Automorphism Groups

Lecture Notes

November 24, 2024

Manuel Bodirsky
 Contents

Chapter 1. Permutation Groups 7

1.1. Structures 7
1.2. Automorphism Groups 10
1.3. Group Actions 14
1.4. Congruences and Primitivity 18
1.5. Semidirect Products 20

Chapter 2. Counting Orbits 27

2.1. Two Integer Sequences 27
2.2. Combinatorial Tools 28
2.3. On the Number of Orbits of n-Subsets 30
2.4. Highly Set-transitive Permutation Groups 31

Chapter 3. Oligomorphic Permutation Groups 33

3.1. sInv-Aut 33
3.2. Countably Categorical Structures 36
3.3. Homogeneous Structures and Amalgamation Classes 38
3.4. Strong Amalgamation and Algebraicity 44
3.5. The Weak Amalgamation Property 48
3.6. First-order Interpretations 48

Chapter 4. Topological Groups 53

4.1. Topological Spaces 53
4.2. Topological Groups 66
4.3. Closed Subgroups 75
4.4. Open Subgroups 77
4.5. Compact Subgroups 78
4.6. Closed Normal Subgroups 80

Chapter 5. Birkhoff’s Theorem and Permutation Groups 83

5.1. Birkhoff’s Theorem 83
5.2. Topological Birkhoff 85
5.3. Continuous Homomorphisms and Interpretability 86
5.4. Topological Isomorphism and Bi-interpretability 87

Chapter 6. Reconstruction of Topology and Automatic Continuity 91

6.1. Reconstruction Notions 91
6.2. The Small Index Property 92
6.3. Consistency of Automatic Continuity 93
6.4. Ample Generics 99
6.5. The Extension Property for Partial Automorphisms 107
6.6. The Strong Small Index Property 109
6.7. Open Problems 112
3
4 CONTENTS

Chapter 7. Ramsey Classes and Topological Dynamics 115

7.1. Ramsey Classes 115
7.2. The Kechris-Pestov-Todorcevic Connection 115
7.3. Compact Spaces for Oligomorphic Groups 116
7.4. Canonical Functions 117
7.5. Model-Complete Cores of Ramsey Structures 119
Chapter 8. Reducts and Closed Supergroups 125
Chapter 9. Restricted Orbit Growth 127
Chapter 10. Homogeneous Structures in Restricted Signatures 129

Bibliography 131
Appendix A. Background Material 137
A.1. Ultrafilter 137
A.2. The Axiom of Choice and its Weaker Versions 139
 CONTENTS 5

Disclaimer: this is a draft and probably contains many typos and mistakes.
Please report them to [email protected].
Recommendations. Chapters 1-4 are suitable for a 3rd year bachelor course or the
beginning of a master course. Chapters 5-9 are for a master course.
Notes concerning text book literature. We use material from the following text
books:
• Cameron’s Permutation groups [40] and Oligomorphic permutation groups [39],
• Dixon and Mortimer’s Permutation groups [48],
• The collection Notes on Infinite Permutation Groups [12] by Macpherson,
Möller, and Neumann,
• Kechris’ Classical descriptive set theory [88],
• Gao’s Invariant descriptive set theory [60], and
• Hodges’ Model theory [72].
Oligomorphic permutation groups are the topic of the short and stimulating book by
Peter Cameron [39]. The book on permutation groups by the same author [40] is
mostly about finite permutation groups, but Section 5 covers oligomorphic permuta-
tion groups. There is even less material on infinite permutation groups in [48]. The
notes in [12] are on infinite permutation groups, but they neglect the topological as-
pect of the topic. Hodges [73] is titled ‘model theory’, but covers many fundamental
things for permutation groups on the way, including topological aspects. Relevant
facts about Polish groups can be found in [60, 88].
Prerequisites. The text is essentially self-contained. Signatures, structures, sub-
structures, etc. are introduced, even though many readers may be familiar with these
concepts. However, we do not introduce first-order logic, but rather refer to logic
bachelor course notes [18]; the same applies for axiomatic set theory (we assume
familiarity for instance with Zorn’s lemma).
The present notes contains a few theorems that are just stated but not proved,
for instance
• the theorem of Ryll-Nardzewski (Theorem 3.2.3); however, we have ex-
tracted all the consequences of the (proof of the) Ryll-Nardzewski theorem
that we need in this text into Theorem 3.1.1 which we do prove. The full
proof of Theorem 3.2.3 can be found in many text books on model theory.
• the theorem of Birkhoff-Kakutani (Theorem 4.2.8); this is covered for ex-
ample in [60, 88].
• a consequence of a theorem of Lusin-Sierpiński (Theorem 6.3.12); this is
again covered in [88] (21.6).
6 CONTENTS

Exercises.
The text contains 153 exercises;
some of them are graded using the
Mandala scale.

Acknowledgements.
The author thanks the participants
of the course on automorphism groups
at TU Dresden
in the winter semester 2022/23.
 CHAPTER 1

Permutation Groups

A permutation of a set X is a bijection between X and X. We use cycle notation

for permutations: e.g., (12)(345) denotes the permutation g such that g(1) = 2,
g(2) = 1, g(3) = 4, g(4) = 5, and g(5) = 3. A permutation group G on a set X is a
set of permutations of X such that the following three conditions hold.
(1) G contains the identity permutation idX .
(2) G contains for every permutation u ∈ G also its inverse u−1 defined by
u−1 (u(x)) = x for all x ∈ X.
(3) G contains for all u, v ∈ G their composition u ◦ v, defined by
(u ◦ v)(x) = u(v(x)) for all x ∈ X.

Examples.
• The set of all permutations of X, denoted by Sym(X). We use Sn as a
shortcut for Sym({1, . . . , n}) and Sω as a shortcut for Sym(N).
• The set of functions {ta : Z → Z | ta (x) = x + a, a ∈ Z} ⊂ Sym(Z).
• The set of all order-preserving permutations of Q.
All of the three examples of permutation groups given above are transitive, i.e., for
any a, b ∈ X there exists u ∈ G such that u(a) = b.
Inspiration “Erlanger Programm” of Felix Klein: Understanding structure (in his
case, geometry) by understanding its symmetry (via permutation groups).
Definition 1.0.1. Let G be a permutation group on a set A and H a permutation
group on a set B. Then a bijection i between A and B is called a (permutation group)
isomorphism if there exists a bijection ξ between G and H such that for all g ∈ G and
a ∈ A we have
i(g(a)) = ξ(g)(i(a)).

1.1. Structures
A signature τ is a set of relation and function symbols. Each symbol is equipped
with an arity k ∈ N. A τ -structure A is a set A (the domain of A) together with
• a relation RA ⊆ Ak for each k-ary relation symbol in τ , and
• a function f A : Ak → A for each k-ary function symbol in τ ; here we allow
the case k = 0 to model constant symbols.
Unless stated otherwise, A, B, C, . . . denote the domains of the structures A, B, C,
A A A A
. . . , respectively. We sometimes write (A; R1 , R2 , . . . , f1 , f2 , . . . ) for the relational
A A A A
structure A with relations R1 , R2 , . . . and functions f1 , f2 , . . . When there is no
danger of confusion, we use the same symbol for a function and its function symbol,
and for a relation and its relation symbol. We say that a structure is infinite if its
domain is infinite.
7
8 1. PERMUTATION GROUPS

Example 1. A directed graph (or, short, digraph is a relational structure over

the signature that contains a single binary relation symbol for the edge relation of
the graph. 4
Example 2. A (simple, undirected)
graph is a pair (V, E) consisting of a set
of vertices V and a set of edges E ⊆ V2 , that is, E is a set of 2-element subsets
of V . Graphs can be modelled using relational structures G using a signature that
contains a single binary relation symbol R, putting RG := E. If we insist that a
structure with this signature satisfies (x, y) ∈ RG ⇒ (y, x) ∈ RG and not (x, x) ∈ RG ,
then we can associate to such a structure an undirected graph and obtain a bijective
correspondence between undirected graphs and structures G as described above. 4
1.1.1. Extensions and substructures. A τ -structure A is a substructure of a
τ -structure B iff
• A ⊆ B,
• for each R ∈ τ , and for all tuples ā from A, ā ∈ RA iff ā ∈ RB , and
• for each f ∈ τ we have that f A (ā) = f B (ā).
In this case, we also say that B is an extension of A. Substructures A of B and
extensions B of A are called proper if A 6= B.
Note that for every subset S of the domain of B there is a unique smallest
substructure of B whose domain contains S, which is called the substructure of B
generated by S, and which is denoted by B[S].
Example 3. A group is a structure G with a binary function symbol · for multi-
plication, a unary function symbol −1 for taking the inverse, and a constant denoted
by 1, called the neutral element, satisfying the sentences
• ∀x, y, z. x · (y · z) = (x · y)·z,
• ∀x. x · x−1 = 1,
• ∀x. 1 · x = x, and ∀x. x · 1 = x.
In this signature, the subgroups of G are precisely the substructures G as defined
above; we also write H ≤ G if H is a subgroup of G. Every group has the trivial
subgroup which is the subgroup {1} that just contains the identity element. To
distinguish groups from permutation groups, we might also refer to a group as an
abstract group. Clearly, every permutation group G gives rise to an abstract group G
where ◦ takes the role of multiplication. 4
Example 4. Let G be a permutation group on a set A. Let A be a structure
with domain A whose signature only contains unary function symbols that denote
permutations of A. Note that every τ -term must have exactly one variable, and every
τ -term t(x) defines over A a permutation tA . Then A is called a G-set if the set of all
permutations obtained in this way is precisely G. (We do allow that several function
symbols denote the same element of G.) 4
1.1.2. Homomorphisms. In the following, let A and B be τ -structures. A
homomorphism h from A to B is a mapping from A to B that preserves each function
and each relation for the symbols in τ ; that is,
• if (a1 , . . . , ak ) is in RA , then (h(a1 ), . . . , h(ak )) must be in RB ; and
• f B (h(a1 ), . . . , h(ak )) = h(f A (a1 , . . . , ak )).
A homomorphism from A to B is called a strong homomorphism if it also preserves
the complements of the relations from A. Injective strong homomorphisms are called
embeddings, and we write e : A ,→ B if e is an embedding of A into B.
Example 5. If G and H are groups and h : G → H is a map, then in order
to verify that h is a homomorphism it suffices to prove that h preserves ◦ (Why?).
 1.1. STRUCTURES 9

Moreover, note that injective homomorphisms are embeddings (this is not true for
structures whose signature contains relation symbols! Find a counterexample!). 4

1.1.3. Isomorphisms. Surjective embeddings are called isomorphisms. The fol-

lowing is immediate from the definitions.

Proposition 1.1.1. G and H are isomorphic (as permutation groups, Defini-

tion 1.0.1) if and only if there exists a G-set A and an H-set B such that A and B
are isomorphic as structures (as introduced above).

Note that if G and H are isomorphic (as permutation groups), then the corre-
sponding abstract groups are isomorphic as well, but the converse need not be true
(see Corollary 5.1.2 for a characterisation of permutation groups that are isomorphic
as abstract groups).

1.1.4. Automorphisms. Homomorphisms and isomorphisms from B to itself

are called endomorphisms and automorphisms, respectively. When f : A → B and
g : B → C, then g ◦ f denotes the composed function x 7→ g(f (x)). Clearly, the
composition of two homomorphisms (embeddings, automorphisms) is again a homo-
morphism (embedding, automorphism). Let Aut(A) and End(A) be the sets of auto-
morphisms and endomorphisms, respectively, of A. The set Aut(A) can be viewed as
a group, and End(A) as a monoid with respect to composition.

1.1.5. Expansions and reducts. Let σ, τ be signatures with σ ⊆ τ . If A is a

σ-structure and B is a τ -structure, both with the same domain, such that RA = RB
for all relations R ∈ σ and f A = f B for all functions and constants f ∈ σ, then A is
called a reduct of B, and B is called an expansion of A.

1.1.6. Disjoint unions. Let τ be a relational signature. A disjoint union A ] B

of two τ -structures A and B is the union of isomorphic copies of A and B with disjoint
domains. That is, for all R ∈ τ we have RA]B = RA ∪ RB . As disjoint unions are
unique up to isomorphism, we usually speak of the disjoint union of A and B. The
disjoint union of a set of τ -structures C is defined analogously (and the disjoint union
of an empty set of structures is the τ -structure with empty domain). A relational
structure is called connected if it is not the disjoint union of two nonempty structures.

Example 6. For digraphs (see Example 1), connectivity in the sense we have
just introduced corresponds to weak connectivity in graph theory. The definition of
strong connectivity for digraphs can be found in Exercise 5. For undirected graphs,
connectivity in the sense introduced above coincides with the notion of connectivity
from graph theory. 4

1.1.7. Direct products. Let A and B be τ -structures. Then the (direct, or

categorical ) product A × B is the τ -structure C with domain A × B such that
• for each k-ary R ∈ τ we have ((a1 , b1 ), . . . , (ak , bk )) ∈ RC if and only if
(a1 , . . . , ak ) ∈ RA and (b1 , . . . , bk ) ∈ RB ;
• for each k-ary f ∈ τ

f C ((a1 , b1 ), . . . , (ak , bk )) = (f (a1 , . . . , ak ), f (b1 , . . . , bk )).

The direct product A × A is also denoted by A2 , and the k-fold product A × · · · × A,

defined analogously, by Ak (it is straightforward to define this also for infinite k).
10 1. PERMUTATION GROUPS

1.1.8. Congruences. Let A be a structure with a purely functional signature

τ . A congruence of A is an equivalence relation E on A such that for every k-ary
f ∈ τ the function f A is a homomorphism from (A; E)k to (A; E).
Example 7. Let G be a permutation group on a set A, and let A be a G-set.
Recall that A is a structure with a purely functional signature (all functions are
unary). Then the congruences of A are equivalence relations on A that are preserved
by all permutations in G. Such equivalence relations are also called congruences of
G, and an important topic of this chapter (see in particular Section 1.4). 4
Example 8. Let G be a group. Then there is a natural bijection between the
congruences of G and the normal subgroups of G; this is treated in Section 4.6. 4

Exercises.
(1) Show that a function f : A → A preserves h : A → A if and only if
f preserves the graph of h, i.e., the binary relation
{(a, h(a)) | a ∈ A}.
(2) Show that isomorphic structures have isomorphic automorphism
groups, but that the converse is false.
(3) Prove Proposition 1.1.1.

(4) Consider the following structures.

Γ1 := (Q; {(x, y) : x = y + 1})


Γ2 := Q; (x, y, u, v) : x − y = u − v ∈ {1, −1}
Γ3 := (Q; {(x, y) : |x − y| = 1})
Show that
{idQ } ( Aut(Γ1 ) ( Aut(Γ2 ) ( Aut(Γ3 ) ( Sym(Q).
(5) A directed graph G (Example 2) is called strongly connected
if for any a, b ∈ G there exists a sequence c0 , c1 , . . . , cn
with c0 = a, cn = b, and (ci , ci+1 ) ∈ E G for all i ∈ {0, . . . , n − 1}.
Show that if G is finite and connected (in the sense of
Example 6) and Aut(G) is transitive, then G is strongly connected.
(6) Show that the previous exercise is false in general
for infinite digraphs G.

1.2. Automorphism Groups

Let G be a permutation group on a set A. When is G the automorphism group
of a structure with domain A? This has the following elegant answer.
Definition 1.2.1. We say that S ⊆ Sym(A) is (locally) closed (or closed in
Sym(A)) if it contains all f ∈ Sym(A) with the property that for all finite F ⊆ A
there exists a g ∈ S such that f (x) = g(x) for all x ∈ F .
Proposition 1.2.2. A permutation group G on a set A is the automorphism
group of a relational structure with domain A if and only if G is closed in Sym(A).
In the proof of this proposition, the following concept is useful. We write sInv(G)
for the set of all relations over A that are strongly preserved by all permutations
g ∈ G, i.e., both g and g −1 preserve the relation.
 1.2. AUTOMORPHISM GROUPS 11

Definition 1.2.3 (Canonical structure). A relational structure with domain A

whose relations are exactly the relations from sInv(G) is called a canonical structure
for G.
Proof of Proposition 1.2.2. For the forwards implication, suppose that G =
Aut(A) and that f ∈ Sym(A) \ G. Then f or f −1 does not preserve a relation R or
function from A. Suppose that (a1 , . . . , an ) ∈ RA but (f (a1 ), . . . , f (an )) ∈ / RA . Then
there is no g ∈ G such that g(x) = f (x) for all x ∈ {a1 , . . . , an }. The proof for f −1
is analogous.
For the reverse implication, let A be a canonical structure for G. Clearly, every
g ∈ G is an automorphism of A. Conversely, let f ∈ Aut(A). By assumption, to
show that f ∈ G, it suffices to show that for every finite tuple (a1 , . . . , an ) of elements
from X there exists an g ∈ G such that f (x) = g(x) for all x ∈ {a1 , . . . , an }. The
relation R := {(g(a1 ), . . . , g(an )) | g ∈ G} is preserved by all operations in G and
hence belongs to the relations of A. Thus, f preserves R. Also id ∈ G, and therefore
(a1 , . . . , an ) ∈ R, and so R contains (f (a1 ), . . . , f (an )) = (g(a1 ), . . . , g(an )) for some
g ∈ G. We therefore have G = Aut(A) as desired. 
1.2.1. The topology of pointwise convergence. The word closed suggests
a topology, and indeed there is corresponding topology on G, called the topology of
pointwise convergence. Topological aspects will be treated properly in our chapter on
topological groups, Chapter 4. However, we already give some of the basic topological
definitions now, specialised to the topology of pointwise convergence on Sym(A),
which is the topology we will be working with in the following sections. A basic open
set is a subset of Sym(A) of the form S(a, b) := {g ∈ Sym(A) | g(a) = b} where
a, b ∈ An for some n ≥ 1, and g(a) := (g(a1 ), . . . , g(an )). A subset of Sym(A) is open
if it is a union of basic open sets.
Proposition 1.2.4. The open subsets of Sym(A) define a topology. A subset of
Sym(A) is closed (in the sense of Definition 1.2.1) if and only if it is the complement
of an open set.
Proof. The empty set and Sym(A) are clearly open. The intersection of two
basic open S(a, b) and S(a0 , b0 ) equals S((a, a0 ), (b, b0 )), again a basic open set. For the
second statement, let S ⊆ Sym(A). Then S is closed if and only if C := Sym(A) \ S
can be written as
[
C= S(a, b)
n∈N,a,b∈An s.t. ∀g∈S:g(a)6=b

and the latter is a union of basic open sets. 

Closed permutation groups on a countable set have at most 2ω many elements.
But they cannot have arbitrary cardinalities smaller than 2ω , which we can prove
even without assuming the continuum hypothesis. To phrase our fundamental result
about the cardinalities of permutation groups on a countable set, we need the following
definition.
Definition 1.2.5 (point stabiliser). Let G be a permutation group over the base
set A and let B ⊆ A. Then the point stabiliser G(B) is the subgroup of G consisting
of all α ∈ G such that α(b) = b for all b ∈ B. If b = (b1 , . . . , bn ) is a sequence of
elements of A, then we also write Gb instead of G({b1 ,...,bn }) .
Theorem 1.2.6 (Corollary 4.1.5 in [73]). Let G ≤ Sym(N) be closed. Then the
following are equivalent.
(1) There is an a ∈ Nn , n ∈ N such that |Ga | = 1.
12 1. PERMUTATION GROUPS

(2) |G| ≤ ω.
(3) |G| < 2ω .

Proof. For the implication from (1) to (2), let b̄1 , b̄2 , . . . be an enumeration of
the tuples b̄ ∈ Nn such that there exists gb̄ ∈ G with gb̄ (ā) = b̄. Then every g ∈ G can
be written as gb̄i ◦ h for some i ∈ N and some h ∈ Gā .
The implication from (2) to (3) is trivial.
For the implication from (3) to (1), suppose that ¬(1). We define inductively
sequences a0 , a1 , a2 , . . . and b0 , b1 , b2 , . . . of tuples of elements of N and a sequence
g0 , g1 , g2 , . . . of elements of G such that for every i ∈ N

(1) gi (bi ) = bi ,
(2) gi (ai ) 6= ai ,
(3) ai contains i as an entry,
(4) bi+1 is the concatenation of all sequences ki ◦ · · · ◦ k0 (a0 , . . . , ai ) where kj ∈
{gj , 1} for all j ∈ {0, . . . , k}.

Initially, b0 := (). If bi has been chosen for i ∈ N, we have by assumption that

|Gbi | ≥ 2 and hence there is some gi ∈ G which fixes bi but is not the identity, so
there is a tuple ai such that gi (ai ) 6= ai , which gives us (1) and (2). We may add
the entry i to ai to ensure (3). Then item (4) determines bi+1 and this concludes the
construction of the sequences.
For any subset S ⊆ N \ {0} and i ∈ N, define

(
gi i∈S
giS :=
1 i∈
/S
fiS := giS ◦ · · · ◦ g1S ◦ g0S .

For each j ≥ i we have fjS (ai ) = fiS (ai ) by the properties (1) and (4). In particular,
(3) implies that fjS (i) = fiS (i). Define hS : N → N by hS (i) := fiS (i) for every i ∈ N.
To see that hS is surjective, let i ∈ N and put j := (fiS )−1 (i). If j ≤ i then

hS (j) = fjS (j) = fiS (j) = fiS ((fiS )−1 (i)) = i.

If j > i, then

hS (j) = fjS ((fiS )−1 (i)) = gjS ◦ · · · ◦ gi+1

S
(i) = i.

We claim that for every finite F ⊆ N there exists g ∈ G such that hS (x) = g(x) for all
x ∈ F . Let i := max (F ). Then note that for every x ∈ F we have hS (x) = fxS (x) =
fiS (x) and fiS ∈ G. In particular, hS is injective. Since G is closed in Sω , we have
that hS ∈ G.
It remains to show that hS 6= hT whenever S 6= T . Let i > 0 be smallest which
S
is, say, in S but not in T . Let j ≥ i be larger than all the entries in (fi−1 )−1 (ai ).
 1.2. AUTOMORPHISM GROUPS 13

Then
hS ((fi−1
S
)−1 (ai )) = fjS ((fi−1
S
)−1 (ai ))
= gjS ◦ · · · ◦ gi+1
S
◦ giS (ai )
= giS (ai ) (by (4) and (1)
= gi (ai ) (since i ∈ S)
6= ai (by (2)
= gjT ◦ · · · ◦ gi+1
T
◦ giT (ai ) (since i ∈
/ T)
= fjT (fi−1
T
)−1 (ai )
= fjT (fi−1
S
)−1 (ai ) (by the choice of i)
= hT ((fi−1
S
)−1 (ai )). 
1.2.2. Aut-sInv. Recall that the automorphism group of a relational structure
A, i.e., the set of all automorphisms of A, is denoted by Aut(A). In the following it
will be convenient to define the operator Aut also on sets R of relations over the same
domain A, in which case Aut(R) denotes the set of all permutations p of A such that
p and its inverse p−1 preserve all relations form R.
For P ⊆ Sym(A), and sets R of relations over the domain A, we present a
description of the closure operator P 7→ Aut(sInv(P )); the closure operator R 7→
sInv(Aut(R)) will be described in Section 3.1.
Definition 1.2.7. For P ⊆ Sym(A), we define
• hP i, the permutation group generated by P , to be the smallest permutation
group on A that contains P .
• P , the closure of P in Sym(A), to be the smallest closed subset of Sym(A)
that contains P .
Example 9. Let P be the set of permutations f of N that have finite support,
that is, the set {i ∈ N | f (i) 6= i} is finite. Then P ( P = Sym(N). 4
Proposition 1.2.8. Let P ⊆ Sym(A) be arbitrary. Then Aut(sInv(P )) = hP i
equals the smallest permutation group that contains P and is closed in Sym(A).
Proof. Let P 0 be the smallest permutation group that contains P and is closed
in Sym(A). Since P ⊆ P 0 and P 0 is a permutation group, we must have hP i ⊆ P 0 , and
therefore also hP i ⊆ P 0 since P 0 is closed in Sym(A). To show the converse inclusion
P 0 ⊆ hP i, it suffices to verify that hP i is a closed subgroup of Sym(A). Since hP i is
clearly closed in Sym(A) we only have to show that hP i contains compositions and
inverses. We do the verification for closure under compositions on finite subsets F of
A. Indeed, when f, g ∈ hP i, then there are f 0 , g 0 ∈ hP i such that f (x) = f 0 (x) for all
x ∈ F and g(x) = g 0 (x) for all x ∈ f (F ). We therefore have g(f (x)) = g 0 (f 0 (x)) for
all x ∈ F , and hence g ◦ f ∈ hP i, as desired.
We now show that hP i ⊆ Aut(sInv(P )). Let p ∈ hP i be arbitrary, and let R be
from sInv(P ). We have to show that p and p−1 preserve R. Let t ∈ R; we have that
p(t) = q1 ◦ · · · ◦ qk (t) for some permutations q1 , . . . , qk ∈ P ∪ P −1 . Since q1 , . . . , qk
preserve R, we have that q(t) ∈ R. The argument for p−1 is analogous.
Finally, we show Aut(sInv(P )) ⊆ hP i. Let p be from Aut(sInv(P )). It suffices
to show that for every finite subset {a1 , . . . , an } of A there is a q ∈ hP i such that
p(ai ) = q(ai ) for all i ≤ n. Consider the relation {(q(a1 ), . . . , q(an )) | q ∈ hP i}. It is
preserved by all permutations in P . Therefore, p preserves this relation, and so there
exists q ∈ hP i as required. 
14 1. PERMUTATION GROUPS

Exercises.
(7) Let G be the permutation group on Z that consists of all shift
operations {x 7→ x + c | c ∈ Z}. Is G closed?
(8) Let G be the permutation group on Z that is generated by the
transpositions τi := (i, −i), for i ∈ Z. What is the cardinality
of G, and what is the cardinality of G?
(9) Let P = {f, g} ⊆ Sym(Z) where f is a transposition and g is
x 7→ x + 1. Determine the cardinalities of hP i, P , and hP i.
(10) The finitary alternating group A on N is the set of all permutations
of N that can be written as a composition of an even number
of transpositions. Determine A.
(11) Let G, H ≤ Sym(A). Show that if H is closed, then G ≤ H
if and only if every relation R ⊆ An which is preserved by H
is also preserved by G.
1.3. Group Actions
We now consider abstract groups, that is, algebraic structures G over a set G
of group elements, with a function symbol for multiplication of group elements, a
function for the inverse of a group element, and the constant for the identity (see
Example 3). The link to permutation groups is given by the concept of an action of
such a group on a set, which is described below.
Definition 1.3.1. Let G be a group and X a set. An action of G on X is a
homomorphism φ from G to Sym(X). An action φ is called faithful if φ is injective.
Example 10 (The componentwise action). If G is a permutation group on a set
X and n ∈ N, then the componentwise action of G on X n is given by
ξ(g)(x1 , . . . , xn ) := (g(x1 ), . . . , g(xn )).
Note that this action is faithful unless n = 0. 4
Example 11.
If G is a permutation group on a set X and n ∈ N, then the setwise
action of G on X
n is given by
ξ(g)({x1 , . . . , xn }) := {g(x1 ), . . . , g(xn )}.
If and n > 0, then this action is faithful; this follows e.g. from the argument given in
Example 76 in Chapter 5. 4
Clearly, to every action of G on X we can associate a permutation group as
considered before, namely the image of the action in Sym(X). Conversely, to every
permutation group G on a set X we can associate an abstract group G whose domain
is G (the permutations), where composition and inverse are defined in the obvious
way, and which acts on X faithfully by φ(g) := g.
In this way we can also use other terminology introduced for permutation groups
(such a transitivity, congruences, primitivity, etc.) for group actions. For instance,
we say that an action ξ : G → Sym(X) is transitive if the permutation group ξ(G) ≤
Sym(X) is transitive. We give an alternative characterisation of action which in many
texts is taken to be the official definition.
Proposition 1.3.2. Let G be a group and X a set. The φ : G → Sym(X) is an
action of G on X if and only if the map · : G × X → X defined by g · x := φ(g)(x)
satisfies
• (gh) · x = g · (h · x) for all g, h ∈ G and x ∈ X, and
• 1 · x = x for every x ∈ X.
 1.3. GROUP ACTIONS 15

The action φ is faithful if and only if for any two distinct g, h ∈ G there exists an
x ∈ X such that g · x 6= h · x.
Proof. The proof is just moving symbols. 

If x ∈ X then the orbit of x with respect to an action of G on X is the set

G · x := {g · x | g ∈ G}. An orbit of k-tuples is an orbit of the componentwise action
of G on X k (Example 10).
Cayley’s theorem states that every group has a representation as a permutation
group.
Theorem 1.3.3 (Cayley’s theorem). Let G be any group. Then G has a faithful
action on G.
Proof. The action ξ on G is by left translation: for g ∈ G, we define ξ(g) by
ξ(g)(h) := gh
for all h ∈ G. It is straightforward to verify that this map is an injective group
homomorphism. 

We close this section with some important examples of group actions.

Example 12 (Action by left translation). A left coset of a subgroup H of G is a
set of the form {gh | h ∈ H} for g ∈ G, also written gH. Clearly, the set of all left
cosets of H partitions G, and is denoted by G/H. The cardinality of G/H is called the
index of H in G. We define an action ξ of G on G/H by setting ξ(f )(gH) := (f g)H.
This action is also called the action of G on G/H by left translation. It is transitive
since for any g1 H, g2 H ∈ G/H the map ξ(g2 g1−1 ) takes g1 H to g2 H. Note that this
example generalises the construction in the proof of Cayley’s theorem since we may
take H = {1}. 4
Example 13 (Action by conjugation). The map ξ : G → G given by
ξ(g)(h) := ghg −1
is called the action of G on G by conjugation. First note that for every g ∈ G
the operation ξ(g) is in fact an automorphism of G, because it preserves the group
structure:
ξ(g)(h1 h2 ) = g(h1 h2 )g −1 = gh1 g −1 gh2 g −1 = ξ(g)(h1 ) · ξ(g)(h2 ).
In fact, ξ is a homomorphism from G to Aut(G), because for all g1 , g2 ∈ G and h ∈ G
we have
ξ(g1 g2 )(h) = (g1 g2 )h(g1 g2 )−1
= g1 (g2 hg2−1 )g2−1
= ξ(g1 )(ξ(g2 )h).
The orbits of this action are called the conjugacy classes, and the stabiliser of h ∈ G
with respect to this action is the centraliser CG (h) of h:
CG (h) = {g ∈ G | gh = hg}. 4

Exercises.
(12) Show that if (a1 a2 . . . )(b1 b2 . . . ) . . . is the cycle representation of g ∈ Sn , and
f ∈ Sn , then (f (a1 )f (a2 ) . . . )(f (b1 )f (b2 ) . . . ) . . . is the cycle representation
of f −1 gf .
16 1. PERMUTATION GROUPS

(13) Let H 1 and H 2 be subgroups of Sym(X). Show that H 1 and H 2 are iso-
morphic as permutation groups if and only if there exists f ∈ Sym(X) such
that
H1 = {f hf −1 | h ∈ H2 }.
(14) Let G be a group with an action φ on A, and let A be
the corresponding G-set, i.e., the structure with domain A
which contains a unary operation for every permutation
in the image of φ. Show that the domains of the substructures
of A are precisely the orbits of the action of G on A.
(15) (Exercise 1 on page 9 of [39]) Let H a subgroup of G.
When is the the action φ of G on G/HT by left translation faithful?
Hint. Show that the kernel of φ is g∈G g −1 Hg.
(16) Let k ≥ 1 and n > k. Show that the setwise action of Sn
on k-element subsets of {1, . . . , n} is faithful.

In the following we review the classical theory how permutation groups can be
built from simpler permutation groups forming various forms of products. The direct
product of a sequence of groups (Gi )i∈I is the product of this sequence as defined in
general in Section 1.1.6; note that the product is again a group. Products appear in
several ways when studying permutation groups; the first is when we want to describe
the relation between a permutation group and its ‘transitive constituents’, described
in the following.

1.3.1. The intransitive action of the direct product. If G acts on a set X

and O ⊂ X is an orbit with respect to this action, then G naturally acts transitively
on O by restriction; we call the corresponding group H the group induced by O, or a
transitive constituent.
Proposition 1.3.4 (see [39]). Let G be a permutation group on X and let I be the
set of orbits of G. Then for every O ∈ I let G[O] := {g|O : g ∈ G} be the
Q permutation
group induced by G on O. Then G is isomorphic to a subgroup of O∈I GO , and
g 7→ g|O is a surjective homomorphism from G to G[O] for each O ∈ I.
Proof. Straightforward from the definitions. 

Definition 1.3.5. Let G1 and G2 be groups acting on disjoint sets X and Y ,

respectively. Then the action of G1 × G2 on X ∪ Y defined by (g1 , g2 ) · z = g1 z if
z ∈ X, and g2 z if y ∈ Y , is called the natural intransitive action of G1 ×G2 on X ∪Y .
If G1 and G2 are the automorphism groups of relational structures A and B with
disjoint domains A and B and signatures τ1 and τ2 , respectively, then the image of the
natural intransitive action on A∪B (as a homomorphism from G1 ×G2 to Sym(A∪B))
can also be described as the automorphism group of a relational structure C: we can
take for C the structure with domain A ∪ B and with signature τ1 ∪ τ2 ∪ {P }, where
P C = A, RC = RA for R ∈ τ1 \ τ2 , and RC = RB for R ∈ τ2 \ τ1 , and RC = RA ∪ RB
if R ∈ τ1 ∩ τ2 .
Exercises.
(17) Give an example of two structures A and B that illustrates why
we need the extra unary predicate in the definition of the structure
C above. In other words: show that if we view A and B as
(τ1 ∪ τ2 )-structures, then the automorphism group of the disjoint
union A]B is in general not the same as the image of the intransitive action
of G1 × G2 on A ∪ B.
 1.3. GROUP ACTIONS 17

(18) Prove Proposition 1.3.4.

1.3.2. The product action. When G1 is a group acting on a set X, and G2 a

group acting on a set Y , there is another important natural action of G := G1 × G2
besides the intransitive natural action of G, which is called the product action of G.
In this action, G acts on X × Y by (g1 , g2 ) · (x, y) = (g1 x, g1 y). If the actions of G1
and G2 are transitive, then the product action is clearly transitive, too.
When G1 and G2 are the automorphism groups of structures A and B, then the
image of the product action of G in Sym(A × B) is the automorphism group of the
following structure, which we call the full product structure of two relational structures
A and B, and denote by A  B. Let σ be the signature of A, and τ be the signature of
B; we assume that σ and τ are disjoint, otherwise we rename the relations so that the
assumption is satisfied. For each k-ary R ∈ σ, the structure AB contains the relation
{((a1 , b1 ), . . . , (ak , bk )) | (a1 , . . . , ak ) ∈ RA , b1 , . . . , bk ∈ B}, and for each k-ary R ∈ τ ,
it contains the relation {((a1 , b1 ), . . . , (ak , bk )) | (b1 , . . . , bk ) ∈ RB , a1 , . . . , ak ∈ A}.
Finally, we also add the relations P1 = {((a1 , b1 ), (a2 , b2 )) | a1 = a2 } and P2 =
{((a1 , b1 ), (a2 , b2 )) | b1 = b2 } to A  B.
Proposition 1.3.6. The automorphism group of C := A  B is G1 × G2 in its
product action on A × B.
Proof. Let h be the product action of G = G1 × G2 on A × B, viewed as a
homomorphism from G to Sym(A × B). Let (g1 , g2 ) be an element of G. Then
h((g1 , g2 )) is the permutation (x, y) 7→ (g1 x, g2 y) of A × B, and this map preserves
C: when ((a1 , b1 ), . . . , (ak , bk )) ∈ RC , for R ∈ σ, then (a1 , . . . , ak ) ∈ RA , and so
(g1 a1 , . . . , g1 ak ) ∈ RA . Therefore, ((g1 a1 , g2 b1 ), . . . , (g1 ak , g2 bk )) ∈ RC . The proof
for the relation symbols R ∈ τ is analogous.
We now show that conversely, every automorphism g of C is in the image of
h. Let Note that P1 and P2 are congruences of the automorphism group of C. Fix
elements a0 ∈ A, b0 ∈ B. Let g1 be the permutation of A that maps a ∈ A to the
point a0 such that g((a, b0 )) = (a0 , b0 ). Similarly, let g2 be the permutation of B that
maps b ∈ B to the point b0 such that g((a0 , b)) = (a0 , b0 ). Since g preserves P1 , P2 , the
definition of g1 and g2 does not depend on the choice of a0 and b0 . Moreover, g1 is
from G1 , since g preserves the relations for the symbols from σ. Similarly, g2 is from
G2 . Then g 0 := h((g1 , g2 )) equals g, since g 0 ((a, b) = (g1 a, g2 b) = g(a, b). Hence, g is
a permutation of A × B that lies in the image of h. 
Note that Proposition 1.3.6 becomes false in general when we omit the relations
P1 and P2 in A  B (consider for example the countably infinite structure without
structure B (that is, B has the empty signature).
Finally we remark that Aut((A  B)  C) and Aut(A  (B  C)) are isomorphic
as permutation groups. We explicitly define the d-fold full product as follows.
Definition 1.3.7 (Full product of d structures). Let B 1 , . . . , B d be structures with
disjoint relational signatures τ1 , . . . , τd . We denote by B 1  · · ·  B d the structure
with domain B := B1 × · · · × Bd that contains for every i ≤ d, and every m-ary
R ∈ (τi ∪ {=}) an m-ary relation defined by
{((x11 , . . . , xd1 ), . . . , (x1m , . . . , xdm )) ∈ B m | (xi1 , . . . , xim ) ∈ RB i } .
If B := B 1 = · · · = B k , then we first rename R ∈ τi into Ri so that the factors have
pairwise disjoint signatures, and then write B [d] for B 1  · · ·  B d .

Exercises.
18 1. PERMUTATION GROUPS

(19) Let G = K × H. Prove that G has a subgroup K ∗ isomorphic to K and a

subgroup H ∗ isomorphic to H such that
(a) G = K ∗ H ∗ := {kh | k ∈ K ∗ , h ∈ H ∗ }
(b) K ∗ ∩ H ∗ = {1};
(c) kh = hk for all k ∈ K ∗ and h ∈ H ∗ .
(20) Prove that the previous exercise provides a characterisation of groups that
are direct products: if G is a group with subgroups K and H, then the map
K × H → G given by (k, h) 7→ kh is an isomorphism between K × H and
G if and only if the three items from the previous exercise hold for K = K ∗
and H = H ∗ .

1.4. Congruences and Primitivity

Recall from Section 1.1.8 that a congruence of a permutation group G on a set
D is an equivalence relation on D that is preserved by all permutations in G. The
equivalence classes of a congruence are also called congruence classes.
Definition 1.4.1. Let G be a permutation group on a set D. A subset S of D is
called a block of G if g(S) = S or g(S) ∩ S = ∅ for every g ∈ G.
Lemma 1.4.2. Let G be a permutation group on a set D. Then S ⊆ D is a block
of G if and only if S is a congruence class of a congruence of G.
Proof. Suppose that S is an equivalence class of the congruence C, and suppose
that g(S) ∩ S contains an element t. That is, there is an element s ∈ S such that
g(s) = t ∈ S. We will show that g(S) = S. Arbitrarily choose r ∈ S. Then (r, s) ∈ C,
and hence (g(r), g(s)) = (g(r), t) ∈ C. Since t ∈ S, it follows that g(r) ∈ S. Hence,
g(S) ⊆ S. But also (r, t) ∈ C, and (g −1 (r), g −1 (t)) = (g −1 (r), s) ∈ C. Since s ∈ S, it
follows that g −1 (r) ∈ S and r ∈ g(S). Hence, S ⊆ g(S).
Now suppose that S ⊆ D is a block of G. Define
C := {(x, y) | ∃g ∈ G : g(x), g(y) ∈ S} ∪ {(x, x) | x ∈ D} .
This relation is clearly reflexive, symmetric, and preserved by G. In order to prove
that C is a congruence it remains to verify transitivity. Let (x, y), (y, z) ∈ C. Then
there are g1 , g2 ∈ G such that g1 (x), g1 (y), g2 (y), g2 (z) ∈ S. Thus, S ∩ g2 (g1−1 (S))
contains g2 (y), and (g2 ◦ g1−1 )(S) = S because S is a block. Since g1 (x) ∈ S this
implies that g2 (g1−1 (g1 (x))) ∈ S. So g2 (x) ∈ S and g2 (z) ∈ S, and (x, z) ∈ C. 
The relation {(x, x) | x ∈ D} is a congruence of every G ≤ Sym(D), and called the
trivial congruence. A congruence is called proper if it is distinct from the equivalence
relation that has only one equivalence class.
Definition 1.4.3. A transitive permutation group G is called primitive if every
proper congruence of G is trivial, and imprimitive otherwise.
Note that if every proper congruence of G is trivial, then it is necessarily transitive,
except in the case where G is a permutation group on a two-element set.
Definition 1.4.4. An orbital is an orbit of pairs, that is, a set of the form
{(g(a), g(b)) | g ∈ G} for a, b ∈ D. If O is an orbital, the orbital digraph of O is the
directed graph with vertex set D and edges O.
Example 14. Let G be the permutation group on {1, 2, 3, 4} generated by (1234).
Then G has four orbitals, depicted in Figure 1.1. 4
In a transitive permutation group on D, the trivial orbital is the orbital
∆D := {(a, a) | a ∈ D}.
 1.4. CONGRUENCES AND PRIMITIVITY 19

Figure 1.1. The four orbitals of the transitive permutation group

generated by (1234).

Theorem 1.4.5 (Higman’s theorem). A transitive permutation group G on a (not

necessarily finite) set X is primitive if and only if the orbital digraph of all non-trivial
orbitals is connected.
Proof. First suppose that G is primitive. Let O be a non-trivial orbital and let
S ⊆ X be a connected component of the orbital digraph of O. Clearly, g(S) ∩ S = ∅
or g(S) ∩ S = S for every g ∈ G, because g preserves O, and hence S is a block. If
S is a singleton set, then by the transitivity of G all components of X are singletons,
contrary to the assumption that O is non-trivial. Hence, the primitivity of G implies
that S = X (Lemma 1.4.2), which shows that the orbital digraph of O is connected.
Now suppose that the orbital digraph of every non-trivial orbital is connected.
Let C be a congruence such that there are distinct a, b ∈ X with (a, b) ∈ C. By
assumption, the orbital digraph of the orbital that contains (a, b) is connected. Hence,
the transitivity of the relation C implies that C has at most one congruence class.
We conclude that G is primitive. 
Yet another perspective on congruences, blocks, and primitivity of transitive per-
mutation groups G involves the following definition.
Definition 1.4.6 (set stabiliser). Let G be a permutation group on X. If S ⊆ X
then the subgroup
G{S} := {g ∈ G | g(S) = S}
is called that set stabiliser of G as S.
Note that the set stabiliser is a point stabiliser (see Definition 1.2.5) for the action
of G on |S|-element subsets of X (Example 11). The following is Theorem 1.5A in [48].
Theorem 1.4.7. Let G be a group which acts transitively on a set D, and let
a ∈ D. Let B be the set of all blocks B of G that contain a, and let S be the set of
all subgroups H of G that contain Ga . Then there is a bijection µ between B and S
given by µ(B) := G{B} ; the inverse mapping is given by µ−1 (H) = {g(a) | g ∈ H}.
Note that the mapping µ is order-preserving in the sense that if B1 , B2 ∈ B then
B1 ⊆ B2 ⇔ µ(B1 ) ⊆ µ(B2 ).
Corollary 1.4.8. Let G be a group acting transitively on a set D with at least
two elements. Then G is primitive if and only if each point stabiliser Ga , for a ∈ D,
is a maximal subgroup of G.
Proof. The statement follows from Theorem 1.4.7. For a self-contained proof,
suppose that a ∈ D is such that Ga is not maximal, i.e., there exists a proper subgroup
H of G that properly contains Ga . Then B := H(a) := {h(a) | h ∈ H} is a block of
G. To see this, let g ∈ G. Suppose that g(B) ∩ B 6= ∅. Then there exist h1 , h2 ∈ H
such that g(h1 (a)) = h2 (a). Hence, h−12 ◦ g ◦ h1 ∈ Ga , so g ∈ h2 Ga h1
−1
⊆ H. It
follows that g(B) = B for all g ∈ G, showing that B is a block. Next, observe that
GB = H since for every g ∈ G we have g(B) = B if and only if g ∈ H, as we have
seen in the previous paragraph. If B = D, then GB = G, contrary to the assumption
20 1. PERMUTATION GROUPS

that H = GB is a proper subgroup of G. If B = {a}, then GB = Ga , contrary to the

assumption that H = GB properly contains Ga . Hence, G must have a non-trivial
proper congruence by Lemma 1.4.2, and hence is not primitive.
Conversely, if G is not primitive, then by Lemma 1.4.2 it has a block B such that
1 < |B| and B 6= D. Let a ∈ B. Clearly, GB is a subgroup of G that contains Ga
(see Exercise 22). Also note that {g(a) | g ∈ GB } = B (see Exercise 23). Hence, if
GB = Ga then B = {a}, contrary to the assumptions on B. If GB = G then the
transitivity of G implies that B = D, contrary to the assumptions on B. Thus, GB
shows that Ga is not a maximal subgroup of G. 

Exercises.
(21) Let G ≤ Sn be primitive. Show that if G contains a transposition, then
G = Sn .
(22) Show that if B is a block of the permutation group G and a ∈ B, then Ga
is contained in GB .
(23) Show that if B is a block of a transitive permutation group G, then
{g|B : g ∈ GB }
is transitive as well.
(24) (from [48], Exercise 1.5.8) Let G ≤ S6 be the group generated by
{(123456), (26)(35)}.
Find all blocks that contain 1.
Find all subgroups of G that contain G1 .
(25) Give examples of permutation groups G ≤ S2n which cannot
be generated by fewer than n elements.
(26) (from [48], Exercise 1.5.14) Suppose that G ≤ Sn has r orbits.
Show that G can be generated by a subset of size n − r
(in particular, every permutation group on n elements
can be generated by n − 1 elements).

1.5. Semidirect Products

Semidirect products can be seen either as a way to construct new groups from
simpler ones (Section 1.5.3), or, equivalently, as a tool to decompose a given group into
simpler constituents (Section 1.5.4). They generalise the concept of direct products
of groups. We first introduce some fundamental concepts for (abstract) groups.

1.5.1. Normal subgroups. A subgroup N of G with domain N is called normal

if gN = N g for all elements g of G; in this case, we write N / G.
Example 15. If G1 and G2 are groups, then the direct product G1 × G2 has a
normal subgroup isomorphic to G1 with the elements {(g1 , 1) | g1 ∈ G1 }. 4
Recall the following equivalent characterisations of normality of subgroups.
Proposition 1.5.1. Let G be a group, and N be a subgroup of G. Then the
following are equivalent.
(1) N is normal.
(2) G has the congruence E = {(a, b) | ab−1 ∈ N }.
(3) There is a homomorphism h from G to some group such that N = h−1 (0).
(4) For every g ∈ G and every v ∈ N we have gvg −1 ∈ N .
 1.5. SEMIDIRECT PRODUCTS 21

Proof. (1) ⇒ (2): to verify that E is a congruence, we have to show that for
all (a1 , b1 ), (a2 , b2 ) ∈ E, (a1 a2 , b1 b2 ) ∈ E. Indeed, (a1 a2 )(b1 b2 )−1 = a1 (a2 b−1 −1
2 )b1 ∈
−1 −1
a1 N b1 = N a1 b1 ⊆ N N = N .
(2) ⇒ (3): g 7→ gN is a group homomorphism from G to G/N .
(3) ⇒ (4): For g ∈ G and v ∈ h−1 (0), we must show that gvg −1 ∈ h−1 (0).
Indeed, h(gvg −1 ) = h(g)h(v)h(g)−1 = h(g)0h(g)−1 = 0.
(4) ⇒ (1): assume that gN g −1 ⊆ N for all g ∈ G. Let a ∈ G be arbitrary.
Applying the assumption for g = a we find that aN ⊆ N a. Applying the assumption
for g = a−1 we find that a−1 N (a−1 )−1 = a−1 N a ⊆ N , and hence N a ⊆ aN . We
conclude that aN = N a. 
Example 16. The alternating group of degree n is the subgroup An of S n which
consists of all even permutations, i.e., the permutations that can be written as a
composition of an even number of transpositions. Then the map sgn that sends
g ∈ S n to 0 if g ∈ An , and to 1 otherwise, is a homomorphism from S n to Z2 and An
is a normal subgroup of Sn . 4
Non-trivial groups without non-trivial proper normal subgroups are called simple.
Exercises.
(27) Show that the group of permutations of N with finite support is a normal
subgroup of Sym(N).
(28) Show that AT5 has no proper non-trivial normal subgroups.
(29) Show that g∈G g −1 Hg is the largest normal subgroup of G which is con-
tained in H (also see Exercise 15).
1.5.2. Semidirect products: motivation. Let G be a group and let K and
H be subgroups of G. We have already defined the set-product KH := {kh | h ∈
K, k ∈ H} in Exercise 19. Note that KH might not be a subgroup (Exercise 30).
However, if H or K is a normal subgroup, then KH is a subgroup. For instance, if
K is a normal subgroup, then
(kh)(k 0 h0 ) = (khkh−1 )(hh0 ) ∈ KH (1)
−1 −1 −1 −1 −1 −1
and (kh) =h k = (h k h)h ∈ KH.
Example 17. Let G = S n , for n ≥ 3, let N be the normal subgroup An of S n
(see Example 16), and let H be the subgroup of G generated by the transposition
(12), i.e., H = {id, (12)}. Then G = N H, because every element g ∈ G is either in
An or can be written as (g(12))(12), which is in HN since g(12) ∈ An and (12) ∈ H.
Clearly, N ∩ H = {id}. However, S n is not isomorphic to An × Zn : for n ≥ 3, we
have
(123) ◦ (12) = (132) 6= (32) = (12) ◦ (123),
while
((123), 1)(1, (12)) = ((123), (12)) = (1, (12))((123), 1)
(see property (c) in Exercise 19). 4
Note the appearance of hkh−1 in (1), which defines a group action of H on K
(generalising Example 13). This group action is in fact a homomorphism from H to
Aut(K). Such homomorphisms from H to Aut(K) will be the starting point of our
first definition of semidirect products in the next section, which is in the setting where
we do not require that K and H are subgroups of the same group G (and which are
therefore called outer semidirect products).
Exercises.
22 1. PERMUTATION GROUPS

(30) Find an example of a group G and two subgroups H and K

such that HK is not a subgroup.

