Lecture notes on Heegaard Floer homology

arXiv:2008.01836v1 [math.GT] 4 Aug 2020

Jennifer Hom

Abstract. These are the lecture notes for a course on Heegaard Floer homology
held at PCMI in Summer 2019. We describe Heegaard diagrams, Heegaard Floer
homology, knot Floer homology, and the relationship between the knot and 3-
manifold invariants.

Introduction
Heegaard Floer homology, defined by Ozsváth-Szabó [25], and knot Floer ho-
mology, defined by Ozsváth-Szabó [23] and independently Rasmussen [38], are
invariants of 3-manifolds and knots inside of them. These notes aim to provide
an overview of these invariants and the relationship between them.
We will describe our manifolds and knots via Heegaard diagrams, which we
define in Section 1. Any two Heegaard diagrams for the same manifold or knot
are related by a sequence of moves, called Heegaard moves, much like any two
projections for a knot are related by a sequence of Reidemeister moves.
From such diagrams, we will construct chain complexes for 3-manifolds (Sec-
tion 2) and knots (Section 3) whose chain homotopy type (and in particular, ho-
mology) is independent of the choice of Heegaard diagram. The proof that these
invariants do not depend on the choice of Heegaard diagram relies on showing
invariance under Heegaard moves.
From the knot invariant associated to a knot K in S3 , one can compute the
3-manifold invariant for any Dehn surgery along K; we discuss this relationship
in the case of integral surgery in Section 4.
Acknowledgements. I would like to thank the organizers of the 2019 Park City
Mathematics Institute for the opportunity to give the lecture series associated
with these notes, and the participants of PCMI for being an attentive audience. I
would also like to thank Miriam Kuzbary and Robert Lipshitz for helpful com-
ments on earlier versions of these notes.

1. Heegaard splittings and diagrams

1.1. Heegaard splittings Our goal is to define an invariant of closed 3-manifolds
and knots inside of them. In order to do this, we will need a way to describe our
manifolds and knots. We will do this using Heegaard diagrams. Throughout, we

The author was partially supported by NSF grant DMS-1552285 and a Sloan Research Fellowship.
2 Heegaard Floer homology

will assume that all of our 3-manifolds are closed and oriented, and that all of
our homeomorphisms are orientation preserving, unless stated otherwise. Much
of what follows comes from [41, Lecture 1] and [30, Sections 2 and 3].

Definition 1.1. A handlebody of genus g is a closed regular neighborhood of a

wedge of g circles in R3 .

Definition 1.2. Let Y be a 3-manifold. A Heegaard splitting of Y is a decomposition

of Y = H1 ∪f H2 where H1 , H2 are handlebodies and f is an orientation reversing
homeomorphism from ∂H1 to ∂H2 . The genus of the Heegaard splitting is the
genus of the surface ∂H1 or equivalently ∂H2 .

Example 1.3. Note that S3 = B3 ∪ B3 . This is a genus 0 Heegaard splitting of S3 .

Example 1.4. Consider S3 as the 1-point compactification of R3 . Consider the circle

consisting of the z-axis and the point at infinity. A regular neighborhood of this
union is a handlebody H1 of genus 1. The complement of H1 is also a handlebody
of genus 1. Together, these two handlebodies form a genus 1 Heegaard splitting
of S3 .

Theorem 1.5. Any closed, orientable 3-manifold Y admits a Heegaard splitting.

Proof. Consider a triangulation of Y. Let H1 be a closed regular neighborhood

of the 1-skeleton of the triangulation; since H1 is a neighborhood of a graph, it
is a handlebody. The complement H2 = Y − H1 is also a handlebody; namely it
is a neighborhood of the dual 1-skeleton, that is, the graph whose vertices are
the centers of the tetrahedra and whose edges are segments perpendicular to the
faces of the tetrahedra. 

Remark 1.6. Alternatively, a self-indexing Morse function on Y gives rise to a

Heegaard splitting of Y; see [30, Section 3], in particular, Exercise 3.5.

Definition 1.7. Two Heegaard splittings Y = H1 ∪f H2 and Y = H1′ ∪f ′ H2′ of Y are

homeomorphic if there exists a homeomorphism φ : Y → Y taking Hi to Hi′ .

Remark 1.8. One may also consider the following stricter notion of equivalence:
two Heegaard splittings Y = H1 ∪f H2 and Y = H1′ ∪f ′ H2′ of Y are isotopic if there
exists a map ψ : Y × [0, 1] → Y such that
(1) ψ|Y×{0} = idY ,
(2) ψ|Y×t is a homeomorphism for all t,
(3) ψ|Y×{1} sends Hi to Hi′ .
Note that if Y = H1 ∪f H2 and Y = H1′ ∪f ′ H2′ are isotopic, then they are homeo-
morphic (via ψ|Y×{1} ). The converse is false, since the homeomorphism φ : Y → Y
need not be isotopic to the identity.

Definition 1.9. Let Y = H1 ∪f H2 be a genus g Heegaard splitting of Y. A stabliza-

tion H1′ ∪f ′ H2′ of H1 ∪f H2 is the genus g + 1 Heegaard splitting of Y where H1′
 Jennifer Hom 3

consists of H1 together with a neighborhood N of a properly embedded unknot-

ted arc γ in H2 , and H2′ consists of H2 − N.

Exercise 1.10. Prove that the homeomorphism type (in fact, isotopy type) of a
stabilization of Y = H1 ∪f H2 is independent of the choice of γ.

The following theorem of Reidemeister and Singer highlights the importance

of stabilizations.

Theorem 1.11 ([39, 42]). Any two Heegaard splittings of Y become isotopic after suffi-
ciently many stablizations.

1.2. Heegaard diagrams We will describe a Heegaard splitting via a Heegaard

diagram H, as defined below.

Definition 1.12. Let H be a handlebody of genus g. A set of attaching circles for H

is a set {γ1 , . . . , γg } of simple closed curves in Σ = ∂H such that
(1) the curves are pairwise disjoint,
(2) Σ − γ1 − · · · − γg is connected,
(3) each γi bounds a disk in H.

Exercise 1.13. Show that Σ − γ1 − · · · − γg is connected if and only if [γ1 ], . . . , [γg ]

are linearly independent in H1 (Σ; Z).

See Figure 1.14 for an example of a set of attaching circles.

Figure 1.14. A set of attaching circles for the “obvious” handle-

body in R3 bounded by Σ.

Definition 1.15. A Heegaard diagram compatible with Y = H1 ∪f H2 is a triple

H = (Σ, α, β) such that
(1) Σ is closed oriented surface of genus g,
(2) α = {α1 , . . . , αg } is a set of attaching circles for H1 ,
(3) β = {β1 , . . . , βg } is a set of attaching circles for H2 .
We call (Σ, α, β) a Heegaard diagram for Y.

Remark 1.16. The convention in the field is to draw α-circles in red and β-circles
in blue.
4 Heegaard Floer homology

See Figure 1.17 for examples of Heegaard diagrams for RP3 . (Another example
of a Heegaard diagram is given in Figure 1.31 below.)
β β1

β2

α α1 α2
3
Figure 1.17. Left, a Heegaard diagram for RP . Right, a stabilization.

Given a Heegaard diagram (Σ, α, β) for Y, we can build a 3-manifold as follows.

Thicken Σ to Σ × [0, 1]. Attach thickened disks along αi × {0}, 1 6 i 6 g, and along
βi × {1}, 1 6 i 6 g.
Exercise 1.18. Verify that since α and β are sets of attaching circles, the boundary
of the resulting 3-manifold is homeomorphic to S2 ⊔ S2 .
Now fill in each of these boundary components with a copy of B3 ; there is a
unique way to do so, since any orientation preserving homeomorphism from S2
to itself is isotopic to the identity. The resulting 3-manifold is homeomorphic to
Y.
∼ H1 (Σ; Z)/h[α1 ], . . . [αg ], [β1 ], . . . , [βg ]i.
Exercise 1.19. Show that H1 (Y, Z) =
Exercise 1.20. Let H = (Σ, α, β) be a Heegaard diagram for Y. Compute H1 (Y; Z)
from H as follows. Choose an order and orientation on the α- and β-circles
and form the matrix M = (Mij ) where Mij is the algebraic intersection number
between the ith α-circle and the jth β-circle. Show that M is a presentation matrix
for H1 (Y; Z).
We now consider three ways to alter a Heegaard diagram that do not change
the associated 3-manifold. These moves are called Heegaard moves and consist
of isotopies, handleslides, and stabilizations/destabilizations. Isotopies and han-
dleslides do not change the genus of the Heegaard diagram, while stabilizations
(respectively destablizations) increase the genus by one (respectively decrease the
genus by one).
Let {γ1, . . . , γg } be a set of attaching circles for a handlebody H where ∂H = Σ.
The set {γ1 , . . . , γg } is isotopic to {γ1′ , . . . , γg′ } if there is a 1-parameter family of
disjoint simple closed curves starting at {γ1 , . . . , γg } and ending at {γ1′ , . . . , γg′ }.
A handleslide of, say, γ1 over γ2 produces a new set of attaching circles {γ1′ , γ2 , . . . , γg }
where γ1′ is any simple closed curve disjoint from γ1 , . . . , γg such that γ1′ , γ1 , and
γ2 cobound an embedded pair of pants in Σ − γ3 − · · · − γg . See Figure 1.21.
 Jennifer Hom 5

γ1 γ2
γ1′

Figure 1.21. A handleslide.

