Geometriae Dedicata 102: 151177, 2003.

# 2003 Kluwer Academic Publishers. Printed in the Netherlands.

151

Equivalent Curves in Surfaces

CHRISTOPHER J. LEININGER
Department of Mathematics, Barnard College at Columbia University, 2990 Broadway
MC 4448, New York, NY 10027-6902. e-mail: [email protected]
(Received: 9 September 2002; accepted in nal form: 24 January 2003)
Abstract. We consider various notions of equivalence for homotopy classes of curves on
hyperbolic surfaces based on topological, algebraic, and geometric structures, and nd the
relationships between these equivalences.
Mathematics Subject Classication. 57M50
Key words. closed curve, hyperbolic surface.

1. Introduction
Given a closed orientable surface S of genus g > 1, there exists distinct homotopy
classes of curves g and g0 on S such that
lengthX g lengthX g0
for every hyperbolic structure X on S (see [15] and [25]). The proof involves writing
down words in the fundamental group which are not conjugate, but because of
certain trace relations, they have the same trace squared with respect to every
representation into PSL2 R (see Section 3).
The hyperbolic metrics are of course very special, and it seems reasonable that if
g and g0 are as above, there should be some purely topological reason for this.
PROBLEM 1.1 ([4]). Find a topological description of pairs of distinct homotopy
classes of curves g and g0 which have the same length with respect to every hyperbolic
structure on S.
More generally, given a family of path metrics GS on S, we ask the following.
PROBLEM 1.2. Do there exist pairs of distinct homotopy classes of curves g and g0
which have the same length with respect to every metric in GS? If so, nd a
topological description of such pairs.
For a generic Riemannian metric on S, there are no homotopy classes of curves
which have the same length (see [2]), so that the answer to the question will generically be no. Even when the answer is yes, a solution to the problem is likely

152

CHRISTOPHER J. LEININGER

to be uninteresting if GS is an arbitrary family of metrics. We are thus primarily

interested in this problem in the situation that GS arises naturally.
The hyperbolic metrics arise (for example) as the unique representatives of their
conformal classes with constant curvature 1, and are often useful in considering
topological problems about surfaces (see e.g. [32]). This is probably the most interesting setting for Problem 1.2. Another family of metrics which occur naturally
(especially in Teichmuller theory) are those metric structures induced by quadratic
differentials on S, holomorphic with respect to a complex structure (see e.g. [1]).
These are Euclidean cone metrics with some additional structure (see Section 4).
We will refer to such metrics as branched at metrics.
We discuss Problem 1.2 in the context of both hyperbolic and branched at
metrics, relate the two, and give a complete solution in the latter case. We also show
that an obvious necessary topological condition with respect to the hyperbolic
metrics is not sufcient, and mention a possible alternative.
To state the main theorem, we make the following denitions (see Sections 2 and 4
for more details).
DEFINITION 1.3. Let g and g0 be homotopy classes of essential closed curves on S.
. g and g0 are hyperbolically equivalent, g h g0 , if for every hyperbolic structure
X on S, lengthX g lengthX g0 .
. g and g0 are branched at equivalent, g bf g0 , if for every branch at structure
Q on S, lengthQ g lengthQ g0 .
. g and g0 are simple intersection equivalent, g si g0 , if for every isotopy class of
essential simple closed curve a on S, ia; g ia; g0 .
. g and g0 are trace equivalent, g tr g0 , if for every representation r : p1 S !
SL2 C; trr^g2 trr^g0 2 , where g^ and g^ 0 are elements of p1 S representing
the conjugacy classes determined by g and g0, respectively.
The respective equivalence classes will be called hyperbolic classes, branched at
classes, simple intersection classes, and trace classes.
The Main Theorem is the following.
THEOREM 1.4. Given a pair of homotopy classes of essential closed curves g and g0
on S, then
g tr g0 , g h g0 ) g si g0 , g bf g0
Moreover, we can nd g and g0 , each of which ll the surface, so that g si g0 yet g 6h g0 .
We say that a homotopy class of essential closed curves lls S if every representative intersects every other essential closed curve.
The outline of the paper is as follows. In Section 2, we give a few standard denitions to x notation. In Section 3, we briey discuss the construction of nontrivial
pairs g and g0 with g h g0 , following Horowitz and Randol. We then prove that

EQUIVALENT CURVES IN SURFACES

153

g tr g0 , g h g0 . One direction is immediate from the trace formula for lengths,
and the other direction follows easily from the main result of [26]. In Section 4,
we prove that g si g0 , g bf g0 . In the process, we determine an interesting
relationship between lengths and heights (see Lemma 4.8). In Section 5, we discuss
Bonahons interpretation of Thurstons compactication of Teichmuller space which
easily implies g h g0 ) g si g0 . In Section 6, we construct the required counterexamples, completing the proof of Theorem 1.4. We also describe a few of the difculties encountered in nding such pairs of curves. In Section 7, we describe a
candidate for a topological characterizations of hyperbolic equivalence and state
a couple questions. We end by citing a few references for more information on
Problem 1.1 and related problems.

2. Background
In this section we recall a few standard denitions and x some notation.
2.1.

TEICHMULLER SPACE AND CLOSED CURVES

Let S denote a closed orientable surface of genus g > 1. Let T eichS denote the
Teichmuller space of S, which we think of as the space of hyperbolic metrics on S
up to isotopy. By the Classical Uniformization Theorem (see [11]), this is equivalently the space of complex structures on S up to isotopy.
We let CrvS denote the set of homotopy classes of (unoriented) primitive essential closed curves on S. We denote the set of elements in CrvS with simple representatives by SS. Of course, SS can be naturally identied with the set of isotopy
classes of essential simple closed curves.
The geometric intersection number of two elements g, g0 2 CrvS is dened by
ig; g0 min
jg g 0 1 Dj
0
0
g 2g;g 2g

where g g 0: S1 S1 ! S S; D  S S is the diagonal, and jAj denotes the

cardinality of a set A
Given an element g 2 CrvS, we also have its associated length function
length() g : T eichS ! R
dened, for every hyperbolic metric X 2 T eichS, by
lengthX g inf lengthX g
g 2g

This is also the length of the geodesic representative of g in X.

We will often nd it useful to x a hyperbolic metric on S for reference. Therefore,
whenever we refer to geodesics on S (or in its universal cover) this is meant with
respect to the xed hyperbolic structure, unless otherwise stated. Once this is done,
each element of CrvS has a unique geodesic representative, and we can identify

154

CHRISTOPHER J. LEININGER

CrvS with the set of closed (primitive) geodesics on S. Moreover, geometric

intersection number is realized by these representatives (see e.g. [6] for a proof in
the case of simple curves).

2.2.

