We investigate the strong stable sets of the geodesic flow on hyperbolic surfaces and their relation to the injectivity radius. We prove that the strong stable set of a vector does not coincide with its stable horocycle if and only if the injectivity radius along the associated geodesic ray is not bounded away from zero and the ray does not diverge through a cusp. In that case, the strong stable set is an uncountable union of stable horocycles. In previous works, we considered the particular cases where the ray meets arbitrarily short closed geodesics or closed horocycles. Here we complete the study by treating the third possible case, where the ray meets closed hypercycles of arbitrarily small length.
Introduction
We consider the geodesic flow gt on the unit tangent bundle T1S of a hyperbolic surface S. We endow T1S with the Sasaki distance dSa induced by the hyperbolic metric of S. For u∈T1S, the stable horocycle Huss is defined as the projection of the stable horocycle of any lift of u to T1H, whereas the strong stable set Wuss is the set
Wuss={v∈T1S:t→+∞limdSa(gtu,gtv)=0}.
Following Bellis [6], we define the asymptotic fineness of the geodesic ray u(R+) as
t≥0liminfInj(u(t)),
where Inj(x) denotes the injectivity radius of S at x∈S. Our main result establishes a link between the asymptotic fineness and strong stable sets.
Theorem 1. Let S be a hyperbolic surface and u∈T1S. Then Huss⊊Wuss if and only if
t≥0liminfInj(u(t))=0
and the geodesic ray u(R+) does not diverge through a cusp. In that case, the set Wuss is an uncountable union of stable horocycles Hviss.
There are several ways for the asymptotic fineness of u(R+) to vanish. Bellis [5] and the authors together with S. Herrero [BCDHV26] study the case where u(R+) meets a sequence of arbitrarily short closed geodesics. In [2], the authors study the case where u(R+) meets a sequence of arbitrarily short closed horocycles. The contribution of this article is to study a third possible case, which is not always covered by the previous works: the ray u(R+) meets closed hypercycles of arbitrarily small length.
Theorem 2. Let S be a hyperbolic surface and u∈T1S. If the geodesic ray u(R+) meets an infinite sequence of closed hypercycles whose lengths tend to zero, then
Huss⊊Wuss,
the set Wuss is an uncountable union of stable horocycles Hviss.
These three cases are exhaustive: if the asymptotic fineness of u(R+) vanishes and u(R+) does not diverge through a cusp, then at least one of them occurs (Proposition 3). However, they are not mutually exclusive.
The proof of Theorem 2 follows the strategy of [BCDHV26] and [2]. We adapt to closed hypercycles the notion of winding used in the previous works. One novelty is the use of the hyperbolic band model, in which the hypercycles around a given geodesic become horizontal lines, to obtain the necessary estimates.
The paper is organized as follows. In § 2 we recall the facts of hyperbolic geometry that will be used. In § 3 we prove the trichotomy (Proposition 3), we deduce Theorem 1 from Theorem 2 and the previous works, and we give an example of the third case. In § 4 we introduce the winding around a closed hypercycle and we prove the Key Proposition (Proposition 5), which controls the distance between a geodesic ray and its winding. § 5 is devoted to the proof of Theorem 2, where we construct uncountably many vectors in Wssu∖Hssu.
Preliminaries
Throughout the paper we use the upper half-plane model H={z=x+iy∈C:y>0} equipped with the metric
ds2=y2dx2+dy2.
The Riemannian distance induced by this metric is denoted by d. We denote the unit tangent bundle of H by T1H, and the basepoint projection by π:T1H→H.
Horocycles and boundary at infinity
The boundary at infinity of H is ∂H=R∪{∞}. Every geodesic ray in H converges to a unique point of ∂H, and two rays define the same point at infinity if and only if they remain at bounded distance from each other.
Let ξ∈∂H, and let (c(t))t≥0 be a geodesic ray converging to ξ. The Busemann function based at ξ is defined by
Bξ(x,y)=t→+∞lim(d(x,c(t))−d(y,c(t))),x,y∈H.
This limit exists and does not depend on the choice of the geodesic ray c asymptotic to ξ. For fixed x∈H, the level sets of the map y↦Bξ(y,x) are called horocycles centered at ξ. Geometrically, horocycles are circles tangent to the real line or horizontal lines.
For u∈T1H, we denote by u(+∞)∈∂H its forward endpoint. The stable horocycle through u is the subset
Hssu={v∈T1H:v(+∞)=u(+∞) and Bu(+∞)(π(v),π(u))=0}.
We will also denote
Hwsu={v∈T1H:v(+∞)=u(+∞)}.
The following lemma can be proved by elementary computations using the formula for the hyperbolic distance [4 Theorem 7.2.1].
