2311 14022

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS

CHARLES GARNET COX AND ANITHA THILLAISUNDARAM


arXiv:2311.14022v1 [math.GR] 23 Nov 2023

Abstract. We investigate invariable generation among key examples of branch groups. In


particular, we prove that all generating sets of the torsion Grigorchuk groups, of the branch
Grigorchuk-Gupta-Sidki groups and of the torsion multi-EGS groups (which are natural gen-
eralisations of the Grigorchuk-Gupta-Sidki groups) are invariable generating sets. Further-
more, for the first Grigorchuk group and the torsion Grigorchuk-Gupta-Sidki groups, every
finitely generated subgroup has a finite invariable generating set. We also show that a branch
Grigorchuk-Gupta-Sidki group is almost 23 -generated, that the diameter of its generating sub-
graph ∆(G) is 2 and its total domination number is 2. Some of our results apply to groups for
which every maximal subgroup is normal. This class, known as MN , includes all nilpotent
groups.

1. Introduction
Of importance in computational Galois theory, the notion of invariable generation gives rise
to many interesting questions. A group G is invariably generated (IG) if there exists a set S such
that for any choice of elements ag ∈ G, where g ∈ G, we have ha−1 s sas : s ∈ Si = G. We then
say that S is an IG-set. All finite groups are invariably generated (known since 1872, from [23]),
leading to a question of the size of the smallest IG-set. In [34] the notion of groups where no
proper subgroup meets every conjugacy class was considered, which is equivalent to the group
being invariably generated. It was arguably not until [25] that invariable generation was studied
again for infinite groups. The picture is different here, since there are groups which are not
IG. From [34] the property of being IG is closed under extensions, and the answer to Wiegold’s
question of whether IG is stable under finite-index subgroups is surprisingly no, answered
independently in [14, 27]. One can also distinguish between groups for which a finite IG-set
exists (making them finitely invariably generated, or FIG for short). Again independently,
[14, 27] show that there are finitely generated groups that are IG but not FIG. Non-abelian
free groups of finite rank are not IG (see [34]) but for any finitely generated group G we have
that G ≀ Z is FIG (from [5]). This means that there are uncountably many FIG groups.
It was proved in [25, Lem. 2.6] that for a finitely generated group G, if all maximal subgroups
of G have finite index, then G is IG. Furthermore, if there exists an integer c such that every
maximal subgroup M of G satisfies |G : M | ≤ c, then G is FIG. It is natural to ask to
what extent are the converse statements to these results true. From [6], we know that the
converse statements do not hold. Indeed, the Houghton groups Hn , for n ≥ 2, are FIG, but
all have maximal subgroups of infinite index; here, the stabiliser of a single point x gives such
a maximal subgroup. Likewise, the 3-generated solvable group constructed by Hall [21] has
maximal subgroups of infinite index and is FIG since all solvable groups are FIG. By taking
a direct product of a FIG group with Hn , for any n ≥ 2, one can construct uncountably
many examples of FIG groups with maximal subgroups of infinite index. Still, apart from such
Date: November 27, 2023.
2010 Mathematics Subject Classification. Primary 20F05; Secondary 20E08, 20E28, 20D15.
Key words and phrases. Invariable generation, maximal subgroups, groups acting on rooted trees, branch
groups, generating subgraph.
1
2 C. G. COX AND A. THILLAISUNDARAM