MEASURED LAMINATIONS AND FOLIATIONS

We denote the space of measured geodesic laminations on S by MLS. A

point L; l 2 MLS consists of a geodesic lamination L along with a transverse
measure l of full support (when no confusion arises, we refer to L; l simply as l).
If a 2 SS and t 2 R , then the geodesic representative for a, along with t times the
transverse counting measure denes a point in MLS. This denes an injection
R SS into MLS, and we identify R SS with its image. The relevant
Theorem here is the following ([7, 31]; see also Section 5).
THEOREM 2.1 (Thurston). R SS is dense in MLS. If g 2 CrvS, then
t; a ! t  ia; g
extends uniquely to a continuous function
i; g : MLS ! R
Remarks 2:2: (1) This is not exactly the statement in [31], but the work there
provides this version easily. This is also a direct consequence of the work in [7]
and [8].
(2) The extension of i; g can be explicitly described by
Z
il; g inf

dl

g 2g

g

The space MLS is naturally homeomorphic to MF S, the space of measure

classes of measured singular foliations on S. A point F ; m 2 MF S consists of a
singular foliation F and a transverse measure m of full support (again, we write
m F ; m when no confusion arises).
The homeomorphism
F : MLS ! MF S
is natural in the following sense [20].
THEOREM 2.3 (Thurston). For any g 2 CrvS, we have
Z
il; g inf
g 2g

dFl:
g

EQUIVALENT CURVES IN SURFACES

155

For obvious reasons, given g 2 CrvS and m 2 MF S, we write

Z
im; g inf dm
g 2g

g

With this convention, we state the following Corollary which will be used later.
COROLLARY 2.4. Let g; g0 2 CrvS: Then the following are equivalent.
1 g si g0 ,
2 il; g il; g0 for every l 2 MLS, and
3 im; g im; g0 for every m 2 MF S:
2.3.

REPRESENTATIONS OF p1 S

For K R or C, we will consider the representation varieties

RK S Hom p1 S; SL2 K:
These are naturally K-algebraic sets (see e.g. [10]).
Given an element g 2 p1 S, we have the corresponding character
wg : RK S ! K
given by
wg r trrg:
and this is a regular function on RK S. Since wg is invariant under conjugation (and
taking inverses), it depends only on g up to conjugacy (and taking inverses) in p1 S.
Since two loops represent conjugate elements in p1 S if and only if they are freely
homotopic, it follows that if g 2 p1 S is any representative of g 2 CrvS we can
dene wg wg .
We will also want to consider the quotient space
pK : RK S ! XK S
under the action of GL2 K by conjugation (this is almost the character variety). The
natural inclusion SL2 R  SL2 C gives us the inclusions RR S  RC S and
XR S  XC S (this last inclusion requires a little care-this is why we quotient by
the action of GL2 K instead of SL2 K:)
Given a point X 2 T eichS; we have the holonomy representation
rX : p1 S ! PSL2 R
which lifts to a representation (see [10])
r^ X : p1 S ! SL2 R:
The representation r is well dened up to conjugation in PGL2 R, and given r, the
lifts r^ are indexed by H1 S; Z=2Z: Choosing continuously varying lifts, we obtain a

156

CHRISTOPHER J. LEININGER

map T eichS ! XR S: Two different points X; Y 2 T eichS have non-conjugate

holonomy representations so that this map is injective and we can view
T eichS  XR S: With this identication T eichS sits inside QFS  XC S,
the set of quasi-Fuchsian representations (discrete, faithful, geometrically nite
representations without parabolics). By Mardens Quasi-conformal Stability
Theorem, [22] T eichS  XR S and QFS  XC S are open sets.
In fact, T eichS inherits a natural real analytic structure and QFS inherits a
natural complex analytic structure (using Frickes coordinates, see e.g. [17] or [19]).
Near any X 2 T eichS there are analytic coordinates so that T eichS  QFS
looks like
U \ R6g6  U \ C6g6
where U  C6g6 is an open set.
For each g 2 CrvS; wg is invariant under the action of GL2 C, so it pushes
forward to a well dened function on XC S. The analytic structures on T eichS
and QFS make wg into an analytic function. The length function on T eichS is
related to wg by the formula
lengthX g 2 cosh1



wg X
:
2

With this notation, we see that g tr g0 means w2g w2g0 . This trace formula
immediately implies the following.
PROPOSITION 2.5. For each g; g0 2 CrvS,
g tr g0 ) g h g0
Remark 2:6: When it is convenient, and no confusion arises, we will use the same
symbol to simultaneously denote an element of CrvS, any of its representatives,
the conjugacy class in p1 S determined by it (up to taking inverses), and any
representative of that conjugacy class.

3. Trace Classes and Hyperbolic Equivalence

In [15], Horowitz studies representations of a free group F into SL2 K, where K is a
commutative ring with 1 and characteristic 0. In an example he constructs, for any
N > 0, a collection of distinct elements g1 ; . . . ; gN 2 F, such that gj and g1
j are nonconjugate for every i 6 j, and so that wgi r wgj r for each i; j 1; . . . ; N and for
all representations r : F ! SL2 K. One of the key tools is the relation
truv truv1 trutrv
for every u; v 2 SL2 K.

EQUIVALENT CURVES IN SURFACES

157

EXAMPLE 3.1 [15]. In the rank-2 free group

F2 ha; bji
let g a2 b1 ab and h a2 bab1 . Since g and h are cyclically reduced, it is easy to
check that g is not conjngate to h1 . Using the above trace relation one can check
that for any representation
r : F2 ! SL2 C
we have
wg r wg rwab rwab1 r  wb2 r  wa r wh r
Randol in [25] uses Horowitzs examples to construct arbitrarily large h -equivalence classes. One way to do this is as follows. Let S0 denote a sphere with g 1
holes. there are obvious maps
i : S0 ! S

and p : S ! S0

such that p  i is the identity (see Figure 1).

Let i and p denote the induced maps on fundamental groups. Let g1 ; . . . ; gN 2
Fg p1 S0 be N elements so that gi is not conjugate to g1
j , for i 6 j, and
wgi wgj , for each i; j 1; . . . ; N (as described above). It follows that i gi and
i gj 1 must be non-conjugate in p1 S for each i 6 j, since i  p is the identity
on Fg. Therefore, i g1 ; . . . ; i gN represent distinct elements of CrvS, which
we refer to as g1 ; . . . ; gN respectively, and gi tr gj for i; j 1; . . . ; N. By Proposition
2.5, g1 ; . . . ; gN 2 CrvS must have gi h gj for i; j 1; . . . ; N.
We now begin the proof of Theorem 1.4. The following Proposition is the rst
step.
PROPOSITION 3.2. If g; g0 2 CrvS, then g tr g0 , g h g0 .
Proposition 2.5 is precisely one of the implications in the statement of this Proposition. The other implication will follow easily from the discussion in Section
2.3 and the following Theorem of [26].

Figure 1. i and p.

158

CHRISTOPHER J. LEININGER

THEOREM 3.3 (Rapinchuck et al. [26]). RC S is irreducible.

