detecting incompressibility of boundary in 3-manifolds

14
Detecting Incompressibility of Boundary in 3-Manifolds COLIN ADAMS Department of Mathematics, Williams College, Williamstown, MA 01267, U.S.A. e-mail: [email protected] (Received: 21 March 2001; in final form: 15 March 2002) Abstract. A construction is presented which can be utilized to prove incompressibility of boundary in a 3-manifold W. One constructs a new 3-manifold DW by doubling W along a subsurface in its boundary. If DW is hyperbolic, and if W has compressible boundary, then DW must have a longitude of ‘length’ less than 4. This can be applied to show that an arc a that is a candidate for an unknotting tunnel in a 3-manifold cannot be an unknotting tunnel. It can also be used to show that a ‘tubed surface’ is incompressible. For knot and link com- plements in S 3 , and a an unknotting tunnel, DW is almost always hyperbolic. Empirically, this construction appears to provide a surprisingly effective procedure for demonstrating that spe- cific arcs are not unknotting tunnels. Mathematics Subject Classification (2000). 57M50. Key words. hyperbolic 3-manifold, unknotting tunnel, boundary incompressibility. 1. Introduction Given a specific compact orientable 3-manifold W with boundary, it is important to know whether the boundary is incompressible. If not, one could compress it to sim- plify the manifold. In theory, one can utilize a triangulation of the surface and normal surface theory to determine the incompressibility of the boundary. However, this can be a difficult and computationally intensive procedure. Here, a method is presented which will often demonstrate easily the incompressibility of the boundary of the manifold. One creates a new manifold DW that is obtained by doubling the original manifold across a particular subsurface in the boundary. If DW is hyperbolic, it will be of finite volume with a finite set of cusps. If the boundary of W is compressible, then on the boundary of any set of cusps in DW with disjoint interiors, there must be a nontrivial closed curve of length less than 4. No such curve immediately implies that the boundary of W is incompressible. The SNAPPEA computer program for investigating hyperbolic structures on 3-manifolds, written by Jeffrey Weeks (cf. [18]) can be used to determine whether or not there are nontrivial closed curves of length less than 4 in the boundary of a disjoint set of cusps in DW. One inputs the manifold DW into the program, which then computes the hyperbolic structure Geometriae Dedicata 99: 47–60, 2003. 47 # 2003 Kluwer Academic Publishers. Printed in the Netherlands.

Upload: colin-adams

Post on 06-Aug-2016

213 views

Category:

Documents


1 download

TRANSCRIPT

Detecting Incompressibility of Boundary

in 3-Manifolds

COLIN ADAMSDepartment of Mathematics, Williams College, Williamstown, MA 01267, U.S.A.e-mail: [email protected]

(Received: 21 March 2001; in final form: 15 March 2002)

Abstract. A construction is presented which can be utilized to prove incompressibility of

boundary in a 3-manifold W. One constructs a new 3-manifold DW by doubling W along asubsurface in its boundary. If DW is hyperbolic, and if W has compressible boundary, thenDW must have a longitude of ‘length’ less than 4. This can be applied to show that an arc

a that is a candidate for an unknotting tunnel in a 3-manifold cannot be an unknotting tunnel.It can also be used to show that a ‘tubed surface’ is incompressible. For knot and link com-plements in S3, and a an unknotting tunnel, DW is almost always hyperbolic. Empirically, this

construction appears to provide a surprisingly effective procedure for demonstrating that spe-cific arcs are not unknotting tunnels.

Mathematics Subject Classification (2000). 57M50.

Key words. hyperbolic 3-manifold, unknotting tunnel, boundary incompressibility.

1. Introduction

Given a specific compact orientable 3-manifold W with boundary, it is important to

know whether the boundary is incompressible. If not, one could compress it to sim-

plify the manifold.

In theory, one can utilize a triangulation of the surface and normal surface theory

to determine the incompressibility of the boundary. However, this can be a difficult

and computationally intensive procedure. Here, a method is presented which will

often demonstrate easily the incompressibility of the boundary of the manifold.

One creates a new manifold DW that is obtained by doubling the original manifold

across a particular subsurface in the boundary. If DW is hyperbolic, it will be of

finite volume with a finite set of cusps. If the boundary of W is compressible, then

on the boundary of any set of cusps in DW with disjoint interiors, there must be

a nontrivial closed curve of length less than 4. No such curve immediately implies

that the boundary of W is incompressible. The SNAPPEA computer program for

investigating hyperbolic structures on 3-manifolds, written by Jeffrey Weeks (cf.

[18]) can be used to determine whether or not there are nontrivial closed curves of

length less than 4 in the boundary of a disjoint set of cusps in DW. One inputs

the manifold DW into the program, which then computes the hyperbolic structure

Geometriae Dedicata 99: 47–60, 2003. 47# 2003 Kluwer Academic Publishers. Printed in the Netherlands.

and displays the cusps. From this, one easily determines whether or not there are

curves of length less than 4 in the boundaries.