Lemma 1. Let u,v∈T1H such that v∈Hwsu. Then d(v(t),u(t)) is non-increasing in t∈R.
Distance on the unit tangent bundle
We consider the distance d1 on the unit tangent bundle of T1H defined as follows [1]:
d1(v,w):=d(v(0),w(0))+d(v(1),w(1)),v,w∈T1H.
Proposition 1. The distance d1 is equivalent to the Sasaki distance on the unit tangent bundle T1H.
Fuchsian groups
Throughout the paper, Γ denotes a torsion-free Fuchsian group acting properly discontinuously by isometries on H. We consider the quotient surface S=Γ\H and its unit tangent bundle T1S=Γ\T1H. We use a tilde to denote lifts to the universal cover: given u∈T1S, we write u∈T1H for a unit tangent vector projecting onto u under the canonical projection T1H→T1S.
The sets Hssu and Hwsu are Γ-equivariant and therefore descend to well-defined subsets in T1S, which we still denote by Hssu and Hwsu.
On the unit tangent bundle T1S of a hyperbolic surface S=Γ\H, the distance d1 descends as follows:
d1(u,v)=γ∈Γinf{d1(u,γv)}.
where u and v are lifts of u and v to T1H. This distance is again equivalent to the Sasaki metric on T1S.
The strong-stable set of a vector u∈T1S is defined as
Wssu={v∈T1S:t→+∞limd1(gtu,gtv)=0}.
We remark that, since the distance d1 is equivalent to the Sasaki distance, they give the same notion of strong-stable set.
Hypercycles and hyperbolic isometries
Let γ be a hyperbolic isometry of H. It fixes exactly two points in ∂H, the repelling fixed point γ− and the attracting fixed point γ+. The geodesic c joining γ− to γ+ is the axis of γ, and γ acts on c as a translation of length
ℓ(γ)=z∈Hmind(z,γz).
The displacement of a point z∈H by γ only depends on its distance to the axis [4 Theorem 7.35.1]:
sinh(2d(z,γz))=cosh(d(z,c))sinh(2ℓ(γ)).(1)
In particular, the displacement is minimal along the axis and increases with the distance to the axis.
For r>0, the set of points of H at distance r from c consists of two curves, one on each side of c, called hypercycles of γ. Each of them is invariant under γ. In the upper half-plane model, they are arcs of Euclidean circles, or Euclidean half-lines, joining γ− to γ+. Given a point q∈H, we denote by H(γ,q) the hypercycle of γ through q.
Proposition 2. Let c1,c2 be two geodesics in H such that d(c1,c2)>0. Then there exist unique p=c1(t1), q=c2(t2) realizing the distance between c1 and c2. Moreover, there is an hypercycle at distance d(c1,c2) from c1 that is tangent to c2 at q.
If γ belongs to Γ, the hypercycle H(γ,q) projects to a closed curve in S=Γ\H, which we call a closed hypercycle. A direct computation shows that the arc of H(γ,q) between q and γq has length
ℓ(γ)coshd(q,c).(2)
If γ is primitive in Γ, this is the length of the closed hypercycle of γ.
The band model
For some computations it is convenient to use the band model
B={x+iy∈C:−2π<y<2π},ds2=cos2(y)dx2+dy2.
The map z↦iez is an isometry from B onto H, which sends the real line {y=0} onto the imaginary axis. The boundary at infinity of B consists of the two lines {y=±π/2} together with the two points ±∞. We will use the following facts.
The real line {y=0} is a geodesic. For λ>0, the translation z↦z+λ is a hyperbolic isometry with axis {y=0} and translation length λ.
The vertical segments {x=k}, k∈R, are geodesics orthogonal to {y=0}.
The horizontal lines {y=k}, k∈(−π/2,π/2)∖{0}, are hypercycles. More precisely, for x∈R and y∈(−π/2,π/2),
coshd(x,x+iy)=cos(y)1.
Equations of the geodesics relative to {y=0} (see Figure 1). Every geodesic other than {y=0} is of one of the following three types.
Figure 1. Geodesics of the band model B. The labels refer to the items of Section 2.5. The geodesics of item (4a) are tangent to the hypercycles {y=±y0} at their points closest to {y=0}, and the vertical geodesic through one of these points is the common perpendicular. The geodesics of item (4c) are asymptotic to {y=0}.
(a) Ultraparallel geodesics. A geodesic at positive distance from {y=0} has an equation of the form
cosh(x−x0)=αsin(y),x0∈R,∣α∣>1.
Its endpoints are x0±r+iϵπ/2, where ϵ is the sign of α and coshr=∣α∣, and its point closest to {y=0} is x0+iy0, where siny0=1/α.
