Lecture Notes on Topics in Complex Geometry

Sébastien Picard

December 15, 2022

Contents
1 Complex Geometry 2
1.1 Complex manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Holomorphic vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Definitions and notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 Bundle constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Constructions from the tangent bundle . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Geometry of bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Chern connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.3 Hermitian geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2 Kähler Manifolds 26
2.1 Hodge theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Kodaira vanishing theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Sheaves and the Lefschetz hyperplane theorem . . . . . . . . . . . . . . . . . . . . . 38

3 Deformations of Complex Manifolds 43

3.1 Families of complex manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Semi-continuity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Stability of Kähler metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4 Calabi-Yau Threefolds 56
4.1 Parameters of threefolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2 Ricci flat metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3 Deformations of complex structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 Quintic threefolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.1 Holomorphic volume form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.2 Hodge numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4.3 Nodal singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4.4 Examples of conifold transitions . . . . . . . . . . . . . . . . . . . . . . . . . 66

5 Conifold Transitions: Local Model 69

1
 5.1 Blowing-up a nodal singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1.1 Blow-up review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1.2 ODP in C3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.1.3 ODP in C4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Smoothing a nodal singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3 Candelas-de la Ossa metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3.1 Metrics on the small resolution . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3.2 Metrics on the smoothings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.3 The cone metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 Special Lagrangian cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.4.1 Definition of special Lagrangian cycles . . . . . . . . . . . . . . . . . . . . . . 85
5.4.2 Examples on the smoothing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

6 Conifold Transitions: Global Geometry 90

6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.2 Topological change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

References 96

1 Complex Geometry
This section is an introduction to complex geometry. For other references in the style of these notes,
see Kodaira’s book [19], Chapter 1 of Siu’s notes [24], Chapter 1 of Song-Weinkove’s notes [25], or
Chapter 1 of Szekelyhidi’s book [26].

1.1 Complex manifolds

Let Ω Ď Cn be a domain. We denote complex variables by

z “ pz 1 , . . . , z n q,

so that
z k “ xk ` iy k , z̄ k “ xk ´ iy k , k P t1, . . . , nu,
2n 1 1 n n
and the real variables on Ω Ď R are px , y , . . . , x , y q. The inverse transformation is
1 k 1 k
xk “ pz ` z̄ k q, yk “ pz ´ z̄ k q. (1.1)
2 2i
By the chain rule, for a smooth function f : Ω Ñ C, we have
ˆ ˙ ˆ ˙
Bf 1 B B Bf 1 B B
“ ´i f, “ `i f. (1.2)
Bz 2 Bx By Bz̄ 2 Bx By

Definition 1.1. Let f : Ω Ñ Ck be a C 1 function with components

f “ pf 1 ppq, . . . , f k ppqq.

2
We say f is holomorphic if
Bf i
“0
Bz̄ k
for all i, k.
Let M be a compact complex manifold.
Ť This means that M is a smooth compact manifold admiting
a finite cover by open sets M “ i Ui with homeomorphisms

zU : U Ñ U Ď Cn , 1
zU ppq “ pzU n
ppq, . . . , zU ppqq

with the following property. For any pair pU, zU q, pV, zV q with U XV ‰ H, then we can write

zVp “ fV U p pzU q, ´1
fV U “ zV ˝ zU

with fV U : zU pU X V q Ñ zV pU X V q a holomorphic bijective function with holomorphic in-

verse.
Example 1.2. P1 “ pC2 zt0uq{ „, where points p P P1 are written as

p “ rZ0 : Z1 s

and rZ0 : Z1 s „ rX0 : X1 s if and only if pZ0 , Z1 q “ λpX0 , X1 q for λ P C˚ . Holomorphic charts are:
‚ U0 “ tZ0 ‰ 0u, with coordinate
Z1
z“ .
Z0

