Download as pdf or txt
Download as pdf or txt
You are on page 1of 88

TIGHT AND EFFICIENT GEODESICS IN THE CURVE

COMPLEX
An abstract of
a Thesis
Presented to the faculty of
the Department of Mathematics
Western Illinois University

In Partial Fulfillment
of the Requirements for the Degree
Master of Science

By

MING ZHU WANG

December 2016

ABSTRACT

The curve complex of a surface is a simplicial complex with vertices denoting


isotopy classes of closed loops (henceforth referred to as curves) on the surface. Two
vertices are joined by an edge if their corresponding curves are disjoint, and the
distance between two curves on the surface is then defined to be the length of the
geodesic (shortest) edge path connecting the corresponding vertices.
In general between any two fixed vertices there is an infinite set of geodesics,
but two finite subsets of geodesics have been identified, namely tight geodesics by
Masur and Minsky and efficient geodesics by Birman, Margalit and Menasco. The
latter group of researchers have shown that these two finite subsets do not precisely
coincide; specifically, via a finite list of examples they show there exist geodesics that
are (1) efficient and tight, (2) tight but not efficient, and also (3) efficient but not tight.
We show that in fact there are infinitely many examples of geodesics satisfying these
three conditions, and prove a series of propositions whereby these infinite families can
be generated.

TIGHT AND EFFICIENT GEODESICS IN THE CURVE


COMPLEX

A Thesis
Presented to the faculty of
the Department of Mathematics
Western Illinois University

In Partial Fulfillment
of the Requirements for the Degree
Master of Science

By

MING ZHU WANG

December 2016

Acknowledgments
First and foremost, I would like to express my sincere gratitude to my advisor Dr.
Douglas LaFountain for his guidance and support of my first attempt at writing any
sort of mathematical literature. His patience, encouragement, and enthusiasm has
contributed tremendously to my understanding and enjoyment for the subject of this
thesis.
I would also like to thank the members of my thesis committee: Dr. Jason Blackford and Dr. Seyfi Turkelli, for taking the time to read through my work with Dr.
LaFountain.
My sincere gratitude also goes out to Dr. Dinesh Ekanayake for his constant
support and ever-helpful advice, as well to Dr. Clifton Ealy for sparking my real
interest in the subject of pure mathematics.
A hearty thanks goes out to my office mates (past and present) in MG 311B:
Mostak Ahammed, Alicia Culbertson, Celestine Enikanlogbon, Julie Herek, Jeremy
Kettering, Shaikh Obaidullah, Gloria Nnabuike, Mitch Riley, Grace Qiu, Patrick
Snyder, Jake Winters, and Jerry Young, for all the memorable conversations, the lastminute study sessions, and even the semi-heated political debates. I am so blessed to
have met all of you.
Last but not least, I would like to thank those who are the dearest to me: Mom,
Dad, and Patrick. Thank you so much for everything.

ii

Contents
Acknowledgments

ii

List of figures

viii

1 INTRODUCTION

2 CLOSED SURFACES AND CURVES

2.1

Closed surfaces of genus g > 1 . . . . . . . . . . . . . . . . . . . . . .

2.2

Covering spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

Simple closed curves . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4

Geometric intersection number . . . . . . . . . . . . . . . . . . . . . .

11

2.5

Mapping class group . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

2.6

Dehn twists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.7

Punctured sphere with marked points . . . . . . . . . . . . . . . . . .

19

2.8

Importance of families of disjoint curves on closed surfaces . . . . . .

23

3 THE CURVE COMPLEX

26

3.1

Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

3.2

Path-connectedness of the curve complex . . . . . . . . . . . . . . . .

27

3.3

Geodesics and metric in the curve complex . . . . . . . . . . . . . . .

29

3.4

Local infinitude of the curve complex . . . . . . . . . . . . . . . . . .

33

iii

3.5

Overview of the curve complex in geometry . . . . . . . . . . . . . . .

4 TIGHT GEODESICS

35
36

4.1

Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

4.2

Finitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

4.3

Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

41

4.4

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

44

5 EFFICIENT GEODESICS

45

5.1

Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

5.2

Finitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

5.3

Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

52

5.4

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

57

6 NEW RESULTS

58

6.1

Dehn twists, tightness, and efficiency . . . . . . . . . . . . . . . . . .

58

6.2

Infinitely many tight and/or efficient geodesics of distance 3 . . . . .

68

Bibliography

76

iv

List of Figures
2.1

Two representations of an annulus . . . . . . . . . . . . . . . . . . . .

2.2

A Mobius strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

A surface of genus 2 . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.4

Construction of the torus . . . . . . . . . . . . . . . . . . . . . . . . .

2.5

R2 /Z2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.6

An octagon-tiling of the Poincare disk. . . . . . . . . . . . . . . . . .

2.7

Construction of the genus-2 surface . . . . . . . . . . . . . . . . . . .

2.8

p : R ! S1

2.9

is essential and simple,

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
is essential but not simple,

is simple but

not essential. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.10 Taking a straight line passing through two points on the integer lattice
to an essential simple closed curve on S1 .
2.11

and

. . . . . . . . . . . . . . .

10

are from the same isotopy class of simple closed curves . . .

10

2.12 1 is a non-separating curve, while

is separating . . . . . . . . . .

11

2.13 A bigon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

2.14 An innermost disk . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

2.15 (i) D d(D), (ii) d(D) D. . . . . . . . . . . . . . . . . . . . . . .

14

2.16 Intersection number is reduced by 2 . . . . . . . . . . . . . . . . . . .

15

2.17 A finite mapping class of order 3

. . . . . . . . . . . . . . . . . . . .

16

2.18 The screw map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

2.19 A Dehn twist about a curve . . . . . . . . . . . . . . . . . . . . . .

18

2.20 Locating the eight marked points on S3 . . . . . . . . . . . . . . . . .

19

2.21 Mapping from S3 to the punctured sphere with marked points . . . .

20

2.22 With the red dot as a marked point, both diamonds in the neighborhood of that marked point on the surface map to the same diamond
on the sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.23 Further clarification of the F map. . . . . . . . . . . . . . . . . . . .

22

2.24 Depiction of a Dehn twist on the sphere . . . . . . . . . . . . . . . .

24

3.1

Curves from distinct isotopy classes on S2 and the corresponding portion of C(S2 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

27

Closed regular neighborhood of two curves with one intersection is


isotopic to S1 with one boundary component . . . . . . . . . . . . . .

28

3.3

The two cases under consideration . . . . . . . . . . . . . . . . . . . .

29

3.4

dS (, ) = 2, where and

are distinct isotopy classes of curves . . .

30

3.5

Curves of distance 1 on S2 . . . . . . . . . . . . . . . . . . . . . . . .

30

3.6

Curves of distance 2 on S2 . . . . . . . . . . . . . . . . . . . . . . . .

31

3.7

A non-essential component of the surface . . . . . . . . . . . . . . . .

31

3.8

Curves of at least distance 3 on S2

. . . . . . . . . . . . . . . . . . .

32

3.9

Alternate representation of Figure 3.8 . . . . . . . . . . . . . . . . . .

32

3.10 (i)
1.

and

1,

(ii)
1.

0
1 , ...
1,

are from dierent isotopy classes but are all distance 1 from
0
1 , ...

each provide a unique path of distance 2 between 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34

4.1

A tight multigeodesic of distance 3 . . . . . . . . . . . . . . . . . . .

38

4.2

A possible configuration of v0 , w0 , and m1 2 M1 in minimal position .

39

4.3

Multiple parallel m1 arcs are replaced with one similarly parallel representative arc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi

40

4.4

A distance 3 multigeodesic that fails to be tight . . . . . . . . . . . .

42

4.5

Confirming non-tightness . . . . . . . . . . . . . . . . . . . . . . . . .

42

5.1

Here,

46

5.2

A possible element of S

is one of many reference arcs for the trio 0 , 1 , n . . . . . .


[0 [ n ] with all possible positions of 1

and reference arcs . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47

5.3

An efficient geodesic of distance 3 . . . . . . . . . . . . . . . . . . . .

50

5.4

A non-efficient geodesic of distance 3 . . . . . . . . . . . . . . . . . .

53

5.5

A surgery on 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

56

5.6

Isotoping 10 on the sphere . . . . . . . . . . . . . . . . . . . . . . . .

56

5.7

An efficient geodesic of distance 3 . . . . . . . . . . . . . . . . . . . .

57

6.1

The sole intersection of

59

6.2

Left and right Dehn twists of 0 about

. . . . . . . . . . . . . . . .

59

6.3

An arc of cobounds a bigon with the negative Dehn twist . . . . . .

60

6.4

The circled portion of must exit the triangular region . . . . . . . .

60

6.5

(i) The existing polygon is split into two smaller polygons by T (0 ).

and 0 . . . . . . . . . . . . . . . . . . . .

(ii) Reference arcs of the smaller polygons are subarcs of those from
the initial polygon . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

6.6

Two variations of Case 1 . . . . . . . . . . . . . . . . . . . . . . . . .

64

6.7

A demonstration of Case 2 . . . . . . . . . . . . . . . . . . . . . . . .

65

6.8

Here, the positioning of

could create four intersections of 1 with a

reference arc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.9

This positioning of

67

avoids opening up the problematic polygon seen

in Figure 6.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

68

6.10 Generating infinitely many tight but not efficient geodesics of distance 3 70
6.11 Extension of the tight but not efficient example seen in Figure 6.10 to S3 71
6.12 Generating infinitely many efficient but not tight geodesics of distance 3 72
vii

6.13 Generating infinitely many tight and efficient geodesics of distance 3 .

74

6.14 Generalizing the tight and efficient example seen in Figure 6.13 to S2

75

viii

Chapter 1
INTRODUCTION
The curve complex of a surface contains vertices denoting isotopy classes of closed
loops, henceforth referred to as curves, on the surface. An edge joins two vertices if
their corresponding curves are disjoint, and the distance between two curves is then
defined as the length of the geodesic edge path joining their corresponding vertices.
In general, between any two fixed vertices there is an infinite set of geodesics, but
two finite subsets of geodesics have been identified, namely tight geodesics of MasurMinsky [9], and efficient geodesics of Birman-Margalit-Menasco [1]. The latter group
of researchers have shown that these two finite subsets do not precisely coincide via
a finite list of examples which exhibit tight and efficient geodesics, tight (but not
efficient) geodesics, and efficient (but not tight) geodesics. The goal of this thesis is
to show that in fact there are infinitely many geodesics of distance 3 in each of the
three categories, and present a series of propositions whereby these infinite families
can be generated.
Before we proceed, we recall some basic definitions for terms that will be used
freely throughout this thesis.
A homeomorphism is a bijective continuous map between topological spaces whose
inverse is also continuous. If the topological space is a dierentiable manifold, we will
1

2
typically have homeomorphisms that are smooth, which we refer to as dieomorphisms.
A homotopy is a continuous one-parameter family of continuous functions. We
say that two continuous functions from one space to another are homotopic if one
can be continuously deformed into the other. To be precise, let f and g be functions
mapping from topological spaces X to Y . We define a homotopy to be a continuous
map:

H : X [0, 1] ! Y
where H(x, 0) = f (x) and H(x, 1) = g(x). With time t as the parameter, the function
f at t = 0 is continuously deformed by H into function g by t = 1.
An isotopy is a continuous one-parameter family of homeomorphisms; a smooth
isotopy is a smooth one-parameter family of dieomorphisms. For the above homotopy H to be an isotopy, we require that for each fixed t, H(x, t) is a homeomorphism
onto its image.
This thesis will proceed as follows: In Chapter 2, we introduce essential simple
closed curves on closed surfaces and prove the Bigon criterion, which gives a simple
method of determining whether curves are in minimal position. In Chapter 3, we
discuss the curve complex of a surface and define the metric on which distance between
isotopy classes of curves can be realized. In Chapters 4 and 5, we prove finitude and
existence properties of tight geodesics and efficient geodesics. Chapter 6 contains our
new results: a series propositions by which infinitely many distance 3 geodesics of the
aforementioned categories can be generated.

