Geodesic intersections and isoxial Fuchsian groups

Choose hyperbolic elements a,b \in \Gamma such that the axis of a (resp. b) is a lift of \alpha (resp \beta) to \mathbb{H} and so that the axes meet in a lift \hat{z} \in \mathbb{H} of z. Since \text{Comm}(\Gamma) is dense in SL(2,\mathbb R), there is some element g \in \text{Comm}(\Gamma) so that g(\hat{z}) \in \hat{U} for some lift of U to \mathbb{H}. By the commensurability of the groups \Gamma and g\Gamma g^{-1} there is a positive integer m such that (gag^{-1})^m,( gbg^{-1})^m \in \Gamma so that the axes of these elements project to closed geodesics \alpha_u, \beta_u on \Sigma meeting in a point z_u as required. \Box

In one direction, if \alpha is simple then \ell_\alpha is one of the Fenchel-Nielsen coordinates for some pants decomposition of \Sigma so there is some (non convergent) sequence \rho_n \in \mathcal{T}(\Sigma) such that \ell_\alpha \rightarrow 0.

Now suppose that \alpha is not simple and we seek a lower bound for its length.

There are two cases depending on whether there exists a closed simple geodesic \beta disjoint from \alpha or not. If there is no such geodesic then \alpha meets every simple closed geodesic \beta \subset \Sigma and it is cusomary to call such a curve a filling curve. Choose \epsilon > 0 and consider the decomposition of the Teichmeuller space into the \epsilon-thin part and its complement. On the thick part \ell_\alpha \geq \epsilon whilst on the thin part, by the inequality (?), it is bounded below by \mathrm{arcsinh} (1/ \sinh(\epsilon/2) ).

If there is an essential simple closed geodesic disjoint from \alpha then we cut along this curve to obtain a possibly disconnected surface with geodesic boundary. We repeat this process to construct a compact surface C(\alpha) such that \alpha is a filling curve in C(\alpha). By construction C(\alpha) embeds isometrically as a subsurface of \Sigma and since \alpha is not simple C(\alpha) is not an annulus. On the other hand, by taking the Nielsen extension of C(\alpha) then capping off with a punctured disc we obtain a conformal embedding C(\alpha) \hookrightarrow C(\alpha)^* where C(\alpha)^* is a punctured surface with a natural Poincaré metric. By the Ahlfors-Pick-Schwarz Lemma there is a contraction between the metrics induced on C(\alpha) from the metric on \Sigma and from the Poincaré metric on C(\alpha)^*. A consequence of this is that the geodesic in the homotopy class determined by \alpha on C(\alpha) is longer than the one in C(\alpha)^*. So, to bound \ell_\alpha it suffices to bound the length of every filling curve on a punctured surface. There are two cases.

• If C(\alpha) has an essential simple closed curve then we have already treated this case above.

• If C(\alpha) has no essential simple closed curves then it is a 3 punctured sphere an the bound is trivial since the Teichmueller space consists of a point.

\Box

It is standard from the theory of negatively curved groups that the lift admits a unique extension to \mathbb{H} \sqcup \partial \mathbb{H}, continuous on the boundary \partial \mathbb{H}, since \mathbb{H}/ \Gamma is compact and so the restriction of the lift to the set of lifts of a base point x \in \Sigma, \Gamma.\{ \hat{x}\} is Lipschitz.

Since the extension is continuous, to prove (1) it suffices to note that the lift of the Fenchel-Nielsen deformation fixes the endpoints of the edges of \partial \overline{P} and these are dense in \Lambda^P.

For (2) let w \in \partial \mathbb{H} and suppose that it is not a point of \partial \overline{P}. Then there is an edge \hat{\alpha} of \partial \overline{P} such that w is a point of the interval determined by the endpoints of this geodesic. It is easy to check using our recursive description of the action of \hat{\tau}_\alpha^t on \mathbb{H} that w converges to the appropriate endpoint of \hat{\alpha}. \Box

Let \hat{\alpha} be a lift of \alpha which separates \hat{\gamma}_1,\hat{\gamma}_2 \subset \mathbb{H}. Let P_1 and P_2 be the pair of complementary regions which have \hat{\alpha} as a common edge and we label these so that \hat{\gamma}_i is on the same side of \hat{\alpha} as P_i for i=1,2. We choose the lift of the base point to be in P_1 and lift the Fenchel-Nielsen deformation.

First consider the orbit \hat{\tau}_\alpha^t(y) of an ideal endpoint y of \hat{\gamma}_2 as t \rightarrow \infty. Since x \in P_1, the region P_2 gets translated and so, for any side \beta of P_2, the sequence \hat{\tau}_\alpha^t(\beta) converges to the endpoint \hat{\alpha}^+. Now there is a pair of edges \beta_1, \beta_2 such that the endpoints of \hat{\gamma}_2 are contained in the closed interval containing the endpoints of \beta_1, \beta_2. Since each of the \beta_i converge to \hat{\alpha}^+ under the deformation it is easy to see that \hat{\tau}_\alpha^t(\hat{\gamma}_2) must converge to \hat{\alpha}^+ too.