(b) Intersecting geodesics. A geodesic intersecting {y=0} has an equation of the form
sinh(x−x0)=βsin(y),x0∈R,β∈R.
Its endpoints are x0+s+iπ/2 and x0−s−iπ/2, where sinhs=β. It meets {y=0} at x0, and the angle θ∈(0,π) from the positive direction of {y=0} to the geodesic satisfies cotθ=β. The case β=0 gives the geodesics of item (2).
(c) Asymptotic geodesics. A geodesic asymptotic to {y=0} has an equation of the form
sin(y)=ϵe−δ(x−x0),x0∈R,ϵ,δ∈{−1,1}.
Its endpoints are x0+iϵπ/2 and +∞ if δ=1, or −∞ if δ=−1.
Vanishing asymptotic fineness
Given x∈S, we denote by Inj(x) the injectivity radius of S at x. The next proposition reduces the study of the vectors whose geodesic ray has vanishing injectivity radius, without diverging through a cusp, to three cases. Each case is governed by the geometry of the thin parts visited by the ray: short closed geodesics, cusps, or short geodesic loops enclosed by hypercycles.
Proposition 3. Let S=Γ\H be a hyperbolic surface and let u∈T1S. Assume that the geodesic ray u(R+) does not diverge through a cusp and that its injectivity radius is not bounded below, that is,
t≥0infInj(u(t))=0.
Then at least one of the following holds:
u(R+) crosses an infinite sequence of closed geodesics whose lengths tend to 0,
u(R+) intersects an infinite sequence of closed horocycles whose lengths tend to 0,
u(R+) intersects an infinite sequence of closed hypercycles whose lengths tend to 0 but it does not meet the associated closed geodesics.
Proof. Since inft≥0Inj(u(t))=0, there exists a sequence of times tn≥0 with Inj(u(tn))→0.
Fix a lift u∈T1H of u. The injectivity radius at u(tn) equals half the length of the shortest geodesic loop based at u(tn), so there exists γn∈Γ∖{id} such that
εn:=d(u(tn),γnu(tn))=2Inj(u(tn))⟶0.
Since Γ is torsion free, each γn is either parabolic or hyperbolic. After passing to a subsequence, either all the γn are parabolic, or all of them are hyperbolic.
Parabolic case. Assume that every γn is parabolic. Let xn∈∂H be its fixed point and let Hn be the horocycle centered at xn passing through u(tn). Denote by ℓ(Hn,γn) the length of the horocyclic arc of Hn between u(tn) and γnu(tn). A direct computation in the upper half-plane model shows that
sinh(2εn)=2ℓ(Hn,γn),
hence ℓ(Hn,γn)→0. The horocycle Hn projects to a closed horocycle of S of length at most ℓ(Hn,γn) passing through u(tn). So u(R+) intersects a sequence of closed horocycles whose lengths tend to 0. As u(R+) does not diverge through a cusp, this shows that u is cusp-recurrent in the sense of [2], and we are in case (2).
Hyperbolic case. Assume that every γn is hyperbolic. Let cn be its axis and ℓ(γn) its translation length. Clearly, ℓ(γn)≤d(u(tn),γnu(tn))=εn tends to 0 when n→∞.
We distinguish two subcases.
Subcase (i):u(R+) meets the closed geodesic obtained by projecting cn onto S for infinitely many n. Passing to a subsequence, this holds for every n. These closed geodesics have length at most ℓ(γn). Since ℓ(γn)→0, we are in case (1).
Subcase (ii):u(R+) meets the closed geodesic obtained by projecting cn onto S for at most finitely many n. Passing to a subsequence, u(R+) does not meet these closed geodesics for any n. The hypercycle H(γn,u(tn)) projects to a closed hypercycle in S passing through u(tn). Let γn′ be a generator of the stabilizer of cn in Γ, so that ℓ(γn′)≤ℓ(γn). By (2.2), the length of the projection of H(γn,u(tn)) is
ℓ(γn′)cosh(d(u(tn),cn)).(3)
Using the inequality x≤2sinh(x/2) and (1), we obtain
which tends to 0 when n→+∞. Hence we are in case (3).
We now deduce Theorem 1 from Proposition 3, Theorem 2 and the previous works.
Proof of Theorem 1. Let S be a hyperbolic surface and u∈T1S. Since Inj is continuous and positive on S, the asymptotic fineness of u(R+) vanishes if and only if inft≥0Inj(u(t))=0.
Assume first that the asymptotic fineness of u(R+) is positive. Then Bellis proved that Huss=Wuss[5 Théorème D]. If the geodesic ray u(R+) diverges through a cusp, then Huss is a closed horocycle around that cusp and the equality Huss=Wuss is clear.
Assume now that the asymptotic fineness of u(R+) vanishes and that u(R+) does not diverge through a cusp. By Proposition 3, at least one of the following cases occurs.