Chapter 2
CLOSED SURFACES AND
CURVES
When imagining a curve, perhaps many of us instinctively picture the graph of a
function in the Euclidean plane; we tend not to think of a loop on the inner ring of a
torus. However, in this chapter it will be such latter closed curves on closed surfaces
which will ultimately interest us. In this chapter, we will also see how the Euclidean
plane is used to construct a torus. More interestingly, we will see how surfaces of
higher genus are constructed from the hyperbolic plane.
We will also see that curves that are not in minimal position, and which cross in
extraneous places, can be pulled apart to reveal that they in fact can intersect fewer
places. We will prove a key result, the Bigon criterion, to allow for a simple method
of detecting when two curves are not in minimal position.
Finally, we will introduce tools that allow us to manipulate existing surfaces and
curves, along with a simpler representation of curves on a surface.

2.1

Closed surfaces of genus g > 1

One of our primary objects of study in this thesis will be two-dimensional manifolds
locally resembling Euclidean 2-space, or surfaces. We will typically think of surfaces
as having a dierential structure, thereby implying the existence of tangent planes
and smooth mappings. This allows for well-defined notions of transversality, and
applications of Sards Theorem such as general position.
At a preliminary stage, it is important to clearly distinguish between a surface and
the ambient 3-D space in which we depict it. For our purposes, the ambient 3-D space
surrounding a surface is of no particular importance; we will only consider points on
the surface itself. We may at times consider surfaces with or without boundary. In the
case of a surface with boundary, note that any boundary components will necessarily
be 1-manifolds.

Figure 2.1: Two representations of an annulus


We say that a surface is closed if it is compact and without boundary. Each point
belonging to a closed surface has a neighborhood homeomorphic to the open ball in
R2 . It is easy to see that the annulus (Figure 2.1), S 1 [0, 1], is not a closed surface,
since it has two boundary components. On the other hand, Figure 2.3 shows a closed
surface without 1-manifold boundary components.

Figure 2.2: A Mobius strip

5
If there is a coherent continuous choice of a basis to the tangent plane at each
point of a surface, then the surface is said to be orientable. Loosely, a surface is nonorientable if one can trace along some path on it that returns to the original point, but
with the basis reflected in the tangent plane. A classic example of a non-orientable
surface is the Mobius strip (Figure 2.2).
The genus of a surface is the least possible number of disjoint, essential, nonperipheral curves needed to cut the surface into a connected, planar surface (with
the additional proviso that the sphere is a genus-0 surface). In this thesis, we will be
primarily concerned with closed orientable surfaces of genus g > 1. As a convention,
the notation Sg will be used to denote such a closed orientable surface of genus g.
It is helpful to picture a surface of genus g as a sphere with g handle attachments,
or the connected sum of g tori (plural of torus, a genus-1 surface). The connected
sum of two surfaces is created by removing an open disk from both surfaces and
identifying the two boundaries. Figure 2.3 is a depiction of S2 .

Figure 2.3: A surface of genus 2


While we will be primarily concerned with the topology of curves on surfaces in
the absence of a specific metric, there are instances where the geometry of a surface is
useful. To begin to incorporate geometry into our discussion, first consider an explicit
construction of a genus-1 surface by appropriately identifying pairs of edges of a unit
square (Figure 2.4). Observe that we can use a Euclidean square with right angles to
achieve this, since at the point on the torus where the angles meet, there is exactly
2
4

angle per corner of the square.

Perhaps more precisely, by tiling the two-dimensional Euclidean space with unit

Figure 2.4: Construction of the torus


squares, it is readily observed that the torus (T2 ) can be realized as a quotient space
of R2 (Figure 2.5). Specifically, the quotient space R2 /Z2 represents all equivalence
classes of points in R2 where two points (x1 , y1 ) and (x2 , y2 ) are equivalent if x1
and y1

x2

y2 are integers. The torus naturally inherits a 2-D Euclidean geometry from

this construction.

Figure 2.5: R2 /Z2


Similar to how R2 is tiled by the unit square, the P oincar
e disk H2 is tiled
by the hyperbolic octagon (Figure 2.6). One may construct a genus-2 surface in a
similar fashion as the torus by appropriately identifying pairs of edges of an octagon

7
(Figure 2.7). Notice that at the central point on the genus 2 surface where the
octagons corners meet, there is

2
8

angle per corner. Therefore, such an octagon

must be hyperbolic with each interior angle

,
4

thereby equipping genus-2 surfaces

with hyperbolic geometry. In general, a genus g > 1 surface is constructed by taking


a regular 4g-gon in hyperbolic space with interior angle sum of 2 and identifying
appropriate sides. Similar to how a genus-1 surface can be obtained from a quotient
map from R2 to T2 , any surface of genus g > 1 is obtainable from a quotient map
p : H 2 ! Sg .

Figure 2.6: An octagon-tiling of the Poincare disk.

Figure 2.7: Construction of the genus-2 surface

2.2

Covering spaces

S is a covering space of a surface S if it maps onto S in a locally homeomorphic way.


More precisely, We define a covering map to be
p : S ! S
where there is an open neighborhood Ux for each x 2 S such that the inverse image
each of which is mapped
of Ux can be written as a disjoint union of open sets in S,
homeomorphically by p onto Ux . For each x 2 S, its inverse image in S is a discrete
space, which we call the fiber over x.
A deck transformation of a covering map is a homeomorphism of the covering
space that permutes the elements of each fiber, and is uniquely determined by its
action on one point. In particular, only the identity deck transformation can fix a
point.

Figure 2.8: p : R ! S 1
As an example, consider the covering map for S = R and S = S 1 defined by
p(t) = eit , where S 1 is the unit circle (Figure 2.8). For a point x 2 S 1 , let Ax be a
small open arc of S 1 containing x. The inverse image of Ax is then a collection of
open intervals Ux1 , Ux2 , ... in R such that for each Uxi , there exists a point xi 2 Uxi
such that p(xi ) = x and |xj

xk | = 2n for some integer n. Intuitively, p is a map

that wraps the real line around the unit circle, and the inverse image of a small arc
of S 1 is a collection of open intervals in R that are 2 apart.
Notice that by definition of T2 and Sg>1 as quotient spaces from the previous
section, both R2 and H2 are covering spaces of T2 and Sg>1 , respectively. In particular,

9
the latter covering map p : H2 ! Sg>1 will be important shortly.

2.3

Simple closed curves

A closed curve in a surface S is a smooth mapping of a circle into S, that is, a smooth
curve whose starting and ending points coincide. If a closed curve does not intersect
itself, then it is said to be simple. If it is not homotopic to a point or boundary
component of S, then the closed curve is said to be essential. For our purposes, any
mention of a curve is implied to be an essential, simple closed curve. Examples of
closed curves can be seen in Figure 2.9.

Figure 2.9: is essential and simple,


not essential.

is essential but not simple,

is simple but

Observe that simple closed curves in Sg lift to infinite non-compact curves in its
covering space. An example can be seen for the covering map p : R2 ! T2 . Consider
a straight line passing through (0, 0) to (p, q) in R2 where p and q are integers.
Projecting onto R2 /Z2 yields an essential simple closed curve on a genus-1 surface
(Figure 2.10).
We say that two simple closed curves are from the same isotopy class of curves if
one curve can be isotoped to the other. For example, the two curves seen in Figure
2.11 are from the same isotopy class as
with

2.

can be pushed along the surface to overlap

However, the three curves seen in Figure 2.9 are each from a dierent isotopy

class as none of the three curves can be pushed along the surface to overlap with any
of the other two curves.

10

Figure 2.10: Taking a straight line passing through two points on the integer lattice
to an essential simple closed curve on S1 .

Figure 2.11:

and

are from the same isotopy class of simple closed curves

11
An additional characteristic of any essential simple closed curve is whether it
is separating or non-separating. If cutting along a curve on the surface yields two
subsurfaces, we say that the curve is separating. Otherwise, it is non-separating. In
Figure 2.12, 1 is a non-separating curve, while
along

is separating. In particular, cutting

separates S2 into two genus 1 surfaces with one boundary component each,

while cutting along 1 yields one genus 1 surface with two boundary components.

Figure 2.12: 1 is a non-separating curve, while

2.4

is separating

Geometric intersection number

Given two isotopy classes a and b of curves in a surface, the geometric intersection
number of a and b, denoted i(a, b), is the minimum number of intersection points
between a representative of a and a representative of b. We may assume that any
intersections of curves are transverse and refer to such curves as transverse curves.
Notice that i(a, b) = i(b, a) and i(a, a) = 0. Representative curves achieving the
geometric intersection number are said to be in minimal position.
It will often be of interest to us to be able to tell when two curves are in minimal
position. To this end, consider the case where two curves cobound a bigon (Figure
2.13), which is defined as a disk with two edges and vertices, the two edges coming
from subarcs of the two curves, respectively. It is readily noticed that if two curves
cobound a bigon, then there is a natural isotopy of the curves through the disk to
reduce their intersection number by 2.

12

Figure 2.13: A bigon


In fact, the converse is also true; its proof is based largely on material found
in Farb and Margalits A Primer on Mapping Class Groups [4], which has been an
excellent guide in aiding the composition of this section.
Theorem 2.4.1. (Bigon Criterion) Let and
Then, and

be transverse curves in a surface S.

are in minimal position if and only if they do not form a bigon.

A lemma is necessary before presenting the proof of Theorem 2.4.1.


Lemma 2.4.1. If transverse curves and

in a surface S do not form bigons, then

their lifts,
and in H2 , intersect at most once.
Proof. Let p : H2 ! S be a covering map. We will show the contrapositive, that if

and in H2 intersect at least twice, then and

will form at least one bigon in S.

Let
and intersect at least twice in H2 . Then, there must be subarcs of
and
that bound disks in H2 . Out of all possible such disks, at least one is innermost in
the sense that it has no arc of
or passing through its interior (Figure 2.14). This
can be seen by noticing that, by compactness and transversality, the intersection of
either
or with each disk bound by subarcs of
and will have finitely many
line segment components. We will refer to one such innermost disk as D, and its
boundary arcs as 1
and 1 .
If we can show that p maps both D and the boundary components of D injectively
into S, then we are done.
We will first show that p is one-to-one on the boundary components of D. To this
end, let v1 and v2 be the points where 1 and 1 intersect. Notice that p(v1 ) 6= p(v2 )

13

Figure 2.14: An innermost disk


since the two intersections are of opposite orientations. Next, let p1 and p2 be points
of 1 and 1 , respectively, where at least one is distinct from v1 and v2 . If p(p1 ) =
p(p2 ) = p, then and

would intersect transversely at p. This implies that 1 and

1 would intersect at p1 and p2 . However, this is not possible since 1 and 1 bound
an innermost disk. Therefore p(p1 ) 6= p(p2 ).
Alternatively, let q1 and q2 be distinct points of 1 , where again at least one is
distinct from v1 and v2 . If p(q1 ) = p(q2 ), then since p is a covering map there exists a
point q3 located in between q1 and q2 such that p(q3 ) = p(v1 ). Then, since v1 2
\ ,
we must have that q3 2
\ . However, by the same innermost disk and transverse
argument, such a point q3 cannot exist. Therefore p(q1 ) 6= p(q2 ).
This covers all conditions to show that p is one-to-one on the boundary components of D.
Now, we will show that p maps D injectively into S.
Let x, y be two points of D such that p(x) = p(y). Then, there is a deck transformation d : H2 ! H2 such that d(x) = y. If we can show that d must be the identity
map, then it will follow that p is injective on D.
Assume d is not the identity map. Denote the boundary of D by @D. Then, since
p is one-to-one on @D, and the deck transformation permutes the elements of fibers
of points in S, the intersection of @D and its deck transformation d(@D) must be
empty. Moreover, d(x) = y implies that the intersection of D and its image under
the deck transformation d(D) must not be empty, as x and y are points in D. This

14
restricts the action of d on D to either d(D) D or D d(D) (Figure 2.15).

