You are on page 1of 15

Torsion-free and distal dp-minimal groups

Atticus Stonestrom (Email: atticusstonestrom@yahoo.com)

Abstract: Let G be a dp-minimal group. We discuss two different hypotheses on G; first we show
that, if G is torsion-free, then it is abelian. Then we investigate the structure of G when it admits
a distal f-generic type, showing in particular that the quotient of G by its FC-center can then be
naturally equipped with the structure of a valued group. As an application of this, we show that,
in this case, G is virtually nilpotent.
arXiv:2308.04209v1 [math.LO] 8 Aug 2023

Notation: Throughout G will be a group definable in an NIP theory T . We take commutators


to associate to the left, so that [a, b] = a−1 b−1 ab and [a1 , . . . , an+1 ] = [[a1 , . . . , an ], an+1 ]. We
use ∼ to denote the relation of being conjugate in G. Conjugation is for us a right action, with
ab = b−1 ab for all a, b ∈ G. We use one piece of non-standard notation: given sets A1 , . . . , An , we
write [A1 , . . . , An ] to denote the set of commutators {[a1 , . . . , an ] : ai ∈ Ai }.

1 Introduction
‘Dp-minimality’, first introduced in [14] and [15] and then later isolated as a notion in [12], is a
kind of abstract ‘one-dimensionality’ condition which has proved a very fruitful strengthening of
the NIP, both from the perspective of ‘pure’ model theory and from the perspective of its algebraic
implications. Here we are concerned with dp-minimal groups. The typical examples, including for
instance superstable groups of U -rank 1, and the multiplicative and additive groups of R, Qp , and
ACVFs, are all virtually abelian, but there exist dp-minimal groups which are not virtually abelian
[20], and it is an open question, asked for example in [2], whether every dp-minimal group is virtually
nilpotent. This is known to be the case under several stronger hypotheses. For instance, C-minimal
valued groups are virtually nilpotent [21], ω-categorical dp-minimal groups are virtually abelian
[5], profinite dp-minimal groups are virtually abelian [4], and left-ordered dp-minimal groups are
abelian [6].
In this paper we will give a positive answer in two other cases. First we will show that any
torsion-free dp-minimal group is abelian; this is the content of Section 4. We will then show that
any dp-minimal group G which admits a distal f-generic type is virtually nilpotent. This is deduced
from a more general analysis, in which we construct a valued group structure on the quotient of
G by its FC-center in such a way that the structure inherited by G00 /Z(G00 ) is plain; this is the
content of Section 5.

2 Preliminaries
2.1 Group theory
Let (H, ·) be a group.

2.1.1 Commutators
Throughout this paper we will use the non-standard notation that, for subsets A1 , . . . , An ⊆ H,
[A1 , . . . , An ] denotes the set of commutators {[a1 , . . . , an ] : ai ∈ Ai }; when Ai = {a} is a singleton
we will just write [A1 , . . . , Ai−1 , a, Ai+1 , . . . , An ]. The following is standard:

Lemma 2.1. Suppose A1 , . . . , An ⊆ H are each fixed setwise under conjugation by elements of H.
Then, for any k ∈ {0} ∪ [n], the set of g ∈ H such that [A1 , . . . , Ak , g, Ak+1 , . . . , An ] = {1} is a
normal subgroup of H.

1
Proof. Noting that [A1 , . . . , Ak , g] = [g −1 , [A1 , . . . , Ak ]g ], and that [A1 , . . . , Ak ] is fixed setwise
under conjugation by elements of H, it suffices to show the claim when k = 0, ie to show that the
set of g ∈ H with [g, A1 , . . . , An ] = {1} is a normal subgroup of H. This follows by induction on n;
in the case n = 1 the group is just CH (A1 ).
For the inductive step, let U denote the set of u ∈ H with [u, A2 , . . . , An ] = {1} and let V
denote the set of v ∈ H with [v, A1 , . . . , An ] = {1}. Since [v h , A1 , . . . , An ] = [v, A1 , . . . , An ]h ,
V is fixed setwise under conjugation by elements of H. Now, by the inductive hypothesis, U is
−1
a normal subgroup of H. If g ∈ V , then [g, A1 ] ⊆ U , whence [g −1 , A1 ] = [g, A1 ]−g ⊆ U , so
−1 h
that g ∈ V . Finally, if g, h ∈ V , then [g, A1 ], [h, A1 ] ⊆ U , whence [g, A1 ] [h, A1 ] ⊆ U ; since
[gh, A1 ] ⊆ [g, A1 ]h [h, A1 ] we thus have gh ∈ V , as desired. ■

2.1.2 Torsion-free groups


We will need some facts about torsion-free groups; see for example 5.2.19 and 14.5.9 in [13].

Fact 2.2. Suppose H is nilpotent and torsion-free. Then H/Z(H) is torsion-free.

Fact 2.3. If H is torsion-free and [H : Z(H)] is finite, then H is abelian.

The following is an easy computation which we put here for convenience.

Lemma 2.4. Suppose H is abelian-by-torsion and that a ∈ H is a non-torsion element. If i, j ∈ Z


are such that ai ∼ aj , then |i| = |j|.

Proof. Let g ∈ H be such that (ai )g = aj . For notational convenience, let f : H → H denote the
conjugation-by-g map. Since H is abelian-by-torsion, there is some n > 0 such that f n (an ) = an .
Let b = an ; then b is still non-torsion, and we have f (b)i = f (ai )n = ajn = bj . By induction on k
k k
it follows that f k (b)i = bj for all k ⩾ 1, since
k+1 k k k k+1
f k+1 (b)i = f (f k (b)i )i = f (bj )i = (f (b)i )j = bj .
n n n
On the other hand, we have b = f n (b), and hence bi = f n (b)i = bj . Since b is non-torsion, thus
in = j n , so that indeed |i| = |j|. ■

2.1.3 Valued groups


Let (P, ⩽) be a totally ordered set with a greatest element. An P-valued group is a group (H, ·)
equipped with a surjective map v : H → P satisfying the following:
1. v(a) = max P if and only if a = 1.
2. v(ag ) < v(bg ) for all g ∈ H and a, b ∈ H with v(a) < v(b).
3. v(a) = v(a−1 ) for all a ∈ H
4. v(ab) ⩾ min{v(a), v(b)} for all a, b ∈ H.
Following [21], we will say that (H, v) is ‘plain’ if we additionally have v([a, b]) > max{v(a), v(b)}
for all a, b ∈ H \ {1}. The following is proved in [21]:

Fact 2.5. Suppose (H, v) is a plain valued group of finite exponent. Then H is locally nilpotent.

2.2 Model theory


2.2.1 NIP groups
Assume in this section that G is NIP. We will use two general facts about NIP groups; the first
follows from the Baldwin-Saxl theorem [1] and the second is a result of Shelah [16].

2
Fact 2.6. Then there is a constant n such that, for any finite subset B ⊂ G(C), there is S ⊆ B of
size at most n with CG (S) ⊆ CG (b) for all b ∈ B. In particular, if (ui )i∈[n] + (ai )i∈ω + (vi )i∈[n] is
indiscernible and W = {ui , vi : i ∈ [n]}, then CG (W ) ⊆ CG (ai ) for all i ∈ ω.

Fact 2.7. There is a partial type G00 (x), without parameters, such that G00 type-defines a bounded
index subgroup of G, and such that any bounded index subgroup of G type-definable with parameters
contains G00 .