Here is another way to think of a handleslide. Suppose that γ1 and γ2 can be

connected by an arc δ in Σ − γ3 − · · · − γg . Let γ1′ be the connected sum of γ1 with
a parallel copy of γ2 , where the connected sum is taken along a neigborhood of
δ. See Figure 1.23.

Exercise 1.22. Prove that these two descriptions of handleslides agree (up to iso-
topy).

γ1 γ2 γ1′ γ2
δ

Figure 1.23. Another way to view a handleslide.

A stabilization of a Heegaard diagram (Σ, α, β) results in a Heegaard diagram

(Σ ′ , α ′ , β ′ )
where
(1) Σ ′ = Σ#T 2 , where T 2 = S1 × S1 ,
(2) α ′ = α ∪ {αg+1 } and β ′ = β ∪ {βg+1 }, where αg+1 and βg+1 are two
simple closed curves supported in T 2 intersecting transversally in a single
point.
We say that (Σ, α, β) is a destablilization of (Σ ′ , α ′ , β ′ ).

Exercise 1.24. Show that a stabilization of a Heegaard diagram corresponds to a

stabilization of the corresponding Heegaard splitting.
6 Heegaard Floer homology

We have the following standard fact about Heegaard diagrams (see, for exam-
ple, [25, Proposition 2.2]).

Theorem 1.25. Let (Σ, α, β) and (Σ ′ , α ′ , β ′ ) be two Heegaard diagrams for Y. Then
after applying a finite sequence of isotopies, handleslides, and stabilizations to each of
them, the two diagrams become homeomorphic (i.e., there is a homeomorphism Σ → Σ ′
taking α to α ′ and β to β ′ , setwise).

We will be interested in pointed Heegaard diagrams, that is, tuples (Σ, α, β, w)

where w is a basepoint in Σ − α − β. We now consider pointed isotopies, where
the isotopies are not allowed to pass over w, and pointed handleslides, where w
is not allowed to be in the pair of pants involved in the handleslide. We have the
following upgraded version of Theorem 1.25.

Theorem 1.26 ([25, Proposition 7.1]). Let (Σ, α, β, w) and (Σ ′ , α ′ , β ′ , w ′ ) be two

pointed Heegaard diagrams for Y. Then after applying a finite sequence of pointed iso-
topies, pointed handleslides, and stabilizations to each of them, the two diagrams become
homeomorphic.

1.3. Doubly pointed Heegaard diagrams We will also be interested in describ-

ing knots inside our 3-manifolds. For simplicity, we will focus on the case where
the ambient 3-manifold is S3 .

Definition 1.27. A doubly pointed Heegaard diagram for a knot K ⊂ S3 is a tuple

(Σ, α, β, w, z), where w, z are basepoints in Σ − α − β, such that
(1) (Σ, α, β) is a Heegaard diagram for S3 ,
(2) K is the union of arcs a and b where a is an arc in Σ − α connecting w
to z, pushed slightly into H1 and b is an arc in Σ − β connecting z to w,
pushed slightly into H2 .

See Figure 1.28 for an example of a doubly pointed Heegaard diagram for the
left-handed trefoil.

a •
w
• b
z
c

Figure 1.28. A doubly pointed Heegaard diagram for the left-

handed trefoil, −T2,3 . (The labelled points a, b, and c will be
used in Section 3.)
 Jennifer Hom 7

Given a knot diagram D for a knot K, one can obtain a doubly pointed Hee-
gaard diagram for K as follows. Suppose that D has c crossings. Forgetting the
crossing data of the diagram D yields an immersed curve C in the plane. The
complement of C is c + 2 regions in the plane, one of which is unbounded. Let
Σ be the boundary of a regular neighborhood of C in R3 ; note that Σ is a surface
of genus c + 1. For each of the bounded regions in the complement of C, we put
a β-circle on Σ. For each crossing of D, we put an α-circle on Σ as in Figure 1.29.
Lastly, we add an α-circle, say αc+1 , corresponding to a meridan of K. Place a
w-basepoint on one side of αc+1 and a z-basepoint on the other side. (Note that
which side one choose for w determines the orientation of K.) See Figure 1.30 for
an example.

Figure 1.29. A knot crossing and the corresponding portion of

the associated doubly pointed Heegaard diagram.

• •
w z

Figure 1.30. Another doubly pointed Heegaard diagram for the

left-handed trefoil, −T2,3.

Exercise 1.32. Show that the above construction yields a doubly pointed Heegaard
diagram for K ⊂ S3 .
8 Heegaard Floer homology

Figure 1.31. A Heegaard diagram for S3+5 (−T2,3 ).

In the aforementioned construction, if we replace the circle αc+1 with an n-

framed longitude and remove the basepoint z, we obtain a pointed Heegaard
diagram for S3n (K). See Figure 1.31 for an example.
We now consider doubly pointed isotopies, which are required to miss both w
and z, and doubly pointed handleslides, where neither w nor z is allowed to be
in the pair of pants involved in the handleslide. We have the following analog of
Theorem 1.26.

Theorem 1.33 ([23, Proposition 3.5]). Let (Σ, α, β, w, z) and (Σ ′ , α ′ , β ′ , w ′ , z ′ ) be

two doubly pointed Heegaard diagrams for K ⊂ S3 . Then after applying a finite sequence
of doubly pointed isotopies, doubly pointed handleslides, and stabilizations to each of them,
the two diagrams become homeomorphic.

Exercise 1.34. Find a sequence of doubly pointed Heegaard moves from Figure
1.28 to Figure 1.30.
In Sections 2 and 3, we will define invariants of 3-manifolds and knots in
S3 . These invariants will be defined in terms of pointed and doubly pointed
Heegaard diagrams, and invariance will follow from the fact that the invariants
remain unchanged under pointed and doubly pointed Heegaard moves.
We conclude this section with some additional exercises.
Exercise 1.35. Find a genus 1 doubly pointed Heegaard diagram for the figure
eight knot.

Exercise 1.36. Find a genus 1 doubly pointed Heegaard diagram for the torus knot
Tp,q .

Exercise 1.37. Let (Σ, α, β) be a Heegaard diagram for Y. What is the manifold
described by (−Σ, α, β)? By (Σ, β, α)?
 Jennifer Hom 9

Exercise 1.38. Let (Σ, α, β, w, z) be a Heegaard diagram for a knot K in S3 . What

is the knot described by (Σ, α, β, z, w)? By (−Σ, β, α, z, w)?

2. Heegaard Floer homology

2.1. Overview From a Heegaard diagram H for Y, we will build chain com-
c
plexes CF(H) and CF− (H) whose chain homotopy types are invariants of Y; the
former is a finitely generated chain complex over F and the latter is a finitely gen-
erated graded chain complex over F[U]. Throughout, F denotes the field Z/2Z
and U is a formal variable of degree −2. We denote the homology of these com-
c
plexes by HF(Y) and HF− (Y) respectively; the former is a graded vector space
over F and the latter is a graded module over F[U]. When discussing properties
that apply to either flavor of Heegaard Floer homology, will write HF◦ rather than
c HF− , HF+ , or HF∞ (the latter two of which are defined below). The material
HF,
in this section draws from [30, Sections 4-8] and [25, Sections 3 and 4].
For simplicity, we will consider the case where Y is a rational homology sphere.
The case b1 (Y) > 0 requires some additional admissibility assumptions on H; see
[25, Section 4.2]. Before defining these invariants, we discuss certain aspects of
their formal structure.
The Heegaard Floer homology of Y splits as a direct sum over spinc structures
on Y:
M
HF◦ (Y) = HF◦ (Y, s)
s∈spinc (Y)

See [30, Section 6] for a discussion of spinc structures in terms of homotopy

classes of non-vanishing vector fields. Note that spinc structures on Y are in (non-
∼ H1 (Y; Z). The above splitting on
canonical) bijection with elements in H2 (Y; Z) =
the level of homology comes from a splitting on the chain level.
c
Both HF(Y) and HF− (Y) are finitely generated and graded. Finitely generated
graded vector spaces are simply a direct sum of graded copies of F. Finitely gen-
erated graded modules over a PID are also completely characterized. (See, for
example, [36, Proposition A.4.3].) Since the only homogenously graded polyno-
mials in F[U] are the monomials Un , any finitely generated graded module over
F[U] is isomorphic to
M M
(2.1) F[U](di ) ⊕ F[U](cj ) /(Unj ),
i j

where F[U](d) denotes the ring F[U] where the element 1 has grading d.
Moreover, by [24, Theorem 10.1], we have that for a rational homology sphere
Y, for all s ∈ spinc (Y),
M
(2.2) ∼ F[U](d) ⊕
HF− (Y, s) = F[U](c ) /Unj ,
j
j
10 Heegaard Floer homology

that is, there is exactly one free summand in HF− (Y, s). We define the d-invariant
of (Y, s) to be
d(Y, s) = max{gr(x) | x ∈ HF− (Y, s), UN x 6= 0 ∀ N > 0}.
Note that d(Y, s) is equal to d in Equation (2.2). The U-torsion part is called
HFred (Y, s); that is, using the notation in Equation (2.2),
M
HFred (Y, s) = F[U](cj ) /Unj .
j