Now consider the orbit of an endpoint y of \hat{\gamma}_1 under the deformation. It suffices to show that, under this deformation, y does not converge to \hat{\alpha}^+. There are two cases according to whether or not y belongs to the limit set \Lambda^{P_1} of the subgroup of \Gamma which stabilises P_1.

1. If y \in \Lambda^{P_1} then it is invariant under the Fenchel-Nielsen deformation.

2. If y \not \in \Lambda^{P_1} then it limits to a point in y_\infty \in \Lambda^{P_1} which is an endpoint of one of the edges of P_1. By hypothesis \hat{\gamma}_1 does not meet \hat{\alpha} and so y_\infty is not \hat{\alpha}^+.

\Box

With the notation above we have the following expression for the angle:

This equation is actually implicit in [pjm] but it is not claimed to be new there and seems to have been well known. The left hand side of (Proof) is clearly an analytic function on \mathcal{T}(\Sigma) and it follows from elementary real analysis the the angle varies real analytically too. \Box

With the notation of subsection ?, there is a convex region P in \mathbb{H} bounded by lifts of \alpha as before. Let \hat{\alpha} be an edge of \partial \overline{P}, and choose a corresponding lift \hat{\beta} which intersects \hat{\alpha}. There is an element of the covering group g \in \Gamma such that

We lift the Fenchel-Nielsen deformation and consider, as before, its extension

Now, arguing as in Lemma 4.2.1, we see that;

• the endpoints of \hat{\alpha} are fixed by \hat{\tau}_\alpha^t,

• the endpoint of \hat{\beta} on the same side of \hat{\alpha} as P converges to a point z \neq \alpha^+ as t \rightarrow -\infty,

• the other endpoint of \hat{\beta} converges to \alpha^+ as t \rightarrow -\infty

It follows that, after possibly changing the orientation of \beta, that the angle between \hat{\alpha} and \hat{\beta}, and hence \alpha \angle_x \beta, tends to 0. Likewise, as t \rightarrow +\infty the angle between \hat{\alpha} and \hat{\beta}, and hence \alpha \angle_x \beta, tends to \pi.

Thus, by continuity, the range of the angle function is ]0,\pi[.

\Box

If \alpha and \beta_1\cup \beta_2 are disjoint then \beta_1\angle_y\beta_2 is constant along the \hat{\tau}_\alpha^t-orbit so the result is trivial.

Suppose now that \alpha and \beta_1\cup \beta_2 are not disjoint and choose x as a basepoint of \Sigma. Then, with the notation of paragraph ?, there is a convex region P in \mathbb{H} bounded by lifts of \alpha. We now consider three cases according to the number of edges of \partial \overline{P} that \hat{\beta_1} \cup \hat{\beta_2} meets.

We first deal with the simplest case. Suppose that \hat{\beta_1} \cup \hat{\beta_2} meets \partial \overline{P} in four distinct edges denoted C_1,C_2,C_3,C_4 \subset \mathbb{H}, and, after possibly relabelling these, \hat{\beta_1} meets C_1,C_2 whilst \hat{\beta_2} meets C_3,C_4 as in Figure ?. Now we deform \rho_0 by a Fenchel-Nielsen twist along \alpha to obtain a 1-parameter family of \rho_t \in \mathcal{T}(\Sigma), t\in \mathbb R. As we have seen above, under such a deformation the length of \alpha does not change nor does the geometry of \partial \overline{P} in particular the positions of the C_i remain unchanged. From our discussion of the \hat{\tau}_\alpha^t and its extension to \mathbb{H} \sqcup \partial \mathbb{H} it is clear that, \forall t \in \mathbb R, \hat{ \hat{\tau}_\alpha^t(\beta_1)} meets C_1,C_2 whilst \hat{ \hat{\tau}_\alpha^t(\beta_2)} meets C_3,C_4. Thus, if the diameters of the circles were small, the angle at \hat{z} cannot not vary much from its value at \rho_0 since the radii of the circles are small. More generally, we can bound the size of the angle using the cross ratio formula. Labeling the endpoints as in Figure ? one has:

Note first that each of the four points lies on the unit circle and so that its diameter, that is 2, is a trivial upper bound for each of the four distances appearing on the left hand side of this equation. Now under the deformation each of the endpoints \hat{\tau}_\alpha^t(\beta_i^\pm) stays in one of four disjoint euclidean discs defined by one of the C_j. In particular, there exists \delta_4 > 0 such that for all t \in \mathbb R

and this is sufficient to obtain bounds on the cross ratio:

If \hat{\beta_1} \cup \hat{\beta_2} meets \partial \overline{P} in just two edges, C_1,C_2 \subset \mathbb{H} say. Although we no longer have a uniform lower bound for | \hat{\tau}_\alpha^t(\beta_1^\pm) - \hat{\tau}_\alpha^t(\beta_2^\pm) | in this case, there still exists \delta_2 > 0 such that for all t \in \mathbb R,

Thus, for all t \in \mathbb R,

Finally, if \hat{\beta_1} \cup \hat{\beta_2} meets \partial P in exactly three edges then it is easy to see that, using the same reasoning as for the two edge case, there is \delta_3 such that

\Box