Proof of Proposition 3:2: It remains to show that if g h g0 , then g tr g0 .
As was noted in Section 2.3, the functions wg and wg0 are real and complex analytic
functions on T eichS and QFS, respectively. Since lengthX g lengthX g0 for
every X 2 T eichS, the trace formula implies wg X wg0 X for every
X 2 T eichS, and hence w2g w2g0 on T eichS. Now because of the local analytic
structure of the inclusion of T eichS  QFS described in Section 2.3, elementary
complex analysis implies w2g w2g0 on the component of QFS containing our
embedding of T eichS (this then holds for any component by choosing a different
embedding).
Since QFS  XC S is open, p1
C QFS is an open subset of RC S (in the usual
C topology), and thus w2g w2g0 on this open set. However, since w2g  w2g0 is a regular
function, w2g  w2g0 0 denes an algebraic subset of RC S containing p1
C QFS.
Some irreducible component, Z, of this algebraic subset contains an open subset
(in the C topology) of p1
C QFS. Since Z and RC S are both irreducible (the
latter by Theorem 3.3), they have well dened dimensions, and since Z contains an
open subset of RC S, these dimensions are equal. It follows that Z RC S
(see e.g. [29]), and hence w2g w2g0 on all of RC S.
&
We state an immediate corollary of Proposition 3.2, which provides an often easily
computable obstruction to g h g0 .
COROLLARY 3.4. Let g; g0 2 CrvS, with g h g0 . Then g and g0 may be oriented so
that they represent the same homology class.
Proof. Suppose that, regardless of the choice of orientations, g and g0 represented
different homology classes. Then there exists a homomorphism f : p1 S ! Z such
that fg 6 fg0 . Let c : Z ! SL2 C be given by
 n

2 0
cn
0 21n
Since trcn 2n 1=2n , we see that w2g cf 6 w2g0 cf, contradicting Proposition
3.2.
&
The following consequence of Proposition 3.2 may potentially provide a topological description of hyperbolic equivalence (see Section 7).
COROLLARY 3.5. Let g; g0 2 CrvS, with g h g0 . Then, for every complete
hyperbolic surface M and continuous map f : S ! M, we have
lengthM fg lengthM fg0 :
Proof. Let g; g0 , and f : S ! M be as in the statement of the theorem. We let [g]
and [g0] denote elements of p1 S representing the conjugacy classes determined by g

EQUIVALENT CURVES IN SURFACES

159

and g0 respectively. fg and fg0 determine conjugacy classes of elements of p1 M,

represented by f g and f g0  respectively. The hyperbolic structure determines a
holonomy representation r : p1 M ! PSL2 R which lifts to a representation into
SL2 R (see [10]), which we also call r.
Now, p  f: p1 S ! SL2 R is a homomorphism and so
w2f g r w2g r  f w2g0 r  f w2f g0  r
by Proposition 3.2. Applying the trace formula for length we see that fg and fg0
have the same length in M.
&
Remark 3:6: Corollary 3.5 also holds if M is a hyperbolic 3-manifold. That is,
if g h g0 , and f : S ! M is a continuous map to a hyperbolic 3-manifold, M, then
fg and fg0 have the same lengths.

4. Branched Flat Equivalence and Simple Intersection Equivalence

4.1.

BRANCHED FLAT METRICS AND HOLOMORPHIC QUADRATIC DIFFERENTIALS

A branched at metric on S is a Euclidean cone metric such that all cone angles are
of the form kp for k 2 Z and k > 2 (see e.g. [9]), and such that the holonomy around
any loop in S minus the singular locus lies in {0, p}.
A branched at metric denes a (complete) geodesic metric on S, such that the
metric pulled back to the universal cover has unique geodesics connecting any two
points (see [1] or [5]). The geodesics in this metric are straight in the complement
of the singular locus and make an angle no less than p at any singular point which
the geodesic hits. Any g 2 CrvS has a length minimizing geodesic representative.
Given a complex structure X 2 T eichS , we let QX denote the vector space of
holomorphic quadratic differentials on S (holomorphic with respect to X ). The
union of the spaces QX , as X ranges over T eichS , forms a complex vector bundle
over T eichS which we denote by QS , and we let ZS denote the zero section.
We claim that any f 2 QS nZS denes a branched at metric (see [1, 13], or [17]).
To see this, we suppose f 2 QX and let z be a local coordinate about a point p0 for
which f p0 6 0. Then f fzdz2 and in a small enough neighborhood of p0 , we
can integrate to obtain a new local coordinate
Z o p
zo
fz dz
z0

In this new local coordinate, f has the simple expressions f dz2 . We call a local
coordinate, obtained in this way, a preferred coordinate. Any two preferred coordinates differ by rotation through 0 or p and translation. That is, if z1 and z2 are two
preferred coordinates about a point p (not a zero of f), then
z1 z2 o
for some o 2 C.

160

CHRISTOPHER J. LEININGER

It now follows that jfj1=2 is the length element for a Riemannian metric at all
points of X, except the zeros of f. In the preferred coordinate z, this has the simple
form jdzj, which is the standard Euclidean metric. To understand the behavior of the
metric in a neighborhood of the zero p of f, we note that there exists a local coordinate z about p such that f zm dz2 . It is now straightforward to show that
the cone angles are of the required type.
Finally to see that the holonomy lies in f0; pg, we note that f denes another type
of geometric structure; a pair of transverse measured foliations. This follows from
the above description of the transition functions with respect to the preferred coordinates z x iy. The horizontal and vertical foliations with transverse measure
dened by jdyj and jdxj respectively, are invariant by these transition functions,
hence dene measured foliations on X. The corresponding measured foliations
are called the
p horizontal and vertical
p foliations of f and will be denoted by
F h f; jIm fj and F v f; jRe fj, respectively. Note that the leaves of both
foliations are geodesics and that they are orthogonal. Using either one of these foliations as a guide, it is clear that the holonomy lies in {0, p}.
The height of g 2 CrvS with respect to f is dened to be
Z
p
hf g inf
jIm fj:
g0 2g g
0

This is, of course, simply the variation of g with respect to the horizontal foliation.
The width is similarly dened to be the variation with respect to the vertical foliation.
It is not hard to show that if g0 is a f-geodesic representative for g, then
Z
p
hf g
jIm fj:
g0

Similarly for the vertical foliation.

Now suppose S is given a branched at structure. This defines both a complex
structure and a non-zero holomorphic quadratic differential, up to multiplication
by a complex unit. An atlas for the complex structure is described by local isometrics
from neighborhoods of the non-singular points to open sets in E2 C. In neighborhoods of the singularities, one can construct coordinate charts modeled on appropriate branched covers.
To describe the quadratic differential, note that because the holonomy lies in
{0, p}, given any non-singular point and any line in the tangent space to that point,
we may parallel translate that line to every non-singular point giving a line eld well
dened on the complement of the singular locus. This allows us to further rene our
rst type of coordinate charts above so that our line eld is always sent to the line
eld spanned by @=@x in C. The transition functions for these new coordinate charts
are of the form z 7!  z x for some x 2 C, and hence preserve dz2 . Letting f denote
the pull back of dz2 under these coordinates and extending by zero over
the singular locus gives a quadratic differential, holomorphic in the complement
of the singular locus. One can check that this is in fact holomorphic everywhere.

EQUIVALENT CURVES IN SURFACES

161

Note that rotating the line eld amounts to multiplying f by a complex scalar with
unit norm.
4.2.

BRANCHED FLAT SIMPLE INTERSECTION

We now set out to prove the following.

