Anda di halaman 1dari 126

Low Dimensional Topology

Brandon Williams
mbw@math.sunysb.edu
November 25, 2008
Contents
1 Transversality and Intersection Theory 3
1.1 Immersions, Submersions and Transversality . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Intersection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Knot Theory 7
2.1 Basic Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The Alexander Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Knot Signatures and 4-Ball Genus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Cyclic and Branched Covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Knots in other 3-Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Handlebodies 26
3.1 Handlebody Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Handle Moves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 The Fundamental Group and Homology of a Handlebody . . . . . . . . . . . . . . . 33
3.4 2-Dimensional Handlebodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.5 Heegaard Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.6 Kirby Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4 Morse Theory 43
4.1 Morse Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Handlebody Decompositions Induced by Morse Functions . . . . . . . . . . . . . . . 46
4.3 Properties of Gradient Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4 Existence of Morse Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5 Kirby Calculus 52
5.1 Surgery on Links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Kirby Moves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 The Rational Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6 4-Manifold Theory 60
6.1 Intersection Forms on 4-Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.2 Kirby Calculus Reinterpreted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
1
7 Morse Homology 66
7.1 The Morse-Smale Condition and Moduli Spaces of Flow Lines . . . . . . . . . . . . . 66
7.2 The Compactness and Gluing Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.3 The Morse Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.4 A Natural Isomorphism between Morse and Singular Homology . . . . . . . . . . . . 73
7.5 Direct Invariance of Morse Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.6 Orientations on Moduli Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.7 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.8 Wittens Complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8 Heegaard Floer Homology 87
8.1 Construction for 3-Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.2 Invariance of Heegaard Floer Homology . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.2.1 Independence of Complex Structures . . . . . . . . . . . . . . . . . . . . . . . 95
8.2.2 Independence of Isotopies of Attaching Circles . . . . . . . . . . . . . . . . . 97
8.2.3 Independence of Handle Slides . . . . . . . . . . . . . . . . . . . . . . . . . . 98
8.2.4 Independence of Stabilizations . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.3 The Surgery Exact Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
8.4 The Absolute Q Grading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.5 Heegaard Floer Knot Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A Homotopy Groups 117
B Duality Theorems 118
C Cohomology with Homotopy 119
D Homeomorphisms of Surfaces 120
E Some Analysis 121
F Some Exercises 124
References 126
2
1 Transversality and Intersection Theory
1.1 Immersions, Submersions and Transversality
Let ) : ` be a smooth map between smooth manifolds. A point j ` is said to be regular
if d)
p
: T
p
` T
f(p)
is surjective, otherwise it is said to be critical. The image of a critical
point is called a critical value, and a regular value in is any point that is not a critical value.
We say that ) is an immersion at j ` if d)
p
is injective, and a submersion at j ` if d)
p
is surjective. Recall the inverse function theorem and its corollaries pertaining to immersions and
submersions.
Theorem 1.1 (Inverse Function Theorem). If d)
p
: T
p
` T
f(p)
is an isomorphism, then ) is a
local dieomorphism at j, i.e. there is a neighborhood l of j and \ of )(j) such that )[
U
: l \
is a dieomorphism.
Corollary 1.1. If ) : `
m

n
is an immersion at j `, then there are coordinates :
l R
m
around j ` and : \ R
n
around )(j) such that )
1
(r
1
, . . . , r
m
) =
(r
1
, . . . , r
m
, 0, . . . , 0).
Proof. First let and be any charts such that (j) = (j) = 0, and let p = )
1
. Then
dp
0
: 1
m
1
n
is injective since ) is an immersion, so we can change the coordinates so that
dp
0
=
_
1
m
0
_
where 1
m
is the : : identity matrix and 0 is the (n :) : zero matrix. We cannot apply
the inverse function theorem to p since : < n, but if we dene G : (l) 1
nm
1
n
by
G(r, .) = p(r) + (0, .), then dG
0
= 1
n
. So G is a local dieomorphism, and G is another
chart around )(j) of . In these coordinates we have that the local representation of ) is just
(r
1
, . . . , r
m
) (r
1
, . . . , r
m
, 0, . . . , 0).
Corollary 1.2. If ) : `
n

m
is an submersion at j `, then there are coordinates : l R
n
around j ` and : \ R
m
around )(j) such that )
1
(r
1
, . . . , r
n
) = (r
1
, . . . , r
m
).
Proof. First let and be any charts such that (j) = (j) = 0, and let p = )
1
. Then
dp
0
: 1
n
1
m
is injective since ) is a submersion, so we can change the coordinates so that
dp
0
=
_
1
m
0
_
where 1
m
is the : : identity matrix and 0 is the : (n :) zero matrix. We cannot apply
the inverse function theorem to p since : < n, but if we dene G : 1
n
(\ ) 1
nm
by
G(r
1
, . . . , r
n
) = (p(r), r
m+1
, . . . , r
n
), then dG
0
= 1
n
. So G is a local dieomorphism, and G
1

is another chart around j of `. In these coordinates we have that the local representation of ) is
just (r
1
, . . . , r
n
) (r
1
, . . . , r
m
).
Corollary 1.3. If is a regular value of ) : `
n

m
, then )
1
() is a submanifold of ` with
codim)
1
() = :.
Proof. Let j )
1
(), then ) is a submersion at j, so by corollary 1.2 we can nd coordinates
: l R
n
around j and : \ R
m
around such that )
1
(r
1
, . . . , r
n
) = (r
1
, . . . , r
m
) and
() = (0, . . . , 0). Thus for r l and )(r) = we have (r) = (0, . . . , 0, r
m+1
, . . . , r
n
) for some
r
m+1
, . . . , r
n
. So restricted to l )
1
() provides a chart of the relatively open set l )
1
()
into R
nm
, making )
1
() into a submanifold of dimension n :.
3
Consider the following simple application of corollary 1.3. For real-valued functions )
1
, . . . , )
k
:
` R, when is the set 7 of common zeros a manifold? If we set ) = ()
1
, . . . , )
k
) : ` R
k
,
then it is clear that )
1
(0) is a manifold if 0 is a regular value of ). Another way of looking at
this is to see that d)
p
= ((d)
1
)
p
, . . . , (d)
k
)
p
) : T
p
` T
f(p)
R
k
, and this is surjective if and only
if (d)
1
)
p
, . . . , (d)
k
)
p
are linearly independent functionals. When this happens we say )
1
, . . . , )
k
are
independent at j, and so we have the following corollary.
Corollary 1.4. If )
1
, . . . , )
k
: ` R are independent at all points where they simultaneously
vanish, then the set of common zeros 7 is a submanifold with dim7 = dim` /.
In the case of the above corollary we say that the submanifold 7 has been cut out by the
functions )
1
, . . . , )
k
. However, not every submanifold can be globally cut out by independent
functions.
Now that we know the pre-image of regular values are submanifolds, we naturally wonder if the
pre-image of submanifolds are submanifolds. Two smooth maps ) : ` and p : 1 are
said to be transversal if at any point j ` and 1 for which )(j) = p() we have
d)
p
(T
p
`) +dp
q
(T
q
) = T
f(p)

and we denote this by ) p. Note that the above sum does not have to be direct, and in fact
most of the time will not be. If p is just the inclusion of a submanifold 1 of , then we say ) is
transverse to 1 and write ) 1. If ) is also an inclusion of a submanifold ` of , then we say
the submanifolds ` and 1 are transversal, and write ` 1.
Proposition 1.1. If ) : ` is transverse to a submanifold 1 in , then )
1
(1) is a subman-
ifold of ` with codim)
1
(1) = codim1. In particular, if two submanifolds ` and 1 of are
transversal, then ` 1 is another submanifold of and codim(` 1) = codim` + codim1.
An important feature of immersions, submersions, transversal intersections and embeddings is
that they can be perturbed slightly, and they will remain in their respective class. To make
this precise we need to dene what we mean by perturbation. Fix a smooth map )
0
: ` ,
and let 1 : ` [1, 1] be a smooth map such that 1(, 0) = )
0
. In fact, we will use the
notation )
t
= 1(, t). Then any of the smooth maps )
t
: ` for t [1, 1] are called smooth
perturbations of )
0
; that is, we just smoothly homotoped )
0
a little bit. We say that a statement
1 dened on the space of smooth functions C

(`, ) is stable if for every ) C

(`, ) such
that 1()) and every family of perturbations 1 of ), then there is a small enough c such that 1()
t
)
for all [t[ < c.
Theorem 1.2. Let ` be compact. The following statements are stable as statements on C

(`, ).
1. 1()) = ) is an immersion.
2. 1()) = ) is an submersion.
3. 1()) = ) is a local dieomorphism.
4. For a xed submanifold 7 , 1()) = ) is transversal to 7.
5. 1()) = ) is an embedding.
6. 1()) = ) is an dieomorphism.
4
Proof.
1. Suppose )
0
: `
m

n
is an immersion, and take a family 1 : `[1, 1] perturbations
of )
0
, i.e. 1(, 0) = )
0
. The result will follow if we prove every point (r, 0) ` [1, 1]
has an open neighborhood l
x
such that d)
p
is injective for all j l. For then we can
take their union l =
x
l
x
and choose an c 0 small enough so that ` (c, c) l.
Since this is now just a local question we might as well assume ` is an open subset of R
m
and is an open subset of R
n
. Since (d)
0
)
x
is injective, we have that the Jacobian matrix
(()
0
)
i
,r
j
(r))
ij
contains at least : linearly independent columns. We can of course assume
that these columns are the rst : columns of the Jacobian. If we consider the Jacobian as a
function of (r, t) A [1, 1], then we see that the determinant of the rst : columns is a
smooth function that is not equal to zero at (0, r), hence we can nd an c 0 small enough
so that it is not zero for (t, r), [t[ < c. Therefore all the )
t
s are immersions for [t[ < c.
2. The proof of this identical to 1 as now the Jacobian has n linearly independent rows.
3. This follows directly from 1 since a local dieomorphism is just an immersion for when
dim` = dim.
4. The condition of being transverse has an equivalent reformulation in terms of local submer-
sions, so this follows from 2.
5. Recall that an embedding is a proper, injective immersion )
0
: ` . Since ` is compact,
)
0
is automatically proper, so we only need to prove that if )
0
is injective, then )
t
is injective
for [t[ less than some small c 0. We prove this by contradiction, so we can assume there is
a sequence of numbers t
i
0 and distinct points r
i
,= j
i
such that )
t
i
(r
i
) = )
t
i
(j
i
). Since `
is compact we can pass to a subsequence such that t
i
0, r
i
r, and j
i
j. If we dene
the function G : ` 1 1, then we have G(r
i
, t
i
) = G(j
i
, t
i
) (by denition), hence
G(r, 0) = lim
i
G(r
i
, t
i
) = lim
i
G(j
i
, t
i
) = G(j, 0)
Since )
0
is injective this implies r = j. The Jacobian of G at (r, 0) is in block form
_
_
_
_
_

d()
0
)
x
.
.
.

0 0 1
_
_
_
_
_
But, d()
0
)
x
is injective, hence dG
(x,0)
has at least : + 1 linearly independent columns and
so G is an immersion at (r, 0). This will remain true in a small neighborhood of (r, 0),
which must contain the points (r
i
, t
i
) and (j
i
, t
i
) for large enough i. This contradicts that
G(r
i
, t
i
) = G(j
i
, t
i
), therefore no such sequence can exist, and so )
t
is injective for [t[ small
enough.
6.
Now let us extend the ideas of transversality to manifolds with boundary. For a manifold `
with boundary, and a smooth map ) : ` , the restriction of ) to the boundary ` will be
denoted by ).
UNFINISHED
5
1.2 Intersection Theory
Recall the following classication of 1-dimensional manifolds and one of its immediate corollaries.
Proposition 1.2. Every compact 1-dimensional manifold is homeomorphic to a disjoint union of
circles and closed intervals.
Corollary 1.5. The number of boundary components of any compact 1-dimensional manifold is
even.
Consider a smooth map ) : ` , with ` compact, and submanifold 7 `. By slightly
perturbing ), if necessary, we can assume ) 7 and so )
1
(7) is a submanifold of `. If
codim7 = dim`, then codim)
1
(7) = codim7 = dim`, hence )
1
(7) is a 0-dimensional
manifold. Since ` is compact, )
1
(7) consists of nitely many points, so we dene the mod-2
intersection number of ) and 7 by 1
2
(), 7) = #)
1
(7) mod 2. This is well-dened by the
following proposition.
Proposition 1.3. If ), p : ` are homotopic, then 1
2
(), 7) = 1
2
(p, 7) mod 2.
Proof. Let 1 : `1 be a homotopy between ) and p. We can perturb 1 slightly, while keeping
1
0
and 1
1
xed, to make 1 7. Then 1
1
(7) is a submanifold of ` 1 with codim1
1
(7) =
codim7 = dim`, hence dim1
1
(7) = 1. The boundary of 1
1
(7) consists of
1
1
(7) (` 1) = 1
1
(7) (` 0 ` 1) = )
1
(7) 0 p
1
(7) 1
Since #1
1
(7) is even, we must have that #)
1
(7) and #p
1
(7) are both even or both odd,
hence 1
2
(), 7) = 1
2
(p, 7) mod 2.
We can dene the mod-2 intersection number of two submanifolds \, 7 ` as 1
2
(\, 7) :=
1
2
(i, 7) where i : \ ` is the inclusion. Although we can slightly perturb \ and 7 to create
more intersection points (or remove intersection points), the above proposition says that points
must be added in pairs.
UNFINISHED
6
2 Knot Theory
One of the reasons that low dimensional topology is so much harder to study than higher dimensions
is the appearance of knots. In high dimensions one can unravel all knots to become trivial, but
this does not happen in 3 dimensions. Fortunately we can get a combinatorial presentation of
3-manifolds from knots, but before that we have to cover some basics of knot theory.
2.1 Basic Denitions
Classically, a knot / is just a smoothly embedded copy of o
1
into o
3
. We could also embed our
knots into R
3
, the theories would turn out to be equivalent, but o
3
is easier to think about since it
is a nice compact 3-manifold. The complement of a knot in its surrounding manifold is called the
knot complement, and its rst homology group is quite simple.
Proposition 2.1. If / is a knot in o
3
, then H
1
(o
3
/) = Z and is generated by any meridian.
Proof. Let l be an open tubular neighborhood of / in o
3
, and let \ be the complement of the
closure of a smaller tubular neighborhood. Then l is homotopy equivalent to o
1
, \ is homotopy
equivalent to the knot complement o
3
/, and l \ is homotopy equivalent to a torus o
1
o
1
.
The Mayer-Vietoris sequences gives us the following exact sequence
0 H
1
(l \ ) H
1
(l) H
1
(\ ) 0
which is just
0 Z Z Z H
1
(o
3
/) 0
This implies that H
1
(o
3
/) = Z. Further, note that this map is of the form (i

, ,

), where i and ,
are the inclusions of l \ into l and \ , respectively. The group H
1
(l \ ) = ZZ is generated
by a meridian [j] and longitude [/], and we can choose this longitude such that ,

[/] = 0. Since [/]


is mapped to an element generating the rst factor of Z H
1
(o
3
/), we have that ,

[j] generates
the second factor.
Given two knots /
1
, /
2
in a manifold o
3
, we say that they are equivalent, or isotopic, if they
are ambiently isotopic. This means that there is a smooth 1-parameter family of maps
t
: o
3
o
3
such that
0
= id,
1
(/
1
) = /
2
, and
t
is a dieomorphism for all time t. A knot that is isotopic
to the standardly embedded o
1
in o
3
is called the unknot or the trivial knot.
There is a nice way to visualize a knot / in o
3
. First, move the knot a little if needed so that
it does not pass through the point at innity. Now / is embedded in R
3
. For a generic plane
in R
3
not intersecting /, the projection 1 of / onto is a closed curve with nitely many double
points, and no other types of singularities. We can decorate these double points as an over or
under crossing depending on whether one strand is above another relative to the plane. Such a
projection 1 is called a knot diagram of /. Conversely, given a closed curve 1 in the plane
which is embedded on the complement of nitely many double points and whose double points are
decorated with under or over crossings, we can nd a knot / such that 1 is a projection of /.
Clearly a single knot / can admit many dierent knot diagrams, and so we hope there is a way
to determine when two diagrams correspond to the same isotopy class of knot. It turns out there
are 3 moves one can apply to a knot diagram, called the Reidemeister moves, such that two
knot diagrams 1
1
and 1
2
are the diagrams of two isotopic knots if and only if one can get from
one diagram to the other by the Reidemeister moves (see gure 2.1). We will use 1
1
, 1
2
and 1
3
7
R
1
R
1
R
2
R
2
R
3
Figure 2.1: Reidemeister Moves
+1 1
Figure 2.2: Crossing Signs
to denote these moves. This reduces a lot of knot theory to combinatorial pictures, from which we
can dene a lot of knot invariants.
A slight generalization of a knot is a link. An n-component link is a smooth embedding of n
copies of o
1
. All of the above denitions and discussion has a direct generalization to links. Clearly
if two links 1
1
and 1
2
are isotopic, then their number of components must be equal.
One can put an orientation on a knot / by simply choosing a non-vanishing tangent vector eld
on /, or equivalently placing an arrow on any knot diagram of /. A knot has precisely 2 orientations
and an n-component link has 2
n
orientations. Given a knot diagram 1 of an oriented knot /, we
can dene the sign of each crossing of 1. We adopt the right hand rule for sign conventions, which
means signs are determined as in gure 2.2. Note that the sign of a crossing is information we get
once we have chosen a knot diagram, not from just a knot. The sum of all the signs of the crossing
of a knot diagram 1 is called the writhe of the diagram, and is denoted by n(1). The writhe of a
knot diagram is not a knot invariant because it is not invariant under 1
1
. It is, however, a ribbon
invariant, but we will not discuss that here.
Let / be a knot in o
3
and 1 a diagram of /. A resolution of a crossing, or a smoothing, is
the process of removing the double point and reconnecting the strands so that there is no crossing.
There are two ways of doing this, called the 0-resolution and 1-resolution, shown in gure 2.3. Each
8
0-resolution
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?

















crossing
1-resolution
Figure 2.3: Ways to resolve a crossing
time one resolves a crossing the number of components of the diagram either stays the same or
increases by one. A resolution of a diagram is the diagram resulting from resolving all crossings,
which is just a nite collection of disjoint circles. Clearly if there are n crossings in the diagram 1,
then there are 2
n
resolutions, which are parameterized by 0, 1
n
. If the knot / has an orientation,
then there is a canonical resolution of 1, called the oriented resolution, where we resolve all the
crossings of sign +1 by a 1-resolution and all crossings of sign 1 by a 0-resolution.
A Seifert surface for a link 1 in o
3
is a smoothly embedded, compact, connected and orientable
surface 1 in o
3
with boundary such that 1 = 1.
Proposition 2.2 (Seiferts Algorithm). Every link 1 in o
3
has a Seifert surface.
Proof. We can construct the surface explicitly. Choose any orientation on 1 and let 1 be a diagram
for 1. Let 1

denote the the oriented resolution of 1. Fill each component of 1

so that there is a
disc for each circle, and connect the discs a twisting band for each crossing. The result is a surface
with boundary equal to exactly 1.
Any link has many Seifert surfaces, and Seiferts algorithm almost never produces the surface
of minimal genus. We dene the genus of a link 1 to be the minimum of genera of Seifert surfaces
of 1, and is denoted by p(1). The four-ball genus of 1 is the minimum of genera of compact,
connected, oriented and smoothly embedded surfaces in 1
4
whose boundary is 1 1
4
= o
3
, and
is denoted by p

(1).
Lemma 2.1. Let /
1
, /
2
be two knots embedded in o
3
. Then the genus satises p(/
1
#/
2
) = p(/
1
) +
p(/
2
).
Proof. We can think of /
1
and /
2
embedded in the same sphere, but situated far apart. Let 1
1
and
1
2
be Seifert surfaces of minimal genus for /
1
and /
2
, respectively. Then the boundary sum 1
1
;1
2
is a Seifert surface of /
1
#/
2
, and so we have proved
p(/
1
#/
2
) p(/
1
) +p(/
2
)
To prove the other inequality we x a Seifert surface 1 of minimal genus for /
1
#/
2
. We can x
a 2-sphere in o
3
so that /
1
#/
2
consists of two points. Then, if the two arcs of /
1
#/
2
between
these two points are labelled
1
and
2
, and we take a small arc in between these two points,
then the curve
1
+ is /
1
and the curve
2
+ is /
2
. By perturbing slightly we can guarantee
that 1 is just a collection of closed curves, as well as the curve . We will perform surgery on
these closed curves one by one to remove these intersections, so that the only intersection points
left is the arc .
9
Let C be one of these closed curves with the additional property that the disc it bounds in
does not contain any of the other closed curves. These ensures that if we push the interior of this
disc a little o of (in either direction) it does not intersect 1. Let

1 be the surface obtained from
1 by cutting out an open annulus around C (so now there are two more boundary components)
and gluing in two discs, each parallel to the disc in bounded by C, and on either side of . The
result is another Seifert surface

1 of /
1
#/
2
. If the curve C does not separate 1, then

1 will have
genus one less that 1 (we just cut open a handle!), which is not possible since 1 is assumed to
have minimal genus. So, we must have that C separates 1, hence

1 consists of a disjoint union of
a 2-sphere and a Seifert surface of /
1
#/
2
that has one less intersection with . Repeating this for
all the closed curves in 1 shows that we can assume that and 1 intersect only at the arc .
Now we can take 1
1
to be the closure of the part of 1 outside of the sphere , and 1
2
the closure
of the part of 1 inside the sphere. Then 1
1
and 1
2
are Seifert surfaces of /
1
and /
2
respectively,
hence
p(/
1
) +p(/
2
) p(/
1
#/
2
)
Given disjoint, oriented knots /
1
and /
2
, let 1 be a common projection. We can assign signs to
the double points of 1 just as before, except now the two strands might belong to dierent knots.
The linking number lk(/
1
, /
2
) of /
1
and /
2
is dened to be the sum of the signs of all crossings
where /
1
passes under /
2
. In order to show that this denition is well-dened we need to see that
it does not depend on the diagram chosen for /
1
/
2
. It is an easy matter to show that lk(/
1
, /
2
)
is invariant under the Reidemeister moves, hence lk(/
1
, /
2
) is an invariant of the isotopy class of
/
1
/
2
. It is also easy to check that lk(/
1
, /
2
) = lk(/
2
, /
1
) and lk(/
1
, /
2
) = lk(/
1
, /
2
), where
/
1
is the knot /
1
with opposite orientation.
There are two other ways of describing the linking number that will be useful later. Let 1 =
/
1
/
2
be the link with components our knots, and let 1 be a Seifert surface for 1. At each point
on the boundary /
1
in 1 we have a tangent vector that determines the orientation of /
1
and
a vector pointing into the surface. Let n be a vector normal to 1 such that the ordered basis
(, , n) matches with the orientation of o
3
. The normal vector n can be extended to the rest of
the surface 1, and it determines an orientation. We can perturb /
2
slightly if needed so that /
2
intersects 1 transversely in nitely many points. If we assign these points the sign +1 when /
2
s
tangent vector points in the same direct as n and 1 when it points in the opposite direction, then
lk(/
1
, /
2
) is equal to the sum of these numbers.
The second way is to notice that the rst homology of the knot complement H
1
(`/
1
) = Z
is generated by the homology class [j] of any meridian j of /
1
. The other knot /
2
determines a
homology class [/
2
] H
1
(`/
1
), which must be an integer multiple of [j]. The linking number
lk(/
1
, /
2
) is precisely this number, i.e. [/
2
] = lk(/
1
, /
2
) [j].
A framing for a knot / in o
3
is a choice of a normal vector eld along /. If A is a framing of /,
then we can push / a little bit in the direction of A to get a new knot /

, and the integer lk(/, /

)
is called the framing number or framing coecient of A. Conversely, given an integer n we
can nd a parallel /

of / such that lk(/, /

) = n, and this curve is unique up to isotopy in o


3
/.
So, we can think of framings as either normal vector elds, parallel knots, or just integers assigned
to knots. A knot with a framing is called a framed knot, and a framed link is dened similarly. For
an oriented, framed link 1 with components 1
1
, . . . , 1
n
we dene its linking matrix [1] to be the
symmetric matrix with (i, ,)-entry equal to lk(1
i
, 1
j
), where we use the convention that lk(1
i
, 1
i
)
is just the framing of the i-th component of 1.
10
Figure 2.4: Examples of the canonical longitude and a non-canonical longitude
Given a knot / in o
3
the meridian j of (/) chosen above is called the canonical meridian,
and is always dened. However, a longitude / of (/) is not so clear how to dene since it can
wrap around / in complicated ways. The ways in which / can wrap around / are parameterized
by linking numbers, so we dene the canonical longitude to be the curve / in (/) such that
lk(/, /) = 0. This curve is unique up to isotopy, and will be useful in our discussion on knot
surgeries. Given a diagram 1 of a knot / one can draw the canonical longitude as a parallel of /
that wraps n(1) full times around /, where n(1) is the writhe of the diagram. Take note that
we are using the linking number to dene the canonical longitude even though our knot need not
be oriented. However, we can given / any orientation and point / in the same direction, and then
the linking number will not depend on the orientation we gave /.
In gure 2.4 we have three knots drawn (the bolded knots) with longitudes (not bolded). The
rst two longitudes are canonical, as one can easily calculate their linking numbers to be 0. However
the third longitude, which seems like a more obvious choice for a longitude than the second, is not
canonical since its linking number is -3.
The fundamental group of the knot complement
1
(o
3
/) is called the knot group, and clearly
is an invariant of the knot. There is a simple algorithm for computing a presentation of
1
(o
3
/)
given a diagram of 1. For a diagram 1 of /, let n+1 be the number of crossings. Let
0
, . . . ,
n
be
the arcs in 1 that start at a crossing, travel along the under-strand, and end at the next crossing
that the strand goes under such that
i
starts at where
i1
ends (we always take the subscripts of
the s to be mod n). Also orient these arcs such that
i
points towards the starting point of
i+1
.
For convenience, suppose that / does not pass through (0, 0, 1) (the point at innity in R
3
), and let
us use this as the base point in o
3
/. In terms of the knot diagram, we can think of the base point
as sitting right where we (the viewers) are, and so loops based at leave from us, travel through
the knot diagram, and then return to us. Hence, a loop in the knot diagram can be represented
by just a curve segment (not a loop), and we just assume the ends are connected to the point at
innity in the obvious way.
Generators of
1
(o
3
/) are just short arc segments r
0
, . . . , r
n
in the diagram such that r
i
passes
under
i
, and oriented so that it crosses under
i
from the right to the left. Each crossing of the
knot diagram gives a relation :
i
on the generators of the fundamental group. Suppose the crossing is
where the arcs
i
,
j
and
j+1
come together for some indices i and , (hence
i
is the over-strand).
If the sign of the crossing is +1, then we add the relation r
i
r
j
= r
j+1
r
i
, and if the sign is 1, then
we add the relation r
j
r
i
= r
i
r
j+1
. Note that we are computing the sign of these crossings with the
orientation on / induced by the orientations on the arcs, which were chosen for convenience. If
we changed those orientations we would get an equivalent presentation of the fundamental group.
Let :
i
denote the relation obtained at the crossing where
i
and
i+1
meet.
11
A
B
C
a
b
c
Figure 2.5: Computing the knot group of the trefoil
Proposition 2.3. The fundamental group of o
3
/ is isomorphic to r
0
, . . . , r
n
: :
0
, . . . , :
n
).
Proof. First we isotope / so that it lies in . = 0 R
3
, except at a crossing where the understand
dips to the . = c plane for a short segment, which we call
i
for the crossing corresponding to
where
i
and
i+1
meet. We will decompose o
3
/ into a bunch of pieces and apply van Kampens
theorem. Let = . c /, and let 1
i
(i = 0, . . . , n) be a small open rectangular box around
the segment
i
, but with
i
remove and with a thin tube connected the box to . Finally, let C
be the complement of the closure of 1
0
1
n
.
First we claim that
1
() is a free group generated by r
0
, . . . , r
n
. To see this let
i
be the arc

i
whose endpoints dive straight down to connect to the plane . = c, and let / be the union of
these arcs. Then
1
(o
3
/) is isomorphic to
1
(o
3
/), so we compute the later group. If we take
a curtain below the arc
i
to the plane . = c and then thicken it we get a ball 1
i
. Removing
the 1
i
s from . c yields a simply connected space, so we determine what happens to the
fundamental group when we add the pieces 1
i

i
back. The piece 1
i

i
is just a punctured disc,
so its fundamental group is generated by a loop going around this hole once. Also the piece 1
i

i
intersects the space where we removed all the 1
j
s from . c is just a disc, hence we added
one generator by adding the piece 1
i

i
. Continuing this we get the claim.
UNFINISHED
This presentation of the knot group is called the Wirtinger presentation.
Example 2.1. Consider the diagram of the trefoil knot / shown in gure 2.5 (we label our arcs
by , 1, C and generators of
1
by o, /, c). The Wirtinger representation of the knot group is

1
(o
3
/) = o, /, c : /o = oc, oc = c/, c/ = /o)
We can eliminate a generator from this presentation by seeing that c = /o/
1
, hence another
presentation of the fundamental group is given by

1
(o
3
/) = o, / : /o/ = o/o)
If we set r = o/o and j = /o, then we see that r
2
= o/oo/o = /o/o/o = j
3
, hence yet another
presentation is given by

1
(o
3
/) =

r, j : r
2
= j
3
_
12
2.2 The Alexander Polynomial
We will now construct some classical invariants of knots and links in o
3
. With some work these
invariants can be extended to links in integral homology spheres, as we will see later. These
combinatorial invariants have very interesting relationships with high-powered invariants, like
Seiberg-Witten invariants and Heegaard Floer homology.
Fix a diagram 1 of a knot 1 in o
3
with n components, and give the link any orientation.
Suppose this diagram has / double points and the oriented resolution of this diagram has c connected
components. Let us compute the Euler characteristic and genus of the Seifert surface 1 constructed
from Seiferts algorithm on 1 (see proposition 2.2). The Euler characteristic of c disjoint discs
is c, and attaching a band to these discs decreases the Euler characteristic by one. Therefore
(1) = c /. On the other hand, an orientable surface of genus p with n boundary components
has Euler characteristic 2 2p n. Setting these two Euler characteristics equal to each other and
solving for p shows that the genus of 1 is
p = 1 +
/ c n
2
(2.1)
One can compute that the rst homology of a surface of genus p with n 1 boundary components
is the free abelian group of rank 2p + n 1 (just look at the handlebody decomposition of such a
surface), therefore, by (2.1), H
1
(1) is free abelian with rank 1 +/ c.
Now, for a general Seifert surface 1 of a link 1 in o
3
, let = (r
1
, . . . , r

) be an ordered
collection of non-intersecting, oriented, closed curves in 1 that form an ordered basis for H
1
(1).
Choose an orientation of 1 by specifying a nowhere zero, normal vector eld n on 1. For each
curve r
i
, let r
+
i
be the curve in o
3
that is r
i
pushed o the surface 1 slightly in the direction of n.
For each pair of curves r
i
, r
+
j
we can compute their linking numbers (they do not intersect), and
so we dene the Seifert matrix of 1 with orientation n to be the matrix [1] = (lk(r
i
, r
+
j
))

i,j=1
.
Of course this matrix depends on our choice of orientation of 1 and basis for H
1
(1), so to include
this dependence in the notation we may sometimes write [1]
,n
. Suppose we have xed our basis
r
1
, . . . , r

with pushed o curves r


+
1
, . . . , r
+

relative to n. If we change the orientation of 1 to


n, then we can translate 1 a little bit so that r
+
1
, . . . , r
+

lie in 1, in which case the curves


r
1
, . . . , r

are now the pushed o curves relative to n. So, the Seifert matrix with respect to
this orientation is (lk(r
+
i
, r
j
)). This is equal to (lk(r
j
, r
+
i
)) by the symmetry of linking numbers,
which is the transpose of the Seifert matrix computed with respect to n. So, reversing orientations
transposed the Seifert matrix. If we x the orientation, then any two bases of H
1
(1) are related by
an invertible matrix l (invertible over Z means det l = 1), so if and are two bases related
by l, then [1]

= l
T
[1]

l.
However, even if we consider two Seifert matrices and 1 equivalent if they are related by
= l
T
1l, for some invertible l, we do not get a knot invariant. We could just analyze how
Seifert matrices transform under the Reidemeister moves, but a nicer way is to nd a set of moves
that can be performed on Seifert surfaces such that any two Seifert surfaces of a knot are related
by these moves. One such move is called stabilization and it increases the genus of a Seifert surface
by one. Let 1 be a Seifert surface of / embedded in o
3
and T

a torus with a small disc cut out


(so its boundary is o
1
) also embedded in o
3
such that 1 T

= . Taking the boundary sum of


these two surfaces 1;T

gives another Seifert surface of /, and is called the stabilization of 1.


We say two Seifert surfaces are stably isotopic if they become isotopic after a nite number of
stabilizations.
Proposition 2.4. Any two Seifert surfaces of a xed knot are stably isotopic.
13
We have already remarked that a change of basis of H
1
(1) corresponds to changing the Seifert
matrix of 1 by l
T
ol, for some integral matrix with det l = 1. Let us see how stabilization
of a Seifert surface changes the Seifert matrix. Let 1 be a Seifert surface of / with generators
r
1
, . . . , r
2g
for H
1
(1). By forming 1

= 1;T

we are adding two generators to the rst homology,


one curve going around the longitude of T

and one curve going around the meridian, so let j and


. denote these curves. Since the torus T

can knot and link with 1 in complicated ways, we have


that column and row of [1

] corresponding to j can consist of arbitrary integers, except that either


lk(j, .
+
) = 1 and lk(., j
+
) = 0, or the opposite. Therefore the Seifert matrix [1

] is given by
_
_
_
_
_
_
_
0
[1]
.
.
.
.
.
.
0
1
0 0 0 0
_
_
_
_
_
_
_
or
_
_
_
_
_
_
_
0
[1]
.
.
.
.
.
.
0
0
0 0 1 0
_
_
_
_
_
_
_
This leads us to dene two operations we can perform on integral matrices. Given an integral,
n n matrix (not necessarily invertible), performing an o
1
move is to change the matrix to
l
T
l for some invertible integral matrix l. Performing an o
2
move is to change the matrix to
the (n + 2) (n + 2) matrix given by
_
_
_
_
_
_
_
0
o
.
.
.
.
.
.
0
1
0 0 0 0
_
_
_
_
_
_
_
(2.2)
where the s can be any numbers. We say that two integral matrices are o-related one can be
brought to the other with a nite sequence of o
1
and o
2
moves. An immediate consequence of
proposition 2.4 is
Corollary 2.1. Any two Seifert matrices of a knot are o-equivalent.
Therefore the o-equivalence class of any Seifert matrix of a knot is an invariant of the knot.
There are simpler or more computable invariants we can derive from the Seifert matrix. Given a
knot /, let o be any Seifert matrix for /. The Laurent polynomial

k
(t) = det
_
t
1/2
o t
1/2
o
T
_
in t
1/2
is called the Alexander polynomial of /. Technically we should call this the Conway
normalization of Alexanders polynomial, as Alexanders original denition used the innite cyclic
covering of the knot complement, and was dened only up to multiplication by a unit. Since the
dimension of o is 2p (where p is the genus of the surface associated to o) we have
k
(t) is actually
a Laurent polynomial in t, and a direct substitution shows
k
(t) =
k
(t
1
). If o is of genus p,
then we can easily see that
k
(t) = t
g
det(o to
T
).
Proposition 2.5. The Alexander polynomial does not depend on the Seifert matrix chosen, hence
is a knot invariant.
14
Proof. If o

is another Seifert matrix for /, then o and o

are related by a nite sequence of o


1
and o
2
moves by corollary 2.1. Let
k
(t) denote the Alexander polynomial with respect to o, and

k
(t) the Alexander polynomial with respect to o

. If o

= l
T
ol, then we see that

k
(t) = det
_
t
1/2
l
T
ol t
1/2
l
T
ol
_
= (det l)
2
det
_
t
1/2
o t
1/2
o
T
_
=
k
(t)
where the last line follows since det l = 1. On the other hand, if o

is the result of performing


the o
2
move on o (see (2.2)), then o
T
can be written as
_
_
_
_
_
_
_
0
o
T
.
.
.
.
.
.
0
0
0 0 1 0
_
_
_
_
_
_
_
and the matrix t
1/2
o

t
1/2
o
T
can be written as
_
_
_
_
_
_
_
0
t
1/2
o t
1/2
o
T
.
.
.
.
.
.
0
t
1/2
0 0 t
1/2
0
_
_
_
_
_
_
_
When computing the determinant of this matrix we can perform a cofactor expansion along the
last column, and then another cofactor expansion along the last row of the resulting matrix, in
which case we see that the determinant of the above is precisely
t
1/2
_
t
1/2
det
_
t
1/2
o t
1/2
o
T
__
=
k
(t)
hence the Alexander polynomial is invariant under the o
2
moves.
We can extend the denition of Alexander polynomial to links in the obvious way, but it will
no longer be independent of the Seifert surface. The problem is that a link 1 with n components
has 2
n
orientations. A Seifert surface 1 for 1 can induce two of these orientations on 1 (and they
are opposite), and conversely we say that a Seifert surface is compatible with an oriented link if the
induced orientation agrees with the links orientation. While it is true that any two Seifert surfaces
compatible with an oriented link are stably equivalent, it is not true that any two Seifert surfaces
of a link are stably equivalent. So, the Alexander polynomial is an invariant of oriented links with
compatible Seifert matrices.
As mentioned above, the Alexander polynomial is not sensitive to the orientation of a knot, i.e.

k
(t) =
k
(t). Unfortunately, the Alexander polynomial is also not sensitive to changing a knot
to its mirror. The mirror of a knot / is the knot / obtained by reecting / through any plane
in R
3
. This is equivalent to changing all over crossings to under crossings (and vice versa) in a
diagram of /.
Proposition 2.6. For any knot /,
k
(t) =
k
(t).
15
Proof. If o is a Seifert matrix of /, then o is a Seifert matrix of /. Let 2: be the order of o.
Then

k
(t) = det
_
t
1/2
o +t
1/2
o
T
_
= det
_
t
1/2
o
T
t
1/2
o
_
= det
_
t
1/2
o t
1/2
o
T
_
=
k
(t
1
) =
k
(t)
There are also scalar invariants of knots that one can derive from Seifert matrices.
Proposition 2.7. If o is the Seifert matrix of some Seifert surface of a knot /, then [ det(o +o
T
)[
is a knot invariant.
Proof. We just have to check that [ det(o + o
T
)[ does not change after applying o
1
and o
2
moves
to o. If we change o by l
T
ol, then clearly
[ det(l
T
ol +l
T
o
T
l)[ = [(det l)
2
det(o +o
T
)[ = [ det(o +o
T
)[
Let o

be the matrix resulting from applying o


1
to o (see (2.2)). Then o

+o
T
is of the form
_
_
_
_
_
_
_
0
o +o
T
.
.
.
.
.
.
0
1
0 0 1 0
_
_
_
_
_
_
_
By adding multiples of the last column to the other columns, and multiples of the last row to the
other rows we can bring this matrix into the form
_
_
_
_
_
_
_
0 0
o +o
T
.
.
.
.
.
.
0 0
0 0 0 1
0 0 1 0
_
_
_
_
_
_
_
(2.3)
and the determinant of this matrix is det(o +o
T
). Therefore the absolute value of this determi-
nant remains invariant under o
1
and o
2
moves.
We call [ det(o+o
T
)[ the determinant of the knot /, and denote it by det(/). The relationship
between the determinant of / and the Alexander polynomial is clear.
Proposition 2.8. [
k
(1)[ = det(/)
Proof. Let o be any Seifert matrix of / and let 2: be the order of o (the order must be even for
a knot). Then
[
k
(1)[ =

det
_
(1)
1/2
o (1)
1/2
o
T
_

det
_
(1)
1/2
o + (1)
1/2
o
T
_

(1)
m
det(o +o
T
)

16
The Alexander polynomial satises a skein relation, which is closely related to another polyno-
mial invariant that is essentially the same as the Alexander polynomial. For an oriented link 1 and
a diagram 1 of 1, choose any crossing in this diagram and let 1
+
denote the diagram resulting
from switching this crossing to a positive crossing, 1

the diagram resulting from switching to a


negative crossing, and 1
0
the diagram resulting from resolving the crossing in the oriented way (see
??).
Theorem 2.1. The Alexander polynomial of an oriented link 1 satises the skein relation

