it suffices to note that the restriction of
h : S ∞ × I → S ∞ to each S k × I is continuous.) The map h1 takes S ∞ onto the sub-
space A ⊂ S ∞ given by the vanishing of the first coordinate. Next, rotate everything
back onto the basis vector e1 = (1, 0, 0, . . .):
(1 − t)y + te1
h′t : A → S ∞ , y 7→
||(1 − t)y + te1 ||
By concatenating these two homotopies, we obtain the desired retraction from S ∞
onto the base point.
Thus S ∞ → RP ∞ is an EZ2 → BZ2 .
(b) We can also view S ∞ as the colimit of odd-dimensional spheres S 2n+1 ⊂ Cn+1 . Let
CP ∞ = colimn→∞ CP n . The principal U(1)-bundles S 2n+1 → CP n define a principal
U(1)-bundle S ∞ → CP ∞ which gives a E U(1) → B U(1).
(c) The sequence of O(n)-bundles StR (k, n) → GrR (k, n) defines a principal O(n)-bundle
StR (k, ∞) = colimn→∞ StR (k, n) → GrR (k, ∞) = colimn→∞ GrR (k, n)
Lemma 2.10. The infinite-dimensional Stiefel manifold StR (k, ∞) is contractible.
Proof. A point in StR (k, ∞) is a k-tuple v = (v1 , . . . , vk ), where each vj ∈ R∞
and vi · vj = δij . Let f (k) : R∞ → R∞ be the kth iteration of the shift operator
introduced above. We define a homotopy by

ht (v1 , . . . , vk ) = Gram ((1 − t)v1 + tf (k) (v1 ), . . . , (1 − t)vk + tf (k) (vk ))
where Gram denotes Gram’s orthogonalization procedure. To see this is well-defined
we need to check that the vectors (1 − t)vj + tf (k) (vj ) are linearly independent for
P
all t. A linear dependence j λj ((1 − t)vj + tf (k) (vj )) = 0 would mean, however,
P P
that the vectors j λj vj and f (k) ( j λj vj ) are proportional, which only happens if
P
j λj vj = 0, a contradiction. At the end of the homotopy ht , the StR (k, ∞) has been
moved into the subspace StR (k, ∞) ∩ span(e1 , . . . , ek )⊥ . We define a homotopy h′t from
this subspace onto the frame (e1 , . . . , ek ) by letting

h′t (w1 , . . . , wk ) = Gram te1 + (1 − t)w1 , . . . , tek + (1 − t)wk .
Concatenation of these two homotopies gives the desired retraction from Gr(k, ∞) onto
the frame (e1 , . . . , ek ).
Thus StR (k, ∞) → GrR (k, ∞) is a classifying O(n)-bundle.
(d) Similarly, we have a classifying U(n)-bundle,
StC (k, ∞) = colimn→∞ StC (k, n) → GrC (k, ∞) = colimn→∞ GrC (k, n)
(e) Using Stiefel manifolds of not necessarily orthonormal frames, we similarly get classi-
fying bundles for GL(n, R), GL(n, C).
(f) Let EG → BG be a classifying G-bundle, where G is a Lie group. Let H ⊂ G be a
closed subgroup. Restricting the action to H, we can take EH = EG with
BH = EG/H = EG ×G (G/H).
(Our assumptions on G, H imply that this is a CW-complex. Of course, this holds
under much more general assumptions.) In particular, we have constructed classifying
bundles for all compact groups G, since any such group may be presented as a subgroup
of U(n), and for all Lie groups admitting a faithful finite-dimensional representation.
(g) Let Σ be a compact connected 2-manifold (other than the 2-sphere or RP 2 ) with base
x0 , and let G = π1 (Σ, x0 ) be its fundamental group. Then the universal covering
Σ̃ → Σ defines a classifying bundle, EG → Σ = BG since the universal cover of such
a surface is diffeomorphic to R2 .
(h) The easiest examples of classifying bundles are: EZ = R as a bundle over BZ = S 1 ,
and ER = R as a bundle over BR = pt.
It is not immediately clear that classifying bundles exist in general. Let us however establish
some properties of such bundles. One important property will be that for any “reasonable”
topological space X, the space PrinG (X) of isomorphism classes of principal G-bundles ober
22 CONTENTS
X is in 1-1 correspondence with the space [X, BG] of homotopy classes of continuous maps
f : X → BG, by the pull-back construction. A sufficient condition will be that X is a CW-
complex. In particular, manifolds are certainly allowed.
Theorem 2.11. Let EG → BG be a universal principal G-bundle, and P → X any prin-
cipal G-bundle where the base X is a paracompact Hausdorff space homotopy equivalent to a
CW-complex. Then the map assigning to each continuous map f : X → BG the pull-back
bundle P = f ∗ EG → X sets up a bijection,
PrinG (X) = [X, BG].
Proof. We may assume that X is a CW-complex. Given any principal G-bundle P → X,
consider the associated bundle P ×G EG. This is a fiber bundle with fibers EG. Since the fibers
are contractible, and X is a CW-complex, one can one can construct a section σ : X → P ×G EG
of this fiber bundle, by induction. The induction starts by choosing pre-images in each point
over the 0-skeleton X 0 ⊂ X. Having constructed the section over the k − 1-skeleton, one wants
to extends the section over the k-skeleton. For any characteristic map Φ : Dk → X k ⊂ X, the
pull-back Φ∗ P → Dk admits a trivialization Dk × G, giving a trivialization of fiber bundles,
Φ∗ (P ×G EG) ∼ = Dk × EG.
The given section of P ×G EG over Xk−1 amounts to a continuous map S k−1 → ∂Dk → EG,
which we would like to extend to a map Dk → EG. Equivalently, we need a map S k−1 ×I → EG
equal to the given map on S k−1 × {0} and equal to a constant map on S k−1 × {1}. Such a
map is obtained by composing the map S k−1 × I → EG × I (equal to our given map on the
first factor) with a contraction, EG × I → EG. This gives the desired section over X k . Since
a map from a CW-complex is continuous if and only if its restriction to all X k is continuous,
this gives the desired section. By a similar argument, one shows that any two sections σ0 , σ1
are homotopic.
The section σ : X → P ×G EG lifts uniquely to a G-equivariant section σ̂ : P → P × EG.
(Indeed, for any p ∈ P with base point x ∈ X, the fiber over σ(x) ∈ P ×G EG contains a
unique point of the form (p, y), and this will be σ̂(p).) Thus we get a commutative diagram,
P / P × EG / EG × P / EG
σ̂ =

B / P ×G EG / EG ×G P / BG
σ =
in which the upper horizontal maps are G-equivariant. The composition of the lower horizontal
maps gives a map f : X → BG, and the diagram shows P ∼ = f ∗ EG. Homotopic sections σ
give rise to homotopic f ’s, and therefore isomorphic G-bundles.
For any principal G-bundle P → X, a map f : X → BG with f ∗ EG ∼ = P is called a
classifying map for P . We have shown that the choice of a classifying map is equivalent to the
choice of a section of the bundle P ×G EG → X.
Theorem 2.12. If EG → BG and E ′ G → B ′ G are two classifying bundles, where BG, B ′ G
are paracompact Hausdorff spaces having the homotopy type of CW complexes, there exists a
homotopy equivalence B ′ G → BG that is covered by a G-equivariant homotopy equivalence
E ′ G → EG. In this sense classifying bundles are unique up to homotopy equivalence.
Proof. We have a classifying map f : B ′ G → BG for E ′ G → B ′ G (viewing EG → BG

as the classifying bundle) and also a classifying map g : BG → B ′ G (viewing E ′ G → B ′ G as
the classifying bundle). The composition f ◦ g : BG → BG is a classifying map for EG itself,
so it must be homotopic to the identity. Similarly g ◦ f is homotopic to the identity.
Examples 2.13. It is known that for any path connected space X with base point x0 , the
fundamental group π1 (X, x0 ) is related to the set of homotopy classes of (not necessarily base
point preserving) maps [S 1 , X] by
[S 1 , X] = π1 (X, x0 )/ Ad(π1 (X, x0 )),
quotient by the conjugation action.8 It follows that the space of principal G-bundles over S 1 is
in 1-1 correspondence with the set of conjugacy classes in π1 (BG). For G = Z2 , we know that
π1 (RP ∞ ) = π1 (RP 2 ) = Z2
which tells us that there are exactly two principal Z2 -bundles over S 1 , up to isomorphism: The
trivial bundle and the non-trivial double covering. For G = U(1), we find π1 (CP ∞ ) = {1},
which means that any principal U(1)-bundle over S 1 is trivial. Of course, these facts are easy
to check directly, and in general the result PrinG (X) = [X, BG] is hardly useful to actually
determine the set PrinG (X).
Consider X = S n , (n > 1), with its standard cover by open sets U, V given as the com-
plement of north pole and south pole. Since U, V ∼ = Dn , and principal G-bundle over U, V
is trivial. If G is path connected, these trivializations are moreover unique up to homotopy.
On the intersection, the trivializations differ by a transition map U ∩ V → G. Homotopic
transition maps give rise to isomorphic bundles, and conversely, any transition map defines a
bundle. Since U ∩ V ≃ S n−1 , it follows that
[S n−1 , G] = PrinG (S n ) = [S n , BG].
In particular, G-bundles over S 2 are classified by [S 1 , G] = π1 (G) (using that the fundamental
group of a topological group is abelian).
Exercise 2.14. Let Σ be a 2-dimensional CW-complex. Show (directly) that if G is simply
connected, PrinG (Σ) consist of only one element, given by the class of the trivial bundle. If
G is furthermore a semi-simple, show that a similar statement holds true for 3-dimensional
CW-complexes. (Use that the fact that π2 (G) is trivial for such groups.) What does all this
imply for the topology of BG? (Remark: Elements in Hn (X) are represented by maps from
n-dimensional simplicial complexes into X. See Hatcher [18, p.109].)
Exercise 2.15. Show that if X is a CW-complex, with X 2 its 2-skeleton, PrinU(1) (X) =
PrinU(1) (X 2 ). (Hint: For n ≥ 2, any map S n → U(1) is homotopic to the constant map, since
any such map can be lifted to the universal cover R → U(1) and R is contractible.)
For G = GL(n, R) (and similarly for GL(n, C)), there is a more geometric way to see the
classifying map, at least if X is compact Hausdorff. Let P → X be a principal GL(n, R)-
bundle and E → X the associated vector bundle. Since X is compact, E is isomorphic to a
8A similar statement holds for higher homotopy groups: [S n , X] = π (X, x )/π (X, x ). See e.g. Davis-
n 0 1 0
Kirk, [10, Theorem 6.57].
24 CONTENTS
direct summand of a trivial vector bundle X × Rn , for n sufficiently large.9 Thus, for each
x ∈ X the fiber Ex defines a k-plane in Rn , i.e. a point in GrR (k, n). This gives a map
X → GrR (k, n) ⊂ GrR (k, ∞), with the property that E is the pull-back of the tautological
k-plane bundle over X. In the particular case where X = M is a k-dimensional embedded
submanifold of Rn , with tangent bundle T M ⊂ M × Rn , the resulting map M → GrR (k, n) is
known as the Gauss map.
2.5. Characteristic classes. Fix a commutative coefficient ring R (usually R = Z, R, Q, Z2 ).
Given P ∈ PrinG (X) with classifying map f : X → BG one obtains a pull-back map (charac-
teristic homomorphism)
c(P ) = f ∗ : H ∗ (BG, R) → H ∗ (X, R),
(depending only on the isomorphism P , since any two classifying maps are homotopic). The
image of this map is the ring of characteristic classes of P . The construction is functorial: If
F̃ : P → P ′ is a morphism of principal bundles covering a map F : X → X ′ on the base, the
diagram
H ∗ (BG, R)
c(P ′ )oo
ooooo c(P )
o
ow oo
∗ ′
H (X , R) / H ∗ (X, R)
F∗
commutes. For instance, if G = S 1 so that BG = CP ∞ , one knows that the cohomology ring
H ∗ (CP ∞ , Z) is a polynomial ring freely over one degree 2 generator α ∈ H 2 (CP ∞ , Z):
H ∗ (CP ∞ , Z) =< α > .
(That is, H q (CP ∞ , Z) vanishes for q odd, and for q = 2r equals Z with generator αr .) Hence,
the ring of characteristic classes of a principal U(1)-bundle P = f ∗ EG over X is a polynomial
ring in the Chern class,
c1 (P ) := c(P )(α) ∈ H 2 (X, Z).
It turns out that, in fact, the Chern class determines the U(1)-bundle up to isomorphism.
The characteristic rings H ∗ (BG, R) are known for many groups G (particularly Lie groups)
and coefficient rings R. We quote some results without proof, see e.g. Milnor-Stasheff [?], or
Bott-Tu [6].
(a) The cohomology ring of the infinite complex Grassmannian is
H ∗ (B U(n), Z) = Z[c1 , . . . , cn ],
a free polynomial ring in generators ci ∈ H 2i (B U(n)), where ci is called the ith Chern
class.
(b) The cohomology ring for SU(n) looks very similar:
H ∗ (B SU(n), Z) = Z[c2 , . . . , cn ]
i.e. it starts with a class c2 ∈ H 4 (SU(n), Z) in degree four.
9Proof: Using a partition of unity, one constructs sections σ , . . . , σ of E for which the images span the
1 n
fiber Ex at any point x. These sections determine a surjective bundle map X × Rn → E. We may identify E
with the orthogonal complement of the fiberwise kernel of this map.
(c) For G = O(n) one has, with Z2 -coefficients,

