Bull Braz Math Soc, New Series (2023) 54:55

https://doi.org/10.1007/s00574-023-00372-6

Representations of Solvable Subgroups of PSL (3, C)

Mauricio Toledo-Acosta1,2

Received: 4 August 2023 / Accepted: 24 September 2023

© The Author(s), under exclusive licence to Brazilian Mathematical Society 2023

Abstract
In this paper, we give a complete description of the representations of all upper triangu-
lar complex Kleinian subgroups of PSL (3, C). In Toledo-Acosta (Bull Braz Math Soc
New Ser 20:1–45, 2021), we show that any solvable group is virtually triangularizable
and can be constructed as the semidirect product of two layers of parabolic elements
and two layers of loxodromic elements. There are five families of purely parabolic dis-
crete groups of PSL (3, C) (Barrera et al. in Linear Algebra Appl 653:430–500, 2022),
therefore, the parabolic part of any upper triangular group belongs to one of these five
families. In this paper we study which loxodromic elements can be combined with the
elements of the parabolic part of an upper triangular discrete subgroup of PSL (3, C)
in each of these five cases. These parabolic elements impose strong restrictions on
the type, and number, of loxodromic elements that can be present in the group. We
show that up to conjugation, there are 16 types of upper triangular complex Kleinian
groups in PSL (3, C) containing loxodromic elements. These results are a further step
towards the completion of the study of elementary discrete subgroups of PSL (3, C).

Keywords Complex Kleinian groups · Solvable groups · Upper triangular groups ·

PSL (3, C)

Mathematics Subject Classification Primary 37F99; Secondary 30F40 · 20H10 ·

22E40

Partially supported by Grants of projects PAPPIT UNAM IN101816, PAPPIT UNAM IN110219,
CONACYT 282937, FORDECYT 265667.

B Mauricio Toledo-Acosta
[email protected] ; [email protected]

1 Departamento de Matemáticas, Universidad de Sonora, Boulevard Luis Encinas y Rosales s/n

Col. Centro, CP 83000 Hermosillo, Mexico
2 Universidad Virtual del Estado de Guanajuato, Hermenegildo Bustos 129 A Sur, Centro, C.P. 36400
Purísima del Rincón, Mexico

0123456789().: V,-vol 123

 55 Page 2 of 28 M. Toledo-Acosta

1 Introduction

Kleinian groups are discrete subgroups of PSL (2, C), the group of biholomorphic
automorphisms of the complex projective line CP1 , acting properly and disconti-
nuously on a non-empty region of CP1 . Kleinian groups have been thoroughly studied
since the end of the 19th century by Lazarus Fuchs, Felix Klein, Henri Poincaré, and
many others. Kleinian groups have played a major role in several fields of mathematics,
such as Riemann surfaces and Teichmüller theory, automorphic forms, holomorphic
dynamics, conformal and hyperbolic geometry, etc. For a detailed study of Kleinian
groups, see Maskit (1987) and Matsuzaki and Taniguchi (1998). In Seade and Ver-
jovsky (2001), José Seade and Alberto Verjovsky introduced the notion of complex
Kleinian groups, which are discrete subgroups of PSL (n + 1, C) acting properly and
discontinuously on an open invariant subset of the complex projective space CPn . In
the last decade, there has been a great effort to complete the study of the dynamics
of complex Kleinian groups in dimension 2 (see, for example Barrera et al. 2011,
2014, 2016, 2018; Cano et al. 2013; Cano and Seade 2014). In this regard, one of the
remaining pieces is the full description of the discrete subgroups of PSL (3, C) which
have simple dynamics. These subgroups of PSL (3, C) are the analogous of elemen-
tary subgroups of PSL (2, C). However, there is no standard definition of elementary
subgroups of PSL (3, C).
In the context of Kleinian groups, elementary groups are discrete subgroups of
PSL (2, C) such that the limit set is a finite set, in which case the limit set has 0, 1, or
2 points. These groups have simple dynamics. In complex dimension 2, the Kulkarni
limit set is either a finite union of complex lines (1, 2 or 3) or it contains an infinite
number of complex lines. On the other hand, the limit set either contains a finite
number of lines in general position (1, 2, 3 or 4) or it contains infinitely many lines in
general position (Barrera et al. 2016). Therefore, one could define elementary groups
in complex dimension 2 as discrete subgroups of PSL (3, C) such that the Kulkarni
limit set contains a finite number of lines, or discrete subgroups of PSL (3, C) such that
the Kulkarni limit set contains a finite number of lines in general position. However,
this definition would not be useful in complex dimension greater than 2 since there is
no similar classification of the Kulkarni limit set in this case.
Another alternative to characterize elementary complex Kleinian groups is solvabil-
ity. This approach was first explored in Toledo-Acosta (2021), and it is continued in
this paper. In the first part (Toledo-Acosta 2021), we studied the dynamics of solvable
subgroups of PSL (3, C). We proved that solvable groups are virtually triangularizable
and, up to a finite-index subgroup, can be decomposed in four layers, via the semidirect
product of four types of elements. We also showed that solvable groups act properly
and discontinuously on the complement of, either, a line, two lines, a line and a point
outside of the line or a pencil of lines passing through a point.
In this second part, we provide a full description of the representations of solvable
subgroups of PSL (3, C), up to a finite index subgroup. We build upon the work done
in Barrera et al. (2022), in this paper the authors prove that, up to conjugation, there
are five families of purely parabolic discrete groups in PSL (3, C) (see Theorem 3.1).
These purely parabolic groups play the role of the two innermost layers of discrete
upper triangular subgroups of U+ (see Theorem 2.6), they form a parabolic core for the

123
Representations of Solvable Subgroups of PSL (3, C) Page 3 of 28 55

group. In this paper, we describe the loxodromic elements which can be combined with
every possible parabolic core in order to generate upper triangular discrete groups.
As we will evidence, the more parameters the parabolic part of a group has, the fewer
loxodromic elements can be added to the group while preserving its discreteness.
We show that up to conjugation, there are 16 types of upper triangular complex
Kleinian groups in PSL (3, C) containing loxodromic elements. These results are a
further step towards the completion of the study of elementary discrete subgroups of
PSL (3, C).
Before stating the main theorem of this paper, it is necessary to introduce some
notation. We denote by PSL (3, C) the group of biholomorphic automorphisms of the
complex projective plane CP2 . We denote by U+ ⊂ PSL (3, C) the upper triangular
subgroups of PSL (3, C). We list the following families of purely parabolic discrete
subgroups of PSL (3, C),
(1) Torus groups,
⎧⎡ ⎤ ⎫
⎨ 1 0 x ⎬
T (W ) = ⎣ 0 1 y ⎦ (x, y) ∈ W
⎩ ⎭
0 0 1

where W ⊂ C2 is a additive subgroup.

(2) Dual torus groups,
⎧⎡ ⎤ ⎫
⎨ 1 x y ⎬
T ∗ (W ) = ⎣ 0 1 0 ⎦ (x, y) ∈ W
⎩ ⎭
0 0 1

where W ⊂ C2 is a discrete additive subgroup with r (W ) ≤ 2.

The names given to these families, and the families defined in Theorem 3.1, were
taken from Barrera et al. (2022). They stem from the fact that they are subgroups of
fundamental groups of such complex surfaces. For a more detailed description of these
families see Barrera et al. (2022) and Kodaira (1964).
For the sake of clarity, we divide the main theorem of this paper into four parts.
These four parts describe, up to conjugation, all upper triangular, non-cyclic, discrete
groups containing loxodromic elements.

Theorem 1.1 Let ˜ ⊂ U+ be a non-cyclic, discrete abelian group containing loxo-

dromic elements, then there exists a finite index subgroup  ⊂ ˜ conjugate to one of
the following groups.
(1)

Diag(α , β , 1) n, m ∈ Z ,
n m

where α, β ∈ C∗ , |α| = 1 or |β| = 1.

