Towards the boundary of the fine curve graph

Abstract.

The fine curve graph was introduced as a geometric tool to study the homeomorphisms of surfaces. In this paper we study the Gromov boundary of this space and the local topology near points associated with minimal measurable foliations. We then give several applications including finding explicit elements with positive stable commutator length, and proving a Tits alternative for subgroups of $\textrm{Homeo}(S)$ containing a pseudo-Anosov map, generalizing a result of Hurtado-Xue.

1. Introduction

Geometry of Curve Complexes

The curve complex was introduced by Harvey [Har81] to study mapping class groups, and has subsequently become one of the most powerful tools to do so. One of the main reasons is that the coarse geometry of the curve graph was shown to be negatively curved in the landmark work of Masur-Minsky [MM99]. More precisely, they show that the curve graph $\mathcal{C}(S)$, i.e. the $1$-skeleton of the curve complex, of any hyperbolic surface is $\delta$-hyperbolic in the sense of Gromov [Gro87].

The curve graph (as a coarse metric object) has a natural bordification given by considering its Gromov boundary (that is in fact present for any coarsely hyperbolic space). Furthermore, this abstract boundary has a natural geometric description as the space of minimal measureable (singular) foliations by Klarreich [Kla22]. This identification means that one can translate coarsely geometric properties of the action on the curve graph into honest topological or dynamical properties of mapping classes acting on the surface.

In previous work [BHW22] we introduced the fine curve graph as a variant of the curve graph which can be used to study surface homeomorphism and diffeomorphism group. The fine curve graph $\mathcal{C}^{\dagger}(S)$ is defined as the graph whose vertices are embedded essential simple curves which are connected by an edge if they are disjoint, and we proved that it is also $\delta$-hyperbolic [BHW22]. In subsequent work [BHM${}^{+}$22] we began to develop a dictionary between properties dynamical properties of homeomorphisms and geometric properties of their action on the fine curve graph.

Classification of Surface Homeomorphisms

Thurston famously classified mapping classes [Thu88] by showing that any mapping class is isotopic to one that is periodic, reducible or pseudo-Anosov. Masur and Minsky [MM99] re-interpreted this classification in terms of the curve graph: pseudo-Anosov mapping classes are exactly those that act on the curve graph loxodromically (i.e. with positive asymptotic translation length). The endpoints at infinity of a quasi-axis are exactly the stable and unstable foliation of the pseudo-Anosov (under the identification given by Klarreich’s theorem). Periodic or reducible mapping classes act elliptically (i.e. with bounded diameter orbits). We remark that the third type of isometries of a hyperbolic space – parabolic isometries, which have unbounded orbits, but zero asymptotic translation length – do not occur when considering the action of mapping class group on curve graphs.

For the action of actual homeomorphisms on $\mathcal{C}^{\dagger}(S)$, only a partial analog of this dictionary exists. A basic tool to study the geometry of the fine curve graph is the existence of coarsely defined 1-Lipschitz projections to the surviving curve graph of suitably punctured surface:

$$\mathcal{C}^{\dagger}(S)\longrightarrow\mathcal{C}^{surv}(S\setminus P).$$

The existence of these maps implies that homeomorphisms of $S$ which fix a finite set of points $P$, and define a pseudo-Anosov mapping class on $S\setminus P$ act as hyperbolic isometries on the fine curve graph. By now, it is known that the converse is also true (see [BHM${}^{+}$22] for the case of the torus, and [GM23] for the general case). However, in contrast to the classical setting, there are also homeomorphisms which act as parabolic isometries on $\mathcal{C}^{\dagger}(S)$ [BHM${}^{+}$22], and a dynamical classification of elliptic or parabolic isometries seems currently out of reach.

Boundary points in the Fine Curve Graph

The central goal of this paper is to prove Klarreich-like results for the fine curve graph. While we cannot describe every boundary point, we at least control a large set of such points. Namely, we show:

Proposition 1.1 (Foliations as boundary points).

Any minimal measurable foliation $\mathcal{F}$ on a punctured hyperbolic surface $S\setminus P$ determines a (unique) point $\xi_{\mathcal{F}}$ in the Gromov boundary $\partial_{\infty}\mathcal{C}^{\dagger}(S)$.

A sequence of smooth curves $\gamma_{n}$ converges to $\xi_{\mathcal{F}}$ (as points in the Gromov hyperbolic space $\mathcal{C}^{\dagger}(S)$) exactly if the curves $\gamma_{n}$ (as curves on the surface) converge into leaves of $\mathcal{F}$: for any $L$ and any $\varepsilon$, and large $n$, every subsegment of length $L$ of $\gamma_{n}$ is contained in an $\varepsilon$–neighbourhood of a leaf of $\mathcal{F}$.

Note that the important aspect of this proposition, which we will make essential use of below, is that the convergence is geometric. We also emphasise that the version stated here is not the most general, since we will have to work with non-smooth curves, where length will be replaced by a more general notion.

Also note that since the action of $\mathrm{Homeo}(S)$ on $\partial_{\infty}\mathcal{C}^{\dagger}(S)$ is minimal [LT24], Proposition 1.1 describes a dense set of boundary points.

Stable commutator lengths

We have previously shown that there exist unbounded quasi-morphisms on the identity component $\operatorname{Homeo}_{0}(S)$ and hence there are elements with positive stable commutator length. However, while in theory one could extract an explicit such element from our proof (and the work of Bestvina-Fujiwara [BF02] we rely on), we could not give robust criteria to certify that specific, dynamically meaningful homeomorphisms have positive scl. The following theorem remedies this, and gives such a geometric criterion.

Theorem 1.2.

Let $F$ be a Thurston representative of a pseudo-Anosov mapping class in $\mathrm{Mcg}(S\setminus P)$ on a punctured surface that is not conjugate to its inverse (up to powers) in $\mathrm{Mcg}(S\setminus P)$. Then $F$ has positive stable commutator length.

If $F$ is a Thurston representative of a point push pseudo-Anosov of a single point, then $F$ in fact has positive stable commutator length in $\mathrm{Homeo}_{0}(S)$ without any further assumptions (see Remark 5.4).

We also consider the case of surfaces with boundary, thereby answering a question posed to us by Monod and Nariman:

Theorem 1.3.

Let $S$ denote a connected surface with (possibly empty) boundary such that $\chi(S)<0$. Then the identity component $Homeo_{0}(S,\partial S)$ admits unbounded quasi-morphisms.

This more or less resolves the problem of which surface homeomorphism groups admit quasi-morphisms, since for $S^{2}$ the annulus and the disk, one knows that they cannot exist – by [BIP08] these groups do not admit any conjugation invariant norms at all. For all other closed orientable surfaces this was previously proved in [BHW22] and extended to non-orientable surfaces with (non-orientable) genus at least three by [KK22].

Stabilizers of points of the Gromov boundary

Given that boundary points can be associated to measurable foliations one can ask about the stabilizer of such points. The following extends results of Hurtado-Xue [HX21, Theorem 1.10] where $C^{1}$-regularity was assumed and only the case of the torus $T^{2}$ was considered:

Theorem 1.4.

Suppose that $G<\mathrm{Homeo}_{+}(S)$ is a group which contains a Thurston representative of a pseudo-Anosov element of $\mathrm{Mcg}(S\setminus P)$ for some finite puncture set $P$. Then either $G$ contains a free group, or has an index $2$ subgroup which fixes a foliation on $S$.

Similarly we can show that certain boundary points cannot be stabilised by parabolic (or hyperbolic) elements, answering a question posed to us by Yair Minsky.

Proposition 1.5.

Suppose that $\mathcal{F}$ is a non-orientable minimal foliation on the hyperbolic surface $S\setminus P$ (for $P$ finite), and $\xi_{\mathcal{F}}$ the corresponding boundary point given by Proposition 1.1. Then

1. $\xi_{\mathcal{F}}$ is not fixed by any parabolic isometry of $\mathcal{C}^{\dagger}(S)$.

2. If $\mathcal{F}$ is not the invariant foliation of a pseudo-Anosov on $S\setminus P$, then $\xi_{\mathcal{F}}$ is not fixed by any hyperbolic isometry of $\mathcal{C}^{\dagger}(S)$.

Acknowledgements

We would like to thank Katie Mann, Sam Nariman and Yair Minsky for stimulating questions and helpful comments. The first and second author are supported by the Special Priority Programme SPP 2026 Geometry at Infinity funded by the DFG. We would also like to thank Federica Fanoni and José Andrés Rodríguez Migueles for comments on an earlier draft. The third author was partially supported by the EPSRC Fellowship EP/N019644/2.