H ∗ (B O(n), Z2 ) = Z2 [w1 , . . . , wn ]
where wi ∈ H i (B O(n), Z) is called the ith Stiefel-Whitney class.
(d) With Q coefficients
H ∗ (B O(n), Q) = Q[p1 , . . . , p[n/2] ]
where pi ∈ H 4i (B O(n), Q) is called the ith Pontrjagin class, and [n/2] is the greatest
integer ≤ n/2.
(e) For G = SO(n), with n odd, the cohomology ring is generated by Pontrjagin classes
as before.10. If n is even, it is generated by the Pontrjagin classes together with the
Euler class e ∈ H n (B SO(n)) subject to one relation, e2 = pn/2 .
(f) For G a compact Lie group, and R = R, the cohomology ring H ∗ (BG) is isomorphic
to the ring (Sg∗ )G of invariant polynomials on g, with degrees doubled. It is a classical
fact that (Sg∗ )G itself is a polynomial ring with a finite set of generators. For instance,
if G = U(n) one can take A 7→ tr(Ak ) (1 ≤ k ≤ n) as generators.
2.6. Equivariant cohomology. Classifying bundles may be used to define the equi-
variant cohomology of a G-space X, using the Borel construction. Fix a classifying bundle
EG → BG.
Definition 2.16. For any G-space X, the equivariant cohomology ring HG ∗ (X, R) of X
∗
with coefficients in a commutative ring R is the cohomology ring H (XG , R) of the associated
fiber bundle
XG := EG ×G X.
The space XG is often called the Borel construction. Note that a different model for
the classifying bundle gives a homotopy equivalent Borel construction, and hence the same
equivariant cohomology ring. Let us describe a few basic properties of this construction:
2.6.1. G-maps. The Borel constructions is functorial with respect to G-maps. That is,
if f : X → Y is a G-equivariant map of G-spaces, one gets a map of Borel constructions
XG → YG , hence a ring homomorphism
f ∗ : HG
∗ ∗
(Y ) → HG (X).
If f0 ≃ f1 : X → Y are homotopic through G-maps then the induced maps in equivariant
cohomology coincide.
Taking Y = pt to be the trivial G-space, it follows that there is a natural homomorphism
∗ ∗
HG (pt) → HG (X).
It turns HG∗ (X) into a module over the ring H ∗ (pt). This is often a better point of view to
G
think about HG ∗ (X), e.g. H ∗ (X) is rarely finitely generated as an abelian group, but it often
G
∗ (pt)-module.
is as a HG
10More precisely, we pull-back the Pontrjagin classes for BO(n) back under the classifying map for
E SO(n) ×SO(n) O(n)
26 CONTENTS
2.6.2. Change of groups. Suppose G is a Lie group and H a closed subgroup. Recall that
EG → BG defines a classifying bundle EH → BH with EH = EG (viewed as an H-space)
and BH = EG/H = EG ×G (G/H). If X is a G-space, viewed as an H-space by restricting
the action, we get a natural map of orbit spaces,
EH ×H X → EG ×G X
taking H-orbits to G-orbits. This map XH → XG induces a map in cohomology, H ∗ (XG ) →
H ∗ (XH ), i.e. a ring (and also a HG
∗ (pt)-module) homomorphism
∗ ∗
HG (X) → HH (X).
In particular, there is a homomorphism HG ∗ (X) → H ∗ (X) to ordinary cohomology.
2.6.3. Products. In ordinary singular cohomology, the ring structure gives rise to a cross
product
H ∗ (X1 ) ⊗ H ∗ (X2 ) → H ∗ (X1 × X2 ), [α1 ] ⊗ [α2 ] 7→ pr∗1 [α1 ] ∪ pr∗2 [α2 ].
The Kuenneth theorem says that under favorable circumstances, e.g. if R = R, this map is
an isomorphism. More generally, if R is a principal ideal domain (e.g. R = Z), the map is
injective with cokernel given by torsion groups. (See e.g. Davis-Kirk [10, p. 56].) Similarly,
in equivariant cohomology we have a cross product HG ∗ (X ) ⊗ H ∗ (X ) → H ∗ (X × X ). This
1 G 2 G 1 2
map, however, is rarely an isomorphism, even if the coefficient ring is R. However, viewing
HG∗ (X) as a module over H ∗ (pt) (as we should), we can also tensor over H ∗ (pt) and get a
G G
cross product
∗ ∗ ∗
HG (X1 ) ⊗HG∗ (pt) HG (X2 ) → HG (X1 × X2 ).
This has a much better chance of being an isomorphism (for coefficients R = R), and often (but
not always) it is. In general, the relationship between the two is given by a certain spectral
sequence (see e.g. Hsiang, [19]).
2.6.4. Equivariant cohomology of principal bundles. In our construction of classifying maps,
we essentially proved the following
Proposition 2.17. The equivariant cohomology ring of a principal G-bundle P → X (with
X a paracompact Hausdorff space homotopy equivalent to a CW-complex) is the cohomology
ring of the base X.
Proof. The associated bundle PG = EG ×G P → BG can also be viewed as a bundle
P ×G EG → X. Since the fibers EG of this bundle are contractible, an argument similar to the
proof of Theorem 2.11 shows that PG retracts onto X. It follows that HG∗ (P ) = H ∗ (X).
2.6.5. Equivariant cohomology of homogeneous spaces.
Proposition 2.18. Let G be a Lie groups and H a closed subgroup. Then
∗ ∗
HG (G/H) = HH (pt) = H ∗ (BH).
The isomorphism is induced by the inclusion eH → G/H. (Note that H ∗ (BH) may be viewed
as a H ∗ (BG)-module.)
Proof. We observed above that EG, viewed a an H-space, is a model for EH with BH =
EG/H = EG ×G (G/H). Thus
∗
HG (G/H) = H ∗ (EG ×G (G/H)) = H ∗ (BH).
3. CONSTRUCTION OF EG BY SIMPLICIAL TECHNIQUES 27
The inclusion of pt = eH into G/H gives a map

BH = EG ×H (eH) → EG ×H (G/H) → EG ×G (G/H)
inducing maps in cohomology
∗ ∗
HG (G/H) → HH (G/H) → H ∗ (BH).

2.6.6. Mayer-Vietoris. Suppose X = U ∪ V is an open cover of the G-space X by G-
invariant open sets. Then we get an open cover XG = UG ∪VG of the Borel construction. By the
usual Mayer-Vietoris-sequence for this cover, we get a Mayer-Vietoris-sequence in equivariant
cohomology,
q q q q q+1
· · · → HG (X) → HG (U ) ⊕ HG (V ) → HG (U ∩ V ) → HG (X) → · · ·
More generally, for any open cover by G-equivariant open sets one has a spectral sequence for
HG∗ (X). In the manifold case, if G acts properly, we can always choose a cover by tubular
neighborhoods of orbits. For example, if M/G is compact, one can use this to prove that
HG∗ (M ) is a finitely generated H ∗ (pt)-module.
G
Example 2.19. Let U(1) act on S 2 by rotation about the z-axis. We want to calculate
k
HU(1) (S 2 ) with coefficients R = Z. Consider the open cover of S 2 given by the complement U
of the fixed point set, and the complement V of the equator. Then HU(1) k (U ) = 0 in degree
k > 0, since U(1) acts freely on U and the quotient retracts onto a point. On the other hand
k
HU(1) k
(V ) = HU(1) k
(pt) ⊕ HU(1) (pt). The Mayer-Vietoris sequence tells us therefore
k
HU(1) (S 2 ) = HU(1)
k k
(pt) ⊕ HU(1) (pt)
for k > 0.
2.6.7. Equivariant characteristic classes. The classifying bundle appears in two important
constructions: Characteristic classes and equivariant cohomology. This can be combined, yield-
ing equivariant characteristic classes. Namely, if P → X is a K-equivariant principal G-bundle,
one obtains a principal G-bundle PK → XK over the Borel construction; the characteristic
classes of this bundle live in H ∗ (XK ) = HK∗ (X) and are called the equivariant characteristic
classes.
3. Construction of EG by simplicial techniques

We will now explain a general construction of classifying bundles, using so-called (semi-
)simplicial techniques. We begin with the case of discrete groups.
3.1. Construction of EG for discrete groups. Let ∆n ⊂ Rn+1 be the standard n-
simplex,
Xn X
n
∆ ={ ti ei | ti = 1}.
i=0 i
Let G be a discrete group.11. Let ||EG|| be the infinite-dimensional simplicial complex, with
one standard n-simplex ∆n for each ordered n + 1-tuple (g0 , . . . , gn ) ⊂ Gn+1 (repetitions of gi
11For this case, the construction is nicely explained in Hatcher [18, p. 89]
28 CONTENTS
are allowed). This simplex is denoted [g0 , . . . , gn ], with elements gi as its vertices. Boundaries
of simplices are identified in the obvious way. A free G-action is given on simplices by
g.[g0 , . . . , gn ] = [g0 g −1 , . . . , gn g −1 ].
This action turns ||EG|| into a principal G-bundle, as is easy to verify.
Lemma 3.1. The space ||EG|| is contractible.
Proof. View the n-simplex [g0 , . . . , gn ] as the 0-face of the n + 1-simplex [e, g0 , . . . , gn ]. It
is the closed face opposite to the vertex [e]. Given x ∈ [g0 , . . . , gn ], let ht (x) = (1 − t)x + te,
using the linear structure in [e, g0 , . . . , gn ]. This defines a homotopy ht from ||EG|| onto [e].
(It is not a strong deformation retraction, since [e] is not fixed under the homotopy ht .)
Thus ||EG|| → ||BG|| = ||EG||/G is a model for the classifying bundle EG → BG.
Remark 3.2. Note that this argument did not involve the group structure of G, thus
it works for any set X, and the corresponding “free” simplicial complex with n-simplices
parametrized by X n+1 .
There is a closely related model, defined as follows. Call an n-simplex [g0 , . . . , gn ] degen-
erate if gj = gj+1 for some j. There is a natural map from such a simplex onto the simplex
[g0 , . . . , gi , gi+2 , . . . , gn ], collapsing the edge [gi gi+1 ] onto a vertex [gi ]. Let ∼ denote the equiv-
alence relation generated by such maps. Notice that
x ∼ x′ ⇒ ht (x) ∼ ht (x′ )
Thus the homotopy ht for ||EG|| induces a retraction of |EG| onto the simplex [e]. (Notice
that this time, it is actually a strong deformation retract.) Furthermore, for g ∈ G we have
x ∼ x′ ⇔ g.x ∼ g.x′
hence the G-action on |EG| is free. One may verify the local triviality condition, hence |EG|
is again a classifying bundle. The space |EG| is a CW-complex, with one cell for each non-
degenerate simplex.
Example 3.3. Let G = Z2 = {e, c}. Then there are 2n+1 n-simplices [g0 , . . . , gn ] for each
n, but only 2 non-degenerate ones: [e, c, e, c, . . .] and [c, e, c, e, . . .]. To construct |EG|, one
starts with two 0-simplices [e], [c]. Next one attaches two 1-simplices [e, c] and [c, e], obtaining
S 1 . One then attaches two 2-simplices [e, c, e] and [c, e, c]. (Notice that one of the three edges
of [e, c, e] is the degenerate edge [e, e], which gets mapped to [e].) The resulting space is S 2
with its standard CW -complex structure. Iterating, one finds that |EG| is just S ∞ with the
usual CW-complex structure. The space ||EG|| is much ’fatter’ and does not have such a nice
geometric interpretation.
3.2. Simplicial spaces. We may re-formulate the construction more systematically, as
follows. For n ≥ 0 denote [n] := {0, . . . , n}. A map f : [n] → [m] is called increasing if
f (i + 1) ≥ f (i) for all i, and strictly increasing if f (i + 1) > f (i) for all i. One may think of [n]
as the vertices of an n-simplex. Any increasing map determines a map of simplices,
X n n
X
∆(f ) : ∆n → ∆m , ti ei 7→ ti ef (i) .
i=0 i=0
Under composition of increasing maps we have ∆(f1 ◦ f2 ) = ∆(f1 ) ◦ ∆(f2 ). The inclusions of
the codimension one faces ∆n−1 → ∆n correspond to the face maps

i i j if j < i
∂ : [n − 1] → [n], ∂ (j) =
j + 1 if j ≥ i
The maps ∆n+1 → ∆n collapsing the edge from ei to ei+1 correspond to the degeneracy maps

i i j if j ≤ i
ǫ : [n + 1] → [n], ǫ (j) =
j − 1 if j > i
The face and degeneracy maps generate the set of all increasing maps. More precisely we have:
Lemma 3.4. Any increasing map f : [n] → [m] can be uniquely written as a composition
f = ∂ ik · · · ∂ i1 ǫj1 · · · ǫjl with i1 < · · · < ik and j1 < · · · < jl .
Proof. Suppose first that f is 1-1. Then f is uniquely described by its image, and clearly
f = ∂ ik · · · ∂ i1 where i1 < · · · < ik is the ordered list of indices that are not in the image. If
f : [n] → [m] is not 1-1, let j ≤ n − 1 be the largest index with f (j) = f (j + 1). Then f may
be uniquely written f = f ′ ǫj , where f ′ (j) < f ′ (j + 1). Using induction, we eventually find
f = gǫj1 · · · ǫjl where j1 < · · · < jl and g is 1-1.
Consider now a sequence of spaces Xn := Gn+1 . Any increasing map f : [n] → [m] induces
a continuous map
X(f ) : Xm → Xn , (g0 , . . . , gm ) 7→ (gf (0) , . . . , gf (n) ).
Under composition of increasing maps, X(f1 ◦ f2 ) = X(f2 ) ◦ X(f1 ). Our model ||EG|| was
defined as a quotient,
∞
a
||X|| = Xn × ∆n / ∼,
n=0
under the equivalence relations,
(x, ∆(f )(y)) ∼ (X(f )(x), y)
for every strictly increasing map f : [n] → [m]. These are exactly the “natural” identifications
made above. The model |EG| is defined similarly,
∞
a
|X| = Xn × ∆n / ∼,
n=0
dividing out the relations `

for all increasing maps. In both cases, the topology is that of a
colimit over the quotients N n
n=0 Xn × ∆ / ∼.
In this reformulation, the construction of a classifying bundle works for any topological
group G. ||EG|| is Milnor’s model [23], while |EG| is introduced in Segal’s paper [27]. As
pointed out by Segal, |EG| → |EG|/G may fail to be locally trivial, in general, but it is locally
trivial if G is somewhat reasonable (e.g., a Lie group). If G is a Lie group, exactly the same
argument as before gives a homotopy from ||EG||, |EG| onto [e]. There are many advantages
to having such a universal construction. For instance, it is immediately clear that any group
homomorphism H → G induces a map of classifying spaces.
30 CONTENTS
The universal construction of the classifying bundle goes back to Milnor, [23]. The sim-
plicial version of Milnor’s construction was developed by Dold-Lashof, Segal, Stasheff, and
Milgram.
Both the fat model ||EG|| and the lean model |EG| have their advantages and disadvan-
tages. One advantage of the lean model is that the map
|E(G × K)| → |EG| × |EK|,
induced by the two projections G × K → G and G × K → K, is a homeomorphism provided
one takes the “correct” product topology. 12 The analogue for the fat model is not true, even
as sets. We’ll return to this issue later.
Exercise 3.5. Convince yourself that the map |E(G × K)| → |EG| × |EK| is a bijection
of sets.
The construction of the classifying bundle generalizes as follows:
Definition 3.6. A simplicial space X• is a collection of topological spaces Xn , n = 0, 1, . . .,
together with continuous maps X(f ) : Xm → Xn for any increasing map f : [n] → [m], such
that X(f ◦ g) = X(g) ◦ X(f ) under composition of such maps, and X(id) = id. A simplicial
map F• : X• → X•′ between simplicial spaces is a collection of continuous maps Fn : Xn → Xn′
intertwining the maps X(f ), X ′ (f ).
If we let ORD denote the category with objects the sets [n] and morphisms the increas-
ing maps f : [n] → [m], we may rephrase the definition as follows: A simplicial space is a
contravariant functor from that category ORD into the category TOP of topological spaces.
A simplicial map is a natural transformation between two such functors. Replacing TOP by
other categories, one similarly defines simplicial sets, manifolds, groups, rings etc.
Definition 3.7. The geometric realization of a simplicial space X• is the quotient space
∞
a
|X| = ∆n × Xn / ∼,
n=0
under the equivalence relations,
(∆(f )(y), x) ∼ (y, X(f )(x))
for all y ∈ ∆n , x ∈ Xm , and any increasing map f : [n] → [m]. One similarly defines the fat
geometric realization ||X|| by only dividing out the strictly increasing maps.
The maps ∂i := X(∂ i ) : Xn → Xn−1 and ǫi := X(ǫi ) : Xn → Xn+1 are called the face and
degeneracy maps of the simplicial space.
Clearly, a simplicial map F• : X• → X• induces a continuous map |F | : |X| → |X ′ |
between the geometric realizations, and similarly between the fat geometric realizations. That
is, geometric realization is a functor from the category of simplicial spaces into the category of
topological spaces.
12The product should be taken in the category of compactly generated spaces. This is similar the problem
that while the direct product of two CW-complexes is again a CW-complex, the topology (given as a colimit) is
slightly different from the product topology.
3.3. Examples of simplicial spaces.