constructions, it is of interest to see how often one finds FIG groups among the groups with
maximal subgroups of infinite index. In this note, we look for such examples among groups
acting on rooted trees, and additionally we investigate which subgroups of such groups are also
FIG.
Now, the automorphism group of an infinite spherically homogeneous rooted tree is well
established as a source of interesting finitely generated infinite groups, such as finitely generated
groups of intermediate word growth, finitely generated infinite torsion groups, finitely generated
amenable but not elementary amenable groups, and finitely generated just infinite groups.
Early constructions were produced by Grigorchuk [16] and Gupta and Sidki [20] in the 1980s,
which then led to a generalised family of so-called Grigorchuk-Gupta-Sidki groups (GGS-groups
for short), and also to the family of multi-EGS groups (here EGS stands for Extended Gupta
Sidki).
For the purpose of stating our results, we briefly recall the construction of some of these
groups here. The first Grigorchuk group acts on the binary rooted tree and is an infinite
2-group generated by the following four elements: the automorphism a, which swaps the two
maximal subtrees, and the recursively-defined automorphisms b, c and d, which are described
as
b = (a, c), c = (a, d) and d = (1, b);
here the notation (x, y) indicates the independent actions on the respective maximal subtrees,
for x and y automorphisms of the binary rooted tree. The first Grigorchuk group was also
generalised to a family of Grigorchuk groups, which are defined similarly.
Next, let p be an odd prime. We recall that a GGS-group acts on the p-adic tree (also called
the p-regular rooted tree), with generators a and b, where a cyclically permutes the p maximal
subtrees rooted at the first-level vertices, whereas b fixes the first-level vertices pointwise and
is recursively defined by the tuple (ae1 , . . . , aep−1 , b) which corresponds to the action of b on
the maximal subtrees, for some exponents e1 , . . . , ep−1 ∈ Fp . A multi-EGS group is, loosely
speaking, a group which is similar to a GGS-group but with many b-type generators. We refer
the reader to Section 2 for the precise definition.
An important type of subgroup of the automorphism group of an infinite spherically homo-
geneous rooted tree is one having subnormal subgroup structure similar to the corresponding
structure in the full group of automorphisms of the tree. These subgroups are termed branch
groups; see Section 2 for the definition.
Of relevance to this paper is the study of maximal subgroups of finitely generated branch
groups. The first results here were by Pervova [29, 30], who proved that the torsion Grig-
orchuk groups and the torsion GGS-groups do not contain maximal subgroups of infinite index.
Pervova’s result was generalised to the torsion multi-EGS groups by Klopsch and Thillaisun-
daram [26]. Francoeur and Thillaisundaram [13] have further shown that the non-torsion
GGS-groups do not have maximal subgroups of infinite index. The first explicit examples of
finitely generated branch groups with maximal subgroups of infinite index were provided by
Francoeur and Garrido [11]. Their examples are the non-torsion Šunić groups acting on the
binary rooted tree. Despite having maximal subgroups of infinite index, we show that these
groups are similar to the Houghton groups in that they are FIG.
Theorem 1.1. Let G be a non-torsion Šunić group acting on the binary rooted tree. Then G
is FIG.
It would be interesting to see if there exists a finitely generated group G within the class
of groups acting on rooted trees, which is IG but not FIG, but G has maximal subgroups of
infinite index.
INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS 3

It was mentioned in the introduction of [25] that the standard generating set {a, b, c} for
the first Grigorchuk group is also an IG-set. A stronger result holds, as seen below. We first
make the following observation.
Theorem 1.2. Let G be a group where every proper subgroup lies in a proper maximal sub-
group. Then all maximal subgroups of G are normal if and only if every generating set is an
IG-set.
In particular, the first hypothesis in this theorem is satisfied by any finitely generated group.
Such a hypothesis is necessary; see Remark 3.3. We note that Myropolska proved in [28,
Prop. 2.2] that for a finitely generated group, all maximal subgroups are normal if and only
if every normal generating set is a generating set. Hence, if G is a finitely generated group
with a non-normal maximal subgroup, then there exists some normal generating set that is
not a generating set. The existence of such a normal generating set also then implies the
existence of a generating set that is not an IG-set. We observe furthermore that since all
maximal subgroups of infinite index are not normal, the existence of maximal subgroups of
infinite index also indicates that there is a generating set that is not an IG-set.
Using Theorem 1.2 plus the work of [13, 26, 29, 30], we have the following.
Corollary 1.3. Let G be any torsion Grigorchuk group, branch GGS-group or torsion multi-
EGS group. Then any generating set for G is an IG-set.
It was shown in [6] that all finite-index subgroups of the Houghton groups Hn , for n ≥ 2,
are FIG. A similar situation holds for the following family of branch groups.
Theorem 1.4. Let G be a torsion multi-EGS group. Then every finite-index subgroup of G is
FIG.
Note that for a nilpotent group, each of its finitely generated subgroups is FIG; cf. Propo-
sition 3.2. Using the subgroup structure of the first Grigorchuk group and of a torsion GGS-
group, we establish the same result for these groups.
Theorem 1.5. Let G be either the first Grigorchuk group or any torsion GGS-group. Then
every finitely generated subgroup of G is FIG.
Despite the many examples of FIG groups, the class of known FIG groups still feels quite
small. Therefore it would be interesting to further investigate and understand the landscape
of FIG groups. Certainly the condition of having maximal subgroups of bounded index is not
necessary due to Z being FIG, but one could ask for examples of FIG groups with all maximal
subgroups of finite, but unbounded, index, where such maximal subgroups do not arise from Z
in an obvious manner (such as taking direct products with Z).
Next, a group G is said to be 32 -generated if for every g ∈ G\{1} there is an h ∈ G such that
hg, hi = G. This has been studied for over 50 years. Note that if a group G is 23 -generated,
then it must have every proper quotient cyclic. It was conjectured that for finite groups this
is also a sufficient condition, which was resolved in [3]. In [7] it was shown that this is not
a sufficient condition for 2-generated groups. We can also ask similar questions for general
2-generated groups by introducing the notion of a group G being almost 23 -generated which
means that if g ∈ G is such that gG′ ∈ G/G′ is part of a generating pair in G/G′ , then
g is part of a generating pair in G. This idea was introduced in [15], where it was proved
that Thompson’s group F is almost 32 -generated. We will use the standard notation that
d(G) := inf{|S| : hSi = G} and that Φ(G) denotes the Frattini subgroup of G (the intersection
of all of the maximal subgroups of G). Recall that MN denotes the class of groups for which
every maximal subgroup is normal.
4 C. G. COX AND A. THILLAISUNDARAM