‚ U1 “ tZ1 ‰ 0u, with coordinate

Z0
z̃ “ .
Z1
The change of coordinates function f10 : zpU0 X U1 q Ñ z̃pU0 X U1 q is
1
f10 : C˚ Ñ C˚ , z̃ “ f10 pzq “ .
z

Example 1.3. Complex projective space in higher dimensions Pn “ pCn`1 zt0uq{ „ is defined
similarly. Points are denoted
p “ rZ0 : Z1 : ¨ ¨ ¨ : Zn s
and rZ0 : Z1 : ¨ ¨ ¨ : Zn s „ rX0 : X1 : ¨ ¨ ¨ : Xn s if and only if pZ0 , Z1 , . . . , Zn q “ λpX0 , X1 , . . . , Xn q
for λ P C˚ . Holomorphic charts are of the form Uk “ tZk ‰ 0u. For example, pU0 , zq has coordinate
ˆ ˙
Z1 Zn
z “ pz 1 , . . . , z n q “ ,..., ,
Z0 Z0

while pU1 , z̃q has coordinate

ˆ ˙
Z0 Z2 Zn
z̃ “ pz̃ 1 , . . . , z̃ n q “ , ,..., ,
Z1 Z1 Z1

3
and the change of coordinates function is z̃ “ f10 pzq with

1 zk
z̃ 1 “ , z̃ k “ , k P t2, . . . , nu.
z1 z1
The coordinates on the other open sets Uk are defined similarly.
Example 1.4. Let P pZ0 , . . . , Zn q be a homogeneous polynomial of degree r, meaning P pλZ0 , . . . , λZn q “
λr P pZ0 , . . . , Zn q. Suppose that P has the property that only the point Z0 “ ¨ ¨ ¨ “ Zn “ 0 solves
BP
BZi “ 0 for all i. Then
X “ tx P Pn : P pxq “ 0u
defines a complex manifold of dimension n ´ 1. To see this, we look at tP “ 0u inside the local
charts Uk Ď Pn . For example, in coordinates pU0 , zq, the equation defining X is

0 “ f pz 1 , . . . , z n q “ P p1, z 1 , . . . , z n q.
Bf
At a point p P tf “ 0u, there must be a coordinate z i such that Bz i ‰ 0, in other words one
BP BP BP
of BZ1 , . . . BZn must be nonzero. The assumption on P is that we cannot have BZ i
“ 0 for all
BP
i P t0, . . . , nu, so we only need to rule out BZ0 ‰ 0 with all other partials zero. This is ruled out by
Euler’s identity
n
ÿ BP
Zi “ rP,
i“0
BZ i

BP BP
which shows that on tP “ 0u X U0 if BZ 0
‰ 0 then one of BZi for i ě 1 must be non-vanishing.
dˇ
ˇ dˇ
ˇ
Euler’s identity follows from dt t“1 P ptZq “ dt t“1
tr P pZq.
Bf
Without loss of generality, Bz n ‰ 0. By the holomorphic implicit function theorem, there exists a

holomorphic function ψ : V Ñ Op , V Ă Cn´1 , p P Op Ă U0 , such that

X X Op “ tpw1 , . . . , wn´1 , ψpwqq : w P V u.

This defines a coordinate chart pV, wq.

There are two types of coordinate change:

Bf Bf
‚ Suppose p P X X U0 satisfies both Bz 1 ‰ 0 and Bz n ‰ 0. Then we can parametrize X in a
Bf
neighborhood of p by either (implicit function on Bzn ‰ 0)

