Locally compact groups

Paul Klee, Bunte Gruppe

Linus Kramer

October 1, 2018
Contents

Preface iii

Conventions v

1 Topological Groups 1
Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Quotients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Connected Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Orbit Spaces of Compact Transformation Groups . . . . . . . . . . . . . . 9
Metrizability of Topological Groups and the Tychonoff Property . . . . . . 13

2 Around the Baire Property 21

The Open Mapping Theorem and Pettis’ Lemma . . . . . . . . . . . . . . 23
Čech Complete Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Čech Complete Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Polish Groups and the Open Mapping Theorem . . . . . . . . . . . . . . . 34
Preface
Conventions

Our conventions in group theory and topology are mostly standard. The neutral element
of a group G is denoted by e or, if the group is abelian and written additively, by 0. For
subsets X, Y ⊆ G we put

XY = {xy | x ∈ X and y ∈ Y }, X −1 = {x−1 | x ∈ X}

and for an integer k ≥ 1 we put

X ·k = {x1 x2 · · · xk | x1 , . . . , xk ∈ X}.

We call X symmetric if X = X −1 . The centralizer of X is denoted by

CenG (X) = {g ∈ G | gx = xg holds for all x ∈ X}.

The center of a group is written as Cen(G) = CenG (G). Our convention for commutators
is that
[a, b] = aba−1 b−1 .
Most of the time, we will consider left actions. Such a left action of a group G on a set
X will be written as
G × X −→ X, (g, x) 7−→ gx.
The stabilizer of a point x ∈ X will be denoted by

Gx = {g ∈ G | gx = x}.

For a subset A ⊆ X we put GA = {ga | g ∈ G and a ∈ A}.

A subset V of a topological space X is called a neighborhood of a point x ∈ X if there
is an open set U with x ∈ U ⊆ V . It is our convention is that all compact or locally
compact spaces are assumed to be Hausdorff.
We consider 0 to be a natural number. The set of natural numbers is thus

N = {0, 1, 2, 3, . . .},
and we denote the set of all positive natural numbers by

N1 = {1, 2, 3, . . .}.

The difference of two sets X, Y is written as

X − Y = {x ∈ X | x 6∈ Y }

and the symmetric difference is written as

X△Y = (X − Y ) ∪ (Y − X) = (X ∪ Y ) − (X ∩ Y ).

The power set of a set X is denoted as

P(X) = {Y | Y ⊆ X}.

We use the axiom of choice without further ado.

1 | Topological Groups

This chapter contains basic results about the point-set topology of topological groups.
Whenever we use the Hausdorff condition, this will be mentioned explicitly. However, we
do assume that locally compact spaces and compact spaces are Hausdorff.
Definition 1.1. A topological group (G, ·, T ) consists of a group (G, ·) and a topology T
on G for which the multiplication map

G × G −→ G
(g, h) 7−→ gh

and the inversion map

G −→ G
g 7−→ g −1

are continuous. We can combine this into one condition by considering the map

κ : G × G −→ G
(g, h) 7−→ g −1h.

If G is a topological group, then κ is continuous. Conversely, if κ is continuous, then the

maps g 7−→ g −1 = κ(g, e) and (g, h) 7−→ gh = κ(κ(g, e), h) are also continuous.
Suppose that G is a topological group. For every a ∈ G, the right translation map

ρa (g) = ga,

the left translation map

λa (g) = ag

and the conjugation map

γa (g) = aga−1
2 Chapter 1

are homeomorphisms of G onto itself, with inverses λa−1 , ρa−1 and γa−1 , respectively. In
particular, the homeomorphism group of G acts transitively on G. It follows that every
neighborhood W of a group element g ∈ G can be written as W = gU = V g, where
U = λg−1 (W ) and V = ρg−1 (W ) are neighborhoods of the identity. In what follows, we
will mostly write G for a topological group, without mentioning the topology T explicitly.
A neighborhood of the identity element e will be called an identity neighborhood for short.

Definition 1.2. We define a morphism f : G −→ K between topological groups G, K to

be a continuous group homomorphism.

Example 1.3. The following are examples of topological groups and morphisms.
(a) The additive and the multiplicative groups of the fields Q, R, C, and the p-adic fields
Qp , endowed with their usual field topologies, are examples of topological groups.
The exponential maps exp : R −→ R∗ and exp : C −→ C∗ are morphisms.
(b) The circle group U(1) = {z ∈ C | |z| = 1} ⊆ C∗ is another example of a topological
group. The map R −→ U(1) that maps t to exp(2πit) = cos(2πt) + i sin(2πt) is a
morphism.
(c) Every morphism f : R −→ R is of the form f (t) = rt, for a unique real r. This
follows from the fact that Q is dense in R, and that an additive homomorphism
f : Q −→ R is determined uniquely determined by the element r = f (1), since Q is
uniquely divisible.
(d) Every morphism f : U(1) −→ U(1) is of the form f (z) = z m , for a unique integer
m ∈ Z.
(e) As a vector space over Q, the group (R, +) has dimension 2ℵ0 . Hence the abelian
ℵ
group R has 22 0 additive endomorphisms, almost all of which are not continuous.
(f) Let H denote the additive group of the reals, endowed with the discrete topology.
Then H is a locally compact group and the identity map id : H −→ R is a continuous
bijective morphism whose inverse is not continuous.
(g) Let F be a field and let GLn F denote the group of invertible n × n-matrices
over F . For an n × n-matrix g, let g # denote the matrix with entries (g # )i,j =
(−1)i+j det(g ′(j, i)), where g ′ (j, i) is the (n − 1) × (n − 1)-matrix obtained by remov-
ing column i and row j from the matrix g. Then gg # = g # g = det(g)1. Hence if
1
F is a topological field, then κ(g, h) = g −1 h = det(g) g # h depends continuously on g
and h, and therefore GLn F is a topological group. In particular, the matrix groups
GLn Q, GLn R, GLn C, and GLn Qp are topological groups.
(h) Every group G, endowed either with the discrete or with the trivial nondiscrete
topology, is a topological group.

Proposition
Q 1.4. Suppose that (Gi )i∈I is a family of topological groups. Then the product
G = i∈I Gi , endowed with the product topology, is again a topological group. For each j,

[Preliminary Version - October 1, 2018]

Topological Groups 3

the projection map prj : G −→ Gj is an open morphism. If H is a topological group and

if there are morphisms fj : H −→ Gj , for every j ∈ J, then there is a unique morphism
f : H −→ G such that prj ◦f = fj holds for all j ∈ J.

Proof. We have to show that the map κ : G × G −→ G that maps (g, h) to g −1 h is contin-
uous. Let κj : Gj × Gj −→ Gj denote the corresponding maps, which are by assumption
continuous. Then we have for each j a continuous map prj ◦κ = κj ◦ (prj × prj ),

κ
G×G G
prj × prj prj
κj
Gj × Gj Gj .

By the universal property of the product topology on G, this implies that κ is continuous.
The remaining claims follow, since these maps on the one hand are continuous (and
open) as claimed by the properties of the product topology, and on the other hand are
group homomorphisms.

The following local criterion for morphisms is often useful.

Lemma 1.5. Let G, K be topological groups and let f : G −→ K be a (not necessarily

continuous) group homomorphism. Then the following are equivalent.
(i) The homomorphism f is continuous and hence a morphism of topological groups.
(ii) The homomorphism f is continuous at one point a ∈ G, i.e. for every neighborhood
W of f (a), there exists a neighborhood V of a such that f (V ) ⊆ W .

Proof. It is clear that (i) ⇒ (ii), because a continuous map is continuous at every point.
Suppose that (ii) holds and that U ⊆ K is open. If g ∈ f −1 (U), then f (a) = f (ag −1g) ∈
f (ag −1)U. Hence there exists a neighborhood V of a with f (V ) ⊆ f (ag −1 )U. Then ga−1 V
is a neighborhood of g, with f (ga−1V ) ⊆ U. Hence f −1 (U) is open.

Below we will look into separation properties for topological groups more closely. At
this point we just record the following.

Lemma 1.6. A topological group G is Hausdorff if and only if some singleton {a} ⊆ G
is closed.

Proof. Suppose that {a} ⊆ G is closed. The preimage of {a} under the continuous map
(g, h) 7−→ g −1 ha is the diagonal {(g, g) | g ∈ G} ⊆ G × G, which is therefore closed
in G × G. Thus G is Hausdorff. Conversely, every singleton set in a Hausdorff space is
closed.

[Preliminary Version - October 1, 2018]

4 Chapter 1

Subgroups
Now we study subgroups of topological groups.

Proposition 1.7. Let H be a subgroup of a topological group G. Then H is a topological

group with respect to the subspace topology. Moreover, the closure H is also a subgroup of
G. If H is normal in G, then H is also normal.

Proof. It is clear from the definition that a subgroup of a topological group is again
a topological group. Let H ⊆ G be a subgroup. The continuity of the map κ from
Definition 1.1 ensures that

κ(H × H) = κ(H × H) ⊆ κ(H × H) = H.

Thus H is a subgroup. Suppose in addition that H ✂ G is normal. For a, g ∈ G we put

γa (g) = aga−1 . Since the conjugation map γa : G −→ G is continuous, we have

γa (H) ⊆ γa (H) = H.

This shows that H is normal in G.

Lemma 1.8. Let G be a topological group and let A ⊆ G be a closed subset. Then the
normalizer of A,
NorG (A) = {g ∈ G | γg (A) = A},
is a closed subgroup.

Proof. For a ∈ A let ca (g) = gag −1. Then ca : G −→ G is continuous and hence
c−1
a (A) = {g ∈ G | gag
−1
∈ A} is closed. Therefore
\
S = {c−1 a (A) | a ∈ A} = {g ∈ G | γg (A) ⊆ A}

is a closed semigroup in G, and NorG (A) = S ∩ S −1 is closed as well.

Lemma 1.9. Let G be a Hausdorff topological group, and let X ⊆ G be any subset. Then
the centralizer
CenG (X) = {g ∈ G | [g, X] = e}
is closed. In particular, the center of G is closed.

Proof. Given x ∈ X, the map g −→ [g, x] = gxg −1x−1 is continuous. Therefore CenG (x) =
{g
T ∈ G | [g, x] = e} is closed, provided that {e} ⊆ G is closed. Then CenG (X) =
{CenG (x) | x ∈ X} is closed as well.

[Preliminary Version - October 1, 2018]

Topological Groups 5

Lemma 1.10. Let G be a Hausdorff topological group. If A ⊆ G is an abelian subgroup,

then A is an abelian subgroup.

Proof. The commutator map (g, h) −→ [g, h] is constant on A×A and hence also constant
on the closure A × A = A × A.

Lemma 1.11. Let G be a topological group and suppose that U ⊆ G is an open subset. If
X ⊆ G is any subset, then UX and XU are open subsets. In particular, the multiplication
map m : G × G −→ G, (g, h) 7−→ gh and the map κ : G × G −→ G, (g, h) 7−→ g −1 h are
open.

Proof.
S For each x ∈ X, theSsets Ux = ρx (U) and xU = λx (U) are open. Hence UX =
{Ux | x ∈ X} and XU = {xU | x ∈ X} are open as well.

Proposition 1.12. Let G be a topological group and let H ⊆ G be a subgroup.

(i) The subgroup H is open if and only if it contains a nonempty open set.
(ii) If H is open, then H is also closed.
(iii) The subgroup H is closed if and only if there exists an open set U ⊆ G such that
U ∩ H is nonempty and closed in U.

Proof. For (i), suppose that H contains the nonempty open set U. Then H = UH is open
by Lemma 1.11. Conversely, if H is open then it contains
S the nonempty open set H. For
(ii), suppose that H ⊆ G is open. Then G − H = {aH | a ∈ G − H} is also open and
therefore H is closed. For (iii), suppose that U ∩ H is nonempty and closed in the open
set U. Then U ∩ H is also closed in the smaller set U ∩ H ⊆ U. Upon replacing G by H,
we may thus assume in addition that H is dense in the ambient group G, and we have to
show that H = G. The set U − H = U − (U ∩ H) is open in U and hence open in G. On
the other hand, H is dense in G. Therefore U − H = ∅ and thus U ⊆ H. By (i) and (ii),
H is closed in G, whence H = G. Conversely, if H is closed, then H is closed in the open
set G.