(a) Any topological space X may be viewed as a simplicial space, by taking all Xn = X
and all maps X(f ) the identity map. The geometric realization of this simplicial space
is just X itself.
(b) Given a topological space X, one may define a simplicial space En X = X n+1 with
(EX)(f ) : (x0 , . . . , xm ) → (xf (0) , . . . , xf (n) )
for any increasing map f : [n] → [m]. The same argument as for EG shows that the
geometric realizations of this simplicial space is contractible. The diagonal embedding
X → X n+1 gives a continuous map from X into the geometric realization, X → |EX|.
Note that if X = G is a group, then E• G is a simplicial group, and the structure as
a principal G-bundle is given by the action of G as a (simplicial) subgroup from the
right.
(c) Let X be a finite simplicial complex with N vertices. Choose an ordering on the set V
of vertices, to identify V = [N ]. Let Sn be the set of all increasing maps φ : [n] → [N ]
such that φ(0), . . . , φ(n) are the vertices of a (possibly degenerate) n-simplex in X,
with
S(f ) : Sm → Sn , S(f )(φ) = φ ◦ ∆(f ).
(Note that S• is a simplicial subset of E• V.) The non-degenerate simplices in Sn are
those given by strictly increasing maps φ : [n] → [N ], thus are in 1-1 correspondence
with the (geometric) simplices in X. Using this fact, it is easy to see that
|S| = X.
In the important special case that X = ∆k is the standard k-simplex, we write Sn :=
∆n [k]. Thus ∆n [k] is simply the set of morphisms f : [n] → [k].
(d) Let G be a topological group. Then Bn G := Gn is a simplicial space, with face maps

 (h2 , . . . , hn ) if i=0
∂i (h1 , . . . , hn ) = (h1 , . . . , hi hi+1 , . . . , hn ) if 0 < i < n

(h1 , . . . , hn−1 ) if i=n
and degeneracy maps
ǫi (h1 , . . . , hn ) = (h1 , . . . , hi , e, hi+1 , . . . , hn ).
It is easy to check that the map πn : En G → Bn G given by
(g0 , . . . , gn ) 7→ (g0 g1−1 , . . . , gn−1 gn−1 )
is a simplicial map. This identifies B• G as the base of the simplicial principal bundle
makes E• G → E• G/G = B• G.
(e) Generalizing this example, suppose X is a G-space. One may define a simplicial space
(XG )• by letting
(XG )n = Gn × X,
with face and degeneracy maps

 (h2 , . . . , hn , x) if i=0
∂i (h1 , . . . , hn , x) = (h1 , . . . , hi hi+1 , . . . , hn , x) if 0 < i < n

(h1 , . . . , hn−1 , hn .x) if i=n
32 CONTENTS
ǫi (h1 , . . . , hn , x) = (h1 , . . . , hi , e, hi+1 , . . . , hn , x).

This is naturally identified with En G ×G X, with quotient map
(g0 , . . . , gn , x) 7→ (g0 g1−1 , . . . , gn−1 gn−1 , gn .x)
Hence the geometric realization of this simplicial space is the Borel construction XG .
(f) Let X be a topological space, and U = {Ua , a ∈ A} an open cover of X indexed by an
ordered set A. Given a0 ≤ · · · ≤ an let
Ua0 ,...,an = Ua0 ∩ · · · Uan
An = {(a0 , . . . , an ) ∈ An+1 | a0 ≤ · · · ≤ an , Ua0 ,...,an 6= ∅}.
Then A• is a simplicial set, with face maps and degeneracy maps inherited from E• A.
Its geometric realization is a simplicial complex, sometimes known as the nerve of the
open cover. Define a
Un X := Ua0 ,...,an
(a0 ,...,an )∈An
(disjoint union), with face maps induced by inclusions, and degeneracy maps the natu-
ral bijections Ua0 ,...,ai ,ai ,...,an → Ua0 ,...,an . There is a natural simplicial map Un X → X
induced by the inclusions of open sets.
Theorem 3.8. [24, Section 7] If X is a paracompact Hausdorff space, the geometric
realization
|UX| → |X| = X
of the map Un X → X is a homotopy equivalence.
Proof. We have to construct a homotopy inverse f : X → |UX| to the given
map g : |UX| → X. Choose a locally finite partition of unity χa subordinate to the
cover Ua . Given x ∈ X, let a0 < . . . < an be an ordered set of indices such that
Xn
χai (x) = 1, and x ∈ Ua0 ,...,an .
i=0
Define f (x) ∈ |UX| to be the image of
Xn

χai (x)ei , x ∈ ∆n × Ua0 ,...,an .
i=0
It is easily checked that f is well-defined (i.e. independent of the choice of a0 < . . . < an
(note that we do allow χai (x) = 0). Since the same collection of indices also works
on a neighborhood of x, it is clear that f is continuous. We have g(f (x)) = x by
construction. The composition f ◦ g : |UX| → |UX| is homotopic to the identity: The
required homotopy is induced by the homotopies I ×(∆n ×Ua0 ,...,an ) → (∆n ×Ua0 ,...,an ),
Xn n
X

t, ( si ei , x) 7→ ((1 − t)si + tχai (x))ei , x)
i=0 i=0

Remarks 3.9. (i) By essentially the same proof, the map ||UX|| → X is a
homotopy equivalence as well.
(ii) It may be useful to visualize the result for an open cover by just two open` sets
U0 , U1 . The geometric realization is obtained from a disjoint union U0 U1 by
gluing in a “cylinder” I × (U0 ∩ U1 ). One obtains an inclusion X → |UX|, where
the partition of unity shows how to embed X over the intersections U0 ∩ U1 . The
linear retraction of the cylinder onto the image of X give the desired homotopy
equivalence.
(iii) The result shows that all homotopy invariant topological invariants of X (in
particular its homology/cohomology) may be studied (at least in principle) in
terms of U• X. This is particularly interesting if U is a good cover, i.e. if all
Ua0 ,...,an are contractible. In this case the topology of Un X is trivial, and all
information on the topology of |UX| lies in the face and degeneracy maps. Below
this will lead us to a simplicial interpretation of Čech cohomology.
Classifying maps can be interpreted in the simplicial construction, as follows. Sup-
pose G is a Lie group, π : P → X is a principal G-bundle over a paracompact Haus-
dorff space, and Ua is a trivializing open cover of X. That is, over each Ua there is
a G-equivariant map φa : π −1 (Ua ) → G. Suppose as before that we have chosen an
ordering of the index set, and define a map
ψ : π −1 (Ua0 ,...,an ) → En G = Gn+1 , x 7→ (φa0 (x), . . . , φan (x)).
ψ is compatible with the face and degeneracy maps, hence it gives a simplicial map
(π −1 U)• P → E• G
where π −1 U is the cover of P by set π −1 (Ua ). Since ψ is G-equivariant, it descends to
a simplicial map
U• X → B• G.
The geometric realization of this map is the classifying map for P , composed with the
map X ≃ |UX| for some partition of unity, is a classifying map for P .
Below we will mostly work with the fat geometric realization, which has simpler properties
in a number of respects. There is, however, one important property of the (lean) geometric
realization: It is well-behaved under products.
Proposition 3.10 (Milnor,Segal). Let X• , X•′ be simplicial spaces, and let (X × X ′ )• be
their direct product, i.e. (X × X ′ )n = Xn × Xn′ . The natural map
|(X × X ′ )• | → |X• | × |X•′ |
induced by the two projections is a bijection of sets. It is a homeomorphism provided the
product on the right hand side is taken in the category of “compactly generated spaces”.
We indicate the main idea in an example (cf. Benson, p.25): Consider the simplicial space
∆• [1]. Its non-degenerate simplices are (0), (1), (01). Elements of (∆[1] × ∆[1])n are pairs of
increasing sequences (t0 . . . tn , t′0 . . . t′n ) where each ti , t′i is 0 or 1. Thus (∆[1] × ∆[1])• has
(1) four 0-simplices (0, 0), (0, 1), (1, 0), (1, 1),
(2) five non-degenerate 1-simplices (00, 01), (01, 00), (01, 01), (01, 11), (11, 01)
(3) two non-degenerate 2-simplices (001, 011), (011, 001).
34 CONTENTS
We hence see that the geometric realization is a square, could into two triangles along the
diagonal from (0, 0) to (1, 1). Similarly, in the general case:
Exercise 3.11. Show that the geometric realization of ∆• [k]×∆• [l] is the product ∆k ×∆l ,
subdivided into k + l-simplices in a certain canonical way. (The fact that ∆k × ∆l admits a
canonical subdivision enters the definition of the cross product in singular homology. See e.g.
Hatcher [18, p. 277].)
Remark 3.12. One can use this to give a very clean proof of the contractibility of the space
|E• X| onto a base point x∗ ∈ X ⊂ |E• X|. Thinking of I as the geometric realization of ∆• [1],
we would like to obtain the retraction as the geometric realization of a simplicial map
∆• [1] × E• X → E• X.
Recalling that ∆• [1] consists of increasing maps φ : [n] → [1], the map is defined as follows:
(φ, (x0 , . . . , xn )) = (x′0 , . . . , x′n )
where x′i = xi if φ(i) = 0, x′i = x∗ if φ(i) = 1. It is straightforward to check that this is a
simplicial map. An inclusion {1} ֒→ I is obtained by geometric realization of the simplicial
maps
{1}• → ∆• [1], 1 7→ (1, . . . , 1)
and similarly for {0}. The geometric realizations of the restricted maps
{1}• × E• X → E• X, ((1, . . . , 1), (x0 , . . . , xn )) 7→ (x∗ , . . . , x∗ ),
{0}• × E• X → E• X, ((0, . . . , 0), (x0 , . . . , xn )) 7→ (x0 , . . . , xn )

are the constant map and the identity map, respectively.
3.4. The homology and cohomology of simplicial spaces. Our goal is to develop
techniques for calculating the equivariant cohomology of a G-manifold M , particularly the
cohomology of BG. Any simplicial space X• defines a double complex
(C• (X• ), d, δ),
where d : Cq (Xp ) → Cq−1 (Xp ) is the usual boundary map and δ : Cq (Xp ) → Cq (Xp−1 ) is
defined in terms of the face maps as
p
X
δ= (−1)i (∂i )∗ .
i=0
Clearly, the two differentials d, δ commute. Define the associated total complex (C• (X), D),
where M
Ck (X) := Cq (Xp )
p+q=k
and D = d + (−1)q δ
on Cq (Xp ). (The sign guarantees that D squares to 0.) We define a
homomorphism ψ : Cq (Xp ) → Cp+q (||X||)
Cq (Xp ) → Cp+q (∆p × Xp ) → Cp+q (||X||),
where the first map is (−1)pq times cross-product with the identity map id∆p (viewed as a
singular chain id∆p ∈ Cp (∆p )), and the second map is push-forward under the quotient map
∆p × Xp → ||X||. Summing over all p + q = k, this gives a group homomorphism
(4) ψ : Ck (X) → Ck (||X||).
ψ is natural with respect to morphisms of simplicial spaces X• → Y• .
Theorem 3.13. [24, Theorem 4.2] The map (4) is a chain homotopy equivalence, and
hence induces an isomorphism in homology. That is, the singular homology of kXk may be
computed as the homology of the total complex associated to the double complex (Cq (Xp ), d, δ).
Sketch of proof. A detailed proof can ` be found in the Bott-Mostow-Perchik article.
Here are some of the ideas involved. Let π : ∞
n=0 ∆ n × X → kXk be the quotient map. For
n
any α ∈ Cq (Xp ), we have
ψ(Dα) = ψ(dα) + (−1)q ψ(δα)
= (−1)p(q−1) π∗ (id∆p ×dα) + (−1)q ψ(δα)
= (−1)pq dπ∗ (id∆p ×α) − (−1)pq π∗ (d id∆p ×α) + (−1)q ψ(δα)
Xp

= dψ(α) − (−1)pq (−1)j π∗ (∂ j × α) − π∗ (id∆p−1 ×(∂j )∗ α)
j=0
The sum is zero due to the identifications given by π. This shows that ψ(Dα) = dψ(α).
The proof that (4) is an isomorphism in homology is similar to the proof that the simplicial
homology of a simplicial complex equals is singular homology. Recall that the topology on ||X||
`
was defined by taking a colimit of spaces ||X||(N ) = π( N n
n=0 (∆ × Xn ). It may be shown (cf.
Bott-Mostow-Perchik) that any compact set in ||X||, and in particular the image of any singular
chain, is contained in some ||X||(N ) with N sufficiently large. This defines natural filtrations of
the chain complexes C• (kXk) and C• (X). One obtains a map between the spectral sequences
(cf. infra) associated to these filtrations, and the main point of the proof is now to show that
these spectral sequences coincide, already at the E1 stage.
If R is any abelian coefficient group (typically R = R, Z, Z2 ), we can consider homology with

coefficients in R, and the theorem shows that the homology groups of ||X|| with coefficients in
R can be computed from a double complex Cp (Xq ; R).
Dually there is a double complex of singular cochains,
(C • (X• ; R), d, δ),
where d is the usual coboundary map (dual to the boundary map, which also denote by d) and
q+1
X
δ= (−1)i (∂i )∗
i=0
(dual to the map δ for the chain complex). Let C k (X; R) be the total complex,
M
C k (X; R) = C q (Xp ; R).
p+q=k
36 CONTENTS
By dualizing the maps from homology, we see that there is a natural homotopy equivalence
C k (kXk; R) → C k (X; R).
Hence there is a canonical isomorphism between the cohomology of the total complex and the
cohomology of the fat geometric realization, ||X• ||.
Example 3.14. The group homology H̃• (G) of a discrete group G is defined as the homology
of a chain complex C̃• G, where C̃k (G) is the free Abelian group generated by elements of Gk ,
and the differential is given as
k−1
X
δ(h1 , . . . , hk ) = (h2 , . . . , hk ) + (−1)i (h1 , . . . , hi hi+1 , . . . , hk ) + (−1)k (h1 , . . . , hk−1 ).
i=1
Note that if we view elements of G as 0-simplices in G, this is just the complex (C0 (Bk G), δ)
inside the double complex Cp (Bq G). Dually, we have the cochain complex C̃ k (G) = Hom(Ck , Z)
and the corresponding group cohomology H̃ k (G). Elements of C̃ k (G) are functions φ : Gk → Z,
and the dual differential (once again denoted δ) is
k
X
(δφ)(h1 , . . . , hk+1 ) = φ(h2 , . . . , hk+1 )+ (−1)i φ(h1 , . . . , hi hi+1 , . . . , hk+1 )+(−1)k+1 φ(h1 , . . . , hk ).
i=1
Theorem 3.15. The inclusion C̃ k (G)