L
+
(t)
L

(t) = (t
1/2
t
1/2
)
L
0
(t)
Further, this relation, along with the normalization
unknot
(t) = 1, uniquely determines the Alexan-
der polynomial as a function from oriented links to Z[t
1/2
, t
1/2
].
Proof. Let 1
0
be the Seifert surface for 1
0
constructed from Seiferts algorithm (see proposition 2.2).
We can construct Seifert surfaces 1

for 1

by gluing in a twisted band. Take a basis r


1
, . . . , r
k1

for H
1
(1
0
), then we can get a basis for H
1
(1

) by adding one extra curve r


k
that goes through the
twisted band once. Let o
0
be the Seifert matrix of 1
0
computed with respect to r
1
, . . . , r
k1
.
The Seifert matrices o

of 1

are of the form


o
+
=
_
_
_
_
_
/
1
o
0
.
.
.
/
k1
o
1
o
k1

_
_
_
_
_
o

=
_
_
_
_
_
/
1
o
0
.
.
.
/
k1
o
1
o
k1
1
_
_
_
_
_
for some integers o
1
, . . . , o
k
, /
1
, . . . , /
k
, . Writing out the entries of t
1/2
o

t
1/2
o
T

we see that
the last column and row of these matrices are equal, except for the bottom-right corner entry
where one matrix has (t
1/2
t
1/2
) and the other has (t
1/2
t
1/2
)( 1). So, if we compute

L
+
(t)
L

(t) by performing a cofactor expansion along the last columns, then the terms from

L
+
(t) will cancel those from
L

(t) except for the last term, which is equal to


(1)
2k
(t
1/2
t
1/2
)
L
0
(t) (1)
2k
(t
1/2
t
1/2
)( 1)
L
0
(t) = (t
1/2
t
1/2
)
L
0
(t)
Therefore
L
(t) satises the skein relation.
Although the Alexander polynomial satises a skein relation, this relation does not quite look
like the relation for the Jones polynomial. To make things more uniform we introduce the Conway
polynomial
L
(.). This is a function : oriented links Z[.] that is uniquely determined by
the normalization
unknot
(.) = 1 and the skein relation

L
+
(.)
L

(.) = .
L
0
(.)
The relationship between the Alexander and Conway polynomial is given by
L
(t) =
L
_
t
1/2
t
1/2
_
.
2.3 Knot Signatures and 4-Ball Genus
A very important invariant of knots is the signature.
Proposition 2.9. Let be a unit complex number not equal to 1, and let o be a Seifert matrix of
the knot /. Then the signature of the Hermitian matrix (1)o+(1)o
T
over C is an invariant
of the knot.
17
Proof. We just have to check that the signature of (1 )o + (1 )o
T
does not change after
applying o
1
and o
2
moves to o. Clearly the signature is invariant under o
1
moves. Let o

be the
Seifert matrix resulting in applying an o
2
move, then we have (1 )o +(1 )o
T
is congruent to
_
_
_
_
_
_
_
0 0
(1 )o + (1 )o
T
.
.
.
.
.
.
0 0
0 0 0 1
0 0 1 0
_
_
_
_
_
_
_
The signature of this block matrix is the sum of the signatures of the blocks. Since the signature
of the bottom-right block is zero, we have that the signature of the above matrix is the signature
of (1 )o + (1 )o
T
, hence the signature is invariant under o
2
moves.
We call the signature of (1 )o + (1 )o
T
the -signature of the knot, and denote it by

(/). If = 1, then we just call this the signature of /, and denote it by (/).
For a knot / in o
3
, consider the 4-ball 1
4
bounded by o
3
. Let p

(/) denote the minimal genus


of a surface 1 that can be smooth embedded in 1
4
with 1 = / o
3
. This knot invariant is
called the smooth 4-ball genus, or sometimes the smooth slice genus. A knot is said to be
slice if p

(/) = 0. The relationship between signature and slice knots is the following.
Theorem 2.2. If / is a slice knot, then (/) = 0.
We will outline the proof of this theorem in a series of lemmas. Suppose / is a slice knot with
slicing disc 1 1
4
. If 1 is a Seifert surface for / in o
3
, then we can glue together 1 1 to obtain
a closed surface in 1
4
.
Lemma 2.2. The surface 1 1 bounds a smoothly embedded, orientable 3-manifold ` in 1
4
with
` o
3
= 1.
Lemma 2.3. Let ` be a compact, orientable 3-manifold such that ` is a surface of genus p.
Then the kernel of the map i

: H
1
(`; Q) H
1
(`; Q) induced by the inclusion is of dimension
p.
Proof. We can put the homology and cohomology long exact sequences of the pair (`, `) into
the following commutative diagram
H
2
(`, `; Q)
//

H
1
(`; Q)
i
//

H
1
(`; Q)

H
1
(`; Q)
i

//
H
1
(`; Q)

//
H
2
(`, `; Q)
where the vertical arrows are the Poincare and Poincare-Lefschetz duality isomorphisms. By the
universal coecient theorem we have that H
1
(`; Q) is dual to H
1
(`; Q), H
1
(`; Q) is dual to
H
1
(`; Q), and i

is dual to i

. By the rank-nullity theorem we have


2p = dimimi

+ dimker i

= dimimi

+ dimker = dimimi

+ dimimi

= 2 dimimi

hence dimimi

= p, and so dimker i

= p.
Lemma 2.4. There is a basis r
i
over Z for H
1
(`) such that i

(r
j
) = 0 for 1 , p.
18
Proof. We think of H
1
(`) = Z
2g
as sitting inside H
1
(`; Q) = Q
2g
. Let l be the p dimensional
subspace of Q
2g
which is the kernel of the map Q
2g
H
1
(`; Q) induced by the inclusion. We can
take a basis r
1
, . . . , r
g
for this vector space such that r
i
Z
2g
, and so let

l be the Z-span of
these vectors.
As a Z-module we can write Z
2g
,

l as ,

l 1,

l, where and 1 are submodules of Z


2g
such
that ,

l is free and 1,

l is torsion. The fact that this latter module is torsion means that for
/ 1 there is an integer n such that n/

l, hence n/ l, or simply / l. Thus a Z-basis for 1
also serves as a Q-basis for l. So, the Z-basis we take for H
1
(`) is any extension of a basis for
1 to a full basis for H
1
(`).
Lemma 2.5. Let 1 be a genus p Seifert surface for a slice knot /. Then there is a basis for H
1
(1)
such that the associated Seifert matrix is of the form
o =
_
0 1
Q 1
_
for some p p matrices 1, Q, 1.
Proof. Let 1 be a slicing disc for / in 1
4
so that there is a 3-manifold ` with ` = 1 1. Let
r
i
be a Z-basis for H
1
(`) as in the previous lemma so that r
i
= 0 in H
1
(`; Q) for 1 i p.
We can assume that the homology classes r
i
are represented by oriented, closed curves in `, and
we will also write r
i
to denote these curves. For each 1 i p we can nd a non-zero integer n
i
such that n
i
r
i
= 0 in H
1
(`). The class n
i
r
i
can be represented by a closed curve in `, which
we also denote by n
i
r
i
, and it bounds a 2-chain in `. This means we can nd maps )
i
: 1
i
1
4
,
where 1
i
is a surface with boundary, such that )
i
(1
i
) = n
i
r
i
. If we push the curve n
i
r
i
o of `
in the normal direction, then we can also pushing the mapping )
i
in this direction. Therefore, for
1 i, ,, p the curves n
i
)
i
and (n
j
)
j
)
+
bound disjoint surfaces in 1
4
, and hence
0 = lk(n
i
)
i
, (n
j
)
j
)
+
) = n
i
n
j
lk()
i
, )
+
j
)
Therefore the upper-left p p block of the Seifert matrix of 1 with respect to the basis r
i
is
zero.
Lemma 2.6. If Q is a non-degenerate quadratic form on an even dimensional space that vanishes
on a subspace of half dimension, then (Q) = 0.
Proof. Suppose the space\ that Q is dened on is of dimension 2p, and let l be a subspace of
dimension p such that Q(n) = 0 for all n l. If \

denote the maximal positive/negative-denite


subspaces of \ with respect to Q, then \ = \
+
\

since Q is non-degenerate. However, we


also have l \
+
= 0 and l \

= 0, hence dim\
+
p and dim\

p. Since we need
dim\
+
+ dim\

= 2p we are forced to have p = dim\


+
= dim\

, which implies (Q) = 0.


Proof of Theorem 2.2. Clearly the combination of the previous lemmas proves the theorem.
We say that two knots /
1
, /
2
in o
3
are concordant if there is a smooth embedding of a cylinder
o
1
1 in o
3
1 such that o
1
i maps to /
i
i (i = 0, 1). Concordance denes an equivalence
relation on the set of knots in o
3
. If a knot / is concordant to the unknot, then we can cap o the
unknot with a disc and obtain a smoothly embedded disc in 1
4
that is bounded by /, hence / is
slice. Therefore a knot is concordant to the unknot if and only if it is slice.
Let (
1
denote the concordance classes of knots in o
3
. If /
1
, /

1
and /
2
, /

2
are concordant pairs
of knots, then /
1
#/
2
and /

1
#/
2
are concordant. The cobordism can be constructed by splicing
19
together the cylinders connected /
1
to /

1
and /
2
to /

2
. Next we claim that for any knot /, the
knot /#/ is slice. Dene the following subsets of R
4
:
R
3
=
_
(r, j, ., 0) R
4
_
R
3

=
_
(r, j, ., 0) R
4
: . 0
_
R
3
+
=
_
(r, j, ., 0) R
4
: . 0
_
R
2
=
_
(r, j, 0, 0) R
4
_
Arrange the knot /#/ in R
3
so that its intersection with the plane R
2
in R
4
consists of two points
on the band used in the connect sum. Let /
+
denote the intersection of /#/ with R
3
+
, which is just
an arc, so that if /

is the reection of /
+
through R
2
, then /
+
/

is isotopic to /#/. Consider


the set of points
1 =
_
(r, j, . cos , . sin ) R
4
: (r, j, ., 0) /
+
, 0
_
That is, we spin all the points on /
+
about the plane R
2
in R
4
through all angles from 0 to .
Clearly 1 is just a disc (being a half spin of an arc) with boundary /#/, and this provides us with
a slicing disc for /#/.
We can now conclude that (
1
forms a group under connect sum with [unknot] serving as the
identity and the mirror of a knot serving as its inverse. Further, if /
1
and /
2
are concordant, then
0 = (/
1
#/
2
) = (/
1
)(/
2
), hence descends to a well-dened group homomorphism : (
1
Z.
2.4 Cyclic and Branched Covers
Now let us discuss Alexanders original denition of his polynomial using the innite cyclic cover of
the complement of a link. To construct this covering space we x an oriented link 1, set A = o
3
1,
and x a Seifert surface 1 (compatible with the orientation of 1). Let Y be the space obtained
from cutting A along 1 so that Y consists of two disjoint copies of 1, denoted by 1
+
and 1

;
that is, Y is homeomorphic to the complement of an open tubular neighborhood of 1 in o
3
. There
is a natural identication : 1

1
+
. Take the countable collection Y i (i Z) of copies of
Y , and let /
i
: Y Y
i
be the obvious homeomorphism. Form the space A

by gluing Y
i
to Y
i+1
via the homeomorphism /
i+1
/
1
i
; that is, we glue Y
i
to Y
i+1
by gluing 1

i to 1
+
i + 1
via the map /
i+1
/
1
i
. The obvious projection : A

A, where [
Y
i
: Y
i
A is given by
(r, i) = r, is an innite cyclic covering space of A.
Let t denote the covering transformation of A

whose restriction t[
Y
i
: Y
i
A

is given by
t(r, i) = (r, i + 1). This map induces an isomorphism (which we also denote by t) t : H
1
(A

)
H
1
(A

), turning H
1
(A

) into a Z[t, t
1
]-module, called the Alexander module of 1. Of course,
we need to show that this module is independent of the various choices in its construction
Proposition 2.10. If A

and A

are the innite cyclic coverings of the link complement A


of a link 1 constructed from two dierent Seifert surfaces, then A

and A

are isomorphic has


covering spaces, and H
1
(A

) and H
1
(A

) are isomorphic as Z[t, t


1
] modules.
Proof. Recall that covering spaces are determined up to equivalence by the image of the fundamental
group of the total space under the covering projection. That subgroup of the fundamental group
of the base space consists of those loops in the base that lift to loops in the total space. So, we
show that the notion of lifting loops to loops does not depend on the Seifert surface 1 used in the
construction of A

.
20
Let : [0, 1] A be any path in A and consider a lift : [0, 1] A

. This path is a loop in


A

if and only if (0) and (1) are both in A

, which means the path intersects the chambers


1

i with algebraic multiplicity zero. Down in A this means that that has zero algebraic
intersection number with 1, which we know is equivalent to and 1 having zero linking number
(which is the sum of the linking numbers with each component of 1). This statement is of course
independent of 1, hence our notion of path lifting is independent of 1, and so these covering spaces
are equivalent.
Next we have to check that any two coverings are equivalent via a t-equivariant homeomorphism
in order to ensure that the rst homology Z[t, t
1
]-modules are isomorphic. Suppose j

: A

A
is another covering space constructed with a dierent Seifert surface 1

. Then there is a homeomor-


phism / : A

such that j

/ = j. Let be a path in A

from a point o to the translate


t o. In the base, the path j is a loop with linking number 1 with 1. Then j

/ = j ,
so / is a path in A

from a point o

to its translate t o

, hence t /(o) = /(t o) and so / is


t-equivariant.
Since the covering space A

A is well-dened given a link 1 we have that the Z[t, t


1
]-
module H
1
(A

) is an invariant of the link, called the Alexander module. Let us connect this
invariant with the Alexander polynomial dened earlier. Given an 1-module `, a presentation of
` is an exact sequence
1

1

` 0
where 1 and 1 are free 1-modules. If we x bases for 1 and 1 then we get a matrix representation
of , called the presentation matrix of `. The :-th elementary ideal of ` is the ideal of 1 generated
by all (: : + 1) (: : + 1) minors of a presentation matrix of `. In particular, the rst
elementary ideal of ` is simply the determinant of any presentation matrix of `. The relationship
between the Alexander polynomial and Alexander module can be stated as the following.
Theorem 2.3. If o is a Seifert matrix for a knot / in o
3
, then to o
T
is a presentation matrix
for H
1
(A

) as a Z[t, t
1
]-module.
Proof. In the notation we used when constructing A

, let us write A

= l \ , where l =
i
Y
2i
and \ =
i
Y
2i+1
. We have H
0
(l \ ) =
i
H
0
(1
i
), and the action of t on this group shifts the
summands of Z by one, so we can identify H
0
(l \ ) with Z[t, t
1
] H
0
(1) generated by 1 1.
Similarly, H
0
(l) H
0
(\ ) =
i
H
0
(Y
i
) and so we can identify this with Z[t, t
1
] H
0
(Y ) generated
by 1 1. With these identications we see that the map H
0
(l \ ) H
0
(l) H
0
(\ ) in the
Mayer-Vietoris sequence is given by 1 1 1 1 +t 1, and so is injective. Therefore the map
H
1
(l) H
1
(\ ) H
(
A

) is the Mayer-Vietoris sequence is surjective.


Let is now x a basis r
i
for H
1
(1) and a dual basis j
i
with respect to the bilinear form :
H
1
(o
3
1)H
1
(1) Z. Let o = (:
ij
) be the Seifert matrix of 1 with respect to this basis. Similar
to the above arguments we have an identication of H
1
(l \ ) with Z[t, t
1
] H
1
(1), the latter
being generated by 1 r
i
, and an identication of H
1
(l) H
1
(\ ) with Z[t, t
1
] H
1
(Y ), the
latter being generated by 1 j
i
. With these identications, the map H
1
(l\ ) H
1
(l)H
1
(\ )
in the Mayer-Vietoris sequence is given by 1 r
i
= 1 r

i
+ t r
+
i
. However, as elements in
H
1
(o
3
1), the cycles r
+
i
and r

i
can be written as
r
+
i
=

j
:
ji
j
j
r

i
=

j
:
ij
j
j
21
Therefore
1 r

i
+t r
+
i
=

j
(1 :
ij
j
j
+t :
ji
j
j
)
With, with respect to our bases we have that the matrix of the map in the Mayer-Vietoris sequence
is given by to o
T
.
We can change the construction of the innite cyclic cover of the link exterior slightly to obtain
a nite cyclic cover. For an integer / 0, take / copies of Y (the link exterior cut open along a
Seifert surface) by setting Y
i
= Y i (i = 0, 1, . . . , / 1). For the space A
k
by gluing Y
i
to Y
i+1
via the homeomorphism /
i+1
/
1
i
(where all induces are taken modulo /). Then the obvious
projection
k
: A
k
A is a /-fold cyclic covering of the link exterior. Note that we have an innite
cyclic covering A

A
k
that maps the points in Y
i
A

identically onto Y
i mod k
. In fact, the
space A
k
is simply the quotient A

t
k
_
.
This collection of covering spaces depends only on the link 1 and not on the various choices
used to dene it.
This collection of covering spaces can now be extending to cyclic branched covers

A
k
o
3
of the 3-sphere, branched over the link 1. A loop in A lifts to a loop in A
k
if and only if it has
linking number 0 (modulo /) with 1. For a component 1
i
from the link 1, let 1
i
be a tubular
neighborhood. We can identify 1
i
with o
1
o
1
by choosing a meridian j
i
and longitude /
i
of
1
i
such that /
i
links zero times with 1
i
. Then j
k
i
(the /-th power of j
i
) and /
i
lift to a loops in
A
k
which identify a torus in A
k
. The restriction of the covering map to this torus gives a /-fold
covering of one torus over another which is equivalent to (., n) (., n
k
). We can extend this to
a branched covering o
1
1
2
o
1
1
2
given by (., n) = (., n
k
), where this is branched over the
core circle o
1
0. So, if we glue solid tori into the boundary components of A
k
we obtain a space

A
k
with a natural projection

A
k
o
3
that is a /-fold covering branched over 1.
2.5 Applications
Seifert matrices are closely related to the intersection form on a compact, orientable surface. In-
tersection forms are dened on any even dimensional manifold, and we give a fuller treatment for
4-manifolds in section 6.1, so we will just briey state the facts. For a closed, oriented surface `,
Poincare duality gives us a non-degenerate, antisymmetric, bilinear form Q
M
: H
1
(`) H
1
(`)
Z dened by Q
M
(, ) = ( , [`]). It follows that the matrix representation of Q
M
in any
basis of H
1
(`) has determinant 1. However, we have the following lemma concerning these types
of matrices.
Lemma 2.7. If is the matrix representation of a non-degenerate, antisymmetric bilinear form,
then there is a matrix l such that = l
T
Jl, where J is a block diagonal matrix of copies of
_
0 1
1 0
_
Therefore the matrix of Q
M
with respect to any basis must have determinant +1. There is
a more geometric denition of the intersection form on `. A homology class o H
1
(`) can
be represented by a closed, smoothly embedded 1-dimensional submanifold of `, call it C
a
. If
o, / H
1
(`) with representative submanifolds C
a
and C
b
, then perturb them a little so that they
intersect transversely, and let o / be the signed sum of their points of intersection, called the
intersection product. Then one can show that Q
M
(, ) = 11() 11().
22
If ` has one boundary component (i.e. ` is a Seifert surface of some knot), then Q
M
is still a
non-degenerate, antisymmetric, bilinear form, and it can still be expressed in terms of intersections
of homology classes. If ` has more than one boundary component, then Q
M
is an antisymmetric,
bilinear form, and can be dened in terms of intersections, but it is now necessarily degenerate,
and so any matrix representation has zero determinant.
Proposition 2.11. Let 1 be a Seifert surface of a knot / with Seifert matrix o in some basis of
H
1
(1). Then the intersection form of 1 in this basis is precisely o o
T
.
Proof. At a positive crossing of r
i
and r
j
we see that r
+
j
is above r
i
. So we have lk(r
i
, r
+
j
)
contributes +1 at this crossing and lk(r
j
, r
+
i
) does not contribute anything (since we only count
crossings where the rst strand passes under). On the other hand, at a negative crossing of r
i
and r
j
we have that r
+
j
passes below r
i
. Hence lk(r
i
, r
+
j
) contributes nothing at this crossing and
lk(r
j
, r
+
i
) contributes +1. Therefore r
i
r
j
= lk(r
i
, r
+
j
) lk(r
j
, r
+
i
).
Corollary 2.2. If / is a knot, then
k
(1) = 1, and if 1 is a link, then
L
(1) = 0.
Proof. Let 1
k
and 1
L
be Seifert surfaces for / and 1. By proposition 2.11 we have
k
(1) =
det Q
M
= 1 and
L
(1) = det Q
L
= 0.
The Alexander polynomial of the connected sum of two knots has a very simple expression.
Proposition 2.12. If /
1
and /
2
are knots, then
k
1
#k
2
(t) =
k
1
(t)
k
2
(t).
Proof. We can assume that /
1
and /
2
are situated in o
3
far apart, and that 1
1
and 1
2
are Seifert
surfaces for /
1
and /
2
respectively that do not intersect. Then by taking the boundary sum 1
1
;1
2
we get a Seifert surface for /
1
#/
2
. If o
1
and o
2
are the Seifert matrices for 1
1
and 1
2
respectively,
then the Seifert matrix of 1
1
;1
2
is
_
o
1
0
0 o
2
_
We clearly now have

k
1
#k
2
(t) = det(t
1/2
o
1
t
1/2
o
T
1
) det(t
1/2
o
2
t
1/2
o
T
2
) =
k
1
(t)
k
2
(t)
The Alexander polynomial has a clear relationship with the genus of a knot.
Proposition 2.13. For a knot / we have deg
k
(t) p(/).
Proof. If o is a Seifert matrix of order 2: for /, then deg
k
(t) :, hence deg
k
(t) p(/).
Proposition 2.14. If det o ,= 0 for some Seifert matrix o of /, then deg
k
(t) = p(/).
Proof. If o is a Seifert matrix associated to a Seifert surface of genus p, then we can write
k
(t) =
t
g
det(oto
T
). The constant term of the polynomial det(oto
T
) is just det o, which is non-zero
by assumption. Therefore the coecient of t
g
in
k
(t) is non-zero, and so deg
k
(t) = p.
A link 1 is said to be a split link if one can nd an embedded 2-sphere in o
3
that separates
some components of 1 from the other components of 1.
Proposition 2.15. If 1 is an oriented, split link 1, then
L
(t) = 0.
23
Proof. Suppose 1 = /
1
/
2
such that /
1
can be separated from /
2
by an embedded 2-sphere. Let
1
1
and 1
2
be Seifert surfaces for /
1
and /
2
respectively, and let p
1
and p
2
be the genera of 1
1
and
1
2
respectively. We can constructed a Seifert surface for 1 by taking the connected sum 1
1
#1
2
;
that is, cut small, open discs out of 1
1
and 1
2
and glue in a tube o
1
1. Recall that a surface of
genus p with n boundary components has rst homology of rank 2p +n 1, hence H
1
(1
1
)

= Z
2g
1
and H
2
(1
2
)

= Z
2g
2
. However, H
1
(1
1
#1
2
)

= Z
2g
1
+2g
2
+1
, so this homology group has one extra
generator in addition to the generators coming from 1
1
and 1
2
. This generator can be represented
by the meridian j of the tube o
1
1. This generator does not link with the push os of any of the
generators from 1
1
and 1
2
, hence there is an entire row and column of zeros in the Seifert matrix
of 1
1
#1
2
, and so
L
(t) = 0.
2.6 Knots in other 3-Manifolds
Much of what we have said so far can be generalized to 3-manifolds other than o
3
. For example,
a knot / in a 3-manifold ` is just a smoothly embedded copy of o
1
into `. Unfortunately,
there is no analogous concept of diagram of knots in a general manifold, but any term we dened
before without the use of a diagram can be dened for knots in a manifold. We will work towards
extending other concepts and results to knots in a manifold, but almost always we are going to need
to require that the manifold is an integral homology sphere so that we have the following property.
Proposition 2.16. If ` is a homology sphere and / an embedded knot, then `/ is a homology
circle, that is H

(`/) = H

(o
1
).
Proof. The proof is the same as in proposition 2.1.
Our rst denition of linking number of two oriented, disjoint knots needed a knot diagram,
so we cannot use this denition in a manifold. However, if ` is an integral homology sphere, and
/
1
and /
2
knots in `, proposition 2.16 says that H
1
(`(/
1
)) = Z is generated by [j], where j
is any meridian of (/) (which is just the boundary of any normal cross-section). So, as we did
before, we dene the linking number lk(/
1
, /
2
) to be the integer such that [/
2
] = lk(/
1
, /
2
) [j]. We
can also dene the canonical longitude / of a knot / in ` to be the closed curve in (`(/))
such that lk(/, /) = 0.
Let / be a knot in ` with canonical longitude /, and orient / and / such that the basis (/, /, n)
is the same as the orientation on ` (` is orientable since it is an integral homology sphere),
where n is the inward point normal vector. The choice of meridian j and longitude / gives an
identication : (`(/)) o
1
o
1
. Let j
0
: (`(/)) o
1
be j
0
=
S
1 , where
S
1 is
projection onto the second factor.
Proposition 2.17. Let 1 = `(/) be the knot complement. The projection j
0
: 1 o
1
extends to a map j : 1 o
1
.
Proof. Let i : 1 1 be the inclusion. We claim that j
0
extends to j if and only if the homotopy
class [j
0
] is in the image of the induced map i

: [1, o
1
] [1, o
1
]. Clearly if such an extension
exists then [j
0
] = i

[j]. Conversely, suppose there is some homotopy class of maps [j] such that
[j
0
] = i

[j]. This means that j


0
and j i are homotopic on 1, but we need them to be equal.
Let H be a homotopy between j i and j
0
and let l ` be a collar of 1 in 1, i.e. l is
dieomorphic to 1 [0, 1] such that 1 0 is mapped identically to 1 l. If we dene
j

: 1 o
1
to be j outside l and H(r, t) on the part of l corresponding to (r, t) 1 1, then
we see that j

is an extension of j
0
.
24
In appendix C we saw that we have a natural isomorphism of groups
[1, o
1
]

= H
1
(1)
since o
1
is a 1(Z, 1) space. The inclusion i induces a map on cohomology, and by naturality of the
above isomorphism, and naturality of the Poincare isomorphisms we get the following commutative
diagram from the long exact sequence of the pair (1, 1)
H
1
(1)
i
//
H
1
(1)
H
1
(1)
PD
OO

//
H
2
(1, 1)
PD
OO
where is the connecting homomorphism. We have that [/] is the Poincare dual of [j
0
], so by
commutativity of the above square we have [j
0
] = 11
1
i

[/] = 0, since [/] is the canonical


longitude, hence [j
0
] ker . But, by exactness we have [j
0
] ker = imi

, hence j
0
extends to a
map j by our previous claim.
This proposition can easily be extended to links. One might wonder if the map j
0
is a bration
of the knot complement over o
1
. This is not always true, so it leads to the notion of a bered
knot, and will be discussed later. We can use proposition 2.17 to show that Seifert surfaces exist
for links in integral homology spheres.
Proposition 2.18. Given a link 1 in an integral homology sphere ` there is a smoothly embedded,
compact, orientable and connected surface 1 such that 1 = 1.
Proof. Let j : 1 1
2
be the map constructed in proposition 2.17, where 1 is the link comple-
ment in `. We can slightly homotopy j such that it is transverse to a point 1
2
, in which
case 1

= j
1
() is a smoothly embedded submanifold of 1 of codimension 1 (by proposition 1.1)
with boundary contained in 1, hence 1 is a surface. We can connect the boundary of 1

to 1
via annuli contained in the tubular neighborhoods of the link components to get a Seifert surface
1.
25
Handle Anatomy
Attaching map : 1
k
1
n
A
Attaching region 1
k
1
nk
Attaching sphere 1
k
0
Belt sphere 0 1
nk
Core 1
k
0
Cocore 0 1
nk
Belt Sphere
Core
Cocore
Attaching Sphere
Attaching Region
D
nk
D
k
Figure 3.1: Anatomy of an n-dimensional /-handle
3 Handlebodies
3.1 Handlebody Decompositions
Let A be a smooth n-dimensional manifold with boundary. The process of attaching an n-
dimensional /-handle to A is to form the quotient space A

/, where / is a copy of 1
k
1
nk
,
and where : 1
k
1
nk
A is a smooth embedding. Notice that 1
k
1
nk
is only part of
the boundary of 1
k
1
nk
. Also, the manifold A

/ is not smooth, as it has corners. However,


there is a canonical way to smooth the corners, so we may assume the quotient space is a smooth
n-manifold with boundary. If two attaching maps ,

: 1
k
1
nk
A are isotopic, then
the resulting manifolds A

/ and A

/ are dieomorphic. Also note that there is an obvious


deformation retract of 1
k
1
nk
onto 1
k
0, hence A

/ deformation retracts onto A

1
k
0
(where is suitably restricted in the second space), hence attaching a /-handle is the same as
attaching a /-cell.
The anatomy of a handle is simple and the pieces have intuitive names. The map is called
the attaching map, while 1
k
1
nk
is called the attaching region (and sometimes the image
of this set under ). The subset 1
k
0 is called the core of the handle, and 0 1
nk
the cocore.
The region 1
k
0 the attaching sphere, and 0 1
nk
the belt sphere. Finally, the number
/ is called the index of the handle. See gure 3.1.
Unfortunately this picture is a little misleading since attaching a /-handle does not always
look like attaching a ball along two disconnected pieces of its boundary. For example, attaching
26
a 4-dimensional 2-handle means attaching a copy of 1
2
1
2
along the solid torus o
1
1
2
. The
boundary of 1
2
1
2
is a union of solid tori o
1
1
2
1
2
o
1
that intersect at their boundaries
o
1
o
1
, and this decomposition of o
3
into two solid tori is clearly the genus 1 Heegaard splitting.
An embedding : 1
k
1
nk
A is the same as embedding the trivial normal bundle
of the attaching sphere (1
k
0) into A. Therefore determines, and is determined by, an
embedding
0
: 1
k
0 A of the attaching sphere and an equivalence ) of the normal bundle

0
(1
k
) with the trivial vector bundle 1
k
R
nk
; this is intuitively plausible, but is also the
content of the Tubular Neighborhood Theorem. The embedding
0
is just a knot in A, and the
vector bundle equivalence ) is called a (normal) framing of 1
k
= o
k1
. If the data (
0
, )) and
(

0
, )

) are smoothly isotopic, then the resulting manifolds A

/ and A

/ are isotopic.
Fix a framing )
0
for o
k1
, i.e. a vector bundle equivalence
o
k1
R
nk
f
0
//

0
(o
k1
)

o
k1

0
//

0
(o
k1
)
If ) is another framing of o
k1
, then )
1
)
0
: o
k1
R
k
o
k1
R
nk
must map (j, ) to
(j, p(j) ), where p : o
k1
GL(n /) is smooth. Without loss of generality we can assume
that p(j) = 1 for a xed base point j o
k1
. Therefore each framing determines an element of
the group
k1
(G1(n /)), which is isomorphic to
k1
(O(n /)) by the deformation retract of
G1(n/) onto O(n/), and it is even true that the set of framings are in one-to-one correspondence
with elements of
k1
(O(n /)). However, the set of framings is not canonically isomorphic to

k1
(O(n /)), since the identication depends on )
0
, but it is a principal homogeneous space for

k1
(O(n /)). Sometimes there is a canonical choice of )
0
, and this will be useful later.
We can get a lot of information about attaching very low index handles and very high index
handles. The attaching sphere for a 0-handle 1
0
1
n
is empty, so attaching a 0-handle to A is
the same as simply taking the disjoint union of A with 1
n
. The attaching sphere of a 1-handle is
a disjoint pair of points, and so if A is connected and non-empty there is a unique isotopy class
of attaching maps (well, except in the case n = 2 and A non-compact, in which case we can also
interchange the two components of the attaching sphere). Further, since
0
(O(n 1)) = Z,2 for
n 2 there are exactly two framings on o
0
, hence there are exactly two manifolds one can obtain
by attaching a 1-handle to a 0-handle (these are the n-dimensional analogs of annulus and Mobius
band).
Similarly, for (n1)-handles, with n ,= 2, there is a unique framing of o
n2
since
n2
(O(1)) = 0.
In the case n = 2 we see that (n 1)-handles are just 1-handles, which were discussed above.
Attaching an n-handle is the same as gluing a copy of 1
n
to A along 1
n
= o
n1
. Since A
is (n 1)-dimensional we see that we can only attach 1
n
to boundary components of A that are
dieomorphic to o
n1
. However, due to issues of exotic n-spheres and exotic dieomorphisms of
n-spheres we can only say that attaching an n-handle is unique in dimensions 6.
The other combinations of / and n, where / is neither small nor close to n, are harder to
understand. If n / 2, then there is only one isotopy class of embeddings o
k1
into (n
1)-dimensional manifolds, hence all embeddings can be unknotted. So, the only thing that
controls what we get from attaching a /-handle to a 0-handle is the framing of an unknotted sphere
in o
n1
, which are of course parameterized by
k1
(O(n /)). Suppose we are attaching a /-
handle /
2
= 1
k
1
nk
to a 0-handle /
1
= 1
n
such that the attaching sphere is the standardly
embedded, unknotted sphere o
k1
o
n1
. Without loss of generality we can assume that we have
27
an identication /
1
= 1
k
1
nk
and that the attaching map of /
1
is of the form : 1
k
1
nk

1
k
1
nk
1
n
such that
D
k is the identity, and the restriction of to any slice 1
nk
is an orthogonal linear map. Let A be the manifold /
1

/
2
. We claim that there is a canonical map
: A o
k
making A into a 1
nk
-bundle. Decompose o
k
as a union of two discs, the upper and
lower hemispheres. Then projects the handle /
1
onto its its rst factor, which is then identied
with the upper hemisphere, and projects /
2
onto its rst factor, which is then identied with the
lower hemisphere. It is clear that this makes A into a 1
nk
-bundle, and in fact all 1
nk
-bundles
are of this form. Further, consider the framing )
0
corresponding to id
D
k
D
nk. This clearly gives
the trivial bundle A = o
k
1
nk
, and so this is a canonical element to associate to the identity
in O(n /).
If we apply the above discussion to the case / = 2, then we have that 1
n2
-bundles over o
2
are classied by Z if n = 4 and by Z,2 if n 4. In this case we can get a canonical identication
of these bundles with Z (n = 4) and Z,2 (n 4) by associating the trivial bundle with 0. For
n = 4, the integer associated to a 1
2
bundle over o
2
is called the Euler number of the bundle. If
n 4, then we denote the non-trivial 1
n2
bundle over o
2
by 1
n2

o
2
, and we also dene the
associated o
n3
bundle over o
2
by o
n3

o
2
= (1
n2

o
2
).
Let A be a compact n-manifold with boundary written as
+
A .

A. If A is oriented give

+
A the induced orientation and

A the opposite of the induced orientation. A handlebody


decomposition of A is an identication of A with an n-manifold obtained by attaching handles to

A1 such that

A is identied with

A0. A manifold with such a decomposition is called


a relative handlebody built on

A, or if

A = then A is simply called a handlebody.


The theory of handlebodies and Morse theory are closely related, so it is hard to discuss one
without the other. Unfortunately we have now come to a point where we need to assume the reader
has some knowledge of Morse theory. One can read this section concurrently with section 4.
According to the results in section 4 we have that every smooth, compact n-manifold has a
handlebody decomposition using Morse theory, and this can even be extended to non-compact
manifolds with compact boundary. Amazingly every topological n-manifold with n ,= 4 also admits
a handlebody decomposition, but this is not true in dimension n = 4. For example, there is a
4-manifold with intersection form 1
8
which does not admit a smooth structure or a handlebody
decomposition.
Proposition 3.1. Any handlebody decomposition of a pair (A,

A) can be modied (by isotoping


attaching maps) so that handles are attached in order of increasing index, and handles of the same
index can be attached in any order or simultaneously.
Proof. Suppose we attach a /-handle /
1
and then an /-handle /
2
to a manifold A, with / /. First
we will perturb the attaching map of /
2
slightly so that its attaching sphere does not intersect /
1
s
belt sphere (if needed). Note that /
1
s belt sphere, 0 1
nk
, is of dimension n / 1 and /
2
s
attaching sphere, 1

0, is of dimension / 1. The codimension of the intersection of generically


positioned submanifolds is the sum of the submanifolds codimension, so the codimension of /
1
s
belt sphere intersected with /
2
s attaching sphere, in A, is
(n 1) (n / 1) + (n 1) (/ 1) = n 1 n +/ + 1 +n 1 / + 1 = n +/ / n
Hence we can perturb the attaching sphere of /
2
slightly so that the spheres become entirely
disjoint. Now take a smooth vector eld on A /
1
emanating from the cocore of /
1
radially, and
ow the attaching sphere of /
2
along this vector eld. This will push the attaching sphere of /
2
to
A, so we can rst attach /
2
and then /
1
.
28
We will always assume that our handlebodies are ordered so that handles are attached with
increasing index, and we will write A
k
to denote

A 1 union all the handles of index /.


A handlebody decomposition of a pair (A,

A) naturally leads to a dual handlebody decom-


position of (A,
+
A), where /-handles in the rst decomposition become (n /)-handles in the
second decomposition. When we attach a /-handle / to A, the points in 1
k
1
nk
become
interior points of A / and the points in 1
k
1
nk
become boundary points in A /. Other
handles will be attached to the part 1
k
1
nk
of /, so we could reverse this and instead think
of / has an (n /)-handle being attached to those other handles. Intuitively, we are turning the
handlebody upside down. If the handlebody decomposition is given by a Morse function ), then
the dual handlebody is just the handlebody given by the Morse function 1 ).
Example 3.1. We will show how C1
n
, complex projective n-space, can be decomposed into a
handlebody with one 2/-handle for each 0 / n. Recall that C1
n
is the quotient of C
n+1
0 by
C

acting by scalar multiplication. The equivalence class of (.


0
, . . . , .
n
) is denoted by [.
0
: : .
n
].
Complex projective space is an n-dimensional complex manifold (or 2n-dimensional real manifold)
that can be covered with n + 1 charts
i
: C
n
C1
n
dened by

i
(.
0
, . . . , .
n1
) = [.
0
: : .
i1
: 1 : .
i
: : .
n1
]
The inverses
i
: l
i
C
n
of these charts are given by

i
([.
0
, . . . , .
n
]) = (.
0
,.
i
, . . . , .
i1
,.
i
, .
i+1
,.
i
, . . . , .
n
,.
i
)
where l
i
consists of all points [.
0
, . . . , .
n
] such that .
i
,= 0.
Let 1 be the unit disc in C, and and let 1
i
=
i
(1 1) (an embedded complex n-ball
in C1
n
). Every point j C1
n
has a unique representative (.
0
, . . . , .
n
) such that [.
i
[ = 1 for some
0 i n and [.
j
[ 1 for all , (just divide by the coordinate with largest norm), which we will
call the canonical representative of j. It follows that 1
i
covers C1
n
. We also see that j 1
i
if
and only its canonical representative has [.
i
[ = 1, and j 1
i
if and only if [.
j
[ = 1 for some other
, ,= i. From this it follows that two balls 1
i
and 1
j
can intersect only at points in the boundaries
of each ball: j 1
i
1
j
if and only if its canonical representative has [.
i
[ = [.
j
[ = 1 and [.
k
[ 1
for all /. In particular, then ball 1
k
intersects the union
i<k
1
i
at all points whose canonical
representative has [.
k
[ = 1, [.
j
[ = 1 for some 0 , < / and [.