1.5.3. The outer semidirect product. Let H and N be groups and let θ : H →
Aut(N ) be a homomorphism.
Example 18. In our running example, H is Z2 and N is An for n ≥ 3 (see
Example 16). Note that for every t ∈ Sn the map αt : N → N given by g 7→ tgt−1
(conjugation) is from Aut(N ). Pick any t ∈ Sn \ An such that t2 = 1. Then the
map that sends 1 ∈ Z2 to αt and that sends 0 to idN is a homomorphism from H to
Aut(N ). 4
Definition 1.5.2. The semidirect product of N by H with respect to θ, de-
noted by N oθ H (or H nθ N ), is the group G with the elements N × H and group
multiplication defined by
(u, x)(v, y) := (uθ(x)(v), xy)
for all (u, x), (v, y) ∈ G. If the reference to θ is clear, we use o without the subscript.
Note that if θ is the trivial homomorphism that maps every element of H to
idN ∈ Aut(N ), then the semidirect product equals the direct product (we will see in
Exercise 31 that a converse of this statement is true as well). Definition 1.5.2 contains
some claims that we still have to verify. Multiplication is indeed associative:
((u, x)(v, y))(w, z) = (uθ(x)(v), xy)(w, z)
= (uθ(x)(v)θ(xy)(w), (xy)z)
= (uθ(x)(v)θ(x)(θ(y)(w)), x(yz)) (since θ is a homomorphism)
= (uθ(x)(vθ(y)(w)), x(yz)) (since θ(x) ∈ Aut(N ))
= (u, x)(vθ(y)(w), yz)
= (u, x)((v, y)(w, z)).
In the following, we write x(v) instead of θ(x)(v) for better readability. Clearly,
(1, 1) is a neutral element, and the inverse of (u, x) is (x−1 (u−1 ), x−1 ):
(u, x)(x−1 (u−1 ), x−1 ) = (ux(x−1 (u−1 )), xx−1 )
= (uu−1 , 1) = (1, 1)
Note that H ∗ := {(1, x) | x ∈ H} is a subgroup of G that is isomorphic to H, and
that N ∗ := {(u, 1) | u ∈ N } is a subgroup of G isomorphic to N . The next proposition
collects some further important properties of semidirect products (compare them with
the properties of direct products in Exercise 19!).
Proposition 1.5.3. Let G = N o H. Then
• G = N ∗H ∗,
• N ∗ ∩ H ∗ = {(1, 1)}, and
• N ∗ / G,
Proof. To see that G = N ∗ H ∗ it suffices to observe that (u, x) can be written as
(u, 1)(1, x), and obviously N ∗ ∩ H ∗ = {(1, 1)}. Finally, for (u, x) ∈ N ∗ and (v, y) ∈ G
we have
(u, x)(v, 1)(x−1 (u−1 ), x−1 ) = (ux(v), x)(x−1 (u−1 , x−1 )
= (ux(v)x(x−1 (u−1 )), xx−1 )
= (ux(v)u−1 , 1) ∈ N
 1.5. SEMIDIRECT PRODUCTS 23

which implies that N ∗ is a normal subgroup (Proposition 1.5.1). 

Note that the action of H ∗ on N ∗ by conjugation in G reflects the original action

of H on N , that is,
(1, x)(u, 1)(1, x)−1 = (x(u), x)(x−1 (1), x−1 )
= (x(u)x(x−1 (1)), xx−1 )
= (x(u), 1).
Usually, H ∗ and N ∗ are identified with H and N , so we then consider H and N
as subgroups of the semidirect product G = N o H.

1.5.4. The inner direct product. This section provides a characterisation of

the groups G that can be obtained as a semidirect product of two proper subgroups
of G.
A sequence of groups N , G, H with homomorphisms α : N → G and β : G → H
is called exact at G if the kernel of β equals the image of α. A sequence of groups
G1 , G2 , . . . with homomorphisms αi : Gi → Gi+1 is called exact if it is exact at Gi for
all i ≥ 2. A short exact sequence is an exact sequence of the form
α β
1 −→ N −→ G −→ H −→ 1.
Note that in this case, being exact at N implies that α is injective, and being exact
at H implies that β is surjective. Hence, N can be considered as a normal subgroup
of G and H is isomorphic to G/N . In this case G is called a group extension of N
(by H).

Example 19. Let G = S n , N = An , and H = Z2 . The inclusion map from

An to Sn is an injective homomorphism α : An → S n . Let β be the map sgn from
Example 16, which is a surjective homomorphism from S n to Z2 . 4

Proposition 1.5.4. Let G be a group, H ≤ G, and N / G. Then the following

are equivalent.
(1) H is a complement for N in G, i.e., G = N H := {nh | n ∈ N, h ∈ H} and
N ∩ H = {1G }.
(2) For every g ∈ G there exists a unique n ∈ N and h ∈ H such that g = nh.
(3) There is a homomorphism µ : G → H that fixes H pointwise and whose
kernel is N .
(4) The restriction of the factor map σ : G → G/N to H is an isomorphism
between H and G/N .
(5) There exists a short exact sequence
α β
1 −→ N −→ G −→ H −→ 1
that splits, i.e., there is a homomorphism ρ : H → G such that β ◦ ρ = idH .
(6) G is isomorphic to the semidirect product N oθ H where θ is the action of
H on N by conjugation in G.

Proof. (1) implies (2): Suppose that n1 , n2 ∈ N and h1 , h2 are such that n1 h1 =
n2 h2 . Then in particular, n1 H = n2 H, which is the case if and only if n−1 2 n1 ∈
N ∩ H = {1}, and hence n1 = n2 . Similarly we deduce that h1 = h2 .
(2) implies (3). Let µ : G → H be the function that maps g ∈ G to the unique
h ∈ H such that g = nh for some n ∈ N . Then µ is a homomorphism: if g1 = n1 h1
24 1. PERMUTATION GROUPS

and g2 = n2 h2
µ(g1 g2 ) = µ(n1 h1 n2 h2 )
= µ(n1 h1 n2 h1−1 h1 h2 )
= h1 h2 (since h1 n2 h−1
1 ∈ N)
= µ(n1 h1 )µ(n2 h2 )
= µ(g1 )µ(g2 ).
−1
Then µ (1) = N and for any h ∈ H we have µ(h) = µ(1h) = h.
(3) implies (4): The restriction of σ to H is a homomorphism from H to G/N .
It is injective since for u ∈ H we have σ(u) = 1G/N if and only if u ∈ N if and only if
µ(u) = 1H if and only if u = 1H . It is surjective since for every [g]N ∈ G/N we have
that µ(g) = µ(µ(g)), so [g]N = [µ(g)]N = σ(µ(g)).
(4) implies (5): let τ : H → G/N be the restriction of the factor map σ which
is an isomorphism by (4). Then β := τ −1 σ : G → H is a surjective homomorphism
whose kernel is N . Choosing ρ : H → G to be the inclusion map we obtain β ◦ ρ =
τ −1 σρ = idH .
(5) implies (6): we may assume that α and ρ are inclusion maps. Define θ : H →
Aut(N ) as n 7→ hnh−1 . We claim that N oθ H is isomorphic to G, the isomorphism
ξ being (n, h) 7→ nh. We verify that ξ is a homomorphism:
ξ((n1 , h1 )(n2 , h2 )) = ξ(n1 h1 n2 h−1
1 , h1 h2 )

= n1 h1 n2 h−1
1 h1 h2
= n1 h1 n2 h2 = ξ(n1 , h1 )ξ(n2 , h2 ).
The homomorphism ξ is injective: if ξ(n1 , h1 ) = n1 h1 = 1 then
1 = β(n1 h1 ) = β(n1 )β(h1 ) = 1 · β(h1 ),
and hence h1 = 1 since β is injective. Since n1 h1 = 1 this implies that n1 = 1, too.
To show that ξ is surjective let g ∈ G. We claim that N g = N β(g). It suffices to
show that gβ(g)−1 ∈ N , i.e., lies in the kernel of β. And indeed,
β(gβ(g)−1 ) = β(g)β(β(g))−1 ) = β(g)β(g)−1 = 1.
Hence, there exists n ∈ N such that g = nβ(g). Then ξ(n, β(g)) = nβ(g) = g which
shows that ξ is surjective.
(6) implies (1): this is Proposition 1.5.3. 
If the equivalent conditions in Proposition 1.5.4 apply, then G is called a split
extension of N (by H). We also say that G splits over N .
Example 20. Revisiting Example 19, we note that the short exact sequence
sgn
1 → An → Sn → Z2 → 0 splits: any homomorphism ρ from Z2 to Sn that maps 1
to an element t ∈ Sn \ An such that t2 = 1 satisfies sgn ◦ρ = idZ2 . Proposition 1.5.4
then implies that S n is isomorphic to An o Z2 . 4
Here is an example of a short exact sequence that does not split.
Example 21. Let G := (Z6 ; +). Then h : G → (Z2 ; +) given by h(g) := g
mod 2 is a surjective homomorphism, and there is an isomorphism i between Z3 and
the kernel N of h. We then have the short exact sequence
i h
1 −→ N −→ G −→ Z2 −→ 1.
However, there is no homomorphism r : Z2 → Z6 such that h ◦ r = idH since any non-
constant homomorphism s : Z2 → Z6 would have to map 1 to 3 since s(1) + s(1) = 0
 1.5. SEMIDIRECT PRODUCTS 25

implies that s(1) = 3, but then h ◦ s(1) = h(3) = 0 6= 1. So the sequence does not
split, and the equivalent conditions from Proposition 1.5.4 do not apply. 4

Exercises.
(31) Show that in a semidirect product N oθ H, the subgroup H is normal if and
only if θ : H → Aut(N ) is trivial (in the sense that it maps every h ∈ H to
idN ), and in this case N oθ H = N × H.
1.5.5. Application: the wreath product. We will now describe a natural op-
eration to construct new structures from known structures, and then describe how the
semidirect product helps to explain the automorphism groups of the new structures.
We start with a simple example.
Example 22. Let A the disjoint union of two copies of the 5-element clique
K 5 = ({1, 2, . . . , 5}; 6=). Note that G = Aut(A) has a normal subgroup N which is
isomorphic to S 5 × S 5 . Also note that G/N is isomorphic to Z2 , and that G is in fact
isomorphic to (S 5 )2 o Z2 . 4
Generalising Example 22, we may start from any two structures A and B with
disjoint relational signatures σ and τ . We will define a new structure A[B]; the idea
is that we replace the elements of A by copies of B.
Formally, we create a copy B a of B for every element a of A such that all the B a
have pairwise disjoint domains. Let E be a binary relation symbol that is not already
in σ ∪ τ . Then A[B] is the σ ∪ τ ∪ {E}-structure C defined as follows. The τ -reduct
of C equals the disjoint union of the B a . The relation E C is the equivalence relation
such that E(x, y) holds for x, y ∈ C if and only if x and y lie in the same copy of B
in C. For every relation symbol R ∈ σ of arity k we set
RC := {(c1 , . . . , ck ) | there is (a1 , . . . , ak ) ∈ RA and ci ∈ B ai for i ∈ {1, . . . , k}}.
In order to describe Aut(C) we need the following definition.
Definition 1.5.5 (Wreath product). Let G be a group and let H be a group
acting on a set A. Let N := GA . Note that for every h ∈ H and n ∈ N the map
(na )a∈A 7→ (nh−1 (a) )a∈A is an automorphism of N , and that the map θ that sends
h ∈ H to this automorphism is a homomorphism from H to Aut(N ). Define
G Wr H := N oθ H.
Proposition 1.5.6. Let G be a group acting on a set B and H be a group acting
on a set A. Then G Wr H has the following action on B × A:
(n, h) · (b, a) := (nh(a) (b), h(a)).
Proof. We verify the two conditions from Proposition 1.3.2. Let (b, a) ∈ B × A.
First note that
A
1G Wr H (b, a) = (1G , 1H ) · (b, a) = (1G (b), 1H (a)) = (b, a).
If (n, h), (n0 , h0 ) ∈ G Wr H, then
(n0 , h0 ) · ((n, h) · (b, a)) = (n0 , h0 ) · (nh(a) (b), h(a))
= (n0h0 (h(a)) (nh(a) (b)), h0 h(a))
= (n0h0 (h(a)) nh(a) (b), h0 h(a))
= (n0h0 (h(a)) θ(h0 )(n)h0 h(a) (b), h0 h(a))
= ((n0 θ(h0 )(n))h0 h(a) (b), h0 h(a))
= (n0 θ(h0 )(n), h0 h) · (b, a) = (n0 , h0 )(n, h) · (b, a). 
26 1. PERMUTATION GROUPS

Figure 1.2. A digraph; the task of Exercise 33 is to determine its

automorphism group. Undirected edges represent directed edges in
both directions.

If G and H are permutation groups, then we also use G Wr H for the permutation
group on B × A induced by this action. Note that if H acts on B with |B| > 1 and
|A| > 1, then the permutation group G Wr H is imprimitive, with block {(b, a) | b ∈
B} for every a ∈ A.
Proposition 1.5.7. For any two structures A and B we have
Aut(A[B]) = Wr(Aut(A), Aut(B)).
Exercises.
(32) Give an example of two structures A and B that illustrates why we need the
extra equivalence relation in the definition of the structure A[B].
(33) Describe the automorphism group of the digraph depicted in Figure 1.2.
(34) Show that there is no tree whose automorphism group is isomorphic to
(Z3 ; +). Hints.
• Show that every tree has a center, i.e., a vertex or an edge that is fixed
by every automorphism.
• Find an explicit description of point stabiliser of the automorphism
group of a tree.
 CHAPTER 2

Counting Orbits

It is natural to explore the theory of infinite permutation groups by starting with

large permutation groups. We first introduce several properties of permutation groups
that express certain aspects of ‘being large’. A permutation group G on a set A is
• k-transitive if for s, t ∈ Ak with pairwise distinct entries there is an g ∈ G
such that g(s) = t; (recall: the action of G on tuples is componentwise, i.e.,
g(s1 , . . . , sk ) := (g(s1 ), . . . , g(sk ))
• transitive if it is 1-transitive;
• k-set transitive if for all S, T ⊆ A of cardinality k there is a g ∈ G such that
g(S) = {g(s) | s ∈ S} = T .
• highly set-transitive if it is k-set transitive for all k ≥ 1.
• highly transitive if it is k-transitive for all k ≥ 1.
An example of a highly transitive permutation group is Sym(N). Clearly, if G is highly
transitive, then it is also highly set-transitive. An example of a highly set-transitive
but not highly transitive permutation group is Aut(Q; <); see Section 3.2.
It is easy to see that a 2-set transitive permutation group G on an infinite set is
also transitive. We prove the contraposition: assume that G has more than one orbit.
There must be an orbit O with two distinct elements c1 , c2 . Let c3 be an element not
from O. Then there is no automorphism that maps {c1 , c2 } to {c1 , c3 }, and hence G
is not 2-set transitive. This fact will be generalised by Proposition 2.1.1 below.

2.1. Two Integer Sequences

For B ⊆ A, the orbit of B under G is the orbit of B under the action of G on
subsets of A of cardinality |B| from Example 11, i.e., the set {g(a) | g ∈ G, a ∈ B}.
Proposition 2.1.1 (Cameron [39]). Let G be a permutation group on an infinite
set. The number fG (n) of orbits of n-subsets forms a non-decreasing sequence.
We will show this proposition in Section 2.3. Being highly set-transitive is equiv-
alent to fG (n) = 1 for all n ∈ N. There is another important sequence attached to a
permutation group. If A is a set and n ∈ N, we write An6= for the set of all n-tuples
with pairwise distinct entries from A.
∗
Definition 2.1.2. Let G be a permutation group. Then fG (n) denotes the number
n
of orbits of the componentwise action on A6= .
∗
So, G is highly transitive if fG (n) = 1 for all n ∈ N. Note that
∗
fG (n) ≤ fG (n) ≤ n!fG (n) (2)
since there are n! different orderings of n elements. These two sequences correspond
∗
to two different counting paradigms in combinatorics: labelled (in the case of fG ) and
unlabelled enumeration (in the case of fG ).

27
28 2. COUNTING ORBITS

Exercises.
(35) Show that Proposition 2.1.1 is false if G
is a permutation group on a finite set.
(36) Prove that (k + 1)-transitivity implies k-transitivity, for all k ≥ 1.
∗
(37) Show that if G is a permutation group on an infinite set, then fG (k) ≤
∗
fG (k + 1), for all k ≥ 1.
∗ ∗ ∗
(38) Show that if there exists a k such that fG (k) = fG (k+1), then fG (k+1) = 1.
(39) Show that a permutation group G on a set A is highly transitive if and only
if G = Sym(A) = Aut(A; =).
(40) Let (A; E) be a countably infinite structure where E denotes an equivalence
relation with infinitely many infinite classes. Describe the automorphism
group Aut(A; E). How many orbits of n-subsets are there?
(41) (Exercise 3 on page 57 in [39]) Let (A; E2 ) be a countably infinite
structure where E2 denotes an equivalence relation with infinitely
many classes of size two, and let (A; E 2 ) be a structure where E 2
denotes an equivalence relation with two infinite classes.
Show that Aut(A; E2 ) and Aut(A; E 2 ) have the same number
of orbits of n-subsets, for all n.

2.2. Combinatorial Tools

In order to prove Proposition 2.1.1, we need a couple of combinatorial tools.
2.2.1. The Pigeon-hole Principle. If n pigeons fly to less than n holes, there
must be one hole that got more than one pigeon. There is an important infinite
version of the statement: if infinitely many pigeons fly to finitely many holes, one
hole must have gotten infinitely many pigeons. This will be used in the next tools
that we present.
2.2.2. König’s Tree Lemma. A walk in a graph (V, E) (see Example 2) is
a sequence x0 , x1 , . . . , xn ∈ V with the property that {xi , xi+1 } ∈ E for all i ∈
{1, . . . , n − 1}. A walk is a path if all its vertices are distinct. A cycle is a walk of
length at least three of the form x0 , x1 , . . . , xn = x0 such that x1 , . . . , xn are pairwise
distinct. A tree is a connected graph (V, E) (see Section 1.1.6) without cycles. The
degree of a vertex u ∈ V is the number of vertices v ∈ V such that {u, v} ∈ E.
Lemma 2.2.1 (König’s Tree Lemma). Let (V, E) be a tree such that every vertex
in V has finite degree, and let v0 ∈ V . If there are arbitrarily long paths that start in
v0 , then there is an infinitely long path that starts in v0 .
Proof. Since the degree of v0 is finite, there exists a neighbour v1 of v0 such
that arbitrarily long paths start in v0 and continue in v1 (by the infinite pigeon-
hole principle). We now construct the infinitely long path by induction. Suppose
we have already found a sequence v0 , v1 , . . . , vi that can be continued to arbitrarily
long paths in (V, E). Since the degree of vi is finite, vi must have a neighbour vi+1 in
V \{v0 , v1 , . . . , vi } such that v0 , v1 , . . . , vi+1 can be continued to arbitrarily long paths
in (V, E). In this way, we define an infinitely long path v0 , v1 , v2 , . . . in (V, E). 
The degree assumption in Lemma 2.2.1 is necessary, as can be seen from Fig-
ure 2.1.
Proofs using König’s tree Lemma are often referred to as compactness arguments
– the link with topology will become clear in Section 4.1. The following proposition
illustrates one of the many uses of König’s tree Lemma.
 2.2. COMBINATORIAL TOOLS 29

Figure 2.1. A tree with arbitrarily long paths, but no infinite paths.

Proposition 2.2.2. A countably infinite graph G is 3-colourable if and only if

every finite subgraph of G is 3-colourable.
2.2.3. Ramsey’s Theorem. To prove Proposition 2.1.1, we also use an impor-
tant tool from combinatorics: Ramsey theory. We denote the set {0, . . . , n − 1} also
by
[n]. Subsets of a set of cardinality s will be called s-subsets in the following. Let
M M


s denote the set of all s-subsets of M . We also refer to mappings χ : s → [c] as
a coloring of M (with the colors [c]). In Ramsey theory, one writes
L → (m)sc
L


if for every χ : s → [c] there exists an M ⊆ L with |M | = m such that χ is constant
on M


s . In the following, ω denotes the cardinality of N. Note the following.
• For all n ∈ N we have [n + 1] → (2)1n : this is the pigeon-hole principle.
• For all c ∈ N we have N → (ω)1c : this is the infinite pigeon-hole principle.
We first state and prove a special case of Ramsey’s theorem.
Theorem 2.2.3. N → (ω)22 .
This statement has the following interpretation in terms of undirected graphs:
every countably infinite undirected graph either contains an infinite clique (a complete
subgraph) or an infinite independent set (a subgraph without edges).



Proof. Let χ : N2 → [2] be a 2-colouring of the edges of N2 . We define an
infinite sequence x0 , x1 , . . . of numbers from N and an infinite sequence V0 ⊇ V1 ⊇ · · ·
of infinite subsets of N. Start with V0 := N and x0 = 0. By the infinite pigeon-hole
principle, there is a c0 ∈ [2] such that {v ∈ V0 | χ(x0 , v) = c0 } =: V1 is infinite. We
now repeat this procedure with any x1 ∈ V1 and V1 instead of V0 . Continuing like
this, we obtain sequences (ci )i∈N , (xi )i∈N , (Vi )i∈N .
Again by the infinite pigeon-hole principle, there exists c ∈ [2] such that ci = c
for infinitely many i ∈ N. Then P := {xi | ci = c} has the desired property. To see
this, let i < j be such that xi , xj ∈ P . Then xj ∈ Vj ⊆ Vi and hence χ({xi , xj }) =
ci = c. 

We now state Ramsey’s theorem in it’s full strength; the proof is similar to the
proof of Theorem 2.2.3 shown above.
Theorem 2.2.4 (Ramsey’s theorem). Let s, c ∈ N. Then N → (ω)sc .
30 2. COUNTING ORBITS

A proof of Theorem 2.2.4 can be found in [73] (Theorem 5.6.1); for a broader
introduction to Ramsey theory see [63]. It is easy to derive the following finite version
of Ramsey’s theorem from Theorem 2.2.4 via König’s tree lemma.
Theorem 2.2.5 (Finite version of Ramsey’s theorem). For all c, m, s ∈ N there
is an l ∈ N such that [l] → (m)sc .
Proof. A proof by contradiction:
suppose that there are positive integers c, m, s
such that for all l ∈ N there is a χ : [l]
s
→ [c] such that (∗)[l] for all m-subsets M of [l]
the mapping χ is not constant on M s . We construct a tree as follows. The
vertices
are the maps χ : s → [c] that satisfy (∗)[l] . We make the vertex χ : [l]
[l]


s → [c]
adjacent to χ : [l+1] 0


s → [c] if χ is a restriction of χ . Clearly, every vertex in the
tree has finite degree. By assumption, there are arbitrarily long paths that start in
the vertex χ0 where χ0 is the map with the empty domain. By Lemma
2.2.1, the
tree contains an infinite path χ0 , χ1 , . . . We use this to define a map χN : Ns → [c] as
follows. For every n ∈ N, there exists a c0 ∈ [c] and an i0 ∈ N such that χi (S) = c0
for all S ∈ [n] and i ≥ i0 . Define χN (S) := c0 for all S ∈ [n]



s s . Then χN satisfies
(∗)N , a contradiction to Theorem 2.2.4. 
Here comes a variant of Ramsey’s theorem.
Lemma 2.2.6. Let X be an infinite set. Suppose that χ : Xs → [c] is surjective.

Then there exist infinite sets X1 , . . . , Xc ⊆ X and k1 , . . . , kc ∈ [c] such that ki ∈ χ Xsi

/ χ Xsi for all i < j ≤ c.

for all i ≤ c and kj ∈
Proof. Ramsey’s theorem states that there exists an infinite set X1 such that
χ Xs1 is constant; we define k1 to be this constant. Our proof proceeds by induction.

Suppose we have already found X1 , . . . , Xr and k1 , . . . , kr such that ki ∈ χ Xsi for

/ χ Xsi for all i < j ≤ c. Let S ∈ Xs such that χ(S) ∈

all i ≤ r and kj ∈ / {k1 , . . . , kr },
and let Y ⊆ Xr \ S be infinite. Let S0 , S1 , . . . be an enumeration of all the subsets
of S such that Si ⊆ Si ⇒ i ≤
j (the enumeration extends the
inclusion order). For
Y Y
i = 0, 1, . . . , define χi : s−|Si|
→ [c] as follows: for B ∈ s−|Si|
, set
χi (B) = χ(B ∪ Si ) .
Now
by Ramsey’s theorem, there exists an infinite set Zi and ` ∈ [c] such that
χi Zsi = {`i }. Note that for i = 0 we have Si = ∅ and `i ∈ {k1 , . . . , kr } since
Y
= Y0 = {∅},



Zi ⊆ Y ⊆ Xr . On the other hand, for i = 2s we have Si = S and s−|S i|
and `i = χi (∅) = χ(∅ ∪ Si ) = χ(S) ∈ / {k1 , . . . , kr }. Let i0 be smallest such that
`i0 ∈/ {k1 , . . . , kr }. Then X r+1 := Z i0 and
kr+1 := `i0 satisfy the desired properties:
Clearly, kr+1 ∈ χ Xr+1 Xr+1


n , and k j ∈
/ χ s for r + 1 < j ≤ c by the minimal choice of
i0 . 

2.3. On the Number of Orbits of n-Subsets

Let G be a permutation group on a countably infinite set D. We want to
prove that the number of orbits of n-subsets in a permutation group G forms a
non-decreasing sequence (Proposition 2.1.1).
Proof of Proposition 2.1.1. Let O1 , . . . , Oc be distinct orbits of n-subsets
(we do not assume that these are all orbits of n-subsets, that is, our proof also covers
the situation that the group is not oligomorphic). We show that there are at least c
orbits
of (n + 1)-subsets, using Ramsey’s theorem in the form of Lemma 2.2.6. Let
χ: Dn → [c] be the map that assigns to a subset of D from Oi the number i ∈ [c],
 2.4. HIGHLY SET-TRANSITIVE PERMUTATION GROUPS 31

and i = 1 if the subset lies in none of O1 , . . . , Oc ; note that χ is surjective. By

Lemma 2.2.6,
there exist infinite sets XX1 ,i
. . . , Xc ⊆ D and k1 , . . . , kc ∈ [c] such that
ki ∈ χ Xni for all i ≤ c and kj ∈ / χ n for all i < j ≤ c. For each i ≤ c, let
Xi


Bi ∈ n+1 be such that there exists an Si ⊂ Bi with χ(Si ) = ki . The sets B1 , . . . , Bc
lie in distinct orbits of n + 1-subsets, because no permutation from G can map Bi to
Bj for i < j ≤ c since Bj ⊆ Xj does not contain n-subsets of color ki . This proves
that G has at least as many orbits of n + 1-subsets as orbits of n-subsets. 

Exercises.
(42) Let (X; <) be a partially ordered set on a countably infinite set X.
Show that (X; <) contains an infinite chain, or an infinite antichain.
(43) Show that an infinite sequence of elements of a totally ordered set
contains one of the following:
• a constant subsequence;
• a strictly increasing subsequence;
• a strictly decreasing subsequence.
Derive the Bolzano-Weierstrass theorem (every bounded sequence in Rn has
a convergent subsequence), using the completeness property of Rn .
(44) Show that for every permutation group on an infinite set A
with finitely many orbitals there are pairwise distinct x, y, z ∈ A
such that (x, y), (y, z), (x, z) lie in the same orbital.

2.4. Highly Set-transitive Permutation Groups

In this section, we present a classification of highly set-transitive closed subgroups
of Sym(X) for countably infinite X. For x1 , . . . , xn ∈ Q we write − x− −−−→
1 · · · xn if x1 <
· · · < xn .
Theorem 2.4.1 (of [37]). Let G be a highly set-transitive permutation group that
is closed in Sym(X) for a countably infinite set X. Then G is isomorphic (as a
permutation group) to one of the following:
(1) Aut(Q; <);
(2) Aut(Q; Betw) where Betw is the ternary relation
(x, y, z) ∈ Q3 | −−→ ∨ −−→ ;

xyz zyx
(3) Aut(Q; Cycl) where Cycl is the ternary relation
(x, y, z) | −−→ ∨ −−→ ∨ −−→ ;

xyz yzx zxy
(4) Aut(Q; Sep) where Sep is the 4-ary relation
(x1 , y1 , x2 , y2 ) | −
x−−−−−→ −−−−−−→ −−−−−−→ −−−−−−→

1 x2 y1 y2 ∨ x1 y2 y1 x2 ∨ y1 x2 x1 y2 ∨ y1 y2 x1 x2

∨− x−−−−−→ −−−−−−→ −−−−−−→ −−−−−−→

2 x1 y2 y1 ∨ x2 y1 y2 x1 ∨ y2 x1 x2 y1 ∨ y2 y1 x2 x1 ;

(5) Aut(Q; =).

The relation Sep is the so-called separation relation; note that Sep(x1 , y1 , x2 , y2 )
holds for elements x1 , y1 , x2 , y2 ∈ Q iff all four points x1 , y1 , x2 , y2 are distinct and the
smallest interval over Q containing x1 , y1 properly overlaps with the smallest interval
containing x2 , y2 (where properly overlaps means that the two intervals have a non-
empty intersection, but none of the intervals contains the other). Illustrations of the
four proper closed subgroups of Sym(Q; <) that contain Aut(Q; <) can be found in
Figure 2.2.
To give you some ideas of the proof of Theorem 2.4.1, we give a proof of the
following.
32 2. COUNTING ORBITS

Sym(Q)

Aut(Q;Sep)

Aut(Q;Betw) Aut(Q;Cycl)

Aut(Q;<)

Figure 2.2. Illustrations of Aut(Q; <), Aut(Q; Betw),

Aut(Q; Cycl), and Aut(Q; Sep).

Proposition 2.4.2. Let G be a permutation group on a countably infinite set

D such that G is 3-set transitive but not 2-transitive. Then G is isomorphic to a
permutation group that is contained in Aut(Q; <).
Proof. By Proposition 2.1.1, G is 2-set transitive, and hence there are at most
three orbitals (Definition 1.4.4): to see this, fix distinct u, v ∈ D. Let O1 := {(x, y) ∈
D2 | ∃α ∈ G : α(u, v) = (x, y)}, O2 := {(x, y) ∈ D2 | ∃α ∈ G : α(v, u) = (x, y)}, and
O3 := ∆D . Then the orbits of pairs are either O1 , O2 , O3 , or O1 ∪ O2 , O3 . Since G is
not 2-transitive, there are exactly the three orbitals O1 , O2 , O3 . The structure (D; O1 )
is a special directed graph called tournament: for any two distinct x, y ∈ D we have
that either (x, y) ∈ O1 or (y, x) ∈ O1 . By 3-set transitivity, all of the three-element
substructures of (D; O1 ) are isomorphic. There are only two possibilities: either
these substructures are transitive, or they are directed 3-cycles. An easy case analysis
shows that there is no 4-element tournament such that all 3-element substructures are
directed 3-cycles. Hence, the first possibility must hold. Hence, (D; O1 ) is transitive,
and therefore a linear order. The linear order is dense: for all (a, c) ∈ O1 there exists a
b ∈ D such that O1 (a, b) and O1 (b, c). To see this, let u, v, w ∈ D be such that O1 (u, v)
and O1 (v, w) (and hence O1 (u, w)). There exists a g ∈ G such that g(u, w) = (a, c).
Then b := g(v) has the desired properties since O1 (a, g(v)) and O1 (g(v), c). Similarly,
one can show that (D; O1 ) is unbounded, that is, for every b ∈ D there exist a, c ∈ D
such that O1 (a, b) and O1 (b, c). We will see later (Proposition 3.2.1) that every
countable dense unbounded linear order is isomorphic to (Q; <), which implies the
statement. 

Exercises.
(45) Show that Aut(Q; Cycl) strictly contains Aut(Q; <).
(46) Label each edge in Figure 2.2 by its index (that is, the index
of the group at the bottom in the group at the top of the edge).
 CHAPTER 3

Oligomorphic Permutation Groups

A permutation group G over a countable set X is oligomorphic if G has only

finitely many orbits of n-tuples for each n ≥ 1.
Examples and counterexamples:
• all permutation groups on finite sets;
• Aut(Q; <), and all its supergroups;
• Aut(Z; <) is a non-example: it has only one orbit, but infinitely many or-
bitals (orbits of pairs). To see this, note that (u, v) and (x, y) are in the
same orbit if and only if u − v = x − y.
• Aut(Q; +) is another non-example: it has two orbits and infinitely many
orbits of pairs.
Lemma 3.0.1. Let G ≤ Sym(N) be oligomorphic and a ∈ Nn for n ∈ N. Then Ga
is oligomorphic as well.
If follows from this observation and Theorem 1.2.6 that every closed oligomorphic
permutation group must have continuum cardinality.

3.1. sInv-Aut
There is a surprising link between oligomorphicity of permutation groups and first-
order logic. Let A be a τ -structure and let φ(x1 , . . . , xk ) be a first-order formula with
free variables from x1 , . . . , xk , i.e., a formula built in the usual way with existential and
universal quantifiers, conjunction, disjunction, negation, equality, relation symbols
from τ , terms build from function symbols in τ , variables x1 , . . . , xk , and the quantified
variables. If a1 , . . . , ak ∈ A then we write A |= φ(a1 , . . . , ak ) if the formula φ evaluates
in A to true when instantiating xi with ai . For a rigorous definition, we refer to
course notes (e.g., [18]) or text books (e.g., [71]) in mathematical logic, or text books
in model theory (e.g., [72, 116, 152]). We say that φ(x1 , . . . , xk ) defines over A the
relation
{(a1 , . . . , ak ) ∈ Ak | A |= φ(a1 , . . . , ak )}.
Example 23. The formula x1 < x2 < x3 ∨ x3 < x2 < x1 defines the relation
Betw over (Q; <), and the formula ∀y(x1 < y ∨ x1 = y) defines the relation {0} over
(Q≥0 ; <). 4
Theorem 3.1.1. Let A be a structure such that Aut(A) is oligomorphic. Then
R ∈ sInv(Aut(A)) if and only if R is first-order definable over A.
The proof of Theorem 3.1.1 can be found below. One direction of the equivalence
is easy to show, and holds for general relational structures.
Proposition 3.1.2. Let A be a structure. If R is first-order definable in A, then
R ∈ sInv(Aut(A)).
Proof. Straightforward induction over the syntactic structure of first-order for-
mulas and their semantics. 
33
34 3. OLIGOMORPHIC PERMUTATION GROUPS

Corollary 3.1.3. There is no linear order that is first-order definable over

(C; +, ∗).
Proof. Let < be a linear order on C, and suppose without loss of generality
that −i < i. The map a 7→ a (complex conjugation) is an automorphism of C, and
exchanges −i to i. We thus found an automorphism that violates <, and Proposi-
tion 3.1.2 implies that < is not first-order definable. 
Corollary 3.1.4. Let A be a structure such that Aut(A) is oligomorphic. Let
A0 be a structure with domain A0 = A such that all relations of A0 are first-order
definable in A. Then Aut(A0 ) is oligomorphic as well.
It follows for example that Aut(Q; Betw) is oligomorphic.
Exercises.
(47) Show that the relation {(x, y) ∈ R2 | x = y 2 } is not first-order
definable over the structure (R; +).

(48) Show that the assumption of Theorem 3.1.1 that Aut(A) is

oligomorphic is necessary, i.e., find a structure A and a relation R
such that R is preserved by Aut(A) but not definable in A.
In order to show Theorem 3.1.1, we first show some useful lemmata. Let x1 , . . . , xn
be variables. If ψ(x1 , . . . , xn ) is a τ -formula and Ψ is a set of τ -formulas φ(x1 , . . . , xn ),
and A is a τ -structure, then we say that ψ and Ψ are equivalent over A if for all a ∈ An
we have A |= ψ(a) if and only if A |= φ(a) for all φ ∈ Ψ.
Lemma 3.1.5. Let A be a τ -structure such that Aut(A) is oligomorphic, let a ∈
An , and let Ψ be the set of all formulas φ(x1 , . . . , xn ) such that A |= φ(a). Then there
exists a formula ψ(x1 , . . . , xn ) which is equivalent to Ψ over A.
Proof. Proposition 3.1.2 implies that two n-tuples that lie in the same orbit
of Aut(A) satisfy the same formulas φ(x1 , . . . , xn ). Hence, by the oligomorphicity of
Aut(A) there are only finitely many tuples b1 , . . . , bm such that any two tuples from
b1 , . . . , bm do not satisfy the same formulas; we may suppose that b1 = a. For every
i ∈ {2, . . . , m} there exists a formula φi that holds in a but not in bi ; then φ2 ∧· · ·∧φm
is equivalent to Ψ. 
Lemma 3.1.6. Let A be a structure with an oligomorphic automorphism group,
and let s̄, t̄ ∈ Ak be tuples that satisfy the same first-order formulas in A. Then for
every a ∈ A there exists b ∈ A such that (s̄, a) ∈ Ak+1 and (t̄, b) ∈ Ak+1 satisfy the
same first-order formulas in A.
Proof. Let Ψ be the set of all first-order formulas satisfied by (s̄, a) in A. By
Lemma 3.1.5, there exists a τ -formula ψ which is equivalent to Ψ over A. Then
A |= ∃y.ψ(s̄, y), and A |= ∃y.ψ(t̄, y) by assumption. So there exists b ∈ A such that
A |= ψ(t̄, b). This shows that (t̄, b) satisfies all first-order formulas satisfied by (s̄, a)
in A. The converse holds as well since negation is part of first-order logic. 
Proof of Theorem 3.1.1. One implication of the statement has been shown in
Proposition 3.1.2. Conversely, suppose that R ∈ sInv(Aut(A)). Let τ be the signature
of A. Then R is a union of orbits of Aut(A) on Ak ; since Aut(A) is oligomorphic,
there exists an m ∈ N such that R = O1 ∪ · · · ∪ Om . We show that orbits of k-tuples
are first-order definable in A; this is sufficient because if ψi defines Oi over A, then
ψ1 ∨· · ·∨ψm defines R. So let O be such an orbit and let a = (a1 , . . . , ak ) ∈ O, and let
Ψ be the set of all first-order τ -formulas ψ(x1 , . . . , xk ) in the language of A such that
A |= ψ(a1 , . . . , ak ). We prove that if a tuple b̄ = (b1 , . . . , bk ) satisfies every formula in
 3.1. SINV-AUT 35

Ψ then b̄ ∈ O by constructing an automorphism of A that maps a to b. This is done

by a back-and-forth argument, using Lemma 3.1.6 for going forth, and again using
Lemma 3.1.6 for going back (see the proof of Proposition 3.2.1). By Lemma 3.1.5,
there is a τ -formula ψ which is equivalent to Ψ over A, and ψ defines O in A. 

Corollary 3.1.7. Let B and C be structures on the same domain such that
Aut(B) and Aut(C) are oligomorphic. Then Aut(B) = Aut(C) if and only if B and
C are (first-order) interdefinable in the sense that all relations of B have a first-order
definition in C and vice-versa.
If two countable structures have oligomorphic automorphism groups that are
isomorphic as permutation groups (Definition 1.0.1), then this corresponds to a model-
theoretic relation between the structures.
Definition 3.1.8. Two structures A and B are called bi-definable if there exists
a bijection f : A → B such that every R ⊆ An is definable in A if and only if f (A) is
definable in B.
Corollary 3.1.9. Let A and B be countable structures such that Aut(A) and
Aut(B) are oligomorphic. Then Aut(A) and Aut(B) are isomorphic as permutation
groups if and only if A and B are bi-definable.
Proof. Exercise 49. 

Corollary 3.1.10. Let A be such that Aut(A) is oligomorphic and let P ⊆

Sym(A). Then hP i = Aut(A) if and only if the set of relations that are definable in
A equals sInv(P ).

Proof. Suppose that hP i = Aut(A). Every relation that is first-order definable

in A is strongly preserved by Aut(A), and hence in particular by P ⊆ hP i = Aut(A).
Conversely, if R is strongly preserved by P , then it is also strongly preserved by hP i,
and hence first-order definable by Theorem 3.1.1.
Conversely, suppose that the set of relations that are definable in A equals
sInv(P ). We then have
hP i = Aut(sInv(P )) (by Proposition 1.2.8)
= Aut(A) (by assumption). 

Corollary 3.1.11. Let A and B be structures with the same domain and oligo-
morphic automorphism groups. Then Aut(A) ≤ Aut(B) if and only if every relation
that is first-order definable in B is first-order definable in A.
Proof. Exercise 50. 

It follows from Theorem 3.1.1 that if B is a structure with an oligomorphic au-

tomorphism group G, then the congruences of G are exactly the first-order definable
equivalence relations in B. Another application of Theorem 3.1.1 can be found in the
following example.
Example 24. The center of G is the set
C(G) := {α | αβ = βα for all β ∈ G}.
If G = Aut(A) is oligomorphic, then the center contains precisely those automor-
phisms of A that are preserved by all automorphisms of A. Hence, by Theorem 3.1.1,
C(G) consists precisely the automorphisms of A that are first-order definable in A. 4
36 3. OLIGOMORPHIC PERMUTATION GROUPS

The following example demonstrates that Theorem 3.1.1 fails if we keep the as-
sumption that Aut(A) has only finitely many orbits of n-tuples, for every n ∈ N, but
do not require that the domain of A is countable (which is part of the definition of
oligomorphicity).
Example 25. Consider the structure (R; ≺) where ≺ is the binary relation
Q × (R \ Q) ∪ {(x, y) | x < y, x, y ∈ Q or x, y ∈ R \ Q}.
Let a ∈ Q and b ∈ R \ Q. First note that every formula φ(x) that holds on a
also holds on b; this can be shown by induction over the shape of formulas, using
that we can extend isomorphisms between finite substructures (R; ≺) by one more
point in the image or pre-image. However, there is no automorphism of (R; ≺) that
maps a to b, because the set {x | x ≺ a} is countable, but the set {x | x ≺ b} is
uncountable. Hence, the unary relation consisting of all elements of Q is preserved by
all automorphisms of (R; ≺) but not first-order definable in (R; ≺). It is also easy to
see that the automorphism group of (R; ≺) has finitely many orbits of n-tuples, for
all n ∈ N (the expansion with the unary relation Q is homogeneous). 4
Remark 3.1.12. Note that if G1 and G2 act oligomorphically on A and B, re-
spectively, then the natural intransitive action of G1 × G2 is also oligomorphic: when
a(n) is the number of orbits of the componentwise action of G1 on An , and b(n) is
the number of orbits of the componentwise action ofP G2 on B, then the number of
orbits of the componentwise of G1 × G2 on A ∪ B is 0≤i≤n a(i)b(n − i), and hence
finite for all n.
If A and B have the same signature τ , then the automorphism group of the τ -
structure A × B (see Definition 1.1.6) contains the image of the product action of
Aut(A) × Aut(B) on A × B. The number of orbits of n-tuples of this action can
be bounded by an bn where an is the number of orbits of n-tuples in A and bn is the
number of orbits on n-tuples in B. Hence Aut(A × B) is oligomorphic.

Exercises.
(49) Prove Corollary 3.1.9.
(50) Prove Corollary 3.1.11. Hint: Exercise 11 and Theorem 3.1.1.
(51) Show that the assumption of Example 24 that Aut(A) oligomorphic
is necessary (there is even a directed graph A such that C(Aut(A))
contains automorphisms that are not definable in A).
(52) Show that (Z2 )N is isomorphic to the automorphism group
of a countable structure.
(53) Show that (Z2 )N is not isomorphic to the automorphism group
of an ω-categorical structure.

3.2. Countably Categorical Structures

Let τ be a countable signature. A set of (first-order) τ -sentences is called a τ -
theory. A model of a τ -theory T is a τ -structure A such that A satisfies all sentences
in T . Theories that have a model are called satisfiable. For every τ -structure A, we
denote by Th(A) the theory of A, that is, the set of all τ -sentences that are satisfied
by A.
A satisfiable τ -theory T is called ω-categorical (or ℵ0 -categorical, which we use
interchangeably) if all countable models of T are isomorphic. A structure is called
ω-categorical if its first-order theory is ω-categorical. Note that the theory of a finite
structure does not have countable models, and hence is ω-categorical.
 3.2. COUNTABLY CATEGORICAL STRUCTURES 37

Cantor [43] proved that the linear order of the rational numbers (Q; <), which
we will use as a running example in this section. We will see many more examples
of ω-categorical structures later. One of the standard approaches to verify that a
structure is ω-categorical is via a so-called back-and-forth argument. To illustrate, we
give the back-and-forth argument that shows that (Q; <) is ω-categorical; much more
about this important concept in model theory can be found in [73, 132].
Proposition 3.2.1. The structure (Q; <) is ω-categorical.
Proof. Let A be a countable model of the first-order theory T of (Q; <). It is
easy to verify that T contains (and, as this argument will show, is uniquely given by)
• ∃x. x = x (no empty model)
• ∀x, y, z ((x < y ∧ y < z) ⇒ x < z) (transitivity)
• ∀x. ¬(x < x) (irreflexivity)
• ∀x, y (x < y ∨ y < x ∨ x = y) (totality)
• ∀x ∃y. x < y (no largest element)
• ∀x ∃y. y < x (no smallest element)
• ∀x, z ∃y (x < y ∧ y < z) (density).
An isomorphism between A and (Q; <) can be defined inductively as follows.
Suppose that we have already defined f on a finite subset S of Q and that f is an
embedding of the structure induced by S in (Q; <) into A. Since <A is dense and
unbounded, we can extend f to any other element of Q such that the extension is
still an embedding from a substructure of Q into A (going forth). Symmetrically, for
every element v of A we can find an element u ∈ Q such that the extension of f that
maps u to v is also an embedding (going back ). We now alternate between going forth
and going back; when going forth, we extend the domain of f by the next element of
Q, according to some fixed enumeration of the elements in Q. When going back, we
extend f such that the image of A contains the next element of A, according to some
fixed enumeration of the elements of A. If we continue in this way, we have defined
the value of f on all elements of Q. Moreover, f will be surjective, and an embedding,
and hence an isomorphism between A and (Q; <). 
A second important running example of this section is the countable random
graph (V; E). This (simple and undirected) graph with a countably infinite number
of vertices has the following extension property: for all finite disjoint subsets U, U 0 of
V there exists a vertex v ∈ V \ (U ∪ U 0 ) such that v is adjacent to all vertices in U
and to no vertex in U 0 . The existence of such a graph will be show in Example 29.
Proposition 3.2.2. The random graph (V; E) is ω-categorical.
Proof. Note that the extension property of (V; E) given above is a first-order
property; a back-and-forth argument similar to the one given in the proof of Proposi-
tion 3.2.1 shows that every countably infinite graph with this property is isomorphic
to (V; E). 
The reason why we treat ω-categoricity in this course is the following theorem.
An accessible proof can be found in Hodges’ book (Theorem 6.3.1 in [73]).
Theorem 3.2.3 (Engeler, Ryll-Nardzewski, Svenonius). A countable structure B
is ω-categorical if and only if Aut(B) is oligomorphic.
If the signature of B is countable, there is another characterisation of ω-categoricity
of B via the property of Theorem 3.1.1.
Corollary 3.2.4. Let B be a structure with a countable signature and a countable
domain. Then B is ω-categorical if and only if sInv(Aut(B)) equals the set of relations
with a first-order definition over B.
38 3. OLIGOMORPHIC PERMUTATION GROUPS

Proof. The forwards implication is the content of Theorem 3.1.1. Conversely,

suppose that Aut(B) are infinitely many orbits of n-tuples, for some n. Then the union
of any subset of the set of all orbits of n-tuples is preserved by all automorphisms of
B; but there are only countably many first-order formulas over a countable language,
so not all the invariant sets of n-tuples can be first-order definable in B. 

Exercises.
(54) Show that in every ω-categorical structure B there exists a unique
finest equivalence relation which is definable in B and has finitely
many classes, and a unique coarsest equivalence which is definable
in B and has finite classes.

The following exercises are taken from Peter Cameron’s book “Oligomorphic permu-
tation groups”.
(55) Write down sentences φn , ψn (over the signature {=}) such that
(a) any model of φn has at least n elements;
(b) any model of ψn has exactly n elements.
(56) Write down a sentence, using equality and one binary relation
symbol, all of whose models are infinite.
Is this possible with equality alone?

3.3. Homogeneous Structures and Amalgamation Classes

A relational1 structure A is called homogeneous (sometimes also called ultra-
homogeneous [73]) if every isomorphism between finite substructures of A can be
extended to an automorphism of A.
Proposition 3.3.1. Let A be homogeneous with a finite signature. Then Aut(A)
is oligomorphic (and hence A is ω-categorical).
Proof. By the homogeneity of A, the orbit of an n-element subset B of Aut(A)
is given by the substructure induced by A on B. But there are finitely many non-
isomorphic substructures of A of size n. As we have seen earlier (2), this also bounds
the number of orbits of n-tuples of Aut(A). 
Example 26. The following structures are homogeneous.
• (Q; <).
• the canonical structure of a permutation group (Definition 1.2.3).
• every expansion of a countable structure A with an oligomorphic automor-
phism group by all first-order definable relations is homogeneous. 4
A versatile tool to construct countable homogeneous structures from classes of
finite structures is the amalgamation technique à la Fraı̈ssé. We present it here for
the special case of relational structures; this is all that is needed in the examples we
are going to present. For a stronger version of Fraı̈ssé-amalgamation for classes of
structures that might involve function symbols, see [73].
In the following, let τ be a countable relational signature. The age of a τ -structure
A is the class of all finite τ -structures that embed into A. A class C has the joint
embedding property (JEP) if for any two structures B 1 , B 2 ∈ C there exists a structure
C ∈ C that embeds both B 1 and B 2 .