Figure 2.15: (i) D d(D), (ii) d(D) D.


Without loss of generality, assume d(D) D. Then, since D is homeomorphic to
a closed disk, the Brouwer Fixed-Point theorem can be applied to conclude that there
exists a point x0 2 D such that d(x0 ) = x0 . Since a deck transformation is uniquely
determined by its action on one point, the existence of a fixed point implies that d
must be the identity map. It follows that p is injective on D.

We are now free to use Lemma 2.4.1 to prove Theorem 2.4.1.


Proof of the Bigon Criterion. Assume and

form a bigon. Then an isotopy of

can be defined to reduce the number of intersections of with


that and

by 2, which implies

were not in minimal position.

We will prove the contrapositive of the other direction, namely that if and
are not in minimal position, then they form a bigon.
Assume and

are not in minimal position. If we can show that


and intersect

at least twice in H2 , then by applying Lemma 2.4.1 we will have shown that and
form a bigon.
As and

are not in minimal position, there is an isotopy of that reduces

intersections with . Let h : S 1 [0, 1] ! S be this isotopy, where h( , 0) maps to


and h( , 1) maps to the isotoped , which transversely intersects
than did.

in fewer points

15
Notice that h 1 ( ) in S 1 [0, 1] consists of properly embedded arcs. In particular,
the endpoints of proper arcs are on the boundary of S 1 [0, 1]. Since and

are

not in minimal position, there is at least one point of intersection, and therefore at
least one endpoint of an arc belonging to h 1 ( ) must be on S 1 {0}. If all such
arcs have its other endpoint on S 1 {1} then the number of intersections would not
be reduced; therefore, there must exist some properly embedded arc of h 1 ( ) with
both endpoints on S 1 {0}. Call this arc c1 (Figure 2.16).

Figure 2.16: Intersection number is reduced by 2


Let c2 be the part of S 1 {0} that cobounds a disk D1 together with c1 . The
mapping h restricted from D1 to S is continuous, which implies that h(c1 [ c2 ) is
homotopic to a point in S and lifts to a closed curve in H2 .
Then, for the covering map p : H2 ! S, we will have p 1 (c1 ) and p 1 (c2 ) .
Since c1 and c2 share two points in S, their lifts in S will intersect at least twice.
Therefore, by Lemma 2.4.1, and

form a bigon in S.

We will borrow an important lemma from [4] without proof:


Lemma 2.4.2. Any three distinct curves on a surface can be isotoped to be in mutually
minimal position.

16
All intersections from this point forward are assumed to be essential, in the sense
that no bigons are formed.

2.5

Mapping class group

Consider a surface S and the set of homeomorphisms from S to itself. The mapping
class group of a surface S is the group of isotopy classes of orientation-preserving
homeomorphisms from S to itself that fix boundary points of a surface pointwise.
The group structure is the functional composition group structure of the set of homeomorphisms. We will denote the mapping class group of a surface S by M od(S) and
refer to its elements as mapping classes.
Often, M od(S) will be written as a quotient group:
Homeo+ (S)/Homeo0 (S)
where Homeo+ (S) is the group of orientation preserving homeomorphisms of S, and
Homeo0 (S) refers to the homeomorphisms of S that are isotopic to the identity
mapping.

Figure 2.17: A finite mapping class of order 3


The mapping class that rotates the surface shown in Figure 2.17 by an angle of
2
3

is non-trivial and of order 3.


As a complete example, we will consider the mapping class group for the torus:

M od(T2 ).

17
First, we will acquaint ourselves with the (p, q) notation of a curve on T2 . Recall
that T2 can be written as the quotient space R2 /Z2 , and as such, every curve on T2
is isotopic to a line segment with one endpoint at (0, 0), and another at (p, q) 2 Z2 .
Notice that (p, q) is a distinct curve only when gcd(p, q) = 1. For example,
the curve (4, 2) directly overlaps with the curve (2, 1). A homeomorphism of T2
must therefore take simple closed curves to simple closed curves, and hence lift to
a linear transformation which takes points (p, q) to (p0 , q 0 ) in Z2 , where gcd(p, q) =
gcd(p0 , q 0 ) = 1.
Such linear transformations are precisely SL(2, Z).

B 1 1 C
Thus, M od(T2 )
= SL(2, Z), which implies that M od(T2 ) is generated by @
A
0 1
0
1
B 1 0 C
and @
A, the generators of SL(2, Z). The matrices correspond to what we will
1 1
soon observe to be Dehn twists about (1, 0) and (0, 1), respectively.
We take a result from [4] as a fact:
Theorem 2.5.1. Any two non-separating curves on a surface S are related by some
element in the mapping class group. Any two separating curves that separate S into
two homeomorphic pairs of disjoint components, respectively, are related by some
element in the mapping class group
As a consequence, when we draw a simple non-separating curve, we will put it in
a position that is easily examined, and is unique up to the mapping class group.
At this point, we want to provide a way to understand the mapping class group
not just of T2 , but of Sg for all g > 1, using special homeomorphisms termed Dehn
twists.

18

2.6

Dehn twists

We begin by defining the screw map on the annulus A = S 1 [0, 1].

Figure 2.18: The screw map


Let T : A ! A be a homeomorphism of A defined by T(, t) = ( + 2t, t), which
can be intuitively perceived as fixing one boundary component of A while rotating
the other boundary component by 2, dragging the surface and its arcs along with
it (Figure 2.18). Note that the screw map fixes both boundary components of the
annulus.
Using the screw map, we can define the Dehn twist about a curve on a surface
S.
For a curve on a surface S, fix an annular neighborhood A and define

to be

an embedding of A onto A .
Let T : S ! S be a homeomorphism of S defined by T =

on A ,

and equal to the identity mapping everywhere else on S. Figure 2.19 depicts a Dehn
twist applied to the curve

about the curve on a genus-1 surface.

Figure 2.19: A Dehn twist about a curve


It is important to notice that the isotopy class of T is neither dependent on the
choice of A nor the choice of . In fact, if is isotopic to some other curve

on S,

19
then T is isotopic to T . The isotopy classes of Dehn twists about distinct curves
of S, up to isotopy, are well-defined elements of M od(S).
For example, M od(T2 ) is generated by Dehn twists about (1, 0) and (0, 1).
It is true, but not at all obvious, that for any genus g surface, its mapping class
group can be generated by finitely many Dehn twists about certain curves on the
surface. This is a key result of Dehn, which may be found in [3].

2.7

Punctured sphere with marked points

Until now, we have basically restricted our attention to elementary curves on genus
1 surfaces that can be represented visually with relative ease. However, difficulties
in visualization quickly arise when attempting to depict more intricate curves on
surfaces of higher genus. In this section, we will introduce an alternative, simplified
representation of curves on a genus g surface as line segments and loops on a plane,
where in fact the plane will be a once-punctured sphere (identified with the plane via
stereographic projection). Our understanding of this mapping is taken from [1].
To see how this works, as an example we take S3 , the genus 3 surface, and define
a total of eight marked points on S3 precisely at the intersection points of the surface
with the axis indicated in Figure 2.20.

Figure 2.20: Locating the eight marked points on S3


We then consider a branched covering map F from S3 onto the sphere that is twoto-one almost everywhere, except for being one-to-one at marked points. Specifically,
in a neighborhood of each marked point, the map resembles the mapping from z to

20
z 2 in a neighborhood of the origin in the complex plane (Figure 2.22). Observe that
the marked point is mapped to itself in a one-to-one manner, but the two diamonds
map to a single diamond in a two-to-one manner.
The totality of the map F can be further understood by its action on simple closed
curves as follows, and for this we refer the reader to both Figure 2.21 and 2.23.

Figure 2.21: Mapping from S3 to the punctured sphere with marked points
Specifically, Figure 2.21 shows on the left a purple curve on S3 which passes
through two marked points; the green curve does as well. The map F is two-to-one
away from the marked points and thus maps both the purple and green curves to
purple and green arcs connecting two marked points on the sphere. This is shown
in more detail on rows 1 and 3 of Figure 2.23. On the other hand, curves on S3
which do not contain marked points, such as the blue non-separating curves and the
orange separating curve, map to curves on the sphere. In the case of the disjoint blue
non-separating curves, they will each map to the same curve; and in the case of the
orange separating curve, it will wrap twice around the orange curve on the sphere.

21
This is shown in more detail in rows 2 and 4 of Figure 2.23.
The entire map F can therefore be understood by its analogous actions on analogous curves.

Figure 2.22: With the red dot as a marked point, both diamonds in the neighborhood
of that marked point on the surface map to the same diamond on the sphere
F in Figure 2.21 maps S3 to the punctured sphere. The map P punctures the
sphere and flattens the surface. The result is indeed that all curves on S3 appear as
either arcs joining two marked points, or loops encircling at least two marked points.
This example for S3 can be extended to surfaces of arbitrary genus g > 1 in the
obvious way, where Sg will map to a sphere with 2g + 2 marked points.
The following relationships are fundamental when studying curves using the sphere
with marked points representation:
(i) A loop encircling exactly two marked points on the once-punctured sphere is
isotopic to the line segment connecting those marked points. This can be seen by
observing that the neighborhood of a line segment connecting two marked points lifts
to an annulus on S, which has a unique isotopy class of core curve.
(ii) A line segment connecting two marked points on the once-punctured sphere
lifts to a non-separating curve in S3 .
(iii) A loop encircling 2k + 1, k

1, marked points on the sphere lifts to a

separating curve in S3 .
(iv) A subdisk on the sphere with two or more marked points in its interior lifts
to an essential component on S3 .

22

Figure 2.23: Further clarification of the F map.

23
(v) A subdisk on the sphere with one or no marked points lifts to a (non-essential)
polygon.
(vi) Two line segments sharing a marked point on the sphere lifts to two essentially
intersecting curves on the surface
With the above knowledge, we can picture a Dehn twist of one curve about another
on the punctured sphere with marked points.
Referring to Figure 2.24, we show two essentially intersecting curves on the surface
as two line segments and three marked points down on the sphere, where the line
segments share one marked point on one end and have individual marked points on
each of their other ends. Therefore these two curves intersect just once on Sg for this
example. Call the curves and

as indicated.

In performing a Dehn twist of one curve about the other, the original curve will
pass its original docking point, swing around to the outer endpoint of the curve it is
twisting about, and dock there. This depiction of the Dehn twist will become highly
relevant in our new results.