Corollary 1.13. Let G be a topological group and let V ⊆ G be a neighborhood of some

element g ∈ G. Then V generates an open subgroup of G.

Corollary 1.14. Suppose that G is a Hausdorff topological group and that H ⊆ G is a

subgroup. If H is locally compact in the subspace topology, then H is closed. In particular,
every discrete subgroup of G is closed.

Proof. Let C ⊆ H be a compact set which is an identity neighborhood in the topological

group H. Then there exists an open identity neighborhood U in G such that U ∩ H ⊆ C.
Since C is compact, C is closed in G and hence U ∩ H = U ∩ C is closed in U and
nonempty. Now we may apply Proposition 1.12(iii).

[Preliminary Version - October 1, 2018]

6 Chapter 1

The product AB of closed subsets A, B in a topological group G need not be closed.

An example
√ is the additive group of the reals (R, +), with the closed subgroups A = Z
and B = 2Z. Then A + B is a countable dense subgroup of R which is not closed.
However, we have the following.
Lemma 1.15. Let G be a Hausdorff topological group, and let A, B ⊆ G be closed subsets.
If either A or B is compact, then AB ⊆ G is closed.
The proof uses Wallace’s Lemma.
Lemma 1.16 (Wallace). Let X1 , . . . Xk be Hausdorff spaces containing compact sets
Aj ⊆ Xj , for j = 1, . . . , k. If W ⊆ X1 × · · · × Xk is an open set containing A1 × · · · × Ak ,
then there exist open sets Uj with Aj ⊆ Uj ⊆ Xj , for j = 1, . . . , k, such that

A1 × · · · × Ak ⊆ U1 × · · · × Uk ⊆ W.

Proof. There is nothing to show for k = 1. Suppose that k = 2. We put A = A1 and

B = A2 and we fix a ∈ A. For every point b ∈ B, we choose an open neighborhood
Ub × Vb of (a, b) such that Ub × Vb ⊆ W . Since {a} × B is compact, finitely many such
neighborhoods Ub1 × Vb1 , . . . , Ubm × Vbm cover {a} × B. We put Ua = Ub1 ∩ · · · ∩ Ubm and
Va = Vb1 ∪ · · · ∪ Vbm . Then {a} × B ⊆ Ua × Va ⊆ W . Now we let a ∈ A vary. Since
A is compact, finitely many such strips Ua1 × Va1 , . . . , Uan × Van cover A × B. We put
U = Ua1 ∪ · · · ∪ Uan and V = Va1 ∩ · · · ∩ Van . Then A × B ⊆ U × V ⊆ W and the claim
is proved for k = 2.
For k ≥ 3 we apply the previous argument to A = A1 and B = A2 × · · · × Ak , and
we obtain open sets U ⊆ X1 and V ⊆ X2 × · · · × Xk with A × B ⊆ U × V ⊆ W . By
induction, we find now open sets U2 , . . . Uk such that A2 × · · · × Ak ⊆ U2 × · · · Uk ⊆ V .
Therefore A1 × · · · × Ak ⊆ U × U2 × · · · Uk ⊆ W .
Proof of Lemma 1.15. Suppose that A is compact and B is closed, and that g ∈ G − AB.
We have to show that G−AB contains a neighborhood of g. By assumption, A−1 g∩B = ∅.
If we put κ(g, h) = g −1 h, then κ(A × {g}) ⊆ G − B. By Wallace’s Lemma 1.16 there
exists an open neighborhood V of g such that κ(A × V ) ⊆ G − B, i.e. A−1 V ∩ B = ∅.
Hence V ∩ AB = ∅ and the claim follows. The case where B is compact and A is closed
follows by taking inverses.

Quotients
Suppose that H is a subgroup of a topological group G. We endow the set G/H of left
cosets with the quotient topology with respect to the natural map

p : G −→ G/H, g 7−→ gH.

[Preliminary Version - October 1, 2018]

Topological Groups 7

Thus a subset of G/H is open if and only if its preimage is open. The next result is
elementary, but important.

Proposition 1.17. Let G be a topological group and let H be a subgroup. Then the
quotient map
p : G −→ G/H
is open. The quotient G/H is Hausdorff if and only if H is closed in G, and it is discrete
if and only if H is open in G.

Proof. Suppose that U ⊆ G is an open set. Then p−1 (p(U)) = UH is open by Lemma 1.11,
hence p(U) is open by the definition of the quotient topology.
If G/H is Hausdorff, then {H} ⊆ G/H is closed, hence H = p−1 ({H}) ⊆ G is closed as
well. Conversely, suppose that H ⊆ G is closed. The map p × p : G × G −→ G/H × G/H
is open, because p is open and because a cartesian product of two open maps is again
open. The open set W = {(x, y) ∈ G × G | x−1 y ∈ G − H} maps under p × p onto the
complement of the diagonal in G/H × G/H. Hence the diagonal {(gH, gH) | g ∈ G} is
closed in G/H × G/H, and therefore G/H is Hausdorff.
If H is open, then every coset gH is open and hence G/H is discrete. Conversely,
if G/H is discrete, then H, being the preimage of the open singleton {H} ⊆ G/H, is
open.
If G is Hausdorff and if K ⊆ G is compact then more can be said, see Corollary 1.28
below.

Corollary 1.18. Suppose that G is a Hausdorff topological group which is compact and
that H ⊆ G is a closed subgroup. Then H is open if and only if H has finite index in G.

Proof. A discrete topological space is compact if and only if it is finite.

Corresponding remarks apply to the set H\G of right cosets, by taking inverses.

Proposition 1.19. Let G be a topological group. If N ✂ G is a normal subgroup, then the

factor group G/N is a topological group with respect to the quotient topology on G/N. The
quotient map p : G −→ G/N is an open morphism. The factor group G/N is Hausdorff
if and only if N is closed. In particular, G/N is a Hausdorff topological group.

Proof. We put κ̄(gN, hN) = g −1 hN and p(g) = gN. Then the diagram
κ
G×G G
p×p p

κ̄
G/N × G/N G/N

[Preliminary Version - October 1, 2018]

8 Chapter 1

commutes, and p ◦ κ is continuous. Since p is open, p × p is also open and hence a quotient
map. It follows from the universal property of quotient maps that κ̄ is continuous, and
therefore G/N is a topological group. The remaining claims follow from Propositions 1.7
and 1.17.

The next result is the Homomorphism Theorem for topological groups.

Lemma 1.20. Let f : G −→ K be a morphism of topological groups, and put N = ker(f ).

Then f factors through the open morphism p : G −→ G/N via a unique morphism f¯,

f
G K.
p
f¯
G/N

If f is open, then f¯ is also open.

Proof. The group homomorphism f¯ exists uniquely by the Homomorphism Theorem for
groups. Since p is a quotient map, f¯ is continuous and thus a morphism of topological
groups. If f is open and if W ⊆ G/N is an open set, then f (p−1 (W )) = f¯(W ) is open as
well.

Corollary 1.21. Suppose that G, K are topological groups and that K is Hausdorff. If
f : G −→ K is a morphism of topological groups, then f factors through the open mor-
phism p : G −→ G/{e},
f
G K.
p
f¯

G/{e}

Connected Components
Definition 1.22. Let x be a point in a topological space X. The connected component
of x is the union of all connected subsets of X containing x. This union is closed and
connected. We call a topological space X totally disconnected if the only connected
nonempty subsets of X are the singletons.
The connected component the identity element of a topological group G will be denoted
by G◦ , and we call G◦ the identity component of G. Since the homeomorphism group of
G acts transitively on G, the group G is totally disconnected if and only if G◦ = {e}. We
note that a totally disconnected group is automatically Hausdorff.

[Preliminary Version - October 1, 2018]

Topological Groups 9

Proposition 1.23. Let G be a topological group. Then the identity component G◦ is a

closed normal subgroup, and G/G◦ is a totally disconnected Hausdorff topological group.
Proof. We put κ(g, h) = g −1 h and we note that a continuous image of a connected set is
connected. Since G◦ ×G◦ is connected and contains the identity element, κ(G◦ ×G◦ ) ⊆ G◦ .
This shows that G◦ is a subgroup. By the remark above, G◦ is closed. For every a ∈ G,
the set γa (G◦ ) = aG◦ a−1 is connected and contains the identity, whence aG◦ a−1 ⊆ G◦ .
This shows that G◦ is a closed normal subgroup.
It remains to show that G/G◦ is totally disconnected. We consider the canonical
morphism p : G −→ G/G◦ and we put H = (G/G◦ )◦ and N = p−1 (H). Then N is a
closed normal subgroup of G containing G◦ . We claim that N = G◦ . If we have proved
this claim, then H = {G◦ } and thus G/G◦ is totally disconnected. The restriction-
corestriction map p|HN : N −→ H is open, hence H carries the quotient topology with
H
respect to p|N : N −→ H. Suppose that V ⊆ N is closed and open in N and contains the
identity. Since G◦ is connected and contains e, we have vG◦ ⊆ V for all v ∈ V . Hence
V = p−1 (p(V )), and therefore p(V ) is closed and open in H. But H is connected, whence
H = p(V ) and thus V = N. It follows that N is connected, whence N = G◦ .
Corollary 1.24. Let f : G −→ K be a morphism of topological groups. If K is totally
disconnected, then f factors through the open morphism p : G −→ G/G◦ ,
f
G K.
p
f¯
◦
G/G

Proof. Since f (G◦ ) ⊆ K is connected, G◦ is contained in the kernel of f .

Orbit Spaces of Compact Transformation Groups

We recall that compact spaces are by definition Hausdorff.
Definition 1.25. A compact transformation group consists of a continuous action

K × X −→ X

of a compact group K on a Hausdorff space X. We denote the K-orbit of x ∈ X by

K(x) = {gx | g ∈ K} ⊆ X.

We endow the orbit space

K\X = {K(x) | x ∈ X}

[Preliminary Version - October 1, 2018]

10 Chapter 1

with the quotient topology with respect to the map

q : X −→ K\X

that maps x to its orbit K(x).

Example 1.26. Let G be a Hausdorff topological group and suppose that K ⊆ G is a

compact subgroup.
(i) The group K acts by left multiplication on G, and the orbit space is the set of right
cosets K\G = {Kg | g ∈ g}.
(ii) If H ⊆ G is a closed subgroup, then K acts by left multiplication on G/H. The
orbit space K\(G/H) can be identified with the set of double cosets

K\G/H = {KgH | g ∈ G}

via the map K(gH) = {kgH | k ∈ K} 7−→ KgH. The topology on K\G/H
can be either viewed as the quotient topology with respect to the canonical map
q : G/H −→ K\G/H or as the quotient topology with respect to the canonical map
r : G −→ K\G/H. This makes no difference, because a composition of quotient
maps is again a quotient map.
(iii) The group K also acts on G from the left by conjugation. The orbit space for this
action is the space of K-conjugacy classes.

We recall that a continuous map between Hausdorff spaces is called proper if the
preimage of every compact set is compact.

Proposition 1.27. Suppose that K × X −→ X is a compact transformation group. Then

we have the following.
(i) The orbit space K\X is Hausdorff.
(ii) The map q : X −→ K\X is open, closed and proper.
In particular, K\X is compact (locally compact) if and only if X is compact (locally
compact).
−1
Proof. WeS first show that q is open and closed. If U ⊆ X is open, then q (q(U)) =
KU = {gU | g ∈ K} is open, hence q is an open map. Suppose that A ⊆ X is
closed. We have to show that q −1 (q(A)) = KA ⊆ X is closed. For z ∈ X − KA we have
K(z) ∩ KA = ∅. By Wallace’s Lemma 1.16 there exits an open neighborhood U of z such
that KU ∩ KA = ∅. Thus X − KA is open, and hence KA is closed.
Next we show that K\X is Hausdorff. Suppose that x, y ∈ X are points with q(x) 6=
q(y). Since the orbits K(x), K(y) ⊆ X are compact and disjoint, there exist disjoint open
sets U, V ⊆ X with K(x) ⊆ U and K(y) ⊆ V . In particular, K(y) ∩ U = ∅. Now we use

[Preliminary Version - October 1, 2018]

Topological Groups 11

that q is open and closed. The point q(x) is in the open set q(U), and the point q(y) is
not in the closed set q(U ) = q(U). Therefore K\X is Hausdorff.
Now we show that q is proper. Suppose that B ⊆ K\X is compact. Then A =
−1
q (B) is a closed set which is a union of K-orbits. Let U be an open S covering of
A. For every a ∈ A we find a finite subset Ua ⊆ U such that K(a) ⊆ Ua , because
K(a) ⊆ A is compact. S By Wallace’s S Lemma 1.16 we find an open neighborhood Va of a
such that KVa ⊆ Ua . Since B ⊆ {q(Va ) | a ∈ A}, there exists finitely many points
a1 , . . . , an ∈ A such that B ⊆ q(Va1 ) ∪ · · · ∪ q(Van ). It follows that
[ [
A ⊆ KVa1 ∪ · · · ∪ KVan ⊆ Ua1 ∪ · · · ∪ Uan .