1The entire theory can be adapted to general signatures that might also contain function sym-
bols; to keep the exposition simple, we restrict our focus to relational signatures in this section.
 3.3. HOMOGENEOUS STRUCTURES AND AMALGAMATION CLASSES 39

Proposition 3.3.2. Let C be a class of finite τ -structures. Then C is the age of

a (countable) relational structure if and only if C
• is closed under isomorphisms and substructures,
• contains only countably many structures up to isomorphism,
• has the JEP.
Remark 3.3.3. The Proposition is false if we drop the second item: take e.g.
τ := {R1 , R2 , . . . }, for Ri unary, and put all finite τ -structures into C. Then C
satisfies the first and the third, but not the second item.
The union of two relational τ -structures B 1 , B 2 is the τ -structure C with domain
B1 ∪B2 and relations RC := RB 1 ∪RB 2 for all R ∈ τ . The intersection of B 1 and B 2 is
defined analogously. Let B 1 , B 2 be τ -structures such that B 1 [B1 ∩ B2 ] = B 2 [B1 ∩ B2 ];
the pair (B 1 , B 2 ) is then called an amalgamation diagram. Then B 1 ∪B 2 is also called
the free amalgam of B 1 , B 2 . More generally, a τ -structure C is an amalgam of B 1
and B 2 if for i ∈ {1, 2} there are embeddings fi of B i to C such that f1 (a) = f2 (a)
for all a ∈ B1 ∩ B2 .
Definition 3.3.4. An isomorphism-closed class C of finite τ -structures
• has the free amalgamation property if for all B 1 , B 2 ∈ C the free amalgam
of B 1 and B 2 is contained in C;
• has the amalgamation property if every amalgamation diagram (B 1 , B 2 ) of
structures B 1 , B 2 ∈ C has an amalgam C ∈ C;
• is an amalgamation class if it contains at most countably many non-isomorphic
structures, has the amalgamation property, and is closed under isomorphisms
and taking induced substructures.
Note that since we only look at relational structures here (and since we allow
structures to have an empty domain), the amalgamation property of C implies the
joint embedding property.
Example 27. Let C be the class of all finite linear orders. Then C is clearly closed
under isomorphisms and induced substructures, and has countably many isomorphism
types. To show that it also has the amalgamation property, let B 1 , B 2 ∈ C and let
C be the free amalgam of B 1 and B 2 . Then C is an acyclic finite graph; therefore,
any depth-first traversal of C leads to a linear ordering of the elements that is an
amalgam (even a strong amalgam, but not a free amalgam) in C of B 1 and B 2 . It
follows that C is an amalgamation class. 4
Theorem 3.3.5 (Fraı̈ssé [58, 59]; see [73]). Let τ be a countable relational signa-
ture and let C be an amalgamation class of τ -structures. Then there is a homogeneous
and at most countable τ -structure C whose age equals C. The structure C is unique
up to isomorphism, and called the Fraı̈ssé-limit of C (often denoted by Flim(C)).
Proof. We first prove uniqueness. Let C and D be countable homogeneous
of the same age. We have to show that C and D are isomorphic, and construct
the isomorphism f : C → D by a back-and-forth argument, similarly to the proof
that (Q; <) is ω-categorical in Proposition 3.2.1. Let C = {c1 , c2 , . . . } and D =
{d1 , d2 , . . . }. Suppose f is already defined on a finite subset F of C.
• Going forth: Let i ∈ N be smallest so that ci ∈ / F . Then there is e : C[F ∪
{ci }] ,→ D. Note that the finite structures D[e(F )] and D[f (F )] are iso-
morphic via f ◦ e−1 . By the homogeneity of D, this isomorphism can be
extended to an automorphism a of D. Then a ◦ e|F = f and we extend
f by setting f (ci ) := a(e(ci )). Clearly, the extension of f thus defined is
40 3. OLIGOMORPHIC PERMUTATION GROUPS

an embedding of C[F ∪ {ci }] into D because it is the composition of the

embedding e with the automorphism a.
• Going back: let i ∈ N be smallest so that di ∈ / f [F ]. Analogously, find c ∈ C
so that the extension f (c) := di is an isomorphism.
A structure C is called weakly homogenous if for all B ∈ Age(C), substructure A of
B, and e : A ,→ C there is g : B ,→ C which extends e. Note that in the proof above,
we only needed weak homogeneity of C and D to construct f . It follows that weak
homogeneity implies homogeneity.
We now prove the existence of the homogeneous structure C from the statement
of the theorem. We will construct a sequence (C i )i∈N of τ -structures C i ∈ C such
that
• C 0 is the τ -structure with the empty domain,
• (C i )i∈N is a chain, i.e., C i is a substructure of C j if i ≤ j, and
• if A, B ∈ C, with A substructure of B and e : A ,→ C i for some i ∈ N, then
there are j ∈ N and g : B ,→ C j which extends e.
S
Then C := i∈N C i is weakly homogeneous, and hence homogeneous, by the com-
ments above.
Also note that Age(C) = C. Here, the inclusion ⊆ is clear. For the converse
inclusion, first note that for every A ∈ C there is B ∈ C such that A ,→ B and
C 0 ,→ B by the JEP. So B ,→ C j for some j ∈ N, and hence A ,→ B ,→ C.
Let P be a countable set of representatives for all (A, B) ∈ C 2 such that A is
a substructure of B. Let α : N2 → N be a bijection such that α(i, j) ≥ i for all
i, j. Suppose C k already constructed. Let (Ak,i , B k,i , fk,i )i∈N be a list of all triples
(A, B, f ) where
• (A, B) ∈ P and
• f : A ,→ C k .
Let i, j be such that k = α(i, j). Construct C k+1 as amalgam of C k and B i,j so that
fi,j extends to B i,j ,→ C k+1 . 
Example 28. Let C be the class of all finite partially ordered sets. Amalgamation
can be shown by computing the transitive closure: if C is the free amalgam of B 1 and
B 2 over A, then the transitive closure of C gives an amalgam in C. The Fraı̈ssé-limit
of C is called the homogeneous universal partial order. 4

Exercises.
(57) Recall that a tournament is a directed graph without self-loops such that
for all pairs x, y of distinct vertices exactly one of the pairs (x, y), (y, x) is
an arc in the graph. Show that the class of all finite tournaments has the
amalgamation property.
Example 29. Let C be the class of all finite graphs. It is even easier than in the
previous examples to verify that C is an amalgamation class, since here the free amal-
gam itself shows the amalgamation property. The Fraı̈ssé-limit of C is the countable
random graph (V; E) (also called the Rado graph) which we already encountered in
Section 3.2. 4
Proposition 3.3.6. For every k, there is a permutation group which is k-transitive
but not (k + 1)-transitive.
Proof. Let R be a relation symbol of arity k + 1. Let C be the class of all finite
{R}-structures where R denotes a relation that only contains tuples with pairwise
distinct entries. The class C is clearly closed under isomorphism, substructures, and
 3.3. HOMOGENEOUS STRUCTURES AND AMALGAMATION CLASSES 41

T3 T4 T5

Figure 3.1. Henson’s family of finite tournaments that are pairwise

homomorphically incomparable.

has only countably many isomorphism classes of structures. It also has the free
amalgamation property. Note that any two structures with at most k elements are
isomorphic, since R denotes the empty relation in those structures. Since the Fraı̈ssé-
limit is homogeneous, its automorphism group is therefore k-transitive. On the other
hand, the class contains non-isomorphic structures of size k+1 (e.g., a structure where
R denotes the empty relation and a structure where R is non-empty), and hence the
automorphism of Flim(C) is not k + 1-transitive. 

Example 30. Let C be the class of all finite triangle-free graphs, that is, all
graphs that do not contain K3 as a subgraph. Again, we have the free amalgamation
property. The Fraı̈ssé-limit is up to isomorphism uniquely described as the triangle-
free graph A such that for any finite S, T ⊂ A such that S is stable (i.e., induces a
graph with no edges; such a vertex subset is sometimes also called an independent
set) there exists v ∈ A \ (S ∪ T ) which is connected to all points in S, but to no point
in T . 4
We now introduce a convenient tool to describe classes of finite τ -structures. If
N is a class of τ -structures, we say that a structure A is N -free if no B ∈ N embeds
into A. The class of all finite N -free structures we denote by Forb(N ).
Example 31. Henson [65] used Fraı̈ssé limits to construct 2ω many homogeneous
directed graphs. Note that for all classes N of finite tournaments, Forb(N ) is an
amalgamation class, because if A1 and A2 are directed graphs in Forb(N ) such that
A = A1 ∩ A2 is an induced substructure of both A1 and A2 , then the free amalgam
A1 ∪ A2 is also in Forb(N ).
Henson in his proof specified an infinite set T of tournaments T 3 , T 4 , . . . with the
property that T i does not embed into T j if i 6= j. The tournament T n , for n ≥ 3, in
Henson’s set T has vertices 0, . . . , n + 1, and the following edges (see Figure 3.1):
• (i, i + 1) for 0 ≤ i ≤ n;
• (0, n + 1);
• (j, i) for j > i + 1 and (i, j) 6= (0, n + 1).
Note that this property implies that for two distinct subsets N1 and N2 of T the
two sets Forb(N1 ) and Forb(N2 ) are distinct as well. Since there are 2ω many subsets
42 3. OLIGOMORPHIC PERMUTATION GROUPS

of the infinite set T , there are also that many distinct homogeneous directed graphs;
they are often referred to as Henson digraphs. 4
The structures from Example 31 can be used to prove various negative results
about homogeneous structures with finite signature. A better behaved class of homo-
geneous structures are those whose age is finitely bounded (this is the same terminology
as in [110]).
Definition 3.3.7. A class C of finite relational τ -structures (or a structure with
age C) is called finitely bounded if τ is finite and there exists a finite set of finite
τ -structures N such that C = Forb(N ).
Fraı̈ssé’s theorem (Theorem 3.3.5) has a converse.
Proposition 3.3.8. The age of every homogeneous relational structure has the
amalgamation property.
Proof. Let A be a homogeneous structure and let (B 1 , B 2 ) be an amalgamation
diagram with B 1 , B 2 ∈ Age(A). Then there are embeddings ei B 1 ,→ A, for i ∈ {1, 2},
and by the homogeneity of A there exists an automorphism α ∈ Aut(A) such that
α(e1 (x)) = e2 (x) for every x ∈ B1 ∩ B2 . Let C be the substructure of A with
domain α(e1 (B1 )) ∪ e2 (B2 ). Then the embedding f1 : B1 → C given by α ◦ e1 and
the embedding e2 show that C is an amalgam of (B 1 , B 2 ). 

Exercises.
(58) Is the class of finite forests (i.e., simple acyclic graphs)
an amalgamation class?

(59) Let C be the class of all finite graphs G such that there is no
embedding from the 5-cycle C 5 into G.
Is C an amalgamation class?

(60) Show that the assumption that the domain of A is countable is

necessary for the third item in Example 26, that is, show that there
exists an uncountable structure whose automorphism group has
finitely many orbits of n-tuples, for all n, but whose expansion
by all first-order definable relations is not homogeneous.

(61) Show the age C of a structure has the amalgamation property

if and only if it has the 1-point amalgamation property, i.e., if for
all A, B 1 , B 2 ∈ C and embeddings e1 : A ,→ B 1 and e2 : A ,→ B 2
such that |B1 | = |B2 | = |A| + 1 there are a C ∈ C and embeddings
fi : B i ,→ C for i ∈ {1, 2} such that f1 ◦ e1 = f2 ◦ e2 .

(62) Let D be the tournament obtained from the directed cycle C ~3

of length three by adding a new vertex u, and adding the edges
~ 3 . Let D0 be the tournament
(u, v) for every vertex v of C
obtained from D by flipping the orientation of each edge.
Show that the class of all finite tournaments that embeds
neither D nor D0 is an amalgamation class.

(63) Let P be a unary relation symbol. Let D be the class of all finite
{P, <}-structures A such that <A is a linear order.
(a) Show that D is an amalgamation class.
 3.3. HOMOGENEOUS STRUCTURES AND AMALGAMATION CLASSES 43

(b) Let B be the Fraı̈ssé-limit of the class D, and define E ⊆ B 2

by (u, v) ∈ E if
• u < v and (u ∈ P ⇔ v ∈ P ), or
• u > v and not (u ∈ P ⇔ v ∈ P ).
Show that (B; E) is a tournament.
(c) Show that the class Age(B; E) equals the class of all
tournaments that can be obtained from tournaments T
in Age(Q; <) by performing the following operation: pick
u ∈ T and reverse all edges between u and other elements of T
(we ‘switch edges at u’).
(d) Show that (B; E) is homogeneous.
(e) Show that Age(B; E) equals the class C from Exercise 62.

Solution suggestion. First note that every tournament obtained from

the tournament D or the tournament D0 from Exercise 62 by switching
edges at a vertex (see Part (c)) is isomorphic to D or to D0 , and hence
not in the age of (B; E); it follows that Age(B; E) ⊆ C. For the converse
inclusion, let (V ; E) ∈ C. Note that for any fixed u ∈ V , the set

V1 := {u} ∪ {v ∈ V | (u, v) ∈ E}

induces a linear order in (V ; E), because (V ; E) is a tournament that

does not embed D. Choose u such that |V1 | is maximal after having
applied a finite number of switches at vertices from V . From now on,
(V ; E) denotes the graph after having performed this finite number of
switches. If V1 = V , then (V ; E) ∈ Age(B; E) by Part (c) of the exercise
and we are done. Also the set V2 := {v ∈ V | (v, u) ∈ E} induces a
linear order in (V ; E), because (V ; E) is a tournament that does not
embed D0 . Let w ∈ V2 be the maximal element with respect to E.
Let w0 ∈ V1 be minimal with respect to E such that (w0 , w) ∈ E; such
an element must exist due to the choice of u since otherwise V1 ∪ {w}
induces a linear order in (V ; E). Also note that w0 6= u because (w, u) ∈
E. Let w00 ∈ V1 be the predecessor of w0 with respect to E; such an
element must exist because w0 6= u. Let V10 := {v ∈ V1 | (w0 , v) ∈ E}.
Note that (w, w00 ), (w00 , w0 ), (w0 , w) ∈ E; hence, for every v ∈ V10 we
have (v, w) ∈ E since otherwise {v, w, w0 , w00 } induce a copy of D0 in
(V, E). Then after switching edges at each v ∈ V10 ∪ {w0 }, the set
V1 ∪ {w} induces a linear order (with least element w, then w0 , followed
by the elements of V10 , then u, followed by all other elements of V1 ; the
final vertex is w00 ), and we again reach a contradiction to the choice of
u such that |V1 | is maximal.
(f) Show how solutions to the previous sub-exercises provide
a solution to Exercise 62.
(g) Show that (B; E) is isomorphic to the tournament whose vertices are
a countable dense subset S ⊆ R2 of the unit circle without antipodal
points, and where the edges are oriented in clockwise order, i.e., put
((u1 , u2 ), (v1 , v2 )) ∈ E if and only if u1 v2 − u2 v1 > 0.
44 3. OLIGOMORPHIC PERMUTATION GROUPS

3.4. Strong Amalgamation and Algebraicity

A strong amalgam of B 1 , B 2 is an amalgam C of B 1 , B 2 with embeddings fi : B i →
C such that f1 (B1 ) ∩ f2 (B2 ) = f1 (B1 ∩ B2 ) = f2 (B1 ∩ B2 ). We say that a class C has
the strong amalgamation property if all B 1 , B 2 ∈ C have a strong amalgam in C. A
class C is called a strong amalgamation class if it is an amalgamation class with the
strong amalgamation property.
Examples:
• The age of the Random Graph.
• The age of (Q; <).
• The age of all other structures we have seen so far.
• An example of an amalgamation class that does not have strong amalgama-
tion: let P be a unary relation symbol, and let C be the class of all finite
{P }-structures where P contains at most one element. Then C is an amal-
gamation class, and the Fraı̈ssé-limit is a countably infinite structure where
P contains only one element. But C does not have strong amalgamation.
Picture with an amalgamation diagram that fails to have a strong amalgam.
• Another non-example is the (N; E2 ) (where E2 is an equivalence relation
with infinitely many classes of size two; see Exercise 41).

3.4.1. Algebraicity. If C is a strong amalgamation class, then the automor-

phism group of the Fraı̈ssé-limit of C has a remarkable property. If A is a structure
and B ⊆ A, then we say that R ⊆ An is (first-order) definable in A with parameters
from B (also: definable in A over B) if R is definable in the expansion of A by a
constant for each element of B. We say that R is parameter definable if R is definable
in A with parameters from A.
Definition 3.4.1 (Model-theoretic algebraic closure). Let A be a structure and
B ⊆ A. Then aclA (B) denotes the elements of A that lie in finite sets that are
first-order definable over A with parameters from B.
Example 32. Let A = (N; E2 ) where E2 is an equivalence relation with infinitely
many classes of size two. Then we have aclA (∅) = ∅, and for every a ∈ A we have
aclA ({a}) = {a, b} where b is the unique element that is distinct from a and lies in
the same E2 class as a. 4
Definition 3.4.1 and Theorem 3.1.1 motivates the following definition.
Definition 3.4.2 (Group-theoretic algebraic closure). Let G ≤ Sym(A) be a
permutation group and B ⊆ A finite. Then aclG (B) denotes the set of elements of A
that lie in finite orbits in G(B) .
Lemma 3.4.3. Both the model theoretic and the group theoretic algebraic closure
are closure operators. If G is oligomorphic and the automorphism group of an (ω-
categorical) structure A, then the algebraic closure of B in G coincides precisely with
the model-theoretic algebraic closure of B in A, i.e.,
aclA (B) = aclAut(A) (B). (3)
In this case, the algebraic closure of a finite set is finite.
Proof. Both for the group theoretic and the model theoretic version, the ver-
ification that B ⊆ acl(B), B ⊆ C ⇒ acl(B) ⊆ acl(C), and acl(acl(B)) = acl(B) is
easy and left to the reader. The statements about acl in an ω-categorical structure A
follow from Theorem 3.1.1. 
 3.4. STRONG AMALGAMATION AND ALGEBRAICITY 45

B A

Figure 3.2. An illustration of Lemma 3.4.3.

We say that B ⊆ A is algebraically closed in A if aclA (B) = B. We say that

A has no algebraicity if all finite B ⊆ A are algebraically closed in A. We use the
analogous terminology for permutation groups G instead of structures, i.e., B ⊆ A is
algebraically closed in G if aclG (B) = B, and G has no algebraicity if all finite B ⊆ A
are algebraically closed in G.
Theorem 3.4.4 (See (2.15) in [39]). Let A be a homogeneous structure. Then
the age of A has strong amalgamation if and only if G := Aut(A) has no algebraicity.
Proof. We first show that strong amalgamation of the age implies no algebraic-
ity of G. Let t ∈ An , F = {t1 , . . . , tn }, and u ∈ A \ F . We want to show that the
orbit of u in Gt is infinite. Let m ∈ N and let B be the structure induced by F ∪ {u}
in A. Then there exists a strong amalgam B 0 ∈ Age(A) of B with B over A[F ]. We
iterate this, taking a strong amalgam of B with B 0 over A[F ], showing that, because
of homogeneity, there are m distinct elements in A \ F that lie in the same orbit as u
in Gt . Since m ∈ N and u ∈ A \ F were chosen arbitrarily, the group Gt has no finite
orbits outside F .
For the other direction, we rely on the following lemma of Peter Neumann (see
Figure 3.2).
Lemma 3.4.5. Let G be a permutation group on D without finite orbits, and let
A, B ⊂ D be finite. Then there exists g ∈ G with g(A) ∩ B = ∅.
Proof. The proof is by induction on |A|. The induction base A = ∅ is trivial.
We assume the result for any set A0 with |A0 | < |A|. We may also assume without
loss of generality that there exists a ∈ A \ B, because for any a0 ∈ A we choose k ∈ G
such that k(a0 ) ∈ / B since G has no finite orbits. If g 0 ∈ G shows that the statement
holds for k(A) instead of A, then g  := g 0 k is the map we are looking for.
Let h1 , . . . , hk ∈ G be such that h1 (a), . . . , hk (a) = (G◦a)∩B. By the inductive
assumption, there exists h ∈ G such that

h(A \ {a}) ∩ B ∪ h1 (B) ∪ · · · ∪ hk (B) = ∅.
If h(a) ∈/ B, then h(A) ∩ B = ∅ and we are done. Otherwise, h(a) = hi (a) for some
i ∈ {1, . . . , k}. Define g := h−1
i h. Then
• g(a) = a ∈ / B since a ∈ A \ B.
• g(A \ {a}) ∩ B = h−1 i h(A \ {a}) ∩ B = ∅, since h(A \ {a}) ∩ hi (B) = ∅.
Therefore, g(A) ∩ B = ∅. 
We now continue with the reverse implication of Theorem 3.4.4. Suppose G
has no algebraicity, and let B 1 , B 2 ∈ Age(A). By the homogeneity of A we can
furthermore assume that B 1 , B 2 are substructures of A; that is, the structure induced
by B1 ∪B2 in A is an amalgam, but possibly not a strong one. Let t ∈ An be such that
{t1 , . . . , tn } = B1 ∩ B2 . Since G has no algebraicity, Gt has no finite orbits outside
46 3. OLIGOMORPHIC PERMUTATION GROUPS

B1 ∩ B2 . By the lemma, there exists g ∈ Gt such that (B1 \ B2 ) ∩ g(B2 \ B1 ) = ∅.

Then the substructure of A induced on B1 ∪ g(B2 \ B1 ) is a strong amalgam of B 1
and B 2 . 
Exercises.
(64) Give an example of a homogeneous structure with a transitive
automorphism group whose age does not have strong amalgamation.
(65) Let G be a permutation group on a set A. Show that the following are
equivalent.
(a) There exists a structure A with finite relational signature such that
G = Aut(A);
(b) There exists a relation R ⊆ An such that G = Aut(A, R);
(c) There exists a structure A with finite relational signature
such that G = Aut(A), and all relations in A
have pairwise distinct entries.
(66) If in the previous exercise we additionally assume that the domain A
is finite, we can combine the two conditions, and the above items
are equivalent to
(d) There exists a relation R ⊆ An such that G = Aut(A, R)
and R has pairwise distinct entries.
(67) Prove the equivalence of (a) and (d) in the previous exercises
for oligomorphic G.
(68) Prove the equivalence of (a) and (d) in the previous exercises
for an arbitrary permutation group G on an infinite set.
(69) For Equation 3, can we replace the assumption that Aut(A)
is oligomorphic by the assumption that A is homogeneous?

3.4.2. Generic superpositions. For strong amalgamation classes there is a

powerful construction to obtain new strong amalgamation classes from known ones.
Definition 3.4.6. Let C1 and C2 be classes of finite structures with disjoint re-
lational signatures τ1 and τ2 , respectively. Then the generic superposition of C1 and
C2 , denoted by C1 ∗ C2 , is the class of (τ1 ∪ τ2 )-structures A such that the τi -reduct of
A is in Ci , for i ∈ {1, 2}.
The following lemma has a straightforward proof by combining amalgamation in
C1 with amalgamation in C2 .
Lemma 3.4.7. If C1 and C2 are strong amalgamation classes, then C1 ∗ C2 is also
a strong amalgamation class.
If A1 and A2 are countable homogeneous structures with countable signatures and
without algebraicity, then A1 ∗ A2 denotes the (up to isomorphism unique) Fraı̈ssé
limit of the generic superposition of the age of A1 and the age of A2 .
Example 33. For i ∈ {1, 2}, let τi = {<i }, let Ci be the class of all finite τi -
structures where <i denotes a linear order, and let Ai be the Fraı̈ssé limit of Ci . Then
A1 ∗ A2 is known as the random permutation (see e.g. [34, 107, 147]). 4
Proposition 3.4.8. Let A1 , A2 be countable homogeneous structures with count-
able signatures and without algebraicity. Then for i ∈ {1, 2} the τi -reduct of A1 ∗ A2
is isomorphic to Ai .
 3.4. STRONG AMALGAMATION AND ALGEBRAICITY 47

Proof. Let a0 , a1 , a2 , . . . be an enumeration of the elements of A1 ∗ A2 , and

let b0 , b1 , b2 , . . . be an enumeration of the elements of A1 . Suppose that f is an
isomorphism between the τ1 -reduct B of a finite substructure of A1 ∗ A2 and a finite
substructure of A1 .
• Going forth: let j ∈ N be smallest so that aj ∈ / B. Then the τ1 -reduct of
A1 ∗ A2 [B ∪ {aj }] has an embedding e into A1 . By the homogeneity of A1
we may assume that e extends f .
• Going back: let j ∈ N be smallest so that bj ∈ / f (B), and let k ∈ N be such
that ak ∈ / B. By definition, the age of A1 ∗ A2 contains a structure C
– whose τ2 -reduct equals the τ2 -reduct of A1 ∗ A2 [B ∪ {ak }].
– the extension of f that maps ak to bj is an isomorphism between the
τ1 -reduct of C and A1 [f (B) ∪ {bj }].
Let e be an embedding of C into A1 ∗ A2 ; by the homogeneity of A1 ∗ A2 we
may assume that e is the identity on B. Then f (e(ak )) := bj is an extension
of f whose image contains bj , as desired.
If we repeat going forth and going back for countably many steps, each time extending
the isomorphism f as described above, the result is an embedding that is defined on
all of A1 ∗ A2 (by the way we defined going forth) and which is surjective onto A1
(by the way we defined going back). An isomorphism between A1 ∗ A2 and A2 can
be constructed analogously. 
Exercises.
(70) Construct an permutation group G on a set X
with precisely n! orbits of n-element subsets.
Extra question (I don’t know the answer to it): does this property
characterise G uniquely up to isomorphism?
(71) Show that the random graph can be partitioned into two subsets
so that both parts are isomorphic to the random graph.
(72) Show that the previous statement is not true for all partitions
of the random graph into two infinite subsets.
(73) Show that for every partition of the random graph, one of
the parts induces a subgraph which is isomorphic
to the random graph.
(74) Show that an amalgamation class C has the free amalgamation property if
and only if C = Forb(N ) for some set of finite structures N whose Gaifman
graphs are cliques. The Gaifman graph of a structure A
is a graph with vertex set A where two distinct u, v ∈ A are adjacent
if and only if A contains a relation with a tuple where both u and v
appear in some entry.
(75) Let A be a homogeneous relational structure whose age has the free amalga-
mation property. Show that if B, C ∈ Age(A), then there is g ∈ Aut(A)(B∩C)
such that A[g(B) ∪ C] is isomorphic to B ∪ C.
(76) Show that if C1 ∗ C2 is a strong amalgamation class, then C1 and C2
are strong amalgamation classes.
(77) Let C1 , C2 be amalgamation classes with Fraı̈ssé-limits A1 , A2
such that Aut(A1 ) and Aut(A2 ) are oligomorphic. Show that if C1
does not have the strong amalgamation property, then C1 ∗ C2
is not an amalgamation class, unless Aut(Flim(C2 )) = Sym(A2 ).
48 3. OLIGOMORPHIC PERMUTATION GROUPS

(78) Show that there are permutation groups G1 , G2

on a countably infinite set such that both G1 and G2
are isomorphic (as permutation groups) to Aut(Q; <),
but G1 ∩ G2 = {id}.

3.5. The Weak Amalgamation Property

Definition 3.5.1. A class K of finite structures satisfies the weak amalgamation
property (WAP) if every A ∈ K has an extension A0 ∈ K which is determined on A,
that is, for all B 1 , B 2 ∈ K and embeddings ei : A0 → B i for i ∈ {1, 2} there exists
C ∈ K and embeddings fi : B i → C such that f1 ◦ e1 |A = f2 ◦ e2 |A .
Example 34. A class of relational structures with JEP and WAP, but not the AP
is the age of (Z; {(x, y) | x = y + 1}) (Exercise 79). An ω-categorical structure whose
age has the WAP but not the AP is the structure (Q+ 0 ; <, {0}) (Exercise 80). 4

Exercises.
(79) Prove that the age of (Z; {(x, y) | x = y + 1}) does not have the AP,
but the WAP.
(80) Prove that the age of (Q; <, {0})
does not have the AP, but the WAP.
(81) Prove that the age of (Q; {(x, y, z) | x < y, z 6= x, z 6= y})
does not have the AP, but the WAP.
(82) A countable ω-categorical τ -structure B is called model-complete if
every first-order τ -formula is over B equivalent to an existential
τ -formula. Show that B is model-complete if and only if Aut(B)
is dense in Emb(B) ⊆ B B , where Emb(B) is the set of all
self-embeddings of B.
(83) Let A be an ω-categorical structure. The a model companion of A
is a structure B with the same age as A which is model-complete
(see the previous exercise). Show that every ω-categorical structure
has a model companion, and that the model companion is unique
up to isomorphism. We refer to B as the model companion of A.
(84) (Todor Tsankov, personal communication 2012) Prove that the age
of every ω-categorical structure A has the WAP.
(85) A structure A with a finite relational signature is called homogenizable if
there exists a definable expansion B of A by finitely many new relations
such that B is homogeneous. We say that A is boundedly homogenizable
if it is homogenizable and for every a ∈ An , n ∈ N, there exists b ∈ Am ,
m ∈ N, such that the type of (a, b) in A is equivalent to a quantifier-free
formula. Find a structure which is model-complete, homogenizable, but not
boundedly homogenizable.
More on the WAP can be found in Example 81.

3.6. First-order Interpretations

Many ω-categorical structures can be derived from other ω-categorical structures
via interpretations (our definition follows [73]).
 3.6. FIRST-ORDER INTERPRETATIONS 49

Definition 3.6.1. A relational σ-structure B has a (first-order2) interpretation

I in a τ -structure A if there exists a natural number d, called the dimension of I, and
• a τ -formula δI (x1 , . . . , xd ) – called the domain formula,
• for each atomic σ-formula φ(y1 , . . . , yk ) a τ -formula φI (x1 , . . . , xk ) where the
xi denote disjoint d-tuples of distinct variables – called the defining formulas,
• a surjective map h : D → B, where
D := {(a1 , . . . , ad ) ∈ Ad | A |= δI (a1 , . . . , ad )}
– called the coordinate map,
such that for all atomic σ-formulas φ and all tuples ai ∈ D
B |= φ(h(a1 ), . . . , h(ak )) ⇔ A |= φI (a1 , . . . , ak ) .
Note that the coordinate map h determines the defining formulas up to logical
equivalence; hence, we sometimes identify I with h. Note that the kernel of h coincides
with the relation defined by (x = y)I , for which we also write =I , the defining formula
for equality.
We say that B is interpretable in A with finitely many parameters if there are
a1 , . . . , an ∈ A such that B is interpretable in the expansion of A by the singleton
relations {ai } for all 1 ≤ i ≤ n.
Lemma 3.6.2 (see Theorem 7.3.8 in [72]). Let A be an ω-categorical structure.
Then every structure B that is interpretable in A with finitely many parameters is
ω-categorical or finite.
Proof. An easy consequence of Theorem 3.2.3. 
Note that in particular all reducts of an ω-categorical structure and all expansions
of an ω-categorical structure by finitely many constants are again ω-categorical.
Example 35. Allen’s interval algebra: studied in temporal reasoning in computer
science. Domain: closed bounded intervals over the rational numbers. Relations: con-
tainment, disjointness, precedence, etc. Formally a structure A that is best described
by a first-order interpretation I in (Q; <):
• the dimension of the interpretation is two;
• the domain formula δI (x, y) is x < y;
• for each inequivalent {<}-formula φ with four variables a binary relation R
such that (a1 , a2 , a3 , a4 ) satisfies φ if and only if ((a1 , a2 ), (a3 , a4 )) ∈ R.
By Lemma 3.6.2, A is ω-categorical. 4
3.6.1. Composing interpretations. Interpretations can be composed. In or-
der to conveniently treat these compositions, we first describe how an interpreta-
tion of a σ-structure B gives rise to interpreting formulas for arbitrary σ-formulas
ψ(x1 , . . . , xn ). Replace each atomic σ-formula φ(y1 , . . . , yn ) in ψ by the formula
φI (y1,1 , . . . , y1,d , . . . , yn,1 , . . . , yn,d ); we write ψI (x1,1 , . . . , x1,d , . . . , xn,1 , . . . , xn,d ) for
the resulting τ -formula, and call it the interpreting formula for ψ. Note that if ψ
defines the relation R in B, then φI defines I −1 (R) in A. For all d-tuples a1 , . . . , ak ∈
I −1 (B)
B |= φ(I(a1 ), . . . , I(ak )) ⇔ A |= φI (a1 , . . . , ak ) .
2‘First-order’ refers to the usage of first-order logic in the definition; one may replace first-order
logic by fragments or extensions of first-order logic, or by entirely different logics, and in this way
one obtains different forms of interpretations that have been studied in the literature. We often omit
the specification of ‘first-order’ in this text since all interpretations studied here are first-order.
50 3. OLIGOMORPHIC PERMUTATION GROUPS

Definition 3.6.3. Let C, B, A be structures with the relational signatures ρ, σ,

and τ . Suppose that
• C has a d-dimensional interpretation I in B, and
• B has an e-dimensional interpretation J in A.
Then C has a natural (d · e)-dimensional interpretation I ◦ J in A: the domain of
I ◦ J is the set of all de-tuples in A that satisfy the τ -formula (>I )J , and we define
I ◦ J(a1,1 , . . . , a1,e , . . . , ad,1 , . . . , ed,e ) := I(J(a1,1 , . . . , a1,e ), . . . , J(ad,1 , . . . , ed,e ))
(which is a partial function, composed from partial functions, and which is defined on
all tuples where all partial functions that appear in the composition are defined).
Let φ be a τ -formula which defines a relation R over A. Then the formula (φI )J
defines in A the preimage of R under I ◦ J.

3.6.2. Bi-interpretations. Two interpretations I1 and I2 of B in A are called

homotopic 3 if the relation {(x̄, ȳ) | I1 (x̄) = I2 (ȳ)} is first-order definable in D. Note
that idC is an interpretation of C in C, called the identity interpretation of C (in C).
Definition 3.6.4. Two structures A and B with an interpretation I of B in A
and an interpretation J of A in B are called mutually interpretable. If both I ◦ J and
J ◦ I are homotopic to the identity interpretation (of A and of B, respectively), then
we say that A and B are bi-interpretable (via I and J).
Example 36. The directed graph C := (N2 ; E) where

E := ((u1 , u2 ), (v1 , v2 )) | u2 = v1
and the structure D := (N; =) are bi-interpretable. The interpretation I1 of C in D
is 2-dimensional, the domain formula is true, and the coordinate map is the identity.
The interpretation I2 of D in C is 1-dimensional, the domain formula is true, and the
coordinate map sends (x, y) to x.
Then I2 (I1 (x, y)) = z is definable by the formula x = z, and hence I1 ◦ I2 is
homotopic to the identity interpretation of D. Moreover, I1 (I2 (u), i2 (v)) = w is
first-order definable by
E(w, v) ∧ ∃p (E(p, u) ∧ E(p, w)) ,
so I2 ◦ I1 is also homotopic to the identity interpretation of C. 4
Example 37. It is easy to see that Allen’s interval algebra (Example 35) is bi-
interpretable with (Q; <). 4
We show that the property to have essentially infinite language is preserved by
bi-interpretability.
Proposition 3.6.5. Let B and C be ω-categorical structures that are first-order
bi-interpretable. Then B has essentially infinite signature if and only if C has.
Proof. Let τ be the signature of B. We have to show that if C has finite
signature, then B is interdefinable with a structure B 0 with a finite signature. Let σ ⊆
τ be the set of all relation symbols that appear in all the formulas of the interpretation
of C in B. Since the signature of C is finite, the cardinality of σ is finite as well.
We will show that there is a first-order definition of B in the σ-reduct B 0 of
B. Suppose that the interpretation I1 of C in B is d1 -dimensional, and that the
interpretation I2 of B in C is d2 -dimensional. Let θ(x, y1,1 , . . . , yd1 ,d2 ) be the formula

3We follow the terminology from [3].

 3.6. FIRST-ORDER INTERPRETATIONS 51

that shows that I2 ◦ I1 is homotopic to the identity interpretation. That is, θ defines
in B the (d1 d2 + 1)-ary relation that contains a tuple (a, b1,1 , . . . , bd1 ,d2 ) iff
a = h2 (h1 (b1,1 , . . . , b1,d2 ), . . . , h1 (bd1 ,1 , . . . , bd1 ,d2 )) .
Let φ be an atomic τ -formula with k free variables x1 , . . . , xk . We will specify a
σ-formula that is equivalent to φ over B 0 .
^
1
∃y1,1 , . . . , ydk1 ,d2 i
θ(xi , y1,1 , . . . , ydi 1 ,d2 )
i≤k
1 k 1 k
, . . . , ydk1 ,d2 )


∧ φI1 I2 (y1,1 , . . . , y1,d 2
, y2,d2
, . . . , y2,d 2

is equivalent to φ(x1 , . . . , xk ) over B. Indeed, by surjectivity of h2 , for every element

ai of B there are elements ci1 , . . . , cid2 of C such that h2 (ci1 , . . . , cid2 ) = ai , and by
surjectivity of h1 , for every element cij of C there are elements bi1,j , . . . , bid1 ,j of B such
that h1 (bi1,j , . . . , bid1 ,j ) = cij . Then
B |= R(a1 , . . . , ak ) ⇔ C |= φI2 (c11 , . . . , c1d2 , . . . , ck1 , . . . , ckd2 )
⇔ B 0 |= φI1 I2 (b11,1 , . . . , bk1,d2 , b12,d2 , . . . , bk2,d2 , . . . , bkd1 ,d2 ) 
We will come back to bi-interpretability in Section 5.3.
 CHAPTER 4

Topological Groups

We start with a very brief introduction to concepts from topology that will be
relevant in what follows.

4.1. Topological Spaces

Topological spaces are mathematical structures that are used to capture the no-
tions of “closeness” and “continuity” on a very general level. A topological space is a
set S together with a collection of subsets of S, called the open sets of S, such that
(1) the empty set and S are open;
(2) arbitrary unions of open sets are open;
(3) the intersection of two open sets is open.
Complements of open sets are called closed. Note that S and the empty set are both
open and closed.

Example 38. On every set S, there is the trivial topology where the only open
sets are S and the empty set. 4

Example 39. Every set S can be equipped with the discrete topology: in this
topology, every subset of S is open (and hence also closed). 4

Example 40. The standard topology on R: a set U ⊆ R is open exactly if for

every x ∈ U there exists an  > 0 such that the every y ∈ R with x −  < y < x +  is
contained in U . The standard topology on Q is defined analogously. 4

Example 41. The standard topology on Rd , d ∈ N: a set U ⊆ Rd is open exactly

if for every x ∈ U there exists an  > 0 such that the -ball around x is contained in
U. 4

Example 42. Every set S can be equipped with the cofinite topology: in this
topology, the open sets are the empty set and every cofinite subset of S. The only
closed subsets in this topology are the finite sets and the entire set S. 4

Example 43. The topology of pointwise convergence on Sym(N); see Proposi-

tion 1.2.4. 4

For E ⊆ S, the closure of E, denoted by Ē, is the intersection over all closed sets
over S that contain E. A subset E of S is called dense (in S) if its closure is the full
space S. The interiour of E is the largest open set contained in E; and is denoted
by Int(E). Equivalently, Int(E) := S \ S \ E.
The subspace of S induced on E is the topological space on E where the open
sets are exactly the intersections of E with the open sets of S. When we work with
permutation groups G ⊆ Sym(X) then we will always work with the topology on G
inherited from Sym(X) in this way.
53
54 4. TOPOLOGICAL GROUPS

4.1.1. Countability, separation, and connection. A basis (or base) of S is

a collection of open subsets of S such that every open set in S is the union of sets
from the collection. Clearly, a topology is uniquely given by any of its bases. For
p ∈ S, a collection of open subsets of S is called a basis at p if each set from the
collection contains p, and every open set containing p also contains an open set from
the collection. A topological space S is called
• first-countable if for all s ∈ S there exists a countable basis at s;
• second-countable if it has a countable basis;
• separable if it contains a countable dense set.
Note that being second-countable implies first-countable (to obtain a countable basis
at s ∈ S, select all members of the countable basis of S that contain s.) Every second-
countable space is also separable: if {U1 , U2 , . . . } is a countable basis of nonempty
sets, choosing any xn ∈ Un gives a countable dense set {x1 , x2 , . . . }. (Suppose for
contradiction that there exists p ∈ S \ {x1 , x2 , . . . }. Then there is a closed set V
containing {x1 , x2 , . . . } but not p. Hence, there is an open set U that contains p but
no point in {x1 , x2 , . . . }. Since {U1 , U2 , . . . } is a basis, there exists an i ∈ N such that
Ui ⊆ U . But then xi ∈ Ui ⊆ U , a contradiction.)

Example 44. We revisit the examples that we have seen above.

• The standard topology on R and, more generally, on Rd for d ∈ N are
second-countable (and in particular first-countable), a countable basis being
the set of all open balls with rational center and rational radius.
• The topology of pointwise convergence on Sym(N) is second-countable: there
are countably many basic open sets S(a, b) with a, b ∈ Nn for n ∈ N.
• The discrete topology on any set S is first-countable (for every p ∈ S, the
set {{p}} is a basis at p), but if S is uncountable, S is not separable (we
have Ū = U for all U ⊆ X), and therefore also not second-countable.
• The cofinite topology on an uncountably infinite set S is not first-countable,
but separable: the closure of any countably infinite subset of S is S. 4

A topological space S is called Hausdorff if for any two distinct points u, v of S

there are disjoint open sets U and V that contain u and v, respectively.

Example 45. The standard topology on R, Rd , Q, and the topology of pointwise

convergence on Sym(N) are Hausdorff. The trivial topology on a set with at least two
elements is not Hausdorff. The cofinite topology on an infinite set is a more interesting
example that is not Hausdorff: the intersection of any two non-empty open sets is
infinite, so in particular not empty. So we cannot separate two distinct points with
open disjoint sets. 4

We have seen that R and Sym(N) share some countability and separation prop-
erties. But in some other respects, these two spaces are very different. A topological
space S is called disconnected if it is the union of two or more disjoint nonempty open
subsets, and connected otherwise. A subset of S is said to be connected if it is con-
nected under its subspace topology. The inclusion-wise maximal connected subsets
of a non-empty topological space are called the connected components of that space.
Picture in R2 !

Example 46. The standard topology on Q is disconnected: the two sets

{x ∈ Q | x < π} and {x ∈ Q | x > π}

 4.1. TOPOLOGICAL SPACES 55

are open, and for any irrational π they partition Q. On the other hand, R and Rd are
connected1. 4
A topological space S is totally disconnected if all connected subsets of X are
one-element sets.
Example 47. The topology of pointwise convergence on Sym(N) is totally dis-
connected: if f, g ∈ Sym(N) are distinct, there exists an i ∈ N such that S f (i) 6= g(i).
Then S(i, f (i)) is an open set that contains f , and Sym(N)\S(i, f (i)) = j6=f (i) S(i, j)
is a disjoint set that contains g, and it is open as a union of basic open sets. Hence,
no set that contains more than one element is connected. 4

Exercises.
(86) Show the claim from Example 46 that the standard topology on R
is connected.
(87) On which sets X is the cofinite topology separable?
(88) Show that R with the standard topology is not homeomorphic
to a closed subgroup of Sym(N).
(89) Show that the cofinite topology on an uncountably infinite set
is not first-countable.
4.1.2. Continuity and convergence. A map between two topological spaces
is called continuous if the pre-images of open sets are open, and open if images of
open sets are open. A bijective open and continuous map is called a homeomorphism.
There are equivalent characterisations of continuity of maps from a first-countable
space S to a topological space T that are often easier to work with. For a sequence
(sn )n≥1 of elements of S, we say that sn converges against s if for every open set U
of S that contains s there exists an n0 such that sn ∈ U for all n ≥ n0 . Note that if
T is Hausdorff, then s is unique, and called the limit of (sn )n≥1 , and we write
lim sn = s.
n→∞
For x ∈ S, we say that f is continuous at x if for every open V ⊆ T containing f (x)
there is an open U ⊆ S containing x whose image f (U ) is contained in V .
Proposition 4.1.1. Let S be a first-countable and T an arbitrary topological
space. Then for every f : S → T the following are equivalent.
(1) f is continuous.
(2) For all sequences sn , if sn converges against s, then f (sn ) converges against
f (s).
(3) f is continuous at every x ∈ S.
Proof. The implication from (1) to (2) is true even without the assumption that
S is first-countable. Let (sn )n≥1 be such that limn→∞ sn = s, and let V be open so
that f (s) ∈ V . Then U := f −1 (V ) is open, and s ∈ U . So there exists an n with
sn ∈ U . For this n we have f (sn ) ∈ V . So limn→∞ f (sn ) = f (s).
For the implication from (2) to (3), we show the contraposition. Suppose that f is
not continuous at some s ∈ S. That is, there exists an open set V containing f (s) such
that all open sets U that contain s have an image that is not contained in V . Since S is
first-countable, there exists a countable collection Un of open sets containing s so that
any open set that contains s also contains some Un . Replacing Un by ∩nk=1 Uk where
1This relies on the fact that the real numbers are (by definition!) Dedekind-complete: every
non-empty subset S of R with an upper bound in R has a least upper bound.
56 4. TOPOLOGICAL GROUPS

necessary, we may assume that U1 ⊃ U2 ⊃ · · · . If Un ⊆ f −1 (V ), then f (Un ) ⊆ V , in

contradiction to our assumption; so we can pick an sn ∈ Un \ f −1 (V ) for all n, and
obtain a sequence that converges to s. But sn ∈ / f −1 (V ) for all n, and so f (sn ) does
not converge to f (s) ∈ V .
Finally, the implication from (3) to (1) again holds in arbitrary topological spaces.
Let V ⊆ T be open. We want to show that U := f −1 (V ) is open. When s is a point
from U , then because f is continuous at s, and V contains f (s) and is open, there
S an open set Us ⊆ S containing s whose image f (Us ) is contained in V . Then
is
s∈U Us = U is open as a union of open sets. 
Q
4.1.3. Product spaces. The product i∈I Si of a Q family of topological spaces
(Si )i∈I is the topological space
Q on the cartesian product i∈I Si where the open sets
are unions of sets of the form i∈I Ui where Ui is open in Si for all i ∈ I, and Ui = Si
for all but finitely many i ∈ I. When I has just two elements, say 1 and 2, we also
write S1 × S2 for the product. We denote by S k for the k-th power S × · · · × S of S,
equipped with the product topology as described above.
We also write S I to a |I|-th power of S, where the factors are indexed by the
elements of I. In this case, we can view each element of T := S I as a function from
I to S in the obvious way. The product topology on T is also called the topology of
pointwise convergence, due to the following.
Proposition 4.1.2. Let S be a topological space, and I be a set. Let (fn )n∈N be
a sequence of elements of the product space T := S I . Then limn→∞ fn = f if and
only if limn→∞ fn (j) = f (j) in S for all j ∈ I.
Proof. Suppose first that limn→∞ fn = f in T . Let Q j ∈ I be arbitrary and let V
be an open set that contains f (j). Then the set U := i∈I Ti where Ti = V if i = j,
and Ti = S otherwise, is open in T and contains f . So there is an n0 such that fn ∈ U
for all n ≥ n0 . But then fn (j) ∈ V for all n ≥ n0 , and so limn→∞ fn (j) = f (j).
Now suppose that limn→∞ fn (j) = f (j) in S for all j ∈ I, and let V be an open
Q f . Then there exists a finite J ⊆ I and open subsets (Vj )j∈J
set of T that contains
of S such that f ∈ i∈I Ti where Ti = Vi if i ∈ J and Ti = S otherwise. For each
j ∈ J there exists an nj so that fn (j) ∈ Vj for all n ≥ nj . Then fn ∈ V for all
n ≥ max j∈J nj , and hence limn→∞ fn = f . 
Example 48. When we equip the natural numbers N with the discrete topology,
then NN with the topology of pointwise convergence is called the Baire space. Note
that the open sets are exactly the unions of sets of the form
S(ā, b̄) := {g ∈ N → N | g(ā) = b̄}
for some ā, b̄ ∈ Nk , k ∈ N. The closed subspace 2N is called the Cantor space. The
subspace on Sym(N) is precisely the topology that we introduced in Proposition 1.2.4.
4
Theorem 4.1.3 (Baire). NN is homeomorphic to the irrational numbers P.
Proof. It suffices to construct a mapping from ZN to P. Note that the sets of
the form {x ∈ ZN | x1 . . . xk = s} for s ∈ Zk , k ∈ N form a basis of open sets for
ZN . Let (qn )n∈N be an enumeration of the rational numbers. Inductively construct a
sequence of open intervals (Is )s∈Zk ,k∈N satisfying the following:
(1) Is = R if s ∈ Z0 ;
(2) if s ∈ Zk , for k > 0, then Is is an open interval in R with rational endpoints
of length less than 1/k;
(3) for every n ∈ Z we have I(s,n) ⊆ Is ,
 4.1. TOPOLOGICAL SPACES 57

(4) the right end point of I(s,n) is the left end point of I(s,n+1) ,
(5) {I(s,n) | n ∈ Z} covers Is ∩ P,
(6) the n-th rational qn is an endpoint of Is for some s ∈ Zk with k ≤ n + 1.
T
Define the function f : ZN → P as follows. Given x ∈ ZN the set n∈N Ix1 ...xn must
consist of a singleton irrational:
• it is nonempty because Ix1 ...xn xn+1 ⊆ Ix1 ...xn ;
• it is a singleton because the length of Ix1 ...xn tends to zero for n → ∞.
So we can define f by
\
{f (x)} := Ix1 ...xn .
n∈N

The function f is injective because if s and t are not prefixes of each other then Is
and It are disjoint, and f is surjective because for every u ∈ P and k ∈ N there is a
unique s ∈ Zk with u ∈ Is . Finally, f is a homeomorphism because
f ({x ∈ ZN | x1 . . . xk = s}) = Is ∩ P
and the sets of the form Is ∩ P form a basis for P. 