A rational homology sphere Y with HFred (Y, s) = 0 for all s ∈ spinc (Y) is called
an L-space.
Remark 2.3. Different grading conventions exist in the literature. We have chosen
our grading convention so that HF− (S3 ) =∼ F[U](0) , as opposed to the perhaps
more common F[U](−2) . Our grading convention choice simplifies certain formu-
las, such as the Künneth formula [24, Theorem 1.5]:
CF− (Y1 #Y2 , s1 #s2 ) ≃ CF− (Y1 , s1 ) ⊗F[U] CF− (Y2 , s2 ).
Note that our choice of grading convention also impacts the gradings on HFred (Y, s).
c
The chain complexes CF(H, s) and CF− (H, s) fit in the following U-equivariant
short exact sequence:
·U c
(2.4) 0 → CF− (H, s) −−→ CF− (H, s) → CF(H, s) → 0,
yielding the U-equivariant exact triangle:
·U
HF− (Y, s) HF− (Y, s)

c
HF(Y, s).
Remark 2.5. Equation (2.2) and the above exact triangle imply that for a rational
c
homology sphere Y, we have dim HF(Y) > |H1 (Y; Z)|; cf. Exercise 2.17. It follows
c
that dim HF(Y) = |H1 (Y; Z)| if and only if Y is an L-space.
We may also consider HF∞ (Y, s) = H∗ (CF∞ (Y, s)) where
CF∞ (Y, s) = CF− (Y, s) ⊗F[U] F[U, U−1 ].
Note that CF− (Y, s) ⊂ CF∞ (Y, s) and define CF+ (Y, s) to be the quotient
CF∞ (Y, s)/CF− (Y, s).
We have the following short exact sequence:
0 → CF− (Y, s) → CF∞ (Y, s) → CF+ (Y, s) → 0.
The above short exact sequence on the chain level induces the following U-equivariant
exact triangle:
HF− (Y, s) HF∞ (Y, s)
(2.6)
HF+ (Y, s).
 Jennifer Hom 11

Exercise 2.7. Use Equation (2.2) to show that when Y is a rational homology sphere,
HF∞ (Y, s) =∼ F[U, U−1 ]. Use this fact combined with Equation (2.6) to show that
if
M
HF− (Y, s) =∼ F[U](d) ⊕ F[U](c ) /Unj , j
j

then
M
∼ T+
HF+ (Y, s) = ⊕ F[U](cj +1) /Unj ,
(d+2)
j

where T + = F[U, U−1 ]/UF[U].

Remark 2.8. In light of Exercise 2.7, one may define HFred (Y, s) in terms of HF− (Y, s)
or HF+ (Y, s); this choice also affects the grading of HFred (Y, s). The most common
convention in the literature is to define HFred in terms of HF+ .

A smooth cobordism W from Y0 to Y1 (that is, a compact smooth 4-manifold

W with ∂W = −Y0 ⊔ Y1 ) induces a map from the Heegaard Floer homology of Y0
to the Heegaard Floer homology of Y1 . More specifically, given a spinc structure
t on W, we have a homomorphism
F◦W,t : HF◦ (Y0 , t|Y0 ) → HF◦ (Y1 , t|Y1 ),
which is defined via a handle decomposition of W (but does not depend on the
choice of handle decomposition). The cobordism map F◦W,t has a grading shift
depending only on W and t. See [28] for further details or [31, Section 3.2] for an
expository overview.
Let Z be a compact 3-manifold with torus boundary, and γ0 , γ1, and γ∞ three
simple closed curves in ∂Z such that
#(γ0 ∩ γ1 ) = #(γ1 ∩ γ∞ ) = #(γ∞ ∩ γ0 ) = −1.
See Figure 2.10 for an example of such a triple of curves (which justifies the choice
of subscripts on γ). Let Yi be the result of Dehn filling Z along γi for i = 0, 1, ∞;
that is, Yi is the union of Z and a solid torus, where γi is a meridian of the solid
torus. Then by [24, Theorem 9.12], we have an exact triangle
b
F0
c 0)
HF(Y c 1)
HF(Y
(2.9)
b
F∞ c ∞) b
F1
HF(Y

where b
Fi is the cobordism map associated to the corresponding 2-handle cobor-
dism. The analogous exact triangle also holds for HF+ .
12 Heegaard Floer homology

γ1
γ∞

γ0

Figure 2.10.

Remark 2.11. For the analogous exact triangle for the minus or infinity flavors,
one must work over the formal power series ring F[[U]] and semi-infinite Laurent
polynomials F[[U, U−1 ] respectively; see [14, Section 2].

2.2. The Heegaard Floer chain complex Now that we have completed our overview
of the structural properties of Heegaard Floer homology, we turn to the construc-
tion of the Heegaard Floer chain complex associated to a Heegaard diagram.
Let H = (Σ, α, β, w) be a pointed Heegaard diagram for Y where Σ has genus
g, and as usual α = {α1 , . . . , αg } and β = {β1 , . . . , βg }. We further require that the
α- and β-circles intersect transversally.
Consider the g-fold symmetric product
Symg (Σ) = Σ×g /Sg ,
where Sg denotes the symmetric group on g-elements. Points in Symg (Σ) consist
of unordered g-tuples of points in Σ.

Exercise 2.12. Even though the action of Sg on Σ×g is not free, show that the
quotient Symg (Σ) is a smooth manifold. (Hint: Fix a complex structure on Σ.
Then use the Fundamental Theorem of Algebra to define a map between ordered
and unordered g-tuples of complex numbers.)

We have two half-dimensional subspaces of Symg (Σ):

T α = α1 × · · · × αg and Tβ = β1 × · · · × βg .
c
The chain complex CF(H) is freely generated over F by Tα ∩ Tβ , that is, intersec-
tion points between Tα and Tβ . Note that points in Tα ∩ Tβ can be viewed in
Σ as unordered g-tuples of intersection points between the α- and β-circles such
that each α-circle, respectively β-circle, is used exactly once.

Exercise 2.13. Find all of the generators Tα ∩ Tβ in Figure 1.31, viewed as un-
ordered g-tuples of points in Σ.

We will also be interested in the following subspace of Symg (Σ):

Vw = {w} × Symg−1 (Σ).
 Jennifer Hom 13

This subspace can be viewed as unordered g-tuples of points in Σ such that at

least one point is w.
c
The differential ∂ : CF(H) c
→ CF(H) will count certain holomorphic disks in
g
Sym (Σ). Let D denote the unit disk in C, and let eα , respectively eβ , denote the
arc in ∂D with Re(z) > 0, respectively Re(z) 6 0. A Whitney disk from x to y is a
continuous map φ : D → Symg (Σ) such that
(1) φ(−i) = x,
(2) φ(i) = y,
(3) φ(eα ) ⊂ Tα ,
(4) φ(eβ ) ⊂ Tβ .
See Figure 2.14. Let π2 (x, y) denote the set of homotopy classes of Whitney disks
from x to y.
y

x
Figure 2.14. A schematic of a Whitney disk.

Exercise 2.15. Following [30, Section 5.1], there is a rather straightforward obstruc-
tion to the existence of a Whitney disk from x to y.
(1) Viewing x, respectively y, as an unordered g-tuple {x1 , . . . , xg }, respec-
tively {y1 , . . . , yg }, of points in α ∩ β, we may choose a collection of arcs
a ⊂ α such that ∂a = y1 + · · · + yg − x1 − · · · − xg and a collection of arcs
b ⊂ β such that ∂b = x1 + · · · + xg − y1 − · · · − yg . Then a − b is a 1-cycle in
Σ. Using Exercise 1.19, verify that the image of ǫ(x, y) = [a − b] ∈ H1 (Y; Z)
is well-defined.
(2) Show that H1 (Symg (Σ); Z) = ∼ H1 (Σ; Z). (Hint: See [25, Lemma 2.6].) Com-
bined with Exercise 1.19, conclude that
H1 (Symg (Σ); Z) ∼ H1 (Y; Z),
=
H1 (Tα ; Z) ⊕ H1 (Tβ ; Z)
and that if ǫ(x, y) 6= 0 ∈ H1 (Y; Z), then there cannot exist a Whitney disk
in Symg (Σ) from x to y.