Theorem 1.6. Let G ∈ MN with d(G) = 2 and suppose G′ ≤ K ≤ Φ(G) for some finitely
generated subgroup K. Then G is almost 23 -generated.
Given a 2-generated group G, we can define the generating graph Γ(G) with vertex set G\{1}
and an edge between x and y if and only if hx, yi = G. The generating graph has been well
studied, especially for finite groups. Clearly 32 -generation corresponds to the connectivity of
the generating graph. It was proved in [3] that for a finite group G, either Γ(G) has an isolated
vertex or Γ(G) is connected with diameter at most 2. For an arbitrary 2-generated group G
it is natural to consider the subgraph ∆(G) resulting from removing the isolated vertices from
Γ(G). The structure of ∆(G) even for finite groups is less straightforward since, for instance,
there is no upper bound on its possible diameter; cf. [8]. Given G in MN with d(G) = 2,
we will see in Lemma 5.1 that vertices in ∆(G) are connected if and only if their images in
∆(G/G′ ) are connected. We note that [1, Thm. 1.1] describes all such generating graphs.
Specifically, that ∆(Z2 ) is connected with infinite diameter and for any other abelian group A
with d(A) = 2 we have that ∆(A) has diameter at most 2. Note that the abelianisation of our
group G cannot be cyclic; cf. Corollary 5.2.
Theorem 1.7. Let G ∈ MN with d(G) = 2 and suppose G′ ≤ K ≤ Φ(G) for some finitely
generated subgroup K.
(i) If G/G′ ∼
= Z2 , then ∆(G) is connected but has infinite diameter.
(ii) If G/G′ ∼
= C2 × C2 , then ∆(G) has diameter 1.
(iii) Otherwise d(G/G′ ) = 2 and ∆(G) has diameter 2.
Note that if G is either a finitely generated nilpotent group or a finitely generated torsion
group, then G′ is finitely generated. However, it is not clear to us whether every G ∈ MN
with d(G) = 2 has a finitely generated subgroup K where G′ ≤ K ≤ Φ(G).
We recall that the total domination number γt (Γ) of a graph Γ is the least size of a set S
of vertices of Γ such that every vertex of Γ is adjacent to a vertex in S. Such a set S of least
size is called a total dominating set for Γ. These concepts have been recently introduced for
finite graphs, and the total domination number of ∆(G) has been studied for G a finite simple
group or a finite 2-generated nilpotent group; see for example [4] and [22] respectively. As for
infinite groups G, not much is known about the total domination number and total dominating
sets for ∆(G), apart for the Tarski monsters, which have a total dominating set of size 2, and
certain finite-index subgroups of the second Houghton group [7, Prop. 5.5], which have no finite
total dominating sets.
Corollary 1.8. Let G ∈ MN with d(G) = 2 and G/G′ ∼ = Cp × Cp for some prime p. Then
γt (∆(G)) = 2. Moreover, any lift of any generating set for G/G′ is a total dominating set for
∆(G).
Note that the hypotheses for the above results apply to branch GGS-groups. Hence we
deduce the following:
3
Corollary 1.9. Let G be a branch GGS-group. Then G is almost 2 -generated, ∆(G) has
diameter 2, and γt (∆(G)) = 2.

Organisation. Section 2 contains preliminary material on groups acting on the p-adic tree,
including the formal definitions of the Grigorchuk groups, the GGS-groups, the multi-EGS
groups and the Šunić groups. In Section 3, we prove Theorem 1.2. In Section 4, we apply the
results from the previous section to prove Theorems 1.1, 1.4, and 1.5. Lastly, in Section 5 the
remaining results are proved.
INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS 5

Acknowledgements. We thank Dominik Francoeur, Benjamin Klopsch and Jeremy Rickard


for very helpful conversations. This research was supported by a London Mathematical Society
Research in Pairs (Scheme 4) grant. The first author would also like to thank the welcoming
staff at Lund University during his two-week stay in Sweden.

2. Preliminaries
Here we recall the notion of branch groups and establish prerequisites for the rest of the
paper. For more information, see [2, 18].