PROPOSITION 4.1. If g, g0 2 CrvS, then g bf g0 , g si g0 .
We begin with the following, obvious Lemma.
allf 2 QSnZS.
LEMMA 4.2. Let g, g0 2 CrvS. If g si g0 then hf g hf g0 , forp
Proof. From the previous sub-section we know hf g ijRe fj; g. Corollary
2.4 implies the result.
&
Remark 4:3: It follows from the work of Hubbard and Masur in [16] that g si g0
if and only if hf g hf g0 , for all f 2 QSnZS, although we do not need this
stronger version.
Next, we prove the following well known lemma constituting half of Proposition 4.1.
LEMMA 4.4. If g, g0 2 CrvS, then g bf g0 ) g si g0 .
Proof. The proof of this lemma follows from analyzing the limiting behavior
of the Teichmuller deformation associated to an appropriate quadratic differential
(see for example [18]).
Explicitly, we begin by xing a curve a 2 SS. One can easily construct
fa 2 QSnZS such that the horizontal foliation F h fa ; jImfa j represents
(1, a) under the inclusion R SS  MF S. If we remove the critical trajectories
of such fa , the result is an open Euclidean annulus with core curve isotopic to a,
length 1, and girth R, for some R > 0. We deform such a fa to obtain a family
ffa;K gK20;1  QSnZS, by compressing the horizontal direction by a factor of
K. More precisely, let e1 , e2 be an orthonormal basis for the tangent space at a
non-singular point x 2 S, with e1 tangent to the horizontal foliation (and hence e2
tangent to the vertical). The original metric at x is thus given by dij , and the deformed
metric is given by g12 g21 0, g11 K2 , and g22 1. For every K 2 0; 1, removing the critical trajectories of fa;K we obtain an open Euclidean annulus with
p
length 1 and girth KR, so that F h fa;K ; jRe fa;K j also represents (1, a).
We now let g 2 CrvS denote any element, and claim that
lim lengthfa;K g hfa;K g ia; g:

K!0

The second equality holds by construction. To see the rst, let g0 denote a fa -geodesic representative for g. Then g0 consists of ia; g geodesic arcs traversing the

162

CHRISTOPHER J. LEININGER

length of the annulus and some number of geodesic arcs, each of which is contained
in the set of critical trajectories. It is not hard to see that g0 remains geodesic after the
deformation, that is, it is also a fa;K -geodesic. As K ! 0, the length of the arcs
contained in the critical trajectories converges to 0, and the length of the others
converges to 1.
To complete the proof, let g, g0 2 CrvS be such that g bf g0 , and let fa;K be as
above. Then
ia; g lim lengthfa;K g lim lengthfa;K g0 ia; g0 :
K!0

K!0

Since a was arbitrary, g si g0 .

&

To prove the other implication of Proposition 4.1 we will need the following
DEFINITION 4.5. Given a nite list of points Z z1 ; . . . ; zk 2 Ck , dene the
height of Z to be

hZ

k
X

jImzj j:

j1

Remark 4:6: We are considering lists of numbers instead of sets to allow the
possibility of a number showing up with multiplicity greater than one.
LEMMA 4.7. Let Z z1 ; . . . ; zk 2 Ck . Then HZ y heiy Z is continuous and
piecewise smooth. If y1 ; . . . ; yn denotes the points in 0, p where H0Z y does not exists,
then
k
X

jzj j

j1

n 

1X
lim H0Z y  lim H0Z y :
y!yl
2 l1 y!yl

Proof. HZ is clearly continuous.

To prove the remainder of this lemma, it will be convenient to dene the function
f : C ! C by

fz

z if Imz > 0 or Imz 0 and Rez 5 0

:
z otherwise

Then
HZ y

k
X

Im feiy zj :

j1

Note that for any z 2 C , the function heiy z Im f eiy z is smooth (as a function of y) exactly when eiy z 2
= R, i.e. when argz y 2
= Zp. Moreover, away from such

163

EQUIVALENT CURVES IN SURFACES

d
d
values of y, the derivative is dy
heiy z dy
Im feiy z Re feiy z. If we let yz be any
value of Zp  argz, then

lim feiy z jzj and

y!y
z

lim feiy z jzj

y!y
z

As y ! yz from the right we have Im feiy z ! 0 so that

lim

y!y
z

d
Im feiy z lim Re feiy z lim feiy z jzj
dy
y!yz
y!yz

Putting this together with the analogous result when y ! yz from the left, we obtain
lim

y!y
z

d
d
Im feiy z  lim Im feiy z jzj  jzj 2jzj
y!yz dy
dy

We now return to the proof. We may clearly assume that no coordinate of Z is 0.

Also, note that we may replace Z by fz1 ; . . . ; f zk without loss of generality, so
that all coordinates of Z lie in the open upper half plane union the positive real axis.
Further, if zi is a (positive) real multiple of zj j 6 i, then we may replace the two
coordinates, zi and zj , by the single coordinate zi zj . We therefore assume that
no two coordinates lie on the same line through 0 in C.
It follows that the arguments Argz1 ; . . . ; Argzk , taken in 0; p are distinct. HZ
is now seen to be smooth except when y mp  Arg zj and m 2 Z. Letting
yj p  Arg zj for each j 1; . . . ; k; fy1 ; . . . ; yk g is precisely the set of points in
0; p where HZ is not smooth.
We have
k 
X
j1


d
Im feiy zl 
y!yj dy
j1 l1

!
k
X
d
iy
Im fe zl
 lim

2jzj j
y!yj dy
j1

k X
k
 X
lim H0Z y  lim H0Z y

y!yj

y!yj

lim

where the last equality follows from Equation (1) when j l and from the fact that
Im feiy zl is smooth at yj whenever j 6 l. This completes the proof of the
Lemma.
&
The following Lemma, which seems interesting in its own right, will easily
complete the proof of Proposition 4.1.
LEMMA 4.8. Let g 2 CrvS and f 2 QSnZS. Then Hg y he2iy f g is continuous
and piecewise smooth. If y1 ; . . . ; yn denotes the set of points in 0; p for which H0g does
not exist, then
lengthf g

n 

1X
lim H0g y  lim H0g y :
y!yl
2 l1 y!yl

164

CHRISTOPHER J. LEININGER

Proof. Let g0 denote a f-geodesic representative for g. This is also an e2iy

f-geodesic, since the metric dened by e2iy f is independent of y. Let g0;1 ; . . . ; g0;k
denote the straight arcs (with endpoints on the zeros of f) which make up g0 . For
every y 2 R and j 2 f1; . . . ; kg, we can develop g0;j to a straight arc g0;j;y  C using
preferred coordinates dened by e2iy f.
If g0;j;y and g00;j;y are any two developing images of g0;j by e2iy f, then there is o 2 C,
such that g00;j;y cg0;j;y , where cz z o. We may therefore assume that each
g0;j;y has one endpoint at 0. Denote the other endpoint of g0;j;y zj;y .
Set Zy z1;y ; . . . ; zk;y and note that hZy he2iy f g since the horizontal foliation of e2iy f is obtained by pulling back the horizontal foliation of C by a preferred
coordinate, and since geodesics realize the height.
If we let z denote a preferred coordinate for f, then eiy z is a preferred coordinate
for e2iy f. Therefore, we can take Zy eiy z1;0 ; . . . ; eiy zk;0 eiy Z0 .
So, Hg y hZy heiy Z0 HZ0 y. Since
lengthf g lengthf g0

k
X

lengthf g0;j

j1

Lemma 4.7 applied to Z0 completes the proof.

k
X
j1

lengthC g0;j;0

k
X

jzj;0 j

j1

&

Proof of Proposition 4:1: By Lemma 4.4, we need only prove that if g, g0 2 CrvS
and g si g0 , then g bf g0 .
Fix f 2 QSnZS arbitrarily. Let Hg y and Hg0 y be as in the statement of
Lemma 4.8. By Lemma 4.2, Hg y Hg0 y for every y 2 R, and so applying Lemma
4.8, we obtain lengthf g lengthf g0 . Since f was arbitrary, we are done.
&
Remark 4:9: As a conseqnence of this, Corollary 5.4 below, and the existence of
arbitrarily large hyperbolic classes in CrvS, we see that there are arbitrarily large
branched at classes in CrvS.