A choice of complex structure on Σ induces one on Symg (Σ). Given φ ∈

π2 (x, y), let M(φ) denote the moduli space of holomorphic representatives of
φ. ([25, Proposition 3.9] ensures that under suitably generic conditions, M(φ) is
smooth.) The expected dimension of M(φ) is called the Maslov index of φ and de-
noted µ(φ). There is an R-action on M(φ) coming from complex automorphisms
of D that preserve i and −i. (One way to see this R-action is to use the Riemann
14 Heegaard Floer homology

mapping theorem to change the unit disk to the infinite strip [0, 1] × iR ⊂ C where
eα corresponds to {1} × iR and eβ to {0} × iR. Then the R-action corresponds to
b
vertical translations.) Let M(φ) b
= M(φ)/R. If µ(φ) = 1, then M(φ) is a compact
zero dimensional manifold [25, Theorem 3.18]. Let nw (φ) denote the algebraic
intersection between φ(D) and Vw .
c
We define a relative Z-grading, called the Maslov grading, on CF(H) as follows.
(Recall that a relative grading defines the difference between the gradings of two
elements.) Let φ ∈ π2 (x, y). Define
gr(x) − gr(y) = µ(φ) − 2nw (φ)
By [25, Proposition 2.15], this relative grading is well-defined whenever ǫ(x, y) =
0, where ǫ is the function defined in Exercise 2.15.
We may use the function ǫ to partition the intersection points in Tα ∩ Tβ .
These equivalence classes are in bijection with H1 (Y; Z) and hence are in bijection
with spinc (Y); see [25, Section 2.6] for more details. In particular, we have a
c
splitting CF(H) c
= ⊕s∈spinc (Y) CF(H, s).
c c
The differential ∂ : CF(H, s) → CF(H, s) is defined to be
X X
∂x = b
#M(φ) y.
y∈Tα ∩Tβ φ∈π2 (x,y)
µ(φ)=1
nw (φ)=0

Note that in the first summation, it suffices to only consider y with ǫ(x, y) = 0;
c
that is, the differential respects the splitting CF(H) c
= ⊕s∈spinc (Y) CF(H, s). It
follows from the definition of the relative Maslov grading that the differential ∂
lowers the Maslov grading by one. By [25, Theorem 4.1], ∂2 = 0. Let
c
HF(H, c
s) = H∗ (CF(H, s)).
c
Remark 2.16. We may define a relative Z/2Z grading on CF(H) that agrees with
the mod 2 reduction of the relative Maslov grading as follows. Following [24,
Section 5], choose an orientation of Tα and Tβ and define the relative Z/2Z
grading between x, y ∈ Tα ∩ Tβ to be the product of their local intersection
numbers. (Here, we are identifying Z/2Z with {±1} under multiplication.)

Exercise 2.17. Let H be a Heegaard diagam for a rational homology sphere Y. Use
c
Remark 2.16 to prove that χ(HF(H)) c
= ±|H1 (Y; Z)| and conclude that dim HF(H) >
|H1 (Y; Z)|. (Hint: Use Exercise 1.20.) Compare with Remark 2.5.

We now define the chain complex CF− (H), which is freely generated over
F[U] by Tα ∩ Tβ . Here, U is a formal variable with gr(U) = −2. We no longer
require that Whitney disks miss the basepoint (i.e., we remove the nw (φ) = 0
requirement), and instead use the variable U to count the algebraic intersection
number nw (φ) of φ(D) and Vw :
 Jennifer Hom 15

X X
∂x = b
#M(φ)Unw (φ)
y.
y∈Tα ∩Tβ φ∈π2 (x,y)
µ(φ)=1

The definition of ∂ is extended to all elements of CF− (H) by F[U]-linearity. As in

c the chain complex CF− (H) splits over spinc (Y), that is, CF− (H) =
the case of CF,
⊕s∈spinc (Y) CF− (H, s). By [25, Theorem 4.3], ∂2 = 0. Let
HF− (H, s) = H∗ (CF− (H, s)).

Theorem 2.18 ([25, Theorem 1.1]). Let H be a pointed Heegaard diagram for Y. Then
c
the isomorphism type of HF(H, s) and HF− (H, s) is an invariant of Y and s.
c
In order to prove the above theorem, one must show that HF(H, s) is independent
of the choice of Heegaard diagram, basepoint, and complex structure. Indeed,
Ozsváth-Szabó show that Heegaard moves induce homotopy equivalences on the
associated chain complexes, as do changes in complex structure.
c
Remark 2.19. Note that CF(H) = CF− (H)/(U = 0), giving rise to the short exact
sequence in (2.4).

In [28, Theorem 7.1], Ozsváth-Szabó prove that the relative Z-grading may
be lifted to a well-defined absolute Q-grading; this is done by considering a
cobordism from S3 to Y and considering holomorphic triangles associated to a
Heegaard triple.
We conclude this section with some computations.
c
Exercise 2.20. Compute HF(L(p, q), s) and HF− (L(p, q), s) for all s ∈ spinc (L(p, q)).
c
In general, computing HF(Y) and HF− (Y) from a Heegaard diagram is not easy.
c
In Section 4, we will see how to compute HF(Y) and HF− (Y) when Y is surgery
on a knot in S3 .

Exercise 2.21. Recall that pq ± 1 surgery on the torus knot Tp,q is a lens space. Use
c 3 (Tp,q )) for n > pq − 1.
Exercise 2.20 and Equation (2.9) to compute HF(S n

Exercise 2.22. Let H = (Σ, α, β, w) be a Heegaard diagram for Y. Let H ′ =

(−Σ, β, α, w). Show that the chain complex CF− (H) is isomorphic to CF− (H ′ ).

Exercise 2.23. Use Exercise 1.37 and the fact that d(−Y, s) = −d(Y, s) [24, Propo-
sition 4.2] to determine the relationship between HF− (Y) and HF− (−Y) as abso-
lutely graded modules.

In general, direct computations of Heegaard Floer homology are difficult. Sarkar

and Wang [40] proved that HF c can be computed combinatorially using nice dia-
grams. Lipshitz-Ozsváth-Thurston [13] give an alternative combinatorial method
for computing HF c using bordered Floer homology, a version of Heegaard Floer
homology for 3-manifolds with boundary. Manolescu-Ozsváth-Thurston [17]
16 Heegaard Floer homology

show that all flavors of Heegaard Floer homology are algorithmically computable
in terms of a link surgery description for a 3-manifold and a grid diagram for that
link. In Section 4, we will see how to compute HF− of surgery on a knot in S3 in
terms of the knot Floer complex, a knot invariant which we describe in Section 3
below.
Lastly, we note that Theorem 2.18 merely states that the isomorphism type of
HF− (Y) is an invariant of Y. Juhász-Thurston-Zemke [10] prove that Heegaard
Floer homology in fact assigns a concrete group (that is, not just a group up to
isomorphism) to a 3-manifold Y equipped with a basepoint.

3. Knot Floer homology

3.1. Overview Let K be a knot in S3 . The simplest version of the knot invariant
[
is HFK(K), a bigraded vector space over the field F = Z/2Z:
M
[
HFK(K) = [ m (K, s),
HFK
m,s∈Z

[
where HFK(K) is supported in finitely many bigradings. Knot Floer homology
categorifies the Alexander polynomial [23, Equation (1)] in the sense that the
[
graded Euler characteristic of HFK(K) is ∆K (t):
X
∆K (t) = m
(−1) dim HFK [ m (K, s) ts .
m,s

Moreover, knot Floer homology strengthens two key properties of the Alexander
polynomial. Let
X
∆K (t) = a0 + as (ts + t−s )
s>0
denote the symmetrized Alexander polynomial. While the Alexander polynomial
gives a lower bound on the genus of K in the following manner:
g(K) > max{s | as 6= 0},
knot Floer homology actually detects g(K) [22]:
(3.1) [
g(K) = max{s | HFK(K, s) 6= 0}.
Similarly, while the Alexander polynomial obstructs fiberedness in that
K is fibered ⇒ ag(K) = ±1,
knot Floer homology actually detects fiberedness [3, 19]:
[
K is fibered ⇔ HFK(K, g(K)) = F.
3.2. The knot Floer complex We now modify the constructions in Section 2 to
the case of doubly pointed Heegaard diagrams in order to define knot invari-
ants. As mentioned above, for simplicity, we restrict ourselves to knots in S3 .
(With mild modifications, the constructions described here apply to any null-
homologous knot in a rational homology sphere.) Some of this material comes
 Jennifer Hom 17

from [30, Section 10]; see [23] for more details and proofs. Many of our conven-
tions and notations come from [45]; see especially [45, Section 1.5]. Knot Floer
homology was independently defined by Rasmussen in [38].
Recall that in the differential of the chain complex CF− , the variable U recorded
information about the basepoint w. Now that we have two basepoints, we will
work with the two variable polynomial ring F[U, V]. We endow this ring with a
bigrading gr = (grU , grV ). We call grU the U-grading and grV the V-grading. The
variables U and V have grading
gr(U) = (−2, 0) and gr(V) = (0, −2).
It will often be convenient to consider the following linear combination of grU
and grV ,
1
A = (grU − grV ),
2
called the Alexander grading. Note that A(U) = −1 and A(V) = 1.
Let H be a doubly pointed Heegaard diagram for a knot K ⊂ S3 . Let CFK F[U,V] (H)
be the free F[U, V]-module generated by Tα ∩ Tβ . This module is relatively bi-
graded as follows. Let φ ∈ π2 (x, y) and define
grU (x) − grU (y) = µ(φ) − 2nw (φ)
grV (x) − grV (y) = µ(φ) − 2nz (φ).
By [24, Proposition 7.5] (see also [45, Section 5.1]), this relative grading is well-
defined. The relative gradings grU and grV can be lifted to absolute Z-gradings,
using the absolute grading on HF− (S3 ); we describe this process below.
The differential ∂ : CFKF[U,V] (H) → CFK F[U,V] (H) is defined to be
X X
∂x = b
#M(φ)U nw (φ) nz (φ)
V y,
y∈Tα ∩Tβ φ∈π2 (x,y)
µ(φ)=1

