Notes on quotients and group actions

Erik van den Ban

Fall 2006

1 Quotients
Let X be a topological space, and R an equivalence relation on X. The set of equivalence
classes for this relation is denoted by X/R and the canonical map by p : X → X/R. The
quotient X/R is equipped with the quotient topology. This is the finest topology for which
the map p : X → X/R is continuous. Thus, a subset U ⊂ X/R is open if and only if its
preimage p−1 (U ) is open in X. The quotient topology is uniquely determined by the following
universal property. For every topological space Y and every continuous map f : X → Y that
is constant on the equivalence classes, the factored map f¯ : X/R → Y is continuous.

Lemma 1.1 If X/R is Hausdorff, then R is closed in X × X. If p : X → X/R is open, the

converse is also true.

Proof: The map p × p : X × X → X/R × X/R is continuous (with respect to the product
topologies). Since X/R is Hausdorff, the diagonal ∆ is closed in X/R × X/R. Therefore, its
preimage is closed in X × X. We now observe that
R = (p × p)−1 (∆).
Conversely, assume that p is open. Then p × p is open as well. Since X × X \ R is open, its
image under p × p is open. This image is equal to (X/R × X/R) \ ∆. Hence ∆ is closed, and
it follows that X/R is Hausdorff. 

Remark 1.2 In general, the quotient topology for p × p on X/R → X/R is finer than the
product topology (of the quotient topologies on the two factors X/R). If the two topologies
coincide, then R closed implies X/R Hausdorff. The topologies coincide if p is open.

Let M be a manifold, and R an equivalence relation on M. In the rest of this section we

will address the question whether the quotient space M/R carries a manifold structure such
that the natural projection map p : M → M/R has reasonable properties. The map p should
at least be smooth. It seems also reasonable to require that p be a submersion. We will first
show that if such a manifold structure exists, it is necessarily unique. The following lemma
will be a useful tool.

Lemma 1.3 Let M, N, Z be manifolds, and let the following be a commutative diagram of
maps between sets:
Z
α . & β
f
M −→ N

1
Assume that α is a surjective submersion, and that β is smooth. Then f is smooth. If in
addition β is a submersion, then so is f. If in addition β is surjective, then so is f.

Proof: Let m ∈ M. Then m = α(z) for some z ∈ Z. Using the normal form of α at z, we see
that there exists an open neighborhood O of m in M and a smooth map s : O → Z such that
s(m) = z and α ◦ s = I on O. It follows that f = f ◦ α ◦ s = β ◦ s on O. Hence, f is smooth on
O. 

Corollary 1.4 Let p : M → N be a surjective map between sets. Assume that M has a
structure of smooth manifolds. Then N has at most one structure of smooth manifold for
which p is a submersion.

Proof: For j = 1, 2, let Nj be N equipped with a structure of smooth manifold such that
p : M → Nj is a submersion. Let I : N1 → N2 be the identity map. Then by the above
lemma, I is smooth. Similarly, the inverse of I is smooth, hence I is a diffeomorphism. 
We return to the situation of a smooth manifold M equipped with an equivalence relation
R. We will briefly say that M/R is a smooth manifold, if and only if M/R has a structure
of smooth manifold for which p : M → M/R is a smooth submersion. This structure is then
necessarily unique.
In the following lemma we view R as a subset of M × M. Let prj denote the projection
map M × M → M on the j-th component, for j = 1, 2. Then the equivalence class of an
element a ∈ M may be described as [a] = pr1 ((M × {a}) ∩ R).

Lemma 1.5 Assume that M/R is a smooth manifold. Then R is a closed submanifold of
M. Moreover, the projection maps prj |R : R → M are surjective submersions, for j = 1, 2.

Proof: By definition, the projection map p is a submersion, hence α := I × p : M × M →

M ×M/R is a submersion as well. The graph Γ = {(x, p(x)) | x ∈ M } is a smooth submanifold
of M × M/R. We note that R = α−1 (Γ). Hence R is a closed submanifold of M × M
and α|R : R → Γ is a surjective submersion. Since pr1 ◦ (I, p) = I on M, it follows that
pr1 : Γ → M is a surjective submersion. It follows that the composed map pr1 ◦ α is a
surjective submersion R → M. We now observe that the composed map equals pr1 |R . 

Remark 1.6 The above is a special instance of a more general result, namely that the pull-
back of a submersion is a submersion. More precisely, let f : M → N be a smooth map, and
p : Z → N a submersion. Then set-theoretically the pull-back f ∗ Z may is defined as the set
{(m, z) ∈ M × Z | f (m) = p(z)}. This set equals the inverse image of graph(f ) under the
submersion I × p. Since graph(f ) is a smooth submanifold, it follows that f ∗ Z is a smooth
submanifold. Finally, by a similar argument as in the above proof it follows that projection
onto the first factor induces a submersion f ∗ p : f ∗ Z → M.

The following result, due to R. Godement, expresses that the converse is also true.

Theorem 1.7 Let R be an equivalence relation on M and assume that R is a closed sub-
manifold of M × M such that pr1 |R is a surjective submersion onto M. Then M/R has a
unique structure of smooth manifold such that p : M → M/R is a submersion.

2
 We assume the conditions of the theorem to be fulfilled, and will present the proof in a
number of lemmas, following the exposition of J.-P. Serre, p. 92. in [Serre]. 1 We begin by
observing that, by symmetry of the relation, the projection pr2 |R : R → M is a submersion
onto M as well.