5. Geodesic Currents
In this section, we briey discuss Thurstons compactication of Teichmuller
space in the context of Bonahons work on geodesic currents [8] as it applies to
our situation.
Following [8], we dene a geodesic cnrrent as follows. We have on S a xed hyperbolic metric (cf. Section 2.1), which provides us a hyperbolic metric on the universal
cover p : S~ ! S (thus making it isometric to H2 ). We denote the circle at innity by
~ The endpoints of a geodesic naturally
S11 and the space of geodesics on S~ by GS.
1
~
provide an identication of GS with S1 S11 nD=  where D  S11 S11 is the
~
diagonal and x; y  y; x. The fundamental group p1 S acts isometrically on S,
~ We dene a geodesic current on S to be a p1 S-invariant
and hence it also acts on GS.
~ (measuring compact sets nitely). We denote the
positive Borel measure on GS
space of geodesic currents on S by CurrS, topologized with the weak* topology.

EQUIVALENT CURVES IN SURFACES

165

We can naturally embed R CrvS into CurrS as follows. The metric on S

allows us to view g 2 CrvS as a geodesic so that p1 g is a p1 S-invariant discrete
~ We dene the corresponding geodesic current (denoted simply by
subset Xg  GS.
t  g) to be t times the atomic measure corresponding to Xg (so, t  gE t  jE \ Xg j,
for any Borel set E ).
We may also embed MLS into CurrS. The measured lamination L; l pulls
~ l
~ on S.
~ as a (closed) p1 S-invariant
~ We view L
back to a measured lamination L;
~ as follows. A geodesic arc a in S~ denes a
~ and dene a measure on L
subset of GS,
~
subset Ea  L consisting of those geodesics transversely intersecting a. We dene the
~ a . The sets of the form Ea are enough to well dene a
measure of Ea to be lE
~
~ and denote simply by l 2 CurrS.
measure on L which we extend to all of GS
Next, we briey describe the (proper) embedding of T eichS into CurrS as
dened in [8]. The Liouville measure, L, is the IsomH2 -invariant positive Borel measure on GH2 (the space of geodesics on H2 ) with the following dening property.
If a and b are two disjoint arcs on the circle at innity with endpoints a0 , a1 and b0 ,
b1 respectively (in either the disk or upper half plane model) and Ea;b is the set
of geodesics with one endpoint in a and the other in b, then



a0  b0 a1  b1

LEa;b log
a0  b1 a1  b0
Now, given a hyperbolic structure X on S, we view X as a hyperbolic surface along
with a diffeomorphism f : S ! X. The map f lifts p1 S-equivariantly to a diffeomorphism of the universal covers
f~: S~ ! H2 :
This map extends (by a p1 S-equivariant homeomorphism) to a map on circles at
innity, and hence induces a homeomorphism
~ ! GH2
fX : GS
~ which
Pulling back the Liouville measure via fX describes a Borel measure on GS


is invariant under p1 S. That is, LX fX L is a geodesic current. The map
T eichS ! CurrS
dened by X 7!LX is a proper embedding [8].
For the remainder of this section, we will identify R CrvS, MLS, and
TeichS with their images in CurrS.
In [7], a bilinear function is dened
I : CurrS CurrS ! R
This function enjoys several nice properties (for the denition of I and a proof of the
following see [7] and [8] (there, I is called i)).

166

CHRISTOPHER J. LEININGER

THEOREM 5.1 (BONAHON).

1: The function I is continuous.
2: Given t  g, t0  g0 2 R CrvS, l 2 MLS, and X 2 T eichS, we have
.
.
.
.

It  g; t0  g0 tt0  ig; g0
Il; t  g t  il; g
ILX ; t  g t  lengthX g
ILX ; LX p2 jwSj

3: If m 2 CurrSnf0g satises Im; m 0, then m 2 MLS.

Using this function and this embedding of T eichS, one can recover Thurstons
compactication (see [32] and [12] for the original denition). Briey, one rst notes
that by the nal item in Theorem 5.1(2), the inclusion of T eichS into CurrS
remains embedded after passing to the projectivization of CurrS, denoted
PCurrS CurrSnf0g=R . The space PCurrS is compact, and therefore the
closure of the image of T eichS in PCurrS is compact (see [8]).
Now suppose fXi g is a sequence in T eichS which diverges. We wish to identify
the limit point in PCurrS as an element of PMLS, the projectivization of
MLS. By passing to a subsequence (using compactness), we may assume that
the image of the fXi g in PCurrS converges. Therefore, there exists ti 2 R so that
ti  Xi ! l 2 CurrSnf0g. Since the embedding of T eichS into CurrS is proper (or
by an easy geometric argument), it must be that ti ! 0. So by Theorem 5.1
t2i p2 jwSj t2i IXi ; Xi Iti Xi ; ti Xi ! Il; l
as i ! 1. Since the rst term is obviously approaching 0, we see that Il; l 0 and
therefore l 2 MLS.
As in [8], one can check that this is exactly Thurstons compactication.
THEOREM 5.2 (Thurston). T eichS is compactied by PMLS. In this compactication, which we denote by T eichS, a seqnence fXi g  T eichS converges to
l 2 PMLS if and only if for every pair g, g0 2 CrvS we have
lim

lengthXi g
il; g

0
il;
g0
g
i

i!1 lengthX

provided il; g0 6 0. Moreover, T eichS is a dense open set in T eichS.

Remark 5:3: In the usual statement of this Theorem g and g0 are taken to lie in
SS, but the work in [8] described above easily implies this version. The usual
6g6
, the closed
statement also includes the additional information that T eichS B
ball in R6g6 , and that T eichS is the interior (see 32 or 12).
The following is now immediate from this Theorem.
COROLLARY 5.4. Let g; g0 2 CrvS. Then g h g0 ) g si g0 .

EQUIVALENT CURVES IN SURFACES

167

Proof. Let a 2 SS be an arbitrary simple closed curve, and let us denote the
image of 1; a in MLS by a. By Theorem 5.2, there is a sequence Xi 2 T eichS
such that Xi ! a (it is not difcult to explicitly construct such a sequence). Since
g h g0 , we have
lengthXi g
ia; g

0
i!1 lengthX g
ia; g0
i

1 lim

which completes the proof.