pw1 , . . . , wn´1 , ψpwqq

4
 Bf
or (implicit function on Bz 1 ‰ 0)
pψ̃pw̃q, w̃1 , . . . , w̃n´1 q.
The change of coordinates is

w̃1 “ w2 , w̃2 “ w3 , ... w̃n´1 “ ψpwq,

which is holomorphic.
‚ Suppose p P X X pU0 X U1 q. Then we can use coordinates coming from either U0 or U1 . As before
we denote pU0 , zq and pU1 , z̃q with z̃ 1 “ 1{z 1 and z̃ k “ z k {z 1 for k ě 2. The submanifold appears
as the equation
0 “ P p1, z 1 , z 2 , . . . , z n q
on U0 , and
0 “ P pz̃ 1 , 1, z̃ 2 , . . . , z̃ n q
BP
on U1 . Suppose BZ n
‰ 0. As before, the implicit function theorem gives coordinates wi and w̃i , so
that the equations become
0 “ P p1, w1 , w2 , . . . , ψpwqq
on U0 , and
0 “ P pw̃1 , 1, w̃2 , . . . , ψ̃pw̃qq
on U1 . The change of coordinates is

w̃1 “ 1{w1 , w̃2 “ w2 {w1 , ... w̃n´1 “ wn´1 {w1 .

BP BP
Other situations when e.g. BZ0 ‰ 0 and BZ1 ‰ 0 can also be worked out in a similar way.

5
1.2 Holomorphic vector bundles
1.2.1 Definitions and notation
Ť
We recall the cocycle definition of a rank r complex vector bundle. Let M “ i Ui be a finite cover-
ing of open coordinate charts, together with matrix-valued functions on the nonzero overlaps

tU V : U X V Ñ GLpr, Cq

satisfying
tU V ppq “ tU W ppqtW V ppq, p P U X V X W.
We call the tU V transition functions. They satisfy tU U “ Irˆr and t´1
U V “ tV U . We define a complex
vector bundle E by ˆď ˙
r
E“ Ui ˆ C { „,
i

where the relation is as follows. For pp, uq P U ˆ Cr and pp, vq P V ˆ Cr , we identify pp, uq „ pp, vq
if
u “ tU V ppqv.
This is written using matrix notation. In terms of components, this is written

uk “ rtU V ppqsk ` v ` ,

where repeated indices are summed. Here we write v “ pv 1 , . . . , v r q and the components of the
matrix tU V are denoted tU V i j , e.g. for 2x2,
» fi » fi » fi
u1 tU V 1 1 tU V 1 2 v1
– fl “ – fl – fl .
u2 tU V 2 1 tU V 2 2 v2

The Ui ˆ Cr are the trivializations of the bundle. The projection map π : E Ñ M is given by
πpp, uq “ p.

6
We denote E|p “ π ´1 ppq to be the fiber over p, and note that E|p is a vector space of dimension r.
For two points pp, uq, pp, vq in the same trivialization Ui ˆ Cr , the vector space structure is

app, uq ` bpp, vq “ pp, au ` bvq,

and one can check that this is well-defined.

‚ Note: we will call a complex vector bundle of rank 1 a line bundle.
Definition 1.5. A complex vector bundle π : E Ñ M over a complex manifold is holomorphic if
the transition functions tU V are holomorphic.
‚ Note: taking U , V to be coordinate charts, then U X V is viewed as an open set in Cn . That
tU V : U X V Ñ GLpr, Cq is holomorphic means that each entry of the matrix is a holomorphic
function of p P U X V Ď Cn .
Example 1.6. We will denote the trivial holomorphic line bundle by OX Ñ X. This means that
the transition functions are tU V “ 1.
Definition 1.7. Let E Ñ M be a rank r complex vector bundle with trivializations Ui . A smooth
section, denoted s P ΓpEq, is given by local vector-valued smooth functions tUi , sUi u with sU : U Ñ
Cr satisfying
sU “ tU V sV .

7
A section s P ΓpEq defines a well-defined map s : M Ñ E such that

sppq P E|p .