Lemma 1.8 The projection map p : M → M/R is open. The quotient topology is Hausdorff.

Proof: Let U ⊂ M. Then p−1 (p(U )) = pr1 ((M × U ) ∩ R). Since pr1 |R is a submersion, we
see that for U open, the set p−1 (p(U )) is open in M, hence p(U ) is open in M/R. Therfore,
p is an open map. The last assertion now follows by application of Lemma 1.1. 

Lemma 1.9 Each equivalence class of R is a closed submanifold of M of dimension dim R−

dim M.

Proof: The idea is that [x] = pr1 (pr−1

2 ({x})), for every x ∈ M. By symmetry, pr2 : R → M
is a submersion as well. Thus, if x ∈ M, then pr−12 ({x}) is a smooth submanifold of R, of
dimension dim R − dim M. Now pr2 ({x}) = (M × {x}) ∩ R hence pr−1
−1
2 ({x}) is a smooth
submanifold of M × {x}. Thus, the projection onto the first coordinate, which equals [x], is
a smooth submanifold of M. 
The next lemma expresses that locally near a point a ∈ M, we may parametrize the equiv-
alence classes of R by a smooth submanifold through a that is transversal to [a]. Eventually
this will give a local coordinatisation of the orbit space near [a].

Lemma 1.10 Let a ∈ M. Then there exists an open neighborhood U 3 a, a closed submani-
fold S of U and a submersion q : U → S such that for every x ∈ U, the set [x] ∩ U intersects
S in the single point q(x).

Proof: We fix a submanifold S 0 of M through a such that Ta S 0 ⊕ Ta [a] = Ta M. We will

eventually show that all assertions hold for a suitable neighborhood U of a in M, and for
S = S 0 ∩ U.
The manifold S 0 has codimension dim[a] = dim R − dim M. Since pr2 |R is a submersion
from R onto M it follows that Z := pr2 (S 0 )∩R = (M ×S 0 )∩R is a submanifold of codimension
dim R − dim M in R. Therefore, Z has the same dimension as M.
We observe that [a] × {a} is a submanifold of Z and that the diagonal ∆S 0 of S 0 × S 0 is
a submanifold of Z. The dimensions of these submanifolds add up to dim M. Moreover, the
tangent spaces at (a, a) are given by Ta [a] × {0} and ∆Ta S 0 , hence have trivial intersection. It
follows that T(a,a) Z is the direct sum of [a] × {0} and ∆Ta S 0 . We thus see that pr1 |Z : Z → M
is a local diffeomorphism at (a, a). It follows that there exists an open neighborhood O of a
in M such that pr1 is a diffeomorphism from Z × (O × O) onto an open neighborhood U 0 of
a in M. Let s denote the inverse of this isomorphism, then pr1 ◦ s = IU 0 and q = pr2 ◦ s is a
submersion from U 0 onto an open subset of S 0 ∩ O.
Accordingly, s(x) = (x, q(x)) for all x ∈ U 0 , and we see that U 0 ⊂ O. If x ∈ U 0 ∩ S 0 , then
(x, x) ∈ (O × O) ∩ Z and pr1 (x, x) = pr1 s(x), hence s(x) = (x, x) and we see that q(x) = x.
In particular this implies that q(q(x)) = q(x) as soon as q(x) ∈ U 0 .
1
J.-P. Serre, Lie algebras and Lie groups; Lecture Notes in Mathematics, 1500, Springer-Verlag, New York,
1992

3
 We now put U = U 0 ∩ q −1 (U 0 ∩ O) and S = S 0 ∩ U and claim that U and S satisfy the
requirements of the theorem. Indeed, clearly S ⊂ U. If x ∈ U, then q(x) ∈ U 0 and therefore
q(q(x)) = q(x) ∈ U 0 ∩ O from which we see that q(x) ∈ U ∩ S 0 = S.
Finally, let x ∈ U. By what we showed, q(x) ∈ S 0 ∩ U = S. Moreover, if y ∈ S is such
that y ∈ [x], then it follows that (x, y) ∈ (O × O) ∩ ((M × S) ∩ R) = (O × O) ∩ Z, hence
(x, y) = pr1 (x, y), from which we deduce that y = q(x). This shows that the intersection of
[x] ∩ U and S consists of the single point q(x). 

Remark 1.11 A submanifold S with the properties mentioned above is called a slice through
the classes of R in U.
If V ⊂ M is a subset, then by RV we denote the restriction of R to V. Thus, in terms of
graphs, RV = (V × V ) ∩ R. The classes of RV in V are the sets [x] ∩ V, for x ∈ V. It follows
that the inclusion V ,→ M factors to an inclusion V /RV ,→ M/R. The projection V → V /RV
will be denoted by pV . Then the diagram
V ,→ M
pV ↓ ↓p
V /RV ,→ M/R
commutes.
Corollary 1.12 Let a ∈ M. Then there exists an open neighborhood U of a in M such that
U/RU is a manifold.
Proof: Let U, S, q be as in the above lemma. Then the map q : U → S factors to a bijection
q̄ : U/RU → S. Accordingly, q = q̄ ◦ pU . We use the bijection q̄ to transfer the manifold
structure of S to a manifold structure on U/RU . Then pU becomes a submersion. 
A subset V ⊂ M is called saturated if and only if it is a union of equivalence classes for
R. Equivalently, this means that V = p−1 (p(V )). The next step is to extend the manifold
structures on open neighborhoods to manifold structures on saturated open neighborhoods.
Lemma 1.13 Let U ⊂ M be an open subset such that U/RU is a manifold. Let V =
p−1 (p(U )). Then the natural map q : V → U/RU is a surjective submersion.
Proof: Let pU : U → U/RU and pV : V → V /RV denote the projection maps. The inclusion
map i : U → V factors to a bijection j : U/RU → V /RV . The map q is given by j ◦ q = pV .
From this we infer that the following diagram commutes
pr
(M × U ) ∩ R →2 U
pr1 ↓ ↓ pU
q
V → U/RU .
The top arrow and right vertical arrow represent surjective submersions, and so does the left
vertical arrow. By application of Lemma 1.3 it follows that q is a surjective submersion. 