and is extended to all elements of CFKF[U,V] (H) by F[U, V]-linearity. Note that
the differential preserves the Alexander grading.
Setting V = 1 and forgetting grV (that is, only considering the grading grU ), we
recover CF− (S3 ), whose homology is isomorphic to F[U](0) , where the subscript
(0) now denotes grU (1); this determines the absolute U-grading. Note that setting
V = 1 corresponds to forgetting the z-basepoint. To determine the absolute V-
grading, we simply reverse the roles of U and V in the above construction. That
is, we set U = 1, forget grU , and only consider grV , recovering CF− (S3 ), whose
homology is isomorphic to F[V] where grV (1) = 0. This corresponds to forgetting
the w-basepoint.

Theorem 3.2 ([23, Theorem 3.1]). Let H be a doubly pointed Heegaard diagram for a
knot K ⊂ S3 . The chain homotopy type of CFK F[U,V] (H) is an invariant of K ⊂ S3 .

Note that [23, Theorem 3.1] is phrased in terms of filtered chain complexes;
see [45, Section 1.5] for a description of the translation between filtered chain
18 Heegaard Floer homology

complexes and modules over F[U, V]. We will often abuse notation and write
CFKF[U,V] (K) rather than CFK F[U,V] (H).
Example 3.3. Figure 1.28 shows a doubly pointed Heegaard diagram for the left-
handed trefoil. We have
∂a = Ub
∂b = 0
∂c = Vb.
Setting V = 1, we see that the homology, which is isomorphic to F[U], is generated
by [a + Uc], implying that grU (a) = grU (Uc) = 0. Setting U = 1, we see that the
homology, which is isomorphic to F[V], is generated by [c + Va], implying that
grV (c) = grV (Va) = 0. It follows that the generators a, b, c have the following
gradings:
grU grV A
a 0 2 −1
b 1 1 0
c 2 0 1
∼
Then H∗ (CFKF[U,V] (−T2,3 )) = F[U, V](0,0) ⊕ F(1,1) , where the subscript denotes
gr = (grU , grV ) of 1. The F[U, V] summand is generated by Va + Uc and the F
summand by b.
Exercise 3.4. Let H = (Σ, α, β, w, z) be a doubly pointed Heegaard diagram for
K ⊂ S3 .
(1) Let
(CFK F[U,V] (H) ⊗F[V] F[V, V −1 ], s)
denote the summand of CFK F[U,V] (H) ⊗F[V] F[V, V −1 ] in Alexander grad-
ing s, thought of as an F[W]-module, where W = UV. Note that multi-
plication by UV preserves the Alexander grading on CFKF[U,V] (H). Let
CF− (H) = CF− (Σ, α, β, w). Show that there is an isomorphism of F[W]-
modules
∼ CF− (H),
(CFK F[U,V] (H) ⊗F[V] F[V, V −1 ], s) =
where the right-hand side is viewed as a module over F[W], rather than
F[U]. (Hint: Consider the map CF− (H) → (CFKF[U,V] (H) ⊗F[V] F[V, V −1], s)
given by W n x 7→ Un V n+s−A(x) x.) Conclude that
∼ F[U, V, V −1].
H∗ (CFKF[U,V] (H) ⊗F[V] F[V, V −1 ]) =
(2) Repeat part (1) reversing the roles of U and V.
The knot Floer complex behaves nicely under connected sum, reversal, and
mirroring. By [23, Theorem 7.1], we have that
CFKF[U,V] (K1 #K2 ) ≃ CFKF[U,V] (K1 ) ⊗F[U,V] CFKF[U,V] (K2 ).
 Jennifer Hom 19

Let Kr denote the reverse of K and mK the mirror. By [23, Section 3], we have that
(3.5) CFKF[U,V] (mK) ≃ CFK F[U,V] (K)∗ ,
where C∗ = HomF[U,V] (C, F[U, V]) and
(3.6) CFK F[U,V] (Kr ) ≃ CFK F[U,V] (K).

Exercise 3.7. Let H = (Σ, α, β, w, z) be a doubly pointed Heegaard diagram for

K ⊂ S3 . Show that H1 = (Σ, α, β, z, w) and H2 = (−Σ, β, α, w, z) are both dia-
grams for Kr . Show that CFK F[U,V] (H) is isomorphic to CFK F[U,V] (H2 ) (cf. Exer-
cise 2.22). Conclude that Equation (3.6) holds.

Remark 3.8. It follows from Equation (3.6) and Exercise 3.7 that CFKF[U,V] (K) is
chain homotopy equivalent to complex C ′ obtained from CFKF[U,V] (K) by ex-
changing the roles of U and V. (Note that one should then also exchange the
roles of grU and grV .)

Exercise 3.9. Compute CFKF[U,V] (T2,3 ) two ways: from a doubly pointed Hee-
gaard diagram and by applying Equation (3.5) to Example 3.3, and confirm that
the two answers agree.

Exercise 3.10. Compute CFKF[U,V] for the figure eight knot and for the torus knot
T3,4 using the doubly pointed Heegaard diagrams from Exercises 1.35 and 1.36.

3.3. Algebraic variations As usual, let H be a doubly pointed Heegaard dia-

gram for a knot K ⊂ S3 . There are several algebraic modifications we may make
to CFK F[U,V] (H). Since the chain homotopy type of CFKF[U,V] (H) is an invari-
ant of the knot K, it follows that these algebraic modifications also yield knot
invariants.
The first modification we consider is setting both U = 0 and V = 0, resulting in
a bigraded chain complex CFKF (H) over the field F. Setting U = V = 0 is equiva-
lent to requiring that nw (φ) = nz (φ) = 0 in the definition of the differential. We
[
denote the homology of CFK F (H) by HFK(K), which is a bigraded vector space.
[
It is common to use grU and A as the bigrading on HFK(K). (Of course, this
1
is the same information as grU and grV , since A = 2 (grU − grV ).) We write
[ m (K, s) to denote the summand of HFK(K)
HFK [ with grU = m and A = s. The
grading grU is often called the Maslov grading.

Example 3.11. Setting U = V = 0 in Example 3.3 results in

∂a = ∂b = ∂c = 0.
[
Thus, HFK(K) has dimension 3, generated by a, b, and c, with gradings given in
the table in Example 3.3. That is,

F if (m, s) = (0, −1), (1, 0), or (2, 1)
[
HFKm (K, s) =
0 otherwise.
20 Heegaard Floer homology

[
Theorem 3.12 ([23, Equation (1)]). The graded Euler characteristic of HFK(K) is equal
to the Alexander polynomial of K:
X
∆K (t) = [ m (K, s) ts .
(−1)m dim HFK
m,s

Recall from [11] (see also [30, Theorem 11.3]) that the Alexander polynomial of
K can be computed in terms of the Kauffman states of a diagram for K. Note that
the Kauffman states of the left diagram in Figure 1.30 are in bijection with the
Heegaard Floer generators of the right diagram. This observation, together with
a computation of the bigradings, is at the heart of the proof of Theorem 3.12; see
[30, Sections 11-13] for details.
Another algebraic modification is to set a single variable, say V, equal to zero,
resulting in a chain complex CFKF[U] (H) over the PID F[U]. This corresponds
to requiring that nz (φ) = 0 in the definition of the differential. The homology
of CFK F[U] is a F[U]-module, denoted HFK− (K). As a finitely generated graded
module over a PID, HFK− (K) is isomorphic to a direct sum of free summands
and U-torsion summands as in Equation (2.1).
It is common to view HFK− (K) as bigraded by grU and A. The action of U
lowers grU by 2 and A by 1.
∼ H∗ (CFKF[U] (K) ⊗F[U]
Exercise 3.13. Let K ⊂ S3 . Prove that HFK− (K) ⊗F F[U, U−1 ] =
−1 ∼ c 3 −1 ∼ −1
F[U, U ]) = HF(S ) ⊗F F[U, U ] = F[U, U ] and conclude that there is a
unique free summand in HFK− (K).

Example 3.14. Setting V = 0 in Example 3.3 results in the free F[U]-module gener-
ated by a, b, and c with differential
∂a = Ub
∂b = 0
∂c = 0.
Hence
∼ F[U](2) ⊕ F(1)
HFK− (K) =
where [c] is a generator for the F[U]-summand, [b] is a generator for the F-
summand, and the subscript denotes grU .