123
 55 Page 4 of 28 M. Toledo-Acosta

(2)
⎧⎡ ⎤ ⎫
⎨ μ(w) wμ(w) 0 ⎬
⎣ 0 μ(w) 0 ⎦ w∈W
⎩ ⎭
0 0 μ(w)−2

where W ⊂ C is a discrete additive group such that r (W ) ≤ 3 and μ : (W , +) →

(C∗ , ·) is a group morphism satisfying that for every sequence of distinct elements
{wk } ⊂ W it holds that μ(wk ) → 0 or μ(wk ) → ∞.
Theorem 1.2 Let ˜ ⊂ U+ be a non-cyclic, discrete non-abelian group containing
loxodromic elements. If the parabolic part of ˜ is conjugate to a torus group, then
there exists a finite index subgroup  ⊂ ˜ conjugate to the group
⎧⎡ 2n n ⎤ ⎫
⎨ α β 0 0 ⎬
T (W )  ⎣ 0 α n β 2n 0 ⎦ n ∈ Z ,
⎩ ⎭
0 0 1

where r (W ) = 3 and α, β ∈ C∗ , α = 1, αβ 2 ∈
/ R.
Theorem 1.3 Let ˜ ⊂ U+ be a non-cyclic, discrete non-abelian group containing
loxodromic elements. If the parabolic part of ˜ is conjugate to a dual torus group
T ∗ (W ) such that r (W ) = 2, then there exists a finite index subgroup  ⊂ ˜ conjugate
to one of the following groups.
(1)
⎡ p γ ⎤
12 γ13
T ∗ (W )  ⎣ 0 1
q ⎦
p ,
0 0 1

where W ∼
= (1, 0), (0, 1) , p ∈ Z\ {−1, 0, 1}, q ∈ Z\ {0} and γ12 , γ13 ∈ C.
(2)
⎡ p γ12 γ13
⎤ ⎡ ⎤
p2 μ12 μ13 
1
T ∗ (W )  ⎣ 0 1 p11 ⎦ , ⎣ 0
q q2 ⎦
1 p2 ,
0 0 1 0 0 1

where W ∼ = (1, 0), (0, 1) , p1 , p2 ∈ Z\ {−1, 0, 1}, q1 , q2 ∈ Z\ {0} and

γ12 , γ13 , μ12 , μ13 ∈ C satisfying the conditions of Lemma 3.9(2).
(3)
 ⎡ ⎤
pq γ12 γ13 
T ∗ (W )  γ := ⎣ 0 q r ⎦ ,
0 0 p

where W ∼ = (1, 0), (0, 1) , p, q ∈ Z\ {0} such that γ is loxodromic, r ∈ Z and

γ12 , γ13 ∈ C.

123
Representations of Solvable Subgroups of PSL (3, C) Page 5 of 28 55

(4)

⎡ p + qRe(y) + iqIm(y) γ ⎤
12 γ13
T ∗ (W )  ⎣ 0 1 γ23 ⎦ ,
0 0 1

W ∼ / R. Also, p, q ∈ Z such that | p| + |q| = 0 and

= (0, 1), (0, y) , with y ∈
|( p + qRe(y)) + iqIm(y)| = 1, γ12 , γ23 ∈ C and γ23 ∈ C∗ .
(5)

 ⎡ ⎤
α γ12 γ13 
T ∗ (W )  γ := ⎣ 0 α −2 z p,q γ23 ⎦ ,
0 0 αz −1
p,q

W ∼ = (0, 1), (0, y) , with y ∈ / R. Also, β ∈ C∗ , z p,q = ( p + qRe(y)) + iqIm(y),

| p| + |q| = 0, γ12 , γ23 , γ23 ∈ C. Finally, γ must be loxodromic.
(6)

 ⎡ ⎤
p + q x γ12 γ13
T ∗ (W )  γ := ⎣ 0 1 0 ⎦ ,
0 −1
0 ( p + q x) β −3

W ∼= (1, 0), (x, 0) , with x ∈ / R. Also, p, q ∈ Z satisfying the conditions of

Lemma 3.11(ii), β ∈ C∗ , γ12 , γ13 ∈ C, and γ is loxodromic.
(7)

 ⎡ ⎤
p + q x μ12 0
T ∗ (W )  γ , μ := ⎣ 0 1 0 ⎦ ,
−1 −3
0 0 ( p + q x) β2

W ∼= (1, 0), (x, 0) , with x ∈

/ R, and γ defined by (6). Also, p , q ∈ Z satisfying
the conditions of Lemma 3.11(iii), and μ is loxodromic.

Theorem 1.4 Let ˜ ⊂ U+ be a non-cyclic, discrete non-abelian group containing

loxodromic elements. If the parabolic part of ˜ is conjugate to a dual torus group
T ∗ (W ) such that r (W ) = 1, then there exists a finite index subgroup  ⊂ ˜ conjugate
to one of the following groups.
(1)

 ⎡ ⎤
pα γ12 γ13
T ∗ (W )  γ := ⎣ 0 α 0 ⎦ ,
0 0 p −1 α −2

where W ∼
= (1, 0) , p ∈ Z\ {0, 1}, α ∈ C∗ , |α| = 1, and γ12 , γ13 ∈ C.

123
 55 Page 6 of 28 M. Toledo-Acosta

(2)
 ⎡ β j pq ⎤
qβ 1− p μ13
T ∗ (W )  γ , μ := ⎣ 0 β 0 ⎦ ,
0 0 q −1 β −2

where W ∼ = (1, 0) , γ defined by (1). Also, β ∈ C∗ , j, q ∈ Z, q = 0 such that μ

is loxodromic, and μ13 ∈ C, if p 2 α 3 = 1, μ13 = 0.
(3)
 ⎡ δkpr ⎤
rδ 1− p η13
T ∗ (W )  γ , μ, η := ⎣ 0 δ 0 ⎦ ,
0 0 q δ −2
−1

where W ∼ = (1, 0) and γ , μ are given by (1) and (2). Also, δ ∈ C∗ , r ∈

Z\ {−1, 0, 1}, r δ 3 = 1 such that η is loxodromic, and η13 , k ∈ C satisfying
the conditions of Lemma 3.12.
(4)
⎡ p γ ⎤
12 γ13
T ∗ (W )  ⎣ 0 1 γ23 ⎦ ,
0 0 1

where W ∼
= (0, 1) , p ∈ Z\ {−1, 0, 1}, γ12 , γ13 ∈ C, γ23 ∈ C∗ .
(5)
 ⎡ ⎤
α γ12 γ13 
T ∗ (W )  γ := ⎣ 0 pα −2 γ23 ⎦ ,
0 0 p −1 α

where W ∼
= (0, 1) , α ∈ C∗ , p ∈ Z\ {0}, p 2 = α 3 and γ is loxodromic.
(6)
 ⎡ jp ⎤
β 0 1− p β
T ∗ (W )  γ , μ = ⎣ 0 qβ −2 0 ⎦ ,
0 0 −1
q β

where W ∼ = (0, 1) and γ is given by (5). Also, β ∈ C∗ , q, j ∈ Z\ {0}, q 2 = β 3

and μ is loxodromic.
The paper is organized as follows: in Sect. 2 we present a brief background necessary
for the next sections, we also summarize the necessary results from Toledo-Acosta
(2021). In Sect. 3, we state and prove the technical results describing or dismissing
combinations of parabolic and loxodromic elements. These results together provide a
full description of the representation of upper triangular discrete triangular subgroups
of PSL (3, C). Finally, in Sect. 4, we prove the main theorems.

123
Representations of Solvable Subgroups of PSL (3, C) Page 7 of 28 55

2 Notation and Background

2.1 Complex Kleinian Groups

The complex projective plane CP2 is defined as

CP2 = (C3 \{0})/C∗ ,

where C∗ := C\{0} acts by the usual scalar multiplication. Let [ ] : C3 \{0} → CP2
be the quotient map. We denote the projectivization of the point x = (x1 , x2 , x3 ) ∈ C3
by [x] = [x1 : x2 : x3 ]. We denote by e1 , e2 , e3 the projectivization of the canonical
base of C3 .
Let M3 (C) be the group of all square matrices of size 3 × 3 with complex coef-
ficients, and let SL (3, C) ⊂ M3 (C) (resp. GL (3, C)) be the subgroup of matrices
with determinant equal to 1 (resp. not equal to 0). The group of biholomorphic auto-
morphisms of CP2 is given by

PSL (3, C) ∼
= PGL (3, C) := GL (3, C) /{scalar matrices}.

If g ∈ PSL (3, C) (resp. z ∈ CP2 ), we denote by g ∈ SL (3, C) any of its lifts (resp.
z ∈ C3 ). We denote again by [ ] : C3 \{0} → CP2 the quotient map. We denote by
Fix(g) ⊂ CP2 the set of fixed points of an automorphism g ∈ PSL (3, C). We will
denote by Diag(a1 , a2 , a3 ) to the diagonal element of PSL (3, C) with diagonal entries
a1 , a2 , a3 .
As in the case of automorphisms of CP1 , we classify the elements of PSL (3, C) in
three classes: elliptic, parabolic and loxodromic. However, unlike the classical case,
there are several subclasses in each case (see Section 4.2 of Cano et al. (2013)). We
now give a quick summary of the subclasses of elements we will be using.

Definition 2.1 An element g ∈ PSL (3, C) is said to be:

• Elliptic if it has a diagonalizable lift in SL (3, C) such that every eigenvalue has
norm 1.
• Parabolic if it has a non-diagonalizable lift in SL (3, C) such that every eigenvalue
has norm 1.
• Loxodromic if it has a lift in SL (3, C) with an eigenvalue of norm distinct of 1.
Furthermore, we say that g is:
– Loxo-parabolic if it is conjugate to an element h ∈ PSL (3, C) such that
⎛ ⎞
λ 1 0
h = ⎝ 0 λ 0 ⎠ , |λ| = 1.
0 0 λ−2

– A complex homothety if it is conjugate to an element h ∈ PSL (3, C) such that

h = Diag λ, λ, λ−2 , with |λ| = 1.

123
 55 Page 8 of 28 M. Toledo-Acosta

– A rational (resp. irrational) screw if it is conjugate to an element h ∈

PSL (3, C) such that h = Diag (λ1 , λ2 , λ3 ), with |λ1 | = |λ2 | = |λ3 | and
λ1 λ−1
2 =e
2πiθ with θ ∈ Q (resp. θ ∈ R\Q).