Indeed, in this formalism we set sppq “ pp, sU ppqq when p P U , and the condition sU “ tU V sV
ensures that if p P U X V then pp, sU ppqq „ pp, sV ppqq.
Remark 1.8. If E Ñ M is a holomorphic bundle and the sU are holomorphic functions, then we
say s is a holomorphic section and write s P H 0 pM, Eq.
Remark 1.9. In components, the transformation law is

skU “ tU V k ` s`V (1.3)

e.g. for 3x3, this notation means

» fi » fi » fi
s1 t 1 tU V 1 2 tU V 1 3 s1V
— U ffi — U V 1 ffi — ffi
— 2 ffi —
—sU ffi “ —tU V 2 1 tU V 2 2 tU V 2 3 ffi —s2V ffi ,
ffi — ffi
– fl – fl – fl
s3U tU V 3 1 tU V 3 2 tU V 3 3 s3V

on U X V .
There is another viewpoint on this from the perspective of basis vectors rather than vector compo-
nents. For a trivialization U ˆ Cr , let

eU r
a ppq “ pp, p0, . . . , 0, 1, 0, . . . , 0qq P U ˆ C ,

where the 1 is at the ath position. Then teU U U

1 ppq, . . . er ppqu is a basis for E|p , and we say that tea u
is a local frame over U .
On an overlap U X V , the same basis vector can be written in two different ways. We note that
eU V b
a „ eb tV U a . Instead of the „ notation, this is usually just written

eU V b
a “ eb tV U a . (1.4)

Here is how to see this in the 2 ˆ 2 case and a “ 1. By definition, eU V b

1 „ eb tV U 1 if and only if

eU V b
1 “ tU V eb tV U 1 .

which is » fi ¨» fi » fi ˛ ¨ » fi˛
1 1 0 1 ‹
˝– fl tV U 1 1 ` – fl tV U 2 1 ‚ “ rtU V s ˝rtV U s – fl‚
– fl “ rtU V s ˚ ‹ ˚
0 0 1 0

and this holds since tU V tV U “ I2ˆ2 .

For a section s P ΓpEq, we will sometimes make the frame explicit and write

s “ sa ea .

8
In the 2 ˆ 2 case, this notation means
» fi » fi » fi
s1 ppq 1 0
– fl “ s1 ppq – fl ` s2 ppq – fl .
s2 ppq 0 1

Note that on U X V , then

saU eU a V
a “ sV ea

so we simply write s “ sa ea . Indeed, substituting the transformation laws (1.3), (1.4) gives

saU eU a b V c c b V b V
a “ rtU V b sV srec tV U a s “ δ b sV ec “ sV eb .

In terms of linear algebra, this is just the statement that the same vector v will appear in different
components v a using different bases ea .
Ť
Example 1.10. Let M “ i Ui be a complex manifold with holomorphic coordinate charts pU, zq.
The holomorphic tangent bundle T 1,0 M Ñ M is the holomorphic bundle defined by transition
matrices
Bz k
tU V k i “ U .
BzVi
Sections X P ΓpT 1,0 M q are denoted
B
X “ X p pzq .
Bz p
On an overlap of coordinate charts pU, zq, pŨ , z̃q, components transform as
Bz̃ p `
X̃ p “ X
Bz `
while the basis transforms as
B Bz p B
k
“ k p.
Bz̃ Bz̃ Bz
It follows that
B B
Xp p
“ X̃ p p
Bz Bz̃
on overlaps.
Let E Ñ M , F Ñ M be two holomorphic
Ť vector bundles with transition functions tU V , t̃U V with
respect to a trivialization M “ i Ui . An isomorphism of holomorphic bundles h : E Ñ F is given
by a collection thU : U Ñ GLpr, Cqu of holomorphic invertible matrices satisfying

hU “ t̃U V hV t´1
UV . (1.5)

This definition is such that h is a well-defined isomorphism from fibers of E to fibers of F .

hppq : E|p Ñ F |p

This amounts to the statement that if v „ w in E, then hv „ hw in F . Indeed, if vU “ tU V wV

then
hU vU “ pt̃U V hV t´1
U V qptU V wV q “ t̃U V phV wV q.