Corollary 1.14 Let U, V be as above. If U/RU is a smooth manifold, then so is V /RV .

Proof: With notation as in the proof of the above lemma, j : U/RU → V /RV is a bijection
and j ◦ q = pV . We equip V /RV with the manifold structure obtained by transport of structure
under j. Then pV becomes a submersion. 

4
Remark 1.15 It follows from the above that the unique map r : V → S defined by [x] ∩ S =
{r(x)} is a smooth surjective submersion. Accordingly, the slice S for U is also a slice for V,
the smallest saturated open set containing U.

Corollary 1.16 Every point a ∈ M is contained in a saturated open neighborhood V such

that V /RV is a manifold.

Completion of the proof of Theorem 1.7. It follows from the preceding corollary
that M has a covering with saturated open neighborhoods Ui such that Ui /RUi is a smooth
manifold for each i.
Let p : M → M/R be the projection map, then p|Ui : Ui → M/R factors to a bijection from
Ui /RUi onto the open subset p(Ui ) of M/R. We transfer the manifold structure from Ui /RUi
to a manifold structure on p(Ui ). Let i, j be indices such that p(Ui ) ∩ p(Uj ) 6= ∅. Since Ui and
Uj are saturated, p(Ui ) ∩ p(Uj ) = p(Ui ∩ Uj ). The manifold structure from p(Ui ) induces a
manifold structure Mi on p(Ui )∩p(Uj ) for which the projection map p : Ui ∩Uj → p(Ui )∩p(Uj )
is a submersion. Likewise, the manifold structure from p(Uj ) induces a manifold structure
Mj on p(Ui ) ∩ p(Uj ) such that p : Ui ∩ Uj → p(Ui ) ∩ p(Uj ) is a surjective submersion. By
application of Lemma 1.3 it follows that the two manifold structures coincide.
We conclude that M/R has a unique manifold structure for which all inclusion maps
Ui /RUi → M/R are diffeomorphisms onto open subsets. For this manifold structure, the
projection map p : M → M/R is a submersion. 

2 Quotients for group actions

In this section we apply Theorem 1.7 to suitable group actions. Let G be a group acting
(from the right) on a locally compact Hausdorff topological space M by homeomorphisms.
Associated to this group action is the equivalence relation R on M defined by (x, y) ∈ R ⇐⇒
xG = yG. Thus, the classes of R are the right G-cosets. We write M/G = M/R for the
associated coset space.

Lemma 2.1 The quotient map p : M → M/G is open.

Proof: Let U be open. Then p−1 p(U ) = ∪g∈G U g, which is a union of open sets hence open.
It follows that p(U ) is open. 

Corollary 2.2 Let G be a group acting on M by homeomorphisms. Then the quotient

topology on M/G is Hausdorff if and only if the map

α : M × G → M × M, (x, g) 7→ (x, xg). (2.1)

has closed image.

Proof: Since p : M → M/G is open, the quotient topology is Hausdorff if and only if R is
closed in M × M, see Lemma 1.1. We now observe that R equals the image of α. 

5
 The action of G on M is called free if for each x ∈ M the stabilizer subgroup Gx = {g ∈
G | xg = x} is trivial. It is called free and properly discontinuous if for each x ∈ M there
exists a neighborhood U such that U ∩ U g = ∅ for every g ∈ G, g 6= e. Equivalently, this
means that the sets U g, for g ∈ G are mutually disjoint.

Lemma 2.3 Let G have a free action on M by homeomorphisms. Then the following con-
ditions are equivalent:
(a) The action is properly discontinuous and the relation R is closed in M × M.
(b) The map α is proper (G equipped with the discrete topology).

If any of these conditions is fulfilled, each G-orbit is closed in M.