2.8

Importance of families of disjoint curves on


closed surfaces

For the interested reader, we briefly discuss a few reasons why disjoint curves on
closed surfaces are of interest in geometric topology.
We have briefly observed the relationship between the mapping class group of a
surface and how its curves are aected by a homeomorphism of the surface. While
some mapping classes are periodic, many others are not. Based on whether collections
of disjoint curves are fixed on the surface, the category of non-periodic mapping classes
is further divided into two subcategories: reducible mapping classes, which fix some
finite union of disjoint curves on the surface, and pseudo-Anosov mapping classes,

24

Figure 2.24: Depiction of a Dehn twist on the sphere

25
which do not fix any family of disjoint curves. However, pseudo-Anosov mapping
classes do fix a particular set on the surface called a geodesic lamination [2].
Another interesting use for families of disjoint curves on closed surfaces is constructing closed oriented 3-manifolds, which are spaces locally resembling 3-dimensional
Euclidean space. In fact, they can be used to construct any closed oriented 3-manifold
in the following way: Consider a genus g handlebody, which is the 3-dimensional solid
whose boundary is the genus g surface. The compressing discs of a genus g handlebody
are discs inside the handlebody such that the set of boundaries of the compressing
discs are disjoint closed curves that chop the surface into thrice-punctured spheres. In
particular, this set of disjoint boundary curves forms what is known as a pants decomposition, which contains precisely 3g
let 1 , ..., 3g

3 elements. For a given genus g handlebody,

be the boundary curves of its compressing discs, and let

1 , ...,

3g 3

be the curves that bound the compressing discs of another genus g handlebody. Then
by taking a non-trivial mapping class
of one handlebody to another so that

from Sg to itself, we can glue the surfaces


i

= (i ) for all i = 1, ..., 3g

3. The result

of the gluing is a 3-manifold. The decomposition of the 3-manifold back into two
handlebodies is called a Heegaard splitting of the 3-manifold. A wonderful fact in
3-manifold topology is that any 3-manifold admits a Heegaard splitting [10].
Finally, families of disjoint curves on closed surfaces may be used to parametrize
Teichm
uller space, the space of hyperbolic structures on a surface. In particular, the
Frenchel-Nielsen coordinates for a point of the Teichm
uller space consist of 3g
positive real numbers called lengths, and 3g

3 positive real numbers called twists.

The lengths are simply the lengths of the closed curves in the pants decomposition of
Sg , while the twists are the degrees of rotation between glued pairs of thrice-punctured
spheres.

Chapter 3
THE CURVE COMPLEX
Given a set of isotopy classes of curves, a natural question arises of how to define a
metric within such a set. To this end, we will define a new simplicial complex, within
which each isotopy class of curves is a vertex, and conditions of intersections and
disjointness between isotopy class representatives dictate where edges may be placed.
One result that we will observe is that given any two isotopy classes of curves, we
can find a path of edges connecting them. Additionally, we will demonstrate a key
property of higher distance curves, which is that cutting along curves of distance
greater than or equal to 3 will chop the entire surface up into finitely many polygons.

3.1

Definition

The curve complex of a surface S, denoted C(S), is a simplicial complex whose vertices
are isotopy classes of curves in S. A collection of k + 1 vertices form a k-simplex
in C(S) if and only if there exist a representative from each isotopy class such that
they are mutually disjoint. Since vertices in C(S) represent curves in S, we will often
conflate the two, and speak of a curve v.
Specifically, an edge joins two vertices v1 and v2 if and only if i(v1 , v2 ) = 0. We will
26

27
often focus our attention on the 1-skeleton of C(S), which is given by the collection
of its edges and vertices.

Figure 3.1: Curves from distinct isotopy classes on S2 and the corresponding portion
of C(S2 ).
As an example, refer to Figure 3.1. Observe that the only curves that are not mutually disjoint are those represented in purple and green. Therefore the corresponding
portion of the curve complex contains four vertices, one for each distinct isotopy class
of curves, with an edge between each pair except those of the purple and green curves.
The two triangles indicated can be filled in by 2-simplices in the curve complex, since
the three curves represent their vertices are mutually disjoint.

3.2

Path-connectedness of the curve complex

For nearly all surfaces Sg , we will find C(Sg ) to be connected in the sense that given
any pair of vertices in the curve complex, there exists an edge path from one vertex
to the other.
By the following theorem, all surfaces of genus g > 1 are shown to have a connected
curve complex. We borrow inspiration for this proof from Masur and Minskys work
on the complex of curves [8].
Theorem 3.2.1. If g > 1, then C(Sg ) is connected.
Proof. We will show that if g > 1, and and

are two vertices in C(Sg ), then there

exists a sequence of vertices { = v0 , v1 , ..., vk = } in C(Sg ) such that i(vi , vi+1 ) = 0,


for all i 2 {0, 1, ..., k

1}. The proof will be based on an inductive argument.

28
If i(, ) = 0, then we are done. Thus, suppose i(, ) = 1. Choosing a representative pair from and

that are in minimal position, we can take a closed regular

neighborhood of [ . Call this neighborhood N , and notice that it is homeomorphic


to a torus with one boundary component (Figure 3.2).

Figure 3.2: Closed regular neighborhood of two curves with one intersection is isotopic
to S1 with one boundary component
Let

be the isotopy class of the boundary component of N . If

is not essential,

then S must be a torus, which is not a surface under consideration. Therefore

is

an essential curve, which implies that its isotopy class must be a vertex in C(Sg ).
Since N is a neighborhood of and ,

must be disjoint from both and . Thus

we have found a path from to , namely: !

! . This therefore concludes

the base case i(, ) = 1.


Now suppose i(, )

2. If we can find an isotopy class

such that i(, ) <

i(, ) and i( , ) < i(, ), then by induction there exists a path from to
path from

and a

to , and therefore there exists a path from to .

To find such a , first consider a 2 and b 2


position. For i(, )

such that a and b are in minimal

2, we can look at two consecutive intersections of a with b,

which will look one of two ways:


(i) a intersects b with the same orientation each time
(ii) a intersects b with dierent orientations each time
Consider case (i). Define c 2

to be the curve portrayed in Figure 3.3 (i);

namely, c is isotopic to the union of an arc of a and an arc of b as indicated. It


is evident that i(, ) = 1. Moreover, we claim that c must be essential. If c is
not essential, then c bounds a disk, which would be isotopic to a bigon cobounded

29

Figure 3.3: The two cases under consideration


by an arc of a and an arc of b. This would imply that a and b are not in minimal
position, which is a contradiction. Therefore, we have that c is an essential curve and
i(, ) < i(, ). Additionally, since c will cross b once less than a crosses b, it follows
that i( , ) < i(, ). Thus for case (i), this is the desired .
Now, consider case (ii). Define c 2

to be the curve portrayed in Figure 3.3 (ii),

where again c is isotopic to the union of an arc of a and an arc of b as indicated.


Again, c cannot be null-homotopic, else a and b form a bigon. Therefore c is an
essential curve. It is evident that i(, ) < i(, ), and i( , ) < i(, ); the former
due to i(, ) = 0, and the latter due to the representative of

crossing b twice less

than a crosses b.
Therefore, by induction we see that the curve complex must be connected for
Sg>1 .

3.3

Geodesics and metric in the curve complex

At this point, we consider the fundamental question of defining the distance between
two curves in a given surface S. Intuitively, two curves from the same isotopy class
are distance-0 apart. Distance between curves in S as a metric in C(S) will be defined
in the following way:

30
Let C0 (S) be the set of all vertices in C(S). Recall that each vertex in C(S) is a
distinct isotopy class of curves on S.
Let dS : C0 (S) C0 (S) ! Z

be the function that defines the distance between

two isotopy classes of curves on S, where dS (v1 , v2 ) is the minimum number of edges
that form a path connecting v1 and v2 in C1 (S). See Figure 3.4 for an example.

Figure 3.4: dS (, ) = 2, where and

are distinct isotopy classes of curves

Since all Sg such that g > 1 have a connected curve complex, dS (v1 , v2 ) is defined
for any two vertices v1 , v2 in C(Sg ). Additionally, the reflexive, symmetric, and triangle
inequality properties of a metric are all satisfied. Notice that each edge path that
defines distance is a geodesic in the sense that the edge path is minimal. Additionally,
i(v1 , v2 ) = 0 directly implies dS (v1 , v2 ) = 1.
Isotopy classes of curves satisfying higher distances become increasingly difficult
to find. As an example, consider the following increasing list of requirements to be
fulfilled as we attempt to locate curves of distance 1, 2, and 3 on a surface S.
In Figure 3.5 we see representatives of two isotopy classes and

that satisfy

the condition for dS (, ) = 1. The only requirement here is the existence of disjoint
representatives of the isotopy classes.

Figure 3.5: Curves of distance 1 on S2

31
In order for dS (, ) = 2, two requirements need to be satisfied: and
not have disjoint representatives, and there must exist a third isotopy class

must
with a

representative that is disjoint from representatives from both and . As Figure 3.6
shows, these conditions are not difficult to satisfy in S.

Figure 3.6: Curves of distance 2 on S2


It is not immediately obvious how to identify curves of distance 3 on S. However,
based on the requirements for dS (, ) = 2 an important conclusion can be made for
dS (, )

3:

Curves possessing distance greater than two not only lack disjoint representatives,
but also lack a third isotopy class with the representative that is disjoint from both
and

representatives. This implies that given two curves of distinct isotopy classes,

one can conclude that they are at least distance 3 if and only if cutting along the curves
on the surface yields only non-essential components of the surface (i.e. polygons), such
as the one seen in Figure 3.7.

Figure 3.7: A non-essential component of the surface


Notice that in such a case, all curves disjoint from both isotopy classes are nonessential, and therefore do not exist as a vertex in C(S). Thus, if 0 2 and

32
chop the surface up into polygons, we say that 0 and
that 0 and

fill the surface, or simply

fill.

Figure 3.8 depicts two curves of distance at least 3 on S2 .

Figure 3.8: Curves of at least distance 3 on S2


While cutting along curves on a surface such as those in Figure 3.8 is definitely
tedious, the process is simplified significantly by performing the cut on the oncepunctured sphere with marked points representation of the surface. For this reason,
we will favor this alternative representation over the standard representation. As an
example, see Figure 3.9 for the alternate representation of Figure 3.8.

Figure 3.9: Alternate representation of Figure 3.8


To see that the two curves 1 and

of Figure 3.8 do in fact fill on the surface

S2 , we notice that cutting along the curves on the sphere with marked points representation seen in Figure 3.9 yields only disks and disks containing one marked point
each both of which lift to disks in S2 . Thus 1 and

indeed fill.

33

3.4

Local infinitude of the curve complex

Let S be a closed surface of genus g > 1. Given two vertices in C(S), one would like
to find the shortest path from one to the other. Although one would hope that an
algorithm may be written for this purpose, the locally infinite nature of the curve
complex prevents this from occurring naturally.
To be precise, C(S) is locally infinite in the sense that there are infinitely many
edges leaving any given vertex. Namely, by application of Dehn twists, infinitely
many dierent isotopy classes of curves may be found disjoint from any given curve
(Figure 3.10 (i)). As a result, it becomes possible to generate infinitely many ways to
get between curves of distance 2 (Figure 3.10 (ii)). Due to this unfortunate property,
there is no obvious way to determine a geodesic path between any two given curves
in S, thereby creating difficulties for those wishing to work within the curve complex.
In order to circumvent the issue of local infinitude, two sets of preferred geodesics
have already been identified: tight geodesics of Masur and Minsky [9], and efficient
geodesics of Birman, Margalit, and Menasco [1].
Each set is preferred in the sense that there exists at least one geodesic, and at
most finitely many such geodesics, between any two fixed vertices of C(S).
Our motivation for this thesis stems from an observation made by Birman, Margalit, and Menasco:
We do not know if between any two vertices there always exists a geodesic that is
efficient and tight.
For this thesis, the above question remains unanswered. However, in the following
chapters, we will take a closer look at tight and efficient geodesics, and ultimately
prove the existence of infinitely many geodesics between curves of distance 3 that are
tight and efficient, tight but not efficient, and efficient but not tight.