There are other algebraic modifications one may consider, such as setting Un =
0 or UV = 0.
3.4. Computations How does one compute CFK F[U,V] in practice? For small
crossing knots, CFKF[U,V] can be computed via grid diagrams [15, 16]; see [2]
[ for knots up to 12 crossings computed using grid diagrams.
for a table of HFK
For an excellent textbook on the subject of grid diagrams, see [36]. The invariant
CFKF[U,V]/(UV=0) can be algorithmically computed following [34] (see also [35]);
such computations are significantly faster than computations with grid diagrams.
 Jennifer Hom 21

At the time of writing, a computer implementation of this algorithm is available

at https://web.math.princeton.edu/~szabo/HFKcalc.html.
For certain special families of knots, we can compute CFK F[U,V] directly from
the definition (in the case of (1, 1)-knots, described below) or from other easier to
compute knot invariants such as the Alexander polynomial and signature (in the
case of alternating knots and knots admitting L-space surgeries).
A knot in S3 that admits a genus 1 doubly pointed Heegaard diagram is called
a (1, 1)-knot. For a (1, 1)-knot K, the complex CFKF[U,V] (K) can be computed by
counting embedded disks in the universal cover of Σ = T 2 , similar to Example
3.3; see [4].
If K is alternating (or more generally, quasi-alternating; see [27, Definition 3.1]),
[
then [21, Theorem 1.3] states that HFK(K) is completely determined by the Alexan-
der polynomial and signature of K. Moreover, [37, Lemma 7] (which is completely
[
algebraic) states that if K is alternating, then HFK(K) completely determines the
chain homotopy type of CFKF[U,V] (K).

Exercise 3.15. Let K be an alternating knot. A key ingredient in the proof of [21,
[ m (K, s) 6= 0, then m = s + σ(K) , where σ(K) denotes
Theorem 1.3] is that if HFK 2
[
the signature of K. (If HFK(K) is supported on a single diagonal with respect to
the Maslov and Alexander gradings, we say K is homologically thin.) Show that
this fact combined with Theorem 3.12 completely determines the bigraded vector
[
space HFK(K) when K is an alternating knot.

If K admits a lens space surgery (or more generally, an L-space surgery), it

follows from [26, Theorem 1.2] that CFKF[U,V] (K) is completely determined by the
Alexander polynomial of K, as follows. If K admits an L-space surgery, then the
non-zero coefficients in ∆K (t) are all ±1 and they alternate in sign [26, Corollary
1.3]. Let
n
X
∆K (t) = (−1)i tai ,
i=0
for some decreasing sequence (ai ) and even n. Let bi = ai − ai−1 . If K admits a
positive L-space surgery, then CFK F[U,V] (K) is generated by x0 , . . . , xn where for
i odd,
∂xi = Ubi xi−1 + V bi+1 xi+1 ,
and for i even, ∂xi = 0. The absolute grading is determined by grU (x0 ) = 0
and grV (xn ) = 0. (Note that there is no loss of generality in considering only
positive L-space surgeries. Indeed, if K admits a negative L-space surgery, then
mK admits a positive L-space surgery and one can apply Equation (3.5).)

Exercise 3.16. Compute CFKF[U,V] (T3,4 ) using the fact that T3,4 admits a positive
L-space surgery and the above description of CFKF[U,V] (K) in terms ∆K (t) for
knots admitting L-space surgeries. Compare with Exercise 3.10.
22 Heegaard Floer homology

[
Exercise 3.17. Suppose K admits a positive L-space surgery. Express HFK(K) in
[
terms of ∆K (t), and verify that HFK(K) satisfies Theorem 3.12.

(For the relationship between (1, 1)-knots and L-space knots, see [5].)

4. Heegaard Floer homology of knot surgery

4.1. Large surgery In this section, we discuss the relationship between the knot
Floer complex CFK F[U,V] (K) and HF− (S3n (K)), where S3n (K) denotes n-surgery
on K ⊂ S3 .
We begin with some observations about CFKF[U,V] (K), which is a chain com-
plex over F[U, V]. Let W = UV. Note that multiplication by W preserves the
Alexander grading. Hence as a chain complex over F[W], the complex CFKF[U,V] (K)
splits as a direct sum over the Alexander grading. (However, note that neither
multiplication by U nor by V respects this splitting.)
Following [23, Section 4], one may identify spinc -structures on S3n (K) with
Z/nZ. Recall that HF− (S3n (K)) is a module over a polynomial ring in a single
variable.

Theorem 4.1 ([23, Theorem 4.4], cf. [38, Section 4]). Let n > 2g(K) − 1 and |s| 6
⌊ n2 ⌋. Then
∼ H∗ (CFK F[U,V] (K, s))
HF− (S3 (K), [s]) =
n

as relatively Z-graded modules over a polynomial ring in a single variable W. That is, on
the left-hand side, W = U while on the right-hand side, W = UV. The relative grading
on the right-hand side may be taken to be either grU or grV .

Remark 4.2. See [23, Corollary 4.2] for the absolutely graded version of Theorem
4.1.

Example 4.3. Let Y = S3+1 (−T2,3 ). (It follows from [18, Proposition 3.1] that
∼ −Σ(2, 3, 7).) We will use Theorem 4.1 to compute HF− (Y). Since
S3+1 (−T2,3 ) =
Y is an integer homology sphere, there is a unique spinc -structure on Y. From
Example 3.3, we have that CFKF[U,V] (−T2,3 , 0) is generated over F[W], where
W = UV, by
Va, b, Uc.
The differential is given by
∂(Va) = W · b
∂b = 0
∂(Uc) = W · b.
Note that the elements Va and Uc are in the same relative grading, while the
relative grading of b is one greater than the relative grading of Va. We have that
HF− (Y) =∼ H∗ (CFK F[U,V] (−T2,3 , 0)) =
∼ F[W](0) ⊕ F(1) ,
 Jennifer Hom 23

where the F[W]-summand is generated by Va + Uc and the F-summand is gen-

erated by b. (The absolute gradings are computed following [23, Corollary 4.2].)

Example 4.4. Let Y = S3+3 (−T2,3 ). By Theorem 4.1,

HF− (Y, [s]) = ∼ H∗ (CFK F[U,V] (−T2,3 , s)

for s = −1, 0, 1. By Example 4.3, we have

∼ H∗ (CFKF[U,V] (−T2,3 , 0)) =
HF− (Y, [0]) = ∼ F[W] ⊕ F.

From Example 3.3, we have that CFKF[U,V] (−T2,3 , −1) is generated over F[W] by
a, Ub, U2 c.
The differential is given by
∂a = Ub
∂(Ub) = 0
∂(U2 c) = W · Ub.
Hence H∗ (CFK F[U,V] (−T2,3 , −1)) =∼ F[W], generated by U2 c + Wa. Similarly,
∼ F[W], generated by V 2 a + Wc; we leave this calcu-
H∗ (CFKF[U,V] (−T2,3 , 1)) =
lation to the reader. See Figure 4.5.

U2 c Ub a Uc b

U3 Vc U2 Vb UVa U2 Vc UVb Va

U3 V 2 b U2 V 2 a U2 V 2 b UV 2 a
. .
.. ..
(a) (b)
c

UVc Vb

U2 V 2 c U2 Vb V 2a
. ..
(c)

Figure 4.5. Different Alexander graded summands of

CFK F[U,V] (−T2,3 ). Top left, A−1 = CFKF[U,V] (−T2,3 , −1).
Top right, A0 = CFKF[U,V] (−T2,3 , 0). Bottom, A1 =
CFK F[U,V] (−T2,3 , 1).
24 Heegaard Floer homology

The proof of Theorem 4.1 relies on relating the Heegaard diagrams for (S3 , K)
and S3n (K); see, for example, Figures 1.30 and 1.31. The rough idea is that for
each generator of the Heegaard diagram for (S3 , K) and for each spinc -structure
on S3n (K), there is a canonical “nearest” generator obtained by replacing the inter-
section point on the meridian with a nearby intersection point on the n-framed
longitude. The remainder of the proof relies on the relationship between spinc -
structures and the Alexander grading, as well as a count of holomorphic triangles
in a Heegaard triple. See [23, Section 4] for more details.

Exercise 4.6. Compute HF− (S3+1 (41 )), where 41 denotes the figure eight knot, us-
ing CFK F[U,V] (41 ) as computed in Exercise 3.10.

Exercise 4.7. Compute HF− (S3+5 (T3,4 )) using CFKF[U,V] (T3,4 ) as computed in Ex-
ercise 3.10.