Proof: First assume (a). Since R is closed, and p is open, the quotient space M/G is
Hausdorff. It follows that for every x, y ∈ M with xG 6= yG there exist open neighborhoods
U 3 x and V 3 y such that U G ∩ V G = ∅. On the other hand, if xG = yG, then we may
choose an open neighborhood U of x such that the U g, for g ∈ G, are mutually disjoint. Let
y = xh and put V = U h. Then U g ∩ V 6= ∅ for only a single g. Thus in any case, given two
elements x, y ∈ M there exist open neighborhoods U of x and V of y such that U g ∩ V 6= ∅
for at most finitely many g ∈ G. Let now C ⊂ M be compact and let x ∈ C. Then by using
compactness, it follows that x has an open neighborhood U such that U g ∩ C 6= ∅ for only
finitely many g ∈ G. By using compactness once more, it follows that the collection of g ∈ G
such that C ∩ Cg 6= ∅ is finite.
Finally, let K ⊂ M × M be compact. Put C = pr1 (K) ∪ pr2 (K). Then C is compact and
K ⊂ C × C. Let F be the finite set of g ∈ G such that Cg ∩ C 6= ∅. Then α(x, g) ∈ K implies
that x ∈ C and g ∈ F. Hence α−1 (K) is a closed subset of the compact set C × F. It follows
that α−1 (K) is compact. Hence (b).
For the converse implication, assume (b). Then R = image(α) is closed. Moreover, let
x ∈ M. Then x has a compact neighborhood C. Then α−1 (C ×C) = {(x, g) ∈ C ×G | xg ∈ C}
is compact. Therefore the projection of this set onto the second coordinate is compact, hence
a finite subset F ⊂ G. We note that F equals {(g ∈ G | C ∩ Cg 6= ∅}. Since the action is free,
we may chose, for every f ∈ F \ {e} open neighborhoods Vf of x and Wf of xf such that
Vf ∩ Wf = ∅. Put Uf = Vf ∩ Wf f −1 . Then Uf ∩ Uf f = ∅. Let U be an open neighborhood of
x contained in C and in ∩f ∈F \{e} Uf . Then U g ∩ U = ∅ ⇒ g = e. 

Lemma 2.4 Let M be a smooth manifold, equipped with a free and properly discontinuous
action of a group G of diffeomorphims. Assume moreover that M/G is Hausdorff. Then
M/G is a manifold. Moreover, p : M → M/G is a smooth covering, for which G acts by
covering maps on M.

Proof: Let x ∈ M. Then there exists an open set U 3 x such that the set U g, for g ∈ G, are
mutually disjoint. Since M is second countable, it follows that G is at most countable. Hence
G may be viewed as a zero dimensional manifold, and accordingly, M × G may be viewed
as a manifold of the same dimension as M. Since G acts by diffeomorphisms, the map (2.1)
is an injective immersion. Since α is proper, it is an embedding onto the closed submanifold
R. It follows that M/G = M/R is a smooth manifold. The assertion that G acts by covering
transformations is immediate. 

6
 We will now consider the situation of a right action by a Lie group G on M. We first
recall some basic facts from the theory of Lie groups. A Lie group is a group (G, ·, e)
equipped with the structure of a smooth manifold such that the multiplication map m :
(x, y) 7→ xy, G × G → G and the inversion map ι : x 7→ x−1 , G → G are smooth. Given
a ∈ G the maps ra : G → G, x 7→ xa and la : G → G, x 7→ ax are diffeomorphisms from G
onto itself. A vector field v on G is said to be right-invariant if and only if ra∗ v = v for all
a ∈ G. The linear space of such vector fields is denoted by XR (G). Let g = Te G. Then for
every X ∈ g there is a unique right invariant vector field XR ∈ XR (G) with XR (e) = X. It
is given by XR (a) = dra (e)X. Accordingly, the map XR (G) → g defined by v 7→ v(e) is a
linear isomorphism. The space XR (G) is closed under Lie brackets of vector fields; in other
words, XR (G) is a sub Lie algebra of the algebra X(G) of all vector fields on G. By transport
of structure, there exists a unique structure of Lie algebra on g for which the isomorphism
XR (G) → g becomes an isomorphism of Lie algebras. Equippee with this structure, g is called
the Lie algebra of the Lie group G.
A one-parameter subgroup of G is a smooth group homomorphism from (R, +, 0) to G.
For every X ∈ g there exists a unique one-parameter subgroup αX : R → G with αX 0 (0) = X.

In fact, it coincides with the integral curve of the vector field XR with initial point e. The
exponential map exp : g → G is defined by

exp X := αX (1).

The exponential map is smooth; moreover, the one-parameter subgroup αX may be expressed
in terms of the exponential map by

αX (t) = exp tX.

In particular, this implies that


d 
exp(s + t)X = exp sX exp tX, exp tX = X.
dt t=0

The latter formula is equivalent to d exp(0) = Ig . In particular, we see that the exponential
map is a local diffeomorphism at 0.
The right action of G on M is said to be smooth if the action map M × G → M,
(m, g) 7→ mg is smooth. In this setting, every X ∈ g defines a smooth vector field XM on M,
given by the formula 
d 
XM (m) =  m exp tX. (2.2)
dt t=0
The action of G on M is called proper if the map (2.1) is proper. The situation of Lemma 2.4
may be viewed as a special case of the next result, with G a zero-dimensional (hence discrete)
group.

Lemma 2.5 Let (m, g) 7→ mg be a smooth action of a Lie group G on a smooth manifold
M. If the action is free and proper, then M/G has a unique structure of smooth manifold such
that the natural projection map p : M → M/G is a submersion.

Proof: We first observe that freeness of the action is equivalent to injectivity of α.

7
 For Y ∈ g, the Lie algebra of G, let YM denote the vector field on M given by (2.2). Let
now x ∈ M, g ∈ G; then the derivative of α at (x, g) is given by

d
dα(x, g)(X, dlg (e)Y ) = d1 α(x, g)X +  α(x, g exp tY )
dt t=0
= (X, drg (x)X) + (0, YM (xg)).

From this it follows that α is an immersion. Since α is proper, it follows that α is an embedding
onto a closed submanifold. Now R is the image of α, hence it remains to be verified that
pr1 : R → M is a submersion. This is immediate from the fact that pr1 ◦ α equals the
projection map M × G → M, hence is a submersion. The result now follows by application
of Theorem 1.7. 