The ray u(R+) crosses closed geodesics whose lengths tend to 0. The conclusion follows from [BCDHV26, Main Theorem].
The ray u(R+) intersects closed horocycles whose lengths tend to 0, so u is cusp-recurrent and the conclusion follows from [2 Theorem 2].
The ray u(R+) intersects closed hypercycles whose lengths tend to 0. The conclusion follows from Theorem 2.
In all cases, Huss⊊Wuss and Wuss is an uncountable union of stable horocycles.
The following example shows that case (3) of Proposition 3 can occur while cases (1) and (2) do not.
Example 3. There exist a hyperbolic surface S without cusps and a vector u∈T1S such that
liminft≥0Inj(u(t))=0,
for every sequence of closed geodesics (ck) with ℓ(ck)→0, we have d(ck,u(R+))→+∞.
Let ℓn be a sequence of positive numbers converging to 0 and let D>0. Let Pn be the pair of pants with boundary geodesic components of lengths ℓn, D and D [Bus92, Chapter 3]. For n≥0, Pn is glued to Pn+1 through a geodesic boundary of length D. Then funnels are attached to the remaining boundary components. We thus obtain a complete hyperbolic surface with no cusps called a flute surface.
Let αn be the boundary component of Pn of length ℓn. Let kn be a sequence of integers. We consider a vector u∈T1S whose geodesic ray u(R+) has the following behavior:
it starts in P0, that is, u(0)∈P0,
it visits the pairs of pants P0,P1,P2,… in this order,
for every n≥1, in Pn the ray turns kn times around αn and does not turn around the other two boundary geodesics.
By choosing the numbers of turns kn appropriately, we can guarantee that the distance between u(R+) and αn tends to infinity, while u(R+) visits points of the collar of αn where the injectivity radius is small, so that the injectivity radius along the ray vanishes.
To conclude one shows that there exists a constant ε0>0 such that every closed geodesic of S distinct from the geodesics αn has length at least $\varepsilon_0.
Winding around a closed hypercycle
Definition 4. Let γ be an hyperbolic isometry of H with axis c and let α:R→H be a unit speed parametrization of a hypercycle of c. The orientation of α is positive with respect to γ if there exists D>0 such that for all t∈R,
γα(t)=α(t+D).
Definition 5. Let γ be an hyperbolic isometry of H with axis c and let u~∈T1H such that d(c,u~(R))>0. Let q=u~(t1) be the point in u~(R) that realizes the distance to c. Let α:R→H be the positive unit speed parametrization of the hypercycle of c through q such that α(0)=q. We say that u~ is in good position with respect to γ if t1>0 and
dtdu~(t1)=dtdα(0).
Definition 6. Let u~∈T1H be in good position with respect to a hyperbolic isometry γ. We define the winding of u~ around γ (see Figure 2), denoted by Windγ(u~)∈T1H, as the unique vector satisfying
Windγ(u~)(0)=u~(0).
Windγ(u~)(+∞)=γu~(+∞).
Figure 2. The winding Windγ(u~) of u~ around γ, in the disk model.
Definition 7. The winding time associated to γ is the real number τγ,u~ defined by
τγ,u~=Bu~(+∞)(γ−1u~(0),u~(0)).
Proposition 4 (bound of the winding time). Let u~∈T1H be in good position with respect to a hyperbolic isometry γ. Let q=u~(t1) be the point in u~(R) that realizes the distance to the axis c of γ. Then
For the other inequality, we observe that γu~(+∞) is between u~(+∞) and the attracting fixed point xγ+ of γ. Then the geodesic ray [u~(0),γu~(+∞)) is closer than u~(R+) to the axis c of γ. Let q′ be the point of [u~(0),γu~(+∞)) closest to c. We have d(q′,c)≤d(q,c). Then d(q′,γq′)≤d(q,γq).
Since q′ is in the geodesic ray [u~(0),γu~(+∞)), we have
Proposition 5 (Key Proposition). Let u~∈T1H be in good position with respect to a hyperbolic isometry γ. Let q be the point in u~(R) closest to the axis of γ. Assume that d(q,γq)≤1. Then, for every t≥0,
Proof. For simplicity, we denote v~=Windγ(u~) and τ=τγ,u~. Let q=u~(t1), t1≥0, be the point that realizes the distance of u~(R) to the axis of γ. We observe that u~(0) and γu~(+∞) lie in the same half-plane in H∖(γ−,γ+). Hence, the geodesic generated by v~ does not intersect the axis of γ. Let q′=v~(t1′), t1′∈R, be the point that realizes the distance of v~(R) to the axis of γ. The first step is to control the distance between q and q′.