4.1.4. Metric spaces. Important examples of topologies come from metric spaces.
A pseudometric space is a pair (M, d) where M is a set and d is a pseudometric on
M , i.e., a function
d: M × M → R
such that for any x, y, z ∈ M , the following holds:
(1) d(x, y) ≥ 0 (non-negativity)
(2) d(x, y) = d(y, x) (symmetry)
(3) d(x, z) ≤ d(x, y) + d(y, z) (subadditivity or triangle inequality)
If d additionally satisfies
(4) d(x, y) = 0 ⇔ x = y (indiscernibility)
then d is called a metric, and (M, d) is called a metric space. If M 0 ⊆ M then the
restriction of d to M 0 is clearly a metric, too. Every metric on M gives rise to a
topology on M , namely the topology with the basis

{y ∈ M | d(x, y) < } | 0 <  ∈ R, x ∈ M .
A topological space S is metrisable if there exists a metric d on S which is compatible
with the topology, i.e., the topology equals the topology that arises from the metric
as described above.
Example 49. The discrete metric ρ on X is defined by

1 if x 6= y,
ρ(x, y) =
0 if x = y
for any x, y ∈ X. In this case (X, ρ) is called a discrete metric space or a space of
isolated points. 4
Example 50. The distance function d(x, y) = |x−y| (absolute difference) defines
a metric on R, Rd , and on Q. The topology that arises from this metric is precisely
the standard topology on R, Rd , and on Q. 4

Proposition 4.1.4. A metric space is second countable if and only if it is sepa-

rable.
58 4. TOPOLOGICAL GROUPS

2nd countable Metrisable

separable 1st countable

Proof. Suppose that X is separable, and let A be a countable set which is

dense in X. Then open balls with rational radii and centres from A form a countable
basis. Why? Conversely, when X is second-countable, we choose for every U from a
countable base of X one element; this will give a countable dense subset of X. 
4.1.5. Complete Metrics. A sequence (sn )n∈N of elements of a metric space
(M, d) is called a Cauchy sequence if
∀ > 0 ∃n0 ∈ N ∀n, m > n0 : d(sn , sm ) <  ;
in this case, we write
lim d(sn , sm ) = 0 .
n,m→∞
A metric space (M, d) is called complete if every Cauchy sequence converges against
an element of M . A topological space S is called completely metrizable if it has
a compatible complete metric. It is called Polish if S is separable and completely
metrizable.
Example 51. The standard distance metric on R is complete. The same metric
on Q is not complete. Since Q is dense in R and countable, it follows that R is
separable and hence Polish. 4
Example 52. The symmetric group Sym(D) on a countably infinite set D has
the following compatible metric d. Let b1 , b2 , . . . be an enumeration of D. Then for
elements f, g ∈ G we define d(f, g) := 0 if f = g, and otherwise d(f, g) := 1/2n where
n is the least natural number such that f (bn ) 6= g(bn ). In fact, d is an ultrametric,
that is, it satisfies
d(x, z) ≤ max(d(x, y), d(y, z))
for all x, y, z. This metric is not complete: to see this, let f be an arbitrary injective
non-surjective mapping from D → D. For each n, there exists a permutation hn of
D such that hn (bi ) = f (bi ) for all i ≤ n. Hence, the sequence (hn )n≥1 is Cauchy, but
it does not converge to a permutation. 4
Example 53. Similarly to the previous example, the Baire space (Example 48)
can be equipped with a compatible metric: define d(f, g) := 0 if f = g, and otherwise
d(f, g) := 1/2n where n is the least natural number such that f (n) 6= g(n). For the
Baire space, this is a complete metric. The restriction of this metric to the Cantor
space shows that the Cantor space is completely metrisable as well. 4
Example 54. The topology on Sym(D) on a countably infinite set D is also
completely metrizable. Again, let b1 , b2 , . . . be an enumeration of D. We define
a compatible complete metric d0 on Sym(D) by setting d0 (f, g) = 0 if f = g, and
otherwise d0 (f, g) = 1/2n where n is the least natural number such that f (bn ) 6= g(bn )
or f −1 (bn ) 6= g −1 (bn ). To prove that this metric is complete, let (gm )m∈N be a Cauchy
sequence with respect to d0 . For b ∈ D, define g(b) := gi (b) if i ∈ N is such that
gr (b) = gs (b) for all r, s ≥ i. Note that this is well-defined. Also note that such an
i exists for every b ∈ D: if b = bj for j ∈ N, then (gm ) being Cauchy with respect
to d0 implies that there exists i such that for all r, s ≥ i we have d0 (gr , gs ) < 1/2j
which means that gr (bj ) = gs (bj ). Note that the map g : D → D is injective, and
 4.1. TOPOLOGICAL SPACES 59

also surjective. To show that b ∈ D lies in the image of g, let j be such that b = bj .
Then there exists i0 ∈ N such that c := gr−1 (b) = gs−1 (b) for all r, s ≥ i, again by the
assumption that (gm ) is Cauchy with respect to d0 . Then g(c) = b by the definition
of g. Since Sym(D) is also separable, we have that Sym(D) is Polish. 4
Example 55. The Hilbert cube [0, 1]N . Here, the interval [0, 1] is equipped with
the usual topology inherited from R, and [0, 1]N carries the product topology. 4

Exercises.
(90) Every metrisable space X has a compatible metric d
which satisfies d(x, y) ≤ 1 for all x, y ∈ X.
(91) Let (X, d) be a complete metric space and S ⊆ X. Then S is
closed in X if and only if (S, d|S 2 ) is complete.
(92) Show that every metrisable space is Hausdorff.
(93) Show that every metrisable space X is regular : for any x ∈ X
and any open set U that contains x there is an open set O
that contains x such that Ō ⊆ U .
(94) Show that every metrisable space X is normal : for any
disjoint closed C, F ⊆ X there are disjoint open O, U ⊆ X
such that C ⊆ O and F ⊆ U .
(95) Let X be a metrisable space and Y ⊆ X be closed. Then Y is
a countable intersection of open sets.
(96) True or false: the closure of an open ball of radius r in
a metric space is the closed ball of radius r
in that metric space.
(97) Show that a countable product of metrisable spaces
is metrisable.
(98) Every separable metrisable space is homeomorphic to a subspace
of the Hilbert cube [0, 1]N .
For a metric space (X, d) and A ⊆ X, define diamd (A) := supb,c∈A d(b, c). Note
that for every  > 0 and every open subset U of X we find an open V ⊆ U such that
diamd (V ) < .
2
Definition 4.1.5. Countable intersections of open sets are called Gδ .
In the proof of the following lemma we use the axiom of dependent choice (Ap-
pendix A.2).
Lemma 4.1.6. Every Polish subspace Y of a Polish space X is Gδ .
Proof. Let dX be a complete compatible metric on X and let dY be a complete
compatible metric on Y . Define Vn ⊆ X as the union of all open sets U ⊆ X that
satisfy
(1) U ∩ Y 6= ∅;
(2) diamdX (U ) < n1 ;
(3) diamdY (U ∩ Y ) < n1 ;
T
It suffices to show that Y = n∈N Vn . Let x ∈ Y and n ∈ N; we want to show that
x ∈ Vn . Let U1 ⊆ Y be an open set that contains x such that diamdY (U1 ) < n1 . By the
definition of the relative topology, there is an open set U2 in X such that U2 ∩Y = U1 .
Let U3 be an open subset of X that contains x such that diamdX (U3 ) < n1 . Then
U := U2 ∩ U3 satisfies all of the three conditions above. Hence, x ∈ Vn .
2The origin of the notation is G for the German word Gebiet and δ for intersection.
60 4. TOPOLOGICAL GROUPS

T
Conversely, suppose that x ∈ n∈N Vn . Then for every n ∈ N, there exists an
open subset Un of X that contains x and satisfies the three conditions given above.
By the first condition, Un ∩ Y 6= ∅. Choose xn ∈ Un ∩ Y . Then the second condition
implies that limn→∞ xn = x. The third condition implies that (xn )n∈N is Cauchy
with respect to dY . Since dY is complete, limn→∞ xn = x ∈ Y . 
Remark 4.1.7. The converse of Lemma 4.1.6 is true as well: every Gδ set is
Polish. We do not need this statement in the following and therefore omit the proof.
We finally state an important property of completely metrizable spaces.
Theorem 4.1.8 (The Baire Category Theorem). Every Polish3 space S is Baire,
i.e., has the property that countable intersections of dense open sets are dense.
Proof.
T Let (Un )n∈N be a sequence of open dense sets. We want to show that
U := Un is dense. It is sufficient to show that any non-empty open set W in S
contains an element of U . Since U1 is dense, there is x1 ∈ U1 ∩ W . Hence, there is
an r1 with 0 < r1 < 1 such that {z ∈ S | d(x1 , z) ≤ r1 } ⊆ U1 ∩ W where d is the
compatible complete metric. We can continue recursively to find a sequence (xn )n∈N
of elements of S and a sequence (rn )n∈N of elements of R such that
{z ∈ S | d(z, xn ) ≤ rn } ⊆ Un ∩ Brn−1 (xn−1 ) (4)
as follows: if we have defined x1 , . . . , xn and r1 , . . . , rn satisfying (4), then density of
Un+1 guarantees that Brn (xn ) ∩ Un+1 is non-empty. Using the axiom of dependent
choices, we may choose an element xn+1 from this set such that
there is an rn+1 ∈ R with {z ∈ S | d(z, xn+1 ) ≤ rn+1 } ⊆ Un+1 ∩ Brn (xn ). (5)
Instead of using this axiom we can instead use the assumption that there exists a
countable set {u1 , u2 , . . . } which is dense in S: we may then choose xn+1 := uk for k
smallest possible so that (5) is satisfied. Since xm ∈ Brn (xn ) for all m > n, we have
that (xn )n∈N is Cauchy, and hence converges to some limit x by the completeness of d.
For any n, the set {z ∈ S | d(z, xn ) ≤ rn } is closed and hence contains x. Therefore,
x ∈ W and x ∈ Un for all n. 
4.1.6. Metric Completions. Let (M1 , d1 ) and (M2 , d2 ) be two metric spaces.
An isometry between (M1 , d1 ) and (M2 , d2 ) is a function i : M1 → M2 such that
d1 (x1 , x2 ) = d2 (i(x1 ), i(x2 )) (note that i must be injective, but it is not required to
be surjective). Two metrics are called isometric if there exists a bijective isometry
between them.
Metric spaces have the advantage that we can use Cauchy sequences to talk about
points that aren’t really there. More formally:
Definition 4.1.9. A completion of a metric space (M, d) is a complete metric
space (M ∗ , d∗ ) together with an isometry i : M → M ∗ such that i(M ) is dense in M ∗ .
Proposition 4.1.10. Every metric space has a completion.
Proof. Let (M, d) be a metric space. Let C be the collection of all Cauchy
sequences in M . Define an equivalence relation ∼ on C by setting (xn ) ∼ (yn ) if
limn→∞ d(xn , yn ) = 0 for (xn ), (yn ) ∈ C. Define
• M ∗ to be the set of all equivalence classes of ∼,
X ∗ := [(xn )] | (xn ) ∈ C ,


3Using the axiom of dependent choices (DC), this assumption can be weakened from Polish to
completely metrisable (in fact, the modified statement is then equivalent to DC; see Appendix A.2).
 4.1. TOPOLOGICAL SPACES 61

• d∗ : X ∗ × X ∗ → R by setting for [(xn )], [(yn )] ∈ M ∗

d∗ ([(xn )], [(yn )]) := lim d(xn , yn ),
n→∞

• i : M → M ∗ by setting for x ∈ M
φ(x) := [(x, x, . . . )].
Claim 1. d∗ is well-defined. Let (x0n ) and (yn0 ) be two Cauchy sequences such that
(x0n ) ∼ (xn ) and (yn0 ) ∼ (yn ). By the triangle inequality
d(xn , yn ) ≤ d(xn , x0n ) + d(x0n , yn0 ) + d(yn0 , yn )
and thus
|d(xn , yn ) − d(x0n , yn0 )| ≤ |d(xn , x0n ) − d(yn0 , yn )|
which tends to 0 for n → ∞, and proves that limn→∞ d(xn , yn ) = limn→∞ d(x0n , yn0 ).
Claim 2. d∗ is a metric on M ∗ . This is straightforward.
Claim 3. i is an isometry:
d∗ (i(x), i(y)) = lim d(x, y) = d(x, y) .
n→∞

Claim 4. i(M ) is dense in M ∗ . Let [(xn )] ∈ M ∗ and  > 0. Since (xn ) is Cauchy,
there exists an n0 ∈ N such that d(xm , xn ) <  for all m, n ≥ n0 . For z := i(xn0 ) we
have
d∗ ([(xn )], i(z)) = lim d(xn , xn0 ) ≤  .
n→∞

Claim 5. d∗ is complete. Let [(x1n )], [(x2n )], . . . be Cauchy in (M ∗ , d∗ ). Diagonal

argument: We define a function k : N → N as follows. Set k(1) = 1, and k(2) such
that d(x2k(2) , x2l ) < 1/2 whenever l ≥ k(2). For s ∈ N, choose k(s) such that
• N (s) ≥ N (s − 1)
• d(xkk(s) , xkk(l) ) < 1/k whenever l ≥ k(s).
Then (xsk(s) ) is a Cauchy sequence:

d(xsk(s) , xtk(t) ) ≤ lim sup(d(xsk(s) , xsj ) + d(xsj , xtj ) + d(xtj , xtk(t) ))

j∈N
≤ 1/s + 0 + 1/t
which tends to 0 for n, m → ∞.
Moreover, limm→∞ [(xm n
n )] = [(xk(n) )]: let  > 0 and choose n0 ∈ N such that
1/n0 < /2 and if n, m ≥ n0 then d(xnk(n) , xm
k(m) ) < /2. Now, for m ≥ n0 :

d∗ ([(xm n m n
n )], [(xk(n) )]) = lim d(xn , xk(n) )
n→∞
= lim sup d(xm m m n
n , xk(m) ) + lim sup d(xk(m) , xk(n) )
n→∞ n→∞
≤ 1/n0 + /2 <  

We will in the following we refer to the completion of (M, d) because completions

are essentially unique:
Proposition 4.1.11. Let (M1∗ , d∗1 , i1 ) and (M2∗ , d∗2 , i2 ) be two completions of
(M, d). Then there is a unique bijective isometry f between M1∗ and M2∗ such that
f ◦ i1 = i2 .
62 4. TOPOLOGICAL GROUPS

Proof. Let x ∈ M1∗ . Since i1 (M ) is dense in M1∗ for each n ∈ N there exists
xn ∈ i1 (M ) with d∗1 (xn , x) ≤ n1 . Let yn := i2 (i−1
1 (xn )). Since i1 and i2 are isometries,
we have d2 (yn , ym ) = d1 (xn , xm ) for all m, n ∈ N. The sequence (xn )n∈N converges
against x, so it is Cauchy, and it follows that (yn )n∈N is Cauchy, too. Since M2∗ is
complete the sequence (yn )n∈N must converge to some y ∈ M2∗ .
Claim 1. The map f : M1∗ → M2∗ defined by f (x) := y is well-defined. Suppose
that (x0n )n∈N is another sequence of elements of M1 that converges to x. For n ∈ N,
let yn0 := i2 (i−1 0
1 (xn ); we have to show that limn→∞ yn = y.
0

Let  > 0. Since limn∈N yn = y there exists m ∈ N such that d∗2 (yn , y) < /2 for
all n ≥ m. There is also a k ∈ N such that for all n ≥ k we have d∗1 (xn , x) < /4 and
d∗1 (x0n , x) < /4. Hence, d1 (xn , x0n ) ≤ d∗1 (xn , x) + d∗1 (x, x0n ) < /4 + /4 = /2. Since i1
and i2 are isometries we have d2 (yn , yn0 ) = d1 (xn , x0n ). Hence, d2 (yn , yn0 ) < /2 for all
n ≥ k. So for n ≥ max (k, m) we have d∗2 (yn0 , y) ≤ d2 (yn0 , yn )+d2 (yn , y) < /2+/2 = ,
showing that limn→∞ yn0 = y.
Claim 2. f is an isometry. Let x, x0 ∈ M1∗ and let (xn )n∈N and (x0n )n∈N be
sequences of elements of i1 (M ) that converge to x and x0 , respectively. Define yn :=
i2 (i−1 0 −1 0
1 (xn )) and yn := i2 (i1 (xn )), and we have seen that limn→∞ yn = f (x) and
limn∈N yn0 = f (x0 ). Then
d∗2 (f (x), f (x0 )) = lim d2 (yn , yn0 ) = lim d1 (xn , x0n ) = d(x, x0 ).
n∈N n∈N

Claim 3. f is surjective. Let y ∈ M2∗ . Since i2 (M ) is dense in M2∗ there is a a

sequence (yn )n∈N of elements of i2 (M ) converging to y. Similarly as above it can be
shown that (i1 (i−1 ∗
2 (yn )))n∈N is a sequence in i(M1 ) that converges to some x ∈ M1 ,
and that f (x) = y. 
If the isometry i of a metric completion (M ∗ , d∗ ) of (M, d) is not specified, we
typically assume that M ⊆ M ∗ and i is the identity. Clearly, the completion of a
separable metric space is separable, too.
Example 56. (R; d) is the completion of (Q; d). 4
Example 57. Let d be the ultrametric on Sym(N) from Example 52. Then the
metric completion of (Sym(N), d) equals the set of all injections from N to N. 4
4.1.7. Uniform continuity. Given metric spaces (X, d1 ) and (Y, d2 ), a function
f : X → Y is called uniformly continuous if
∀ > 0 ∃δ > 0 ∀x, y ∈ X (d1 (x, y) < δ ⇒ d2 (f (x), f (y)) < ).
For comparison: continuity of f with respect to the topologies induced by d1 and d2
only requires that
∀ > 0, x ∈ X ∃δ > 0 ∀y ∈ X (d1 (x, y) < δ ⇒ d2 (f (x), f (y)) < ).
Example 58. The function x 7→ x2 from R → R is continuous, but not uniformly
continuous: given an arbitrarily small positive real , uniform continuity requires the
existence of a positive number δ such that for all x1 , x2 with |x1 − x2 | < δ we have
|f (x1 ) − f (x2 )| < . But (x + δ 0 )2 − x2 = 2xδ 0 + (δ 0 )2 is larger than  for sufficiently
large x. 4
Example 59. An endomorphism ξ of the Baire space (Example 53) is uniformly
continuous if for every finite F ⊆ N there exists a finite G ⊆ N such that for all
f, g ∈ NN if f |G = g|G then ξ(f )|F = ξ(g)|F .
For comparison: an endomorphism of the Baire space is continuous if and only
if for every finite F ⊆ N and every f ∈ NN there exists a finite G ⊆ N such that if
g ∈ NN is such that f |G = g|F then ξ(f )|F = ξ(g)|F . 4
 4.1. TOPOLOGICAL SPACES 63

Proposition 4.1.12. A uniformly continuous map f between metric spaces maps

Cauchy sequences to Cauchy sequences.
Proof. Let (sn )n∈N be a Cauchy sequences, and let  > 0. By uniform continuity
of f there exists δ > 0 such that d(f (x) − f (y)) <  for d(x − y) < δ. Since sn is
Cauchy, there exists an n0 > 0 such that d(sn − sm ) < δ for all n, m > n0 . Hence,
d(f (sn ) − f (sm )) <  for all n, m > n0 . Therefore, (f (sn ))n∈N is Cauchy. 
4.1.8. Compactness. A topological space S is called S compact if for an arbitrary
collection U = {Ui }i∈A of open subsets of S with S S = i∈A Ui (also called an open
cover ) there is a finite subset B of A such that S = i∈B Ui (the collection {Ui }i∈B
is called a subcover of U). Clearly, finite spaces are compact. We state some closure
properties for compactness.
Proposition 4.1.13. Closed subsets of compact spaces are compact.
Proof. Let T be a compact space and let C be a closed subspace of T . Let U be
an open cover of C. By assumption, T \ C is open in T . Hence, U ∪ {T \ C} is an open
cover of T . As T is compact, there is a finite subcover of U, say {U1 , U2 , . . . , Ur }.
This also covers C by the fact that it covers T . If T \ C is among U1 , U2 , . . . , Ur , then
it can be removed and the remaining sets still cover C. Thus we have found a finite
subcover of U which covers C, and hence C is compact. 
The following is more substantial (actually, equivalent to the axiom of choice).
Theorem 4.1.14 (Tychonoff). Products of compact spaces are compact.
Proof. We only prove the statement for countable products of compact spaces;
this is all that will be needed in this text anyway. We first show that if X and Y are
compact, then so is X × Y . Let U be a collection of open subsets of X × Y such that
no finite subset of U covers X × Y ; we will show that U does not cover X × Y .
Claim 1. There exists x0 ∈ X such that for every open U ⊆ X that contains x0
the set U × Y is not finitely covered by U. Suppose otherwise that for every x ∈ X
there exists an open set Ux ⊆ X that contains x such that U × Y is covered by finitely
many elements of U. By the compactness of X, finitely many of the Ux cover X, so
finitely many sets of the form Ux × Y cover X × Y , contradicting the assumptions.
Claim 2. There exists y0 ∈ Y such that for every open U ⊆ X that contains x0
and every open V ⊆ Y that contains y0 no finite subset of U covers U × V . Otherwise,
for every y ∈ Y there is an open Uy ⊆ X containing x0 and an open Vy ⊆ Y containing
y such that Uy × Vy is covered by finitely many elements
S of U. By theTcompactness of
Y , there is a finite subset F ⊆ Y such that Y = y∈F Vy . Set U := u∈F Uy . Then
U is open and contains x0 , and
[ [
U ×Y = U × Vy ⊆ Uy × Vy
y∈F y∈F

is covered by finitely many elements of U, contradicting Claim 1.

It follows that no basic open set containing (x0 , y0 ) is covered by finitely many
elements of U. In particular, no basic open set containing (x0 , y0 ) can be contained in
an element of U, so (x0 , y0 ) is not covered by U. This finishes the proof that X × Y is
compact. To prove the statement for countable products, we first slightly generalise
the proof of Claim 2 to get the following.
Claim 3. Suppose that U is a family of open subsets of X × Y × Z where Y is
compact, and suppose that there is an x0 ∈ X such that for every open U ⊆ X that
contains x0 the set U × Y × Z is not covered by finitely many elements of U. Then
64 4. TOPOLOGICAL GROUPS

there exists an y0 ∈ Y such that for every open U ⊆ X that contains x0 and every
open V ⊆ Y that contains y0 , the set U × V × Z is not finitely coveredQ by U.
We finally prove that if X1 , X2 , . . . are compact, then X = i∈N Xi is compact.
Let U be a family of open sets that that no finite subset of U covers X. We will
construct an element x = (x1 , x2 , . . . ) of X that is not covered by U. Note first
that there is an x1 ∈ X1 such that for every open U1 ⊆ X1 that contains x1 the set
U1 × X2 × X2 × · · · is not finitely covered; the proof is as the proof of Claim 1, with
X2 × X3 × · · · playing the role of Y . Next, we can find x2 ∈ X2 such that such that
for every open U1 ⊆ X1 that contains x1 and every open U2 ⊆ X2 that contains x2
the set U1 × U2 × X3 × X4 × · · · is not covered by finitely many elements of U; this
follows from Claim 3 applied to X1 ×X2 ×(X3 ×X4 ×· · · ). Continuing in this way, we
inductively define x1 , x2 , x3 , . . . such that for each n and all open Ui ⊆ Xi for i ≤ n
such that Ui contains xi , the set U1 × · · · × Un × Xn+1 × · · · is not covered by finitely
many elements of U. The element (x1 , x2 , . . . ) ∈ X is then not covered by U. 

Exercises.
(99) Prove that a finite union of compact sets is compact.
In order to discuss which subsets of R and of Sym(N) are compact (with respect
to the subspace topology), we need the following definition for metric spaces.
Definition 4.1.15. A subset S of a metric space (M, d) is bounded if it is con-
tained in an open ball of finite radius, i.e., if there exists x ∈ M and a real  > 0 such
that for all s ∈ S, we have d(x, s) < .
The open ball of radius  and center x will be denoted by Bx () in the following.
Example 60. A subset of Rd is compact if and only if it is closed and bounded
– this is the theorem of Heine-Borel. 4
Which subsets of Sym(N) are compact?
Proposition 4.1.16. Any compact subset S of a Hausdorff topological space X
is closed in X.
Proof. If S is compact but not closed then there exists a ∈ S̄ \ S. For each
x ∈ S there exists an open set Ux that contains x but does not intersect an open set
Vx that contains a, because X is Hausdorff. Then U := {Ux | x ∈ S} is an open cover
of S, and by compactness of S there exists a finite subcover {Ux1 , . . . , Uxn } of U. But
then V := Vx1 ∩ · · · ∩ Vxn is open and contains a, and hence contains a point b in S
since a ∈ S̄. Since V is disjoint from each of Ux1 , . . . , Uxn , we have b ∈
/ Ux1 ∪ · · · ∪ Uxn ,
in contradiction to {Ux1 , . . . , Uxn } being a cover of S. 

With compactness, we ask for much: in general topological spaces, we might

in general even have open covers without countable subcovers! However, this can’t
happen if S is second-countable.
Proposition 4.1.17 (Lindelöf). Let S be second-countable. Then every open
cover of S has a countable subcover.
Proof. Let U = {Uα }α∈A be an open cover of S. Each Uα can be written as a
union β∈Jα Vβα of basic open sets Vβα . Then V := {Viα }α∈A,β∈Jα covers S. Since
S
S is second-countable, there are only countably many basic open sets, so V be we
written as {V1 , V2 , . . . }. By construction, for each i ∈ N there exists β(i) ∈ A such
that Vi ⊆ Uβ(i) . Then {Uβ(i) }i∈N forms a countable subcover of U. 
 4.1. TOPOLOGICAL SPACES 65

The set S is totally bounded if for every real  > 0 there exists a finite collection
of open balls in M of radius  whose union contains S. Clearly, a totally bounded
space is bounded, but the converse is not true: the discrete metric is bounded, but
not totally bounded: for  = 1/2, we need infinitely many open -balls (points!) to
cover the infinite set.
Proposition 4.1.18. If a metric space is totally bounded, then it is separable.
Proof. If X is totally
S bounded then for each positive
S n ∈ N there exists a finite
An ⊆ X such that X = x∈An Bx (1/n). Let A := n≥0 An . Clearly A is countable.
We claim that Ā = X. Let x ∈ X. For any n ∈ N there is some yn ∈ An such that
x ∈ Byn (1/n). This gives a sequence (yn ) with d(x, yn ) < 1/n. Thus lim yn = x
which proves the claim, and separability of X. 
In the proof of the following theorem, we assume the axiom of countable choice
(see Appendix A.2).
Theorem 4.1.19. For a metric space (X, d), the following are equivalent.
(1) X is compact;
(2) Every collection of closed sets in X with the finite intersection property
(every finite subcollection has a nonempty intersection) has a nonempty in-
tersection;
(3) If F1 ⊇ FT2 ⊇ F3 ⊇ · · · is a decreasing sequence of nonempty closed sets in
X, then n≥1 Fn is nonempty;
(4) X is sequentially compact, that is, every sequence in X has a convergent
subsequence;
(5) X is totally bounded and complete.
Proof. (1) ⇒ (2). Suppose that C is a collection of closed sets with empty
intersection. Then U := {X \ C | C ∈ C} is an open cover of X, and hence contains
a finite subcover of X. The complements of the members of the subcover in X give
the collection with the finite intersection property.
(2) ⇒ (3). A decreasing sequence of non-empty closed sets obviously has the
finite intersection property.
(3) ⇒ (4). Let (xn )n∈N be a sequence of points in X, and let Fn be the closure of
the set {xn , xn+1 , . . . }. Then F1 ⊇ F2 T ⊇ F3 ⊇ · · · and all the sets Fn are nonempty
and closed. Therefore, by (3), the set n≥1 Fn contains at least one point a. Then
(xn )n∈N contains a subsequence converging to a: to see this, let d be the compatible
metric, and set n0 = 1. Now suppose that nk has already been defined for k ∈ N. Since
a is in the closure of {xnk +1 , xnk +2 , . . . } there exists an n ∈ {nk + 1, nk + 2, . . . } such
that d(xn , a) < 1/(k + 1). Let nk+1 the the smallest such n. Then limk→∞ xnk = a.
(4) ⇒ (5). To prove that X is complete, let (xn ) be any Cauchy sequence in X.
By (4), there is a subsequence converging to some point a ∈ X. But then the whole
sequence (xn ) converges to a. This shows that X is complete.
Now suppose that X is not totally bounded, i.e., there exists a number  > 0 such
that X has no finite covering by open balls of radius . Then we can define a sequence
(xn )n≥1 of points in X having d(xi , xj ) ≥  for all i 6= j, by the following inductive
construction: First let x1 be any point in X. Then, supposing that x1 , . . . , xn−1 have
been chosen, we know Bx1 () ∪ · · · ∪ Bxn−1 () is not the whole space. Hence we can
choose a point xn satisfying d(xi , xn ) ≥  for all i < n. On the other hand, the
sequence (xn ) cannot have any convergent subsequence; for if it had a subsequence
(xnk ) converging to a, then there would exist an integer k0 such that d(xnk , a) < /2
for all k ≥ k0 , and hence by the triangle inequality d(xnk , xnk0 ) <  for all k, k 0 ≥ k0 ,
contrary to the definition of the sequence (xn ).
66 4. TOPOLOGICAL GROUPS

(5) ⇒ (4). Let (xi )i∈N be a sequence of elements from X. Let S = {xn | n ∈ N}.
If S is finite then the statement is trivial so assume that S is infinite. Since X is
totally bounded, there exists a finite cover of X with open balls of radius 1 := 1.
One of those balls, call it B1 , must contain infinitely many elements from S. Again by
total boundedness, there exists a finite cover of X with open balls of radius 2 := 1/2,
and again, one of those balls must contain infinitely many elements from B1 ∩ S; this
ball we call B2 . We continue this process, producing a sequence of balls (Bk ) of radius
1/k so that Bk ∩ Bk−1 ∩ · · · ∩ B1 contains infinitely many elements of the sequence
xi . Pick now indices n1 < n2 < n3 < · · · such that yk := xnk ∈ Bk . It is easy to
see that (yi ) is Cauchy and so by the completeness assumption on X it must have a
convergent subsequence.
(4) ⇒ (1). Let U be an open cover of X. From the implication (4) ⇒ (5) we have
that X is totally bounded, and Proposition 4.1.18 implies that X is second-countable.
By Proposition 4.1.17 (Lindelöf) we can therefore assume that U is countable, U =
{U1 , U2 , . . . }. Suppose for contradiction that U does not have a finite subcover. Pick
xn ∈ X \ (U1 ∪ · · · ∪ Un ) arbitrarily. Then by assumption, the sequence (xn ) has a
subsequence (yn ) which converges to some y0 ∈ X. Since U is a cover of X there is
some m ∈ N with y ∈ Um . But then yj ∈ / Um for all j ≥ m, which is a contradiction.


We mention that sequential compactness and compactness are not equivalent in

general topological spaces; for example, [0, 1]R is compact by Theorem 4.1.14, but it
can be shown that it is not sequentially compact.

Local compactness. A topological space S is called locally compact if every

p ∈ S is contained in an open set which is itself contained in a compact subset of S.
Clearly, every compact space S is also locally compact (take S itself as compact open
set that contains p).

Example 61. R is locally compact, but not compact. 4

Example 62. The discrete space on S is locally compact, but only compact if S
is finite. 4

4.2. Topological Groups

A topological group is an (abstract) group G together with a topology on the
elements G of G such that (x, y) 7→ xy is continuous from G2 to G and x 7→ x−1
is continuous from G to G. Two topological groups are said to be isomorphic if
the groups are isomorphic, and the isomorphism is a homeomorphism between the
respective topologies.

Example 63. The groups (R, +) and (Q, +) are topological groups with respect
to their standard topology. (Why?) 4

Example 64. The elements of the group G := Sym(N) form a (non-closed!)

subset of the Baire space NN (Example 48), and the topology induced by the Baire
space on Sym(N) is also called the topology of pointwise convergence. Observe that a
set of permutations of a set X is a closed subset of Sym(X) if and only if it is locally
closed as defined in Proposition 1.2.2.
Composition is continuous as a map from G2 → G. If U ⊆ G is a basic open set
S(ā, c̄) for ā, c̄ ∈ Nn (we use the terminology from Example 48), then the preimage of
 4.2. TOPOLOGICAL GROUPS 67

U is
{(f, h) ∈ G2 | f ◦ h ∈ S(ā, c̄)} = {(f, h) ∈ G2 | ∃b̄ (h ∈ S(ā, b̄) and f ∈ S(b̄, c̄))}
[

= S(b̄, c̄) × S(ā, b̄)
b̄∈Nn

which is open as a union of open sets. The preimage of S(ā, b̄) under the inverse map
is S(b̄, ā), which is open, too. 4
Proposition 4.2.1. Let G be a topological group, g ∈ G, and U ⊆ G open. Then
gU is open, too. If U is an open subgroup, then it is also closed.
Proof. As a consequence of Proposition 4.1.2, for every g ∈ G the function
tg : G → G defined by tg (x) := gx is continuous. The pre-image of U under the
function tg−1 is gU . Therefore, this set is open as the pre-image of an open set under
Scontinuous function. The second part follows since the complement of U in G equals
a
g∈G\U gU , a union of open sets, hence open. 

Remark 4.2.2. Proposition 4.2.1 also implies that the topology on G is given by a
basis at 1G : if B is a basis of open sets at the identity, and g ∈ G, then {gU | U ∈ B}
is a basis at g.

Exercises.
(100) Let G be a topological group and let A, B ⊆ G. If A is open,
then so is AB := {ab | a ∈ A, b ∈ B}.
(101) Show that a group G with a topology on G is a topological group
if and only if the map (x, y) 7→ xy −1 is continuous from G2 to G.

(102) Show that for all n ∈ N the groups GL(n, R) and GL(n, C)
of invertible real or complex matrices are topological groups
with respect to the standard topology.

A topological group G is Hausdorff (first-countable, metrizable, Polish) if the

topology of G is Hausdorff (first-countable, metrizable, Polish, respectively). Note
that G is first-countable if and only if G has a countable basis at the identity (see
Remark 4.2.2).

4.2.1. Continuous group actions. Recall from Section 1.3 that an action of
a group G on a set S is a homomorphism from G to Sym(S).
Definition 4.2.3. An action ξ of a topological group G on a topological space S
is called continuous if (g, s) 7→ ξ(g)(s) is continuous as a map from G × S → S.
Example 65. Recall the faithful action of G on G by left multiplication from the
proof of Cayley’s theorem, Theorem 1.3.3). This is the special case of Example 12
where H = {1}. This action is continuous since it equals the group composition which
is continuous by definition. 4
If S is a topological space, then Homeo(S) ⊆ Sym(S) denotes the set of all
homeomorphisms of S. We view Homeo(S) as a topological space with the subspace
topology inherited from S S which carries the product topology4.

4Note that it is not clear (and depends on S) whether Homeo(S) with this topology is a topo-
logical group.
68 4. TOPOLOGICAL GROUPS

Proposition 4.2.4. Every continuous action of a topological group G on a topo-

logical space S is a continuous homomorphism from G into Homeo(S).
Proof. Suppose that ξ : G → Sym(S) is a continuous action of G on S, so the
map χ(g, s) := ξ(g)(s) is continuous from G × S to S. For every g ∈ G, the map tg
defined by s 7→ χ(g, s) is continuous. The inverse of tg is s 7→ χ(g −1 , s), which is also
continuous. Hence, tg is a homeomorphism. Q To show that ξ is continuous, let U be
a basic open subset of Homeo(S), i.e., U = s∈S Us where Us is open in S for all
s ∈ S, and there exists a finite set F such that Us = S for all s ∈ S \ F . Note that
for fixed s, the map ts : G → S given by g 7→ ξ(g)(s) is continuous, and hence for all
s ∈ F the set {g ∈ G | ξ(g)(s) ∈ Us } is open. Therefore,
ξ −1 (U ) = {g ∈ G | ξ(g)(s) ∈ Us for all s ∈ F }
\
= {g ∈ G | ξ(g)(s) ∈ Us }
s∈F
is a finite intersection of open sets and hence open. 
If S carries the discrete topology (in which case Homeo(S) = Sym(S)), the state-
ment of Proposition 4.2.4 can be strengthened to obtain an equivalent characterisation
of continuity of actions.
Lemma 4.2.5. Let G be a topological group and ξ an action of G on a set S
equipped with the discrete topology. Then ξ is continuous if and only if ξ is continuous
as a map from G to Sym(S).
Proof. The forward implication follows from Proposition 4.2.4. For the converse
implication, we show that the function χ : G × S → S given by (g, s) 7→ ξ(g)(s) is
continuous. Since S is discrete, it suffices to show that for every s0 ∈ S there exists
an open U ⊆ G and an open T ⊆ S such that χ(U, T ) contains s0 and is contained
in χ−1 ({s0 }), because then χ−1 ({s0 }) is a union of open sets U × T . Let g ∈ G and
s ∈ S be such that χ(g, s) = s0 . Since S is discrete, in particular T := {s} is open.
Let U := ξ −1 (S(s, s0 )) which is by assumption an open subset of G and contains g.
Then χ(U, T ) = {χ(u, s) | u ∈ S(s, s0 )} = {s0 }. 
An important example of a continuous action is the action by conjugation from
Example 13.
Example 66. Let G be a topological group. Then the conjugation action ξ : G →
G given by ξ(g)(h) := ghg −1 is continuous since composition and inverse in a topo-
logical group are continuous. 4
Further important examples of continuous actions of a topological group arise
from the action by left translation (Example 12). We also view G/H as a topological
space, with the quotient topology. We first define p : G → G/H by setting p(g) = gH
(the projection map). Define U ⊆ G/H to be open if and only if p−1 (U ) is open in
G (in this way, p will necessarily be continuous). In other words, a set of left-cosets
is open in G/H if and only if their union is open in G.
Proposition 4.2.6. Let H be an open subgroup of a topological group G. Then
the action of G on G/H by left translation is continuous.
Proof. Let ξ be the action of G on G/H by left translation. Let S ⊆ G/H be
open, and let gH ∈ S. It suffices to show that there are open subsets U ⊆ G and
T ⊆ G/H such that gH ∈ {ξ(u)(t) | u ∈ U, t ∈ T } ⊆ S. By the definition of the
quotient topology p−1 (S) is open in G. Since composition in G is continuous, the
set {(g1 , g2 ) ∈ G2 | g1 g2 ∈ p−1 (S)} is open in G2 . This set contains (1, g), because
 4.2. TOPOLOGICAL GROUPS 69

p(1g) = gH ∈ S. So there exists an open U ⊆ G containing 1 and an open V ⊆ G

containing g such that {uv | u ∈ U, v ∈ V } ⊆ p−1 (S). Then T := p(V ) = {vH | v ∈
V } is open in G/H, and
gH ∈ {ξ(u)(t) | u ∈ U, t ∈ T } = {uvH | u ∈ U, v ∈ V } ⊆ S. 
Also, if G ≤ Sym(X) then for every
n ∈ N the componentwise action of G on
X n and the setwise action of G on X n are continuous. A general result about the
continuous actions of a permutation group is Theorem 5.2.1 in Chapter 5. We present
an example of a discontinuous group action of an oligomorphic permutation group.
Example 67. Let K be the class of all finite structures (A; E0 , E1 , . . . ) where Ei
denotes an equivalence relation on Ai6= with at most two equivalence classes. Clearly,
K is closed under substructures and isomorphism, and countable up to isomorphism.
It is easy to verify that it also has the amalgamation property (Section 3.3). Let A
be the Fraı̈ssé-limit of K. Then Aut(A) has a continuous homomorphism ξ1 to (Z2 )N
(which is equipped with the product topology for Z2 discrete): for α ∈ Aut(A) we
define ξ1 (α) = (αi )i∈N where αi = 0 if α fixes the equivalence classes of Ei and αi = 1
otherwise. This map is clearly a group homomorphism, and it is continuous. Indeed,
if U is an open subset of (Z2 )N , then it is a union of sets of the form U = {u ∈ (Z2 )N |
u1 = x1 , . . . , un = xn } for some n ∈ N and x ∈ (Z2 )n . Then
 
\ [
ξ1−1 (U ) =  Aut(A) ∩ S(a, b)
i∈{1,...,n},xi =0 a,b∈Ai6= ,Ei (a,b)
 
\ [
∩  Aut(A) ∩ S(a, b)
i∈{1,...,n},xi =1 a,b∈Ai6= ,¬Ei (a,b)

is open as a finite intersection of unions of open sets in Aut(A).

To construct a discontinuous group homomorphism, let U be an ultrafilter on N
(Appendix A.1), and let ξ2 : (Z2 )N → Z2 be the function given by
(
0 if {i ∈ N | αi = 0} ∈ U
(αi )i∈N :=
1 otherwise.
Claim 1. ξ2 is a group homomorphism: we have
ξ2 (α + β) = 0
⇔ {i | αi + βi = 0} ∈ U
⇔ {i | αi = βi = 0} ∪ {i | αi = βi = 1} ∈ U
⇔ {i | αi = βi = 0} ∈ U or {i | αi = βi = 1} ∈ U
⇔ ({i | αi = 0} ∈ U ∧ {i | βi = 0} ∈ U) ∨ ({i | αi = 1} ∈ U ∧ {i | βi = 1} ∈ U)
⇔ ξ2 (α) + ξ2 (β) = 0.
Claim 2. ξ2 is continuous if and only if U is principal. If there exists a ∈ N such
that U = {Y ⊆ N | a ∈ Y }, then
ξ2−1 (0) = {α ∈ (Z2 )N | {i ∈ N | αi = 0} ∈ U}
= {α ∈ (Z2 )N | αa = 0}
is a basic open set in (Z2 )N . We have the analogous statement for ξ2−1 (1), which
proves the continuity of ξ2 . Conversely, if ξ2−1 (0) is open, then it is a union of sets of
the form U = {α ∈ (Z2 )N | αa1 = b1 , . . . , αak = bk } for some k ∈ N, a1 , . . . , ak ∈ N
and b1 , . . . , bk ∈ Z2 . If ξ2−1 (0) contains U , then {ai | i ∈ {1, . . . , k}, bi = 0} ∈ U.
70 4. TOPOLOGICAL GROUPS

Hence, there exists i ∈ {1, . . . , k} such that {ai } ∈ U (see Lemma A.1.3) and U is
principal.
In the same way it can be shown that for a non-principal ultrafilter U, the map
ξ2 ◦ ξ1 is a discontinuous group homomorphism from an oligomorphic permutation
group to Z2 . 4

Proposition 4.2.7 (Proposition 13 and Proposition 14 in [35]). Let G be a

topological group, and let H be a subgroup of G. Then
• H is open in G if and only if G/H is discrete;
• H is closed in G if and only if G/H is Hausdorff.

Proof. Part 1: G/H is discrete if each left coset gH is open, which is the case
if and only if H is open.
Part 2: If G/H is Hausdorff, then every coset of H that is distinct from H is
contained in an open set O that does not intersect H. The union of all those open
sets is open, and defines the complement of H in G/H. Hence, H is closed.
Conversely, suppose that H is closed. Then the equivalence relation
[
R := {(x, y) | x−1 y ∈ H} = (gH)2
g∈G

on G is closed in G × G, since Sit is the inverse image of H under the continuous

mapping (x, y) 7→ x−1 y. Hence, g∈G {gH}2 is closed in G/H, and
[
U := (G/H)2 \ {gH}2
g∈G

is open in G/H. Let g1 H, g2 H ∈ G/H be distinct, that is, (g1 , g2 ) ∈ / R. By the

definition of the product topology there exist open sets U1 , U2 of G/H such that
{g1 H} × {g2 H} ⊆ U1 × U2 ⊆ U . The sets U1 and U2 are disjoint: otherwise, if
gH ∈ U1 ∩ U2 for some g ∈ G, then {gH} × {gH} ∈ U . So (g, g) ∈ G2 \ R, a
contradiction to the fact that R contains (g, g) for all x ∈ G. This proves that G/H
is Hausdorff. 

Example 68 (The Logic Action). Let τ be a relational signature, and let C be a

class of finite τ -structures which is closed under substructures and isomorphisms, has
the JEP, and contains only finitely many non-isomorphic structures with n elements
for each n ∈ N. Let XC be the space of all structures with domain N whose age is
contained in C. The basic open sets in XC are given by elements A of C together with
a map α : A → N as follows:

{B ∈ XC | α : A ,→ B is an embedding}.

Note that this topology is compact because it is a closed subset of a product of finite
spaces (TODO: explain more, Proposition 4.1.13, Theorem 4.1.14).
We now define the so-called logic action of Sym(N) on XC : for g ∈ Sym(N)
and B ∈ XC , define g(B) to be the unique structure B 0 in XC such that g is an
isomorphism between B and B 0 . Clearly, this action is continuous. 4

Exercises.
(103) Show that the componentwise
action of G ≤ Sym(X) on X n and
the setwise action of G on X
n are continuous.
 4.2. TOPOLOGICAL GROUPS 71

4.2.2. Topologically faithful actions. Recall that an action on S is called

faithful if it is an injective homomorphism to Sym(S). A faithful action of a closed
subgroup of Sym(N) on a discrete space S is called topologically faithful if it is con-
tinuous and its image is closed in Sym(S).
Example 69. Let H be an open subgroup of G. By Proposition 4.2.7 the quotient
space G/H is discrete, so in particular the image of the action of G on G/H by left
translation is closed. 4
We also give an example of a continuous injective homomorphism from a closed
subgroup G of Sym(N) to Sym(N) whose image is not closed. That is, we have a
faithful continuous action of G which is not topologically faithful; G is even oligomor-
phic.
Example 70. This example is due to Dugald Macpherson and can be found in
Hodges’ model theory [72] (on page 354). Let Q be the structure (Q; <, P ) where
• < is the usual strict order of the rational numbers, and
• P ⊆ Q is such that both P and O := Q \ P are dense in (Q; <).
Let P be the substructure induced by P in Q. It is easy to see (and follows from
more general principles that will be presented in Corollary 5.3.3) that the mapping
which sends f ∈ Aut(Q) to f |P induces a continuous homomorphism ξ from Aut(Q)
to Aut(P ) whose image is dense in Aut(P ). We claim that ξ is not surjective. To
prove this, we consider Dedekind cuts (S, T ) of P , that is, partitions of P into subsets
S, T with the property that for all s ∈ S and t ∈ T we have s < t. Note that for
each element o ∈ O we obtain a Dedekind cut (S, T ) with S := {a ∈ P | a < o} and
T := {a ∈ P | a > o}. But since there are uncountably many Dedekind cuts (they
are in bijection with the real numbers, see 56) and only countably many elements of
O, there also exists a Dedekind cut (S 0 , T 0 ) which is not of this form. By a standard
back-and-forth argument, there exists an α ∈ Aut((P, <)) that maps S to S 0 and
T to T 0 . Suppose for contradiction that there is β ∈ Aut(Q) with β|P = α. Then
s < β(o) < t for all s ∈ S 0 , t ∈ T 0 , in contradiction to the assumptions on (S 0 , T 0 ). 4
Some other groups have the remarkable property that every faithful continuous
action is topologically faithful; this is for example known for Sym(N) (due to [61]; see
Theorem 1.3 in [162] for a more recent and more powerful result in this context).

4.2.3. Metrics on topological groups. The (ultra-) metric d on Sym(D) from

Example 52 is left-invariant, i.e., d(gh1 , gh2 ) = d(h1 , h2 ) for all g, h1 , h2 ∈ G, because
if n ∈ N is smallest such that h1 (n) 6= h2 (n), then n is also smallest such that
g(h1 (n)) 6= g(h2 (n)).
Theorem 4.2.8 (Birkhoff, Kakutani; see Theorem 9.1 in [88]). A topological
group G is metrisable if and only if G is Hausdorff and first-countable. Every metris-
able topological group has a compatible left-invariant metric.
Lemma 4.2.9. Let G be a group with a left-invariant metric d. Then for all
g, h ∈ G
d(gh, 1G ) ≤ d(g, 1G ) + d(h, 1G ).
Proof. d(gh, 1G ) = d(h, g −1 ) ≤ d(h, 1G ) + d(1G , g −1 ) = d(g, 1G ) + d(h, 1G ). 

Proposition 4.2.10. Let ξ : G → H be a continuous homomorphism between

topological groups with compatible left-invariant metrics d1 and d2 . Then f is uni-
formly continuous.
72 4. TOPOLOGICAL GROUPS

Proof. Let  > 0. Since ξ is continuous, there exists a δ > 0 such that for all
g ∈ G with d1 (1G , g) < δ we have d2 (1H , ξ(g)) < . Let g1 , g2 ∈ G be such that
d1 (g1 , g2 ) < δ. Then d1 (1G , g1−1 g2 ) < δ, and hence
d2 (ξ(g1 ), ξ(g2 )) = d2 (1H , ξ(g1 )−1 ξ(g2 )) = d2 (1H , ξ(g1−1 g2 )) < 
which shows uniform continuity of ξ. 
We have seen in Example 52 an example of a left-invariant metric d on Sym(D)
which is not complete.
Lemma 4.2.11. Let G be a topological group with a compatible left-invariant metric
d. Then
d0 (g, h) := d(g, h) + d(g −1 , h−1 )
is a compatible metric, too.
Proof. Clearly, d0 is non-negative, indiscernible, symmetric, and subadditive.
We have to show that d0 induces the same topology on G as d. Let  > 0. The set
S := {g ∈ G | d(1, g) < } is open with respect to d and contains the identity 1 ∈ G.
This set contains the set S 0 := {g ∈ G | d0 (1, g) < } which is open with respect to d0
and also contains 1. Conversely, the set S 0 contains the set {g ∈ G | d(1, g) < /2}
(which is open with respect to d and contains 1): if g is such that d(1, g) < /2, the
by the left-invariance of d we have that d(g −1 , 1) < /2, and hence
d0 (1, g) ≤ d(1, g) + d(1, g −1 ) < /2 + /2 < ,
so g ∈ S2 . Since the topology on G is given by a basis of open sets at 1, the statement
follows. 
Lemma 4.2.12. Let G be a topological group, let d be a compatible left-invariant
metric, and let d0 be the compatible metric defined in Lemma 4.2.11. Let (gi )i∈N and
(hi )i∈N be Cauchy sequences in (G, d0 ). Then (gi−1 hi )i∈N is Cauchy in (G, d) and in
(G, d0 ).
Proof. Let  > 0. Then there exists an n0 ∈ N such that
d0 (hn , hm ) = d0 (h−1 G
m hn , 1 ) < /3 (6)
for all n, m ≥ n0 . By the continuity of the multiplication operation and since d is a
compatible metric there exists a δ > 0 such that for all k ∈ G with d(k, 1G ) < δ
d(h−1 G
n0 khn0 , 1 ) < /3.