9
Example 1.11. We return to the example M “ P1 “ U0 YU1 . There are two charts pU0 , zq, pU1 , z̃q
and z̃ “ z ´1 . Therefore
B Bz̃ B 1 B B
“ “´ 2 “ ´z̃ 2 .
Bz Bz Bz̃ z Bz̃ Bz̃
Said otherwise, a section of T 1,0 M may be written on U0 X U1 as vpzq Bz
B B
or ṽpz̃q Bz̃ with
ṽ “ ´z̃ 2 v.
so that the transition function is t10 “ ´z̃ 2 .
For example, defining BzB
over U0 extends to a global vector field V over M by setting ´z̃ 2 Bz̃
B
over
B
U1 . However, even though Bz is nowhere vanishing over U0 , this vector field must acquire a zero at
z̃ “ 0 in U1 . In component notation, this vector field V is given by the data
V “ tpU0 , vpzqq, pU1 , ṽpz̃qqu P H 0 pM, T 1,0 M q
with
vpzq “ 1, ṽpz̃q “ ´z̃ 2 .

Example 1.12. Let k P Z. Define the bundle Opkq Ñ P1 with trivializations pU0 , zq, pU1 , z̃q by
setting t10 “ z̃ k , so that sections transform as
s̃ “ z̃ k s.
The previous example, combined with (1.5) and suitable choice of hU0 , hU1 , shows that T 1,0 P1 –
Op2q. Let k ą 0.
‚ There are no holomorphic sections of Op´kq. Suppose such a section appears as a holomorphic
function spzq over the trivialization U0 . Then over U1 , that same section takes the form s̃ “ z̃ ´k s,
which in the z̃ coordinates belonging to U1 is
s̃pz̃q “ z̃ ´k spz̃ ´1 q.
ř8
Writing spzq “ i“0 ai z i , we see that s̃pz̃q must have a pole and cannot be holomorphic.
‚ Holomorphic sections of Opkq correspond to homogeneous polynomials P pZ0 , Z1 q of degree k: any
section σ P H 0 pP1 , Opkqq is σ “ tpU0 , sq, pU1 , s̃qu locally the form s “ P pZ0 , Z1 q{Z0k over U0 , and
of the form s̃ “ P pZ0 , Z1 q{Z1k over U1 . Indeed, let σ P H 0 pP1 , Opkqq be an arbitrary holomorphic
section. Then spzq, s̃pz̃q are both holomorphic and
s̃pz̃q “ z̃ k spz̃ ´1 q. (1.6)
ř8 ř8
After writing s̃ “ k“0 bk z̃ k , s “ k“0 ak z k and comparing coefficients, we see that spzq “
a0 ` a1 z ` ¨ ¨ ¨ ` ak z k . It follows that
„ 
1 k k´1 k
s “ k a0 Z0 ` a1 Z0 Z1 ` ¨ ¨ ¨ ` ak Z1
Z0
since on U0 “ tZ0 ‰ 0u the coordinate is z “ Z1 {Z0 . The transformation (1.6) implies
„ 
1 k k´1 k
s̃ “ k a0 Z0 ` a1 Z0 Z1 ` ¨ ¨ ¨ ` ak Z1
Z1
since on U1 “ tZ1 ‰ 0u the coordinate is z̃ “ Z0 {Z1 . Therefore the homogeneous polynomial
corresponding to this section is P “ a0 Z0k ` a1 Z0k´1 Z1 ` ¨ ¨ ¨ ` ak Z1k .

10
Example 1.13. In higher dimensional projective space, define Opkq Ñ Pn by pUi , tij q where
Ui “ tZi ‰ 0u and tij : Ui X Uj Ñ C˚ is
ˆ ˙k
Zj
tij “
Zi

For example on Op1q Ñ P2 , with coordinates pU0 , zq, pU1 , z̃q with z “ pZ1 {Z0 , Z2 {Z0 q and z̃ “
pZ0 {Z1 , Z2 {Z1 q, then t10 “ Z0 {Z1 “ z̃ 1 .

