Automorphisms of relatively hyperbolic groups and the Farrell–Jones conjecture Authors: Naomi Andrew, Yassine Guerch, Sam Hughes Published: 2026 ABSTRACT AbstractWe prove the fibred Farrell–Jones conjecture (FJC) in A-, K-, and L-theory for a large class of suspensions of relatively hyperbolic groups, as well as for all suspensions of one-ended hyperbolic groups. We deduce two applications: FJC for the automorphism group of a one-ended group hyperbolic relative to virtually polycyclic subgroups;FJC is closed under extensions of FJC groups with kernel in a large class of relatively hyperbolic groups. Along the way we prove a number of results about JSJ decompositions of relatively hyperbolic groups which may be of independent interest.Mathematics Subject Classification: Primary 18F25, Secondary 20F28, 20F65, 20F67, 20E08 Introduction Let G be a group. The Farrell–Jones conjecture (FJC) is one of the most prominent open conjectures in algebraic and differential topology. In its simplest form the K-theoretic conjecture predicts that a certain assembly map is an isomorphism. Here E__G is the classifying space for the family of virtually cyclic subgroups, KR is the algebraic K-theory spectrum for the ring R, and Kn(RG) is the algebraic K-theory of the group ring RG. There are variants of the conjecture for Waldhausen’s A-theory and for L-theory. The conjecture for L-theory, as well as a detailed account of the Farrell–Jones conjecture, and the objects involved can be found in W. Lück’s book project [41]. For recent progress on A-theory the reader should consult [18]. Computing the algebraic K-theory of a group ring RG is a very difficult problem. In principle, knowing that FJC holds for G gives a method of computing Kn(RG) using equivariant algebraic topology. It also has a number of other applications, for example, to the Borel Conjecture [11] and to computing the Whitehead group Wh(G). Knowledge of Wh(G) is a fundamental step in classifications of higher dimensional manifolds with fundamental group G. From this point onward, by the Farrell–Jones conjecture for G, we mean the most general setting, that is, the fibred Farrell–Jones conjecture with respect to the family of virtually cyclic subgroups VC . See for example [18, 41] for a discussion of these terms. We will denote the classes of groups satisfying the FJC for X-theory by FJCX where X is A-, K-, or L-theory. The class FJCK of groups satisfying FJC for algebraic K-theory is large: containing hyperbolic groups [12], many relatively hyperbolic groups [2], CAT(0) groups [50] (see also [11, 37]), soluble groups [51], GLn(Z) [14] and more generally lattices in connected Lie groups [5] and S-arithmetic groups [48], as well as mapping class groups [3], normally poly-free groups [9], and suspensions of virtually torsion free hyperbolic groups [6]. The class enjoys many closure properties: it passes to arbitrary subgroups, finite index overgroups, and directed colimits. For more information the reader is referred to the surveys [1, 13, 40, 42]. One property that is not known is whether FJCX is closed under extensions 1→N→Γ→Q→1. One direction of interest is to put conditions on N so that Γ is in FJCX whenever Q is. By [5, Theorem 2.7] and [18, Theorem 1.1(ii)] this reduces to understanding cyclic extensions of G; which is to say the suspensions NΦ=N⋊ΦZ, where Φ is some automorphism of N defining this suspension. Intuitively, a group G is hyperbolic relative to P if its geometry is hyperbolic “away from the subgroups P∈P.” One (of many: see [33] for the definitions as well as proofs of their equivalence) way to formalise this uses the notion of coning off a Cayley graph: take a vertex for every coset gP of each element of P, and add an edge from each element of gP to the new vertex. The group G is hyperbolic relative to P if the resulting graph is δ-hyperbolic in the sense of Gromov, and fine: every edge is contained in finitely many cycles of a given finite length. An automorphism of G lies in the subgroup Aut(G,P) if it preserves the conjugacy classes of every subgroup P∈P. For more information on Aut(G,P) see [25, 44]. Recently, Bestvina, Fujiwara and Wigglesworth [6] proved the suspension of a virtually torsion free hyperbolic group satisfies the Farrell–Jones conjecture. We extend this result to a large class of relatively hyperbolic groups. Theorem A Let (G,P) be a virtually torsion-free or one-ended relatively hyperbolic group with P finite and let Φ∈Aut(G,P). If for every [P]∈P we have PΦ∈FJCX, then GΦ∈FJCX. These hypotheses include, for instance, all suspensions of toral relatively hyperbolic groups and more generally one-ended or virtually torsion-free groups that are hyperbolic relative to virtually polycyclic or soluble subgroups. Note that this removes the assumption of virtual torsion-freeness in [6] for one-ended hyperbolic groups. This is pertinent since it is a well known question of Gromov whether every hyperbolic group is residually finite (and hence virtually torsion-free). With infinitely ended groups more care is needed, we discuss this further in Sect. 1.2. Applications Our first application is a result on extensions with relatively hyperbolic kernel. A group is non-relatively hyperbolic or NRH if it is not hyperbolic relative to a collection of proper subgroups. Corollary B Let (N,P) be a virtually torsion-free or one-ended relatively hyperbolic group such that P consists of finitely many conjugacy classes of groups which are NRH and whose suspensions P⋊ΨZ are in FJCX for all automorphisms Ψ of P. Let 1→N→Γ→Q→1 be a short exact sequence. If Q is in FJCX, then Γ is in FJCX. The assumption that peripheral subgroups are NRH is needed for Corollary B, since it requires Theorem A to hold for arbitrary automorphisms. The key point being that Aut(N;P) has finite index in Aut(N) under this extra hypothesis. It is a major open problem whether Out(FN) satisfies FJCX. Whilst we do not solve this, using Theorem A we are able to show automorphism groups of one-ended groups hyperbolic relative to virtually polycyclic groups satisfy FJCX. In particular, Aut(G) and Out(G) for G a one-ended hyperbolic group satisfy FJCX. Theorem C If G is a one-ended group hyperbolic relative to finitely many conjugacy classes of virtually polycyclic groups, then Aut(G) and Out(G) are in FJCX. Remarks on the proofs As is usual for (relatively) hyperbolic groups, there are two main flavours to our arguments, depending on the number of ends of G. In both cases we apply a result of Knopf [36] allowing us to deduce that a group acting acylindrically on a tree satisfies the Farrell–Jones conjecture if and only if its vertex groups do, though the source of the trees is different in each case. For one-ended relatively hyperbolic groups, we have access to the powerful machinery of JSJ decompositions developed (in this generality) by Guirardel and Levitt [28]. We consider three related trees: the canonical JSJ decomposition Tcan relative to the peripheral subgroups P, a refinement Tϕ of Tcan which better suits the study of an outer automorphism ϕ and another tree that we call TPer. This tree is the canonical JSJ tree relative to the (non-elementary) periodic subgroups of the outer automorphism ϕ. That is, we require that the periodic subgroups of every representative Φadg, are elliptic. Our main structural result about this tree is Theorem 5.21: even without assuming that the periodic subgroups are finitely generated, they agree exactly with the rigid vertices of TPer. We prove that this ensures that the induced action of G⋊ΦZ is acylindrical, and then analyse vertex groups that can appear in this new action. We remark that the strong uniqueness properties of the JSJ decomposition imply that for a one-ended, torsion free hyperbolic group, the rigid vertex groups of the tree considered in [6] agree with those in our TPer. We consider the case when G is infinitely ended and has a finite index subgroup which is a free product of one-ended groups. Being virtually torsion free is sufficient but not necessary for this to occur. Once we have a free product splitting we have the following combination-type theorem. Theorem D Let G=G1∗…∗Gk∗FN be a free product of finitely generated groups, let F′={[G1],…,[Gk]} and let Φ∈Aut(G,F′). For every i∈{1,…,k}, denote by Φi an element of the outer class of Φ preserving Gi. If for every i∈{1,…,k} the group Gi⋊ΦiZ is in FJCX, then G⋊ΦZ is in FJCX. Theorem D is proved by induction on the Grushko rank k+N. There are two kinds of induction step, depending on whether the maximal periodic free factor system is sporadic or not. A free factor system (G,F) is sporadic if G≅G1∗G2 or G≅G1∗Z. This division might seem unusual to experts; a more standard division (for instance, in [6] as well as throughout the Out(Fn) literature) depends instead on if the automorphism is polynomially or exponentially growing. Polynomially growing automorphisms are always sporadic in this sense, but so are some exponentially growing automorphisms. The non-sporadic case uses Dahmani and Li’s work on relative hyperbolicity for suspensions of free factors [16], whereas in the sporadic case we use the fact that these splittings are rigid. These rigidity arguments hold equally well for sporadic Stallings–Dunwoody decompositions, and so we are still able to obtain some results without first passing to a finite index free product: see Proposition 8.1. Fixed and Periodic Subgroups, the classes AC(VNil) versus FJCX, and localising invariants Some previous results of this flavour