We will also need some of the theory of definably amenable NIP groups, as developed in [8],
[9], and [3]. Recall that G is said to be ‘definably amenable’ if, for some (any) M ≼ C, there is
a translation-invariant Keisler measure on G(M ). Definably amenable NIP groups admit a good
notion of genericity: a C-definable subset D ⊆ G is said to be left (resp. right) f-generic if there
do not exist k ∈ ω and a sequence (gi )i∈ω such that the family (gi D)i∈ω (resp. (Dgi )i∈ω ) is k-
inconsistent. A partial type is said to be left (right) f-generic if it implies only left (right) f-generic
formulas. Now the following is from [3]:

Fact 2.8. A type q(x) ∈ SG (C) is left (right) f-generic if and only if it is invariant under left (right)
translation by all elements of G00 (C).

It is proved in [9] that, if G is definably amenable, then for any small M ≺ C there is a left
f-generic type p(x) ∈ SG00 (C) invariant over M . Then the type p−1 is right f-generic, and using
Fact 2.8 it follows that the pushforward of px ⊗ p−1
y under the map (x, y) 7→ xy is bi-f-generic, so
one has:

Fact 2.9. If G is definably amenable, then for any small M ≺ C there is a bi-f-generic M -invariant
type q(x) ∈ SG00 (C).

2.2.2 Dp-minimality
There are numerous equivalent characterizations of dp-minimality; see for example [7] for an intro-
duction. A partial type π(x) over a small parameter set is said to be ‘dp-minimal’ if the following
holds for some (any) parameters B over which π is defined: for any infinite sequences I, J mutually
indiscernible over B, and any realization a |= π(x), one of I and J is indiscernible over (a, B). Then
π(x) is not dp-minimal if and only if there are L(C)-formulas ϕ(x, y) and ψ(x, z) and tuples bi ∈ Cy
and cj ∈ Cz such that, for all i, j ∈ ω, the partial type
^ ^
π(x) ∧ ϕ(x, bi ) ∧ ψ(x, cj ) ∧ ¬ϕ(x, bk ) ∧ ¬ψ(x, cl )
k̸=i l̸=j

is consistent. We say that the definable group G is dp-minimal if the formula x ∈ G is dp-minimal.
The following was noted in [18] for definable subgroups of dp-minimal groups, and the analogue
is true for invariant subgroups:

Fact 2.10. Suppose G is dp-minimal, and that A, B ⩽ G are subgroups invariant over some small
model M . Then A ∩ B has bounded index in one of A and B.

Proof. Suppose for contradiction that A ∩ B has unbounded index in both A and B. Then there
are sequences I = (ai )i∈ω in A and J = (bj )j∈ω in B, each indiscernible over M , and such that
ai B ∩ aj B = Abi ∩ Abj = ∅ for all i ̸= j. By Ramsey and M -invariance of A, B we may assume
that I and J are mutually indiscernible over M , and then by dp-minimality one of I and J must
be indiscernible over (M, a0 b0 ). In the former case, there is an automorphism fixing (M, a0 b0 ) and
taking a0 to a1 , and since a−1 −1
0 (a0 b0 ) ∈ B it follows by invariance that a1 (a0 b0 ) ∈ B. But then
−1 −1
a1 a0 ∈ B, a contradiction. A symmetric argument, using that (a0 b0 )b0 ∈ A, gives the result in
the latter case. ■

We will also use the following, from [18], [22], and [11]:

3
Fact 2.11. Any dp-minimal group is abelian-by-torsion. In fact, if G is dp-minimal, then there is
n > 0 such that, for all a, b ∈ G, one of [an , b] = 1 and [a, bn ] = 1 holds.

Fact 2.12. Any dp-minimal group is definably amenable.

Fact 2.13. If G is dp-minimal and q(x) ∈ SG (C) is left or right f-generic, then q is M -invariant
for every small M ≺ C.

2.2.3 Distal types


Distality was introduced in [19] to isolate ‘order-like’ behavior in NIP theories. Here we will be
concerned with distal types, which are one of several ways of treating the notion; see [17] for an
introduction to distality emphasizing this perspective. Recall that, for a parameter set B ⊂ C and
disjoint tuples of variables x, y, two types p(x) ∈ Sx (B) and q(y) ∈ Sy (B) are said to be ‘weakly
orthogonal’, denoted p(x) ⊥w q(y), if p(x) ∧ q(y) determines a unique complete type in S(x,y) (B).
Now a B-invariant global type p(x) ∈ Sx (C) is said to be ‘distal over B’ if the following holds: for
any tuple c, if I |= p⊗ω |(B,c) then tp(c/B, I) ⊥w p|(B,I) . It is proved in [17] that, if p(x) is a global
type invariant over sets B, C ⊂ C, then p(x) is distal over B if and only if it is distal over C. So
one may speak of a ‘distal type’ without ambiguity.
Although we will not use the following for our main results, it will be convenient to cite in some
later discussion; see Proposition 9.13 in [17]:

Fact 2.14. A global dp-minimal type invariant over some small model is either generically stable
or distal.

3 Generic Nilpotence
In this section assume that G is NIP.
It is well known that, in a stable group, to check virtual nilpotence of class at most (c − 1) it
suffices to check that an independent c-tuple of principal generics satisfies [x1 , . . . , xc ] = 1; see eg
[23]. We point out here an analogous fact in the NIP setting, which follows quickly from Fact 2.7.

Lemma 3.1. Suppose there is a left or right f-generic type p(x) ∈ SG00 (C) invariant over some
small model and such that p⊗c (x1 , . . . , xc ) ⊢ [x1 , . . . , xc ] = 1. Then G has a definable finite-index
subgroup that is nilpotent of class at most (c − 1).

Proof. By induction on c. In the case c = 0, G must be finite, so the result follows. Now sup-
pose the result holds for c, and that p(x) ∈ SG00 (C) is an M -invariant f-generic type satisfying
p⊗(c+1) (x1 , . . . , xc+1 ) ⊢ [x1 , . . . , xc+1 ] = 1. Let q(y) be the pushforward of p⊗c (x1 , . . . , xc ) under
the map Gc → G, (x1 , . . . , xc ) 7→ [x1 , . . . , xc ]; then p(x) ⊗ q(y) ⊢ [x, y] = 1.
If b |= q|M and a |= p(M,b) , then a ∈ CG (b); since tp(a/b) is f-generic it follows that CG (b) must
have finite index in G. In particular, G00 (x) ⊢ x ∈ CG (b), so by compactness there is a ∅-definable
set D ⊇ G00 such that b ∈ CG (D). Since D is ∅-definable, thus q(y) ⊢ y ∈ CG (D).
Now, let H = CG (CG (D)). Then H ⊇ D ⊇ G00 ; thus H has finite index in G, so it suffices
to show that H has a definable finite-index subgroup that is nilpotent of class at most c, and we
moreover have p(x) ⊢ x ∈ H and q(y) ⊢ y ∈ H. Since we also have q(y) ⊢ y ∈ CG (D), it follows
that q(y) ⊢ y ∈ Z(H).
Let p̄ be the pushforward of p under the quotient map H → H/Z(H); this is well-defined since
p(x) ⊢ x ∈ H, and by the above we have p̄⊗c (x1 , . . . , xc ) ⊢ [x1 , . . . , xc ] = 1. Since p is concentrated
on H 00 and the pullback of a type-definable subgroup of bounded index under a quotient map is
type-definable and bounded index, p̄ is concentrated on (H/Z(H))00 . Moreover, the image of an
f-generic set under a quotient map is still f-generic, so p̄ is f-generic. By the inductive hypothesis
applied to the interpretable group H/Z(H), it follows that H/Z(H) has a finite-index definable
subgroup that is nilpotent of class at most (c − 1), and taking its pullback to H gives the result. ■

4
4 Torsion-free case
In this section assume that G is dp-minimal.