To do the next computations we use the band model B (see § 2.5). We can assume that the axis of γ is the geodesic y=0 and that γ has the form γ(z)=z+λ, where λ>0 is the translation length of γ.
The geodesics at positive distance from y=0 are those of item (4a). Without loss of generality, we can assume that the equation of u~(R) has x0=0 and α>1. Hence,
u~(R)={(x,y)∈R×(−π/2,π/2)∣cosh(x)=αsin(y)},
which lies in the upper half-strip y>0. The endpoints at infinity of u~(R) are
u~(−∞)=−r+π/2i,u~(+∞)=r+π/2i,
where r>0 satisfies cosh(r)=α. The point q realizing the distance to the axis y=0 is q=y0i with α=1/sin(y0). The initial point u~(0) lies between u~(−∞) and q, so it has negative real part.
The geodesic v~(R) is also at positive distance from the axis y=0 and in the upper half-strip. Let
cosh(x−x0′)=α′sin(y),x0′∈R,α′>1,
be the equation of v~(R). The endpoints of v~(R) are
v~(−∞)=x0′−r′+π/2i,v~(+∞)=x0′+r′+π/2i,
where r′>0 satisfies cosh(r′)=α′. The point q′ realizing the distance to the axis is q′=x0′+y0′i with α′=1/sin(y0′).
By definition, v~(+∞)=γu~(+∞), which implies r+λ=x0′+r′. The point v~(−∞) must lie between −r+π/2i and π/2i.
From the position of the endpoints of u~(R) and v~(R), we obtain the following estimates
(1) 0<y0′<y0,
(2) 0<r′−r<λ/2,
(3) 0<x0′<λ.
Consider the function f(t)=arcsin(1/cosh(t)) defined for t∈R. We have f(r)=y0 and f(r′)=y0′. An easy computation shows that f is 1-Lipschitz, hence
∣y0−y0′∣≤∣r−r′∣<2λ.
Introducing q′′=x0′+y0i, we have
d(q,q′)≤d(q,q′′)+d(q′′,q′).
First, we notice that γq=λ+y0i. Since 0<x0′<λ, the point q′′=x0′+y0i belongs to the hypercycle segment between q and γq. This implies that d(q,q′′)≤d(q,γq).
The points q′′ and q′ lie on the vertical geodesic {x=x0′} (item (2)). Hence d(q′′,q′) is the length of the vertical segment joining them and, since 0<y0′<y0,
Together with the previous bounds, these imply the statement. □
Proof of Theorem 2
Rays intersecting arbitrarily short hypercycles
Let S=Γ\H be a hyperbolic surface and let u∈T1S be such that the geodesic ray u(R+) meets closed hypercycles of arbitrarily small length. Notice that each closed hypercycle has a closed geodesic in its free homotopy class, which is shorter. The case where u(R+) meets infinitely many of these closed geodesics falls in the first case of Proposition 3 and the conclusion of Theorem 2 follows from [BCDHV26]. Therefore, we may assume that u(R+) does not meet the closed geodesics associated to these hypercycles.
Recall that H(γ,q) denotes the hypercycle of γ through q (Section 2.4). If γ is in good position with respect to u~ and q is the point of u~(R) closest to the axis of γ, then u~(R) is tangent to H(γ,q) at q.
Proposition 6. Let S=Γ\H be a hyperbolic surface and let u∈T1S be such that u(R+) meets closed hypercycles of arbitrarily small length without meeting the associated closed geodesics. Then, up to conjugating Γ by an isometry of H, there
exist a lift u~ of u with u~(0)=i and u~(+∞)=∞, and a sequence (γn)n≥0 of hyperbolic isometries in Γ such that
u~ is in good position with respect to each γn,
the point qn=u~(tn) of u~(R) closest to the axis of γn is such that (tn)n≥0 is increasing and tends to +∞,
0<γn−<γn+<γn+1− for all n≥0,
the hypercycles H(γn,qn) are pairwise disjoint,
d(qn,γnqn) tends to 0.
Proof. By hypothesis, there are times sn≥0 such that u(sn) belong to a hypercycle whose length tend to 0 when n tends to +∞.
Up to conjugating the group Γ by an isometry of H, we can assume that u lifts to u~∈T1H such that u~(0)=i and u~(+∞)=∞. For every n, there exists a primitive hyperbolic isometry γn such that the hypercycle H(γn,u~(sn)) of γn through u~(sn) project to the closed hypercycles whose length tend to 0. In particular, εn:=d(u~(sn),γnu~(sn)) is smaller than the length of the hypercycle, so it also tends to 0.
Since ℓ(γn)≤εn→0, up to passing to a subsequence, the isometries γn can be taken to be pairwise distinct. It can also be assumed that their fixed points are different from 0 and ∞.