have concluded the stronger property that the suspension is in the class AC(VNil). This class consists of groups having particularly nice actions on compact Euclidean retracts, see [3] for a precise definition. Every group in this class satisfies the Farrell–Jones conjecture [3]. This class has similar closure properties to the class of groups satisfying the Farrell–Jones conjecture, except that FJCX is closed under directed colimits while AC(VNil) is not known to be. However, for the majority of the paper we work directly with the class FJCX. The reason for this is that we have to understand the periodic subgroups of certain automorphisms as an ascending union of fixed subgroups, and consider the action of the automorphism on this subgroup. In general our hypotheses do not guarantee that this union stabilises—we do not have a virtual neatness property to rely on. However, there are hypotheses that ensure virtual neatness, and if we assume these then again the suspensions will be in AC(VNil). One set of sufficient conditions is, Theorem 5.24 Let G be a hyperbolic group relative to a collection P of slender groups and let Φ∈Aut(G). There exists N∈N∗ such that Per(Φ)=Fix(ΦN) and Per(Φ) is finitely generated. If we add these hypotheses to our main theorem, we can prove the suspensions lie in AC(VNil). Theorem E Suppose (G,P) is one-ended or virtually torsion free, and hyperbolic relative to finitely many conjugacy classes of virtually polycyclic subgroups. Then for every automorphism Φ of G, Γ:=G⋊ΦZ is in AC(VNil). Except for replacing each periodic subgroup with the fixed subgroup of a power, the proof of this theorem is identical to the proof of Theorem A. We discuss this in a little more detail after completing that proof. Remark Following work of Bunke, Kaprowski, and Winges [10] our results apply equally well to the Farrell–Jones conjecture for localising invariants, that is, with coefficients in H:Cat∞,∗Lex→M a lax monoidal finitary localising invariant with values in a stably monoidal and cocomplete stable ∞-category which admits countable products. We refer the reader to the introduction of loc. cit. for more information. Structure of the paper Section 2 introduces the relevant background results on the Farrell–Jones conjecture. Section 3 contains definitions and results on free products and their automorphisms, needed for Sect. 6. Section 4 collects results on JSJ decompositions of one-ended relatively hyperbolic groups, and provides a lemma on acylindricity when passing to the action of a suspension. The one-ended case of Theorem A is proved in Sect. 5 by careful analysis of a certain JSJ tree. From this analysis, we also deduce Theorem 5.24. In Sect. 6 we prove Theorem D and the infinitely-ended case of Theorem A. Using these results we prove Theorems A and E. In Sect. 7, we deduce Theorem C from Theorem A. Finally, in Sect. 8 we extend Theorem A as far as possible with our current techniques to groups which are infinitely ended but do not split as free products. The arguments of the last three sections are almost independent of Sects. 4 and 5, apart from requiring the background information on trees of cylinders from Sect. 4.1. Background on the Farrell–Jones conjecture For full context and background on the Farrell–Jones conjecture, see for instance Lück’s book project [41]. In this section, we recall some properties of the class FJCX of groups which satisfy the Farrell–Jones conjecture for X-theory where X is A, K, or L. Theorem 2.1 The class FJCX is closed under the following operations: taking subgroups; taking subgroups; taking finite index overgroups; taking finite index overgroups; finite direct products; finite direct products; finite free products; finite free products; directed colimits. directed colimits. Proof The cases of K- and L-theory are given in [29, Theorem 2.1]. The case of A-theory is [18, Theorem 1.1(ii)]. □ While FJCX is not known to be closed under extensions, there is a partial result which we make use of. Theorem 2.2 Let 1→N→Γ→Q→1 be a short exact sequence with N∈FJCX. If for every infinite cyclic subgroup C of Q, the preimage of C in Γ belongs to FJCX, then Γ belongs to FJCX. Proof The cases of K- and L-theory are given in [5, Theorem 1.7]. The case of A-theory is given in [18, Theorem 1.1(ii)]. □ Here is an easy, mild strengthening of commensurability towards virtual isomorphism. Lemma 2.3 Let 1→N→Γ→Q→1 be a short exact sequence with N finite. Then if Q is in FJCX then so is Γ, and if Γ is residually finite and in FJCX then so is Q. Proof For the first statement apply Theorem 2.2 to the short exact sequence, noting that both finite groups and virtually cyclic groups are in FJCX. For the second, observe that if Γ is residually finite then there is a finite index subgroup Γ0 of Γ whose intersection with N is trivial, and then Γ0≅Q0 for some finite index subgroup Q0 of Q. The result follows from commensurability. □ We refer for instance to the work of Bowditch [15] for the definition of a relatively hyperbolic group. Theorem 2.4 (Bartels) Let G be a group hyperbolic relative to a collection {[P1],…,[Pn]} of conjugacy classes of subgroups. If, for every i∈{1,…,n}, we have Pi∈FJCX, then G∈FJCX. Proof This result is due to Bartels. The cases of K- and L-theory are [2, Corollary 4.6]. The case of A-theory is also ostensibly due to Bartels combined with some recent developments on the A-theoretic FJC. We sketch the relevant details. The key here is that Bartels’ space Δ for a relatively hyperbolic group pair (G,P) is finitely P-amenable (see [2, Theorem 3.1]). By [36, Proof of Theorem 1.8(a)] this implies that G is strongly transfer reducible over F. The result now follows from [18, Theorem 6.19]. □ Let G be a group acting by isometries on a tree T. Recall that the action is acylindrical if there exists K≥0 such that the stabiliser of any geodesic path of length at least K is of bounded size. Theorem 2.5 (Knopf) Let G be a group acting acylindrically by isometries on a tree T. If every vertex stabiliser belongs to FJCX, then G belongs to FJCX. Proof The result is due to S. Knopf. For K-theory we refer to [36, Corollary 4.2]. The result for L-theory is [36, Corollary 4.3], note that here one has the additional hypothesis that index 2 overgroups of the stabilisers in G must satisfy FJCL. But this follows from Theorem 2.1. For A-theory, as in Bartels’ result, one combines finite F-amenability [36, Proposition 4.1] with the recent developments for A-theory [36, Proof of Theorem 1.8(a)] and [18, Theorem 6.19]. □ Let G be a group and let Φ∈Aut(G). Let Per(Φ)=⟨Fix(Φn)⟩n≥1 be the periodic subgroup of Φ. At several points in our arguments we will need the following lemma. Lemma 2.6 Let G be a group belonging to FJCX and let Φ∈Aut(G). The group Per(Φ)⋊ΦZ belongs to FJCX. Proof Note that which is an increasing union of subgroups. Therefore, by Theorem 2.1 (5), it suffices to prove that, for every n∈N, the group Fix(Φn!)⋊ΦZ belongs to FJCX. Let n∈N. Note that Fix(Φn!)⋊Φn!Z is a finite index subgroup of Fix(Φn!)⋊ΦZ. By Theorem 2.1 (2), the group Fix(Φn!)⋊ΦZ belongs to FJCX if and only if the group Fix(Φn!)⋊Φn!Z belongs to FJCX. The group Fix(Φn!)⋊Φn!Z is isomorphic to Fix(Φn!)×Z. By Theorem 2.1, the group Z belongs to FJCX. Since G∈FJCX and since FJCX is closed under taking subgroups by Theorem 2.1 (1), the group Fix(Φn!) belongs to FJCX. Since FJCX is closed under taking direct products by Theorem 2.1 (3), the group Fix(Φn!)×Z and hence the group Per(Φ)⋊ΦZ belongs to FJCX. □ Free products of groups and their automorphisms Free products of groups Let N∈N, let G1,…,Gk be countable groups and let G=G1∗…∗Gk∗FN. Let F={[G1],…,[Gk]} be the set consisting of the conjugacy classes of the Gi. We refer to (G,F) as a free product. An element g∈G is peripheral if there exists [A]∈F with g∈A. Otherwise, g is nonperipheral. A subgroup P of G is peripheral if every element of P is peripheral, and is nonperipheral otherwise. A free factor system of (G,F) is a set F′={[A1],…,[Aℓ]} of conjugacy classes of proper subgroups of G such that: for every i∈{1,…,k}, there exists [A]∈F′ such that Gi⊆A; for every i∈{1,…,k}, there exists [A]∈F′ such that Gi⊆A; there exists a subgroup B of G such that G=A1∗…∗Aℓ∗B. there exists a subgroup B of G such that G=A1∗…∗Aℓ∗B. The set of free factor systems of G is equipped with a partial order where F1≤F2 if, for every [A1]∈F1, there exists [A2]∈F2 with A1⊆A2. A free factor system F′ is sporadic if either F′={[A1],[A2]} and G=A1∗A2 or F′={[A1]} and G=A1∗Z. Otherwise, the free factor system F′ is nonsporadic. The free product (G,F) is sporadic (resp. nonsporadic) if F is. We denote by Aut(G,F) the subgroup of automorphisms of G preserving F and by Out(G,F) the subgroup of outer automorphisms of G preserving F. An automorphism Φ∈Aut(G,F) is fully irreducible if no power of Φ fixes a proper free factor system of (G,F). A (G,F)-tree is a tree equipped with an action of G without inversion such that, for every [A]∈F, the group A is elliptic in T. A Grushko (G,F)-tree is a (G,F)-tree T with trivial edge stabilisers and such that, for every v∈VT, the conjugacy class of the stabiliser Gv of v is trivial or contained in F. Let F′ be a sporadic free factor system of (G,F). There is a unique, up to unique G-equivariant homeomorphism, reduced Grushko (G,F′)-tree TF′, which we call the Bass-Serre tree of (G,F′). (In this context, reduced means that no vertices have trivial stabiliser.) The tree TF′ has a unique orbit of edges. The tree TF′ is canonical in the sense that every element Φ∈Aut(G,F′) induces a G-equivariant homeomorphism of TF′. Therefore, for every Φ∈Aut(G,F′), the group G⋊ΦZ acts by homeomorphisms on TF′. Growth under an automorphism of a free product Let (G,F) be a free product and let T be a Grushko (G,F)-tree. We turn T into a metric graph by assigning length 1 to every edge of T. Let g∈G. The translation length of g in T is ‖g‖T=infx∈Td(x,gx). The translation length of g only depends on the conjugacy class of g. Let Φ∈Aut(G,F). An element g∈G has ‖.