2.1. The p-adic tree and its automorphisms. Let T be the p-adic tree, for a prime p. Let
X = {1, 2, . . . , p} be an alphabet on p letters. The set of vertices of T can be identified with
the free monoid X ∗ , and we will freely use this identification without special mention. The
root of T corresponds to the empty word ∅, and for each word v ∈ X ∗ and letter x, an edge
connects v to vx. There is a natural length function on X ∗ , and the words w of length |w| = n,
representing vertices that are at distance n from the root, form the nth layer of the tree. The
boundary ∂T consisting of all infinite simple rooted paths is in one-to-one correspondence with
the p-adic integers.
For u a vertex, the full rooted subtree of T , that has its root at u and includes all vertices v
with u a prefix of v, is isomorphic to the full tree T .
We observe that every automorphism of T fixes the root, and that the orbits of Aut T on
the vertices of the tree T are precisely its layers. The image of a vertex u under f ∈ Aut T
is denoted by uf . For a vertex u, considered as a word over X, and a letter x ∈ X we have
(ux)f = uf x′ where x′ ∈ X is uniquely determined by u and f . This gives a permutation f (u)
of X so that
(ux)f = uf xf (u) .
The permutation f (u) is called the label of f at u. The automorphism f is called rooted if
f (u) = 1 for all u 6= ∅. The automorphism f is called directed, with directed path ℓ for some
ℓ ∈ ∂T , if the support {u | f (u) 6= 1} of its labelling is infinite and contains only vertices at
distance 1 from ℓ.
The section of f at a vertex u is the unique automorphism fu of T given by the condition
(uv)f = uf v fu for v ∈ X ∗ .

2.2. Subgroups of Aut T . Let G ≤ Aut T . The vertex stabiliser stG (u) is the subgroup
consisting of elements in G that fix the vertex u. For n ∈ N, the nth level stabiliser StG (n) =
∩|v|=n stG (v) is the subgroup of automorphisms that fix all vertices at level n. We emphasise
the difference in the notation of vertex stabilisers and level stabilisers, which aims to avoid
confusion from the fact that the first-level vertices of T are often identified with the elements
of X. Every g ∈ StAut T (1) can be identified with its sections g1 , . . . , gp , via the natural
embedding:
p
ψ : StAut T (1) −→ Aut T × · · · × Aut T
g 7−→ (g1 , . . . , gp )
The subgroup ristG (u), consisting of all automorphisms in G that fix all vertices v of T not
having u as a prefix, is called the rigid vertex stabiliser of u in G. The rigid nth level stabiliser
is the product
Ypn
RistG (n) = ristG (ui ) E G
i=1
of the rigid vertex stabilisers of the vertices u1 , . . . , upn at level n.
6 C. G. COX AND A. THILLAISUNDARAM

Let G be a subgroup of Aut T acting spherically transitively, i.e. transitively on every layer
of T . Then G is a branch group if RistG (n) has finite index in G for every n ∈ N.

2.3. The Grigorchuk groups Gω . The group Gω acts on the binary rooted tree and is
defined via the parameter ω = (ω1 , ω2 , . . . ), such that ωi ∈ {0, 1, 2} for any i ∈ N. We denote
by Ω the set of all such ω and for n ∈ N, let σ n ω be the sequence (ωn+1 , ωn+2 , . . . ). For an
account on these groups see [16] or [17].
For ω ∈ Ω, the Grigorchuk group Gω is generated by a, bω , cω , dω , where a is the rooted
automorphism corresponding to the cycle (1 2), and the directed automorphisms bω , cω , dω lie
in the first level stabiliser and are defined with directed path 22 . . . as follows. For n ∈ N ∪ {0},
let un = 2 . n. . 2 ∈ X ∗ . Then bω (un 1) is trivial if ωn+1 = 2 and is equal to a otherwise; the
automorphism cω (un 1) is trivial if ωn+1 = 1 and is equal to a otherwise; and finally dω (un 1) is
trivial if ωn+1 = 0 and is equal to a otherwise. If ω = {0, 1, 2, 0, 1, 2, . . . }, then Gω is the first
Grigorchuk group.

2.4. The GGS-groups. Let T be the p-adic tree, for an odd prime p. By a we denote
the rooted automorphism, corresponding to the p-cycle (1 2 · · · p) ∈ Sym(p), that cyclically
permutes the vertices at the first level. Given a non-zero vector e = (e1 , e2 , . . . , ep−1 ) ∈ (Fp )p−1 ,
we recursively define a directed automorphism b ∈ StAut T (1) via
ψ(b) = (ae1 , ae2 , . . . , aep−1 , b).
We call the subgroup Ge = ha, bi of Aut T the GGS-group associated to the defining vector e.
If Ge = ha, bi is a GGS-group corresponding to the defining vector e, then G is an infinite
P
torsion group if and only if p−1j=1 ej = 0 in Fp ; compare [18]. Furthermore, it was shown
in [10, Lem. 4.2] and [9, Thm. 3.7] that a GGS-group Ge is branch if and only if e is not the
constant vector. For further information on GGS-groups, we refer the reader to [9, 10].