We will be particularly interested in the applications of this result to unknotting

tunnels. Let M be a compact 3-manifold with one or two toroidal boundary compo-

nents. A properly embedded arc is said to be an unknotting tunnel if the complement

of a regular neighborhood of the arc is a genus two handlebody. The handlebody,

which has comprssible boundary, will correspond to our 3-mainfold W.

A manifold with an unknotting tunnel is said to be of tunnel number one. More

generally, a compact manifold with boundary is said to be tunnel number n if there

exists a set of n disjoint properly embedded arcs such that the complement of an

open regular neighborhood of the set of arcs is a handlebody, and no such set with

fewer arcs exists. Such a set of arcs is called an unknotting tunnel system. A knot or

link in the 3-sphere is said to be tunnel number n if its exterior is tunnel number n.

A good reference for unknotting tunnels is [16].

Torus knots are tunnel number one and their unknotting tunnels were classified in

[5]. Thinking of a torus knot exterior as obtained by gluing two solid tori together

along an annulus in each of their boundaries, the unknotting tunnels correspond

to an arc in the annulus cutting the annulus into a disk and an arc that starts at

the knot, drills in to the core curve of one of the solid tori, travels around the core

curve and then goes back out to the knot.

Two bridge knots and links are tunnel number one. The unknotting tunnels for

two-bridge links (as opposed to knots) were classified in [4] and [10] and consist only

of the so-called upper and lower tunnels, horizontal arcs that connect the two max-

ima and the two minima when the 2-bridge knot or link is drawn so that the two

maxima occur at the same height at the top and the two minima occur at the same

height at the bottom. The unknotting tunnels for two-bridge knots are classified in

[9], and consist of the upper and lower tunnels and their so-called duals.

Tunnel number one knots in S3 are known to be prime. (cf. [14, 17]). The tunnel

number one knots and links in the 3-sphere with nonsimple complements have been

classified along with their unknotting tunnels in [6, 7, 12]. In [13], the authors utilize

2-fold covers to determine the tunnel numbers of prime knots through ten crossings,

showing that all but 87 of the 249 knots are tunnel number one, with the remainder

of tunnel number two. In the case of 2-cusped hyperbolic 3-manifolds, it was shown

in [2] that unknotting tunnels are isotopic to geodesics of length (relative to a max-

imal set of cusps) less than ln(4). Hence, for such a manifold, one need only check a

finite number of candidates to obtain a complete classification of the unknotting

tunnels.

However, determining whether or not a candidate arc in a 3-manifold is an

unknotting tunnel is generally difficult. In this paper, we will describe a technique

and its variations that seem very effective in showing that a particular candidate is

not an unknotting tunnel. Although it utilizes hyperbolic 3-manifold theory, there

is no assumption that the original manifold is hyperbolic. The constructions are

described below.

48 COLIN ADAMS

A compact manifold with torus boundary components will be said to be hyper-

bolic if there is a complete hyperbolic metric on its interior. Such a manifold will

have a finite hyperbolic volume. Because it is hyperbolic, there is a covering map

p: H3 ! intðMÞ. W.Thurston showed that a compact orientable 3-manifold is hyper-

bolic if and only if it is irreducible, boundary irreducible and it does not contain any

essential tori or annuli.

Each torus boundary component corresponds to a cusp in the hyperbolic struc-

ture. Expanding that cusp to be as large as possible while preserving the fact it

has embedded interior yields a so-called maximal cusp (see [1]). A set of cusps that

do not overlap in their interiors and such that any expansion of one of the cusps

would cause such an overlap is called a maximal set of cusps. Note that if the bound-

ary of any maximal set of cusps does not contain a nontrivial closed curve of length

less than 4, then no set of cusps with disjoint interiors contains a nontrivial closed

curve of length less than 4 in its boundary.

Given a 3-manifold W with boundary and a surface F contained in the boundary,

the double ofW along F is the 3-manifold DW that results if we reflectW through F

to obtain a copy RW and take the union of W and RW, which share F.

THEOREM 1.1. Let W be a 3-manifold with compressible boundary. Let fc1; . . . ; ckg

be a system of disjoint nontrivial simple closed curves on @W. Let DW be the mani-

fold that results if W is doubled along the surface F in its boundary given

by F ¼ @W � [NðciÞ. Suppose that DW is hyperbolic. Then it has the following

properties:

ðiÞ The surface F is the totally geodesic fixed-point set of an orientation reversing invo-

lution R.

ðiiÞ There exists a longitude curve of length less than 4 in the boundary of some cusp in

any set of cusps of DW with disjoint interiors.

The first conclusion is immediate. It is the second conclusion in which we are inter-

ested. It will have applications to unknotting tunnels and to incompressibility of

tubed surfaces. Although the hypothesis that the resulting manifold DW be hyper-

bolic seems restrictive, we will see that often, it is not.