2. (Not Just) Background

In this section, we recall some background about fine curve graphs, and singular flat structures. In both cases, we also prove a few new results which are adapted to our purposes.

2.1. Hyperbolicity of the fine curve graph

Let $S$ be a closed, connected, oriented surface of genus at least 1. The fine curve graph $\mathcal{C}^{\dagger}(S)$ is the graph whose vertices are essential simple closed curves (up to reparametrisation), and edges correspond to disjointness (except if $S$ is a torus, in which case edges correspond to a single intersection point).

We have

Theorem 2.1 ([BHW22]).

The fine curve graph is Gromov hyperbolic.

We will require a specific construction of quasigeodesics, which is implicit in previous work, but has not appeared in a concrete form.

Namely, if $\alpha,\beta$ are two essential simple closed curves intersecting transversely, we define a bicorn (defined by $\alpha,\beta$) to be an essential simple closed curve of the form

$$a\cup b,\quad\quad a\subset\alpha,b\subset\beta,$$

where $a,b$ are subarcs. We emphasise that both simplicity and essentialness are requirements – in general, an arbitrary choice of arcs $a,b$ will not yield a bicorn, even if they share endpoints.

We need the following result of Rasmussen

Theorem 2.2 ([Ras20]).

There is a constant $\delta>0$ so that for any finite type surface $S$, the nonseparating curve graph $\mathcal{NC}(S)$ is $\delta$–hyperbolic. In addition, if $\alpha,\beta\subset S$ are two curves in minimal position, then the set of all nonseparating bicorns formed by $\alpha,\beta$ is Hausdorff close to a geodesic in the curve graph of $S$ joining $\alpha$ to $\beta$ (with uniform constants).

The last part of this theorem is not stated as such in [Ras20], so we briefly explain why it follows from his proof. Namely, Rasmussen obtains hyperbolicity of the nonseparating curve graphs using the following criterion by Bowditch [Bow14, Proposition 3.1]:

Proposition 2.3.

Let $X$ be a graph, $D>0$ a number, and suppose for each pair of vertices $x,y\in V(X)$ we have chosen a connected subgraph $\mathcal{L}(x,y)$ containing $x,y$.

Suppose that if $d(x,y)\leq 1$, the diameter of $\mathcal{L}(x,y)$ is at most $D$, and in addition for all $x,y,z$ we have

$$\mathcal{L}(x,y)\subset N_{D}(\mathcal{L}(x,z)\cup\mathcal{L}(z,y)).$$

Then $X$ is $\delta(D)$–hyperbolic, and for each $x,y$ the subgraph $\mathcal{L}(x,y)$ is $B(D)$–Hausdorff close to a geodesic joining $x$ to $y$.

In [Ras20], Rasmussen shows that putting $\mathcal{L}(\alpha,\beta)$ to be the graph spanned by bicorns defined by $\alpha,\beta$, the prerequisites of the above proposition are satisfied. Hence, Theorem 2.2 follows.

2.2. Boundaries of Gromov hyperbolic spaces

Here, we briefly recall the definition of the boundary at infinity of a hyperbolic space in terms of Gromov products, which is suitable for spaces that are not locally compact. We only very briefly recall the required notions, and refer the reader to e.g. [BH99, III.H.3] for details.

For a metric space $X$ and a basepoint $x_{0}\in X$ we recall the Gromov product

$$(x\cdot x^{\prime})_{x_{0}}=\frac{1}{2}\left(d(x_{0},x)+d(x_{0},x^{\prime})-d(x,x^{\prime})\right).$$

If $X$ is $\delta$-hyperbolic, then the Gromov product is (up to uniform additive constants) the distance from $x_{0}$ to a geodesic joining $x$ and $x^{\prime}$. In other words, the Gromov product measures when geodesics from $x_{0}$ to $x$ and $x^{\prime}$ start to diverge.

A sequence $(a_{i})$ in $X$ is called admissible if

$$(a_{i}\cdot a_{j})_{x_{0}}\to\infty,\quad i,j\to\infty,$$

and two admissible sequences $(a_{i}),(b_{j})$ are equivalent if

$$(a_{i}\cdot b_{j})_{x_{0}}\to\infty,\quad i,j\to\infty.$$

We then define the Gromov boundary $\partial_{\infty}X$ to be the set of admissible sequences up to equivalence. One then extends the Gromov product to boundary points by setting

$$(\xi\cdot\eta)_{x_{0}}=\sup_{(a_{i})=\xi,(b_{j})=\eta}\left(\liminf_{i,j\to\infty}(a_{i}\cdot b_{j})_{x_{0}}\right)$$

One then defines a topology on $X\cup\partial_{\infty}X$. Namely, for $\xi\in\partial_{\infty}X$, a neighbourhood basis is given by sets of the form

$$U(\xi,K)=\{\eta\in X\cup\partial_{\infty}X\quad(\eta\cdot\xi)_{x_{0}}>K\}.$$

Note that if $X$ is proper, the boundary defined above agrees with the more common one in terms of geodesic rays. For non-proper spaces, it is not clear that every Gromov boundary point (in the sense above) is the endpoint of a geodesic ray.

If $Y\subset X$ is any subset, we denote by $\partial_{\infty}Y$ the subset of the boundary defined by accessible sequences with terms in $Y$. If $Y$ is quasiconvex, and therefore itself hyperbolic, this is just (an embedding of) the Gromov boundary of $Y$ (into the Gromov boundary of $X$).

2.3. Singular Flat Structures

Recall that a singular flat structure $q$ on a surface $S$ is a flat metric outside a finite set of points $P\subset S$, and each point $P$ is a cone point of angle $k\pi$ for some $k\in\mathbb{N}$. Alternatively, it is given by an atlas into the complex plane whose transition functions are of the form $z\mapsto\pm z+c$. Another equivalent data is a quadratic differential for (some) complex structure on $X$. We refer the reader to [Str84] for background on such structures and the equivalence. Here, we only briefly recall the central results about these structures which we will need in the sequel.

In a singular flat metric, the notion of horizontal and vertical segment make sense, and one has in fact two transverse singular foliations

$$\mathcal{F}^{h}_{q},\mathcal{F}^{v}_{q}$$

whose leaves are formed by all horizontal (respectively vertical) segments. In fact, both of these carry a natural transverse measure, given (locally) by integrating the vertical (respectively horizontal) component of a path.

A leaf of one of these foliations is called regular if it contains no singular point. For any point $p$ on a regular leaf $l$, the leaf $l$ is the only leaf through $p$. Conversely, singular leaves are not determined by their points (as there are choices to be made how to continue through singularities). Since there are only countably many singular leaves, almost every point is regular (and almost every point on any path transverse to the foliation is also regular).

Good covers and CAT(0) geometry

Given a quadratic differential $q$ on a surface $S$, a straight segment is an embedded segment $I\subset S$ which intersects the singularities of $q$ at most in its endpoints, and which is a straight line in the flat charts. A geodesic for $q$ is a concatenation of straight segments $s_{i}$ so that the angle between $s_{i}$ and $s_{i+1}$ at the joining point is at least $\pi$ on both sides. If $q$ does not have angle-$\pi$ singularities, these are indeed locally geodesic for the metric, and are very well-behaved (since the induced metric is locally CAT(0)).

However, we explicity need to allow angle-$\pi$-singularities (as these appear e.g. in the invariant foliations of point-pushing pseudo-Anosovs) – and then the geodesics for $q$ are much less well-behaved, since they can e.g. run into an angle-$\pi$ singularity and return in exactly the same way.

To control geodesics, we will therefore frequently use the following trick. Given a surface $S$ and a quadratic differential $q$, there is a branched cover $\hat{S}\to S$ (branched over $P$), and a lift $\hat{q}$ so that every singularity of $\hat{q}$ has cone angle at least $3\pi$. We call the data $(\hat{S},\hat{q})$ a good cover for $q$. Good covers always exist, and while they are not unique, there are only finitely many for a given degree.

If $q$ has minimal horizontal (or vertical) foliation, the same will be true for $\hat{q}$. Furthermore, geodesics for $q$ will lift to geodesics for $\hat{q}$. The reason we consider good covers is the following (compare [Str84, Theorem 14.2.1]).

Lemma 2.4.

Suppose $q$ is a quadratic differential on $S$ and $(\hat{S},\hat{q})$ is a good cover. Then no two geodesic segments (for the flat metric defined by $\hat{q}$) bound a disk.

Proof.

This follows from the fact that the metric is locally CAT($0$), or alternately from the Gauß-Bonnet formula (for conical metrics, e.g. [Str84, Theorem 14.1]). ∎

This has the following useful corollary

Corollary 2.5.