&

Now, let g; g0 2 CrvS, and consider the following function

Fg;g0 : T eichS ! R
given by
Fg;g0 X

lengthX g
lengthX g0

From Section 2.3, we see that Fg;g0 is not only continuous on T eichS, but it is real
analytic. This function does not necessarily extend continuously over T eichS
in general (even when g si g0 ).
However, if we assume that both g and g0 (individually) ll S, so that for every
l 2 MLS; il; g 6 0 6 il; g0 , then there is an extension (denoted Fg;g0 which is
continuous by Theorem 2.1 and Theorem 5.2.
Moreover, if in addition we assume that g si g0 , then Fg;g0 1 on PMLS. In
particular, Fg;g0 is asymptotic to 1 in every direction, and hence Fg;g0 extends by 1
over any compactication of T eichS.
Despite this behavior, in Section 6 we construct pairs of curves g and g0 so that g
and g0 both ll the surface and g si g0 , yet g 6h g0 .

6. Simple Intersection Equivalence

In this section, we construct a pair of closed curves which both (individually) ll S
and are simple intersection equivalent, yet are not hyperbolically equivalent. The rst
difculty that one encounters in constructing such a pair of curves is verifying that
they are simple intersection equivalent. We discuss this difculty at some length as it
strongly contrasts the behavior of simple curves.
6.1.

DISTINGUISHING CURVES

The work in this sub-section is not necessary for the proof of the main theorem, but
is included because of the obvious relevance to constructing the required counterexamples.

168

CHRISTOPHER J. LEININGER

It is well known (see e.g. [24]) that for any surface S, there exists a nite set
fa1 ; . . . ; an g  SS such that if g; g0 2 SS, and
iaj ; g iaj ; g0
for each j 1; . . . ; n, then g g0 .
Of course, when g; g0 2 CrvS (and S has genus at least 2) this fails by Corollary
5.4 and the existence of g h g0 with g 6 g0 (Section 3). In fact the situation is worse,
as the following indicates.
THEOREM 6.1. Given any proper compact subset K  PMLS there exists two
curves g; g0 2 CrvS and a 2 SS such that
il; g il; g0
for every l with l 2 K, yet ia; g 6 ia; g0
In particular, this theorem says that there is no nite list of simple closed curves
such that for any g; g0 2 CrvS, the intersection number with each curve in this list
can be used to decide whether or not g and g0 are simple intersection equivalent.
Proof. Since K is a proper compact subset of PMLS there exists a lamination
l 2 MLS such that l 2
= K and so that some complementary region of l is an ideal
triangle. It follows from [31] (in particular Section 9.7) that l may be chosen so that
there are train tracks t and tK with t carrying l; tK carrying every element of K, and
so that t and tK meet efciently, in the sense that they meet transversely and so that
any trainpaths of t and tK lift to paths in the universal cover which meet transversely
in at most one point. This efciency guarantees that if g is carried by t and gK is
carried by tK , then ig; gK is determined by the respective weights they dene on the
branches of t and tK . In fact, it is not hard to see that this holds even if we allow g
to be a non-simple closed curve carried by t (the denition of a train track carrying
a closed curve is essentially the same as that given for a lamination in [31]).
Now let g be an oriented simple closed curve fully carried by t (i.e. g denes positive weights on all branches of t). Note that t must have a complementary region
which is a triangle. It follows that there must be some branch which contains strands
of g with disagreeing orientations. To see this, note that if this werent the case, then
the boundary of the triangular regions could be oriented to agree with the orientation of g. Thus two sides of the triangle would have to meet at a vertex with one
orientation heading into the vertex and the other heading out (see Figure 2). A
branch on the opposite side of this switch would have to contain strands with both
orientations (because the switches are trivalent).
Next choose two strands in such a branch with opposite orientation which are
adjacent (here we are thinking of g as embedded in a small I-bundle neighborhood
of t). Cut these strands and re-glue them with a cross (see Figure 3). We claim that
the resulting immersed 1-manifold is actually a curve, g0 2 CrvS and it is not in SS
(i.e. it is not homotopic to a simple curve).

EQUIVALENT CURVES IN SURFACES

169

Figure 2. Incompatible.

Figure 3. Exchange.

Assuming this claim for the moment, we see that g0 is also carried by t, and that g
and g0 dene the same weights on the branches of t. It follows that for any measured
lamination m carried by tK (in particular, any m 2 K), that im; g im; g0 . Because
g 2 SS and g0 is not, one can easily nd a 2 SS such that ia; g 6 ia; g0 .
To complete the proof of the proposition, we must prove:
(i) g0 has one component (as an immersed 1-manifold), and
(ii) the single self-intersection point cannot be removed.

Proof of i. Let g0 g1 g2 denote the 2 oriented arcs which remain after cutting
g. Re-gluing will result in a disconnected 1-manifold if the two ends of g1 are glued
together (and hence also the two ends of g2 ). Since we have chosen a pair of strands
which are oppositely oriented, the negative boundary components of g0 are glued
together and hence g1 is glued to g2 . Therefore, g0 is connected and (i) holds.
Proof of ii. Again, we think of g as embedded in an I-bundle neighborhood of t.
By construction, g0 has exactly one (transverse) self intersection point. An exercise in
the topology of surfaces shows that a curve with one transverse self intersection
point is homotopic to a simple closed curve if and only if one of the complementary
regions of the curve is a disk (the complementary regions being the components of
the path metric completion of the surface minus the curve). The components of the

170

CHRISTOPHER J. LEININGER

preimage of g0 in S~ (i.e. the complete lifts of g0 ) are easily seen to be embedded

(a component of the preimage is a trainpath and so the intersection with the I-bundle
neighborhood of a branch of p1 t is a single arc). A complementary region which
is a disk would lift to the universal cover and no component of the preimage could
be embedded. Therefore, no complementary region is a disk and hence g0 is not
homotopic to a simple closed curve.
&
6.2.

PAIRS OF PANTS

There is a situation where we can easily decide whether or not a pair of curves are
simple intersection equivalent without knowing that they are hyperbolically equivalent. We will use this in our construction in the next subsection.
Let P denote a pair of pants, i.e. a sphere with 3 holes. There are exactly 6 isotopy
classes of essential properly embedded arcs which we call l1 ; l2 ; l3 and w1 ; w2 ; w3 (see
[24]). Representatives of each isotopy class are shown in Figure 4. Note that for each
i 2; 3 there is a homeomorphism of P taking l1 to li . Similarly, for each i 2; 3
there is a homeomorphism of P taking w1 to wi .
Choose a basepoint x and pair of generators a and b for p1 P; x as shown in the
Figure 5 (so p1 P; x Fa; b, the free group on a and b). As in the closed case, we
denote the set of essential closed curves in P by CrvP. Any element of CrvP may
then be represented by a cyclically reduced word, unique up to cyclic permutation
and taking inverses (see e.g. [21]). We write such an element as ax1 by1 . . . axn byn with
xj 6 0 and yj 6 0 for each j 1; . . . ; n, except possibly when n 1, in which case we
write the element as ax1 ; by1 , or ax1 by1 .
The geometric intersection number of g 2 CrvP with l1 is given by
ig; l1

n
X

jxj j:

j1

where g is given as ax1 by1 ; . . . ; axn byn as above. To see this, we construct a cover
p : P~ ! P using cut and paste techniques so that
. g lifts to a curve g~ in this cover
. g~ can be homotoped so that every component l~1 of p1 l1 intersects g~ at most
once.