Let cn be the axis of γn. By hypothesis u~(R+) does not intersect cn. Since the fixed points of γn are different from 0 and ∞, the geodesics u~(R) and cn are not asymptotic. Let qn=u~(tn) be the point of u~(R) closest to cn. By (2.1), the displacement d(z,γnz) is smaller the closer z lies to cn. In particular,
d(qn,γnqn)≤d(u~(sn),γnu~(sn))=εn⟶0.
We claim that tn>0 for n large enough. Otherwise, if tn≤0, then tn≤0≤sn, so u~(0)=i is closer to cn than u~(sn). By the property of the displacement function, we would get
d(i,γni)≤d(u~(sn),γnu~(sn))=εn,
which if it happened for infinitely many n would contradict the discreteness of Γ. Up to taking a subsequence, we assume tn>0. Moreover, qn does not lie on cn, thus u~(R) stays at positive distance from cn.
Observe that u~(R) is tangent to the hypercycle of γn through qn. Replacing γn by γn−1 if necessary, the orientation of u~(R) agrees with the one induced by γn on the hypercycle through qn. Together with tn>0, this shows that u~ is in good position with respect to γn.
Moreover, tn→+∞. Otherwise, up to passing to a subsequence, the points qn stay in a compact subset of H, and d(qn,γnqn)→0 with pairwise distinct γn would contradict the discreteness of Γ. Up to passing to a subsequence, we may assume that (tn)n≥0 is increasing.
The axis cn does not meet u~(R), hence it is contained in one of the two half-planes {x>0} or {x<0}. Up to taking a subsequence and conjugating again the group Γ by the reflection z↦−z, which fixes u~(R) pointwise and preserves good position, we may assume that they all lie in {x>0}. Then γn−>0 and γn+>0, and good position forces the attracting fixed point to be on the right of the repelling one, so that 0<γn−<γn+.
The geodesic realizing the distance between u~(R) and cn is orthogonal to both, and it passes through the point qn. Hence it is a Euclidean half-circle centered at 0. One of the end points xn of this geodesic is between γn− and γn+, so γn−<xn<γn+. By definition, qn=u~(tn)=ietn, then xn=etn tends to +∞, and so does γn+. We claim that γn− also tends to +∞. Otherwise, up to passing to a subsequence, we may assume that γn− converges to some ξ∈R. Then the axes cn converge to the geodesic with endpoints ξ and +∞. Picking a point z on this geodesic, it is accumulated by points zn on cn. Since ℓ(γn) tends to 0, we have d(z,γnz)→0, which contradicts the fact that Γ is discrete. Hence, we must have limnγn=+∞.
The hypercycle H(γn,qn) is an arc of the Euclidean circle through γn−, γn+ and qn, which is tangent to the imaginary axis at qn (see Figure 3). Then the Euclidean distance between 0 and H(γn,qn) tends to +∞. For this we can observe that this distance is greater than the distance from 0 to the Euclidean circle of radius γn− centered at γn−+γn−i. This allows us to extract a subsequence such that the hypercycles H(γn,qn) are pairwise disjoint and γn+<γn+1− for all n≥0. □
Figure 3. The geodesic u~(R), the axis cn of γn, and the hypercycle H(γn,qn) tangent to u~(R) at qn.
From now on we fix the vector u∈T1S, its lift u with u(0)=i and u(+∞)=∞, and the sequence (γn)n≥0 of hyperbolic isometries given by Proposition 6.
Construction of the vector wα in Hssu
Let (αn)n≥0 be a subsequence of (γn)n≥0. We write βn=α0⋯αn∈Γ, and we denote by αn−, αn+ the repelling and attracting fixed points of αn, so that 0<αn−<αn+<αn+1−. For each n≥0, let qnα be the point of u(R) closest to the axis of αn, and set
δnα=d(qnα,αnqnα).
Since (αn) is a subsequence of (γn), we have δnα→0. We introduce the vectors vn∈T1H defined by
vn(0)=i,vn(+∞)=βn(∞).
Lemma 2.The sequence (vn)n≥0 converges to a vector vα∈T1H defined by vα(0)=i and vα(+∞)=limn→+∞βn(∞).
Proof. Since every vn is based at i, it suffices to prove that the sequence (βn(∞))n≥0⊂∂H converges. Each αi is a hyperbolic isometry with repelling and attracting fixed points $0<\alpha_i^-<\alpha_i^+$. Since ∞ lies outside [αi−,αi+], it is attracted by αi toward αi+, so that αi+<αi∞<+∞. Moreover, αi preserves the order on the interval (αi+,+∞), hence αi+<x<y implies αi+<αix<αiy. Using the nesting αi+<αi+1−, we obtain for every $n\geq 1
0<αn−1αn∞<αn−1∞.