If M is a compact orientable 3-manifold with boundary consisting of a nonempty

set of tori, and if fa1; . . . ; ang is an unknotting tunnel system for M, then

W ¼ M � [NðaiÞ is a genus g handlebody. Hence it has compressible boundary. If

we apply the construction above, and DW is hyperbolic, there will be a nontrivial

closed curve of length less than 4 in the boundary of any set of cusps with disjoint

interiors. However, we have lots of choice in the curves fc1; . . . ; ckg.

We now give three explicit choices that result from the general construction. The

first two are relevant to unknotting tunnels and the third to tubing surfaces.

We describe the first construction for one unknotting tunnel a in a manifold M,

although it can also be applied to an unknotting tunnel system. Note that in our

situation, @M must consist of one or two tori. Choose two disjoint simple nontrivial

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 49

closed curves b1 and b2 on the boundary of the manifold such that each passesthrough one of the endpoints of a. They will be parallel if both endpoints of a areon the same boundary component. Remove an open regular neighborhood of

a [ b1 [ b2 fromM, calling the resulting 3-manifoldW. Since a is an unknotting tun-nel, W must be a handlebody. That part of the boundary of W that was not part of

the boundary of M is a 4-holed sphere, call it F. Double the manifold W along F to

obtain a new manifold, denoted D1W. This corresponds to a choice of curves c1 and

c2 in the boundary of W that are core curves of the two annuli in @W � F.

The manifold D1W will have two torus boundary components, and the obvious

reflective involution. In the case that D1W is hyperbolic, the four-holed sphere F will

be the fixed point set of the reflective involution and will thereforre appear in the

hyperbolic structure as a totally geodesic surface.

For convenience, call the intersection of the four-holed sphere with the bound-

ary of either cusp a meridian and a perpendicular curve on the cusp boundary a

longitude. Note that the reflective involution implies that there does exist a per-

pendicular curve in the maximal cusp boundary. In the case that we start with a

knot or link exterior in the 3-sphere, and we choose b1 and b2 to be meridians inthe traditional sense, D1W will be the exterior of a 2-component link in the

3-sphere with our meridian and longitude corresponding to the traditional meri-

dian and longitude.

Since each of the two cusps in the manifold puncture the four-punctured sphere

twice, upper bounds on the size of cusps in hyperbolic surfaces from [3] imply that

the meridian of each cusp, as measured in the boundary of the corresponding maxi-

mal cusp must have length at most 5. We will see an example where 5 is realized.

From this upper bound on the length of the meridian, we obtain the following

corollary.

COROLLARY 1.2. Let a be an unknotting tunnel in a 3-manifold M such that D1W is

hyperbolic. Then at least one of the cusps in any set of cusps with disjoint interiors in

D1W must have volume strictly less than 10.

In the case that the arc a is an unknotting tunnel in a knot or link exterior in S3, it

turns out that D1W is almost always a hyperbolic manifold.

THEOREM 1.3. Let a be an unknotting tunnel for a nontrivial knot or link in the

3-sphere and let b1 and b2 be meridian curves. Then D1W is hyperbolic if and only if a is

not the upper or lower tunnel for a 2-bridge knot or link. When D1W is hyperbolic, the

longitude of some cusp boundary in any set of cusps with disjoint interiors has length

strictly less than 4.

By Theorem 1.1, D1W must always have a reflective involution R. In the case

when there is only one unknotting tunnel, rather than an unknotting tunnel system,

D1W also has an orientation preserving involution J as follows. A genus two handle-

50 COLIN ADAMS

body has an involution that extends to the regular neighborhood of the unknotting

tunnel with 1-dimensional fixed point set intersecting the boundary of the regular

neighborhood of the unknotting tunnel at two points. Upon doubling, we obtain

an involution with circular fixed point set B intersecting F at two points.

In Figure 1, we construct D1W for the knot 52, using one of its dual tunnels. In this

case, the longitude of the maximal cusp corresponding to the component bounding

the twice-punctured disk has length 2. As a side note, this component has meridian

of length exactly 5, the upper bound. It is worth noting that the resulting manifold

D1W also corresponds to known unknotting tunnels for 810 and 820. In fact, any

given D1W is always generated by an infinite set of manifold and candidate unknot-

ting tunnel pairs.

The upper and lower tunnels of the two-bridge knots and links are treated as

exceptional cases in the statement of Theorem 1.3, as in that case, D1W becomes

the exterior of the trivial link of two components. The construction automatically

removes any rational tangles around the unknotting tunnel, and in the case of the

upper and lower tunnels for 2-bridge knots and links, that leaves no knotting.

One might wonder whether a longitude of length less than 4 in D1W forces a to bean unknotting tunnel. In the vast majority of examples of arcs that were not unknot-

ting tunnels, SNAPPEA (the hyperbolic structures program written by Jeffrey

Weeks [18]) generated a D1W with all longitudes of length at least 4. However, there

is at least one example with longitude less than 4 which does not correspond to an

unknotting tunnel.