Suppose that $\gamma_{1},\gamma_{2}$ are closed geodesics on a good cover $\hat{S}$. Let $\widetilde{S}$ be the universal cover of $\hat{S}$, and $\widetilde{\gamma}_{i}$ be lifts. Then the intersection

$$\widetilde{\gamma}_{1}\cap\widetilde{\gamma}_{2}$$

consists of a common geodesic segment.

Furthermore, singular flat geodesics between points in $\widetilde{S}$ are unique and realise distance.

Proof.

If the intersection were disconnected, we could find a bigon bounded by subsegments of the $\widetilde{\gamma}_{i}$. Similarly, if geodesics would not be unique, we could find a bigon bounded by two distinct geodesics joining the same endpoints. ∎

Note that on $S$ itself, these claims are false if $q$ has angle–$\pi$ singularities – geodesics can bound bigons containing angle-$\pi$-singularities – hence the need for good covers.

Corollary 2.6.

Let $\widetilde{S}$ be the universal cover of a good cover $\hat{S}$. Then the metric on $\widetilde{S}$ induced by the lifted quadratic differential is Gromov hyperbolic.

Suppose that $\gamma$ is a bi-infinite geodesic for this metric. Then $\gamma$ separates $\widetilde{S}$ into two disks, each of which is convex.

Proof.

Since $\hat{S}$ is a compact surface, the metric on $\widetilde{S}$ induced by the lifted quadratic differential is quasi-isometric to the hyperbolic metric on the disk.

The separation property is inherited from the corresponding fact for hyperbolic quasigeodesics. The convexity claim then follows from Corollary 2.5. ∎

These results in particular imply that leaf spaces of the horizontal and vertical foliations of $\widetilde{q}$ equipped with the transverse measures are real trees $T_{h},T_{v}$. We denote by

$$\pi_{h}:\widetilde{S}\to T_{h},\quad\pi_{v}:\widetilde{S}\to T_{v}$$

the corresponding projections. Since the trees are dual to the laminations, the diameter of the (vertical) projection $\pi_{v}(X)$ of the projection of a set $X$ measures the “horizontal width” (i.e. how much the set crosses the vertical leaves).

Strictly speaking, these projections depend on the choice of a good cover – we will always pick some good cover throughout, and the ambiguity will not be relevant.

2.4. Hyperbolic elements and quasi-invertibility

We recall the notion of coarse equivalence introduced by Bestvina-Fujiwara [BF02]. For this consider hyperbolic isometries $f,g$ of a general $\delta$-hyperbolic space $(X,d)$. These then have invariant quasi-axes given for example by taking the orbit of an arbitrary point say $A=(\alpha_{i}),A^{\prime}=(\beta_{j})$ of quality $L,L^{\prime}$ respectively. Then we say $f\sim g$ are quasi-equivalent if there is a (fixed) constant $B=B(L,L^{\prime},\delta)$ such that for any $D>0$ there is an isometry moving a segment of length $D$ in $A$ within a $B$-neighbourhood of $A^{\prime}$, respecting the direction of $A$ and $A^{\prime}$. One can check, using the hyperbolicity condition, that this definition is well-defined and indeed gives an equivalence relation.

Since quasi-geodesics determine points on the Gromov boundary, we can set $\xi_{\pm}=\lim_{i\to\pm\infty}\alpha_{i}$ and $\eta_{\pm}=\lim_{j\to\pm\infty}\beta_{j}$. Then this definition is equivalent to the following: There exist $g_{k}\in\textrm{Isom}(X,d)$ so that

$$\lim_{k\to\infty}g_{k}(\xi_{\pm})=\eta_{\pm}.$$

In the special case $f\sim f^{-1}$ we say that $f$ is quasi-invertible.

The following is consequence of the fact that the action of the mapping class group on the curve graph satisfies the WPD condition and can be deduced from [BF02, Propositions 6, 11]:

Theorem 2.7.

Let $G\subset Mcg(\Sigma)$ be a subgroup of the mapping class group of a finite type surface that is not virtually cyclic and contains a pseudo-Anosov element. Then $G$ contains non-quasi-invertible pseudo-Anosovs.

Most importantly for us is the following, which states that if one can exclude quasi-invertibility then one can construct quasi-morphisms and certify positive stable commutator length.

Theorem 2.8 ([BF02] Proposition 5).

Let $G\curvearrowright(X,d)$ is an action by isometries on a $\delta$-hyperbolic space. If $f\not\sim f^{-1}$ is a non-quasi-invertible isometry, then there is a quasi-morphism that is unbounded on the group generated by $f$. In particular, $f$ has positive stable commutator length.

3. The Target Lemma

In this section, we fix a singular flat structure $q$ on a surface $S$, a good cover $(\hat{S},\hat{q})$, and the corresponding lifted structure $\widetilde{q}$ on the universal cover of $\hat{S}$. We assume throughout that the vertical foliation of $q$ (and therefore $\hat{q}$) is minimal, and so vertical geodesic rays eventually visit every open set. The goal of this section is to prove a version of quantitative minimality: if $X$ is a set which is “horizontally wide, but vertically small”, then a vertical geodesic hits $X$ after a bounded time.

The following notions will make this precise:

Definition 3.1.

Let $\gamma$ be a (continuous) path on $S$, and let $\widetilde{\gamma}$ be a lift of $\gamma$ to the universal cover $\widetilde{S}$.

1. The total ($q$-)width (respectively total ($q$-)height) is the diameter of the projection $\pi_{v}(\widetilde{\gamma})\subset T_{v}$ in the vertical dual tree (respectively the diameter $\pi_{h}(\widetilde{\gamma})\subset T_{h}$ in the horizontal dual tree).

2. The ($q$-)size of $\gamma$ is the diameter of $\widetilde{\gamma}$ for the metric defined by $\widetilde{q}$.

Remark 3.2.

Given the good cover $(\hat{S},\hat{q})$, width and size are clearly well-defined since the deck group of the universal cover of $\hat{S}$ acts by isometries, preserving the horizontal and vertical foliations and their transverse measures.

Choosing a different good cover can only change these quantities by a multiplicative constant depending on the degree of the good cover. For us, the specific values are rarely relevant, so the explicit choice of good cover will not have any influence.

Remark 3.3.

Note that size and width are well-behaved with respect to smoothing, and tubular neighbourhoods. Namely, if $\gamma^{\prime}$ is contained in an $\varepsilon$–neighbourhood of $\gamma$ (with respect to the flat metric), then both total width and size of $\gamma^{\prime}$ are at most $2\varepsilon$ larger than the corresponding quantity for $\gamma$.

The following is a first instance of the quantitative minimality mentioned in the beginning of this section.

Lemma 3.4 (Fast Return Lemma).

Let $q$ be a singular flat structure whose vertical foliation is minimal. Then for any $B,\varepsilon>0$ there is a $L>0$ so that the following holds. Suppose $g\subset S$ is a geodesic of length at most $B$, and width at least $\varepsilon$. Then any vertical unit speed geodesic of length at least $L$ intersects $g$.

Proof.

This is a compactness argument: if such a constant does not exist, one could find a sequence of geodesics $g_{n}$ of length at most $B$ and width at least $\varepsilon$, and vertical geodesics $\sigma_{n}$ of length $n$ which are disjoint from $g_{n}$. Taking a (sub)limit, we would find a nontrivial geodesic $g_{\infty}$ which still has width at least $\varepsilon$, and a vertical (possibly singular) ray $\sigma_{\infty}$ which does not intersect $g_{\infty}$. This contradicts minimality of the vertical foliation. ∎

Lemma 3.5 (Target Lemma).

Let $q$ be a singular flat structure on $S$ with minimal vertical foliation. Then, for any $B,\varepsilon$ there is a number $L>0$ with the following property. Suppose that $\gamma:[0,1]\to S$ is a path so that the total width of $\gamma$ is at least $\varepsilon$ and the size of $\gamma$ is at most $B$.

Then any vertical geodesic for $q$ of length at least $L$ intersects $\mathrm{im}(\gamma)$.

Remark 3.6.

Observe that the Target Lemma 3.5 is (in some sense) optimal: by starting with any transverse arc of total width $\varepsilon$ and pushing part of it to follow the vertical foliation (pushing more subarcs out of the way if intersections would be created) we can generate an arc where some first return time is enormous, without changing the total width (but, making the size enormous at the same time).

It is likely that the target lemma could be proved by a contradiction argument directly, but we instead opt to reduce it to the Fast Return Lemma above.

Proof.

Denote by $(\hat{S},\hat{q})$ the chosen good cover of $(S,q)$. Since we defined width and size using the good cover, it suffices to show the conclusion of the lemma for $(\hat{S},\hat{q})$ itself.