Lemma 4.1. Suppose that I = (ai )i∈ω is a sequence in G such that (i) ai ̸∼ aj for all i ̸= j, and
̸ j and
(ii) [ai , aj ] = 1 for all i, j. Then, for every sequence J = (bj )j∈ω in G, there is some i =
bi bj
k ∈ ω with [ak , ak ] = 1.
b
Proof. Suppose that J is otherwise, so that [abki , akj ] ̸= 1 for all i ̸= j and k ∈ ω. By Ramsey
and compactness, we may then assume that I and J are mutually indiscernible sequences. Now,
let n be given Fact 2.6. Pick ui , vi so that (ui )i∈[n] + I + (vi )i∈[n] is still J-indiscernible and let
W = {ui , vi : i ∈ [n]}; then I and J are mutually indiscernible over W . Moreover, D := CG (W ) is
W -definable, and by condition (ii) and Fact 2.6 we have ak ∈ D ⊆ CG (ak ) for all k ∈ ω.
Now, by condition (i), we have ai ∼ ab00 iff i = 0, so I is not indiscernible over ab00 and hence
not over (W, ab00 ). Thus, by dp-minimality, J must be indiscernible over (W, ab00 ). Since a0 ∈ D, we
have b0 |= ab00 ∈ Dy , and thus also ab00 ∈ Db1 . But D ⊆ CG (a0 ), so that ab00 ∈ CG (a0 )b1 = CG (ab01 ),
ie [ab00 , ab01 ] = 1, a contradiction. ■

Lemma 4.2. Suppose that a ∈ G(C) is non-torsion, and that J = (gi )i∈ω is an a-indiscernible
sequence in G. Then [ag0 , ag1 ] = 1.

Proof. Let s ⊆ ω be the set of all k ∈ ω such that [akg0 , akg1 ] ̸= 1; since J is a-indiscernible, note
that, if k ∈ s, then [akgi , akgj ] ̸= 1 for all i ̸= j. On the other hand, we have [ai , aj ] = 1 for all
i, j, and by Lemma 2.4 and Fact 2.11 we have ai ̸∼ aj for all i ̸= j ∈ ω, so it follows from Lemma
4.1 that s must be finite. On the other hand, letting b = ag0 and c = ag1 , we have [bk , ck ] = 1 for
all k ∈/ s, and it follows that [b, c] = 1. (For example, we may find four pairwise coprime numbers
p, q, r, s ∈ ω \ s. Then bp and bq each lie in CG (cpq ), so b ∈ CG (cpq ), and likewise b ∈ CG (crs ). Now
cpq and crs each lie in CG (b), so c ∈ CG (b), as needed.) ■

Lemma 4.3. Let M be a small model. Suppose that p(x) ∈ SG (C) is an M -invariant type whose
realizations are non-torsion, and that q(y) ∈ SG00 (C) is an M -invariant right f-generic type. Then
q(y) ⊗ p(x) and p(x) ⊗ q(y) each imply [x, xy ] = 1.

Proof. To see that q(y) ⊗ p(x) implies [x, xy ] = 1, let a |= p|M and let J = (gi )i∈ω |= q ⊗ω |(M,a) .
−1
Then J is a-indiscernible, so by Lemma 4.2 [ag0 , ag1 ] = 1 and hence [a, ag1 g0 ] = 1. But q is right
f-generic and concentrated on G00 , so g1 g0−1 |= q|(M,a) , and the result follows.
To see that p(x) ⊗ q(y) implies [x, xy ] = 1, let J = (gi )i∈ω |= q ⊗ω |M and let a |= p|(M,J) . Since p
is M -invariant and J is M -indiscernible, also J is (M, a)-indiscernible. So again by Lemma 4.2 we
−1
have [a, ag1 g0 ] = 1. Also, again by right f-genericity, g1 g0−1 |= q|M , and now since a |= q|(M,g1 g−1 )
0
the result follows. ■

Lemma 4.4. Suppose there is an M -invariant bi-f-generic type concentrated on G00 whose real-
izations are non-torsion. Then there is a definable finite-index subgroup H ⩽ G that is nilpotent of
class at most 2.

Proof. Let q(x) ∈ SG00 (C) be bi-f-generic with non-torsion realizations, and let (ai )i∈ω + (bi )i∈ω
be a Morley sequence of q over M . By Lemma 4.3, we have [a1 , ab11 ] = 1 and [b1 , ba1 1 ] = 1. In
other words, [a1 , b1 ] ∈ CG (a1 ) ∩ CG (b1 ), so that ⟨a1 , b1 ⟩ is nilpotent of class at most 2. Now, by
dp-minimality, one of (ai )i∈ω and (bi )i∈ω is indiscernible over [a1 , b1 ]. If (bi )i∈ω is, then, since
[a1 , b1 , b1 ] = 1, we have [a1 , b1 , b2 ] = 1, so the result follows by Lemma 3.1. So we may assume
instead that (ai )i∈ω is indiscernible over [a1 , b1 ]. Since [a1 , b1 , a1 ] = 1, thus also [a1 , b1 , a2 ] = 1;
since q is bi-f-generic and concentrated on G00 , and a2 |= q|(M,a1 ) , we have aa2 1 |= q|(M,a1 ) , and
since b1 |= q|(M,a1 ,a2 ) it follows that [a1 , b1 , aa2 1 ] = 1.

5
b−1
Now, again by dp-minimality, one of (ai )i∈ω and (bi )i∈ω is indiscernible over [b1 , a11 ]. If
b−1 b−1
(bi )i∈ω is, then, since [b1 , a11 , b1 ] = 1, also [b1 , a11 , b2 ] = 1, so that [b1 , a1 , bb21 ] = 1. Since q is
bi-f-generic and concentrated on G00 , and b2 |= q|(M,a1 ,b1 ) , we have bb21 |= q|(M,a1 ,b1 ) , so it follows
that [b1 , a1 , b2 ] = 1. But then [a1 , b1 , b2 ] = 1 and so again by Lemma 3.1 we are done. So we
b−1 b−1
may assume that (ai )i∈ω is indiscernible over [b1 , a11 ]. Since [b1 , a11 , a1 ] = 1, it follows that
b−1
[b1 , a11 , a0 ] = 1, and hence that [b1 , a1 , ab01 ] = 1.
Altogether we have shown that, if the desired conclusion does not hold, then [a, c, ba ] = 1 and
[c, b, ac ] = 1 for all (a, b, c) |= q ⊗3 |M . By the Hall-Witt identity, we have

[c, b, ac ][a, c, ba ][b, a, cb ] = 1,

so it follows that [b, a, cb ] = 1. By one final application of bi-f-genericity of q, we have cb |= q|(M,a,b) ,


so that [b, a, c] = 1, whence [a, b, c] = 1 and the result follows from Lemma 3.1. ■

Theorem 4.5. If G is torsion-free, then it is abelian.