‖T-polynomial growth under iteration of Φ if there exists P∈Z[X] such that, for every n∈N: Note that any elliptic element of G in T has ‖.‖T-polynomial growth under iteration of Φ. A subgroup P of G is a ‖.‖T-polynomial subgroup of Φ if there exists an automorphism Ψ∈Aut(G,F) contained in the outer class of some power of Φ such that Ψ(P)=P and every element of P has ‖.‖T-polynomial growth under iteration of Ψ. Let PT(Φ) be the set of conjugacy classes of maximal ‖.‖T-polynomial subgroups of Φ. By for instance [21, Theorem 1.1], the set PT(Φ) does not depend on T. When Φ is fully irreducible, the set PT(Φ) satisfies some additional properties. Recall that a subgroup A of G is malnormal if, for every g∈G-A, we have A∩gAg-1={e}. Proposition 3.1 [16, Proposition 1.13] Let (G,F) be a nonsporadic free product and let Φ∈Aut(G,F) be fully irreducible. Let T be a Grushko (G,F)-tree. The set PT(Φ) is finite. The set PT(Φ) is finite. For every [A]∈PT(Φ), the subgroup A is malnormal in G. For every [A]∈PT(Φ), the subgroup A is malnormal in G. Let P={[P1],…,[Pℓ]} be a finite set of conjugacy classes of malnormal subgroups of G. Let Φ∈Aut(G) be an automorphism such that, for every i∈{1,…,ℓ}, there exists gi∈G such that adgi∘Φ(Pi)=Pi. The suspension of P is the set {[Pi⋊adgi∘ΦZ]} considered as a set of conjugacy classes of subgroups of G⋊ΦZ. The following result is due to Dahmani–Li [16]. Theorem 3.2 [16, Corollary 2.3] Let (G,F) be a nonsporadic free product and let Φ∈Aut(G,F) be fully irreducible. Let T be a Grushko (G,F)-tree and let PT(Φ) be the set of conjugacy classes of maximal ‖.‖T-polynomial subgroups of Φ. There exists n∈N such that the group G⋊ΦnZ is hyperbolic relative to the suspension of PT(Φ). In the rest of the section, we give a precise description of the set PT(Φ) for a fully irreducible automorphism. We first need a result, which can be found for instance in the work of Francaviglia–Martino–Syrigos [20] concerning the existence of a limiting tree of a fully irreducible automorphism. Lemma 3.3 [20, Lemma 2.14.1] Let (G,F) be a nonsporadic free product and let Φ∈Aut(G,F) be a fully irreducible automorphism. There exist a Grushko (G,F)-tree S, an R-tree T equipped with an isometric action of G and a constant λ>1 such that, for every g∈G, we have Let (G,F) be a free product. Note that, for any subgroup A of G, the free factor system F induces a free factor system F|A of A. Following the terminology of for instance Guirardel–Horbez [22, Definition 3.2], an R-tree equipped with an isometric action of G is an arational (G,F)-tree if the following holds: the tree T is not a Grushko (G,F)-tree; the tree T is not a Grushko (G,F)-tree; for every [A]∈F, the group A is elliptic in T; for every [A]∈F, the group A is elliptic in T; for every free factor system F1 such that ϕk∈Out(G,P∪H0(t)∪HQH(t)). The result now follows from Theorem 4.2 (10). □ Recall that Tcan and Tϕ are compatible. Let T0ref be their minimal refinement. Since both Tcan and Tϕ are (A,P∪H0∪HQH)-trees, so is T0ref. Lemma 5.5 The tree T0ref is a JSJ tree over elementary subgroups relative to P∪H0∪HQH. In particular, T0ref and Tϕ have the same set of elliptic subgroups. Proof We check [28, Definition 2.11]. We first prove that T0ref is universally elliptic in (A,P∪H0∪HQH)-trees, that is every edge group of T0ref is elliptic in every (A,P∪H0∪HQH)-trees. Let Ge be an edge group of T0ref. By minimality of T0ref, the group Ge is an edge group of either Tcan or Tϕ. As Tcan is a JSJ (A,P∪H0)-tree, it is universally elliptic in (A,P∪H0)-trees, hence it is universally elliptic in (A,P∪H0∪HQH)-trees. Similarly, Tϕ is universally elliptic in (A,P∪H0∪HQH)-trees. Thus, Ge is elliptic in every (A,P∪H0∪HQH)-trees and T0ref is universally elliptic in (A,P∪H0∪HQH)-trees. We now prove that T0ref dominates every universally elliptic (A,P∪H0∪HQH)-trees, that is, every vertex group of T0ref is elliptic in every universally elliptic (A,P∪H0∪HQH)-tree. Since Tϕ is a JSJ (A,P∪H0∪HQH)-tree, it dominates every universally elliptic (A,P∪H0∪HQH)-trees. Since T0ref collapses onto Tϕ, every vertex group of T0ref is elliptic in Tϕ hence is elliptic in every universally elliptic (A,P∪H0∪HQH)-tree. This shows that T0ref dominates every universally elliptic (A,P∪H0∪HQH)-trees. Together with the above paragraph, we obtain that T0ref is a JSJ tree over elementary subgroups relative to P∪H0∪HQH. □ Lemma 5.6 Let v∈Tϕ and let [H]∈PerNP(ϕ). Suppose that v is QH with fibre. Then H∩Gv is elementary. Suppose that v is QH with fibre. Then H∩Gv is elementary. Suppose that v is rigid. If H∩Gv is nonelementary, then Gv⊆H. Suppose that v is rigid. If H∩Gv is nonelementary, then Gv⊆H. Proof Suppose first that v is QH with fibre. Suppose towards a contradiction that H∩Gv is nonelementary. By Lemma 5.5, the group Gv is an elliptic group of the refined tree T0ref. We still denote by v the vertex in T0ref fixed by Gv. Let p:T0ref→Tcan be the natural projection. Since Gv is a flexible vertex, the image p(v) is not a rigid vertex of Tcan (recall that every rigid vertex stabiliser of Tcan is also elliptic in every (A,P∪H0∪HQH)-tree). Hence p(v) is a flexible vertex of Tcan. Therefore, if H∩Gv is nonelementary, then H∩Gv∈HQH. This contradicts Theorem 4.2 (9). Hence H∩Gv is elementary. Suppose now that v is rigid. Let Ψ∈ϕ be such that Per(Ψ)=H. Suppose that the intersection H∩Gv is nonelementary. Then Ψ preserves Gv since v is the unique vertex in Tϕ fixed by H∩Gv. By Lemma 5.4, there exists k∈N∗ such that Ψk acts as a global conjugation on Gv by an element g∈Gv. Taking a larger k if necessary, we may also assume that Ψk acts trivially on a nonelementary subgroup H′⊆H∩Gv. Note that Θ=adg-1∘Ψk acts trivially on Gv. Thus, Ψk and Θ acts trivially on the same nonelementary subgroup H′⊆Gv. By Lemma 5.1, there exists N∈N such that ΨkN=ΘN. In particular, Ψ has a power which fixes Gv elementwise. Thus, we have Gv⊆H. □ Remark 5.7 Note that, for every Φ∈NP(ϕ), the isometry FΦ of Tϕ is elliptic. Indeed, if FΦ is loxodromic, then Φ can only fix an element g∈G which is loxodromic and whose axis is the same as the one of FΦ. Since the action of G on Tϕ is acylindrical, the element g is contained in a unique maximal virtually cyclic subgroup. In particular, Per(Φ) is a virtually cyclic group and Φ∉NP(ϕ). The rest of the section is dedicated to the proof of some properties of the set PerNP(ϕ) and of the action of GΦ on Tϕ. To this end we prove that PerNP(ϕ) is finite (see Lemma 5.12). We need the following lemmas regarding the intersection of characteristic sets of isometries in Tϕ. Lemma 5.8 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let v∈VTϕ be non-elementary and let e1 and e2 be distinct edges adjacent to v. Let ϕ∈Out(G,P). Suppose that there is a representative Φ∈ϕ such that, for every i=1,2, we have Φ(Gei)=Gei. Then v is rigid and Gv is fixed elementwise by some power of Φ. Proof We first prove that v is rigid. Indeed, suppose towards a contradiction that v is QH with finite fibre. Then Ge1 and Ge2 are virtually cyclic. Thus, for every i∈{1,2}, the automorphism Φ has a power Φk fixing an infinite order element gi∈Gei. By Lemma 5.3 and as the groups Ge1 and Ge2 are virtually cyclic, the group ⟨g1,g2⟩ is a non-elementary subgroup. As ⟨g1,g2⟩⊆Per(Φ)∩Gv, this contradicts Lemma 5.6. Thus, the vertex v is rigid. By Lemma 5.4, after taking a power Φℓ, it acts on Gv as global conjugation by an element g∈Gv. We claim that, after taking a further power of Φ, the element g is trivial. Indeed, note that, by Lemma 5.3 the stabiliser of an edge adjacent to v is almost malnormal in Gv: for every edge e′ of Tϕ adjacent to v and every g′∈Gv, if g′Geg′-1∩Ge is infinite then g′∈Ge. Moreover, if e and e′ are two distinct edges adjacent to v, then Ge∩Ge′ is finite. Since Φ preserves Ge1 and Ge2, the power Φℓ must act by conjugating by an element of the finite intersection Ge1∩Ge2. This becomes trivial on taking a further power of Φ, which concludes the proof of the claim. Thus we have shown that Φ has a power fixing Gv elementwise, which concludes the proof. □ Corollary 5.9 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). Let [H]=[Per(Φ)]∈PerNP(ϕ) and let THϕ be the minimal H-invariant subtree of Tϕ. If THϕ contains an edge e, then the endpoints of e are respectively elementary and rigid. If v is the rigid endpoint of e, then Ge⊆Gv⊆H and Gv is fixed elementwise by some power of Φ. Proof Note that, since Tϕ is bipartite, the endpoints of e are respectively elementary and rigid or QH with fibre. It suffices to prove that an endpoint v of e cannot be QH with fibre. Since THϕ contains an edge, it follows that H is not elliptic in Tϕ. Thus, THϕ is the union of the axes of elements of H. Then, e is contained in the axis of an element g of H. Recalling the bipartite structure of Tϕ, let v be the rigid or QH vertex adjacent to e. Then there exists an edge e′≠e adjacent to v and contained in the axis of g. Let Φ∈ϕ be such that Per(Φ)=H and let N∈N∗ be such that ΦN(g)=g. Since ΦN is elliptic in Tϕ by Remark 5.7, it fixes