3 Fiber bundles associated with group actions

Let B, F be smooth manifolds. A fiber bundle over a manifold B (the base manifold)
with fiber modelled on F is a smooth manifold M together with a submersion p : M → B
such that for every a ∈ B there exists an open neighborhood U 3 a and a diffeomorphism
τ : p−1 (U ) → U × F (trivialization) such that the following diagram commutes:
τ
U ×F −→ π −1 (U )
pr1 & .π
U

In particular, this means that every fiber p−1 (x), for x ∈ B is diffeomorphic to F.
Let G be a Lie group. A principal fiber bundle with structure group G over B is a
fiber bundle p : M → B (with fiber modelled on G) together with a smooth right action of G
on M such that for every a ∈ B there exists an open neighborhood U 3 a and a trivialization
τ : p−1 (U ) → U × G in which the action becomes trivialized in the sense that the following
diagram commutes, for all g ∈ G :
τ
p−1 (U ) −→ U × G
αg ↓ ↓1×rg
−1 τ
p (U ) −→ U × G

Here αg denotes the right action M → M, x 7→ xg and rg denotes the right multiplication
G → G, x 7→ xg.
Principal fiber bundles appear naturally in connection with proper free actions of G.

Theorem 3.1
(a) Let M be a smooth manifold, equipped with a proper and free action of G. Then M/G has
a unique structure of smooth manifold such that the canonical projection p : M → M/G
is a submersion. Furthermore, p : M → M/G is a principal fiber bundle with structure
group G.
(b) Let p : M → B be a principal fiber bundle with structure group G. Then the action of
G on M is proper and free, and the map p factors to a diffeomorpism M/G ' B.

8
Proof: Let M be as in (a). Then the first assertion follows from Lemma 2.5. For the second
assertion, we refer to the standard literature. One may also consult my lecture notes on Lie
groups. Assertion (b) is a rather straightforward consequence of the definition of a principal
fiber bundle.

Remark 3.2 A covering projection M → B may be viewed as a fiber bundle with discrete
fiber. In this spirit, if G is discrete, we retrieve the old equivalence of free and proper discrete
actions to coverings.

We shall now discuss the associated bundle construction. Let M be a smooth manifold
equipped with a proper and free smooth right action of a Lie group G. Then the natural map
p : M → M/G defines a principal fiber bundle with structure group G. Let F be a smooth
manifold equipped with a smooth left action of G. Then we may define a right action of G on
M × F by (y, z)g = (yg, g −1 z). This action is proper and free, since the action of G on M is
already so. It follows that M ×G F has a unique structure of smooth manifold such that the
projection map q : M × F → M ×G F is a submersion. We note that p ◦ pr1 : M × F → M/G
induces a map π : M ×G F → M/G which is smooth by Lemma 1.3.

Lemma 3.3 The map π : M ×G F → M/G defines a fiber bundle with fiber modelled on F.

Proof: We must show that π : M ×G F → M/G admits suitable local trivializations. For
this, let a ∈ M/G. Then the principal fiber bundle p : M → M/G admits a local trivialization
τ over some open neighborhood U of a. The map s : x 7→ τ −1 (x, e) defines a section of p over
U. We define the map ψ : U × F → p−1 (U ) by ψ(x, z) = q(s(x), z)). Then clearly, p ◦ ψ = pr1 .
Thus, if ψ is a diffeomorphism onto, it follows that ψ −1 is a local trivialization of π over U.
We will finish the proof by showing that ψ is indeed a diffeomorphism onto. We leave it to
the reader to verify that ψ is a bijection. Thus, it remains to be shown that ψ is a local
diffeomorphism. For dimensional reasons it suffices to show that ψ is a submersion.
Fix x ∈ U. From the fact that M → M/G is a principal bundle, it follows that the map
j : X 7→ XM (s(x)) is an injective linear map g → Ts(x) M such that

im (ds(x)) ⊕ im (j) = Ts(x) M.

Let y ∈ F and consider the map k : g → T(s(x),y) (M × F ) given by k(X) = XM ×F (s(x), y) =

(j(X), −XF (y)). Then it follows that

T(s(x),y) (M × F ) = Ts(x) M × Ty F = im (ds(x) × I) ⊕ imk.

We now observe that dq(s(x), y) vanishes on im k. Hence, dψ(x, y) = dq(s(x), y) ◦ (ds(x) × I)

is surjective from Tx (M/G) × Ty F onto Tψ(x,y) (M ×G F ). It follows that ψ is submersive at
every point of U × F. 

4 Background on Riemannian manifolds

Let M be a smooth manifold. A Riemannian metric on M is a family of positive definite
inner products gx on Tx M, for x ∈ M, such that the map g : x 7→ gx is a smooth covariant
two-tensor. Equivalently, smoothness means that for every pair of smooth vector fields X, Y ∈
X(M ) the function g(X, Y ) is smooth.

9
 A Riemannian manifold is a smooth manifold M equipped with a Riemannian metric g.
Let N be a second smooth manifold, and ϕ : N → M a smooth map. Then the pull-back
ϕ∗ (g), defined by
ϕ∗ (g)x (ξ, η) = gϕ(x) (dϕ(x)ξ, dϕ(x)η),
defines a covariant two-tensor on N. If ϕ is a local diffeomorphism, then ϕ∗ (g) is a Riemannian
metric on N.
If N is equipped with a Riemannian metric h, then the map ϕ is called an isometry if and
only if the derivative dϕ(x) : Tx N → Tϕ(x) M is an isometric linear map, for every x ∈ N. If
ϕ is a local diffeomorphism this means precisely that ϕ∗ (g) = h.
If γ : [0, 1] → M is a C 1 -curve , its length L(γ) is defined by the formula
Z 1
L(γ) := kγ 0 (t)kγ(t) dt,
0

where k · ka denotes the norm associated with ga , for a ∈ M. Given two points a, b ∈ M,
let Ca,b denote the set of C 1 -curves γ : [0, 1] → M with initial point γ(0) = a and end point
γ(1) = b. We put
d(a, b) = inf L(γ).
γ∈Ca,b

It can be shown that d defines a metric on M whose topology coincides with the original
topology of M.