In other words, we may assume that the foliations of $q$ have no angle-$\pi$ singulatities, so that Corollaries 2.5 and 2.6 apply.

Lift $\gamma$ to a path $\widetilde{\gamma}$ in the universal cover. Denote by $l_{0},l_{1}$ the (possibly singular) vertical leaves through the endpoints of $\widetilde{\gamma}$ which are limits of regular leaves through points of $\widetilde{\gamma}$. Note that these are disjoint bi-infinite geodesics, each of which separates the disk (Corollary 2.6).

By possibly replacing $\gamma$ with a subpath, we may assume that

1. $\widetilde{\gamma}$ intersects $l_{0},l_{1}$ only in its endpoints, and

2. the distance between $l_{0},l_{1}$ in $T_{v}$ is at least $\varepsilon$.

Denote by $\sigma$ the shortest geodesic joining $l_{0}$ and $l_{1}$ and observe that by (2) it therefore also has total width $\geq\varepsilon$ (note that it might be the case that any geodesic joining $l_{1},l_{0}$ is singular, and necessarily not just horizontal). On the other hand, note that the length of $\sigma$ is at most $B$, since $\widetilde{\gamma}$ also joins $l_{0},l_{1}$, and the latter is a diameter $B$ subset by assumption.

The Gromov hyperbolicity of the singular flat metric on $\widetilde{S}$ implies that some point on $\widetilde{\gamma}$ has distance at most $K_{1}=K_{1}(B,\varepsilon)$ from $\sigma$ (as any path outside a $k$–neighbourhood of the shortest connection between the two geodesics has diameter bounded from below by an exponential function of $k$). Thus, the distance between any point on $\widetilde{\gamma}$ and any point on $\sigma$ is at most $K_{2}=K_{1}+2B$.

Next note that for any point $p\in\sigma$, one of the vertical half-leaves through $p$ intersects $\widetilde{\gamma}$ (since both $\widetilde{\gamma}$ and $\sigma$ join the vertical geodesics $l_{0},l_{1}$). Since geodesics in $\widetilde{S}$ are unique and realise distance (Corollary 2.5), there is in fact a vertical segment of length at most $K_{2}$ starting in $p$ which intersects $\widetilde{\gamma}$. This shows that any vertical segment starting in a point of $p\in\sigma$ which extends length $K_{2}$ in both directions needs to intersect $\widetilde{\gamma}$.

Since $\sigma$ has length at most $B$ and width at least $\varepsilon$, the Fast Return Lemma 3.4 applies to $\sigma$, and guarantees that any vertical geodesic of length $\geq L$ intersects $\sigma$.

Combining the previous two observations, we conclude that any vertical geodesic of length at least $2L+2K_{2}$ intersects $\gamma$, proving the lemma. ∎

Remark 3.7.

There is a version of Lemma 3.5 (with essentially the same proof) for a measured geodesic lamination $\lambda$ on $S\setminus P$. Here, again, we can bound the return time of leaf segments of $\lambda$ to a transversal $I$, supposing lower bounds on the total transverse measure, and upper bounds on the diameter of a lift.

4. Constructing Boundary Points from Foliations

In this section we work with the nonseparating fine curve graph $\mathcal{NC}^{\dagger}(S)$, which is the full subgraph whose vertex set consists of non-separating simple closed curves. This is purely for convenience, and to be able to cite certain results from the literature. The main results will remain true for the full fine curve graph as well, as the two are quasi-isometric.

We aim to show that a minimal foliation (possibly with angle-$\pi$-singularities) naturally defines a unique Gromov boundary point of the fine curve graph. To construct the boundary point, we use the following construction, which may be of independent use later.

Lemma 4.1.

Let $P\subset S$ be a finite set. Suppose that $\alpha,\alpha^{\prime}$ are two nonseparating curves in minimal position on the punctured surface $S\setminus P$. Then the sequence of all bicorns formed by $\alpha,\alpha^{\prime}$ are a uniform (unparametrised) quasigeodesic in $\mathcal{NC}^{\dagger}(S)$.

Proof.

Curves on a bicorn path can be arranged to stay in minimal position with respect to each other (relative to $P$), and therefore their pairwise distances in the fine nonseparating curve graph are equal to the ones in the (usual) nonseparating curve graph [BHW22]. Now, the claim follows from Theorem 2.2. ∎

Given any set $L\subset S$, define $\mathcal{D}(L)$ to be the set of all curves disjoint from $L$:

$$\mathcal{D}(L)=\{\alpha\in\mathcal{NC}^{\dagger}(S):\alpha\cap L=\emptyset\}.$$

Lemma 4.2.

The set $\mathcal{D}(L)$ is $K$–quasiconvex for some constant $K$ independent of $L$.

Proof.

The set $\mathcal{D}(L)$ is clearly invariant under bicorn surgery. Hence, the result follows from Lemma 4.1. ∎

Proposition 4.3.

Let $S$ be a closed surface, and $P\subseteq S$ be a finite set of points. Consider a minimal singular foliation $\mathcal{F}$ whose angles are multiples of $\pi$, and so that angle-$\pi$-singularities only occur in $P$.

Then there is a unique point of the Gromov boundary $\xi_{\mathcal{F}}\in\partial_{\infty}\mathcal{NC}^{\dagger}(S)$ contained in every $\partial_{\infty}\mathcal{D}(L)$ where $L$ is a finite union of leaf segments of $F$.

In fact, if $L_{i}$ is a sequence of leaf segments of unboundedly increasing length, then any sequence $\alpha_{i}\in\mathcal{D}(L_{i})$ is an admissible sequence defining the boundary point $\xi_{\mathcal{F}}$.

Proof.

Let $\alpha_{0}$ be a basepoint in $\mathcal{NC}^{\dagger}(S)$ and consider an increasing sequence $L_{1}\subset L_{2}\subset\cdots$ of leaf segments with diverging length.

We first claim that the distance of $\mathcal{D}(L_{i})$ to the basepoint $\alpha_{0}$ diverges. Namely, suppose that $X\to S$ is any finite (unbranched!) cover. Since $\mathcal{F}$ is minimal, and $X$ is a finite cover, there is a length $n$ so that there is a lift $\widetilde{L}_{n}$ of $L_{n}$ which is filling with a lift $\widetilde{\alpha}_{0}$. This implies that the lift of every curve $\delta\in\mathcal{D}(L_{n})$ to $X$ intersects $\alpha_{0}$. By the covering criterion [BHM${}^{+}$22, Lemma 4.3] this proves the claim.

Now, take $\alpha_{i}$ to be a non-separating, simple curve disjoint from $L_{i}$. Since $\mathcal{D}(L_{i})$ is quasiconvex by Lemma 4.2, the distance of $\mathcal{D}(L)$ to a basepoint gives a coarse lower bound for the Gromov product of any two points in $\mathcal{D}(L)$. Hence, $(\alpha_{i})$ gives an admissible sequence converging to a boundary point $\xi$. By the same argument, a different choice of $\alpha^{\prime}_{i}$ gives an equivalent sequence, hence the same boundary point. This shows that $\xi_{\mathcal{F}}$ is the only Gromov boundary point contained in the boundary of all $\mathcal{D}(L_{i})$.

To finish the proof, it suffices to observe that if $L$ is any finite union of leaf segments, we can choose $\alpha_{i}$ to be disjoint from $L\cup L_{i}$. ∎

In fact, being disjoint from a long leaf segment is not necessary for convergence to the boundary point. Here, and below, we will need to choose a quadratic differential $q$ – but the choice will eventually not matter.

Corollary 4.4.

Suppose $\xi_{\mathcal{F}}$ is the boundary point defined by a minimal measured foliation $\mathcal{F}$, and let $q$ be a quadratic differential with vertical foliation $\mathcal{F}$.

Let $(\alpha_{i})$ be a sequence of curves with size diverging to infinity, and assume that for all $B,\varepsilon>0$ there is an $N=N(B,\varepsilon)$ so that any subsegment of $\alpha_{i},i>N$ of size $\leq B$ has width at most $\varepsilon$.

Then $\alpha_{i}$ converges to $\xi_{\mathcal{F}}$ in the Gromov boundary.

Proof.

Let $\mathcal{D}(\varepsilon,B)$ be the set of all curves with the property that $\beta\in\mathcal{D}(\varepsilon,B)$ if any size-$\leq B$-subsegment of $\beta$ has width at most $\varepsilon$.

Since there are only finitely many isotopy classes of simple closed curves of size at most $B$, and $\mathcal{F}$ is minimal, there is an $\varepsilon_{B}$ so that any curve of size at most $B$ has width at least $\varepsilon_{B}$. In particular, if $\varepsilon<\varepsilon_{B}$, all curves in $\mathcal{D}(\varepsilon,B)$ have size at least $B$.