2.5. The multi-EGS groups. Let T be the p-adic tree, for an odd prime p. As before,
the rooted automorphism a corresponding to the p-cycle (1 2 · · · p) ∈ Sym(p), that cyclically
permutes the vertices at the first level. For j ∈ {1, . . . , p} let rj ∈ {0, 1, . . . , p − 1}, with
rj 6= 0 for at least one index j. We fix the numerical datum E = (E(1) , . . . , E(p) ), where each
(j) (j)
E(j) = (e1 , . . . , erj ) is an rj -tuple of Fp -linearly independent vectors
(j) (j) (j) 
ei = ei,1 , . . . , ei,p−1 ∈ (Fp )p−1 , i ∈ {1, . . . , rj }.
The multi-EGS group associated to E is the group
GE = ha, b(1) , . . . , b(p) i
(j)
= {a} ∪ {bi | 1 ≤ j ≤ p, 1 ≤ i ≤ rj }
(j) (j)
where, for each j ∈ {1, . . . , p}, the generator family b(j) = {b1 , . . . , brj } consists of commuting
(j)
directed automorphisms bi ∈ StAut(T ) (1) along the directed path

∅, (p − j + 1), (p − j + 1)(p − j + 1), . . . ∈ ∂T
that satisfy the recursive relations
 (j) (j) (j) (j) 
(j) (j)
ψ(bi ) = aei,j , . . . , aei,p−1 , bi , aei,1 , . . . , aei,j−1 .
The multi-EGS groups are infinite and act spherically transitively on T . APmulti-EGS
group GE is a torsion group if and only if for every e = (e1 , . . . , ep−1 ) ∈ E, we have p−1
j=1 ej ≡ 0
INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS 7

(mod p); compare [33, Lem. 3.13]. From [33, Lem. 3.3], we have that if E is not the set of
non-zero constant vectors, then the multi-EGS group GE is branch.
2.6. The Šunić groups. Let T be the p-adic tree, for a prime p, and for m ≥ 2 let f (x) =
xm + am−1 xm−1 + · · · + a1 x + a0 be an invertible polynomial over Fp . The Šunić group Gp,f
is generated by the rooted automorphism a corresponding to the p-cycle (1 2 · · · p) ∈ Sym(p),
and by the m directed generators b1 , . . . , bm defined as follows:
ψ(b1 ) = (1, . . . , 1, b2 )
ψ(b2 ) = (1, . . . , 1, b3 )
..
.
ψ(bm−1 ) = (1, . . . , 1, bm )
ψ(bm ) = (1, . . . , 1, b−a0 −a1
1 b2
−am−1
· · · bm ).
From [32, Lem. 1 and 6], the groups Gp,f are branch, and from [11, Cor. 2.13], the non-
torsion Šunić groups G2,f acting on the binary rooted tree contain an element b ∈ hb1 , . . . , bm i
such that ψ(b) = (a, b). In [11, Thm. 8.1], it was shown that for a non-torsion Šunić group G2,f ,
every maximal subgroup of infinite index is conjugate to the maximal subgroup H(q) =
h(ab)q , b1 , . . . , bm i for some odd prime q.

3. On invariable generation and maximal subgroups


Proof of Theorem 1.2. For the forward direction, let S be a generating set for G. If hsigi | si ∈
S, gi ∈ Gi =6 G, then hsigi | si ∈ S, gi ∈ Gi ≤ M for some maximal subgroup M . As all
maximal subgroups are normal, it follows that hSi ≤ M , a contradiction.
For the reverse direction, suppose M is a non-normal maximal subgroup. Then there exists
an x ∈ M such that xg ∈ / M for some g ∈ G. Hence hM, xg i = G, and so the set {xg } ∪ M is
a generating set but not an IG-set. 
We remark that from the proof above, it follows that the “if” statement holds more generally
without the assumption that every proper subgroup lies in a proper maximal subgroup.
We will see in the next section that the non-torsion Šunić groups, which act on the bi-
nary rooted tree and have non-normal maximal subgroups, have finite non-IG generating sets.
However, it is still interesting if the following holds:
Problem 3.1. Can a finitely generated group G ∈
/ MN still satisfy the condition that every
finite generating set is an IG-set?
We remark from the proof above that such a G would need to have all non-normal maximal
subgroups being infinitely generated (and so also not of finite index).
Recall that for a nilpotent group, all maximal subgroups are normal; see [31, Thm. 12.1.5].
Therefore, Theorem 1.2 immediately shows that finitely generated nilpotent groups have every
generating set an IG-set. This was recorded for finite nilpotent groups; see [24, Prop. 2.4],
where a stronger result was achieved which showed that for a finite group, being nilpotent
is equivalent to every generating set being an IG-set. We now see that all nilpotent groups
(without the assumption of being finitely generated) have every generating set being an IG-set.
Proposition 3.2. Let G be a nilpotent group. Then every generating set of G is an IG-set.
Proof. We proceed by induction on c, the nilpotency class of G. If c = 1, then the result
is clear. So suppose the result is true for groups of class at most c − 1, and assume that
8 C. G. COX AND A. THILLAISUNDARAM