pointwise the axis of g. In particular, it fixes e and e′. By Lemma 5.8, we see that v is rigid and that Gv is fixed elementwise by a power of Φ. □ Corollary 5.10 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). Let [H]∈PerNP(ϕ) and let THϕ be the minimal H-invariant subtree of Tϕ. The tree THϕ does not contain a QH with fibre vertex. Proof Suppose first that THϕ is reduced to a point v. Then H⊆Gv and v is not QH with fibre by Lemma 5.6. Suppose now that THϕ is not reduced to a point. Then any vertex v of THϕ is adjacent to an edge and the result follows from Corollary 5.9. □ Lemma 5.11 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). Let n∈N∗ and let Ψ,Θ∈ϕn be such that FΨ and FΘ are elliptic isometries of Tϕ. Suppose that there exist g,h∈G loxodromic in Tϕ, such that Ψ(g)=g, Θ(h)=h and Ax(g)∩Ax(h) contains an edge e. There exists N∈N∗ such that ΨN=ΘN. Proof First note that FΨ (resp. FΘ) fixes pointwise the axis of g (resp. h). Therefore, both Ψ and Θ preserves the stabilisers of the endpoints of e. By construction of Tϕ, one of the endpoints v of e is either a rigid or a QH with fibre vertex. Moreover, both Ψ and Θ preserve the subgroup associated with an edge adjacent to v distinct from e. Therefore, we can apply Lemma 5.8: there exists N∈N∗ such that both ΨN and ΘN act as the identity on the nonelementary subgroup Gv. By Lemma 5.1, up to taking powers of Ψ and Θ, we have ΨN=ΘN. □ Lemma 5.12 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). The set ⋃n∈NPerNP(ϕn) is finite. Consequently, there exists N∈N∗ such that any subgroup whose conjugacy class is in ⋃n∈N∗PerNP(ϕn) belongs to a subgroup whose conjugacy class is in PerNP(ϕN). Proof Note that, for every N∈N∗ and every Ψ∈ϕ, we have Per(ΨN)=Per(Ψ). Thus, we will generally take a power of the considered automorphisms if needed. Let Tϕ be the above described tree associated with ϕ. Up to taking a power of ϕ, we may suppose that ϕ acts trivially on G\Tϕ, that the homomorphism ⟨ϕ⟩→Out(Ge) is trivial for every e∈ETϕ and that the homomorphism ⟨ϕ⟩→Out(Gv) is trivial for every rigid vertex v. Counting elliptic subgroups: Let [H]∈⋃n∈N∗PerNP(ϕn). Suppose first that every element of H is elliptic in Tϕ. By Lemma 4.3, H is elliptic in Tϕ. Since [H]∈⋃n∈N∗PerNP(ϕn), the group H is not contained in the stabiliser of a vertex which is elementary. By Corollary 5.9, it is also not contained in the stabiliser of a QH with finite fibre vertex. Therefore, the group H is contained in the stabiliser of a vertex v which is rigid. By Lemma 5.6, we have in fact H=Gv. In particular, the stabiliser of any vertex of Tϕ contains at most one conjugacy class of elliptic subgroups in ⋃n∈N∗PerNP(ϕn). Since the action of G on Tϕ has finitely many orbits of vertices, the set ⋃n∈NPerNP(ϕn) contains only finitely many conjugacy classes of elliptic subgroups of Tϕ. Counting subgroups containing loxodromic elements: Suppose now that H contains a loxodromic element h. Let n∈N∗, let Ψ∈ϕn be such that Per(Ψ)=H and let N∈N∗ be such that ΨN(h)=h. By Remark 5.7, the isometry FΨN is elliptic. Let D be a finite fundamental domain for the action of G on Tϕ. Up to taking a conjugate of H, we may suppose that D contains an edge e of the axis of h. Let Θ∈ϕn with [Per(Θ)]∈PerNP(ϕn) and let h′∈G loxodromic be such that ΘN(h′)=h′ and that Ax(h′) contains e. As above the isometry FΘN is elliptic in Tϕ. By Lemma 5.11, there exists m∈N∗ such that Ψm=Θm. Therefore, we see that Per(Ψ)=Per(Θ). Hence the conjugacy class of H in PerNP(ϕn) is entirely determined by the edges of the fundamental domain D contained in translates of axes of elements of H. Since D is finite, and since the natural map PerNP(ϕn)→PerNP(ϕ(n+1)!) is injective, there exist only finitely many [H]∈⋃n∈N∗PerNP(ϕn) containing the conjugacy class of a loxodromic element. (In fact, their number is bounded above by the number of edges in the fundamental domain.) As we have ruled out every case, we see that the set ⋃n∈N∗PerNP(ϕn) is finite. The second assertion follows from the first since the set ⋃n∈N∗PerNP(ϕn!) is a nondecreasing sequence of sets exhausting ⋃n∈N∗PerNP(ϕn). □ Definition 5.13 Let ϕ∈Out(G,P∪H0(t)) and let N∈N∗ be the integer given by Lemma 5.12. If N=1, we say that ϕ is almost rotationless. Note that Lemma 5.12 implies that every element of Out(G,P∪H0(t)) has an almost rotationless power. Lemma 5.12 has also the following corollary regarding isogredience classes of automorphisms. Recall that two automorphisms Φ,Ψ of a group G are in the same isogredience class if there exists g∈G such that Φ=adgΨadg-1. The isogredience class of Φ is contained in its outer class. Lemma 5.12 immediately implies the following. Corollary 5.14 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). There exist only finitely many isogredience classes of automorphisms in ϕ with a nonelementary periodic subgroup. Notice that Lemma 5.12 implies that there exist only finitely many isomorphism classes of nonelementary periodic subgroups for automorphisms contained in a given outer class. We ask for the following generalisation when the peripherals are virtually polycyclic. Question 5.15 Let G be a hyperbolic group relative to virtually polycyclic groups. Do there exist only finitely many isomorphism classes of nonelementary periodic subgroups for automorphisms of G? The case where the peripherals are abelian and G is torsion free follows from the work of Guirardel–Levitt [26, 27]. The Periodic JSJ tree We now need to understand vertex stabilisers of a JSJ tree given by Theorem 4.2 and its acylindricity in order to apply Theorem 2.5. We set H=⋃n∈N∗PerNP(ϕn), and refer to the JSJ tree over elementary subgroups relative to P∪H0∪H as TPer. (Again, while this notation is neither standard nor entirely unambiguous, we use it consistently through our proofs.) We highlight the fact that, if the subgroups in H are finitely generated, then the acylindricity and the understanding of the vertex stabilisers of TPer mostly follow from Theorem 4.2 (10) (using our Lemma 5.12 for the other assumption). The main technical difficulty is thus to show that, when the periodic subgroups are not finitely generated, we still have a complete understanding of the stabilisers of rigid vertices. Remark 5.16 Note that, by Lemma 5.12, there exists an almost rotationless power ϕN of ϕ such that every subgroup of G whose conjugacy class is in H is contained in a subgroup whose conjugacy class is in PerNP(ΦN). Thus, every (A,P∪H0∪H)-tree is an (A,P∪H0∪PerNP(ΦN))-tree, and conversely. Thus the JSJ tree of G relative to P∪H0∪H is also the JSJ tree relative to P∪H0∪PerNP(ΦN) (see [28, Definition 2.12]). Therefore, we only need to work with almost rotationless automorphisms and we will still get results regarding the periodic JSJ tree associated with an arbitrary automorphism. Let ϕN be an almost rotationless power of ϕ. Let PerNP(ϕN)={[H1],…,[Hk]} where, for every i∈{1,…,k}, the group Hi is not elementary and there exist Φi∈ϕN such that Hi=Per(Φi). Note that, for every i∈{1,…,k}, Theorem 4.2 gives that the group Hi fixes a unique rigid vertex vi in TPer since Hi is nonelementary. Recall the construction of Tϕ at the beginning of Sect. 5.1. Note that the trees Tϕ and TPer are compatible. Let Tref be the unique minimal tree which refines Tϕ and TPer. We denote by pϕ:Tref→Tϕ and pPer:Tref→TPer the associated G-equivariant projections. The tree Tref satisfies the following properties: a subgroup H of G stabilises a point in Tref if and only if H stabilises a point in both Tϕ and TPer. Moreover, for every edge e∈ETref, at least one of the images pϕ(e) or pPer(e) is not reduced to a point. Note that, since the actions of G on Tϕ and TPer are acylindrical, by minimality of Tref, the action of G on Tref is also acylindrical. (Any sufficiently long path in Tref will project to a path of length at least 3 in at least one of Tϕ and Tref; edge stabilisers are not changed by the projection map, and so the stabiliser of the path must have been finite to begin with.) Moreover, by uniqueness of Tref, and since the outer automorphism ϕ preserves Tϕ and TPer, we see that ϕ also preserves Tref. For i∈{1,…,k}, let FΦi be the isometry of Tref induced by Φi. As in Remark 5.7, one can show that FΦi is elliptic in Tref (this uses the acylindricity of the action of G on Tref). For every i∈{1,…,k}, let Tiref be the minimal tree of Hi in Tref. It might be that every element of some Hi is elliptic in Tref: in this case Lemma 4.3 implies that the whole subgroup Hi is also elliptic. Then Hi stabilises a vertex in both TPer and Tϕ; since Hi is non-elementary these vertices are unique. But since every edge of Tref survives in the projection to at least one of Tϕ and TPer, there cannot be an edge of Tref stabilised by Hi, and we may take the unique fixed vertex as the minimal invariant tree in this case. Lemma 5.17 Suppose that ϕ is almost rotationless. Let Φ∈NP(ϕ), H=Per(Φ), v a vertex in its minimal invariant tree THref, and FΦ the isometry of Tref induced by Φ. There exists n∈N∗ such that FΦn fixes v. Proof If THref consists of a single vertex v, then this is the unique vertex stabilised by H. Since Φ preserves H, FΦ must also fix v. Otherwise, the vertex v is contained in the axis of some g∈H. In particular, since g is periodic, there exists n∈N∗ with Φn(g)=g, and hence FΦn preserves this axis. Since FΦ is elliptic in Tref, the isometry FΦn fixes elementwise the axis of g. In particular, FΦn fixes the vertex v. □ Proposition 5.18 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let Φ and Ψ be