– Strongly loxodromic if it is conjugate to an element h ∈ PSL (3, C) such

that h = Diag (λ1 , λ2 , λ3 ), where the elements {|λ1 |, |λ2 |, |λ3 |} are pairwise
different.
If  ⊂ PSL (3, C) is a group, we denote by  p ⊂  the subgroup generated by all
parabolic elements of . We will say that  p is the parabolic part of .
An element γ ∈  is called a torsion element if it has finite order. The group  is a
torsion free group if the only torsion element in  is the identity. We denote the rank
of a group  by r ().
 
Definition 2.2 Let G be a group. The derived series G (i) of G is defined inductively
as
 
G (0) = G, G (i+1) = G (i) , G (i) .

One says that G is solvable if, for some n ≥ 0, we have G (n) = {id}.

2.2 Upper Triangular Groups

We denote the upper triangular subgroup of PSL (3, C) by

⎧⎡ ⎤ ⎫
⎨ a11 a12 a13 ⎬
U+ = ⎣ 0 a22 a23 ⎦ a11 a22 a33 = 1, ai j ∈ C .
⎩ ⎭
0 0 a33

Let λ12 , λ23 , λ13 : (U+ , ·) → (C∗ , ·) be the group morphisms given by
  −1   −1   −1
λ12 αi j = α11 α22 , λ23 αi j = α22 α33 , λ13 αi j = α11 α33 .
 
If  ⊂ U+ is a group, we simplify the notation by writing Ker λi j instead of
Ker λi j ∩ .
The following theorem states that, up to finite index, we can consider upper trian-
gular subgroups of PSL (3, C) instead of solvable subgroups of PSL (3, C).
Theorem 2.3 (Theorem 3.6 of Wehrfritz (1973)) Let G ⊂ GL (3, C) be a discrete
solvable subgroup, then G is virtually triangularizable.
The following proposition can be found in Barrera et al. (2022).
Proposition 2.4 (Lemma 4.10 of Barrera et al. (2022)) Let  ⊂ U+ be a discrete
group, then  contains a finite index subgroup 0 ⊂  such that the groups 0 ,
λ12 (0 ), λ23 (0 ) are torsion free and finitely generated.
Now, we recall some results from Toledo-Acosta (2021) that will be used in Sect.
3.

123
Representations of Solvable Subgroups of PSL (3, C) Page 9 of 28 55

Proposition 2.5 (Corollary 4.8 of Toledo-Acosta (2021)) Let  ⊂ U+ be a non-

abelian, torsion-free discrete subgroup, then  cannot contain complex homotheties.
Recall the definition of the following purely parabolic subgroup of  which deter-
mines the dynamics of .

Core() = Ker() ∩ Ker(λ12 ) ∩ Ker(λ23 ).

We denote the elements of Core() by

⎡ ⎤
1 x y
gx,y = ⎣0 1 0⎦.
0 0 1

For the sake of simplicity, we also denote

⎡ ⎤
1 0 x
h x,y = ⎣ 0 1 y ⎦ , x, y ∈ C.
0 0 1

Theorem 2.6 (Theorem 5 of Toledo-Acosta (2021)) Let  ⊂ U+ be a non-abelian,

torsion-free, complex Kleinian group, then  can be written in the following way

 = Core()  ξ1  · · ·  ξ p  η1  · · ·  ηm  γ1  · · ·  γn

where

λ23 () = λ23 (γ1 ), . . . , λ23 (γn ) , n = r (λ23 ()) .

λ12 (Ker(λ23 )) = λ12 (η1 ), . . . , λ12 (ηm ) , m = r (λ12 (Ker(λ23 )) .
(Ker(λ12 ) ∩ Ker(λ23 )) = (ξ1 ), . . . , (ξ p ) , p = r ( (Ker(λ12 ) ∩ Ker(λ23 ))) .

Theorem 2.7 (Theorem 7 of Toledo-Acosta (2021)) Let  ⊂ U+ be a non-abelian,

torsion free, complex Kleinian group, then r () ≤ 4. Using the notation of Theorem
2.6, it holds k + p + m + n ≤ 4, where k = r (Core()).

Table 1 The decomposition of a non-abelian subgroup of U+ in four layers

Parabolic Loxodromic
     
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 x y 1 x y α x y α x y
⎣0 1 0⎦ ⎣0 1 z ⎦ ⎣0 β z ⎦ ⎣0 β z ⎦
0 0 1 0 0 1 0 0 β 0 0 γ
z = 0 α  = β, z  = 0 β = γ
Loxo-parabolic Loxo-parabolic Complex homothety Strongly loxodromic
Core() A\Ker() Ker(λ23 )\A
Parabolic Part  p Third Layer Fourth Layer

123
 55 Page 10 of 28 M. Toledo-Acosta

Theorem 2.6 gives a decomposition of the group  in four layers, the first two
layers are made of parabolic elements and the last two layers are made of loxodromic
elements. The description of these four layers are summarized in Table 1.

3 Representations of Upper Triangular Groups

In this section, we prove the technical lemmas necessary to give an explicit description
of all upper triangular groups. These results will be organized and summarized in Sect.
4. For further details, see Toledo-Acosta (2019).
The following result from Barrera et al. (2022) describes the parabolic part of a
complex Kleinian group  ⊂ U+ . Our results will describe the loxodromic elements
compa-tible with these parabolic parts.

Theorem 3.1 (Theorem 3.1 of Barrera et al. (2022)) Let ˜ ⊂ PSL (3, C) be a complex
Kleinian group without loxodromic elements, then there exists a finite index subgroup
 ⊂ ˜ conjugate to one of the following groups.

(1) An elliptic group,

⎧⎡ ⎤ ⎫
⎨ μ(w) wμ(w) 0 ⎬
Ell(W , μ) = ⎣ 0 μ(w) 0 ⎦ w∈W
⎩ ⎭
0 0 μ(w)−2

where W ⊂ C is a discrete subgroup and μ : (W , +) → (S1 , ·) is a group

morphism.
(2) A torus group T (W ), where W ⊂ C2 is a additive subgroup.
(3) An abelian Kodaira group,

⎧⎡ 2 ⎤ ⎫
⎨ 1 x L(x) + x2 + w ⎬
Kod0 (W , R, L) = ⎣ 0 1 x ⎦ x ∈ R, w ∈ W
⎩ ⎭
0 0 1

where W ⊂ C is an additive discrete subgroup, R ⊂ C is an additive subgroup

and L : R → C is an additive function such that
• If R is discrete, then r (W ) + r (R) ≤ 4.
• If R is not discrete, then r (W ) ≤ 1, r (W ) + r (R) ≤ 4 and

lim L(xn ) + wn = ∞
n→∞

for any sequence {wn } ⊂ W , and any sequence {xn } ⊂ R such that xn → 0.
(4) A dual torus group, where W ⊂ C2 is a discrete additive subgroup with r (W ) ≤ 2.

123
Representations of Solvable Subgroups of PSL (3, C) Page 11 of 28 55

(5) An Inoue group,

⎧⎡ ⎤k ⎡ ⎤l ⎡ ⎤m ⎫
⎪
⎨ 1 1 0 1 qp 1s 1 x y ⎪
⎬
Ino(w) = ⎣ 0 1 0 ⎦ ⎣ 0 1 0 ⎦ ⎣ 0 1 1 ⎦ k, l, m ∈ Z ,
⎪
⎩ 0 0 1 ⎪
⎭
0 0 1 0 0 1

where w = (x, y, p, q, s); x, y ∈ C, p, q, s ∈ Z such that q, s = 0, p, q are

co-primes and q 2 divides s.
(6) An extended Kodaira groups of type 1 or 2,
⎧⎡ ⎤⎡ ⎤n ⎡ ⎤m ⎫
⎨ 1 0 w 1 1 0 1 x+z y ⎬
m, n ∈ Z
Kodi (W , x, y, z) = ⎣ 0 1 0 ⎦ ⎣ 0 1 1 ⎦ ⎣ 0 1 z⎦ ,
⎩ w∈W ⎭
0 0 1 0 0 1 0 0 1

with i = 1, 2. W ⊂ C is a non-trivial, additive discrete group, and x, y, z ∈ C

are such that x ∈ W \ {0} and, either:
Type i =1: z ∈
/ R.
Type i =2: z ∈ R\Q and r (W ) = 1.

Given a non-abelian, torsion free complex Kleinian group  ⊂ U+ , the parabolic

part of  is conjugate to one of the six options given by Theorem 3.1. According to
Theorem 2.6,  p correspond to the semidirect product of elements of the first two
layers. Therefore, in order to give the full description of , we have to check whether
we can add one, two or three loxodromic elements of the third or fourth layer, taking
into account that the rank of  has to be at most four. In Table 2 we summarize all
these possible cases.
We denote by N3 and N4 the groups generated by the elements of the third and
fourth layer respectively. Given m, n not simultaneously 0, we say that a discrete group
 ⊂ U+ belongs to case kmn if r ( p ) = k, r (N3 ) = m and r (N4 ) = n. Additionally,
we say that  belongs to subcase kmn( j) if  p is conjugate to a group given by case
(j) of Theorem 3.1.
Observe that purely parabolic cases 400, 300, 200 and 100 have already been
described in Barrera et al. (2022) (these are the cases listed in Theorem 3.1).