In fact, many of the examples of arcs that are not unknotting tunnels have a long-

itude of length exactly 4 for the following reason. For many of the arcs under

consideration, there is an annulus properly embedded in W which touches F along

an arc in its boundary and, hence, which doubles to a twice-punctured disk in

D1W. This occurs for the example in Figure 1 and for the dual tunnels in any of

the two-bridge knot complements, for instance. Hence, there is an incompressible

boundary-incompressible thrice-punctured sphere properly embedded in D1W, one

puncture of which corresponds to a longitude of one of the cusps. As is well known,

the thrice-punctured sphere is totally geodesic. But a maximal cusp in the thrice-

punctured sphere has boundary of length exactly 4. Hence, when this cusp in

D1W is maximized, either its point of tangency with itself occurs in the thrice-

Figure 1. Construction of D1W for dual tunnel in 52 knot complement.

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 51

punctured sphere, in which case the length of this longitude is exactly 4, or the point

of tangency of this cusp with itself does not occur in the thrice-punctured sphere, in

which case the length of the longitude is strictly less than 4.

The construction for D1W generalizes to an n-arc unknotting tunnel system in a

manifold M. In the case M is a link exterior in the 3-sphere, D1W is the exterior

of a 2n-component link in the connected sum of n � 1 copies of S2 S1. Again, at

least one cusp boundary in any set of cusps with disjoint interiors must have longi-

tude of length strictly less than 4.

We now describe a second construction based on the general framework. Let

fa1; . . . ; ang be an unknotting tunnel system for the manifold M, which has non-

empty toroidal boundary components. Let W be the complement of the open neigh-

borhoods of the unknotting tunnel system. Let F ¼ @W � [@NðaiÞ. Let D2W be the

double of W along F. Note that D2W is obtained by doubling M along its toroidal

boundaries and then removing neighborhoods of the doubled arcs in the system.

When D2W is hyperbolic, at least one maximal cusp must have longitude of length

less than 4. In the case that there is a single unknotting tunnel and M is a knot ex-

terior in the 3-sphere, D2W is usually hyperbolic:

THEOREM 1.4. If a is an unknotting tunnel for a nontrivial knot in the 3-sphere, then

D2W is hyperbolic if and only if a is not an unknotting tunnel for a torus knot or cable

knot. When D2W is hyperbolic, then a longitude on the single maximal cusp must have

length strictly less than 4.

Note that the cable knots that are tunnel number one are a relatively restricted

class, corresponding to certain cables on torus knots (cf. [6, 12]). Although the fact

that D2W has only one cusp is an advantage to this construction, the disadvantage is

that D2W will not be a knot or link exterior even though M may be one. Note that

between the two constructions, the only tunnel number one knots in S3 such that

their doubles under each of these constructions are not hyperbolic are the 2-braid

knots. In all other cases, either D1W or D2W will be hyperbolic.

The third construction was suggested by William Menasco. Let G be a properly

embedded incompressible boundary incompressible surface with boundary in a com-

pact connected orientable manifold W with toroidal boundary components. From

such a surface, one would like to construct a closed incompressible surface by adding

tubes that run around the toroidal boundary components from one boundary com-

ponent of G to another. See [8, 11, 15] for more on this tubing operation. One needs

to know that the resulting surface is incompressible.

Let G0 be the surface obtained by tubing together pairs of boundary components

of G by disjoint annuli A1; . . . ;An in @W. Let W0 be the manifold that is obtained

when W is cut open along G. Note that it may have more than one component.

52 COLIN ADAMS

Let D3W be the manifold obtained when W0 is doubled along the copies of G in its

boundary.

THEOREM 1.5. Let G be a properly embedded orientable incompressible boundary-

incompressible surface in a connected orientable 3-manifold W, Let G0 be obtained by

tubing together pairs of boundary components of G. If D3W is hyperbolic and G0 is

compressible, then there must be a longitude with length less than 4 in the cusp

boundary in any set of cusps with disjoint interiors in one of the components of D3W.

The remainder of the paper is devoted to proofs of the above theorems and a

corollary.

2. Proofs

THEOREM 1.1. Let W be a 3-manifold with compressible boundary. Let fc1; . . . ; ckg

be a system of disjoint nontrivial simple closed curves on @W. Let DW be the manifold

that results if W is doubled along the surface F in its boundary given by

F ¼ @W � [NðciÞÞ. Suppose that DW is hyperbolic. Then it has the following properties:

ðiÞ The surface F is the totally geodesic fixed-point set of an orientation reversing invo-

lution R.

ðiiÞ There exists a longitude curve of length less than 4 in the boundary of some cusp in

any set of cusps of DW with disjoint interiors.

Proof. Part (i) is immediate from the construction, since the fixed point set of an

involution must be totally geodesic. It remains to prove Part (ii).