Let n1 ≥ n0 be such that d0 (gn , gm ) < δ for all n, m ≥ n1 . Then for all n, m ≥ n1
d(gm gn−1 , 1G ) = d(gn−1 , gm
−1
) ≤ d0 (gn , gm ) < δ
and hence
d(h−1 −1 G
n0 gm gn hn0 , 1 ) < /3. (7)
Therefore
d(gn−1 hn , gm
−1
hm )
= d(h−1 −1 G
m gm gn hn , 1 )

≤ d(h−1 G −1 −1 G −1 G
m hn0 , 1 ) + d(hn0 gm gn hn0 , 1 ) + d(hn0 hn , 1 ) (Lemma 4.2.9)
≤ /3 + /3 + /3 =  (by (6), (7)), and (6))
which proves that (gi−1 hi )i∈N
is Cauchy in (G, d). Note that by symmetry also the
sequence (h−1
i gi )i∈N is Cauchy in (G, d), and it follows that both sequences are also
Cauchy in (G, d0 ). 
 4.2. TOPOLOGICAL GROUPS 73

Lemma 4.2.13. Let G be a topological group with a compatible left-invariant metric

d, and let d0 be the metric defined in Lemma 4.2.11. Let (G∗ , d∗ ) be the metric
completion of (G, d0 ). Then the group multiplication can be extended uniquely to G∗
such that G∗ becomes a topological group with the compatible complete metric d∗ .
Proof. To define an extension of the multiplication operation of G to G∗ , pick
representatives (gn )n∈N and (hn )n∈N of elements of G∗ (recall our construction of G∗
in Proposition 4.1.10). Then (gn hn )n∈N is Cauchy with respect to d0 by Lemma 4.2.12.
Define
[(gn )n∈N ] · [(hn )n∈N ] := [(gn hn )n∈N ].
To show that this is well-defined, let (gn0 )n∈N and (h0n )n∈N be Cauchy sequences in
(G, d0 ) such that limn∈N d(gn , gn0 ) = 0 and limn∈N d(hn , h0n ) = 0.
Let  > 0. We will show that d(gn0 h0n , gn hn ) ≤ . Since (hn )n∈N is a Cauchy
sequence, there exists n0 ∈ N such that d(hn , hm ) < /3 for all n, m ≥ n0 . By the
continuity of the multiplication operation there exists a δ > 0 such that for all k ∈ G
with d(k, 1G ) < δ
d(h−1 G
n0 khn0 , 1 ) < /3.
Let n1 ≥ n0 be such that d0 (gn0 , gn ) < δ for all n ≥ n1 . Then for n ≥ n1 and by
Lemma 4.2.9
d(gn0 h0n , gn hn ) = d(h−1 −1 0 0 G
n gn gn hn , 1 )

≤ d(h−1 G −1 −1 0 G −1 0 G
n hn0 , 1 ) + d(hn0 gn gn hn0 , 1 ) + d(hn0 hn , 1 )
≤ /3 + /3 + /3 = .
This shows that [(gn0 h0n )n∈N ] = [(gn hn )n∈N ] and hence the multiplication on G∗ is
indeed well-defined.
The multiplication operation defined on G∗ is associative and has the neutral
element [(1G )n∈N ]. The inverse of [(gn )n∈N ] is [(gn−1 )n∈N ] (Lemma 4.2.12 implies that
(gn−1 )n∈N is Cauchy). We use Proposition 4.1.1 to verify that the multiplication and
taking inverses in G∗ is continuous with respect to the topology induced by d∗ : if
limm→∞ limn→∞ gn,m = g and limm→∞ limn→∞ hn,m = h then
  h i
−1 −1
lim [ lim gn,m ] · [ lim hn,m ] = lim lim gn,m · lim hn,m
m→∞ n→∞ n→∞ m→∞ n→∞ n→∞
h i
−1
= lim lim gn,m · lim hn,m
m→∞ n→∞ n→∞
 
= −1
lim gn,m hn,m = [g −1 · h]
n,m→∞

so that we indeed obtain a topological group G∗ . 

Lemma 4.2.14 (Proposition 2.2.1 in [60]). Let G be a Polish group and U a
subgroup. Then U is Polish if and only if U is closed in G.
Proof. Suppose first that U is closed in G. Let S ⊆ G be a countable set that
is dense in G and let d be a compatible metric on G. Let n ∈ N. For every u ∈ U
1
there exists su ∈ S such that d(u, su ) < n+1 . From all the elements v ∈ U such that
su = sv , select one. Let T be the set of all elements of U selected like this, for all
values of n ∈ N. Then T is countable because S is countable. Also note that T is
dense in U , so U is separable. The restriction of a complete compatible metric for G
metric to U is certainly also a compatible metric for U , and it is complete since U is
closed.
Conversely, we assume that U is Polish. Let Ū be the closure of U in G; note that
Ū is Polish. By Lemma 4.1.6 we have that U is a countable intersection of open sets.
74 4. TOPOLOGICAL GROUPS

Suppose for contradiction that there exists a g ∈ Ū \ U . Since U is dense in Ū , the

coset gU is dense in Ū and also a countable intersection of open sets. By the Baire
category theorem (Theorem 4.1.8) applied to the Polish group Ū , the intersection of
gU and U is dense in Ū , and in particular non-empty, which is impossible. 

Lemma 4.2.15. Let G be a Polish group with a compatible left-invariant metric

d. Then d0 (g, h) := d(g, h) + d(g −1 , h−1 ) is a compatible complete metric for G.

Proof. Let (G∗ , d∗ ) be the metric completion of (G, d0 ) (see Proposition 4.1.10).
By Lemma 4.2.13, G∗ can be viewed as a topological group G∗ with the compatible
complete metric d∗ , and since G∗ is also separable we conclude that G∗ is Polish. Since
G is Polish, too, Lemma 4.2.14 implies that G is closed in G∗ . Therefore, G = G∗ .
This shows that d0 is a compatible complete metric on G. 

Exercises.
(104) Show that Sym(N) does not admit a compatible complete
and left-invariant metric.
(105) Show that a Polish group has a compatible and complete
left-invariant metric if and only if every compatible
left-invariant metric is complete.
(106) Show that no oligomorphic permutation group G on a countably
infinite set has a compatible complete and left-invariant metric.

Hint. We may assume that G is closed, so that we can write G = Aut(A).

Then apply the compactness theorem of first-order logic to construct an
elementary self-embedding of A which is not surjective.

Alternative direct solution. Let d be the complete left-invariant metric

from Example 52. By the previous exercise, it suffices to show that d is not
complete. We define the sequence (ai )i∈N of elements of N inductively as
follows:
• Let a1 be any natural number in an infinite orbit of G.
• A0 := ∅.
• AN := {0, 1, 2, 3, . . . , an }.
• Let an+1 be any element strictly larger than an such that an and an+1
are in the same infinite orbit with respect to GAn−1 and the orbit of
an+1 with respect to GAn is infinite. This is possible, because the
infinite GAn−1 -orbit containing an only splits into finitely many GAn
orbits which need to contain an infinite orbit.
• For every n ∈ N, let gn , hn ∈ G be such that h0 = idN , hn+1 = hn ◦ gn ,
and gn with n ≥ 1 is an element of GAn−1 mapping an+1 to an .
For n < m, we have

d(hn , hm ) = d(g1 g2 . . . gn , g1 g2 . . . gm )
= d(1, gn+1 . . . gm ) ≤ 2−an ≤ 2−n

using the left invariance of d and that gn+1 . . . gm ∈ GAn . This shows that
(hn )n∈N is Cauchy with respect to d. But this sequence is not convergent
since the preimage of a1 after n − 1 steps is an which is a non-convergent
sequence.
 4.3. CLOSED SUBGROUPS 75

4.3. Closed Subgroups

In this section we give a topological characterisation of those topological groups
that appear as automorphism groups of countable structures. Since the automorphism
group of a structure A is a closed subgroup of Sym(A) (Proposition 1.2.2), and since
Sym(A) is Polish (Example 54), it follows from Lemma 4.2.14 that Aut(A) is Polish.
But also (R; +) is Polish, and it certainly isn’t the automorphism group of a countable
structure, so we need to identify more properties of closed subgroups of Sym(N).
A topological group is non-archimedian if it has a basis at the identity consisting
of open subgroups.
Example 71. The group (R; +) is archimedian: for all a, b ∈ R with 0 < a ≤ b
there exists an n ∈ N such that na := a + a + · · · + a ≥ b. Hence, the open interval
(−b, b) does not contain any non-trivial open subgroup, since if the subgroup contains
a ∈ (0, b), then it also contains elements larger than b. This implies that (R; +) is not
non-archimedian in the sense above and motivates the terminology. 4
Example 72. The group Sym(D) is non-archimedian: the point stabilisers Ga
for a ∈ Dn , n ∈ N, are open subgroups of G and they form a basis at the identity. 4
Remark 4.3.1. Note that if G and H are non-archimedian, and if we want to
verify that a given group homomorphism h : G → H is continuous, it suffices to verify
that the preimage of every open subgroup of H is open in G (see Remark 4.2.2).
Likewise, h is open if the image of every open subgroup of G is open in H.
Automorphism groups of countable structures can be characterised in topological
terms.
Theorem 4.3.2 (Section 1.5 in [9]; also see Theorems 2.4.1 and 2.4.4 in [60]).
Let G be a topological group. Then the following are equivalent.
(1) G is topologically isomorphic to the automorphism group of a countable re-
lational structure.
(2) G is topologically isomorphic to a closed subgroup of Sym(N).
(3) G is Polish and admits a compatible left-invariant ultrametric.
(4) G is Polish and non-archimedian.
Proof. The equivalence of (1) and (2) has been shown in Proposition 1.2.2.
The implication from (2) to (3) has been explained in the paragraphs preceding the
statement of the proposition. So it suffices to show (3) ⇒ (4) ⇒ (2).
For the implication from (3) to (4), let d be a left-invariant ultrametric on G.
Let Un = {g ∈ G | d(g, 1) < 2−n }, for n ∈ N. We claim that the set of all those Un
forms a basis at the identity consisting of open subgroups. Since d is a left-invariant
ultra-metric, for g, h ∈ Un we have
d(g −1 h, 1) = d(h, g) ≤ max (d(h, 1), d(1, g)) < 2−n .
Hence, g −1 h ∈ Un and Un is indeed a subgroup.
For the implication from (4) to (2), let {B1 , B2 , . . . } be an at most countable basis
at the identity (which exists since G is metrisable). Each Bi has an open subset Vi
which is a subgroup, since G has a basis at the identity consisting of open subgroups.
Then {V1 , V2 , . . . } is a countable basis of the identity consisting of open subgroups.
Each Vi has at most countably many cosets since G is separable. So the set of all
cosets of those groups gives an at most countable basis B := {U1 , U2 , . . . } that is
closed under left multiplication. If B is infinite, we define the map ξ : G → Sym(N)
by setting
ξ(g)(n) = m ⇔ gUn = Um .
76 4. TOPOLOGICAL GROUPS

If |B| = n0 is finite, we define the map ξ : G → Sym(N) similarly, but set ξ(g)(n) = n
for all n > n0 .

Claim 1. ξ is a homomorphism. We have ξ(f g) = ξ(f )ξ(g) since

ξ(f )ξ(g)(n) = m ⇔ f (gUn ) = Um ⇔ f gUn = Um ⇔ ξ(f g)(n) = m .

Claim 2. ξ is injective: when f, g ∈ G are distinct, then there are disjoint open
subsets U and V with f ∈ U and g ∈ V , because the topology is metrisable and
therefore Hausdorff; since B is a basis, we can assume that U = Un for some n ≥ 1.
If f Un = gUn , then g ∈ Un = U since f ∈ Un , contradicting the assumption that U
and V are disjoint. Hence, ξ(f )(n) 6= ξ(g)(n), and so ξ(f ) 6= ξ(g). So ξ is indeed an
isomorphism between G and a subgroup of Sym(N).

Claim 3. ξ is continuous. Let V ⊆ Sym(N) be an open set. Then V is a union

of basic open sets S(ā, b̄) for some ā, b̄ ∈ Nn . Let i ≤ n and g, h ∈ G be such that
g ◦ h ∈ Ubi . Since composition in G is continuous and Ubi is open, there is an open
subset Gg,h of G containing g and an open set Hg,h of G containing h such that
(g, h) ∈ Gg,h × Hg,h ⊆ ◦−1 (Ubi ). We then have
\
ξ −1 (S(ā, b̄)) = {g ∈ G | gUai = Ubi }
i≤n
\ [
= Gg,h .
i≤n g∈G,h∈Uai

This set is a finite intersection of a union of open sets and thus open. Hence, ξ −1 (V )
is a union of open sets and therefore open as well, which concludes the proof that ξ
is continuous.

Claim 4. The map ξ is open and the image of ξ is closed in Sym(N). Let
• d1 be the left-invariant compatible metric on G (Theorem 4.2.8),
• d01 be the compatible complete metric on the Polish group G defined as
d01 (g, h) = d1 (g, h) + d1 (g −1 , h−1 ) (see Lemma 4.2.15), and
• d02 be the compatible complete metric on Sym(N) from Example 54.
We will show that ξ −1 is Cauchy-continuous as a map from (ξ(G), d02 ) to (G, d01 ). This
clearly implies both parts of the claim.
Let g1 , g2 , . . . be a sequence in G such that ξ(g1 ), ξ(g2 ), . . . converges against
h ∈ Sym(N). We have to show that g1 , g2 , . . . is d01 -Cauchy. Since d1 is left-invariant,
−1
limn,m→∞ d1 (gn , gm ) = 0 if and only if limn,m→∞ d1 (gm gn , 1) = 0. Let  > 0 be
arbitrary. Since B is a basis, there exists Uk ∈ B such that

Uk ⊆ {g ∈ G | d1 (g, 1) < /2}

and Uk Uk−1 ⊆ {g ∈ G | d1 (g, 1) < }. Since limn→∞ ξ(gn ) = h, there exists an n0
such that ξ(gn )(k) = ξ(gm )(k) = h(k) for all n, m > n0 . Then gn Uk = gm Uk , and so
−1
gm gn ∈ Uk Uk−1 ⊆ {g ∈ G | d1 (g, 1) < } .

Hence, d1 (gm gn−1 , 1) < , and limn,m→∞ d1 (gm

−1
gn , 1) = 0. Similarly one can show
that limn,m→∞ d1 (gm gn , 1) = 0, using the fact that ξ(gn−1 ) = ξ(gn )−1 , and hence
−1

limn→∞ ξ(gn−1 ) = h−1 . Thus, limn,m→∞ d1 (gn , gm ) = 0 = limn,m→∞ d1 (gn−1 , gm −1

),
0
and therefore limn,m→∞ d1 (gn , gm ) = 0. 
 4.4. OPEN SUBGROUPS 77

4.4. Open Subgroups

We present a simple but useful characterisation of the open subgroups of permu-
tation groups. Recall the definition of point stabilisers from Definition 1.2.5.
Lemma 4.4.1. Let G be subgroup of Sym(N) and let U be a subgroup. Then the
following are equivalent.
(1) U is open in G;
(2) U contains Gt for some t ∈ Nn , n ∈ N;
(3) U contains an open subset of G.
Proof. (1) ⇒ (2): Since U is open in G it must contain G ∩ S(a, b) for some
a, b ∈ Nn since these sets form a basis of the topology of Sym(N). Every element of
Ga can be written as αβ with α ∈ G ∩ S(b, a) ⊆ U and β ∈ G ∩ S(a, b) ⊆ U . Hence,
U contains Ga .
(2) ⇒ (3): trivial since Gt is open in G. S
(3) ⇒ (1): Let H be an open subset of U . Then U = β∈U βH. Since βH are
open, it follows that U is open, too. 

It follows that all open subgroups of Sω have countable index.

Remark 4.4.2. We have seen in Proposition 4.2.1 that every open subgroup of a
topological group is closed. The converse is false: for example, when E is an equiva-
lence relation on a countably infinite set B with two infinite classes, then Aut(B; E)
is a closed subgroup of Sym(B) (we already saw in Proposition 1.2.2 that the closed
subgroups of the automorphism group of a structure A correspond precisely to arbi-
trary expansions of A), but does not contain the point stabiliser of some finite subset
of B.
Next, we present another characterisation of the open subgroups U of subgroups
G of Sω which more explicitly describes all elements of U . Let U ≤ G, n ∈ N, and
a ∈ B n be such that U contains Ga . Define EU,a to be
{(α(a), αβ(a)) | β ∈ U, α ∈ G}.
Then EU,a is a congruence of the componentwise action of G on B n : it is preserved by
G, and an equivalence relation on B n . Reflexivity is clear. For symmetry, let α ∈ G
and β ∈ U , so that (α(a), αβ(a)) ∈ EU,a ; we have to show that (αβ(a), α(a)) ∈ EU,a .
Let γ := αβ ∈ G. Then (αβ(a), α(a)) = (γ(a), γβ −1 (a)) ∈ EU,a since β −1 ∈ U . For
transitivity, let (u, v), (v, w) ∈ E := EU,a . Then u = α(a), v = αβ(a) = α0 (a), and
w = α0 β 0 (a), for α, α0 ∈ G, β, β 0 ∈ U . Hence, β −1 α−1 α0 ∈ Ga ⊆ U and α−1 α0 ∈ U .
So we obtain that (u, w) = (α(a), α0 β 0 (a)) = (α(a), α(α−1 α0 β 0 )(a)) ∈ EU,a .
Lemma 4.4.3. Let G ≤ Sym(B) and U ≤ G. Then the following are equivalent.
(1) U is open in G.
(2) There is a ∈ B n , n ∈ N, such that Ga ≤ U = {α ∈ G | (a, α(a)) ∈ EU,a }.
(3) U = G{S} is the set stabiliser of a block S of the componentwise action of
G on B n for some n ∈ N.
Proof. (1) ⇒ (2). Let U be an open subgroup of G. Then Lemma 4.4.1 implies
that U must contain Ga for some n ∈ N and some a ∈ B n . Moreover, we claim that
U = {α ∈ G | (a, α(a)) ∈ EU,a }. (8)
Indeed, if β ∈ U , then
(a, β(a)) ∈ EU,a = {(α(a), αγ(a)) | γ ∈ U, α ∈ G}
78 4. TOPOLOGICAL GROUPS

as witnessed by γ = β and α = idB . This proves ⊆ in (8). Conversely, if α ∈ G

is such that (a, α(a)) ∈ EU,a , then (a, α(a)) = (α0 (a), α0 β(a)) for some β ∈ U and
α0 ∈ G. Hence, α0 ∈ Ga and α−1 α0 β ∈ Ga , and since Ga ⊆ U we have that α ∈ U .
This proves ⊇ in (8).
(2) ⇒ (3). Let S be the congruence class of EU,a which contains a. Lemma 1.4.2
shows that S is a block of the componentwise action of G on B n . We claim that
GS = {α ∈ G | (a, α(a)) ∈ EU,a }.
Let β ∈ GS . Then β(a) ∈ S and hence (a, β(a)) ∈ EU,a . This shows that β ∈ {g ∈
G | (a, α(a)) ∈ EU,a }. Conversely, suppose that α ∈ G is such that (a, α(a)) ∈ EU,a ;
then α(a) must be in the same congruence class of EU,a as a, which is S, and hence
α ∈ GS .
(3) ⇒ (1). Let S ⊆ B n be a block and let C be the corresponding congruence
of G. Arbitrarily pick an s ∈ S. To show that U = G{S} is open it suffices by
Lemma 4.4.1 that G{S} contains Gs . Let α ∈ Gs and t ∈ S. Then (s, t) ∈ C and
hence (αs, αt) ∈ C. Since αs = s ∈ S we conclude that αt ∈ S. So α ∈ G{S} . 
Lemma 4.4.3 has the following consequence.
Corollary 4.4.4. Every oligomorphic subgroup of Sym(N) has countably many
open subgroups.
Proof. An oligomorphic group G has for each n finitely many congruences of the
componentwise action of G on B n , and at most countably many congruence classes
for each congruence. 
So in particular Sym(N) itself has only countably many open subgroups. We
mention another fact about open subgroups of oligomorphic permutation groups.
Corollary 4.4.5 (Lemma 2.4 in [74]). Let G be an oligomorphic subgroup of
Sym(N) and let U be an open subgroup of G. Then U is contained in only finitely
many subgroups of G.
Proof. By Lemma 4.4.1, U contains Ga for some a ∈ Nn and n ∈ N. So it
suffices to show the statement for U = Ga . Let H be any subgroup of G that contains
Ga . By (8), H is of the form
{α ∈ G | (a, α(a)) ∈ EH,a }.
As G is oligomorphic, there are finitely many congruence relations of G on B n , which
implies the statement. 

Exercises.
(107) Let A be a countable ω-categorical structure. Let B ⊆ A be finite and let
C := aclA (B). Then Aut(A){C} is open in Aut(A).

4.5. Compact Subgroups

Proposition 4.5.1. Let G be a compact subgroup of Sym(N). Then all orbits of
G are finite.
Proof. Let O be an infinite orbit of G, and fix a ∈ O. Then the sets S(a, b) for
b ∈ O form an open partition of G, and hence no finite sub-collection of those sets
can cover G. Hence, if O has infinite orbits, then G is not compact. 
Proposition 4.5.2. Let G be a closed subgroup of Sym(N). Then G is compact
if and only if all orbits of G are finite.
 4.5. COMPACT SUBGROUPS 79

Subgroups of Sym(N)

compact

finite locally compact oligomorphic

closed and countable

Figure 4.1. An illustration of the relationship between basic finite-

ness properties of subgroups of Sym(N): finite, countable, compact,
locally compact, and oligomorphic.

Proof. The forward implication follows from Proposition 4.5.1. For the other
direction, suppose that the orbits O1 , O2 , . . . of G are finite. We write G|Oi for
the permutation
Q group formed by the restrictions of G to the finite set Oi . Then
G = i G|Oi is a closed subset of a product of finite subgroups of G, and hence
compact by Tychonoff’s theorem (Theorem 4.1.14) and Proposition 4.1.13. We do
not use the entire strength of Tychonoff’s theorem, and show two alternative proofs.

Second Proof. Let {Ui }i∈A be an open cover of G. Since Sym(N) and hence
G are second-countable, we can assume that A = N (Proposition S 4.1.17). Suppose
for contradiction that for no finite B ⊆ A we have that G ⊆ i∈B Ui . Consider the
following rooted tree. The vertices on level n are the restrictions of the permutations
in G to {1, . . . , n}. Adjacency between a vertex on level n and vertices on level n + 1
is defined by restriction. Clearly, for every n there are finitely many vertices on level
n since the orbit of (1, . . . , n) with respect to the componentwise action of GSis finite.
A vertex on level n is good if it is the restriction of a function from G \ i≤n Un .
Clearly, the restriction of a good vertex is good. Moreover, by assumption there are
good vertices on all levels. By König’s tree lemma, there is an infinite branch of good
vertices in the tree. This branch defines an injection from N to N. In fact, it must
be a bijection since the finiteness of the orbits implies that the map is surjective onto
each orbit. This map is from G since G is closed, but it does not lie in any of the Ui ,
a contradiction.

Third proof. Our final proof uses Theorem 4.1.19. Clearly G is complete since
it is closed in Sym(N). To prove total boundedness of G, let  > 0. Choose n ∈ N
such that 1/2n < . Since all orbits of G are finite, the orbit O of (1, . . . , n) with
respect to the componentwise action of G on Nn is finite too; choose f1 , . . . , fk ∈ G
so that O = {f1 (1, . . . , n), . . . , fk (1, . . . , n)}. Then by construction Bf1 (), . . . , Bfk ()
covers all of G. 

It follows from Proposition 4.5.1 that a compact subgroup G of Sym(N) must have
infinitely many orbits, and in particular cannot be oligomorphic. In fact, oligomor-
phicity is already ruled out by local compactness; this can be seen from Lemma 3.0.1
and the following.
80 4. TOPOLOGICAL GROUPS

Corollary 4.5.3. Let G ≤ Sym(N) be locally compact. Then there exists a ∈ Nn ,

n ∈ N such that Ga has only finite orbits. If G is additionally closed, then also the
converse holds.
Proof. If G is locally compact there exists an open set U ⊆ G that contains 1G
and that is contained in a compact set K ⊆ G. Since U is open and contains 1G there
exists a finite tuple ā ∈ Nn , n ∈ N, such that Ga ⊆ U . Then Ga is a closed subset
of the compact set K and hence compact by Proposition 4.1.13. The statement then
follows from Proposition 4.5.1.
Conversely, if G is closed, then so is Ga , and hence Ga is compact by Theo-
rem 4.5.2. To show local compactness of G, let α ∈ G. Then αGa is an open and
compact set which contains α, and hence G is locally compact. 

Also note that locally compact subgroups of Sym(N) contain all closed countable
subgroups of Sym(N) by Theorem 1.2.6; see Figure 4.1.
Exercises.
(108) Prove that every countable compact subgroup of Sym(N) is finite.
(109) Prove that every closed subgroup of Sym(N) which is not locally compact has
a homeomorphism ξ to Sym(N) such that ξ and ξ −1 are uniformly continuous
(with respect to the metric inherited from the Baire space).

4.6. Closed Normal Subgroups

Example 73. Let E be an equivalence relation on a countably infinite set D such
that all equivalence classes B1 , B2 , . . . of E have finite size. Then the oligomorphic
group Aut(D; E) has the closed normal subgroup Aut(D; B1 , B2 , . . . ) (which is not
oligomorphic). 4
Proposition 4.6.1. Let G be a closed subgroup of Sym(B). If E is an equivalence
relation on B n , n ∈ N, which is preserved by G, then the subgroup of G that preserves
each equivalence class of E is closed and normal. Conversely, every closed normal
subgroup of G is the intersection of closed normal subgroups that arise in this way.
Proof. Let C be the expansion of B by a unary relation for each equivalence
class of E. Then Aut(C) is closed by Proposition 1.2.2, and it is a normal subgroup
of Aut(B): when g ∈ Aut(B) and h ∈ Aut(C), then g ◦ h ◦ g −1 preserves each
equivalence class of E, and thus is an automorphism of C. Normality of Aut(C)
follows from Proposition 1.5.1.
For the second part, suppose that G has a closed normal subgroup N . Consider
the relation
Rn := {(x, y) | x, y ∈ B n and there is h ∈ N such that h(x) = y} .
This relation is obviously an equivalence relation, and it is preserved by all the au-
tomorphisms of B. For this, we have to show that for all g ∈ G and all (x, y) ∈ Rn
we have that (g(x), g(y)) ∈ Rn . So suppose that x, y ∈ B n such that h(x) = y for
some h ∈ N . Then g(y) = g(h(x)) ∈ (gN )(x) = (N g)(x) = N (g(x)) by normality
of N . Hence there exists an h0 ∈ N such that h0 (g(x)) = g(y), which shows that
(g(x), g(y)) ∈ Rn .
Let C be the structure that contains for all n the n-ary relations given by the
equivalence classes of the relations Rn for all n ≥ 0. We claim that N is precisely
the automorphism group of C. As in the first part we can verify that every h ∈ N is
an automorphism of C. The converse follows by local closure as follows. Let g be an
automorphism of C, and let x, y be from B n so that g(x) = y. Since g preserves the
 4.6. CLOSED NORMAL SUBGROUPS 81

equivalence classes of Rn , there exists an h ∈ N such that h(x) = y. Hence, g lies in

the closure of N , which implies that g is from N since N is closed. 
Example
 74. The automorphism group G of the structure B = (Q; Betw), where
Betw = (x, y, z) | (x < y < z) ∨ (z < y < x) , is 2-transitive and therefore primitive.
However, the relation

((x1 , x2 ), (y1 , y2 )) | (x1 < x2 ∧ y1 < y2 ) ∨ (x1 > x2 ∧ y1 > y2 ) ∨ (x1 = x2 ∧ y1 = y2 )
is an equivalence relation on Q2 that is preserved by G. And indeed, G has a closed
normal subgroup N that is isomorphic to the automorphism group of (Q; <), and
G/N has two elements, corresponding to the automorphisms that reverse the order
<, and the automorphisms that preserve the order. 4
Theorem 4.6.2. The group Sym(N) is topologically simple, i.e., it has no proper
non-trivial closed normal subgroups.
Proof. The statement follows from a stronger statement about all the normal
subgroups of Sym(N). There are four such subgroups, which is an old result that has
been discovered independently by several authors [5, 127, 141]: besides the trivial
subgroup {idN } and the full subgroup Sym(N), there is only the finitary alternat-
ing group A (Exercise 10) and the subgroup P of permutations with finite support
(Example 9, Exercise 27). Clearly, the latter two are not closed.
A self-contained proof of Sebastian Meyer goes as follows. Suppose that H is a
closed non-trivial normal subgroup of Sym(N). Then there exists f ∈ H \ {idN } be-
cause H is non-trivial. Let a ∈ N be such that f (a) 6= a. Let c ∈ N \ {a, f −1 (a), f (a)}.
Since H is a normal subgroup of Sym(N), for any g ∈ Sym(N) we have that g −1 f g ∈ H
and (g −1 f −1 g)f ∈ H. Let g = g −1 ∈ Sym(N) be the map that exchanges a and c and
fixes all other elements of N. Then
g −1 f −1 gf (a) = g −1 (a) = c
and for x ∈ N \ {a, f −1 (a), c, f −1 (c)} we have
g −1 f −1 gf (x) = g −1 (x) = x.
To show that H = Sym(N), let k ∈ Sym(N) and M ⊆ N be finite; we have to show
that H contains an operation h such that h(x) = k(x) for every x ∈ M . Choose
h ∈ H such that M 0 := {x ∈ M | h(x) = k(x)} is largest possible. If M 0 = M then
we are done; otherwise there is b ∈ M such that h(b) 6= k(b). Let p and q be distinct
elements from
N \ {k(x), h(x) | x ∈ M }.
Let g ∈ Sym(N) be any permutation that maps h(b) to a, k(b) to c, and p to f −1 (a).
0

Moreover, if c 6= f (c) then g 0 maps q to f −1 (c). Then (g 0 )−1 (g −1 f −1 gf )g 0 h ∈ H.

Moreover, for each x ∈ M 0 we have that h(x) ∈ / {p, q, h(b), k(b)}. This implies for
every x ∈ M 0 that g 0 h(x) ∈
/ {a, f −1 (a), c, f −1 (c)} and
(g 0 )−1 (g −1 f −1 gf )g 0 h(x) = h(x) = k(x)
because g −1 f −1 gf is the identity on g 0 h(x) ∈
/ {a, f (a), c, f (c)}. We also have that
(g 0 )−1 (g −1 f −1 gf )g 0 h(b) = (g 0 )−1 (g −1 f −1 gf )(a) = (g 0 )−1 (c) = k(b)
which contradicts the choice of h so that M 0 is largest possible. 
82 4. TOPOLOGICAL GROUPS

Exercises.
(110) Let G ≤ Sym(N) be oligomorphic, and let G◦ be the intersection
of the closed subgroups of ∆ of finite index in G. Show that G◦ is
oligomorphic and that (G◦ )◦ = G◦ .
(111) Can the group Aut(A) from Example 67 be written
as a semidirect product N oθ ZN 2 ? Here, N is the normal
subgroup of Aut(A) consisting of all automorphisms that fix
all the equivalence classes of all the equivalence relations
E0 , E1 , . . . (see Proposition 4.6.1 and Proposition 1.5.4).
 CHAPTER 5

Birkhoff ’s Theorem and Permutation Groups

Birkhoff’s theorem from universal algebra, specialised to permutation groups, de-

scribes how a given group can act on a given set (Section 5.1). In this chapter we also
present a topological generalisation where we will be interested in continuous actions
of topological groups (Section 5.2). As an application, this gives rise to a topolog-
ical characterisation of bi-interpretability for ω-categorical structures (Section 5.3;
bi-interpretability has already been defined in Section 3.6.2). More specifically, we
obtain the following corollary which has been credited to Coquand by Ahlbrandt and
Ziegler [3].
Theorem 5.0.1. Let A and B be countable ω-categorical structures, each with at
least two elements. Then Aut(A) and Aut(B) are isomorphic as topological groups if
and only if A and B are bi-interpretable.

5.1. Birkhoff ’s Theorem

In order to apply Birkhoff’s theorem to permutation groups, we represent permu-
tation groups G as G-sets as in Example 4 and 7. If K is a class of τ -structures, then
we write
• P(K) for the class of all products of structures in K;
• S(K) for the class of all substructures of structrues in K;
• H(K) for the class of all homomorphic images of structures in K.
If A is a G-set, then we also write Gr(A) for G; recall that by definition G equals
the permutation group on A consisting of all unary term functions over A.
Theorem 5.1.1 (Birkhoff’s theorem for permutation groups). Let G be a subgroup
of Sym(B), for |B| ≥ 2, and let B be a G-set with signature τ .
• For every homomorphism ξ : G → Sym(A) there exists a G-set A on A such
that A ∈ HSP({B}) and tA = ξ(tB ) for every τ -term t.
• If A ∈ HSP({B}) then the function ξ that maps tB to tA for every τ -term t
is well defined and a homomorphism from G onto Gr(A).
Proof. Let ξ : G → Sym(A) be a homomorphism. Let I be a well-ordered set
such that B A = {ci | i ∈ I}. For a ∈ A, define ca := (ci (a))i∈I . Let S be the smallest
A
substructure of B B that contains {ca | a ∈ A}. So the elements of S are precisely
those that can be written as tS (ca ) for some τ -term t(x) and some a ∈ A. Define
µ : S → A by
µ(tS (ca )) := ξ(tB )(a).
Claim 1. µ is well-defined. Suppose that tS (ca ) = rS (ca0 ). We first show that
t = rB . Let b ∈ B. Note that there is some i ∈ I such that ci (a) = b and ci (a0 ) = b.
B

Hence,
tB (b) = tB (ci (a)) = tS (ca )i
= rS (ca0 )i = rB (ci (a0 ))) = rB (b).

83
84 5. BIRKHOFF’S THEOREM AND PERMUTATION GROUPS

Hence, tB = rB and therefore tS = rS . Since rS is injective, tS (ca ) = rS (ca0 ) implies

that ca = ca0 . Since |B| ≥ 2 this implies that a = a0 . Therefore
ξ(tB )(a) = ξ(rB )(a) = ξ(rB )(a0 )
and hence µ is well-defined.
Claim 2. µ is surjective. Let a ∈ A and choose the τ -term t := x (just a variable)
so that rB = 1G ; then
µ(ca ) = µ(tS (ca )) = ξ(tB )(a) = a
since ξ(1G ) = 1Sym(A) .
Let A be the τ -algebra where g ∈ τ denotes ξ(g B ).
Claim 3. µ is a homomorphism from S to A. Let f ∈ τ and let s ∈ S, and let
s = tS (ca ) for some τ -term t and some a ∈ A. Then
µ(f S (s)) = µ(f S (tS (ca ))) = µ(f (t)S (ca ))
= f (t)A (a)
= f A (tA (a))
= f A (tS (ca )) = f A (µ(s))
A
Hence, A is the homomorphic image of the subalgebra S of B B , so A ∈ HSP(B).
For the second statement, we first show that ξ is well-defined. Suppose that r
and t are τ -terms such that rB = tB , and let µ : S → A be a homomorphism from
a subalgebra S of a power of B to A. Then rS = tB . Hence, for all s ∈ S we have
µ(rS (s)) = µ(tS (s)) and rA (µ(s)) = tA (µ(s)) since µ is a homomorphism. Since µ is
surjective it follows that rA = tA .
To show that ξ is a homomorphism, we have to prove that
ξ(1G ) = 1Sym(A) (9)
B −1 B −1
ξ((g ) ) = ξ(g ) (10)
B B B B
and ξ(g h ) = ξ(g )ξ(h ). (11)
To show (9), consider the τ -term x that just consists of a variable. Then
ξ(1G ) = ξ(idB ) = ξ(xB ) = xA = idA = 1Sym(A) .
To show (11), let g and h be τ -terms. Then
ξ(g B hB ) = ξ((gh)B ) = (gh)A = g A hA = ξ(g B )ξ(hB ).
Finally, since Gr(B) is a permutation group, there exists a τ -term t such that
tB = (g B )−1 . Then
idA = ξ(idB ) = ξ(g B tB ) = ξ(g B )ξ(tB ) = g A tA
and hence tA = (g A )−1 . We then have
ξ(g B )−1 = ξ(tB ) = tA = (g A )−1 = ξ(g B )−1 . 

Corollary 5.1.2. Let G ≤ Sym(B) and H ≤ Sym(C), with |B|, |C| ≥ 2. Then
the following are equivalent:
(1) G and H are isomorphic as abstract groups;
(2) There exists a G-set B and an H-set C such that
HSP(B) = HSP(C).
 5.2. TOPOLOGICAL BIRKHOFF 85

5.2. Topological Birkhoff

In the following, uniform continuity will refer to the left-invariant ultrametric of
Sym(A).

Theorem 5.2.1 (Topological Birkhoff for permutation groups). Let G be a sub-

group of Sym(B), for |B| ≥ 2, and let B be a G-set with signature τ .
• For every continuous homomorphism ξ : G → Sym(A) such that ξ(G) has
finitely many orbits, there is an A ∈ HSPfin ({B}) such that tA = ξ(tB ) for
every τ -term t.
• If A ∈ HSPfin ({B}) then the function ξ that maps tB to tA for every τ -term t
is well-defined and a (uniformly) continuous surjective homomorphism from
G to Gr(A).

Proof. The proof is an adaptation of the proof of Theorem 5.1.1. Let F =

{a1 , . . . , ak } ⊆ A be a finite set that contains one element from each orbit of ξ(G) on
A. By Proposition 4.2.10, ξ is uniformly continuous, and hence there exists a finite
F 0 ⊆ B such that

for all f, g ∈ G if f |F 0 = g|F 0 then ξ(f )|F = ξ(g)|F . (12)

Choose F 0 to be smallest possible; note that this implies that F 0 contains at most
F
one element from each orbit of G. Let C be (F 0 ) and let m := |C| = |F 0 |k . Let
c1 , . . . , cm be the elements of C, and for j ≤ k define cj = (c1 (aj ), . . . , cm (aj )). Let S
be the substructure of B m generated by c1 , . . . , ck ; so the elements of S are precisely
those of the form tS (cj ) for a τ -term t and j ≤ k. Define a function µ : S → A by
setting
µ(tS (cj )) := ξ(tB )(aj ).
Claim 1. µ is well-defined. Suppose that tS (cj ) = rS (cl ) for j, l ≤ k. We first
show that tB |F 0 = rB |F 0 . Let b ∈ F 0 . Note that there is some i ≤ m such that
ci (aj ) = b and ci (al ) = b. Hence,

tB (b) = tB (ci (aj )) = tB ((cj )i )

= rB ((cl )i ) (since tS (cj ) = rS (cl ))
= rB (ci (al )) = sB (b).

Hence, tB |F 0 = rB |F 0 . Therefore, using (12), we obtain

ξ(tB )|F = ξ(rB )|F .

It also follows from tS (cj ) = rS (cl ) that for all i ≤ m the elements (cj )i and (cl )i lie
in the same orbit of G (here we use the assumption that |B| ≥ 2). By our assumption
on F 0 this means that l = j. Hence,

ξ(tB )(aj ) = ξ(rB )(aj ) = ξ(rB )(al ),

and µ is indeed well-defined.

Claim 2. µ is surjective. This follows immediately from the assumption that F

contains an element from each orbit of ξ(G) on A.

Let A be the τ -structure where g A := ξ(g B ) for every g ∈ τ .

86 5. BIRKHOFF’S THEOREM AND PERMUTATION GROUPS

Claim 3. µ is a homomorphism. Let f ∈ τ and let s ∈ S. Since S is generated

by c1 , . . . , ck there exists a τ -term t and a j ≤ k such that s = tS (cj ). Then
µ(f S (s)) = µ(f S (tS (cj ))) = µ((f (t))S (cj ))
= (f (t))A (aj )
= f A (tA (aj ))
= f A (µ(tS (cj ))) = f A (µ(s)).
It follows that A is the homomorphic image of the subalgebra S of B m , and so
A ∈ HSPfin (B).
For the second statement we have already seen in Theorem 5.1.1 that if A ∈
HSPfin (B) ⊆ HSP(B) then the natural homomorphism ξ : G → Gr(A) exists. It thus
remains to show that ξ is uniformly continuous.
Let F be a finite subset of A. We have to find a finite subset F 0 of B such that
for all k ∈ N and f, g ∈ G if f |F 0 = g|F 0 then ξ(f )|F = ξ(g)|F . By assumption, there
exists a surjective homomorphism µ from a subalgebra S of B n , for some n ∈ N, to
A. For each a ∈ F pick an s ∈ S such that µ(s) = a; let F 0 ⊆ B be the (finite) set
of all entries of all the tuples in s. Now let f, g ∈ Gr(B) be such that f |F 0 = g|F 0 .
Choose τ -terms r, t such that rB = f and tB = g. Clearly, rS |S∩Gn = tS |Gn . Since
F ⊆ µ(S∩Gn ) it follows that rA |F = tA |F , which proves the statement since rA = ξ(f )
and tA = ξ(g). 

Corollary 5.2.2 (Topological Birkhoff for permutation groups). Let G be a

subgroup of Sym(B), for |B| ≥ 2, and let B be a G-set with signature τ . Then the
following are equivalent.
• There exists a continuous homomorphism ξ : G → Sym(A) such that ξ(G)
has finitely many orbits.
• There exists an A ∈ HSPfin ({B}) such that Gr(A) = ξ(G).

5.3. Continuous Homomorphisms and Interpretability

In this section we prove Theorem 5.0.1: we give a characterisation of topolog-
ical isomorphism of automorphism groups of ω-categorical structures in terms of
bi-interpretability. We first establish the link between pseudo-varieties and full in-
terpretations.
Definition 5.3.1. Let I be a d-dimensional interpretation of A in B. Then I is
called full if a relation R ⊆ Ak is first-order definable in A if and only if the relation
h−1 (R), defined as
{(b11 , . . . , bd1 , . . . , b1k , . . . , bdk ) ∈ B d1 d2 | (b11 , . . . , bd1 ), . . . , (b1k , . . . , bdk ) ∈ S and
(h(b11 , . . . , bd1 ), . . . , h(b1k , . . . , bdk )) ∈ R}
is first-order definable in B.
Observe that any structure with an interpretation in a structure B is a reduct of
a structure with a full interpretation in B.
Theorem 5.3.2. Let B be a structure and let B 0 be an Aut(B)-set. If A has a
full interpretation in B then there is an A0 ∈ HSPfin (B 0 ) such that Gr(A0 ) = Aut(A).
If B is finite or countably infinite ω-categorical, then the converse implication holds
as well.
 5.4. TOPOLOGICAL ISOMORPHISM AND BI-INTERPRETABILITY 87

Proof. Suppose first that A has a d-dimensional full interpretation I in B. Since

I −1 (A) is definable in B, it is preserved by all operations in B 0 , and therefore induces
d
a subalgebra C 0 of B 0 . Let K be the kernel of I. Since I −1 (=A ) is definable in B,
all operations of B 0 preserve K = I −1 (=A ), so K is a congruence of C 0 . Thus, I is a
surjective homomorphism from C 0 to A0 := C 0 /K. We verify that Gr(A0 ) = Aut(A).
0
We have to show that for every f ∈ τ , the relation R is preserved by f A . Since R is
definable in A by the assumption that the interpretation I is full, we have that I −1 (R)
0
is definable in B. This in turn implies that f B preserves I −1 (R), which implies the
claim.
For the converse implication, suppose that B is finite or countably infinite and
ω-categorical. Then there is an algebra A0 ∈ HSPfin (B 0 ) such that Gr(A0 ) = Aut(A).
d
So there exists a finite number d ≥ 1, a subalgebra C 0 of B 0 , and a surjective
0 0
homomorphism I from C to A . We claim that I is a d-dimensional interpretation
of A in B. All operations of B 0 preserve C (viewed as a d-ary relation over B) since
d
C 0 is a substructure of B 0 . By Theorem 3.1.1, this implies that C has a definition
in B, which becomes the domain formula of the interpretation. Since I is an algebra
homomorphism, the kernel K of I is a congruence of C 0 . It follows that K, viewed
as a 2d-ary relation over B, is preserved by all operations from B 0 . Theorem 3.1.1
implies that K has a definition in B. This definition becomes the interpreting formula
of the equality relation on A.
To see that I is a full interpretation, let R ⊆ Ak be a relation of A, let τ be
0
the signature of B 0 , and let f ∈ τ be arbitrary. By assumption, f A preserves R.
0
Therefore, f B preserves I −1 (R). Hence, all automorphisms of B preserve I −1 (R), and
because B is ω-categorical, the relation I −1 (R) has a definition in B (Theorem 3.1.1),
which becomes the interpreting formula for R(x1 , . . . , xk ). We have verified that I is
an interpretation of A in B. To see that I is a full interpretation, let R ⊆ Ak be a
relation such that I −1 (R) is definable in B. Then I −1 (R) is preserved by Gr(B 0 ) and
R is preserved by Gr(A0 ). By assumption Gr(A0 ) = Aut(A), and hence R is preserved
by all automorphisms of A and definable in A by Theorem 3.1.1. 
The following corollary is a direct consequence of Theorem 5.3.2 and Theo-
rem 5.2.1.
Corollary 5.3.3. Let B be a countable ω-categorical, with |B| ≥ 2, and A an
arbitrary structure. Then A has a full interpretation in B if and only if A is at most
countable ω-categorical and there exists a continuous homomorphism from Aut(B) to
Aut(A) whose image is dense in Aut(A).
Proof. Let B 0 be a Aut(B)-set. By Theorem 5.3.2, there is a full interpretation
of A in B if and only if there is an A0 ∈ HSPfin (B 0 ) such that Gr(A0 ) = Aut(A).
Corollary 5.2.2 shows that this is the case if and only if there exists a continuous
homomorphism from Aut(B) to Aut(A) whose image is dense in Aut(A). 
Remark 5.3.4. In Corollary 5.3.3 we cannot in general require surjectivity of the
homomorphism (instead of requiring that the image being dense) as we have seen in
Example 70.

5.4. Topological Isomorphism and Bi-interpretability

We continue to state some consequences of the topological Birkhoff theorem.
Theorem 5.4.1. Let C, D be ω-categorical structures with |C|, |D| ≥ 2. Them
(1) ⇔ (2) ⇔ (3):
88 5. BIRKHOFF’S THEOREM AND PERMUTATION GROUPS

(1) Aut(C) and Aut(D) are isomorphic as topological groups.

(2) There is an Aut(C)-set C 0 and an Aut(D)-set D0 such that
HSPfin (C 0 ) = HSPfin (D0 ).
(3) C and D are bi-interpretable.
Proof. The equivalence between (1) and (2) follows from Theorem 5.2.1. For
the implication from (2) to (3), we assume that there is a d1 ≥ 1, a substructure S 1
of C d1 , and a surjective homomorphism h1 from S 1 to D. Moreover, we assume that
there is a d2 ≥ 1, a subalgebra S 2 of Dd2 , and a surjective homomorphisms h2 from
S 2 to C. The proof of Theorem 5.3.2 shows that h1 is an interpretation of D in C
and h2 is an interpretation of C in D. Because the statement is symmetric it suffices
to show that the (graph of the) function h1 ◦ h2 : (S2 )d1 → D defined by
(y1,1 , . . . , y1,d2 , . . . , yd1 ,1 , . . . , yd1 ,d2 ) 7→ h1 (h2 (y1,1 , . . . , y1,d2 ), . . . , h2 (yd1 ,1 , . . . , yd1 ,d2 ))
is definable in D. Theorem 3.1.1 asserts that this is equivalent to showing that h1 ◦ h2
is preserved by Aut(D) = Gr(D0 ). So let b be an element of (S2 )d1 . Then indeed
0 0 0
f D ((h1 ◦ h2 )(b)) = h1 f C (h2 (b1 )), . . . , f C (h2 (bd1 ))


0 0
= h1 h2 (f D (b1 )), . . . , h2 (f D (bd1 ))


0
= (h1 ◦ h2 )(f D (b)) .
For the implication from (3) to (1), suppose that C and D are bi-interpretable via
an interpretation I1 : C d1 → D and I2 : Dd2 → C. Let A0 be an Aut(A)-set. As we
have seen in the proof of Theorem 5.3.2, the domain S1 of I1 induces a structure
S 1 in (A0 )d1 and I1 is a surjective homomorphism from S 1 onto an Aut(D)-set D0
with the same signature τ as C 0 . Similarly, the domain S2 of I2 induces in (D0 )d2
a structure S 2 and I2 is a homomorphism from S 2 onto an Aut(C)-set A00 with the
same signature as D0 .
We claim that HSPfin (C 0 ) = HSPfin (D0 ). The inclusion ‘⊇’ is clear since D0 ∈
HSPfin (C 0 ). For the reverse inclusion it suffices to show that C 0 = A00 since A00 ∈
0 00
HSPfin (D0 ). Let f ∈ τ ; we show that f C = f A . Let a ∈ C. Since I2 ◦ I1 is surjective
onto C, there are c = (c1,1 , . . . , cd1 ,d2 ) ∈ C d1 d2 such that a = I2 (I1 (c)). Then
00 00
f A (a) = f A (I2 ◦ I1 (c))
0 0
= I2 f D (I1 (c1,1 , . . . , cd1 ,1 )), . . . , f D (I1 (c1,d2 , . . . , cd1 ,d2 ))


0
= I2 ◦ I1 (f C (c))
0 0
= f C (I2 ◦ I1 (c)) = f C (a)
where the second and third equations hold since I2 and I1 are algebra homomorphisms,
0
and the fourth equation holds because f C preserves h2 ◦h1 , because I2 ◦I1 is homotopic
to the identity. 
Example 75. The structures
C := (N2 ; {(u1 , u2 ), (v1 , v2 ) | u1 = v1 }) and D := (N; =)
are mutually interpretable, but not bi-interpretable. To see this, observe that Aut(C)
has a proper non-trivial closed normal subgroup N . To specify N , let Pi := {(u1 , u2 ) |
u1 = i} for i ∈ N. Then Aut(C, P1 , P2 , . . . ) is a non-trivial (Aut(C)/N is isomorphic
to Aut(D)) proper closed subgroup, and it can be verified that N is normal (see
Proposition 4.6.1), whereas Aut(D), the symmetric permutation group of a countably
infinite set, has no proper non-trivial closed normal subgroups (it has exactly three
proper non-trivial normal subgroups [141], none of which is closed). 4
 5.4. TOPOLOGICAL ISOMORPHISM AND BI-INTERPRETABILITY 89

We also mention that Theorem 5.4.1 in combination with Proposition 3.6.5 shows
that for every ω-categorical structure B, having essentially infinite signature only
depends on the automorphism group of B viewed as a topological group.


Example 76. Let B be the graph with domain Nk and the edge relation
E B = (S, T ) | |S ∩ T | = 1 .


We claim that B and (N; =) are bi-interpretable. Let I be the k-dimensional inter-
pretation of B in (N; =}) whose domain is Nk6= and which maps (x1 , . . . , xk ) ∈ Nk6= to
{x1 , . . . , xk }. Clearly, the relations
I −1 (B) = Nk6=
I −1 (=B ) = (x, y) | x, y ∈ Nk6= and {x1 , . . . , xn } = {y1 , . . . , yn }


I −1 (E B ) = (x, y) | x, y ∈ Nk6= and {x1 , . . . , xn } ∩ {y1 , . . . , yn } = 1



are definable in (N; =).