34

Figure 3.10: (i) 1 , 10 , ... are from dierent isotopy classes but are all distance 1 from
0
1 . (ii) 1 , 1 , ... each provide a unique path of distance 2 between 1 and 1 .

35

3.5

Overview of the curve complex in geometry

For the interested reader, we briefly describe the history of the curve complex. The
curve complex was introduced by Harvey in [6], but the question of its geometry was
taken up by Masur and Minsky in [8], where they proved that the curve complex
is

hyperbolic, which means that for any triangle defined in the curve complex,

there is a

> 0 such that each of the three sides of the triangle is contained in the

-neighborhood of the other two sides.


In [8], Masur and Minsky also studied the action of the mapping class group on the
curve complex. Since mapping classes act on curves on a surface, they subsequently
induce isometries of the curve complex. Any isometry can be classified by studying
its orbits. If every orbit of an isometry is bounded, then we say that the isometry
is elliptic. If the orbits are all quasi-isometric embeddings of the integers, then the
isometry is said to be hyperbolic. If an isometry is neither elliptic nor hyperbolic, then
it is parabolic. Masur and Minsky show in [8] that periodic and reducible elements of
the mapping class group act elliptically on the curve complex, while pseudo-Anosov
elements act hyperbolically; the latter implying the curve complex has infinite diameter.

Chapter 4
TIGHT GEODESICS
In this section, we present the set of preferred geodesics called tight geodesics discovered by Masur and Minsky in [9]. We will see that between any two vertices in C(S),
there exists at least one and at most finitely many tight geodesic paths joining the
two vertices. Examples seen in this section are largely inspired by those found in [1].

4.1

Definition

A multicurve is a set of k + 1 vertices with pairwise disjoint representative curves on


a surface S, which form a k-simplex in C(S).
A sequence of multicurves is a multigeodesic {M0 , M1 , ..., Mn }, if for all vi 2 Mi
and vj 2 Mj , i 6= j, we have dS (vi , vj ) = |i

j|. Furthermore, we stipulate that M0

and Mn are necessarily single curves and therefore contribute one vertex each to the
curve complex of S.
We say that a multigeodesic is tight if for each i 2 {1, ..., n

1}, Mi can be

represented as the union of essential boundary components of a regular neighborhood of vi


Mi

[ vi+1 , where vi

and vi+1 are minimally intersecting representatives of

and Mi+1 , respectively. We will denote this by Mi = @F (Mi


36

[ Mi+1 ), where

37
F (Mi

[ Mi+1 ) is the union of a regular neighborhood of Mi

[ Mi+1 along with

any disk components in the complement of that regular neighborhood.


Figure 4.1 (i) presents a tight multigeodesic of distance 3 for S3 in the oncepunctured sphere with marked points representation. In that figure, M0 = 0 , M1 =
1 [ (1 ), M2 = 2 [ (2 ), and M3 = 3 ; in other words, M1 and M2 are 1-simplices.
We first explain why dS3 (0 , 3 ) = 3. First, 0 is disjoint from M1 , and intersects
M2 and 3 essentially. (Curves meeting at a marked point have that marked point
lifted to an essential intersection on the surface.) Second, M1 is disjoint from M2 , and
intersects 3 essentially. Third, M2 is disjoint from 3 . As a result, dS3 (0 , 3 ) 3.
3 observe that cutting along 0 [ 3 in Figure 4.1

Now to see why dS3 (0 , 3 )

(i) yields components of the sphere that all lift to non-essential components of the
surface. Thus dS3 (0 , 3 )

3, which implies dS3 (0 , 3 ) = 3.

Now to see that the multigeodesic is tight, observe that @F (M0 [ M2 ) = M1 and
@F (M1 [ M3 ) = M2 .
These last two conditions are met in the following way: Figure 4.1 (iii) shows the
filled in regular neighborhood of M0 [ M2 , and Figure 4.1 (iv) fills in the outer disk on
the sphere, where the point at infinity yields the claimed disk. This is F (M0 [ M2 ),
whose boundary is @F (M0 [ M2 ) = M1 , using the fact that a curve enclosing exactly
two marked points is isotopic to an arc connecting the two marked points. Similarly
by symmetry, we see that the boundary of F (M1 [ M3 ) is @F (M1 [ M3 ) = M2 .
Notice that a key characteristic of tight multigeodesics is that if a curve intersects
essentially an element of Mi , then that curve must also intersect an element of Mi

1[

Mi+1 . In other words, if there exists a curve that intersects an element of Mi but does
not intersect any element of Mi

[ Mi+1 , then the multigeodesic fails to be tight.

38

Figure 4.1: A tight multigeodesic of distance 3

39

4.2

Finitude

Masur and Minskys proof showing that there are only finitely many tight multigeodesics between any two vertices of distance 3 or greater can be found in [9].
Due to the highly involved nature of the proof found in [9], we will instead present
a proof of finitude for just the distance 3 case.
Theorem 4.2.1. There are finitely many tight multigeodesics between any two isotopy
classes of curves of distance 3.
Proof of Finitude. Let v and w be vertices in C(S), and k = {v = M0 , M1 , M2 , M3 =
w} be a tight multigeodesic. We will show that there are finitely many choices for
M1 and M2 , which would imply finitely many distinct sequences for k.
Since k is a multigeodesic, we have dS (v, w) = 3. Then, for v0 2 v and w0 2 w in
minimal position, cutting along v0 [w0 chops S into even-sided polygons, finitely many
of which are non-rectangular. Consider one example of a non-rectangular polygon
(Figure 4.2). Since M1 is disjoint from M0 , we must have M1 being arcs joining
distinct sides of M3 . Although more than one segment of M1 can join two distinct
sides of M3 , the segments must be parallel to avoid self-intersection.

Figure 4.2: A possible configuration of v0 , w0 , and m1 2 M1 in minimal position


If a bound exists for the number of arcs of M1 in each non-rectangular polygon,

40
then since M2 = @F (M1 [ M3 ) there is clearly a bound for the number of choices
for M2 , and vice versa. Note that rectangular polygons are straight passageways
for M1 and M2 , as M1 may only cross M3 sides, and M2 may only cross M0 sides.
In other words, they can be seen as simply thickening the sides of non-rectangular
polygons, and therefore do not increase the number of possible choices for M1 or M2 .

Figure 4.3: Multiple parallel m1 arcs are replaced with one similarly parallel representative arc
Assume for the moment that on a given non-rectangular polygon, two distinct
sides of M3 are joined by arbitrary many parallel segments of M1 . Since k is tight,
M2 = @F (M1 [ M3 ). By the definition of F (Mi

[ Mi+1 ), M2 will not form bound-

aries for the disk components found between parallel segments of M1 . Therefore,
in determining where the segments of M2 lie in the polygon, one can replace each

41
parallel family of M1 segments with one representative from the family. Since there
are finitely many combinations for pairs of distinct M3 sides in any regular even-sided
polygon, there are finitely many parallel families of M1 segments, and so there is
a bound on the number of segments of M2 joining two sides of M0 . Therefore the
number of choices for M2 and M1 is bounded and finite.

As an example to illustrate the reasoning in the above proof, Figure 4.3 (i) presents
a possible configuration for v0 , w0 , and m1 2 M1 in minimal position.
(ii) shows where m2 2 M2 must be positioned in order to satisfy tightness. The
key in this figure is that the existence of multiple parallel arcs of m1 does not aect
the possible arcs of m2 which can be formed.
(iii) has non-essential components of m2 isotoped out of the polygon, thereby
keeping only necessary components of m1 . In other words, when positioning M2
segments in a polygon, it is possible for both endpoints of a segment to lie on the
same M0 side. In such instances, those segments of M2 can be isotoped out of the
polygon.

4.3

Existence

As a result of the previous theorem, not all multigeodesics are tight, or can be made
tight by just adding in curves. As an instructive example, consider the multigeodesic
seen in Figure 4.4, which lifts to S3 .
We may quickly confirm its identity as a distance 3 multigeodesic by seeing that
the following are true:
- 0 is disjoint from 1 .
- 1 is disjoint from 2 .
- 2 is disjoint from 3 . Thus dS3 (0 , 3 ) 3.

42

Figure 4.4: A distance 3 multigeodesic that fails to be tight


- Cutting along 0 [ 3 yield squares with one marked point each, which all lift
to non-essential components of the surface. Thus dS3 (0 , 3 )

3, which implies

dS3 (0 , 3 ) = 3.

Figure 4.5: Confirming non-tightness


To see that this multigeodesic is not tight, we first cut along 1 [ 3 , as seen in
Figure 4.5 (ii). If we swing the cut around on the sphere as seen in Figure 4.5 (iii),
it is evident that @F (1 [ 3 ) 6= 2 . In particular, we can find an arc on the sphere
that intersects 2 without intersecting 1 [ 3 .
Despite the fact that not all multigeodesics are tight, we would like to show that
for each pair of vertices v, w in C(S) for which d(v, w) = N , there exists a tight
multigeodesic {M0 , M1 , ..., MN } where M0 = v and MN = w. This proof comes from
[9].
Theorem 4.3.1. Let v and w be vertices in C(S). Then, there exists a multigeodesic
{v = M0 , M1 , ..., MN = w} that is tight.

43
Proof of Existence. Let h = {v = W0 , ..., w = WN } be a multigeodesic that is not
tight. We will describe a process of adjusting h until it becomes tight.
To this end we begin with a special case of a length 3 multigeodesic. Namely, let
M1 , M2 , M3 , M4 be a multigeodesic, and suppose for the sake of argument that we
know M3 = @F (M2 [ M4 ). On the other hand, assume M2 6= @(M1 [ M3 ). We would
like to replace M2 by M20 = @F (M1 [ M3 ) and also show that M3 = @F (M20 [ M4 ).
In other words, we would need to know that @F (M20 [ M4 ) = @F (M2 [ M4 ).
Since we still have that dS (v1 , v4 ) = 3 and by disjointness dS (v1 , v20 ) = 1, we
cannot have dS (v20 , v4 ) = 1. Therefore M1 , M20 , M3 , M4 is still a multigeodesic. Now,
M20 = @F (M1 [ M3 ) implies that M20 is disjoint from M3 = @F (M2 [ M4 ). Since M20
is not disjoint from M4 , M20 must be contained in F (M2 [ M4 ), which implies that
F (M20 [ M4 ) F (M2 [ M4 ). Now, if we can show that F (M2 [ M4 ) F (M20 [ M4 ),
then we are done.
Let be an essential curve in the subsurface defined by F (M2 [ M4 ). We need
to show that intersects F (M20 [ M4 ). If intersects M4 , then it must intersect
F (M20 [ M4 ), and we are done. On the other hand, if does not intersect M4 , then
it must intersect M2 . Additionally, since M1 and M4 chop the surface into polygons,
must also intersect M1 .
Since M2 is disjoint from both M1 and M3 , it must be in the exterior of F (M1 [M3 ),
as intersects M2 . This implies that intersects the exterior of F (M1 [ M3 ). Since
M1 is in the interior of F (M1 [ M3 ) and intersects M1 , must therefore intersect
@F (M1 [ M3 ) = M20 . Thus indeed intersects F (M20 [ M4 ). We conclude that if a
multigeodesic is tight at an index i, then a neighboring multigeodesic may be made
tight at both i and i

1 by an appropriate replacement.