Figure 4. Arcs in Pants.

EQUIVALENT CURVES IN SURFACES

171

Figure 5. Pants.

It follows that
ig; l1 i~g; p1 l1

n
X

jxj j:

j1

Similarly,
ig; l2

n
X

jyj j:

j1

A similar type of argument can be used to show that

ig; w1 2n
with the exception that
iax1 ; w1 iby1 ; w1 0:
Formulae for ig; l3 and ig; wi , for i 2; 3, can be obtained by choosing a different
basis for p1 P; x.
Now suppose that the pair of pants, P, is an incompressible subsurface of a closed
surface S. Fixing a hyperbolic structure on S, we may assume that P has geodesic
boundary. If a 2 SS, and g; g0 2 CS, and g and g0 both have representatives lying
entirely in P, then the geodesic representatives of g and g0 also lie in P. Taking the
geodesic representative for a, we see that a \ P is a nite union of arcs of types
l1 ; l2 ; l3 ; w1 ; w2 ; w3 . It follows that if ig; li ig0 ; li and ig; wi ig0 ; wi , in P, for
each i 1; 2; 3, then ig; a ig0 ; a. This proves the following.
PROPOSITION 6.2. If g; g0 2 CrvS have representatives contained in an incompressible pair of pants, P  S, then g si g0 if and only if g and g0 have the same
geometric intersection number with each of the 6 essential arcs in P.
&
6.3.

COUNTEREXAMPLES

Before we construct our counterexamples, we will need the following.

PROPOSITION 6.3. Suppose g; g0 2 CrvS and p : S~ ! S is a nite sheeted cover in
which both g and g0 lift to curves g and g~ 0 respectively. If g~ si g~ 0 , then g si g0 .

172

CHRISTOPHER J. LEININGER

Figure 6. The immersed curve g0 .

Proof. We assume g~ si g~ 0 and let a 2 SS be arbitrary. We write a~ p1 a, and

note that

a~

k
a

a~ j

j1

~ for each j 1; . . . ; k. We then have

where a~ j 2 SS
ia; g i~a; g~

k
X
j1

i~aj ; g~

k
X
~aj ; g~ 0 i~a; g~ 0 ia; g0

&

j1

To construct our counterexample, we begin with a surface S of genus 2 and the

curve g0 2 CrvS shown in Figure 6. The surface S is obtained by considering the
sphere with 4 holes shown and gluing boundary components B to B 0 by a dilation
and A to A0 by a reection through the vertical line cutting the gure in half.
In the surface S, we refer to these curves as B and A respectively. The arcs then
match up to give the closed curve we call g0 2 CrvS. One can check that cutting
S along g0 gives 2 octagons, and we provide S a hyperbolic metric so that each is
a regular all-right octagon. The full preimage of g0 in H2 (the universal cover) gives
a tessellation by regular all-right octagons, so that in particular, every complete
geodesic in H2 transversely intersects g0 . Therefore, every geodesic in S transversely
intersects g0 , and g0 lls S.
Now consider the 2-fold covering space p : S~ ! S corresponding to the kernel
of the homomorphism induced by mod 2 intersection number with B. One easily
checks that g0 has 0 mod 2 intersection number with B, and hence has two distinct
~ Figure 7 shows this two fold cover (the gluings are as indicated). A and B
lifts to S.
both lift in this cover (one of the lifts of B, labeled B1 , is drawn as a dotted circle to
help clarify the picture). Both lifts of g0 are shown; one is drawn solid, which we call
g~ 0 , and the other is dashed.

EQUIVALENT CURVES IN SURFACES

173

Figure 7. A 2-fold cover of S.

Now we consider an incompressible pair of pants P  S~ with two components of

@P being g~ 0 and B0 . We take P as a regular neighborhood of g~ 0 [ B0 [ r, where r is
a short arc connecting g~ 0 to B0 . In Figure 8 we have redrawn S~ indicating P as the
shaded subsurface.
We orient g~ 0 and B0 , pick a point x 2 intP, and choose a basis fa; bg for p1 P; x
as in the previous section, so that a and b are freely homotopic (as oriented curves) to
g~ 0 and B0 , respectively.
~ be curves contained in P given by the following words in
Now, let g~ ; g~ 0 2 CrvS

~
Figure 8. Pants in S.

174

CHRISTOPHER J. LEININGER

a and b.
g~ a2 bab1 ;

g~ 0 a2 bab

and set
g p~g;

g0 p~g0 :

Note that g~ and g~ 0 are lifts of g and g~ 0 . For if this were not the case, then g~ , say,
~ and the covering transformation would have to take
would be a 2:1 cover of g in S,
g~ onto itself. This would imply that the covering map restricts to a self map of P,
which is impossible since there are no xed point free orientation preserving
involutions of a pair of pants.
CLAIM 1. Both g and g0 ll S.
Proof. This follows from the fact that g0 lls S. To see this, consider our pair of
pants P  S~ with its hyperbolic structure (with geodesic boundary), and represent all
curves by geodesics. Since neither of g~ nor g~ 0 is a multiple of a boundary component
of P, it is easy to see that any curve which transversely intersects g~ 0 must transversely
intersect both g~ and g~ 0 .
Now any complete geodesic s in S transversely intersects g0 since g0 lls S. Any
such point of intersection lifts to a point of intersection of g~ 0 with the preimage of
s. This gives rise to points of transverse intersection of both g~ and g~ 0 with this lift
of s, by the remarks of the previous paragraph. These points push down to points
of intersection of s with g and g0 . Therefore both g and g0 ll S.
CLAIM 2. g si g0 .
Proof. We show that g~ si g~ 0 and apply Proposition 6.3 to prove this claim. By
Proposition 6.2, we need only check that i~g; li i~g0 ; li and i~g; wi i~g0 ; wi , for
i 1; 2; 3. Using the method described in the previous subsection, one easily obtains
i~g; l1 3 i~g0 ; l1 ; i~g; l2 2 i~g0 ; l2 ; i~g; l3 3 i~g0 ; l3
i~g; w1 4 i~g0 ; w1 ; i~g; w2 4 i~g0 ; w2 ; i~g; w3 6 i~g0 ; w3
CLAIM 3. g 6h g0 .
Proof. By Corollary 3.4, it sufces to show that g 6 g0  2 H1 S; Z. We have
g p ~g p 2~g0  B0  ~g0   B0  p ~g0  g0 
g0  p ~g0  p 2~g0  B0  ~g0  B0  p ~g0  2B0  g0  2B
So, if g g0  then we obtain
g0  g0  2B , 2B 0
or
g0  g0   2B , B g0 :

EQUIVALENT CURVES IN SURFACES

175

Since B 6 0, and since B cannot possibly be g0  (g0  has non-trivial algebraic

intersection number with A, for example) both of these give a contradiction, hence
g 6 g0  2 H1 S; Z and g 6h g0 . In a similar fashion, one can construct many
other examples.
We have proved
LEMMA 6.4. There exists curves g; g0 2 CrvS, each of which ll S, such that g si g0 ,
yet g 6h g0 .
&
Proposition 3.2, Proposition 4.1, Corollary 5.4, and Lemma 6.4 combine to prove
Theorem 1.4.