Hence A is compact.
Since q is proper, X is compact if and only if K\X is compact. If X is locally compact
and if U ⊆ X is an open neighborhood of x ∈ X with compact closure, then q(U) is a
compact neighborhood of q(x) because q is open. Hence K\X is locally compact. If
K\X is locally compact and if V ⊆ K\X is an open neighborhood of q(x) with compact
closure, then q −1 (V ) is a compact neighborhood of x because q is proper.

Corollary 1.28. Suppose that G is a Hausdorff topological group and that K ⊆ G is a

compact subgroup. Then the map q : G −→ K\G, g 7−→ Kg is continuous, open, closed,
and proper.

The same result holds of course for the map p : G −→ G/K.

Corollary 1.29. Let G be a Hausdorff topological group, with a closed subgroup H. If

two of the three spaces G, H, G/H are compact, then all three spaces are compact.

Lemma 1.30. Suppose that K × X −→ X is a compact transformation group and that

x ∈ X is a fixed point of K. Then every neighborhood V of x contains a K-invariant
neighborhood U of x.

Proof. Suppose that x is a fixed point of the K-action and that V is a neighborhood
of x. By Wallace’s Lemma 1.16 there exists an open neighborhood U of x such that
KU ⊆ V .

Corollary 1.31. Suppose that G is a Hausdorff topological group and that K ⊆ G is a

compact subgroup. Then there are arbitrarily small identity neighborhoods V ⊆ G which
are invariant under conjugation by elements of K.

Proof. The compact group K acts as a compact transformation group on G via conjuga-
tion, and e is a fixed point for this action.

[Preliminary Version - October 1, 2018]

12 Chapter 1

Recall that a Hausdorff space X is called a Tychonoff space, or completely regular, or

a T3 1 -space if for every x ∈ X and every neighborhood V of x there exists a continuous
2
map
ϕ : X −→ [0, 1]

with ϕ(x) = 0 and ϕ(y) = 1 for all y ∈ X − V .

Proposition 1.32. Suppose that K × X −→ X is a compact transformation group. If X

is a Tychonoff space, then K\X is a Tychonoff space as well.

We first need a little observation.

Lemma 1.33. Suppose that X is a set, that u, v : X −→ R are two functions which
are bounded from below, and that ε > 0. If |u(x) − v(x)| ≤ ε holds for all x ∈ X, then
| inf u(X) − inf v(X)| ≤ ε.

Proof. For every n ≥ 1 there exists a point xn ∈ X such that u(xn ) − inf u(X) ≤ 2−n .
Then v(xn ) ≤ inf u(X) + ε + 2−n and thus inf v(X) ≤ inf u(X) + ε.

Proof of Proposition 1.32. Suppose that X is Tychonoff, that x ∈ X, and that V ⊆ K\X
is a neighborhood of q(x). There exists a neighborhood U of x with q(U) ⊆ V . Since
X is a Tychonoff space, there exists a continuous map ψ : X −→ [0, 1] with ψ(x) = 0
and ψ(y) = 1 for all y ∈ X − U. We put ψ̃(z) = min ψ(K(z)). Then ψ̃ is constant on
the K-orbits and hence ψ̃ descends to a map ϕ : K\X −→ [0, 1] via ϕ(K(z)) = ψ̃(z) =
min ψ(K(z)). We have ϕ(q(x)) = 0 and ϕ(q(z)) = 1 if q(z) 6∈ V . It remains to show that
ϕ is continuous. For this, it suffices to show that ψ̃ is continuous because K\X carries
the quotient topology with respect to the map q,

ψ̃
X [0, 1]
q
ϕ

K\X.

Given z ∈ X, we may consider the continuous map h : K × X −→ [0, 1] with h(g, y) =

|ψ(gz)−ψ(gy)|. Since h(g, z) = 0 holds for all g ∈ K, there exists by Wallace’s Lemma 1.16
for every ε > 0 a neighborhood Wε of z such that h(g, y) ≤ ε holds for all (g, y) ∈ K ×Wε .
It follows from Lemma 1.33 that | inf ψ(K(z)) − inf ψ(K(y))| ≤ ε holds for all y ∈ Wε ,
and this shows that ψ̃ is continuous at z.

[Preliminary Version - October 1, 2018]

Topological Groups 13

Metrizability of Topological Groups and the Tychonoff Property

In this section we prove several important results about the metrizability and about
separation properties for topological groups and their coset spaces. We first review length
functions and invariant pseudometrics on groups.

Definition 1.34. A pseudometric d on a set X is a map d : X × X −→ R satisfying

d(x, x) = 0, 0 ≤ d(x, y) = d(y, x), d(x, z) ≤ d(x, y) + d(y, z),

for all x, y, z ∈ X. If d(x, y) = 0 implies that x = y, then d is called a metric.

A metric or pseudometric d on a group G is called left invariant if the left translation
map λa : G −→ G is an isometry of the metric or pseudometric space (G, d), for every
a ∈ G. In other words, we require for a left invariant metric that d(x, y) = d(ax, ay) holds
for all a, x, y ∈ G. Similarly, a pseudometric d is called right invariant if d(xa, ya) =
d(x, y) holds for all x, y, a ∈ G. If d is a left invariant pseudometric, then d′ (x, y) =
d(x−1 , y −1 ) is a right invariant pseudometric, and vice versa.

Every pseudometric d on a set X induces a topology Td on X, such that U ⊆ X is in Td

if and only if for every x ∈ U there exists some ε > 0 such that {y ∈ X | d(x, y) ≤ ε} ⊆ U.
This topology Td is Hausdorff if and only if the pseudometric d is a metric. The following
facts are elementary.

Lemma 1.35. Suppose that d, d′ are pseudometrics on a set X, and that d ≤ d′ .

(i) The metric d is continuous with respect to Td′ and thus Td ⊆ Td′ .
(ii) If d′ is continuous with respect to Td , then Td = Td′ .
(iii) If T is a topology on X, then T = Td holds if and only if d is continuous with respect
to T and T ⊆ Td .

Definition 1.36. A length function ℓ on a group is a map

ℓ : G −→ R≥0

with the properties

ℓ(e) = 0, ℓ(g) = ℓ(g −1), ℓ(gh) ≤ ℓ(g) + ℓ(h),

for all g, h ∈ G. Then

|ℓ(gh) − ℓ(g)|, |ℓ(hg) − ℓ(g)| ≤ ℓ(h)
holds for all g, h ∈ G. Moreover, the set

K = {g ∈ G | ℓ(g) = 0}

[Preliminary Version - October 1, 2018]

14 Chapter 1

is a subgroup of G. For k ∈ K and g ∈ G we have ℓ(kg) ≤ ℓ(g) = ℓ(k −1 kg) ≤ ℓ(kg) and
therefore
ℓ(kg) = ℓ(g)
holds for all k ∈ H and g ∈ G. In particular, ℓ(γk (g)) = ℓ(g) holds for all k ∈ K and
g ∈ G. If ℓ is a length function, then
d(g, h) = ℓ(g −1h)
is a left invariant pseudometric on G and d′ (g, h) = ℓ(gh−1) is a right invariant pseudo-
metric on G. Conversely, if d is a left or right invariant pseudometric on a group G, then
ℓ(g) = d(e, g) is a length function.
Proposition 1.37. Suppose that a left invariant metric d metrizes the topological group
G. Then the metric
dc (x, y) = d(x, y) + d(x−1 , y −1)
also metrizes G. Moreover, the metric completion G b of (G, dc ) is a topological group that
contains G as a dense subgroup.
Proof. It is clear that dc is a continuous metric on G. Since d ≤ dc , every d-open set is also
dc -open, and thus dc metrizes G. Let (G, b dc ) denote the metric completion of (G, dc ). We
wish to extend the map κ : G × G −→ G that maps (x, y) to x−1 y to a map G b ×G b 7−→ G.
b
The problem is that κ need not be uniformly continuous with respect to dc . We first prove
an auxiliary result.
Claim. Given x, y ∈ G b and ε > 0, there exists δ > 0 such that dc (x−1 −1
1 y1 , x2 y2 ) ≤ ε holds
for all x1 , x2 , y1, y2 ∈ G with dc (x1 , x), dc (x2 , x), dc (y1 , y), dc(y2 , y) ≤ δ.
ε
We choose x0 , y0 ∈ G such that dc (x0 , x), dc (y0 , y) ≤ 10 . Then we choose δ > 0 with
ε
δ ≤ 10 in such a way that
ε
d(zx0 , x0 ), d(zy0 , y0) ≤
10
holds for all z ∈ G with dc (z, e) ≤ 2δ. We have
d(x1 x−1 −1 −1
2 , e) = d(x1 , x2 ) ≤ dc (x1 , x2 ) ≤ dc (x1 , x) + dc (x, x2 ) ≤ 2δ

and therefore
d(x−1 −1 −1
1 y1 , x2 y2 ) = d(x2 x1 y1 , y2 )
≤ d(x2 x−11 y1 , y0 ) + d(y0 , y2 )
= d(y1 , x1 x−1
2 y0 ) + d(y0 , y2 )
≤ d(y1, y0 ) + d(y0 , x1 x−1
2 y0 ) + d(y0 , y2 )
≤ d(y1, y) + d(y, y0) + d(y0, x1 x−1 2 y0 ) + d(y0 , y) + d(y, y2 )
ε ε ε ε
≤ δ + 10 + 10 + 10 + δ ≤ 2 .

[Preliminary Version - October 1, 2018]

Topological Groups 15

In the same way we have d(y1−1x1 , y2−1x2 ) ≤ 2ε and thus dc (x−1 −1

1 y1 , x2 y2 ) ≤ ε and the claim
follows.
We view the map κ as a subset of G×G×G ⊆ G× b G×b G,b and we let κ̂ denote its closure
b b b
in G× G× G. If (xn )n≥1 and (yn )n≥1 are sequences in G converging to points x, y ∈ G, b then
the claim shows that (x−1 n yn )n≥1 is a Cauchy sequence in G. Hence limn (xn , yn , xn yn )
−1

exits in G b×G b × G,
b and hence (pr1 × pr2 )(κ̂) = G b × G.
b
We claim that κ̂ is a continuous map. Suppose that (x, y, z) ∈ κ̂. Given ε > 0 there
exists by the claim δ > 0 such that for all x′ , y ′ ∈ G with dc (x′ , x), dc (y ′ , y) ≤ δ we have
dc (κ(x′ , y ′), z) ≤ ε. Since κ̂ is the closure of κ, it follows that dc (z ′ , z) ≤ ε holds for all
(x′ , y ′, z ′ ) ∈ κ̂ with dc (x′ , x), dc (y ′ , y) ≤ δ. Hence κ̂ is a continuous map at (x, y).
Since κ̂ is continuous, so are the maps i(x) = κ̂(x, e) and m(x, y) = κ̂(i(x), y). Now
m(m(x, y), z) = m(x, m(y, z)) holds on the dense subset G × G × G, and thus everywhere.
Hence m is an associative multiplication. Similarly, m(x, e) = m(e, x) = x holds on the
dense subset G and hence everywhere, and likewise m(i(x), x) = e = m(x, i(x) holds on
G and hence everywhere. Therefore G b is a topological group which contains G as a dense
subgroup.

We need also the following result about metrics on quotients.