→ C 0 (Bk G) induces an isomorphism from the group
cohomology H̃ • (G) of G to the cohomology H • (BG) of the classifying space BG.
Proof. This will be “immediate” once we have the spectral sequence set-up, but we can
easily give a direct proof. Indeed, since G is discrete the d-cohomology is trivial.
We first show that the map H̃ k (G) → H k (BG) is surjective. Let α ∈ Z k (BG) represent a
class in H k (BG). Write α = α0,k +α1,k−1 +· · · . Since Dα = 0, we have in particular dα0,k = 0.
Since the d-cohomology is trivial, we can write α0,k = dβ 0,k−1 . Replacing α with α − Dβ 0,k−1 ,
and denoting the new form by α again, we achieve α0,k = 0. The new form has dα1,k−1 = 0,
so again we may subtract a D-coboundary to achieve α1,k−1 = 0. Iterating, we find that α
is D-cohomologous to a form in C̃ k (G) = C 0 (Bk G), closed under both d and δ. By a similar
argument, one shows that the map H̃ k (G) → H k (BG) is injective.
Example 3.16. Let X be a paracompact Hausdorff space, with cover U = {U | a ∈ A}
where A is an ordered set. Given a coefficient ring R consider the subcomplex
Č k (X, U, R) ⊂ C 0 (Uk X, R),
consisting of locally constant functions, in other words the kernel of d : C 0 (Uk X, R) →
C 1 (Uk X, R). If U is “good”, that is, each non-empty intersection Ua0 ...ak is contractible, ele-
ments of C 0 (Uk X, R) are in fact constant functions, thus are collections of elements fa0 ,...,ak ∈ R,
one for each non-empty intersection Ua0 ,...,ak with a0 ≤ · · · ≤ ak , and the differential reads,
k+1
X
(δf )a0 ···ak+1 = (−1)i fa0 ···âi ···ak .
i=0
The cohomology of this complex is called the Čech cohomology of X with respect to the cover
U.
4. SPECTRAL SEQUENCES 37
Theorem 3.17 (Isomorphism between Čech and singular cohomology). Suppose the cover
is “good”, that is, each non-empty intersection Ua0 ...ak is contractible. Then Ȟ k (X, U, R) is
canonically isomorphic to the singular cohomology H k (X, R). More precisely, the inclusion
map Č k (X, U, R) → C k (UX, R) gives an isomorphism
Ȟ k (X, U, R) → H k (||UX||, R) = H k (X, R).
Proof. This follows since the d-cohomology of the double complex C q (Up X) is trivial in
positive degree, as for the previous theorem.
Example 3.18. Let X be a simplicial complex, with a given ordering on its set of vertices,
and S• the corresponding simplicial set. Then |S• | = X, while kS• k is homotopy equivalent
to X. In this case, (C0 (S• ), δ) may be identified with the simplicial chain complex. Let
C∆k (X) := C 0 (S ) denote the simplicial cochain complex. Again, we easily see that the inclusion
k
into the double complex gives an isomorphism H∆ k (X) = H k (||S||) = H k (X).
4. Spectral sequences
Suppose (C •,• , d, δ) is a bi-complex: That is, C •,• is a bi-graded R-module for some coeffi-
cient ring R, and d, δ are two commuting differentials with
d : C p,q → C p,q+1 , δ : C p,q → C p+1,q .
One can then introduce the total complex
M
Ck = C p,q
p+q=k
with differential D = d + (−1)q δ. Our goal is to compute the cohomology of the complex
(C • , D). Cocycles for the differential D are elements of the form
α = α0,k + α1,k−1 + · · · + αk,0
satisfying a system of equations,
(5) dα0,k = 0
1,k−1
dα = ±δα0,k
···
0 = δαk,0 .
It is convenient to picture this system of equations in an array of boxes, labeled by (p, q), with
p increasing in horizontal direction and q increasing in vertical direction.
4.1. The idea of a spectral sequence. It seems natural to find solutions by induction.
(In a sense, a spectral sequence will be similar to a “power series” ansatz for solving an ordinary
differential equation. ) Let us consider a solutions having their first non-trivial term in the p, q
position, and extending downwards. That is, consider the system
(6) dαp,q = 0
p+1,q−1
dα = ±δαp,q
···
0 = δαk,0 .
38 CONTENTS
The first equation says that αp,q is a d-cocycle. This leads us to consider the d-cohomology
E1 := H(C, d), with bigrading
ker d ∩ C p,q
E1p,q := .
im d ∩ C p,q
The second equation has a solution if and only if δαp,q is exact. Put differently, the class of
αp,q in E1p,q should be closed under the differential d1 : E1p,q → E1p+1,q which is induced by δ.
This leads us to consider
ker d1 ∩ E1p,q
E2p,q := .
im d1 ∩ E1p,q
Classes in E2p,q are represented αp,q admitting an extension to an element αp,q +αp+1,q−1 solving
the first two equations. Iterating this idea will lead to the concept of spectral sequence, (Er , dr )
with Er+1 the cohomology of Er . Before we explain how to continue the sequence, let us explain
the abstract notion of a spectral sequence and what it actually computes.
4.2. What does the spectral sequence compute?
Definition 4.1. A spectral sequence is a sequence of bigraded differential complexes
(Erp,q , dr ), r = 0, 1, 2, . . . ,
where dr raises the total degree by 1 and the p-degree by r, with
p,q ker dr ∩ Erp,q
Er+1 = .
im dr ∩ Erp,q
We will only consider first quadrant bigraded spectral sequences, i.e. Erp,q = 0 unless
p, q ≥ 0. In this case
dr : Erp,q → Erp+r,q−r+1
is zero for r sufficiently large (namely, r > q + 1). Hence the spectral sequence stabilizes:
p,q
Er+1 = Erp,q for r suffiently large. The limiting groups are denoted E∞ p,q
.
Consider now a (first quadrant) double complex C p,q as above. Define the horizontal
L L
filtration of C •,• by direct sums F i (C) := i≥p q≥0 C i,q . Note that D : F i (C) → F i (C),
hence we obtain a filtration of the D-cohomology H k = H k (C, D):

F 0 (H k ) ⊃ F 1 (H k ) ⊃ F 2 (H k ) ⊃ · · · ,
where F p (H p+q ) is the subspace represented by D-closed zig-zags starting in the (p, q)-position
and extending downwards.
Theorem 4.2. There is a bigraded spectral sequence Erp,q with E2 -term E2p,q = Hδ (Hd (C p,q )),
such that
p,q
E∞ = F p H q+p /F p+1 H q+p .
One writes (abusing notation)
E2p,q ⇒ H p+q (C).
Remark 4.3. One could also switch the role of d and δ and gets another spectral sequence
computing the same cohomology.
The result may seem a little disappointing at first sight: Of course, in reality we would
rather get H k = H p+q itself, rather than the associated graded group of some filtration of
H p+q .
Remarks 4.4. (cf. Davis-Kirk [10, p.240]) Suppose V is an R-module with a decreasing
filtration F 0 (V ) ⊃ F 1 (V ) ⊃ · · · , and let Gr• (V ) be the associated graded module,
Grj (V ) = F j (V )/F j+1 (V ).
Suppose the filtration is finite, i.e. F j (V ) = 0 for j sufficiently large.
(a) If Gr• (V ) = 0 then V = 0. More generally, suppose we are given a morphism A :
V ′ → V ′′ of filtered R-modules, and consider the induced map Gr(A) : Gr• (V ′ ) →
Gr• (V ′′ ). If Gr(A) is an isomorphism then so is A. (This follows from the first part,
by considering induced filtration on the kernel and P cokernel of A.)
(b) If R is a field, and dim V < ∞ then dim V = i dim Gri (V ). Thus Gr• (V ) carries
almost the same information as V in this case.
(c) In general, one cannot recover V from Gr• (V ). For example, let R = Z. Consider the
following two filtered Z-modules,
Z ⊃ 2Z ⊃ 0, Z ⊕ Z2 ⊃ Z ⊃ 0.
In both cases the associated graded module is Z2 ⊕ Z, but the groups are non-
isomorphic. Similarly
Z4 ⊃ Z2 ⊃ 0, Z2 ⊕ Z2 ⊃ Z2 ⊃ 0.
Thus, if our coefficient ring is a field, the spectral sequence really is going to compute
the cohomology. In general, we can still say that for any morphism of double complexes
C •,• → (C ′ )•,• , and the induced map in cohomology is an isomorphism if one has Er = (E ′ )r
for r sufficiently large.
Remark 4.5. More generally, there are spectral sequences for the cohomology of chain
complexes (C • , d) with a given filtration F 0 (C • ) ⊃ F 1 (C • ) ⊃ · · · by subcomplexes. In our
case, the filtration came from a bigrading.
4.3. How does one construct the spectral sequence? The spectral sequence of a
double complex C •,• , with filtration F p (C) as defined as above, may be defined as follows. Let
Zrp,q := {a ∈ F p C p+q | Da ∈ F p+r C p+q+1 }
Elements of Zrp,q are represented by zig-zags of length r, starting in the (p, q) position and
solving the first r of our system of equations. Define a submodule
p+1,q−1 p−r+1,q+r−2
Brp,q := Zr−1 + D(Zr−1 ).
p+1,q−1
Here Zr−1 ⊂ Zrp,q may be viewed as those zig-zags for which the first term αp,q is zero.
p−r+1,q+r−2
Note that D(Zr−1 ) ⊂ C p,q since D of any element in Zr is just ±δ of the tail of the
p−r+1,q+r−2
zig-zag, and the tail of an element in Zr−1 sits in the (p − 1, q) position. Let
Erp,q = Zrp,q /Brp,q .
If a ∈ Zrp,q , its differential Da is obviously contained in Zrp+r,q−r+1 . Since
p+1,q−1
DBrp,q = DZr−1 ⊂ Brp+r,q−r+1
the class of Da in Erp+r,q−r+1 depends only on the class of a in Erp,q . This defines
dr : Erp,q → Erp+r,q−r+1 .
40 CONTENTS
Since dr is induced from D, it is immediate that dr squares to 0, so dr is a differential. Its

kernel dr is represented by zig-zag’s a ∈ Zrp,q of length r that can be extended to a zig-zag Zr+1p,q
of lengthe r + 1. It is straightforward (but slightly tedious) to check that indeed, Er+1 is the
p,q
cohomology for Er . By construction, E∞ is represented by elements in the p, q-position that
may be extended down to a D-cocycle.
As a first example, let us re-examine the isomorphism between Čech cohomology and
singular cohomology. Given a good cover U, view the Čech complex Č p (X, U; R) as the zeroth
row of a double complex C p,q . Let C̃ p,q = C q (U p X; R). The inclusion map C p,q → C̃ p,q induces
maps between the spectral sequences. Already at the E1 stage the two spectral sequences
coincide. It follows that the map in cohomology H(C, D) → H(C̃, D) is an isomorphism. But
H(C, D) = H(Č • (X, U; R), δ) while H(C̃, D) = H(||UX||; R) = H(X; R).
4.4. The simplicial de Rham theorem. As an example of spectral sequence techniques,

we will now proof the simplicial analogue of de Rham’s isomorphism between singular coho-
mology (with coefficients in R) and de Rham cohomology. Let us recall briefly that the proof of
this isomorphism uses an intermediate complex of smooth singular chains Cksm (X) ⊂ Ck (X).
One proves that the inclusion map is a chain homotopy equivalence, hence dually the map
C k (X, R) → Hom(Cksm (X), R) is a cochain homotopy equivalence. Integration over smooth
chains defines a map Ωk (X) → Hom(Cksm (X), R), and the essence of de Rham’s theorem is
that this is a cochain homotopy equivalence as well.
Suppose now X• is a simplicial manifold, and define a double complex of differential forms,
(Ω• (X• ), d, δ),
where again δ is an alternating sum of pull-backs. Let Ωk (X) be the associated total complex.
Theorem 4.6 (Simplicial de Rham theorem). The cohomology of the complex Ωk (X) is
canonically isomorphic to the singular cohomology of kXk with coefficients in R.
Proof. Define a map
Ωp (Xq ) → Hom(Cpsm (Xq ), R)
by integration over chains. This is a homomorphism of double complexes, and by the usual de
Rham theorem the induced map in d-cohomology is an isomorphism. Hence, the associated
spectral sequences coincide already at the E1 -term. It follows that the above map induces an
isomorphism of cohomology groups for the total complex. Similarly, the maps
C p (Xq , R) = Hom(Cp (Xq ), R) → Hom(Cpsm (Xq ), R)
give an isomorphism for the total cohomology.
4.5. The cohomology H i (BG) for i ≤ 4. We will now give a systematic computation
of H i (BG) (0 ≤ i ≤ 4) for a compact connected simple Lie group G. We will quickly need
some basic facts about the cohomology of Lie groups, so let us briefly review those facts. We
introduce the following notation. For ξ ∈ g let ξ L denote the left-invariant vector field on
G equal to ξ at e, and ξ R the right-invariant vector field. Let Ω(G)L , Ω(G)R be the spaces
of left-/right- invariant differential forms and Ω(G)L×R the space of bi-invariant differential
forms. The space Ω(G)L is generated by the components of the left Maurer-Cartan form
θL ∈ Ω1 (G)L ⊗ g which is defined by ι(ξ L )θL = ξ. Similarly let θR be the right-Maurer-Cartan

form. These satisfy structure equations,
dθL + 12 [θL , θL ] = 0, dθR − 21 [θR , θR ] = 0.
Note also that [θL , [θL , θL ]] = 0 by the Jacobi identity (plug in left-invariant vector fields to
see this.) Both inclusions
Ω(G)L×R → Ω(G)L → Ω(G)
induce isomorphisms in cohomology: Indeed, if [α] ∈ H(G), then [lg∗ α] = lg∗ [α] = [α] by
homotopy invariance, since G is connected. Thus, the average of α under the left action is
cohomologous to α. Similarly, if we also average under the right-action we get a bi-invariant
form cohomologous to α.
Lemma 4.7 (Koszul). The de Rham differential on Ω(G)L is given by the formula
X
d|Ω(G)L = 21 hea , θL i LeLa .
a
Proof. Since both sides are graded derivations, it suffices to check on θL . We have
X X
1
2 hea , θL i LeLa θL = 12 hea , θL iιeLa dθL
a a
1X a L
= − he , θ iιeLa [θL , θL ]
4 a
X
= − 12 hea , θL i[ea , θL ]
a
1 L L
= − 2 [θ , θ ] = dθL .