Lemma 4.1 Let M be a smooth manifold. Then there exists a Riemannian metric on M.

Proof: Let n = dim M. Let {(Ui , χi )}i∈I be a locally finite covering of M by means of
coordinate charts with compact closure. Let h denote the standard Euclidean metric on Rn ,
then gi = χ∗i (h) is a Riemannian
P metric on Ui . Let {ψi } be a partition of unity subordinate to
the covering {Ui } then g = i ψi gi is a smooth covariant 2-tensor on M. It is readily verified
that g is symmetric and positive definite at every point, hence defines a Riemannian metric.

It follows from the above that every smooth manifold is metrizable.
Let (M, g) be a Riemannian manifold. Then for every a ∈ M there exists an open
neighborhood U 3 a such that for every p, q ∈ M there exists a unique curve γ with initial
point p and end point q which minimizes the length functional, i.e., d(p, q) = L(γ). This
unique curve is called a local geodesic. In local coordinates x1 , . . . , xn , it can be described by
means of a second order ordinary differential equation of the form

d2 γ k X k dγ i dγ j
+ Γij = 0.
dt2 dt dt
ij

Here the Christoffel symbols Γkij are smooth functions on the coordinate patch associated
with the given local coordinates. These symbols are explicitly expressible in terms of the
components gij of the metric in the same local coordinates.
The reader should keep in mind the example of Rn equipped with the standard Euclidean
metric. In that situation the Christoffel symbols vanish, and the local geodesics are the curves
of the form t 7→ a + tv, for a, v ∈ Rn .

10
 A geodesic on M is defined to be any C 2 (hence C ∞ -) curve γ : J → M, with J an open
interval, satisfying the above equation in local coordinates.
domain of definition cannot be increased. It follows from the existence and uniqueness
theorem for the initial value problem for ordinary differential equations of the second order,
that for every a ∈ M and every ξ ∈ Ta M there exists a unique maximal geodesic γa,ξ : Ja,ξ →
M, with Ja,ξ ⊂ R an open interval containing 0, such that
γ(0) = a, and γ 0 (0) = ξ.
As in the theory of vector fields, it can be shown that the set D of (a, ξ, t) ∈ T M × R with
t ∈ Ja,ξ is open. For a ∈ M, we define the open subset Da of Ta M to consist of all ξ with
(a, ξ, 1) ∈ D. Moreover, we define the exponential map expa : Da → M by
expa ξ = γa,ξ (1).
From this definition it can be deduced that
γa,ξ (t) = expa tξ,
for all t ∈ Ja,ξ . In particular, it follows by differentiation with respect to t at t = 0, that
d expa (0) = ITa M .
Hence, expa is a local diffeomorphism at a. Here we mention that M is said to be geodesically
complete if Ja,xi = R for all (a, ξ) ∈ T M. This implies that expa is defined on all of Ta M.
Using the material described in the above summary, we can now prove the following result.

Lemma 4.2 Let M, N be two Riemannian manifolds, and let ϕ : M → N be an isometric

local diffeomorphim. Let a ∈ M and put b = ϕ(a). Then there exists an open neighborhood U
of 0 in Ta M such that
ϕ ◦ expa = expb ◦ dϕ(a) on U.

Proof: Let U, V be open balls centered at 0 in Ta M and Tb N respectively, such that expa is
a diffeomorphism from U into M, expb is a diffeomorphism from V into N, and ϕ(expa U ) ⊂
expb (V ). Then for every ξ ∈ U, the curve t 7→ expa (tξ), [0, 1] → M is a geodesic. Since ϕ is
an isometry, it follows that c : t 7→ ϕ(expa ξ)) is a geodesic in N, with c(0) = b. Put η = c0 (0).
Then it follows that c(t) = expb tη. By the chain rule it follows that η = dϕ(a)ξ. Hence,
ϕ(expa tξ) = expb tdϕ(a)ξ. Now substitute t = 1 to finish the proof. 

Corollary 4.3 Let M, N be two connected Riemannian manifolds, and let ϕ, ψ : M → N be

two isometric local diffeomorphims. Then the following assertions are equivalent.
(a) ϕ = ψ;
(b) There exists a point a ∈ M such that ϕ(a) = ψ(a), and dϕ(a) = dψ(a).

Proof: Clearly, (a) implies (b). Conversely, assume (b). We consider the set A of points
x ∈ M with ϕ(x) = ψ(x) and dϕ(x) = dψ(x). Then A is non-empty and closed. We will
finish the proof by showing that A is open as well. Fix p ∈ A and put q = ϕ(p) = ψ(p).
By the previous result, we may fix an open ball U around 0 in Tp M such that expp is a
diffeomorphism from U into M and such that ϕ ◦ expp = expp ◦ dϕ(p) on U, and a similar
equation with ψ in place of ϕ. Since dϕ(p) = dψ(p), it follows that ϕ = ψ on expp (U ), which
is an open neighborhood of p. We now observe that expp U ⊂ A. 