Lemma 1.38. Let ℓ be a length function on a group G and let H ⊆ G be a subgroup.

Then d(x, y) = ℓ(xy −1 ) is a right invariant pseudometric on G and

dH (xH, yH) = inf ℓ(xHy −1 ) = inf d(x, yH) = inf{d(xh1 , yh2 ) | h1 , h2 ∈ H}

is a pseudometric on G/H. If G is a topological group and if ℓ is continuous, then dH is

continuous on G/H.

Proof. It is clear that d is a right invariant pseudometric and that dH is symmetric and
nonnegative. For x, y, z ∈ G and h, h′ ∈ H we have

dH (xH, zH) ≤ d(x, zh′ ) ≤ d(x, yh) + d(yh, zh′ ) = d(x, yh) + d(y, zh′ h−1 ),

whence
dH (xH, zH) ≤ dH (xH, yH) + dH (yH, zH).
Thus dH is a pseudometric on G/H.
Suppose that G is a topological group and that ℓ is continuous. Then d is continuous
and the map d′ (x, y) = dH (xH, yH) is a pseudometric on G. Since 0 ≤ d′ ≤ d, the
pseudometric d′ is also continuous. The map p : G −→ G/H is open and therefore the
map p × p : G × G −→ G/H × G/H is also open and hence a quotient map. Therefore
dH is continuous on G/H × G/H.

[Preliminary Version - October 1, 2018]

16 Chapter 1

Theorem 1.39 (Birkhoff–Kakutani). Let G be a Hausdorff topological group. Then

the following are equivalent.
(i) The topology on G is metrizable by a left invariant metric.
(ii) The topology on G is metrizable by a right invariant metric.
(iii) The topology on G is metrizable.
(iv) Some nonempty open subset U ⊆ G is metrizable.
(v) The identity element e ∈ G has a countable neighborhood basis.
If a group Γ acts as a group of automorphisms on the group G and if e has a countable
neighborhood basis consisting of Γ-invariant sets, then the metric d can be chosen to be
left invariant and in addition Γ-invariant, i.e.
d(gx, gy) = d(x, y) = d(γ(x), γ(y))
holds for all g, x, y ∈ G and γ ∈ Γ.
Corollary 1.40. If a topological group G is metrizable and if K ⊆ G is a compact
subgroup, then there exists a left invariant metric d on G which is in addition K-invariant
from the right, i.e. d(x, y) = d(gxk, gyk) holds for all k ∈ K and x, y, g ∈ G. In particular,
every metrizable compact group admits a complete metric which is left and right invariant.
Proof. If G is metrizable, then e has by Corollary 1.31 a countable neighborhood basis
consisting of K-invariant neighborhoods. Hence the metric d on G can be chosen to be left
invariant and invariant under the conjugation action of K. In a metrizable compact space
every metric is complete because every Cauchy sequence has a convergent subsequence.

The Birkhoff–Kakutani Theorem makes no claims about metric completeness. In-

deed, there exist completely metrizable topological groups which admit no left invariant
complete metric. The proof of Theorem 1.39 relies on the following technical lemma.
Lemma 1.41. Let G be a topological group. Suppose that (Kn )n∈Z is a family of symmetric
identity
S neighborhoods with the property that Kn·3 ⊆ Kn+1 holds for all n ∈ Z, and with
h n∈Z Kn i = G. Then there exists a continuous length function ℓ : G −→ R≥0 such that
{g ∈ G | ℓ(g) < 2n } ⊆ Kn ⊆ {g ∈ G | ℓ(g) ≤ 2n }.
holds for all n ∈ Z. In particular,
\
Kn = {g ∈ G | ℓ(g) = 0}.
n∈Z

If a group Γ acts as a group of automorphisms on the group G such that all sets Kn are
Γ-invariant, then the length function ℓ can be chosen to be invariant under the Γ-action
on G.

[Preliminary Version - October 1, 2018]

Topological Groups 17

Proof. First of all we note that for every g ∈ G there exist integers n1 , . . . , nk with
g ∈ Kn1 · · · Knk , because the union of the Kn generates G and because every Kn is
symmetric. For g ∈ G we put

ℓ(g) = inf{t ≥ 0 | there is some k ≥ 1 and n1 , . . . , nk ∈ Z

with t = 2n1 + · · · + 2nk and g ∈ Kn1 Kn2 · · · Knk }.

If g ∈ Km1 · · · Kmr and h ∈ Kn1 · · · Kns , then gh ∈ Km1 · · · Kmr Kn1 · · · Kns . It follows
that ℓ satisfies the triangle inequality. Since each Kn is symmetric, we have ℓ(g) = ℓ(g −1 )
for all g ∈ G. Finally, ℓ(e) = 0 since e ∈ Kn holds for every n ∈ Z. This shows that ℓ is
a length function. Moreover, ℓ(g) ≤ 2n if g ∈ Kn . If all sets Kn are Γ-invariant, then ℓ is
also Γ-invariant.
Next we show the continuity of ℓ. Let g ∈ G be any element, and let ε > 0. We
choose n ∈ Z in such a way that 2n ≤ ε. Since ℓ(a) ≤ 2n holds for all a ∈ Kn , we have
|ℓ(g) − ℓ(h)| ≤ 2n ≤ ε for all h ∈ gKn . Hence ℓ is continuous.
Suppose that ℓ(g) < 2n . Then there exists k ≥ 1 and numbers n1 , . . . , nk ∈ Z with
g ∈ Kn1 · · · Knk and with 2n1 + · · · + 2nk < 2n .
Claim. Suppose that 2n1 + · · · + 2nk < 2n . Then Kn1 · · · Knk ⊆ Kn .
Proof of the claim. We note that nj < n holds for j = 1, . . . , k. Hence Kn1 · · · Knk ⊆
Kn−1 · · · Kn−1 . This proves the claim for k = 1, 2, 3. For k ≥ 4 we proceed by induction
on k. Suppose that k ≥ 4. If 2n1 + · · · + 2nk < 2n−1 , then Kn1 · · · Knk−1 ⊆ Kn−1 by the
induction hypothesis, and Knk ⊆ Kn−1 , whence Kn1 · · · Knk ⊆ Kn−1 Kn−1 ⊆ Kn . There
remains the case where 2n−1 ≤ 2n1 + · · · + 2nk < 2n . We choose the smallest r ∈ {1, . . . , k}
with 2n−1 ≤ 2n1 + · · · + 2nr . Then 2n1 + · · · + 2nr−1 < 2n−1 and 2nr+1 + · · · + 2nk < 2n−1.
By the induction hypotheses Kn1 · · · Knr−1 ⊆ Kn−1 and Knr+1 · · · Knk ⊆ Kn−1 . Thus
Kn1 · · · Knk ⊆ Kn−1 Knr Kn−1 ⊆ Kn−1 Kn−1 Kn−1 ⊆ Kn . Note that this is true also for the
extremal cases r = 1 and r = k. Hence the claim is true.
n n
T It follows that {g ∈ G | ℓ(g) < 2 } ⊆ Kn ⊆ {g ∈ G | ℓ(g) ≤ 2 }, and in particular
n∈Z Kn = {g ∈ G | ℓ(g) = 0}.

Proof of Theorem 1.39. It is clear that (i) ⇔ (ii) ⇒ (iii) ⇒ (iv). If a nonempty open
set U ⊆ G is metrizable, then every left translate of U is metrizable and hence some
identity neighborhood is metrizable. In a metrizable space, every point has a countable
neighborhood basis. Therefore we have (iv) ⇒ (v).
It remains to show that (v) ⇒ (i). Let {Vn | n ≥ 1} be a countable neighborhood basis
of e. We put Wn = Vn ∩ Vn−1 . If the Vn are all Γ-invariant, then the Wn are Γ-invariant
as well. We now put Kn = G for n ≥ 0, and K−1 = W1 . We then proceed inductively as
follows. Given K−n , we choose m ≥ 1 in such a way that Wm·3 ⊆ K−n ∩ Vn+1 holds. This
is possible because the Wm form a neighborhood basis of e. We put K−(n+1) = Wm .

[Preliminary Version - October 1, 2018]

18 Chapter 1

Let ℓ denote a continuous length function as in Lemma 1.41 for the family (Kn )n∈Z .
If the Vn are Γ-invariant, then the Kn are also Γ-invariant and hence we may assume
in this case that ℓ is Γ-invariant. Then d(g, h) = ℓ(g −1 h) is a continuous left invariant
pseudometric on G. Let U ⊆ G be an open set and suppose that g ∈ U. There exists
an integer n ≥ 1 such that gVn ⊆ U. It follows that the set {h ∈ G | d(g, h) < 2−n } is
contained in gK−n ⊆ gVn ⊆ U. Therefore the pseudometric d metrizes the topology of G.
Since G is Hausdorff, d is a metric.
The following is another important consequence of Lemma 1.41.
Theorem 1.42. Suppose that G is a topological group and that H ⊆ G is a closed
subgroup. Then G/H is a Tychonoff space. In particular, every Hausdorff topological
group is a Tychonoff space.
Proof. If H ⊆ G is a closed subgroup, then G/H is Hausdorff by Proposition 1.17. Let
p : G −→ G/H denote the canonical projection and suppose that W ⊆ G/H is a neigh-
borhood of the point x = aH. Since p is continuous, there exists a symmetric identity
neighborhood K0 ⊆ G with p(K0 a) ⊆ W . We put Kn = G for all n ≥ 1, and we
choose inductively symmetric identity neighborhoods Kn for n = −1, −2, −3, . . . such
that Kn·3 ⊆ Kn+1 . Let ℓ denote a continuous length function for the family (Kn )n∈Z as in
Lemma 1.41. Then {g ∈ G | ℓ(g) < 20 } ⊆ K0 . Let dH denote the corresponding continu-
ous pseudometric on G/H, as in Lemma 1.38. Then ϕ(gH) = max{dH (aH, gH), 1} is a
continuous function on G/H, with ϕ(aH) = 0. If gH 6∈ p(K0 a), then K0 a ∩ gH = ∅ and
therefore dH (aH, gH) ≥ 1, whence ϕ(gH) = 1.
Proposition 1.43. Suppose that G is a metrizable topological group and that H is a
closed subgroup. Then G/H is metrizable.
Proof. Since G is metrizable, there exists by Theorem 1.39 a right invariant metric d on
G that metrizes G. From Lemma 1.38 we see that dH is a continuous pseudometric on
G/H. Suppose that W ⊆ G/H is an open set containing gH. Since p is continuous, there
exists ε > 0 such that the set {x ∈ G | d(g, x) < ε} is mapped by p : G −→ G/H into
W . Then {xH | dH (gH, xH) < ε} ⊆ W , which shows that dH determines the topology
on G/H. Since G/H is Hausdorff, dH is a metric.
We extend the previous result to certain double coset spaces. This will become relevant
in the next chapter.
Lemma 1.44. Let G be a Hausdorff topological group and let ℓ be a continuous length
function on G, with its associated right invariant pseudometric d(x, y) = ℓ(xy −1). Suppose
that H ⊆ G is a closed subgroup. Suppose also that K = {g ∈ G | ℓ(g) = 0} is compact
¯
and that the metric d(Kx, Ky) = d(x, y) metrizes the quotient topology on K\G. Then
¯
the pseudometric dH (KxH, KyH) = dH (xH, yH) metrizes the topology on K\G/H.

[Preliminary Version - October 1, 2018]

Topological Groups 19

Proof. We consider the diagram of quotient maps

G
q p

K\G r G/H

p̂ q̂
K\G/H.

Since G/H is Hausdorff and K is compact, K\G/H is Hausdorff by Proposition 1.27.