Let J be the 2-dimensional interpretation of (N; =) in B whose domain is E B and
which maps (S, T ) ∈ E B to S∩T . Let n ∈ N be at least 2. Let m := max (n, k 2 −k+2).
Let Rn ⊆ B n be the relation with the definition
^
Rn (U1 , . . . , Un ) :⇔ ∃Un+1 , . . . , Um E(Ui , Uj ).
i,j∈{1,...,m},i6=j

Claim. If Rm (U1 , . . . , Um ) then |U1 ∩ · · · ∩ Un | = 1. To see this, choose u1 ∈ N

such that S1 := {U ⊆ {U1 , . . . , Um } | u1 ∈ U } is maximal. If |S1 | = m, then in
particular U1 ∩ · · · ∩ Un = {u1 } and then there is nothing to be shown. Otherwise,
if Uj ∈
/ S1 , then |S1 | ≤ k because Uj has to intersect every V ∈ S in an element
uV which is distinct from u1 . If V, V 0 ∈ S are distinct then uV 6= uV 0 because
V ∩ V 0 = {u1 }. Hence, |Uj | = k implies that |S1 | ≤ k. We now choose u2 ∈ N such
that S2 := {U ⊆ {U1 , . . . , Un } \ S1 | u2 ∈ U } is maximal. We claim that |S2 | ≤ k − 1.
To see this, pick any V ∈ S1 . Then every U ∈ S2 must intersect V in a distinct point
which is different from u1 . We may now repeat the argument k − 1 times, and obtain
that m ≤ k + (k − 1) + · · · + (k − 1) = k(k − 1) + 1, which is a contradiction to our
choice of m ≥ k 2 − k + 2.
Note that if (S, T ), (U, V ) ∈ E B , then S ∩ T = U ∩ V if and only if
B B

∃P, Q R4 (U, V, P, Q) ∧ R4 (S, T, P, Q) .
It follows that the relations
J −1 (N) = E B
J −1 (=N ) = (S, T ), (U, V ) | (S, T ), (U, V ) ∈ E B , S ∩ T = U ∩ V


are definable in B.
Finally, note that for all (x1 , . . . , xk ), (y1 , . . . , yk ) ∈ Nk6= and z ∈ N we have


J I(x1 , . . . , xk ), I(y1 , . . . , yk ) = z
if and only if z = {x1 , . . . , xk } ∩ {y1 , . . . , yk }, which is definable in (N; =).
Moreover, for all (S1 , T1 ), . . . , (Sk , Tk ) ∈ E B and V ∈ B we have


I J(S1 , T1 ), . . . , J(Sk , Tk ) = V
B
if and only if there are U1 , . . . , Uk such that (Si , Ti , Ui , V ) ∈ R4 for all i ∈ [k] and
/ E B for all {i, j} ∈ [k]


(Ui , Uj ) ∈ 2 , and hence is definable in B. This concludes the
proof of the claim above.
Then Theorem 5.4.1 implies that there exists an injective (continuous)

homomor-
N
phism from Aut(N; =) to Aut(B). Thus, the action of Sω on k is faithful, as claimed
90 5. BIRKHOFF’S THEOREM AND PERMUTATION GROUPS

earlier in Example 11. The same argument works for any permutation group instead
of Sω . 4

Exercises.
(112) Show that the line graph of an undirected graph G
has an interpretation in G.
(113) Show that every finite structure has an interpretation
in every structure with at least two elements.
(114) Show that the automorphism group of infinitely many disjoint
copies of the 2-element clique K2 is not topologically
isomorphic to infinitely many disjoint copies of the
three-element clique K3 .
 CHAPTER 6

Reconstruction of Topology and Automatic

Continuity

The question what information about a structure can be recovered from its au-
tomorphism group when considered as an abstract group, has long attracted the
attention of research in model theory and the theory of infinite permutation groups;
a very incomplete list of references is [6, 47, 50, 52, 54, 60, 69, 74, 88, 104, 130, 138,
139, 158, 162].

6.1. Reconstruction Notions

We study the question whether we can reconstruct the topology of closed sub-
groups of Sym(N) from the abstract group.

Definition 6.1.1. Let G be a closed subgroup of Sym(N). We say that

• G is reconstructible (or that G has reconstruction) iff for every other closed
subgroup H of Sym(N), if there exists an isomorphism between H and G,
then there also exists a group isomorphism between H and G which is a
homeomorphism;
• G has automatic homeomorphicity (AH) iff every group isomorphism be-
tween G and a closed subgroup of Sym(N) is a homeomorphism;
• G has automatic continuity (AC) iff every homomorphism from G to Sym(N)
is continuous.

Obviously, automatic homeomorphicity implies reconstruction. Less obviously,

we will later see in Proposition 6.3.18 that automatic continuity implies automatic
homeomorphicity.

Example 77. We have seen an example of a closed oligomorphic permutation

group without automatic continuity in Example 67 (clearly, there are closed subgroups
of Sym(N) that are isomorphic to Z2 with the discrete topology). The example still
has reconstruction; this can be shown using the results of Rubin (see Remark 5.4.3
in [110]). 4

A more involved example of a closed oligomorphic permutation group without

reconstruction has been found by Evans and Hewitt [54]. In the next section we
present a fundamental concept for proving automatic continuity (and hence automatic
homeomorphicity and reconstruction), the so-called small index property. A very
powerful method to show the small index property is the concept of ample generics,
introduced in Section 6.4. To verify that an automorphism group has ample generics,
we need to study the important combinatorial property of classes of finite structures,
called EPPA (or the Hrushovski property; Section 6.5). The structure of this chapter
is therefore organised along the following chain of implications.
91
92 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

Reconstruction
⇐ Automatic Homeomorphicity
⇐ Automatic Continuity (Proposition 6.3.18)
⇔ The Small Index Property (Proposition 6.2.1)
⇐ Ample Generics (Theorem 6.4.4)
⇐ EPPA / the Hrushovski property (Section 6.5)
For many automorphism groups (e.g., for the automorphism group of the Rado graph),
this chain of implications is to the best of my knowledge the only known way to prove
reconstruction.

6.2. The Small Index Property

Recall that a subgroup of Sym(N) is open if it contains the stabiliser Ga for some
a ∈ Nn , n ∈ N (Lemma 4.4.1). Clearly, these groups have countable index, so all
open subgroups of Sym(N) have countable index. A topological group G has the
small index property (SIP) if the converse holds, i.e., if every subgroup of G of at
most countable index is open.
Proposition 6.2.1 (Folklore). Let G be a closed subgroup of Sym(N). Then G
has automatic continuity if and only if it has the small index property.
Proof. Let G be a topological group with automatic continuity, and let U be
a subgroup of G of at most countable index. We have to show that U is open.
Let ξ : G → G/U be the action of G on the left cosets of U in G by left trans-
lation (Example 12), where G/U is equipped with the discrete topology. By au-
tomatic continuity, ξ is continuous. In particular, the pre-image of the open set
{α ∈ Sym(G/U ) | α(U ) = U } is open. But this pre-image is precisely U , which
proves the SIP.
Now suppose that G has the SIP, and let ξ : G → Sω be a homomorphism. We
show that for every basic open set U := S(a, b) ∩ ξ(G) for a, b ∈ Nn , n ∈ N, the set
ξ −1 (U ) is open, too. Writing Sa for the stabiliser subgroup {α ∈ ξ(G) | α(a) = a}, we
have U = βSa for β ∈ U . Since ξ is a homomorphism, ξ −1 (U ) = ξ −1 (β)ξ −1 (S a ). The
subgroup S a of ξ(G) has countable index, and therefore ξ −1 (S a ) is a subgroup of G
of countable index, too. By assumption, ξ −1 (S a ) is open, and since multiplication by
ξ −1 (β) is continuous, ξ −1 (β)ξ −1 (S a ) = ξ −1 (U ) is open, which establishes continuity
of ξ. 
The small index property has been verified for the following groups:
(1) Sym(N) [47, 137, 142];
(2) the automorphism groups of countable vector spaces over finite fields [52];
(3) Aut(Q; <) and the automorphism group of the atomless Boolean algebra [158];
(4) the automorphism group of the ω-categorical dense semi-linear order giving
rise to a meet-semilattice [51];
(5) the automorphism group of the countable random graph [74] (see Exam-
ple 29);
(6) all automorphism groups of ω-categorical ω-stable structures [74];
(7) the automorphism groups of the Henson graphs [69] (see Example 31).
On the other hand, the small index property is not known for the automorphism
groups of the countable universal homogeneous tournament, the countable universal
poset, or the homogeneous universal permutation (see [110]).
 6.3. CONSISTENCY OF AUTOMATIC CONTINUITY 93

6.3. Consistency of Automatic Continuity

In this section we prove that it is consistent with Zermelo-Fraenkel set theory
and the axiom of dependent choice (see Appendix A.2) that every closed subgroup
of Sym(N) has then SIP and automatic continuity (Theorem 6.3.17) and hence au-
tomatic homeomorphicity (Proposition 6.3.18). We first introduce basic concepts
from descriptive set theory; these concepts will be relevant again in Secrtion 6.4 and
Section 7.2.1.
6.3.1. Nowhere dense and meager. Let X be a topological space.
Definition 6.3.1. A set A ⊆ X is called somewhere dense if its closure contains
a nonempty open set, and nowhere dense otherwise.
The following is immediate from the definitions.
Proposition 6.3.2. Let A ⊆ X. Then the following are equivalent.
(1) A is nowhere dense;
(2) For every nonempty open set U ⊆ X there is a nonempty open set V ⊆ U
such that V and A are disjoint;
(3) X \ A is dense.
Proof. (1) ⇒ (3): Let U be an open subset of X. Since A is nowhere dense, Ā
does not contain U . So X \ Ā contains an element of U . This shows that X \ Ā is
dense.
(3) ⇒ (2): Let U ⊆ X be nonempty open. Since X \ Ā is dense there exists
x ∈ U \ Ā. So there exists a non-empty open set V ⊆ U such that V and A are
disjoint.
(2) ⇒ (1). Let U be non-empty open. Then by assumption there is a nonempty
open set V ⊆ U which is disjoint from A. Hence, A cannot be dense in U , showing
that A is nowhere dense. 
Clearly, A is nowhere dense if and only if its closure A is nowhere dense, and
subsets of nowhere dense sets are nowhere dense.
Definition 6.3.3. Let X be a topological space. A set A ⊆ X is meager (in X)1
if it is a countable union of nowhere dense sets. Otherwise, A is called non-meager2
The complement of a meager set is called comeager.
Clearly, every subset of a meager set is meager, and every countable union of
meager sets is meager.
Remark 6.3.4. The intuition will be that meager sets are ‘negligible’, and cor-
respondingly that comeager sets capture the notion of ‘almost all’. This methaphor
works particularly well if the topology is Baire, as we will see below (Section 6.3.2,
Exercise 117).
Example 78. The set of meager subsets of R with respect to the standard topol-
ogy is strictly larger than the set of nowhere dense subsets of R. Consider for example
the set Q of rational numbers, which is a meager subset of R, because it is a countable
union of one-element sets (which are nowhere dense). However, the rational numbers
are dense in R and therefore in particular somewhere dense. 4
Lemma 6.3.5. A set is comeager if and only if it contains a countable intersection
of dense open sets.
1Sometimes, meager sets are also called of first category.
2Non-meager sets are also called of second category.
94 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

somewhere dense meager

dense non-meager nowhere dense

G𝛿 comeager non-empty open

countable intersection
of dense open

dense open

Figure 6.1. Properties of subsets of X from Section 6.3.1 and Sec-

tion 6.3.2, ordered by inclusion; we assume that X is non-empty.
Blue edges hold in Baire spaces.
T
Proof. If A ⊆ X is comeager, then X \A is meager and hence A = i∈N (X \Ai )
T
for nowhere dense sets Ai . Hence, i∈N (XT\ Ai ) ⊆ A, and X \ Ai is open and dense.
Conversely, suppose that
T A contains
S i Ui where Ui is open and dense. Then
X \ A is contained in X \ i Ui = i (X \ Ui ). Note that X \ Ui is nowhere dense, so
X \ A is meager since it is a subset of a meager set. Hence, A is comeager. 
Corollary 6.3.6. The intersection of countably many comeager sets is comeager.
6.3.2. Baire Spaces. Recall from Theorem 4.1.8. that a topological space X is
called Baire if every countable intersection of open dense sets is dense.
Example 79. The standard topology on R is Baire. This follows from Theo-
rem 4.1.8 since R is Polish (Example 51). Similarly, the Baire space and the Cantor
space (Example 53) are Baire. 4
Example 80. Clearly, the empty space is Baire. It is the only space which is
Baire and meager (see Proposition 6.3.7 below). 4

Proposition 6.3.7. Let X be a topological space. Then the following statements

are equivalent:
(1) X is Baire.
(2) Every comeager set in X is dense.
(3) Every nonempty open set in X is non-meager.
Proof. (1) implies (2). Let A ⊆ X be comeager. By Lemma 6.3.5, A contains
a countable intersection of dense open sets, which is dense since X is Baire.
(2) implies (3). If U is meager, then X \ U is comeager and by (2) it is dense.
Hence, U is either empty or not
T open.
(3) implies (1). Let A = i∈N Ai . If A is not dense, X \ A contains a nonempty
open set U . By (3), U is non-meager. If all of the Ai would be dense open, then X \ A
and hence U would be meager. 
The following example was brought to my attention by Tsankov in 2012.
Example 81. Consider the logic action of Sym(N) on XC as defined in Exam-
ple 68. Then it follows from results of Ivanov [80] that the following are equivalent:
 6.3. CONSISTENCY OF AUTOMATIC CONTINUITY 95

• C has the WAP.

• XC has a dense Gδ orbit.
Suppose that C is the age of an ω-categorical structure B. Recall that Age(B) has the
WAP (Exercise 84). Then the dense Gδ orbit equals the orbit of the model companion
of B. TODO: add proofs for all these claims. 4

Exercises.
(115) Let S be a countable Polish space.
Is S meager or non-meager?
(116) Show that R \ Q is non-meager.
(117) Show that in non-empty Baire spaces
comeager sets are non-meager.3
(118) Prove Lemma 6.3.6.
(119) Let S1 , S2 , . . . be comeager subsets of Sω .
Show that ∩i∈N Si is non-empty.
(120) Show that a dense Gδ set (Definition 4.1.5) is comeager.

6.3.3. The Baire Property. Let X be a topological space (typically a Baire

space).
Definition 6.3.8. A set A ⊆ X is said to have the Baire property4 (with respect
to X) if for some open set U ⊆ X the symmetric difference U ∆A between U and A
is meager.
Example 82. Using the Axiom of Choice (AC), we construct a subset of R which
does not have the Baire property. For x, y ∈ R, define x ∼ y if x − y ∈ Q; clearly, ∼ is
an equivalence relation. Let A ⊆ R be such that it contains exactly one member from
each ∼-class (here we use AC; such a set is called a Vitali set). Clearly, {A+q | q ∈ Q}
forms a countable partition of R. The map x 7→ x + q is a homeomorphism of R, and
hence the sets A + q for q ∈ Q cannot be meager (otherwise, R would be meager,
contrary to Theorem 4.1.8). Suppose for contradiction that A has the Baire property,
and let U be an open set and M a meager set such that U ∆A = M . Then U 6= ∅
since A is not meager, so there are a, b ∈ R such that a < b and (a, b) ⊆ U . Let
r = b − a and let x ∈ R be such that |x| < r. Then (a + x, b + x) ∩ (a, b) is an open
interval. Since M + x ∪ M is meager, so
S := ((a + x, b + x) ∩ (a, b)) \ ((M + x) ∪ M ) 6= ∅.
Observe that S is contained in (A + x) ∩ A. Hence, if we choose x > 0 to be rational
we obtain a contradiction to the fact that the sets of the form A + q for q ∈ Q are
pairwise distinct. 4
Example 83. Non-principal ultrafilters U on N, viewed as a subset of the Cantor
space 2N , do not have the Baire property. First note that X 7→ N \ X is a homeomor-
phism of 2N that sends U to its complement 2N \U. Hence, if U would be meager, then
2N \ U would be meager, and hence 2N would be meager as a union of meager sets, in
contradiction to the Cantor space being meager (this follows from Proposition 6.3.7
3Note that this is certainly a property that we wish for the metaphor from Remark 6.3.4.
4As pointed out in [118], instead of the expression ‘A has the Baire property’ it would be more
suggestive to call A Baire measurable, to indicate an analogy with measure theory. But we are stuck
with the standard terminology here.
96 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

since the Cantor space is Baire; Example 79). For the same reason, U cannot be
comeager. Note that the symmetric difference of a set in the non-principal ultrafilter
U with a finite set must again be in U. Then Theorem 8.46 in [88] implies that U
cannot have the Baire property. 4
Let U, A ⊆ X. We say that A is meager in U if A ∩ U is meager in the subspace
U of X, and otherwise that A is non-meager in U . If U = A we then also say that
A is non-meager in its relative topology. Note that if U is open, then A is meager in
U if and only if A ∩ U is meager in X. The assumption that U is open is necessary,
as demonstrated by R ⊆ C, because R is non-meager in its relative topology, but R
is nowhere dense in C. Similarly, we say that A is comeager in U if U \ A is meager
in X.
Proposition 6.3.9 (‘Localisation’). Let X be a topological space and suppose that
A ⊆ X has the Baire property. Then A is meager or there is a non-empty open set
U ⊆ X such that A is comeager in U . If X is a Baire space, then the two alternatives
are mutually exclusive.
Proof. Let U ∆A = M where U is open and M is meager. If U is empty then
A = M is meager. So suppose that U is non-empty. Since U \ A ⊆ M is meager in
X, it is also meager in U , and hence A is comeager in U .
For the second statement, suppose that X is a Baire space, that A ⊆ X is meager,
and that U ⊆ X is open such that A is comeager in U . Then C := U \ A is meager in
X, and hence U ⊆ A ∪ C is meager. By item (3) of Proposition 6.3.7 we can therefore
conclude that U = ∅. This shows that the second alternative does not apply. 
It is sometimes convenient to use the following logical notation:
∀∗ x.A(x)
stands for A is comeager, and is pronounced ‘A holds for comeagerly many x’. Simi-
larly,
∃∗ x.A(x)
stands for A is non-meager, and pronounced ‘A holds for non-meagerly many x’. The
notation has the obvious extension to localised expressions of the form
∀∗ x ∈ U.A(x) and ∃∗ x ∈ U.A(x)
The following result is sometimes called the Fubini theorem for category because
of the analogy to Fubini’s theorem about the order of integration in double integrals
in analysis.
Theorem 6.3.10 (Kuratowski-Ulam). Let X, Y be second-countable and suppose
that A ⊆ X × Y has the Baire property. Then
∃∗ (x, y).A(x, y) ⇔ ∀∗ x∃∗ y.A(x, y) ⇔ ∀∗ y∃∗ x.A(x, y) (13)
∗ ∗ ∗ ∗ ∗
and ∀ (x, y).A(x, y) ⇔ ∀ x∀ y.A(x, y) ⇔ ∀ y∀ x.A(x, y) (14)
Proof. Clearly, (13) and (14) are equivalent by taking complements. Moreover,
the two equivalences in (13) are symmetric, so we only show the first equivalence.
If S ⊆ X × Y and x ∈ X, then Sx denotes the set {y ∈ Y | (x, y) ∈ S}. To show
the forward implication in (13), we first prove that if a set B ⊆ X × Y is nowhere
dense, then ∀∗ x(Bx is nowhere dense in Y ). We may assume that B is closed, since
replacing B by its closure can only increase the sets Bx ; hence, if we can prove that
for comeagerly many x ∈ X some supersets of the Bx are nowhere dense in Y , then
the same holds for the sets Bx . Let U := (X × Y ) \ B. It suffices to show that
∀∗ x(Ux is dense), because if Ux is dense then B x = Bx = X \ Ux does not contain a
 6.3. CONSISTENCY OF AUTOMATIC CONTINUITY 97

non-empty open set and hence ∀∗ (Bx is nowhere dense). Let {V1 , V2 , . . . } be a basis
of non-empty open sets for Y . Then
Un := {x ∈ X | ∃y ∈ Vn .(x, y) ∈ U }
is the projection of an open set and hence open. Moreover, Un is dense in X: if
G ⊆ X is nonempty open, then U ∩ (G × Vn ) 6= ∅, because otherwise B would
T contain
the nonempty open set G × Vn , contrary to being nowhere dense. If x ∈ n∈N Un ,
T for every n ∈ N we have Ux ∩ Vn 6= ∅, i.e., Ux is dense. We are done because
then
n∈N Un is comeager by Corollary 6.3.6.
S
Let A be meager. Then A = n∈N Bn where Bn is nowhere dense. For each
n ∈ N, let Xn be the set of all x ∈ X such that (Bn )x is nowhere dense;
T by what
we have seen above, Xn is comeager. By Corollary 6.3.6, we have that n∈N Xn is
comeager. It follows that
∀∗ x ∀n ∈ N (Bn )x is nowhere dense

which implies that

∀∗ x(Ax is meager)
S
since Ax = n∈N (Bn )x .
We now prove the converse implication in (13). Suppose that A is non-meager.
Since A has the Baire Property, there exists an open set U and a meager set M
such that A = U ∆M . Since A is non-meager, U is non-meager. By the definition of
the product topology, U is a union of sets of the form S × T for S ⊆ X open and
T ⊆ Y open. Since X and Y are second-countable, we may assume that this union
is countable. Hence, there must exist some open S ⊆ X and some open T ⊆ Y such
that S × T ⊆ U and S × T is non-meager (otherwise U would be meager). S
Then both S and T are non-meager:Sotherwise, if S is meager and S = n∈N Fn
where Fn is nowhere dense, then S×T = n (Fn ×T ), so it is enough to show that Fn ×
T is nowhere dense. Indeed, since U = X \ Fn is dense open in X (Proposition 6.3.2),
we have that U × Y is dense open in X × Y . Since X × Y \ Fn × T contains U × Y ,
it is dense, and hence Fn × T is nowhere dense.
For every x ∈ S we have Ax ⊇ V \ Mx . By the forward implication, Mx is
meager for comeagerly many x ∈ X. Therefore, Ax is non-meager for comeagerly
many x ∈ S. This implies that Ax is non-meager for non-meagerly many x ∈ X, i.e.,
∃∗ x∃∗ y.A(x, y). 

Remark 6.3.11. Note that the assumption that A has the Baire property was only
used in the proof of the converse implication in (13); the forward implication holds in
general.
The following is beyond the scope of this text, but we need it in the important
Proposition 6.3.18 below.
Theorem 6.3.12 (Lusin-Sierpiński; see Kechris [88] (21.6)). Let X be Polish, let
Y be a metric space, and let f : X → Y be continuous. Then the image of every open
set in X under f has the Baire property in Y .
The axiom of dependent choices (DC) is a weak form of the axiom of choice
(AC) that is still sufficient to develop most of real analysis (see Appendix A.2). Over
Zermelo-Fraenkel set theory (ZF), the Axiom of Dependent Choices is equivalent to
the version of the Baire Category Theorem (Theorem 4.1.8) where we only require
that S is completely metrisable (instead of Polish). The relevance of the following for
reconstruction has been pointed out by Lascar [104, Theorem 2.7].
98 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

Theorem 6.3.13 (Solovay [150], Shelah [143]). If ZF is consistent, then it is

consistent with ZF+DC that every subset of R has the Baire property.
Recall that the irrational numbers are homeomorphic to the Baire space (Theo-
rem 4.1.3). Hence, we obtain the following.
Corollary 6.3.14. If ZF is consistent, then it is consistent with ZF+DC that
every subset of NN has the Baire property.
6.3.4. The Baire property and permutation groups. We now discuss the
concepts from the previous section (nowhere dense, meager, Baire) in the context
of permutation groups. Every closed subgroup of Sω is Polish (Theorem 4.3.2) and
hence Baire (Theorem 4.1.8). In particular, G itself is non-meager; this follows from
Proposition 6.3.7 (3) since G is non-empty.
Lemma 6.3.15. Let G be a Polish group. If U is a subgroup of G of countable
index, then U is non-meager.
Proof. If U is meager, then all cosets of U are meager, too. The set G is the
union of all the cosets of U , but is non-meager. Since a countable union of a meager
set is meager, U must have uncountable index. 

The following lemma is sometimes called Pettis’ lemma; it is Lemma 2.6 in [104]
and a consequence of Theorem 2.3.2 in [60] or Lemma 9.9 in [88].
Lemma 6.3.16. Let G be a closed subgroup of Sym(N) and let H be a subgroup
of G. Then H is meager, open, or for every non-empty open U ⊆ G the set U \ H is
non-meager (in particular, G\H is dense). Consequently, if H has the Baire property
then H is either meager or open.
Proof. The second statement clearly follows from the first: if U has the Baire
Property then there exists an open U ⊆ Sym(N) such that H∆U is meager, and in
particular U \ H is meager, so either U is empty and H is meager, or H is open by
the first statement.
To show the first statement, let H be non-meager and suppose that there exists
a non-empty open U ⊆ G such that U \ H is meager. Then U contains f Ga for some
f ∈ G, a ∈ Nn , and n ∈ N.
Claim. The subgroup H ∩ Ga of Ga is comeager in Ga . The set f Ga is non-
meager. If f Ga ⊆ U \H, then U \H would not be meager, contrary to our assumptions.
Hence, there exists exists an h ∈ H ∩ f Ga . Since f Ga \ H ⊆ U \ H is meager, we
have that h−1 f Ga \ h−1 H is meager. But h−1 H = H and h−1 f Ga = Ga and hence
Ga \ H is meager, which proves the claim.
The claim implies that all cosets of H ∩Ga in Ga are comeager. Since intersections
of comeager sets are comeager (Corollary 6.3.6) and in particular non-empty, there can
be only one coset. Therefore, H contains Ga and hence is open by Lemma 4.4.1. 

Theorem 6.3.17 (Lascar [104]). Assume that ZF is consistent. Then it is consis-

tent with ZF+DC that every closed subgroup G of Sω the SIP and hence has automatic
continuity.
Proof. Assume that every subset of G has the Baire property; this is consistent
with ZF+DC by Corollary 6.3.14. Let U be a subgroup of G of countable index.
Then U cannot be meager by Lemma 6.3.15, so it must be open by Lemma 6.3.16.
Automatic continuity follows by Proposition 6.2.1. 
 6.4. AMPLE GENERICS 99

So we need the Axiom of Choice to find closed subgroups of Sω without automatic

continuity. The following statement demonstrates that the machinery of this section
can also be used for permutation groups to prove statements in ZF.
Proposition 6.3.18 (Corollary 2.8 in [104]). Let φ be an continuous isomorphism
between closed subgroups of Sω . Then φ is a homeomorphism.
Proof. Let φ : G → H for closed G, H ≤ Sω . Let U be an open subgroup of
G. We have to show that φ(U ) is open in H (Remark 4.3.1). Theorem 6.3.12 asserts
that φ(U ) has the Baire property. Recall that U has countable index in G, so φ(U )
has countable index in H. Hence, it cannot be meager according to Lemma 6.3.15
and hence must be open because of Lemma 6.3.16. 

Remark 6.3.19. Proposition 6.3.18 shows that automatic continuity of closed

subgroups of Sω is a property of the abstract group in the sense that if two closed
subgroups of Sω are isomorphic as abstract groups, and one has automatic continuity,
then so has the other.

Exercises.
(121) Prove the statement in Remark 6.3.19.

6.4. Ample Generics

In this section we introduce a powerful method for proving that certain auto-
morphism groups G have the small index property (without any set-theoretic as-
sumptions). The idea is to isolate elements of G that look typical ; the notion of
automorphisms looking the same is captured by conjugation (Example 13).
Definition 6.4.1 (Generic elements). An element of a Polish group G is called
generic if it lies in a comeager orbit with respect to the action of G on G by conjuga-
tion.
Note that there is always at most one comeager orbit, since the intersection of
two comeager sets is comeager (Corollary 6.3.6) and in particular non-empty. The
following example is taken from Truss [159].
Example 84. Sω has a generic element. In fact, g ∈ Sω is generic if and only if
g has no infinite cycles and infinitely many cycles of any given finite length. To see
this, let C be the set of permutations of the specified cycle type. Clearly, all elements
in C are conjugate in Sω ; moreover, conjugate elements have the same cycle type
(see Exercise 12). Hence, for the backwards implication we have to prove that C is
comeager. Observe that for every x ∈ N
Dx = {g ∈ Sω | x lies in a finite cycle of g}
is dense and open. Similarly, for all m, n ≥ 1 the set
D(m, n) := {g ∈ Sω | g has at least m cycles of length n}
is dense and open. Now
\ \
C= Dx ∩ D(m, n)
x∈N m,n≥1

and it follows that C is comeager. For the forward implication, recall that there is
at most one comeager orbit, which shows that the elements of C are the only generic
elements of Sω . 4
100 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

Example 85. The automorphism group G of an equivalence relation on a count-

ably infinite set with two infinite classes has no generic elements. Clearly, G has a
normal subgroup of index two. But Polish groups with a generic element have no
proper normal subgroups of countable index. To see this, let C be the conjugacy
class of the generic element g. Suppose for contradiction that G has a proper nor-
mal subgroup N of countable index. If g ∈ N , then by the normality of N we have
C ⊆ N , so N is comeager. Let f ∈ G \ N . Then f N is comeager and disjoint from N ,
contradicting the fact that the intersection of two comeager sets is comeager (Corol-
lary 6.3.6). Hence, C ∩ N = ∅ and N is meager. Since every coset of N is meager, G
is a countable union of meager sets and hence meager, a contradiction. 4
Example 86. The automorphism group of Aut(Q; <) has generic elements. They
can be described explicitly; this is beyond the scope of this text but can be found
in [159] (Theorem 4.1). Their existence can also be derived from the techniques that
will be developed in the following sections; see Example 89. 4
The diagonal conjugacy action of G on Gn is the action given by
g · (g1 , . . . , gn ) := (gg1 g −1 , . . . , ggn g −1 ).
An element g ∈ Gn is called n-generic if the orbit of g under this action is comeager.
A Polish group G has ample generics if Gn contains an n-generic element for each
n ∈ N>0 . We will see many examples of structures with ample generics later.
Example 87. Let G = Aut(Q; <). Then G does not have a 2-generic element;
this is due to Hodkinson and can be found in [160] (Theorem 2.4). Hence, G does
not have ample generics. Another proof of this fact was found by Siniora (Lemma
6.1.1 in [146]; see Corollary 6.4.13). 4
A proof of the following result can be found e.g. in [60] (Theorem 3.2.4).
Theorem 6.4.2 (Effros). Let G be a Polish group which acts continuously on a
Polish space X. Then for every x ∈ X the following are equivalent:
(1) G · x is non-meager in its closure.
(2) The map g 7→ g · x is open from G onto G · x.
(3) The map from G/Gx onto G · x that maps gGx to g · x, for every g ∈ G, is
a homeomorphism.
(4) G · x is Gδ , i.e., a countable intersection of open subsets of X.
(5) G · x is comeager in its closure.
Lemma 6.4.3 (Lemma 6.6 in [92]). Let G be a closed subgroup of Sω with ample
generics. Let A ⊆ G be non-meager and B ⊆ G be non-meager in any non-empty
open set. Let n ∈ N and x̄ ∈ Gn be n-generic and V ⊆ G be open such that 1 ∈ V .
Then there are a ∈ A, b ∈ B, and v ∈ V such that (x̄, a) and (x̄, b) are (n + 1)-generic
and v · (x̄, a) = (x̄, b).
Proof. Since G has ample generics, the action of G on Gn+1 by conjugation
has a comeager orbit C, i.e., ∀∗ (z̄, y)C(z̄, y). By the theorem of Kuratowski-Ulam
(Theorem 6.3.10; see Remark 6.3.11), we have that
Z := {z̄ | ∀∗ y.C(z̄, y)}
is comeager, and hence Z must intersect the comeager orbit of x̄. Moreover, Z is
preserved by the action of G on Gn , so the orbit of x̄ is contained in Z. It follows in
particular that ∀∗ y.C(x̄, y), i.e., the set Cx̄ := {y ∈ G | C(x̄, y)} is comeager.
Let y ∈ Cx̄ . Let Gx̄ be the stabiliser of x̄ for the conjugation action of G on Gn .
Note that Cx̄ = Gx̄ · y: indeed, since C is an orbit of the conjugation action of G on
 6.4. AMPLE GENERICS 101

X n+1 , for any z ∈ G we have (x̄, z) ∈ C if and only if there exists g ∈ G such that
g · (x̄, z) = (x̄, y), i.e., g ∈ Gx̄ such that g · z = y. It follows that for any y ∈ Cx̄ the
set Gx̄ · y is comeager. Fix a ∈ A ∩ Cx̄ .
By Effros’ theorem (Theorem 6.4.2, (2)), the set (Gx̄ ∩ V ) · a is open in Gx̄ · a.
By assumption, A is non-meager in (Gx̄ ∩ V ) · a. Hence, (Gx̄ ∩ V ) · a ∩ B 6= ∅. Fix
b ∈ (Gx̄ ∩ V ) · a ∩ B. Then for some v ∈ Gx̄ ∩ V we have v · a = b. 

Theorem 6.4.4 (Theorem 5.3 in [74], Theorem 6.9 in [92]). Let G be a closed
subgroup of Sω with ample generics. Then G has the small index property.
Proof. Suppose that H ≤ G has countable index. Then H is non-meager by
Lemma 6.3.15. If there exists a non-empty open set U such that U \ H is meager,
then H is open by Lemma 6.3.16.
Otherwise, G \ H is non-meager in every non-empty open set. In this case we will
reach a contradiction as follows. We will apply Lemma 6.4.3 to A = H and B = G\H
to construct for every a ∈ 2N an element ha ∈ G such that for all a, b ∈ 2N , if a 6= b
then ha and hb lie in different cosets of H in G, contradicting the assumption that H
has countable index.
Let a ∈ 2N . For every n ∈ N and s = (s0 , s1 , . . . , sn ) ∈ {0, 1}n+1 we inductively
define xs , fs , hs ∈ G such that
(1) xs is generic,
(2) xs ∈ H if sn = an and xs ∈ G \ H if sn = 1 − an ,
(3) fs = 1G if sn = 1 − an ,
(4) hs = f(s0 ) ◦ f(s0 ,s1 ) ◦ · · · ◦ f(s0 ,...,sn ) ,
(5) d(hs , hs f(s,1−an ) ) < 2−n (we use any compatible complete metric d on G,
e.g., the metric d0 from Lemma 4.2.15),
(6) if sn = (s0 , an ) for s0 ∈ Nn , then fs x(s0 ,an ) fs−1
0 = x(s0 ,1−an ) .
For n = 1, and if a0 = 0, we apply Lemma 6.4.3 for A := H, B := G \ H, and V = G
and set x0 := a, x1 := b, f1 := 1G , and f1 := v. If a0 = 1, we set x0 := b, x1 = a,
f1 := v −1 , and f0 := 1G .
For n > 1, suppose that xs and fs , for s ∈ {0, 1}n , are already defined. First
consider the case that an = 0. Again by Lemma 6.4.3, this time applied to V :=
{f ∈ G | d(hs , hs f(s,1) ) < 2−n }, there are x(s,0) ∈ H and x(s,1) ∈ G \ H such that
(xs , s(s,0) ) and (xs , s(s,1) ) are generic, and f(s,0) · x(s,0) = x(s,1) . The inductive step
for an = 1 is analogous. By (3) and (4), the sequence (hsn )n∈N is Cauchy, and by the
completeness of d converges to some ha ∈ G.
Claim 1. a 7→ ha is a continuous map from 2N to G.
Claim 2. If a, b ∈ 2N are such that for some n ∈ N we have a1 = b1 , . . . , an = bn ,
an+1 = 0, and bn+1 = 1, then ha · H ∩ hb · (G \ H) 6= ∅. Indeed, we have
ha · x(a1 ,...,an ,0) = h(a0 ,...,an ,0) · x(a1 ,...,an ,0)
= h(a0 ,...,an ) f(a0 ,...,an ,1) · x(a1 ,...,an ,0)
= h(a0 ,...,an ) · x(a1 ,...,an ,0) (as f(a0 ,...,an ,1) = 1G )
and
hb · x(a1 ,...,an ,1) = h(a0 ,...,an ,1) · x(a1 ,...,an ,1)
= h(a0 ,...,an ) f(a0 ,...,an ,0) · x(a1 ,...,an ,1)
= h(a0 ,...,an ) · x(a1 ,...,an ,0) .
So ha · x(a1 ,...,an ,0) = hb · x(a1 ,...,an ,1) which proves the claim since x(a1 ,...,an ,0) ∈ H
and x(a1 ,...,an ,1) ∈ G \ H.
102 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

Claim 3. If a, b ∈ 2N are distinct, then ha and hb lie in different cosets of H. Let

n ∈ N be smallest so that an 6= bn . By the previous claim we have
−1
ha ◦ H ◦ h−1
a ∩ hb ◦ (G \ H) ◦ hb 6= ∅

and hence
h−1 −1
b ha Hha hb ∩ G \ H 6= ∅,
so h−1 ha ∈
/ H. This shows that ha and hb are in different cosets of H.
Claim 3 contradicts the assumption that H has countable index in G. 

Exercises.
(122) Discuss whether and why the adjective generic in the name of
the concept of a generic superposition (Section 3.4.2) is appropriate.

(123) Show that the set Dx from Example 84 is indeed dense

and open.
(124) Show that the set D(m, n) from Example 84 is dense and open.
(125) Let V be a countably infinite set and let E be an equivalence relation on
V with countably many countable classes. Prove that Sym(V ; E) does not
have ample generics.

6.4.1. Dense conjugacy classes and the JEP. We have seen in the previous
section that if G ≤ Sω has ample generics, then it has the small index property: but
how do we prove that G has ample generics? To this end, we present in this section and
the following sections an elegant characterisation of the existence of ample generics
of Kechris and Rosendal [92] which builds on ideas from [74] and [159].
If G has a generic element α, then the orbit of α is in particular dense (since G
is Polish and by Proposition 6.3.7). So we will first focus on understanding whether
G has a dense conjugacy class.
Definition 6.4.5. Let K be an amalgamation class. Then Kp denotes the class
of all pairs
(A, e : B → C)
such that A ∈ K and e is an isomorphism between substructures B and C of A.
An embedding of (A, e : B → C) ∈ Kp into (A0 , e0 : B 0 → C 0 ) ∈ Kp is an embed-
ding f : A ,→ A0 such that f (B) ⊆ B 0 , f (C) ⊆ C 0 , and f ◦ e = e0 ◦ (f |B ). Note that
the definition of the joint embedding property (JEP) and the amalgamation prop-
erty (AP) were purely categorical in the sense that their definition only requires a
notion of embedding; so JEP and AP are naturally defined not only for structures
and embeddings, but also for classes of the form Kp as introduced above.
We say that a continuous action ξ : G → Sym(X), for some set X, is topologically
transitive if for any two non-empty open subsets U, V ⊆ X there exists g ∈ G such
that g(U ) ∩ V 6= ∅. The implication from (1) to (2) in the following theorem can
already be found in [158]; the equivalence of (1) and (2) is Theorem 2.1 in [92].
Theorem 6.4.6. Let K be an amalgamation class and let L be its Fraı̈ssé limit.
Then the following are equivalent.
(1) G := Aut(L) has a dense conjugacy class.
(2) Kp has the JEP.
(3) The action of G on G by conjugation is topologically transitive.
 6.4. AMPLE GENERICS 103

Proof. (1) ⇒ (2). Fix an element α ∈ G having a dense conjugacy class in

G. To show that Kp satisfies the JEP, let (Ai , ei : Bi → Ci ) ∈ Kp for i ∈ {1, 2}; we
assume that Ai is a substructure of L. By the homogeneity of L the embedding ei
has an extension in G, so by the density of the conjugacy class of α there is a βi ∈ G
such that ei = βi−1 αβi |B . Let

A := L[β1 A1 ∪ β2 A2 ] B := L[β1 B1 ∪ β2 B2 ] e := α|B .

Then hi := βi |Ai is an embedding of (Ai , ei ) into (A, e).

(2) ⇒ (3). Suppose that Kp has the JEP and let U1 , U2 ⊆ G be non-empty
open. For i ∈ {1, 2}, the set Ui contains a set of the form {α ∈ G | ei = α|Bi } for
some isomorphism ei : Bi → Ci between finite substructures of L. Let Ai := Bi ∪ Ci .
Since Kp has the JEP, there exists (A, e) and embeddings f1 : (A1 , e1 ) ,→ (A, e) and
f2 : (A2 , e2 ) ,→ (A, e). By the homogeneity of L there exist β, γ1 , γ2 ∈ G such that β
extends e, γ1 extends f1 , and γ2 extends f2 . Then (γ2−1 γ1 · U1 ) ∩ U2 6= ∅, because
• γ2−1 βγ2 ∈ U2 since it extends e2 , and
• γ2−1 βγ2 ∈ γ2−1 γ1 · U1 = γ2−1 γ1 U1 γ1−1 γ2 since γ1−1 βγ1 extends e1 .
(3) ⇒ (1). There are countably many basic open sets of the form S(a, b) in G,
for a, b ∈ Ln , n ∈ N. For each of them, the set

Da,b := G · S(a, b) = {α ∈ G | ∃g ∈ G such that gαg −1 (a) = b}

is clearly open, and dense since the conjugation action is topologically transitive.
Since G is a Baire space, the countable intersection C over all the Da,b is dense, and
in particular non-empty. Let f ∈ C; then the conjugacy class of f is dense, because
for every a, b ∈ Ln , n ∈ N, we have that f ∈ Da,b , and hence there is g ∈ G such that
g · f ∈ Sa,b . 

Corollary 6.4.7. Sω , the automorphism group of the random graph, and more
generally the automorphism group of Fraı̈ssé-limits of classes with free amalgamation
have a dense conjugacy class.

Proof. Let K be the class of all finite structures with the empty signature so
that the automorphism group of the Fraı̈ssé-limit of K is isomorphic (as a permutation
group) to Sym(N). By Theorem 6.4.6, it suffices to verify that Kp has the JEP. So
let (Ai , ei : Bi → Ci ) ∈ Kp for i ∈ {1, 2}. We may assume that A1 ∩ A2 = ∅ and
define A := A1 ∪ A2 , B := B1 ∪ B2 , C := C1 ∪ C2 , and e : B → C as the common
extension of both e1 and e2 . Then the identity map fi : Ai → A is an embedding of
(Ai , e) into (A, e) showing the JEP. The same proof works for the other groups in the
statement. 

Exercises.
(126) Let E 2 be the equivalence relation on N with two infinite classes,
and let K := Age(N; E 2 ). Show that Kp does not have the JEP.
(127) Directly show that Aut(N; E 2 ) does not have a dense conjugacy
class (without using Theorem 6.4.6).
(128) Show that Aut(Q; Cycl) (see Section 2.4) does not have a dense
conjugacy class.
(129) Let V be the countably infinite vector space over F2 .
Show that Aut(V ) has a dense conjugacy class.
104 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

6.4.2. Non-meager Conjugacy Classes and the WAP. Suppose that G is

a closed subgroup of Sω with a generic element α, i.e., the orbit of α with respect to
the conjugation action is comeager. This means in particular that the orbit of α is
non-meager. If the orbit of α is dense, then the converse is true as well, by Effros’
theorem (Theorem 6.4.2): indeed, if G · x is nonmeager in its closure, then G · α is
is comeager in its closure by the implication (1) ⇒ (5) in Effros’ theorem. Thus, if
G · α = G, then G · α is comeager in G.
This section presents an equivalent characterisation of non-meager orbits of con-
tinuous actions of automorphism groups on Polish spaces; in the next section, this
will be applied to characterise the existence of generic elements.
Proposition 6.4.8 (Proposition 3.2 in [92]). Let G ≤ Sym(N) be closed and
ξ : G → X be a continuous action of G on a Polish space X. Let x ∈ X. Then the
following are equivalent.
(1) The orbit G · x is non-meager in X.
(2) for every open subgroup V of G, the set V · x is non-meager in X.
(3) For every open subgroup V of G, the set V · x is somewhere dense in X.
(4) For every open subgroup V of G, we have x ∈ Int(V · x).
Proof. (1) implies (2). Suppose that G · x is non-meager in X and V ≤ G is
an open subgroup of G. Since S open subgroups of G have countable index, we finde
g0 , g1 , . . . such that G = i∈N gi V . So some gn V · x is non-meager. Hence, V · x is
non-meager in X.
Clearly, (2) implies (3).
(3) implies (4). Let V ≤ G be open. Then V · x is somewhere dense, i.e., V · x
contains a non-empty open set U . Take g ∈ V such that g·x ∈ U . Then g −1 V ·x = V ·x
is dense in the open set g −1 · U which contains x. Hence, x ∈ Int(V · x).
(4) implies
S (1). We prove the contraposition and suppose that G·x is meager, i.e.,
G · x ⊆ n∈N Fn where F0 , F1 , F2 , . . . are closed nowhere dense subsets S of X. Then
for each n ∈ N the set Kn := {g ∈ G | g · x ∈ Fn } is closed. Note that n∈N Kn = G
and hence some Kn must be non-meager, i.e., Kn must have non-empy interior. So
there is a non-empty open subgroup V ≤ G and g ∈ G such that Kn contains gV .
Then gV · x ⊆ Fn and hence V is nowhere dense. Therefore V · x is nowhere dense
and V · x has empty interior, which proves that (4) does not hold. 
As in the case of the JEP, also the WAP only depends on the notion of embed-
dings, and hence makes also sense for the class Kp from Definition 6.4.5. We now
study the question whether classes of the form Kp from Section 6.4.1 have the AP or
the WAP.
Example 88. Let K be the class of finite structures with the empty signature.
Then Kp does not have the AP, but it has the WAP. To see why the amalgamation
property fails, let A ∈ K be such that A contains a single element a, and let e be the
isomorphism with the empty domain. Let B 1 be the extension of A by two elements
and let e1 : B1 → B1 be the bijection which exchanges the two elements, and let B 2
be the extension of A with three elements and e2 : B2 → B2 be the bijection which
cyclically shifts the three elements. Then (A, e) embeds via the inclusion map into
(B i , ei ), for i ∈ {1, 2}. If (C, g) is such that there is an embedding fi from (B i , ei )
into (C, g), then f1 (a) 6= f2 (a) because f1 (a) must lie in a cycle of g of length 2 and
f2 (a) must lie in a cycle of g of length 3.
To verify that Kp has the WAP, let (A, e : B → C) ∈ Kp . Let e0 be an extension
of e to a permutation of A. Clearly, (A, e) embeds into (A, e0 ). For i ∈ {1, 2}, let
(Ai , ei : Bi → Ci ) ∈ Kp be such that there exists an embedding fi : (A, e0 ) → (Ai , ei );
 6.4. AMPLE GENERICS 105

we may assume without loss of generality that A1 ∩ A2 = A and that e1 |A = e2 |A =

idA . Let A∗ be the free amalgam of A1 and A2 , and for i ∈ {1, . . . , n} let e∗ be the
union of e1 and e2 ; this is well-defined by the stipulation that e1 |A = e2 |A . Note that
idAi , for i ∈ {1, 2}, is an embedding of (Ai , ei ) into (A∗ , e∗ ), proving that (A, e0 ) is
determined on (A, e). This shows the WAP for Kp . 4
The proof of the WAP in the previous example works more generally if every
(A, e) ∈ Kp can be embedded into (A0 , e0 ) such that e0 is an automorphism of A0 ,
since (A0 , e0 ) is always determined on (A, e). The following example shows a situation
where the WAP cannot be shown in this way by extending e to an automorphism.
Example 89. Let K := Age(Q; <). Then Kp does not have the AP: to see this,
suppose that A ∈ K contains a single element a, and let e be the isomorphism with
the empty domain, just as in Example 88. Let B ∈ K be with two elements a, b, let
e1 : {a} → {a}, and let e2 : {a} → {b} be partial isomorphisms. Then for i ∈ {1, 2}
the identity id{a} is an embedding of (A, e) into (B, ei ). If C is such that there are
embeddings gi : B → C, for i ∈ {1, 2}, and g1 (a) = g2 (a), then g1 (e1 (a)) = g1 (a) 6=
g2 (b) = g2 (e2 (a)).
To see that Kp has the WAP, let (A, e) ∈ Kp . To find (A0 , e0 ) ∈ Kp which is
determined on (A, e), we want to extend e so that it is defined on all of A; to this end,
we might have to choose A0 to be a proper extension of A. We may suppose that A is a
substructure of (Q; <) and by the homogeneity of (Q; <) we find α ∈ Aut(Q; <) which
extends e. Let A0 := (Q; <)[A ∪ α(A)] and e0 := α|A . Then idA is an embedding of
(A, e) into (A0 , e0 ). To verify that (A0 , e0 ) is determined on (A, e), let fi , for i ∈ {1, 2},
be an embedding of (A0 , e0 ) into (B i , ei ). Let C be the free amalgam of B 1 and B 2 over
A. Then repeatedly identify e1 (x) with e2 (x) in C for all x ∈ B1 ∩ B2 ; the resulting
structure C 0 is well-defined by the property that both e1 and e2 are order-preserving.
In C the relation symbol < denotes an acyclic relation which can be extended to a
linear order; let D be the resulting structure in K. Then there exists an embedding
gi of B i into D and g1 ◦ f1 |A = g2 ◦ f2 |A . 4
Theorem 6.4.9 (Theorem 3.4 in [92], [80]). Let K be an amalgamation class and
let L be its Fraı̈ssé limit. Then the following are equivalent.
• G := Aut(L) has a generic element, i.e., a comeager conjugacy class.
• Kp has the JEP and the WAP.
Proof. Suppose that the orbit of α with respect to the action of G on G by
conjugation is comeager. Then the orbit is in particular dense (since G is Polish and
by Proposition 6.3.7) and hence Theorem 6.4.6 shows that Kp satisfies the JEP.
To show that Kp satisfies the WAP, let (A, e) ∈ Kp be given. We may suppose
that A is a substructure of L. Since α is in a dense conjugacy class we may assume
that α is an extension of e. For the open subgroup V := {β ∈ G | β|A = idA } of G, we
know by Proposition 6.4.8 (4) that V · α contains an open set U that contains α. We
may suppose that U is of the form S(a, f (a)) for some a ∈ Ln , n ∈ N, whose entries
contain all elements of A, and some isomorphism f between finite substructures of L.
Since α extends e, we know that f extends e as well.
Let A0 be the substructure induced by L on dom(f ) ∪ im(f ). For i ∈ {1, 2}, let
ei : (A0 , f ) → (B i , fi ) for some (B i , fi ) ∈ Ki . Since L is homogeneous we may assume
that B i is a substructure of L and ei is the identity on A0 , and consequently fi extends
f . By the density of V · α in U there is βi ∈ V such that βi−1 αβ extends fi . Let
C := L[β(B1 ) ∪ β(B2 )] and let g be the restriction of α to β1 (dom(f1 )) ∪ β2 (dom(f2 )).
Then βi |Bi is an embedding of (B i , fi ) into (C, g). Finally, for every a ∈ A we have
β1 (f1 (a)) = β2 (f2 (a)), so (C, g) is indeed an amalgam of (B 1 , f1 ) and (B 2 , f2 ) over
(A, f ).
106 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

Conversely, suppose that Kp has the JEP and the WAP. We will construct an
element α ∈ G such that G · α is dense in G and non-meager in its closure.
Let e1 , e2 , . . . be an enumeration of all isomorphisms e between finite substruc-
tures of L such that any two fi : (L[dom(e) ∪ im(e)], e) ,→ (B, g) can be amalgamated.
Claim. For every open subgroup V ≤ G TODO.
By Proposition 6.4.8 (4) ⇒ (1) we have that G · α is non-meager in G · α. Theo-
rem 6.4.2 (1) ⇒ (5) now implies that G · α is comeager in G · α = G. 
6.4.3. WAP and Ample Generics. This section finally presents the charac-
terisation of those homogeneous structures L whose automorphism group has ample
generics. Let L be the Fraı̈ssé-limit of the amalgamation class K. We introduce the
class Kpn for n ≥ 1, which consists of tuples (A, e1 : B1 → C1 , . . . , en : Bn → Cn ) where
A ∈ K and ei an isomorphism between substructures B i and C i of A. Embeddings
between elements of Kpn are defined analogously as embeddings between elements of
Kp , and again properties like the JEP and the AP make sense.
Theorem 6.4.10 (Theorem 6.2 in [92]). Let K be an amalgamation class and let
L be its Fraı̈ssé limit. Then the following are equivalent.
• Aut(L) has an n-generic element.
• The class Kpn has the JEP and the WAP.
Proof. The proof is similar to the proof of Theorem 6.4.9. 
Lemma 6.4.11. Sym(N) has ample generics.
Proof. Let K be the class of all finite structures over the empty signature. We
have already seen in Corollary 6.4.7 that Kp has the JEP; the proof that Kpn has the
JEP is analogous. Moreover, we have already seen in Example 88 that Kp has the
WAP; the proof that Kpn has the WAP is analogous: again we may find for every
(A, e) ∈ Kpn an extension (A0 , e0 ) ∈ Kpn which is determined on (A, e) by choosing e0
to be an automorphism of A0 . Now the statement follows from Theorem 6.4.10. 
Corollary 6.4.12. Sym(N) has the small index property, automatic continuity,
and automatic homeomorphicity.
Proof. We have just seen that Sym(N) has ample generics, so it follows from
Theorem 6.4.4 that Sym(N) has the small index property. Automatic continuity
follows from Proposition 6.2.1, and automatic homeomorphicity follows from Propo-
sition 6.3.18. 
Corollary 6.4.13. Aut(Q; <) does not have a 2-generic element.
Proof. Let K := Age(Q; <). By Theorem 6.4.10 it suffices to show that Kp2 does
not have the WAP. Let a1 , a2 ∈ Q be such that a1 < a2 , let A := (Q; <)[{a1 , a2 }],
let e1 (a1 ) = a2 and e2 (a1 ) = a2 . We claim that there is no (A0 , e01 , e02 ) ∈ Kp2 which
is determined on (A, e1 , e2 ). Note that in A0 , for i ∈ {1, 2} we have a1 < e0i (a1 ),
and if (e0i )k (a1 ) is defined for k ≥ 1, then (e0i )k−1 (a1 ) < (e0i )k (a1 ). Let k ≥ 1 be
smallest so that (e01 )k (a1 ) is undefined, and let b := (e01 )k−1 (a1 ). Let B 1 be the
extension of A0 by one new element c which is larger than all elements in A0 , and
let e1,1 be the extension of e01 given by e1,1 (b) = c, and let e1,2 be the extension
of e02 given by e1,2 (c) = c. Let B 2 be the extension of A0 by two new elements d1
and d2 such that d1 is larger than all elements in A0 and d2 is larger than d1 . Let
e2,1 be the extension of e01 given by e2,1 (b) = d1 and let e2,2 be the extension of e02
given by e2,2 (d1 ) = d2 . Note that (B 1 , e1,1 , e1,2 ), (B 2 , e2,1 , e2,2 ) ∈ Kp2 . Now suppose
for contradiction that there exists (C, e001 , e002 ) and embeddings fi : (B i , ei,1 , ei,2 ) ,→
 6.5. THE EXTENSION PROPERTY FOR PARTIAL AUTOMORPHISMS 107

(C, e001 , e002 ) such that f1 |A = f2 |A . Then in particular f1 (a1 ) = f2 (a1 ), and hence
f1 (b) = f1 ((e1,1 )k (a1 )) = f2 ((e2,1 )k (a1 )) = f2 (b), and f1 (e1,1 (b)) = f2 (e2,1 (b)). How-
ever, e1,2 (e1,1 (b)) = e1,1 (b) = c, whereas e2,1 (b) = d1 < d2 = e2,2 (e2,1 (b)). Hence,
f1 (e1,1 (b)) = f2 (e2,1 (b)) < f2 (e2,2 (e2,1 (b)) = f1 (e1,2 (e1,1 (b)) = f1 (e1,1 (b))
a contradiction. 