Table 2 All the possibilities for the rank of each layer of 

Case r( p ) r(N3 ) r(N4 ) Case r( p ) r(N3 ) r(N4 ) Case r( p ) r(N3 ) r(N4 )

400 4 0 0 202 2 0 2 111 1 1 1

310 3 1 0 201 2 0 1 110 1 1 0
301 3 0 1 200 2 0 0 103 1 0 3
300 3 0 0 130 2 0 0 102 1 0 2
220 2 2 0 121 1 2 1 101 1 0 1
211 2 1 1 120 1 2 0 100 1 0 0
210 2 1 0 112 1 1 2

123
 55 Page 12 of 28 M. Toledo-Acosta

We will separately describe subcases kmn(1), . . . , kmn(6) in each of the following

subsections.
We will repeatedly use the arguments given in the following lemma in the following
subsections. The proof of the lemma is straightforward.
Lemma 3.2 Let  ⊂ U+ a discrete non-abelian group and let  p ⊂  be the subgroup
generated by all the parabolic elements of .
(i) Using the decomposition given in Theorem 2.6, if  = γ1  · · ·  γ N for 1 <
m ≤ N , it follows that γm−1 γ1 ...γm−1 γm = γ1 1 ...γm−1
p pm−1
for some p1 , ..., pm−1 ∈
Z. This is a consequence of the normality condition  of the
 semidirect product
(ii) Let γ1 , γ2 ∈  be two loxodromic elements, then γ1 , γ2 ∈  p .

3.1 The Parabolic is an Elliptic Group

The next lemma states that subgroups of U+ whose parabolic part is an elliptic group
are abelian. Representations of abelian discrete subgroups of U+ have already been
described in Section 4 of Toledo-Acosta (2021).
Lemma 3.3 Let  ⊂ U+ be a complex Kleinian group such that  p is conjugate to
an elliptic group Ell(W , μ) for some subgroup W ⊂ C, and a group morphism μ as
in the case (1) of Theorem 3.1. Then  is abelian.
Proof We can assume that  p = Ell(W , μ), on the other hand, μ−3 (w) = 1 for any
w ∈ W . Then λ23 (γ ) has infinite order for any γ ∈  p , it follows from Lemma 4.13
of Barrera et al. (2022) that  is abelian. 


3.2 The Parabolic Part is a Torus Group

The following lemma discards subcases 110(2), 111(2), 112(2), 120(2), 121(2), 130(2),
210(2), 211(2), 220(2) and 310(2).
Lemma 3.4 Let  ⊂ U+ be a complex Kleinian group such that  p is conjugate to a
torus group T (W ) for some discrete group W as in the case (2) of Theorem 3.1. Then
 cannot contain elements in its third layer.
Proof Let us assume that  p = T (W ) for some additive subgroup W ⊂ C2 , then
 p = h x,y (x, y) ∈ W . Let us suppose that there exists a loxodromic element
γ ∈ N3 ,
⎡ ⎤
α −2 γ12 γ13
for some α, γ23 ∈ C∗ , |α| = 1
γ = ⎣ 0 α γ23 ⎦ ,
γ12 , γ13 ∈ C.
0 0 α

Conjugating  by an appropriate element of PSL (3, C), we can assume that γ12 = 0
and γ23 = 1. Assume without loss of generality that |α| > 1, and consider the sequence
of distinct elements γ k h x,y γ −k ⊂ , it follows that γ k h x,y γ −k → h α −3k x,y ∈
PSL (3, C), contradicting that  is discrete. 


123
Representations of Solvable Subgroups of PSL (3, C) Page 13 of 28 55

The following lemma discards subcases 101(2), 102(2) and 103(2).

Lemma 3.5 Let  ⊂ U+ be a complex Kleinian group such that  p is conjugate to

a torus group T! (W ) for some discrete group W as in the case (2) of Theorem 3.1. If
T (W ) = h x,y with y = 0,  contains no loxodromic elements in its fourth layer.
!
Proof Assume that  p = h = h x,y , for some x,y ∈ C such that y = 0. Let us
suppose that there exists a loxodromic element γ = γi j ∈ N4 that, up to conjugation,
has the form
⎡ ⎤
α 0 γ13
α, β ∈ C∗
γ = ⎣0 β γ23 ⎦ ,
αβ 2 = 1.
0 0 α −1 β −1

Since h is a normal subgroup of h, γ , we have that γ hγ −1 = g n for some n ∈ Z.

Comparing the entries 13 and 23 in the previous equation yields α 2 βx = nx and
αβ 2 y = ny.
If x = 0, y = 0, the equation above implies that α = β, which in turn implies
that γ is a complex homothety, contradicting Proposition ??. If x = 0, y = 0, the
sequence τk := γ k hγ−k = h 0,ζ , where ζ = (αβ 2 )k y, contains a subsequence of
distinct elements τk j such that, either τk j → id or τk j → h (depending on whether
αβ 2 = 1 or αβ 2 = 1). 


In the previous lemma, we omit the case y = 0 because it will be covered by Lemma
3.13. The following lemma describes case 301(2).

Lemma 3.6 Let  ⊂ U+ be a complex Kleinian group such that  p is conjugate to

a torus group T (W ) for some discrete group W as in the case (2) of Theorem 3.1. If
r ( p ) = 3 and r (N4 ) = 1 then one generator of N4 is
" #
γ = Diag α 2 β, αβ 2 , 1 , with α = 1 and αβ 2 ∈
/ R.

Proof One can assume that  p = T (W ). Take three generators of T (W ) denoted by

h i = h xi ,yi , i = 1, 2, 3. Let γ ∈ N4 be a loxodromic
 generator
 of N4 , up to a suitable
conjugation, one can assume that γ = Diag α 2 β, αβ 2 , 1 , for some α, β ∈ C∗ . Since
−1 k k k
 p is normal in  we  that γ h i γ = h 1 h 2 h 3 for some ki, j ∈ Z, i = 1, 2, 3.
 have
i,1 i,2 i,3

Let us define A = ki, j ⊂ PSL(3, Z), then

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
x1 x1 y1 y1
A ⎣ x2 ⎦ = α 2 β ⎣ x2 ⎦ , A ⎣ y2 ⎦ = αβ 2 ⎣ y2 ⎦ .
x3 x3 y3 y3

This means that α 2 β and αβ 2 are eigenvalues of A, with respective eigenvectors

[x1 : x2 : x3 ] and [y1 : y2 : y3 ]. A direct calculation shows that α = 1 and αβ 2 ∈
/ R.
Using (1) of Theorem 4.4 of Cano and Seade (2014) we conclude the proof. 


123
 55 Page 14 of 28 M. Toledo-Acosta

3.3 The Parabolic Part is an Abelian Kodaira Group

The following lemma discards all subcases kmn(3).

Lemma 3.7 Let  ⊂ U+ be a complex Kleinian group such that  P is conjugate to an

abelian Kodaira group Kod0 (W , R, L) for some R, L, W as in the case (3) of Theorem
3.1. Then  cannot contain loxodromic elements.

Proof Assume that  p = Kod0 (W , R, L). Let us suppose that  contain a loxodromic
element γ ∈ . If γ ∈ N3 then

⎡ ⎤
α −2 γ12 γ13
for some α ∈ C∗ , |α| = 1
γ =⎣ 0 α γ23 ⎦ ,
γ12 , γ13 ∈ C and γ23 ∈ C∗ .
0 0 α

Since  p is a normal subgroup of , we have γ gγ −1 ∈  p for any g ∈  p . If

⎡ 2 ⎤
1 x L(x) + x2 + w
g = ⎣0 1 x ⎦ , then
0 0 1
⎡ " # ⎤
2 γ23 x
1 α −3 x α L(x) + x2 + w −
⎢ 1−α 3 ⎥
γ gγ −1 = ⎣ 0 1 x ⎦, (3.1)
0 0 1

and, since γ gγ −1 ∈  p , then α −3 x = x, which is impossible, given that |α| = 1.

Now, if γ ∈ N4 , then

⎡ ⎤
α γ12 γ13
γ = ⎣0 β γ23 ⎦
0 0 α β −1
−1

with α, β ∈ C∗ , α −1 β −1 = β (or equivalently, αβ 2 = 1). Again, since  p is a normal

subgroup of , we have γ gγ −1 ∈  p for any g ∈  p . If g is given by (3.1), then

⎡ " # ⎤
x2
1 αβ −1 x α 2 β L(x) + + w + αγ23 x(β − α)
⎢ 2 ⎥
γ gγ −1 = ⎣ 0 1 αβ 2 x ⎦,
0 0 1

and, since γ gγ −1 ∈  p , then αβ −1 = 1 and αβ 2 = 1, but this last condition is a

contradiction. This contradictions prove the lemma. 


123
 Representations of Solvable Subgroups of PSL (3, C) Page 15 of 28 55

3.4 The Parabolic Part is a Dual Torus Group

Now we deal with subcases kmn(4). The following proposition describes the form of
 p , up to conjugation. Its proof is straight-forward, and its based on Lemma 2.3 and
2.5 of Barrera et al. (2022).