The pseudometric dH (xH, yH) = inf ℓ(xHy −1) is continuous on G/H. For k ∈ K and
x ∈ G we have ℓ(kx) = ℓ(x) and therefore

dH (k1 xH, k2 xH) = inf ℓ(k1 xHy −1k2−1 ) = inf ℓ(xHy −1) = dH (xH, yH)

holds for all k1 , k2 ∈ K and all x, y ∈ G. Hence

d¯H (KxH, KyH) = dH (xH, yH)

is a pseudometric on K\G/H. Since q̂ × q̂ : G/H −→ K\G/H is an open map, d¯H is

continuous.
We claim that the continuous pseudometric d¯H determines the topology on K\G/H.
Suppose that W ⊆ K\G/H is an open set containing KgH. Since d¯ metrizes K\G, there
¯
exists ε > 0 such that the set V = {Kx | d(Kg, Kx) < ε} ⊆ K\G is mapped by p̂ into
W . If d¯H (KgH, KyH) < ε, then there exist h ∈ H with d(g, yh) = d(Kg,
¯ Kyh) < ε and
therefore p̂(Kyh) = KyH ∈ p̂(V ) ⊆ W . This shows that the pseudometric d¯H topologizes
K\G/H. Since K\G/H is Hausdorff, d¯H is a metric.

[Preliminary Version - October 1, 2018]

20 Chapter 1

[Preliminary Version - October 1, 2018]

2 | Around the Baire Property

We recall some notions related to the Baire Category Theorem.

Definition 2.1. Let X be a topological space. A subset N ⊆ X is called nowhere dense

if its closure N has empty interior. Equivalently, there exists a dense open set U ⊆ X
which is disjoint from N. Any subset of a nowhere dense set is again nowhere dense. A
countable union of nowhere dense sets is called a meager set (a set of first category in the
older literature). It follows that subsets of meager sets are meager, and that countable
unions of meager sets are again meager. A subset M ⊆ X is meager ifTand only if there
exists a countable family of dense open sets (Un )n≥1 in X with M ∩ n≥1 Un = ∅. A
topological space X is called T a Baire space if for every countable family of dense open
sets (Un )n≥0 , the intersection n≥0 Un is again dense. This condition can be phrased in
several ways.

Lemma 2.2. Let X be a topological space. Then the following are equivalent.
(i) X is a Baire space.
(ii) The union of countably many closed subsets with empty interiors has empty interior.
(iii) Every nonempty open subset of X is nonmeager.
(iv) The complement of every meager set is dense.

In descriptive set theory, the Polish space NN is often called the Baire space. We will
not use this terminology.

Proof. Suppose that X is a Baire space and that (An )n≥1 is a family of closed
T sets with
empty interiors.
S Then each of the open sets Un = X −A n is dense, and thus n≥1 n is also
U
dense. Thus n≥1 An has empty interior. Hence (i) implies (ii). Suppose that (ii) holds.
Then in particular, every meager set has empty interior. Hence every nonempty open set
is nonmeager, and (iii) holds. Suppose that (iii) holds and that M ⊆ X is meager. Then
the interior of M is meager and therefore empty. Hence X T − M is dense.
T If (iv) holds
and if (Un )n≥1 is a T
family of dense open sets, then M = X − n≥1 Un = n≥1 (X − Un ) is
meager and hence n≥1 Un is dense. Therefore X is a Baire space.
22 Chapter 2

In particular, every open subset of a Baire space is again a Baire space in the subspace
topology. The class of Baire spaces is, however, not closed under products or passage to
closed subsets, and this is one main reason why we consider later the subclass of Čech
complete spaces. In particular, we will see that every completely metrizable space and
every locally compact space is a Baire space, by Baire’s Category Theorem. In any case,
we have the following elementary result. We recall that a countable intersection of open
sets in a topological space is called a Gδ -set.
Lemma 2.3. Suppose that X is a Baire space and that A ⊆ X is a dense subspace. If A
is a Gδ -set, then A is a Baire space. If X 6= ∅, then A is not meager.
T
Proof. We put A = n≥1 Un , where each Un ⊆ X is open. Since A is dense, each set Un
is dense. Suppose that (Bn )n≥1 is a sequence of relatively open dense subsets of A. Then
there exist open setsT Wn ⊆ X Tsuch that Bn = A ∩ Wn . In particular, each set Wn is
dense in X. Hence n≥1 Bn = n≥1 (Un ∩ Wn ) ⊆ ASis dense in X and therefore dense in
A. Hence A is a Baire space. Moreover, X − A = n≥1 (X − Un ) is meager. If A is also
meager, then X itself is meager and therefore empty.
It follows for example that the space of irrational numbers R − Q ⊆ R is a Baire
space. On the other hand, Q is not a Baire space, since Q is nonempty and meager. The
following result is fundamental for the proof of Pettis’ Lemma below.
Theorem 2.4 (Banach’s Category Theorem). Let X be a topological space, let U
be a collection of open subsets S
of X and let A ⊆ X be a subset. If for each U ∈ U the
set A ∩ U is meager, then A ∩ U is meager. In particular, an arbitrary union of open
meager subsets is again meager.
Proof. We may assume that U 6= ∅. Let C denote the set consisting of all collections
W of subsets of X with the following properties. The members of W are open, pairwise
disjoint, and each W ∈ W is contained in some U ∈ U . The set C is partially ordered by
inclusion and (C, ⊆) is nonempty
S S and inductive. By Zorn’s Lemma, C contains a maximal
element W . Put M = U − W . Then M is closed and we claim that M has empty
interior. If a nonempty open set V is contained in M, then V intersects some U ∈ U
nontrivially. But then W ∪ {V ∩ U} ∈ C, contradicting the maximality of W . Thus M
has empty interior. Since M is closed, M is nowhere dense and in particular meager.
For every member W of W the set A ∩ W is meager. Hence there exists S a countable
family of nowhere
S dense sets NW,n ⊆ W , for n ≥ 1, with A ∩ W = n≥1 NW,n . We
put Nn = {NW,n | W ∈ W } and we claim that Nn is nowhere dense. If a nonempty
open set V is contained in Nn , then there exists a member W ∈ W which intersects V
nontrivially. Now Nn ∩ W ⊆ NW,n , and thus V ∩ W ⊆ NW,n , contradicting S the fact
S that
NW,n is nowhere dense. Hence
S N n is nowhere
S dense. It follows that A ∩ W = n≥1 Nn
is meager, and so is A ∩ U ⊆ M ∪ n≥1 Nn .

[Preliminary Version - October 1, 2018]

Around the Baire Property 23

The Open Mapping Theorem and Pettis’ Lemma

There exist topological spaces which are neither meager nor Baire spaces. An example of
such a space is the subspace Q ∪ [0, 1] ⊆ R. For topological groups this cannot happen.
Proposition 2.5. Let G be a topological group. Then the following are equivalent.
(i) G is a Baire space.
(ii) G is not meager.
(iii) G contains a non-meager subset.
Proof. It is clear that (i) ⇒ (ii) ⇒ (iii), and we show that ¬(i) ⇒ ¬(iii). If G is not a Baire
space,Sthen there exists an open nonempty meager subset U ⊆ G by Lemma 2.2. Then
G = {gU | g ∈ G} is also meager by Banach’s Category Theorem 2.4, and therefore
every subset of G is meager.
There are several open mapping theorems for topological groups. The following is the
most basic version. We note that every Hausdorff topological group which is compactly
generated is σ-compact.
Theorem 2.6 (The Open Mapping Theorem, I). Let f : G −→ K be a morphism
of Hausdorff topological groups. If G is σ-compact and if f (G) ⊆ K is not meager, then
f is open and K is locally compact.
In particular, every Hausdorff topological group which is σ-compact and a Baire space
is locally compact.
Proof. We assume that f (G) ⊆ K is not Smeager, and we consider first the case where
f is in addition injective. We write G = n∈N An , with An compact. For every n ∈ N,
f (A )
the restriction-corestriction f |An n : An −→ f (An ) is a continuous bijection and hence a
homeomorphism.
S Moreover, each f (An ) is compact and therefore closed. Since f (G) =
n∈N f (An ) is not meager, there exists by Lemma 2.2 an index m ∈ N such that f (Am )
contains a nonempty open set V . It follows from Proposition 1.12 that f (G) ⊆ K is open.
Put U = f −1 (V ). Then U ⊆ G is open, and the restriction-corestriction f |VU : U −→ V is
a homeomorphism. It follows from Lemma 1.5 that the corestriction f |f (G) : G −→ f (G)
has a continuous inverse. Hence f is open. Moreover, f (G) is an open locally compact
set in K, and thus K is locally compact.
In the general case we put N = ker(f ) and we consider the commutative diagram
f
G K.
p
f¯
G/N

[Preliminary Version - October 1, 2018]

24 Chapter 2

The quotient G/N is σ-compact. Since p is open by Proposition 1.17 and since f¯ is open
by the previous argument, f = f¯ ◦ p is open.

Corollary 2.7 (The Closed Graph Theorem, I). Let G, K be Hausdorff topological
groups and suppose that f : G −→ K is a group homomorphism. Assume also that G and
K are σ-compact, and that G is a Baire space. Then the following are equivalent.
(i) f is continuous.
(ii) the graph of f is closed in G × K.

Proof. The graph of a continuous map whose range is a Hausdorff space is closed, hence
(i) ⇒ (ii). Suppose that (ii) holds. Then H = {(g, f (g)) | g ∈ G} is a closed subgroup
of G × K and hence σ-compact. The map h : H −→ G that maps (g, f (g)) to g is
continuous and bijective. By Proposition 2.6, h is open and hence the map g 7−→ (g, f (g))
is continuous.

For two sets A, B we write the symmetric difference as

A△B = (A − B) ∪ (B − A) = (A ∪ B) − (A ∩ B).

Definition 2.8. A subset A of a topological space X is called Baire measurable (or almost
open, or a subset with the Baire property) if there exists an open set U ⊆ X such that
the symmetric difference M = U△A is meager.

The reader should be warned that a Baire measurable subset is not the same as a
subspace which is a Baire space.

Theorem 2.9 (Pettis’ Lemma). Let G be a topological group. If A ⊆ G is Baire

measurable and nonmeager, then A−1 A is an identity neighborhood.

Proof. Let U ⊆ G be open such that M = U△A is meager. Then U is not meager and
in particular nonempty. We choose g ∈ U and an open identity neighborhood V with
gV V −1 ⊆ U. We claim that V ⊆ A−1 A. If h ∈ V , then g ∈ U ∩ Uh, whence U ∩ Uh 6= ∅.
Thus (A ∩ Ah)△(U ∩ Uh) ⊆ (A△U) ∪ (A△U)h is meager. But then A ∩ Ah 6= ∅, because
otherwise U ∩ Uh would be a nonempty meager open set in the Baire space G. Hence
V ⊆ A−1 A.

From Pettis’ Lemma we derive several results about the continuity of homomorphisms.
For this, the following is a useful property of Baire measurable sets.

Proposition 2.10. The Baire measurable sets in a topological space X form a σ-algebra
which contains all open sets. In particular, every Borel set in X is Baire measurable.

[Preliminary Version - October 1, 2018]

Around the Baire Property 25

Proof. Every open set is Baire measurable, and in particular the empty set is among the
Baire measurable sets. Suppose that A ⊆ X is Baire measurable. We claim that the
complement B = X − A is also Baire measurable. Let U ⊆ X be open such that U△A is
meager and put V = X − U. We note that M = U − U is nowhere dense. The symmetric
difference of two subsets of a given set is not changed if we replace both sets by their
complements. Hence
V △B = U△A ⊆ (U△A) ∪ M
is meager, and therefore B is Baire measurable. Suppose that (An )n≥0 is a family of
Baire measurable sets. For every An there is an open set Un such
S that the symmetric
S
difference M
S n = (An ∪ Un ) − (An ∩ Un ) is meager. We put A = n≥0 An , M = n≥0 Mn
and U = n≥0 Un . Then An − U ⊆ An − Un ⊆ Mn , whence A − U ⊆ M. Likewise,
U − A ⊆ M, and therefore
U△A ⊆ M
is meager. This shows that the Baire measurable sets form a σ-algebra. Since every open
set is Baire measurable, every Borel set is Baire measurable.

Corollary 2.11. Suppose that G is a topological group and that H ⊆ G is a subgroup. If

H contains a Baire measurable set (eg. a Borel set) which is not meager, then H is open.

Proof. By Theorem 2.9, H = H −1 H is an identity neighborhood and hence open by

Proposition 1.12.

Corollary 2.12. Suppose that G is a topological group which is a Baire space. If H ⊆ G

is a dense subgroup which is a Gδ -set, then H = G.

Proof. By Lemma 2.3, H is not meager. Hence H is open by Theorem 2.9 and therefore
closed by Proposition 1.12.