two representatives of an almost rotationless ϕ∈Out(G,P∪H0(t)). Let H=Per(Φ) and K=Per(Ψ) be two non-elementary periodic subgroups of ϕ, perhaps conjugate. Then their minimal invariant trees THref and TKref have non-empty intersection if and only if H=K. Proof If H=K then the minimal invariant trees THref and TKref are equal, so only one direction needs proof. Consider the minimal invariant trees of H and K in Tϕ. If THϕ and TKϕ do not intersect, then neither do THref and TKref, so for the remainder of the proof we assume there is an intersection here. If the intersection contains an edge, Lemma 5.11 implies that H=K, so from now on assume the intersection is a single vertex v=THϕ∩TKϕ. Each of H and K stabilise a unique rigid vertex in TPer, and whenever H≠K we will construct an (A,P∪H0∪PerNP(ϕ))-tree where H and K stabilise different vertices. This prevents ⟨H,K⟩ being contained in a rigid vertex group, by Theorem 4.2 . But then the rigid vertices stabilised by H and K are distinct, and the trees in Tref (containing THref and TKref as subtrees) collapsing to them must be disjoint. We distinguish two cases, according to the nature of v. Note that v is not QH with fibre by Corollary 5.10. Case 1: v is a rigid vertex. By Lemma 5.6 if H or K is elliptic and Corollary 5.9 otherwise, there exists N∈N∗ such that Gv⊆Fix(ΦN) and Gv⊆Fix(ΨN). Hence both ΨN and ΦN fix elementwise the same nonelementary subgroup. By Lemma 5.1, we see that H=K. Case 2: v is elementary. Let Tv be the set of minimal trees TKϕ with [K]∈PerNP(ϕ) which contain v. Since, for every [K]∈PerNP(ϕ), the group K is nonelementary, no tree S∈Tv is reduced to a point. By Lemma 5.11, for all distinct S,S′∈Tv, the intersection S∩S′ is reduced to v. Let E(v) be the set of edges adjacent to v. We have a partition where for every e∈E, the edge e is not contained in any S∈Tv and, for every S∈Tv and every e∈ES, the edge e is contained in S. Let Sv be the tree with one central vertex v0 adjacent to all the other vertices and that the leaves vS are indexed by the trees S∈Tv. We suppose that the stabiliser of v0 is equal to Gv and, for every S∈Tv, that the stabiliser of vS is equal to the setwise stabiliser Stab(ES) of ES. Let T′ be the tree obtained from T by blowing up Sv at v and attaching for every e∈E, the edge e to v0 and for every S∈Tv and every e∈ES, by attaching the edge e to vS. Note that the tree T′ obtained is an (A,P∪H0)-tree. Moreover, for any [K1],[K2]∈PerNP(ϕ) with TK1ϕ≠TK2ϕ∈Tv the minimal trees of TK1′ and TK2′ of K1 and K2 in T′ are disjoint. Let U be the (A,P∪H0∪H)-tree obtained from T′ by collapsing the minimal tree of every H with [H]∈PerNP(ϕ). Then K and H fix distinct points in U. Thus, U is an (A,P∪H0∪H)-tree where H and K fix distinct points. By Theorem 4.2 (7), the groups H and K fix distinct rigid vertices in TPer. Therefore, the minimal trees THϕ and TKϕ of H and K in Tref are disjoint. □ Lemma 5.19 Suppose ϕ is almost rotationless. For every H∈PerNP(ϕ), the stabiliser of the vertex vH of TPer is equal to the global stabiliser GTHref of the minimal tree THref of H in Tref. Proof First note that, since the projection pPer:Tref→TPer is G-equivariant, the tree THref collapses onto vH. Thus, we have GTHref⊆GvH. Conversely, let U be the (A,P∪H0∪H)-tree obtained from Tref by collapsing, for every H∈PerNP(ϕ), the tree THref. For every H∈PerNP(ϕ), let wH be the vertex of U fixed by H. Note that, by Proposition 5.18, for any H,K∈PerNP(ϕ) and every g∈G such that gHg-1≠K, the trees gTHref and TKref are disjoint. Thus, the stabiliser of wH is equal to GTHref. Since the vertex vH of TPer is rigid in every (A,P∪H0∪H)-tree (see Theorem 4.2 (7)), the group GvH is elliptic in U and contains H. As H fixes a unique point in U, which is wH, we see that GvH⊆GTHref. □ Lemma 5.20 Suppose that ϕ is almost rotationless. For every H∈PerNP(ϕ), and the unique vertex vH∈VTPer it stabilises, we have GvH=H. Proof Let H∈PerNP(ϕ), let Φ∈ϕ be the automorphism satisfying Per(Φ)=H, and FΦ the associated isometry of Tref. By Lemma 5.19, it suffices to show that the stabiliser GTHref of THref is equal to H. By Proposition 5.18 and the fact that gTHref=TgHg-1ref for every g∈G, the group GTHref is equal to the normaliser N(H) of H in G. Suppose first that H fixes a (necessarily unique) vertex v∈VTref. Then v has nonelementary stabiliser. By Lemma 5.6, the point pϕ(v) is the unique rigid vertex fixed by H in Tϕ. Thus, N(H) fixes pϕ(v). By Lemma 5.6, we have N(H)⊆Gpϕ(v)⊆H and H=N(H). Suppose now that H does not fix a vertex. By Lemma 4.3, H contains a loxodromic element, and so THref is an infinite subtree of Tref. Let g∈N(H). Let v1 and v2 be vertices in THref, at distance greater than the acylindricity constant K. Observe that the acylindricity assumption implies that there are only finitely many elements g′ with the property that gvi=g′vi for i=1,2 (since they will be contained in the coset g(Gv1∩Gv2), which is finite). Call this set F. By Lemma 5.17, up to taking a power of Φ, the isometry FΦ fixes v1,gv1,v2 and gv2. Thus, for i=1,2 we have that Φ preserves the set of elements sending vi to gvi, and in particular their finite intersection F. As F is finite and contains g, a power of Φ fixes g. This shows that g∈H=Per(Φ) (recalling that the periodic subgroup is invariant under taking powers) and N(H)=H. □ Theorem 5.21 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). Let ϕN be an almost rotationless power of ϕ. For every rigid vertex v∈VTPer, there exists [H]∈PerNP(ϕN) such that Gv=H. Conversely, for every [H]∈PerNP(ϕN), there exists a rigid vertex v∈VTPer with Gv=H. Proof Up to taking a power of ϕ, we may assume that ϕ is almost rotationless (see Remark 5.16). Let v∈VTPer be rigid. We claim that there exists [H]∈PerNP(ϕ) with H≤Gv. Consider the action of Gv in Tref. If Gv fixes a vertex in Tref, then Gv also stabilises a non-elementary vertex w of Tϕ. If w is rigid, then ⟨ϕ⟩0→Out(Gw) has finite image, so Gv is periodic for some ϕk, and so for ϕ (since we assume it is almost rotationless). Otherwise w is a QH with fibre vertex. Note that all elementary subgroups of Gw are virtually cyclic. In particular, the edges adjacent to v∈VTPer are all virtually cyclic, hence finitely generated. Thus, Gv is a finitely generated subgroup of Gw. Since w is a QH with fibre vertex, it is locally quasi-convex. Thus, Gv is hyperbolic. Note that, since Gv is hyperbolic and a subgroup of Stab(w), the restriction P∪H0∪PerNP(ϕ)|Gv of P∪H0∪PerNP(ϕ) to Gv is a finite family of virtually cyclic subgroups of Gv by Lemma 5.2. Thus, we can apply [25, Theorem 3.9] to show that Out(Gv,P∪H0(t)∪PerNP(ϕ)|Gv∪Incv) (which has Out(Gv,P∪(H0∪PerNP(ϕ)|Gv∪Incv)(t)) as a finite index subgroup) is infinite if and only if Gv has an (A,P∪H0∪PerNP(ϕ)|Gv∪Incv)-splitting. Since v is rigid, no such splitting of Gv exists. Therefore, Out(Gv,P∪H0(t)∪PerNP(ϕ)|Gv∪Incv) is finite. In particular, the group Gv is a nonelementary periodic subgroup of some power of ϕ. Now suppose that Gv is not elliptic in Tref. By Lemma 4.3, Gv contains a loxodromic element in Tref and the minimal invariant tree Tvref⊆pPer-1(v) of Gv in Tref is infinite. If this is contained in THref for some H∈PerNP(ϕ), then v is the unique vertex of TPer stabilised by H, and by Lemma 5.20Gv (as the stabiliser of this vertex) is equal to H. So suppose Tvref is not contained in any THref. Since the THref are disjoint by Proposition 5.18, there is some edge of Tvref contained in no THref, and hence collapsing all the THref will give an (A,P∪H0∪PerNP(ϕ))-tree where Gv is not elliptic, contradicting Theorem 4.2 (7). Conversely, let [H]∈PerNP(ϕ). By construction of TPer, the group H fixes a vertex v of TPer. Since H is nonelementary, such a vertex is unique and is rigid by Theorem 4.2 (10). By Lemma 5.20, we have Gv=H. □ Corollary 5.22 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). Let H=⋃n∈N∗PerNP(ϕn). For every geodesic edge path γ of TPer of length 3 and every automorphism Ψ∈ϕ preserving γ, there exist a vertex v of γ and g∈Gv of infinite order fixed by a power of Ψ. For every geodesic edge path γ of TPer of length 3 and every automorphism Ψ∈ϕ preserving γ, there exist a vertex v of γ and g∈Gv of infinite order fixed by a power of Ψ. The group GΦ acts acylindrically on TPer. The group GΦ acts acylindrically on TPer. Proof Let γ be a geodesic edge path of length 3 in TPer preserved by an automorphism Ψ∈ϕ. Suppose that there exists an edge e of γ whose stabiliser is virtually cyclic (this applies in particular when one of the vertices of γ is QH with fibre). Hence Out(Ge) is finite. Thus, Ψ has a power acting as the identity on an infinite cyclic subgroup of Ge. Thus, we may suppose that γ only contains elementary and rigid vertices. Since γ has length 3 and since TPer is bipartite, it contains an interior vertex v which is rigid. Let e1,e2 be the two edges of γ adjacent to v. Then Ge1∩Ge2 is finite by Theorem 4.2 (8) and for every i∈{1,2} the group Gei is its own normaliser in Gv. Let ϕN be an almost rotationless power of ϕ. By Theorem 5.21, there exists [H]∈PerNP(ϕN) such that Gv=H. Let ΦH∈ϕN be such that Per(ΦH)=H. Since Ψ preserves H, and since H is its own normaliser by Lemma 5.20, there exists h∈H such that ΨN=adhΦH. As ΨN preserves both Ge1 and Ge2, we see that h∈N(Ge1)∩N(Ge2), which is finite as explained above. Up to taking a power of ΦH and ΨN, we may assume that both ΦH and ΨN act trivially on N(Ge1)∩N(Ge2). Taking further powers of ΨN and ΦH shows that there exists M∈N such that ΨM=ΦHM. Let g∈H be of infinite order, which