1.2.2 Bundle constructions

We now describe some bundle constructions. Let E Ñ M be a complex vector bundle of rank r
with trivializations tU V .
‚ Conjugate bundle. The complex vector bundle Ē Ñ M has trivializations tU V . Note that if
E Ñ M is a holomorphic vector bundle, then Ē is not a holomorphic bundle. However, the next
constructions do produce holomorphic bundles if E is holomorphic.
‚ Dual bundle. Define E ˚ Ñ M to be the bundle of rank r defined by trivializations pt´1 T
U V q . We use
i
the following index notation: components of sections s P ΓpM, Eq are denoted s , and components
of sections ϕ P ΓpM, E ˚ q are denoted ϕi , so that the transformation laws reads

siU “ tU V i k skV , ϕU V k
i “ ϕk tV U i .

This is the dual bundle because sections s P ΓpM, Eq and ϕ P ΓpM, E ˚ q can be paired together to
form a function
ϕpsq :“ pϕi si q P C 8 pM, Rq.
This is because the transformation laws imply

ϕU i V i
i sU “ ϕi sV

and so pϕi si qppq is independent of the choice of trivialization. In matrix notation Q “ rtU V s, the
transformation laws for s P ΓpEq and ϕ P ΓpE ˚ q are

s ÞÑ Qs, ϕ ÞÑ rQ´1 sT ϕ, ϕpsq “ ϕT s.

In terms of local frames, if tei u is a local frame for E, we denote the corresponding dual frame on
E ˚ by tei u. This is defined as ei pej q “ δ i j , and a section ϕ P ΓpE ˚ q is written as

ϕ “ ϕi ei .

The pairing ϕpsq can then be seen by the formula for the dual frame: ϕpsq “ pϕi ei qpsk ek q “
ϕj sj .
‚ Determinant bundle. The line bundle det E Ñ M is defined by the trivializations det tU V .
‚ Tensor product. If E Ñ M , Ẽ Ñ M are vector bundles, then the bundle E b Ẽ Ñ M has
trivializations tU V b t̃U V . In components, if indices i, j denotes indices on E and indices α, β
indices on Ẽ, then
siα i α
U “ tU V j t̃U V β sV
jβ
.

11
‚ Endomorphism bundle. We will later encounter sections of E ˚ b E “ End E ˚ , and our convention
for h P ΓpEnd E ˚ q will be
h “ hα β eα b eβ
so that the transformation law for components reads

rhU sα β “ tV U µ α rhV sµ ν tU V β ν ,

which in matrix notation Q “ rtU V s, for u P ΓpEq and h P ΓpEnd E ˚ q, is

u ÞÑ Qu, h ÞÑ rQT s´1 hQT . (1.7)

Note that h defines a map h|p : E ˚ |p Ñ E ˚ |p by hα β ϕβ . Verifying that this map is well-defined is
a similar calculation as (1.5).
In fact, h P ΓpEnd E ˚ q also defines an endomorphism of E by acting on the right as uT h, or in
index notation uα hα β . That uT h transforms like a section follows from

puT hq ÞÑ pQuqT ppQT q´1 hQT q “ uT hQT “ QpuT hq.

Thus uT h P ΓpEq. Thus, we will sometimes view h P ΓpEnd E ˚ q with h “ hα β as h P ΓpEnd Eq.
‚ Divisor bundle. Let Y Ă X be an analytic hypersurface. This means that near each p P Y , there
is neighborhood U such that U X Y is locally given by the vanishing set of a holomorphic function.
The theory of holomorphic functions (see e.g. [13]) implies that there exists the notion of a local
defining function: this means that f is holomorphic with