Assume that we have fixed a choice of cusps fC1;C2; . . . ;Cng with disjoint inter-

iors. Lifting the manifold DW to hyperbolic 3-space, fC1;C2; . . . ;Cng lift to a set

of horoballs with disjoint interiors, and the surface F lifts to a set of totally geodesic

planes, with boundaries passing through the centers of the horoballs. Since W has

compressible boundary, there exists a compressing disk D, which lifts to a disk D0

with interior in the complement of the horoballs and with boundary in the union

of the boundaries of the horoballs and the totally geodesic planes. The boundary

of D cannot be contained entirely in F, or in the boundary of one of the cusps, as

they are each incompressible. Hence @D0 must lie in p�1ð@C1 [ � � � @Cn [ FÞ, and pass

through an alternating sequence of horospheres and totally geodesic planes. Assume

we have isotoped D to minimize the number of times that @D0 intersects horospheres

and totally geodesic planes in p�1ð@C1 [ � � � @Cn [ FÞ. Then there must be a finite

sequence of horospheres alternating with geodesic planes, each of which intersects

the next in the sequence, such that the sequence begins and ends at the same horo-

sphere. Choose that horosphere to be centered at infinity in the upper-half-space

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 53

model, and then take a smallest horosphere H in the cyclic sequence (smallest in the

Euclidean sense). If there is a subsequence of equally small horospheres in the

sequence, choose H on the end of the subsequence. In order that one of its neighbor-

ing horospheres be at least as big as (or bigger than) H, it must be the case that

the distance on H from its peak to its intersection by the connecting totally geodesic

plane be less than or equal to (or less than) 1. Hence the distance on the horosphere

between the two geodesic planes that connect this horosphere to its two neighboring

horospheres must be strictly less than 2. However, that distance appears in the cusp

boundary in the manifold as exactly half a longitude of one of the cusps, beginning

and ending on F. When we reflect through F this path becomes a longitude in the

boundary of one of the cusps of length less than 4. &

Given a properly embedded arc a in a manifold M that is a candidate for an

unknotting tunnel, then in order to show that a is not an unknotting tunnel, it isenough to check that if W ¼ M � NðaÞ, and D1W is hyperbolic, then there exists a

choice of cusps in D1W with disjoint interiors, neither of which has a longitude

shorter than 4. Although not as effective a means of eliminating unknotting tunnel

candidates, one further has:

COROLLARY 1.2. Let a be an unknotting tunnel in a 3-manifold M such that D1W is

hyperbolic. Then at least one of the cusps in any set of cusps with disjoint interiors in

D1W must have volume strictly less than 10.

Proof. The area contained within a maximal set of cusps in a totally geodesic

four-punctured sphere is at most 12 (cf. [3]). Any single cusp in the set has an area of

at least 1. Since either cusp in D1W intersects F twice, and each intersection must

have the same area, 5 is an upper bound on the meridian of either cusp. At least one

cusp must have longitude strictly less than 4. This cusp then has area of its boundary

less than 20 and hence volume less than 10. &

THEOREM 1.3. Let a be an unknotting tunnel for a nontrivial knot or link in the

3-sphere and let b1 and b2 be meridian curves. Then D1W is hyperbolic if and only if a is

not the upper or lower tunnel for a 2-bridge knot or link. When D1W is hyperbolic, the

longitude of some cusp boundary in any set of cusps with disjoint interiors has length

strictly less than 4.

Proof. Let M be the original knot or link exterior. We consider W as a handle-

body contained in D1W, and RW its reflection across F, the 4-holed sphere. Let F 0 be

the sphere obtained in S3 by capping off the four holes of F.

We first note that F is incompressible and boundary-incompressible in D1W. It is

enough to show that as a surface on the boundary ofWðor RW Þ, it is incompressible

and boundary-incompressible. Note that W is a cube-with-wormholes contained in

S3. For convenience, denote the cube in S3, which is bounded by F 0, by C and the

two arcs, the removal of whose neighborhood generates the wormholes, byV1 andV2.

54 COLIN ADAMS

If F compressed inW, then the boundary of the compressing disk D must separate

two of the holes in F from the other two. Hence D separates C into two balls, each of

which contains one of V1 and V2. Then V1 and V2 must be trivally knotted in each

ball since, otherwise, we could construct an essential torus in W, which cannot exist

in a handlebody. However, this forces V1 and V2 to generate a rational tangle in C.

This means that a is the upper or lower tunnel in a 2-bridge knot or link exterior andDW is the exterior of the trivial two- component link, a case we have excluded from

consideration.

If F boundary-compressed, it must do so in W. This would imply that V1 and V2can be isotoped into F. That in turn implies F is compressible, a contradiction.

In order to show that D1W is hyperbolic, work of Thurston shows it suffices to

demonstrate that D1W is irreducible, boundary-irreducible, and it contains no essen-

tial tori or annuli.

Suppose D1W is reducible. Let S be a reducing sphere. Then S must intersect

F, since the handlebody W is irreducible. After minimizing the intersections of S

and F, we can take an innermost intersection curve on S. This yields a disk D0 in

W with boundary of F. By incompressibility of F, the boundary of the disk

must be trivial in F, and by irreducibility of W, the disk D0 the disk D0 could

have been isotoped to eliminate another intersection curve, a contradiction to

minimality.