Exercises.
(130) Let K be the class of all finite partial orders. Show that the automorphism
group of the Fraı̈ssé-limit of K does not have ample generics.

6.5. The Extension Property for Partial Automorphisms

The results in this section can be used to prove that for some class of finite
structures K, the class Kpn from the previous section has the WAP. If a class of
finite structures has the so-called Extension Property for Partial Automorphisms,
then this is a combinatorial statement that is often of independent interest, and
found e.g. applications in theoretical computer science [64].
Definition 6.5.1 (EPPA / Hrushovski property). A class C of finite relational
τ -structures has the EPPA (the extension property for partial automorphisms, also
known as the Hrushovski property) if every structure A ∈ C has an extension B ∈ C
such that every partial isomorphism of A extends to an automorphism of B.
Clearly, the class of all finite linear orders, or the class of all finite partial orders
do not have the EPPA.
Theorem 6.5.2 (of [77]). The class of all finite undirected graphs has the EPPA.
Proof. Let X be a
finite set and n a positive integer. Let G(X, n) denote the

graph with vertex set X n , the set of n-element subsets of X, and where a, b ∈ n
X

are adjacent iff a ∩ b 6= ∅. An induced subgraph G0 of G(X, n) is called poor if

• every x ∈ X in contained in at most two vertices of G0 , and
• any two vertices of G0 intersect in at most one point.
Note that every permutation α of X induces an automorphism of G(X, n), which will
be denoted by α∗ .
Claim 1. For every finite graph G there exists a finite set X and a positive
integer n such that G is isomorphic to a poor subgraph of G(X, n). We only prove
the claim for d-regular graphs G; the argument can
be adapted to the general case.
Let X be the edge set of G. Define f : V (G) → X d by a 7→ {x ∈ X | a ∈ x}; this
map is an isomorphism between G and a poor subgraph of G(X, d).
Claim 2. For every isomorphism g between two poor subgraphs G0 and G1 of
G(X, n) there exists a permutation α of X such that α∗ extends g.
First define α(x) for x ∈ X belonging to two elements a, b ∈ V (G0 ) to be the
unique element of g(a) ∩ g(b); then define α(x) for the elements of X belonging to one
element of V (G0 ), and then for the others. 
Theorem 6.5.3. Let K be an amalgamation class with the EPPA. Then the
Fraı̈ssé limit of K has ample generics.
Proof. We use the EPPA to verify the assumptions of Theorem 6.4.10. Let
(A, e1 : B1 → C1 , . . . , en : Bn → Cn ) ∈ Kpn be given. Let D ∈ K be the structure from
Definition 6.5.1, i.e., a structure such that every partial isomorphism of A extends to
an automorphism of D. In particular, for each i ∈ {1, . . . , n}, there exists an extension
e0i of ei to an automorphism of D. Then (A, e1 , . . . , en ) embeds into (D, e01 , . . . , e0n ),
108 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

and we may amalgamate embeddings from (D, e01 , . . . , e0n ) into other structures of Knp ,
which shows that Kpn has the WAP. 

Corollary 6.5.4. The automorphism group of the random graph has ample
generics and the small index property.
Proof. Combine Theorem 6.5.2 and Theorem 6.5.3 to obtain ample generics,
and Theorem 6.4.4 to obtain the SIP. 

Herwig [68] showed EPPA for the class of all finite τ -structures, for any finite
relational signature τ (also see [70] and [81] for other proofs), properly generalising
Corollary 6.5.4. Herwig [69] showed EPPA for the class of all Kn -free graphs. More
generally, Lascar and Herwig [70, Theorem 3.2] showed EPPA for all classes that
are described by homomorphically forbidding finitely many structures, i.e., classes C
such that there exists a finite set of structures F such that A ∈ C if no structure in
F admits a homomorphism to A. Hodkinson and Otto [75] (also see Corollary 4.6
in [145]) obtained the following, also properly generalising Corollary 6.5.4.
Theorem 6.5.5. Let B be a homogeneous structure with finite relational signa-
ture whose age C has the free amalgamation property. Then C has the EPPA. Conse-
quently, Aut(B) has ample generics, the small index property, automatic continuity
and homeomorphicity, and reconstruction.

Exercises.
(131) Let B be a homogeneous structure with finite relational signature. Show
that the age of B has the EPPA if and only if Aut(B) is the closure of the
union of a countable chain G1 ≤ G2 ≤ · · · of compact subgroups of Aut(B).
(132) Let A be a finite substructure of a τ -structure B. We write A ≤homog B if
for every n ∈ N and any two a, b ∈ An we have that a and b lie in the same
orbit of (the componentwise action of) Aut(B) on B n if and only if they lie
in the same orbit of (the componentwise action of) Aut(A) on An .
(a) Prove that if A ≤homog B and B ≤homog C, then A ≤homog C.
(b) Provide a counterexample to the transitivity of ≤ from
the previous exercise if we replace the action of Aut(A)
on An by the action of Aut(B)A on B n .

(c) Show that if A ≤homog B and B is ω-categorical, then for every

relation R ⊆ B n with a definition in B the relation R ∩ An
is definable in A.
Let (Ai )i∈N be such that Ai ≤homog
S Ai+1 for all i ∈ N. We write limi∈N Ai
for the τ -structure with domain i∈N Ai whose relations are the unions of the
respective relations of the Ai . Write A for the class of all countably infinite
ω-categorical structures of the form limi∈N Ai such that Ai ≤homog A for all
i ∈ N.
(d) Show that no structure in A contains a linear order.
(e) Give three examples of structures in A with pairwise
non-isomorphic automorphism group.
(f) Show that if A ∈ A, then the expansion of A
by all definable relations is also in A and homogeneous.
 6.6. THE STRONG SMALL INDEX PROPERTY 109

(g) Show that if A is homogeneous, then A ∈ A if and only

if A is ω-categorical and for every first-order τ -sentence φ
such that A |= φ there exists a finite τ -structure B
such that B |= φ and B ≤homog A.
(h) Show that no homogeneous A ∈ A is finitely axiomatisable, i.e.,
there is no finite set of τ -sentences which is equivalent to Th(A)
(in the sense that it has the same models as Th(A)).
(i) Show that for every homogeneous and ω-categorical A,
we have A ∈ A if and only if Age(A) has EPPA.

6.6. The Strong Small Index Property

A permutation group G ≤ Sym(N) has the strong small index property (SSIP) if
every countable index subgroup of G lies between the pointwise and the set stabiliser
of a finite subset of N. The strong small index property for Sym(N) itself was shown
in [47]; it also follows from the Small Index Property of Sym(N) (Corollary 6.4.12)
via a more general result, which we present below (Theorem 6.6.6).
Example 90. An example of an oligomorphic permutation group which has the
small index property, but not the strong small index property is the automorphism
group G of an equivalence relation E on a set V with (at least) two infinite classes:
it has the open subgroup H which fixes one equivalence class of E. Then H has
countable index, but is not contained in the set-stabiliser of G of some finite subset
of V . On the other hand, G has the small index property. 4
We first start with some results that illustrate what the SSIP can be useful for.
Proposition 6.6.1. Let ξ : Sym(N) → Sym(N) be a homomorphism such that
ξ(Sym(N)) is a primitive permutation group G. Then there exists an n ∈ N such that


G is isomorphic (as a permutation group) to the setwise action of Sym(N) on Nn .
Proof. Recall from Corollary 1.4.8 that the primitivity of G implies that for any
a ∈ N the point stabiliser Ga is a maximal subgroup of G, and hence H := ξ −1 (Ga ) is
a maximal subgroup of Sym(N). The strong small index property of Sym(N) implies
that H is contained in the set-wise stabiliser Sym(N){F } for some finite F ⊆ N. By
the maximality of H, this means that H equals Sym(N){F } . Let n := |F |. Let i



be the map from Nn → N that maps for each α ∈ Sym(N) the set α(F ) ∈ Nn to
ξ(α)(a) ∈ N. Note that i is well-defined because if α, β ∈ Sym(N) are such that
α(F ) = β(F ), then α−1 β ∈ Sym(N){F } = H, and hence ξ(α−1 β) ∈ Ga . Thus,
α−1 β(a) = a and α(a) = β(a). Moreover,
i is an isomorphism between the image


of the setwise action of Sym(N) on Nn and G: for α ∈ Sym(N) and S ∈ Nn , let
β ∈ Sym(N) be such that β(F ) = S. Then we have
i(α(S)) = i(αβ(F )) = ξ(αβ)(a) = ξ(α)ξ(β)(a) = ξ(α)i(β(F )) = ξ(α)i(S). 
In fact, under the additional assumption of having no algebraicity, the strong small
index property allows a very strong form of reconstruction of countable ω-categorical
structures from their abstract automorphism group.
Theorem 6.6.2 (Paolini and Shelah [130] (Corollary 2)). Let A and B be count-
able ω-categorical structures with the strong small index property and without alge-
braicity. Then Aut(A) and Aut(B) are isomorphic as abstract groups if and only if
A and B are bi-definable (Definition 3.1.8). Moreover, if ξ : Aut(A) → Aut(B) is
an abstract group isomorphism, then there is a bijection f : A → B witnessing the
bi-definability of A and B such that ξ(α) = f αf −1 .
110 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

6.6.1. Weak Elimination of Imaginaries. The difference between the strong

small index property and the small index property is closely related to an important
property in model theory.
Definition 6.6.3. A structure A has weak elimination of imaginaries if for every
R ⊆ Ak which is parameter-definable over A there exists an inclusion-wise smallest
algebraically closed subset B ⊆ A such that R is definable in A over B.
Example 91. Let A be the relational structure with a single equivalence relation
with two infinite classes C1 and C2 . Clearly, A has no algebraicity. Let a, b ∈ A be
distinct but from the same equivalence class C1 . Then C1 is definable in A over {a},
and definable in A over {b}, but not definable in A without parameters. So A does
not have weak elimination of imaginaries. 4
The next example illustrates that whether an ω-categorical structure A has weak
elimination of imaginaries is not captured by the topological automorphism group of
A. Weak elimination of imaginaries of A is rather a property of the action of the
automorphism group.
Example 92. Let B be the relational structure obtained from the structure A in
the previous example (Example 91) as follows: add two new elements p and q, and
add a new binary relation {(a, p) | a ∈ C1 } ∪ {(b, q) | b ∈ C2 }. Note that A and B are
bi-interpretable, and hence Aut(A) and Aut(B) are isomorphic as topological groups.
However, B has weak elimination of imaginaries. As in Example 92, if a, b are distinct
elements of C1 , then C1 is definable in B over {a} and definable in B over {b}, but
neither {a} nor {b} is algebraically closed in B. Instead, aclB ({a}) = {a, p, q} and
aclB ({b}) = {b, p, q}, and there is a unique smallest algebraically closed subset of B
over which C1 is definable, namely {p, q}. (By adding the elements p and q we have
‘eliminated’ the ‘imaginary elements’ C1 and C2 .) 4
Lemma 6.6.4. Let A be a countable ω-categorical structure. Then the following
are equivalent.
(1) A has weak elimination of imaginaries.
(2) Every R ⊆ Ak which is definable in A over B1 ⊆ A and definable in A over
B2 ⊆ A is also definable in A over acl(B1 ) ∩ acl(B2 ).
(3) For all finite B1 , B2 ⊆ A that are algebraically closed in A we have
hAut(A)(B1 ) ∪ Aut(A)(B2 ) i = Aut(A)(B1 ∩B2 ) . (15)
(4) For every open subgroup H of Aut(A) there exists a unique finite alge-
braically closed B ⊆ A of smallest cardinality such that Aut(A)(B) ⊆ H.
(5) For every open subgroup H of Aut(A) there exists a finite algebraically closed
finite set B ⊆ A such that Aut(A)(B) ≤ H ≤ Aut(A){B} .
(6) Every open subgroup H ≤ Aut(A) =: G has a finite index subgroup which is
the point stabiliser G(B) for some finite B ⊆ A.
Proof. We show (1) ⇔ (2), (2) ⇔ (3), (3) ⇒ (4) ⇒ (5) ⇒ (6) ⇒ (3).
(1) ⇒ (2). Let B be the inclusion-wise smallest algebraically closed subset of
A such that R is definable in A with parameters from B. Then B ⊆ acl(B1 ) and
B ⊆ acl(B2 ), so R is also definable in A over acl(B1 ) ∩ acl(B2 ).
(1) ⇐ (2). Every parameter-definable set is definable over a finite set B. Since in
ω-categorical structures, the algebraic closure of a finite set is finite (Lemma 3.4.3),
we may assume that B is algebraically closed. Suppose that B1 and B2 are inclusion-
wise minimal algebraically closed sets such that R is first-order definable in A over
each of them. By assumption, R is also definable over acl(B1 ) ∩ acl(B2 ) = B1 ∩ B2 .
 6.6. THE STRONG SMALL INDEX PROPERTY 111

The minimality of B1 and B2 then implies that B1 = B2 is the smallest algebraically

closed set such that R is definable in A over B.
(2) ⇒ (3). Let B1 , B2 ⊆ A be finite and algebraically closed in A. The inclusion
⊆ in (15) holds in general. For the converse inclusion, first observe that the group on
the left is a closed subgroup of Sym(A). So by Exercise 11, it suffices to show that
every relation that is preserved by hAut(A)(B1 ) ∪ Aut(A)(B2 ) i is also preserved by
Aut(A)(B1 ∩B2 ) . Every relation R that is preserved by hAut(A)(B1 ) ∪ Aut(A)(B2 ) i is in
particular preserved by Aut(A)(B1 ) , and by Theorem 3.1.1 it is first-order definable
in A over B1 . Similarly, R is first-order definable in A over B2 . By assumption, R is
also definable in A over B1 ∩ B2 , and hence preserved by Aut(A)(B1 ∩B2 ) .
(2) ⇐ (3). If R ⊆ Ak is definable in A over B1 ⊆ A and over B2 ⊆ A, then it is also
definable over finite subsets B10 ⊆ B1 and B20 ⊆ B2 . Since in ω-categorical structures
the algebraic closure of a finite set is finite, we may assume that B10 and B20 are
algebraically closed. The relation R is preserved by Aut(A)(B10 ) and by Aut(A)(B20 ) .
Then (2) implies that R is preserved by Aut(A)(B10 ∩B20 ) . By Theorem 3.2.3 and
Theorem 3.1.1, R is definable in A over B10 ∩ B20 ⊆ acl(B1 ) ∪ acl(B2 ).
(3) ⇒ (4). Let H ≤ Aut(A) be open. Then there exists a finite B ⊆ A such that
Aut(A)(B) ≤ H (Lemma 4.4.1). By ω-categoricity, the algebraic closure of B is finite
(Lemma 3.4.3), so we may assume without loss of generality that B is algebraically
closed. Suppose for contradiction that there are two distinct algebraically closed finite
sets B1 , B2 of minimal cardinality with Aut(A)(B1 ) ≤ H and Aut(A)(B2 ) ≤ H. By
assumption, Aut(A)(B1 ∩B2 ) = hG(B1 ) ∪ G(B2 ) i ⊆ H. Note that B := acl(B1 ∩ B2 ) ⊆
acl(B1 ) = B1 and acl(B1 ∩ B2 ) ⊆ acl(B2 ) = B2 , so |B| < |B1 | = |B2 |, a contradiction.
(4) ⇒ (5). It suffices to show that H ⊆ Aut(A){B} . Let h ∈ H. Then G(h(B)) =
hG(B) h−1 ⊆ hHh−1 = H. Clearly, h(B) is finite and algebraically closed, so by
assumption we have that B ⊆ h(B). Hence, h(B) = B and h ∈ G{B} .
(5) ⇒ (6) Clearly, G(B) is an open subgroup of H and has finite index in G{B} .
(6) ⇒ (3). The inclusion ⊆ is clear. The group K := hAut(A)(B1 ) ∪ Aut(A)(B2 ) i
is open in Aut(A) (Lemma 4.4.1). By assumption, there exists a finite C ⊆ A which
is algebraically closed in A such that Aut(A)(C) ≤ K and the index of Aut(A)(C) in
K is finite. We will prove that C ⊆ B1 ∩ B2 . First note that every orbit of c ∈ C in
K is finite by the assumption that the index of Aut(A)(C) in K is finite. However,
the orbit of every x ∈ A \ B1 in Aut(A)B1 is infinite since B1 is algebraically closed.
Therefore, C ⊆ B1 . Similarly, C ⊆ B2 . Thus, Aut(A)B1 ∩B2 ≤ Aut(A)C ≤ K =
hA(B1 ) ∪ Aut(B2 ) i. 
Lemma 6.6.5. Let A be a countable ω-categorical structure such that Aut(A) has
the SIP. Then A has weak elimination of imaginaries if and only if Aut(A) has the
SSIP.
Proof. Let H ≤ Aut(A) be of countable index. By the SIP, H is open. The
equivalence then follows from the equivalence (1) ⇔ (5) in Lemma 6.6.4. 
We close this section by verifying weak elimination of imaginaries for a large class
of homogeneous structures A, namely for those whose age has the free amalgamation
property. In the next proof, if B 1 , B 2 , C are finite substructures of A, then we write
B1 ↓C B2 if A[B1 ∪ B2 ] equals B 1 ∪ B 2 and B1 ∩ B2 ⊆ C.
Theorem 6.6.6 (Lemma 2.7 in [111], Corollary 1 in [129]). Let A be a homo-
geneous ω-categorical structure whose age has the free amalgamation property. Then
Aut(A) has the strong small index property.
Proof. By Theorem 6.5.5, G := Aut(A) has the small index property. To verify
weak elimination of imaginaries, we verify (2) in Lemma 6.6.4. Let B1 , B2 ⊆ A
112 6. RECONSTRUCTION OF TOPOLOGY AND AUTOMATIC CONTINUITY

be finite, and suppose that R ⊆ Ak is definable in A over B1 and over B2 . Since

A has no algebraicity, we have to show that R is also definable over B1 ∩ B2 . By
Theorem 3.1.1, it suffices to show that R is preserved by G(B1 ∩B2 ) , so let g ∈ G(B1 ∩B2 ) .
There exists h1 ∈ G(B1 ) with h1 g(B1 ) ↓B1 B2 (see Exercise 75). There is h2 ∈ G(B2 )
with h2 h1 g(B1 ) ↓B2 B1 , and h3 ∈ G(B1 ) so that h3 (B1 ) ↓B2 B1 . By homogeneity,
there is h4 ∈ G(B1 ) so that h4 h2 h1 g|B1 = h3 |B1 . Thus, h−1 3 h4 h2 h1 g ∈ G(B1 ) , so
g ∈ hG(B1 ) ∪ G(B2 ) i. 
6.6.2. G-finiteness. This section is under construction.
Definition 6.6.7 (Lascar [103]). An oligomorphic group H ≤ Sym(N) is called
G-finite if for every open subgroup U ≤ H, the intersection of the open subgroups of
finite index in U is of finite index in U .
Example 93. Let A be the structure from Example 67. Then Aut(A) is not
G-finite: Aut(A) itself has the stabilisers of the equivalence classes of Ei , for i ∈ N,
as open subgroups of index two (see Lemma 4.4.3). Their intersection has infinite
index. 4

Exercises.
(133) Let A be a countable ω-categorical structure. Show that A has no algebraic-
ity and weak elimination of imaginaries if and only if Aut(A) has no fixed
points and for all finite B1 , B2 ⊆ A we have that hGB1 ∪ GB2 i = GB1 ∩B2 .
(134) If G is an oligomorphic permutation group on a set B, then let’s write G∗ for
the closed normal subgroup of G consisting of all α ∈ G that fix all blocks
of congruences of the action of G on B n , for some n ∈ N, that have finitely
many classes. Show that an oligomorphic permutation group on a set B is
G-finite if and only if for every finite A ⊆ B, the index of (GA )∗ in GA is
finite.
(135) Show that there are uncountably many closed oligomorphic
permutation groups, up to isomorphism of abstract groups.

Hint. Show that the automorphism groups of the countable homogeneous

digraphs from Example 31 are pairwise non-isomorphic as permutation groups,
then as topological groups, and finally as abstract groups. Use statements
of this text (even the ones that have not been proven here but just cite the
literature).
(136) Is there a theorem for extending partial homomorphisms between finite
graphs to endomorphisms of supergraphs (similarly as Theorem 6.5.2 for
partial isomorphisms and automorphisms)?
(137) Show that if A is a structure with a first-order interpretation
in (N; =), and B ⊆ A is finite, then G(acl(B)) is open in Aut(A).

(138) Show that every structure with a first-order interpretation

in (N; =) has an automorphism group which is G-finite.

6.7. Open Problems

We list open problems from the literature that fall into the context of this chapter.
(1) Does the automorphism group of the countable universal homogeneous tour-
nament (T; E) have the small index property (see [110])?
(2) Does the class of all finite tournaments have the EPPA (see [69])?
 6.7. OPEN PROBLEMS 113

Structure Reconstruction SIP Ample Generics EPPA

(N; 6=) Yes Yes Yes Yes
Rado Yes Yes Yes [74] Yes [77]
K3 -free Yes Yes Yes Yes [69]
Henson Yes Yes Yes Yes [69]
(T; E) Yes [139] Open [110] Open Open [69]
(Q; <) Yes Yes [158] No [160] No
(P; ≤) Yes [139] Open [110] No [98] No
(Q; <1 , <2 ) ? Open [110] No No
Example 67 Yes [110, 139] No No No
Evans-Hewitt No [54] No No No

Figure 6.2. Some open problems in the context of this chapter; see
Section 6.7. If there is no reference then the result is trivial or can
be deduced from other entries in the table and/or the results of this
chapter.

(3) Does the automorphism group of the countable universal homogeneous poset
(P; ≤) have the small index property (see [110])?
(4) Does the automorphism group of the countable universal homogeneous per-
mutation (Q; <1 , <2 ) have the small index property (see [110])?
(5) Is there for every n ∈ N a countably infinite structure which has n-generic
automorphisms, but not (n + 1)-generic automorphisms (see [146])? We
only know that the answer is positive for n = 1 (Example 87).
(6) Is the automorphism group of every homogeneous structure with a finite
relational language G-finite (Macpherson [110])?
 CHAPTER 7

Ramsey Classes and Topological Dynamics

This section is under construction. [53, 78, 79, 89, 90, 102, 119, 121–126, 148,
149, 153–155, 163, 164]

7.1. Ramsey Classes

Conjecture 7.1 (see, e.g., [15]). Every homogeneous structure with a finite
relational signature has a finite homogeneous Ramsey expansion.

7.2. The Kechris-Pestov-Todorcevic Connection

In topological dynamics one studies continuous actions of topological groups G
on compact Hausdorff spaces X. Such an action is called a G-flow, and X is called a
G-space (if the reference to the underlying G-flow is clear). A point x ∈ X such that
g · x = x for all g ∈ G is called a fixed point of the G-flow.
By the Kechris-Pestov-Todorcevic correspondence [89], a homogeneous structure
B is Ramsey (with respect to colorings of embeddings) if and only if its automorphism
group G := Aut(B) is extremely amenable, meaning that every G-flow has a fixed
point.
7.2.1. Universal Minimal Flows. Let X be a G-space and suppose that Y ⊆
X is preserved by the action of G on X. Then Y naturally gives rise to a G-flow by
restricting the action to Y , and Y is then called a G-subspace of X. A G-space is
called minimal if X and ∅ are the only closed G-subspaces of X.
Lemma 7.2.1. Every G-space contains a minimal G-subspace.
Proof. Apply Zorn’s Lemma: TODO. 
A function f : X → Y between a G-space X and a G-space Y is called equivariant
if f (g · x) = g · f (x) for all g ∈ G and x ∈ X. If f is bijective, then the inverse is
also equivariant, and f is called an isomorphism, and the G-spaces X and Y are
called isomorphic. A factor of a G-space X is a G-space Y such that there exists a
continuous G-equivariant surjective map from X to Y .
Definition 7.2.2. A G-space is called universal if every minimal G-space X is
a factor of UG .
The following was shown by Ellis, with new proofs by Auslander, Uspenskij, and
Gutman and Li.
Theorem 7.2.3. For every G-space there exists a universal minimal G-space,
which is unique up to isomorphism.
Proof. For the existence, let S be a set of minimal Q G-spaces such that every
minimal G-space is isomorphic to an element of S. Then X∈S X (with the diagonal
action) is a universal G-space:
For the uniqueness, we follow a proof of Gutman and Li. They show that a
universal minimal G-space is coalescent, i.e., every surjective 
115
116 7. RAMSEY CLASSES AND TOPOLOGICAL DYNAMICS

An active field of research studies the question for which Polish groups G the
universal minimal flow M (G) is metrizable. If M (G) is metrizable then it has a
comeagre orbit (Ben Yaacov, Melleray, Tsankov 2017): TODO.

7.3. Compact Spaces for Oligomorphic Groups

Definition 7.3.1. Let G be a topological group with a continuous action on a
topological space A. Let ∼ be the orbit equivalence relation on A where a ∼ b if there
exists α ∈ G such that a = αb. We write A/G for the quotient space A/∼ with the
quotient topology.
The following statement is taken from [26].
Proposition 7.3.2. Let A, B be countably infinite sets and let G be a permutation
group on B. Equip B with the discrete topology and B A with the product topology.
Then B A /G is compact if and only if G is oligomorphic.
Proof. Suppose that A = N. We first prove that if G is oligomorphic, then
B A /G is compact. Let U := {Ui | i ∈ I} be a family of open subsets of B A /G
such that no finite subset of U covers B A /G. For n ∈ N, let ∼n be the equivalence
relation on B A where f ∼n g if there exists an α ∈ G such that f (a) = αg(a) for all
a ∈ {0, . . . , n − 1}. Note that each equivalence class of ∼n is a union of elements of
B A /G, and that the oligomorphicity of G implies that ∼n has finitely many classes
for each n ∈ N. If each of the finitely many equivalence classes of ∼n were contained
in the complement of Ui for some i ∈ I, then we would have found a finite subset
of U that covers B A /G, contrary to our assumptions. S So for each n there exists a
∼n -equivalence class which is not contained in i∈I Ui .
Consider the following tree: the vertices of S the tree are the equivalence classes
of ∼n , for all n ∈ N, that are not contained i∈I Ui . Let the equivalence class of
f : {0, . . . , n − 1} → N be adjacent to the equivalence class of g : {0, . . . , n} → N
if f is the restriction of g. Clearly, the resulting tree is finitely branching and by
KHonigs tree lemma contains an infinite path. From this infinite path F1 , F2 , . . . one
can construct a function f ∈ B A inductively as follows. Initially, pick any function
f1 from F1 . By the definition of edges in the tree there exists an α ∈ G such that αf1
is the restriction of some g2 ∈ F2 . We define f2 to be α−1 g2 which is an extension
of f1 and in F2 . We continue with f2 instead of f1 , and iterate to obtain an infinite
sequence of functions f1S , f2 , . . . which converges against some f ∈ B A . Note that
f /∼ is not contained in i∈I Ui which finishes the proof that B A /G is compact.
For the other direction, assume that G is not oligomorphic. Pick an n ≥ 1 such
that the componentwise action of G on B n has infinitely orbits, and enumerate these
orbits by (Oi )i∈ω . For each i ∈ ω let Ui consist of all classes f /∼ in B A /G with the
property that f |{1,...,n} belongs to Oi ; this is well defined since for all f, g ∈ B A with
f ∼ g we have that f |{1,...,n} belongs to Oi if and only if g|{1,...,n} belongs to Oi .
Then B A /G is the disjoint union of the Ui . But each Ui is open, and hence B A /G is
not compact. 

Corollary 7.3.3. Let B be an ω-categorical structure. Then End(B)/ Aut(B)

is compact.
Proof. End(B) is a closed subset of B B which is preserved by Aut(B). Since
B is ω-categorical, Aut(B) is an oligomorphic permutation group by the theorem
of Engeler, Svenonius, and Ryll-Nardzewski (Theorem 3.2.3). Proposition 7.3.2 im-
plies that B B / Aut(B) is compact. Note that End(B)/G is a closed subspace of
B B / Aut(B), so the statement follows from Proposition 4.1.13 
 7.4. CANONICAL FUNCTIONS 117

If G is an oligomorphic permutation group on a countable set B, then the space

B B /G is not Hausdorff, as the following example shows.
Remark 7.3.4. Consider any function f in B B which lies in the closure of G but
not in G; Exercise 139 shows that if G is oligomorphic, such functions must exist.
Then f is inequivalent to every element of G, but f /∼ cannot be separated S
from idB /∼
by open sets: if U is an open subset of B B /G that contains f /∼ , then U is open
in B B and hence must contain a basic open set Ta,b where a, b ∈ B n for some n ∈ N
and f (a) = b. Since f is in the closure of G there also exists an α ∈ G with αa = b,
and α ∼ idB . So every open set that contains f /∼ also contains idY /∼ .
We will work with a certain compact Hausdorff space.
Definition 7.3.5. Let G y B be a permutation group, and A be a set. On B A ,
define an equivalence relation ≈ by setting f ≈ g if f ∈ G g. We also write B A G
instead of B A /≈ .
Note that in Definition 7.3.5, transitivity and symmetry follow from the fact that
G is a group. Also note that f ≈ g if and only if f = g holds locally modulo G.
Lemma 7.3.6. If G y B is oligomorphic, then the space B A G is a compact
Hausdorff space.
Proof. Since B A G is a quotient of B A /G , and since B A /G is compact (Propo-
sition 7.3.2), the compactness of B A G follows from Proposition 4.1.13. To prove
that B A G is Hausdorff, let s1 /≈ and s2 /≈ be elements of B A G. If these two
elements are distinct, there exists t ∈ An such that s1 (t), s2 (t) ∈ B n lie in differ-
ent orbits of n-tuples under G. Then s1 ∈ U1 := {u ∈ B A | u(t) = s1 (t)} and
s2 ∈ U2 := {u ∈ B A | u(t) = s2 (t)}, and U1 and U2 are open and disjoint. 

Exercises.
(139) Show that if G is an oligomorphic permutation group on a set A,
then the closure of G in AA contains some non-surjective maps
(this is related to Exercise 106).

7.4. Canonical Functions

The material in this section stems from [27]. If f : Q → Q is any function from the
order of the rational numbers to itself, then there are arbitrarily large finite subsets
of Q on which f “behaves regularly”; that is, it is either strictly increasing, strictly
decreasing, or constant. A direct (although arguably unnecessarily elaborate) way to
see this is by applying Ramsey’s theorem (see Section 2.2.3): two-element subsets of
Q are coloured with three colours according to the local behaviour of f on them. In
particular, it follows that
{β f α | α, β ∈ Aut(Q; <)} ⊆ QQ
equipped with the pointwise convergence topology, contains a function which behaves
regularly everywhere. This function of regular behavior is called canonical.
More generally, a function f : A → B between two structures A, B is called canon-
ical when it behaves regularly in an analogous way, that is, when it sends orbits of
n-tuples of Aut(A) to orbits of n-tuples of Aut(B). Similarly as in the example above,
canonical functions can be obtained from f , in the fashion stated above, when A has
sufficient Ramsey-theoretic properties (for example, the Ramsey property) and when
Aut(B) is sufficiently rich (for example ω-categorical) [23, 24, 30].
118 7. RAMSEY CLASSES AND TOPOLOGICAL DYNAMICS

Remark 7.4.1. The concept of canonical functions has turned out useful in nu-
merous applications: for classifying first-order reducts they are used in [1, 2, 19,
28, 107, 128, 133], for complexity classification for constraint satisfaction problems
(CSPs) in [20, 21, 25, 31, 96], for decidability of meta-problems in the context of the
CSPs in [30], for lifting algorithmic results from finite-domain CSPs to CSPs over
infinite domains in [22], for lifting algorithmic results from finite-domain CSPs to
homomorphism problems from definable infinite structures to finite structures [94],
and for decidability questions in computations with atoms in [95].
As indicated above, the technique is available for a function f : A → B whenever A
is a Ramsey structure and B is ω-categorical, and the existence of canonical functions
in the set
{β f α | α ∈ Aut(A), β ∈ Aut(B)} ⊆ B A
was originally shown under these conditions by a combinatorial argument [23,24,30].
It is natural to ask for a perhaps more elegant proof of the existence of canonical func-
tions via topological dynamics, reminiscent of the numerous proofs of combinatorial
statements obtained in this fashion (cf. the survey [11] for ergodic Ramsey theory; [91]
mentions some applications of extreme amenability). In this section we present such
a proof, taken from [27].
7.4.1. Canonicity. Let G ≤ Sym(A) and H ≤ Sym(B). A function f : A → B
is called canonical with respect to (G, H) if for every k ≥ 1, t ∈ Ak , and α ∈ G
there exists β ∈ H such that f α(t) = β f (t). Hence, functions that are canonical
with respect to (G, H) induce for each integer k ≥ 1 a function from the orbits of
the componentwise action of G of Ak to the orbits of the componentwise action of
H on B k . For oligomorphic permutation groups we have the following equivalent
characterisations of canonicity.
Proposition 7.4.2. Let G y A and H y B be permutation groups, where H y
B is oligomorphic. Then for any function f : A → B the following are equivalent.
(1) f is canonical with respect to (G, H);
(2) for every α ∈ G we have f α ∈ Hf := {βf | β ∈ H};
(3) for every α ∈ G there are e1 , e2 ∈ H such that e1 f α = e2 f .
A stronger condition is to require that for all α ∈ G there is an e ∈ H such that
f α = ef . To illustrate that this is strictly stronger, already if G = H, we give an
explicit example.
Example 94 (Trung Van Pham). Let G := Aut(Q; <). Note that (Q; <) and
(Q \ {0}; <) are isomorphic, and let f be such an isomorphism. Then f , viewed as
a function from Q → Q, is clearly canonical with respect to (G, G). But f does not
satisfy the stronger condition above: there is no e ∈ G such that f α = ef . To see
this, choose b, c ∈ Q such that f (b) < 0 < f (c). By transitivity there exists an α ∈ G
such that α(b) = c. Note that 0 < f α(b) < f α(c). Morever, the image of f α equals
the image of f , and hence any e ∈ G such that f α = ef must fix 0. Since e must also
preserve <, it cannot map f (b) < 0 to f α(b) > 0. Hence, there is no e ∈ G such that
f α = ef . 4
In Proposition 7.4.2, the implications from (1) to (2) and from (3) to (1) follow
straightforwardly from the definitions. For the implication from (2) to (3) we need
a lift lemma, which is in essence from [29]. This lemma has been applied frequently
lately [8, 20, 22], in various slightly different forms.
Let H y B be a permutation group, and let f, g ∈ B A , for some A. We say that
f = g holds locally modulo H if for all finite F ⊆ A there exist β1 , β2 ∈ H such that
 7.5. MODEL-COMPLETE CORES OF RAMSEY STRUCTURES 119

β1 f |F = β2 g|F . We say that f = g holds globally modulo H (modulo H) if there

exist e1 , e2 ∈ H (e1 , e2 ∈ H, respectively) such that e1 f = e2 g. Of course, if f = g
holds globally modulo H, then it holds locally modulo H. On the other hand, if
f = g holds locally modulo H, then it need not hold globally modulo H. To see this,
let f (x, y) : ω 2 → ω be an injection, set g := f (y, x), and let H be the group of all
permutations of ω. Then f = g holds locally modulo H, but not globally. However,
there exist injections e1 , e2 ∈ ω ω such that e1 f = e2 g, so f = g holds globally modulo
H. This is true in general, as we see in the following lift lemma.
Lemma 7.4.3. Let H y B be an oligomorphic permutation group, let I be an
index set, and let Ai be a set for every i ∈ I. Let fi , gi be functions in B Ai such that
fi = gi holds locally modulo H for all i ∈ I. Then there exist e, ei ∈ H such that
e fi = ei gi holds globally for all i ∈ I.
Proof. For simplicity of notation, assume that the Ai are countable; then B Ai
is a metric space (otherwise, we would have to work with more general topological
notions than sequences). We have fi ∈ H gi ; so let (βij gi )j∈ω be a sequence converging
to fi for all i ∈ I. Setting A := B we see that AA G is compact by Lemma 7.3.6.
Therefore, the set
{([δ]≈ , ([δ βij ]≈ )i∈I ) | j ∈ ω, δ ∈ H}
is a subset of a compact space, (AA G) × (AA G)I . Hence, it has an accumulation
point ([e]≈ , ([ei ]≈ )i∈I ). Clearly, e, ei ∈ H for all i ∈ I, and the functions ei prove the
lemma. 
The implication from (2) to (3) in Proposition 7.4.2 now is a direct consequence
of Lemma 7.4.3.
7.4.2. Canonisation. The following is the canonisation theorem, first proved
combinatorially in [30] in a slightly more specialized context.
Theorem 7.4.4. Let G y A, H y B be permutation groups where G is extremely
amenable and H is oligomorphic, and let f : A → B. Then
H f G := {β f α | α ∈ G, β ∈ H}
contains a canonical function with respect to (G, H).
Proof. The space H f G/∼ is a closed subspace of the compact Hausdorff space
Y X G from Lemma 7.3.6, and hence is a compact Hausdorff space as well. We define
a continuous action of G on this space by
(α, [g]≈ ) 7→ [g α−1 ]≈ .
Clearly, this assignment is a function, it is a group action, and it is continuous. Since
G is extremely amenable, the action has a fixed point [g]≈ . Any member g of this
fixed point is canonical: whenever α ∈ G, then [g α]≈ = [g]≈ , which is the definition
of canonicity. 

7.5. Model-Complete Cores of Ramsey Structures

Model companions and model-complete cores are a powerful method to construct
new structures from known ones. They inherit quite a number of important prop-
erties from the structures we start from. We will see that for first-order reducts of
homogeneous Ramsey structures the theory of model companions and model complete
cores is particularly well behaved (Section 7.5.3). We present the theory for model
complete cores; the theory for model companions can be seen as a special case, as we
will see Remark 7.5.3.
120 7. RAMSEY CLASSES AND TOPOLOGICAL DYNAMICS

7.5.1. Model Complete Cores. The results from this section are from [14];
we follow the presentation in [16]. An ω-categorical structure C is called
• model complete if every embedding from C into C preserves all first-order
formulas.
• a core if every endomorphism of C is an embedding.
Proposition 7.5.1 (see [16]). Let C be an ω-categorical structure. Then the
following are equivalent.
(1) C is a model-complete core.
(2) Every endomorphism of C preserves all first-order formulas.
(3) For every n ∈ N, the orbits of n-tuples of Aut(C) are primitively positively
definable in C.
(4) For every e ∈ End(B), n ∈ N, and a ∈ C n there exists i ∈ End(C) such that
i(e(a)) = a.
Theorem 7.5.2 (from [14]; see [16]). For every ω-categorical structure B there
exists an model-complete core structure C which is homomorphically equivalent to B;
the structure C is unique up to isomorphism, and ω-categorical. We may assume that
C is an induced substructure of B.
Remark 7.5.3. Saracino’s theorem.
7.5.2. Range Rigid Functions. The results in this section are from [120]. Let
G be a permutation group on a set X. A function g : X → X is called range-rigid
with respect to G if for all β ∈ G we have
g ∈ {α ◦ g ◦ β ◦ g | α ∈ G}.
In particular, the identity map is range-rigid. Note that a function g : X → X
is range-rigid with respect to G if and only if for all t ∈ X n , n ∈ N, if there exists
s ∈ X n and α ∈ G such that αg(s) = t, then t and g(t) lie in the same orbit. In
other words, g preserves the orbits of n-tuples that have non-empty intersection with
g(X)n .
Example 95. Let U, V be unary relation symbols, and let B be a countably
infinite {U, V }-structure where U = {u}, V = {v}, and u 6= v. Let w ∈ B \ {u, v}.
Then the map g : B → B with g(w) = g(u) = u and g(x) = v for all x ∈ B \ {u, w}
is range-rigid with respect to Aut(B), but not canonical with respect to B. The
identity map is canonical with respect to Aut(B) but not range-rigid with respect to
Aut(B). 4
We can use canonisation to find range-rigid functions in sufficiently rich sets of
operations.
Theorem 7.5.4. Let G be a closed oligomorphic extremely amenable permutation
group on a countable set X and let M be a non-empty closed transformation semigroup
on X such that G ◦ M ◦ G ⊆ M . Then M contains a function which is range-rigid
with respect to G and canonical with respect to G.
Proof. Pick any f ∈ M . Applying Proposition 7.4.2 to f , we obtain a function
f 0 ∈ {αf β | α, β ∈ G} ⊆ M which is canonical with respect to G. Since G is oligo-
morphic, for every n ∈ N there are finitely many orbits of n-tuples for every n, so we
may compose f 0 with itself sufficiently many times to obtain a function such that for
all t ∈ X n whose orbit contain a tuple of the form g(s) for s ∈ X n we have that g(t)
lies in the same orbit as t. A standard compactness argument shows that there is one
function that does it for all n; note that the resulting map is range-rigid with respect
to G and canonical with respect to G. 
 7.5. MODEL-COMPLETE CORES OF RAMSEY STRUCTURES 121

Lemma 7.5.5. Let A be a homogeneous structure and let g : A → A be range-rigid

with respect to Aut(A). Then the age C of the substructure induced by the image of g
in A has the amalgamation property.
Proof. Let B 1 , B 2 ∈ C. We claim that the restriction gi of g to Bi , for i ∈ {1, 2},
shows that A[g(B1 ∪ B2 )] ∈ C is an amalgam of B 1 and B 2 . It suffices to show that
the restriction gi of g to Bi is an embedding. By the definition of C there exists a
finite set Bi0 ⊆ A such that Bi = g(Bi0 ). Since g is range-rigid, we have in particular
that g ∈ {α ◦ g ◦ g | α ∈ Aut(A)}, and hence there exists an αi ∈ Aut(A) such that
g ◦ g(a) = αi ◦ g(a) for all a ∈ Bi0 . Hence, for all b ∈ Bi we have gi (b) = αi (b), which
shows that gi is an embedding of B i into C. 

Definition 7.5.6. Let A be a homogeneous structure and let g : A → A be range-

rigid with respect to Aut(A). We denote by Ag the Fraı̈ssé-limit of the age of the
substructure induced by g(A) in A (which has the amalgamation property by Theo-
rem 3.3.5). By the homogeneity of A, we may assume that Ag is an induced substruc-
ture of A.
Lemma 7.5.7. Let A be a homogeneous structure and let g : A → A be range-rigid
with respect to Aut(A). If A is ω-categorical, then Ag is ω-categorical as well.
Proof. If A is ω-categorical, then A has for every n only finitely many inequiv-
alent atomic formulas with n variables. Since the age of Ag is contained in the age of
A, the same is true for Ag , and the statement thus follows from the homogeneity of
Ag . 

Lemma 7.5.8. Let A be a homogeneous τ -structure and let g : A → A be range-

rigid with respect to Aut(A). If A is finitely bounded, then so is Ag .
Proof. Suppose that F is a finite set of finite τ -structures such that Forb(F) =
Age(A). Let m ≥ 1 be the maximum arity of the relations of B. Let F 0 be the union
of F with the set of all structures on the set {1, . . . , m} that do not embed into Ag .
Clearly, Age(Ag ) ⊆ Forb(F 0 ). Let D ∈ Forb(F 0 ); we have to show that D ∈ Age(Ag ).
Since Forb(F 0 ) ⊆ Forb(F) = Age(A) we may assume that D is a substructure of A and
claim that the restriction of g to D is an embedding. Let t ∈ Dm ; since D ⊆ Forb(F 0 )
we have that the substructure induced by {t1 , . . . , tm } in D is in Age(Ag ). By the
homogeneity of A there exists β ∈ Aut(A) and s ∈ Am such that t = βg(s). Since g is
range rigid, there exists α ∈ Aut(A) such that αgβg(s) = g(s), and hence t = βg(s)
and g(t) = g(βg(s)) = α−1 g(s) = α−1 β −1 t lie in the same orbit of Aut(A), which
proves the claim. 