[ 1 for all /. This means that


the restriction of
k
to (1 1) 1 1 1
k
1
nk
is an embedding into
i<k
1
i
,
therefore
ik
1
i
can be obtained from
i<k
1
i
by attaching a 2/-handle via
k
.
By replacing R for C in the above construction we get a decomposition of R1
n
into a handlebody
with one /-handle for each 0 / n. We are going to explicitly compute the attaching maps for
R1
1
, R1
2
, C1
1
and C1
2
. From the description above we have that R1
1
is constructed by gluing
1
1
= [r, 1] : 1 r 1 to 1
0
= [1, r] : 1 r 1 via the map
1
restricted to 1 R. In
this case
1
: 1, +1 1
0
and
1
(1) = [1, 1] and
1
(1) = [1, 1]. This is the attaching map,
and it attaches one interval to another by identifying endpoints, hence R1
1
is just a circle.
The real projective plane R1
2
can be written as a union of a 0-, 1- and 2-handle. From the
construction we described we have the following balls
1
0
= [1, r, j] : 1 r, j 1
1
1
= [r, 1, j] : 1 r, j 1
1
2
= [r, j, 1] : 1 r, j 1
We attach a 1-handle to 1
0
via the map
1
[
DD
, which maps (1, r) to [1, 1, r] = [1, 1, r]
1
1
. Explicitly, this means that (1, r) is mapped to [1, 1, r] and (1, r) maps to [1, 1, r], so
29
Figure 3.2: Attaching the 1-handle in R1
2
Figure 3.3: Attaching the 2-handle in R1
2
this is identifying edges of two squares with a twist (see gure 3.2). We attach a 2-handle to
1
0
1
1
via
2
[
(DD)
which maps (1, r) to [1, r, 1] = [1, r, 1] 1
0
and maps (r, 1) to
[r, 1, 1] = [r, 1, 1] 1
1
. This glues a disc into 1
0
1
1
as in gure 3.3, where have not yet
identied the nal 2 sides of the square since that would make it dicult to draw 1
0
1
1
1
2
.
Now consider C1
1
, which is the union of a 0-handle and 2-handle. By denition we have
1
0
= [1, .] : [.[ 1
1
1
= [., 1] : [.[ 1
We attach a 2-handle to 1
0
via
1
which maps . to [., 1] = [1, 1,.] = [1, .] 1
0
. This means we
attach two discs along the map . ., so C1
1
is just the 2-sphere.
Finally, consider C1
2
, which is the union of a 0-handle, 1-handle and 2-handle. We have
1
0
= [1, ., n] : [.[, [n[ 1
1
1
= [., 1, n] : [.[, [n[ 1
1
2
= [., n, 1] : [.[, [n[ 1
We attach a 2-handle to 1
0
via
1
restricted to 1 1, which maps (., n) to [., 1, n] =
[1, 1,., n,.] = [1, ., .n] 1
0
. The 4-handle is attached to 1
0
1
1
via the restriction of
2
to (11), which maps (., n) to [1, .n, .] on the 11 part, and maps (., n) to [.n, 1, n] on
the 1 1 part.
30
There is a natural handlebody decomposition on the product of two handlebodies. Consider
the simplest case
A = 1
n
1

1
/
1
/
1
= 1
k
1
1
n
1
k
1

1
: 1
k
1
1
n
1
k
1
1
n
1
Y = 1
n
2

2
/
2
/
2
= 1
k
2
1
n
2
k
2

2
: 1
k
2
1
n
2
k
2
1
n
2
Then A Y has the following handlebody decomposition
A Y = 1
n
1
1
n
2

1
id
/
1
1
n
2

id
2
1
n
1
/
2
/
1
/
2
Here we are thinking of /
1
1
n
2
as an (n
1
+ n
2
)-dimensional /
1
-handle since it is just 1
k
1

1
n
1
k
1
1
n
2
(the second two factors are like a (n
1
+ n
2
)-ball), and 1
n
1
/
2
is an (n
1
+ n
2
)-
dimensional /
2
-handle since it is just 1
n
1
1
k
2
1
n
2
k
2
(here the factors are out of order, but
that does not matter). We also think of /
1
/
2
as an (n
1
+n
2
)-dimensional (/
1
+/
2
)-handle since
it is just 1
k
1
1
n
1
k
1
1
k
2
1
n
2
k
2
, and the attaching region of this handle is
1
k
1
1
n
1
k
1
1
k
2
1
n
2
k
2
1
k
2
1
n
2
k
2
1
k
1
1
n
1
k
1
so the attaching map is
1
id
2
id.
Note that if the handlebody decompositions of A and Y are induced by Morse functions ) and p,
respectively, then the product handlebody decomposition of AY is the same as the decomposition
induced by the Morse function / : AY R dened by /(r, j) = )(r) +p(j). The critical points
(r, j) A Y of / are precisely where d/
(x,y)
= d)
x
+dp
y
= 0, hence d)
x
= 0 and dp
y
= 0 and so
r must be a critical point of ) and j a critical point of p. If the critical point r is of index / and j
of index |, then (r, j) is of index / +|, for the Hessian is of the form
H
h
(r, j) =
_
_
_

2
f
x
i
x
j
_
0
0
_

2
g
y
i
y
j
_
_
_
=
_
H
f
(r) 0
0 H
g
(j)
_
3.2 Handle Moves
Any manifold admits many dierent handlebody decompositions, so we want to nd a set of moves
that we could perform on handles in order to bring one handlebody presentation into another. First
we show how to add a (/ 1)-handle and / in a specic way such that the result manifold does not
change; that is, the handles cancelled each other. Let A be an n-manifold with boundary. Write
1
k
as a union of upper and lower hemispheres, o
k1
= 1
k1
+

S
k2 1
k1

. Attach 1
k
1
nk
to
A via an embedding of 1
k1

1
nk
in A (this is not attaching a /-handle). This does not
change the dieomorphism type of A since it is just take the boundary sum of A with an n ball.
We are going to show how to decompose this attached copy of 1
k
1
nk
into a (/ 1)-handle
and /-handle. Let J be a small neighborhood of 1
k1
+
in 1
k
, so that J = 1
k1
+
1
1
, and let 1
k
0
be the disc left from slicing J o of 1
k
. Then J 1
nk
is a (/ 1)-handle attached to A, and
1
k
0
1
nk
is a /-handle attached to (AJ 1
nk
). This process of creating a pair of canceling
handles is called handle creation/cancellation.
Sometimes it is possible to nd a (/ 1)-handle and /-handle that are situated in A so that
they cancel like the above.
Proposition 3.2. A (/ 1)-handle / and a /-handle /

can be cancelled if the attaching sphere of


/

intersects the belt sphere of / transversely in a single point.


31
Proof.
The process of nding two handles that cancel each other and canceling them from the han-
dlebody is called handle cancellation. We can use handle cancellations to show that connected
handlebodies can be made to have only one 0-handle and one n-handle.
Proposition 3.3. If A is a compact and connected n-manifold, then A admits a handlebody
decomposition with exactly one 0-handle and one n-handle.
Proof. First we eliminate all 0-handles except for one. The 0-handles A
0
of A is just a disjoint union
of n-balls, and since A is connected and since 1-handles are the only handles with disconnected
attaching spheres, we have that each ball of A must be connected to some other ball via a 1-handle.
The attaching sphere of a 1-handle is a pair of points, and the belt sphere of a 0-handle is o
n1
, so
it is clear by proposition 3.2 that if a 1-handle connects two 0-handles then we can cancel one of
the 0-handles and the 1-handle. Repeating this for all 1-handles connecting two 0-handles we can
cancel all 0-handles except for one.
Now that we have eliminated all 0-handles except for one, take the dual handlebody decompo-
sition. This is a handlebody decomposition with a unique n-handle, but possibly many 0-handles.
Apply the procedure from before to eliminate all 0-handles except for one and we are left with a
handlebody with a unique 0- and n-handle.
Let (/
1
,
1
) and (/
2
,
2
) be two /-handles with attaching maps, and suppose they are attached
to A simultaneously: A

2
(/
1
/
2
). Let be the image of the attaching sphere of /
1
in A
and 1 the image of the belt sphere in A. We can isotope the attaching map of /
1
, which can also
be thought of as isotoping the attaching sphere and the framing of the attaching sphere, and the
resulting manifold will not change. By isotoping the attaching sphere we are, intuitively, sliding
the handle /
1
around (A /
2
). Now, slide the attaching sphere of /
1
over the handle /
2
until
the attaching sphere of /
1
intersects the belt sphere of /
2
. Since the attaching sphere of /
1
is of
dimension / 1 and the belt sphere of /
2
is of dimension n/ 1, we have that they will intersect
at a point j, and T
p
T
p
1 is a codimension 1 subspace of T
p
(A /
2
). This means there are
two directions we can move past 1 (corresponding to the two components of the complement of
the hyperplane T
p
T
p
1). One direction will slide /
1
back to where we started. If we push
the other way, and continue sliding /
1
until the attaching sphere lies entirely in A, then we get a
new handlebody which is the result of sliding the handle /
1
over /
2
. Without loss of generality
we can also assume that as we slide /
1
over /
2
, the attaching sphere of /
1
stays inside some disc
1
k
pt of /
2
.
We have not change the manifold A /
1
/
2
by sliding the handle /
1
over /
2
(since we just
isotoped the attaching sphere around), but the framing of /
1
might have changed in the process.
For example, gure 3.4 shows that sliding a trivially attached 1-handle across a twisted 1-handle
results in another twisting.
It turns out these two moves on handlebody decompositions can always determine when two
handlebody decompositions give the same manifold.
Proposition 3.4. For any two relative handlebody decompositions of a compact pair (A,

A)
(with handles attaching in increasing order), it is possible to bring one decomposition into the
other via a nite sequence of creation/cancellation of handle pairs, handle slides, and isotopies of
attaching maps.
32
Figure 3.4: A handle slide that changes the framing of a handle
3.3 The Fundamental Group and Homology of a Handlebody
The combinatorial presentation of a manifold as a handlebody immediately gives a presentation of
the fundamental group of the manifold. Suppose A is a handlebody with unique 0-handle, and give
an orientation to the core of each 1-handle. We can make the oriented cores 1
1
0 into oriented
loops in A
1
by connecting the end points with an oriented, straight line segment to the origin in
A
0
= 1
n
. These loops are generators of the free group
1
(A
1
, r
0
). As we attach 2-handles we
will be adding relations to
1
(A
1
, r
0
) to get
1
(A
2
, r
0
). In particular, express the attaching circle
1
2
0 of a handle in terms of the loops in A
1
to get the relation that attaching this 2-handle adds.
Handles of higher index do not add any other relations since their attaching spheres are simply
connected, so in particular
1
(A
2
, r
0
) =
1
(A, r
0
). Note that the above description of
1
(A) is
analogous to the presentation of
1
(A) in terms of a CW-decomposition of a space.
Next we develop a homology theory in terms of handlebodies, which is similar to cellular ho-
mology derived from a CW-complex. For a relative handlebody (A,

A), let the /-chains of our


homology theory be C
k
(A,

A) = H
k
(A
k
, A
k1
), where H
k
(, ) is just regular singular homol-
ogy, hence C
k
(A,

A) can be thought of as formal linear combinations of /-handles (once the


cores of the handles are given a xed orientation). The boundary operator
k
: C
k
(A,

A)
C
k1
(A,

A) is dened by the long exact sequence of the triple (A


k
, A
k1
, A
k2
). In particular,
we have the following pieces of long exact sequences
H
k
(A
k
, A
k1
)

k
H
k1
(A
k1
)
H
k1
(A
k1
)
j
H
k1
(A
k1
, A
k2
)
where
k
is the connecting morphism and ,

is induced by inclusion of the pairs (A


k1
, )
(A
k1
, A
k2
). Then we dene
k
= ,


k
, and clearly
2
= 0 since
k
,

= 0. If A is oriented,
then there is a more geometric denition of . Let / be a /-handle with attaching sphere , and
let 1
i
be the belt sphere of each (/ 1)-handle /
i
. Then

k
/ =

i
( 1
i
)/
i
(3.1)
where 1
i
is the intersection number computed in
+
A
k1
. Since ker im we dene
H
k
(A) = ker
k
, im
k+1
, and call this the /-th handlebody homology group of A (this term
is not standard). We can get a homology theory H
k
(A,

A; G) with coecients in any abelian


group G by dening C
k
(A,

A; G) = C
k
(A,

A)
Z
G with boundary operator id. We can
33
also dualize this construction to get handlebody cohomology. In particular, set C
k
(A,

A; G) =
Hom(C
k
(A,

A), G) and let


k
: C
k
(A,

A; G) C
k+1
(A,

A; G) be the induced dierential,


and dene H
k
(A,

A; G) = ker
k
, im
k1
. This is called the /-th handlebody cohomology
group of A. If there is risk of confusing the handlebody chain and homology groups with other
homology theories (such as singular homology), we may sometimes write them as C
HB
k
(A) and
H
HB
k
(A). Further, if

A = , then we will write these groups as C


k
(A; G), H
k
(A; G), and so on.
Since handle moves do not change the dieomorphism type of A we have that the homology and
cohomology groups will be the same. However, handle moves will add extra relations or change
our canonical generators. Suppose a /-handle / and (/ 1)-handle /
i
form a canceling pair so
that 1
i
= 1, where is the attaching sphere of / and 1
i
is the belt sphere of /
i
. Then
/ = /
i
+

j=i
o
j
/
j
for some integers o
j
. When we cancel these handles we will add the relations
/ = 0 and /
i
=

j=i
o
j
/
j
to the chain groups. On the other hand, sliding a handle / across a
handle /

changes the canonical generators of C


k
(A) by replacing the basis element / with / /

.
Example 3.2. Since o
n
has a handlebody decomposition 0-handle 1-handle, we can easily see
that its handlebody homology is H
k
(o
n
) = Z for / = 0, n and 0 for 0 < / < n. A more interesting
example is writing o
3
as a union of a 0, 1, 2 and 3-handle, o
3
= /
0
/
1
/
2
/
3
. The standardly
embedded solid torus A
1
= o
1
1
2
in o
3
is a 0-handle 1-handle. The attaching circle of the
2-handle we attach to A
1
is just the longitude o
1
, where is any point on the boundary
of 1
2
, and let A
2
be the result of this attachment. Then A
2
is just o
2
, and so we attach the
3-handle in the only way possible, and the result is A
3
= o
3
.
We have C
i
(A) = Z/
i
). Since there are no (1)-handles, we have
0
= 0. The attaching
sphere of /
1
intersects the belt sphere of /
0
at two points, but dierent signs, so
1
(/
1
) = 0. The
attaching sphere of /
2
intersects the belt sphere of /
1
at one point, which we can arrange to have
intersection number +1, hence
2
(/
2
) = /
1
. Finally, the attaching sphere of /
3
intersects the belt
sphere of /
2
at two points, but with dierent signs, so
3
(/
3
) = 0. Taking homology we see again
that H
0
(o
3
) = H
3
(o
3
) = Z and H
1
(o
3
) = H
2
(o
3
) = 0.
Example 3.3. We saw in example 3.1 that C1
n
admits a handlebody decomposition with one
/-handle for each even integer / between 0 and 2n. So C
k
(C1
n
) = Z for all even 0 / 2n, and
0 for all other /. This implies that all the dierentials are zero, hence H
k
(C1
n
) = C
k
(C1
n
).
3.4 2-Dimensional Handlebodies
Two dimensional handlebodies have simple descriptions since there are so few ways to attach 2-
dimensional /-handles (/ = 0, 1, 2). The 0- and 2-handles are just 2-discs, and a 1-handle is a band
1 1. The attaching region of this band consists of two disjoint intervals 1 . 1, and there are two
ways to attach the band to a 0-handle: without a twist, or with a twist. Once the 1-handles are
attached, there is a unique way to attach the 2-handles so that we get a closed surface (just ll
the holes!), so we can draw diagrams of 2-dimensional handlebodies by just drawing the 0-handle
1-handles. Some examples of these decompositions are shown in gure 3.5, where the rst image
is R1
2
, the second is T
2
#R1
2
, and the third is T
2
.
3.5 Heegaard Diagrams
Heegaard splittings are an old method for constructing 3-manifolds, and we will see that it is equiv-
alent to 3 dimensional handlebody. Although we get nice combinatorial pictures for 3 dimensional
handlebodies, called Heegaard diagrams, they were still dicult to use. However, much later,
34
Figure 3.5: Examples of 2-dimensional handlebody diagrams
many invariants were built from Heegaard diagrams, including a recent variation of Lagrangian
Floer homology, which provides powerful invariants of 3- and 4-manifolds.
We dene a handlebody of genus p to be a 0-handle 1-handles such that the result is
orientable, i.e. the interior (in the Jordan sense) of an orientable, genus p surface. If H
1
and
H
2
are handlebodies of genus p, then for any orientation-reversing homeomorphism : H
1

H
2
of their boundaries, we can form the orientable, compact 3-manifold H
1

H
2
. Conversely,
given a 3-manifold `, it is always possible to nd a pair of embedded submanifolds H
1
, H
2
, both
homeomorphic to a handlebody of genus p, such that H
1
= H
2
(this is proved below). In either
case we call this a Heegaard splitting, or Heegaard decomposition, of the 3-manifold.
Theorem 3.1. Every compact, orientable 3-manifold admits a Heegaard splitting.
Proof. We will give two dierent proofs.
The rst proof is more geometric, using the fact that every 3-manifold has a triangulation.
Let 1 be a triangular of `, and let 1

and 1

denote the rst and second barycentric


subdivision of 1 respectively. Let us recall how the barycentric subdivision is dened. Take
a 2-face in the triangulation and connect, with a new edge, the barycenter of with each
vertex of and the midpoint of each edge of . This divides into 6 new 2-faces. Then
we can divide each 3-face into 24 new 3-faces by coning the barycenter of over each
subdivided face of . Performing this operation on all faces of 1 gives us a new triangulation
of `, called the rst barycentric subdivision 1

. Performing this operation again gives us


the second barycentric subdivision 1

.
Let H
1
be the union of the 3-faces in 1

that do not intersect the 1-faces of 1, and let H


2
be the closure of the complement of H
1
. Then H
1
and H
2
are handlebodies which intersect
precisely at their boundaries. This gives a Heegaard splitting of `.
The second proof is Morse theoretic, and is the beginning of more versatile methods in low
dimensional topology via Morse theory. As we will see later (theorem 4.6), any smooth
manifold admits a self-indexing Morse function ) : ` [0, n] with unique index 0 and index
n critical points. Let ) : ` [0, 3] be such a Morse function for our 3-manifold, and let
H
1
= )
1
[0, 3,2] and H
2
= )
1
[3,2, 3]. If ) has p index 1 critical points, then H
1
is a 0-
handle with p 1-handles attached, hence a genus p handlebody. If we dualize the handlebody
decomposition we see that H
2
is a 0-handle with / 1-handles attached, where / is the number
of index 2 critical points. Since H
2
= H
1
= a surface of genus p, we have that p = /, and
35
Figure 3.6: Standardly embedded handlebody
so H
2
is also a handlebody of genus p. Therefore ` has been expressed as a union of two
handlebodies along their boundary.
The proof of theorem 3.1 is a little discouraging since any manifold admits many triangula-
tions, and hence many splittings. But, there is a nice combinatorial picture to help us understand
the structure of a Heegaard splitting. Let H
1
and H
2
be two copies of the standardly embedded
handlebody (see gure 3.6), and let be the standardly embedded surface of genus p. Take an
orientation-preserving homeomorphism
1
: H
1
and an orientation-reversing homeomor-
phism
2
: H
2
, and form the compact, orientable 3-manifold ` = H
1

1
2

1
H
2
. Let

1
, . . . ,
g
be the image of the standard circles in (the thin circles in gure 3.6) under
1
, and

1
, . . . ,
g
the image of the standard circles in under
2
. The curves
1
, . . . ,
g
are collectively
known as the -curves, and are mutually disjoint, and the curves
1
, . . . ,
g
are collectively known
as the -curves, and are also disjoint.
We now have a picture associated to any Heegaard decomposition: its a surface with 2p curves
drawn on it. Many times we will take one of the collections of curves to be the standard curves,
which would mean that
1
or
2
is just the identity, in which case we will not even draw them.
The data of the surface with these 2p embedded curves is called the Heegaard diagram of
the splitting. We would like to know that these embedded curves somehow uniquely determine
the 3-manifold obtained from gluing the handlebodies together, but rst we state a small technical
lemma that will be used repeatedly in these notes.
Lemma 3.1. Let A and Y be two n-manifolds with homeomorphic boundaries, and let 7 = A
f
Y
be the space formed by gluing A and Y together along a homeomorphism ) : A Y . If
: A A is any homeomorphism that extends to a homeomorphism : A A, then 7 is
homeomorphism to 7

= A
f
Y .
Proof. There are embeddings of A and Y into 7 by sending a point to its equivalence class in 7.
If r A, j Y are in the interior, then the corresponding points in 7 are [r] = r and [j] = j,
whereas if r A, j Y are boundary points, then [r] = r, )(r) and [j] =
_
j, )
1
(j)
_
. Let

A and

Y denote the images of A and Y under these embeddings so that

A

Y is just the set of
equivalence classes with representative in A or Y . We dene a map p : 7 7

on a point . in
A or Y by the following
p(.) =
_
[
1
(r)] .

A
[j] .

Y
where r A and j Y are representatives of .. This map is clearly well-dened. In order for it
to be continuous we need [
1
(r)] = [j] in 7

for any . = r, j in

A

Y (i.e. j = )(r)). This is
easy to see for on one side we have
[
1
(r)] = [
1
(r)] =
_

1
(r), )((
1
(r))
_
=
_

1
(r), )(r)
_
36
and on the other side we have
[j] = [)(r)] =
_

1
(r), )(r)
_
Therefore the map is continuous, and now it is clear that it is even a homeomorphism.
Theorem 3.2. The homeomorphism type of the manifold resulting from gluing two handlebodies
together depends only on the image of the standard meridians under the gluing map, hence equal
Heegaard diagrams describe homeomorphic manifolds.
Proof. Suppose we are gluing genus p handlebodies H
1
and H
2
together via a homeomorphism
) : H
1
H
2
. We can glue these handlebodies together in a two step process. First, let

1
, . . . ,
g
denote a small tubular neighborhood of each standard meridian in H
1
(hence these are
annuli), and let 1
1
, . . . , 1
g
denote the thickened discs these annuli bound in H
1
. Then we can rst
attach the discs 1
i
to H
2
via ) to obtain the intermedia space `

. Now we are left with attaching


H
1
(1
1
1
g
) to `

, but this is topologically a ball, and so any homeomorphism from the


boundary of this ball to `

(which is just a sphere) can be extended to the interior of the ball,


and hence any gluing map will give a homeomorphic space by the above gluing lemma. Therefore
the homeomorphism type of H
1

f
H
2
depends only on the images of the
i
curves.
We can also dene Heegaard diagrams axiomatically. Let us temporarily say that an H-diagram
is a surface of genus p and embedded curves
1
, . . . ,
g
,
1
, . . . ,
g
such that the -curves are
disjoint and the complement of their union is connected, and similarly for the -curves.
Theorem 3.3. An H-diagram (,
1
, . . . ,
g
,
1
, . . . ,
g
) corresponds to the Heegaard diagram of
some splitting.
Proof. Let
i
,
i
be curves with the above properties. First we claim that cutting along the
-curves yields a 2-sphere with 2p holes. Suppose that with the -curves removed is a surface
of genus / with 2p holes. Then its Euler characteristic would be 2 2/ 2p. On the other
hand, removing circles from surfaces that do not separate the surface does not change the Euler
characteristic, so we should have 22p = 22/2p, hence / = 0, and so
1
, . . . ,
g
is a sphere
with 2p holes.
Since with the removed -curves is just a sphere with 2p holes, we can homeomorphically
deform it so that it is in the standard position of a handlebody of genus p with its standard circles
cut out. Now it is easy to nd a handlebody of genus p with homeomorphism of its boundary
onto taking its standard circles to the -circles. This can be done for the curves in the same
way.
Therefore H-diagrams and Heegaard diagrams are equivalent notions, so we will refer to both as
Heegaard diagrams from now on. The condition that the complement of the union of the -curves
in is connected is equivalent to saying that the homology classes of
1
, . . . ,
g
span a rank p
submodule of H
1
()

= Z
2g
.
There is also a Morse theoretic way of thinking of Heegaard diagrams. Let ) : ` [0, 3]
be a self-indexing Morse function with unique maximum and minimum, and let H
1
= )
1
[0, 3,2],
H
2
= )
1
[3,2, 3] and = )
1
(3,2). By theorem 3.1 we know that H
1
and H
2
are handlebodies
of the same genus, say p. Let 1
1
, . . . , 1
g
be the index 1 critical points of ), and Q
1
, . . . , Q
g
be the
index 2 critical points. Let
i
consist of the points that ow to 1
i
under
g
), and let
i
consist
of the points that ow to Q
i
under +
g
). Then (
1
, . . . ,
g
,
1
, . . . ,
g
) is a Heegaard diagram
for `.
37
Figure 3.7: Heegaard diagrams of o
1
o
2
, o
3
, R1
3
respectively
Example 3.4. There is only one Heegaard splitting of genus 0. In this case we have a manifold
` expressed as ` = 1
1
1
2
where 1
1
, 1
2
are 3-balls. Let : 1
1
o
3
be an embedding of the
rst 3-ball onto the upper-hemisphere of o
3
. The restriction of to 1
1
= 1
2
can be extended
to a homeomorphism from 1
2
to the lower-hemisphere of o
3
(by Alexanders lemma). Pasting
these homeomorphisms together gives a homeomorphism of ` with o
3
, hence all genus 0 Heegaard
splittings are o
3
.
Example 3.5. The standard Heegaard splitting of o
3
is to take the standardly embedded solid
torus H
1
in o
3
, and let H
2
be the complement of H
1
in o
3
. The gluing map is just the identity.
The standard orientation on o
3
induces orientations on H
1
and H
2
. If we let n denote the outward
pointing normal vector eld on H
1
, then the orientation of H
1
induces an orientation on H
1
by
choosing the equivalence class of bases at a tangent space such that adding n gives the equivalence
class of bases from the induced orientation on H
1
. Do something similar to get an orientation on
H
2
induced by H
2
. Clearly the two orientations induced on H
1
= H
2
are dierent, therefore
the gluing map is orientation reversing. This is why we require the gluing map to be orientation
reversing in our denition of Heegaard splitting.
Example 3.6. We can also classify all genus 1 splittings. All such splittings are given by an
orientation reversing homeomorphism ) : T T, and the isotopy classes of these maps are in
bijective correspondence with integer matrices
=
_
:
j :
_
det = j: +: = 1
so j and are relatively prime or one is 0 and the other is 1. We use here because we required
our gluing homeomorphism to be orientation reversing, so the determinant of this matrix should be
1, which can be written simply as j: +: = 1 with our sign choice. This homeomorphism maps
the meridian j to the curve j + j , i.e. the meridian wraps times around the meridian
and j times around the longitude. Let ` be the manifold resulting from gluing to copies of the
solid torus via ). It is easy to see that ` depends only on the image of j under ) by using an
argument similar to theorem 3.2. Thus, ` is completely determined by the relatively prime pair
of integers (j, ), and we call this space the (j, ) Lens space and denote it by 1(j, ).
The Heegaard diagram of 1(j, ) is just a curve on the torus that wraps times around
the meridian and j times around the longitude. One can prove that 1(j, ) = 1(j, ) and
1(j, ) = 1(j, + nj) for any integer n, so we can assume j 0 and 0 j 1. But, if j = 0
then we are forced to have = 1, and if = 0 we are also forced to have j = 1. In these cases
we have 1(1, n) = o
3
for all integers n and 1(0, 1) = o
1
o
2
, so disregarding these examples
we can assume j 2 and 1 j 1. We also have 1(2, 1) = R1
3
.
Example 3.7. Heegaard diagrams can be quite complicated, for example see gure 3.8. We will
show later that this 3-manifold is simply connected.
38

2
Figure 3.8: Heegaard diagram of a simply connected 3-manifold
Let us reconcile the above theory of Heegaard splittings of 3-manifolds with our general theory of
handlebodies. A handlebody of genus p is a 0-handle 1-handles such that the result is orientable.
This is just the rst step in building a 3-dimensional handlebody, so let us denote it by A
1
. Next we
could attach some 2-handles (we certainly cannot attach a 3-handle yet, unless p = 0). Attaching a
2-handle is completely determined by the image of the attaching sphere 1
2
0 (which we will call
the attaching circle in this case) and the framing of the attaching circle, which are parameterized
by
1
(O(1)) = 0. Therefore, attaching the 2-handles is equivalent to specifying a collection of
embedded circles in A
1
, which is just a Heegaard diagram, and let A
2
be the result. If A
2
is a
collection of o
3
s, then there is a unique way to cap these o with 3-handles so that the result is
a closed 3-manifold. This is precisely the case we draw p embedded curves on A
1
such that the
complement of their union is connected, so such a picture uniquely determines a 3-manifold.
Now we see that a Heegaard splitting is simply a 3-dimensional handlebody, and a Heegaard
diagram is just a collection of curves drawn on A
1
, we would like to translate our machinery of
handle moves, handlebody homology and computation of fundamental groups into the Heegaard
setting. We start with handle moves. We x A
1
to be the standardly embedded handlebody of
genus p since all the information in a 3-dimensional handlebody is in the attaching circles of the
2-handles. In the handlebody theory we could isotopy the attaching maps, which just means that
we can isotopy the and curves on the surface of genus p. To create a canceling pair of 1-
and 2-handles simply increase the genus of A
1
by one (i.e. add a 1-handle) and add the attaching
circle that goes once around the longitude of the new 1-handle but does not venture into the old
part of A
1
. Attaching the 2- and 3-handles to this new Heegaard diagram will give the original
manifold connect summed with o
3
(recall the Heegaard diagram of o
3
, gure 3.7), hence it will be
dieomorphic to the original manifold.
Handle slides are a little more dicult to understand. Sliding the 1-handles is very simple, and
does not produce anything interesting since we required A
1
to be orientable, hence allowed only
one type of framing on the 1-handles. To get a feeling for sliding 2-handles, consider sliding two
handles that are attaching to 1
3
. Figure 3.9 shows how to do this in the obvious way. Here we
sliding the attaching circle of one handle through the disc 1
2
1 of the other handle. A better
way to look at this is just to look at the attaching circles. Then we see that sliding 2-handles is
equivalent to taking a parallel of one circle, and forming the banded sum of this circle with the
other attaching circle (see gure 3.10).
Now to slide a 2-handle /
1
across another 2-handle /
2
in a Heegaard diagram make a parallel
39
Figure 3.9: Sliding 3-dimensional 2-handles
Figure 3.10: Sliding 3-dimensional 2-handles
copy of the attaching circle
2
of /
2
and form the band sum of the attaching circle
1
of /
1
with
the parallel of /
2
. If we let denote the band sum of
1
and the parallel of
2
, then note that

1
,
2
, and bound a pair of pants in A
1
. Conversely, if there are two attaching circles
1
,
2
on
A
1
and another curve on A
1
such that does not intersect any other attaching curves and

1
,
2
, and bound a pair of pants, then we can discard any one of these three curves without
changing the resulting 3-manifold. See gure 3.11 for a non-trivial example of this.
Next we see how to compute the fundamental group of a 3-manifold A from one of its Heegaard
diagrams. Choose a basis r
1
, . . . , r
g
for the free group
1
(A
1
), and let :
i
be the word in r
1
, . . . , r
g
representing the i-th attaching circle. Then a presentation of
1
(A) is simply r
1
, . . . , r
g
: :
1
, . . . , :
g
).
Example 3.8. Consider the Heegaard diagram in gure 3.8. Let r
1
denote the standard generator
around the rst hole (counter-clockwise orientation) and r
2
the generator around the second hole
(counter-clockwise orientation). One can easily check that the curve
1
is represented by the
curve r
3
1
r
2
2
and
2
is represented by the curve r
4
1
r
3
2
. Therefore a presentation for the manifold
` associated to this Heegaard splitting is
1
(`) =

r
1
, r
2
: r
3
1
r
2
2
= r
4
1
r
3
2
= 1
_
. The rst relation
implies that r
3
1
r
3
2
= r
2
, and plugging this into the second relation gives r
1
r
2
= 1. Therefore r
1
and r
2
are inverses in
1
(`), hence 1 = r
3
1
r
2
2
= r
1
and r
2
= 1, and so ` is simply connected.
Now we compute the homology of a 3-manifold A from one of its Heegaard diagrams. We
clearly have H
0
(A) = H
3
(A) = Z, C
1
(A) = C
2
(A) = Z
g
(if A has p 1- and 2-handles). Further,

1
= 0 since the attaching spheres of the 1-handles intersect 1
3
at two points of opposite sign,
and
3
= 0 for similar reasons. Hence all the homology information of A is contained in
2
. Given
40
Figure 3.11: Sliding 3-dimensional 2-handles
a 2-handle /,
2
/ is a linear combination of the one handles /
i

g
i=1
, where the coecient of /
i
is
the intersection number of /s attaching circle with /
i
s belt sphere. The belt sphere of /
i
is just
the meridian that bounds a disc (the non-bolded curves in gure 3.6).
Example 3.9. If we label the left and right 1-handles of gure 3.8 by /
1
and /
2
respectively, and
label the 2-handle corresponding to the black attaching circle by /
1
and the other 2-handle by /
2
,
then a close examination of this diagram shows that

2
/
1
= 3/
1
+ 2/
2

2
/
2
= 4/
1
+ 3/
2
We could also see this by just abelianizing the relations of
1
. As a linear transformation H
2
(A) =
Z
2
Z
2
= H
1
(A) it is invertible since its determinant is +1, therefore
2
is an isomorphism. This
implies that H
1
(A) = 0 = H
2
(A), and so A is a homology sphere. In fact, A is homeomorphic to
o
3
, as can be checked by sliding handles.
3.6 Kirby Diagrams
A Kirby diagram is a diagram we can associate to a 4-dimensional handlebody A. We start with
a 4-ball 1
4
, and think of its boundary 1
4
= o
3
as R
3
. We attach 1-handles to this via
an embedding of 1
3
. 1
3
. There is the issue of framing (there are two framings), but for now let
us choose the framing such that A
1
= ;:o
1
1
3
(here we are attaching : 1-handles). We can
represent this in R
3
by drawing 2: disjoint embedded 3-balls in R
3
and pairing them
so that when we identify pairs in an orientation reversing manner we get A
1
= ;:o
1
1
3
.
To attaching a 2-handle to A
1
we specify a knot in A
1
(the attaching circle) and label it with
a framing. So, a diagram of A
2
consists of : pairs of 3-balls and a collection of arcs that either
start and end on the boundary of the 3-balls, or form a closed loop. We can specify a framing on
a knot component by a nowhere zero normal vector eld, or equivalently, a parallel knot. Once we
specify which parallel corresponds to 0
1
(O(2)) = Z we can get any other framing by adding
full left or right twists. We will choose the parallel which has linking number 0 with the knot to
be the framing corresponding to 0.
It turns out that there is essentially a unique way to attach 3-handles and 4-handles so that we
obtain a closed manifold, so our diagram of A can stop here. Things are more complicated when
A ,= , but we will not deal with that now.
(todo: need more on this...)
Example 3.10. Let us nd a Kirby diagram of o
2
o
2
. The 2-sphere has the obvious decomposition
o
2
= 1

1
+
, where 1

is a 0-handle, 1
+
a 2-handle, which we think of as the southern and
northern hemispheres. The attaching map for the 2-handle is just the identity. The handlebody
decomposition of o
2
o
2
is just
o
2
o
2
=
0handle
..
1

2handle
..
1
+
1

2handle
..
1

1
+

4handle
..
1
+
1
+
41
First we determine the attaching maps for the 2-handles. One is the map
1
: 1
+
1


(1

) dened by
1
(r, j) = (r, j), and the other is the map
2
:
1
2
1
0
1
0
1
2
UNFINISHED
Example 3.11. Let us nd a handlebody decomposition of o
2

o
2
. Recall that this is the boundary
of the non-trivial 1
3
bundle over o
2
.
42
4 Morse Theory
4.1 Morse Functions
We start with a smooth n-manifold ` and a smooth function ) : ` R. Take a chart (r, l),
that is an open set l ` and a homeomorphism r : l R
n
. If we write r in components
r = (r
1
, . . . , r
n
), then we dene the derivative of ) at some point j l with respect to this chart
by
)
r
i
(j) = 1
i
() r
1
)(r(j))
where 1
i
is the regular derivative of ) r
1
: R
n
R in the i-th component. Recall that if (j, \ )
is another chart, then the derivative of ) with respect to these two charts on the overlap are related
by
)
j
i
=
n

j=1
)
r
j
r
j
j
i
(4.1)
Another way of saying this is to rst dene the Jacobian matrix of two charts (r, l) and (j, \ ).
This is the matrix whose (i, ,)-entry is equal to
x
i
y
j
, and it is denoted by J
x
y
. If we let
f
x
denote
the column vector of derivatives of ), then (4.1) simply says
)
j
= J
x
y

)
r
Since
f
x
i
is just another smooth map from l to R we can take a second derivative with respect to
a chart. The second derivatives calculated with respect to dierent charts are related by

2
)
j
i
j
j
=

2
)
r
k
r

r
k
j
i
r

j
j
(4.2)
As with rst derivatives there is another way of looking at this. Let us dene the Hessian of )
with respect to the chart r be the matrix whose (i, ,)-entry is

2
f
x
i
x
j
, and it is denoted by H
x
f
.
Since partial derivatives commute we have that H
x
f
is symmetric. We now see that (4.2) can be
written as
H
y
f
= (J
x
y
)
t
H
x
f
J
x
y
(4.3)
This tells us that even though the Hessian as a matrix is not well-dened (i.e. it depends on the
chart), it is well-dened as a bilinear form. There is also a nice coordinate free way of describing
the Hessian. Let : (T`) (T

`) (T

`) be any connection on T

`, and for any j `


dene the map H
f
(j) : T
p
` T

p
` by
H
f
(j)(A) =
X
(d)) (4.4)
This does not depend on the choice of connection, for if

is another connection, then the dierence


is tensorial in both arguments and d)
p
= 0, hence
X
(d))

X
(d)) = 0 at j.
Suppose that a point j ` is a critical point of ), i.e. d)
p
= 0. We say that j is a non-
degenerate critical point if det H
x
f
(j) is nonzero for some chart r, hence every chart by (4.3).
We dene the index of a critical point j to be the number of negative eigenvalues of the Hessian
at j, which is well-dened by (4.3), and is denoted by j(j). Equivalently, we call the critical
point j non-degenerate if H
f
(j) does not have 0 for an eigenvalue, and j(j) is the number of
43
negative eigenvalues of H
f
(j). Finally, we dene a Morse function is a smooth function with all
non-degenerate critical points.
It turns out that Morse functions look very nice around critical points. The Morse lemma gives
a precise statement of this, and will be very important in our discussion of cell decompositions
of spaces via Morse functions. First we recall a basic fact from calculus using the fundamental
theorem.
Lemma 4.1. Let l be a convex subset of R
n
containing 0 and ) : l R a smooth function with
)(0) = 0. Then we can write
)(r
1
, . . . , r
n
) =
n

i=1
r
i
p
i
(r
1
, . . . , r
n
)
for some p
i
: l R with p
i
(0) =
f
x
i
(0).
Proof. For xed r
1
, . . . , r
n
we apply the fundamental theorem of calculus and chain rule to get
)(r
1
, . . . , r
n
) =
_
1
0
d)(tr
1
, . . . , tr
n
)
dt
dt =
_
1
0
n

i=1
r
i

)
r
i
(tr
1
, . . . , tr
n
) dt
So let p
i
(r
1
, . . . , r
n
) =
_
1
0
f
x
i
(tr
1
, . . . , tr
n
) dt, then clearly p
i
(0) =
f
x
i
(0) and we are done.
Lemma 4.2 (Morse Lemma). Let ) : ` R be a smooth function and j ` a non-degenerate
critical point. Then there is a chart j = (j
1
, . . . , j
n
) on a neighborhood l of j such that j
i
(j) = 0
and
) = )(j) (j
1
)
2
(j

)
2
+ (j
+1
)
2
+ + (j
n
)
2
on l, where j is the index of j.
Proof. First note that nding a chart so that ) can be expressed in the above form is equivalent
to nding a chart (r, l) such that on l we have
() r
1
)(t
1
, . . . , t
n
) = t
2
1
t
2

+t
2
+1
+ t
2
n
where t
i
are real numbers. Therefore we can, without loss of generality, assume that ) is dened on
some open set \ of R
n
(say containing zero), and by performing some translations we can assume
j = 0 and )(0) = 0.
Next it is clear that if ) can be represented in the above form for some chart (j, l), then we
have