4.2. Integer surgery Note that Theorem 4.1 requires that surgery coefficient n to
be greater than or equal to 2g(K) − 1. In [32, Theorem 1.1], Ozsváth-Szabó provide
a recipe for computing the Heegaard Floer homology of any integer surgery along
K ⊂ S3 ; they improve this to a formula for rational surgery in [33].
In these notes, we will work with the minus flavor, as in [14, Theorem 1.1].
One disadvantage to working with the minus flavor is that one must work with
completed coefficients (cf. Remark 2.11), that is, we work over the power series
rings F[[U]] and F[[U, V]]. To this end, for a pointed Heegaard diagram H for Y,
let
CF− −
F[[U]] (H) = CF (H) ⊗F[U] F[[U]]

and let HF−

F[[U]]
(Y) = H∗ (CF−
F[[U]]
(H)). Note that since F[[U]] is flat over F[U],
− ∼ HF− (Y) ⊗F[U] F[[U]]. Similarly, let
we have that HFF[[U]] (Y) =
CFKF[[U,V]](K) = CFK F[U,V] (K) ⊗F[U,V] F[[U, V]].
Let
As = CFKF[[U,V]](K, s)
denote the part of CFKF[[U,V]] (K) in Alexander grading s. (Note that this is not
quite a grading in the usual sense, as CFKF[[U,V]] (K) is a direct product rather
than a direct sum of its homogenously graded pieces. We will abuse notation
and still refer to s as the Alexander grading, and similarly for grU and grV .) We
have that
Y
∼
CFKF[[U,V]](K) = As .
s
Let
Bs = (CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1]], s)
denote the part of CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1]] in Alexander grading s. By
Exercise 3.4, we have that Bs ≃ CF− 3
F[[W]] (S ) for every s, where W = UV. Note
 Jennifer Hom 25

that
V : Bs → Bs+1 and V −1 : Bs+1 → Bs .
Since the composition of these two maps is the identity, we see that V maps Bs
isomorphically to Bs+1 . We have
Y
∼
CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1 ]] = Bs .
s

with V, V −1 , and W acting as follows

W W W
V V V V

... Bs−1 Bs Bs+1 ...

V −1 V −1 V −1 V −1

By Remark 3.8, we have that CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1 ]] is chain ho-
motopy equivalent to CFKF[U,V] (K) ⊗F[U,V] F[[U, U−1 , V]] after exchanging the
roles of U and V. (Note that this chain homotopy equivalence reverses the Alexan-
der grading.) Moreover, in any fixed Alexander grading, both complexes are
homotopy equivalent to Bs , so let
φs : (CFK F[U,V] (K) ⊗F[U,V] F[[U, U−1 , V]], s) → (CFK F[U,V] (K) ⊗F[U,V] F[[U, V, V −1 ]], s)
denote an Alexander grading-preserving chain homotopy equivalence between
Q
these two F[UV]-modules. Let φ = s φs . (Note that φ is not F[U, V]-equivariant,
although it is F[UV]-equivariant.)
Let
ιV : CFKF[[U,V]] (K) → CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1]]
and
ιU : CFKF[[U,V]] (K) → CFKF[U,V] (K) ⊗F[U,V] F[[U, U−1 , V]]
denote inclusion. Moreover, since multiplication by V is invertible in CFKF[U,V] (K) ⊗F[U,V]
F[[U, V, V −1]], we have that
V n : CFK F[U,V] (K) ⊗F[U,V] F[[U, V, V −1 ]] → CFKF[U,V] (K) ⊗F[U,V] F[[U, V, V −1]]
is a (relatively graded) isomorphism. Note that V n |Bs : Bs → Bs+n .
Consider the chain map
Dn : CFK F[[U,V]](K) → CFK F[U,V] (K) ⊗F[U,V] F[[U, V, V −1]]
where Dn = ιV + V n ◦ φ ◦ ιU .
Recall that given two chain complexes (X, ∂X ), (Y, ∂Y ) and a chain map f : X →
Y, the mapping cone of f is the chain complex Cone(f) = X ⊕ Y with the differential
∂(x, y) = (∂X x, f(x) + ∂Y y). (Note that we are working over a ring of characteristic
two; otherwise, one needs to insert some appropriate minus signs in the definition
of the mapping cone.)
26 Heegaard Floer homology

V −2 c V −1 c

UV −1 c V −1 b Uc b

U2 c Ub a U2 Vc UVb Va

U3 Vc U2 Vb UVa U3 V 2 c U2 V 2 b UV 2 a

U3 V 2 b U2 V 2 a U3 V 3 b U2 V 3 a
. .
.. ..
(a) (b)

Figure 4.8. Different Alexander graded pieces of

∼ Q Bs . Left, B−1 .
CFK F[U,V] (−T2,3 ) ⊗F[U,V] F[[U, V, V −1]] = s
Right, B0 . Note that the two complexes are isomorphic, with the
isomorphism provided by multiplication by V ±1 .

Exercise 4.9. Let f : X → Y be a chain map. Show that (Cone(f), ∂) is a chain

complex.

The surgery formula expresses HF− (S3n (K)) in terms of the mapping cone of
Dn .

Theorem 4.10 ([14, Theorem 1.1], cf. [32, Theorem 1.1]). We have the following
isomorphism of F[W]-modules
∼ H∗ (Cone(Dn )),
HF− (S3 (K)) ⊗F[W] F[[W]] =
n

where on the left-hand side HF− (S3n (K) is viewed as a module over W = U and on the
right-hand side, W = UV.

There is a similar formula for rational surgeries; see [33]. Note that
ιV |As : As → Bs and V n ◦ φ ◦ ιU |As : As → Bs+n .
Hence
Dn |As : As → Bs ⊕ Bs+n ,
and so Dn may be depicted as follows:
As−n As As+n

... V n φιU ιV V n φιU ιV ...

Bs Bs+n
 Jennifer Hom 27

Exercise 4.11. Use (3.1) to conclude that for s > g(K), the maps
ιV : As → Bs and V n ◦ φ ◦ ιU : A−s → B−s+n
are homotopy equivalences. Use this to conclude that Cone(D1 ) is homotopy
equivalent to
A1−g A2−g Ag−2 Ag−1

VφιU ιV ... VφιU ιV

B2−g Bg−1

and that Cone(D−1 ) is homotopy equivalent to

A1−g Ag−1

V −1 φιU ιV ... V −1 φιU ιV

B−g B1−g Bg−2 Bg−1 .

Similar statements hold for arbitrary positive (respectively negative) surgery co-
efficients.

Remark 4.12. We consider the knot Floer complex as a module over the bigraded
ring F[U, V], whereas Ozsváth-Szabó originally defined the knot Floer complex
as a Z-filtered chain complex over F[U]. The two formulations are equivalent, as
described in [45, Section 1.5]. Indeed, our notation As and Bs in this section lines
up with the (minus flavor over the power series ring) of the complexes As and Bs
in [32], while our ιV (respectively V n φιU ) is denoted v (respectively h) in [32].

Remark 4.13. Theorem 4.10 has analogous versions for the plus and hat flavors of
Heegaard Floer homology. Both the plus and hat flavors have the advantage that
one may pass to homology before taking the mapping cone. For the hat flavor,
this is because the chain complexes are all vector spaces, while for the plus flavor,
this is because the maps ιV,∗ and (V n ◦ φ ◦ ιU )∗ are surjective for knots in S3 . See
[32, Section 5] for some sample calculations.

Exercise 4.14. Compute HF− (S3+1 (T3,4 )) using Exercise 4.11 and the computation
from Exercise 4.7.

4.3. Applications We conclude with some applications of the mapping cone for-
mula.
Recall the Cosmetic Surgery Conjecture, which posits that for a nontrivial knot
K ⊂ S3 and r, r ′ ∈ Q, if S3r (K) and S3r ′ (K) are homeomorphic as oriented man-
ifolds, then r = r ′ . Using the mapping cone formula, Ni-Wu [20] prove the
following:
28 Heegaard Floer homology

Theorem 4.15 ([20, Theorem 1.2]). Suppose K is a nontrivial knot in S3 such that
∼ S3 ′ (K) as oriented manifolds where r, r ′ are distinct rational numbers. Then
S3r (K) = r
r = −r ′ and r is of the form p/q where p, q are relatively prime integers with q2 ≡ −1
(mod p).

Remark 4.16. Hanselman [6, Theorem 2] recently improved this result to r = ±2

or ±1/q, using bordered Floer homology, a version of Heegaard Floer homol-
ogy for 3-manifolds with parametrized boundary. Hanselman’s result relies on a
reinterpretation of bordered Floer homology as immersed curves [7].

Heegaard Floer homology can also be used to obstruct manifolds from being
obtained by surgery on a knot in S3 . Lickorish [12] and Wallace [43] proved
that every closed oriented 3-manifolds can be obtained by surgery on a link in S3 .
∼ Z/pZ, it follows that if H1 (Y; Z) is not cyclic, then
However, since H1 (S3p/q (K)) =
Y cannot be obtained by surgery on a knot in S3 . For example, H1 (RP3 #RP3 ; Z) = ∼
3 3
Z/2Z ⊕ Z/2Z, hence RP #RP is not surgery on any knot in S . 3

This H1 obstruction vanishes for integer homology spheres. Auckly [1] gave
an example of a hyperbolic integer homology sphere that cannot be obtained via
surgery on a knot in S3 ; moreover, his techniques can be used to give infinitely
many such examples. Using the mapping cone formula, Hom-Karakurt-Lidman
[8] prove the following:

Theorem 4.17 ([8, Theorem 1.1]). The Brieskorn homology spheres Σ(n, 2n − 1, 2n +
1), n > 8, n even, cannot be obtained by surgery on a knot in S3 .