Suppose D1W is boundary-reducible. Then there is a disk D embedded in DW with

nontrivial boundary on one of the two toroidal boundary components. D must inter-

sect F as otherwise, D exists in M, a contradiction to the incompressibility of a mer-

idian in a knot or link exterior in the 3-sphere. By incompressibility of F, the closed

intersection curves in D \ F can be eliminated. Choose D to have the minimum pos-

sible number of intersection curves with F. Taking an outermost intersection arc in

D, we obtain a sub-disk of D that lies entirely inW and that has boundary consisting

of one arc in @W � F and one arc in F. This contradicts the boundary-incompressi-

bility of F.

Suppose there exists an essential torus T in D1W. If T does not intersect F, then it

lies entirely in W, and is essential there. But this is not possible for a handlebody.

On the other hand, if T intersects F, then we can eliminate trivial intersection

curves on T and F. Assume that we have minimized the number of remaining inter-

section curves. Then T intersects W in a set of annuli.

Let A be such an annulus. Suppose first that A is knotted in S3, in the sense that

the torus formed from A and an annulus in F 0 with the same boundary is knotted

in S3. Then A splits C into a knot exterior E and a ball B. Since A is incompres-

sible, B must contain at least one of V1 and V2. Since A cannot split a knot exter-

ior from W, E must contain one of V1 and V2 as well, say V2. Since W is a

handlebody, it cannot contain a knot exterior with incompressible boundary. That

is to say, V1 must be prime, i.e. unknotted in B. Hence, A is parallel to @NðV1Þ

through B.

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 55

On the other hand, if A is unknotted, then A splits the ball bounded by F 0 into a

ball B0 and a solid torus S0. Then B0 must contain one of V1 and V2 by incompres-

sibility of A, say V1, and S0 must contain V2 as otherwise, the number of intersection

curves of T with F could be lowered. By the fact W is a handlebody, V1 must be

unknotted in the ball. Hence, A is parallel into @NðV1Þ.

Thus T intersects W in a set of annuli, each of which is individually parallel to

@NðV1Þ or @NðV2Þ. Then, in D1W;T is made up of annuli to each side of F that

are parallel to @NðV1Þ, @NðV2Þ and their reflections through F. If a given annulus

A in W is parallel to @NðV1Þ then the annulus that is its continuation in RW at

one of its boundary components must be parallel to R@NðV1Þ, since this bound-

ary component wraps once around the hole corresponding to R@NðV1Þ. Thus T is

boundary-parallel in D1W, contradicting our assumption that it is essential.

Suppose that D1W contains an essential annulus A. We first show it must intersect

F. Otherwise it would be contained entirely inW or RW, sayW. Then A has bound-

aries that are meridians on R@NðV1Þ or R@NðV2Þ. But then it becomes a sphere when

capped off with meridianal disks from NðV1Þ and/or NðV2Þ, meaning that both of its

boundary components are meridianal curves on the same strand, say V1, and that V1cannot be knotted inside the ball cut off by A. But then A is parallel to @NðV1Þ, con-

tradicting the fact A is not boundary-parallel.

So A must intersect F. After removing all trivial intersection curves in A \ F, all

the remaining curves are either closed or arcs. If they are all closed, we can choose

an annulus A0 on A containing no other intersection curves such that one of its

boundary components, call it g, is a boundary of A and the other is an intersection

curve with F. Then g is a meridian on @NðV1Þ or @NðV2Þ, say @NðV1Þ for convenience.

Then A0 capped off with a meridian disk in NðV1Þ is a disk and therefore it separates

the ball bounded by F into two balls. Again, V1 must be unknotted to the side of A0

that does not contain V2 since a handlebody does not contain a knot exterior with

incompressible boundary. We can isotope A through that side to eliminate an inter-

section curve. Repeating this process, we eventually eliminate all intersection curves,

a contradiction.

Otherwise, A \ F consists of essential arcs in A. Hence, there exists a disk D con-

tained in A, such that its boundary consists of four arcs, two contained in F and two

contained in @W � F. The double of D is an annulus in D1W. Suppose first that for

the two arcs that are contained in @W � F, one is contained in @NðV1Þ and one con-

tained in @NðV2Þ. Then each arc must begin at one boundary component of @NðViÞ

and end at the other. Hence V1 and V2 are parallel in C. Since they cannot be knot-

ted, as that would create an essential torus in W, it must be the case that V1 and V2are two unknotted parallel arcs in C. HenceM is a 2-bridge knot or link and a is theupper or lower tunnel.

Suppose that both of the arcs on the boundary of D that are in @W � F are in

@NðV1Þ, for instance. Then D cuts a ball from W. Since A is boundary-incompres-

sible, V2 must be contained in the ball. Since W contains no essential tori,

V2 must be trivially knotted within the ball. This again implies V1 and V2 are

56 COLIN ADAMS

parallel, and trivial, and that M is a 2-bridge knot or link exterior and a is theupper or lower tunnel. &

THEOREM 1.4. If a is an unknotting tunnel for a nontrivial knot in the 3-sphere, then

D2W is hyperbolic if and only if a is not an unknotting tunnel for a torus knot or cable

knot. When D2W is hyperbolic, then a longitude on the single maximal cusp must have

length strictly less than 4.