Since the left-invariant vector fields generate right-translations, this formula shows that d
is in fact 0 on Ω(G)L×R ! Using Ω(G)L×R = (∧g∗ )G this proves,
Proposition 4.8. The de Rham cohomology of a compact connected Lie group is isomor-
phic to (∧g∗ )G . Every cohomology class has a unique bi-invariant representative.
The structure of the algebra (∧g∗ )G is completely known, by the Hopf-Koszul-Samelson
theorem (see e.g. Greub-Halperin-Vanstone [16]). Some basic facts are easy to figure out
by hand, however. For instance, if p ∈ (S m g∗ )G is an invariant polynomial on g, one has a
corresponding bi-invariant form on G:
αp = θL · p′ ([θL , θL ])
here p′ ∈ S m−1 (g∗ ) ⊗ g∗ is the gradient of p, defined by ξ · p′ (ζ) = d |t=0 p(ζ + tξ). (We leave
dt
it as an exercise that this form is indeed bi-invariant). In particular, if B is an invariant inner
product on g, the polynomial p(ξ) = B(ξ, ξ) defines a form of degree 3, which we prefer to
normalize as follows:
1
η = B(θL , [θL , θL ]).
12
It is easy to check that if G is simple, H i (G) = 0 for i = 1, 2 while H 3 (G) = R with generator
this form η.
42 CONTENTS
Now consider the double complex for BG, i.e. C p,q = Ωq (Gp ). The E1 -term is H q (Gp ).
Note this is 0 for p + q ≤ 4 and q > 0, except for the term H 3 (G) = R which we just computed.
(Draw a picture of the E1 -term!) To compute the E2 -term, we have calculate the image of [η]
under δ : H 3 (G) → H 3 (G2 ) = R ⊕ R. By the explicit formulas for the face maps,
δ : Ωq (G) → Ωq (G2 ), δα = pr∗1 α + pr∗2 α − µ∗ α
where µ : G × G → G is group multiplication and pri are the two projections. For η, one may
verify that

δη = d 21 B(pr∗1 θL , pr∗2 θR ) .
Thus d1 [η] = [δη] = 0, and E21,3 = E1p,q = R. Also, the d1 = δ-cohomology of the row p = 0 is
trivial. Thus for p+q ≤ 4, E21,3 = E20,0 = R, and E2p,q = 0 otherwise. Note that the differentials
d2 from the remaining entries E2p,q with p + q ≤ 4 are all zero, and similarly for d3 , d4 , . . .. We
conclude that E2p,q = E∞p,q
for p + q ≤ 4. We have thus shown (real coefficients)
H 1 (BG) = H 2 (BG) = H 3 (BG) = 0, H 4 (BG) = R.
We can be more precise: Let β = 21 B(pr∗1 θL , prL
∗ θ R ) ∈ Ω2 (G2 ). Then δβ = 0 (as one verifies
2
by direct calculation) and therefore η + β ∈ q p
p+q=4 Ω (G ) is a D-cocycle representing a
generator of H 4 (BG).
Remark 4.9. If G is simple and simply connected, it is known that H i (G, Z) has no torsion
in degree i ≤ 4. Therefore, the above argument also works with Z coefficients, using singular
cochains, and one finds that H 4 (BG, Z) = Z while H i (BG, Z) = 0 for 1 ≤ i ≤ 3.
4.6. Product structures. Recall that the front p-face of an n-simplex p ≥ n, is the
simplex spanned by the first p + 1 vertices e0 , . . . , ep , while the back p-face is spanned by the
last p + 1 vertices en−p , . . . , en . We recall that these enter the definition of the cup product on
′
singular cochains: Given cochains α ∈ C q (X), β ∈ C q (X), one defines the value of α ∪ β ∈
′ ′
C q+q (X) on a singular q+q ′ -simplex σ : ∆q+q → X to be the value of α on the front q-simplex,
′ ′ ′
∆q → ∆q+q → X, times the value of β on the back q ′ -simplex ∆q → ∆q+q → X.
The front-face and back-face correspond to the following two morphisms in the category
Ord:
φnp : [p] → [n], i 7→ i, ψpn : [p] → [n], i 7→ i + n − p.
If X• is a simplicial space, we obtain corresponding maps
X(φnp ), X(ψpn ) : Xn → Xp .
We can use these to define a product structure on the double complex of singular cochains
C • (X• ), by composition,
′ ′ ′
C q (Xp ) ⊗ C q (Xp′ ) → C q (Xp+p′ ) ⊗ C q (Xp+p′ ) → C q+q (Xp+p′ ).
′ ′
Here the first map is (−1)q p times pull-back under the map X(φp+p ) × X(φp+p
′
p p′ ), while the
q ′p
second map is the usual cup product. The sign (−1) is necessary in order that D becomes
a derivation for the product structure. More precisely we have:
5. THE CHERN-WEIL CONSTRUCTION 43
Proposition 4.10. Both differentials d, δ on the double complex are (graded) derivations
′
with respect to the product structure. That is, if α ∈ C q (Xp ) and β ∈ C q (Xp′ ), we have
d(αβ) = (dα)β + (−1)q+p α(dβ)
′
δ(αβ) = (−1)q (δα)β + (−1)p α(δβ)
D(αβ) = (Dα)β + (−1)q+p α(Dβ).
Since D is a derivation, the product structure descends to the cohomology H • (C • (X• ), D)).
By a similar formula, we have a ring structure on the simplicial de Rham complex.
Theorem 4.11. The isomorphism H • (C • (X• ), D) → H • (||X||) is an isomorphism of
graded rings.
See e.g. Dupont, [14].
In general, if a double complex (C p,q , d, δ) has a product structure relative to which d, δ
satisfy
d(αβ) = (dα)β + (−1)q+p α(dβ)
′
δ(αβ) = (−1)q (δα)β + (−1)p α(δβ)
′ ′
for α ∈ C p,q , β ∈ C p ,q , the differential D becomes a derivation for the total complex. Further-
more, each dr : Er → Er in the spectral sequence becomes a graded derivation, for the product
structure induced from Er−1 . Passing to the limit, we obtain a product structure on E∞ . On
the other hand, the product structure on the cohomology of the total complex is compatible
with the filtration, hence it descends to a product structure on the associated graded group,
which we saw is E∞ . The two product structures on E∞ coincide.
The upshot is: The total cohomology of the double complex Ωp (Xq ), with the ring structure
just introduced, is isomorphic to the cohomology algebra of ||X|| as a ring.
5. The Chern-Weil construction

5.1. Connections and curvature on principal bundles. Let G be a Lie group and
π : P → B be a (smooth) principal G-bundle. A differential form α on P is called horizontal
if ιξP α = 0 for all ξ ∈ g, and basic if it is both invariant and horizontal. It is well-known (and
easy to see, using local trivializations) that the pull-back map π ∗ : Ω(B) → Ω(P ) with image
the basic forms. That is, any basic form descends to a unique form on B. More generally, if V
is a G-represenations, a V -valued form α ∈ Ω(P ) ⊗ V is called basic if it is in (Ω(P )hor ⊗ V )G .
Basic V -valued forms descend to forms with values in the associated vector bundle P ×G V .
Let V P ⊂ T P be the vertical subbundle, i.e. Vp P = ker(dp π) for p ∈ P . There is an exact
sequence of vector bundles,
0 → V P → T P → π ∗ T B → 0.
A connection on P is a G-equivariant splitting of this sequence, i.e. a G-equivariant surjective
bundle homomorphism T P → V P , restricting to the identity on V P . This can be reformulated
as follows: The bundle V P is trivialized by the generating vector fields for the G-action on P :
P × g → V P, (p, ξ) 7→ ξP (p).
44 CONTENTS
Thus, a connection on P is a G-equivariant bundle map T P → P × g taking ξP to ξ. It is thus

given by a connection 1-form θ ∈ Ω1 (P ) ⊗ g with defining properties,
g ∗ θ = Adg θ, ιξP θ = ξ.
Any two connections differ by a form in Ω1 (P ) ⊗ g that is invariant (i.e. lies in (Ω1 (P ) ⊗ g)G
and horizontal (i.e. annihilated by all ιξP .) That is, the space of connections is an affine
space, with underlying vector space the horizontal and invariant g-valued 1-forms. Given a
connection, one may define the corresponding horizontal bundle HP = ker θ, and the splitting
identifies HP = π ∗ T B.
The curvature of a connection θ is defined as
F θ = dθ + 21 [θ, θ].
It is an invariant and horizontal g-valued 2-form. To remove any possible ambiguities, let us
a of g by
write out the curvature in terms of a basis ea of g: Introduce structure constants fbc
P a P a θ
P θ a θ a
[eb , ec ] = a fbc ea , and write θ = a θ ea . Then F = a (F ) ea where (F ) are 2-forms,
X
(F θ )a = dθa + 12 a b
fbc θ ∧ θc .
bc
One of its basic properties is the Bianchi identity

dθ F θ := (d + [θ, ·])F θ = 0.
(The proof relies on the fact that [θ, [θ, θ]] = 0, which in turn follows from the Jacobi identity.)
There are many geometric interpretations of the curvature – for example, it measures the failure
of the horizontal lift Lift : X(B) → X(P ) to be a Lie algebra homomorphism.
We are interested in the role of F θ in the Chern-Weil construction of characteristic classes
on B. If p ∈ S m g∗ is a polynomial on g, we may form c̃θ (p) := p(F θ ) ∈ Ω2m (P ). More
accurately, we may view the curvature as a map g∗ → Ω2 (P ), as such it extends to an algebra
homomorphism c̃θ : Sg∗ → Ωeven (P ). Note that the image of cθ lies in the space of forms on P
that are horizontal and invariant, i.e. basic. The space Ω(P )basic of basic forms is isomorphic
to Ω(B), by pull-back. Hence, c̃θ descends to an algebra homomorphism
cθ : Sg∗ → Ωeven (B).
Theorem 5.1 (Chern-Weil construction). If p ∈ (Sg∗ )G is an invariant polynomial, the
form cθ (p) is closed. Its cohomology class does not depend on the choice of θ.
We will postpone the (not very difficult) proof, since we will prove a more general result
further down. The cohomology classes [cθ (p)] are called the characteristic classes of the prin-
cipal bundle P . Indeed, we will see that they are exactly the characteristic classes (for real
coefficients) obtained from the classifying map for P .
5.2. g-differential algebras. Cartan’s idea [8, 9] was to introduce an algebraic model
for the space of differential forms on the classifying bundle EG → BG, and to re-phrase the
Chern-Weil construction in those terms. (We will show how his model is related to the simplicial
model discussed earlier.)
L
Definition 5.2. A differential graded algebra is an graded algebra A = ∞ i
i=0 A with a
differential d of degree +1, such that d is a derivation. It is called a g-differential algebra if, in
addition, there are derivations Lξ of degree 0 and ιξ of degree −1, for all ξ ∈ g, satisfying the
relations of contractions, Lie derivative and differential on a manifold with a g-action:
[ιξ , ιξ′ ] = 0, [Lξ , ιξ′ ] = ι[ξ,ξ′ ]g
[d, Lξ ] = 0, [Lξ , Lξ′ ] = L[ξ,ξ′ ]g ,

[d, d] = 0, [d, ιξ ] = Lξ
(using graded commutators).
There is an obvious notion of homomorphism of g-differential algebras. Sometimes one also
considers g-differential spaces (i.e. one doesn’t require algebra structures.) A first example of
a g-differential algebra is the algebra of differential forms on a manifold with a g-action.
Definition 5.3. A connection on a g-differential algebra is an element θ ∈ A1 ⊗g satisfying
ιξ θ = ξ and Lξ θ = −[ξ, θ]g. The curvature of the connection θ is the element F θ ∈ A2 ⊗ g
defined as
F θ = dθ + 21 [θ, θ].
Note that the condition Lξ θ = −[ξ, θ]g is the infinitesimal version of the G-invariance
condition for a principal connection. It is equivalent to the global condition if G is connected.
An example is therefore the connection on the space of differential forms on a principal G-
bundle, if G is connected. More generally, if P is a manifold with a Lie algebra action of g, the
existence of a connection on Ω(P ) implies that the action is locally free (and e.g. for g compact
the converse holds true). As in the case of principal bundles, if a connection exists, the space
of connections is an affine space with underlying vector space the space (A1hor ⊗ g)inv .
We will often take the equivalent point of view that a connection is an equivariant map
θ : g∗ → A1 with ιξ θ(µ) = hµ, ξi, and the curvature is an equivariant map F θ : g∗ → A2 .
Definition 5.4. Let A be a g-dga. One defines subalgebras of horizontal, invariant and
basic elements by
\ \
Ahor = ker(ιξ ), Ainv = ker(Lξ ), Abasic = Ahor ∩ Ainv .
ξ ξ
As in the case of principal bundles, F θ takes values in horizontal elements. If we extend

Fθ to an algebra homomorphism Sg∗ → Aeven , p 7→ p(F θ ) then p(F θ ) is basic provided p ∈
(Sg )inv .
∗
For any principal bundle π : P → B, with G connected, a form on P is basic if and only if
it is the pull-back of a form on the base: π ∗ : Ω(B) → Ω(P )basic is an isomorphism.
Lemma 5.5. The basic subalgebra Abasic is invariant under d. Its cohomology is called the
basic cohomology of A.
Proof. Follows from [ιξ , d] = Lξ and [Lξ , d] = 0.
Note that by contrast, the subalgebra of horizontal elements is not d-invariant.

46 CONTENTS
5.3. The Weil algebra. The Weil algebra W g is most quickly defined in terms of gener-
ators and relations:
Definition 5.6. The Weil algebra W g is the commutative graded algebra, freely generated
by elements µ ∈ g∗ of degree 1 and elements µ ∈ g∗ of degree 2 (linear over the two copies of
g∗ ), with contractions, Lie derivatives and differential given on degree 1 generators by
ιξ µ = hµ, ξi, Lξ µ = − ad∗ξ µ, dµ = µ.
The canonical connection on W g is given by θW : g∗ → W g, µ 7→ µ.
This is well-defined: The relations among contractions, Lie derivatives and differential force
us to put
ιξ µ = ιξ dµ = Lξ µ = − ad∗ξ µ
as well as Lξ µ = −ad∗ξ µ and dµ = 0. It is easily checked that these definitions are consistent,
essentially because W g is free over the given set of generators.
Remark 5.7. Note that the construction of an algebra and a differential makes sense for
g∗ replaced with any vector space V , and defines a commutative graded differential algebra.
This is usually called the Koszul algebra.
As an algebra, W g is simply the tensor product of Sg∗ (corresponding to the degree 2
generators) and ∧g∗ (corresponding to the degree 1 generators). We may also view W g as a
symmetric algebra (in the graded sense) over the graded vector space g∗ ⊕ g∗ , where the first
copy corresponds to degree 1 generators and the second copy to degree 2 generators.
Proposition 5.8 (Acyclicity). There exists a canonical homotopy operator h : W g → W g
with [h, d] = id −Π, where Π : W g → W g is projection onto W 0 g = R.
Proof. Let σ be the degree -1 derivation given on generators by σµ = 0 and σµ = µ.
The commutator [σ, d] is a derivation of degree 0, equal to the identity on generators. It hence
extends to the Euler operator on W g, given on a product of generators simply by multiplication
by the number of generators in that product. In particular, it is invertible on W + g = ker Π.
It is easy to check that [σ, d] commutes with σ and d. The homotopy operator is defined by
h = 0 on W 0 g and h ◦ [σ, d] = σ on W + g.
Proposition 5.9 (Universal property). If A is any g-dga with connection θ, there is a
unique homomorphism of g-dga’s cθ : W g → A such that the following diagram commutes,
WO g /A.
cθ }}>
}}
}}}
}
g∗
Proof. The homomorphism takes µ to θ(µ) and µ to dθ(µ). It is straightforward to check
that this has the correct properties.
These two properties show clearly that we should think of W g as the algebraic analogue of
Ω(EG), with cθ the analogue to pull-back under a classifying map Ω(EG) → Ω(P ), and h the
algebraic analogue of the homotopy operator for a contraction of EG to a base point.
Proposition 5.10. Let A be a locally free g-dga, and θ0 and θ1 two connections on g. Let
c0 , c1 : W g → A be the two characteristic homomorphisms defined by θ0 , θ1 . There exists an
operator h : Wg → A of degree −1 with the following properties,
[h, d] = c0 − c1 , [h, ιξ ] = 0, [h, Lξ ] = 0.
Proof. Consider the connection θ on A ⊗ Ω([0, 1]) given by θ = (1 − t)θ0 + tθ1 , where t is
the coordinate on [0, 1]. It defines a characteristic homomorphism
c : W g → A ⊗ Ω([0, 1])
that pulls back to c0 , c1 at t = 0, t = 1. Let h be ± the composition of c with fiber integration
over [0, 1]. For the appropriate choice of sign, Stokes’ theorem gives [h, d] = c0 − c1 . The
identities [h, ιξ ] = 0, [h, Lξ ] = 0 hold because the g-action has no component in the M -
direction.
In general, given two homomorphisms of g-differential spaces, c0 , c1 : A → A′ , we define

a g-chain homotopy to be an operator h : A → A′ of degree -1 having the properties in this
Proposition.
5.4. The algebraic Chern-Weil construction. To determine the basic subcomplex, it

is convenient to replace the degree 2 generators µ by the curvatures µ̂ = F θ (µ). If ea denotes
a basis of g∗ dual to the basis ea of g introduced above, we have
X
a b
µ̂ = µa ea + 21 fbc e ∧ ec ,
bc
so these are again generators, and we get another isomorphism W g = Sg∗ ⊗ ∧g∗ where now
W
Sg∗ is generated by the curvatures µ̂ = F θ (µ), rather than µ. It is immediate from this
description that (W g)hor is the subalgebra Sg∗ generated by the curvatures, and therefore
(W g)basic = (Sg∗ )inv .
Proposition 5.11. In terms of the isomorphism W g = Sg∗ ⊗ ∧g∗ , where Sg∗ is generated
∧g∗
by the curvatures µ̂, we have ιW
ξ = 1 ⊗ ιξ and the Weil differential is
∗
1 ∧g∗ ba ∧g∗
dW = ea (LSg
ea + 2 Lea ) + e ιea
L Proof. We use that if A is a commutative graded algebra, then the space Der(A) :=
i
i Der (A) of graded derivations of A is a left-A-module. In particular the right hand side of
∗ ∧g∗ ∧g∗
the formula for dW defines a derivation, since (LSg ea , Lea , ιea are all derivations of W g. Two
derivations agree if and only if they agree on generators. On generators ea , the right hand side
gives
(eb (LSg
∗
1 ∧g∗ bb ∧g∗ a = 1 b ∧g∗ a bb ∧g∗ a
eb + 2 Leb ) + e ιeb )e 2 e Leb e + e (ιeb )e
a b c
= − 21 fbc e e + eba
= ea .
Similarly, one verifies (eb (Leb + 21 Leb ) + ebb ιeb )ea by direct computation.
Sg∗ ∧g∗ ∧g∗