Proposition 3.8 Let  ⊂ U+ be a complex Kleinian group such that  p is conjugate

to a dual torus group T ∗ (W ) for some discrete additive subgroup W ⊂ C2 , as in (4)
of Theorem 3.1. Then  p is conjugate to one of the following groups:
!
(i) 1 = g1,0 , g0,1 !.
(ii) 2 = g0,1 , g0,y !, with y ∈
/ R.
(iii) 3 = g1,0 ,! gx,0 , with x ∈
/ R.
(iv) 4 = g1,0 .
!
(v) 5 = g0,1 .

In the context of Lemma 2.3 and 2.5 of Barrera et al. (2022), 1 is a dual torus
group of type I, and the rest are dual torus group of type II.
The following lemmas study the different combinations of  p and elements of the
third and fourth layer of . In all of these lemmas,  is a discrete subgroup of U+ ,
and  p denotes the parabolic part of .
!
Lemma 3.9 Suppose that  p is conjugate to 1 = g1,0 , g0,1 .
(i) If rank(N3 ) = 1, one generator of N3 is given by
⎡ ⎤
p γ12 γ13
for p ∈ Z\ {−1, 0, 1} , q ∈ Z\ {0}
γ =⎣0 1
q ⎦
, (3.2)
p γ12 , γ13 ∈ C.
0 0 1

(ii) If rank(N3 ) = 2, one generator of N3 is given by (3.2), and the other is given by
⎡ ⎤
p2 μ12 μ13
μ=⎣ 0 2 ⎦
q
1 p2 , (3.3)
0 0 1

with p2 ∈ Z\ {−1, 0, 1}, q2 ∈ Z\ {0}, μ12 , μ13 ∈ Q\ {0} and

p1 p2 (m + j p1 p2 )
μ12 = −
(1 − p1 )( p1 q2 + p2 q1 )
 
p2 mq1 + j p12 ( p2 q1 − (1 − p1 )q2 )
μ13 =
(1 − p1 )2 ( p1 q2 + p2 q1 )

where j, m ∈ Z are integers such that p1 p2 + p1 q2 | m + j p1 p2 and p1 p2 +

p1 q2 | mp1 q2 − j p1 p22 q1 .

123
 55 Page 16 of 28 M. Toledo-Acosta

(iii) If rank(N3 ) = 0 and rank(N4 ) = 1, one generator of N4 is given by

⎡ ⎤
pq γ12 γ13 for p, q ∈ Z\ {0} such that
γ =⎣ 0 q r ⎦ , γ is loxodromic, (3.4)
0 0 p r ∈ Z, γ12 , γ13 ∈ C.

(iv) It’s not possible to have rank(N4 ) = 2.

(v) If rank(N3 ) = 1, then there are no elements in the fourth layer N4 .

Proof We can assume that  p = 1 . Let g ∈  p be an element with the form g = gn,m
for some n, m ∈ Z. We prove each conclusion separately:

(i) Let γ ∈ N3 be a generator of the third layer of . Then

⎡ ⎤
α −2 γ12 γ13
γ =⎣ 0 α γ23 ⎦ , for some γ23 , α ∈ C∗ , |α| = 1.
0 0 α

Since  p is a normal subgroup of , then γ gγ −1 ∈  p , which means that

α −3 n, α −4 (mα − nγ23 ) ∈ Z for any n, m ∈ Z (see (i) of Lemma 3.2). In
particular, for n = 0, m = 1 we get α −3 = p for some p ∈ Z\ {0}, for
4
n = 1, m = 0 we get γ23 = qp − 3 for some q ∈ Z\ {0}. Then

⎡ 2 ⎤ ⎡ ⎤
γ12 γ13
1 1
p3 p p 3 γ12 p 3 γ13
⎢ 1 4 ⎥ ⎢ ⎥
p − 3 qp − 3 ⎦ = ⎣ 0
q
γ =⎣ 0 1 p ⎦,
1
−3 0 0 1
0 0 p

and since γ12 and

 γ13 are arbitrary, we get (3.2).
(ii) Let μ = μi j be a second generator of N3 . Conjugating  by an adequate
element of PSL (3, C), we can assume that

⎡ ⎤
p1 0 0
γ =⎣ 0 1 ⎦.
q1
p1
0 0 1

 
Using the proof of (i), μ has the form given by (3.3). Observe that if μ, γ = id
then
 μ 12 = μ13 = 0 and then  would be abelian. Then, we can assume that
μ, γ ∈  p \ {id} (see (ii) of Observation 3.2), comparing the entry 13 of the
last expression yields

qμ12 + (1 − p1 ) p1 μ13
= j (3.5)
p12 p2

123
Representations of Solvable Subgroups of PSL (3, C) Page 17 of 28 55

for some j ∈ Z\ {0}. Since  p  γ is normal in  we have

⎧⎡ ⎤ ⎫
⎪ knq
⎨ p 1 n m + p1 1
k ⎪
⎬
⎢ ⎥
μγ μ−1 ∈ p  γ = ⎣ 0 1 q1
k p1 ⎦ m, n ∈ Z .
⎪
⎩ 0 0 ⎪
⎭
1

Comparing the entries 12 and 13 in the last expression yields

μ12 (1 − p1 ) = n
& '
q1 q2 p1 q 2 q1
μ12 − + μ12 + μ13 (1 − p1 ) = m + n (3.6)
p1 p2 p2 p1

for some m, n ∈ Z. Combining both expresions in (3.6) we get

mp2 + ((1 − p1 )q2 − p2 q1 ) μ12 + p2 μ13 (1 − p1 ) = 0. (3.7)

Solving (3.5) and (3.7) we get the desired expressions for μ12 and μ13 .
(iii) If γ ∈ N4 then
⎡ ⎤
α γ12 γ13
γ = ⎣0 β γ23 ⎦ (3.8)
0 0 α β −1
−1

for some α, β ∈ C∗ such that γ is loxodromic and α = β 2 , γ12 , γ13 , γ23 ∈

C. Using (i) of Lemma 3.2 it follows that γ gγ −1 ∈  p , which means that
αβ −1 , α 2 (mβ − nγ23 ) ∈ Z for any m, n ∈ Z. If n = 0, m = 1 then α 2 β = q
for some q ∈ Z. If n = 1, m = 0 then αβ −1 = p ∈ Z and α 2 γ23 = r ∈ Z. All
this together yields (3.4).
(iv) Assume that rank(N4 ) = 2 and let γ , μ be two generators of N4 . Using an
adequate conjugation we can assume that
⎡ ⎤ ⎡ ⎤
q1 p1 0 γ13 q2 p2 μ12 μ13
γ =⎣ 0 q1 0 ⎦ , μ=⎣ 0 q2 μ23 ⎦ .
0 0 p1 0 0 p2

Then, using normality and Theorem 2.6 we have μγ μ−1 ∈  p  γ . Com-

paring the entries 12, we have

q1 μ12 (1 − p1 )
= −q1 μ12 (1 − p1 )
q2

and therefore, q2 = −1. Then λ13 (μ) is a torsion element in λ13 , this contra-
diction proves that we cannot have rank(N4 ) = 2.

123
 55 Page 18 of 28 M. Toledo-Acosta

(v) Let γ be a generator of N3 , using an adequate conjugation we can assume that

⎡ ⎤
p 0 γ13
with p, q ∈ Z\ {0}
γ = ⎣ 0 1 p ⎦,
q
| p| = 1.
0 0 1

Let us assume that there is an element μ ∈ N4 , then

⎡ ⎤
α μ12 μ13
μ = ⎣0 β μ23 ⎦ ,
0 0 α −1 β −1
 
with αβ 2 = 1. Since γ , μ ∈  p , then comparing the entries 23 yields
q(1 − αβ 2 ) p −1 α −1 β −2 = 0. This means that, either q = 0 or αβ 2 = 1. Both
conclusions contradict the hypotheses. Thus, there are no elements in N4 .


!
Lemma 3.10 Suppose that  p is conjugate to 2 = g0,1 , g0,y , with y ∈ / R.
(i) If rank(N3 ) = 1 then one generator of N3 is given by
⎡ ⎤
( p + qRe(y)) + iqIm(y) γ12 γ13
γ =⎣ 0 1 γ23 ⎦ , (3.9)
0 0 1

for p, q ∈ Z such that | p| + |q| = 0 and |( p + qRe(y)) + iqIm(y)| = 1,

γ12 , γ23 ∈ C and γ23 ∈ C∗ .
(ii) If rank(N3 ) = 0 and rank(N4 ) = 1 then one generator of N4 is given by
⎡ ⎤
α γ12 γ13
γ = ⎣ 0 α −2 z p,q γ23 ⎦ , (3.10)
0 0 αz −1
p,q

for β ∈ C∗ , z p,q = ( p + qRe(y)) + iqIm(y), | p| + |q| = 0, γ12 , γ23 , γ23 ∈ C.