We recall that a topological space is called Lindelöf if every open covering has a
countable subcovering. Examples of Lindelöf spaces are second countable spaces and σ-
compact spaces. A not necessarily continuous map between topological spaces is called
Baire measurable (Borel measurable) if the preimage of every open set is Baire measurable
(a Borel set). Every continuous map is thus Baire measurable and Borel measurable. The
reader should be aware that there are other, non-equivalent notions of Baire measurable
functions in the literature.

Theorem 2.13. Let G, K be topological groups and let f : G −→ K be a group homo-

morphism. Assume also that K is Lindelöf and that G is a Baire space. The following
are equivalent.
(i) f is a morphism of topological groups.

[Preliminary Version - October 1, 2018]

26 Chapter 2

(ii) f is Borel measurable.

(iii) f is Baire measurable.
(iv) The group K has arbitrarily small identity neighborhoods whose preimages are Baire
measurable.

Proof. A closed subset of a Lindelöf space is again Lindelöf. Replacing K by f (G),

we may thus assume in addition that f (G) is dense in K. It is clear that (i) ⇒ (ii)
⇒ (iv) and that (i) ⇒ (iii) ⇒ (iv). Suppose that (iv) holds. Let V ⊆ K be an identity
neighborhood. We claim that f −1 (V ) contains an identity neighborhood. We choose an
identity neighborhood U ⊆ K such that U −1 U ⊆ V , and such that E = f −1 (U) is Baire
measurable. Since
S K is Lindelöf and f (G) S is dense, we find elements g1 , g2 , g3 , . . . in G
such that K = n≥1 f (gn )U. Hence G = n≥1 gn E. Since G is not meager, E cannot be
meager. Hence E −1 E is an identity neighborhood by Theorem 2.9, and f (E −1 E) ⊆ V .
It follows that f is continuous at the identity element of G. By Lemma 1.5, the map f is
continuous and hence a morphism of topological groups.

Čech Complete Spaces

We introduce another piece of terminology which allows us to unify methods from com-
pletely metrizable spaces and locally compact spaces.

Definition 2.14. A Tychonoff space X is called Čech complete if there exits a countable
family (Un )n≥1 of open coverings of X such that the following holds. If F is a family of
closed subsets of X which has the finite intersection
T property, and if for every n ≥ 1 there
exists F ∈ F and U ∈ Un with F ⊆ U, then F 6= ∅. In this case (Un )n≥1 is called a
complete sequence of open coverings for X.

It follows right from the definition that a closed subspace of a Čech complete space is
again Čech complete.

Lemma 2.15. Every completely metrizable space is Čech complete.

Proof. Every metric space is a Tychonoff space. Suppose that (X, d) is a complete metric
space. We let Un denote the collection of all open subsets of diameter at most 2−n , for
n ≥ 1. Suppose that F is a family of closed sets having the finite intersection property,
and that for each n ≥ 1 there exists a member Fn ∈ F which has diameter at most 2−n .
We choose points xn ∈ F1 ∩ F2 ∩ · · · ∩ Fn . Then d(xn , xn+1 ) ≤ 2−n and thus (xn )n≥1 is
a Cauchy sequence in X. We put x = limn xn . Suppose that E is an arbitrary member
of F . For every n ≥ 1 we may choose a point yn ∈ E ∩ Fn . T Then d(xn , yn ) ≤ 2−n and
thus limn yn = limn xn = x. Therefore x ∈ E, and hence x ∈ F . Therefore (Un )n≥1 is
a complete sequence of open coverings of X.

[Preliminary Version - October 1, 2018]

Around the Baire Property 27

Lemma 2.16. Let K be a compact space and suppose that X ⊆ K is a Gδ -set. Then X
is Čech complete. In particular, every locally compact space is Čech complete.
Proof. Since every compact
T space is normal, X is a Tychonoff space. Let U1 , U2 , . . . ⊆ K
be open with X = n≥1 Un . We put Un = {X ∩ V | V ⊆ K is open and V ⊆ Un }, for
n = 1, 2, . . .. Then Un is an open covering of X, since K is regular. Suppose that E is
a collection of relatively closed subsets of Y having the finite intersection property, and
that for every n there exists
T Vn ∈ Un and En ∈ E with En ⊆ Vn . Since K is compact,
there exists a point z ∈ {E | E ∈ E}. Since z ∈ En ⊆ Un holds for every n, we have
z ∈ X. Hence (Un )n≥1 is a complete sequence of open coverings of X.
Theorem 2.17 (Baire’s Category Theorem). Every Čech complete space is a Baire
space. In particular, every completely metrizable space and every locally compact space is
a Baire space.
Proof. Let (Un )n≥1 be a complete sequence of open coverings of X. Suppose that (Vn )n≥1
is a family of open dense T subsets of X, and that W ⊆ X is a nonempty open set. We
have to show that W ∩ n≥1 Vn 6= ∅.
We put W0 = W , and we choose inductively points xn and open sets Wn for n =
1, 2, 3, . . . as follows. We choose a point xn ∈ Wn−1 ∩ Vn and an element Un ∈ Un
with xn ∈ Un . Since X is regular, there is an open neighborhood
T Wn of xn such that
Wn ⊆ Wn−1 ∩ Vn ∩ Un . T We put F = {Wn | n ≥ 1}. Then F 6= ∅ because X is Čech
complete and thus W ∩ n≥1 Vn 6= ∅.

In view of Baire’s Category Theorem, we investigate Čech complete spaces more

closely. First we improve Lemma 2.15 and Lemma 2.16. We recall that every Tychonoff
space X embeds into its Čech-Stone compactification βX. The Čech-Stone compactifi-
cation has the following universal property. If f : X −→ K is a continuous map from
a Tychonoff space X to a compact space K, then there is a unique continuous map
βf : βX −→ K that extends f ,
f
X K.
βf
βX
It follows that X is dense in βX and that β is a functor from the category of Tychonoff
spaces and continuous maps to the category of compact spaces and continuous maps.
Proposition 2.18. Let X be a Tychonoff space. The following are equivalent.
(i) X is Čech complete.
(ii) Whenever K is a compact space and ι : X −→ K is a topological embedding with
dense image, the image ι(X) ⊆ K is a Gδ -set.

[Preliminary Version - October 1, 2018]

28 Chapter 2

(iii) There exists a compact space K and a topological embedding ι : X −→ K such that
the subset ι(X) ⊆ K is a Gδ -set.
Proof. We show first that (i) ⇒ (ii). Let ι : X −→ K be a topological embedding and let
(Un )n≥1 be a complete sequence of open coverings of X. We put
[
Wn = {V ⊆ K | V is open and ι−1 (V ) ∈ Un }.
T
Then ι(X) T ⊆ W n holds for every n ≥ 1 and we claim that ι(X) = n≥1 Wn . Suppose
that z ∈ n≥1 Wn . We let A denote the collection of all closed neighborhoods of z in K.
The set E = {A ∩ ι(X) | A ∈ A} has the finite intersection property because z is in the
closure of ι(X). By the definition of Wn , there exists for every n ≥ 1 an open set Vn ⊆ K
containing z with ι−1 (Vn ) ∈ Un . Since K is regular, thereTexists a closed neighborhood
An ∈ A T with zT∈ An ⊆ Vn . Since ι(X) is Čech complete, E 6= ∅. On the other hand,
{z} = A ⊇ E ⊆ ι(X). Hence z ∈ ι(X). This shows that ι(X) is a Gδ -set in K.
Since every Tychonoff space X embeds with a dense image into its Čech-Stone com-
pactification βX, we see that (ii) ⇒ (iii). In Lemma 2.16 we showed that (iii) ⇒ (i).
Corollary 2.19. If X is a Čech complete space and if A ⊆ X is a Gδ -set, then A is Čech
complete.
Now we turn to metric completeness versus Čech completeness. We recall the following
classical fact.
Lemma 2.20. Suppose that X is completely metrizable and that A ⊆ X is a Gδ -set.
Then A is completely metrizable.
T
Proof. Let U1 , U2 , . . . ⊆ X be open, with A = n≥1 Un and let d be a complete metric
on X. We may assume that Un 6= X holds for all n ≥ 1. For each n ≥ 1 the map
fn : x 7−→ d(x, X − Un ) is continuous. Hence the map h : A −→ X × RN1 that maps
x ∈ A to h(x) = (x, 1/f1 (x), 1/f2 (x), . . .) is continuous as well, with a continuous inverse.
Hence h is a topological embedding. The product space X ×RN1 is completely metrizable.
Therefore it suffices to show that h(A) is closed. If (xn )n≥1 is a sequence in A and if
limn h(xn ) = (z, t1 , t2 , . . .), then limn xn = z and tj = limn 1/fn (z), which shows that
fn (z) 6= 0 for all n. Thus z ∈ A.
Proposition 2.21. A metrizable space X is Čech complete if and only if X is completely
metrizable.
Proof. We showed in Lemma 2.15 that every completely metrizable space is Čech com-
plete. Conversely, suppose that (X, d) is metric and Čech complete. Let X b denote the
metric completion of (X, d), and consider the natural embedding ι : X −→ β X.b Then X
b and hence by Proposition 2.18 a Gδ -set in β X.
is dense in β X b Hence X is a Gδ -set in X,
b
and therefore X is completely metrizable by Lemma 2.20.

[Preliminary Version - October 1, 2018]

Around the Baire Property 29

Corollary 2.22. Every locally compact metrizable space is completely metrizable.

Proposition 2.23. If X is a completely metrizable space, then a subspace A ⊆ X is
completely metrizable if and only if A is a Gδ -set.
Proof. If A ⊆ X is a Gδ -set, then A is completely metrizable by Lemma 2.20. Conversely,
if A ⊆ X is completely metrizable and if we put Y = A, then A is dense in βY and
therefore a Gδ -set in βY by Proposition 2.18. Therefore A is a Gδ -set in Y . On the other
hand, the closed set Y is a Gδ -set in X and therefore A is also a Gδ -set in X.
Theorem 2.24. Every product of Čech complete spaces is a Baire space and every count-
able product of Čech complete spaces is again Čech complete.
Proof. First of all we note that a product (of any length) of Tychonoff spaces is again a
Tychonoff space. Suppose that (Xn )n≥1 is a countable sequence Q of Čech complete spaces.
We put Kn =QβXn and consider the compact space K = n≥1 Kn . The natural map
Q
n≥1 Xn −→ n≥1 Kn is an embedding. It remains toTshow that its image is a Gδ -set.
For each n ≥ 1 let U1,n , U2,n , . . . ⊆ Kn be open, with k≥1 Uk,n T = Xn , and put
QWk,n =
K1 × · · · × Kn−1 × Uk,n × Kn+1 × · · · . Then Wn,k ⊆ K is open and k,n≥1 Wk,n = n≥1 Xn .
Q
Hence n≥1 Xn is a Gδ -set in K and thus Čech complete.
Suppose now that (X Qj )j∈J is a family of Čech complete spaces. We have to show
that the product X = j∈J T Xj is a Baire space. Let Un ⊆ X be open and dense, for
n = 1, 2, . . . and put A = n≥1 Un . We claim that A ⊆ X is dense. For Q this we show
that A intersects every nonempty basic open set W in X. We put W = j∈J Wj , where
Wj ⊆ Xj is nonempty and open, and Wj =QXj outside a finite nonempty index set J0 ⊆ J.
The canonical projection map p : X −→ j∈J0 Xj is continuous and open and hence the
Q
sets p(Un ) are open and dense. T Since the finite product j∈J0 Xj is Čech complete and
in particular a Baire space, n≥1 p(Un ) is dense. Hence there exist points xj ∈ Wj , for
j ∈ J0 , such that (xj )j∈J0 ∈ p(Un ) holds for all n ≥ 1. We choose arbitrary
T points xj ∈ Xj
for all j ∈ J − J0 . Then (xj )j∈J ∈ Un ∩ W holds for all n. Hence W ∩ n≥1 Un 6= ∅.
We recall that a continuous map between Hausdorff spaces is called proper if the
preimage of every compact set is compact. The following is elementary.
Lemma 2.25. Suppose that f : X −→ Y is a continuous map between Hausdorff spaces,
that A ⊆ X is a subset and that the restriction f |A : A −→ Y is closed and proper. Then
A ⊆ X is closed.
Proof. Suppose that x ∈ X − A. Then B = A ∩ f −1 (f (x)) is compact, because the
restriction f |A : A −→ Y is proper. Since X is Hausdorff and B ⊆ X is compact, there
exists an open set U ⊆ X with B ⊆ U and x 6∈ U. The set f (A − U) is closed and
does not contain f (x). Hence f −1 (f (A − U)) is closed in X and does not contain x. In
particular, x 6∈ A − U ⊆ f −1 (f (A − U)). On the other hand, x 6∈ U, whence x 6∈ A.