G is of class c. Let G be generated by a set X = {xi | i ∈ I}, for some indexing set I.
By the induction hypothesis, the set {xi [xi , gi ]γc (G) | i ∈ I, gi ∈ G}, for gi ∈ G, generates
G/γc (G). Equivalently, for every g ∈ G\γc (G), there is an element hg ∈ hxi [xi , gi ] | i ∈ Ii
and zg ∈ γc (G) such that hg = gzg . Further, since γc (G) is generated by the elements [g, x],
where g ∈ γc−1 (G)\γc (G) and x ∈ G\γ2 (G), we can find hg , hx ∈ hxi [xi , gi ] | i ∈ Ii such
that hg = gzg and hx = xzx , where zg , zx ∈ γc (G). Then [hg , hx ] = [gzg , xzx ] = [g, x], since
γc (G) ≤ Z(G). 
Remark 3.3. The above does not generalise to solvable groups. Take, for example, C2 ⋉(Q×Q)
where the C2 = hti permutes the two copies of Q. This has one maximal subgroup, M = Q × Q
which is normal, but not every generating set is an IG-set. Indeed, {t} ∪ {(q, q ′ ) ∈ Q2 } is an
IG-set whereas {t(q, q ′ ) : q, q ′ ∈ Q} has conjugates which all lie in H = hti ⋉ {(q, q) : q ∈ Q}.
Alternatively, infinitely generated simple groups without maximal subgroups have generating
sets that are not IG-sets.
Further, we record the following additional tool for identifying IG groups, which is of inde-
pendent interest.
Lemma 3.4. Let G be a group and suppose that S ⊆ G is a generating set for G. Write Se for
the subset obtained from S where each element of S is replaced by any conjugate of it. Then S
e ≥ G′ for all choices of S.
is an IG-set if and only if hSi e
e = G ≥ G′ for all choices of S.
Proof. The forward direction is clear, since if hSi = G, then hSi e
e = hsgs | s ∈ Si = hs[s, gs ] | s ∈ Si for some fixed choice
For the reverse direction, since hSi
e ≥ G , then it follows that hSi
of {gs | s ∈ S} ⊆ G, if hSi ′ e = hs | s ∈ Si = G. Since Se was
arbitrary, we have that S is an IG-set, as required. 

4. Applications to certain branch groups


We know from [11] that the non-torsion Šunić groups acting on the binary rooted tree have
maximal subgroups of infinite index. Furthermore, as mentioned in Subsection 2.6, every
maximal subgroup of infinite index is conjugate to H(q) := h(ab)q , b1 , . . . , bm i for some odd
prime q.
Lemma 4.1. Let G be a non-torsion Šunić group acting on the binary rooted tree. Then no
conjugate of ab is in H(q) for any odd prime q.
Proof. Suppose for a contradiction that a conjugate of ab is in H(q) for some odd prime q.
Equivalently, suppose there is a maximal subgroup M of G of infinite index, with ab ∈ M .
For convenience, for a vertex u we write Mu for the restriction of stM (u) to the subtree rooted
at u. Note that ab ∈ Mu for every vertex u; compare [11, Lem. 8.12]. Then by [11, Lem. 8.15],
there is a vertex v of the tree such that Mv = H(q). It follows that ab ∈ H(q), which gives the
desired contradiction. 
Theorem 4.2. Let G = ha, b1 , . . . , bm i, for some m ≥ 2, be a non-torsion Šunić group acting
on the binary rooted tree. Then {a, b1 , . . . , bm } is not an IG-set, but {ab, b1 , . . . , bm } is an
IG-set. Hence G is FIG.
q−1
Proof. For the first claim, note that a(ba) 2 = (ab)q−1 a = (ab)q b ∈ H(q) = h(ab)q , b1 , . . . , bm i.
Hence {a, b1 , . . . , bm } is not an IG-set. Now, by Lemma 4.1 no conjugate of ab is in H(q) for
any odd prime q. Therefore it follows that no conjugate of ab is in any maximal subgroup of
infinite index. It is straightforward that {(ab)g0 , bg11 , . . . , bgmm }, for any g0 , . . . , gm ∈ G, is not
in any maximal subgroup of finite index, since these maximal subgroups are all normal and
contain G′ . Therefore h(ab)g0 , bg11 , . . . , bgmm i = G. 
INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS 9

Hence we have proved Theorem 1.1.


For G a FIG group, let dI (G) denote the minimal cardinality of a set of invariable generators
for G. It was shown in [24, Prop. 2.5] that dI (G) − d(G) can be arbitrarily large for finite
groups. Certainly, from Theorem 1.2, we have that dI (G) = d(G) for any finitely generated
group G ∈ MN . For certain families of infinite groups not in MN , such as the Houghton
groups Hn for n ≥ 2, it was shown in [6, Thm. A] that also d(Hn ) = dI (Hn ). From Corollary 1.3
and Theorem 4.2, the following is immediate.
Corollary 4.3. Let G be either a non-torsion Šunić group acting on the binary rooted tree, a
torsion Grigorchuk group, a branch GGS-group or a torsion multi-EGS group. Then dI (G) =
d(G).
We end this section with the proofs of the results concerning subgroups.