Proof. By Fact 2.12 and Fact 2.9, let q(x) ∈ SG00 (C) be an M -invariant bi-f-generic type, and
let (a, b) |= q ⊗2 |M . As in Lemma 4.4, we have that ⟨a, b⟩ is nilpotent of class at most 2, so that
[ak , b] = [a, b]k = [a, bk ] for all k ∈ ω. On the other hand, by Fact 2.11, for some n > 0, one of
[an , b] = 1 and [a, bn ] = 1 holds, whence [a, b]n = 1. Since G is torsion-free, thus [a, b] = 1, and by
Lemma 3.1 it follows that G00 is abelian.
Now, to show the desired result, it suffices by Fact 2.3 to show that G00 ⊆ Z(G). Thus fix
arbitrary g |= G with g ̸= 1. Let N be any small model containing (M, g), and let c |= q|N . By
Lemma 4.3 applied to the N -invariant type tp(g/N ), we have [g, g c ] = 1. On the other hand,
c |= G00 , and G00 is a normal abelian subgroup of G, so also [c, cg ] = 1. Thus ⟨c, g⟩ is nilpotent of
class at most 2, and so again since G is torsion-free and by Fact 2.11 it follows that [c, g] = 1. So
c ∈ CG (g), and by f-genericity of q it follows that CG (g) has finite-index in G, whence G00 ⊆ CG (g).
Since g was arbitrary, indeed G00 ⊆ Z(G), and the result follows. ■

5 Distal case
We will now turn to dp-minimal groups admitting a distal f-generic type. This includes all dp-
minimal groups definable in a distal structure, ie in a structure in which every invariant type is
distal, but also includes more general examples too. For instance, if P is a total order with no least
element, then any dp-minimal P-valued group will have distal f-generic types. (Indeed no f-generic
type in such a group can be generically stable, and so by Fact 2.14 the claim follows.)
Let us make a few initial remarks. First note that, if G is any definably amenable NIP group
with a generically stable f-generic type, then every global f-generic type of G is generically stable.
Indeed, suppose p(x) ∈ SG (C) is (say) left f-generic and generically stable. Let M be a small model
containing a representative of every coset of G00 and over which p(x) is invariant. By Fact 2.8,
every left translate of p(x) is of form gp(x) for some g ∈ G(M ), whence every left translate of p(x)
is generically stable and M -invariant and hence finitely satisfiable in M . It follows that G is fsg.
By Fact 2.8 and item (iii) of Corollary 3.3 in [10], it follows that a global type of G is left or right
f-generic if and only if it is bi-generic; call this (∗). By (∗), p is generic, so G is generically stable
in the sense of Section 6 of [9], and by the results proved there the global bi-generic types of G
are precisely the translates of p. Again by (∗), it follows that the left or right f-generic types are
precisely the translates of p, so the claim follows.
By Fact 2.13 and Fact 2.14, it follows that, if G is dp-minimal, then either every left/right f-
generic type of G is generically stable, or every left/right f-generic type of G is distal. In particular,
if G has any distal left/right f-generic type, then all of its left/right f-generic types are distal. For
the rest of the section, let us assume that G has this property. Using Fact 2.9 and Fact 2.13, there

6
is then a bi-f-generic type q(x) ∈ SG00 (C) which is distal and invariant over every small model; we
will fix this type q for the entire section.
Many of the results we will prove here were proved for C-minimal valued groups in [21]. In
particular, the results of Section 5.3 and Theorem 5.22 were both proved in the C-minimal case
there. The main difference here is that we do not start with a valued group structure on the group
G, and so must produce one just from the group structure.

5.1 Conjugacy classes are thin


Recall that a C-definable subset X ⊆ G with 1 ∈ X = X −1 is said to be ‘thick’ if, for any sequence
(ai )i∈ω indiscernible over the parameters defining X, we have a−1
i aj ∈ X for all i, j ∈ ω. In this
section we will show that, for any u ∈ G \ {1}, the set G \ (uG ∪ u−G ) is thick.

Lemma 5.1. Let B ⊂ C be a parameter set. Suppose u, g |= G are such that tp(u/B) ⊥w tp(g/B)
and that tp(u/B) ⊥w tp(ug /B), and that there is some n > 0 with [g n , u] = 1. If [g, u] ≡B u, then
u = 1.

Proof. Suppose [g, u] ≡B u. We prove by induction on k that [g k , u] ≡B u for each k > 0; taking
k = n will then give the result. The base case k = 1 is by hypothesis. For the inductive step,
assume that [g k , u] ≡B u. Then, since tp(u/B) ⊥w tp(g/B), it follows that [g k , u]g ≡B ug . By the
case k = 1, we also have [g, u] ≡B u. Now since tp(u/B) ⊥w tp(ug /C), the 2-tuple ([g, u], ug ) has
the same type over B as ([g, u], [g k , u]g ) does. But we have ug · [g, u] = u, and so the result follows
from the identity [g k , u]g [g, u] = [g k+1 , u]. ■
00
Lemma 5.2. Let M be a small model, and suppose that v ∈ G(M ) and that u ∈ v G . If J + g |=
q ⊗(ω+1) |(M,u) , then tp(u/M, J) ⊥w tp(g/M, J) and tp(u/M, J) ⊥w tp(ug /M, J).

Proof. That tp(u/M, J) ⊥w tp(g/M, J) follows immediately from distality of q. Now, let h |= G00
be such that u = v h , and let J ′ + g ′ |= q ⊗(ω+1) |(M,u,h) . We have (u, J ′ , g ′ ) ≡M (u, J, g), so

it suffices to show that tp(u/M, J ′ ) ⊥w tp(ug /M, J ′ ). Since h |= G00 and g ′ |= q|(M,u,h,J ′ ) ,

it follows by left f-genericity of q that hg |= q|(M,u,h,J ′ ) . Since v ∈ G(M ), we thus have in
′ ′ ′
particular that v g ≡(M,J ′ ) v hg = ug . On the other hand, again since v ∈ G(M ), and since

tp(u/M, J ′ ) ⊥w tp(g ′ /M, J ′ ), we have tp(u/M, J ′ ) ⊥w tp(v g /M, J ′ ), so the result follows. ■

Lemma 5.3. Let M be a small model, and suppose that I = (ai )i∈ω is an M -indiscernible sequence
G00
such that, for some v ∈ G(M ), we have a−1
i aj ∈ v for all i < j. Then I is constant.

Proof. By Ramsey and compactness, it suffices to show the result when I has order type R, so
replace I by a sequence (ai )i∈R with the same properties. Now, I claim that, for some n > 0,
q(x) ⊢ [xn , a−1 n
i aj ] = 1 for all i, j ∈ R; call this (⋆). If q(x) ⊢ x = 1 for some n > 0 then the
claim is clear. Otherwise the realizations of q(x) are non-torsion, whence by Lemma 4.4 G00 is
nilpotent of class at most 2; in particular then [bn , c] = [b, cn ] for all b, c |= G00 and all n ∈ ω. Since
a−1
i aj |= G
00
and q is concentrated on G00 , (⋆) then follows from Fact 2.11.
Let J + g |= q ⊗(ω+1) |(M,I) , and let u = a−1 0 a1 . By Lemma 5.2, we have tp(u/M, J) ⊥
w

tp(g/M, J) and tp(u/M, J) ⊥w tp(ug /M, J); moreover, by 5.2 applied to the bi-f-generic distal
−1
type q −1 , we also have tp(u/M, J) ⊥w tp(g −1 /M, J) and tp(u/M, J) ⊥w tp(ug /M, J). Call these
facts (∗).
Note also that, since I is M -indiscernible and q is M -invariant, I is (M, J, g)-indiscernible. Thus
u ≡(M,J,g) a−1i aj for all i < j ∈ R.
Now, (ai )i∈[0,1] and (aj )j∈[3,4] are mutually indiscernible over (M, a2 , a5 , J, g). By dp-minimality,
one of them is thus indiscernible over (M, a2 , a5 , J, g, a0 ga−13 ). First suppose (ai )i∈[0,1] is indis-
cernible over that set. Note that

g −1 · a−1 −1 −1
0 · (a0 ga3 ) · a5 = a3 a5 ≡(M,J) u.