[Preliminary Version - October 1, 2018]

30 Chapter 2

The complement of a Gδ -set in a topological space is called an Fσ -set. An Fσ -set is

thus a countable union of closed sets.

Corollary 2.26. Suppose that X, Y are Tychonoff spaces and that f : X −→ Y is a

continuous, surjective, closed, and proper map. Then X is Čech complete if and only if
Y is Čech complete.

Proof. We consider the induced map βf : βX −→ βY between the Čech-Stone com-

pactifications. We claim that βf (βX − X) = βY − Y . To show this, we put Z =
(βf )−1 (Y ) ⊆ βX and we consider the restriction-corestriction (βf )|YZ : Z −→ Y . By
Lemma 2.25, X is closed in Z. In the other hand, X is dense in Z, whence Z = X
and thus βf (βX − X) ⊆ βY − Y . Since Y is dense in βY and f (X) = Y , we have
βf (βX − X) = βY − Y . Now βf is a closed continuous map and hence βf (βX − X) is
an Fσ -set if and only if βY − Y is an Fσ set.
A Hausdorff space X is called paracompact if every open covering of X has a locally
finite refinement. Every metric space is paracompact.

Proposition 2.27. Suppose that f : X −→ Y is a continuous open surjective map, that

X is Čech complete and that Y is paracompact. Then Y is Čech complete.

Proof. We subdivide the proof into several steps.

Claim. There exists a Tychonoff space Z containing X as a Gδ -subspace and a continuous
closed proper map F : Z −→ Y which extends f .
We consider the map βf : βX −→ βY between the Čech-Stone compactifications, and
we put Z = βf −1(Y ) ⊇ X. We denote the restriction-corestriction of βf by

F : Z −→ Y.

If B ⊆ Y is compact, then F −1 (B) = βf −1(B) is compact, hence F is proper. If A ⊆

βf (X) is closed, then A is compact and F (A ∩ Z) = βf (A) ∩ Y is closed in Y , hence F
is closed.
Claim. Suppose that U ⊆ Z is open and that F (U ∩ X) = Y . Then there exists an open
set U ′ ⊆ Z with F (U ′ ∩ X) = Y and with U ′ ⊆ U.
Suppose that U ⊆ Z is open and that F (X ∩ U) = Y . We choose, for every y ∈ Y ,
an open set set Vy ⊆ Z as follows. We choose an element x ∈ X ∩ U with f (x) = y,
and then an open neighborhood Vy ⊆ U of x with Vy ⊆ U. This is possible because
Z is regular. Since Vy ∩ X is open in X and since f is open, Uy = f (Vy ∩ X) is an
open neighborhood of y. Since Y is paracompact, the open cover {Uy | y ∈ Y } has
a locally finite refinement W . For every W ∈ W we choose y(W ) ∈ Y such that
W ⊆ Uy(W ) . We put UW = Vy(W ) ∩ F −1 (W ) and we note that UW ⊆ U is open. We

[Preliminary Version - October 1, 2018]

Around the Baire Property 31

S
also put U ′ = {UW | W ∈ W }. If W ∈ W , then W ⊆ f (Vy(W ) ∩ X) and thus
f (UW ∩ X) = W . Hence F (U ′ ∩ X) = Y . Suppose that z ∈ Z is in the closure of U ′ .
There exists an open neighborhood O of f (z) such that set WO = {W ∈ W | O∩W 6= ∅}
is finite. Hence
[ [
z ∈ {UW | W ∈ WO } = {UW | W ∈ WO } ⊆ U.

Claim. There existsSa closed Gδ -set C ⊆ X such that f |C : C −→ Y is surjective.

We write X = n≥1 Un , where Un ⊆ Z is open. we construct a sequence of open
sets Zn as follows We put Z0 = Z. Given Zn−1 , Twe choose T an open set Zn such that
Zn ⊆ Un ∩ Zn , with F (X ∩ ZN ) = Y . We put C = n≥0 Zn = n≥0 Zn . Thus C ⊆ X is a
T
closed Gδ -set. The set f −1 (y) ⊆ X is compact for every y, hence Zn ∩ f −1 (y) 6= ∅.
The claim of the proposition is true.
The set C ⊆ X is a Gδ -set in the Čech complete space X, and therefore Čech complete.
Since C ⊆ Z is closed, F |C = f |C is continuous, proper, closed and surjective. Therefore
Y is Čech complete.

Čech Complete Groups

From the results of the previous section we see that the class of Čech complete groups
has many favorable properties. It is closed under passage to closed subgroups, to open
subgroups and even to Gδ -subgroups, and also closed under passage to countable products.
It remains to consider quotients.

Proposition 2.28. Let G be a Hausdorff topological group and Y ⊆ G be any Gδ -set

containing e (for example, Y = G). The following are equivalent.
(i) The group G is Čech complete.
(ii) There is a continuous length function ℓ on G such that K = {g ∈ G | ℓ(g) = 0} is
compact and contained in Y , and such that K\G is Čech complete and metrizable
¯
by the metric d(Kx, Ky) = ℓ(xy −1 ).
(iii) There is a compact subgroup K ⊆ Y such that K\G is Čech complete.
(iv) There is a compact subgroup K ⊆ G such that K\G is Čech complete.
If G is σ-compact, then the subgroup K in (iii) can in addition be chosen to be normal.

A corresponding result holds of course for the left coset space G/K.

Proof. It is clear that (ii) ⇒ (iii) ⇒ (iv). If (iv) holds, then G is Čech complete by
Corollary 2.26 and hence (iv) T⇒ (i). It remains to show that (i) ⇒ (ii). Suppose that G is
Čech complete and that Y = n≥1 Vn , where the Vn ⊆ G are open identity neighborhoods.

[Preliminary Version - October 1, 2018]

32 Chapter 2

If G is σ-compact, then there exists

S an ascending sequence L1 ⊆ L2 ⊆ L3 ⊆ · · · of
compact subsets of G such that G = n≥1 Ln .
Let (Un )n≥1 be a complete sequence of open coverings. For every n we choose Un ∈
Un with e ∈ Un . Next we choose inductively closed symmetric identity neighborhoods
·3
K−n ⊆ Un ∩ Vn , such that K−(n+1) ⊆ K−n holds, for n = 1, 2, . . ..
S
If G = n≥1 Ln is an ascending union of compact sets, then we choose the K−n in
addition in such a way that

(c) aK−(n+1) a−1 ⊆ Kn

holds for all a ∈ Ln . This is possible by Wallace’s Lemma 1.16.

For n ≥ 0 we put Kn = G. Let ℓ : G −→ R denoteTa continuous length function for the
the sequence (Kn )n∈Z as in Lemma 1.41. Put K = n∈Z Kn = {g ∈ G | ℓ(g) = 0} ⊆ Y .
If (c) holds, then K is a normal subgroup of G.
We claim that the closed subgroup K ⊆ G is compact. T If E is a family of closed
subsets of K having the finite intersection property, then E 6= ∅, because each E ∈ E
is contained in some Un ∈ Un . Hence K is compact and K\G is Čech complete by
Corollary 2.26.
It remains to show the metrizability of K\G. We put d(x, y) = ℓ(xy −1 ) and we note
that d(x, y) = 0 holds if and only if x ∈ Ky. Therefore d descends to a metric d¯ on K\G.
Since the map q : G −→ K\G is open, the map q × q : G × G −→ K\G × K\G is also
open and in particular a quotient map. Hence d¯ is a continuous metric on K\G.
Suppose that W is an open neighborhood of Ke ∈ K\G and put U = q −1 (W ). We
claim that there exists n ≥ 1 with K−n ⊆ U. Otherwise, the sets A Tn = K−n − U are
nonempty, closed and nested, A1 ⊇ A2 ⊇ A3 ⊇ · · · . Then A = n≥1 An 6= ∅, and
A ⊆ K − U = ∅, a contradiction. Hence there exists n ≥ 1 with q(K−n ) ⊆ W , and
¯
therefore {Kg | d(K, Kg) < 2−n } ⊆ W . Therefore d¯ metrizes K\G.

Corollary 2.29. If G is a Čech complete topological group G and if there exists an

identity neighborhood V ⊆ G which contains no compact subgroup besides {e}, then G is
completely metrizable.

Lemma 2.30. Suppose that G is a Čech complete group and that H ⊆ G is a subgroup
which is a Gδ -set. Then H is closed.

Proof. Since the closure H ⊆ G is Čech complete, we may in addition assume that H is
dense in G. The claim follows now from Corollary 2.12.

Lemma 2.31. Suppose that G is a Čech complete topological group and that H is a Čech
complete dense subgroup. Then H = G.

[Preliminary Version - October 1, 2018]

Around the Baire Property 33

Proof. The subspace H ⊆ βG is dense and therefore a Gδ -set in βG by Proposition 2.18.

Hence H is a Gδ -set in G, and therefore closed by Lemma 2.30.
Proposition 2.32. Let G be a completely metrizable topological group and let d be a left
invariant metric which metrizes G. Then the metric dc (x, y) = d(x, y) + d(x−1 , y −1 ) is
complete.
Proof. By Proposition 1.37 the metric completion G b of (G, dc ) is a topological group.
b b by Proposition 2.23. Thus
Since G ⊆ G is dense and Čech complete, G is a Gδ -set in G
G=G b by Lemma 2.31.

Corollary 2.33. If G is a completely metrizable topological group, then every left and
right invariant metric which metrizes A is complete.
The result applies in particular to abelian groups, where every left invariant metric is
automatically right invariant.
Proof. If d is left and right invariant, then dc = 2d is complete by Proposition 2.32 hence
d is complete as well.
Theorem 2.34. If G is a Čech complete group and if H is a closed subgroup, then G/H
and H are Čech complete. If G is completely metrizable, then G/H and H are completely
metrizable.
Proof. If G is Čech complete, then the closed subgroup H is also Čech complete. We have
to show that G/H is Čech complete. By Proposition 2.28 there exists a continuous length
function ℓ on G such that K = {g ∈ G | ℓ(g) = 0} is compact and K\G is metrizable by
¯
the metric d(Kx, Ky) = ℓ(xy −1 ). We consider the diagram of quotient maps and spaces

G
q p

K\G r G/H

p̂ q̂
K\G/H.

Since p, q, q̂ are open maps by Proposition 1.17 and Proposition 1.27, r, p̂ are open as
well. The space K\G/H is metrizable by Lemma 1.44 and in particular paracompact.
Since K\G is Čech complete, we conclude from Proposition 2.27 that K\G/H is Čech
complete. Then G/H is Čech complete by Corollary 2.26.
Suppose that G is completely metrizable. Then H is also completely metrizable.
Moreover, G/H is metrizable by Proposition 1.43 and therefore completely metrizable.

[Preliminary Version - October 1, 2018]

34 Chapter 2

Polish Groups and the Open Mapping Theorem

A topological space X is called Polish if it is separable and completely metrizable. Equiv-
alently, X is second countable and completely metrizable, or metrizable, Čech complete
and second countable. A topological group is called Polish if its underlying topology is
Polish. For example, the topological group ZN is Polish since Z the discrete group is
completely metrizable and second countable.

Lemma 2.35. A subspace A of a Polish space X is Polish if and only if A is a Gδ -set.

Proof. Any subspace of a second countable space is second countable. By Proposition 2.23,
the subspace A ⊆ X is completely metrizable if and only if it is a Gδ -set.

Proposition 2.36. Suppose that G is a Polish group and that H ⊆ G is a subgroup.

Then H is Polish if and only if H is closed in G. If H is closed, then G/H is also Polish.
In particular, G/H is again a Polish group if H is a closed normal subgroup.