48 CONTENTS
From this formula for the Weil differential, we see that the restricted differential on (W g)basic
is in fact 0. Thus Hbasic (W g) = H((W g)basic ) = (Sg∗ )inv .
To summarize, we see that if A is a g-dga with connection, the characteristic homomorphism
θ
c : W g → A restricts to a chain map
cθ : (Sg∗ )g → Abasic
(where (Sg∗ )g carries the 0 differential), and that the two chain maps defined by two connections
are related by a chain homotopy cθ : (Sg∗ )g → Abasic of degree -1. Hence one obtains an algebra
homomorphism
(Sg∗ )g → Hbasic (A)
independent of the choice of connection. This is Cartan’s algebraic analogue to the Chern Weil
construction.
Proposition 5.12. Let A be a locally free commutative g-dga. Then the mapping
A → A ⊗ W g, α 7→ α ⊗ 1
is a g-chain homotopy equivalence. Given a connection θ, a homotopy inverse is given by the
map
A ⊗ W g → A, (α ⊗ w) 7→ α cθ (w).
Proof. We have to show that the composition
W g ⊗ A → A → W g ⊗ A, w ⊗ α 7→ (cθ w)α 7→ 1 ⊗ (cθ w)α
is g-homotopic to the identity. For this, it suffices to show that the two maps
W g → W g ⊗ A, w 7→ 1 ⊗ cθ w
W g → W g ⊗ A, w 7→ w ⊗ 1
are g-chain homotopic. But this follows, since the first map is the characteristic homomorphism
for the connection 1 ⊗ θ on W̃ g ⊗ A, while the second map is the characteristic homomorphism
for the connection θW ⊗ 1.
5.5. The Weil model of equivariant cohomology. Recall that if M is a G-manifold,

we defined the equivariant cohomology ring of M to be the cohomology ring of the borel
construction MG = EG ×G M . If EG were a finite-dimensional principal bundle, and thus MG
were a manifold, this would be the cohomology of the de Rham complex
Ω(MG ) ∼
= (Ω(EG × M ))basic = (Ω(EG) ⊗ Ω(M ))basic .
Thinking of W g as an algebraic model for Ω(EG), and of a g-dga as the algebraic counterpart
of Ω(M ), this motivated the following definition:
Definition 5.13. The equivariant cohomology of a g-dga A is the basic cohomology of
W g ⊗ A:
Hg(A) = Hbasic (W g ⊗ A)
This is known as the Weil model of equivariant cohomology, especially in the case A = Ω(M )
for a G-manifold M . In this case we will write Hg(M ) := Hg(Ω(M )). We will show later
that (for G a compact connected Lie group) it is equivalent to the Borel model, i.e. that
Hg(M ) ∼= HG (M ) := H(MG ). Taking A = R = Ω(pt) to be the trivial g-dga,
Hg(pt) = Hbasic (W g) = (Sg∗ )inv .
Any homomorphism of g-dga’s A1 → A2 induces an algebra homomorphism Hg(A1 ) → Hg(A2 );
in particular (taking A1 = R and A2 = A), Hg(A) is a module over (Sg∗ )inv . For any g-dga,
the inclusion map A → W g ⊗ A, α 7→ 1 ⊗ α is a homomorphism of g-dga’s, hence it induces
an algebra homomorphism Hbasic (A) → Hg(A). Above we proved that if A is locally free and
commutative then the map A → W g ⊗ A is a g-homotopy equivalence. That is,
Proposition 5.14. If A is a locally free g-dga, the map A → W g ⊗ A, α 7→ 1 ⊗ α induces
an algebra isomorphism Hbasic (A) = Hg(A).
Thus, if P is a principal G-bundle, the pull-back map Ω(B) → Ω(P )basic ⊂ Ω(P ) induces an
isomorphism Hg(Ω(P )) = H(Ω(B)) ≡ H(B). This provides further evidence for the conjecture
Hg(M ) = HG (M ), since indeed
HG (P ) = H(EG ×G P ) = H(P ×G EG) = H(B)
(using that P ×G EG → B is a fiber bundle with contractible fibers EG.)
5.6. The Cartan model of equivariant cohomology. The Weil model of equivariant
cohomology has the advantage of a good conceptual explanation, W g playing the role of differ-
ential forms on EG. For computational purposes, it is usually more convenient to work with
an equivalent model known as the Cartan model. Let A be a g-dga (we usually have in mind
the algebra of differential forms on a G-manifold A = Ω(M )). The first step in calculating
Hg(A) is to determine the basic subcomplex (W g ⊗ A)basic . Identify W g = Sg∗ ⊗ ∧g∗ , where
Sg∗ is the symmetric algebra generated by the curvature variables. To simplify notation, we
will denote the degree 2 generators by v a := eba . We have
W g = (Sg∗ ⊗ ∧+ g∗ ) ⊕ Sg∗ ,
L
where ∧+ g∗ = i ∗ ∗
i>0 ∧ g , hence a projection W g → Sg (“setting the connection variables
equal to 0”). Extend to a projection
W g ⊗ A → Sg∗ ⊗ A.
Theorem 5.15 (Cartan). The projection W g ⊗ A → Sg∗ ⊗ A restricts to an algebra iso-
morphism
(W g ⊗ A)basic → (Sg∗ ⊗ A)inv .
This isomorphism takes the differential dW + dA to the equivariant differential
dg = d − v a ⊗ ιA
a.
(where d = dA ).
Proof. We prove this result using an elegant trick due to Kalkman [20]. Consider the
following derivation on W g ⊗ A,
ψ = er ⊗ ιA
r .
50 CONTENTS
Note that ψ has degree 0 and is g-equivariant. Hence exp(ψ) is a g-equivariant algebra auto-
morphism. Let us compute
exp(ψ) ◦ (ιW A W A W A
a + ιa ) ◦ exp(−ψ) = Ad(exp ψ)(ιa + ιa ) = exp(adψ )(ιa + ιa ).
P j
Write exp(adψ ) = ∞ 1
j=0 j! adψ . We find,
adψ ιa = adψ (ιW A

a + ιa )
= −ιA
a
ad2ψ ιa = 0.
Thus exp(ψ) takes ιW A W
a + ιa to ιa . Thus,
exp(ψ) : (W g ⊗ A)basic → ((W g)hor ⊗ A)inv = (Sg∗ ⊗ A)inv .
On the other hand, observe that the automorphism exp(ψ) does not change the projection of
an element onto Sg∗ ⊗ A, hence on (W g ⊗ A)hor it coincides with that projection. It remains
to work out the induced differential. Consider the formula for the Weil differential,
∗ ∗ ∗
dW = ea (LSg 1 ∧g a ∧g
ea + 2 Lea ) + v ιea .
To compute the induced differential, we must consider
∗ ∗ ∗
dW + dA = ea (LSg 1 ∧g a ∧g
ea + 2 Lea ) + v ιea + d
A
∗ ∗ ∗
= ea (LSg 1 ∧g a ∧g A a A A
ea + 2 Lea ) + v (ιea + ιea ) − v ιea + d .
on an element of (W g ⊗ A)basic , followed by projection to Sg∗ ⊗ A. The term involving
∗ Sg∗ ∗
(ι∧g A a
ea + ιea ) disappears on horizontal elements, while the terms involving e (Lea and e Lea
a ∧g
disappear after projection onto Sg∗ ⊗ A. Hence the differential on (Sg∗ ⊗ A)inv is induced from
A
the term −v a ιAea + d , which commutes with the projection.
If we identify Sg∗ with polynomials on g, the algebra (Sg∗ ⊗ A)inv becomes the algebra of
g-equivariant polynomial maps α : g → A. In terms of these identifications, the differential
reads
(dgα)(ξ) = d(α(ξ)) − ιξ α(ξ).
In the case A = Ω(M ), one often calls
ΩG (M ) := (Sg∗ ⊗ Ω(M ))G
the complex G-equivariant differential forms, and dG = d − ι(ξM ) the equivariant differential.
Recall that the generators of the symmetric algebra have degree 2, hence the grading on ΩG (M )
is given by M
ΩkG (M ) = (S i g∗ ⊗ Ωj (M ))G .
2i+j=k
Let us verify that after all these computations, the differential still squares to 0:
d2G α(ξ) = d(dG α(ξ)) − ιξ dG α)(ξ)
= ddα(ξ) − dιξ α(ξ) − ιξ dα(ξ) − ιξ ιξ α(ξ)
= −Lξ α(ξ)
= −α([ξ, ξ]) = 0.
It is also interesting to re-examine the proof that Hg(A) = Hbasic (A) for any commutative
locally free g-dga. Suppose θ = θa ea ∈ A1 ⊗ g is a connection on A. In the following Lemma
and its proof we do not use the summation convention.
θ :=
Q a
Q a
Lemma 5.16. The operator Phor a ιea θ = a (1 − θ ιea ) is a projection operator onto
the space Ahor of horizontal elements.
Proof. Note that the operators ιea θa (a = 1, . . . , dim g) are a family of pairwise commuting
projection operators. Hence their product is again a projection operator. For any α ∈ A we
have Y
θ
ιr Phor α = ιr ιea θa = 0,
a
since ιr commutes with terms a 6= r and its product with the terms a = r is zero. If α is
horizontal, we have
Y
θ
Phor α= (1 − θa ιea )α = α,
a
showing that Phor is projection onto Ahor .
Note that the horizontal projection operator may also be written,

Phor = exp(−y a ιea )|ya =θa
where y a are degree 1 variable corresponding to ea , and the notation indicates that we first
apply the operator, and then set y a = θa in the resulting expression. Let F θ : g∗ → A2 be the
curvature of θ. Recall that for p ∈ Sg∗ , we defined p(F θ ) as the image of p under the algebra
homomorphism SgP ∗ → Aeven defined by F θ . (“Plugging in the curvature for the variable ξ”.)
P
Similarly, for α = I pI ⊗ αI ∈ Sg∗ ⊗ A we define α(F θ ) = I pI (F θ )αI .
Theorem 5.17 (Cartan). Let A be a locally free g-dga, and θ a connection. Define a map
Carθ : (Sg∗ ⊗ A)inv → A, α 7→ Phor (α(F θ )).
Then Carθ is an algebra homomorphism taking values in Abasic . Furthermore, Carθ is a chain
map, and induces an isomorphism in cohomology. In fact, the projection α 7→ 1 ⊗ Carθ (α) is
a projection, chain homotopic to the identity, by an explicit homotopy operator.
Note that the Cartan map extends the Chern-Weil homomorphism (Sg∗ )inv → Abasic : For
an invariant polynomial p ∈ (Sg∗ )inv , p(F θ ) is already basic and so the horizontal projection
operator can be omitted.
Proof. We will prove this result by comparing with the Weil model. Identify (Sg∗ ⊗ A)inv
with (W g⊗A)basic . We had shown above that the map W g⊗A → A taking w ⊗α to 1⊗cθ (w)α
is a g-chain homotopy inverse to the map A → W g ⊗ A, α 7→ 1 ⊗ α. Given α ∈ (Sg∗ ⊗ A)inv ,
the corresponding element of (W g ⊗ A)basic is obtained by applying the exp(−ea ⊗ ιA
a )α. The
θ
element c (w)α is exactly
exp(−y a ⊗ ιA θ θ
a )|y a =θa α(F ) = Car (α).