U X Y “ tf “ 0u

and any other local holomorphic function g vanishing on Y factors as gpzq “ hpzqf pzq with h a local
holomorphic function. The notion of local defining function is not unique: if f1 and f2 are local
defining functions, then f1 “ hf2 where h is a holomorphic function non-vanishing on U .
We can associate a line bundle OpY q Ñ X in the following way. In a coordinate chart U , the
submanifold Y appears as Y X U “ tfU pzq “ 0u where fU pzq is a local holomorphic function. The
transition function of OpY q is given by tU V “ fU {fV on U X V . If Y X U “ H, we can take
fU “ 1.
‚ Note: if another choice of local defining function is taken, by (1.5) it follows that this defines an
isomorphic bundle.
‚ Note: there is a global section s P H 0 pOpY qq given by the local data pU, sU q with sU “ fU , since
sU “ tU V sV is tautology.
Example 1.14. Let P pZ0 , . . . , Zn q be a homogeneous polynomial of degree k, and let Y “ tP “
0u Ă Pn . Then OpY q “ Opkq. To see this, in the local chart U0 Ă Pn the equation in coordinates
pU0 , zq is 0 “ P pZ0 , . . . , Zn q{Z0k “ s0 and in the local chart U1 the equation in coordinates pU1 , z̃q
is 0 “ P pZ0 , . . . , Zn q{Z1k “ s1 . The transition function t10 is then
„ k
P pZ0 , . . . , Zn q{Z1k Z0
t10 “ “
P pZ0 , . . . , Zn q{Z0k Z1
which matches with the transition functions of Opkq.

12
Let Y Ď X be a smooth analytic hypersurface. This means that at p P U , the local defining function
U X Y “ tf pzq “ 0u has the property that Bi f ppq ‰ 0 for some coordinate direction Bi . In this case,
there exists new holomorphic local coordinates tz̃ i u such that (after possibly shrinking U )
U X Y “ tz̃ n “ 0u.
To see this, let tz i u be the original holomorphic coordinates and suppose after relabeling that
Bf
Bz n ppq ‰ 0. By the holomorphic implicit function theorem, after possibly shrinking U we have

U X Y “ tpw1 , . . . , wn´1 , ψpwqq : w P V u

where ψ : V Ñ C is a holomorphic function and V Ď Cn´1 . New coordinates are then given
by
z̃ 1 “ z 1 , . . . , z̃ n´1 “ z n´1 , z̃ n “ z n ´ ψpzq,
and these satisfy z̃ n pqq “ 0 if and only if q P V .
‚ Canonical bundle. The canonical bundle of a complex manifold is KX “ pdet T 1,0 M q˚ . The tran-
Bz̃ ´1
sition functions on pU, zq an overlap pŨ , z̃q are pdet Bz q . Sections Ω P ΓpKX q are denoted
Ω “ f dz 1 ^ ¨ ¨ ¨ ^ dz n ,
p
and the transformation law is f˜ “ detp Bz
Bz̃ q qf .

Proposition 1.15. (Adjunction formula) Let Y Ď X be a smooth analytic hypersurface.

KY “ pKX b OpY qq|Y

Proof. Locally Y is given by tz n “ 0u for suitable holomorphic coordinates pz 1 , . . . , z n q. On an

overlap of open sets, suppose both z̃ n “ 0 and z n “ 0 carve out Y . Then z̃ n pzq is a holomorphic
function of z which vanishes on z n “ 0, and so the theory of holomorphic functions implies that we
can write
z̃ n pzq “ z n f pzq
for a holomorphic function f . For i ď n ´ 1, we compute
Bz̃ n n Bf Bz̃ n Bf
i
“ z i
, n
“ f ` zn n .
Bz Bz Bz Bz
The transition function for T 1,0 X restricted to Y “ tz n “ 0u is then
» fi
ˇ
Bz̃ ˇ
ˇ A ˚
“– fl (1.8)
Bz ˇY 0 f
where A is transition function for T 1,0 Y using coords z 1 , . . . , z n´1 . Therefore
Bz̃
det “ pdet Aqf (1.9)
Bz
and the transition functions give the bundle isomorphism
detpT 1,0 Xq|Y “ pdet T 1,0 Y q b L,
where the line bundle L has transition function f . We can take the inverse to get the formula with
KY “ pdet T 1,0 Y q´1 . Note that L has the same transition function as OpY q. By definition, if
locally z n “ 0 and z̃ n “ 0 carve out Y , the transition function of OpY q is z̃ n {z n which in this case
is f pzq.