11
Corollary 4.4 Let M be a connected Riemannian manifold, a ∈ M, and let I(M, a) be the
group of isometric diffeomorphisms of M onto itself, fixing the point a. Then ν : ϕ →
7 dϕ(a)
defines an injective group homomorphism from I(M, a) into O (Ta M ).

Proof: From the chain rule, it follows that ν is a group homomorphism. Injectivity follows
by application of the previous corollary. 

Remark 4.5 The group I(M ) of all isometries of M is a topological group for the compact
open topology. It can be shown, see, e.g., [Hel]2 , Ch. IV., Thm. 2.5, that I(M ) is a locally
compact group. Moreover, the subgroup I(M, a) is compact. It follows that ν maps I(M, a)
homeomorphically onto a closed subgroup of O (Ta M ). Hence, I(M, a) carries the structure
of a compact Lie group.

5 Actions of Lie groups

In this section, we assume that a smooth left action of a Lie group G on a smooth manifold
M is given. As before we consider the associated linear map g → X(M ) defined by

d 
XM (m) =  exp tXm, (m ∈ M ),
dt t=0

for X ∈ g. Here we warn the reader that X 7→ XM is an anti-homomorphism of Lie algebras

(the analogous map for a right action is a homomorphism). We note that the flow ϕ of the
vector field XM is given by
ϕt (m) = exp tX m.
Let now a ∈ M, then the natural map G → M, given by g 7→ ga factors to a bijection from
G/Ga onto the orbit Ga. The following basic result of Lie group theory is very important in
this context.

Lemma 5.1 Every subgroup H of the Lie group G which is closed in the sense of topology
has a unique structure of Lie group for which the inclusion map iH : H → G is a smooth
embedding onto a closed submanifold.

Proof: For a proof, see, e.g. [BtD]3 or my lecture notes on Lie groups. 
In the setting of the above lemma, let h := Te H be equipped with the Lie algebra structure
associated with H. Then the inclusion map iH : H → G is a group homomorphism and an
embedding of manifolds. From this it is not difficult to obtain that diH (e) : h → g is an
injective homomorphism of Lie algebras. Accordingly, we identify h with a subalgebra of g.
Furthermore, the following diagram commutes
i
H
H −→ G
↓ ↓ (5.3)
diH (e)
h −→ g.
2
[Hel]: S. Helgason, Differential Geometry and Symmetric Spaces, Academic Press, 1962
3
[BtD]: T. Bröcker, T. tom Dieck, Representations of compact Lie groups. Graduate Texts in Mathematics,
98. Springer-Verlag, New York, 1985

12
 In the setting of the above lemma, with H a closed subgroup of the Lie group G, the
action of H on G by right multiplication is readily seen to be proper and free, turning the
natural map π : G → G/H into a principal fiber bundle. From a local trivialization we see
that dπ(e) : g → TeH (G/H) is a submersion with kernel ker dπ(e) = Te (π −1 (eH)) = Te H = h.
It follows that dπ(e) factors to a linear isomorphism from g/h onto TeH (G/H). From now on
we shall use this identification without further comment.
The isotropy group Ga is closed, hence a closed Lie subgroup in the above sense. This
implies that G → G/Ga is a principal fiber bundle with structure group Ga . The induced
map G/Ga → M is readily seen to be smooth. Moreover, the following result is valid.

Lemma 5.2 The natural map j : G/Ga → M is an injective immersion with image Ga.

Proof: We consider the map J : G → M, g 7→ ga. Its tangent map at e is given by

dJ(e)X = XM (a). If X ∈ g and dJ(e)X = 0 then it follows that XM has a zero at a, hence
its integral curve t 7→ exp tXa is constant. It follows that exp tX ∈ Ga for all t. Differentiating
at t = 0 we find that X ∈ ga (the Lie algebra of Ga ).
Conversely, if X ∈ ga , then expGa tX ∈ Ga , hence exp tX ∈ Ga , by commutativity of
the diagram (5.3). This implies that exp tXa = a for all t, hence dJ(e)(X) = XM (a) = 0.
It follows that ker dJ(e) = ga . Since j ◦ π = J, it follows that dj(eGa ) ◦ dπ(eGa ) = dJ(e).
Since dπ(eGa ) factors to a linear isomorphism from g/f ga = g/ ker dJ(e) onto TeGa (G/Ga ),
it follows that dj(eGa ) is injective.
For g ∈ G let lg : G/Ga → G/Ga denote the left multiplication by g on G/Ga , and let
αg : M → M denote the left multiplication by g on M. Both maps are diffeomorphisms, and
αg ◦ j = j ◦ lg . Using this homogeneity, and the chain rule for differentiation we infer that
dj(gGa ) is injective for all g ∈ G. 

Corollary 5.3 In the setting of the above lemma, the map X 7→ XM (a) factors to a linear
isomorphism from the quotient space g/ga onto the tangent space Ta (Ga) of the orbit Ga.

Proof: See the proof of the previous result. 