7. Alternatives and Related Questions

Suppose g; g0 2 CrvS and g h g0 . Let f : S ! T 2 be a continuous function from S to
the torus T 2 . It follows from Corollary 3.4 that the map f : p1 S ! p1 T 2 Z2
must have f g f g0 . On a torus this implies fg fg0 in CrvT2 , and in
particular, fg si fg0 (si is dened as in the hyperbolic case).
Now, let S0 be any orientable surface of genus g > 1, and let f : S ! S0 be any continuous function. By Corollary 3.5, fg h fg0 and so by Theorem 1.4, fg si fg0 .
This leads us to make the following
DEFINITION 7.1. Given g; g0 2 CrvS, say that g and g0 are strongly simple
intersection equivalent, g ssi g0 , if for every orientable surface S0 and continuous
map f : S ! S0 , we have fg si fg0 .
The preceding arguments prove
THEOREM 7.2. If g; g0 2 CrvS and g h g0 , then g ssi g0 .
Strong simple intersection equivalence is certainly stronger than simple intersection equivalence as the counterexamples in the previous section show.
QUESTION 7.3. Is is true that if g; g0 2 CrvS have g ssi g0 , then g h g0 ?
It is not clear what the answer to this question should be. The initial problem
encountered is that there seems to be no real way to check that a pair of curves
are strongly simple intersection equivalent without assuming that they are hyperbolically equivalent.
In particular, we pose the following
QUESTION 7.4. Given a pair of curves, g; g0 2 CrvS, is there an algorithm to
decide whether or not g ssi g0 ?

176

CHRISTOPHER J. LEININGER

We have not touched on any other possible characterizations of hyperbolic

equivalence, and we have also not mentioned anything about the higher dimensional
case (in particular, the 3-dimensional case [23]). For more on this see [14, 28], and
particularly Conjecture 4.1 of [3].

Acknowledgements
I would like to thank Jim Anderson, Andrew Casson and my advisor Alan Reid for
useful conversations.

References
1. Abikoff, W.: The real analytic structure of Teichmuller space, Springer Lect. Notes Math.
820 (1980).
2. Abraham, R.: Bumpy metrics, In: Global Analysis, Proc. Sympos. Pure Math. 14, Amer.
Math. Soc., Providence, R.I., 1970, pp. 13.
3. Anderson, J.: Variations on a theme of Horowitz, to appear, Proceedings of the Workshop
on Kleinian Groups and Hyperbolic 3-manifolds, Warwick, September 2001.
4. Basmajian, A.: Selected problems in Kleinian groups and hyperbolic geometry, Contemp.
Math. 211 (1997), 915.
5. Bridson, M. and Haeiger, A.: Metric Spaces of Non-Positive Curvature, Springer-Verlag
Berlin, 1999.
6. Bleiler, S. and Casson, A.: Automorphisms of Surfaces after Neilsen and Thurston,
Cambridge University Press.
7. Bonahon, F.: Bouts des Varietes hyperboliques de dimension 3, Ann. Math. 124 (1986),
71158.
8. Bonahon, F.: The geometry of Teichmuller space via geodesic currents, Invent. Math. 92
(1988), 139162.
9. Cooper, D., Hodgson, C. and Kerckhoff, S.: Three-dimensional orbifolds and cone-manifolds, Math. Soc. Japan, Tokyo, 2000. x170 pp.
10. Culler, M. and Shalen, P. B.: Varieties of gronp representations and splittings of 3-manifolds, Ann. Math. 117 (1983), 109146.
11. Farkas, H. and Kra, I.: Riemann Surfaces, Springer-Verlag New York, 1992.
12. Fathi, A., Laudenbach, F., and Poenaru, V. et al.: Travaux de Thurston sur les surfaces,
Asterisque 6667 (1979), 1284.
13. Gardiner, F.: Teichmuller Theory and Quadratic Differentials, Wiley, 1987.
14. Ginzburg, D. and Rudnick, Z.: Stable multiplicities in the length spectrum or Riemann
surfaces, Israel J. Math. 104 (1998), 129144.
15. Horowitz, R.: Characters of free groups represented in the two-dimensional special linear
group, Comm. Pure Appl. Math. 25 (1972), 635649.
16. Hubbard, J. and Masur, H.: Quadratic differentials and foliations, Acta Math. 142 (1979),
221274.
17. Imayoshi, Y. and Taniguchi, M.: An Introduction to Teichmuller Spaces, Springer, Tokyo,
1992.
18. Kerckhoff, S. P.: The asymptotic geometry of Teichmuller space, Topology 19 (1980),
2341.
19. Kerckhoff, S. P.: Earthquakes are analytic, Comment. Math. Helvetici 60 (1985),
1730.

EQUIVALENT CURVES IN SURFACES

177

20. Levitt, G.: Foliations and laminations on hyperbolic surfaces, Topology 22 (1983),
119135.
21. Lyndon, R. C. and Schupp, R. E.: Combinatorial Group Theory, Springer-Verlag, Berlin,
1977.
22. Marden, A.: The geometry of nitely generated Kleinian groups, Ann. Math. 99 (1974),
383461.
23. Masters, J. D.: Length multiplicities of hyperbolic 3-manifolds, Israel J. Math. 119 (2000),
928.
24. Penner, R. C. and Harer, J. L.: Combinatorics of Train Tracks, Princeton University Press,
Princeton NJ, 1992.
25. Randol, B.: The length spectrum of a Riemann surface is always of unbounded multiplicity, Proc. Amer. Math. Soc. 78 (1980), 455456.
26. Rapinchuk, A. S., Benyash-Krivetz, V. V. and Chernousov, V. I.: Representation varieties
of the fundamental groups of compact orientable surfaces, Israel J. Math. 93 (1996),
29z71.
27. Ratcliffe, J.: Foundations of Hyperbolic Manifolds, Springer-Verlag New York, Inc., 1994.
28. Schmutz Schaller, P.: Geometry of Riemann surfaces based on closed geodesics, Bull.
Amer. Math. Soc. 35 (1998), 193214.
29. Shafarevich, I. R.: Basic Algebraic Geometry 1, Varieties in Projective Space, SpringerVerlag, Berlin, 1994.
30. Strebel, K.: Quadratic Differentials, Springer-Verlag Berlin Heidelberg, 1984.
31. Thurston, W. P.: The Geometry and Topology of 3-manifolds, Princeton University
mimeographed notes (1979).
32. Thurston, W. P.: On the geometry and dynamics of diffeomorphisms of surfaces, Bull.
Amer. Math. Soc. (N.S.) 19 (1998), 417431.