exists since H is nonelementary. Then Ψ has a power which fixes g. This proves Assertion (1). We now prove that the action of GΦ on TPer is acylindrical. Note that, if a finite index subgroup of GΦ acts acylindrically on TPer, so does GΦ. Thus, we may assume that ϕ is almost rotationless. By Theorem 4.2 (2), the action of G on TPer is acylindrical. Thus, by Lemma 4.5 (2) (which we can apply by Assertion (1)), it suffices to prove that, for every n∈N∗, every Ψ∈ϕn and every g∈Fix(Ψ), the element g is elliptic in T. Let g be as above. If g is peripheral, then g fixes a point by construction of TPer. Suppose now that g is nonperipheral. Suppose towards a contradiction that g is loxodromic in TPer. Since Ψ(g)=g, the characteristic set of the isometry FΨ contains the axis of g. By Lemma 4.4, up to taking a power of ϕ and changing the representative Ψ, the isometry FΨ fixes pointwise the axis of g. By Lemma 4.5 (1), up to taking a power of Ψ, the automorphism Ψ fixes elementwise a nonabelian free group of loxodromic elements. In particular, since every peripheral element fixes a point in TPer, we see that Ψ fixes a nonabelian free group of nonperipheral elements, hence a nonelementary subgroup. Since ϕ is almost rotationless, for every n≥1, we have PerNP(ϕ)=PerNP(ϕn), so that [Per(Ψ)]∈PerNP(ϕ). Thus, there exists [H]∈PerNP(ϕ) such that g∈Fix(Ψ)⊆H. In that case, the element g is elliptic in TPer by construction of TPer, a contradiction. Therefore, the element g is elliptic in TPer and we can apply Lemma 4.5 (2) to prove that the action of GΦ on TPer is acylindrical. □ Let G be a finitely generated group and let Φ∈Aut(G). Suppose that Φ has a power which preserves the conjugacy class of a malnormal subgroup F of G. We then denote by FΦ the group F⋊adg∘ΦnFZ, where nF is the minimal positive integer such that ΦnF preserves the conjugacy class of F and g∈G is such that adg∘ΦnF(F)=F. Since F is malnormal, the group FΦ does not depend on g. Note that the group FΦ only depends on the outer class of Φ. Corollary 5.23 Let G be a hyperbolic group relative to P. Let H0 be a finite set of conjugacy classes of finitely generated subgroups of G such that G is one-ended relative to P∪H0. Let ϕ=[Φ]∈Out(G,P∪H0(t)). If for every [P]∈P the group PΦ is in FJCX, then GΦ is in FJCX. Proof Consider the action of GΦ on TPer. By Corollary 5.22, the action of GΦ on TPer is acylindrical. Up to taking a power of Φ (which does not change the conclusion by Theorem 2.1), we may assume that ϕ=[Φ] is almost rotationless. By Theorem 2.5, it suffices to show that, for every v∈VTPer, the stabiliser (GΦ)v of v in GΦ belongs to FJCX. Note that, for every v∈VTPer, the group (GΦ)v can be written as a semi direct product Gv⋊Z, where Gv is the stabiliser of v in G. Suppose first that Gv is elementary. If Gv⊆P for some [P]∈P, then (GΦ)v is a subgroup of PΦ. In particular, it belongs to FJCX by Theorem 2.1 (1). If Gv is infinite virtually cyclic, then (GΦ)v belongs to FJCX. Suppose now that Gv is QH with fibre. Then Gv fits in a short exact sequence where Σv is a hyperbolic 2-orbifold and F is finite. Moreover, up to taking a power of Φ (which is possible by Theorem 2.1 (2)), since Out(Gv) has a finite index subgroup acting as the identity on F, there exists Ψ∈ϕ preserving Gv and fixing F elementwise. Thus, we have a short exact sequence The groups F and π1(Σv)⋊Z belong to FJCX. Indeed, the Farrell–Jones conjecture holds for all fundamental groups of (not necessarily compact) connected manifolds (possibly with boundary) of dimension at most 3, by [41, Theorem 16.1(i)(e)]. As explained in loc. cit. the result is deduced from Corollary 1.3 and Remark 1.4 of [5] where [46, 47] is used. Moreover, for every virtually cyclic group Q⊆π1(Σv)⋊Z, the preimage of Q in (GΦ)v is virtually cyclic, and hence belongs to FJCX. Thus, the group (GΦ)v belongs to FJCX. Suppose that Gv is rigid. By Theorem 5.21, since ϕ is almost rotationless, there exists [H]∈PerNP(ϕ) such that Gv=H. Suppose that Φ∈ϕ is such that Per(Φ)=H. Then (GΦ)v is isomorphic to Per(Φ)⋊ΦZ. By Lemma 2.6, we have (GΦ)v∈FJCX. As we have ruled out every case, for every v∈VTPer, the group (GΦ)v belongs to FJCX. By Theorem 2.5, the group GΦ belongs to FJCX. □ An aside on slender peripherals We also isolate here an interesting consequence of Theorem 5.21 for automorphisms of groups hyperbolic relative to slender groups. Recall that a group is slender if all its subgroups are finitely generated. Theorem 5.24 Let G be a hyperbolic group relative to a collection P of slender groups and let Φ∈Aut(G). There exists N∈N∗ such that Per(Φ)=Fix(ΦN) and Per(Φ) is finitely generated. Proof See also the proof of [25, Theorem 8.2]. We claim that it suffices to prove that Per(Φ) is finitely generated. Indeed, suppose that Per(Φ) is generated by a1,…,an. For every i∈{1,…,n}, let ki be such that Φki(ai)=ai. Let N=k1…kn. Then Per(Φ)=Fix(ΦN). So we prove that Per(Φ) is finitely generated. Note that slender groups are NRH groups, so that Aut(G)=Aut(G,P). Note also that, since P is a set of conjugacy classes of slender groups, every elementary subgroup of G is finitely generated. Thus, we may suppose that Per(Φ) is not elementary. Let ϕ=[Φ]∈Out(G) and let H=PerNP(ϕ). Let (Hi)i≥1 be an increasing sequence of finite set of conjugacy classes of finitely generated subgroups of G such that H=⋃i≥1Hi. By [45], for every i≥1, the group G is finitely presented relative to P∪Hi. Thus, the Stallings–Dunwoody JSJ deformation space Di of G over finite groups relative to P∪Hi exists (see [23]). We claim that there exists i0 such that, for every i≥i0, we have Di=Di0. The proof is identical to [28, Lemma 8.2] replacing Linnell’s accessibility Theorem by [23, Proposition 4.2]. We sketch the argument. Let i≥1. By [28, Lemma 2.15], since edge groups are all finite and hence all universally elliptic, for every Ti∈Di, there exists a sequence of refinements of trees Ti1→…→Tii-1→Tii=Ti such that, for every j≤i, we have Tij∈Dj. Let j≤i. Notice that, since edge groups are finite, Tij-1 and Tij are not in the same deformation space if and only if they do not have the same infinite elliptic subgroups if and only if the number of orbits of vertices with infinite stabiliser is greater in Tij-1 than in Tij. By [23, Proposition 4.2], the number of orbits of vertices of Ti1 with infinite stabiliser is bounded by a constant J depending only on D1. Thus, the number of j≤i such that Tij-1 and Tij are not in the same deformation space is bounded by J. This forces the spaces Di to eventually stabilise, hence the claim. Let Di0 be the stabilised deformation space. As P∪H=P∪⋃i≥1Hi, this shows that any tree in Di0 is a (P∪H)-tree with finite edge groups. Suppose first that G is one-ended relative to P∪H. Then Di0 consists of the trivial tree and G is also one-ended relative to P∪Hi0. Moreover, since Hi0 is a finite set and all groups whose conjugacy class is in Hi0 are finitely generated, up to taking a power of ϕ, we have ϕ∈Out(G,P∪Hi0(t)). Let TPer be the JSJ tree associated with P∪Hi0∪H. Since edge stabilisers of TPer are elementary, they are all finitely generated. Thus, every vertex stabiliser of TPer is also finitely generated. By Theorem 5.21, the group Per(Φ) is equal to the stabiliser of a vertex of TPer, hence is finitely generated. Suppose now that G is not one-ended relative to P∪H and consider a tree S in Di0. Recall that edge stabilisers in S are all finite. Since Per(Φ) is infinite, it fixes a unique vertex v. Since the deformation space of S is invariant by Φ, the group Φ(Gv) also fixes a unique vertex w in S. As Per(Φ)⊆Φ(Gv) and as Per(Φ) only fixes v, we see that v=w. This shows that Φ(Gv)=Gv. Note that, by minimality of S, the group Gv is hyperbolic relative to the restriction Pv of P in Gv and one-ended relative to Pv∪(Hi0)v∪Hv. Since P is a set of conjugacy classes of slender groups, so is Pv. Thus, the conclusion follows from the one-ended case applied to the restriction Φ|Gv. □ We remark that Minasyan–Osin [44, Corollary 1.3] also proved that the fixed subgroup of the automorphism of any hyperbolic group relative to slender groups is finitely generated. A combination theorem for the Farrell–Jones conjecture Let G be a finitely generated group and let Φ∈Aut(G). If F is a malnormal subgroup of G whose conjugacy class is Φ-periodic, recall the definition of FΦ from just above Corollary 5.23. In this section, we prove the following combination theorem. Theorem D Let G=G1∗…∗Gk∗FN be a free product of finitely generated groups, let F′={[G1],…,[Gk]} and let Φ∈Aut(G,F′). If for each i∈{1,…,k}, the group (Gi)Φ is in FJCX, then G⋊ΦZ is in FJCX. The proof of Theorem D is by induction on k+N. Let F′≤F be a maximal proper Φ-periodic free factor system. Up to taking a power of Φ (which does not change the conclusion of Theorem D by Theorem 2.1 (2)), we may suppose that Φ∈Aut(G,F). We will distinguish between two cases, according to whether F is sporadic or not. The nonsporadic case Lemma 6.1 Let G=G1∗…∗Gk∗FN be a free product of groups, let F={[G1],…,[Gk]} and let Φ∈Aut(G,F) be fully irreducible. If for each i∈{1,…,k}, the group (Gi)Φ is in FJCX, then G⋊ΦZ is in FJCX. Proof Let S be the Grushko (G,F)-tree given by Lemma 3.3. Let PS(Φ) be the ‖.