Applying α0,…,αn−2 successively, which preserve the order on the relevant intervals, yields
0<βn(∞)<βn−1(∞).
Hence (βn(∞))n≥0 is a decreasing sequence of positive numbers, and therefore converges. □
The ray βn−1vn(R+) issues from βn−1i and points to βn−1(βn∞)=∞; it is therefore the vertical geodesic ray starting at βn−1i. The next lemma places this ray in good position with respect to αn+1 and compares its displacement with δn+1α. This is the geometric point where the present case departs from [BCDHV26]: the ray does not cross the axis of αn+1 but remains tangent to one of its hypercycles.
Lemma 3.For every n≥0, the vector βn−1vn is in good position with respect to αn+1. Moreover, if qn+1′ denotes the point of βn−1vn(R) closest to the axis of αn+1, then
d(qn+1′,αn+1qn+1′)≤δn+1α.
Proof. For n≥0, let Rn be the closed region of the first quadrant bounded by the segment of the imaginary axis between 0 and qnα, the arc of the hypercycle H(αn,qnα) between qnα and αn−, and the segment [0,αn−] of ∂H. Since the hypercycles H(αn,qnα) are pairwise disjoint and qnα=u(tnα) with (tnα) increasing, we have Rn⊂Rn+1 for all n≥0. Moreover, since H(αn,qnα) is invariant under αn and αˉn−1 moves the points of ∂H toward its attracting fixed point αˉn, one checks that αˉn−1(Rn)⊂Rn.
We prove by induction that βn−1i∈Rn for all n≥0. For n=0 we have α0−1(i)∈α0−1(R0)⊂R0, because i∈R0. Assume βn−1i∈Rn. Then
Since βn−1i∈Rn⊂Rn+1, the vertical ray βn−1vn(R+) lies on the left of αˉn, hence it stays at positive distance from the axis of αn+1, and is tangent to a hypercycle of αn+1 enclosed by H(αn+1,qn+1α). The orientation of βn−1vn(R+) agrees with the one induced by αn+1, so βn−1vn is in good position with respect to αn+1. Finally, the point qn+1′ realizing the distance to the axis lies on a hypercycle enclosed by H(αn+1,qn+1α), hence closer to the axis than qn+1α. As the displacement function decreases with the distance to the axis, we conclude
By Lemma 3, the winding Windαn+1(βn−1vn) is defined, and it is the vector based at βn−1i pointing at αn+1∞. On the other hand, βn−1vn+1 is based at βn−1i and points at βn−1(βn+1∞)=αn+1∞. Hence, for all n≥0,
βn−1vn+1=Windαn+1(βn−1vn).
By the definition of the winding time (Definition 7) applied to αn+1 and βn−1vn, together with Proposition 4 and Lemma 3,
By construction grnβn−1vn∈Hssu, so its projection satisfies grnvn∈Hssu.
Proposition 7. Assume that ∑n≥0δnα<+∞. Then the sequence (rn)n≥0 converges to a real number rα, and ∣rn∣≤∑i≥0δiα for all n≥0; hence ∣rα∣≤∑i≥0δiα.
using ∣r0∣=∣τα0,u∣≤δ0α (Proposition 4). Letting n→+∞ gives the last inequality.
Construction of the vector wα in Wssu
Throughout this subsection we fix a subsequence (αn)n≥0 of (γn)n≥0 such that δnα≤1 for all n and ∑n≥0δnα<+∞, and we write Δ=∑n≥0δnα.
By Proposition 7 the winding times rn converge to rα, and by Lemma 2 the vectors vn converge to vα. We set, for n≥0,
wn=grnvn∈T1H,wα=grαvα=n→+∞limwn∈T1H.
We denote by vn, vα, wn, wα the projections to T1S of vn, vα, wn, wα, respectively. We remark that, since βn−1wn=grnβn−1vn∈Hssu, the vectors wn lie in Hssu and wn converges to wα.
Our goal is to show that in fact wα∈Wssu. For this we show that the distance between gtwn and gtu tends to zero uniformly in n and then we take the limit in n.
Proposition 8. For every ε>0 there exists a time T>0 such that
d1(gtu,gtwn)<εfor all t>T and all n≥0.
Proof. Fix ε>0. Since ∑kδkα<+∞, we may fix N with 17∑k>Nδkα<ε/2. We will use that, if n>N, then
Recall that βk−1vk+1=Windαk+1(βk−1vk) and ταk+1,βk−1=rk+1−rk, and that, by Lemma 3, the point of βk−1vk(R) closest to the axis of αk+1 has displacement at most δk+1α≤1. Applying the Key Proposition 5 to βk−1vk and αk+1 and then projecting to T1S, we obtain
for all t≥−rk. Since ∣rk∣≤Δ, this holds for all t≥Δ in particular. Therefore the second term in (5.2) is bounded by 17∑k=Nn−1δk+1α<ε/2 for all t≥Δ.