In the following we assume that the Lie group G is connected, so that its orbits Ga are
connected injectively immersed submanifolds of M. The action of G on the manifold M is
said to be foliated if dim Ga is constant as a function of a ∈ M. In this case all G-orbits have
the same dimension. We will show that they are actually the leaves of a foliation on M.
To see this, we note that every G-orbit is an injectively immersed submanifold of dimension
dim G − d, where d = dim Ga . Thus, a 7→ Ea := Ta (Ga) defines a subbundle E of the tangent
bundle. It suffices to show that E is locally integrable. To see this, let a ∈ M and let v ⊂ g
be a linear subspace such that f v ⊕ ga = g. Moreover, let S be a submanifold of M such that
a ∈ S and Ta S ⊕ Ta (Ga) = Ta M. We consider the map ϕ : v × S → M, (X, z) 7→ exp X z.
Its derivative at (0, a) equals (X, ζ) 7→ X + ζ hence is a linear isomorphism. It follows that
there exist open neighborhoods O of 0 in v and V of a in S such that the restricted map
ϕ : O × V → M is a diffeomorphism onto an open subset U ⊂ M. Fix X0 ∈ O and y ∈ V
and put b = exp X0 y. The map X 7→ ϕ(X0 + X, y) maps into the G-orbit Gb = Gy. Hence,
its derivative D1 ϕ(X0 , y) maps into Eb . For dimensional reasons, it follows that D1 (ϕ(X0 , y)
has image Eb . We conclude that Dϕ maps the subbundle pr∗1 T O isomorphically onto E|U .
Therefore, E is locally integrable. 

13
 We now come to the important slice theorem for proper actions.

Theorem 5.4 Let the smooth left action of the Lie group G on M be foliated and proper.
Then for every a ∈ M there exists a smooth submanifold S of M such that:
(a) a ∈ S and S is Ga -invariant;
(b) the natural map β : G × S → M, (g, y) 7→ gy, factors to a diffeomorphism β̄ from
G ×Ga S onto an open G-invariant neighborhood of Ga in M.

Remark 5.5 Since β̄ intertwines the G-actions, it follows that β brings the action of G on
a neighborhood of the orbit Ga in the standard form G ×Ga S.

Proof: First we note that Ga is compact by properness of the action. By the method of
averaging we obtain a Ga -invariant Riemannian metric on M. We now select an open ball U
centered at 0 in Ta M such that expa : U → M is a diffeomorphism onto an open subset. Let
v be the orthocomplement of Ta (Ga) in Ta M and put V = U ∩ v. Then S = expa (V ) is a
smooth submanifold of M. Moreover, Ga acts by isometries, so if g ∈ Ga , then the tangent
map dαg (a) of αg : m 7→ gm leaves V invariant. From expa ◦ dαg (a) = αg ◦ expa we see that
the action by Ga leaves S invariant.
Since the action of Ga on G by right multiplication is proper and free, it follows that
G ×Ga S has a unique structure of smooth manifold for which the natural map q : G × S →
G ×Ga S is a submersion. It follows that the natural map β : G × S → M factors to a smooth
map β̄ : G ×Ga S → M.
Select a subspace c ⊂ g complementary to ga , i.e., c ⊕ ga = g. Then the map c → G/Ga ,
given by X 7→ exp XGa is a local diffeomorphism at 0. It follows that the map c×S → G×F S,
(X, y) 7→ q(exp X, y) is a local diffeomorphism at (0, a). It follows from the proof of the
previous lemma that the restriction of β to exp c × S is a local diffeomorphism at (e, a). We
conclude that β̄ is a local diffeomorphism at eGa × a. Choosing the radius of U sufficiently
small, we may arrange that β̄ maps an open neighborhood of eGa × S diffeomorphically
onto an open neighborhood of a in M. By homogeneity it follows that the map β̄ is a local
diffeomorphism of G ×Ga S onto an open G-invariant neighborhood of Ga in M.
By properness of the action, there exists a compact subset C ⊂ G such that gS ∩ S 6= ∅ ⇒
g ∈ C. Since Ga is compact, we may replace C by CGa to arrange that C is right Ga -invariant.
The map β̄ is injective on the image Ca := q(C × {a}) in G ×Ga S. In view of Lemma 5.6
below, we may select an open neighborhood O of Ca on which β̄ is a diffeomorphism, and by
replacing U by a smaller neighborhood, we may arrange that β̄ is injective on q(C × S).
We will now complete the proof by showing that β̄ is injective on G ×Ga S. Indeed, assume
that g1 , g2 ∈ G and y1 , y2 ∈ S and that β̄(q(g1 , y1 )) = β̄(q(g2 , y2 )). Then g2−1 g1 y1 = y2 , and
therefore g2−1 g1 ∈ C. Since β̄ is injective on q(C × S) it follows that q(g2−1 g1 , y1 ) = q(e, y2 ).
This implies that q(g1 , y1 ) = q(g2 , y2 ), and we see that β̄ is injective. 

Lemma 5.6 Let X, Y be manifolds, and f : X → Y a local diffeomorphism. Let C ⊂ X be

compact, and assume that f |C is injective. Then there exists an open neighborhood U of C
such that f |U is a diffeomorphism from U onto an open subset of Y.

Proof: There exists a decreasing sequence {Uj } of relatively compact open neighborhoods of
C whose intersection is C. Assume that f is not injective on Uj , for each j. Then there exist

14
sequences of points xj , yj ∈ Uj with xj 6= yj such that f (xj ) = f (yj ). The sequence (xj , yj ) in
M × M has a limit point (x, y). Clearly x, y ∈ C and f (x) = f (y), hence x = y. Fix an open
neighborhood O of x on which f is injective. Then xj , yj ∈ O for some j, and f (xj ) = f (yj ),
hence xj = yj , contradiction. It follows that f is injective on a suitable open neighborhood
of C. The result follows. 

15