Returning our attention to h, we replace Wi by @F (Wi


of i 2 {1, 2, ..., N

[ Wi+1 ) for any choice

1}. Then, by the result of our above argument, we can continue to

repeat the process for new values of i and move closer to satisfying all requirements

44
for a tight multigeodesic. Therefore, h will be fully adjusted to a tight multigeodesic
after N

4.4

1 steps.

Applications

Agan, for the interested reader we briefly mention some applications of tight geodesics.
As mentioned previously, Masur-Minsky in [8] showed that C(S) is a -hyperbolic
space. This result led them to try to apply the techniques of hyperbolic spaces and
groups to study C(S) and the action of the mapping class group on it, namely to study
such questions as the word problem and the conjugacy problem for the mapping
class group. However, the fact that the curve complex is locally infinite resulted
in a technical challenge when trying to use distance arguments in C(S) to prove
algebraic statements about the mapping class group. Thus in [9], Masur-Minsky used
a hierarchical structure in C(S) which is similar to those of locally finite complexes,
including the aforementioned finiteness result for tight geodesics with given endpoints,
and a convergence criterion for sequences of geodesics. They then applied these ideas
to establish, among other things, a linear bound on the shortest word conjugating
two pseudo-Anosov mapping classes in the mapping class group.

Chapter 5
EFFICIENT GEODESICS
While Masur and Minskys tight geodesics have interesting applications in geometry,
efficient geodesics of Birman, Margalit, and Menasco have properties of computability
for distance. In particular, with efficient geodesics, the distance between two vertices
in C(S) is able to be computed for small distances. In this chapter, we will see that
there exists at least one and at most finitely many efficient geodesic paths joining
any two vertices in C(S). Additionally, we will see how non-efficient geodesics can be
made efficient through a process of surgery.

5.1

Definition

Let v0 , v1 , ..., vn be a geodesic of n

3 in C(S). For each vertex in the geodesic,

choose a representative i with the extra requirement that 0 , 1 and n are pairwise
in minimal position. Since n is at least 3, cutting along 0 and n splits S into
even-sided polygons, which well refer to as 0
If

is an arc such that

n polygons.

and 1 are in minimal position, and the interior of

disjoint from 0 [ n , then we say

is

is a reference arc for the trio 0 , 1 , n (Figure

5.1).
45

46

Figure 5.1: Here,

is one of many reference arcs for the trio 0 , 1 , n

The geodesic v0 , v1 , ..., vn is initially efficient if the number of intersections of 1


with

is less than n for any choice of reference arc . Loosely, a geodesic is initially
[0 [ n ] too much.

efficient if 1 does not intersect the polygons in S

We say that the geodesic is efficient if for k = 0, 1, ..., n

3 we have that

vk , vk+1 , ..., vn is initially efficient, meaning that the number of intersections of k+1
with any reference arc
the k

is less than n

k. Note that here,

n polygons. Additionally, vn , vn 1 , vn 2 , vn

meaning that the number of intersections of n


3, where here

is a reference arc for the n

is a reference arc for

must also be initially efficient,

with any reference arc


3

is less than

polygons.

Two things immediately simplify the above definition:


First, when defining a collection of reference arcs for the trio 0 , 1 , n , each 1
arc in each 0

n polygon can be replaced with straight line segments. Note that

the reference arcs in question are only for a particular trio of representative curves.
Second, since the key reference arcs are those that have the most intersections with
1 , the only arcs that need be considered are straight line segments that connect
midpoints of 0 edges of each non-rectangular 0

n polygon. As 0 is disjoint

from 1 , and therefore has restricted entry and exit points to edges of n , straight
line segments joining midpoints of 0 edges will not form bigons with arcs of 1 , which

47

Figure 5.2: A possible element of S


reference arcs

[0 [ n ] with all possible positions of 1 and

48
are also assumed to be straight line segments. Additionally, since each straight line
segment from one 0 edge to another in a rectangular 0
one in a non-rectangular 0

n polygon is parallel to

n polygon, we are guaranteed to find the reference arc

having the most essential intersections with 1 when looking just at non-rectangular
polygons.
Figure 5.2 (i) is an example of one of possibly many polygons resulting from
cutting along 0 and n on the surface. Part (ii) depicts all possible positions of
1 through the given polygon. Each arc of 1 may be weighted by multiple parallel
strands of 1 , which will preserve distance but can jeopardize efficiency. Part (iii)
shows all reference arcs within the given polygon that must be considered.
We emphasize that in the process of checking efficiency for a geodesic, reference arcs are determined individually for all trios: (0 , 1 n ), (1 , 2 , n ), ... ,
(n 3 , n 2 , n ), and (n , n 1 , n 3 ). For example, in checking initial efficiency for
v1 , v2 , ..., vn (n

4) one cuts S along 1 [ n and considers all reference arcs to be

straight line segments joining midpoints of 1 edges in non-rectangular polygons. If


the number of intersections of 2 with is less than n

1 for any choice of , then

v1 , v2 , ..., vn is initially efficient.


Figure 5.3 depicts an efficient geodesic of distance 3 in the once-punctured sphere
with marked points representation for S3 . Notice that this geodesic extends to the
tight multigeodesic of distance 3 seen in Figure 4.1 by adding in additional curves.
To realize efficiency, see that the following are true:
- Cutting along 0 [ 3 yields either polygons containing no marked points or one
marked point, both of which lift to non-essential components of the surface. One
such polygonal region contains one arc of 1 and one arc of 2 (Figure 5.3 (ii)).
Observe that this region does not contain a marked point. Notice there are two other
bigons, each containing a marked point, which also contain 1 or 2 . These will lift
to rectangles in S, and thus are not under consideration when checking efficiency.

49
- This region (Figure 5.3 (iii)) will lift to hexagons on the surface containing one
arc each of 1 and 2 . This implies that any reference arc for the trio 0 , 1 , 3
will intersect 1 at most once. Similarly, any reference arc for the trio 3 , 2 , 0 will
intersect 2 at most once. As initial efficiency requires that the number of intersections
of a trios middle element with any reference arc be less than 3, we see that both trios
are initially efficient. Therefore, efficiency is achieved.

5.2

Finitude

In order to prove that there are finitely many efficient geodesics between v0 and vn ,
we must show that there are finitely many candidates for 1 , 2 , ..., n 1 . Due to the
way efficiency is defined, we will see that it is enough to simply find a bound on the
number of candidates for 1 ; a similar argument will then apply to 2 , ..., n 1 . This
theorem and proof are from [1].
Theorem 5.2.1. There are finitely many efficient geodesics between any two isotopy
classes of curves of distance n

3.

Proof of Finitude. Consider the trio 0 , 1 , n and a surface S = Sg>1 . We begin


by cutting S along 0 , which yields a new (possibly disconnected) surface S 0 having
two boundary components. Since cutting along 0 [ n on S would yield even-sided
polygons, n on S 0 becomes a collection of arcs with endpoints that are either both
on one boundary component of S 0 , or one on each boundary component of S 0 . Notice
that cutting along two parallel n arcs component yields a rectangular polygon.
A collection of reference arcs in S 0 will be formed in the following way: For each
family of parallel arcs of n , define one reference arc similarly parallel. For each arc of
n which is not parallel to any other arcs of n , define a single reference arc parallel
to that arc of n . This guarantees that cutting along the defined reference arcs on S 0
yields only 2k-gons, where k

3. Then, with the intent of cutting S 0 into hexagons

50

Figure 5.3: An efficient geodesic of distance 3

51
only, define additional reference arcs to cut any 2k-gons with k > 3 into hexagons.
The collection of reference arcs defined according to the above method is disjoint,
and the interior of each arc in the collection is also disjoint from 0 in S 0 .
We will use a simple Euler characteristic argument to show that there are precisely
6g

6 reference arcs in the collection.


It is well-known that (Sg ) = 2

2g, and this 2

2g remains the Euler charac-

teristic of S 0 , namely the surface S split along 0 . To see this, consider the usual
calculation (S 0 ) = V

E + F , where V, E, F refer to the number of vertices, edges,

and faces, respectively. In particular, edges can naturally be arcs of 0 and reference
arcs, vertices can be there they intersect, and F can be the interiors of the hexagons.
Notice that we are splitting S along 0 , which is a simple closed curve, and all
vertices exist on 0 . This implies that there is one vertex for each 0 edge. So in the
Euler characteristic calculation (V

E + F ), the vertices and edges belonging to 0

cancel each other out. In other words, all vertices and all 0 edges together contribute
nothing to the final Euler characteristic calculation. Thus indeed Sg0 = 2
Moreover, it is also true that each hexagon contributes

1
2

2g.

to (S 0 ). To see this,

first observe that the interior of each hexagon contributes 1 to F . Additionally, each
hexagon is comprised of three 0 edges and three reference arc edges. By the above
argument, only the three reference arc edges on each hexagon will contribute to the
final Euler characteristic calculation. However, since each reference arc edge bounds
3
2

two hexagons on opposite sides, the three reference arc edges contribute
Euler characteristic calculation. Therefore each hexagon contributes
(S 0 ). Letting h be the number of hexagons, we obtain that

1
h
2

=2

3
2

to the final

+1 =

1
2

to

2g.

Now, let r be the number of reference arcs. As each hexagon has three reference
arc edges and each reference arc edge bounds two hexagons, we obtain the equation
3h = 2r. This implies h = 23 r.
Thus, if we do an Euler characteristic count based on the number of hexagons, we

52
obtain: (

1 2
)( r)
2 3

=2

2g, which yields r = 6g

6.

We can now explain why for an efficient geodesic there are only finitely many
candidates for 1 . First, any such 1 will be completely determined by the number
of 1 arcs joining any two of the three edges in each of the above hexagons. As
the reader can check, this in turn will be completely determined by the number of
intersections of 1 with each of the reference arc edges of each of the hexagons. Since
initial efficiency is assumed, the number of intersections of 1 with each reference arc
in the collection is bounded above by n. Therefore, there exist only finitely many
candidates for 1 , namely less than n6g 6 . By the convenient definition of efficiency,
this argument can be repeated to show that there are also finitely many candidates
for 2 through n 1 .

5.3

Existence

By the above theorem, we see that just as not all geodesics are tight, it is also the
case that not all geodesics are efficient. It will be informative to look at an example
of this.
Consider the sequence of curves s = {0 , 1 , 2 , 3 } that is observed in Figure
5.4 (i), depicted on the punctured sphere with marked points representation for S2 .
Notice that s satisfies all conditions to be a distance n = 3 geodesic.
Consider the outer hexagon obtained by cutting along 0 [ 3 (Figure 5.4 (ii)).
Arcs of 1 and 2 are positioned appropriately according to the intersections. The
hexagon on the punctured sphere is lifted to the surface to become a 12-gon (Figure
5.4 (iii)), where we identify s to be non-efficient by locating a reference arc
intersects 1 four times, which is more than n

which

1.

Although not all geodesics are efficient, we would like to prove the existence of

53

Figure 5.4: A non-efficient geodesic of distance 3


at least one efficient geodesic between v0 and vn . We outline below a sketch of the
proof from [1], with some details provided. In order to accomplish this, several new
concepts must be introduced.
First, we will create a standard collection of representative curves, i 2 vi , chosen
in the following way:
(i) Each i is in minimal position with both 0 and n .
(ii) i and i+1 are disjoint for i = 0, 1, ..., n

1.