2
)
j
i
j
j
(j) =
_
2 i = ,, , j
2 i = ,, , j
Therefore the Hessian H
x
f
is already diagonalized in this coordinate system, and clearly has j
negative entries along the diagonal. Therefore j really will be the index of j if we can get ) into
the above form.
Let r = (r
1
, . . . , r
n
) be the standard coordinates on R
n
. By lemma 4.1 we can nd functions
p
i
: \ R such that
)(r
1
, . . . , r
n
) =
n

i=1
r
i
p
i
(r
1
, . . . , r
n
)
44
and p
i
(0) =
f
x
i
(0). However, since j = 0 is assumed to be a critical point we have
f
x
i
(0) = 0, so
we can apply lemma 4.1 to the functions p
i
to obtain functions /
ij
: \ R such that
p
i
(r
1
, . . . , r
n
) =
n

j=1
r
j
/
ij
(r
1
, . . . , r
n
)
and /
ij
(0) =
g
i
x
j
(0) =

2
f
x
i
x
j
(0). Combining the above two expressions yields
)(r
1
, . . . , r
n
) =
n

i,j=1
r
i
r
j
/
ij
(r
1
, . . . , r
n
) (4.5)
If we dene

/
ij
=
1
2
(/
ij
+ /
ji
), then we see that

/
ij
=

/
ji
and yet still ) =

r
i
r
j

/
ij
. Therefore
we can assume that /
ij
= /
ji
. It is also clear that the matrix (/
ij
(0))
ij
is equal to
1
2
H
x
f
.
We now simply mimic the proof of the diagonalization of quadratic forms, applied to (/
ij
), to
show that we can nd coordinates to bring ) into the desired form. By assumption the Hessian is
non-singular, so (/
ij
(0)) is also non-singular. Reordering the coordinate system can assume that
/
11
(0) ,= 0, and so by continuity we have that /
11
(0) ,= 0 on some small neighborhood \

of 0.
Let us also assume that /
11
0 on \

(the other case is treated similarly). We introduce a new


coordinate j
1
: \

R dened by
j
1
=
_
/
11
_
r
1
+
n

i=2
r
i
/
1i
/
11
_
(4.6)
The coordinates (j
1
, r
2
, . . . , r
n
) form a chart (in a possibly smaller neighborhood \

of 0). Squaring
this new coordinate gives
(j
1
)
2
= /
11
_
r
1
+
n

i=2
r
i
/
1i
/
11
_
2
= /
11
_
_
(r
1
)
2
+ 2
n

i=2
r
1
r
i
/
1i
/
11
+
_
n

i=2
r
i
/
1i
/
11
_
2
_
_
= /
11
(r
1
)
2
+ 2
n

i=2
r
1
r
i
/
1i
+
1
/
11
_
n

i=2
r
i
/
1i
_
2
(4.7)
This expression allows us to rewrite (r
1
)
2
as
/
11
(r
1
)
2
= (j
1
)
2
2
n

i=2
r
1
r
i
/
1i

1
/
11
_
n

i=2
r
i
/
1i
_
2
(4.8)
45
Now, using (4.5) and (4.8) we can rewrite ) in coordinates (j
1
, r
2
, . . . , r
n
) as
) = /
11
(r
1
)
2
+ 2
n

i=2
r
1
r
i
/
1i
+
n

i,j=2
r
i
r
j
/
ij
= (j
1
)
2
2
n

i=2
r
1
r
i
/
1i

1
/
11
_
n

i=2
r
i
/
1i
_
2
+ 2
n

i=2
r
1
r
i
/
1i
+
n

i,j=2
r
i
r
j
/
ij
= (j
1
)
2
+
n

i,j=2
r
i
r
j
/
ij

1
/
11
_
n

i=2
r
i
/
1i
_
2
Note that had we assumed /
11
< 0 on \

, then we would have (j


1
)
2
in the above. The terms in
this last expression only involve r
i
s for i 2, so we can repeat the above arguments to nd a new
coordinate j
2
such that (j
1
, j
2
, r
3
, . . . , r
n
) is a coordinate system and
) = (j
1
)
2
(j
2
)
2
terms with r
3
, r
4
, . . . , r
n
Continuing this we get a coordinate system j = (j
1
, . . . , j
n
) such that
) = (j
1
)
2
+ (j
n
)
2
Since the number of minus signs is equal to j, which does not depend on coordinate charts, we are
done.
A simple consequence of the Morse lemma is the following.
Corollary 4.1. The critical points of a Morse function are isolated.
4.2 Handlebody Decompositions Induced by Morse Functions
For a manifold ` with Morse function ) : ` R, let us dene the sets `
a
= )
1
(, o]. If o is
not a critical value, then `
a
is a submanifold with boundary. We will now analyze what happens
to `
a
as o crosses a critical value. Our rst theorem towards this says that the dieomorphism
type of `
a
does not change between critical values. First let us develop the notion of gradient
vector elds and ows.
Let (`, p) be a Riemannian manifold with Morse function ) : ` R. Recall that the metric
gives us a canonical isomorphism of tangent spaces with cotangent spaces by mapping a vector
A T
p
` to the covector p(A, ) T

p
`. The gradient vector eld of ) is the unique vector eld

g
) corresponding to the 1-form d) under this isomorphism, hence
p(A,
g
)) = d)(A)
for any vector eld A. In coordinates (r
1
, . . . , r
n
) we can write

g
) =
n

i,j=2
p
ij
)
r
j

r
i
where p
ij
are the components of the inverse metric. It is clear that
g
) vanishes at precisely the
critical points of ).
46
Theorem 4.1. Let [o, /] be an interval that does not contain any critical values and such that
)
1
[o, /] is compact. Then `
a
is dieomorphic to `
b
. In fact, `
a
is a deformation retract of
`
b
.
Proof. First note that the compactness hypothesis would be immediate if we assumed that ` is
compact. Let A be a vector eld given by
A =

g
)
[[
g
)[[
2
on )
1
[o, /], and vanish identically outside some compact set containing )
1
[o, /]. Since this vector
eld has compact support it generates a unique 1-parameter group
t
: ` ` of dieomorphisms.
For xed j ` such that
t
(j) lies in )
1
[o, /], the function ) (j) is a map R R, and we can
compute (using the chain rule and denition of
g
))
d)(
t
(j))
dt
= d)(d
t
(d,dt)) = p(d
t
(d,dt),
g
)) =
p(
g
),
g
))
p(
g
),
g
))
= 1
So, the function ) (j) : R 1 is linear; in fact, )(
t
(j)) = t + )(
0
(j)) = t + )(j). So, if we
take a point j with )(j) = o, then )(
ba
(j)) = (/ o) + o = /, hence
ba
maps `
a
into `
b
.
Similarly,
1
ba
=
ab
maps `
b
into `
a
, hence `
b
is dieomorphic to `
a
, and this proves the
rst part of the theorem.
For the second part of the theorem we can construct the deformation retract explicitly. Consider
the homotopy :
t
: `
b
`
b
dened by
:
t
(r) =
_
r )(r) o

t(af(x))
(r) o )(r) /
We clearly have :
0
= id and :
1
is a retract of `
b
onto `
a
, therefore `
a
is a deformation retract
of `
b
.
Theorem 4.2. Let ) : `
n
R be a smooth function with a non-degenerate critical point j of
index j. Let c = )(j) and suppose )
1
[c , c + ] is compact, and contains no critical points of
) other than j. Then for small enough , the manifold `
c+
is dieomorphic to `
c
with an
n-dimensional j-handle attached.
Proof. See gure 4.1 for the local behavior of ) around the critical point j. Take a chart (r, l)
around j as in the Morse lemma. Then the lightly shaded region is the set of points (r
1
, . . . , r
n
)
with )(r
1
, . . . , r
n
) c , and the medium shaded region is the set of points with c
)(r
1
, . . . , r
n
) c + . The intersection of the medium shaded region with the (r
1
, . . . , r

)-axis is
the set of points (r
1
, . . . , r

) with (r
1
)
2
+ (r

)
2
, hence a j-dimensional ball. The intersection
of the medium shaded region with the (r
+1
, . . . , r
n
)-axis is the set of points (r
+1
, . . . , r
n
) with
(r
+1
)
2
+ + (r
n
)
2
, hence a (n j)-dimensional ball.
The dark region / in the gure consists of the set of points (r
1
, . . . , r
n
) such that
(r
1
)
2
+ + (r

)
2
(r
+1
)
2
(r
n
)
2
and (r
+1
)
2
+ + (r
n
)
2

for suciently small . Clearly this region is dieomorphic to 1

1
n
, and intersects `
c
at
the points 1

1
n
, i.e.
(r
1
)
2
+ + (r

)
2
(r
+1
)
2
(r
n
)
2
= and (r
+1
)
2
+ + (r
n
)
2

47
x
+1
, . . . , x
n
x
1
, . . . , x

f c
f c +
Figure 4.1: Behavior of ) around a critical point
One can now use the gradient ow of ) to show that `
c+
is dieomorphic to `
c
/, hence a
j-handle was attached.
Theorem 4.2 and gure 4.1 also give an explicit description of the attaching spheres of the
handle attached. If the handle / is attached to `
c
to obtain `
c+
, then the attaching sphere of
/ is precisely where the descending manifold of j intersects the boundary of `
c
.
Theorem 4.3. If ) is a Morse function on a manifold `, and if each `
a
is compact, then `
has a handlebody decomposition with one j-handle for each critical point of index j.
4.3 Properties of Gradient Flows
Let us further investigate the properties of ows of gradient vector elds. For this section all of our
manifolds will be closed.
Proposition 4.1. Every function decreases along its negative gradient ow lines.
Proof. Let ) : ` R be C
1
and n : R ` a ow line of
g
). Since ) n is C
1
we have, by
48
the fundamental theorem of calculus, that for any / o
)(n(o)) )(n(/)) =
_
a
b
d
dt
) n(t) dt
=
_
b
a
d)
u(t)
(dn,dt) dt
=
_
b
a
d)
u(t)
(
g
)(n(t))) dt
=
_
b
a
p(
g
)(n(t)),
g
)(n(t))) dt
=
_
b
a
[[
g
)(n(t))[[ dt
0
Therefore )(n(o)) )(n(/)), and so ) decreased along n.
Proposition 4.2. Letting n be a ow line of ), if n() = j or n() = j, then j is a critical
point of ).
Proof. First suppose n() = j (the other case is similar). We have
lim
t
_
t
0
[[dn,dt(:)[[
2
d: = lim
t
()(n(0)) )(n(t))) = )(n(0)) )(j) <
This forces [[dn,dt(t)[[ 0 as t . Therefore
[[
g
)(j)[[ = lim
t
[[dn,dt(t)[[ = 0
and so j is a critical point of ).
Proposition 4.3. If j is a non-degenerate critical point of ), and n is a negative gradient ow
line with n() = j, then there are constants C 0, c 0 and T 0 such that
d(n(t), j) Cc
t
for all t T.
4.4 Existence of Morse Functions
We have proved quite a bit concerning Morse functions, but we have not yet shown that they always
exist. It turns out that not only do they exist, but in some sense, almost all smooth functions satisfy
the Morse property. We will prove this result and show that Morse functions can be chosen with
special properties, such as with separated critical points and self-indexing critical points.
Since most theorems that involve the phrase almost every invoke Sards theorem in some way,
we begin with this result. Recall that a smooth function ) : ` is said to have a critical
point at j ` if rank )
p
< dim, and a point is said to be a critical value of ) if
= )(j) for some critical point j `.
Theorem 4.4 (Sards Theorem). Let ) : R
n
R
m
be a smooth function. The critical values of )
have Lebesgue measure zero in R
m
.
49
Sards theorem gives us an easy proof that Morse functions are numerous on open sets of R
n
.
Proposition 4.4. Let ) : l R be a smooth function on an open set l R
n
. Then there exists
numbers o
1
, . . . , o
n
such that

)(r
1
, . . . , r
n
) = )(r
1
, . . . , r
n
) o
1
r
1
o
n
r
n
has all non-degenerate critical points. Further, the o
i
s can be chosen arbitrary small.
Proof. Dene a function / : l R
n
by
/ =
_
_
_
/
1
.
.
.
/
n
_
_
_ =
_
_
_
f
x
1
.
.
.
f
x
n
_
_
_
Then the Jacobian of / is
J
h
=
_
/
i
r
j
_
=
_

2
)
r
i
r
j
_
which is just the Hessian H
f
of ). So, the critical points of / are precisely where det H
f
= 0. By
theorem 4.4 we can nd a point o = (o
1
, . . . , o
n
) R
n
that is not a critical value of /. In fact,
there are many such points, and we can choose this point as close to 0 as we want. For this point,
dene

) : l R by

)(r
1
, . . . , r
n
) = )(r
1
, . . . , r
n
) o
1
r
1
o
n
r
n
Let j l be a critical point of

), then
0 =

)
r
i
(j) =
)
r
i
(j) o
i
hence /(j) = o. Since o is not a critical value of / we have that j is not a critical point of /, and
so det H
f
,= 0. Since ) and

) dier only by a linear function we have that their Hessians are equal,
so det H
e
f
= det H
f
,= 0, hence j is a non-degenerate critical point of

).
We can use this proposition to prove the more general result
Theorem 4.5. There exists a Morse function on any closed manifold `.
Proof. Start with any smooth function )
0
: ` R. We will make a sequence of small changes to
this function to remove degenerate critical points over open sets.
Let l
i
be a nite open cover of ` by charts, and let
i
: l
i
R be a collection of bump
functions for l
i
. Recall that this means that there is a compact set 1
i
l
i
such that
i
1 on
1
i
, 0 1 on l
i
and
i
0 outside l
i
. Without loss of generality we can assume that 1
i

covers `. Let /
m
= 1
1
1
m
and set 1
0
= .
We will use an induction argument to remove degenerate critical points from /
m
. Clearly )
0
has no degenerate critical points in /
0
, so let us assume that )
m1
has been constructed with
no degenerate critical points in /
m1
. Let r : l
m
R
n
be a chart for l
m
, then )
m1
r
1
:
r(l
m
) R is a function dened on an open set of R
n
. By proposition 4.4 we can nd small vector
o = (o
1
, . . . , o
n
) such that
)
m1
r
1
(r
1
, . . . , r
n
) o
1
r
1
o
n
r
n
50
has only non-degenerate critical points. We can extend this function to all of ` by multiplying by
the bump function, so we dene
)
m
(j) =
_
)
m1
(j) o r(j)
m
(j) j l
m
)
m1
(j) j , supp
m
This is a well-dened smooth function on ` since
m
0 on the overlap of l
m
and the complement
of supp
m
. Now )
m
does not have any degenerate critical points on 1
m
, and since )
m1
did not
have any degenerate critical points on /
m1
it follows that )
m
is a Morse function on /
m
. This
completes the proof by induction.
The proof of theorem 4.5 actually shows a little more. Since we could pick the numbers o
1
, . . . , o
n
very small at each stage we could actually show that the constructed Morse function is C
2
-close to
the initial function )
0
, hence there is a Morse function close to any smooth function.
(note: Other theorems to prove)
Theorem 4.6. There is a Morse function ) : ` R on any closed manifold ` such that if
j, ` are critical points, then )(j) )() if and only if j(j) j().
Such Morse functions are called self-indexing, and clearly the image of ` under ) is precisely
[0, n], where n is the dimension of the manifold.
51
5 Kirby Calculus
5.1 Surgery on Links
We have seen that Heegaard diagrams provide a nice combinatorial picture for 3-manifolds, even if
they are a little unwieldy to work with. This is mostly due to the fact that self-homeomorphisms
of high genus surfaces are dicult to understand. We will now present another combinatorial
method of constructing 3-manifolds, called surgery on links. It has the advantage that we now only
need to understand self-homeomorphisms of the torus, but the disadvantage that we now need to
understand links better.
Given a 3-manifold `, let / be an embedded knot in ` and (/) a tubular neighborhood,
which is homeomorphic to 1
2
o
1
. Surgery on this knot is the proccess of cutting out the tubular
neighborhood and gluing it back in. Since `(/) has boundary homeomorphic to o
1
o
1
we
can nd a homeomorphism : (1
2
o
1
) (`(/)) and form the quotient space `
k,
=
(`(/))

(1
2
o
1
), which we call the Dehn surgery on / via , or simply surgery on
/ since we are not going to discuss more general surgeries. The construction generalizes to links
by performing it on each link component. Every closed, orientable 3-manifold can be obtained by
such a construction applied to ` = o
3
, as we will see later.
Proposition 5.1. A surgery `
k,
is completely determined by the image of 1
2
under .
Proof. The solid torus 1
2
o
1
can be attached in two steps. Let J be a small segment of o
1
containing . Then we can start by attaching 1
2
J to `(/). This leaves 1
2
J
c
to be
attached along its boundary, which is just a 3-ball being attached along a 2-sphere. There is only
one way to attach this piece since every orientation preserving homeomorphism of o
2
is isotopic to
the identity.
So now a surgery can be described by a knot / and a curve on (`(/)), which is where the
meridian will be mapped under the attaching map. We can come up with a more combinatorial
description of this curve, or more specically, closed curves on the torus. Let and be the
meridian and longitude of the torus T
2
= o
1
o
1
so that
1
(T
2
) = Z Z is generated by []
and []. For j, integers let = j + denote a curve in the homotopy class j[] + [], i.e.
wraps j times around the meridian and times around the longitude. We have the following
characterization of these curves.
Proposition 5.2.
1. If a curve in the homotopy class j[] + [] has no self-intersections, then either j and
are relatively prime, or one of them is 0 and the other 1.
2. If two closed curves on the torus are homotopic and without self-intersections, then they are
isotopic.
Therefore isotopy classes of curves on T
2
are in one-to-one correspondence with pairs of relatively
prime integers.
Let us apply this to a knot / in o
3
. Let j be a canonical meridian of (/), and let / be the
canonical longitude of (/). Recall that this means that lk(/, /) = 0. Choose orientations on j and
/ such that the ordered basis (j, /, n) matches the orientation of o
3
, where n is the outward pointing
normal to (/). If we glue a solid torus 1
2
o
1
into o
3
(/) via : 1
2
o
1
(`(/)),
then we know must map the meridian 1
2
to a curve jj + /, for some relatively prime
52
integers j, (or possibly one equal to 0 and the other 1). Therefore a surgery can now be described
simply by a knot and an extended rational number j, Q . This is also called rational
surgery. If = 1, then the surgery is called an integral surgery. We call the rational number
j, the framing index of /, or simply the framing. This discussion applies equally well to knots
embedded in an integral homology sphere.
Here is a simple proposition relating surgeries to Heegaard splittings, as well as a powerful
corollary.
Proposition 5.3. Let )
1
, )
2
: H H

be homeomorphisms of the surfaces of two handlebodies


such that )
1
= )
2

c
, where
c
is a Dehn twist along a simple closed curve c in H. Set `
i
=
H
f
i
H

for i = 1, 2. Then `
2
is obtained from `
1
by an integral surgery along a knot / `
1
isotopic to c.
Proof. Let / be a knot in H obtained from pushing c a little H. Connect a tubular neighborhood
(/) of / to H via an annulus

= o
1
1 such that o
1
0 corresponds to a longitude /

on
(/) and o
1
1 corresponds to c. There is a self-homeomorphism of H(/) that is the
result of cutting H(/) along , performing a full twist on one of the rims, and then gluing the
rims back together. Clearly is just an extension of
c
to H(/), but also its restricted to the
torus (/) is a Dehn twist around /

. Let `

i
= (H(/))
f
i
H

, i.e. `

i
is `
i
with a solid
torus removed. Then `

1
and `

2
are homeomorphic via the map
/(r) =
_
(r) r H(/)
r r H

The fact that `


1
and `
2
become homeomorphic after removing a solid torus means that `
1
is
obtained from `
2
via a surgery on /. In particular, the meridian j of (/) is wrapped around
j +/

, hence it is an integral surgery.


Theorem 5.1 (Lickorish-Wallace). Every closed, orientable 3-manifold ` can be obtained from
an integral surgery on a link embedded in o
3
.
Proof. By theorem 3.1 we know that ` = H
f
2
H

, where H, H

are handlebodies and )


2
is an
orientation reversing homeomorphism of their boundaries. Similarly o
3
= H
f
1
H

. Then )
1
2
)
1
is an orientation preserving homomeomorphism of a surface of genus p, hence by the Dehn-Lickorish
theorem ((todo: do this theorem)) we have )
1
= )
2

c
1

cn
, where
c
i
is a Dehn twist around
a simple, closed curve c
i
. Therefore by proposition 5.3 we have that ` is a sequence of integral
surgeries on knots in o
3
, or equivalently an integral surgery on a link.
A surgery diagram for o
3
is a link diagram with each component labeled with an element of
Q. The canonical longitude of a knot / can be drawn in a diagram by just drawing a parallel
copy of the knot next to / such that its linking number with / is 0. However, take note that the
rst Reidemeister move will change this canonical longitude, so one needs to be careful. We will
now explain the subtleties of this situation.
Consider the case of integral surgeries on knots in o
3
, so let / be a knot with integral framing
n. This framing uniquely determines (up to isotopy) a longitude / of / such that lk(/, /), which
is the curve we glue the meridian to when we perform the surgery. What is the meaning of the
Reidemeister moves in a surgery diagram? It is clear that performing 1
2
and 1
3
moves does
not change the manifold resulting from performing surgery on the knot, so we will continue calling
these moves 1
1
and 1
2
on surgery diagrams. However, performing an 1
1
move makes the longitude
53
Figure 5.1: Belt Trick
Figure 5.2: Ribbon Reidemeister I
wrap one more time around the knot. This is the so-called belt trick, and is shown in gure 5.1.
This means that the framing of the knot will need to change by 1, depending on which way you
twist. This move on surgery diagrams will be called 1

1
due to its equivalent interpretation as a
Reidemeister move for ribbons, and is shown in gure 5.2. The inverse of the rst 1

1
is called
adding a left twist, and the inverse of the other 1

1
move is called adding a right twist. Now two
surgery diagrams related by a nite number of 1

1
, 1
2
and 1
3
moves give homeomorphic manifolds,
but does the converse hold? Unfortunately no, but there are two more moves one can add, called
Kirby moves, for which the converse holds. This is discussed in section 5.2.
Example 5.1. The 1,0 = surgery on the unknot in o
3
results in o
3
again since we are just
cutting out an unknotted, solid torus and gluing it back in the same way. Performing a 0-surgery
on the unknot means to cut out an unknotted, solid torus H from o
3
and glue it back in such that
the meridian wraps once around the longitude. However, this longitude is precisely the meridian of
the the solid torus o
3
H, so this surgery is really just gluing two solid tori together so that their
meridians are identied, hence this is o
1
o
2
.
More generally, a surgery on the unknot in o
3
with surgery coecient j, (j and relatively
prime or one equal to 0 and the other 1) results in 1(j, ). This is because we cut out an unknotted,
solid torus H from o
3
and glue it back it so that the meridian wraps j times around the longitude
and times around the meridian. However, Hs meridian is precisely o
3
Hs longitude, and Hs
longitude is precisely o
3
Hs meridian, hence we are just gluing together two solid tori so that
the meridian of one gets wrapped j times around the longitude and times around the meridian
of the other torus. This is precisely the Heegaard splitting description of 1(j, ). Compare this
with example 3.6. We can also nd a surgery diagram for 1(j, ) that has only integral surgery
coecients. See gure 5.3.
Let /
i
, 1 i n, denote the components of a surgery diagram of some link, let /
i
be framed
by j
i
,
i
and let ` be the 3-manifold resulting from performing this surgery. If we let 1 =
o
3
((/
1
) (/
n
)) be the knot complement, then H
1
(1) = Z
n
and is generated by the
54
Figure 5.3: Surgery diagrams for o
1
o
2
, 1(j, ) and 1(j 1, )
meridians j
i
of the link. When we glue in a solid torus o
1
1
2
so that 1
2
is wrapped j
i
times around the meridian of /
i
and
i
times around the longitude, when add a relation to H
1
(1).
Namely, we add the relation j
i
j
i
+
i

i
= 0, where
i
is the canonical longitude of /
i
, since we
are essentially lling this curve with a disc. The longitudes can be expressed as a relation in terms
of the j
i
s. Since lk(/
i
,
i
) = 0 we have that
i
bounds a surface 1
i
(i.e. a Seifert surface) in
o
3
(/
i
) (but not necessarily in 1). Cut a small disc out of 1
i
at each intersection point of a
dierent component /
j
with 1
i
, and orient it positively if the intersection is positive and negatively
if the intersection is negative. Since /
j
and 1
i
intersect lk(/
i
, /
j
) times (counted with sign), we
have
i
=

j=i
lk(/
i
, /
j
)j
j
. We have proved the following.
Proposition 5.4. For a surgery diagram with components /
1
, . . . , /
n
and framings j
1
,
1
, . . . , j
n
,
n
,
the rst homology of this surgery is the free abelian group on n generators modulo the relations
j
i
j
i
+
i

i=j
lk(/
i
, /
j
)j
j
= 0.
Corollary 5.1. For a surgery diagram with components /
1
, . . . , /
n
and integral framings j
1
, . . . , j
n
,
let be the linking matrix of this link with respect to any orientation. Then the rst homology of
this surgery is isomorphic to Z
n
, im.
Corollary 5.2. If the linking matrix has det = 1, then the surgery is a homology sphere. In
particular, a surgery on any knot with framing 1 is a homology sphere.
5.2 Kirby Moves
We now describe two basic moves that can be performed on surgery diagrams that do not change
the resulting surgered manifold. If a link 1 can be written as 1 = 1
1
1
2
, where 1
1
and 1
2
are
links that can be separated by a ball, then clearly a surgery on 1 is dieomorphic to the connected
sum of the surgeries on each 1
1
and 1
2
. So, if we add an unknotted, unlinked component to any
link 1 with framing 1, surgery on this new link is the connected sum of the surgery on 1 and the
surgery on the unknot with framing 1, which is just o
3
, hence the manifold is unchanged. The
addition or removal of an unknotted, unlinked component with framing 1 from a surgery diagram
is called the rst Kirby move (1
1
).
Next, suppose our link 1 is expressed as 1 = /
1
/
2
, where /
1
and /
2
are knots with framing
n
1
and n
2
, and are not necessarily unlinked. Let / be a longitude of /
2
dening its framing (i.e.
lk(/
2
, /) = n
2
), and let /

1
be the connected sum of /
1
with / along a band disjoint from 1 and /,
and let 1

= /

1
/
2
. We then give /

1
the framing n
1
+n
2
+2 lk(/
1
, /
2
), where the linking number
is computed with respect to the orientations on /
1
and /
2
that induce an orientation on /

1
(there
are only two, and both have the same linking number). Then surgery on 1 gives the same manifold
as surgery on 1

, but we will wait to show this until later. This process of adding the component
/
1
to /
2
, sometimes also called sliding /
1
across /
2
, is the second Kirby move (1
2
). Note that
55
Figure 5.4: Separating an unknotted component from a strand
the link 1

will of course depend on the band we choose in forming /

1
. Amazingly these two moves
are enough to get between any two surgery diagrams of the same manifold.
Theorem 5.2 (Kirby). Two manifolds presented as a surgery on some links are orientation pre-
servingly homeomorphic if and only if one link can be obtained from the other via a nite sequence
of 1
1
and 1
2
moves.
We shall now develop many secondary moves from 1
1
and 1
2
that are easier to apply directly.
Example 5.2. Suppose there is an unknotted component of 1 with framing 1 that is linked once
with a single strand from another component of 1 with framing n. We claim that we can separate
the unknotted component from the rest of the link, in which case the framing of the linked stranded
changes by 1. See gure 5.4 for the case of the unknot framed by 1, where the bolded curves are
the components of the link and the non-bolded curve is the longitude of the unknot determining its
framing. Note that the linking number of the vertical strand with the unknotted component is +1,
so the framing of the vertical strand changes to n 1 + 2 1 = n + 1. The framing of this strand
does not change from the 2nd frame to the 3rd because the two 1

1
moves performed cancel each
other out. The case where the unknot is framed by +1 is similar.
Example 5.3. Suppose now there is an unknotted component of 1 with framing 1 that is linked
once with two strands from other components of 1. All the steps are shown in gure 5.5. We can
see that separating this unknotted component resulted in a full right twist in the two strands. If
the two strands belong to two dierent link components, then each of their framings will increase
by 1, and if they belong to the same component, the framing will increase by 2.
Example 5.4. The general picture of the previous two examples is that if we can nd an unknotted
component of 1 with framing 1 that is linked once with / strands, then we can separate the unknot
from these strands by adding a full left/right twist in the strands, and changing their framings
accordingly. This type of move (see gure 5.6) is called the Fenn-Rourke move. There is another
name for this operation due to its relation to the 4-manifold theory of Kirby calculus. If we use
the rst Kirby move to introduce an unknotted, unlinked component with framing 1, we say that
we have performed a blow-up operation. Conversely, if we nd an unknotted component with
framing 1 that links some strands once, then the operation of sliding the unknotted component
o the strands and using 1
1
to discard it is called the blow-down operation.
Example 5.5. Suppose our link 1 can be written as 1 =
n
i=0
1
i
such that 1
0
is the unknot with
framing 0 and links with only 1
1
once (the other 1
i
s are free to link with 1
1
in any way). We claim
that we can use 1
2
moves to change any crossing of some component 1
i
(i 1) passing under
1
1
to an over crossing. By changing all such crossings to over crossings we would have completely
separated 1
0
1
1
from all of
n
i=2
1
i
, and we would have completely unknotted 1
1
.
56
Figure 5.5: Separating an unknotted component from two strands
Figure 5.6: Separating an unknotted component from many strands
Suppose a strand of 1
i
crosses underneath 1
1
. We can isotopy 1
0
along 1
1
so that it is close to
this crossing (in a diagram we only need 1
1
and 1
2
moves). Sliding 1
i
over 1
0
and then applying
1
1
and 1
2
allows us to change this crossing to an over crossing (see gure 5.7). If i 1, then the
framing of 1
i
does change from this Kirby move since 1
0
has framing 0 and 1
i
is not not linked
with 1
0
. However, if i = 1, then the framing of 1
1
is changed to n
1
+ 2 lk(1
0
, 1
1
), where n
1
is the
framing of 1
1
.
Once all the crossings have been changed so that 1
0
1
1
is separated from the rest of the link
and 1
1
is unknotted to become 1

1
, we are left to deal with a Hopf link 1
0
1

1
with one component
framed by 0 and the other by some integer, say j. By sliding 1

1
over 1
0
like we did above we see
that the framing of 1

1
changes by 2, so doing this repeatedly we can reduce this to the case when
the Hopf link is framed by 0 and 1, or 0 and 0. Performing surgery on either of these links just
gives o
3
, so we can remove them entirely. In the end, we have separated 1
0
1
1
from the rest of
1 and cancelled it.
Example 5.6. There is a simple way of expressing a surgery on the connected sum of two knots
/
1
#/
2
and the surgery on each knot separately. Suppose /
1
and /
2
have framings j and respec-
tively, and are linked once by a 0-framed unknot /
0
. If we slide /
1
over /
2
to get /

1
, then we see
that /
0
now only links /
2
. So, applying the move in example 5.6 allows us to separate /
0
/
2
from
/

1
and cancel it. The knot that is left, /

1
, is precisely /
1
#/
2
with framing j + . See ??. (todo:
make this pic)
A graph embedded in the plane with vertices
i
decorated by elements j
i
Q induces a
surgery diagram in the following way. Place an unknotted, unlinked circle /
i
at each vertex
i
and
57
Figure 5.7: Changing under crossings to over via a 0-framed unknot
frame it by j
i
. If two vertices are connected by an edge, then link the corresponding circles once.
If the graph is a tree, then there is a unique way to do this. These surgery tries provide a simply
way to express surgery diagrams that are composed of unknotted components linked together in
the simplest way.
We have seen that a surgery on any knot with framing 1 results in an integral homology sphere.
There is a particular homology sphere that is ubiquitous in low dimensional topology called the
Poincare sphere. We will dene it to be the space obtained by surgery on the left-hand trefoil with
framing 1.
Example 5.7. We will show that the Poincare sphere is homeomorphic to performing surgery on
the 1
8
Dynkin diagram where each vertex is framed by 2. We blow down this diagram three
times by adding 3 unknotted components with framings +1 and sliding them over the 3 ends of the
diagram. Then we blow up 8 times to remove all components with framing 1. At this point we
have 3 components with framing 3, 2 and 5 linked to a component with framing 1. Blowing this
up we get two components with framing 2 and 4 linked to a component with framing 1. Blowing
up again gives a component of framing 3 linked with a component of framing 1, and a nal blow of
gives a single knot with framing 2. Performing a few 1

1
moves gives the trefoil with framing 1.
(todo: make pic)
5.3 The Rational Calculus
Now we will extend our moves on integral surgeries to rational surgeries. The rst move we
will consider is called a Rolfsen twist. Recall that the lens spaces 1(j, ) and 1(j, + nj) are
homeomorphic for any integer n; that is, the surgeries on the unknot with rational coecients
p
q
and
p
np+q
are homeomorphic. We will describe how to explicitly visualize this homeomorphic, which
will then lead us to a new move on rational surgeries.
Let H be the standardly embedded solid torus in o
3
and T its boundary torus. We can construct
1(j, ) by a j, surgery on the unknot in o
3
, which is equivalent to gluing two solid tori together
via a map ) : T T that maps the meridian j to the curve j + j/, where / is the longitude. If
: T T is a right Dehn twist around j, then each time the curve j+j/ intersects j (which is j
times) the curve (j +j/) picks up another intersection with /, hence (j +j/) = ( +j)j +j/.
Since extends to a homeomorphism of the solid torus H that T bounds we have that it also
extends to a homeomorphism of H
f
H and H
f
H by our gluing lemma (lemma 3.1). This
shows explicitly that 1(j, ) is homeomorphic to 1(j, j + ). We can generalize this construction
by letting be an n-fold Dehn twist around j (where it is a right twist if n 0 and a left twist if
n < 0), in which case we get an explicit homeomorphism of 1(j, nj + ) with 1(j, ). If : = j,,
then we can also rephrase this as the : surgery on the unknot is homeomorphic to the
1
n+
1
r
surgery
58
n
r =
p
q
n 1
1
1+
1
r
=
p
p+q
Figure 5.8: The Rolfsen Twist
on the unknot for any n.
We can now describe the Rolfsen twist. Suppose we have a rationally framed link in o
3
such
that one component / is framed with an integer n and is linked once with an unknotted component
l which has any rational framing : = j,. We claim that surgery on this link is equivalent to
surgery on the link formed by changing /s framing to n1 and ls framing to
p
p+q
=
1
1+
1
r
. We
will show this by mimicking the above arguments to describe an explicit homeomorphism between
the spaces.
First we remove a tubular neighborhood l of l from o
3
to obtain the space A, and let
) : l A be the gluing homeomorphism for the surgery on l; that is, ) maps the meridian
of l to times around the meridian and j times around the longitude of A. Let be the
homeomorphism of A that cuts the space open along a disc 1 A such that 1 is the longitude
on l, performs a full right twist, then glues everything back, and is the identity outside a small
ball containing 1. Now glue l back into A via the map ). Then clearly this surgery is equivalent
to the original surgery, except now l is framed by
p
p+q
and / is framed by n + 1.
It turns out that the Rolfsen twists are nearly enough to get between any rational surgery
diagrams for the same 3-manifold.
Theorem 5.3. If two 3-manifolds presented as a rational surgery diagram are orientation preserv-
ingly dieomorphic, then one diagram can be brought into the other via a nite sequence of Rolfsen
twists and adding/removing a knot component with framing .
Let us look at many examples using these rational calculus moves.
Example 5.8.
Proposition 5.5. We can convert any rational surgery diagram 1 for a 3-manifold into an integral
surgery diagram 1

for the same 3-manifold by adding unknotted components.


(todo: surgery diagram for 1(j, ) in terms of continued fraction expansion of j,)
(todo: any rational surgery can be turned into an integral surgery)
59
6 4-Manifold Theory
6.1 Intersection Forms on 4-Manifolds
Let ` be a closed, oriented, connected and simply connected 4-manifold, and let [`] denote its
fundamental class. Since ` is simply connected we have H
1
(`) = 0, hence H
3
(`) = 0 by
Poincare duality. Further, the split exact sequence of the universal coecient theorem says that
H
2
(`) = Ext(H
1
(`), Z) Hom(H
2
(`), Z) = Hom(H
2
(`), Z)
Since Hom(, 1) is torsion free if 1 is free, we have that H
2
(`) is torsion free, and hence H
2
(`)
is torsion free by Poincare duality. Dene a bilinear form Q
M
: H
2
(`) H
2
(`) Z by
Q
M
(, ) = ( , [`])
This symmetric bilinear form is called the intersection form of `. This form is non-degenerate,
which means that its matrix in any basis is non-singular, i.e. determinant = 1.
There is a more geometric way of dening this bilinear form in terms of counting intersections
of embedded surfaces. If ` is a closed, smooth, oriented n-manifold, we say that a homology
class o H
k
(`) is represented by an oriented, smoothly embedded /-submanifold i : ` if
i

[] = o. Then we have the following


Lemma 6.1. Let ` be a closed, oriented, smooth 4-manifold. Any homology class o H
2
(`) can
be represented by a smoothly embedded surface in `.
Proof. We will prove this for the case when ` is simply connected, which is the case we need
the most for our later discussions. Since ` is simply connected, the Hurewicz homomorphism
/ :
2
(`) H
2
(`) is an isomorphism. Recall that this map is dened by /([)]) = )

[o
2
], so
every homology class H
2
(`) has a corresponding classifying map ) : o
2
` such that
)

[o
2
] = . Since dim` = 4 we have that ) can be slightly homotoped so that ) is an immersion
with nitely many isolated double points. Locally, a double point looks like two planes in R
4
intersecting at a point 1. We can cut a small disc out of each plane around 1 and connect the
planes with a cylinder that misses 1. Performing this operation on each double point of )(o
2
) we
get a smoothly embedded surface in ` representing .
So, for a 4-manifold ` and two homology classes o, / H
2
(`), let 1
a
, 1
b
` be smoothly
embedded, oriented surfaces representing o and / respectively. We can slightly deform the embed-
dings, if necessary, so that 1
a
and 1
b
intersect transversally in a nite collection of points. Assign
a sign of 1 to each intersection point j depending on whether or not the induced orientation on
T
p
1
a
T
p
1
b
from 1
a
and 1
b
agrees with T
p
`. We call the sum of these numbers the intersection
product of o and /, and denote it by o /. Then
Proposition 6.1. Q
M
(, ) = 11() 11()
If we x a basis c
1
, . . . , c
k
of H
2
(`), then we call the matrix [Q
M
] = c
i
c
j

ij
the matrix form
of Q
M
. It is clearly symmetric.
Example 6.1. Let ` = o
2
o
2
, then H
2
(`) is generated by o = o
2
j
0
and / = r
0
o
2
for
some base point (r
0
, j
0
) `. We can slightly deform o to be completely disjoint from the original
o, hence o o = 0, and similarly / / = 0. We cannot deform o so that it is disjoint from /, but
60
o and / clearly intersect transversally at one point, and by choosing orientations appropriately we
have o / = 1. Therefore the matrix form of Q
M
with respect to the basis o, / is
[Q
M
] =
_
0 1
1 0
_
Example 6.2. We will see soon that the intersection form on C1
2
is (1) and on C1
2
is (1).
More generally, the intersection form on the 1
2
-bundle over o
2
with Euler number n is just (n).
Example 6.3. An easy application of the Mayer-Vietoris sequence shows that H
2
(`#)

=
H
2
(`) H
2
(), and so Q
M#N
= Q
M
Q
N
.
If ` is a compact, oriented, connected and simply connected 4-manifold with boundary, then
H
2
(`) and H
2
(`, `) are still torsion free, and so Poincare-Lefschetz duality gives us a non-
degenerate pairing Q
M
: H
2
(`) H
2
(`, `) Z dened by Q
M
(, ) = ( , [`, `]). If
` is an integral homology sphere, then H
2
(`, `) = H
2
(`), and so Q
M
is a symmetric, non-
degenerate pairing of H
2
(`) with itself. In terms of intersections of surfaces, we can still dene the
intersection product : H
2
(`) H
2
(`) Z in the usual way, and if ` is an integral homology
sphere we still have Q
M
(, ) = 11() 11(). There is a converse to the non-degeneracy of Q
M
.
Proposition 6.2. The intersection form Q
M
on a compact, oriented, connected and simply con-
nected 4-manifold ` with boundary is non-degenerate if and only if ` is an integral homology
sphere.
Proof. If ` is a disjoint union of integral homology spheres, then we have already seen that Q
M
is non-degenerate by Poincare-Lefschetz duality.
Suppose Q
M
is non-degenerate. Since ` is simply connected we have H
1
(`) = 0 by the
universal coecient theorem, hence H
3
(`, `) = 0 by Poincare-Lefschetz duality. A part of the
long exact sequence of the pair (`, `) looks like
0 H
2
(`)
i
H
2
(`)
j
H
2
(`, `) H
1
(`) 0
So we have i

is injective, hence imi

= 0 if and only if H
2
(`) = 0. If imi

were not trivial, then


for any o imi

and / H
2
(`) we have o / = 0 since we can move any surface representing / so
that it does not intersect the boundary at all. But, Q
M
is assume to be non-degenerate, so we do
not allow this to happen, hence imi

= 0, and so H
2
(`) = 0. Finally, this implies H
1
(`) = 0,
therefore ` is an integral homology sphere.
Intersection forms are very important in 4-dimensional topology. In fact, they nearly classify all
simply connected 4-manifolds by a theorem of Freedman. So, it will be useful to recall some basic
facts of non-degenerate, symmetric, bilinear forms Q : Z
n
Z
n
Z. We say that two such forms
Q
1
, Q
2
are isomorphic if there is an isomorphism : Z
n
Z
n
such that Q
1
(r, j) = Q
2
((r), (j)).
The rank of Q is dened to be n. If we think of Q as a bilinear form over R, then we can diagonalize
over the reals and there will be /
+
positive values and /

negative values on the diagonal. These


numbers are invariants of the isomorphism type of Q, as is the signature sign Q = /
+
/

. We call
Q a denite form if /

= 0 or /
+
= 0, and call Q an indenite form otherwise. We say that Q is
even if Q(r, r) = 0 mod 2 for all r, and otherwise we say Q is odd.
There is a nice relationship between surgeries on links and intersection forms on cobordisms
of surgeries. First we will consider a fourth way of dening the linking number of two disjoint,
oriented knots (see the other three denitions in section 2.1). For disjoint, oriented knots /
i
and /
j
61
in o
3
, let 1
i
and 1
j
be Seifert surfaces with their induced orientations. We can push the interiors
of these surfaces into the 4-ball 1
4
that bounds o
3
so that 1
i
o
3
= /
i
and 1
j
o
3
= /
j
. Then,
perturbing 1
i
and 1
j
slightly so that they intersect transversely if necessary, one can compute that
lk(/
i
, /
j
) = 1
i
1
j
.
Let A be a 4-dimensional 2-handlebody; that is, a 4-manifold that can be decomposed as 0-
handle 2-handles 4-handle. This is equivalent to specifying a link 1 = /
1
/
n
in 1
4
with /
i
framed by an integer j
i
. For each component /
i
, let 1
i
denote a surface of /
i
embedded
in 1
4
= o
3
, then we can push this surface a little bit into 1
4
to obtain a surface

1
i
such that

1
i
o
3
=

1
i
= /
i
. Let G
i
be the core disc of /
i
, then by gluing

1
i
and G
i
together along /
i
we
obtain a closed, oriented, smoothly embedded surface
i
in A. Put any orientation on /
i
, then we
get induced orientations on