Finally, γ must be loxodromic.
Proof We can assume that  p = 2 . Let g = g0,n+my ∈  p , with n, m ∈ Z. We prove
each conclusion separately:
(i) Let γ ∈ N3 be a generator of the third layer of . Then
⎡ −2 ⎤
α γ12 γ13
for some γ23 , α ∈ C∗
γ =⎣ 0 α γ23 ⎦ ∈ N3 ,
|α| = 1.
0 0 α

Since  p is a normal subgroup of , then γ gγ −1 ∈  p and then, for any

n, m ∈ Z,

α −3 (n + my) = p + qy (3.11)

123
Representations of Solvable Subgroups of PSL (3, C) Page 19 of 28 55

for some p, q ∈ Z. Denote y = a + bi and α −3 = r (cos θ + i sin θ ). In

particular, (3.11) holds for n = 1, m = 0. Then,
 (3.11) becomes  θ =
 r cos
p+qa, r sin θ = qb. Which means that Re α −3 = p+qa and Im α −3 = qb
and therefore, α −3 = ( p + qRe(y)) + iqIm(y) for p, q ∈ Z such that | p| +
|q| = 0 (otherwise, α = 0) and |( p + qRe(y)) + iqIm(y)| = 1 (otherwise, γ
would not be loxodromic). Since we can re-write γ as
⎡ −3 −1 ⎤
α α γ12 α −1 γ13
γ =⎣ 0 1 α −1 γ23 ⎦ ,
0 0 1

we get (3.9).
(ii) If γ ∈ N4 , then γ has the form given by (3.8). Since  p is normal in ,
γ gγ −1 ∈ , and this implies that α 2 β(n + my) = ñ + m̃ y, for some ñ, m̃ ∈ Z.
The same argument and calculations used in the proof of (i) above show that
α 2 β = ( p + qRe(y)) + iqIm(y). Substituting this in the form of γ yields
(3.10).


!
Lemma 3.11 Suppose that  p is conjugate to 3 = g1,0 , gx,0 , with x ∈ / R.
(i)  cannot contain elements in its third layer.
(ii) If rank(N4 ) = 1 then one generator of N4 is given by
⎡ ⎤
p + q x γ12 γ13
γ =⎣ 0 1 0 ⎦ , p, q ∈∗Z; | p| + |q| = 0 (3.12)
−1 −3 β ∈ C , γ12 , γ13 ∈ C,
0 0 ( p + q x) β

such that γ is loxodromic. Also

p − q |x|2 ∈ Z, p + q + 2qRe(x) ∈ Z. (3.13)

(iii) If rank(N4 ) = 2 then one generator is given by (ii) and the other is given by
⎡ ⎤
p2 + q2 x μ12 0
μ=⎣ 0 1 0 ⎦, (3.14)
0 0 ( p2 + q2 x)−1 β2−3

where p2 , q2 ∈ Z such that | p2 | + |q2 | = 0 and both integers satisfy (3.13),

β2 ∈ C∗ is such that γ is loxodromic and p2 − q |x|2 ∈ Z. Furthermore,
μ12 ∈ C such that

μ12 = (n + mx − (x + 1)( p2 + q2 x)) (1 − ( p1 + q1 x))−1 (3.15)

for some n, m ∈ Z such that

p1 + q 1 x − 1
μ12 = k1 + k2 x, k1 , k2 ∈ Z. (3.16)
( p1 + q1 x)( p2 + q2 x)

123
 55 Page 20 of 28 M. Toledo-Acosta

Proof We can assume that  p = 2 . Let g = gn+mx ∈  p . We prove each conclusion

separately:
(i) The argument is the same used in the first conclusion of Lemma 3.12.
(ii) If γ ∈ N4 , then γ has the form given by (3.8). Since  p is normal in ,
γ gγ −1 ∈  p , and this implies that αβ −1 (n + mx) = ñ + m̃x, and −α 2 γ23 (n +
mx) = 0, for some ñ, m̃ ∈ Z. Then γ23 = 0, and using the same argument and
calculations as in the first part of the proof of Lemma 3.10 we have

αβ −1 = ( p + qRe(x)) + iqIm(x).

Substituting this in (3.8) yields (3.12). Again, since γ gγ −1 ∈  p for γ as in

(3.12), comparing entries 12 yields

p + ( p + q)x + q x 2 = p + q x, (3.17)

for some p , q ∈ Z. To solve (3.17), we write x = Re(x) + iIm(x) and

assume Im(x) = 0. From the equation corresponding to the imaginary part we
get ( p + q) + 2qRe(x) ∈ Z, substituting this into the equation corresponding
to the real part we get p − q |x|2 ∈ Z. This verifies (3.13).
(iii) Let γ be a generator of N4 given by (ii) and let μ be the second generator of
μ, then μ has the same form given by (ii), that is
⎡ ⎤
p2 + q2 x μ12 0
μ=⎣ 0 1 0 ⎦.
−1 −3
0 0 ( p2 + q2 x) β2

We can conjugate  by a suitable A ∈ PSL (3, C) such that

" #
Aγ A−1 = Diag p1 + q1 x, 1, α −3 ( p + q x)−1 .

   
We have that, either μ, γ = id or μ, γ = gk1 ,k2 for some k1 , k2 ∈ Z.
Both possibilities imply that μ13 = 0. If μ, γ = id then μ12 = 0, and then
μ is diagonal. Since γ is diagonal and the parabolic part is abelian, then 
would be abelian (and therefore, it hasalready been described in Section 4
of Toledo-Acosta (2021)). Then, μ, γ = gk1 ,k2 for some k1 , k2 ∈ Z, this
yields (3.16). On the other hand, since  p  γ is normal in , it follows that
μg1,1 γ μ−1 = gn,m γ k for some n, m, k ∈ Z. A direct calculation shows that
k = 1, the remaining expression yields (3.15).


!
Lemma 3.12 Suppose that  p is conjugate to 4 = g1,0 .
(i)  cannot contain elements of the third layer.

123
Representations of Solvable Subgroups of PSL (3, C) Page 21 of 28 55

(ii) If rank(N4 ) = 1, one generator of N4 is given by

⎡ ⎤
pα γ12 γ13 p ∈ Z\ {0, 1} , α ∈ C∗
γ =⎣ 0 α 0 ⎦ , |α| = 1 (3.18)
0 0 p −1 α −2 γ12 , γ13 ∈ C.

(iii) If rank(N4 ) = 2, one generator of N4 is given by (3.18), and the other satisfies
one of the following conditions:
(iv) If p 2 α 3 = 1 then
⎡ β j pq ⎤
qβ 1− p μ13
μ=⎣ 0 β 0 ⎦ , μ13 ∈ C.
0 0 −1
q β −2

(v) If p 2 α 3 = 1 then
⎡ β j pq ⎤
qβ 1− p 0
μ=⎣ 0 β 0 ⎦.
0 0 q −1 β −2

In both cases, β ∈ C∗ , j, q ∈ Z and q = 0 such that μ is loxodromic.

(vi) If rank(N4 ) = 3, two generators of N4 are given by (ii) and (iii) respectively,
and the third generator η is given by
⎡ δkpr ⎤
rδ 1− p η13
η=⎣ 0 δ 0 ⎦, (3.19)
0 0 q δ −2
−1

where δ ∈ C∗ , r ∈ Z\ {−1, 0, 1}, r δ 3 = 1 and

 
pα 2 γ13 (1 − r 2 δ 3 ) − pqαβ 2 μ13 (1 + r 2 δ 3 )
η13 =
r δ 2 (1 − p 2 q 2 α 3 β 3 )
(1 − p) (α(n + r ) − γ12 (1 − r )) − j2 p 2 qα(1 − r )
k=
pr α(1 − pq)

for n ∈ Z. Also, either

kr (1 − q) = jq(1 − r ) or qr (1 − p) | p (kr (1 − q) − jq(1 − r )) ,

depending on whether μ and η commute.

Proof We assume that  p = 4 , we denote g = g1,0 . We prove each conclusion

separately:

123
55 Page 22 of 28 M. Toledo-Acosta

(i) Assume that  contain an element of the third layer,

⎡ ⎤
α −2 γ12 γ13
γ , α ∈ C∗
γ =⎣ 0 α γ23 ⎦ ∈ N3 , 23 (3.20)
|α| = 1
0 0 α

Using (i) of Lemma 3.2, γ gγ −1 ∈  p . Comparing entries 13 yields γ23 = 0,

contradicting that γ ∈ N3 .
(ii) If rank(N4 ) = 1, let γ ∈ N4 be an element given by
⎡ ⎤
β γ12 γ13
α 2 β = 1, α = β
γ =⎣0 α γ23 ⎦ , (3.21)
γ is loxodromic.
0 0 α −1 β −1

Since  p is a normal subgroup of , then γ gγ −1 ∈  p . Comparing entries 13

yields γ23 = 0 and, for n = 1 in particular, βα −1 ∈ Z, in other words, β = pα
for any p ∈ Z. All of these conditions together with (3.21) imply (3.18).
(iii) Let γ ∈ N4 be one generator ofN4 given by (3.18).
 Up to a suitable conjugation
we can assume that γ = Diag pα, α, p −1 α −2 , with p ∈ Z\ {0} and α ∈ C∗ .
Using part (ii), we know that
⎡ ⎤
qβ μ12 μ13
μ=⎣ 0 β 0 ⎦.
0 −1
0 q β −2

 
Since μ, γ ∈ g , comparing the respective entries we have μ13 (1− p 2 α 3 ) = 0
and μ12 (1 − p) = β j pq for some j ∈ Z. Therefore

β j pq
μ13 = 0 or p 2 α 3 = 1; and μ12 = . (3.22)
1− p

A direct calculation shows that these two conditions are equivalent to prove that
 p  γ is normal in  p  γ  μ . Therefore the only restrictions for μ are
(3.22). This proves this part of the lemma.
(iv) If rank(N4 ) = 3, we denote by γ and μ the first two generators of N4 , they have
the forms given by (ii) and (iii) respectively. Furthermore, η has the form (3.19).
By the normality of  p  γ  μ in  we have ηgγ μη−1 = g n γ m 1 μm 2
for some integers n, m 1 , m 2 ∈ Z. Comparing entries 12 and 13 in both sides in
the previous equation and solving for η13 and k yields the expressions for these
variables. On the other hand, [μ, η] ∈  p , let ξ be the entry 13 of [μ, η], then
 
qβμ13 1 − r 2 δ 3
ξ=   .
r δ2 1 − q 2 β 3

If [μ, η] = id then ξ = 0, if [μ, η] = id the ξ ∈ Z. This completes the proof.