Note that a size-$B$-segment with width at most $\varepsilon$ is contained in an $\varepsilon$–neighbourhood of a length-$B$ leaf segment of $\mathcal{F}$. Hence, arguing as in the first part of the proof of Proposition 4.3, we see that the distance of the sets $\mathcal{D}(\varepsilon,B)$ to a given basepoint in the fine curve graph diverges, as $\varepsilon\to 0,B\to\infty$.

Finally, suppose that $\beta_{1},\beta_{2}\in\mathcal{D}(\varepsilon,B)$, and $\gamma$ is a bicorn formed by $\beta_{1},\beta_{2}$. Any size-$\leq B$ subsegment of $\gamma$ is the union of at most two size-$\leq B$ subsegments of $\beta_{1},\beta_{2}$. Thus,

$$\gamma\in\mathcal{D}(B,2\varepsilon).$$

In particular, if $2\varepsilon<\varepsilon_{B}$, the bicorn itself has size at least $B$, and thus if $B$ is large enough, $\varepsilon$ is small enough, the Gromov product of $\beta_{1},\beta_{2}$ is very large.

On the other hand, given any vertical leaf segment $L$ and any $\varepsilon,B$, we can find a curve $\alpha\in\mathcal{D}(L)\cap\mathcal{D}(\varepsilon,B)$ by taking a very long vertical leaf segment disjoint from $L$ and closing with horizontal segment of size at most $\varepsilon$ (which is possible by minimality).

Together this shows that $\xi_{\mathcal{F}}$ is the unique Gromov boundary point contained in the boundary of all $\mathcal{D}(\varepsilon,B)$, showing the corollary. ∎

We next aim to describe the converse: what it means that a sequence of curves converges to one of those boundary points. From now on we also choose, once and for all, a basepoint $b$ in $\mathcal{NC}(S)$.

The key result is the following lemma. Both in the lemma, and its dependencies, we restrict to smooth curves for convenience (so we can easily talk about transversality). While not strictly necessary, this makes the arguments easier, and it is good enough to conclude the results we want to prove.

Lemma 4.5.

For any $\varepsilon,L$ there is a $K$ so that the following holds. If $\beta$ is a piecewise smooth simple closed curve and the Gromov product satisfies $(\beta\cdot\mathcal{F}_{v})_{b}>K$, then every segment of $\beta$ of size $L$ has total horizontal width at most $\varepsilon$, and thus $\varepsilon$–fellow travels a (possibly singular) leaf of $\mathcal{F}_{v}$.

To prove this, we need a few ingredients.

Lemma 4.6.

Suppose that $q$ is a singular flat structure on $S$. Let $\alpha_{0}$ be a smooth curve. Then for any $L>0$ there is an $R>0$ so that the following is true: if $\beta$ is a curve of size at most $L$, then the distance to $\alpha_{0}$ is at most $R$.

The idea of this proof is that for a curve $\beta$ of bounded size, we can remove bigons with $\alpha_{0}$ with uniformly few surgeries, and for a configuration without bigons bound the intersection number.

Figure 1. Removing bigons in the proof of Lemma 4.6

Proof.

First observe that we may assume that $\beta$ is also smooth and transverse to $\alpha_{0}$ (since modifying a curve in an embedded $\varepsilon$-neighbourhood can change size by at most $2\varepsilon$ and moves at most distance $2$ in the fine curve graph).

Now, let $b$ be a path which lifts $\beta$, and observe that it has diameter at most $L$. A ball of diameter $L$ can intersect at most $K$ distinct lifts of $\alpha_{0}$ (where $K$ depends only on $L$ and $\alpha_{0}$).

Let $a_{1},\ldots,a_{k}$ (with $k\leq K$) be the distinct lifts of $\alpha_{0}$ which $b$ intersects.

Suppose that there is some $a_{i}$ which is intersected more than once by $b$, and let $a_{i}$ be an outermost such. Let $b_{1},\ldots,b_{K}$ denote all subarcs of $b$ which lie on the outer side of $a_{i}$ (i.e. the side not containing the initial point of $b$).

Consider the images $\hat{b}_{1},\ldots,\hat{b}_{K}$ of the $b_{i}$ on the surface $S$. These define arcs of $\beta$ which all end on the same side of $\alpha_{0}$, and do not intersect $\alpha_{0}$ in their interior.

Thus, there is a disjoint union $B$ of (maximal) bigons based at $\alpha_{0}$ bounded by a subset of the $\hat{b}_{i}$ and subarcs of $\alpha_{0}$, so that all $\hat{b}_{i}$ are contained in $B$ (compare Figure 1).

Let $\alpha_{1}$ be the result of surgering $\alpha_{0}$ at $B$, and pushing slightly off $B$. We observe the following points:

1. The distance between $\alpha_{1}$ and $\alpha_{0}$ in the fine curve graph is at most $2$,

2. if a subarc of $\beta$ intersects $\alpha_{1}$, it also intersects $\alpha_{0}$ in the same points, and

3. the subarcs $\hat{b}_{1},\ldots,\hat{b}_{K}$ do not intersect $\alpha_{1}$.

As a consequence, the number of lifts of $\alpha_{1}$ which $b$ intersects is also at most $k$, and in fact the number of such lifts which are intersected more than once has decreased by at least one.

Hence, after repeating this process at most $k$ times, we find a curve $\alpha_{k}$ (of distance at most $2k$ to $\alpha_{0}$) with the property that $b$ intersects at most $k$ lifts of $\alpha_{k}$, and each in at most $1$ point.

Hence, $\beta$ and $\alpha_{k}$ intersect in at most $k$ points, and therefore the distance in the fine curve graph is at most $3k+1$. ∎

Lemma 4.7.

Assume that the flat structure has minimal vertical foliation. Suppose that $\beta_{0}$ is a smooth embedded segment which is not vertical. Then for some $p\in\beta_{0}$, the curve formed by a subarc of $\beta_{0}$ and a first return vertical flow-line starting in $p$ is a nonseparating curve.

Proof.

Denote by $l_{1},\ldots,l_{k}$ the isotopy classes of arcs formed by first-return vertical flow-lines to $\beta_{0}$. Since they are all disjoint, it is clear that there is only a finite number of them. Collapsing $\beta_{0}$ to a point the $l_{1},\ldots,l_{k}$ give an embedded graph on the surface (with a single vertex). By the minimality assumption on the vertical foliation all complementary regions must be discs. Hence we obtain a CW-structure on the surface with $1$-cells $e_{1},\ldots,e_{k}$ that are all loops. In particular, the $e_{k}$ generate the first homology of the surface. Since the latter is nontrivial, at least one of the $e_{k}$ is a nonseparating loop (as separating simple loops are nonhomologous).

∎

Now we are ready to prove the main result. The idea is that if the width of the curve $\beta$ would be large, then some vertical first return would be fast. This in turn would yield a bounded size bicorn beween the vertical foliation and $\beta$ – which violates the Gromov product bound.

Proof of Lemma 4.5.

If $\beta$ does not satisfy the fellow-travelling condition, then there is a subsegment $\beta_{0}\subset\beta$ so that the total horizontal width of $\beta_{0}$ is larger than $\varepsilon$ but whose size is bounded by $L$. We may further assume that $\beta$ is smooth (by replacing $\beta$ by a smooth curve in a very small neighbourhood).

Now, Lemma 3.5 implies that first-return times are bounded by some $B(L,\varepsilon)$. By Lemma 4.7, this implies that there is a first return flowline (of length at most $B$) defining a nonseparating bicorn, whose size is therefore bounded by $B+L$. Choosing a curve $C_{i}$ obtained by closing a very long horizontal segment with a short vertical segment, we may also assume that $C_{i},\beta$ define a nonseparating bicorn of size at most $B+L+1$. By Lemma 4.1, this implies that a quasigeodesic connecting $\beta$ to $C_{i}$ (for $i$ large) will contain a curve whose size is bounded by $B+L+1$. However, this yields an upper distance bound to $C_{0}$ by Lemma 4.6. Hence, for large $K$, this violates the Gromov product bound. ∎

With this lemma in hand, we can now prove our main result. Note that the last conclusion relies only on the foliation $\mathcal{F}_{v}$ itself, not the choice of $q$, since size of sets is coarsely equal for any two choices of metrics.

Proposition 4.8.

Suppose that a sequence of curves $\alpha_{i}$ converges to the boundary point $\xi_{\mathcal{F}_{v}}$ in the Gromov boundary. Then the size of the $\alpha_{i}$ converges to infinity, and they converge to leaf segments uniformly on bounded size sets: for any $\varepsilon,B$, any subinterval $J_{i}\subset\alpha_{i}$ of size at most $B$, any accumulation point of the $J_{i}$ in the Hausdorff topology is a leaf segment of $\mathcal{F}_{v}$.