1
i
and G
i
, hence
i
also gets an induced orientation.
Proposition 6.3. The surfaces
1
, . . . ,
n
form a basis for H
2
(A).
Proof. First consider the space obtained from attaching one 2-handle Y = 1
4

(1
2
1
2
). Let
be a small open neighborhood of (1
2
1
2
), and let l = 1
4
and \ = (1
2
1
2
) . Then
l is homotopy equivalent to 1
4
, \ is homotopy equivalent to 1
2
1
2
, and l \ is homotopy
equivalent to o
1
1
2
. The Mayer-Vietoris sequence gives the following exact sequence.
H
2
(l) H
2
(\ ) H
2
(Y )

H
1
(l \ ) H
1
(l \ )
which implies that the connecting homomorphism : H
2
(Y ) H
1
(l \ ) is an isomorphism,
where clearly H
1
(l\ )

= Z. To understand the generator of H
2
(Y ) let us recall how the connecting
homomorphism is dened. A cycle c 7
2
(Y ) can be expressed as c = n + with n C
2
(l) and
C
2
(\ ). This implies n = , hence n 7
1
(l \ ), and so we dene [c] = [n].
The attaching sphere knot / of the 2-handle is the generator of H
1
(l \ ), so we need to
determine what element of H
2
(Y ) gets mapped to /. By our description of we see that can
be represented by a 2-chain in l and a 2-chain in \ whose boundaries are /. Therefore we can
take these chains to be the Seifert surface of / pushed into 1
4
and the core disc of the 2-handle
1
2
1
2
, and so is can be represented by gluing these surfaces together.
Repeating this argument for each 2-handle in A we get a set of n generators of H
2
(A), which
we know to be of rank n from its handlebody homology, therefore these generators form a basis.
We call the basis
,
. . . ,
n
the canonical basis of H
2
(A). Let us compute the matrix of the
intersection form with respect to this canonical basis. First we compute the self intersection number

i

i
. Translate the core disc G
i
to a parallel disc G

i
in the boundary 1
2
1
2
. This new disc
will intersect the boundary of 1
4
at a longitude of /
i
, call it /

i
, with linking number lk(/
i
, /

i
) = j
i
.
Isotope the Seifert surface 1
i
to a new surface 1

i
such that 1

i
= /

i
, and then push this surface
into 1
4
to obtain

1

i
such that

1

i
o
3
=

1

i
= /

i
and so that

1
i
and

1

i
intersect transversally.
By gluing

1

i
to G

i
we obtain a closed surface

i
that represents the same homology class as , and
clearly

i
= (

1
i
G
i
) (

1

i
G

i
) = 1
i
1

i
= lk(/
i
, /

i
) = j
i
Repeating this argument for two dierent components /
i
and /
j
, shows that
i

j
= lk(/
i
, /
j
).
So, we have proved the following.
Proposition 6.4. The intersection form of a 2-handlebody in the canonical basis is equal to the
linking matrix of the associated link.
62
D
4
D
2
D
2
Figure 6.1: Surgery Cobordism
Let ` = A, i.e. ` is the result of performing the surgery on the link 1 in o
3
. We saw in
corollary 5.1 that H
1
(`) could be computed as Z
n
, im[1], where [1] is the linking matrix of 1. As
a corollary we now have
Corollary 6.1. Let A be a 2-handlebody with intersection form Q
X
. Then H
1
(A) is nite if and
only det Q
X
,= 0, in which case [H
1
(A)[ = [ det Q
X
[.
6.2 Kirby Calculus Reinterpreted
Kirby calculus is essentially a 4-dimensional technique, and so we take another look at the Kirby
moves. To bring Kirby calculus into the 4-dimensional world we show that surgeries on manifolds
induce cobordisms. Recall that two compact, oriented, smooth n-manifolds `
1
and `
2
are said
to be cobordant if there is a compact, oriented, smooth (n + 1)-manifold \ such that \ =
`
1
.`
2
, where `
1
is `
1
with the opposite orientation. A manifold ` is said to be cobordant
to zero if it is cobordant to the empty manifold, i.e. \ = `.
Proposition 6.5. The manifold resulting from any surgery on any knot in o
3
is cobordant to zero.
Proof. Let / be a knot in o
3
with tubular neighborhood (/), and let : (o
1
1
2
) (`(/))
be the attaching map that determines the surgery on /. Let c be the image of any meridian 1
2
under so that c is a curve in (/). There is a dieomorphism : 1
2
1
2
(/) such that
(1
2
0) = / and (1
2
) = c, where is some point on the boundary of the 2-disc.
Let \ be the manifold resulting from attaching the 2-handle 1
2
1
2
to the the 4-ball 1
4
via
: 1
2
1
2
(/) 1
4
.
The boundary of the 4-ball has changed by attaching this 2-handle. The part of the boundary
of 1
4
outside of (/) (which is the knot complement) have remained boundary points, and all the
points in (/) have now become interior points of `. The free part 1
2
1
2
of the boundary of
the 2-handle have now become boundary points of ` (see gure 6.1). This means we have removed
a solid torus from ` and glued in another solid torus. In fact, the solid torus was glued in by a
dieomorphism that maps the meridian 1
2
to the curve c, hence this is just performing
the surgery on / determined by . Therefore we have constructed a cobordism from the surgered
manifold to the empty manifold, and so it is cobordant to zero.
Corollary 6.2. Every closed, oriented 3-manifold is cobordant to zero.
63
M
[0, 1]
N(k)
D
2
D
2
Figure 6.2: Surgery Cobordism
Proof. By proposition 6.5 we have that every surgery on o
3
is cobordant to zero. But, by theo-
rem 5.1 every closed, oriented 3-manifold is a surgery on o
3
, hence every such manifold is cobordant
to zero.
Example 6.4. We saw earlier that 1
2
-bundles over o
2
are canonically parameterized by Z by
attaching a 4-dimensional 2-handle along an unknotted circle in 1
4
. Let 1
n
denote this 4-manifold
with boundary, and since 1
n
is the result of performing a surgery on the unknot with framing n,
we see that 1
n
= 1(n, 1). In particular 1
0
is the trivial bundle o
2
1
2
. We also saw that C1
2
is a union of a 0, 2 and 4-handle, and the 2-handle is attached to the 0-handle along an unknotted
circle. So, which of the 4-manifolds 1
n
corresponds to C1
2
1
4
, i.e. C1
2
without its 4-handle? In
C1
2
the 2-handle 1
2
1
2
is attached to 1
4
= 1
2
1
2
via the map : 1
2
1
2
(1
2
1
2
)
dened by (., n) = (., .n). By xing a . and restricting [n[ = 1 we see that a longitude of
o
1
1
2
is wrapped once around the meridian and once around the longitude of o
1
1
2
. So, when
we attach this handle the boundary changes by performing a surgery on the unknot with framing
+1, hence C1
2
1
4
is dieomorphic to 1
1
. We also have 1
1
is dieomorphic to C1
2
1
4
.
A slight generalization of the construction in proposition 6.5 gives an explicit cobordism between
any manifold and any surgery on that manifold.
Corollary 6.3. Any integral surgery on any 3-manifold ` is cobordant to `.
Proof. Mimic the construction in proposition 6.5 by attaching a 2-handle to the component `1
of the 4-manifold ` 1. After the handle is attached one component will be ` and the other
component will be the result of the sugery (see gure 6.2).
For a framed link 1, let \
L
denote the 4-dimensional cobordism of the surgery on 1 with
zero, i.e. \
L
= 0-handle a 2-handle for each component of 1. This association of a 4-manifold
with boundary to any framed link is what allows us to recast Kirby calculus as a 4-dimensional
theory. In fact, this is the way Kirby calculus was originally described. The rst Kirby move
added an unknotted, unlinked component to 1 with framing 1 to get the link 1

. On the level
of 4-manifolds this amounts to \
L
being dieomorphic to \
L
#C1
2
or \
L
#C1
2
, depending on
the framing of the added unknot. This follows from example 6.4 where we saw that 1
1
= C1
2
1
4
and 1
1
= C1
2
1
4
.
64
Next, we claim that performing the second Kirby move on two components of 1 is equivalent
to sliding one handle correspond to one component across the handle corresponding to the other
component. Since this whole operation is technically taking place in a 4-dimensional manifold it
is dicult to picture it. Figures 3.9 and 3.10 showed how to slide 3-dimensional 2-handles over
each other, and the idea in 4-dimensions is similar. Let us attach two 2-handles /
1
, /
2
to the 4-ball
1
4
with attaching circles /
1
, /
2
and framings n
1
, n
2
and let A = 1
4
/
1
/
2
. To slide /
1
over
/
2
we need to nd a disc parallel to the core of /
2
and then isotope the attaching circle of /
1
through this disc. We can take any disc parallel to the core, say 1
2
j where j ,= 0. The
boundary of this disc is precisely where it intersects the 3-sphere, which is at a longitude /

2
of /
2
such that lk(/
2
, /

2
) = n
2
, i.e. its a longitude determined by the framing. As we saw in gure 3.10,
isotoping the attaching circle of /
1
through this disc is the same as form the banded sum of the
attaching circle with the boundary of the disc, so the attaching circle of /
1
after sliding across /
2
is the connect sum of /
1
with /
2
along some band disjoint from /
1
and /
2
. We could compute the
framing of this handle after the slide directly, or appeal to proposition 6.4. Namely, let
1
,
2
be the
canonical basis of H
2
(A), then after a handle slide this canonical basis has changed to
1
+
2
,
2
.
The framing on /
1
after the slide is
(
1
+
2
) (
1
+
2
) =
1

1
+
2

2
+ 2
1

2
= n
1
+n
2
+ 2 lk(/
1
, /
2
)
In example 5.5 we saw how an unknotted component with framing 0 that links one other strand
exact once allows us to separate and cancel those two components from the rest of the link. What
really happened is that we were able to separate the 0-framed unknot and the linked component
from the rest of the link, and then we could completely unknot that linked component so that we
were left with a Hopf link with one component framed by 0 and the other framed by some integer
n. By repeatedly sliding handles we arrived at a Hopf link framed by 0 and 0, or 0 and 1 depending
on the parity of n. The 4-manifolds corresponding to these framed Hopf links are o
2
o
2
and
o
2

o
2
respectively. So, example 5.5 has the following two immediate corollaries.
Corollary 6.4. Let ` be a 4-dimensional handlebody and let /
0
, /
1
be attaching circles of two 2-
handles such that /
1
lies entirely in 1
4
and /
0
is a 0-framed meridian of /
1
. Finally, let be the
handlebody with all the handles of ` except for /
1
and /
2
. Then ` is dieomorphic to #o
2
o
2
if the framing coecient of /
1
is even, and otherwise ` is dieomorphic to #o
2

o
2
.
Corollary 6.5. If ` is a 2-handlebody with odd intersection form, then `#o
2
o
2
is dieomor-
phic to `#o
2

o
2
.
Proof. Since Q
M
is odd we must have that any matrix representation of Q
M
has an odd number of
the diagonal. In particular, if we take the matrix representation with respect to the canonical basis
of H
2
(`) we see that some 2-handle has odd framing, so let / denote the attaching circle for this
handle. Form `#o
2
o
2
by adding a Hopf link /
1
/
2
, both framed as 0, and slide /
1
over / to
obtain /

1
. Then /

1
has odd framing and is linked once by /
2
, so we can separate them from the rest
of the link, and reduce it until it is a Hopf link framed by 0 and 1, hence we have `#o
2

o
2
.
65
7 Morse Homology
7.1 The Morse-Smale Condition and Moduli Spaces of Flow Lines
Let ) : A R be a Morse function on a closed, Riemannian n-dimensional manifold (A, p)
with the Levi-Civita connection

. The metric p and connection

will be used for background
computations, and is not really signicant. The set of critical points of ) of index i will be denoted
by crit
i
()), and we set crit()) =
i
crit
i
()). Recall that at a critical point j crit()), the tangent
space T
p
A splits as a direct sum 1

T
p
A1
+
T
p
A, where 1

T
p
A and 1
+
T
p
A denote the negative
and positive eigenspaces of the Hessian H
f
at j and dim1

T
p
A = j(j). A negative gradient ow
line is a curve n : R A such that dn,dt =
g
) n. Since A is compact,
g
) generates a
1-parameter group of dieomorphisms
t
: A A, i.e
t
is a dieomorphism for all t R such
that
s+t
=
s

t
,
0
= id and for xed point j A we have
d
t
(j)
dt
=
g
)(
t
(j))
For a critical point j crit(j) we dene the ascending and descending manifolds to be
A(j) =
_
r A : lim
t

t
(r) = j
_
D(j) =
_
r A : lim
t

t
(r) = j
_
These sets consist of points that ow to critical points in forward or backward time respectively.
The ascending manifold is sometimes called the stable manifold and the descending manifold is
sometimes called the unstable manifold. A study of the dynamics of the negative gradient ow
allows one to prove the following important statement.
Proposition 7.1. If j crit
i
()), then D(j) and A(j) are smoothly embedded discs of dimension
i and n i respectively. Further, T
p
D(j) = 1

T
p
A and T
p
A(j) = 1
+
T
p
A.
We will not prove this because it is not useful in our route to dening Morse homology. A pair
(), p) is said to be Morse-Smale if ) is a Morse function such that for every pair of critical points
j, crit()) the ascending and descending manifolds intersect transversely D(j) A(). Since
codimension is additive under transverse intersections we have dimD(j) A() = j(j) j(). We
do not yet know such pairs even exist (they do, in fact, in abundance), but we immediately get the
following lemma.
Lemma 7.1. If (), p) is a Morse-Smale pair on A, then a non-constant ow line of
g
) from j
to has j(j) j().
Proof. Since (), p) is Morse-Smale we have D(j) A() is a manifold of dimension j(j) j(). If
j(j) < j(), then this manifold is empty, hence there are no ow lines from j to . If j = , then
this is a 0-dimensional manifold whose points correspond to constant ow lines. Therefore if there
is a non-constant ow line from j to we must have j(j) j().
Example 7.1. If we stand the torus T
2
vertically with the induced metric from its embedding in
R
2
, and let ) : T
2
R be the height function, then ) does not satisfy the Morse-Smale condition.
If we let j, , :, : denote the critical points of ) such that )(j) )() )(:) )(:), then we
can see that D() does not intersect A(:) transversely. Unfortunately this type of failure of the
66
Morse-Smale condition is common for obvious choices of Morse functions, which tend to have
many symmetries. We can perturb ) a little so that it satises the Morse-Smale condition (for
example, slightly tilt T
2
o its vertical axis), but sometimes it is better to consider a more relaxed
requirement than Morse-Smale, such as the Morse-Bott condition.
Example 7.2. Consider the at torus T
2
= R
2
,Z
2
with the smooth function ) : T
2
R dened
by )(r, j) = cos(2r)+cos(2j). This function is clearly well dened, and the critical points are at
the equivalence classes of (0, 0), (1,2, 0), (0, 1,2), (1,2, 1,2) R
2
. It is easy to compute the Hessian
at these critical points and see that they are non-degenerate, so ) is a Morse function. It is also
easy to see that all the ascending and descending manifolds intersect transversally, so ) satises
the Morse-Smale condition.
There is another way of discussing the Morse-Smale condition that is better suited for innite
dimensional generalizations of Morse homology. For two critical points j, crit()), let /
f
(j, )
be the moduli space of ow lines from j to , i.e.
/
f
(j, ) =
_
n : R A :
dn
dt
=
g
)(n(t)), n() = j, n() =
_
We will show how to put a topology on this set momentarily. This set carries a free R action given
by pre-composition with translation: (: n)(t) = n(t + :). Let

/
f
(j, ) denote the quotient of
/
f
(j, ) by this action, which we call the moduli space of unparameterized gradient ow lines.
There is an obvious isomorphism between D(j) A() and /
f
(j, ) as sets since any ow line is
uniquely determined by its initial condition, so we would like to know that /
f
(j, ) is actually a
manifold.
To pursue this we need to develop signicant functional analysis machinery. Let (
p,q
be the
subset of \
1,2
loc
(R, A) such that if n (
p,q
, then
1. n() = j and n() =
2. There is an 1 < 0 suciently negative such that n(, 1] is contained in a coordinate chart
around j and n[
(,R]
\
1,2
((, 1], A).
3. There is an 1 0 suciently positive such that n[1, ) is contained in a coordinate chart
around j and n[
[R,)
\
1,2
([1, ), A).
Then (
p,q
is a Banach manifold modeled on \
1,2
(R, R
n
). Let
E
: c (
p,q
be the Banach bundle
over (
p,q
dened by
c =
_
uCp,q
n 1
2
(n

TA)
The pullback bundle n

TA makes sense because the Sobolev embedding \


1,2
C
0
ensures that
n is continuous, so c is a Banach bundle with characteristic ber 1
2
(R, R
n
). For any metric p on
A we can dene a section 1 : (
p,q
c by
1(n) = (n, n

+
g
) n)
where n

is the weak derivative of n. Technically 1 depends on the Morse function ), metric p and
the points j, , but we omit this from the notation. If we let 0
E
denote the zero section of c (
p,q
,
then we have the following:
Proposition 7.2. 1
1
(0
E
) = /
f
(j, )
67
Proof. If n 1
1
(0
E
), then n

=
g
) n. Since
g
) and n are continuous, we have n

is
continuous, hence n

is equal to the regular derivative dn,dt. Therefore n is C


1
and so is an
integral curve from j to , i.e. n /
f
(j, ). On the other hand, if n /
f
(j, ), then n converges
exponentially to the critical points j and , so n \
1,2
(R, A) and hence n 1
1
(0
E
).
By applying the standard bootstrapping argument we see that n is actually as smooth as p
(assuming ) is smooth), hence n is actually C
k
.
In this context we say that (), p) is a Morse-Smale pair if 1 0
E
, which of course implies that
/
f
(j, ) is a submanifold of the Banach manifold (
p,q
. If we could show that d1 is Fredholm
at the points in 1
1
(0
E
), then we would have that 1
1
(0
E
) = /
f
(j, ) is a nite dimensional
submanifold of (
p,q
. If we work a little harder we can actually generalize the above constructions a
little to show something stronger. Let /
k
denote the Banach manifold consisting of C
k
-metrics on
A. Note that this is just a convex subset of the Banach space (i.e. linear space) of C
k
sections of
o
2
T

A A. Let
e
E
:

c /
k
(
p,q
be the Banach bundle that is the pullback of c (
p,q
along
the projection /
k
(
p,q
(
p,q
. In particular,

c can be written as

c =
_
(g,u)A
k
Cp,q
(p, n) 1
2
(n

TA)
So, the characteristic ber of

c is 1
2
(R, R
n
). We dene a section

1 : /
k
(
p,q


c of this bundle
by

1(p, n) = (p, n, n

+
g
) n). This section depends only on ), but we do not include this in the
notation. Then the most important proposition we want to prove is the following.
Proposition 7.3. The section

1 is transverse to the zero section:

1 0
e
E
. Further, if : /
k

(
p,q
/
k
is the projection onto the second factor, then the restriction of to :

1
1
(0
e
E
) /
k
is
Fredholm of index j(j)j(), i.e. d
(g,u)
is Fredholm of index j(j)j() at each (p, n)

1
1
(0
e
E
).
The immediate corollary of the this proposition is that a generic p /
k
is a regular value of
the restricted map , hence
1
(p) is a nite dimensional submanifold of dimension j(j) j().
Clearly
1
(p) = /
f
(j, ), where this moduli space is taken with respect to p, therefore we have
the following.
Corollary 7.1. The moduli space of ow lines /
f
(j, ) from j to for a Morse-Smale pair is
a (j(j) j())-dimensional submanifold of (
p,q
. The moduli space of unparameterized ow lines

/
f
(j, ) is a (j(j) j() 1)-dimensional manifold.
Proposition 7.3 will actually follow reasonably easily from the fact about the vertical dierential
11
u
of the section constructed for the rst Banach bundle c (
p,q
is Fredholm.
Proposition 7.4. For each n 1
1
(0
E
), the vertical dierential 11
u
: T
u
(
p,q
c
u
is Fredholm
of index j(j) j().
So, now we will work to prove proposition 7.4. The rst thing we can do is introduce coordinates
on our manifolds and bundles so that we can write the vertical dierential in a nicer way. To
put coordinates on (
p,q
x n (
p,q
and a small open set l around n. Let (TA) be a small
enough convex, tubular neighborhood of the zero section of TA so that the exponential map exp :
(TA) A is dened, where this is the exponential map with respect to the background metric
p and Levi-Civita connection

oating around. This exponential map induces coordinates on (
p,q
in the following way. Let n (
p,q
and take another curve (
p,q
close to n. Then there is a unique
68

u
U C
p,q
0
(U) W
1,2
(u

TX)
(U) L
2
(u

TX)
E|
U
Figure 7.1: Trivializing c (
p,q
section () \
1,2
(n

TA) (i.e. a \
1,2
vector eld along n) such that (t) = exp(n(t), ()(t)).
So, for a small neighborhood l around n we have a chart : l \
1,2
(n

TA) = T
u
(
p,q
with
(n) = 0.
Next we trivialize
E
: c (
p,q
over l. For l we can dene the bundle map 1
v
u
:
n

TA

TA that parallel transports a vector \


t
in the ber of n

TA over t along the geodesic


: (n(t), :()(t)) to a vector in the ber of

TA over t. This map allows us to identify a


section in 1
2
(n

TA) with a section in 1


2
(

TA), which are precisely the bers of c over n and


. Now that we have coordinates on (, l) around n in (
p,q
and a trivialization of c[
U
(and hence
coordinates on c[
U
), we can write the coordinate representation 1 : (l) (l) 1
2
(n

TA) of
the section 1 as
1() =
_
, (1
v
u
)
1
_
exp(n, )

+
g
) exp(n, )
__
Computing the vertical dierential 11
u
: (l) 1
2
(n

TA) in these coordinates gives


11
u
()(t) =
d
d:

s=0
1(:(t))
=
d
d:

s=0
_
(1
v
u
)
1
_
exp(n(t), :(t))

+
g
)(exp(n(t), :(t))
__
=

(t)
_
n

(t) +
g
)(n(t))
_
=

(t)
n

(t) +

(t)
(
g
)(n(t)))
where the second from last line follows from the fact that any connection can be recovered from par-
allel translations along its geodesics. Also, since the connection is torsion free we have

(t)
n

(t) =

(t)
(t), so
11
u
() =

u
+

(
g
)(n))
If we trivialize n

TA using parallel translation with respect to



, then vertical dierential 11
u
:
\
1,2
(R, R
n
) 1
2
(R, R
n
) can be written as
(11
u
)(t) =
d
d:

s=t
(:)
t
(t)
where C
k
(R, End(R
n
)) is dened by
t
(\ ) =

V
(
g
)(n(t))). By (4.4) we have

= H
f
(j)

= H
f
()
69
With this set up we will see that proposition 7.4 follows from a more general proposition concerning
spectral ows (see [4]). Before stating this proposition we recall the denition of spectral ow, at
least for the special case we need. Let C
b
(R, End(R
n
)) be a bounded path of operators on R
n
such that

are symmetric and do not have zero as an eigenvalue. Then we dene the spectral
ow of the path to be the number of eigenvalues of
s
that change from negative to positive as
: ranges from to . We denote this integer by SF(), and clearly in this special case we have
SF() = dim1
+
(

) dim1
+
(

)
Proposition 7.5. Let C
b
(R, End(R
n
)) such that

and

are symmetric and do not have


zero as an eigenvalue. Then the operator 1
A
: \
1,2
(R, R
n
) 1
2
(R, R
n
) given by
1
A
=
d
dt

is Fredholm with
ind 1
A
= SF()
Proof.
Applying proposition 7.5 to the vertical dierential 11
u
= d,dt
s
we see that 11
u
is
Fredholm of index
ind 11
u
= SF()
= (dim1
+
(H
f
()) dim1
+
(H
f
(j)))
= dim1

(H
f
(j)) dim1

(H
f
())
= j(j) j()
and this nishes the proof proposition 7.4. Let us see how to use proposition 7.4 to prove proposi-
tion 7.3.
Proof of proposition 7.3. Our derivation of the local expression of the vertical dierential 11
u
at
n 1
1
(0
E
) can be slightly adapted to give the following local expression for the vertical dierential
1

1
(g,u)
: T
(g,u)
(/
k
(
p,q
) T
(g,u,0)

c at a point (p, n)

1
1
(0
e
E
):
1

1
u
(, ) = 11
u
() +

(
g
)(n))
where

denotes the derivative with respect to the variable (remember that /


k
is just a convex,
open subset of a linear space).
Let (p, n)

1
1
(0
e
E
) and consider the dierential d
(g,u)
: T
(g,u)

1
1
(0
e
E
) T
g
/
k
of the restric-
tion of to

1
1
(0
e
E
). We claim that the kernel of d
(g,u)
is isomorphic to the kernel of 11
u
. We
have that
T
(g,u)

1
1
(0
e
E
) = (1

1
(g,u)
)
1
(T
(g,u)
0
e
E
)
which can be written as
T
(g,u)

1
1
(0
e
E
) =
_
(, ) T
g
/
k
T
u
(
p,q
: 11
u
() +

(
g
)(n)) = 0
_
(7.1)
using our coordinates. We clearly have d
(g,u)
(, ) = , hence
ker d
(g,u)
=
_
(0, ) T
g
/
k
T
u
(
p,q
: 11
u
() = 0
_
= ker 11
u
70
Next we claim that the cokernel of d
(g,u)
is isomorphic to the cokernel of 11
u
. (todo: do this).
Finally, we show that the image of d
(g,u)
is closed. Using our coordinates and (7.1), the image
of d
(g,u)
can be written as
d
(g,u)
_
T
(g,u)

1
1
(0
e
E
)
_
=
_
T
g
/
k
: 11
u
() +

(
g
)(n)) = 0 for some T
u
(
p,q
_
=
_
T
g
/
k
:

(
g
)(n)) im11
u
_
So, we see that imd
(g,u)
is the pre-image of im11
u
under the map

(
g
)(n)). This map is
continuous, and 11
u
is closed since it is Fredholm, therefore imd
(g,u)
is closed.
Proposition 7.6. A generic C
k
function ) : A R (/ 2) is Morse.
Proof. We clearly want to use the Sard-Smale theorem somehow, and in fact we will use theo-
rem E.3. In order to use this we need to formulate a Banach bundle such that the zero section cor-
responds to what we want. In particular, let : T C
k
(A)A be the Banach bundles that is the
pullback of the cotangent bundle T

A A along the projection C


k
(A) A A so that the ber
of T over (), j) is just T

x
A. Consider the section : : C
k
(A)A T given by :(), j) = d)
p
T

p
A.
Let (), r)
1
(0
F
) and let us compute the vertical dierential 1:
(f,x)
: T
(f,x)
(C
k
(A)A) T

x
A
at (), r). Note that since C
k
(A) is a linear space we have T
f
C
k
(A) is naturally isomorphic to
C
k
(A), so we describe how 1:
(f,x)
acts on a pair (/, \ ) C
k
(A) T
x
A.
1:
(f,p)
(/, \ ) = d/(A) +
V
(d))
UNFINISHED
7.2 The Compactness and Gluing Theorems
The space of unparameterized ow lines

/
f
(j, ) is not always compact; for example, in gure 7.2
we see a sequence of ow lines (the thin curves) on the deformed sphere that converge to a broken
curve (the bolded curve). A broken ow line from j to is a sequence of critical points :
1
=
j, :
2
, . . . , :
k1
, :
k
= crit()) and curves
1
, . . . ,
k1
in A such that
i
is a ow line from :
i
to :
i+1
. However, there is a canonical compactication of

/
f
(j, ) to a manifold with corners
by adding broken ow lines. There are two parts to this compactication process. The rst
part (called the compactness result) consists of identifying how a sequence can fail to converge in

/
f
(j, ), trying to get a reasonable object out of a diverging sequence, and then adding these
points to

/
f
(j, ) to get a compact space. The second part (called the gluing result) consists
of showing that a broken ow line can be slightly perturbed to an honest ow line so that our
compactication is a manifold with corners.
Recall that a second countable, Hausdor space A is said to be a smooth n-manifold with
corners if every point r A has an open neighborhood l

and homeomorphism

: l


R
nk
[0, )
k
(for some 0 / n) such that the transition functions

are smooth. For


0 / n, we dene the codimension / stratum of A to be the set A
k
of points r A with a
chart : l R
nk
[0, )
k
such that at least one of the last / coordinates of (r) is zero. Note
that A
0
is just the interior of A, and if A
2
= A
3
= = A
n
= , then A is just a manifold with
boundary and A
1
= A. So we will use the notation A to denote A
1
even when the higher strata
are non-empty.
Now we look at what it means for a sequence in

/
f
(j, ) to converge, and how this convergence
can fail. In /
f
(j, ), a sequence of elements converges if it converges in the subspace topology
71
Figure 7.2: Flow lines converging to a broken ow line
induced by the Banach manifold (
p,q
. Since modding out by the R action identies ow lines that
dier by only a translation we have that a sequence ( n
n
) of unparameterized ow lines in

/
f
(j, )
converge to a ow line n if for any sequence of lifts (n
n
) of ( n
n
) and lift n of n there is a sequence
(:
n
) of real numbers such that :
n
n
n
converges to n in /
f
(j, ).
(todo: nish compactications and gluing)

The BIG theorem:
Theorem 7.1. If A is a closed n-manifold and (), p) is a Morse-Smale pair, then for any two
critical points j, , the manifold

/
f
(j, ) has a natural compactication to a smooth manifold with
corners

/
f
(j, ) whose codimension / stratum is

/
f
(j, )
k
=
_
r
1
, . . . , r
k
crit(f)
p, r
1
, . . . , r
k
, q distinct

/
f
(j, :
1
)

/
f
(:
1
, :
2
)

/
f
(:
k1
, :
k
)

/
f
(:
k
, )
In particular, if / = 1, then as oriented manifolds we have

/
f
(j, ) =
_
r crit(f)
p, r, q distinct
(1)
(p)+(r)+1

/
f
(j, :)

/
f
(:, )
The two most important special cases of this theorem are when j() = j(j) 1, in which case
/
f
(j, ) is a compact 0-manifold (hence a nite collection of points), and when j() = j(j) 2,
in which case

/
f
(j, ) has a compactication to a 1-manifold with boundary

/
f
(j, ) such that

/
f
(j, ) consists of just once broken ow lines:

/
f
(j, ) =
_
rcrit
(p)1
(f)

/
f
(j, :)

/
f
(:, )
72
7.3 The Morse Complex
We can now dene the Morse complex
1
associated to a Morse-Smale pair (), p). Let C`
i
(), p)
be the Z,2 vector space generated by the critical points of index i
C`
i
(), p) :=

pcrit
i
(f)
Z,2 j)
We can dene this complex with Z coecients if we work harder and dene orientations on the
moduli spaces /
f
(j, ). It is simpler to work with Z,2 coecients for now, and we will discuss
orientation issues later. The dierential : C`
i
(A) C`
i1
(A) counts the number of ow lines
between critical points. Specically, if j crit
i
()), then
(j) =

qcrit
i1
(f)
#

/
f
(j, )
and extend linearly to C`
i
()). The count #

/
f
(j, ) is of course done modulo 2 (at least until
we can dene orientations).
Proposition 7.7. = 0
Proof. First recall that the number of points in the boundary of a 1-manifold is zero when counted
modulo 2. For j crit
i
()) we can easily compute
(j) =
_
_

rcrit
i1
(f)
#/
f
(j, :) :
_
_
=

rcrit
i1
(f)
#/
f
(j, :) (:)
=

qcrit
i2
(f)
_
_

rcrit
i1
(f)
#/
f
(j, :) #/
f
(:, )
_
_

So, the coecient of crit
i2
()) in (j) is precisely #/
f
(j, ), which is zero since it is a
compact 1-dimensional manifold.
We dene the Morse homology H`

(), p) of the pair (), p) to be the homology of the chain


complex (C`

(), p), ).
7.4 A Natural Isomorphism between Morse and Singular Homology
We would like to show that the homology groups do not depend on our choice of Morse function
or metric. There are many ways of doing this, but the rst way we will consider is to show
that H`

(), p) is naturally isomorphic to some other known homology theory, such as singular
homology, which of course implies that H`

(), p) does not depend on ) or p.


The homology theory we will show is naturally isomorphic to H`

is dened as follows. Recall


that an i-dimensional current on A is a functional on the space of dierential i-forms
i
(A).
1
Some might call this construction the Morse-Smale-Witten complex due to its complicated history. See [1] for
the interesting history of Morse theory and Morse homology
73
The topology on
i
(A) is dened by saying that a sequence of dierential forms
n
converges to
if and only if for every compact set 1 A all partial derivatives of
n
(in any coordinate system)
converges uniformly to the partial derivatives of . A singular simplex :
i
A determines an
i-dimensional current [] by
[]() =
_

We dene the chain groups C


c
i
(A) of our homology theory to be the free abelian group generated
by [], where :
i
A is a singular i-simplex that is transverse to A(j) for all j crit()). The
dierential : C
c
i
(A) C
c
x1
(A) is dened by
[]() =
_

d
where C
i
(A) and
i1
(A). We clearly have
2
= 0 since d
2
= 0. The homology of this
chain complex will be denoted by H
c
i
(A), and it is naturally isomorphic to singular homology.
There is a natural compactication of D(j), for j crit()), that is similar to theorem 7.1 by
adding broken ow lines.
Proposition 7.8. There is a natural compactication of D(j) to a smooth manifold with corners
D(j), whose codimension / stratum is
D(j)
k
=
_
q
1
, . . . , q
k
crit(f)
p, q
1
, . . . , q
k
distinct

/
f
(j,
1
)

/
f
(
1
,
2
)

/
f
(
k1
,
k
) D(
k
)
In particular, for / = 1 we have
D(j) =
_
p=qcrit(f)
(1)
(p)+(q)+1

/
f
(j, ) D()
as oriented manifolds. Further, the projections of D(j) onto the last factor D(
k
) patch together to
form a smooth map c : D(j) A whose restriction to D(j) is the inclusion.
The compactied spaces D(j) are homeomorphic to a closed ball of dimension j(j), and so the
maps c : D(j) A give A a CW-structure with one cell of dimension i for each critical point of
index i. A consequence of this is that (A) is equal to

i
(1)
i
c
i
where c
i
= #crit
i
()).
We now dene maps 1 : C`
i
(A) C
c
i
(A) and : C
c
i
(A) C`
i
(A) that are chain homotopy
inverses of each other. For j crit
i
()) we have D(j) is a closed disc of dimension i.
UNFINISHED
7.5 Direct Invariance of Morse Homology
Now we will show that H`

(), p) does not depend on ) and p directly. This line of reasoning is


more applicable to Floer homologies than section 7.4 since there might not be another well-known
homology theory to compare to. Fix two Morse-Smale pairs ()
0
, p
0
) and ()
1
, p
1
) with associated
Morse complexes (C
0

,
0
) and (C
1

,
1
). Take any path of pairs = ()
t
, p
t
) : t 1 from ()
0
, p
0
)
to ()
1
, p
1
). Two things to note: the pairs ()
t
, p
t
) do not have to be Morse-Smale for all t (in fact,
it may be impossible), and such a path is always possible as all functions ` R are homotopic
and any two metrics can be connected by a straight line of metrics. We will show how such a path
(with some simple requirements) induces a chain map

: C
0

C
1

.
74
Let \
t
denote the negative gradient vector eld of )
t
with respect to p
t
, and dene the vector
eld \

on 1 A by
\

= (t + 1)t(t 1)

t
+\
t
(7.2)
Since the two components of this vector eld are gradient like vector elds, we can dene critical
points, ow lines, ascending and descending manifolds in the regular way. In particular, the rst
coordinate of \

is just the gradient of the function R R dened by t


1
4
(t + 1)
2
(t 1)
2
. This
function only has critical points at 1, 0, 1 of index 0, 1, 0 respectively. Thus, the critical points of
index i of \

are precisely
crit
i
(\

) = 0 crit
i1
()
0
) 1 crit
i
()
1
) (7.3)
We will say that the path is admissible if the ascending and descending manifolds of the critical
points of \

intersect transversely. Just as almost all pairs (), p) are Morse-Smale we also have
that almost all paths of pairs are admissible.
For critical points 1, Q of \

we dene /

(1, Q) to be the moduli space of ow lines of \

from 1 to Q, and

/

(1, Q) the moduli space of unparameterized ow lines. If is admissible,


then /

(1, Q) is a (j(1) j(Q))-dimensional manifold, and so



/

(1, Q) is a (j(1) j(Q) 1)-


dimensional manifold. The spaces

/

(1, Q) are without boundary, but are not generally compact.


We can compactify them via broken lines as before. We state this result only for the two special
cases we need. If j(1) = j(Q)+1, then

/

(1, Q) is already compact, and since it is 0-dimensional


we have that it is just a nite collection of points. If j(j) = i = j() + 1, then j((0, j)) = i + 1
and j((1, )) = i 1, then

/

((0, j), (1, )) is a 1-dimensional manifold with a compactication

((0, j), (1, )) by adding broken ow lines that pass through one intermediate critical points of
index i. These points correspond to eq. (7.3), and so we have

((0, j), (1, )) =


_
rcrit
i1
(f
0
)

((0, j), (0, :))



/

((0, :), (1, ))


=
_
r

crit
i
(f
1
)

((0, j), (1, :

))

/

((1, :

), (1, ))
Note that a ow line between two points in the 0 A slice must stay in that slice for all time,
and similarly for the 1 A slice. The ow lines in the these slices correspond to the ows of )
0
and )
1
in A, so we have

((0, j), (0, :)) =



/
f
0
(j, :)

((1, :

), (1, )) =

/
f
1
(:

, )
For j crit
i
()
0
) and crit
i
()
1
) we have j((0, j)) = i+1 and j((0, )) = i, hence

/

((0, j), (1, ))


is a compact 0-dimensional manifold. We now dene

: C
0
i
C
1
i
on j crit
i
()
0
) by

(j) =

qcrit
i
(f
1
)
#

((0, j), (1, ))


and extend this linearly to all of C
0

(where

/

((0, j), (1, )) is counted modulo 2).