‖S-maximal polynomial subgroups of Φ. By Theorem 3.2, up to taking a power of Φ (which does not change the conclusion of Lemma 6.1 by Theorem 2.1 (2)) the group G⋊ΦZ is hyperbolic relative to the suspension of PS(Φ). By Proposition 3.4, for every [P]∈PS(Φ), either [P]∈F or P is infinite cyclic. In either case PΦ is contained in FJCX. By Theorem 2.4, the group G⋊ΦZ is contained in FJCX. □ The sporadic case This section follows [6, Proof of Proposition 4.1]. Let (G,F) be a sporadic free product and let Φ∈Aut(G,F). Since (G,F) is sporadic, the automorphism Φ induces a G-equivariant homeomorphism of the Bass–Serre tree TF associated with F. This induces an action of G⋊ΦZ on TF. However, this action is not necessarily acylindrical. In order to apply Theorem 2.5, we will consider the action of G⋊ΦZ on the tree of cylinders of TF associated with an admissible relation that we now describe. Let t be a generator of the Z-factor. Up to taking a finite index subgroup of G⋊ΦZ (which is possible by Theorem 2.1 (2)), we may suppose that t fixes an edge e. In that case, edge stabilisers of the action of G⋊ΦZ are all infinite cyclic, generated by conjugates of t. Therefore, the commensurability relation is an admissible equivalence relation, and we define the tree of cylinders Tc of TF relative to this admissible relation. Lemma 6.2 Let Y be the cylinder of TF containing e. The stabiliser of Y in G⋊ΦZ is isomorphic to ⟨Fix(Φn)⟩n≥1⋊ΦZ. Proof Note that any element h∈G⋊ΦZ can be written uniquely as w-1tj, where w∈G and j∈Z. Let w-1tj∈Stab(Y) with w∈G and j∈Z and let e′=w-1tje. Then we have Ge′=⟨w-1tw⟩. Moreover, since e,e′∈EY, by definition of the commensurability relation, there exist n,m∈N such that In particular, we see that n=m and Φm(w)=w. □ Lemma 6.3 The action of G⋊ΦZ on Tc is acylindrical. Proof The proof is identical to [6, Lemma 4.6]. Let v,v′∈VTc with dTc(v,v′)≥6. We may suppose that v and v′ correspond to vertices w and w′ in TF up to considering adjacent vertices in the path between them. Even after this operation, we have dTc(v,v′)≥4. Let g∈Gv∩Gv′. Then g fixes the path in TF between w and w′. Since dTc(v,v′)≥4, the path in TF between w and w′ must contain two edges in distinct cylinders. Hence g fixes two edges in distinct cylinders. Since edge stabilisers in TF are infinite cyclic and since we are considering the commensurability relation, two edges in TF are in the same cylinder if and only if the intersection of their stabilisers is nontrivial. In particular, this shows that g is trivial and that the action of G⋊ΦZ on Tc is acylindrical. □ Lemma 6.4 Let (G,F) be a sporadic free product of groups and let Φ∈Aut(G,F). If for each [A]∈F, the group AΦ is in FJCX, then G⋊ΦZ is in FJCX. Proof Let TF be the Bass–Serre tree associated with F and let Tc be its tree of cylinders relative to the commensurability relation. We want to apply Theorem 2.5 to the action of G⋊ΦZ on Tc. This action is acylindrical by Lemma 6.3. Thus, it suffices to prove that every vertex stabiliser belongs to FJCX. Recall that we have a bipartition of VTc=V0Tc∐V1Vc, where vertices in V0Tc correspond to vertices of TF and vertices in V1Tc correspond to cylinders of Tc. If v∈V0Tc, then its stabiliser is isomorphic to AΦ. In that case, the stabiliser of v belongs to FJCX by hypothesis. Suppose now that v∈V1Tc. By Lemma 6.2, the stabiliser of v is isomorphic to Per(Φ)⋊ΦZ. By Lemma 2.6, the group Per(Φ)⋊ΦZ belongs to FJCX. Thus, every vertex stabiliser of the action of G⋊ΦZ on Tc belongs to FJCX. By Theorem 2.5, the group G⋊ΦZ belongs to FJCX. □ End of the proof of Theorem D Let G=G1∗…∗Gk∗FN be a free product of groups, let F′={[G1],…,[Gk]} and let Φ∈Aut(G,F′). We prove by induction on k+N that G∈FJCX. Suppose first that k+N=1. If N=0, then G=G1 and G⋊ΦZ∈FJCX by hypothesis. If k=0, then G=Z, G⋊ΦZ is solvable and the result follows from [51]. This proves the base case. Suppose now that k+N≥2 and let F be a maximal Φ-periodic free factor system. We may assume, up to taking a power of Φ, that F is Φ-invariant, so that we can view Φ as an element of Aut(G,F). This is possible by Theorem 2.1 (2) as, for every n∈N, the group G⋊ΦnZ is a finite index subgroup of G⋊ΦZ. By induction hypothesis, for every [A]∈F, the group AΦ belongs to FJCX. Combining the nonsporadic case (Lemma 6.1) and the sporadic case (Lemma 6.4), we conclude that G⋊ΦZ belongs to FJCX. This concludes the proof. □ Proving Theorem A We first record a corollary of Theorem D. Corollary 6.5 Let (G,P) be a virtually torsion-free relatively hyperbolic group with P finite and let Φ∈Aut(G,P). If for every [P]∈P we have PΦ∈FJCX, then GΦ∈FJCX. Proof By Theorem 2.1 we may assume G is torsion-free. Let F be the minimal free factor system of G such that, for every [P]∈P, there exists [A]∈F with P⊆A. Since Φ∈Aut(G,P), by minimality of F, we have Φ∈Aut(G,F). Let [A]∈F. We denote by PA the peripheral structure of A induced by P. Since G is torsion-free, the group A is one-ended hyperbolic relative to PA. By Corollary 5.23 the group AΦ belongs to FJCX. By Theorem D, the group GΦ belongs to FJCX. □ Finally, combining Corollaries 5.23 and 6.5 proves our first theorem from the introduction. Theorem A Let (G,P) be a virtually torsion-free or one-ended relatively hyperbolic group with P finite and let Φ∈Aut(G,P). If for every [P]∈P we have PΦ∈FJCX, then GΦ∈FJCX. We now discuss the (minor) changes to the proof used to prove the following theorem. Theorem E Suppose (G,P) is one-ended or virtually torsion free, and hyperbolic relative to finitely many conjugacy classes of virtually polycyclic subgroups. Then for every automorphism Φ of G, Γ:=G⋊ΦZ is in AC(VNil). Proof Knopf’s work on acylindrical actions of trees applies equally well in the setting of AC(VNil) (see [36, Corollary 4.2] and [6, Theorem 2.4], note that Knopf does not state this but it is implicit in her work). We use the same trees as every step of the proof of Theorem A. Whenever a vertex group is identified as Per(Φ)⋊ΦZ, use Theorem 5.24 to further identify it as some Fix(Φk)⋊ΦZ. As this has a finite index subgroup isomorphic to Fix(Φk)×Z, and AC(VNil) passes to both subgroups and direct products, this vertex group lies in AC(VNil). □ Remark 6.6 Note that in both Theorems A and E we may replace the hypothesis ‘virtually torsion free’ with the more general assumption that G has a finite index subgroup whose Dunwoody decomposition has no non-trivial edge groups. We do not pursue this here. Proofs of the applications Our first application is to extensions of groups with relatively hyperbolic kernel. Corollary B Let (N,P) be a virtually torsion-free or one-ended relatively hyperbolic group such that P consists of finitely many conjugacy classes of groups which are NRH and whose suspensions P⋊ΨZ are in FJCX for all automorphisms Ψ of P. Let 1→N→Γ→Q→1 be a short exact sequence. If Q is in FJCX, then Γ is in FJCX. Proof Since for all [P]∈P the group P is NRH we have that Aut(G,P) is a finite index subgroup of Aut(G). Let Φ∈Aut(G). The suspension GΦ has a finite index subgroup GΦn such that Φn∈Aut(G,P). Now, Theorem A implies that GΦn is in FJCX. It follows from Theorem 2.1 that GΦ is in FJCX. The result now follows from Theorem 2.2. □ Our other application is that Aut(G) is in FJCX for G a one-ended group hyperbolic relative to finitely many conjugacy classes of polycyclic subgroups. Before we prove this, we collect some results. Theorem 7.1 [25, Theorem 4.3] Let (G,P) be a relatively hyperbolic group. Suppose for every [P]∈P, the group P is finitely generated. If G is one-ended relative to P, then there is a short exact sequence where T is a quotient of a finite direct product where each factor is virtually cyclic or contained in some P for P∈P; T is a quotient of a finite direct product where each factor is virtually cyclic or contained in some P for P∈P; MCG0(Si) maps onto a finite index subgroup of the extended mapping class group MCG∗(Si) with finite kernel (they are virtually isomorphic). MCG0(Si) maps onto a finite index subgroup of the extended mapping class group MCG∗(Si) with finite kernel (they are virtually isomorphic). Proposition 7.2 If G is a virtually polycyclic group, then G, Out(G) and Aut(G) are in FJCX. Proof By [7, Theorem 1.1] we see that Out(G) is an arithmetic group. Hence, Out(G) is in FJCX by [5]. Technically they only prove the conjecture for K- and L-theory but it follows for A-theory by [48, 36, Proof of Theorem 1.8(a)], and [18, Theorem 6.19]. Alternatively, one may use [18] and [38]. Since G is virtually soluble it is in FJCX by [51] (for K- and L-theory), [38] (for A-theory), and Theorem 2.1. Now, a virtually polycyclic group is poly-{virtually cyclic}, so any extension G⋊Z is also virtually polycyclic. Thus, G⋊Z is in FJCX. Further, note that G/Z(G) is virtually polycyclic and so in FJCX. Combining these observations with Theorem 2.2 shows that Aut(G) is in FJCX. □ Proposition 7.3 The mapping class group of a hyperbolic 2-orbifold is in FJCX. Proof This follows from the result for (orientable) surfaces [3] and assembling results in the literature. Note that Bartels–Bestvina only prove FJC for K- and L-theory but the result for A-theory follows (as usual) from [36, Proof of Theorem 1.8(a)], and [18, Theorem 6.19]. Let S be a hyperbolic orbifold, and let Σ be an orientable surface covering S with finite degree so that π1(Σ) is characteristic in π1(S). (This can be achieved by taking any covering surface, passing to its orientation cover if necessary, and then taking the characteristic core of the corresponding subgroup and realising the covering surface.) By [35] there is an injective map from Autgeom(π1(S)) to Autgeom(π1(Σ)), where these geometric automorphism groups are exactly the lifts of the mapping class groups. Restricting to the image, Inn(π1(S)) will be normal, and by the third isomorphism theorem the