Proof.

Note that (by replacing the $\alpha_{i}$ by curves in sufficiently small tubular neighbourhoods) it suffices to show the proposition for smooth curves $\alpha_{i}$.

By Lemma 4.6, the size of the $\alpha_{i}$ needs to diverge, since otherwise the distance to the basepoint would be bounded.

Since the $\alpha_{i}$ converge to $\xi_{\mathcal{F}_{v}}$, applying Lemma 4.5 we see that for any $\varepsilon,L$ and large enough $i$, any length–$L$ subsegment is contained in an $\varepsilon$–neighbourhood of a leaf.

This implies that any accumulation point of any segments (of bounded length) of the $\alpha_{i}$ are contained in leaf segments. ∎

Proposition 4.9.

Suppose that $F:S\to S$ is a homeomorphism which preserves the boundary point $\xi_{\mathcal{F}_{v}}$ defined by $\mathcal{F}_{v}$. Then $F$ preserves the foliation $\mathcal{F}_{v}$, i.e. $F$ maps every leaf segment of $\mathcal{F}_{v}$ to a leaf segment, and thus also every singularity of $\mathcal{F}_{v}$ to a singularity.

Proof.

First observe that there is a number $K>0$ so that the following holds: if $\sigma$ is any smooth segment of length at most $L$, then a lift of $F\sigma$ to the universal cover (of the good cover) has diameter at most $KL$. Namely, we can cover $S$ with a finite number $B_{1},\ldots,B_{k}$ of balls, so that the lift of any path $\sigma$ of length $\leq L$ can be covered with $cL$ lifts of the $B_{i}$ (for some constant $c$). Now, a lift of $F$ expands each lift of a ball to a set of bounded diameter, hence the claim.

To show the proposition, it suffices to show that any regular leaf segment $\sigma$ of $\mathcal{F}_{v}$ is mapped into a leaf of $\mathcal{F}_{v}$. To this end, choose a sequence of curves $C_{i}$ obtained by closing very long leaf segments of $\mathcal{F}_{v}$ containing $\sigma$ by short horizontal segments. The curves $C_{i}$ converge to $\xi_{\mathcal{F}_{v}}$ (in the sense that they are an admissible sequence defining this Gromov boundary point) by Corollary 4.4. Note that the image $F(\sigma)$ has size at most $K\,l(\sigma)$ and at least $l(\sigma)/K$ by the claim above.

On the other hand, the curves $F(C_{i})$ also converge (as vertices in $\mathcal{C}^{\dagger}(S)$) to $\mathcal{F}_{v}$, since $F$ preserves this boundary point by assumption. Hence, the Gromov product with $\mathcal{F}_{v}$ diverges. Applying Lemma 4.5 to any $\varepsilon$ and $L=Kl(\sigma)$ we see that eventually $F(\sigma)$ is contained in an $\varepsilon$–neighbourhood of a leaf. Since this is true for any $\varepsilon$, we obtain that $F(\sigma)$ is contained in a leaf as claimed. ∎

Using completely analogous arguments, we could also show

Proposition 4.10.

Suppose that $\lambda$ is a minimal filling geodesic lamination for some hyperbolic metric on $S\setminus P$, $P$ finite.

Then there is a point $\xi_{\lambda}\in\partial_{\infty}\mathcal{C}^{\dagger}(S)$ defined by $\lambda$. If a sequence of curves $\alpha_{i}$ converges to $\xi_{\lambda}$, then the $\alpha_{i}$ converge to leaf segments uniformly on bounded size sets.

If a homeomorphism $F$ fixes $\xi_{\lambda}$ (as a boundary point), then $F$ preserves $\lambda$ (as a lamination).

Proof.

We only briefly indicate which arguments need to be adapted. We choose a transverse measure on $\lambda$ to define width, and use the hyperbolic metric to define size. There is no need to choose a good cover. To define the boundary point, we argue exactly as in Proposition 4.3, with $L$ being leaf segments of the lamination. The proof of Corollary 4.4 also works exactly the same (closing long leaf segments of $\lambda$ with short geodesic segments, instead of horizontal segments). Lemma 4.7 is true, replacing the vertical flow by the geodesic flow along the lamination $\lambda$. Finally the proofs of Lemma 4.5 and Proposition 4.9 go through mutatis mutandis. ∎

We emphasise that the stabilisers of $\xi_{\lambda}$ for a lamination are very large – any homeomorphism supported on a complementary region of $\lambda$ will fix $\xi_{\lambda}$.

5. Applications to pseudo-Anosovs

In this section, we collect some applications of the previous results to the action of point-pushing pseudo-Anosovs on the fine curve graph.

Throughout, let $\Psi$ be a point-pushing pseudo-Anosov relative to a finite set of points $P$, which is not quasi-invertible as a mapping class of $S\setminus P$ in the sense of Bestvina-Fujiwara. Note that many such pseudo-Anosovs exist (cf. Theorem 2.7). We denote by $F$ the Thurston representative of $\Psi$, and by $\mathcal{F}_{h},\mathcal{F}_{v}$ the horizontal and vertical foliations (which then determine the fixed points at infinity for the action of $F$ on the fine curve graph).

First, we observe the following corollary of Proposition 4.9.

Corollary 5.1.

Any homeomorphism which commutes with $F$ (or conjugates $F$ to a positive power of itself) preserves the foliations $\mathcal{F}_{h},\mathcal{F}_{v}$ (as foliations on the surface). In particular, it preserves the set $P$ of angle–$\pi$–singularities.

If the induced mapping class on $S\setminus P$ is trivial, then $F$ is the identity. In particular, if the centraliser of $\Psi$ is cyclic, then the centraliser of $F$ is cyclic as well.

Proof.

The first claim is immediate from Proposition 4.9. To see the second claim, consider the universal cover $\mathbb{H}^{2}$ of $S\setminus P$, and a lift $\widetilde{F}$ of $F$. Since $F$ preserves $\mathcal{F}_{h},\mathcal{F}_{v}$, the lift $\widetilde{F}$ preserves lifts of these foliations.

Then, since $F$ defines the trivial mapping class of $S\setminus P$, the lift $\widetilde{F}$ acts trivially on the boundary at infinity of the hyperbolic plane. Note that no two regular leaves of lifts of, say, $\mathcal{F}_{h}$ can share endpoints, as otherwise there would be a foliated strip with one isolated leaf contradicting minimality. This implies that $\widetilde{F}$ actually fixes each individual leaf of $\widetilde{\mathcal{F}}_{h},\widetilde{\mathcal{F}}_{v}$ setwise. A dense set of points $p\in\mathbb{H}^{2}$ has the property that there is exactly one regular leaf of each of $\widetilde{\mathcal{F}}_{h},\widetilde{\mathcal{F}}_{v}$ through $p$. Hence, $\widetilde{F}$ needs to fix all those $p$, and therefore is the identity homeomorphism. ∎

We next come to one of our main results, which gives the first explicit examples of elements in $\mathrm{Homeo}(S)$ that have positive stable commutator length. Moreover, this even holds if one considers commutators in the full group of orientation preserving homeomorphisms, generalizing our results in [BHW22].

Theorem 5.2.

Suppose that $\psi\in\mathrm{Mcg}(S\setminus P)$ is a pseudo-Anosov surface braid (multi-point-push), which is not conjugate to its inverse.

Then the Thurston representative $F$ of $\psi$ has positive scl in $\mathrm{Homeo}_{+}(S)$.

To prove this theorem, we will show that a quasi-axis cannot be quasi-inverted. By the main result of Bestvina-Fujiwara [BF02] this will imply the proposition (cf. Section 2.4). The main technical work lies in the following lemma.

Lemma 5.3.

Let $\mathcal{F}_{h},\mathcal{F}_{v}$ be minimal singular foliations on $S$ with angle-$\pi$ singularities in a finite set $P$. For any $\varepsilon>0$ there is a $K>0$ so that the following holds. If $f$ is a homeomorphism so that

$$(f(\mathcal{F}_{v})\cdot\mathcal{F}_{h})_{b}>K,\quad(f(\mathcal{F}_{h})\cdot\mathcal{F}_{v})_{b}>K$$

then $d(f(p),p)<\varepsilon$ for any $p\in P$ (for the singular flat metric defined by the foliations).