52 CONTENTS
This Theorem (at least the first part) is contained in Cartan’s paper “La transgression dans
un groupe de Lie”, Théorème 4 (p.64). Since Cartan’s proof was a little cryptic, the result has
been re-proved several times. See e.g. Kumar-Vergne [21, p. 171–176], Duistermaat [11, p.
227–234], Guillemin-Sternberg, [17, p. 53–59], Nicolaescu, [25, p. 17–38]. (The argument in
Nicolaescu’s paper seems more or less identical to the one presented here.)
5.7. Examples of equivariant differential forms. Cartan’s model of equivariant co-

homology is very popular in differential geometry, particularly symplectic geometry.
Suppose M is a G-manifold. We will study equivariant forms in low degrees. Since
Ω1G (M ) = Ω1 (M )inv , an equivariant 1-form on M is simply an invariant 1-form. Since
Ω2G (M ) = Ω2 (M )inv ⊕ (g∗ ⊗ Ω0 (M ))inv
an equivariant 2-form is a sum ω + Ψ, where ω is an invariant 2-form and Ψ : M → g∗
an equivariant function, viewed as an element of (g∗ ⊗ Ω0 (M ))inv . The equivariant 2-form is
equivariantly closed if and only if
0 = dG (ω + Ψ)(ξ) = dω − ιξM ω + dhΨ, ξi,
which gives two equations dω = 0 and ιξM ω = dhΨ, ξi. Thus ω should be closed and invariant,
while the second condition is the familiar moment map condition from symplectic geometry,
with −Ψ as the moment map! (The minus sign is a matter of convention.) Indeed, if ω is not
only closed but also non-degenerate, the map Ψ determines the generating vector fields by this
equation. The equivariant 2-form is exact if there exists an invariant 1-form ν with ω = dν
and Ψ = −ι(ξM )ν.
An equivariant 3-form is an element of
Ω3G (M ) = Ω3 (M )inv ⊕ (g∗ ⊗ Ω1 (M ))inv
It therefore has the form η + β(ξ), where η is an invariant 3-form and β is an equivariant map
from g to Ω1 (M ). The equivariant 3-form is closed if and only if
dη = 0, ι(ξM )η = dβ(ξ), ι(ξM )β = 0.
For example, if M = G where G acts by conjugation, and B is an invariant inner product on
g (possibly indefinite), one may check that
1
η= B(θL , [θL , θL ]), β(ξ) = 12 B(θL + θR , ξ)
12
defines an equivariant 3-cocycle.
5.8. Equivariant formality. In general, the map HG (M ) → H(M ) induced by the chain
map ΩG (M ) → Ω(M )G , α(ξ) 7→ α(0) need not be surjective: Not every cohomology class
admits an equivariant extension. One defines,
Definition 5.18. A compact G-manifold M is called equivariantly formal if the map
HG (M ) → H(M ) is onto.
There are many equivalent conditions for a G-manifold to be equivariantly formal. For
example, M is equivariantly formal if and only if HG (M ) = (Sg∗ )G ⊗ H(M ) as graded vector
spaces, or equivalently if and only if HG (M ) is a free Sg∗ -module.
6. EQUIVALENCE BETWEEN THE SIMPLICIAL AND WEIL MODEL 53
Examples 5.19. (a) If (M, ω) is a compact symplectic manifold, with a symplectic

G-action admitting a moment map, then M is equivariantly formal. (This result is
due (independently) to Ginzburg and Kirwan, and is proved using Morse theory.
(b) The conjugation action of a compact Lie group on itself is equivariantly formal. (There
are explicit generators of the cohomology, and explicit equivariant extensions. See e.g.
Jeffrey’s paper “group cohomology construction ...”)
(c) The left-action of a compact Lie group G on itself, and more generally free G-actions
are not equivariantly formal.
(d) Suppose G is compact and simply connected, and that M is connected. Then one may
show that the map HG i (M ) → H i (M ) is an isomorphism in degree i ≤ 2, while in
degree i = 3 there is a short exact sequence,

3
0 → HG (M ) → H 3 (M ) → H 3 (G.x) → 0
for any x ∈ M . (The cohomology groups H 3 (G.x) are all isomorphic.) Hence, if the
action is equivariantly formal one must have H 3 (G.x) = 0.
(No time for proofs..)
5.9. The Künneth formula. (No time ...See Kumar-Vergne [21].)
6. Equivalence between the simplicial and Weil model

6.1. A non-commutative version of the Weil algebra. As explained above, the Weil
algebra W g is the universal commutative locally free g-dga. It seems natural to ask if there
exists a similar universal object if one drops commutativity. This is indeed the case: Simply
omit “commutative” from the definition of W g.
Thus, we let W̃ g be the g-dga which is freely generated by degree 1 elements µ ∈ g∗ and
degree 2 elements µ. Thus, while W g was the symmetric algebra over the graded vector space
Eg∗ with Eg1∗ = g∗ , Eg2∗ = g∗ , Egi ∗ = 0 otherwise, W̃ g is the tensor algebra:
W g = S(Eg∗ ), W̃ g = T (Eg∗ ).
The contractions, Lie derivatives and differential are defined on degree 1 generators, by the
exact same formulas as for W g:
ιξ µ = hµ, ξi, Lξ µ = − ad∗ξ µ, dµ = µ,
and again the formulas on degree 2 generators are determined by the relations. It is immediate
that these formulas extend to derivations of W̃ g, and that g∗ → W̃ g, µ 7→ µ is a connection.
Some essential properties of W g carry over to W̃ g:
Theorem 6.1 (Acyclicity of W̃ g). The inclusion R → W̃ g as multiples of the identity in-
duces an isomorphism in cohomology. In fact, there is a canonically defined homotopy operator
h : W̃ • g → W̃ •−1 g with the property [h, d] = id −Π where Π is projection onto the degree 0
part.
Theorem 6.2 (Locally free g-dga’s). Suppose A is a locally free g-dga.
54 CONTENTS
(a) For any connection θ on A, there exists a unique homomorphism of g-dga’s cθ : W̃ g →

A (called the characteristic homomorphism) such that the diagram
W̃O g /A
cθ }}>
}}
}}}
}
g∗
commutes.
(b) If θ0 , θ1 are two connections on A, there is a canonically defined g-homotopy operator
h : W̃ i g → Ai−1 , i.e. a linear map such that [h, ιξ ] = 0, [h, Lξ ] = 0 and
[h, d] = cθ1 − cθ0 .
(c) The inclusion map A → W̃ g ⊗ A, α 7→ 1 ⊗ α is a g-homotopy equivalence. For any
connection θ, a g-homotopy inverse is given by the map
W̃ g ⊗ A → A, w ⊗ α 7→ (cθ w)α.
In particular the connection on W g defines a characteristic map
W̃ g → W g,
which is simply the quotient map from the tensor algebra to the symmetric algebra.
The proofs of the two theorems are similar to the commutative setting.
6.2. Uniqueness property of Weil algebras. There does not seem to be a simple
description of the horizontal or basic subalgebras of W̃ g. We will however prove the following:
L
Theorem 6.3. Suppose W = i≥0 W i is a g-dga with connection, such that there exists
a linear operator h : W → W of degree −1 with [h, d] = id −Π where Π is a projection onto
R ⊂ W ). Assume [h, Lξ ] = 0. Then the characteristic map
W̃ g → W
induces an isomorphism in basic cohomology. More generally, for any g-dga B the map
W̃ g ⊗ B → W ⊗ B
induces an isomorphism in basic cohomology. If W is commutative, similar statements hold
for W g.
Remarks 6.4. (a) The theorem is analogous to the result that if EG → BG is a
classifying bundle, and E ′ → B ′ is another principal G-bundle with contractible total
space, then E ′ → B ′ is also a classifying bundle and the classifying map E ′ → EG
is a G-equivariant homotopy equivalence. Unfortunately, the analogy is not perfect:
We would prefer a stronger statement that W̃ g → W (resp. W g → W if W is
commutative) is a g-homotopy equivalence.
(b) An immediate consequence of this theorem is that Hbasic (W̃ g) = Hbasic (W g) = (Sg∗ )inv .
It shows furthermore that in the definition of equivariant cohomology, the Weil algebra
W g may be replaced by any other locally free g-dga with trivial cohomology.
Theorem 6.3 will easily follow from the following Lemma:
6. EQUIVALENCE BETWEEN THE SIMPLICIAL AND WEIL MODEL 55
Lemma 6.5. Let W be as in Theorem 6.3, any A any locally free g-dga. Then the inclusion
A ֒→ W ⊗ A, α 7→ 1 ⊗ α induces an isomorphism in basic cohomology.
Proof of Theorem 6.3. The characteristic map cθ : W̃ g → W can be written as a
composition of two maps,
W̃ g → W̃ g ⊗ W → W,
where the first map is given by z 7→ z ⊗ 1 and the second map is z ⊗ w 7→ cθ (z)w. The second
map is a g-chain homotopy equivalence by part (b) of Theorem 6.2. Hence the map obtained by
tensoring with the identity map for any g-dga B is a g-chain homotopy equivalence as well, and
in particular induces an isomorphism in basic cohomology. Lemma 6.5 applied to A = W̃ g ⊗ B
shows that the first induces an isomorphism in basic cohomology as well.
We now turn to the proof of Lemma 6.5. Note that this is slightly weaker than the corre-
sponding statement for W̃ g.
Proof. The proof is modeled after Guillemin-Sternberg, [17][page 46] The idea is to apply
h ⊗ 1 to W ⊗ A to show that the W factor does not change the basic cohomology. This does not
directly work, however, since h need not commute with the contraction operators ιW ξ . To get
around this difficulty we use the Kalkman trick: Let θ be a connection on A, and let ψ be the
nilpotent degree 0 operator ψ = θa ιW a
a . (Note that this need not be a derivation, since θ are
elements of a non-commutative algebra.) Then exp ψ is g-equivariant and intertwines ιξ + ιAW
ξ
with ιAa . Thus, after applying exp ψ the operator h no longer interferes with contractions on W .
Unfortunately, h is no longer a homotopy operator since ψ changes the differential! Fortunately,
the change of dW + dA can be controlled. Introduce a filtration
W−1 ⊂ W0 ⊂ W1 ⊂ · · ·
on W by setting W−1 = R and M
Wi = Wj
j≤i
W
for i ≥ 0. Then ιW
ξ lowers the filtration degree by 1, LWξ preserves it, d raises it by 1, and h
lowers it by −1. Let W ⊗ A be equipped with the filtration induced from the filtration on W
(the grading on A plays no role). It is easily checked that the twisted differential
D := Ad(exp ψ)(dW + dA )
has the form D = dW + · · · where the dots indicate additional terms that preserve the filtration
degree. Suppose now that α ∈ (W ⊗ Ahor )inv is D-closed and has filtration degree N ≥ 0. Then
α − D(h ⊗ 1)α is cohomologous to α, lies in (W ⊗ Ahor )inv , and has filtration degree N − 1.
Indeed, the equation [d, h] = id −Π implies
[D, h ⊗ 1] = id −Π ⊗ 1 + · · · ,
where the dots lower the filtration degree by at least 1. Iterating, it follows that
(id −D(h ⊗ 1))N +1 α
is cohomologous to α and has filtration degree −1, thus lies in Abasic . This shows that the map
Hbasic (W ⊗ A) → Hbasic (A) is onto. By a similar argument, given α ∈ Abasic with α = Dβ
for some β ∈ (W ⊗ A)basic , we may add a cocycle to β to obtain an element of (W0 ⊗ A)basic .
Hence the map is also injective.
56 CONTENTS
6.3. Equivalence of simplicial and Weil model of equivariant cohomology. An-

other example comes from the simplicial model for the classifying bundle: Recall that we
defined a non-commutative product structure on the double complex
C p,q = Ωq (Gp+1 ).
The principal G-action on E• G = G•+1 is given by the diagonal action from the right, and this
defines Lie derivatives and contractions
ιξ : C p,q → C p,q−1 , Lξ : C p,q → C p,q .
L
Letting W i = p,q
p+q=i C , this defines a g-dga. A connection 1-form is given by the left-
invariant Maurer-Cartan form θL ∈ Ω1 (G ⊗ g, viewed as an element
θL ∈ C 0,1 ⊗ g.
Finally, the D-cohomology is acyclic, since it is isomorphic to the cohomology of EG. (One may
also directly construct a homotopy operator.) The basic subcomplex of Ω• (E• G) is isomorphic
to Ω• (B• G). It follows that the characteristic maps
W g ← W̃ • g → C • (G•+1 )
induce isomorphisms in basic cohomology, and more generally, if M is a G-manifold,
Hbasic (W̃ g ⊗ Ω(M )) .

jj SSS
jjjjjjj SSS
SSS
jjj SSS
t jjj
j SS)
Hbasic (W g ⊗ Ω(M )) H ∗ (EG ×G M )
This proves that the Weil/Cartan model does indeed compute the equivariant cohomology of
M , defined in terms of the Borel model.
We should point out that there is, in fact, a canonical homomorphism of g-differential
spaces (not of algebras) W g → C • (G•+1 ). Indeed, we have
Theorem 6.6. The symmetrization map W g = S(Eg∗ ) → W̃ g = T (Eg∗ ) is a homomor-
phism of g-differential spaces, i.e. it intertwines d, ιξ , Lξ . For any g-dga A, the induced map
in basic cohomology
Hbasic (W g ⊗ A) → Hbasic (W̃ g ⊗ A)
is inverse to the algebra isomorphism Hbasic (W̃ g⊗A) → Hbasic (W g⊗A) induced by W̃ g → W g.
Proof. Since W g → W̃ g → W g (symmetrization followed by the quotient map) is the
identity, and we already know that Hbasic (W̃ g ⊗ A) → Hbasic (W g ⊗ A) is an isomorphism,
it’s enough to show that W g → W̃ g intertwines ιξ , Lξ , d. This in fact follows from a more
general statement: If E is any graded vector space, and A End(E) any endomorphism, then A
extends to derivations of both the symmetric algebra S(E) and the tensor algebra T (E), and
the symmetrization map intertwines the two derivations. This directly applies to d, Lξ since
these are both induced from endomorphisms of Eg∗ . For ιξ , one may apply a small trick and
replace Eg∗ with Eg∗ ⊕ R, and consider the contraction operators defined by
ι̃ξ (µ) = hµ, ξi c, ι̃ξ (µ) = − ad∗ξ µ
7. LOCALIZATION 57
where c is a generator of R. The statement above applies to A = ι̃ξ , and taking the quotient by
the ideal generated by c − 1 we find that W g → W̃ g intertwines the contractions as well.
Thus, in particular we have a canonical map
(Sg∗ )inv ⊂ (W g)basic → (W̃ g)basic → Ω• (G•+1 )basic = Ω• (G• ).
For example, if one is given an inner product on g, the corresponding quadratic polynomial
ξ 7→ ||ξ||2 gives rise to an element of Ω• (G• ) of total degree 4. One may check (with some
effort) that this is the element of Ω3 (G) ⊕ Ω2 (G2 ) described earlier.
7. Localization
R
Let M be a compact oriented G-manifold. The integration map : Ω(M ) → R extends to
a map from the Cartan complex,
Z
: ΩG (M ) → ΩG (pt) = (Sg∗ )G .
By Stokes’ theorem, the integral vanishes on equivariant coboundaries, since (dG β)(ξ) equals
d(β(ξ)) up to terms of lower differential form degree. Hence it induces an integration map in
cohomology, Z
: HG (M ) → HG (pt) = (Sg∗ )G .
The localization formula of Berline-Vergne [4, 3, 2] and Atiyah-Bott [1] gives an explicit
expression for the integral of any equivariant cocycle in terms of fixed point data, provided
G is a compact Lie group. The formula generalizes the Duistermaat-Heckman formula [12]
from symplectic geometry, and also Bott’s formulas for characteristic numbers [5]. Berline-
Vergne’s proof used differential-geometric ideas, while Atiyah-Bott’s proof was more topological
in nature. The proof given below is essentially Berline-Vergne’s proof, except that we use “real
blow-ups” to replace their limiting arguments.
7.1. Statement of the localization formula. As pointed out in Berline-Vergne’s paper
[4], the localization formula holds in a wider context than that of equivariant cohomology.
Indeed, for fixed ξ ∈ g consider any differential form α ∈ Ω(M ) such that α is annihilated by
the derivation dξ = d − ιξM on Ω(M ). For example, if β ∈ ΩG (M ) is an equivariant cocycle,
then α := β(ξ) is annihilated by dξ .
Example 7.1. Let ω + Φ be a closed equivariant 2-form on M . Then α := exp(ω + hΦ, ξi)
is dξ -closed. It does not strictly speaking define an equivariant differential form, however, it is
not polynomial in ξ. This is the setting for the Duistermaat-Heckman theorem.
R
The fixed point formula expresses the integral M α as a sum over the zeroes of the vector
field ξM . Note that the zeroes of ξM are also the fixed point sets for the action of the torus
generated by ξ (i.e. of {exp(tξ)|, t ∈ R}), and in particular are smooth embedded submani-
folds. To simplify the discussion, we will first assume that the set of zeroes is isolated, i.e.
0-dimensional.
−1
For any zero x ∈ ξM (0), let Ax (ξ) : Tx M → Tx M denote the infinitesimal action of ξ.
That is,
d
Ax (ξ)(v) := |t=0 exp(tξ)∗ v.
dt
58 CONTENTS
Choose a G-invariant Riemannian metric on M , then Ax is skew-adjoint for such a Riemannian

metric. Using the orientation on M , we may therefore define the Pfaffian, 13
det1/2 (Ax (ξ)).