(iii) There are no triple intersections. In other words, i \ j \ k = ; for distinct


choices of i, j, k 2 {0, 1, ..., n}.
A reference arc

for the standard collection of representative curves satisfies the

following conditions:
(i) The interior of

is disjoint from 0 [ n .

(ii) The endpoints of

may not intersect 1 through n 1 .

(iii) For i 6= j, there are no triple intersections of , i , and j .


(iv)

is in minimal position with all of 1 through n 1 .

Notice that

is immediately also a reference arc for the triple 0 , 1 , n , however

the reverse does not necessary hold.


Let N represent the number of intersections
chosen orientation traverse
natural numbers

has with 1 [2 [...[n 1 . For some

from one endpoint to another and note the sequence of

= (j1 , j2 , ..., jN ), where ji denotes the index of that is the ith

54
intersection on . We call

the intersection sequence of the i s along .

We define the complexity of an oriented path p = {v0 , v1 , ..., vn } in C(S) as the sum
P
(p) = ni=11 i(v0 , vi ) + i(vi , vn ). An intersection sequence is said to be reducible
if there is another path v00 , v10 , ..., vn0 in C(S), with v00 = v0 and vn0 = vn , where the
complexity of this alternate path is less than the complexity of the initial path.
Notice that if less than n entries of

is 1 for every reference arc , then initial

efficiency is satisfied for the given path.


The proof of existence depends largely on Proposition 3.1 found in [1] It is as
follows:
Proposition 5.3.1. If more than n
then

1 entries of

is 1 for some reference arc ,

is reducible.

Let v and w be vertices in C(S) where dS (v, w)

3. Since complexity is always a

natural number, there exists a geodesic path joining v and w with the least complexity.
We will show that if v0 , v1 , ..., vn is a distance 3 or greater geodesic path from v = v0
to w = vn with the least complexity, then it must be initially efficient.
Theorem 5.3.1. Let v and w be vertices in C(S). Then, there exists a sequence
v = v0 , v1 , ..., vn = w that is efficient.
Sketch of Proof of Existence. Let v0 , v1 , ..., vn be a geodesic of at least distance three,
where v = v0 and w = vn ; we will assume this geodesic is not initially efficient. By the
above reasoning, we therefore must show that it is not of minimal complexity. Since
we assume the geodesic is not initially efficient, there exist 0 2 v0 , 1 2 v1 , n 2 vn
that are in minimal position, and a reference arc
of intersections of 1 with

is greater than n

for the trio such that the number

1.

In order to use Proposition 5.3.1 to show that the complexity can be reduced, we
must show that

can be extended to be a reference arc for the standard collection.

55
Recall that a reference arc of the standard collection must be in minimal position
with each one of the arcs 1 , ..., n 1 . For now, we only know that

is in minimal

position with 1 .
Assume that

is not in minimal position with some i 2 {2 , ..., n 1 }. Then,

and i cobound a bigon.


We may invoke the innermost property from the bigon criterion proof and consider
the innermost bigon bounded by

and the i forming the other arc.

Since the bigon is innermost, no other j 2 {2 , ..., n 1 } (j 6= i) passes through


its interior, and therefore eliminating the bigon via homotopy will not create any new
points of intersection between i and any j . After eliminating the innermost bigon,
another bigon becomes innermost. Therefore the procedure can be repeated until all
bigons are eliminated.
Notice that both 1 and

remained fixed throughout the bigon elimination pro-

cess, and so the number of intersection points between 1 and


n

is still greater than

1.
But now, since we have shown that

satisfies all conditions to be a reference arc

for the standard collection, we may invoke Proposition 5.3.1, which implies that our
geodesic was not of minimal complexity.
To then finally show that an efficient geodesic exists between any two vertices v
and w in C(S), Birman-Margalit-Menasco define the total complexity
of an oriented
path p = {v0 , v1 , ..., vn } to be the n

1 tuple of the complexities of the relevant sub-

paths pi = {vi , vi+1 , ..., vn }, namely


(p) = {(p0 ), (p1 ), ..., (pn )}. They order the

kappas
lexicographically, and show that the geodesic with minimal total complexity
is efficient.
Although we do not provide the proof for Proposition 5.3.1, we will instead demonstrate through the following example how a non-efficient geodesic of distance 3 is made
efficient through surgery, an idea important in the proof of Proposition 5.3.1.

56
Consider the example of the non-efficient geodesic s of distance 3 presented at the
beginning of the section (Figure 5.4).
In order to manipulate the curves of s to yield an efficient geodesic, we perform
a surgery on 1 in the manner depicted in Figure 5.5. Specifically, we locate two
consecutive intersections of 1 with 3 , with opposite orientation, and form the new
10 as indicated. The result of the surgery applied on 1 in s yields the new distance
3 geodesic s0 = {0 , 10 , 2 , 3 } seen in Figure 5.6 (i).

Figure 5.5: A surgery on 1


In Figure 5.6, the curve 10 on the surface of the sphere in (i) can be swung around
to a new position seen in (ii), which is isotopic to a line segment joining two marked
points depicted in (iii).

Figure 5.6: Isotoping 10 on the sphere


Now, cut along 0 [ 3 once again and position arcs of 10 appropriately (Figure
5.7). Observing the lift to the surface in (iii) of Figure 5.7, we now notice that no
reference arc in the 12-gon will intersect 10 more than twice. Since n
have successfully performed surgery on s to yield an efficient geodesic s0 .

1 = 2, we

57

Figure 5.7: An efficient geodesic of distance 3

5.4

Applications

In [1], Birman-Margalit-Menasco state that one of the goals of the paper is to give
an algorithm for distance between two vertices in C(S) that can be implemented
for small distances. Although several algorithms have been developed for the purpose of computing distance between two vertices of C(S), none were seriously created
for performing explicit calculations until the efficient geodesic algorithm designed
by Birman-Margalit-Menasco. In 2014, Menasco, Glenn, Morrell, and Morse implemented the efficient geodesic algorithm as a computer program, MICC, coded in th
language Python [5]. As a result of the MICC program, it has been shown that the
minimal intersection number of two distance 4 curves on S2 is 12. This result is realized by the example shown at the beginning of [1], which at its time of discovery was
the only other explicit portrayal of curves of distance 4 in literature other than the
example of Hempel [7] having intersection number 25. In addition to locating curves
of distance 4 on surfaces of genus g > 1, the efficient geodesic algorithm has also been
used to locate geodesic triangles where any pair of vertices correspond to filling pairs
in minimal position.

Chapter 6
NEW RESULTS
We do not know if between any two vertices there always exists a geodesic that is
efficient and tight. - [1]
In response to the above statement by Birman-Margalit-Menasco, we provide a
first result showing the existence of infinitely many tight and/or efficient geodesics of
distance 3 on arbitrary Sg>1 .
The proof will be based on using an already known distance 3 tight and/or efficient
geodesic {0 , 1 , 2 , 3 } and finding a specific arc

about which Dehn twists of 0

will be performed. The resulting geodesic {T (0 ), 1 , 2 , 3 } will be from a dierent


family than the initial geodesic, but will retain the tight and/or efficient characteristic
of the initial geodesic.

6.1

Dehn twists, tightness, and efficiency

The first section contains four propositions which show that specific Dehn twists can
be used to preserve tightness and/or efficiency of certain geodesics.
We begin with the following set of initial conditions:
(i) , 0 , and are curves that are mutually in minimal position
58

59
(ii)

and 0 intersect once

(iii) 0 and are from dierent isotopy classes


Assuming the above set of initial conditions, we claim the following to be true:
Proposition 6.1.1. T+ (0 ) or T (0 ) is in minimal position with .
Proof of Proposition 6.1.1. We focus on the local piece of the surface where the one
intersection of

and 0 occurs (Figure 6.1).

Figure 6.1: The sole intersection of


One of two Dehn twists of 0 about

and 0

may occur on the surface: the positive (left)

Dehn twist T+ (0 ) or the negative (right) Dehn twist T (0 ) (Figure 6.2).

Figure 6.2: Left and right Dehn twists of 0 about


Since , 0 , and are mutually in minimal position, this places limitations on
where arcs of can co-exist near the intersection point of

and 0 . In particular,

60
we will show that if an arc of cobounds a bigon with one of the Dehn twists, there
cannot exist another arc of that cobounds a bigon with the other Dehn twist.
Without loss of generality, consider the bigon formed by and T+ (0 ) as shown
in Figure 6.3.

Figure 6.3: An arc of cobounds a bigon with the negative Dehn twist
It is important to note that although Figure 6.3 depicts a gap between T+ (0 )
and

[ 0 , the bigon is formed as a result of intersecting 0 and

essentially.

Therefore, any attempt to cobound a bigon with and T (0 ) also requires to


intersect 0 and

essentially.

Figure 6.4: The circled portion of must exit the triangular region
Assume another arc of is placed to cobound a bigon with T (0 ) in the way
shown in Figure 6.4. Notice that the circled portion of is in a triangular region

61
formed by arcs of , 0 , and itself; we may assume this is true without loss of
generality. Since is an essential simple closed curve, continuing from the circled
portion must eventually exit . If the circled portion of continues its trajectory to
exit , then it will cause a self-intersection. If it attempts to exit

via 0 down below,

will create a bigon with 0 . If it attempts to turn back and exit

via , will

form bigons with . As cannot intersect itself, and also must remain in minimal
position with

and 0 , we conclude that such a positioning of the arcs of is not

possible. Therefore cannot cobound bigons with both T (0 ) and T+ (0 ).

Going forward, we will denote the Dehn twist coming from Proposition 6.1.1 as
T (0 ).
Proposition 6.1.2. 0 and fill if and only if T (0 ) and fill.
Proof of Proposition 6.1.2. By Proposition 6.1.1, T (0 ) is in minimal position with
.
Assume 0 and fill the surface. This implies that cutting along 0 [ on the
surface yields a collection of even-sided polygons with alternating 0 sides.
For a given polygon in the collection,

can be positioned through it in one of

three ways:
(i)

goes from 0 to 0 in a polygon, where the 0 sides are identified.

(ii)

goes from to 0 in a polygon, then immediately goes from 0 to in the

adjacent polygon sharing the 0 side.


(iii)

goes from to in a polygon.

We first observe that any instance of (iii) will simply split a pre-existing polygon
into two polygons. As an example, consider the case with

splitting an octagon into

two hexagons (Figure 6.5 (i)).


For this example, we position the arc

from to as seen in the figure. We

62

Figure 6.5: (i) The existing polygon is split into two smaller polygons by T (0 ). (ii)
Reference arcs of the smaller polygons are subarcs of those from the initial polygon

63
perform the twist of 0 about the arc . Then, notice that the initial octagon has
now been split into two hexagons with

T (0 ) sides.

Therefore occurrences of (iii) will not jeopardize the desired result of having T (0 )
and fill.
Now, notice that for any given cutting of the surface along 0 [ , exactly one of
two remaining cases must occur:
Case 1: (i) occurs, which prevents (ii) and (iii) from occurring.
Case 2: (ii) occurs, which prevents (i) from occurring and (ii) from occurring
again.
We will show that both Case 1 and Case 2 continue to yield T (0 ) and that fill
the surface. We first consider Case 1. There is in that case only one 0
containing , so T (0 ) will preserve all of those T (0 )

n polygon

n polygons which did

not contain . For the 2k-gon containing , we observed that T (0 ) will simply
replace that 0

n 2k-gon with a twisted T (0 )

n 2k-gon. Figure 6.6 shows

two variations of Case 1 with an 8-gon as demonstration. Specifically, Figure 6.6 (i)
shows a Dehn twist of 0 about , where

intersects non-consecutive identified sides

of 0 . Although twisted, the resulting T (0 ) polygon is in fact still an 8-gon.


