TITS ALTERNATIVE FOR TWO-DIMENSIONAL ARTIN GROUPS 3
Strategy of the proofs. Our proof of the Tits Alternative is geometric and re-
lies on the action of a two-dimensional Artin group on its Deligne complex (see
Definition 2.4), a particularly well-behaved CAT(0) simplicial complex.
In [MP22], the authors could exploit the dynamics of the acylindrical action
of two-dimensional Artin groups of hyperbolic type on a variation of their Deligne
complex to construct many free subgroups. This approach is no longer possible for
general two-dimensional Artin groups, and the strategy of this paper is therefore
quite different. Instead, given a subgroup Hof AΓ, we will treat separate cases
depending on whether the non-trivial elements of Hall act loxodromically on the
Deligne complex, all act elliptically, or whether Hcontains both loxodromic and
elliptic elements. In particular, we provide a classification result for subgroups of
AΓ(Proposition 4.1). A key case to consider is the case where Hcontains two
elements that act as elliptic isometries of the Deligne complex with disjoint stable
fixed-point sets, where the stable fixed-point set of an isometry is defined as the
union of the fixed-point sets of all its non-trivial powers. We recall some general
questions here:
Question 1.3. Let a,bbe two elliptic isometries of a CAT(0) space Xthat have
disjoint stable fixed-point sets.
(i) Does there exist an element of ⟨a, b⟩that acts loxodromically on X?
(ii) Does there exists a positive integer nsuch that ⟨an, bn⟩is a free group?
Such questions have been studied in the literature, in particular in relation with
the question of the existence of infinite torsion subgroups of CAT(0) groups (ques-
tion (i), see for instance [NOP22, Conjecture 1.5]), and with the Tits Alternative
(question (ii)). A key intermediate result in our proofs of Theorems A and B is the
following:
Proposition C. Let AΓbe a two-dimensional Artin group, and let a,bbe two
elements of AΓthat act elliptically on the Deligne complex DΓwith disjoint fixed-
point sets. Then there exists a positive integer nsuch that ⟨an, bn⟩is a non-abelian
free group.
Moreover, an element of ⟨an, bn⟩acts loxodromically on DΓif and only if it is
not conjugated to a power of anor bn.
Our strategy to prove Proposition C is to construct a tree embedded in the
Deligne complex on which ⟨an, bn⟩acts and such that there is a bijection between
edges of that tree and the reduced words on {a±n, b±n}. To that end, we consider
a geodesic γbetween the fixed-point sets of aand b, and we show that for some
large enough n, the geodesic segment γmakes an angle at least πwith any of its
⟨an⟩-translates or ⟨bn⟩-translates. The CAT(0) geometry of the Deligne complex
then guarantees that the resulting subspace obtained as the union of all ⟨an, bn⟩-
translates of γis a convex subtree of DΓ. (see Figure 1)
Controlling these angles requires a fine understanding of the local structure of
the Deligne complex, and this problem can be translated into a problem about
dihedral Artin groups, a class that is very well understood.
In order to prove Theorem B, the additional assumption that AΓis of hyperbolic
type allows us to exploit the action of the Artin group on its coned-off Deligne com-
plex (see Definition 5.1), a variation of the Deligne complex introduced in [MP22]
on which AΓacts acylindrically. The additional dynamical properties of this action
are a key tool in proving that certain subgroups generated by large powers are free.