Theorem 7.2. Let G be a compact Lie group, and M a compact, oriented G-manifold.
Suppose that the vector field generated by ξ ∈ g has isolated zeroes. Then for all forms α ∈
Ω∗ (M ) such that dξ α = 0, one has the integration formula
Z X α[0] (x)
α[dim M ] = 1/2
.
M ξ (x)=0
det (Ax (ξ))
M
Here α[0] ∈ C ∞ (M ) is the form degree 0 part of α.

7.2. Proof of the Localization formula. In the proof we will use the notion of real
blow-ups. (We learned about this from lecture notes of Richard Melrose, see e.g. [22]. The
concept is also briefly discussed in Duistermaat-Kolk [13], page 125.) Consider first the case
of a real vector space V . Let
S(V ) = V \{0}/R>0
be its sphere, thought of as the space of rays based at 0. Define V̂ as the subset of V × S(V ),
V̂ := {(v, x) ∈ V × S(V )| v lies on the ray parametrized by x}.
Then V̂ is a manifold with boundary. (In fact, if one introduces an inner product on V then
V̂ = S(V ) × R≥0 ). There is a natural smooth map π : V̂ → V which is a diffeomorphism
away from S(V ). If M is a manifold and m ∈ M , one can define its blow-up π : M̂ → M by
using a coordinate chart based at m. Just as in the complex category, one shows that this is
independent of the choice of chart (although this is actually not important for our purposes).
Suppose now that M is a G-space as above. Let π : M̂ → M be the manifold with boundary
obtained by real blow-up at all the zeroes of ξM . It follows from the construction that the vector
field ξM on M lifts to a vector field ξM̂ on M̂ with no zeroes. Choose a ξM̂ -invariant Riemannian
metric g on M̂ , and define
g(ξM̂ , ·)
θ := ∈ Ω1 (M̂ ).
g(ξM̂ , ξM̂ )
Then θ satisfies ι(ξM̂ )θ = 1 and d2ξ θ = LξM θ = 0. Therefore
θ θ X
γ := = = −θ ∧ (dθ)j
dξ θ dθ − 1
j
13Recall that for any vector space V with a given inner product (possibly indefinite), there is a canonical
isomorphism o(V ) ∼= ∧2 (V ), A 7→ λ(A) between skew-symmetric matrices and the second exterior power of V .
Suppose dim V is even. The Pfaffian det1/2 : o(V ) → R is a distinguished choice of square root, where for
n
A invertible, the sign is characterized by the condition that detλ(A)
1/2 (A) is a volume form compatible with the
orientation on V . The Pfaffian changes sign with any change of orientation.

7. LOCALIZATION 59
is a well-defined form satisfying dξ γ = 1. The key idea of Berline-Vergne is to use this form for
partial integration:
Z Z
α = π∗ α
M M̂
Z
= π ∗ α ∧ dξ γ
M̂
Z
= dξ (π ∗ α ∧ γ)
ZM̂
= d(π ∗ α ∧ γ)
M̂
X Z
= π∗ α ∧ γ
S(Tp M )
p∈M T
X Z
= α[0] (p) γ
S(Tp M )
p∈M T
Thus, to complete the proof we have to carry out the remaining integral over the sphere. We
will do this by a trick, defining a dξ -closed form α where we can actually compute the integral
by hand. L
For any zero x ∈ M , choose a decomposition Tx M = Vi , where each Vi is a 2-dimensional
subspace invariant under Ax (ξ). Choose orientations on Vi such that the product orientation
is the given orientation on Tx M . Then the Pfaffian of Ax (ξ) is the product of the Pfaffians
for the restrictions to Vi . On each Vi , introduce polar coordinates ri , φi compatible with the
orientation. Given ǫ > 0 let χ ∈ C ∞ (R≥0 ) be a cut-off function, with χ(r) = 1 for r ≤ ǫ and
χ(r) = 0 for r ≥ 2ǫ. Let
n
Y n
Y
α = (2π)−n dξ (χ(rj )dφj ) = (2π)−n χ(rj )ιξ dφj − χ′ (rj )drj ∧ dφj .
j=1 j=1
Note that this form is well-defined (even though the coordinates are not globally well-defined),
compactly supported and dξ -closed. Its integral is equal to
Z Yn
α= (−χ′ (rj )drj ) = 1.
Tp M j=1
On the other hand

n
Y
α[0] (0) = (2π)−n ιξ dφj .
j=1
which (as one easily verifies) is just the Pfaffian of Ax (ξ). Choosing ǫ sufficiently small, we can
consider α as a form on M , vanishing at all the other fixed points. Applying the localization
formula we find Z Z
1= α = det1/2 (Ax (ξ)) γ,
M S(Tp M )
60 CONTENTS
thus Z
γ = det−1/2 (Ax (ξ)).
S(Tp M )
Q.E.D.
The above discussion extends to non-isolated fixed points, in this case the Pfaffian det1/2 (Ax (ξ))
is replaced by the equivariant Euler class of the normal bundle of the fixed point manifold.
7.3. The Duistermaat-Heckman formula. One often applies the Duistermaat-Heckman
theorem in order to compute Liouville volumes of symplectic manifolds with Hamiltonian group
action. Consider for example a Hamiltonian S 1 = R/Z-action on a symplectic manifold (M, ω),
with isolated fixed points. That is, the action is defined by a Hamiltonian H ∈ C ∞ (M ) with
periodic flow, of period 1. Then
Z
ωn 1 X etH(p)
etH = n Q .
M n! t 1 j aj (p)
p∈M S
where aj (p) are the weights for the S 1 actions at the fixed points. Notice by the way that
the individual terms on the right hand side are singular for t = 0. This implies very subtle
relationships between the weight, for example one must have
X H(p)k
Q =0
1 j aj (p)
p∈M S
for all k < n. For the volume one reads off,
1 X H(p)n
Vol(M ) = Q .
n! 1 j aj (p)
p∈M S
Bibliography
1. M. F. Atiyah and R. Bott, The moment map and equivariant cohomology, Topology 23 (1984), no. 1, 1–28.
2. N. Berline, E. Getzler, and M. Vergne, Heat kernels and Dirac operators, Grundlehren der mathematischen
Wissenschaften, vol. 298, Springer-Verlag, Berlin-Heidelberg-New York, 1992.
3. N. Berline and M. Vergne, Classes caractéristiques équivariantes. formule de localisation en cohomologie
équivariante., C. R. Acad. Sci. Paris Ser. I Math. 295 (1982), 539–541.
4. , Z’ero d’un champ de vecteurs et classes caractéristiques équivariantes, Duke Math. J. 50 (1983),
539–549.
5. R. Bott, Vector fields and characteristic numbers, Mich. Math. J. 14 (1967), 231–244.
6. R. Bott and L. Tu, Differential forms in algebraic topology, Graduate Texts in Mathematics, vol. 82, Springer-
Verlag, New York, 1982.
7. T. Bröcker and T. tom Dieck, Representations of compact Lie groups, Graduate Texts in Mathematics,
vol. 98, Springer-Verlag, Berlin-Heidelberg-New York, 1985.
8. H. Cartan, La transgression dans un groupe de Lie et dans un fibré principal, Colloque de topologie (espaces
fibrés) (Bruxelles), Centre belge de recherches mathématiques, Georges Thone, Liège, Masson et Cie., Paris,
1950, pp. 73–81.
9. , Notions d’algèbre différentielle; application aux groupes de Lie et aux variétés où opère un groupe de
Lie., Colloque de topologie (espaces fibrés) (Bruxelles), Georges Thone, Liège, Masson et Cie., Paris, 1950.
10. James F. Davis and Paul Kirk, Lecture notes in algebraic topology, Graduate Studies in Mathematics, vol. 35,
American Mathematical Society, Providence, RI, 2001. MR 2002f:55001
11. J. J. Duistermaat, The heat kernel Lefschetz fixed point formula for the spin-c Dirac operator, Progress in
Nonlinear Differential Equations and Their Applications, vol. 18, Birkhäuser, Boston, 1996.
12. J. J. Duistermaat and G. J. Heckman, On the variation in the cohomology of the symplectic form of the
reduced phase space, Invent. Math. 69 (1982), 259–268.
13. J. J. Duistermaat and J. A. C. Kolk, Lie groups, Springer-Verlag, Berlin, 2000.
14. J. L. Dupont, Simplicial de Rham cohomology and characteristic classes of flat bundles, Topology 15 (1976),
no. 3, 233–245.
15. M. Goresky and R. MacPherson, Stratified Morse theory, Ergebnisse der Mathematik und ihrer Grenzgebiete,
vol. 14, third series, Springer-Verlag, Berlin-Heidelberg-New York, 1988.
16. Werner Greub, Stephen Halperin, and Ray Vanstone, Connections, curvature, and cohomology, Academic
Press [Harcourt Brace Jovanovich Publishers], New York, 1976, Volume III: Cohomology of principal bundles
and homogeneous spaces, Pure and Applied Mathematics, Vol. 47-III. MR 53 #4110
17. V. Guillemin and S. Sternberg, Symplectic techniques in physics, Cambridge Univ. Press, Cambridge, 1990.
18. Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002. MR 2002k:55001
19. Wu-yi Hsiang, Cohomology theory of topological transformation groups, Springer-Verlag, New York, 1975,
Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 85. MR 54 #11363
20. J. Kalkman, A BRST model applied to symplectic geometry, Ph.D. thesis, Universiteit Utrecht, 1993.
21. S. Kumar and M. Vergne, Equivariant cohomology with generalized coefficients, Astérisque 215 (1993),
109–204.
22. R. B. Melrose, The atiyah-patodi-singer index theorem, Research Notes in Mathematics, vol. 4, A K Peters,
Ltd., wellesley, 1993.
23. J. Milnor, Construction of universal bundles. II, Ann. of Math. (2) 63 (1956), 430–436.
61
62 BIBLIOGRAPHY
24. M. Mostow and J. Perchik, Notes on Gelfand-Fuks cohomology and characteristic classes, Proceedings of the
eleventh annual holiday symposium, New Mexico State University (1973), Lectures delivered by R. Bott,
pp. 1–126.
25. L. Nicolaescu, On a theorem of Henri Cartan concerning the equivariant cohomology, An. Stiint. Univ. Al.
I. Cuza Iasi. Mat. (N.S.) 45 (1999), no. 1, 17–38 (2000), see also arXiv:math.DG/0005068.
26. A. L. Onishchik and E. B. Vinberg, Lie groups and algebraic groups, Springer-Verlag, Berlin-Heidelberg-New
York, 1990.
27. G. Segal, Classifying spaces and spectral sequences, Inst. Hautes Études Sci. Publ. Math. (1968), no. 34,
105–112.
28. R. Sjamaar, Singular orbit spaces in Riemannian and symplectic geometry, Ph.D. thesis, Rijksuniversiteit
te Utrecht, 1990.

Group Actions On Manifolds: Eckhard Meinrenken

Uploaded by

Copyright:

Available Formats

Group Actions On Manifolds: Eckhard Meinrenken

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Group Actions On Manifolds: Eckhard Meinrenken

Uploaded by

Copyright:

Available Formats

Group actions on manifolds

Lecture Notes, University of Toronto, Spring 2003

1. Terminology and notation 4

1. Terminology and notation

definition of the Lie bracket.

Theorem 1.11. Suppose ξ 7→ ξM is a Lie algebra action of g on a manifold M . Then this

1.4. Proper actions.

open neighborhood of m. Since φ(U ) = φ(U h) for h ∈ Gm , it is no loss of generality to assume

h.m = limi hi .m = m, showing H ⊂ Gm . Since Gm is compact, H is compact also. Thus we

exists a G-equivariant diffeomorphism from νN to a neighborhood of N in M , restricting to the

be the corresponding decomposition of M , where Mi is the pre-image of Xi . (Of course, if G

1.6. Principal orbit type theorem.

Proof. Note first that by definition of “depth”,

where the upper horizontal map is G-equivariant.

StR (k, n) = {(v1 , . . . , vk ) ∈ Rkn | vi · vj = δij }

as a principal O(k)-bundle over the Grassmann manifold GrR (k, n) of k-planes in Rn .

We will need two basic constructions with principal bundles π : P → B.

2.2. Pull-backs. Let π : P → B be a principal G-bundle. If f : X → B is a continuous

which in fact defines f ∗P up to isomorphism.

Proof. We have a classifying map f : B ′ G → BG for E ′ G → B ′ G (viewing EG → BG

(c) For G = O(n) one has, with Z2 -coefficients,

The inclusion of pt = eH into G/H gives a map

3. Construction of EG by simplicial techniques

dividing out the relations `

3.3. Examples of simplicial spaces.

ǫi (h1 , . . . , hn , x) = (h1 , . . . , hi , e, hi+1 , . . . , hn , x).

{0}• × E• X → E• X, ((0, . . . , 0), (x0 , . . . , xn )) 7→ (x0 , . . . , xn )

If R is any abelian coefficient group (typically R = R, Z, Z2 ), we can consider homology with

Theorem 3.15. The inclusion C̃ k (G)

hence we obtain a filtration of the D-cohomology H k = H k (C, D):

Since dr is induced from D, it is immediate that dr squares to 0, so dr is a differential. Its

4.4. The simplicial de Rham theorem. As an example of spectral sequence techniques,

θL ∈ Ω1 (G)L ⊗ g which is defined by ι(ξ L )θL = ξ. Similarly let θR be the right-Maurer-Cartan

5. The Chern-Weil construction

Thus, a connection on P is a G-equivariant bundle map T P → P × g taking ξP to ξ. It is thus

One of its basic properties is the Bianchi identity

[d, Lξ ] = 0, [Lξ , Lξ′ ] = L[ξ,ξ′ ]g ,

As in the case of principal bundles, F θ takes values in horizontal elements. If we extend

Note that by contrast, the subalgebra of horizontal elements is not d-invariant.

In general, given two homomorphisms of g-differential spaces, c0 , c1 : A → A′ , we define

5.4. The algebraic Chern-Weil construction. To determine the basic subcomplex, it

5.5. The Weil model of equivariant cohomology. Recall that if M is a G-manifold,

adψ ιa = adψ (ιW A

Note that the horizontal projection operator may also be written,

5.7. Examples of equivariant differential forms. Cartan’s model of equivariant co-

Examples 5.19. (a) If (M, ω) is a compact symplectic manifold, with a symplectic

degree i = 3 there is a short exact sequence,

5.9. The Künneth formula. (No time ...See Kumar-Vergne [21].)

6. Equivalence between the simplicial and Weil model

(a) For any connection θ on A, there exists a unique homomorphism of g-dga’s cθ : W̃ g →

6.3. Equivalence of simplicial and Weil model of equivariant cohomology. An-

Hbasic (W̃ g ⊗ Ω(M )) .

Choose a G-invariant Riemannian metric on M , then Ax is skew-adjoint for such a Riemannian

det1/2 (Ax (ξ)).

Here α[0] ∈ C ∞ (M ) is the form degree 0 part of α.

V̂ := {(v, x) ∈ V × S(V )| v lies on the ray parametrized by x}.

orientation on V . The Pfaffian changes sign with any change of orientation.

On the other hand