quotient is isomorphic to C/(Inn(π1(S))/Inn(π1(Σ))), where C is a subgroup of Autgeom(π1(Σ))/Inn(π1(Σ)), the mapping class group of Σ. The quotient of inner automorphism groups is finite (in fact isomorphic to the deck transformations π1(S)/π1(Σ)), so we have realised MCG(S) as an extension Since mapping class groups of surfaces are residually finite by [30] (and residual finiteness passes to subgroups), we may apply Lemma 2.3 to obtain the conclusion. □ Proposition 7.4 If G is a one-ended group hyperbolic relative to finitely many conjugacy classes of virtually polycyclic groups, then Out(G) is in FJCX. Proof By Theorem 7.1 there is a finite index subgroup Out0(G) fitting into a short exact sequence We want to apply Theorem 2.2 to this short exact sequence. First we check the kernel T: this is a quotient of a direct product of virtually polycyclic groups, and hence is itself virtually polycyclic, and so in FJCX by Proposition 7.2. Now consider the image. The subgroups Out(Pj,IncPj(t)) are subgroups of Out(Pj) for a virtually polycyclic Pj, and hence are in FJCX by Proposition 7.2. Each MCG(Si) maps with finite kernel onto the mapping class group of a hyperbolic 2-orbifold. By Lemma 2.3 it is enough to consider these mapping class groups. These are in FJCX by Proposition 7.3. Then the product is in FJCX by Theorem 2.1. To apply Theorem 2.2 it remains to check the preimages of elements. These are of the form T⋊Z, which are in FJCX by Proposition 7.2 since they are again virtually polycyclic. □ Theorem C If G is a one-ended group hyperbolic relative to finitely many conjugacy classes of virtually polycyclic groups, then Aut(G) and Out(G) are in FJCX. Proof Since P contains only finitely many conjugacy classes of virtually polycyclic groups, by [17, Corollary 1.14], we may modify P such that that for every [P]∈P, the group P is NRH. Observe that since G is hyperbolic relative to finitely many conjugacy classes of virtually polycyclic subgroups its centre is finite. Hence, Inn(G) is quasi-isometric to G and again hyperbolic relative to finitely many conjugacy classes of NRH virtually polycyclic subgroups by [4]. By Proposition 7.4, the group Out(G) is in FJCX. The result now follows from applying Corollary B to the short exact sequence □ Theorem C allows us to prove that the outer automorphism groups of some small complexity relatively hyperbolic groups also belong to FJCX. Corollary 7.5 Let G=A∗CB, where A and B are one-ended hyperbolic groups relative to finitely many conjugacy classes of virtually polycyclic groups and C is a finite group. The groups Out(G) and Aut(G) are in FJCX. Proof We prove the result for Out(G), the proof for Aut(G) being identical to the proof of Theorem C (this uses Proposition 8.1 when C is nontrivial and Theorem A otherwise). Let Out0(G) be the index (at most) 2 subgroup of Out(G) preserving the conjugacy classes of A and B. By [21], every element ϕ∈Out0(G) has a representative Φ∈ϕ such that Φ(A)=A and Φ(B)=B. Moreover, the map sending ϕ to Φ defines an isomorphism between Out0(G) and Aut(A,C)×Aut(B,C). By Theorem C, the groups Aut(A,C) and Aut(B,C) belong to FJCX. By Theorem 2.1, the groups Out0(G) and Out(G) belong to FJCX. □ Corollary 7.6 Let G=A∗C, where A is a one-ended hyperbolic group relative to finitely many conjugacy classes of virtually polycyclic groups and C is a finite group. The groups Out(G) and Aut(G) are in FJCX. Proof As above we only prove the result for Out(G). Let t be a stable letter for the HNN extension A∗C. By [39], the group Out(G) has an index 2 subgroup Out0(G) such that any element ϕ∈Out0(G) has a representative Φ∈ϕ such that Φ(A)=A and Φ(t)=ta for some a∈A. Moreover, the map sending ϕ to Φ induces an isomorphism between Out0(G) and A⋊Aut(A,C). Thus, Out0(G) fits in a short exact sequence The group Aut(A,C) belongs to FJCX by Theorem C. Moreover, for every infinite cyclic subgroup Q∈Aut(A,C), the preimage of Q in Out0(G) belongs to FJCX by Theorem A. Thus, by Theorem 2.2, the groups Out0(G) and Out(G) belong to FJCX. □ One more infinitely ended case is known, since Out(F2)≅GL(2,Z) is an arithmetic group, and the same extension arguments as above will give the result for Aut(F2). Our techniques do not seem to extend to Out(Fn), which will be necessary to make any further progress. Torsion in the infinitely ended case The aim in this section is to prove as much of Sect. 6 as possible without the assumption that G is virtually torsion free. To this end we prove a generalisation of the sporadic case (of Sect. 6.2) to graphs of groups with one finitely stabilised edge. Proposition 8.1 Suppose G≅A∗CB or G≅A∗C with C a finite group. Further suppose that Φ∈Aut(G) preserves the conjugacy classes of A and B, and that the restrictions of Φ to A is such that A⋊Φ|AZ belongs to FJCX (and similarly for B, if applicable). Then G⋊ΦZ belongs to FJCX. The proof being largely an elaboration of the arguments in Sect. 6.2, here we indicate the necessary changes and references. Proof First, we have to argue that Φ preserves the action on the Bass–Serre tree T for this splitting. This follows from [39], or for the two vertex case already from [21], which give that these one edge splittings are rigid: the unique reduced tree in their deformation spaces, together with the hypothesis that Φ preserves the vertex groups. So we may consider the action of G⋊ΦZ on T. Use t to denote the generator of the Z factor. After possibly passing to a finite index subgroup (by taking the square of Φ, if necessary) we may suppose the quotient graphs are the same for both actions, and that t stabilises an edge e. Edge stabilisers are virtually cyclic, and admit a map to Z with a conjugate of t is mapped to the generator. For Ge, we may take this preimage to be t. Note that by work of Wall [49, Lemma 4.1] virtually cyclic groups act on the line and have a unique maximal finite normal subgroup which is the kernel of this action. Since Ge surjects onto Z, in fact this is the unique maximal finite subgroup. Following the proof in the free splitting case, we need to take a tree of cylinders to ensure that we have an acylindrical action. Just as in that case, the commensurablilty relation is admissible, and we take the tree of cylinders Tc relative to this relation. We need to adapt the proofs of Lemmas 6.2 and 6.3 to the new situation. In the first case, we assume that e and w-1tje=w-1e are edges in the same cylinder. Then ⟨t⟩ and ⟨w-1tw⟩ are finite index subgroups of the respective stabilisers, and again the commensurability relation implies that there are powers n,m∈N so that As before, we see that n=m and w is periodic. To adapt the proof of Lemma 6.3, make the adjustments of the first paragraph of that proof and then consider the whole intersection Gv∩Gv′. This group fixes edges in two distinct cylinders, so is contained inside some Ge∩Ge′, where this intersection is between two virtually cyclic subgroups that are not commensurable. In particular, this means the intersection is finite; in fact its cardinality is bounded by the size of the unique maximal finite subgroup (in either—they are conjugate). This means the action of G⋊ΦZ on Tc is acylindrical. To finish the proof, we recall the bipartite nature of Tc, and observe that vertex stabilisers are either isomorphic to A⋊Φ|AZ (or the same for B—the original vertex stabilisers), or a cylinder stabiliser Per(Φ)⋊ΦZ. The first kind is in FJCX by hypothesis; the second by Lemma 2.6. □ With this in hand, one can begin to try and run the induction argument of Sect. 6 on a Stallings–Dunwoody decomposition of a more general infinitely ended group. However, there seems as yet to be no analogy for the relative hyperbolicity argument used in the non-sporadic case, and so the induction will not be able to proceed if at some stage we encounter a maximal periodic “Stallings–Dunwoody type splitting” that has more than one edge, and at least one edge with non-trivial stabiliser. A proof of the following conjecture, the analogy of Theorem 3.2 for general infinite ended groups, should complete the proof of Theorem A with no assumption on torsion. Conjecture 8.2 Suppose G is the fundamental group of a non-sporadic graph of groups with finite edge stabilisers, and Φ∈Aut(G) is fully irreducible relative to this splitting. Then G⋊ΦNZ is hyperbolic relative to the suspensions of polynomially growing subgroups of Φ. It may be necessary to assume accessibility in the previous conjecture but for now we do not. As in the free product case, the correct notion of growth should be with respect to the translation length function for the action of G on the Bass–Serre tree. The correct definition of fully irreducible appears to be that in any splitting (strictly) dominated by ours every power of Φ does not preserve the set of elliptic subgroups. Acknowledgements This work has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 850930). The second author was supported by the LABEX MILYON of Université de Lyon. The authors would like to thank Damien Gaboriau, Dominik Kirstein, Gilbert Levitt, and Ric Wade for helpful conversations. The authors would also like to thank the anonymous referee for a number of helpful comments and clarifications and for suggesting an elegant simplification of some of our arguments. Funding Open Access funding enabled and organized by Projekt DEAL. Data availability Not applicable. Declarations Conflict of interest The authors declare they have no financial interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement Not applicable.