Figure 2. The “bells” in the proof of Lemma 5.3. The solid black foliation is the vertical one, bending around the angle–$\pi$ singularity; the horizontal foliation is drawn dashed. In red, the image of the horizontal foliation under the homeomorphism is drawn. It cannot cross the vertical foliation too much (without bounding the Gromov product), and so it needs to exit at the bottom. But then, the (blue) image of the vertical foliation will have to also exit at the bottom, crossing the horizontal (dashed) foliation.

Proof.

We argue the contrapositive. Namely, we will construct a neighbourhood $V=V_{P}$ of the angle-$\pi$ singularities, and show that if the Hausdorff distance $d(f(P),V)\geq\varepsilon$, then there is a $K=K(\varepsilon)>0$ so that one of the claimed inequalities fails. To do so, let $q$ be the singular flat structure corresponding to the two foliations, and choose some good cover.

The subset $V$ will consists of small neighbourhoods of each of the points $p_{i}$ of the puncture set and each of these will in turn consist of the union of two “bells” $U_{i},U_{i}^{\prime}$. We begin by describing one of them. Denote by $\nu$ the vertical leaf segment of length $\varepsilon$ starting at a point $p_{i}$ in the puncture set $P$, and by $\eta$ the horizontal leaf extending length $\varepsilon/2$ in both directions through the regular endpoint of $\nu$. We let $U_{i}$ be the region swept out by the vertical arcs starting and ending on $\eta_{i}$ and bounding a disk containing $p_{i}$ together with a subinterval of $\eta_{i}$ (see Figure 2). The boundary of $U_{i}$ consists of the base arc $\eta_{i}$, and a top arc. We define the other bell $U^{\prime}_{i}$ analogously, reversing the roles of vertical and horizontal.

The strategy of the proof is as follows: we will argue that leaves of the image of the vertical foliation $f(\mathcal{F}_{v})$ need to leave the bell $U_{i}$ through the base, and similarly leaves of $f(\mathcal{F}_{h})$ leave $U^{\prime}_{i}$ through its base. This will lead to a contradiction, as there are two such leaves that cross transversely once in the union of the bells.

We now give more details. If, for some point $p\in P$, we have that $f(p)\notin V$, then (by a counting argument), $f(P)$ is disjoint from some component $U_{i}\cup U^{\prime}_{i}$ of $V$.

Now observe that by construction, any segment $\sigma$ contained in $U_{i}$ has size at most $2\varepsilon$. If such a segment $\sigma$ joins $p_{i}$ to the top arc, then $\sigma$ has total width $\varepsilon$. Consider now the foliation $f(\mathcal{F}_{h})\cap U_{i}$ on $U_{i}$, and observe that it does not have an angle-$\pi$-singularity. In particular, any leaf segment can be continued until it leaves the bell $U_{i}$ (since the only way leaf segments in a singular foliation cannot be continued is if they reach an angle-$\pi$ singularity, and leaves of minimal foliations eventually leave every disk).

Let $\sigma$ be any maximal leaf segment of $f(\mathcal{F}_{h})\cap U_{i}$ through $p_{i}$. If $\sigma$ would have an endpoint on the top arc of $U_{i}$, it would be an arc of total width $\geq\varepsilon$ and size at most $2\varepsilon$. Let $\alpha_{i}$ be a sequence of curves formed by closing long horizontal leaf segments containing $f^{-1}(\sigma)$, which defines the boundary point $\mathcal{F}_{h}$ (as in Corollary 4.4). Then for all large $i$, the curves $f(\alpha_{i})$ contain $\sigma$ and, by Lemma 4.5, this yields an upper bound for the Gromov product $(f(\alpha_{i})\cdot\mathcal{F}_{v})_{b}<K(\varepsilon)$. By definition, we then also get $(f(\mathcal{F}_{h}))\cdot\mathcal{F}_{v})_{b}<B(\varepsilon)$. By choosing $K$ large enough, this is impossible. Hence, $\sigma$ leaves $U_{i}$ through the base.

An analogous argument with horizontal and vertical exchanged shows that any maximal leaf segment $\tau$ of $f(\mathcal{F}_{v})\cap U^{\prime}_{i}$ through $p$ leaves $U^{\prime}_{i}$ through its base.

But, since we assume $f(P)\cap V_{i}=\emptyset$, we can choose leaf segments $\sigma,\tau$ as above, which only intersect in $p_{i}$ and cross there transversely – since $f(\mathcal{F}_{v})\cap(U^{\prime}_{i}\cup U_{i})$ and $f(\mathcal{F}_{h})\cap(U^{\prime}_{i}\cup U_{i})$ are singular foliations on a disk without angle-$\pi$ singularities, and therefore leaf segments do not bound bigons.

This contradicts the fact that both $\sigma,\tau$ leave their bells through the bottom since any embedded arc disconnects a disk (compare Figure 2). ∎

We are now ready to prove the theorem.

Proof of Theorem 5.2.

Let $C_{0}$ be any smooth non-separating simple closed curve disjoint from the punctures $P\subset S$ of the pseudo-Anosov map $F$ and set $C_{i}=F^{i}(C_{0})$. Then $(C_{i})$ is a quasi-geodesic axis of quality $L$ for the action of $F$ on $\mathcal{NC}^{\dagger}(S)$ and considering isotopy classes $([C_{i}])$ we similarly have a quasi-geodesic axis for the action of the mapping class on $\mathcal{NC}(S\setminus P)$ of quality $L^{\prime}\geq L$.

Suppose that there is a sequence $g_{k}$ of homeomorphisms so that as points in $\partial_{\infty}\mathcal{NC}^{\dagger}(S)$

$$g_{k}\mathcal{F}_{h}\to\mathcal{F}_{v},g_{k}\mathcal{F}_{v}\to\mathcal{F}_{h}.$$

By hyperbolicity of $\mathcal{NC}^{\dagger}(S)$, the following holds: for any $N$ and any $k$ large enough, we have that if $i\geq N^{\prime}(N)>N$, then $g_{k}(C_{i})$ is within distance $B$ of $C_{-j}$ for some $j>N$ and, analogously, $g(C_{-i})$ is within distance $B$ of $C_{j}$ for some $j>N$. Here, $B$ only depends on the hyperbolicity constant and the quality $L$ of the quasigeodesic.

The quasi-axis $([C_{i}])$ in $\mathcal{C}(S\setminus P)$ is not quasi-invertible by our assumption on $\Psi$. In particular, there are $N,K$ so that there is no mapping class $\varphi$ of $S\setminus P$ so that

$$(\varphi C_{i}\cdot C_{-j})_{b}>K-B,\quad(\varphi C_{-i}\cdot C_{j})_{b}>K-B.$$

for all $i,j\geq N$

Now, let $\varepsilon$ so that the $2\varepsilon$–neighbourhood of the puncture set $P$ is disjoint from $C_{i}$ for all $|i|\leq N,N^{\prime}$. Then by Lemma 5.3, for $k$ large enough, we have that $g_{k}(P)$ is within $\varepsilon$ of $P$, and also $g_{k}(C_{N^{\prime}}),g_{k}(C_{-N^{\prime}})$ are within distance $B$ of $C_{-i},C_{j}$ for $i,j>N$.

Now, we can change $g_{i}$ by an isotopy supported in a $2\varepsilon$-ball about $P$ to obtain a homeomorphism $g^{\prime}$ with $g^{\prime}(C_{N^{\prime}})=g_{i}(C_{N^{\prime}}),g^{\prime}(C_{-N^{\prime}})=g_{i}(C_{-N^{\prime}})$ which preserves $P$. Hence, $g^{\prime}$ defines a mapping class $\varphi$ which violates the claim above. ∎

Remark 5.4.

Suppose $F$ is a Thurston representative of a point pushing pseudo-Anosov $\psi$ of a single point $p\in S$. Then $F$ has positive stable commutator length in $\mathrm{Homeo}_{0}(S)$ without any further assumptions.

Namely, in the above proof, the $g_{i}$ are then isotopic to the identity, and therefore so is $g^{\prime}$. Hence, $g^{\prime}$ is a representative of a point push (of the same point $p$).

Applying [BF02] to the action of the point-pushing group $\pi_{1}(S,p)$ on the curve graph $\mathcal{C}(S-p)$ this would imply that $\psi$ is conjugate to its inverse in $\pi_{1}(S,p)$. However, this is impossible since surface groups are bi-orderable.

Finally, we prove a kind of Tits alternative which generalizes [HX21, Theorem 1.10] to all regularities and genera.

Theorem 5.5.

Suppose that $G<\mathrm{Homeo}_{+}(S)$ is a group which contains a Thurston representative of a pseudo-Anosov element of $\mathrm{Mcg}(S\setminus P)$ for some finite $P$.

Then either $G$ contains a free group, or $G$ has an index $2$ subgroup which fixes a foliation on $S$.