Proof. Let M be the original knot exterior. First note that if a is an unknottingtunnel for a torus knot or a cable knot, then by the classification of such unknotting

tunnels in [5, 6, 12], one can see that there is either a disk or annulus in W that

doubles to an essential annulus or essential torus in D2W, and it is therefore not

hyperbolic.

We now prove the theorem in the other direction. We first show that F is

incompressible and boundary-incompressible in D2W. It is incompressible since

@M is incompressible in M. If F boundary-compressed, then there would be a

disk D with one arc of its boundary in F and the remaining arc in @NðaÞ. Thena would be isotopic into F. However, this implies M itself is a solid torus, a

contradiction.

Now, we will show that D2W is irreducible, boundary-irreducible, and contains

no essential tori or annuli. That it is irreducible follows immediately from the fact

M is irreducible and F is incompressible. If it boundary-reduced, and the reducing

disk D did not intersect F, there would be a nonseparating sphere in M, which

cannot occur in S3. Assume D has minimal intersections with F. Take as intersec-

tion curve in F \ D that is outermost on D. This generates a disk D0 with one

boundary in @NðaÞ and the other in F, contradicting the boundary incompressi-

bility of F.

Suppose T is an essential torus in D2W. It must intersect F since there are no essen-

tial tori in a handlebody. After minimizing intersections, we can assume all intersec-

tion curves are nontrivial on F and T, and F cuts T into a set of annuli. Let A be such

an annulus in W. It is incompressible in W, but since a handlebody cannot contain

an essential annulus, it must be boundary-compressible in W. Note that both of

its boundary curves must either be trivial or nontrivial on @M by incompressibility

of @M.

Suppose first that the boundaries of A are trivial on @M. Capping each boundary

component off with a disk in @M, we obtain a sphere which must bound a ball inM.

If a is contained in the ball, it is unknotted in the ball since a handlebody cannot con-tain a knot exterior with incompressible boundary. So A is parallel through the ball

to @N(a).If a is not in the ball, then since A is incompressible, the two boundary compo-

nents of A are concentric on @M, and at least one of the ends of a is containedinside the disk on @M bounded by the inner boundary component of A. Then A is

boundary-parallel in W, a contradiction to the minimality of intersection curves

between T and F.

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 57

Suppose now that the boundary curves of A are nontrivial on @M. If they are both

meridians, then by capping off A with meridian disks, we have a sphere bounding

two balls in S3. If to one side, the arc of K contained in the ball is unknotted, then

A is parallel to @M in M. If it remains parallel in W, we have a contradiction to

minimality of intersections of F and T. If it does not, then the ball also contains

a, a contradiction to a being a valid unknotting tunnel. If the arcs of K are knottedto each side, the knot is composite, a contradiction to being tunnel number one (cf.

[14, 19]).

If the boundary curves of A are not meridians, then we will show that they consist

of exactly one longitude. First note that A is either boundary-compressible or bound-

ary-incompressible in M. In the first case, A is boundary-parallel in M. Then by

minimality of intersections of F and T, a would have to be contained in the solidtorus that gives the isotopy of A to @M. But this contradicts the fact a is an unknot-ting tunnel. So A is boundary-incompressible in M.

Suppose that the curves in @A are not longitudinal on @N(K). Let b be the corecurve of A. Then b is a (p, q)-cable on K, q5 2. Each half of the annulus A realizes

the cabling annulus for this (p, q)-cable on K. But such an annulus is unique up to

isotopy, so the one half of the annulus is isotopic to the other half. This yields a

boundary-compression of A, a contradiction. But if @A is longitudinal on @N(K),

then K is a torus knot or cable knot.

So the only possibility other than K being a torus knot or cable knot is that each of

the annuli that make up T are parallel into @N(a) or its reflection. Hence, there mustbe exactly two such annuli, and T is boundary-parallel in D2W, a contradiction.

Suppose now that A is an essential annulus in D2W. First suppose that A does not

intersect F. Then both boundary components are meridians on @N(a). Capping offthe two boundaries of A by meridian disks in N(a) yields a sphere. Since a handle-body cannot contain a knot exterior with incompressible boundary, a is unknottedin the sphere, and A must be parallel into @N(a) in W, a contradiction.

Now suppose that A intersects F in a minimal set of simple closed curves. Since

each boundary component of A must be a meridian of N(a), which is trivial in M,

A must intersect F in a trivial curve as well. This intersection curve must contain

an endpoint of a. Again, capping A off with disks yields a sphere inM, which bounds

a ball in M. That part of a that is inside the sphere must be unknotted. Then we canslide one component of @A down along @N(a) to lower the number of intersectioncurves of A\F, a contradiction to minimality.

Finally, suppose that the intersection curves of A with F are essential arcs in A.

They cut A into a set of disks, each of which has two arcs in its boundary in F

and the remaining two arcs in its boundary in @N(a). Let D be such a disk in W.