7
Hence, by indiscernibility, also g −1 a−1 −1 −1 g −1
1 a0 ga3 a5 ≡(M,J) u, ie (a1 a0 ) (a3 a5 ) ≡(M,J) u. But we
−1 −1 −1
have a1 a0 ≡(M,J,g) u , hence (a1 a0 ) ≡(M,J) u , and a3 a5 ≡(M,J) u. Since tp(u−g /M, J) ⊥w
−1 g −g

tp(u/M, J), it follows that u−g u ≡(M,J) u, ie that [g, u] ≡(M,J) u. By Lemma 5.1, using (∗) and
(⋆), we must have u = 1, so that indeed I is constant.
Now suppose (aj )j∈[3,4] is indiscernible over (M, a2 , a5 , J, g, a0 ga−1
3 ). As above, note that

a−1 −1
2 · (a0 ga3 ) · a3 · g
−1
= a−1
2 a0 ≡(M,J) u
−1
.
−1
By indiscernibility, thus a−1 −1
2 a0 ga3 a4 g
−1
≡(M,J) u−1 , ie (a−1 −1
2 a0 )(a3 a4 )
g
≡(M,J) u−1 . But we
−1 −1
have a−1 −1
3 a4 ≡(M,J,g) u, hence (a3 a4 )
g
≡(M,J) ug , and a−1
2 a0 ≡(M,J) u
−1
. Since
−1
tp(ug /M, J) ⊥w tp(u/M, J),
−1 −1
it follows that u−1 ug ≡(M,J) u−1 , so that u−g u ≡(M,J) u, ie [g −1 , u] ≡(M,J) u. By Lemma 5.1
applied to the bi-f-generic distal type q −1 , using (∗) and (⋆), we must have u = 1, so that again I
is constant. ■

Corollary 5.4. Suppose I = (bi )i∈ω is an indiscernible sequence such that b−1 −1
i bj ∼ b0 b1 for all
i < j. Then I is constant.
Proof. By Ramsey and compactness, we may find a model M such that (i) G(M ) contains a
representative of every coset of G00 , (ii) I is M -indiscernible, and (iii) for some v ∈ G(M ), we have
G00 λ−1
b−1
i bj ∈ v
G
for all i < j. By (i), there is some λ ∈ G(M ) such that b−1 i bj ∈ v for all i < j.
−1 00
λ G
Let ai = bi . Then (ai )i∈ω is M -indiscernible, and ai aj ∈ v for all i < j; by Lemma 5.3, (ai )i∈ω
must be constant, so I is too. ■

By considering a Morley sequence we hence obtain the following:


Corollary 5.5. Let M ≺ C, and suppose p(x) ∈ SG (C) is M -invariant with p⊗2 (x, y) ⊢ p(x−1 y)
and p⊗2 (x, y) ⊢ x ∼ y. Then p(x) ⊢ x = 1.

5.2 Conjugacy classes determine types


In this section we will show that any type not concentrated on finitely many conjugacy classes is
uniquely determined by the G00 -conjugacy class of any of its realizations.
Lemma 5.6. Suppose B ⊂ C is a small set of parameters and that a |= G begins a B-indiscernible
sequence of non-conjugates. Then ag ≡B a for all g |= G00 .
Proof. By Ramsey and compactness, we may assume without loss of generality that B = M is
a model. I first claim that, for any J + g |= q ⊗(ω+1) |(M,a) , we have ag ≡(M,J) a; it suffices for
this to show that ag ≡(M,J) a for some J + g |= q ⊗(ω+1) |(M,a) . To see this, let I = (ai )i∈ω be
an M -indiscernible sequence with a0 = a and ai ̸∼ aj for all i ̸= j. Let J |= q ⊗ω |(M,I) and let
K = (gi )i∈ω |= q ⊗ω |(M,I,J) . Since I is M -indiscernible and q is M -invariant, I and K are mutually
indiscernible over (M, J), and thus by dp-minimality one of them is indiscernible over (M, J, ag1 ).
Since ai ∼ ag1 = ag01 if and only if i = 0, K must therefore be indiscernible over (M, J, ag1 ), and
−1 −1
hence there is some a∗ ≡(M,J) a such that ag∗0 = ag1 . But now ag1 g0 = a∗ , so that ag1 g0 ≡(M,J) a.
00 −1
Since q is right f-generic and concentrated on G , and g1 |= q|(M,a,J,g0 ) , we have g1 g0 |= q|(M,a,J) ,
so the claim follows.
Now, let J + g |= q ⊗(ω+1) |(M,a) . By the claim, ag ≡(M,J) a, and by the claim applied to the
−1
bi-f-generic distal type q −1 also ag ≡(M,J) a. On the other hand, by distality of q, we have
−1
tp(a/M, J) ⊥w tp(g/M, J), so it follows that ag∗ ≡(M,J) a ≡(M,J) ag∗ for all a∗ ≡(M,J) a. In other
words, if r(x) = tp(a/M, J), then g lies in the normalizer NG (r) := {h ∈ G : r(x)h = r(x)}. NG (r)
is an (M, J)-invariant subgroup of G; since tp(g/M, J) = q|(M,J) and q is f-generic, it follows by
Fact 2.8 that G00 ⊆ NG (r), so that ah ≡(M,J) a for all h |= G00 . The result follows ‘a fortiori’. ■

8
Corollary 5.7. If r(x) ∈ SG (B) for some B ⊂ C and r is not concentrated on finitely many
conjugacy classes, then r(x)g = r(x) for all g |= G00 . If B = M is a model, then the same holds
provided r is not concentrated on a single conjugacy class.

Corollary 5.8. Suppose p(x) ∈ SG (C) is an M -invariant type not concentrated on any conjugacy
class. If (ai )i∈ω |= p⊗ω |M , then (agi i )i∈ω |= p⊗ω |M for all gi |= G00 .

Proof. If (bi )i∈ω is a Morley sequence of p over M , then each bi starts an (M, b̸=i )-indiscernible
sequence of non-conjugates, so by Lemma 5.6 bgi ≡(M,b̸=i ) bi for all g |= G00 . Thus by induction on
n ∈ ω we have (agi i )i<n + (ai )i⩾n |= p⊗ω |M for all gi |= G00 , and the result follows. ■

5.3 Commutators
In this section we will show that, for any a |= G, the set of commutators [a, G00 ] is a group; this
was proved for C-minimal valued groups in [21].

Lemma 5.9. Let a |= G such that q(x) ⊢ [a, x] ̸= 1. Then [a, q], ie the pushforward of q under the
map x 7→ [a, x], is not concentrated on a single conjugacy class.

Proof. Suppose otherwise for contradiction that [a, q](x) ⊢ x ∼ u. Let M be any small model
containing a, u, and let (gi )i∈ω |= q ⊗ω |M . Since q is right f-generic and concentrated on G00 , we
have gj gi−1 |= q|M for all i < j and hence [a, gj gi−1 ] ∼ u for all i < j. For i ∈ ω, let bi = agi , and
consider the M -indiscernible sequence I = (bi )i∈ω . For i < j, we have

b−1
i bj = a
−gi gj
a = [a, gj gi−1 ]gi ∼ u.

By Lemma 5.4, I is constant, so that ag0 = ag1 and hence g1 g0−1 ∈ CG (a). Since g1 g0−1 |= q|M , this
contradicts that q(x) ⊢ [a, x] ̸= 1. ■

By Corollary 5.7, and since [a, q] is just tp(1/C) if q(x) ⊢ [a, x] = 1, we get the following:

Corollary 5.10. For any M ≺ C and any a ∈ G(M ), we have [a, q|M ] = [a, q|M ]g for all g |= G00 .

Lemma 5.11. Let a |= G. Then [a, g][a, q] = [a, q] for all g |= G00 , and [a, G00 ][a, q|M ] ⊆ [a, G00 ]
for any small model M with a ∈ G(M ).