123
Representations of Solvable Subgroups of PSL (3, C) Page 23 of 28 55

!
Lemma 3.13 Suppose that  p is conjugate to 5 = g0,1 .
1. If rank(N3 ) = 1, one generator of N3 is the element
⎡ ⎤
p γ12 γ13
p ∈ Z\ {−1, 0, 1}
γ = ⎣ 0 1 γ23 ⎦ , (3.23)
γ12 , γ13 ∈ C, γ23 ∈ C∗ .
0 0 1

2. It’s not possible to have rank(N3 ) = 2.

3. If rank(N3 ) = 0 and rank(N4 ) = 1, one generator of N4 is the element
⎡ ⎤
α γ12 γ13 α ∈ C∗ , p ∈ Z\ {0}
γ = ⎣ 0 pα −2 γ23 ⎦ , p 2 = α 3 , (3.24)
0 0 p −1 α γ is loxodromic.

4. If rank(N3 ) = 1,  cannot contain elements in its fourth layer.

5. If rank(N3 ) = 0 and rank(N4 ) = 2, one generator of N4 is given by (3.24), and
the other by
⎡ jp ⎤
β 0 1− p β β ∈ C∗ , q, j ∈ Z\ {0}
μ = ⎣ 0 qβ −2 0 ⎦ , q 2 = β 3 , μ is loxodromic. (3.25)
0 0 q −1 β

6. It is not possible to have rank(N4 ) = 3.

Proof Assume that  p = 5 and let g = g0,n ∈  p , for n ∈ Z. We verify each
conclusion:
(i) Let γ ∈ N3 , then γ has the form given by (3.20). As before, the normality of  p
in  implies that γ gγ −1 ∈  p , and this implies that α −3 n ∈ Z for all n ∈ Z.
1
Taking n = 1, yields α = p − 3 . Substituting this in the form of γ verifies (3.23).
(ii) Let us suppose that rank(N2 ) = 2 with N2 = γ , μ where γ has the form
(3.23). Then μ must have the same form and then we denote
⎡ ⎤
q μ12 μ13
μ = ⎣ 0 1 μ23 ⎦ ,
0 0 1
   
for some q ∈ Z. Lemma 3.2 (ii) states
 that γ , μ ∈  p \ {id} or γ , μ = id. In
either case, the entry 12 of γ , μ satisfies

γ12 (1 − q) − μ12 (1 − p)
= 0. (3.26)
pq

This means that

&( )' &( )'
q μ12 p γ12
Fix = Fix ,
0 1 0 1

123
55 Page 24 of 28 M. Toledo-Acosta

and therefore, conjugating  by A = g−γ12 (1− p)−1 ,0 and using (3.26), we can
assume that
⎡ ⎤ ⎡ ⎤
p 0 γ13 q 0 μ13
γ = ⎣ 0 1 γ23 ⎦ , and μ = ⎣ 0 1 μ23 ⎦ ,
0 0 1 0 0 1

for some p, q ∈ Z\ {0}. Then

 
If γ , μ = id ⇒ γ13 (1 − q) = μ13 (1 − q). (3.27)
  γ13 1 − q − μ13 (1 − p)
If γ , μ = id ⇒ ∈ Z.
pq

Since  p  γ is normal in , then μgγ μ−1 = g n γ k for some n, k ∈ Z.

Comparing entries 23 of both expressions yields k = 1, comparing entries 13
we have

q(1 + γ13 ) + μ13 (1 − p) = 1 + γ13 . (3.28)

   
If γ , μ = id, (3.27) yields q = 1, which cannot happen. If γ , μ = id, (3.27)
implies (γ13 1 − q − μ13 (1 − p)) ( pq)−1 = j, for some j ∈ Z. Combining this
equation with (3.28), we have q = (1 + j p)−1 ∈ Z, which cannot happen.
Therefore there cannot be an element in N3 \ γ .
(iii) The proof is similar to the proof of (i) above.
(iv) If rank(N3 ) = 1, one generator of N3 is the element γ given by (3.23). Let us
assume that there is an element μ ∈ N4 , given by
⎡ ⎤
α μ12 μ13
α, β ∈ C∗ , αβ −2 = 1
μ = ⎣0 β μ23 ⎦ ,
μ is loxodromic.
0 0 α −1 β −1
 
Since γ , μ ∈  p and comparing entries 23, it follows γ23 α −1 β −2 (1 − αβ 2 ) =
0. This means that, either γ23 = 0 or αβ 2 = 1. Both conditions contradict the
hypotheses, then  cannot contain
 elements in N4 .
(v) If rank(N4 ) = 2, let γ = γi j and μ be two generators of N4 , then γ has the
form given by (3.24), we can conjugate  by an adequate
 element
 of PSL (3, C)
and assume that γ12 = γ13 = γ23 = 0 in (3.24). Since μ, γ ∈  p = 5 , μ has
the form
⎡ ⎤
β 0 μ13
β ∈ C∗ , q ∈ Z\ {0}
μ = ⎣ 0 qβ −2 0 ⎦, 2
q = β 3 , μ is loxodromic.
0 0 q −1 β
   
If μ, γ = id, then  would be abelian. Therefore, μ, γ = id, comparing
 
entries 13 of μ, γ and g yields μ13 p −1 β −1 (1 − p) = j, for some j ∈ Z\ {0}.
Then μ13 = j pβ(1 − p)−1 , this verifies (3.25). 


123
Representations of Solvable Subgroups of PSL (3, C) Page 25 of 28 55

3.5 The Parabolic Part is an Inuoe Group

Using the notation of (5) of Theorem 3.1, it holds r (Ino(w)) = 3. This dismisses
all subcases 1mn(5) and 2mn(5). In the following lemma, we dismiss the remaining
subcases 310(5) and 301(5).

Lemma 3.14 Let  ⊂ U+ be such that  p is conjugate to Ino(w), for some w as in

the case (5) of Theorem 3.1. Then  cannot contain loxodromic elements.

Proof We can assume that  p = Ino(w). We denote the 3 generators of  p by

g1 , g2 , g3 . Since d = 0, it follows from Proposition 3.8 that Ino(w) is conjugate
to
⎧ ⎡ ⎤m ⎫
⎪
⎪ 1 x − psq y sy ⎪
⎪
⎪
⎪ ⎢ ⎥ ⎪
⎪
⎨ ⎢ ps ⎥ ⎬
l ⎢0 1 − q s ⎥
G = g1,0 g0,1 ⎢
k
⎥ k, l, m ∈ Z .
⎪
⎪ ⎣ ⎦ ⎪
⎪
⎪
⎪ 0 − q ps 2
1+ q ps ⎪
⎪
⎩ ⎭

We consider these generators of  p instead, we still denoted them by g1 , g2 , g3 .

Let us suppose first that there is an element γ ∈ N3 . Observe that g1 , g2 is normal
in  and therefore γ g1 g2 γ −1 ∈ g1 , g2 , Lemma 3.9 yields
⎡ ⎤
p γ12 γ13
γ =⎣0 1
q ⎦ , p ∈ Z\ {−1, 0, 1}
p q ∈ Z\ {0} .
0 0 1

Since  p is normal in  p  γ then γ g3 γ −1 = g1k g2n g3m for some k, n, m ∈ Z.

Comparing entries 32 yields m = 1, substituting in the equation resulting of comparing
entries 22 yields q = 0, which is a contradiction.
Now, assume that there is a loxodromic element γ ∈ N4 . Using the same argument
as in the case of an element in the third layer we know that
⎡ ⎤
ab γ12 γ13
a, b, c ∈ Z
γ =⎣ 0 a c ⎦,
p, q = 0.
0 0 b
 
Considering the entry 12 of γ , g1 ∈ g1 , g2 yields 1− b1 ∈ Z, then b = ±1. If b = 1
then γ is a complex homothety, which cannot happen by Proposition 2.5. If b = −1
then λ12 (γ ) is a torsion element in the torsion free group λ12 (). This concludes the
proof. 