There are three possibilities. First, @D could be a trivial loop on @W. Then A is

boundary-compressible.

Second, D could be a separating compressing disk in the handlebody W. Then D

cuts W into two solid tori. There is an annulus contained in N(a)[D that cuts M

into two solid tori. The only nontrivial knot exteriors in the 3-sphere that consist

58 COLIN ADAMS

of two solid tori glued together along an annulus in the boundary of each are the

torus knot exteriors, and then the unkotting tunnels are the standard ones.

Third, D could be a nonseparating compressing disk forW. Then the complement

of D in W is a solid torus. When N(a) is added to D, we obtain a nonseparating

annulus in M that cuts M into a solid torus. The only knot or link exteriors in S3

that are obtained by gluing together two parallel annuli on the boundary of a solid

torus are cable spaces, which have more than one boundary component, a contradic-

tion to M being a knot exterior.

The fact that the single maximal cusp must have a longitude of length less than 4

follows immediately from Theorem 1.1(ii). This completes the proof. &

THEOREM 1.5. Let G be a properly embedded orientable incompressible boundary-

incompressible surface in a connected orientable 3-manifold W, Let G0 be obtained by

tubing together pairs of boundary components of G. If D3W is hyperbolic and G0 is

compressible, then there must be a longitude with length less than 4 in the cusp

boundary in any set of cusps with disjoint interiors in one of the components of D3W.

Proof. The copies of the surface G becomes totally geodesic in D3W, and the

existence of a compressing disk in D3W for G0 causes the existence of a closed path

through an alternating sequence of geodesic planes and horospheres that bounds a

disk in H3. As in the proof of Theorem 1.1, this forces the existence of a longitude of

length less than 4. &

Acknowledgements

Thanks to Abigail Thompson and William Menasco for helpful conversations. The

work in this paper would not have been possible without the explicit examples that

one can compute using the SNAPPEA hyperbolic structures program, written by

Jeffrey Weeks. It is available at http://humber.northnet.org/weeks/. This work was

supported in part by NSF Grant DMS-9803362.

References

1. Adams, C.: The noncompact hyperbolic 3-manifold of minimal volume, Proc. Amer.Math. Soc. 100 (1987), 601–606.

2. Adams, C.: Unknotting tunnels in hyperbolic 3-manifolds, Math. Ann. 302 (1995),177–195.

3. Adams, C.: Maximal cusps, collars and systoles for hyperbolic surfaces, Indiana Univ.Math. J. 47(2) (1998), 419–437.

4. Adams, C. and Reid, R.: Unknotting tunnels in two-bridge knot and link complements,Comment. Math. Helv. 71 (1996), 617–627.

5. Boileau, M., Rost, M. and Zieschang, H.: On Heegaard decompositions of torus exteriors

and related Seifert spaces, Math. Ann. 279 (1988), 553–581.6. Eudave Muoz, M.: On nonsimple 3-manifolds and 2-handle addition, Topology Appl.

55(2) (1994), 131–152.

DETECTING INCOMPRESSIBILITY OF BOUNDARY IN 3-MANIFOLDS 59

7. Eudave Muoz, M. and Uchida, Y.: Non-simple links with tunnel number one, Proc. Amer.

Math. Soc. 124(5) (1996), 1567–1575.8. Finkelstein, E. and Moriah, Y.: Tubed incompressible surfaces in knot and link comple-ments, Topology Appl. 96(2) (1999), 153–170.

9. Kobayashi, T.: Classification of unknotting tunnels for two bridge knots, In: Proc.Kirbyfest (Berkeley, CA, 1998), Geom. Topol. Monogr. 2, Geom. Topol., Coventry,1999, pp. 259–290.

10. Kuhn, M.: Tunnels of 2-bridge links, J. Knot Theory Ramification 5(2) (1996), 167–171.11. Menasco, W.: Closed incompressible surfaces in alternating knot and link complements,

Topology 23(1) (1984), 37–44.12. Morimoto, K. and Sakuma, M.: On unknotting tunnels for knots,Math. Ann. 289 (1991),

143–167.13. Morimoto, K., Sakuma, M. and Yokata, Y.: Identifying tunnel number one knots,

J. Math. Soc. Japan 48(4) (1996), 667–687.

14. Norwood, F. H.: Every two-generator knot is prime, Proc. Amer. Math. Soc. 86(1) (1982),143–147.

15. Oertel, U.: Closed incompressible surfaces in complements of star links, Pacific J. Math.

111(1) (1984), 209–230.16. Sakuma, M.: The topology, geometry and algebra of unknotting tunnels, Chaos, Solitons

and Fractals 9(4/5) (1998), 739–748.17. Scharlemann, M. G.: Tunnel number one knots satisfy the Poenaru conjecture, Topology

Appl. 18(2–3) (1984), 235–258.18. Weeks, J.: The SNAPPEA hyperbolic structures program, available at http://humber.

northnet.org/weeks/.

60 COLIN ADAMS