Lemma 7.5.9. Let A be a homogeneous structure and let g : A → A be range-rigid

with respect to Aut(A). If A is Ramsey, then so is Ag .
A

Proof. Let S and M be finite substructures of Ag and let χ : Sg → {0, 1} be a
colouring of the copies of S in Ag with the colours 0 and 1. Let B be the substructure
of A induced by g(A). By the homogeneity of Ab and since Age(B) = Age(A b ) there
exists an embedding f : B → Ab . We then define a colouring χ0 of A


→ {0, 1} by
0 0 A

S
χ (S ) := χ(Ag [f ◦ g(S )]). Since A is Ramsey, there exists M ∈ M such that χ0 is
0 0
0

constant c ∈ {0, 1} on MS .
Claim 1. M 00 := B g [f ◦ g(M 0 )] is isomorphic to M . First note that M embeds
into B because Ag and B have the same age; since A is homogeneous, there exists
α ∈ Aut(A) which maps this copy of M in B = A[g(B)] to M 0 . By the range rigidity
122 7. RAMSEY CLASSES AND TOPOLOGICAL DYNAMICS

of g, there exists β ∈ Aut(A) such that β ◦ g(M 0 ) = α(M 0 ); hence, the restriction of
g to M 0 is an embedding, and the claim follows.
00
00

Claim 2. χ is constant on MS . If S 00 ∈ MS , then χ(S 00 ) = χ0 (B[g −1 ◦
f −1 (S 00 )]) = c since g −1 ◦ f −1 (S 00 ) is a copy of S in M 0 . 
7.5.3. Range-Rigidity and Model-Complete Cores. A subset S of a monoid
M is called a left ideal of M if M S = S. Note that for every f ∈ M , then set T := M f
is a (closed) left ideal, because M T = M M f ⊆ M f = T .
Lemma 7.5.10. Let M be a monoid and f ∈ M . Then the following are equivalent.
(1) f lies in an inclusion-wise minimal closed left-ideal of M ;
(2) M f is an inclusion-wise minimal closed left-ideal;
(3) f ∈ M ef for every e ∈ M .
Proof. (1) implies (2). Suppose that f ∈ S for some inclusion-wise minimal
closed left-ideal S of M . Then T := M f ⊆ M S = S = S, so T = S by the minimality
of S, and hence T is an inclusion-wise minimal closed left-ideal S of M .
(2) implies (1). We have f ∈ M f since M is a monoid and thus contains 1.
(2) implies (3). Let e ∈ M . Then M ef ⊆ M f is closed, non-empty, and a left-
ideal since M M ef = M ef . By the minimality of M f we have that M ef = M f , so
f ∈ M ef since 1 ∈ M .
(3) implies (2). It suffices to show minimality of T := M f . Suppose that there
exists a non-empty closed left-ideal I of M contained in M f . Let t ∈ I. Then t ∈ M f
so there are r1 , r2 , · · · ∈ M such that limi∈N ri f = t. By assumption, for every i ∈ N
we have that f ∈ M ri f , so there exist si,1 , si,2 , · · · ∈ M such that limj∈N si,j ri f = f .
We claim that limj∈N sij t = f . Indeed,
lim sij t = lim sij lim ri f
j∈N j∈N i∈N
= lim lim sij ri f = f.
i∈N j∈N

Hence, f ∈ M I = I. Therefore, I = M I ⊇ M f ⊇ I, showing that M f is a minimal

closed left-ideal of M . 
The following lemma has been shown in [7] (Lemma 5.4); the proof uses the
equivalence relation ≈ from Definition 7.3.5.
Lemma 7.5.11. Let X be countable and let M be a closed transformation monoid
on X that contains a closed oligomorphic permutation group G. Then M has a
inclusion-wise minimal non-empty closed left-ideal S.
Proof. Let T be the set of all non-empty topologically closed subsets T of
M G (Definition 7.3.5) with the property that whenever [f ]≈ ∈ T and m ∈ M , then
[m◦f ]≈ ∈ T . Since M G is compact (Lemma 7.3.6), arbitrary descending chains in T
have a non-empty intersection in T (the proof of the implications (1) ⇒ (2) ⇒ (3) for
Theorem 4.1.19 works in arbitrary topological spaces). Hence, T contains a minimal
element T0 by Zorn’s Lemma. Then S0 := {f ∈ M | [f ]≈ ∈ T0 } is closed and
non-empty, and M ◦ S0 = S0 , and minimal with these properties. 
Lemma 7.5.12. Let B be a first-order reduct of a homogeneous structure A. Sup-
pose that
• S is a minimal non-empty closed left-ideal of End(B), and
• there exists g ∈ S which is range-rigid with respect to Aut(A).
Then the substructure C of B with domain Ag is the model-complete core of B.
 7.5. MODEL-COMPLETE CORES OF RAMSEY STRUCTURES 123

Proof. Since A[g(A)] and Ag have the same age and Ag is homogeneous, there
exists an embedding f : A[g(A)] → Ag . Then f ◦ g is a homomorphism of B into C,
because every relation of B has a quantifier-free definition in A, and the same formula
defines the corresponding relations of C in Ag . Since C is even a substructure of B,
they are homomorphically equivalent. So it suffices to prove that C is a model-
complete core. Let e ∈ End(C), n ∈ N, and t ∈ (Ag )n .
Claim 1. We may choose f so that there exists s ∈ An with (f ◦ g 2 )(s) = t.
Indeed, there exists s ∈ An such that g(s) satisfies the same atomic formulas as
t in A, because A[g(A)] and Ag have the same age. Since g is range-rigid with
respect to Aut(A), there exists an automorphism of A that maps g(g(s)) to g(s). The
homogeneity of Ag implies that there exists α ∈ Aut(Ag ) which maps (f ◦ g 2 )(s) to t.
Therefore, α ◦ f is an embedding of A[g(A)] into Ag which has the required property.
Claim 2. There exists h ∈ End(B) such that h◦(e◦f ◦g)(g(s)) = g(s). Otherwise,
S 0 := End(B) ◦ {e ◦ f ◦ g 2 } ⊆ S does not contain g. Since End(B) ◦ S 0 = S 0 , this
contradicts the minimality of S. Then
t = (f ◦ g 2 )(s) = (f ◦ g) ◦ h ◦ (e ◦ f ◦ g)g(s) (Claim 2)
= (f ◦ g ◦ h) ◦ e(t).
The restriction of f ◦ g ◦ g to Ag therefore proves condition (2) of Proposition 7.5.1
for C, and hence C is a model-complete core. 

Lemma 7.5.13. Let X be countable and let G ≤ Sym(X) be closed, oligomorphic,

and extremely amenable. Let M be a closed transformation monoid containing G.
Then M contains a function which is range-rigid and canonical with respect to G and
is contained in a minimal closed left-ideal S of M .
Proof. By Lemma 7.5.11 M contains a minimal non-empty closed left-ideal S0 .
Let S1 be the smallest closed transformation semigroup which contains S0 such that
G ◦ S1 ◦ G = S1 . By Theorem 7.5.4, S1 contains a function g which is range-rigid and
canonical with respect to G.
If thus suffices to show that every element of S1 lies in a minimal closed left-
ideal of M . This is clear for every element of S0 ⊆ S1 . Now suppose that f lies
in a minimal closed left-ideal of M , and let h ∈ M . We claim that hf and f h are
contained in minimal closed left-ideals of M . This then implies that every element of
S1 is contained in a minimal closed left-ideal of M .
By Lemma 7.5.10 (2), we have that M f is a minimal closed left-ideal. Note that
M f = M hf and hence hf lies in a minimal closed left-ideal by Lemma 7.5.10 (2).
To show that f h lies in a minimal closed left-ideal, we use Lemma 7.5.10 (3). Let
e ∈ M . Lemma 7.5.10 (3) implies that f ∈ M ef . Thus, f h ∈ M ef h = M e(f h), so
Lemma 7.5.10 (3) implies that f h indeed lies in a minimal closed left-ideal of M . 

The following theorem summarises many of the previous statements.

Theorem 7.5.14. Let B be a first-order reduct of an ω-categorical homogeneous
Ramsey structure A and let C be the model-complete core of B. Then
(1) B has an endomorphism g which is range-rigid with respect to Aut(A) and
canonical with respect to Aut(A).
(2) Ag is ω-categorical and Ramsey.
(3) If A is finitely bounded, then Ag is finitely bounded as well.
(4) The substructure of B with domain Ag is isomorphic to C; we identify C
with Ag along this isomorphism. Then C is a first-order reduct of Ag .
124 7. RAMSEY CLASSES AND TOPOLOGICAL DYNAMICS

Proof. By Lemma 7.5.13 applied to M = End(B) and G = Aut(A), we obtain

a function g ∈ End(B) which is range-rigid with respect to Aut(A) and canonical
with respect to Aut(A), which shows (1), and additionally contained in a minimal
closed left-ideal of End(B). By Lemma 7.5.7, Ag is ω-categorical, and it is Ramsey
by Lemma 7.5.9, which shows (2). Lemma 7.5.8 implies item (3).
By Lemma 7.5.12, the substructure of B with domain Ag is isomorphic to the
model-complete core C of B. Since A is homogeneous, all relations of B have a
quantifier-free definition in A, and hence the same formulas define the relations of C
in Ag ; this completes the proof of item (4). 
We present an immediate consequence of this theorem (whose statement does not
involve the concept of range-rigidity).
Corollary 7.5.15. Let B be a first-order reduct of an ω-categorical homogeneous
Ramsey structure A and let C be the model-complete core of B. Then C has an ω-
categorical homogeneous Ramsey expansion A0 . If A has a finite signature, then so
has A0 . If A is finitely bounded, then so is A0 .
 CHAPTER 8

Reducts and Closed Supergroups

[2, 10, 32, 33, 37, 87, 107, 128, 133, 134, 156, 157]
Conjecture 8.1 (Thomas [156]). Let A be a homogeneous structure with a fi-
nite relational signature. Then Aut(A) has only finitely many closed supergroups in
Sym(A). Equivalently, A has only finitely many first-order reducts up to interdefin-
ability.
Cameron’s theorem for highly set-transitive permutation groups on a countably
infinite set, Thomas’ result about the closed supergroups of the automorphism group
of the random graph.
Exercises.
(140) Let (V ; T ) be the Fraı̈ssé-limit of the class of all finite tournaments (see
Exercise 57). Show that
Aut(V ; {(x, y, u, v) | T (x, y) ⇔ T (u, v)})
is isomorphic to a semidirect product of Z2 and Aut(V ; T ).

(141) Let (V ; E) be the Rado graph (Example 29), and let R ⊆ V 4 be

the relation {(a, b, c, d) | E(a, b) ⇔ E(c, d)}.
Show that N := Aut(V ; E) is a closed normal subgroup
of G := Aut(V ; R) of index 2, but that G is not isomorphic
to a semidirect product of N and Z2 (see Proposition 1.5.4).

125
 CHAPTER 9

Restricted Orbit Growth

[36, 38, 41, 55, 112–115, 135, 136, 144]

127
 CHAPTER 10

Homogeneous Structures in Restricted Signatures

[4, 42, 44–46, 65, 85, 99–101, 105, 108, 140]

Exercises.
(142) Show that up to isomorphism, there is only one countable
homogeneous linear order.

Theorem 10.0.1 (Woodrow). Up to isomorphism, there are only two countable

homogeneous tournaments that do not embed the 4-element tournament D from Ex-
ercise 62.

129
 Bibliography

[1] L. Agarwal. The reducts of the generic digraph. Annals of Pure and Applied Logic, 167:370–391,
2016.
[2] L. Agarwal and M. Kompatscher. 2ℵ0 pairwise nonisomorphic maximal-closed subgroups of
Sym(N) via the classification of the reducts of the Henson digraphs. Journal of Symbolic Logic,
83(2):395–415, 2018.
[3] G. Ahlbrandt and M. Ziegler. Quasi-finitely axiomatizable totally categorical theories. Annals
of Pure and Applied Logic, 30(1):63–82, 1986.
[4] R. Akhtar and A. H. Lachlan. On countable homogeneous 3-hypergraphs. Arch. Math. Log.,
34(5):331–344, 1995.
[5] R. Baer. Die Kompositionsreihe der Gruppe aller eineindeutigen Abbildungen einer unendlichen
Menge auf sich. Studia Mathematica, 5:15–17, 1934.
[6] S. Barbina and D. Macpherson. Reconstruction of homogeneous relational structures. Journal
of Symbolic Logic, 72(3):792–802, 2007.
[7] L. Barto, M. Kompatscher, M. Olšák, T. V. Pham, and M. Pinsker. Equations in oligomor-
phic clones and the constraint satisfaction problem for ω-categorical structures. Journal of
Mathematical Logic, 19(2):#1950010, 2019.
[8] L. Barto and M. Pinsker. The algebraic dichotomy conjecture for infinite domain constraint sat-
isfaction problems. In Proceedings of the 31th Annual IEEE Symposium on Logic in Computer
Science – LICS’16, pages 615–622, 2016. Preprint arXiv:1602.04353.
[9] H. Becker and A. Kechris. The Descriptive Set Theory of Polish Group Actions. Number 232
in LMS Lecture Note Series. Cambridge University Press, 1996.
[10] J. H. Bennett. The reducts of some infinite homogeneous graphs and tournaments. PhD thesis,
Rutgers university, 1997.
[11] V. Bergelson. Ergodic Ramsey theory: a dynamical approach to static theorems. In Proceedings
of the International Congress of Mathematicians, volume II, pages 1655–1678, Zürich, 2006.
European Mathematical Society.
[12] M. Bhattacharjee, D. Macpherson, R. G. Möller, and P. M. Neumann. Notes on Infinite Per-
mutation Groups. Springer Lecture Notes in Mathematics, 1998.
[13] C. E. Blair. The baire category theorem implies the principle of dependent choices. Bull Acad.
Polon. Sci. Ser. Sci. Math. Astron. Phys., 25, 1977.
[14] M. Bodirsky. Cores of countably categorical structures. Logical Methods in Computer Science
(LMCS), 3(1):1–16, 2007.
[15] M. Bodirsky. Ramsey classes: Examples and constructions. In Surveys in Combinatorics. Lon-
don Mathematical Society Lecture Note Series 424. Cambridge University Press, 2015. Invited
survey article for the British Combinatorial Conference; ArXiv:1502.05146.
[16] M. Bodirsky. Complexity of Infinite-Domain Constraint Satisfaction. Lecture Notes in Logic
(52). Cambridge University Press, Cambridge, United Kingdom; New York, NY, 2021.
[17] M. Bodirsky. Model theory, 2022. Course Notes, TU Dresden, https://wwwpub.zih.
tu-dresden.de/~bodirsky/Model-theory.pdf.
[18] M. Bodirsky. Introduction to mathematical logic, 2023. Course notes, TU Dresden, https:
//wwwpub.zih.tu-dresden.de/~bodirsky/Logic.pdf.
[19] M. Bodirsky, P. Jonsson, and T. V. Pham. The reducts of the homogeneous binary branching
C-relation. Journal of Symbolic Logic, 81(4):1255–1297, 2016. Preprint arXiv:1408.2554.
[20] M. Bodirsky, P. Jonsson, and T. V. Pham. The Complexity of Phylogeny Constraint Satisfac-
tion Problems. ACM Transactions on Computational Logic (TOCL), 18(3), 2017. An extended
abstract appeared in the conference STACS 2016.
[21] M. Bodirsky, B. Martin, M. Pinsker, and A. Pongrácz. Constraint satisfaction problems for
reducts of homogeneous graphs. SIAM Journal on Computing, 48(4):1224–1264, 2019. A con-
ference version appeared in the Proceedings of the 43rd International Colloquium on Automata,
Languages, and Programming, ICALP 2016, pages 119:1-119:14.

131
132 BIBLIOGRAPHY

[22] M. Bodirsky and A. Mottet. Reducts of finitely bounded homogeneous structures, and lifting
tractability from finite-domain constraint satisfaction. In Proceedings of the 31th Annual IEEE
Symposium on Logic in Computer Science (LICS), pages 623–632, 2016. Preprint available at
ArXiv:1601.04520.
[23] M. Bodirsky and M. Pinsker. Reducts of Ramsey structures. AMS Contemporary Mathematics,
vol. 558 (Model Theoretic Methods in Finite Combinatorics), pages 489–519, 2011.
[24] M. Bodirsky and M. Pinsker. Minimal functions on the random graph. Israel Journal of Math-
ematics, 200(1):251–296, 2014.
[25] M. Bodirsky and M. Pinsker. Schaefer’s theorem for graphs. Journal of the ACM, 62(3):52
pages (article number 19), 2015. A conference version appeared in the Proceedings of STOC
2011, pages 655-664.
[26] M. Bodirsky and M. Pinsker. Topological Birkhoff. Transactions of the American Mathematical
Society, 367:2527–2549, 2015.
[27] M. Bodirsky and M. Pinsker. Canonical functions: a proof via topological dynamics. Homoge-
neous Structures, A Workshop in Honour of Norbert Sauer’s 70th Birthday, Contributions to
Discrete Mathematics, 16(2):36–45, 2021.
[28] M. Bodirsky, M. Pinsker, and A. Pongrácz. The 42 reducts of the random ordered graph.
Proceedings of the LMS, 111(3):591–632, 2015. Preprint available from arXiv:1309.2165.
[29] M. Bodirsky, M. Pinsker, and A. Pongrácz. Projective clone homomorphisms. Journal of Sym-
bolic Logic, 86(1):148–161, 2021.
[30] M. Bodirsky, M. Pinsker, and T. Tsankov. Decidability of definability. Journal of Symbolic
Logic, 78(4):1036–1054, 2013. A conference version appeared in the Proceedings of the Twenty-
Sixth Annual IEEE Symposium on. Logic in Computer Science (LICS 2011), pages 321-328.
[31] M. Bodirsky and M. Wrona. Equivalence constraint satisfaction problems. In Proceedings of
Computer Science Logic, volume 16 of LIPICS, pages 122–136. Dagstuhl Publishing, September
2012.
[32] B. Bodor, P. J. Cameron, and C. Szabó. Infinitely many reducts of homogeneous structures.
Algebra Universalis, 79(2):43, 2018. arXiv:1609.07694.
[33] B. Bodor, K. Kalina, and C. Szabo. Permutation groups containing infinite linear groups and
reducts of infinite dimensional linear spaces over the two element field. Communications in
Algebra, 45(7):2942–2955, 2017. Preprint arXiv:1506.00220.
[34] J. Böttcher and J. Foniok. Ramsey properties of permutations. Electronic Journal of Combi-
natorics, 20(1), 2013.
[35] N. Bourbaki. General Topology, Volume 1. Springer, 1998.
[36] S. Braunfeld. Monadic stability and growth rates of ω-categorical structures. Proceedings of
the London Mathematical Society, Series B, 124(3), 2022. Preprint arXiv:1910.04380.
[37] P. J. Cameron. Transitivity of permutation groups on unordered sets. Mathematische
Zeitschrift, 148:127–139, 1976.
[38] P. J. Cameron. Orbits of permutation groups on unordered sets, II. Journal of the London
Mathematical Society, 2:249–264, 1981.
[39] P. J. Cameron. Oligomorphic permutation groups. Cambridge University Press, Cambridge,
1990.
[40] P. J. Cameron. Permutation Groups. LMS Student Text 45. Cambridge University Press, Cam-
bridge, 1999.
[41] P. J. Cameron. Some counting problems related to permutation groups. Discrete Mathematics,
225(1-3):77–92, 2000.
[42] P. J. Cameron. Homogeneous permutations. Electronic Journal of Combinatorics, 9(2), 2002.
[43] G. Cantor. Über unendliche, lineare Punktmannigfaltigkeiten. Mathematische Annalen,
23:453–488, 1884.
[44] G. Cherlin. Homogeneous digraphs I. The imprimitive case. Logic Colloquium 1985, 1987.
[45] G. L. Cherlin. Combinatorial problems connected with finite homogeneity. Contemporary
Mathematics, 131:3–30, 1993.
[46] G. L. Cherlin. The classification of countable homogeneous directed graphs and countable
homogeneous n-tournaments. AMS Memoir, 131(621), January 1998.
[47] J. Dixon, P. M. Neumann, and S. Thomas. Subgroups of small index in infinite symmetric
groups. Bulletin of the London Mathematical Society, 18(6):580–586, 1986.
[48] J. D. Dixon and B. Mortimer. Permutation Groups. Springer, New York, 1996.
[49] F. R. Drake. Set Theory, An Introduction to Large Cardinals. North-Holland Publishing Co.,
Amsterdam, 1974.
[50] M. Droste, C. W. Holland, and D. Macpherson. Automorphism groups of infinite semilinear
orders (I). Proceedings of the London Mathematical Society, 58:454 – 478, 1989.
 BIBLIOGRAPHY 133

[51] M. Droste, C. W. Holland, and D. Macpherson. Automorphism groups of infinite semilinear

orders (II). Proceedings of the London Mathematical Society, 58:479 – 494, 1989.
[52] D. M. Evans. Subgroups of small index in general linear groups. Bulletin of the London Math-
ematical Society, 18:587–590, 1986.
[53] D. M. Evans. A closed oligomorphic permutation group without oligomorphic extremely
amenable subgroups. Announced at the workshop “Homogeneous structures” in Banff, 2015.
[54] D. M. Evans and P. R. Hewitt. Counterexamples to a conjecture on relative categoricity. Annals
of Pure and Applied Logic, 46(2):201–209, 1990.
[55] J. Falque and N. M. Thiéry. The orbit algebra of a permutation group with polynomial profile
is Cohen-Macaulay. Preprint ArXiv:1804.03489, 2018.
[56] U. Felgner. Die unabhängigkeit des Boolschen Primidealtheorems vom Ordnungser-
weiterungssatz. Habilitationsschrift, Universität Heidelberg, 1972.
[57] A. Fraenkel. Eine abgeschwaechte fassung des auswahlaxioms. Journal of Symbolic Logic,
2(1):1–25, 1937.
[58] R. Fraı̈ssé. Sur l’extension aux relations de quelques propriétés des ordres. Annales Scientifiques
de l’École Normale Supérieure, 71:363–388, 1954.
[59] R. Fraı̈ssé. Theory of Relations. Elsevier Science Ltd, North-Holland, Amsterdam, 1986.
[60] S. Gao. Invariant Descriptive Set Theory. Pure and applied mathematics. Taylor and Francis,
2008.
[61] E. D. Gaughan. Topological group structures of infinite symmetric groups. Proceedings of the
National Academy of Sciences of the United States of America, 58:907–910, 1967.
[62] C. Good and I. J. Tree. Continuing horrors of topology without choice. Topol. Appl., 63:79–90,
1995.
[63] R. L. Graham, B. L. Rothschild, and J. H. Spencer. Ramsey theory. Wiley-Interscience Series
in Discrete Mathematics and Optimization. John Wiley & Sons, Inc., New York, 1990. Second
edition.
[64] M. Grohe. The structure of fixed-point logics. PhD-thesis at the Albert-Ludwigs Universität,
Freiburg i. Br., 1994.
[65] C. W. Henson. Countable homogeneous relational systems and categorical theories. Journal of
Symbolic Logic, 37:494–500, 1972.
[66] H. Herrlich. Choice principles in elementary topology and analysis. Comment. Math. Univ.
Carolin., 38(3):545–552, 1997.
[67] H. Herrlich and K. Keremedis. The baire category theorem and choice. Topology and its Ap-
plications, 108:157–167, 2000.
[68] B. Herwig. Extending partial isomorphisms on finite structures. Combinatorica, 15:365–371,
1995.
[69] B. Herwig. Extending partial isomorphisms for the small index property of many ω-categorical
structures. Israel Journal of Mathematics, 107:93–123, 1998.
[70] B. Herwig. Extending partial automorphisms and the profinite topology on free groups. Trans-
actions of the American Mathematical Society, 352(5):1985–2021, 2000.
[71] M. Hils and F. Loeser. A first Journey through Logic, volume 89 of Student Mathematical
Library. American Mathematical Society, Providence, RI, September 2019.
[72] W. Hodges. Model theory. Cambridge University Press, Cambridge, 1993.
[73] W. Hodges. A shorter model theory. Cambridge University Press, Cambridge, 1997.
[74] W. Hodges, I. Hodkinson, D. Lascar, and S. Shelah. The small index property for ω-stable ω-
categorical structures and for the random graph. Journal of the London Mathematical Society,
S2-48(2):204–218, 1993.
[75] I. Hodkinson and M. Otto. Finite conformal hypergraph covers and Gaifman cliques in finite
structures. Bulletin of Symbolic Logic, 9:387–405, 2003.
[76] P. Howard and J. E. Rubin. The boolean prime ideal theorem plus countable choice do not
imply dependent choice. Mathematical Logic Quarterly, 42(1):410–420, 1996.
[77] E. Hrushovski. Extending partial isomorphisms of graphs. Combinatorica, 12(4):411–416, 1992.
[78] J. Hubička and J. Nešetřil. Bowtie-free graphs have a Ramsey lift. arXiv:1402.2700, 2014.
[79] J. Hubička and J. Nešetřil. All those Ramsey classes (Ramsey classes with closures and forbid-
den homomorphisms). Advances in Mathematics, 356:106791, 2019.
[80] A. A. Ivanov. Generic expansions of ω-categorical structures and semantics of generalized
quantifiers. J. Symbolic Logic, 64:775–789, 1999.
[81] M. K. Jan Hubička and J. Nešetřil. All those eppa classes (strengthenings of the Herwig-Lascar
theorem). Trans. Amer. Math. Soc., 375:7601–7667, 2022.
[82] T. Jech. The Axiom of Choice. North-Holland Publishing Company, Amsterdam, London, New
York, 1973.
134 BIBLIOGRAPHY

[83] T. Jech. Set theory. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2003. The
third millennium edition, revised and expanded.
[84] T. Jech and K. Hrbáček. Introduction to Set Theory, Third edition. Monographs and Textbooks
in Pure and Applied Mathematics. CRC Press, 1999.
[85] T. Jenkinson, J. K. Truss, and D. Seidel. Countable homogeneous multipartite graphs. Euro-
pean Journal of Combinatorics, 33:82–109, 2012.
[86] R. B. Jensen. Independence of the axiom of dependent choices from the countable axiom of
choice. Journal of Symbolic Logic, 31:294, 1966.
[87] M. Junker and M. Ziegler. The 116 reducts of (Q, <, a). Journal of Symbolic Logic, 74(3):861–
884, 2008.
[88] A. Kechris. Classical descriptive set theory, volume 156 of Graduate Texts in Mathematics.
Springer, 1995.
[89] A. Kechris, V. Pestov, and S. Todorčević. Fraı̈ssé limits, Ramsey theory, and topological dy-
namics of automorphism groups. Geometric and Functional Analysis, 15(1):106–189, 2005.
[90] A. Kechris and M. Sokić. Dynamical properties of the automorphism groups of the random
poset and random distributive lattice. Fund. Math., 218(1):69–94, 2012.
[91] A. S. Kechris. Dynamics of non-archimedean Polish groups. In Proceedings of the European
Congress of Mathematics, Krakow, pages 375–397. European Math. Society, 2014.
[92] A. S. Kechris and C. Rosendal. Turbulence, amalgamation and generic automorphisms of ho-
mogeneous structures. Proceedings of the London Mathematical Society, 94(3):302–350, 2007.
[93] J. L. Kelley. The tychonoff product theorem implies the axiom of choice. Fund. Math., 37:75–76,
1950.
[94] B. Klin, E. Kopczynski, J. Ochremiak, and S. Toruńczyk. Locally finite constraint satisfaction
problems. In 30th Annual ACM/IEEE Symposium on Logic in Computer Science, LICS 2015,
Kyoto, Japan, pages 475–486, 2015.
[95] B. Klin, S. Lasota, J. Ochremiak, and S. Torunczyk. Homomorphism Problems for First-Order
Definable Structures. In A. Lal, S. Akshay, S. Saurabh, and S. Sen, editors, 36th IARCS An-
nual Conference on Foundations of Software Technology and Theoretical Computer Science
(FSTTCS 2016), volume 65 of Leibniz International Proceedings in Informatics (LIPIcs),
pages 14:1–14:15, Dagstuhl, Germany, 2016. Schloss Dagstuhl–Leibniz-Zentrum fuer Infor-
matik.
[96] M. Kompatscher and T. V. Pham. A Complexity Dichotomy for Poset Constraint Satisfaction.
In 34th Symposium on Theoretical Aspects of Computer Science (STACS), volume 66 of Leibniz
International Proceedings in Informatics (LIPIcs), pages 47:1–47:12, 2017.
[97] M. Krom. Equivalents of a weak axiom of choice. Notre Dame Journal of Formal Logic, 22(3),
1981.
[98] A. Kwiatkowska and A. Panagiotopoulos. The automorphism group of the random poset does
not admit a generic pair, 2020. ArXiv 2012.04376.
[99] A. H. Lachlan. Countable homogeneous tournaments. Transactions of the American Mathe-
matical Society (TAMS), 284:431–461, 1984.
[100] A. H. Lachlan. Stable finitely homogeneous structures: A survey. In Algebraic Model Theory,
NATO ASI Series, volume 496, pages 145–159, 1996.
[101] A. H. Lachlan and R. E. Woodrow. Countable ultrahomogeneous undirected graphs. Transac-
tions of the AMS, 262(1):51–94, 1980.
[102] C. Laflamme, J. Jasinski, L. N. V. Thé, and R. Woodrow. Ramsey precompact expansions of
homogeneous directed graphs. Electron. J. Combin., 21(4), 2014.
[103] D. Lascar. On the category of models of a complete theory. Journal Symbolic Logic, 47(2):249–
266, 1982.
[104] D. Lascar. Autour de la propriété du petit indice. Proceedings of the London Mathematical
Society, 62(1):25–53, 1991.
[105] B. Latka. Finitely constrained classes of homogeneous directed graphs. Journal of Symbolic
Logic, 59(1):124 – 139, 1994.
[106] H. Läuchli. The independence of the ordering principle from a restricted axiom of choice.
Fundamenta Mathematicae, 54:31–43, 1964.
[107] J. Linman and M. Pinsker. Permutations on the random permutation. Electronic Journal of
Combinatorics, 22(2):1–22, 2015.
[108] D. C. Lockett and J. K. Truss. Homogeneous coloured multipartite graphs. European Journal
of Combinatorics, 42:217–242, 2014.
[109] G. Lolli. On Ramsey’s theorem and the axiom of choice. Notre Dame Journal of Formal Logic,
18:599–601, 1977.
 BIBLIOGRAPHY 135

[110] D. Macpherson. A survey of homogeneous structures. Discrete Mathematics, 311(15):1599–

1634, 2011.
[111] D. Macpherson and K. Tent. Simplicity of some automorphism groups. J. Algebra, 342:40–52,
2011.
[112] H. Macpherson. Infinite permutation groups of rapid growth. J. London Math. Soc., 35(2):276–
286, 1987.
[113] H. D. Macpherson. The action of an infinite permutation group on the unordered subsets of a
set. Proceedings of the LMS, 46(3):471–486, 1983.
[114] H. D. Macpherson. Growth rates in infinite graphs and permutation groups. Proc. London
Math. Soc., 51(3):285–294, 1985.
[115] H. D. Macpherson. Orbits of infinite permutation groups. Proc. London Math. Soc., 51(3):246–
284, 1985.
[116] D. Marker. Model Theory: An Introduction. Springer, New York, 2002.
[117] A. R. D. Mathias. The order-extension principle. In Axiomatic set theory, Proceedings of
Symposia of Pure Mathematics, pages 179–184. American Mathematical Society, 1974.
[118] J. Melleray. Polish groups and Baire category methods. Confluentes Mathematici, 8(1):89–164,
2016.
[119] J. Melleray, L. N. V. Thé, and T. Tsankov. Polish groups with metrizable universal minimal
flows. International Mathematics Research Notices, 2016:1285–1307, 2015.
[120] A. Mottet and M. Pinsker. Cores over Ramsey structures. Journal of Symbolic Logic, 86(1):352–
361, 2021.
[121] M. Müller and A. Pongrácz. Topological dynamics of unordered ramsey structures. Fundamenta
Mathematicae, 230(1):77–98, 2015. ArXiv:1401.7766.
[122] J. Nešetřil. Ramsey theory. Handbook of Combinatorics, pages 1331–1403, 1995.
[123] J. Nešetřil. Ramsey classes and homogeneous structures. Combinatorics, Probability & Com-
puting, 14(1-2):171–189, 2005.
[124] J. Nešetřil. Metric spaces are Ramsey. European Journal of Combinatorics, 28(1):457–468,
2007.
[125] J. Nešetřil and V. Rödl. The partite construction and Ramsey set systems. Discrete Mathe-
matics, 75(1-3):327–334, 1989.
[126] J. Nešetřil and V. Rödl. Mathematics of Ramsey Theory. Springer, Berlin, 1998.
[127] L. Onofri. Teoria delle sostituzioni che operano su una infinità numerabile di elementi, memoria
iii. Annali di Matematica Pura ed Applicata, 7(1):103–130, 1929. par.141, p.124.
[128] P. P. Pach, M. Pinsker, G. Pluhár, A. Pongrácz, and C. Szabó. Reducts of the random partial
order. Advances in Mathematics, 267:94–120, 2014.
[129] G. Paolini and S. Shelah. The strong small index property for free homogeneous structures.
Submitted, 2018. Preprint available at ArXiv 1703.10517.
[130] G. Paolini and S. Shelah. Reconstructing structures with the strong small index property up
to bi-definability. Fundamenta Mathematicae, 247:25–35, 2019.
[131] D. Pincus. Adding dependent choice to the boolean prime ideal theorem. Logic Colloq., 76:547–
565, 1977.
[132] B. Poizat. A Course in Model Theory: An Introduction to Contemporary Mathematical Logic.
Springer, 2000.
[133] A. Pongrácz. Reducts of the Henson graphs with a constant. Annals of Pure and Applied Logic,
168(7):1472–1489, 2017.
[134] A. Pongrácz. Reducts of the Henson graphs with a constant. Annals of Pure and Applied Logic,
168(7):1472–1489, 2017.
[135] M. Pouzet. Caractérisation topologique et combinatoire des ages les plus simples. D. Sc. thesis,
1978.
[136] M. Pouzet and N. M. Thiéry. Some relational structures with polynomial growth and their as-
sociated algebras I: Quasi-polynomiality of the profile. The electronic journal of combinatorics,
20(2), 2013.
[137] E. B. Rabinovich. Embedding theorems and de Bruijn’s problem for bounded symmetry groups.
Dokl. Akad. Nauk. Belor. S.S.R., 21(9):784–7, 1977. Russian.
[138] C. Rosendal. Automatic continuity of group homomorphisms. Bulletin of Symbolic Logic,
15(2):184–214, 2009.
[139] M. Rubin. On the reconstruction of ω-categorical structures from their automorphism groups.
Proceedings of the London Mathematical Society, 3(69):225–249, 1994.
[140] J. H. Schmerl. Countable homogeneous partially ordered sets. Algebra Universalis, 9:317–321,
1979.
136 BIBLIOGRAPHY

[141] J. Schreier and Stanislaw Marcin Ulam. Über die Permutationsgruppe der natürlichen Zahlen-
folge. Studia Mathematica, 4:134–141, 1933.
[142] S. W. Semmes. Endomorphisms of infinite symmetric groups. Abstracts of the American Math-
ematical Society, 2:426, 1981.
[143] S. Shelah. Can you take Solovay’s inaccessible away? Israel Journal of Mathematics, 48(1):1–
47, 1984.
[144] P. Simon. On ω-categorical structures with few finite substructures, 2018.
[145] D. Siniora and S. Solecki. Coherent extension of partial automorphisms, free amalgamation
and automorphism groups. J. Symb. Log., 85(1):199–223, 2020.
[146] D. N. Siniora. Automorphism groups of homogeneous structures. Ph.D. thesis, University of
Leeds, 2017.
[147] M. Sokić. Ramsey property of posets and related structures. PhD thesis, University of Toronto,
2010.
[148] M. Sokić. Ramsey property, ultrametric spaces, finite posets, and universal minimal flows.
Israel Journal of Mathematics, 194(2):609–640, 2013.
[149] M. Sokić. Directed graphs and Boron trees. Journal of Combinatorial Theory, Series A,
132:142–171, 2015.
[150] R. M. Solovay. A model of set theory in which every set of reals is Lebesgue measurable. Annals
of Mathematics, 92:1–56, 1970.
[151] A. Tarski. Prime ideal theorem for set algebras and ordering principles. Bulletin of the Amer-
ican Mathematical Society, 60:390–39, 1954.
[152] K. Tent and M. Ziegler. A course in model theory. Lecture Notes in Logic. Cambridge University
Press, 2012.
[153] L. N. V. Thé. More on the Kechris-Pestov-Todorcevic correspondence: precompact expansions.
Fund. Math., 222(1):19–47, 2013. Preprint arXiv:1201.1270.
[154] L. N. V. Thé. Universal flows of closed subgroups of S∞ and relative extreme amenability.
Asymptotic Geometric Analysis, Fields Institute Communications, 68:229–245, 2013.
[155] L. N. V. Thé. A survey on structural Ramsey theory and topological dynamics with the Kechris-
Pestov-Todorčević correspondence in mind. Accepted for publication in Zb. Rad. (Beogr.),
2014. Preprint arXiv:1412.3254v2.
[156] S. Thomas. Reducts of the random graph. Journal of Symbolic Logic, 56(1):176–181, 1991.
[157] S. Thomas. Reducts of random hypergraphs. Annals of Pure and Applied Logic, 80(2):165–193,
1996.
[158] J. K. Truss. Infinite permutation groups. II. Subgroups of small index. Journal of Algebra,
120(2):494–515, 1989.
[159] J. K. Truss. Generic automorphisms of homogeneous structures. Proc. London Math. Soc.,
3(65):121–141, 1992.
[160] J. K. Truss. On notions of genericity and mutual genericity. The Journal of Symbolic Logic,
72(3):755–766, 2007.
[161] A. N. Tychonoff. über die topologische Erweiterung von Räumen. Mathematische Annalen,
102(1):544–561, 1930.
[162] I. B. Yaacov and T. Tsankov. Weakly almost periodic functions, model-theoretic stability,
and minimality of topological groups. Transactions of the AMS, 368(11):8267–8294, 2016.
arXiv:1312.7757.
[163] A. Zucker. Amenability and unique ergodicity of automorphism groups of Fraı̈ssé structures.
Fund. Math., 841:41–62, 2014. Preprint, arXiv:1304.2839.
[164] A. Zucker. Topological dynamics of closed subgroups of Sω . Preprint, arXiv:1404.5057, 2014.
 APPENDIX A

Background Material

A.1. Ultrafilter
Let X be a set. A filter on X is a certain set of subsets of X; the idea is that the
elements of F are (in some sense) ‘large’; it helps thinking of the elements F ∈ F as
being ‘almost all’ of X.
Definition A.1.1. A filter F on X is a set of subsets of X such that
(1) ∅ ∈/ F and X ∈ F;
(2) if F ∈ F and G ⊆ X contains F , then G ∈ F.
(3) if F1 , F2 ∈ F then F1 ∩ F2 ∈ F.
Note that filters have the finite intersection property:
A1 , . . . , An ∈ F ⇒ A1 ∩ · · · ∩ An 6= ∅ (FIP)
Lemma A.1.2. Every subset S ⊆ P(X) with the FIP is contained in a smallest
filter that contains S; this filter is called the filter generated by S.
Proof. First add finite intersections, and then all supersets to S. 
Example 96. For a non-empty subset Y ⊆ X, the family
F := {Z ⊆ X | Y ⊆ Z}
is a filter, the filter generated by a {Y }; such filters are called principal. 4
Example 97. The Fréchet filter : for an infinite set X this is the filter F that
consists of all cofinite subsets of X, i.e.,
F := {Y ⊆ X | X \ Y is finite}. 4
A filter F is called a ultrafilter if F is maximal, that is for every filter G ⊇ F we
have G = F.
Lemma A.1.3. Let F be a filter. Then the following are equivalent.
(1) F is a ultrafilter.
(2) For all A ⊆ X either A ∈ F or X \ A ∈ F.
(3) For all A1 ∪ · · · ∪ An ∈ F there is an i ≤ n with Ai ∈ F.
Proof. (1) ⇐ (2): No A ⊆ X can be added to F. Hence, F is maximal.
(2) ⇐ (3): Note that A ∪ (X \ A) = X ∈ F.
(1) ⇒ (3): If there is an i ≤ n such that F ∪ {Ai } has the FIP, then by
Lemma A.1.2 there is a filter that contains this set, and hence F was not maxi-
mal. Otherwise, there are S1 , . . . , Sn ⊆ F with Ai ∩ Si = ∅. Then Si ⊆ X \ Ai and
thus S1 ∩ · · · ∩ Sn ⊆ X \ (A1 ∪ · · · ∪ An ) ∈
/ F, a contradiction. 
A filter F is principal if it
T contains a inclusionwise minimal element. Note that
this is the case if and only if F ∈ F.
Lemma A.1.4. Let F be a filter on a set X. Then the following are equivalent.
137
138 A. BACKGROUND MATERIAL

(1) F
is a principal ultrafilter;
(2) F
contains {a} for some a ∈ X.
(3) F
is of the form {Y ⊆ X | a ∈ Y } for some a ∈ X.
(4) F
is an ultrafilter and contains a finite set.
T
Proof. (1) ⇒ (2): let A := F ∈ F. If |A| > 1 then we can write A = B1 ∪ B2
for B1 , B2 ⊆ X non-empty. But then Lemma A.1.3 (3) implies that B1 ∈ F or
B2 ∈ F, in contradiction to the definition of A. So A = {a} for some a ∈ X.
(2) ⇒ (3). Clearly, {Y ⊆ X | a ∈ Y } ⊆ F since F is closed under supersets, and
F ⊆ {Y ⊆ X | a ∈ Y } since F does not contain the empty set.
(3) ⇒ (4): Clearly F contains a finite set; use Lemma A.1.3 (2) to check that F
is an ultrafilter. T
(4) ⇒ (1): If A ∈ F is finite, then B := F is finite, and hence B is the
intersection of finitely many elements in F, and hence in F since F is a filter. This
shows that F is principal. 
Are there non-principal ultrafilters?
Lemma A.1.5 (Ultrafilter Lemma). Every filter F is contained in a ultrafilter.
Proof. Let M be the set of all filters on X that contain F, partially ordered
by containment. Note that unions of chains of filters in this partial order are again
filters. By Zorn’s lemma, M contains a maximal filter. 
Non-principal ultrafilters are also called free ultrafilters. In particular the Fréchet
filter is contained in an ultrafilter, which must be free:
Lemma A.1.6. An ultrafilter is free if and only if it contains the Fréchet filter.
Proof. Let U be a free ultrafilter on X and let x ∈ X. Either {x} ∈ U or
X \ {x} ∈ U. As U is free, {x} ∈
/ U (Lemma A.1.4). Hence, X \ {x} ∈ U for every
x ∈ X. Let F ⊆ X be finite. Then
\
X \F = (X \ {x}) ∈ U .
x∈F

Now let U be a principal ultrafilter, i.e., there is x ∈ X with {x} ∈ U (Lemma A.1.4).
Then the element X \ {x} of the Fréchet filters is not in U. 

Exercises.
(143) Show that a set of subsets of a set X can be extended to an ultrafilter if and
only if it has the FIP.
(144) Show that a set F of subsets of a set X can be extended to a free ultrafilter
if and only if the intersection of every finite subset of F is infinite.
(145) Show that every filter F on a set X is the intersection of all ultrafilters on
X that extend F.
(146) Show that if U is a free ultrafilter on X, and S ∈ U and T ⊆ X are
such that the symmetric difference S∆T is finite, then S ∈ U.
|X|
(147) Show that there are 22 many ultrafilters on an infinite set X.
Hint: first show that there is a family F of 2|X| subsets of X such that for
any A1 , . . . , An , B1 , . . . , Bn ∈ F
A1 ∩ · · · ∩ An ∩ (X \ B1 ) ∩ · · · ∩ (X \ Bn ) 6= ∅.
(148) True or false: if U and V are free ultrafilters on an infinite set X, is there is
a permutation π of X such that S ∈ U if and only if π(S) ∈ V?1
1Thanks to Lukas Juhrich for the idea for this exercise.
 A.2. THE AXIOM OF CHOICE AND ITS WEAKER VERSIONS 139

A.2. The Axiom of Choice and its Weaker Versions

‘Das “Auswahlaxiom” (...) fordert in der gewöhnlichen (...) Fassung, daß zu jeder
Menge M , deren Elemente (...) paarwise fremde und nicht-leere Mengen sind,
mindestens eine “Auswahlmenge” existiere, die mit jedem Element von M genau ein
Element gemeinsam hat. Die nächstliegende und mehrfach verwendete Methode, um
ein schwächeres Postulat als die vorstehende Fassung zu formulieren, besteht darin,
daß man entweder über die Mächtigkeit der Menge M , oder über die Mächtigkeit
ihrer Elemente, oder über beides gleichzeitig, einschränkende Bedingung macht, also
nur in diesen eingeschränkten Fällen die Existenz einer Auswahlmenge verlangt.’
Abraham Adolf Halevi Fraenkel [57]
‘Let us sum things up: Topology with “choice” may be as unreal as a soap-bubble
dream, but topology without “choice” is as horrible as a nightmare.’
Horst Herrlich [66]
The proofs of several statements in this text used the Axiom of Choice (Defini-
tion A.2.1). Sometimes these statements can also be proved using weaker forms of
the Axiom of Choice, and sometimes they are even equivalent to these weaker forms.
• The Baire Category Theorem (Theorem 4.1.8);
• The Ultrafilter Lemma (Lemma A.1.5);
• The equivalence of compactness and sequential compactness (Theorem 4.1.19);
• Tychonoff’s theorem (Theorem 4.1.14).
We have collected known facts about some weaker forms of the Axiom of Choice
in Figure A.1. We first state these weaker forms and then provide references for the
equivalences and implications that are displayed in Figure A.1. We have also collected
what we know about non-implications.
Definition A.2.1
S (Axiom of Choice). For every nonempty set A there exists a
function f : A → A such that for every S ∈ A we have f (S) ∈ S.
The Axiom of Countable Choice is defined as the axiom of choice, but restricted
to countable sets A.
Definition A.2.2 (Axiom of Dependent Choices). Let R ⊆ A2 be a relation with
the property that for every a ∈ A there exists b ∈ A with (a, b) ∈ R. Then for every
c ∈ A there exists a function f : N → A such that f (0) = c and for every i ∈ N we
have (f (i), f (i + 1)) ∈ R.
Definition A.2.3 (Order Extension Principle). Every partial order can be ex-
tended to a linear order.
Definition A.2.4 (Ordering Principle). Every set can be linearly ordered.
The following equivalences are known to hold in Zermelo-Fraenkel set theory (ZF):
• The Axiom of Choice, Zorn’s Lemma, and the Well-ordering Theorem are
equivalent; see, e.g., [84].
• The Axiom of Choice implies Tychonoff’s theorem [161], and Tychonoff’s
theorem implies the Axiom of Choice [93].
• The equivalence to the existence of surjections with right inverse is Exer-
cise 151.
• The Axiom of Dependent Choices can be used to prove the Baire Category
Theorem (Theorem 4.1.8) as we have seen in Section 4.1.5; the implica-
tion also holds for pseudo-metrics instead of metrics (also see the discussion
in [66]); conversely, the Baire Category Theorem for pseudo-metric spaces
implies the Axiom of Dependent Choices [13].
140 A. BACKGROUND MATERIAL

Every surjection has Well-Ordering

Axiom of Choice
a right inverse Theorem

Tychonoff Theorem Zorn’s lemma

Ultrafilter Compactness
Lemma of first-order logic Axiom of Baire Category
Dependent Choices Theorem
Boolean Prime Ideal
Theorem
Tychonoff Theorem Tychonoff Theorem Tychonoff Theorem
for Hausdorff spaces for Finite Spaces for Countable Products

Order Extension Principle Baire Category

Axiom of
Theorem for 2nd
Countable Choice
Countable Spaces
Ordering Principle
compactness <-> separable <->
Axiom of Choice from Finite Sets sequential compactness second-countable

Axiom of Countable Choice Compactness of first-order logic

from Finite Sets in countable signatures
Tychonoff Theorem for Countable
Ramsey’s theorem
Products of Finite Spaces
König’s tree lemma

ZF

Figure A.1. The extensions of Zermelo-Fraenkel set theory by the

Axiom of Choice or some of its weak versions, and their relationships.
Blue arcs indicate implications, red arcs indicate strict implications.

• The Axiom of Countable Choice is equivalent to the Baire Category The-

orem for second-countable pseudo-metric spaces [66, Theorem 2.4], and to
Theorem 4.1.19 [66, Theorem 2.4].
• For the equivalence of the Boolean Prime Ideal Theorem (which we do not
need here) and the Ultrafilter Lemma, see [82].
• The equivalence of the Ultrafilter Lemma and the compactness theorem of
first-order logic is well-known; see [17].
• References for the equivalence between the Ultrafilter Lemma, Tychonoff’s
theorem for Hausdorff spaces, and Tychonoff’s theorem for finite spaces can
be found in [66, Theorem 3.4].
• The equivalence of the Axiom of Countable Choice from Finite Sets and
König’s Tree Lemma can be found in [49] (page 203), the equivalence to
Tychonoff’s theorem for countable products of finite spaces in [97], and the
equivalence to Ramsey’s theorem in [109]. A proof that countable choice
from finite sets suffices to prove the compactness theorem for countable
signatures can be found in [71], and the converse is easy to prove.
The following implications are known to hold in Zermelo-Fraenkel set theory (ZF):
• The Axiom of Choice implies the Axiom of Dependent Choices (Exercise 152).
 A.2. THE AXIOM OF CHOICE AND ITS WEAKER VERSIONS 141

• The Axiom of Dependent Choices implies Tychonoff’s theorem for countable

products of compact spaces (see the proof of Theorem 4.1.14).
• Tychonoff’s theorem for countable products implies the Axiom of Countable
Choice [62].
• The Axiom of Choice implies the Ultrafilter Lemma (Lemma A.1.5).
• The Ultrafilter Lemma implies the Order Extension Principle [151].
• The Order Extension Principle implies the Ordering Principle (Exercise 149).
• The Ordering Property implies the Axiom of Choice for families of non-
empty finite sets (Exercise 150).
• Trivially, the Axiom of Countable Choice from Finite Sets is implied by the
Axiom of Countable choice, and is implied by the Axiom of Choice from
Finite Sets.
The following can be proved within ZF alone, without any additional choice axioms:
• Choice for finite families.
• The Baire category theorem for Polish spaces (Theorem 4.1.8).
• The compactness theorem for countable signature [71].
• König’s tree lemma for countable trees (which is not the same as König’s
tree lemma because we cannot prove in ZF that a finitely branching tree is
countable).
Independence Results. There are models of ZF that show the following.
• The Ultrafilter Lemma and the Axiom of Dependent Choices do not imply
the Axiom of Choice [131]. In particular, the Ultrafilter Lemma does not
imply the Axiom of Choice, and the Axiom of Dependent Choices does not
imply the Axiom of Choice.
• The Axiom of Dependent Choices does not imply the Ultrafilter lemma;
this follows from Example 83 in combination with Theorem 6.3.13. Also see
Remark 2.11 (3) in [67].
• the Ultrafilter Lemma and the Axiom of Countable Choice do not imply the
Axiom of Dependent Choices [76]. In particular, the Axiom of Countable
Choice does not imply the Axiom of Dependent Choices [86], and the Ul-
trafilter Lemma does not apply the Axiom of Dependent Choices. It also
follows that Tychonoff’s theorem for countable products alone does not im-
ply the Axiom of Dependent Choices: this is because the Ultrafilter Lemma
together with Tychonoff’s theorem for countable products does imply the
Axiom of Dependent Choices (see Remark 1.7 (1) in [67]).
• The Boolean Prime Ideal Theorem does not follow from the Ordering Ex-
tension Property [56].
• The Ordering Extension Property does not follow from the Ordering Prin-
ciple [117].
• The Ordering Principle does not follow from the Axiom of Choice for families
of non-empty finite sets [106].
• The so-called basic Fraenkel model of ZF does not satisfy the Axiom of
Choice from Finite Sets; there are also models of ZF that do not satisfy
the Countable Axiom of Choice from two-element sets (see Section 4.3-4.5
in [82]).
Unkown relations. We do not know whether ZF together with the Axiom of
Countable Choice implies Tychonoff’s theorem for countable products of compact
spaces [66]. It seems unlikely that the Ultrafilter Lemma implies the Axiom of Count-
able Choice, but I am not aware of any reference. I also don’t know whether the Axiom
of Dependent Choices implies the Order Extension Principle, whether the Axiom of
142 A. BACKGROUND MATERIAL

Countable Choice from Finite Sets is equivalent to the Axiom of Countable Choice,
or whether it is equivalent to the Axiom of Choice from Finite Sets.
Exercises.
(149) Show that the Order Extension Property implies the Ordering
Property.
(150) Show that the Ordering Property implies the Axiom of Choice
for families of non-empty finite sets.
(151) Show that the Axiom of Choice is equivalent to the statement
that every surjective function f : A → B has a right inverse,
i.e., a function g : B → A such that g(f (x)) = x for all x ∈ A.
(152) Show that the Axiom of Choice implies the Axiom
of Dependent Choices.
(153) (Exercise 5.7 in [83]) Show that the Axiom of Dependent
Choices implies the Axiom of Countable Choice
(without quoting the facts stated above).
Hint. Given (An )n∈N , consider the set A of all choice functions
on some Sn := {Ai | i ≤ n}, ordered by extensions.