The proof of Theorem 4.17 relies on a surgery obstruction which relates the
d-invariant of surgery on a knot in S3 with HFred ; see [8, Theorem 1.2] for the
precise statement.
The mapping cone formula has also played a role in results that make no ref-
erence to Dehn surgery. For example, one could ask whether there exist simply-
connected, positive definite symplectic 4-manifolds. Recall that a manifold is geo-
metrically simply-connected if it admits a handle decomposition with no 1-handles.

Theorem 4.18 ([9, Theorem 1.1]). Let X be a closed, geometrically simply-connected

4-manifold with b+
2 (X) > 2. If the intersection form of X is positive definite, then X is
not symplectic.

Remark 4.19. Yasui [44] has recently obtained another proof of Theorem 4.18, us-
ing Seiberg-Witten theory.

The proof of Theorem 4.18 relies on the Ozsváth-Szabó closed 4-manifolds

invariant [29]. Their invariant in defined as a composition of certain cobordism
maps, which, in certain special cases, may be computed using the mapping cone
formula.
 Jennifer Hom 29

References
[1] D. Auckly, Surgery numbers of 3-manifolds: a hyperbolic example, Geometric topology (Athens, GA,
1993), 1997, pp. 21–34. ←28
[2] J. A. Baldwin and W. D. Gillam, Computations of Heegaard-Floer knot homology, J. Knot Theory
Ramifications 21 (2012), no. 8, 1250075, 65. ←21
[3] P. Ghiggini, Knot Floer homology detects genus-one fibred knots, Amer. J. Math. 130 (2008), no. 5,
1151–1169. ←16
[4] H. Goda, H. Matsuda, and T. Morifuji, Knot Floer homology of (1, 1)-knots, Geom. Dedicata 112
(2005), 197–214. ←21
[5] J. E. Greene, S. Lewallen, and F. Vafaee, (1, 1) L-space knots, Compos. Math. 154 (2018), no. 5,
918–933. ←22
[6] J. Hanselman, Heegaard Floer homology and cosmetic surgeries in S3 , 2019. preprint, arXiv:1906.06773.
←28
[7] J. Hanselman, J. Rasmussen, and L. Watson, Bordered Floer homology for manifolds with torus bound-
ary via immersed curves, 2016. preprint, arXiv:1604.03466. ←28
[8] J. Hom, Ç. Karakurt, and T. Lidman, Surgery obstructions and Heegaard Floer homology, Geom.
Topol. 20 (2016), no. 4, 2219–2251. ←28
[9] J. Hom and T. Lidman, A note on positive-definite, symplectic four-manifolds, J. Eur. Math. Soc. (JEMS)
21 (2019), no. 1, 257–270. ←29
[10] A. Juhász, D. Thurston, and I. Zemke, Naturality and mapping class groups in Heegaard Floer homol-
ogy, 2012. preprint, arXiv:1210.4996v4. ←16
[11] L. H. Kauffman, Formal knot theory, Mathematical Notes, vol. 30, Princeton University Press,
Princeton, NJ, 1983. ←20
[12] W. B. R. Lickorish, A representation of orientable combinatorial 3-manifolds, Ann. of Math. (2) 76
(1962), 531–540. ←28
[13] R. Lipshitz, P. S. Ozsváth, and D. P. Thurston, Computing HF d by factoring mapping classes, Geom.
Topol. 18 (2014), no. 5, 2547–2681. ←15
[14] C. Manolescu and P. Ozsváth, Heegaard Floer homology and integer surgeries on links, 2010. preprint,
arXiv:1011.1317v3. ←12, 24, 26
[15] C. Manolescu, P. Ozsváth, and S. Sarkar, A combinatorial description of knot Floer homology, Ann. of
Math. (2) 169 (2009), no. 2, 633–660. ←20
[16] C. Manolescu, P. Ozsváth, Z. Szabó, and D. Thurston, On combinatorial link Floer homology, Geom.
Topol. 11 (2007), 2339–2412. ←20
[17] C. Manolescu, P. Ozsváth, and D. Thurston, Grid diagrams and Heegaard Floer invariants, 2009.
preprint, arXiv:0910.0078v3. ←16
[18] L. Moser, Elementary surgery along a torus knot, Pacific J. Math. 38 (1971), 737–745. ←22
[19] Y. Ni, Knot Floer homology detects fibred knots, Invent. Math. 170 (2007), no. 3, 577–608. ←16
[20] Y. Ni and Z. Wu, Cosmetic surgeries on knots in S3 , J. Reine Angew. Math. 706 (2015), 1–17. ←28
[21] P. Ozsváth and Z. Szabó, Heegaard Floer homology and alternating knots, Geom. Topol. 7 (2003),
225–254. ←21
[22] P. Ozsváth and Z. Szabó, Holomorphic disks and genus bounds, Geom. Topol. 8 (2004), 311–334. ←16
[23] P. Ozsváth and Z. Szabó, Holomorphic disks and knot invariants, Adv. Math. 186 (2004), no. 1, 58–
116. ←1, 8, 16, 17, 18, 19, 20, 22, 23
[24] P. Ozsváth and Z. Szabó, Holomorphic disks and three-manifold invariants: properties and applications,
Ann. of Math. (2) 159 (2004), no. 3, 1159–1245. ←9, 10, 11, 14, 15, 17
[25] P. Ozsváth and Z. Szabó, Holomorphic disks and topological invariants for closed three-manifolds, Ann.
of Math. (2) 159 (2004), no. 3, 1027–1158. ←1, 6, 9, 13, 14, 15
[26] P. Ozsváth and Z. Szabó, On knot Floer homology and lens space surgeries, Topology 44 (2005), no. 6,
1281–1300. ←21
[27] P. Ozsváth and Z. Szabó, On the Heegaard Floer homology of branched double-covers, Adv. Math. 194
(2005), no. 1, 1–33. ←21
[28] P. Ozsváth and Z. Szabó, Holomorphic triangles and invariants for smooth four-manifolds, Adv. Math.
202 (2006), no. 2, 326–400. ←11, 15
[29] P. Ozsváth and Z. Szabó, Holomorphic triangles and invariants for smooth four-manifolds, Adv. Math.
202 (2006), no. 2, 326–400. ←29
30 References

[30] P. Ozsváth and Z. Szabó, An introduction to Heegaard Floer homology, Floer homology, gauge theory,
and low-dimensional topology, 2006, pp. 3–27. ←2, 9, 13, 17, 20
[31] P. Ozsváth and Z. Szabó, Lectures on Heegaard Floer homology, Floer homology, gauge theory, and
low-dimensional topology, 2006, pp. 29–70. ←11
[32] P. Ozsváth and Z. Szabó, Knot Floer homology and integer surgeries, Algebr. Geom. Topol. 8 (2008),
no. 1, 101–153. ←24, 26, 27, 28
[33] P. Ozsváth and Z. Szabó, Knot Floer homology and rational surgeries, Algebr. Geom. Topol. 11 (2011),
no. 1, 1–68. ←24, 26
[34] P. Ozsváth and Z. Szabó, Bordered knot algebras with matchings, 2017. preprint, arXiv:1707.00597.
←21
[35] P. Ozsváth and Z. Szabó, Algebras with matchings and knot Floer homology, 2019. preprint,
arXiv:1912.01657. ←21
[36] P. S. Ozsváth, A. I. Stipsicz, and Z. Szabó, Grid homology for knots and links, Mathematical Surveys
and Monographs, vol. 208, American Mathematical Society, Providence, RI, 2015. ←9, 21
[37] I. Petkova, Cables of thin knots and bordered Heegaard Floer homology, Quantum Topol. 4 (2013), no. 4,
377–409. ←21
[38] J. A. Rasmussen, Floer homology and knot complements, ProQuest LLC, Ann Arbor, MI, 2003. Thesis
(Ph.D.)–Harvard University. ←1, 17, 22
[39] K. Reidemeister, Zur dreidimensionalen Topologie, Abh. Math. Sem. Univ. Hamburg 9 (1933), no. 1,
189–194. ←3
[40] S. Sarkar and J. Wang, An algorithm for computing some Heegaard Floer homologies, Ann. of Math. (2)
171 (2010), no. 2, 1213–1236. ←15
[41] N. Saveliev, Lectures on the topology of 3-manifolds, De Gruyter Textbook, Walter de Gruyter & Co.,
Berlin, 1999. An introduction to the Casson invariant. ←2
[42] J. Singer, Three-dimensional manifolds and their Heegaard diagrams, Trans. Amer. Math. Soc. 35 (1933),
no. 1, 88–111. ←3
[43] A. H. Wallace, Modifications and cobounding manifolds, Canadian J. Math. 12 (1960), 503–528. ←28
[44] K. Yasui, Geometrically simply connected 4-manifolds and stable cohomotopy Seiberg-Witten invariants,
2018. preprint, arXiv:1807.11453. ←29
[45] I. Zemke, Link cobordisms and absolute gradings on link Floer homology, 2017. preprint,
arXiv:1701.03454. ←17, 18, 27

School of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332

E-mail address: [email protected]