Proof. By Lemma 2.35, a subgroup H ⊆ G is Polish if and only if it is a Gδ -set. In this

case, H is closed by Lemma 2.30. The quotient G/H is separable because G is separable
and if H is closed, then G/H is completely metrizable by Theorem 2.34.

Our next aim is the Open Mapping Theorem for Polish groups. This requires some
preparation.

Lemma 2.37. Suppose that X is a Polish space. Then there exists a continuous surjective
map NN −→ X

Proof. Let dX be a complete metric on X. On NN we use the metric d(a, b) = 2−m , where
m = inf{k ∈ N | ak 6= bk }, with 2−∞ = 0. Let Z = {z0 , z1 , z2 , . . .} ⊆ X be a countable
dense subset. Given a ∈ NN , we define a sequence (sak )k≥0 as follows. We put sa0 = z0 and
(
zak if dX (zak , sak ) ≤ 2−k
sak+1 =
sak else.

It follows that dX (sak , xak+1 ) ≤ 2k and therefore (sak )k≥0 is a Cauchy sequence in X. We put
f (a) = limk sak . If ak = bk for k = 0, . . . , m, then sak = sbk for k = 0, . . . m and therefore
dX (f (a), f (b)) ≤ 2−m+4 . Hence f is continuous. Given x ∈ X, we define a sequence a by
am = min{k ∈ N | d(x, zk ) ≤ 2−m−1 }. Then f (a) = x.

We now introduce the Alexandroff–Suslin Operation A.

[Preliminary Version - October 1, 2018]

Around the Baire Property 35

Definition 2.38. Put S = N ∪ N2 ∪ N3 ∪ · · · and suppose that P is a collection of sets.

A Suslin scheme is a map u : S −→ P , s 7−→ us . Given an element a ∈ NN , we put

ua = u(a0 ) ∩ u(a0 ,a1 ) ∩ u(a0 ,a1 ,a2 ) ∩ · · ·

and
[ [
A(u) = {ua | a ∈ NN } ⊆ P.
S
P
The map A : P S −→ 2 is called the Alexandroff–Suslin operation.
Lemma 2.39. Let X be a Hausdorff space and suppose that f : NN −→ X is a continuous
map. Then there exists a Suslin scheme u : S −→ P , where P denotes the collection of
all closed subsets of X, with A(u) = f (NN ).
Proof. For m ≥ 1 and s ∈ Nm we put Es = {b ∈ NN | (b0 , . . . , bm−1 ) = s}, and us =
f (Es ) ⊆ X. Thus we have set up a Suslin scheme u.
If x ∈ A(u), then there exists a ∈ NN with

x ∈ ua = u(a0 ) ∩ u(a0 ,a1 ) ∩ u(a0 ,a1 ,a2 ) ∩ · · · .

Hence \
x∈ f ({b | bk = ak for k = 0, . . . , n}).
n≥0

Let V be a closed neighborhood of x. For every n ≥ 0 there exists an element bn ∈ NN

such that bk,n = ak for k = 0, . . . , n, and with f (bn ) ∈ V . Now limn bn = a and therefore
limn f (bn ) = f (a). It follows that f (a) ∈ V , whence x = f (a).
Conversely, if x = f (a), then a ∈ E(a0 ,...,an ) for every n ≥ 0 and thus f (a) ∈ ua .
Lemma 2.40. Suppose that X is a topological space and that A ⊆ X is a subset. Then
there exists a Baire measurable subset B with A ⊆ B ⊆ A such that every Baire measurable
subset Z ⊆ B − A is meager.
S
Proof. We put U = {U ⊆ X | U ∩ A is meager} and we put W = U . By Banach’s
Category Theorem 2.4, the set Y = A ∩ W is meager. We put

B = (X − W ) ∪ A = (X − W ) ∪ Y.

Then B is Baire measurable and A ⊆ B. The set X − A is a member of U and thus

X − W ⊆ A, which shows that B ⊆ A.
Suppose that Z ⊆ B − A is Baire measurable. Then there exists an open set V ⊆ X
such that M = Z△V is meager. Since Z ∩ A = ∅, the set V ∩ A is meager and therefore
V ⊆ W . On the other hand, Z ⊆ X − W and therefore Z ∩ V = ∅, whence Z ⊆ M.

[Preliminary Version - October 1, 2018]

36 Chapter 2

Lemma 2.41. Suppose that X is a topological space. Let P denote the set of all Baire
measurable subsets of X and suppose that u : S −→ P is a Suslin scheme. Then A(u) ⊆ X
is Baire measurable.
Proof. We define a Suslin scheme v : S −→ P by putting

v(n1 ,...,nm ) = u(n1 ) ∩ u(n1 ,n2 ) ∩ · · · ∩ u(n1 ,...,nm )

and we note that va = ua holds for all a ∈ NN , whence A(u) = A(v). The Suslin scheme
v has the additional property that

v(n1 ) ⊇ v(n1 ,n2 ) ⊇ · · · ⊇ v(n1 ,...,nm ) ⊇ · · · .

At this stage is is convenient to introduce formally the empty tuple s = (). We put
Es = {a ∈ NN | (a0 , . . . , am−1 ) = s}, with the understanding that E() = NN , and we put
v() = X. For s ∈ S ∪ {()} we put
[
As = {va | a ∈ Es }.

Then A() = A(v), and As ⊆ vs . By Lemma 2.40 there exists a Baire measurable set B
with As ⊆ B such that every Baire measurable subset Z ⊆ B − As is meager. Then the
Baire measurable set Bs = B ∩ vs ⊇ As has the same property.
For n ∈ N and s = (n1 , . . . , nm ) we put s.n = (n1 , . . . , nm , n) and ().n = (n). Then
[
As = As.n .
n≥0
S S S
The set
S Ms = Bs − n≥0 Bs.n is meager, because As = n≥0 As.n ⊆ n≥0 Bs.n . Hence
M = {Ms | s ∈ S} is meager as well. We claim that

B() − M ⊆ A() = A(v).

Once we have proved this, it follows that B() − A() ⊆ M is meager, and hence that A() is
Baire measurable. S
Suppose that x ∈ B() − M. Then x ∈SB() − M() = n≥0 B(n) and thus x ∈ B(a0 ) for
some a0 ∈ N. Now x ∈ B(a0 ) − M(a0 ) = n≥0 B(a0 ,n) and therefore x ∈ B(a0 ,a1 ) for some
a1 ∈ N. Continuing inductively we find a ∈ NN such that

x ∈ B(a0 ) ∩ B(a0 ,a1 ) ∩ B(a0 ,a1 ,a2 ) ∩ · · · ⊆ va ⊆ A(v).

Proposition 2.42. Suppose that X is a Polish space, that Y is a Hausdorff space and
that f : X −→ Y is a continuous map. Then f (X) = A ⊆ Y is Baire measurable.

[Preliminary Version - October 1, 2018]

Around the Baire Property 37

Proof. By Lemma 2.37, there exist a continuous surjective map g : NN −→ X. Hence

A = f (g(NN )). By Lemma 2.39 there exists a Suslin scheme u : S −→ P , where P denotes
the set of closed subsets of Y , with A = A(u). By Lemma 2.41 the set A(u) is Baire
measurable.
Theorem 2.43 (The Open Mapping Theorem, II). Suppose that G is a Polish
group, that K is a Hausdorff topological group and that f : G −→ K is a continuous
homomorphism. If f (G) is not meager, then f is open.
Proof. We assume that f (G) is not meager and we put N = ker(f ). Replacing G by
the Polish group G/N, we may assume that f is injective. By Proposition 2.42, f (G)
is Baire measurable. Since f (G) is not meager, f (G) is open by Pettis’ Lemma 2.9. If
U ⊆ G is an identity neighborhood, then there exist a closed identity neighborhood V ⊆ G
with V −1 V ⊆ U. Then E = f (V ) is Baire measurable by Proposition 2.42. Since G is
separable, f (G) is a countable union of translates of E, and therefore E is not meager.
Hence E −1 E ⊆ f (U) is an identity neighborhood in K. It follows that the inverse of
the corestriction f |f (G) : G −→ f (G) is continuous at the identity, and hence continuous
everywhere by Lemma 1.5. Hence f is open.
Corollary 2.44 (The Closed Graph Theorem, II). Suppose that G and K are Polish
groups and that f : G −→ K is a group homomorphism. Then the following are equivalent.
(i) f is continuous.
(ii) The graph of f is closed in G × K.
(iii) The graph of f is a Gδ -set in G × K.
Proof. The graph of a continuous map is closed, hence (i) ⇒ (ii). A closed set in a metric
space is a Gδ -set, hence (ii) ⇒ (iii). Suppose that H = {(g, f (g)) | g ∈ G} is a Gδ -set
in G × K. Then H is a Polish group by Lemma 2.35. The map h : H −→ G that maps
(g, f (g)) to g is continuous and bijective and hence open by Theorem 2.43. Hence the
map g 7−→ (g, f (g)) is continuous, and therefore f is continuous.
Theorem 2.45 (The Open Mapping Theorem, III). Suppose that f : G −→ K is a
morphism of topological groups, that G is Polish, and that K is metrizable. If for every
identity neighborhood V ⊆ G, the set f (V ) has nonempty interior, then f is open.
Proof. In view of Theorem 2.43 it suffices to prove that f (G) is not meager. Suppose to
the contrary that f (G)
S is meager. Then there exist closed sets An with empty interiors,
S for
−1
n ≥ 1, with f (G) ⊆ n≥1 An . The sets Bn = f (An ) ⊆ G are closed, and G = n≥1 Bn .
Since G is a nonempty Baire space, at least one set Bm contains a nonempty open set
U ⊆ Bm . Let g ∈ U. Then g −1 U is an identity neighborhood in G and therefore
W = f (g −1U) is a closed identity neighborhood in K. But then f (U) = f (g)W ⊆ Am
is a neighborhood of f (g), a contradiction. Therefore f (G) is not meager and hence f is
open.

[Preliminary Version - October 1, 2018]

38 Chapter

[Preliminary Version - October 1, 2018]

Bibliography

[1] S. Banach, Théorème sur les ensembles de première catégorie, Fundamenta Mathe-
maticae 16 (1930), 395–398.

[2] Y. Cornulier and P. de la Harpe, Metric geometry of locally compact groups, EMS
Tracts in Mathematics, 25, European Mathematical Society (EMS), Zürich, 2016.
MR3561300

[3] A. Deitmar and S. Echterhoff, Principles of harmonic analysis, second edition, Uni-
versitext, Springer, Cham, 2014. MR3289059

[4] J. Dugundji, Topology, Allyn and Bacon, Inc., Boston, MA, 1978. MR0478089

[5] E. Hewitt and K. A. Ross, Abstract harmonic analysis. Vol. I, second edition,
Grundlehren der Mathematischen Wissenschaften, 115, Springer-Verlag, Berlin, 1979.
MR0551496

[6] K. H. Hofmann and S. A. Morris, The structure of compact groups, third edition,
revised and augmented., De Gruyter Studies in Mathematics, 25, De Gruyter, Berlin,
2013. MR3114697

[7] K. H. Hofmann and S. A. Morris, Open mapping theorems for topological groups,
preprint 2009.

[8] I. Kaplansky, Lie algebras and locally compact groups, reprint of the 1974 edition,
Chicago Lectures in Mathematics, University of Chicago Press, Chicago, IL, 1995.
MR1324106

[9] C. Kuratowski, Topologie. I et II, reprint of the fourth (Part I) and third (Part II)
editions, Éditions Jacques Gabay, Sceaux, 1992. MR1296876

[10] J. C. Oxtoby, Measure and category, second edition, Graduate Texts in Mathematics,
2, Springer-Verlag, New York, 1980. MR0584443
[11] B. J. Pettis, On continuity and openness of homomorphisms in topological groups,
Ann. of Math. (2) 52 (1950), 293–308. MR0038358

[12] M. Stroppel, Locally compact groups, EMS Textbooks in Mathematics, European

Mathematical Society (EMS), Zürich, 2006. MR2226087

[13] T. Tao, Hilbert’s fifth problem and related topics, Graduate Studies in Mathematics,
153, American Mathematical Society, Providence, RI, 2014. MR3237440
Index

topological group, 1