Proof. First note the following general fact. Let r ∈ SG (M ) be a complete type over some model
M ≺ C such that a ∈ G(M ) and such that [a, r] = [a, r]g for all g |= G00 . Then, from the identity
[a, g][a, c]g = [a, cg], it follows that [a, g][a, r] = [a, rg] for all g |= G00 .
The desired result now follows; indeed by Corollary 5.10 and by the remark above we have
[a, g][a, q|M (x)] = [a, q|M (x)g] for all M ≺ C with a ∈ G(M ) and all g |= G00 ; this shows already
that [a, G00 ][a, q|M ] ⊆ [a, G00 ]. Furthermore, since q is right f-generic, we have qg = q for all
g ∈ G00 (C), so that q|M (x)g = q|M (x) whenever G(M ) ∋ g, and hence indeed [a, g][a, q] = [a, q]. ■

Lemma 5.12. For any a |= G, the set of commutators [a, G00 ] is a type-definable subgroup of G00 ,
and it is ‘strongly connected’, ie has no proper type-definable subgroups of bounded index.

Proof. Certainly [a, G00 ] is type-definable, and it is a subset of G00 since G00 is a normal subgroup
of G. Now, let M be any small model containing a. By Corollary 5.10, we have [a, q|M ]g = [a, q|M ]
for all g |= G00 ; likewise, by Corollary 5.10 applied to the bi-f-generic distal type q −1 , we also have
[a, q|−1 g −1 00
M ] = [a, q|M ] for all g |= G . Thus, for any b |= q|M , since q is concentrated on G
00
we have
−1 −1
[a, b−1 ] = [a, b]−b ∈ [a, q|M ]−b = [a, q|M ]−1 and
−1 −1
[a, b]−1 = [a, b−1 ]b ∈ [a, q|−1
M ]
b
= [a, q|−1
M ],

9
so that [a, q|−1
M ] = [a, q|M ]
−1
, and it follows that [a, q|M ][a, q|−1
M ] is closed under taking inverses.
Now, by Fact 2.8, we have G ⊆ q|−1
00 00
M (x)q|M (x). Thus, for any g |= G , there are b, c |= q|M with
g = b−1 c, and then we have

[a, g] = [a, c][a, b−1 ]c ∈ [a, q|M ][a, q|−1 c −1


M ] = [a, q|M ][a, q|M ],

so that [a, G00 ] ⊆ [a, q|M ][a, q|−1


M ]. Finally, by Lemma 5.11 applied to q and to q
−1
, we have

[a, G00 ][a, q|M ][a, q|−1 00 −1 00


M ] ⊆ [a, G ][a, q|M ] ⊆ [a, G ].

Thus [a, q|M ][a, q|−1 00


M ] = [a, G ], and it is symmetric and closed under multiplication; in other words,
it is a group.
To see that [a, G00 ] is strongly connected, note that, by Lemma 5.11, [a, q] is invariant under
left translation by all elements of [a, G00 (C)]. If H ⩽ [a, G00 ] is type-definable and of bounded index
in [a, G00 ], then, since [a, q] is concentrated on [a, G00 ], it must be concentrated on some right coset
of H, and then left-invariance of [a, q] forces H = [a, G00 ]. ■

Corollary 5.13. For any a, b |= G, one of [a, G00 ] ⊇ [b, G00 ] and [a, G00 ] ⊆ [b, G00 ] holds.

Proof. By Fact 2.10, [a, G00 ] ∩ [b, G00 ] has bounded index in one of [a, G00 ] and [b, G00 ]. Since the
latter two groups are strongly connected, [a, G00 ] ∩ [b, G00 ] must be equal to one of them, and the
result follows. ■

Corollary 5.14. For any type-definable set X ⊆ G, the set of commutators [X, G00 ] is a type-
definable strongly connected subgroup of G00 .

Proof. [X, G00 ] is equal to a|=X [a, G00 ], and by Lemma 5.12 and Corollary 5.13 this is the union
S

of a chain of subgroups of G00 . So indeed [X, G00 ] is a group, and it is certainly type-definable.
To see that it is strongly connected, suppose that H ⩽ [X, G00 ] is a bounded-index type-definable
subgroup. Then, for every a |= X, H ∩ [a, G00 ] has bounded index in [a, G00 ], hence must be equal
to [a, G00 ]. So H ⩾ [a, G00 ] for all a |= X and hence H = [X, G00 ], as desired. ■

The following now has an identical proof as Proposition 2.4.1 of [21]:

Lemma 5.15. [G00 , G00 ] has finite exponent.

Proof. Fix a, b |= G00 ; by Lemma 5.12, [a, b, G00 ] is now a normal subgroup of G00 . Working
in the quotient G00 /[a, b, G00 ], it follows from the identities [ak+1 , b] = [a, b][a, b, ak ][ak , b] and
[a, bk+1 ] = [a, bk ][a, b][a, b, bk ] that [ak , b] and [a, bk ] are each equivalent to [a, b]k modulo [a, b, G00 ]
for each k ∈ ω. By Fact 2.11, there is some n > 0 such that one of [an , b] and [a, bn ] is trivial, and
it follows that [a, b]n ∈ [a, b, G00 ].
Thus there is some g |= G00 with [a, b]n = [a, b, g], so that [a, b]n+1 = [a, b]g ∼ [a, b]. By
Lemma 2.4 and Fact 2.11, it follows that [a, b] must be a torsion element. Since a, b |= G00 were
arbitrary, thus every element of [G00 , G00 ] has finite order, and so by compactness [G00 , G00 ] has
finite exponent. ■

5.4 Valued group structure


Suppose that (H, ·) is an arbitrary group and that N P H is a normal subgroup such that the
collection P = {[a, N ] : a ∈ H} is a family of subgroups of H totally ordered by the reverse inclusion
relation, which we will denote ≼. Note that (P, ≼) has a maximal element {1} = [1, N ]. Now, define
a map v : H/CH (N ) → P by v(a) = [a, N ]; this is well-defined since [CH (N )a, g] = {[a, g]} for all
a ∈ H and g ∈ N . By definition, v(a) ≼ v(b) if and only if [a, N ] ⊇ [b, N ].

Lemma 5.16. In the above circumstances, (H, v) is a P-valued group.

10
Proof. For property (1), note that, for a ∈ H, we have v(a) = max P if and only if [a, N ] = {1},
ie if and only if a ∈ CH (N ). For property (2), suppose that v(a) ≺ v(b) and that g ∈ H is
arbitrary. We have [a, N ] ⊋ [b, N ], and hence [a, N ]g ⊋ [b, N ]g . But N is a normal subgroup of
H, so [a, N ]g = [ag , N g ] = [ag , N ], and likewise [b, N ]g = [bg , N ], so that v(ag ) ≺ v(bg ) as desired.
−1
For property (3), fix a ∈ H. We have [a−1 , N ] = [N, a]a = [N, a]. But [a, N ] is by hypothesis
a subgroup of H, hence in particular is closed under inverses, and so [a−1 , N ] = [a, N ], whence
v(a) = v(a−1 ).
So we need only show property (4), ie that v(ab) ≽ min{v(a), v(b)} for all a, b ∈ H. First
suppose that v(b) = min{v(a), v(b)}, ie that [a, N ] ⊆ [b, N ]; we wish to show that [ab, N ] ⊆ [b, N ].
By the identity [ab, g] = [a, g]b [b, g], we have [ab, N ] ⊆ [a, N ]b [b, N ]. But [a, N ]b ⊆ [b, N ]b = [b, N ],
so we have [ab, N ] ⊆ [b, N ][b, N ] ⊆ [b, N ], where the second inclusion uses that [b, N ] is closed under
multiplication. In other words v(ab) ≽ v(b), as desired.
Now suppose that v(a) = min{v(a), v(b)}. Using that v(c) = v(c−1 ) for all c ∈ H, we have
v(ab) = v(b−1 a−1 ) and v(b−1 ) = v(b) ≽ v(a) = v(a−1 ). Thus v(a−1 ) = min{v(b−1 ), v(a−1 )} and so,
by the case in the previous paragraph, we have v(b−1 a−1 ) ≽ v(a−1 ). But this means v(ab) ≽ v(a),
as desired. ■