Proof of Theorem 1.4. Let G be a torsion multi-EGS group. By [26, Cor. 5.7], every finite-
index subgroup of G has only finitely many maximal subgroups, all of finite index. Then by
[25, Lem. 2.6(ii)], we are done. 

Proof of Theorem 1.5. Let G be either the first Grigorchuk group or any torsion GGS-group.
By [19, Thm. 1] and [12, Thm. C], the group G has the subgroup induction property. Then by
[12, Thm. B], a finitely generated subgroup of G has only maximal subgroups of finite index,
and also only finitely many such maximal subgroups. Then by [25, Lem. 2.6(ii)], the result
follows. 

5. Almost 32 -generation
Recall that a group G is almost 23 -generated if whenever g ∈ G is such that gG′ ∈ G/G′
is part of a generating pair in G/G′ , then g is part of a generating pair in G. To prove
Theorem 1.6, we shall use an auxiliary result below.
Let G be a group and S ⊆ G. Then S weakly generates G if the image of S in G/G′ is
a generating set for G/G′ . The following result uses [31, Thm. 5.2.12] which says that, for a
group G, the Frattini subgroup Φ(G) is equal to the set of non-generators of G; i.e. if {g} ∪ S
is a generating set for a group G, with g ∈ Φ(G), then hSi = G.
Lemma 5.1. Let G be a group with G′ ≤ Φ(G) and suppose G′ ≤ K ≤ Φ(G) for some finitely
generated subgroup K. Let S ⊆ G. Then S weakly generates G if and only if S generates G.
Proof. Suppose S weakly generates G. Let X denote a finite generating set for K and let π
denote the abelianisation map from G. Take g ∈ G. Then π(g) = w(π(s1 ), . . . , π(sn )) is a
word in the images of some s1 , . . . , sn ∈ S. Consider ŵ := w(s1 , . . . , sn ). Then π(g) = π(ŵ)
and, since ker π = G′ , we have that g = ŵσ where σ ∈ G′ ≤ hXi. Hence g ∈ hS ∪ Xi. But
X ⊂ Φ(G) and so, since X is finite, we therefore have that g ∈ hSi. Since g was arbitrary, it
follows that S generates G. The reverse direction is immediate. 

For convenience, we also note an immediate consequence of the above.


Corollary 5.2. Let G be a group with G′ ≤ Φ(G) and suppose G′ ≤ K ≤ Φ(G) for some
finitely generated subgroup K. Then d(G) = d(G/G′ ).
Proof of Theorem 1.6. From [28, Thm. A], the condition that G is in MN is equivalent to
G′ ≤ Φ(G). Hence Lemma 5.1 applies and every lift of a generating set of G/G′ must be a
generating set of S, i.e. G is almost 32 -generated. 
10 C. G. COX AND A. THILLAISUNDARAM

Lemma 5.1 can also be interpreted in terms of the generating graph. It states that g and h
are connected in ∆(G) if and only if their images in G/G′ are connected. Therefore to study
the diameter of the generating graphs of groups for which every generating set of G is an
IG-set, it is sufficient to study the generating graphs of the possible abelianisations. The result
of Theorem 1.7 then immediately follows from [1, Thm. 1.1], where these generating graphs
were studied. The remaining corollaries are also then immediate.
Remark 5.3. For a group G, let
G∗ = {g ∈ G | g ∈
/ N where G/N is non-cyclic} = {g ∈ G | G/hgiG is cyclic}.
It is conceivable that one could alternatively define almost 32 -generated as follows: for every
g ∈ G∗ , the element g belongs to a generating pair hg, hi, where h ∈ G∗ too. Note that for
Thompson’s group F , not all elements in F ∗ are part of a generating pair. This is because the
only non-cyclic quotient is F/F ′ ∼
= Z × Z, so F ∗ = F \F ′ , and not all elements in Z × Z are
part of generating pairs. However, groups in MN with abelianisation Cp × Cp would satisfy
this alternative definition.