3.6 The Parabolic Part is an Extended Kodaira Group

Observe that 3 ≤ r (Kod1 (W , x, y, z)) ≤ 4 depending on whether r (W ) = 1

or r (W ) = 2 respectively. If  ⊂ U+ is a discrete groups with parabolic part

123
 55 Page 26 of 28 M. Toledo-Acosta

conjugated to a non-abelian Kodaira group of type 1 Kod1 (W , x, y, z) such that

r (Kod1 (W , x, y, z)) = 4, then  cannot contain any loxodromic elements (Theo-
rem 2.7). Because of this, we can only consider non-abelian Kodaira group of type 1,
Kod1 (W , x, y, z), such that r (W ) = 1. On the other hand, non-abelian Kodaira group
of type 2, Kod2 (W , x, y, z), always satisfy that r (W ) = 1.
The last paragraph implies that subcases 1mn(6) and 2mn(6) cannot happen. There-
fore, we just need to describe subcases 310(6) and 301(6), but these cases cannot
happen either, as we prove in the following lemma.

Lemma 3.15 If  is such that  p is conjugate to a non-abelian Kodaira group of type

1 or 2 Kodi (W , x, y, z), for some W , x, y, z as in Theorem 3.1(6), then  cannot
contain loxodromic elements.

Proof Assume that  p = Kodi (W , x, y, z) for i = 1, 2, with W , x, y, z as in the

hypothesis of the lemma. Denote by φ1 , φ2 , φ3 the respective generators of  p given
in Theorem 3.1(6).
Let us assume first that there is a loxodromic element γ ∈ N3 in the third layer of
, then
⎡ ⎤
α −2 γ12 γ13
γ =⎣ 0 α γ23 ⎦ (3.29)
0 0 α

for some α ∈ C∗ such that |α| = 1 and γ23 = 0. A direct calculation shows that
⎧⎡ ⎤ ⎫
⎨ 1 mn(x + z) ψk,m,n ⎬
p = ⎣ 0 1 mnz ⎦ k, m, n ∈ Z . (3.30)
⎩ ⎭
0 0 1

The explicit expression for ψk,m,n is not relevant to the proof. Using the normality of
 p in  and comparing the entries 23 of γ φ1 φ2 φ3 γ −1 ∈  p and (3.30), we get that
z + 1 = mnz for some m, n ∈ Z. Since z = 0, it follows that 1 + z −1 ∈ Z, and then
z ∈ Q. This contradicts the conditions for z in non-abelian Kodaira groups of type 1
and 2. Thus,  cannot contain elements in its third layer.
Now, assume that γ ∈ N4 is a loxodromic element in the fourth layer, we denote
⎡ ⎤
α γ12 γ13
γ = ⎣0 β γ23 ⎦ , α, β ∈ C∗ .
0 0 α −1 β −1

Since  p is normal in , and comparing entries 12 and 23 of γ φ1 φ3 γ −1 ∈  p and

(3.30), we get

αβ −1 = mn, and αβ 2 = mn (3.31)

123
Representations of Solvable Subgroups of PSL (3, C) Page 27 of 28 55

for some m, n ∈ Z. From (3.31) it follows that β 3 = 1, and α = pβ for some p ∈ Z.

Therefore,
⎡ ⎤
pβ γ12 γ13
γ =⎣ 0 β γ23 ⎦ , β 3 = 1. (3.32)
0 0 p β −2
−1

Since φ1 is a normal subgroup of , it follows that γ φ1 γ −1 ∈ φ1 . Since r (W ) = 1,

Lemma 3.13(3) implies that
⎡ ⎤
α γ12 γ13
q ∈ Z\ {0} , α ∈ C∗
γ = ⎣ 0 qα −2 γ23 ⎦ , (3.33)
γ is loxodromic.
0 0 q −1 α −1

Comparing the diagonal entries of both (3.32) and (3.33), and since β 3 = 1 it follows
that p = ±1, q = 1, α = p and β = 1. Then γ is either parabolic or induces a torsion
element in λ12 (), both possibilities cannot occur. This finishes the proof. 


4 Proofs of the Main Theorems

In this section, we prove Theorems 1.1, 1.2, 1.3 and 1.4. In all proofs,  ⊂ U+ is
a non-cyclic, upper triangular discrete group containing loxodromic elements. We
assume, without loss of generality, that the groups , λ12 (), λ23 () are torsion free
(Proposition 2.4).
Proof of Theorem 1.1 Suppose that  is abelian. If  p = {id}, then  contains only
loxodromic elements and therefore contains only diagonal elements. Propositions 24
and 25 of Toledo-Acosta (2021) verify (1).
If  p = {id}, Theorem 3.1 and Lemma 3.3 imply that  p is conjugate to an elliptic
group. Since  contains loxodromic elements, Propositions 20 and 23 of Toledo-
Acosta (2021) yield (2).
Proof of Theorem 1.2 If  is not abelian and its parabolic part  p is conjugate to a
torus group T (W ), then Lemma 3.6 yields the conclusion. Lemmas 3.4 and 3.5 prove
that this is the only possible case.
Proof of Theorem 1.3 If  is not abelian and its parabolic part  p is conjugate to a dual
torus group T ∗ (W ) with r (W ) = 2, then  p is conjugate to either 1 , 2 or 3 as
given by Proposition 3.8 (i),(ii),(iii).
If  p ∼
= 1 , Lemma 3.9 verifies conclusions (1), (2) and (3). If  p ∼ = 2 , Lemma
3.10 verifies conclusions (4) and (5). If  p ∼
= 3 , Lemma 3.11 verifies conclusions
(6) and (7).
In all cases, the same lemmas prove that there are no other possible cases.
Proof of Theorem 1.4 If  is not abelian and its parabolic part  p is conjugate to a dual
torus group T ∗ (W ) with r (W ) = 1, then  p is conjugate to either 4 , 5 as given by
Proposition 3.8 (iv),(v).

123
 55 Page 28 of 28 M. Toledo-Acosta

If  p ∼
= 4 , Lemma 3.12 verifies conclusions (1), (2) and (3). If  p ∼
= 5 , Lemma
3.13 verifies conclusions (4), (5) and (6).
In all cases, the same lemmas prove that there are no other possible cases.

Using Lemmas 3.7, 3.14, 3.15 we conclude that there are no upper triangular,
non-cyclic, discrete groups  ⊂ U+ containing loxodromic elements such that  p is
conjugated to an abelian Kodaira group, an Inoue group, or an extended Kodaira group
of type 1 or 2. This verifies that the four Theorems 1.1, 1.2, 1.3 and 1.4 describe, up to
conjugation, all upper triangular, non-cyclic, discrete groups containing loxodromic
elements.
Acknowledgements The author would like to thank to the UCIM UNAM, CIMAT, FAMAT UADY, CINC
UAEM and their people for their hospitality and kindness during the writing of this paper. The author
particularly thanks Angel Cano for the helpful conversations.

Data Availability No datasets were generated or analysed during the current study.

References
Barrera, W., Cano, A., Navarrete, J.P.: Subgroups of PSL(3, C) with four lines in general position. Conform.
Geom. Dyn. 15, 160–176 (2011)
Barrera, W., Cano, A., Navarrete, J.P.: One line complex Kleinian groups. Pac. J. Math. 272(2), 275–303
(2014)
Barrera, W., Cano, A., Navarrete, J.P.: On the number of lines in the limit set for discrete subgroups of
PSL(3, C). Pac. J. Math. 281(1), 17–49 (2016)
Barrera, W., Cano, A., Navarrete, J.P., Seade, J.: Discrete parabolic groups in PSL(3, C). Linear Algebra
Appl. 653, 430–500 (2022)
Barrera, W., Gonzalez-Urquiza, A., Navarrete, J.P.: Duality of the Kulkarni limit set for subgroups of
PSL(3, C). Bull. Braz. Math. Soc. 49, 261–277 (2018)
Cano, A., Navarrete, J.P., Seade, J.: Complex Kleinian Groups. Progress in Mathematics, vol. 303. Springer,
Basel (2013)
Cano, A., Seade, J.: On discrete groups of automorphisms of P2 . Geom. Dedic. 168, 9–60 (2014)
Kodaira, K.: On the structure of compact complex analytic surfaces, i. Am. J. Math. 86(4), 751–798 (1964)
Maskit, B.: Kleinian Groups, Grundlehren der Matematischen Wissenschaften, vol. 287. Springer, Berlin
(1987)
Matsuzaki, K., Taniguchi, M.: Hyperbolic Manifolds and Kleinian Groups. Clarendon Press, Oxford (1998)
Seade, J., Verjovsky, A.: Actions of discrete groups on complex projective spaces. Contemp. Math. 269,
155–178 (2001)
Toledo-Acosta, M.: Three generalizations regarding limit sets for complex kleinian groups. Centro de
Investigación en Matemáticas, PHD Thesis. arXiv:1912.02369 (arXiv preprint) (2019)
Toledo-Acosta, M.: The dynamics of solvable subgroups of PSL(3, C). Bull. Braz. Math. Soc. New Ser.
20, 1–45 (2021)
Wehrfritz, B.: Infinite Linear Groups: An Account of the Group-Theoretic Properties of Infinite Groups of
Matrices. Springer, Berlin (1973)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

123