Now, let us return to the group G. The subgroup N = G00 is normal in G, and by Lemma 5.12
and Corollary 5.13 the set P = {[a, G00 ] : a |= G} is a chain of subgroups of G. So, by Lemma 5.16,
the map v : a 7→ [a, G00 ] makes G/CG (G00 ) into a P-valued group. Note that, by Fact 2.7, the
group CG (G00 ) has a purely group-theoretic description; recall that the FC-center of G, which we
will denote FC(G), is the set of all g ∈ G with [G : CG (g)] < ∞. This is a characteristic subgroup
of G, and by Fact 2.7 we have g ∈ FC(G) if and only if G00 ⊆ CG (g). So CG (G00 ) = FC(G) and
thus the map v makes G/FC(G) into a valued group.
We have FC(G) ∩ G00 = CG (G00 ) ∩ G00 = Z(G00 ). Thus G00 /Z(G00 ) inherits the valued group
structure from G/FC(G).

Theorem 5.17. (G00 /Z(G00 ), v) is a plain valued group.

Proof. We need only show that v is a plain valuation. Hence fix a, b ∈ G00 \ Z(G00 ); we wish to
show that v([a, b]) ≻ max{v(a), v(b)}. Since v([a, b]) = v([b, a]), we may assume without loss that
v(a) = max{v(a), v(b)}, so that [a, G00 ] ⊆ [b, G00 ]. We wish to show that v([a, b]) ≻ v(a), ie that
[a, b, G00 ] ̸⊇ [a, G00 ]. To see this, let M be any small model and let g |= q|(M,a,b) ; it suffices to show
that [a, g] ∈ / [a, b, G00 ]. Suppose otherwise for contradiction that [a, g] = [a, b, h] for some h |= G00 .
Now, [a, g] |= [a, q]|(M,a,b) , and b |= G00 , so by Lemma 5.11 we have [a, b][a, g] |= [a, q]|(M,a,b) . But
[a, b][a, g] = [a, b][a, b, h] = [a, b]h . Thus [a, q] is concentrated on the conjugacy class [a, b]G , and by
Lemma 5.9 we must have q(x) ⊢ x ∈ CG (a). But then, by f-genericity of q, CG (a) has finite index
in G, contradicting that a ∈ / FC(G). ■

Now by Fact 2.5 we obtain the following:

Corollary 5.18. Any finite-exponent subgroup of G00 is locally nilpotent. In particular, by Lemma
5.15, [G00 , G00 ] is locally nilpotent.

5.5 Local nilpotence to nilpotence


Lemma 5.19. Let M ≺ C be any model, and suppose p(x) ∈ SG (C) is an M -invariant type with the
property that, for some (any) (a, b) |= p⊗2 |M , the group ⟨a, b⟩ is nilpotent of class at most (n − 1).
Then there is some σ ∈ Sn such that p⊗n (x1 , . . . , xn ) ⊢ [xσ(1) , . . . , xσ(n) ] = 1.

Proof. Let (ai )i∈ω + (ui )i∈ω + (bi )i∈ω be a Morley sequence of p over some model M . Write
I = (ai )i∈ω and J = (bi )i∈ω and K = (ui )i∈ω for notational convenience.
By hypothesis, ⟨a0 , b1 ⟩ is nilpotent of class at most (n−1). We will prove by backwards induction
on k that, for every k ∈ [n] ∪ {0}, and for all c1 , . . . , ck ∈ {a0 , b1 }, there is some τ ∈ Sn−k such that

11
[c1 , . . . , ck , uτ (n−k) , . . . , uτ (1) ] = 1. The base case n = k is clear. Now, suppose we have shown the
result for k + 1, and fix c1 , . . . , ck ∈ {a0 , b1 }. I and J are mutually indiscernible over K, and so by
dp-minimality one of I and J is indiscernible over (K, [c1 , . . . , ck ]).
First suppose I is indiscernible over that set. By the inductive hypothesis, there is some
ν ∈ Sn−k−1 such that [c1 , . . . , ck , a0 , uν(n−k−1) , . . . , uν(1) ] = 1. By indiscernibility of I over
(K, [c1 , . . . , ck ]), it follows that [c1 , . . . , ck , a1 , uν(n−k−1) , . . . , uν(1) ] = 1. Now if we define τ ∈ Sn−k
by taking τ (n − k) = 1 and taking τ (i) = ν(i) + 1 for i ∈ [n − k − 1] then we get the desired result
by indiscernibility of a0 + a1 + u1 + · · · + un−k + b1 .
Now suppose J is indiscernible over that set. By the inductive hypothesis, there is some
ν ∈ Sn−k−1 such that [c1 , . . . , ck , b1 , uν(n−k−1) , . . . , uν(1) ] = 1. By indiscernibility of J over
(K, [c1 , . . . , ck ]), it follows that [c1 , . . . , ck , b0 , uν(n−k−1) , . . . , uν(1) ] = 1. Now if we define τ ∈ Sn−k
by taking τ (n − k) = n − k and τ (i) = ν(i) for i ∈ [n − k − 1] then we get the desired result by
indiscernibility of a0 + u1 + · · · + un−k + b0 + b1 .
So the induction goes through, and when k = 0 we get that [uσ(n) , . . . , uσ(1) ] = 1 for some
σ ∈ Sn , giving the desired result. ■

Lemma 5.20. Suppose that H ⩽ G00 is a subgroup of G type-definable over some M ≺ C. Suppose
that p(x) ∈ SG (C) satisfies the hypotheses of Lemma 5.19, and that p(x) is concentrated on H and
invariant under left-translation by elements of H(C). Then H is nilpotent of class at most (n − 1).
Proof. Note that, since p(x) is invariant under left-translation by elements of H(C), we have the
following: if B ⊂ C is any parameter set and O ⩽ H is an (M, B)-invariant subgroup such that
p|(M,B) (x) ⊢ x ∈ O, then O = H. Moreover, again by translation-invariance, and since p(x) is
concentrated on H, we have p⊗2 (x, y) ⊢ p(x−1 y). Thus, by Corollary 5.5, either H = {1}, in which
case the result is clear, or p(x) is not concentrated on any conjugacy class; assume we are in the
latter case.
By Lemma 5.19, there is some σ ∈ Sn such that p⊗n (x1 , . . . , xn ) ⊢ [xσ(1) , . . . , xσ(n) ] = 1. Let
(a1 , . . . , an ) |= p⊗n |M ; by Corollary 5.8, we have (ag11 , . . . , agnn ) |= p⊗n |M for all gi |= G00 , and so in
particular [aH H H
σ(1) , . . . , aσ(n) ] = {1}. For each k ∈ {0} ∪ [n] and l ∈ [n], define Akl = al if l ⩽ n − k
and Akl = H otherwise. We prove by induction on k that [Akσ(1) , . . . , Akσ(n) ] = {1}; we have
already shown the base case k = 0.
Now, for the inductive step, suppose we have shown the result for k, and let i = σ −1 (n − k).
Note that each Akl is setwise invariant under conjugation by elements of H. Hence, by Lemma 2.1,
the set of h |= H such that
[Akσ(1) , . . . , Akσ(i−1) , h, Akσ(i+1) , . . . , Akn ] = {1}
is a subgroup of H; call it U . By the inductive hypothesis, U ⊇ Akσ(i) = aH n−k , and hence in
particular an−k ∈ U . On the other hand, if l > n − k, then Akl = H. So every Akσ(j) is either equal
to H, hence invariant over M , or of form aH l where l < n − k, hence invariant over (M, al ). So
U is invariant over (M, a1 , . . . , an−k−1 ), and since an−k |= p|(M,a<n−k ) it follows from the remarks
in the first paragraph that U = H. Since A(k+1),σ(i) = H and A(k+1),σ(j) = Akσ(j) for j ̸= i, the
inductive step follows.
Now taking k = n we get [H, n−1 H] = {1}, and the result follows. ■

Corollary 5.21. [G00 , G00 ] is nilpotent.