To bound the first term in (5.2) or to treat the case n≤N, we may observe again that wi∈Hssu, so d1(gtwi,gtu)→0 as t→+∞. Hence there exists T≥Δ such that d1(gtwi,gtu)<ε/2 for all t>T. This concludes the proof. □
Since wα=limn→+∞wn, letting n→+∞ in Proposition 8 gives, for every ε>0 and every t>T,
d1(gtwα,gtu)=n→+∞limd1(gtwn,gtu)≤ε.
Hence limt→+∞d1(gtwα,gtu)=0, that is, wα∈Wssu.
Construction of the vector wα in Wssu∖Hssu
We denote by Σ the set of subsequences α=(αn)n≥0 of (γn)n≥0 such that
δnα≤1for all n≥0andn≥0∑δnα<+∞.
By Proposition 8, every α∈Σ satisfies wα∈Wssu. Since δn=d(qn,γnqn)→0, the set Σ is nonempty. For α∈Σ we set
ξ(α)=wα(+∞)=n→+∞limβn(∞)∈R,
which is a positive number by Lemma 2.
Proposition 9. The image ξ(Σ) is uncountable.
Proof. We use the following property, which follows from the dynamics of the αn as in the proof of Lemma 2: for every α∈Σ and every n≥0,
αnαn+1⋯αm(∞)>αn+for all m≥n,
hence
αn−1−1⋯α0−1(ξ(α))≥αn+.(6)
Assume, for contradiction, that ξ(Σ) is countable and enumerate it as {x0,x1,…}. We construct α′=(αn′)∈Σ with ξ(α′)=xi for all i.
Choose α0′=γk0 with δ0′<1 and α0′+>x0. Inductively, having chosen α0′,…,αn′, pick αn+1′=γkn+1 with kn+1>kn such that
δn+1′<2n+11andαn+1′+>αn′−1⋯α0′−1(xn+1).
This is possible since γk+→+∞ and δk→0. Since ∑n≥0δnα′<∑n≥02−n<+∞ and δnα′≤1, the resulting sequence α′ belongs to Σ.
By (5.3) and the choice of αn′, we have for n≥1
αn−1′−1⋯α0′−1(ξ(α′))≥αn′+>αn−1′+⋯α0′+(xn),
whence ξ(α′)=xn. Finally ξ(α′)>α0′+>x0, so ξ(α′)=x0. This contradicts ξ(α′)∈ξ(Σ)={xi}. □
Corollary 1.There exists a subsequence α∈Σ such that wα∈/Huws. Moreover, Wuss contains an uncountable union of pairwise distinct stable horocycles.
Proof. If w∈T1S belongs to Huws, then w(+∞)∈Γu(+∞)=Γ∞ for any lifts. Since Γ is countable and ξ(Σ) is uncountable by Proposition 9, the set ξ(Σ) meets uncountably many Γ-orbits in ∂H. Hence there exists α∈Σ with ξ(α)=wα(+∞)∈/Γ∞, that is, wα∈/Huws. By Proposition 8, all these wα lie in Wuss, and their forward endpoints lie in distinct Γ-orbits, so the stable horocycles Hwαss⊂Wuss are pairwise distinct and uncountable in number. □
Declaration on the use of AI
During the preparation of this work the authors used Claude (Anthropic) to assist with the writing and the proofreading of the manuscript. The mathematical content and the arguments are entirely the authors’ own. The authors have reviewed all the text and take full responsibility for the content of the article.
References
IMERL, CMAT, Universidad de la República, IFUMI, IRL CNRS 2030, Uruguay
[1]Werner Ballmann. Lectures on spaces of nonpositive curvature, volume 25 of DMV Seminar. Birkhäuser Verlag, Basel, 1995. With an appendix by Misha Brin.DOI
[2]Sergi Burniol Clotet and Françoise Dal’Bo. On the non-expansiveness of the geodesic flow on surfaces with cusps, 2026. Preprint, arXiv:2603.24310.arxiv.org/abs/2603.24310
[4]Alan F. Beardon. The Geometry of Discrete Groups, volume 91 of Graduate Texts in Mathematics. Springer, New York, 1983.DOI
[5]Alexandre Bellis. Étude topologique du flot horocyclique : le cas des surfaces géométriquement infinies. PhD thesis, Université de Rennes, May 2018.DOI
[6]Alexandre Bellis. On the links between horocyclic and geodesic orbits on geometrically infinite surfaces. Journal de l’École polytechnique. Mathématiques, 5:443–454, 2018.DOI