75
[0, 1]
(X, f
0
, g
0
)
(X, f
1
, g
1
)
p
q
r
r

Figure 7.3: Counting ow lines from (0, j) to (1, )


Proposition 7.9.

is a map of chain complexes, i.e. the following diagram commutes


C
0
i

0
//

C
0
i1

C
1
i

1
//
C
1
i1
Proof. We will use the idea from the proof of proposition 7.7 by showing the coecient of
crit
i
()
1
) in (

)(j) is equal to a count of the number of boundary components of a


compact 1-manifold, and hence is zero (modulo 2).
For j crit
i
()
0
) we have


0
(j) =

rcrit
i1
(f
0
)

qcrit
i1
(f
1
)
#

/
f
0
(j, :) #

((0, :), (1, ))

(j) =

crit
i
(f
1
)

qcrit
i1
(f
1
)
#

((0, j), (1, :

)) #

/
f
1
(:

, )
(7.4)
Now it is clear that the coecient of crit
i1
()
1
) in (

)(j) is precisely the number


of boundary components in the compact 1-manifold

/

((0, j), (1, )), hence zero. Therefore

is
a chain map.
We now have that admissible paths induce maps on Morse homology

: H`

()
0
, p
0
)
H`

()
1
, p
1
). If we take to be the constant path at a Morse-Smale pair (), p), then we expect the
induced map

to be the identity. This is indeed the case, as is shown in the following proposition.
Proposition 7.10. Let be the constant path at the Morse-Smale pair (), p). Then is admissible
and

= id
Proof. Let \ be the vector eld on [0, 1] A induced by the constant path , i.e. \ = (t +
1)t(t 1)

t

g
). Since the vector eld (t + 1)t(t 1)

t
on [0, 1] is directed from 0 to 1, we
have that the descending manifold of a critical point (0, j) of \ is D(0, j) = [0, 1) D(j) and the
ascending manifold of a critical point (1, ) is A(1, ) = (0, 1] A(). These submanifolds clearly
76
intersect transversally. On the other hand, the ascending manifold of (0, j) is just 0 A(j) and
the descending manifold of (1, ) is just 1 D(), which do not intersect at all. Therefore \ is
admissible.
For j, crit
i
()), a ow from (0, j) to (1, ) in [0, 1] A projects to a ow from j to in the
0 A slice. This implies j = since there are no ows between critical points of equal index,
unless it is the constant ow. Therefore

(j) = j.
Next, we would like to know that

depends only on the homotopy type of the path . We


will use a dierent model for homotopies than the typical 1-parameter family of maps. Let 1 be
a 2-dimensional manifold with corners such that 1 has two edges, call them c and c

, and two
vertices, call them and n. Such a space is sometimes called a digon. If
1
,
2
are two paths
between pairs ()
0
, p
0
) and ()
1
, p
1
), then a homotopy is just a family of pairs = ()
d
, p
d
) : d 1
such that when restricted to c it coincides with
1
and when restricted to c

it coincides with
2
.
This means that ()
v
, p
v
) = ()
0
, p
0
) and ()
w
, p
w
) = ()
1
, p
1
).
Put a metric p on 1 such that the length of the edges is 1, and let

) : 1 R be a smooth
function with only two critical points, one at of index 2 and one at n of index 0. Further, assume
that the negative gradient vector eld

\ =
e g

) with respect to p is tangent to the edges and is


equal to (t +1)t(t 1) on the edges. Finally, dene the vector eld \

on 1A by \

=

\ +\
d
,
where \
d
is the negative gradient vector eld of )
d
with respect to p
d
. Note that these constructions
are just a higher dimensional version of what we constructed to dene \

. We see that \

restricted
to the slices c A and c

A coincide with the vector elds \

1
and \

2
, respectively. We can
again dene critical points, ow lines, ascending and descending manifolds of \

. Also, just like


last time, we will say that is admissible if the ascending and descending manifolds of critical
points intersect transversely, and almost all families are admissible.
The critical points of index i of \

are precisely
crit
i
(\

) = crit
i2
()
0
) +n crit
i
()
1
)
For critical points 1, Q of \

let /

(1, Q) be the moduli space of ow lines of \

from 1 to Q, and

(1, Q) the moduli space of unparameterized ow lines. For any admissible family we have
that /

(1, Q) is a (j(1) j(Q))-dimensional manifold, and



/

(1, Q) is a (j(1) j(Q) 1)-


dimensional manifold. Compactifying these spaces in full generality can be quite complicated, so
we state the results for the two special cases that we need.
If j(1) = j(Q) + 1, then

/

(1, Q) is already compact, and since it is 0-dimensional we have


that it is just a nite collection of points. If 1 = (, j) and Q = (n, ) with j(j) = i = j(),
then j((, j)) = i + 2 and j((n, )) = i, hence

/

((, j), (n, )) is a 1-dimensional manifold and


has a natural compactication to a 1-dimensional manifold

/

((, j), (n, )) with boundary by


adding broken ow lines. The boundary of this space has more points than one might think at
rst. The broken ow lines that pass through one intermediate critical point of index i + 1 are of
course included in the boundary, and these intermediate critical points are precisely
crit
i+1
(\

) = crit
i1
()
0
) n crit
i+1
()
1
)
But, we also have boundary points coming from those ow lines which stay in the cA and c

A
for all time. These ow lines are in correspondence with

/

1
((, j), (n, )) and

/

2
((, j), (n, )),
77
hence

((, j), (n, )) =



/

1
((, j), (n, ))

/

2
((, j), (n, ))

_
rcrit
i1
(f
0
)

((, j), (, :))



/

((, :), (n, ))

_
r

crit
i+1
(f
1
)

((, j), (n, :

))

/

((n, :

), (n, ))
Note that the ow lines in 1 A that stay in the A slice correspond to the ows in A with
respect to ()
0
, p
0
), and similarly for ow lines in the nA slice, so in the above boundary we have

((, j), (, :)) =



/
f
0
(j, :)

((n, :

), (n, )) =

/
f
1
(:

, )
Let us dene the map H

: C`
i
()
0
, p
0
) C`
i+1
()
1
, p
1
) on j crit
i
()
0
) by
H

(j) =

qcrit
i+1
(f
1
)
#

((, j), (n, ))


and extend linearly to C`
i
()
0
, p
0
).
Proposition 7.11. For any two admissible paths
1
,
2
from ()
0
, p
0
) to ()
1
, p
1
), the induced maps

1
and

2
are chain homotopic.
Proof. Let = ()
d
, p
d
) : d 1 be an admissible homotopy between
1
and
2
. We claim that
H

dened above is a chain homotopy; that is, if we set 1 =


1
H

+H

1
+

2
then
1(j) = 0 for all j crit()). We show this by realizing the coecient of crit
i
()
1
) in 1(j), where
j crit
i
()
0
), as the number of boundary components of

/

((, j), (n, )), which we know to be


zero since the space is a compact, 1-dimensional manifold.
We can easily compute the following
H


0
(j) =

rcrit
i1
(f
0
)

qcrit
i
(f
1
)
#

/
f
0
(j, :) #

((, :), (n, ))

1
H

(j) =

crit
i+1
(f
1
)

qcrit
i
(f
1
)
#

((, j), (n, :

)) #

/
f
1
(:

, )
Let us x j crit
i
()
0
) and crit
i
()
1
), then j((, j) = i+2 and j((n, )) = i, so

/

((, j), (n, ))


is a 1-dimensional manifold and has its natural compactication described above. We see that by
denition the coecient of in

1
(j) is precisely #

1
((, j), (n, )), and similarly for the coe-
cient of in

2
(j). Now it is clear that the coecient of in 1(j) is precisely #

((, j), (n, )),


hence zero.
Now we have that

depends only on the homotopy class of the path . However, all paths
between any two pairs are homotopic (since C

(`) and the space of metrics is contractible), so

does not even depend on . This means that just by virtue of the construction of Morse
homology we always get a natural map

: H`

()
0
, p
0
) H`

()
1
, p
1
) for any Morse-Smale pairs
()
0
, p
0
) and ()
1
, p
1
). We want to show that this naturally dened map is an isomorphism, which
means that Morse homology depends only on A.
78
D
(X, f
1
, g
1
)
(X, f
0
, g
0
)
p
q
r
r

v
w
Figure 7.4: Counting ow lines from (, j) to (n, ) in 1 A
In order to show this we need composition of paths
2

1
(where
1
s ending point is
2
s
starting point) to correspond to composition of induced maps

; that is, we need to show


that

1
is chain homotopic to

. To show this we need to introduce a new apparatus


similar to the digon from above. First, slightly homotope
2

1
so that it is admissible. Then,
let T be a trigon; that is, a 2-manifold with 3 corners, n, , n, and 3 edges, c
uv
, c
vw
, c
uw
. Let
= ()
d
, p
d
) : d T be a family of pairs such that ()
u
, p
u
) =
1
(0), ()
v
, p
v
) =
1
(1) =
2
(0),
()
w
, p
w
) =
2
(1), and such that restricted to c
uv
is
1
, restricted to c
vw
is
2
, and restricted
to c
uw
is
2

1
. Put a metric p on T such that the edges are of length one, and take a smooth
function

) : T R whose only critical points are n, , n with j(n) = 2, j() = 1 and j(n) = 0,
and the negative gradient

\ =
e g

) is tangent to the edges of T and like (1 + t)t(1 t). Dene


the vector eld \

on T A by \

=

\ + \
d
, where \
d
=
g
d
)
d
. We can dene the critical
points, ow lines, ascending and descending manifolds of \

in the regular way. We say that is


an admissible family if the ascending and descending manifolds of \

intersect transversely.
The critical points of index i of \

are precisely
crit
i
(\

) = n crit
i2
()
u
) crit
i1
()
v
) n crit
i
()
w
)
For critical points 1, Q of \

let /

(1, Q) be the moduli space of ow lines from 1 to Q, and

(1, Q) the moduli space of unparameterized ow lines. For any admissible family we have
that /

(1, Q) is a (j(1) j(Q))-dimensional manifold and



/

(1, Q) is a (j(1) j(Q) 1)-


dimensional manifold. Like the spaces

/

considered above, the space



/

(1, Q) may have


boundary, and their compactications can be quite complicated. The description of this space can
vary greatly depending on where 1 and Q are, so we state the results we need in the only two
special cases used.
If j(1) = j(Q)+1, then

/

(1, Q) is already compact, and since it is a 0-dimensional manifold


we have that it is just a nite set of points. If 1 = (n, j) and Q = (n, ) with j(j) = i = j(), then
j((n, j)) = i + 2 and j((n, )) = i, hence

/

((n, j), (n, )) is a 1-dimensional manifold and has a


natural compactication to a 1-dimensional manifold

/

((n, j), (n, )) with boundary by adding


broken ow lines. The broken ow lines that pass through one intermediate critical point of index
i + 1 are contained in the boundary, and these points are precisely
crit
i+1
(\

) = n crit
i1
()
u
) crit
i
()
v
) n crit
i+1
()
w
)
79
But, we also have boundary points coming from those ow lines that stay in the c
uw
A slice for
all time. These ow lines correspond to

/

1
((n, j), (n, )), hence

((n, j), (n, )) =



/

1
((n, j), (n, ))

_
rcrit
i1
(fu)

((n, j), (n, :))



/

((n, :), (n, ))

_
r

crit
i+1
(fw)

((n, j), (n, :

))

/

((n, :

), (n, ))

_
scrit
i
(fv)

((n, j), (, :))



/

((, :), (n, ))


Note that ow lines that stay in the n A slice correspond to

/
fu
(j, :), and similarly for ows
that stay in the n A slice. Therefore we have

((n, j), (n, :)) =



/
fu
(j, :)

((n, :

), (n, )) =

/
fw
(:

, )
Further, the ows from (n, j) to (, :) must stay in the c
uv
A slice and similarly for ows from
(, :) to (n, ), so we have

((n, j), (, :)) =



/

1
((n, j), (, :))

((, :), (n, )) =



/

2
((, :), (n, ))
See gure 7.5.
We dene G

: C`
i
()
u
, p
u
) C`
i+1
()
w
, p
w
) on j crit
i
()
u
) by
G

(j) =

qcrit
i+1
(fw)
#

((n, j), (n, ))


and extend linearly to all of C`
i
()
u
, p
u
).
Proposition 7.12.

1
is chain homotopic to

1
.
Proof. Let = ()
d
, p
d
) : d T be an admissible family between
1
,
2
and
2

1
. We claim
that the map G

dened above is a chain homotopy between

1
and

1
, i.e. if we set
1 =
w
G

+ G

1
+

1
, then the coecient of crit
i
()
w
) in 1(j) is zero
for all j crit
i
()
u
).
We can easily compute
G


u
(j) =

qcrit
i
(fw)

rcrit
i1
(fu)
#

/
fu
(j, :) #

((n, :), (n, ))

w
G

(j) =

qcrit
i
(fw)

crit
i+1
(fw)
#

((n, j), (n, :

)) #

/
fw
(:

, )

1
(j) =

qcrit
i
(fw)

scrit
i
(fv)
#

1
((n, j), (, :)) #

2
((, :), (n, ))
Now it is clear that the coecient of crit
i
()
w
) in 1(j), where j crit
i
()
u
), is precisely
#

((n, j), (n, )), hence zero.


80
u
v
w
p
q
r
r

s
T
Figure 7.5: Counting ow lines from (n, j) to (n, ) in T A
Combining propositions 7.10 to 7.12 we get invariance of Morse homology.
Corollary 7.2. For any two Morse-Smale pairs ()
0
, p
0
) and ()
1
, p
1
), the Morse homologies H`

()
0
, p
0
)
and H`

()
1
, p
1
) are naturally isomorphic.
7.6 Orientations on Moduli Spaces
There are (at least) two ways to introduce orientations on our moduli spaces /
f
(j, ) and

/
f
(j, )
so that we can dene the Morse complex with Z coecients. One way is geometric and technically
the simplest way to dene orientations, but it does not generalize to the innite dimensional ver-
sions of Morse homology. The other way is very technical and uses high powered machinery from
functional analysis, but also has obvious interpretations in the innite dimensional setting. We will
give a supercial description of both of these orientations, and explain why they should give the
same end result.
For the geometric orientations we must assume that our Riemannian manifold (A
n
, p) is ori-
ented. Let us also x a Morse function ) on A so that (), p) is a Morse-Smale pair. We stated
in proposition 7.1 that the ascending manifold A(j) and descending manifold D(j) of a critical
point j crit
i
()) are smoothly embedded discs of dimension i and ni, respectively. Let us dene
M
f
(j, ) to be the intersection D(j) A(), which is a submanifold of dimension j(j) j(). We
get an R action on M
f
(j, ) via the 1-paremeter group of dieomorphisms
t
: A A generated
by
g
): for t R and r M
f
(j, ) we dene t r =
t
(r). This action is free and smooth, so

M
f
(j, ) = M
f
(j, ),R is a (j(j) j() 1)-dimensional manifold.
Proposition 7.13. The map : /
f
(j, ) M
f
(j, ) dened by (n) = n(0) is a dieomorphism,
and for each t R the following diagram commutes
/
f
(j, )
t
//

/
f
(j, )

M
f
(j, )
t
//
M
f
(j, )
where t on the horizontal arrows denotes the R action on the respective spaces. In particular,

M
f
(j, ) is dieomorphic to

/
f
(j, ).
Let us put an arbitrary orientation on T
p
D(j) for all j crit()). Since T
p
D(j)T
p
A(j) = T
p
A,
and T
p
A is oriented by assumption, we get an induced orientation on T
p
A(j). These orientations
81
at j uniquely extend to coherent orientations on all of A(j) and D(j), and so now all ascending
and descending submanifolds are oriented. Now the spaces M
f
(j, ) get an induced orientation, as
described in section 1, so we can orient /
f
(j, ) by requiring the isomorphism in proposition 7.13
to be orientation preserving.
UNFINISHED
7.7 Applications
The Morse inequalities are set of nice results that put topological restrictions on a space according
to what type of Morse functions the space admits, and they nearly come for free with the denition
of Morse homology. It was these inequalities that inspired Floer to dene his innite dimensional
generalization of Morse homology in order to prove the Arnold conjecture. Since H`

(), p) is
isomorphic to H

(A) we already have that the Euler characteristic of A can be computed by


(A) =

(1)
i
#crit
i
()) for any Morse function ) on A. The Morse inequalities strengthen this
quite a bit.
First we need to prove a general result concerning nitely generated chain complexes.
Theorem 7.2 (Euler-Poincare Theorem). Take any nitely generated chain complex
0 C
n
n
C
n1

n1


2
C
1

1
C
0
0
and let H
i
= H
i
(C

). Then
n

i=0
(1)
i
rank C
i
=
n

i=0
rank H
i
Proof. At each i = 0, . . . , n we have a short exact sequence
0 ker
i
C
i

i
im
i
0
hence rank C
i
= rank ker
i
+ rank im
i
. We also have a short exact sequence
0 im
i+1
ker
i
H
i
0
hence rank ker
i
= rank im
i+1
+rank H
i
. There is a common rank ker
i
in each of the equations
we derived from these short exact sequences, and solving for it gives
rank ker
i
= rank C
i
rank im
i
= rank im
i+1
+ rank H
i
Therefore
n

i=0
(1)
i
(rank C
i
rank im
i
) =
n

i=0
(1)
i
(rank im
i+1
+ rank H
i
)
On the left side we have that rank im
i
appears with sign (1)
i+1
, and on the right rank im
i
appears with sign (1)
i1
, hence they cancel and we are left with
n

i=0
(1)
i
rank C
i
=
n

i=0
(1)
i
rank H
i
82
Theorem 7.3 (Morse Inequalities). Let ) : A
n
R be a Morse function. Let c
i
= #crit
i
()) be
the number of critical points of index i, and let /
i
= rank H
i
(A) be the i-th Betti number of A.
Then
c
k
c
k1
+ + (1)
k
c
0
/
k
/
k1
+ + (1)
k
/
0
(7.5)
for all 0 / n.
Proof. We have rank H`
i
(), p) = rank H
i
(A) = /
i
and rank C`
i
(), p) = #crit
i
()). Since
H`
i
(), p) is a subquotient of C`
i
(), p), we clearly have c
i
/
i
. Let 0 / n, and let (C

)
denote the chain complex (C`

) truncated at /:
0 C
k

C
k1



C
1

C
0
0
The chain complex (C

) is nitely generated, so by the Euler-Poincare theorem we have


k

i=0
(1)
i+k
rank C
i
=
k

i=0
(1)
i+k
rank H
i
(C

) (7.6)
The inserted signs (1)
k
are necessary so that the last term (i = /) in the sums are non-negative.
The homology of the chain complex (C

) is equal to the homology of (C1

) in all gradings
except possibly the last, so we have rank H
i
(C

) = /
i
for all 0 i / 1. For the last
term we have H
k
(C

) = ker
k
, hence H
k
(C`

) is a quotient of H
k
(C

), and so we have
/
k
rank H
k
(C

). By replacing the last term rank H


k
(C

) in (7.6) with the smaller term /


k
we obtain the Morse inequality
c
k
c
k1
+ + (1)
k
c
0
/
k
/
k1
+ + (1)
k
/
0
7.8 Wittens Complex
In a paper on supersymmetry in physics, Witten was able to rediscover the Morse inequalities
(see theorem 7.3) and Morse (co)homology by considering a simple supersymmetric theory on the
exterior bundle

A of a compact, orientable manifold A


n
. First we discuss the mathematical
content of his results without mentioning any of the physics.
Let us recall some of the basic results of Hodge theory. If we put a metric p on A, then
we get an induced volume form vol
g
, and we get a star operator :
k
(A)
nk
(A) which
satises
2
= (1)
k(nk)
id

k
(X)
. The de Rham operator d :
k
(A)
k+1
(A) has a formal
adjoint :
k
(A)
k1
(A), which can be dened by = (1)
kn+n+1
d. We dene the
Laplace operator :
k
(A)
k
(A) by = d + d. This operator has discrete spectrum, and
decomposes
k
(A) into eigenspaces

k
(A) =

R
1
k

where 1
k

=
_

k
(A) : =
_
The fundamental theorem of Hodge theory is the following.
Theorem 7.4 (Hodges Theorem). 1
k
0

= H
k
dR
(A)
83
So, we can think of (1

0
, d) has a nite dimensional chain complex (the dierential happens to
be zero) that computes the homology of the innite dimensional chain complex (

(A), d). We
clearly have d(1
k

) 1
k+1

, so for 0 we also get a chain complex


0 1
0

d
1
1

d

d
1
n

0
However, this sequence is exact. To see this, let 1
k

such that d = 0, then


d
_

_
=
1
d =
1
=
1
=
therefore is exact. This means that the homology of the complex (1

, d) is zero. Now, for xed

0
0, let

0
=

0
1
k

That is,

0
consists of all the eigenforms of with eigenvalue less than or equal to
0
. Now,
the restriction of d to

0
gives a nite dimensional complex (

0
, d), and since this complex is a
direct sum of exact sequences (when (0,
0
]) and a sequence with homology equal to de Rham
homology (when =
0
), we have
H

0
, d) = H

dR
(A)
If we have chosen
0
large enough so that there is an eigenvalue of in (0,
0
], then we have re-
placed the innite dimension complex (

, d) with a non-trivial nite dimensional complex (

0
, d)
without changing the homology.
We now repeat these same ideas with a slight variation of Hodge theory. For a xed Morse
function ) : A R and real parameter t we dene the operator d
t
:
k
(A)
k+1
(A) by
d
t
= c
ft
dc
ft
that is, we conjugate the de Rham operator by the linear isomorphism c
ft
. The function )
will remained xed in our discussion, but we will be taking t to be large soon, so we only put t in our
notation. We clearly have d
2
t
= 0, and this operator has the formal adjoint
t
:
k
(A)
k1
(A)
dened by

t
= c
ft
c
ft
and so we dene the Laplacian
t
= d
t

t
+
t
d
t
. We can also dene the homology groups H

t
(A) =
ker d
t
, imd
t
, but the isomorphism
k

k
given by c
ft
descends to an isomorphism
H

dR
(A)

= H

t
(A), so this homology theory does not give us anything new. Like before we get an
eigenspace decomposition

k
(A) =

R
1
k

(t) where 1
k

(t) =
_

k
(A) :
t
=
_
We also have a Hodge-like theorem that says 1
k
0
(t)

= H
k
dR
(A). Also like before, for 0 we have
an exact sequence
0 1
0

(t)
dt
1
1

(t)
dt

dt
1
n

(t) 0
So, if for xed
0
0 we dene
k

0
(t) =

0
1
k

(t), then the nite dimensional complex


(

0
(t), d
t
) has the same homology as (

, d). At this point Witten claims that for suciently


84
large t 0, the dimensions dim
k

0
(t) become independent of t, and so let us write (

0
(), d

)
to denote this stabilized chain complex. In fact, Witten shows that
dim
k

0
() = #crit
k
()) = c
k
As we saw in theorem 7.3, whenever we have a nite dimensional complex whose chain groups
correspond to critical points of a Morse function, and whose homology computes the homology
of A, then we get the Morse inequalities for free. So, we have proved the Morse inequalities by
considering a variation of Hodge theory.
Now let us see the physical ideas behind the above results. Using the denition of d
t
and
t
,
one can easily compute that these operators act on a /-form
k
(A) by
d
t
= d +td)
t
= ti
df

If we choose coordinates (r, l) on A, and dene the operators

i
= i
x
i
and
i
= dr
i
,
then one can show that
t
can be written as

t
= +t
2
[[d)[[
2
+t
n

i=1

2
)
r
i
r
j
[

i
,
j
]
As t gets large we expect the eigenforms of this operator to be increasingly concentrated at the
critical points of ). This leads us to look around the critical points. If j crit()), then we can
choose normal coordinates (r, l) around j such that ) is of the form
) = )(j) +
n

i=1

i
(r
i
)
2
+O(r
3
)
where the
i
s are the eigenvalues of the Hessian of ) at j. In these coordinates, if we ignore terms
of order 3 and above, then Laplacian
t
can be approximated by

t
=
n

i=1
_


2
(r
i
)
2
+t
2

2
i
(r
i
)
2
+t
i
[

i
,
i
]
_
This looks like a simple harmonic oscillator, plus some other operator. Recall that for o 0, the
simple harmonic oscillator
H =
d
2
dr
2
+o
2
r
2
has spectrum (2/ + 1)o : / = 0, 1, . . .. So, let us write our approximation of the Laplacian as

t
=
n

i=1
(H
i
+t
i
1
i
)
where H
i
=
2
,(r
i
)
2
+ t
2

i
(r
i
)
2
and 1
i
= [

i
,
i
]. The H
i
s commute with themselves,
the 1
i
s commute with themselves, and the H
i
s commute with the 1
j
s. Therefore the spec-
trum of

t
is the sum of the spectrum of the H
i
s and 1
i
s. The spectral of H
i
is given by
(2
i
+ 1)t[
i
[ :
i
= 0, 1, . . .. One can easily compute that
1
i
dr
i
1
dr
i
k
=
_
+dr
i
1
dr
i
k
i i
1
, . . . , i
k

dr
i
1
dr
i
k
i , i
1
, . . . , i
k

85
so the eigenvalues of this operator are just +1s and 1s. In fact, the number of +1 eigenvalues is
precisely /. Now, the spectrum of

t
can be written as the set
_
t
n

i=1
((2
i
+ 1)[
i
[ +n
i

i
) :

i
= 0, 1, . . .
n
i
= 1
#n
i
= +1 = /
_
What choice of
i
and n
i
give zero? If we rewrite the summand as 2
i
[
i
[ + [
i
n
i
[ + n
i

i
, then
we see that we are forced to have
i
= 0 and n
i
= sign(
i
). Therefore, there is exactly one zero
eigenvalue of

t
, and all other eigenvalues blow up to since they are proportional to t. Further,
associated eigenform is a /-form, where
/ = #n
i
= +1 = #
i
< 0 = j(j)
So, at each critical point j crit()), we are able to associate a j(j)-form such that

t
= 0.
Since

t
is only an approximation of
t
, we do not necessarily have that
t
= 0. Therefore, the
number of /-forms which satisfy

t
= 0 at some critical point of index / (i.e. the number of
critical points of index /) is greater than or equal to the number of /-forms with = 0 (i.e. the
/-th Betti number of A). This proves the weak Morse inequalities.
86
8 Heegaard Floer Homology
8.1 Construction for 3-Manifolds
The Morse homology discussed in the previous section has inspired variations that usually go under
the heading of Floer-type theories. These theories start with some dierential equation (in Morse
homology this was the negative gradient ow equation), then one forms a graded group freely
generated by some points in a space or some other objects, and then one denes a dierential
to count the number of solutions to the dierential equation that start at one point and end at
another. Hopefully the dierential squares to zero so that one can dene the homology groups, and
hopefully the homology groups do not depend on the various choices used in the construction of
this homology theory. We now follow this plan to build a homology theory for 3-manifolds from
Heegaard diagrams.
General Idea
First, we will give a general idea of the object we are trying to construct. The details in this
construction are very dicult, like it was for Morse homology, requiring a lot of machinery from
analysis and symplectic geometry. Let (,
1
, . . . ,
g
,
1
, . . . ,
g
) be a genus p Heegaard diagram
for a 3-manifold Y , and assume the and curves are in general position so that they intersect
transversally. Let us also x a base point . in the complement of the and curves. We can
form the symmetric product Sym
g
() of , dened to be the quotient of
g
under the action
of o
g
permuting the components of
g
, i.e. Sym
g
() consists of p-tuples of unordered points of
. Even though o
g
does not act freely on , we do have that Sym
g
() is a smooth manifold, and
in fact a complex structure j on induces a complex structure Sym
g
(j) on Sym
g
().
Since the curves never intersect each other (and similarly for the curves), we have that the
products
T

=
1

g
T

=
1

g
are smoothly embedded submanifolds of Sym
g
(), and T

, hence the intersect consists of


nitely many points. The base point . determines a codimension 2 submanifold \
z
of Sym
g
()
dened by
\
z
= . Sym
g1
() Sym
g
()
Clearly \
z
is disjoint from the tori T

and T

.
Let D C denote the unit disc, and let c
1
= D Re(.) 0 and c
2
= D Re(.) 0.
For two intersection points x, y T

(this means that x and y are unordered p-tuples of


intersection points of the and curves) we dene a Whitney disc connecting x to y to be a
continuous map n : D Sym
g
() such that n(i) = x, n(i) = y, n(c
1
) T

and n(c
2
) T

. Let

2
(x, y) denote the homotopy classes of Whitney discs from x to y.
We dene a function n
z
:
2
(x, y) Z to be the algebraic intersection number of a disc
: D Sym
g
() with the submanifold \
z
. For a Whitney disc , let /() denote the moduli
space of pseudoholomorphic representatives of , i.e. the set of representatives n : D Sym
g
()
of that satises the Cauchy-Riemann equation
n
:
+i
n
t
= 0
If we introduce the proper perturbations of our set up, then /() will be a smooth manifold, and
we denote its dimension by j(). There is a natural free R acts on this space, and so we denote
the quotient by

/() = /(),R.
87
With these denitions we can give an intuitive (but slightly incorrect) denition of Heegaard
Floer homology. Let

C1(, ) be the Abelian group freely generated by the points in T

.
We dene a dierential

:

C1(, )

C1(, ) on x T

by

x =

yTT

2
(x,y)

nz()=0
()=1

/() y
That is, we are counting the number of pseudoholomorphic discs between two points in Sym
g
().
One can show that

2
= 0, so we can dene the homology groups

H1(, ). One can also show
that

H1(, ) does not depend on the Heegaard diagram or complex structures chosen in order
to dene it, hence

H1(, ) is a topological invariant. Now we work to properly dene these
homology groups.
Morse Functions and Pointed Heegaard Diagrams
Let us x a Heegaard diagram (,
1
, . . . ,
g
,
1
, . . . ,
g
) for our 3-manifold Y , where is a surface
of genus p. Recall that this means we attach a 2-handle for each and curve, the result being a
3-manifold with boundary two disjoint 2-spheres, and so we attach two 3-handles in the only way
possible. So, the and curves determine handlebodies l
0
and l
1
of genus p such that l
0
l
1
= .
We also have the following Morse theoretic picture of this set up. Let ) : Y [0, 3] be a self-
indexing Morse function with p critical points of index 1, denote them by 1
1
, . . . , 1
g
, and hence p
critical points of index 2, denote them by Q
1
, . . . , Q
g
. Then l
0
= )
1
[0, 3,2], l
1
= )
1
[3,2, 3] and
= )
1
(3,2), and the attaching circles can be dened by
i
= A(1
i
) and
i
= D(Q
i
) .
We say that a Morse function is compatible with a Heegaard diagram if the Heegaard diagram is
equal to the induced Heegaard diagram from the Morse function.
We can make this Heegaard diagram into a pointed Heegaard diagram by choosing a base
point . in the complement of the and curves. We can consider pointed Heegaard moves
on pointed Heegaard diagrams by requiring that isotopies of attaching circles never pass through
., and the pair of pants dening a handle slide is not allowed to contain ..
The Algebraic Topology of Symmetric Products
Let 1 denote the fat diagonal in Sym
g
(); that is, 1 consists of those ordered p-tuples of points in
such that at least points are equal. The embedded tori T

and T

miss the diagonal, and since


the and curves intersect transversally, we have T

in Sym
g
(). For a complex structure
j on , we get an induced complex structure Sym
g
(j) on Sym
g
() by requiring the quotient map
:
g
Sym
g
() to be holomorphic, where
g
is given the product complex structure j
g
. The
homotopy and homology groups of Sym
g
() in low dimensions is stated in the following lemmas.
Lemma 8.1. If is a genus p surface, then
1
(Sym
g
())

= H
1
(Sym
g
())

= H
1
().
Proof. One can show that
1
(Sym
g
()) is abelian, so that gives the rst isomorphism. For a xed
point r , the inclusion r r Sym
g
() induces an isomorphism H
1
()
H
1
(Sym
g
()) on homology. To construct the inverse of this map we take a generic loop : o
1

Sym
g
(). If we pullback the p!-fold branched cover :
g
Sym
g
() along we get a p!-fold
88
branched cover j :

o
1
o
1
such that the following commutes

o
1
e
//
p

o
1

//
Sym
g
()
The space

o
1
is a disjoint union of circles and has a natural o
g
action on it such that is equivariant.
Let

o
1
=

o
1
,o
g
, and let :

o
1
be the map induced by
1
, where
1
:
g
is the
projection onto the rst factor. Then the map H
1
(Sym
g
()) H
1
() given by [] [

o
1
] is the
inverse map.
Recall that the fundamental group
1
of a space naturally acts on the higher homotopy groups

n
. Let

n
denote the quotient of
n
by the action of
1
.
Lemma 8.2. If is a surface of genus p 1, then

2
(Sym
g
())

= Z. Further, if p 2, then

1
(Sym
g
()) acts trivially on
2
(Sym
g
()), so
2
(Sym
g
())

= Z for p 2.
Proof. The isomorphism

2
(Sym
2
())

= Z is given by counting the algebraic intersection number
of a sphere with the subvariety rSym
g1
() for generic r. We can nd an explicit generator. The
hyperelliptic involution on gives rise to a sphere o
0
Sym
2
(), and so let o = o
0
r
3
r
g

Sym
g
(). Then o maps to 1 under the above isomorphism.
Lemma 8.3. The rst Chern class c
1
of Sym
g
() satises c
1
, [o]) = 1.
Lemma 8.4.
H
1
(Sym
g
())
H
1
(T

) H
1
(T

=
H
1
()
[
1
], . . . , [
g
], [
1
], . . . , [
g
])

= H
1
(Y )
Proof. Since the inclusion of into Sym
g
() induces an isomorphism on rst homology, we see
that the subgroup of H
1
() generated by the curves gets maps to H
1
(T

) in H
1
(Sym
g
()), and
similarly for the curves, and so the rst isomorphism follows. The second isomorphism is obvious
since we building Y from by lling the and curves with discs.
There is a Morse theoretic interpretation of intersection points x T

. Let ) be a
Morse function compatible with our Heegaard diagram with critical points 1
1
, . . . , 1
g
of index 1
and critical points Q
1
, . . . , Q
g
of index 2. If x = r
1
, . . . , r
g
such that r
i

i

(i)
(for some
permutation o
g
), then we have r
i
/(1
i
) T(Q
(i)
), hence r
i
can be identied with an
unparameterized, gradient ow line from 1
i
to Q
(i)
. The intersection point x can therefore be
identied with a p-tuple of ow lines from index 1 critical points to index 2 critical points, which
we will denote by
x
.
The tori T

and T

in Sym
g
() have a nice compatibility with the complex structure Sym
g
(j)
induced by j. First, recall the denition of a totally real submanifold of a complex manifold. If
1 7 is a submanifold of the complex manifold (7, J), then we say that 1 is totally real if
T
p
1 JT
p
1 = 0 for all j 1, i.e. the tangent spaces of 1 do not contain any J-complex lines. It
is clear that the tori T

, T

are totally real when viewed as submanifolds of (


g
, j
g
). Since the
projection :
g
Sym
g
() is holomorphic and a p!-fold cover away from the diagonal 1 in
Sym
g
(), it follows that T

and T

are totally real submanifolds of Sym


g
().
89
Whitney Discs in the Symmetric Product
There is an obstruction to the existence of Whitney discs between two intersection points x, y
T

. By renaming the curves, if necessary, we can assume that r


i

i

i
and j
i

i

(i)
,
for some permutation o
g
. Take a pair of paths o : [0, 1] T

, / : [0, 1] T

that start at x
and end at y. The dierence o / is a loop in Sym
g
(), and hence determines a homology class.
This element vanishes if and only if there is a Whitney disc connecting x to y. Unfortunately, this
homology class depends on the paths o and /, but we can remove this dependence in the following
way. The path o can be written as o(t) = (o
1
(t), . . . , o
g
(t)), where o
i
: [0, 1]
i
, hence any other
paths [0, 1] T

diers from o by winding around the curves some number of times. The ways
this winding can occur is parameterized by H
1
(T

)

= Z
g
. Similar statements hold for the curve
/ and the curves. So, by modding the subgroups H
1
(T

) and H
1
(T

) out of H
1
(Sym
g
()) the
homology class of o / does not depend on o or /, and so the image of this homology class under
the isomorphisms in lemma 8.4 in H
1
(Y ) is denoted by c(x, y).
There is also a way of computing this homology class on . Note that a curve o : [0, 1] T

can be thought of as a p-tuple of sub-arcs of the curves such that the boundary (as a 1-chain)
is o = j
1
+ + j
g
r
1
r
g
. There is a similar description for a path / : [0, 1] T

and
the curves, hence the dierence o / is a 1-cycle in . This cycle depends on the paths used
to dene it, but only up to additional windings around and curves, hence it is a well-dened
element of H
1
(), [
1
], . . . , [
g
], [
1
], . . . , [
g
]). The image of this homology class in H
1
(Y ) under
the isomorphism in lemma 8.4 is equal to c(x, y) that we dened above.
Clearly
2
(x, y) is empty if and only if c(x, y) ,= 0. Further, for x, y, z T

we have the
additive relation c(x, y) +c(y, z) = c(x, z), so we can dene an equivalence relation on T

by
x y if c(x, y) = 0.
The set
2
(x, y) is not a group, but has a natural

2
(Sym
g
()) action given by splicing a sphere
onto a Whitney disc:
:

2
(Sym
g
())
2
(x, y)
2
(x, y)
We can also splice a disc from x to y with a disc from y to z to get a disc from x to z, giving a
map
:
2
(y, z)
2
(x, y)
2
(x, z)
This splicing action makes (x, x) into a group. The algebraic intersection map n
z
is clearly additive
with respect to these splicing actions: n
z
( ) = n
z
() +n
z
().
Since it is dicult to visualize discs in Sym
g
() we would like a way of representing them on
the surface . Recall that the function n
w
:
2
(x, y) Z counts the intersection of the subvariety
n Sym
g
() with a disk
2
(x, y). Let 1
1
, . . . , 1
m
denote the closures of the components of

1

g

1

g
, and choose a point .
i
in the interior of each 1
i
. For a Whitney
disc n : D Sym
g
() we dene the domain of to be the form linear combination
T(n) =
m

i=1
n
z
i
(n)1
i
If n
z
i
() 0 for all i, then we write T() 0. This association does not depend on the points .
i
,
and clearly depends only on the homotopy class of n, hence we get a well-dened map on
2
(x, y).
Domains are additive with respect to the splicing action dened above:
T( ) = T() +T()
T(o ) = T() +

1
i
90
For
2
(x, y), we can think of T() as a 2-chain on in the following way. Let n be a
representative of , then the restriction n[
e
1
can be thought of as a curve from x to y contained in
T

, and n[
e
2
can be thought of as a curve from y to x contained in T

. As noted before, curves


of this type determine 1-chains in with boundaries n[
e
1
= j
1
+ + j
g
r
1
r
g
and
n[
e
2
= r
1
+ +r
g
j
1
j
g
. So, the boundary of the 2-chain T(n) is precisely
T(n) =

i
n
z
i
(n)(n[
e
1
+n[
e
2
)
A 2-chain of the form 1 =

o
i
1
i
is said to be a periodic domain if the boundary T is a sum of
and curves and such that if 1
i
is the component containing ., then o
i
= 0 (i.e. n
z
(1) = 0).
For x T

, a class
2
(x, y) is called periodic if n
z
() = 0. For such a class we have T()
is periodic. The set of periodic classes at x will be denoted by T
x
.
Proposition 8.1. If p 1, then there is an isomorphism

2
(x, x)

= Z H
1
(Y )
where T
x
is identied with H
1
(Y ), and the splitting is given by n
z
(). If x, y T

such
that c(x, y) = 0, then

2
(x, y)

= Z H
1
(Y )
as

2
(Sym
g
()) T
x
-torsors.
Spin
c
Structures
We now introduce Spin
c
structures into our set up. We use Turaevs formulation of Spin
c
structures
on 3-manifolds in terms of homology classes of unit vector elds. We say that two unit vector elds
\
1
and \
2
are homologous if they are homotopic through unit vector elds in the complement of
nitely many 3-balls in Y . A Spin
c
structure on Y is a choice of homology class of non-vanishing
vector elds, and the set of Spin
c
structures on Y is denoted by Spin
c
(Y ). Recall the following fact
about orientable 3-manifolds.
Lemma 8.5. All closed, orientable 3-manifolds are parallelizable.
Proof. Let A be a closed, orientable 3-manifold. If we can nd two linearly independent sections
of TA, then TA can be split as a 2-plane bundle plus a line bundle, which must be orientable and
hence trivial. The rst obstruction to nding two linearly independent sections is the second Stiefel-
Whitney class, which by the Wu formula is given by n
2
= n
2
1
= 0. The only other obstruction lives
in the cohomology of Y with coecients in
2
(\
2
(R
3
)) = 0, hence vanishes.
Since TY is trivial we now x an orthonormal trivialization . This trivialization gives us a one
to one correspondence between unit vector elds \ on Y and maps )
V
: Y o
2
. Fix a generator
j H
2
(o
2
) and dene the map

: Spin
c
(Y ) H
2
(Y ) by

(\ ) = )

V
(j).
Lemma 8.6.

is well-dened and is a one-to-one correspondence.


Proof. First we need to show that if \
1
is homologous to \
2
, then

(\
1
) =

(\
2
). Let )
V
1
, )
V
2
:
Y o
2
be the maps corresponding to \
1
and \
2
. By assumption these maps are homotopic on
the complement of 3-ball in Y . UNFINISHED
91
If \
1
and \
2
are unit vector elds on Y , then we can look at the dierence
(\
1
, \
2
) =

(\
1
)

(\
2
) H
2
(Y )
We have removed the trivialization from the notation because this dierence does not depend
on it. For if
1
,
2
are two trivializations of TY , then there is a map p : Y oO(3) such that

1
(\
p
) = p(j)
2
(\
p
) for all vectors \
p
in T
p
Y . So, if we take two trivializations , p , then the
dierence satises

g
(\ )

(\ ) = p

(n)
where n is the generator of H
2
(oO(3); Z,2) = Z,2, so the dierence does not depend on trivial-
izations. So, for xed \ Spin
c
(Y ) the one-to-one correspondence (, ) : Spin
c
(Y ) H
2
(Y )
makes Spin
c
(Y ) into an ane space over H
2
(Y ). For \
1
, \
2
Spin
c
(Y ) we will write \
1
\
2
as a
short hand for (\
1
, \
2
).
Now let ) be a Morse function on Y compatible with our Heegaard diagram Recall that an
intersection point x T

determined a p-tuple of ow lines from index 2 to index 1 critical


points, which we denoted by
x
. The base point . has a unique ow line going through it from the
index 3 critical point to the index 0 critical point. If we delete tubular neighborhoods of these p +1
ow lines from Y we get a space with boundary consisting of p +1 spheres such that
g
) does not
vanish. The restriction of
g
) to a boundary component is an index 0 vector eld on a sphere,
and hence can be extended to the ball it bounds. Extending
g
) to the interior of these spheres,
and normalizing so that the vector eld is unit length, we dene :
z
(x) to be the Spin
c
structure
this vector eld denes. This gives us a map
:
z
: T

Spin
c
(Y )
Lemma 8.7. For x, y T

we have :
z
(y) :
x
(x) = PD(c(x, y)).
Therefore :
z
(x) = :
z
(y) if and only if x can be connected to y by a Whitney disc. If s Spin
c
(Y )
is represented by a vector eld \ , then we let s be the Spin
c
structure represented by \ . The
rst Chern class of Spin
c
structures is a map c
1
: Spin
c
(Y ) H
2
(Y ) dened by c
1
(s) = s s. This
is just the rst Chern class of the 2-plane bundle over Y given as the orthogonal complement of a
vector eld \ representing s.
Analytic Set-Up of the Moduli Spaces
Our denition of the moduli space of pseudoholomorphic representatives is not going to be sucient
in order to ensure that the spaces /() are smooth manifolds. We need to introduce suitable
perturbations to the objects involved, which we do now. The quotient map
:
g
Sym
g
()
is an honest covering map over Sym
g
() 1. Let us x a Kahler structure (j, ) on the surface ;
that is, j is complex structure on and is a non-degenerate 2-form on such that j tames . The
2-form
0
=
g
on
g
induces a Kahler form

(
0
) on Sym
g
() 1. Let us x a collection of
points .
i

1

g

1

g
, with one .
i
in each component, and an open set
\ Sym
g
() with
_
.
i
Sym
g1
() 1
_
\ Sym
g
()
and such that the closure \ does not intersect the tori T

and T

. We say that an almost complex


structure J on Sym
g
() is (j, , \ )-nearly symmetric if J tames

(
0
) over Sym
g
() \ , and J
92
agrees with Sym
g
(j) over \ . The space of (j, , \ )-nearly symmetric, almost complex structures
will be denoted by (j, , \ ). The tameness condition gives us that the tori T

and T

are totally
real, which is used for energy bounds and Gromov compactness.
We will now describe a more convenient model for Whitney discs. Since the unit disc in C with
two boundary points removed can be holomorphically identied with the innite strip [0, 1] iR,
we redene D to be this innite strip. Then
2
(x, y) consists of the homotopy classes of maps n
from D into Sym
g
() such that n(1 R) T

, n(0 R) T

, and n(: i) = x, n(: +i) = y.