Proof.

Let $F$ be the Thurston representative, and let $\xi_{+},\xi_{-}$ be the boundary points of $\mathcal{C}^{\dagger}(S)$ defined by the stable and unstable foliations. There are three cases. If $G$ preserves the set $\{\xi_{+},\xi_{-}\}$, then there is an index $2$ subgroup which fixes both, and we can apply Proposition 4.9. If $\xi_{+}$ or $\xi_{-}$ are fixed by $G$, we are also done. Otherwise, there is a conjugate of $F$ in $G$ which has fixed points that are both distinct from $\xi_{-},\xi_{+}$. But then, large enough powers of $F$ and this conjugate generate a free group by the ping pong lemma for actions on hyperbolic spaces. ∎

6. Stabilisers of Nonorientable Foliations

In this section we put restriction on possible fixed points of homeomorphisms acting parabolically. This answers a question posed to us by Yair Minsky. In particular, we have the following

Theorem 6.1.

Suppose that $F$ is a foliation on $S$ with angle-$\pi$ singularities at a finite set $P$, which is ending as a foliation on $S\setminus P$. Then the corresponding Gromov boundary point $\xi_{F}$ is not the fixed point of any homeomorphism acting parabolically.

If $\xi_{F}$ is the fixed point of a homeomorphism acting loxodromically, then the homeomorphism fixes $P$ and defines a pseudo-Anosov mapping class on $S\setminus P$ so that $F$ is one of its invariant foliations.

Proof.

We first show the statement about parabolics, and argue by contradiction. Suppose that $\phi$ is a homeomorphism fixing $\xi_{F}$, and suppose that $\phi$ acts as a parabolic isometry on $\mathcal{C}^{\dagger}(S)$. By Proposition 4.9, $\phi$ then preserves the foliation $F$, and in particular the finite set $P$ of angle-$\pi$ singularities.

The induced mapping class $[\phi]\in\mathrm{Mcg}(S\setminus P)$ fixes the isotopy class $[F]$ of the foliation $F$ defined on $S\setminus P$. The mapping class $[\phi]$ is therefore of finite order: no infinite order mapping class can fix an ending foliation at all, and if $[\phi]$ were pseudo-Anosov, then $F$ would act on $\mathcal{C}^{\dagger}(S)$ as a hyperbolic isometry.

Hence, up to replacing $\phi$ by a power (which doesn’t affect if it acts as a parabolic) we may assume that $[\phi]$ is the trivial mapping class. In particular, lifting to the universal cover $\widetilde{X}$ of $S\setminus P$, a lift $\widetilde{\phi}$ of $\phi$ maps each leaf of $\widetilde{F}$ to itself, and thus the same is true for $\phi$ and the leaves of $F$.

Now, consider any point $p\in S$ and a regular half-leaf $\rho:[0,\infty)\to S$ starting in $p=\rho(0)$, so that $\phi(p)=\rho(t)$ for some $t>0$. By minimality and non-orientability of $F$, the half-leaf $\rho$ comes arbitrarily close to $p$ again with opposite orientation at times $t_{i}$, with $\rho(t_{i})\to p$. By continuity of $\phi$ and the fact that any leaf is preserved, this means that there is a time $t_{i}$ so that $\phi(\rho(t_{i}))=\rho(s),s<t_{i}$. Thus, $\phi(\rho[0,t_{i}])\subset\rho[0,t_{i}]$.

As a consequence, we see that through any point $\widetilde{p}$ of any regular leaf of $\widetilde{F}$, there is an embedded leaf segment $I$ which is mapped into itself by any power of $\widetilde{\phi}$. Hence, the images $\widetilde{\phi}^{n}(K)$ of a fundamental domain $K$ are contained in a compact set $C$ independent of $n$.

This implies that, given any curve $\beta$, the size of $\phi^{n}(\beta)$ is bounded – which, by Lemma 4.6, implies that $\phi$ is elliptic.

The claim on loxodromics follows in an analogous way – if $\phi$ acts hyperbolically, but $[\phi]$ is not a pseudo-Anosov, then again $\phi$ would fix every leaf setwise, allowing for the same contradiction. ∎

We remark that in the other direction, if one begins with an orientable foliation, then this is the fixed point of a parabolic element given by flowing along leaves as described in [BHM${}^{+}$22].

7. Lifting under finite covers and Quasi-morphisms

In this section we consider connected surfaces $\Sigma$ with non-empty boundary of negative Euler characteristic (possibly non-orientable). Then as $\Sigma$ is hyperbolic by Hamstrom [Ham66] the identity component $\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)$ of the groups of homeomorphisms fixing the boundary (pointwise) is contractible. Now consider any finite cover $\overline{\Sigma}\to\Sigma$ which is trivial when restricted to some boundary component. This assumption gives a well defined lifting map and composing this with the map induced by collapsing boundary components of $\overline{\Sigma}$ we obtain

$$\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)\stackrel{{\scriptstyle\textrm{lift}}}{{\longrightarrow}}\operatorname{Homeo}_{0}(\overline{\Sigma},\partial\overline{\Sigma})\stackrel{{\scriptstyle\textrm{collapse}}}{{\longrightarrow}}\operatorname{Homeo}_{0}(\widehat{\Sigma},P_{\partial}).$$

Using this map and the fact that (modified) Thurston representatives on $\Sigma$ map to Thurston representatives on $\widehat{\Sigma}$ one can deduce the following:

Theorem 7.1.

Let $\Sigma$ be any hyperbolic surface (possibly non-orientable) but with non-empty boundary. Then the group $\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)$ contains elements with $\mathrm{scl}>0$.

Proof.

First consider a Thurston representative $\varphi_{T}$ of a point pushing pseudo-Anosov of $\Sigma$ relative to some (large) puncture set $P\in int(\Sigma)$ so that the mapping class becomes trivial after filling in the punctures $P$. This then maps to the Thurston representative on the blow down $\widehat{\Sigma}$ of the cover surface $\overline{\Sigma}$ under the map above. Choosing $\overline{\Sigma}$ appropriately, we can assume that the closed surface $\widehat{\Sigma}$ is hyperbolic. Hence, the resulting map $\widehat{\varphi}_{T}$ is again a Thurston representative of some point-pushing map. Choosing $\varphi_{T}$ correctly (using Theorem 2.7 and Theorem 5.2) we can thus ensure that it has positive scl in $\operatorname{Homeo}(\widehat{\Sigma})$. It follows that $\varphi_{T}$ also has positive scl in $\operatorname{Homeo}_{0}(\Sigma)$.

However, $\varphi_{T}$ does not fix the the boundary pointwise, and therefore does not lie in $\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)$. To address this, we argue as follows.

Under our assumption that the map is trivial in the mapping class group of $\Sigma$, the homeomorphism $\varphi_{T}$ has fixed points on each boundary component (prongs of the stable/unstable foliations of the pseudo-Anosov map). We then attach a map of annuli $A$ that correspond to an isotopy of the boundary map the the identity (fixing the prongs) to obtain a ‘fattened’ version of the Thurston representative $F^{\mathrm{fat}}_{T}\in\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)$. This homeomorphism is still isotopic to the identity relative to the boundary. Conjugating $F^{\mathrm{fat}}_{T}$ by maps “compressing the annulus” $A$ to a an annulus $A_{\varepsilon}$ of width $\varepsilon$, we obtain a family of homeomorphisms $F^{\varepsilon}_{T}$ which are all conjugate to each other in $\operatorname{Homeo}_{0}(\Sigma,\partial\Sigma)$. Outside the annulus $A_{\varepsilon}$, the map $F^{\varepsilon}_{T}$ agrees with $F_{T}$.

The maps $F^{\varepsilon}_{T}$ do not converge as $\varepsilon\to 0$ on $\Sigma$, but their images in $\operatorname{Homeo}_{0}(\widehat{{\Sigma}},P_{\partial})$ under the lifting and collapsing maps above do converge to the Thurston representative $\widehat{\varphi}_{T}$ on $\widehat{{\Sigma}}$ (since the annuli $A_{\varepsilon}$ lift and collapse to disks of radius $\varepsilon$). Since stable commutator length is continuous (cf. [BHW22, Theorem 1.5]), this implies that for small $\varepsilon$ the image of $F^{\varepsilon}_{T}$ has positive stable commutator length (and, since they are all conjugate, in fact all of them do). ∎

This result is sharp for orientable surfaces as we know that in the case of the annulus and the disc the groups do not admit quasi-morphisms [BIP08] and it was shown in [BHW22] that if $\Sigma=T^{2}$ is a torus then there are quasi-morphisms. The remaining cases are then the Klein bottle, the Möbius band and the real projective plane.