Figure 6.6 (ii) shows the twist of 0 about a

that intersects consecutive identified

sides of 0 , yielding a twisted 8-gon with T (0 ) sides. As the rest of the surface
remains unchanged, T (0 ) and chop the surface into polygons, and therefore must
fill.
For Case 2, there are just two 0
namely the adjoining ones in which

n polygons which we have not yet considered,


goes from an 0 edge to an n edge. In this case

the result of T (0 ) will take the two adjoining polygons and split them into three
T (0 )

n polygons. For example, a Dehn twist of 0 about

shown in Figure 6.7

simply creates one large polygon and two smaller polygons as a result of opening up
the two initial polygons. As the rest of the surface remains unchanged, T (0 ) and

64

Figure 6.6: Two variations of Case 1

65
must fill.

Figure 6.7: A demonstration of Case 2


We have shown that if 0 and n fill, then T (0 ) and n fill. For the other
direction, without loss of generality, assume that T (0 ) = T+ (0 ). If and T (0 )
fill, then observe that 0 = T (T+ (0 )). Since T (T+ (0 )) is clearly still in minimal
position with , 0 and must fill.
We now focus in on the case of when we have a distance 3 multigeodesic.
Define s = {0 , M1 , M2 , 3 } and s0 = {T (0 ), M1 , M2 , 3 } to be multigeodesics.
Let

be a curve that intersects 0 precisely once, and is disjoint from M1 . Then, we

claim the following to be true:


Proposition 6.1.3. s is tight if and only if s0 is tight.
Proof of Proposition 6.1.3. Assume s is a tight multigeodesic. Then, the components
of @F (0 [ M2 ) are precisely M1 .
As

and 0 intersect, and

is disjoint from M1 , we must have

of F (0 [ M2 ). As a result, we have F (T (0 ) [ M2 ) F (0 [ M2 ).

in the interior

66
Now, without loss of generality, assume that T (0 ) = T+ (0 ). Notice that
T (0 ) intersects

once, and that T (T+ (0 )) = 0 . Then, repeating the above ar-

gument for s0 = {T (0 ), 1 , 2 , 3 }, we have that F (T (T+ (0 )[M2 )) F (T (0 )[


M2 ).
Since we have shown both F (T (0 ) [ M2 ) F (0 [ M2 ) and F (0 [ M2 )
F (T (0 ) [ M2 ), we conclude that F (T (0 ) [ M2 ) = F (0 [ M2 ). This implies that
1 = @F (T (0 ) [ M2 ). Therefore {T (0 ), M1 , M2 , 3 } is tight.
The argument is easily reversed to show that the other direction also holds.

Now assume s = {0 , 1 , 2 , 3 } and s0 = {T (0 ), 1 , 2 , 3 } are geodesics, with


intersecting 0 exactly once, and
Proposition 6.1.4. If
edge of a polygon in S

disjoint from 1 .

has an arc that intersects a 0 edge and a consecutive 3


[0 [ 3 ], then s0 is efficient if s is efficient.

Proof of Proposition 4. First, consider the case where


another 3 edge of a polygon in S

goes from an 3 edge to

[0 [ 3 ]. First, recall how such a

will split one

polygon into two as in Figure 6.5 (i). Observe then, as in Figure 6.5 (ii), that any
reference arc in the new polygon is either a reference arc in the original polygon, or
a subarc of a reference arc in the original polygon. Thus the number of intersections
of 1 or 2 with the new reference arcs will be less than or equal to the the number
of intersections of 1 or 2 with the old reference arcs. Therefore, efficiency will not
be violated within these new polygons.
Let P be a polygon in S

[0 [ 3 ] such that

intersects an 0 edge shared by

P and a neighboring polygon Q.


Assume s is an efficient geodesic. This implies that for any given polygon of
S

[0 [ 3 ], 1 and 2 intersects any choice of reference arc at most twice.

67
If

does not exit P nor Q through an adjacent 3 edge, then efficiency can be

easily violated for s0 (Figure 6.8).

Figure 6.8: Here, the positioning of


reference arc
Therefore, we must have

could create four intersections of 1 with a

exit P or Q through a consecutive 3 edge in order

to avoid opening up a larger polygon (Figure 6.9). More specifically, under this
condition, P and Q are now in S

[T (0 ) [ 3 ] divided into three polygons, one

of which is the original P or Q, and the other two of which are sub-polygons of P
or Q. Thus, as above, all reference arcs either are original ones from P and Q, or
are subarcs of original ones from P and Q. As
polygons of S

only intersects 0 once, all other

[0 [ 3 ] are guaranteed to remain unchanged, or be chopped into

smaller polygons in S

[T (0 ) [ 3 ]. So for any given polygon of S

1 and 2 will still intersect any choice of reference arc at most twice.

[T (0 ) [ 3 ],

68

Figure 6.9: This positioning of


Figure 6.8

6.2

avoids opening up the problematic polygon seen in

Infinitely many tight and/or efficient geodesics


of distance 3

Throughout this thesis, we have introduced examples of distance 3 geodesics that are
tight and efficient (Figure 4.1, 5.3), tight but not efficient (Figure 5.4), and efficient
but not tight (Figure 4.4) . Now, as a result of the four previous propositions, the
possibility of generating infinitely many tight and/or efficient geodesics of distance 3
rests entirely on the existence of such a

as mentioned in the previous section. In

particular, given a tight and/or efficient geodesic of distance 3, we would like to place
an arc

on the surface such that the following conditions are satisfied:

and 0 intersect once

and 1 are disjoint

- For efficient geodesics,


polygon in S

intersects a 0 edge and a consecutive 3 edge of a

[0 [ 3 ]

Theorem 6.2.1. For any Sg>1 there exist infinitely many geodesics of length 3 which

69
are tight but not efficient.
Proof. First, let us take the example of the tight but not efficient distance 3 geodesic
seen in Figure 5.4 on S2 and place the arc
checked to see that

as seen in Figure 6.10. It is easily

and 0 intersect once at the shared marked point, and that

is disjoint from 1 . By Propositions 6.1.2 and 6.1.3, each consecutive Dehn twist
Tk (0 ) in the sequence generates a tight geodesic of distance 3. Moreover, each
geodesic sees an increase in i(Tk (0 ), 3 ), since i( , 3 ) > 0. As Dehn twists about
may be performed as many times as desired, this sequence extends infinitely, thereby
generating infinitely many tight geodesics of distance 3.
Finally, each {Tk (0 ), 1 , 2 , 3 } continues to fail to be efficient. This can be
verified by splitting along T (0 ) and 3 , which yields an octagon lifting to a 16-gon
in which 1 intersects a reference arc four times, similar to the situation in Figure
5.4 (iii). For every Tk (0 ) after that, the same octagon results upon splitting along
Tk (0 ) and 3 . Hence, efficiency fails.
Although the particular example in Figure 6.10 is on a surface of genus 2, notice
that it can be naturally extended to similar examples on all surfaces of genus g > 2.
Refer to Figure 6.11 for an extension to S3 .

Theorem 6.2.2. For any Sg>1 , there exist infinitely many geodesics of length 3 which
are efficient but not tight.
Proof. Now, we will take the example of the efficient but not tight distance 3 geodesic
seen in Figure 4.4 and attempt to place an arc , keeping in mind the extra condition
of requiring

to intersect a 0 edge and a consecutive 3 edge of a polygon in

S = [0 [ 3 ]. Notice that the arc


In particular,

placed as seen in Figure 6.12 is satisfactory.

intersects 0 once at the shared marked point, is disjoint from 1 ,

and satisfies the extra condition by intersecting an 0 edge and a consecutive 3

70

Figure 6.10: Generating infinitely many tight but not efficient geodesics of distance 3

71

Figure 6.11: Extension of the tight but not efficient example seen in Figure 6.10 to
S3
edge of the outer 4-gon, which lifts to an 8-gon on the surface. The argument for
generating infinitely many efficient but not tight geodesics of distance 3 is similar to
the one presented for the tight but not efficient case seen above. By Propositions
6.1.2 and 6.1.4, we are able to perform as many Dehn twists about

as desired and

obtain efficient but not tight geodesics of distance 3. This continually increases the
number of essential intersections of the twisted element with 3 , thereby generating a
member of a new class of efficient but not tight geodesics of distance 3. This example
depicted here in S3 naturally extends to all surfaces of genus g > 1 by positioning
each additional pair of marked points between two new layers of the 3 spiral, or
subtracting a pair for the case of S2 .

Theorem 6.2.3. For any Sg>1 , there exist infinitely many tight and efficient geodesics
of length 3.
Proof. Finally, we take the example of the tight and efficient geodesic of distance 3
seen in Figure 4.1 (i) and position an arc

that intersects 0 once, is disjoint from 1 ,

and satisfies the extra efficiency condition of intersecting consecutive 0 - 3 edges


of a polygon in S = [0 [ 3 ]. Notice that the positioning of

as seen in Figure

6.13 is satisfactory for the aforementioned conditions. The extra efficiency condition
is satisfied on consecutive 0 - 3 edges of the leftmost inner 6-gon, which lifts to
two hexagons on the surface. Each twist about

increases the number of essential

72

Figure 6.12: Generating infinitely many efficient but not tight geodesics of distance 3

73
intersections of the twisted element with 3 . By Propositions 6.1.2, 6.1.3, 6.1.4, a
member of a new class of tight and efficient geodesics of distance 3 is generated with
each twist. As the number of Dehn twists about

is not limited, this process yields

infinitely many tight and efficient geodesics of distance 3. This example for S3 can
be easily generalized to S2 in the way shown in Figure 6.14, and similarly extended
to all Sg>3 .

74

Figure 6.13: Generating infinitely many tight and efficient geodesics of distance 3

75

Figure 6.14: Generalizing the tight and efficient example seen in Figure 6.13 to S2

Bibliography
[1] J. Birman, D. Margalit and W. Menasco, (2016). Efficient geodesics and an
eective algorithm for distance in the complex of curves. Math. Annalen.
[2] S. Bleiler and A. Casson, (1988). Automorphisms of surfaces after Nielsen and
Thurston. Cambridge University Press.
[3] M. Dehn, (1938). Die Gruppe der Abbildungsklassen. Acta Mathematica, Vol. 69,
pp. 135-206.
[4] B. Farb and D. Margalit, (2011). A Primer on Mapping Class Groups. Princeton
University Press.
[5] P. Glenn, W. Menasco, K. Morrell and M. Morse, (2016). MICC: A tool for
computing short distances in the curve complex. J. Symbolic Computation.
[6] W. Harvey, (1978). Boundary structure of the modular group. In: Riemann surfaces and related topics: Proceedings of the 1978 Stony Brook Conference.
[7] J. Hempel, (2001). 3-manifolds as viewed from the curve complex. Topology, Vol.
40, pp. 631-657.
[8] H. Masur and Y. Minsky, (1999). Geometry of the complex of curves I: Hyperbolicity. Invent. Math., Vol. 138, pp. 103-149.

76

77
[9] H. Masur and Y. Minsky, (2000). Geometry of the complex of curves II: Hierarchical structure. Geom. Funct. Anal., Vol. 10, pp. 902-974.
[10] H. Zieschang, (1986). On Heegaard diagrams of 3-manifolds. In: On the geometry
of dierential manifolds.

You might also like