For a path J : [0, 1] (j, , \ ) of nearly symmetric, almost complex structures on Sym
g
() and

2
(x, y), we dene the moduli space
/
Js
() =
_
n :
n
:
+J
s
n
t
= 0
_
There is an R action on /
js
() given by (: n)(:, t) = n(: + i(t + :)). We dene the space of
unparameterized pseudoholomorphic discs to be the quotient

/
Js
() =
/
Js
()
R
Lemma 8.8. If n /
Js
() is a J
s
-holomorphic disc, then T(n) 0.
Proof. Since J
s
is (j, , \ )-nearly symmetric, we have that J
s
is an honest complex structure in
a neighborhood of each .
i
Sym
g1
(). Therefore the disc n must intersect these subvarieties
non-negatively.
Now we can set up the necessary Banach manifolds and vector bundles to show that /
Js
()
is a smooth manifold. For a vector bundle 1 over [0, 1] R and let j, 0 be real numbers and
/ 0 an integer. We can put a -weighted Sobolev norm on the space of sections of 1 dened by
[[[[
W
p,k

=
k

j=0
_
[0,1]R

(j)
(: +it)

c
(t)
d: dt
where : R R is some smooth function with (t) = [t[ for [t[ 1. The space of sections for
which this norm is nite is denoted by \
k,p

(1).
For j 2 let B

(x, y) be the space of maps n : [0, 1] R Sym


g
() in \
p,1
loc
satisfying the
following additional properties:
1. The curve n satises the boundary conditions n(1 R) T

, n(0 R) T

.
2. The curve n converges to the points x and y as t exponentially in the sense that there
is T 0 and functions

\
p,1

([0, 1] (, T], T
x
Sym
g
())
+
\
p,1

([0, 1] [T, ), T
y
Sym
g
())
such that for all t T we have
n(: it) = exp
x
(

(: it)) n(: +it) = exp


y
(
+
(: +it))
The space B

(x, y) is a Banach manifold modeled on \


p,1

([0, 1] R, R
2g
), and the tangent space
at n B

(x, y) is precisely
T
u
B

(x, y) =
_
\
p,1

(n

T Sym
g
()) : (1, t) T

, (0, t) T

_
93
For form the Banach vector bundle
E
: c B

(x, y) dened by (at least as a set)


c =
_
uB

(x,y)
n 1
p

(n

T Sym
g
())
which has characteristic ber 1
p

([0, 1] R, R
2g
). We dene a section 1 : B

(x, y) c of this vector


bundle by
1(n) =
_
n,
n
:
+J
s
n
t
_
Now we have /
Js
(x, y) is 1
1
(0
E
), where 0
E
is the zero section of the vector bundle. The vertical
dierential of this section at a curve n 1
1
(0
E
) is of the form (in coordinates)
11
u
: \
p,1

([0, 1] R, R
2g
) 1
p

([0, 1] R, R
2g
)
11
u
() =

:
+J
s

t
+ (
v
J(:))
n
t
This is a Fredholm operator, and so /
Js
(x, y) is a nite dimensional submanifold of B

(x, y). In
fact, the Fredholm index of 11
u
acting on the unweighted space ( = 0) is precisely the Maslov
index of n. Clearly the components of B

(x, y) are in correspondence with the homotopy classes in

2
(x, y), and the Fredholm index of 11
u
is constant on each of these components.
Theorem 8.1. Fix a Heegaard diagram (, , ), with and curves intersecting transversally,
and x data (j, , \ ). Then for generic path J
s
of (j, , \ )-nearly symmetric, almost complex struc-
tures we have /
Js
(x, y) is a smooth manifold.
Denition of

H1
For simplicity we will restrict to the case /
1
(Y ) = 0 so that H
1
(Y )

= H
2
(Y ) are nite.
Fix a pointed Heegaard diagram (, , , .) for Y such that the and curves intersect
transversally, and choose a Spin
c
structure s Spin
c
(Y ). We also x a generic complex structure
j on and a generic path J
s
of nearly-symmetric almost complex structures on Sym
g
(). Let

C1(, , s) be the Z,2 vector spaces generated by all x T

such that :
z
(x) = s. We give
this vector space a relative Z-grading by
gr(x, y) = j() 2n
z
()
where is any homotopy class in
2
(x, y). Since any other disc


2
(x, y) must be of the form

= /o we have
j( /o) 2n
z
( /o) = j() +/j(o) 2 (n
z
() +/n
z
(o)) = j() 2n
z
() + 2/ 2/
therefore gr(x, y) is independent of the chosen. We dene the dierential :

C1(, , s)

C1(, , s) by
x =

yTT

sz(y)=s

2
(x,y)
()=1
nz()=0
#

/() y
The second sum actually consists of at most one term since there is at most one Whitney disc

2
(x, y) such that j() = 1 and n
z
() = 0. Technically depends on J
s
, and so we may
sometimes write
Js
. Note that if y has non-zero coecient in x, then
gr(x, y) = j() 2n
z
() = 1
so lowers the grading by 1.
94
Theorem 8.2.
2
= 0
Now we have that (

C1(, , s), ) is a chain complex, and so we can dene its homology groups,
which we denote by

H1(, , s). We also dene

H1(, ) =

sSpin
c
(Y )

H1(, , s)
8.2 Invariance of Heegaard Floer Homology
8.2.1 Independence of Complex Structures
Theorem 8.3. Let Y be a closed oriented 3-manifold with xed Spin
c
structure s. Then the
homology group

H1(Y, , , s) is independent of the complex structure j on and the path J
s
.
Proof. Let us x a complex structure j on and take two paths J
s,0
, J
s,1
of (, j, \ )-nearly sym-
metric almost complex structures, and connect them via a path J : [0, 1] R (, j, \ ) such
that J
s,t
= J
s,0
for t 0 and J
s,t
= J
s,1
for t 1. For a Whitney disc
2
(x, y) we dene the
moduli space of time-dependent holomorphic representatives by
/
Js,t
() =
_
n : [0, 1] R Sym
g
() :
n
du
ds
+J
s,t
du
dt
= 0
_
We dene the map

Js,t
:
_

C1(, , s),

J
s,0
_

C1(, , s),

J
s,1
_
by

Js,t
(x) =

yTT

sz(y)=s

2
(x,y)
()=0
nz()=0
#/
Js,t
() y
This map is well-dened, so we have to show that it is a chain map and induces an isomorphism on
homology. To see that

Js,t
is a chain map we look at the boundary components of a compact 1-
dimensional space, namely /
Js,t
() when j() = 1. Composing

Js,t
with the boundary operators
we see that

Js,t

J
s,0
(x) =

Js,t

yTT

sz(y)=s

2
(x,y)
()=1
nz()=0
#

/
J
s,0
() y
=

y,zTT

sz(y)=sz(z)=s

2
(x,y)

2
(y,z)
()=1
(

)=0
nz()=nz(

)=0
#

/
J
s,0
() #/
Js,t
(

) z
95

J
s,1

Js,t
(x) =

J
s,1

yTT

sz(y)=s

2
(x,y)
(

)=0
nz(

)=0
#/
Js,t
(

) y
=

y,zTT

sz(y)=sz(z)=s

2
(x,y)

2
(y,z)
(

)=0
()=1
nz(

)=nz()=0
#/
Js,t
(

) #

/
J
s,1
() z
By Gromovs compactness we see that for
2
(x, z) with j() = 1 we have that /
Js,t
() can
be compactied to a 1-dimensional manifold /
Js,t
() with boundary
/
Js,t
() =
_
_

/
J
s,0
() /
Js,t
(

)
_
_

_
_

=
/
Js,t
(

)

/
J
s,1
()
_
_
where the unions are over those homotopy classes of discs such that j(

) = 0 and j() = 1. Now it


is clear that the number of boundary components of /
Js,t
() (modulo 2) is precisely the coecient
of z in (

Js,t

J
s,0

J
s,1

Js,t
)(x). Therefore

Js,t
is a chain map.
To show that the induced map on homology is an isomorphism we show that

Js,t

J
s,1t
is
chain homotopic to the identity map. To dene the chain homotopy we take a path J
s,t,
between
the juxtaposition (in the t coordinate) J
s,1t
J
s,t
path and the path J
s
(constant in t). We dene
the moduli space of this path to be
/
Js,t,
() =
_
[0,1]
/
Js,t()
()
This is a manifold of dimension j() + 1, and it already has some boundary components corre-
sponding to the = 0 and = 1 ends. In particular,
/
Js,t,
() = /
J
s,1t
() /
Js,t
() /
Js
() (8.1)
We dene the proposed chain homotopy

H
Js,t,
:
_

C1(, , s),

J
s,0
_

C1(, , s),

J
s,0
_
by

H
Js,t,
(x) =

yTT

sz(y)=s

2
(x,y)
()=1
nz()=0
#/
Js,t,
() y
Clearly this map is of degree 1. We want to check that the coecient of z in (

J
s,0


H
Js,t,
+

H
Js,t,

J
s,0
+

Js,t

J
s,1t
+id)(x) is zero by relating it to the number of boundary components
in a compact 1-dimensional manifold. The 1-dimensional manifold /
Js,t,
() (with
2
(x, z)
and j() = 0) can be compactied by adding broken ows. The boundary of this compactication
has three forms: two of these correspond to the boundary components of /
Js,t,
() as shown in
(8.1), and now we get an additional form from once broken ows. These broken ows are of the
form
_
_

/
J
s,0
() /
Js,t,
(

)
_
_

_
_

=
/
Js,t,
(

)

/
J
s,0
()
_
_
96
where the union is over homotopy classes of discs with j() = 1 and j(

) = 1. It is now clearly
that the boundary components of the compactication of /
Js,t,
() correspond to the coecient
of z in the chain homotopy formula, and hence is zero.
8.2.2 Independence of Isotopies of Attaching Circles
Now we need to check that the homology groups

H1(, , s) are invariant under Heegaard moves
in our diagram (, , ). Due to the construction of the chain complex

C1(, , s) we can only
consider pointed moves (those moves which never cross the xed base point . ), so we need to
make sure that pointed moves generate all moves.
Proposition 8.2. Any two Heegaard diagrams (, , , .) and (

, .

) that determine the


same 3-manifold can be transformed into each other via a nite sequence of pointed Heegaard moves.
Proof. We know that we can connect the two diagrams with regular Heegaard moves, so we claim
that each of these moves can be slightly altered to be a pointed move. Clearly stabilization can
be made to miss the base point, and we can slightly isotope the attaching band used in making a
handle slide to also miss the base point. So, we really need to only look at isotopies of the attaching
circles.
Suppose we slightly isotoped the attaching circle
1
to
1
, and in the process it crossed the base
point .. If we surger out the rest of the curves we obtain a torus with 2p 2 marked points. In
this torus
1
can clearly be isotoped to
1
without crossing the base point. However, each marked
point
1
crosses during this isotopy corresponds to a handleslide across the corresponding handle
in . Therefore we can replace the isotopy of
1
to
1
that crossed . with a sequence of handle
slides and isotopies that missed ..
Theorem 8.4. Let Y be a closed, oriented 3-manifold equipped with two Heegaard diagrams
(, , , .) and (,

, .) such that one can be obtained from the other via a pointed isotopy
of the attaching circles. Then

H1(, , )

=

H1(,

)
Proof. If the two Heegaard diagrams are isotopic via pointed isotopies that keep the and curves
transverse at all times, then this is equivalent to varying the metric (or equivalently the complex
structure) on , which we does not change the homology groups by theorem 8.3.
So, we need to consider isotopies that introduce new intersection points between the and
curves. We need to do this in a controlled manner, and the way we do this is via exact Hamiltonian
ows, so let us recall this machinery. Fix a Hamiltonian H
t
: A R such that H
t
= 0 for t , [0, 1].
This gives rise to a 1-parameter family of Hamiltonian vector elds A
t
dened by
(A
t
, ) = dH
t
Now this generates a 1-parameter family of dieomorphisms
t
: A A such that
d
t
dt
= A
t
Let us apply this to a special Hamiltonian. Take a smooth bump function / : R such that
supp / is a small neighborhood of a point on
1
and let ) : R [0, 1] be smooth with supp ) (0, 1).
The dene the Hamiltonian H
t
(r) = )(t)/(r) so that the corresponding 1-parameter family of
dieomorphisms
t
moves a small piece of
1
a little bit, but leaves everything else xed.
97
We now use the isotopies to introduce a pair of canceling intersection points. Suppose
1
is
slightly isotoped via an exact Hamiltonian ow to intersect with
1
transversally in an extra pair
of canceling points. We can assume this isotopy is supported in (
2

g
.). The isotopy

t
of induces an isotopy of the torus T

in Sym
g
() in the obvious way.
We can now dene Whitney discs with dynamic boundary conditions. In particular, a
t
-
Whitney disc from x T

to y
1
(T

) T

is a map n : D Sym
g
() satisfying (here we
are thinking of D as [0, 1] R)
n(1 +it)
t
(T

), for all t R
n(0 +it) T

, for all t R
n(: i) = x
n(: +i) = y
Let
t
2
(x, y) denote the set of homotopy classes of such discs, and let /
t
() be the moduli space
of pseudoholomorphic representatives of
t
2
(x, y). The reason we use these exact Hamiltonian
ows to move the curves is that one can prove that we still have energy bounds on holomorphic
discs, and so we can still form the Gromov compactications.
We can now dene a chain map

t
:

C1(, )

C1(

, )
by the formula

t
(x) =

y
1
(T)T

t
2
(x,y)
nz()=0
()=0
#/
t
() y
UNFINISHED
8.2.3 Independence of Handle Slides
In order to prove invariance under handle slides we need to introduce Whitney triangles and gener-
alize all of our results about Whitney discs to triangles. A Heegaard triple diagram of genus p is a
genus p surface with three p-tuples , , of homologically independent curves in . These curves
determine genus p handlebodies l

, l

, l

and 3-manifolds Y

= l

, Y

= l

, Y

=
l

. This information species a 4-manifold A

with boundary A

= Y

in the following way. Let be the standard 2-simplex with vertices

and edges c

, c

, c

where c
i
connects
j
to
k
and i ,= ,, i ,= /. If we thicken up by , then we can glue on the
pieces c
i
l
i
via
A

=
()

(c

(c

(c

)
(c

) (c

), (c

) (c

), (c

) (c

)
The corners of this manifold can naturally be smoothed out.
For x T

, y T

, z T

, we say that a map n : Sym


g
() is a Whitney
triangle connecting x, y, z if n(

) = x, n(

) = y, n(

) = z and n(c

) T

, n(c

) T

, n(c

)
T

. Let
2
(x, y, z) denote the homotopy classes of such maps.
There is an obstruction in the rst homology group of A

to the existence of Whitney triangles.


Let o, /, c : [0, 1] Sym
g
() be any paths in T

, T

, T

from x to y, y to z and z to z, respectively.


98
Then o + / + c is a 1-cycle in Sym
g
(). Its homology is not well-dened, but its image in the
quotient
H
1
(Sym
g
())
H
1
(T

) +H
1
(T

) +H
1
(T

= H
1
(A

)
is well-dened, and we denote it by c(x, y, z). We have that
2
(x, y, z) is non-empty if and only
if c(x, y, z) = 0. We can also dene the map n
z
:
2
(x, y, z) Z to be the algebraic intersection
number of a triangle with the subvariety . Sym
g1
(). The splicing action of discs can be
extended to splice a disc and a triangle together. In particular, if
2
(x, y) and
2
(y, z, w),
then we can splice these together to get a triangle
2
(x, z, w). Clearly n
z
is additive with
respect to this splicing action: n
z
( ) = n
z
() +n
z
().
We can also extend the denition of domain of a Whitney disc to Whitney triangles. In particu-
lar, T() is a 2-chain in consisting of the formal linear combinations of the regions ().
Such a linear combination is said to be a triply-periodic domain if the coecient of the region con-
taining . is zero and its boundary is a sum of , and curves. The triple (, , , , .) is said
to be weakly admissible if every non-zero triply-periodic domain has both positive and negative
coecients.
(todo: blah blah blah ... more to do on the above)
If (, , , ) is a weakly-admissible Heegaard triple, then we construct the map

:

C1(, )

C1(, )

C1(, )
by counting pseudoholomorphic triangles:

(x y) =

zTT

2
(x,y,z)
nz()=0
()=0
#/() z (8.2)
If we give

C1(, )

C1(, ) the dierential (x y) = x y +x y, then



)

is a chain
map. To verify this we need to show the following

(x y) +

)

(x y) +

)

(x y) = 0
for all x T

and y T

. This holds because there are three types of ends in the moduli
space /() when j() = 1 that correspond to splicing a disc onto one of the three vertices of the
triangle.
Let

1

denote the induced map on homology. These maps satisfy an important associative
law, which we state now. For this we take a Heegaard quadruple (, , , , , .) and so we get four
maps

)

,

)
,
,

)

,

)

from counting pseudoholomorphic triangles and their induced maps on


homology

1

,

1

,

1

,

1

. These maps satisfy

() ) =

1

(

1

())
This is shown by constructing a chain homotopy between the various compositions via counting
pseudoholomorphic rectangles. In particular, we can develop the theory of Whitney rectangles
n : Sym
g
() in the obvious way, and dene the set of homotopy classes of such rectangles by

2
(x, y, z, w). If is a homotopy class of Whitney rectangles, then /() will denote the space of
pseudoholomorphic representatives. We can now dene the map

:

C1(, )

C1(, )

C1(, )

C1(, )
99
by counting pseudoholomorphic rectangles:

(x y z) =

wTT

2
(x,y,z,w)
nz()=0
()=0
#/() w
We claim this map determines a chain homotopy between the maps

)

( ) ) and

(

)

()); that is, we need to show that

(x y z) +

((x y z)) +

)

(x y) z) +

)

(x

)

(y z)) = 0
Note that (x y z) = x y z +x y z +x y z, so the above expands to

(x y z) +

(x y z)+

(x y z) +

(x y z)+

(x y) z) +

)

(x

)

(y z)) = 0
The coecient of w in these six terms correspond to the six types of ends of the moduli space
/(), where
2
(x, y, z, w) is a rectangle with j() = 1. In particular, the rst four terms
correspond to splicing a disc onto one of the four corners of the rectangle, and the last two terms
correspond to the rectangle degenerating into two triangles meeting at a vertex in two dierent
ways.
We could continue these constructions to discuss pseudoholomorphic :-gons, and their asso-
ciative laws, but now we will discuss the handle slide invariance of Floer homology. First we look
at two model calculations that will be used in constructing the isomorphisms.
Example 8.1. . Take the standard Heegaard diagram for #po
1
o
2
: = 1# #1 is the p-fold
connected sum of the torus 1, and
i
,
i
are curves supported in the ith factor that bound discs
in the solid genus p handlebody (i.e. they are meridians) and intersect transversally in a pair of
canceling points. Let . be a base point far from the and curves. Orient these curves in
the same direction and label the positive and negative intersection points of
i
and
i
by n

i
.
We claim that, up to grading shifts,

H1(, )

= H

(T
g
). There are 2
g
intersect points in
T

, and we can write these as ordered pairs x = (r


1
, . . . , r
g
) such that r
i
= n
+
i
or n

i
. In
particular, for an integer 1 i p, take x, y T

such that r
i
= n
+
i
, j
i
= n

i
and r
j
= j
j
for
all , ,= i. The curves
i
and
i
bound two discs in , call them 1
1i
and 1
2i
, and let
1
,
2

2
(x, y)
be discs with T(
1
) = 1
1i
and T(
2
) = 1
2i
. It is clear that j(
1
) = j(
2
) = 1. Clearly, if

1
,

2

2
(y, x) are discs with T(

1
) = 1
1i
and T(

2
) = 1
2i
, then j(

1
) = j(

2
) = 1.
Now for any x, y T

and
2
(x, y) with n
z
() = we can nd the precise value of
j(). Recall that we can think of these discs as a products of the standard discs from r
i
to j
i
in . Let j(x) denote the number of n
+
i
s in x. Then of course we can assume r
i
= n
+
i
and
j
i
= n
+
i
for i j(x), hence we have j() = j(x) j(y). This gives us the relative gradings,
gr(x, y) = j(x) j(y), hence the generators are partitioned into p gradings. If we give the lowest
relative grading an absolute grading of 0, then we see that there are precisely
_
g
i
_
generators in
grading i.
Finally, we claim that = 0, which completes the proof of our claim. For any x T

let y be a generator with gr(x, y) = 1. If


2
(x, y) with n
z
() = 0, then 1 = gr(x, y) =
j() = j(x) j(y), hence x has one more + coordinate than y. This means there is an i for which
100
r
i
= n
+
i
, j
i
= n

i
and r
j
= j
j
for , ,= i. As we saw before there are exact two discs connecting
these points, and #

/ = 1 for both, hence y has zero coecient in x.


The generator of

H1(, , .) with highest grading is represented by the cycle

=
_
n
+
1
, . . . , n
+
g
_
.
We will use

to denote both the cycle and the homology class.


Example 8.2. . We take another Heegaard diagram of #po
1
o
2
, but this time with one handle
slid across another. In particular, let = 1
1
# #1
g
, with each 1
i
a torus, and let
i
be a curve
supported in 1
i
going once around the meridian of 1
i
. For i 1 let
i
be a curve supported in
1
i
going once around the meridian and intersecting
i
in a pair of canceling points. Finally, let
1
be the curve supported in 1
1
#1
2
that is the result of sliding
1
across
2
, and perturbed so that
it intersects
1
in a pair of canceling points and does not intersect either
2
or
2
. Again we can
orient the intersection points of the and curves with n

i
.
We claim that, up to grading shifts,

H1(, )

= H

(T
g
).
UNFINISHED
With these model calculations out of the way we can now prove handle slide invariance. Fix
a Heegaard diagram (, , ) for Y , let be a p-tuple of attaching circles such that
1
is a
handle slide of
1
across
2
and
i
(i 1) is a small perturbation of
i
intersecting in a pair
of canceling points, and is another p-tuple of attaching circles that are small perturbations of
the curves such that they intersect at a pair of canceling points. Then the Heegaard diagrams
(, , ), (, , ), (, , ) all describe Y , and the diagrams (, , ), (, , ), (, , ) all de-
scribe #po
1
o
2
. Let

denote the top grading generators of the Floer homology


of the latter three diagrams. We claim that the map ) :

C1(, )

C1(, ) dened by
) = )

) is a quasi-isomorphism (let 1 denote the induced map on homology). We


already know that the map p :

C1(, )

C1(, ) dened by p = )

) is a quasi-
isomorphism, and the associativity law gives us
1() = 1

)
= 1

( 1

))
= 1

(1

))
The last line is a composition of two isomorphisms, 1

) and 1

), which
was checked in our model calculations, hence 1 is an isomorphism. This concludes the proof of
invariance under handle slides.
8.2.4 Independence of Stabilizations
Theorem 8.5. Let (, , , .) be a pointed Heegaard diagram for Y , and let (#1,

, .) be a
stabilization, where 1 is a torus and

=
g+1
,

=
g+1
with
g+1
,
g+1
homologically
independent curves on 1 intersecting once. Then

H1(, )

=

H1(,
g+1
, ,
g+1
)
Proof. We can arrange this stabilization such that the torus 1 is glued to along a region contained
in the component of () containing the base point .. Let c #1 be the intersection point
of
g+1
and
g+1
, and let

=
g+1
and

=
g+1
. Let .

be the corresponding base point


in #1. We clearly have T

= T

c, so this gives us an identication of the chain


groups

C1(, , .) and

C1(

, .

). If x, y T

and x

= x c , y

= y c, then for
101

2
(x, y) with n
z
() = 0 and


2
(x

, y

) with n
z
(

) we have an identication of the moduli


spaces /(

) = /() c. Therefore the chain complexes are identical.


8.3 The Surgery Exact Sequence
For this section we mostly follow the presentations in [2] and [3]. Let / be a knot in a closed, oriented,
integral homology 3-sphere Y . The knot determines a meridian j in /, uniquely determined (up
to isotopy) by the property that j is not null-homologous in / but is null-homologous in /, and
a longitude / in /, uniquely determiend (up to isotopy) by the property that / is null-homologous
in Y / and #(/ j) = 1. Let Y
0
and Y
1
denote the 0- and +1-surgeries on /, i.e. we ll the
torus Y / with a torus whose meridian is mapped to / and j + /, respectively. The theorem we
will work to prove is the following.
Theorem 8.6. There is an exact sequence relating the Heegaard Floer homologies of Y, Y
0
and Y
1
:

H1(Y
0
)
b
F
0
//

H1(Y
1
)
b
F
1 ||z
z
z
z
z
z
z
z

H1(Y )
b
F
bbD
D
D
D
D
D
D
D
(8.3)
There is a considerable amount of machinery we need to set up before we can prove this. First
we need to construct special Heegaard diagrams for Y, Y
0
and Y
1
. In fact, we claim there is a
pointed Heegaard multi-diagram (, , , , , .) such that
1. The Heegaard diagrams (, , ), (, , ), (, , ) describe Y, Y
0
, Y
1
respectively.
2. The curves
i
,
i
,
i
, for i = 1, . . . , p 1, are small isotopic translates of each other that
pairwise, transversally intersect in a pair of canceling points.
3. The curve
g
is isotopic to the juxtaposition of
g
and
g
.
Note that the last two conditions imply that (, , ), (, , ), (, , ) are all Heegaard diagrams
of #(p1)o
1
o
2
. To construct these special Heegaard diagrams take a self-indexing Morse function
) on Y / with one index 0 critical point, p index 1 critical points and (p 1) index 2 critical
points, and set = )
1
(3,2). Recall that the attaching circles have the following description. The
curves are precisely where the ascending manifolds of index 1 critical points intersect , and
the curves are where the descending manifolds of index 2 critical points intersect . We ll the
missing torus in Y / by adding a 2-handle in three dierent ways, and then adding the 3-handle
in the unique way to obtain a closed manifold. First we attaching a 2-handle with attaching circle
j, thus obtaining Y . Next we attach a 2-handle with attaching circle /, obtaining Y
0
. Finally, we
attach a 2-handle with attaching circle j+/, obtaining Y
1
. Flowing these attaching circles down to
the surface gives us the circles
g
,
g
,
g
respectively. Let
i
and
i
be small isotopies of
i
(for
i = 1, . . . , p 1) so that these curves pairwise intersect transversally at a pair of canceling points.
We have now constructed the Heegaard multi-diagram (, , , , ) with the proposed properties.
Label the intersection points of the , and curves as in example 8.1
j

i
=
i

i

i
=
i

i
n

i
=
i

i
Also, the curves
g
,
g
,
g
pairwise intersect in exactly one point, so let
j
g
=
g

g

g
=
g

g
n
g
=
g

g
102
Recall the intersection points

=
_
j
+
1
, . . . , j
+
g1
, j
g
_
,

=
_

+
1
, . . . ,
+
g1
,
g
_
and

=
_
n
+
1
, . . . , n
+
g1
, n
g
_
are cycles in the top grading of

H1(, ),

H1(, ) and

H1(, ) respectively.
Recall that we have the following chain maps from counting pseudoholomorphic triangles

:

C1(, )

C1(, )

C1(, )

:

C1(, )

C1(, )

C1(, )

:

C1(, )

C1(, )

C1(, )

:

C1(, )

C1(, )

C1(, )
and their induced maps

1

,

1

,

1

,

1

on homology. Let

/

be the chain homotopy


between the two compositions

)

( ) ) and

)

(

)

( )), which is given


by counting pseudoholomorphic rectangles.
Now we can dene the maps in the surgery exact sequence. We dene

1 =

1

1
0
=

1

1
1
=

1

)
First we show that any composition of consecutive maps in this sequence is zero. If

C1(, ),
then one composition looks like

1
0


1() =

1

) =

1

(

1

))
where we used the associativity law in the last equality. So, if we can show that

1

) = 0,
then we will have shown that this rst composition is zero. We can do this directly by examining
the Heegaard triple (, , , ).
We think of this diagram as the rst p 1 torus summands in having three meridian inter-
secting pairwise in a pair of canceling points j

i
,

i
, n

i
(the curves being
i
,
i
,
i
, i = 1, . . . , p 1,
of course), and in the last summand we have three straight curves intersecting pairwise in one point
(the curves being
g
,
g
,
g
). Let us examine each of these summands as a model calculation.
Suppose 1 is a torus with three curves , , isotopic to a meridian and such that they pairwise
intersect in a pair of intersecting points j

, n

and with a base point . far from these curves.


Then there is a unique homotopy class of triangles connecting j
+
,
+
and n
+
such that its domain
is non-negative and n
z
= 0.
Now suppose 1 is a torus with three curves , , such that is isotopic to the meridian, is
isotopic to the longitude, is isotopic to + , and they pairwise intersect at one point, denoted
by j, , n. Then for each integer / 0 there are exactly two homotopy classes

k
of triangles
connecting j, and n such that n
z
(

k
) =
k(k1)
2
. To this we lift to the universal cover of 1, which
is R
2
. The lifts of , , become the lines r = i , j = i , and
_
j = r +i +
1
2
_
, for i Z. Then
we can take
+
k
to be the triangle bounded by
_
r = 0, j = 0, j = r +/ +
1
2
_
and

k
to be the
triangle bounded by
_
r = 0, j = 0, j = r / +
1
2
_
. Further, for each of these triangles there is a
unique holomorphic representative, and so j(

k
) = 0.
Now let us go back to our original situation. We can think of = 1
1
# #1
g
as a sum of tori
with the curves
i
,
i
,
i
supported in 1
i
. Then the situations in the rst p 1 summands of is
identical to the rst example we examined above, and the situation in the last summand of is
103
identical to the last example we examined above. Since the base point . is placed away from
the curves we can see that there are exactly two homotopy classes

of triangles from

and

such that n
z
(

) = 0 = j(

) and T(

) 0. These homotopy classes are obtained


by taking the product of the rst p 1 triangles between the points j
+
i
,
+
i
, n
+
i
, and then taking
the product with one of the homotopy classes between j
g
,
g
, n
g
. Further, each of these homotopy
classes admit the same number of pseudoholomorphic representatives, so #/(
+
) = #/(

).
Consider the quantity

)

), which we saw above that if this quantity is zero, then


we have

1
0


1 = 0. Referring to (8.2) for the denition of

)

we see that only the coecient of

can possibly have a non-zero coecient in



)

). However, the coecient of this


generator is the sum of the number of pseudoholomorphic triangles between

, which
is 0 mod 2. Therefore

)

) = 0.
We can use completely symmetrical arguments to show that the other two compositions

1
1


1
0
and

1
0


1 are both zero, therefore the surgery triangle is a chain complex. In order to show that
this chain complex has trivial homology we will use a slightly technical framework in homolog-
ical algebra. So, we take a moment to provide the necessary homological background for these
constructions.
Homological Algebra Background
Let (
1
,
1
) and (
2
,
2
) be chain complexes of Z,2 vector spaces. We are not assuming these
complexes are graded even though it would be easy to introduce the proper grading conventions.
If )
1
:
1

2
is a chain map, then we can construct the mapping cone of )
1
, denoted by
(`()
1
), ). This is the chain complex with chain group equal to
1

2
and dierential dened
by the matrix
=
_

1
0
)
1

2
_
It is easy to check that
2
= 0.
Proposition 8.3. The obvious maps in the sequence
0
2

`()
1
)


1
0
are chain maps and this sequence is exact. Further, the connecting morphism in the associated long
exact sequence is precisely )
1
.
Proof. The maps in the sequence are dened by (/) = (0, /) and (o, /) = o, which clearly makes
the sequence exact. It is easy to see that these maps are also chain maps. For the second part of
the proposition let us recall how the connecting morphism is constructed in general.
Suppose we have a short exact sequence of chain complexes
0 (,
A
)
f
(1,
B
)
g
(C,
C
) 0
We dene a map : H(C,
C
) H(,
A
) in the following way. If c C such that
C
c = 0, then
since p is surjective we can nd / 1 such that p(/) = c. But then 0 =
C
c =
C
p(/) = p(
B
/),
and since im) = ker p there is an element o such that )(o) =
B
/. Finally we see that
0 =
2
B
/ =
B
)(o) = )(
A
o), and since ) is injective this implies that
A
o = 0. So, we dene
104
([c]) = [o]. It is easy to see that this is a well dened map and gives us an exact sequence
H(,
A
)
f
//
H(1,
B
)
g
zzu
u
u
u
u
u
u
u
u
H(C,
C
)

ddI
I
I
I
I
I
I
I
I
Now applying this construction to our mapping cone short exact sequence we see that for o
1
,
([o]) is dened by nding an element (/, c) `()
1
) such that o = (/, c) = / (so (o, 0) will do just
ne), then nding an element d
2
such that (0, d) = (d) = (o, 0) = (
1
o, )
1
(o)) = (0, )
1
(o))
(so d = )
1
(o) will do just ne). Therefore ([o]) = [)
1
(o)], and this proves the proposition.
Proposition 8.4. The mapping cone is functorial and satises a naturality condition. Specically,
if )
1
:
1

2
and p
1
: 1
1
1
2
are two chain maps between chain complexes, and we have maps

1
,
2
such that the following diagram

1
f
1
//

1
1
g
1
//
1
2
(8.4)
commutes up to homotopy, then there is an induced map `(
1
,
2
) : `()
1
) `(p
1
). Further,
the short exact sequences associated to the cones `()
1
) and `()
2
) t into a commutative (up to
homotopy) diagram
0
//

2
//

`()
1
)
//
M(
1
,
2
)

1
//

//
0
0
//
1
2
//
`(p
1
)
//
1
1
//
0
Proof. We clearly want to dene the induced map `(
1
,
2
) by `(
1
,
2
)(o
1
, o
2
) = (
1
(o
1
),
2
(o
2
)).
It is easy to see that this map has the necessary properties.
Proposition 8.5. Let (
i
,
i
)
iZ
be a collection of chain complexes and let )
i
:
i

i+1
be a
chain map for each i Z satisfying
1. The composition )
i+1
)
i
is chain nullhomotopic via H
i
:
i

i+2
2. the map.

i
:= )
i+2
H
i
+H
i+1
)
i
:
i

i+3
is a quasi-isomorphism.
Then, H(`()
i
))

= H(
i+2
).
Proof. First we see that
i
is a chain map; that is,
i+3

i
=
i

i
. Messing with the compositions
we easily see

i
= )
i+2
H
i

i
+H
i+1
)
i

i
= )
i+2

i+2
H
i
+)
i+2
)
i+2
)
i
+H
i+1

i+1
)
i
=
i+3
)
i+2
H
i
+
i+3
H
i+1
)
i
=
i+3

i
105
We dene a map
i
: `()
i
)
i+2
by
i
(o, /) = H
i
(o) + )
i+1
(/) and a map
i
:
i
`()
i+1
)
by
i
(o
i
) = ()
i
(o
i
), H
i
(o
i
)). It is easy to check that these maps are chain maps, and we have

i+1

i
=
i
. Consider the following diagram

i
f
i
//
=

i+1

i+1
//
=

`()
i
)

i+1
//

i
f
i
//

i+1

i+1

i
f
i
//

i+1
f
i+1
//

i+1

i+2
f
i+2
//

i+2

i+3
f
i+3
//
=

i+4
=

i+3
f
i+3
//

i+4

i+4
//
`()
i+3
)

i+3
//

i+3
f
i+3
//

i+4
All the maps in this diagram are chain maps, and all but two squares are immediately seen to
commute. For the other two we can easily dene chain homotopies between the two dierent
compositions, so all the squares commute up to homotopy. If we pass two homology we see that
the top and bottom rows are exact, and the compositions of the columns (except for the middle
composition) are isomorphisms, so it follows by the ve lemma that
i+2

i
is also an isomorphism.
The fact that
i

i1
=
i1
and
i+2

i
are quasi-isomorphisms implies that
i
is an isomorphism,
and hence
i
is also a quasi-isomorphism. Therefore H(`()
i
))

= H(
i+2
).
Back to the Surgery Exact Sequence
We will now see that proposition 8.5 implies the surgery exact sequence. To do this we take
an innite family of , , curves, denoted by
(i)
,
(i)
,
(i)
, that are generic exact Hamiltonian
perturbations of the , , curves, and assume
(0)
= ,
(0)
= ,
(0)
= . Then let
3i
=

C1(,
(i)
),
3i+1
=

C1(,
(i)
) and
3i+2
=

C1(,
(i)
). Note that
3i
,
3i+1
,
3i+2
are the
Floer complexes for Y, Y
0
and Y
1
respectively. We dene the maps )
i
:
i

i+1
by counting
pseudoholomorphic triangles as before. The fact that )
3i+1
)
3i
, )
3i+2
)
3i+1
, )
3i+3
)
3i+2
are chain
null-homotopic comes from counting holomorphic rectangles, and the homotopies are given by
H
3i
:=

(i)

(i)

(i) (

(i)

(i)

(i)

(i) )
H
3i+1
:=

(i)

(i)

(i+1) (

(i)

(i)

(i)

(i+1) )
H
3i+2
:=

(i)

(i+1)

(i+1) (

(i)

(i+1)

(i+1)

(i+1) )
So we need to verify the second hypothesis of proposition 8.5. First we show that
i
is quasi-
isomorphic to
i+3
via a closest point map.
Proposition 8.6. For a small perturbation

of the curves let


be the top-grading generator


of

H1(,

). Then the chain map


) :

C1(, )

C1(,

)
dened by


)

)
induces an isomorphism on homology.
106
Proof. We assume that the perturbation is taken so that
i
and

i
intersect in a pair of canceling
points. The idea is to approximate ) =

)

) with a map that is clearly an isomorphism.


To do this we introduce R-ltrations on

C1(, ) and

C1(,

).
Let us x an area for on and an intersection point x
0
T

, and dene
T : T

R
by
T(x) = /(T()) n
z
() /()
where
2
(x, x
0
) and /(1) denotes the area of a region. This is clearly well-dened. If
/
1
(Y

) 0, then we need to require that /(1) = 0 for any periodic domain, which is possibly
to nd such an area form for any weakly admissible diagram. We can also assume that the signed
between the curves
i
and

i
is zero. For ,



C1(, ) we will write <

if
max
xTT

x has non-zero coecient in


T(x) < max
x

TT

has non-zero coecient in

T(x

)
We can repeat this construction to get a ltration on

C1(,

) relative to a reference point x

0
.
If the

i
is suciently close to
i
, then for each x T

there is a unique closest intersection


point x

. This association sets up a bijection between the sets T

and T

,
which we can extend to a linear isomorphism
:

C1(, )

C1(,

)
Of course this map is not necessarily a chain map, but we claim that the chain map ) can be
written as ) = + /, where / :

C1(, )

C1(,

) is a linear map such that / < . Since is


an isomorphism, this implies that ) is an isomorphism on the chain level, and since this map is a
chain map we have that the chain complexes

C1(, ) and

C1(,

) are isomorphic.
There is a unique triangle
0

2
(x,

, (x)) such that T(


0
) 0 and is supported in the
region between
i
and

i
. This triangle admits a unique holomorphic representative, so #/(
0
) =
0. We now dene the map / :

C1(, )

C1(,

) by
/(x) =

#/() (x)
where the sum is over all
2
(x,

, (x)) such that n


z
() = 0 and T() ,= /T(
0
) for some
integer / (i.e. is not supported in the domain of
0
. We clearly have ) = +/.
UNFINISHED
Proposition 8.6 gives us a quasi-isomorphism
i
:
i

i+3
obtained by counting triangles.
We claim that
i
is chain homotopic to )
i+2
H
i
+H
i+1
)
i
:
i

i+3
. Let us see this for i = 0,
and the general situation will follow easily. Let us write

for the
(0)
curves for what follows. We
have the following maps
= )

) :

C1(, )

C1(,

)
)
0
= )

) :

C1(, )

C1(, )
)
2
= )

) :

C1(, )

C1(,

)
H
0
= /

) :

C1(, )

C1(, )
H
1
= /

) :

C1(, )

C1(,

)
107
We want to show that )
2
H
0
+H
1
)
0
is chain homotopic to
0
. This comes from the associative
law for pseudoholomorphic rectangles. The associative law for rectangles states that for a Heegaard
multi-diagram (, , , , ,

) the following map


G =

/

() ) +

)

() )+

()

() ) +

)

())+

(

)

())
(8.5)
is null-homotopic as a map
G :

C1(, )

C1(, )

C1(, )

C1(,

)

C1(,

)
The null-homotopy is given by counting pseudoholomorphic pentagons. In particular, for the
Heegaard multi-diagram (, , , , ,

) we dene the map


1

:

C1(, )

C1(, )

C1(, )

C1(,

)

C1(,

)
by counting pseudoholomorphic pentagons
1

(x y z w) =

pTT

2
(x,y,z,w,p)
nz()=0
()=0
#/() p
We claim that 1

+1

is equal to the map G in (8.5). This follows from examining


the ends of /() for
2
(x, y, z, w, p) and j() = 1. In particular there are two kinds of
degenerations of a pentagon: a disc can break o from one of the ve vertices making a pentagon
spliced with a disc, or a triangle can break o from one of the vertices making a rectangle spliced
with a triangle.
We can now construct the chain homotopy between )
2
H
0
+H
1
)
0
and
0
. It is the map
1 :

C1(, )

C1(,

)
dened by
1 = 1

)
For

C1(, ) we have
1

) +1

) = G(

)
and by denition of G we have
G(

) =

) +

)

)+

( )

) +

)

))+

))
We can easily see that the rst term in this expansion is precisely H
1
)
0
(), the second term is
precisely )
2
H
0
(), the third and fth terms are zero, and nally we claim the fourth term is
identied with
0
(). It clearly suces to show that

) =

This follows from a model calculation.