Proof. By Lemma 5.18 [G00 , G00 ] is locally nilpotent. Thus the partial type π(x0 , x1 ) expressing
x0 ∈ [G00 , G00 ] and x1 ∈ [G00 , G00 ], and containing the formula σ∈2[n] [xσ(1) , . . . , xσ(n) ] ̸= 1 for
W
every n ∈ ω, is inconsistent. By compactness, it follows that, for some n ∈ ω, every 2-generated
subgroup of [G00 , G00 ] is nilpotent of class at most (n − 1). Thus any invariant global type p(x)
concentrated on [G00 , G00 ] satisfies the hypotheses of Lemma 5.19.
Now, fix arbitrary a |= G00 . By Lemma 5.11 and the paragraph above, the type pa (x) := [a, q]
satisfies the hypotheses of Lemma 5.20 for the group H = [a, G00 ], and hence by Lemma 5.20
[a, G00 ] is nilpotent of class at most (n − 1).

12
But [G00 , G00 ] = a|=G00 [a, G00 ], so by Lemma 5.13 [G00 , G00 ] is a union of chain of subgroups
S

that are each nilpotent of class at most (n−1), so [G00 , G00 ] is nilpotent of class at most (n−1). ■

Theorem 5.22. G is virtually nilpotent.

Proof. It suffices, for example by Lemma 3.1, to show that G00 is nilpotent. But [G00 , G00 ] is
nilpotent by Corollary 5.21, so the result follows. ■

Corollary 5.23. If G is not virtually nilpotent of class at most 2, then it has finite exponent.

Proof. If the realizations of q(x) are non-torsion, then by Lemma 4.4 G is virtually nilpotent of class
at most 2, so we may assume the realizations of q are torsion. By Theorem 5.22, G00 is nilpotent,
so its torsion elements comprise a group, and it thus follows by Fact 2.8 that G00 is torsion. By
compactness, it is hence of finite exponent, say of exponent n > 0. Since it is a group, we have
G00 (x) ∧ G00 (y) ⊢ (xy)n = 1, and so by compactness there is some ∅-definable set D ⊇ G00 such
that (ab)n = 1 for all a, b |= D. Replacing D by D ∩ D−1 , we may assume that D = D−1 .
Now, I claim that G is torsion; by compactness the result will follow. Otherwise, there is some
non-torsion element b |= G. Since every element of DD−1 is torsion, we have bk ∈ / DD−1 for all
i j
k ̸= 0, and hence b D ∩ b D = ∅ for all i ̸= j ∈ ω. Letting I = (ai )i∈ω be an indiscernible sequence
realizing the EM-type of (bi )i∈ω over ∅, we then have ai D ∩ aj D = ∅ for all i ̸= j; since D ⊇ G00 ,
the ai thus lie in pairwise distinct cosets of G00 . Since I is indiscernible, this contradicts that G00
has bounded index in G, and the result follows. ■

13
References
[1] John Baldwin and Jan Saxl. Logical stability in group theory. Journal of the Australian Math-
ematical Society (Series A), Vol. 21 No. 3 (1976). pp. 267-276.

[2] Artem Chernikov, Anand Pillay, Pierre Simon. External definability and groups in NIP theories.
Journal of the London Mathematical Society, Vol. 90 Iss. 1 (2014). pp 213-240.

[3] Artem Chernikov, Pierre Simon. Definably amenable NIP groups. Journal of the AMS, Vol. 31
Iss. 3 (2018). pp. 609-641.

[4] Tim Clausen. Dp-minimal profinite groups and valuations on the integers. Preprint (2020),
arXiv:2008.08797.

[5] Jan Dobrowolski and Frank Wagner. On ω-categorical groups and rings of finite burden. Israel
Journal of Mathematics, Vol. 236 (2020). pp. 801-839.

[6] Jan Dobrowolski and John Goodrick. Some remarks on inp-minimal and finite burden groups.
Archive for Mathematical Logic, Vol. 58 (2019). pp. 267-274.

[7] Alfred Dolich, John Goodrick, David Lippel. Dp-minimality: basic facts and examples. Notre
Dame Journal of Formal Logic, Vol. 52 No. 3 (2011). pp. 267-288.

[8] Ehud Hrushovski, Ya’acov Peterzil, Anand Pillay. Groups, measures, and the NIP. Journal of
the AMS, Vol. 21 No. 2 (2008). pp. 563-596.

[9] Ehud Hrushovski, Anand Pillay. On NIP and invariant measures. Journal of the EMS, Vol. 13
No. 4 (2011). pp. 1005-1061.

[10] Ehud Hrushovski, Anand Pillay, Pierre Simon. A note on generically stable measures and fsg
groups. Notre Dame Journal of Formal Logic, Vol. 53 No. 4 (2012). pp. 599-605.

[11] Itay Kaplan, Elad Levi, Pierre Simon. Some remarks on dp-minimal groups, in: Groups,
Modules, and Model Theory, Surveys and Recent Developments. Springer (2017).

[12] Alf Onshuus and Alexander Usvyatsov. On dp-minimality, strong dependence, and weight.
Journal of Symbols Logic, Vol. 71 No. 1 (2006). pp. 1-21.

[13] Derek Robinson. A Course in the Theory of Groups. Springer (1996).

[14] Saharon Shelah. Dependent first order theories, continued. Israel Journal of Mathematics, Vol.
173 No. 1 (2009). pp. 1-60.

[15] Saharon Shelah. Strongly dependent theories. Israel Journal of Mathematics, Vol. 204 (2014).
pp. 1-83.

[16] Saharon Shelah. Minimal bounded index subgroup for dependent theories. Proceedings of the
AMS, Vol. 136 No. 3, (2008). pp. 1087-1091.

[17] Pierre Simon. A guide to NIP theories. Cambridge University Press (2015).

[18] Pierre Simon. On dp-minimal ordered structures. The Journal of Symbolic Logic, Vol. 76 No.
2 (2011). pp. 448-460.

[19] Pierre Simon. Distal and non-distal NIP theories. Annals of Pure and Applied Logic, Vol. 164
No. 1 (2013). pp. 294-318.

[20] Patrick Simonetta. An example of a C-minimal group which is not abelian-by-finite. Proceedings
of the AMS, Vol. 131 No. 12 (2003). pp. 3913-3917.

14
[21] Patrick Simonetta. On non-abelian C-minimal groups. Annals of Pure and Applied Logic, Vol.
122 (2003). pp 263-287.

[22] Atticus Stonestrom. On f-generic types in NIP groups. Preprint (2023). arXiv:2303.13470.

[23] Frank Wagner. Stable Groups. Cambridge University Press (2013).

15

You might also like