References
[1] C. Acciarri and A. Lucchini, The generating graph of infinite abelian groups, Bull. Aust. Math. Soc. 100
(1) (2019), 68–75.
[2] L. Bartholdi, R. I. Grigorchuk and Z. Šuniḱ, Branch groups, in Handbook of algebra 3, North-Holland,
Amsterdam, 2003.
[3] T. C. Burness, R. M. Guralnick and S. Harper, The spread of a finite group, Ann. of Math. 193 (2) (2021),
619–687.
[4] T. C. Burness and S. Harper, Finite groups, 2-generation and the uniform domination number, Israel J.
Math. 239 (2020), 271–367.
[5] C. G. Cox, Invariable generation and wreath products, J. Group Theory 24 (1) (2021), 79–93.
[6] , Invariable generation and the Houghton groups, J. Algebra 598 (2022), 120–133.
[7] , On the spread of infinite groups, Proc. Edin. Math. Soc. 65 (1) (2022), 214–228.
[8] E. Crestani and A. Lucchini, The non-isolated vertices in the generating graph of a direct powers of simple
groups, J. Algebraic Combin. 37 (2) (2013), 249–263.
[9] G. A. Fernández-Alcober, A. Garrido and J. Uria-Albizuri, On the congruence subgroup property for GGS-
groups, Proc. Amer. Math. Soc. 145 (8) (2017), 3311–3322.
[10] G. A. Fernández-Alcober and A. Zugadi-Reizabal, GGS-groups: Order of congruence quotients and Haus-
dorff dimension, Trans. Amer. Math. Soc. 366 (2014), 1993–2017.
[11] D. Francoeur and A. Garrido, Maximal subgroups of groups of intermediate growth, Adv. Math. 340 (2018),
1067–1107.
[12] D. Francoeur and P.-H. Leemann, Subgroup induction property for branch groups, arXiv preprint:
2011.13310.
[13] D. Francoeur and A. Thillaisundaram, Maximal subgroups of non-torsion GGS-groups, Canad. Math. Bull.
65 (4) (2022), 825–844.
[14] G. Goffer and N. Lazarovich, Invariable generation does not pass to finite index subgroups, Groups Geom.
Dyn. 16 (2022), 1267–1288.
[15] G. Golan Polak, Thompson’s group F is almost 23 -generated, Bull. London Math. Soc., to appear.
[16] R. I. Grigorchuk, On Burnside’s problem on periodic groups, Funktsional. Anal. i Prilozhen 14 (1) (1980),
53–54.
[17] R. I. Grigorchuk, Degrees of growth of finitely generated groups and the theory of invariant means, Izv.
Akad. Nauk SSSR Ser. Mat. 48 (5) (1984), 939–985.
[18] R. I. Grigorchuk, Just infinite branch groups, in: New horizons in pro-p groups, Birkhäuser, Boston, 2000.
[19] R. I. Grigorchuk and J. S. Wilson, A structural property concerning abstract commensurability of subgroups,
J. London Math. Soc. 68 (2) (2003), 671–682.
[20] N. Gupta and S. Sidki, On the Burnside problem for periodic groups, Math. Z. 182 (3) (1983), 385–388.
[21] P. Hall, On the finiteness of certain soluble groups, Proc. London Math. Soc. 9 (3) (1959) 595–622.
INVARIABLE GENERATION OF CERTAIN BRANCH GROUPS 11

[22] S. Harper and A. Lucchini, Connectivity of generating graphs of nilpotent groups, Algebr. Comb. 3 (5)
(2020), 1183–1195.
[23] C. Jordan, Recherches sur les substitutions, J. Math. Pures Appl. 17 (1872), 351–367.
[24] W. M. Kantor, A. Lubotzky and A. Shalev, Invariable generation and the Chebotarev invariant of a finite
group, J. Algebra 348 (2011), 302–314.
[25] W. M. Kantor, A. Lubotzky and A. Shalev, Invariable generation of infinite groups, J. Algebra 421 (2015)
296–310.
[26] B. Klopsch and A. Thillaisundaram, Maximal subgroups and irreducible representations of generalised
multi-edge spinal groups, Proc. Edin. Math. Soc. 61 (3) (2018), 673–703.
[27] A. Minasyan, Some examples of invariably generated groups, Israel J. Math. 245 (2021), 231–257.
[28] A. Myropolska, The class MN of groups in which all maximal subgroups are normal, arXiv preprint:
1509.08090.
[29] E. L. Pervova, Everywhere dense subgroups of a group of tree automorphisms, Tr. Mat. Inst. Steklova 231
(Din. Sist., Avtom. i. Beskon. Gruppy) (2000), 356–367.
[30] E. L. Pervova, Maximal subgroups of some non locally finite p-groups, Internat. J. Algebra Comput. 15
(5-6) (2005), 1129–1150.
[31] D. J. S. Robinson, A Course in the Theory of Groups, Springer, New York, 1996.
[32] Z. S̆unić, Hausdorff dimension in a family of self-similar groups, Geom. Dedicata 124 (2007), 213–236.
[33] A. Thillaisundaram and J. Uria-Albizuri, The profinite completion of multi-EGS groups, J. Group Theory
24 (2021), 321–357.
[34] J. Wiegold, Transitive groups with fixed-point-free permutations, Arch. Math. (Basel) 27 (1976), 473–475.

Charles Garnet Cox: School of Mathematics, University of Bristol, Bristol BS8 1UG, United
Kingdom
Email address: charles.cox@bristol.ac.uk

Anitha Thillaisundaram: Centre for Mathematical Sciences, Lund University, 223 62 Lund,
Sweden
Email address: anitha.thillaisundaram@math.lu.se

You might also like