108
Proof of theorem 8.6. We now have all of the necessary machinery to prove the surgery exact se-
quence. By proposition 8.5 we have that the chain complex `()
0
) is quasi-isomorphic to
2
. But,
`()
0
) =
0

1
=

C1(, )

C1(, ) and
2
=

C1(, ). The short exact sequence
0

C1(, ) `()
0
)

C1(, ) 0
leads to a long exact sequence

H1(, )
//
H(`()
0
))
zzu
u
u
u
u
u
u
u
u
u

H1(, )
ddI
I
I
I
I
I
I
I
I
This is the surgery exact sequence.
Z,2-Gradings
We can introduce Z,2-gradings in

C1(, ) and the homology groups, and the surgery exact
sequence is homogenous with respect to these gradings.
A Few Simple Applications
8.4 The Absolute Q Grading
If Y is a rational homology sphere, then the relative Z grading on

H1(Y ) can be lifted to an
absolute Q grading. Let \ be the cobordism from o
3
to Y by adding : 2-handles to 1
4
.
UNFINISHED
8.5 Heegaard Floer Knot Homology
The Construction
We can also construct an invariant of knots in 3-manifolds from Heegaard Floer homology. Let /
be a knot in a closed, oriented 3-manifold Y . First we need to come up with a Heegaard diagram
for Y that somehow encodes information about our knot /. To do this let us rst x a Heegaard
diagram (, ,
0
) for the knot complement Y /, where is a p-tuple of curves and
0
is a
(p 1)-tuple of curves. If j / is a meridian of /, then (, , ) is a Heegaard diagram for Y ,
where =
0
j. Let us place to base points ., n very close to j, but on opposite sides,
and away from the rest of the and curves. We can connect n to . with a small arc
1
that
crosses j transversally in one point, and we can connect . to n with an arc
2
supported in the
complement of the curves in (which is topologically just a disc). Then the curve
1

2
(as it
sits in o
3
) is ambiently isotopic to /, and if / is oriented it will inherit an orientation.
We will use the base point n to dene the regular Floer chain complex

C1(, , n) (i.e. the
dierential counts discs that miss n Sym
g1
()), whose homology we know consists of a single
copy of Z,2. Recall that we can lift the relative grading on these Floer chain groups to absolute
gradings on homology. In this absolutely graded theory, the Floer homology of o
3
is supported in
degree 0 and the Floer homology of o
1
o
2
is supported in degrees
1
2
. Let

C1
i
(, , n) denote
the subgroup of

C1(, , n) generated by homogeneous elements of

C1(, , n) of degree i.
109
Next, we use the base point . to dene a ltration on the graded group

C1(, , n). Dene
the function T : (T

)
2
Z acting on x, y T

by
T(x, y) = n
z
() n
w
()
for any disc
2
(x, y). This function can be lifted to a unique function T : T

Z with
the following two properties:
1. T(x, y) = T(x) T(y)
2. #x T

: T(x) = i #x T

: T(x) = i mod 2
Finally we dene the ltration T

on

C1(, , n) by letting T
j

C1(, , n)

C1(, , n) be the
subgroup generated by x with T(x) ,. This ltration has the following nice properties:
1. The ltration is decreasing: T
j

C1(, , n) T
j+1

C1(, , n).
2. T

is bounded: for large enough , we have T


j

C1(, , n) =

C1(, , n) and for small
enough , we have T
j

C1(, , n) = 0.
3. The ltration respects the absolute grading: T
j

C1
i
(, , n)

C1
i
(, , n).
4. The dierential :

C1(, , n)

C1(, , n) respects the ltration:
_
T
j

C1(, , n)
_

T
j

C1(, , n).
All of this information implies numerous things. First of all, we can construct the associated
bigraded vector space

C11
i
(, , ., n, ,) := T
j

C1
i
(, , n),T
j1

C1
i
(, , n)
The dierential on

C1

(, , n) descends to a dierential on

C11

(, , ., n, ), and so we can
dene the homology of this complex, which is a bigraded vector space

H11

(, , ., n, ). Fur-
ther, by the fundamental theorem of spectral sequences, we have that there is a spectral sequence
_
1
r
p,q
, d
r
_
of homological type with 1
2
=

H11(, , ., n) and converging to 1

=

H1(o
3
).
It turns out this entire spectral sequence is an invariant of the knot, and will be useful later for
dening additional scalar invariants.
This bigraded homology group turns out to be a topological invariant of the knot /, and so we
write

H11
i
(Y, /, ,) to denote the homology groups.
Proposition 8.7. If (, ,
0
, j) and (

0
, j

) represent the same knot complement, then


one representation can be obtained from the other via a nite sequence of the following moves (and
their inverses):
1. Handle slides and isotopies amongst the or
0
.
2. Isotopy of j.
3. Handle slides of j across the
0
curves.
4. Stabilizations.
Proof.
110
Theorem 8.7. For an oriented knot / in o
3
with compatible Heegaard diagram (, , , ., n), the
ltered chain homotopy type of

C11(,
0
, ., n) is a topological invariant of /, i.e. it does not
depend on the choice of Heegaard diagram and complex structures used in its construction.
Proof. We proceed like in the proof of invariance of

H1, except now we have to make sure that
all of our maps respect the ltration on

C11. First consider a perturbation of the path of nearly-
symmetric, almost complex structures J
s,0
and J
s,1
. Early we dened a continuation map
Js,t
:
(

C1(, ),
J
s,0
) (

C1(, ),
s,1
) by counting time-dependent pseudoholomorphic discs. Since
the ltrations on both of these complexes are precisely the same ....

Js,t
(x) =

yTT

2
(x,y)
()=0
nz()=0
#/
Js,t
() y
UNFINISHED
Extension to Links
Now we explain how to extend the above construction to links. Let 1 be an :-component link in Y ,
and let j
i
,
i

m1
i=1
be a set of points in 1 such that if we identify j
i
with
i
in 1 we get a connected
graph. We can think of these points as (:1)-embedded 0-spheres. Let \ be any null-cobordism
of Y , and let (Y ) be the boundary of the 4-manifold obtained by attaching (: 1) 1-handles
whose attaching spheres are precisely j
i
,
i
. Clearly (Y ) is dieomorphic to Y #(:1)o
1
o
2
.
Then we can connect (: 1) bands to 1 in (Y ) that start at j
i
, end at
i
and run through the
core of the i-th 1-handle, thus obtaining a knot (1) in (Y ). The isotopy class of this knot in
(Y ) depends only on the isotopy class of 1 in Y .
We now dene the Floer homology of a link 1 in Y as the Floer homology of (1) in (Y ), i.e.

H11(Y, 1) :=

H11((Y ), (1)). In order to get absolute gradings in

H11 for a link, recall the
following absolute grading of the Floer homology of #:o
1
o
2
(with i Z):

H1
i
(#:o
1
o
2
) =
_

_
(Z,2)
(
m
i+m/2
)
: even and :,2 i :,2
(Z,2)
(
m
i+m/21/2
)
: odd and :,2 i
1
2
:,2
0 otherwise
That is, the Floer homology of #:o
1
o
2
is the homology of the :-torus T
m
with gradings shifted
so that the homology is symmetric with respect to grading 0.
Exact Sequences for

H11
We can use the surgery exact sequence discussed from theorem 8.6 to derive exact sequences for
the knot homology invariant. Fix an oriented knot / in o
3
, and let be a knot in o
3
disjoint from
/ such that lk(/, ) = 0. The surgery exact sequence provides us with maps

1 :

H1(o
3
)

H1(o
3
0
())

1
0
:

H1(o
3
0
())

H1(o
3
1
())

1
1
:

H1(o
3
1
())

H1(o
3
)
111
that t into an exact triangle. Further, the map

1
0
and

1
1
are homogenous maps of degree 1,2,
whereas the map

1 is a sum of homogenous maps which do not increase the grading. It can be
shown that the chain maps on

C1 that induce the above maps respect the ltrations on

C1, and
hence the surgery exact sequence passes to an exact sequence in knot Floer homology:

H11(o
3
0
(), /)
b
F
0
//

H11(o
3
1
(), /)
b
F
1 xxp
p
p
p
p
p
p
p
p
p
p

H11(o
3
, /)
b
F
ffN
N
N
N
N
N
N
N
N
N
N
Example 8.3. We can use the above surgery sequence for knot Floer homology to make some simple
computations. Let 1 be the Borromean link in o
3
, and let 1(:, n) be the knot in 1(:, 1)#1(n, 1)
obtained as one of the components of 1 after doing surgery on the other two components with
framing : and n. For example, 1(:, ) is the unknot in 1(:, 1) for any :, and 1(1, 1) is the
right-handed trefoil in o
3
. We will compute later that the knot Floer homology of the right-handed
trefoil is precisely

H11
i
(o
3
, 1(1, 1), ,) =
_
Z,2 (i, ,) = (2, 1), (1, 0), (0, 1)
0 otherwise
Now we will compute

H11(o
1
o
2
, 1(0, 1)). UNFINISHED
For an oriented link 1 in o
3
, let 1
+
, 1

, 1
0
denote the oriented links that are the same as
1 except in the vicinity of a crossing, in which case 1
+
, 1

, 1
0
have that crossing changed to a
positive, negative or oriented smoothing, respectively. If 1 has : components, then 1
+
and 1

will also have : components, but 1


0
will have :+1 components. In order to talk about the Floer
homology of the links 1
+
and 1

we look at their associated knots in the space #(:1)o


1
o
2
,
whereas the associated knot of the link 1
0
lives in #:o
1
o
2
. Let be an unknot placed around
a crossing of 1 such that 1 intersects the spanning disc of in a pair of canceling points, and both
strands at this crossing belong to the same link component. The link 1

sitting inside the space


o
3
1
() is isotopic to 1
+
sitting in o
3
. We claim that 1

sitting in o
3
0
()

= o
1
o
2
is isotopic to
the link in o
1
o
2
obtained by attaching a 1-handle to o
3
near the smoothing in 1
0
, and then
connecting the two components of 1
0
near the smoothing with a compatibly oriented band going
through the 1-handle.
An Euler Characteristic Calculation
The skein exact sequences for Floer knot homology allow us to compute the Euler characteristic of

H11

, and compare it to a classical knot invariant.


Getting Heegaard Diagrams for a Certain Class of Knots
We will now explain how to genus 1 Heegaard diagrams for 2-bridge knots, and see that their Floer
homology can be computed combinatorially. A 2-bridge knot is a knot / i o
3
such that there is a
height function ) : o
3
R such that )[
k
has precisely two maxima.
112
x
1
x
2
x
3
z
w
x
1
x
2
x
3

Figure 8.1: A Heegaard diagram for the trefoil knot


The Trefoil Knot
Figure 8.1 shows a Heegaard diagram for o
3
that is compatible with the left-handed trefoil. The
intersection points of the and curves are labelled r
1
, r
2
and r
3
. Let us gure out the relative
grading of these generators, computed with respect to the base point n. For gr(r
1
, r
2
) note that we
have the obvious disc
12
which does not intersect n and has a unique holomorphic representative
(up to translation), hence gr(r
1
, r
2
) = 1. For gr(r
2
, r
3
) we also have the obvious disc
23
, but
now the orientations are reversed, so j(
23
) = 1, and this disc intersects n negatively, hence
gr(r
2
, r
3
) = 1 2(1) = +1. By additivity of gradings we have gr(r
1
, r
3
) = 2, hence the
gradings of the generators dier by one with r
1
in the top grading and r
3
in the bottom grading.
To pin down the translation indeterminacy of this grading we compute the homology of the complex
(

C1(, n), which we know is Z,2 supported in grading zero. In r


1
we clearly have the coecient
of r
2
is one (we do not need to consider r
3
since its grading is too low and lowers gradings by
only one). Next we have r
2
= 0 since the disc connecting r
2
to r
3
does not support a holomorphic
representative. Finally we have r
3
= 0 since r
3
in the lowest grading. We have now computed
ker = r
2
, r
3
) and im r
2
), hence

H1(, ) = r
3
), which implies gr(r
1
) = 2, gr(r
2
) = 1 and
gr(r
3
) = 0.
Now let us compute the relative Alexander gradings. Recall that this is dened by )(x, y) =
n
z
() n
w
(). For the disc
12
connecting r
1
to r
2
we clearly have )(r
1
, r
2
) = 1. The disc

23
connecting r
2
to r
3
intersects n negatively, hence )(r
2
, r
3
) = 1. And by additivity of this
grading we have )(r
1
, r
3
) = 2. The symmetry property this grading is supposed to have implies
that )(r
1
) = 1, )(r
2
) = 0, )(r
3
) = 1. Finally, we compute the homology of (

C11(, , ., n),
k
).
However, none of the discs connecting the intersection points miss the points . and n, hence the
dierentials are zero. Therefore

H11(, , ., n) = r
1
, r
2
, r
3
), and the gradings look like

H11
ij
1 0 1
2 r
1
)
1 r
2
)
0 r
3
)
It is clear that the Euler characteristic of this bigraded vector space is just the Alexander polyno-
mial:
k
(t) = t
1
1 +t.
113
x
1

x
2
x
3
x
4
x
5
x
2
x
3
x
5
x
1
z
w
x
4
Figure 8.2: A Heegaard diagram for the gure eight knot
The Figure Eight Knot
Figure 8.2 shows a Heegaard diagram for o
3
that is compatible with the gure eight knot. Let us
rst gure out the relative gradings on these generators with respect to the base point n. There
is a disc connecting r
1
to r
2
and it intersects n positively, so gr(r
1
, r
2
) = 1. The obvious disc
connecting r
2
to r
3
does not intersect n, and so gr(r
2
, r
3
) = 1. The obvious disc connecting r
3
to r
4
has orientations reversed, and so its Maslov index is 1 and intersects n negatively, hence
gr(r
3
, r
4
) = 1. Finally, the disc from r
4
to r
5
has orientations reversed and does not intersect n, so
gr(r
4
, r
5
) = 1. To pin down the absolute grading we compute the homology of (

C1(, , n), ).
The dierentials can easily be computed as
r
1
= r
4
r
2
= r
3
r
3
= 0
r
4
= 0
r
5
= r
5
Therefore ker = r
3
, r
4
, r
1
+r
5
) and im r
3
, r
4
), hence

H1(, , n) = r
1
+r
5
). This means
that r
1
and r
5
are in grading zero, and so gr(r
2
) = 1, gr(r
3
) = 0 and gr(r
4
) = 1.
Next we compute the relative Alexander gradings. If we just use the discs we used in computing
the Maslov grading then we see that the relative Alexander gradings are the same as the relative
Maslov gradings. In order for the symmetry property to hold for the maslov gradings we must
have )(r
1
) = )(r
3
) = )(r
5
) = 0, )(r
2
) = 1 and )(r
4
), hence the Alexander gradings are precisely
the Maslov gradings. Further, the dierential
k
on

C11(, , ., n) is zero, so we have computed

H11:

H11
ij
1 0 1
1 r
2
)
0 r
1
, r
3
, r
5
)
1 r
4
)
It is clear that the Euler characteristic of this bigraded vector space is the Alexander polynomial
of the gure eight:
k
(t) = t
1
3 +t.
114
x
1

x
2
x
3
x
4
x
5
x
6
x
7
x
2
x
3
x
5
x
6
x
7
x
1
z
w
x
4
Figure 8.3: A Heegaard diagram for the knot 5
2
A Knot with 5 Crossings
Figure 8.3 shows a Heegaard diagram for o
3
that is compatible with the 5
2
knot (a gure eight
knot with an extra twist). Let us rst gure out the relative gradings on these generators with
respect to the base point n. There is a disc connecting r
1
to r
2
that does not intersect n, so we
have gr(r
1
, r
2
) = 1, and similarly gr(r
2
, r
3
) = 1. However, the disc connecting r
3
to r
4
passes
through n positively, and so gr(r
3
, r
4
) = 1. Continuing this we can easily compute
gr(r
1
, r
2
) = 1 gr(r
2
, r
3
) = 1 gr(r
3
, r
4
) = 1
gr(r
4
, r
5
) = 1 gr(r
5
, r
6
) = 1 gr(r
6
, r
7
) = 1
Next we compute the homology of (

C1(, , n), ) so that we can gure out the absolute gradings


of our generators. We can compute the dierentials to be:
r
1
= r
2
+r
4
+r
6
r
2
= r
3
r
3
= 0
r
4
= r
3
+r
7
r
5
= r
2
+r
4
+r
6
r
6
= r
7
r
7
= 0
It is easy to see that the kernel of is 4 dimensional, generated by r
3
, r
7
, r
1
+r
5
, r
2
+r
4
+r
6
, while
its image is 3 dimensional, generated by r
3
, r
7
, r
2
+ r
4
+ r
6
. Therefore

H1(, , n) is generated
by r
1
+r
5
, and so has absolute grading 0. As a consequence we have gr(r
1
) = gr(r
5
) = 0, gr(r
2
) =
gr(r
4
) = gr(r
6
) = 1 and gr(r
3
) = gr(r
7
) = 2.
Now let us compute the Alexander gradings. We can use the same discs we used to compute
the Maslov gradings. In particular, the disc from r
1
to r
2
intersects . positively but misses n,
hence )(r
1
, r
2
) = 1, and similarly )(r
2
, r
3
) = 1. The disc from r
3
to r
4
intersects n positively,
but misses ., hence )(r
3
, r
4
) = 1. The disc from r
4
to r
5
intersects . negative and misses n,
so )(r
4
, r
5
) = 1. The disc from r
5
to r
6
intersects n negatively and misses ., so )(r
5
, r
6
) = 1.
Finally, the disc from r
6
to r
7
intersects . positively, and so )(r
6
, r
7
) = 1. So, the relative
Alexander gradings are equal to the relative Maslov gradings:
)(r
1
, r
2
) = 1 )(r
2
, r
3
) = 1 )(r
3
, r
4
) = 1
)(r
4
, r
5
) = 1 )(r
5
, r
6
) = 1 )(r
6
, r
7
) = 1
115
In particular, r
1
, r
5
are in the same grading, r
2
, r
4
, r
6
are in the same grading, and r
3
, r
7
are
in the same grading. By the symmetry property of the Alexander grading we are forced to have
)(r
1
) = )(r
5
) = 1, )(r
2
) = )(r
4
) = )(r
6
) = 0 and )(r
3
) = )(r
7
) = 1. Further, the dierential

k
on

C11(, , ., n) is trivial since all discs intersect either . or n, and so we have computed

H11:

H11
ij
1 0 1
0 r
1
, r
5
)
1 r
2
, r
4
, r
6
)
2 r
3
, r
7
)
It is clear that the Euler characteristic of this bigraded vector space is the Alexander polynomial
of the 5
2
knot:
k
(t) = 2t
1
3 + 2t.
UNFINISHED
116
A Homotopy Groups
Let us assume that we are working in the nice category of Hausdor, compactly generated, non-
degenerately based spaces. For based spaces (A, r
0
) and (Y, j
0
), let )
1
, )
2
: A Y be maps (not
necessarily based) and n : 1 Y a path. We say that )
1
and )
2
are freely homotopic along n
if there is a homotopy 1 : A 1 Y from )
1
to )
2
such that 1(r
0
, t) = n(t) for all t 1, and we
denote this by )
1

u
)
2
. That is, although we are allowing the base point of A to move during the
homotopy, we are restricting it to move along a specic path. If )
1
and )
2
are based maps, then n
is necessarily a loop. Here is a simple lemma to show this denition makes sense.
Lemma A.1.
1. For any map )
1
: A Y and path n : (1, 0) (Y, )
1
(r
0
)) there is a map )
2
: A Y such
that )
1

u
)
2
.
2. If )
1

u
)
2
, )
1

v
)
3
and n rel 0, 1, then )
1

const
)
3
.
3. If )
1

u
)
2
and )
2

v
)
3
, then )
1

vu
)
3
.
Proof.
1. By denition of non-degenerately based we have that (A, r
0
) is a bration. Therefore the
homotopy extension problem )
1
n : A 0 r
0
1 Y can be solved to give a map
1 : A 1 Y with the desired properties.
2.
This lemma gives us an action of
1
(Y, j
0
) on the based homotopy classes [A, Y ]

: for [)]
[A, Y ]

and [n]
1
(Y, j
0
), let [n][)] = [p], where p is any map such that )
u
p. This is well-dened
by the above lemma.
Proposition A.1. If Y is path connected, then the forgetful map [A, Y ]

[A, Y ] descends to a
bijection [A, Y ]

,
1
(Y, j
0
) [A, Y ]. If Y is simply connected, then the forgetful map is bijective.
117
B Duality Theorems
Theorem B.1 (Poincare Duality). If ` a closed, oriented n-manifold with fundamental class
[`] H
n
(`; Z), then the homomorphism
11 : H
p
(`; Z) H
np
(`; Z)
dened by [`] is an isomorphism.
This isomorphism is called the Poincare isomorphism. It is natural in the sense that if
) : ` is a map of closed, oriented manifolds such that )

[`] = [], then the following


diagram commutes
H
p
(`; Z)
PD
//
H
np
(`; Z)
f

H
p
(; Z)
PD
//
f

OO
H
np
(; Z)
The duality isomorphism leads to a bilinear pairing
H
p
(`; Z) H
np
(`; Z) Z
which is non-degenerate
We can drop the orientability assumption and get a Z,2-form of Poincare duality.
Theorem B.2. If ` is a closed n-manifold with a Z,2-orientation given by [`] H
n
(`; Z,2),
then the homomorphism
11 : H
p
(`; Z,2) H
np
(`; Z,2)
dened by [`] is an isomorphism.
We can also extend these duality theorems to manifolds with boundary.
Theorem B.3 (Poincare-Lefschetz Duality). Let ` be a compact, oriented n-manifold with bound-
ary with fundamental class [`, `] H
n
(`, `; Z). Then the homomorphisms
11 : H
p
(`, `; Z) H
np
(`; Z)
11 : H
p
(`; Z) H
np
(`, `; Z)
given by capping homology classes with [`, `] are isomorphisms.
Finally, a more general form of duality is Alexander duality.
Theorem B.4 (Alexander Duality). Let A be a closed subset of o
n
which is a CW complex. Then

H
i
(o
n
A)

=

H
ni1
(A)
118
C Cohomology with Homotopy
There is a way of dealing with cohomology classes in terms of homotopy classes of maps. Recall
that an Eilenberg-MacLane space of type (G, n) (where G is abelian if G 2) is a CW-complex
1(G, n) such that
i
(1(G, n)) = G if i = n and 0 otherwise. Hurewiczs theorem implies that
H
i
(1(G, n)) = 0 for all i < n and H
n
(1(G, n)) = G, and so the exact sequence from the universal
coecient theorem gives a natural isomorphism
H
n
(1(G, n); G)

= Hom(H
n
(1(G, n)), G)

= Hom(G, G)
Let H
n
(1(G, n); G) denote the cohomology class associated to id : G G in the above
isomorphisms. Then each homotopy class of map ) : A 1(G, n) gives a cohomology class
)

H
n
(A; G). This association turns out to be a natural isomorphism of abelian groups
[A, 1(G, n)]

= H
n
(A; G)
where [A, 1(G, n)] is given the canonical group structure from the H-group structure on 1(G, n) =
1(G, n 1). Naturality of the isomorphism means that for any map ) : A Y the following
diagram commutes
[A, 1(G, n)]

[Y, 1(G, n)]

oo
H
n
(A; G) H
n
(Y ; G)
f

oo
where )

denotes the induced map both on the homotopy class of maps and on cohomology.
In the special case of n = 2 and G = Z we get another view of cohomology. The 2nd degree co-
homology H
2
(A; Z) is isomorphic to [A, 1(Z)] = [A, C1

]. But, at the same time, the classifying


space of l(1) is 1l(1) = C1

, and so [A, C1

] is in one-to-one correspondence with complex


line bundles over A. So each cohomology class H
2
(A; Z) corresponds to a complex line bundle
1

A.
119
D Homeomorphisms of Surfaces
When constructing combinatorial pictures of 3-manifolds it will be useful to know something about
self-homeomorphisms of surfaces. We will be gluing together manifolds with boundary equal to a
surface of genus p along a self-homeomorphism of that surface, so in particular we only need to
know self-homeomorphisms of surfaces up to isotopy. This motivates the denition of the mapping
class group. For any manifold `, let Homeo(`) denote the group of orientation preserving, self-
homeomorphisms of `, and let Homeo
0
(`) denote the normal subgroup of Homeo(`) that are
isotopic to the identity. The quotient H(`) = Homeo(`), Homeo
0
(`) is called the mapping class
group of `.
The structure of H(`) is usually quite complicated, but in the case of the torus we have a
simple description of its mapping class group, and for a general surface we have an explicit set of
generators. Let be a surface of genus p, and let c be an embedded circle in . Associated to this
curve there are 2 elements of H(), which are inverses of each other, called the Dehn twists along c.
Take a small tubular neighborhood (c) of c in . Since (c) is an annulus, let : (c) be
a homeomorphism onto the standard annulus = . : 1 [.[ 2 such that (c) = . : [.[ = 2.
Then dene the homeomorphism
c
: by
(r) =
_
r r , (c)

1
_
(r) c
2i(|(x)|1)
_
r (c)
This determines an element of H(), and had we twisted in the other direction we would have
gotten the inverse to the above element. Further Dehn twists around isotopic curves are isotopic.
There is a standard family of 3p embedded curves on a surface of genus p, label them by

i
,
i
,
j
for i = 1, . . . , p and , = 1, . . . , p 1.
Theorem D.1 (Dehn-Lickorish). The group H(`) is generated by Dehn twists along the curves

i
,
i
,
i
.
Theorem D.1 gives us an explicit description of the mapping class group of the torus. Let
T = o
1
o
1
be the 2-torus. Let j = o
1
1 and = 1 o
1
be the standard meridian and
longitude on T. A homeomorphism ) : T T induces an isomorphism )

:
1
(T)
1
(T), where

1
(T) = ZZ is generated by j and . Every automorphism of ZZ is given by an integer-valued
matrix whose determinant is 1. If ) is orientation preserving, then the matrix representation of
)

has determinant +1. Further, isotopic homeomorphisms induce the same map on homology,
hence we have a homomorphism
: H(T) SL(2, Z) (D.1)
Proposition D.1. The map is an isomorphism.
Proof. We will only show the surjective part of this proposition. Any matrix SL(2, Z) can be
reduced to the identity by composing with elementary transformations. Over the integers these
transformations are of the form
_
1 1
0 1
_ _
1 0
1 1
_
Under the isomorphism (D.1) these matrices correspond to Dehn twists around j and , which we
know generates H(T) by theorem D.1. Therefore is surjective.
120
E Some Analysis
Let \ and \ be Banach spaces and ) : \ \ a map. We say that ) is a dierentiable at a point
\ if there is a bounded linear map d)
v
: \ \ such that
lim
h0
[[)( +/) )() d)
v
(/)[[
[[/[[
= 0
We call d)
v
the derivative of ) at . If d)
v
exists for all \ we obtain a map d) : \ 1(\, \),
where 1(\, \) is the Banach space of bounded linear maps \ \ given the operator norm: for
1 1(\, \) we set [[1[[ = sup [[1()[[ : \, [[[[ 1. Since 1(\, \) is another Banach space
we can discuss the dierentiable properties of d). In general, we say that ) is of class C
1
if ) is
dierentiable and d) is continuous (i.e. bounded), and inductively say that ) is of class C
k
if d) is
of class C
k1
.
A C
k
Banach manifold modeled on a Banach space \ is a topological space A with a cover
l

and collection of charts r

: l

1 which are bijections onto the open set r

(l

) such that
for any two charts (r

, l

) and (r

, l

) we have that r

r
1

: r

(l

) r

(l

) is
C
k
as a map between open sets of the Banach space 1. If we speak of just Banach manifolds, and
leave out its C
k
class, then we take the manifold to be C

. We can develop the notions of tangent


vectors and dierentials in the same way as nite dimensional manifolds.
Theorem E.1 (Implicit Function Theorem). Let ) : A Y be a C
k
map between Banach man-
ifolds. If j Y is a regular value, then )
1
(j) is a C
k
submanifold of A and T
x
)
1
(j) = ker d)
x
for all r )
1
(j).
A linear map T : \ \ between Banach spaces is said to be Fredholm if the kernel and
cokernel of T are nite dimensional, and the image of T is closed in \. Such maps are our
best chance at transferring problems in the innite dimensional setting to the nite dimensional
world. The index of a Fredholm operator is dened to be ind(T) = dimker T dimcoker T.
The space of Fredholm operators 1(\, \) forms an open set in 1(\, \), and the index function
ind : 1(\, \) Z is locally constant.
Let A be a topological space and 1(r) a statement which evaluates to true or false for each
r A. We say that 1(r) is true for generic r A if the solution set r A : 1(r) contains a
countable intersection of open dense sets. The innite dimensional version of Sards theorem can
now be stated.
Theorem E.2 (Sard-Smale). Let A and Y be separable Banach manifolds. Let ) : A Y be a
C
k
map such that d)
x
: T
x
A T
f(x)
Y is Fredholm of index | for all r A. Assume / 1 and
/ | + 1. Then a generic j Y is a regular value of ), i.e. d)
x
is onto for all r )
1
(j).
Let A be a Banach manifold modeled on \ and 1 another Banach manifold. A smooth surjective
map : 1 A of Banach manifolds is said to be a Banach \-bundle (with \ a Banach space) if
A is equipped with a covering l

and smooth charts

:
1
(l

) l

\ such that for


any two

we have

: (l

)\ (l

)\ is of the form (j, n) (j, p(j)(n)),


where p : l

G1(\) is smooth (here G1(\) is the group of topological linear isomorphisms


\ \). This implies that each ber 1
x
:=
1
(r) has the structure of a Banach space, which
is isomorphic to \ but not in any canonical way. We denote the zero section (i.e. all of the zero
vectors) of 1 by 0
E
.
Let : 1 A be a Banach \-bundle, and let us write the elements of 1 as pairs (r, c) A1
where (c) = r. The tangent space of 1 at a point (r, 0) in the zero section naturally splits as
T
(x,0)
1 = T
x
A 1
x
121
For a point r A let
x
: T
(x,0)
1 1
x
be the projection onto the second factor. If : : A 1
is a section ( : = id) and let r A be a point such that :(r) = 0. Then using the natural
identication above we dene the vertical part of the dierential d: to be the composition 1:
x
:=

x
d:
x
: T
x
A 1
x
(we use a capital 1 for the vertical dierential and a lower case d for the
regular dierential). The following result is very useful.
Theorem E.3. Let A and Y be separable Banach manifolds, 1 A Y a Banach bundle, and
: : A Y 1 a smooth section. If for each (r, j) :
1
(0
E
) we have
1. The vertical dierential 1:
(x,y)
: T
(x,y)
(A Y ) 1
(x,y)
is surjective.
2. The vertical dierential 1:
(x,y)
restricted to T
y
Y is Fredholm of index /.
Then for generic r Y , the set j Y : :(r, j) = 0 is a /-dimensional submanifold of Y . Further,
the restriction of 1:
(x,y)
to T
y
Y is surjective.
Proof. By the rst condition on : we already have that :
1
(0
E
) is a Banach submanifold of AY
by the inverse function theorem. Let : A Y A be the projection onto the rst factor.
Let (r, j) A Y such that :(r, j) = 0. Then, as we remarked before, the tangent space at
(r, j, 0) in 1 splits as T
(x,y,0)
1 = T
x
A T
y
Y 1
(x,y)
, and so the vertical dierential is given by
1:
(x,y)
=
(x,y)
d:
(x,y)
, where
(x,y)
is the projection of T
(x,y,0)
onto 1
(x,y)
.
UNFINISHED
For a region l R
n
we dene 1
p
(l) to be the space of functions ) : l R such that
[[)[[
p
:=
__
U
[)[
p
_
1/p
<
If ) 1
p
([o, /]) we say that p 1
p
([o, /]) is the weak derivative of ) if for every C
1
([o, /])
with (o) = (/) = 0 we have
_
b
a
)

=
_
b
a
p
More generally, if ) 1
p
(l) and is a multi-index, then we say that p 1
p
(l) is the -th weak
derivative of ) if for every C
1
c
(l) we have
_
U
)1

= (1)
||
_
U
p
where 1

=

||
x

1x

k
is the -th dierential operator. Weak derivatives are unique (up to a
set of measure zero), and if a function is already dierentiable then its weak derivative is equal
to its regular derivative. Therefore we also denote the weak derivative by 1

), or just )

for
l R. We can further generalize this ideal to functions in 1
p
(l, R
n
) by dening weak derivatives
for each component. The subset of 1
p
(l, R
n
) with / weak derivatives in 1
p
(l, R
n
) is denoted by
\
k,p
(l, R
n
).
Proposition E.1. If ) \
1,2
(l, R
n
) (l R) such that )

is continuous, then ) C
1
(l).
122
Proof. For a xed o l dene p : l R
n
by
p(r) =
_
x
a
)

(t) dt
Since )

is continuous we have that p is C


1
. But, then p

= )

and so p = ) almost everywhere.


Since \
1,2
embeds into C
0
we have that ) is continuous, hence p = ) everywhere, and so ) is
C
1
.
(todo: Sobolev norms, completions and embeddings)
Lemma E.1. Let A, Y, 7 be Banach spaces, 1 : A Y a bounded linear operator and 1 : A 7
a compact linear operator. If there is a constant C 0 such that
[[r[[
X
C ([[1r[[
Y
+[[1r[[
Z
)
for all r A, then 1 has closed range and nite dimensional kernel.
Proof.
123
F Some Exercises
Exercise F.1. Let /
1
, /
2
be oriented, disjoint knots in o
3
. The four denitions of lk(/
1
, /
2
)
presented in these notes are equivalent.
Solution.
1. Let j
1
be a meridian of /
1
and 1
2
be an oriented Seifert surface of /
2
. Perturb 1
2
slightly, if
necessary, so that /
1
and 1
2
intersect transversally in the interior of 1
2
. We claim that the
integer 1 such that [/
2
] = 1[j
1
] is precisely /
1
1
2
.
Exercise F.2. If / denotes the mirror of /, then
k
(t) =
k
(t).
Solution. Suppose / is the reection of / through a plane in R
3
. If 1 is a Seifert surface of
/, then let 1 denote the Seifert surface of / obtained by reecting through . If r
1
, . . . , r
2g
are
generators for H
1
(1), then let r
1
, . . . , r
2g
denote their reections, and these generate H
1
(1). We
have the relation
lk(r
i
, r
+
j
) = lk(r
i
, r
+
j
)
therefore o = o. So,

k
(t) = det(t
1/2
o t
1/2
o
T
) = det((t
1/2
o t
1/2
o
T
)) =
k
(t)
Exercise F.3.
k
1
#k
2
(t) =
k
1
(t)
k
2
(t)
Proof. If 1
1
, 1
2
are Seifert surfaces for /
1
, /
2
respectively, then 1
1
;1
2
is a Seifert surface for /
1
#/
2
.
We clearly have H
1
(1
1
;1
2
)

= H
1
(1
1
) H
1
(1
2
), so if we pick bases = r
1
, . . . , r
2g
and =
j
1
, . . . , j
2g
for H
1
(1
1
) and H
1
(1
2
) we can take their union to get a basis for H
1
(1
1
;1
2
). Further,
the push-os of the basis elements do not link the the elements, and vice-versa, so the Seifert
matrix o with respect to this basis is o
1
o
2
, where o
i
is the Seifert matrix of 1
i
. Therefore

k
1
#k
2
(t) =
k
1
(t)
k
2
(t).
Exercise F.4. If / is a knot then
k
(1) = 1, and if 1 is a link with more than one component
then
L
(1) = 0.
Proof. If o is a Seifert matrix of some Seifert surface of / with respect to some basis, then
k
(1) =
det(o o
T
). The (i, ,)-entry of o o
T
is lk(r
i
, r
+
j
) lk(r
j
, r
+
i
) = lk(r
i
, r
+
j
) +lk(r
i
, r

j
) = r
i
r
j
.
So, o o
T
is the matrix of the intersection pairing : H
1
(1) H
1
(1) Z in the basis r
i
, which
we know is a non-degenerate, anti-symmetric form, and so it is a direct sum of matrices
_
0 1
1 0
_
in some basis. The determinant of this matrix is clearly 1, and so it follows
k
(1) = 1.
We can repeat the same arguments for a link 1 with more than one component, except now we
claim the intersection pairing : H
1
(1) H
1
(1) Z is degenerate. If 1 is a Seifert surface for
1, then there is a non-trivial 1-cycle going once around one of the boundary components of 1
(since 1 has more than one connected components). However, every embedded curve in 1 can be
isotoped to completely miss the boundary of 1, hence the intersection product of with any other
cycle is zero, hence the intersection pairing is degenerate. It follows that so the determinant of any
matrix representation is 0, hence
L
(t) = 0.
124
Exercise F.5. The Alexander polynomial is zero for any split link.
Proof. If /
1
and /
2
are knots in o
3
that can be separated by a closed 3-ball, then 1
1
#1
2
is a Seifert
surface for /
1
/
2
, where 1
1
, 1
2
are Seifert surfaces for /
1
, /
2
respectively. If r
i
and j
i
are
bases for 1
1
and 1
2
respectively, then r
i
, j
j
, j is a basis for 1
1
#1
2
, where j is a meridian of the
tube used to connected 1
1
and 1
2
. This meridian does not link with any of the push-os of the
r
i
s and j
i
s, and vice-versa, therefore there is an entire column of zeros in t
1/2
o t
1/2
o
T
, and so

k
1
k
2
(t) = 0.
Exercise F.6. Let 1 be an oriented link, and let 1
+
, 1

, 1
0
denote the links obtained from 1 be
changing a xed crossing to a positive crossing, a negative crossing, or resolving the crossing in an
oriented fashion. Then we have the skein relation

L
+
(t)
L

(t) = (t
1/2
t
1/2
)
L
0
(t)
Proof.
Exercise F.7.
1. (/) = (/)
2. (/
1
#/
2
) = (/
1
) +(/
2
)
3. (/) is even for knots /
4. If /
1
and /
2
are concordant, then (/
1
) = (/
2
).
Proof.
1. If o is a Seifert matrix for /, then o is a Seifert matrix for /, so
/ = (o o
T
) = ((o +o
T
)) = (o +o
T
) = (/)
2. If o
1
, o
2
are Seifert matrices for /
1
, /
2
, then o
1
o
2
is a Seifert matrix for /
1
#/
2
, and so the
result follows.
3. First we claim that det(o +o
T
) ,= 0 so that o +o
T
has no zero eigenvalues. The Alexander
polynomial can be written as
k
(t) = o
0
+o
1
(t+t
1
)+ +o
n
(t
n
+t
n
), and since
k
(1) = 1
we must have o
0
is odd, and hence det(/) = det(o +o
T
) is odd. Since the number of positive
eigenvalues plus the number of eigenvalues of o + o
T
is equal to 2p, it follows that their
dierence is even.
4. Recall that /
1
and /
2
are said to be concordant if .....
Exercise F.8. For a knot / we have sign(
k
(1)) = (1)
(k)/2
.
Proof. We have
k
(1) = det(o +o
T
) for some Seifert matrix o of /, and the sign of this is
125
References
[1] R. Bott. Morse theory indomitable. Publications mathematiques de lI.H.

E.S., (68):99114,
1988.
[2] Peter Ozsvath and Zoltan Szabo. On the heegaard oer homology of branched double-covers.
arXiv:math/0309170v1 (math.GT), 2003.
[3] Peter Ozsvath and Zoltan Szabo. Lectures on heegaard oer homology. Clay Mathematics
Proceedings, 2005.
[4] J. Robbin and D. Salamon. The spectral ow and the maslov index. Bull. London Math Soc.,
27:133, 1995.
126