13
1.2.3 Constructions from the tangent bundle
We now list some bundle constructions which come from the holomorphic tangent bundle T 1,0 M .
‚ Complexified tangent bundle. The conjugate of T 1,0 M is denoted T 0,1 M :“ T 1,0 M . A local
frame is given by " *
B B
, . . . , ,
Bz̄ 1 Bz̄ n
and sections denoted V ī Bz̄B i . We can write the px, yq coordinate basis in terms of the pz, z̄q coordinate
basis by the change of variables z k “ xk ` iy k , z̄ k “ xk ´ iy k and the chain rule:
„ 
B B B B B B
“ ` , “ i ´ .
Bxk Bz k Bz̄ k By k Bz k Bz̄ k
It follows that the complexified tangent bundle TC M can be written as a direct sum
TC M “ T 1,0 M ‘ T 0,1 M.

Example 1.16. We will sometimes use the notion of a complex structure J : TC M Ñ TC M . Given
a complex manifold, the complex structure J is defined by setting
J|T 1,0 M “ `iId, J|T 0,1 M “ ´iId,
or in other words, J BzBk “ i BzBk and J Bz̄Bk “ ´i Bz̄Bk . In components,
J p q “ iδ p q , J p̄ q̄ “ ´iδ p q , J p q̄ “ J p̄ q “ 0,
in complex coordinates.
‚ Complexified cotangent bundle. We denote smooth sections of the dual of the holomorphic cotan-
gent bundle pT 1,0 M q˚ by Λ1,0 pM q, and holomorphic sections of this bundle by H 0 pM, pT 1,0 M q˚ q.
A local frame is given by
tdz 1 , . . . , dz n u
meaning dz k pBzi q “ δ k j , and so that a section α P Λ1,0 pM q is written α “ αi dz i and αpV q “ αi V i
for V “ V i Bzi P ΓpM, T 1,0 M q. Transformation laws for components and frames are
Bz p Bz̃ k i
α̃i “ αp , dz̃ k “ dz .
Bz̃ i Bz i
Denote Λ0,1 pM q “ Λ1,0 pM q. Complexified 1-forms Λ1C M can be decomposed as
Λ1C “ Λ1,0 pM q ‘ Λ0,1 pM q,
from the decompositions (1.1)
1 1
dxk “ pdz k ` dz̄ k q, dy k “ pdz k ´ dz̄ k q. (1.10)
2 2i

‚ Differential forms. Let z k “ xk `iy k be local complex coordinates, and write w “ px1 , . . . , xn , y 1 , . . . , y n q P
R2n . A differential form on M appears in w coordinates as
1
η“ ηi ¨¨¨i dwi1 ^ ¨ ¨ ¨ ^ dwik .
k! 1 k

14
From (1.10), we see that this can be written in the complex basis of dz, dz̄. We will use the following
convention for complex components: ÿ
η“ η p,q
p`q“k

with
1
η p,q “ η dz i1 ^ ¨ ¨ ¨ ^ dz ip ^ dz̄ j1 ^ ¨ ¨ ¨ ^ dz̄ jq .
p!q! i1 ¨¨¨ip j̄1 ¨¨¨j̄q
We call η p,q a pp, qq-form, denoted Λp,q pM q.
‚ Exterior derivative. The exterior derivative acting on a function is
Bf i Bf
df “ i
dx ` i dy i
Bx By
which in complex coordinates becomes
Bf i Bf
df “ dz ` i dz̄ i .
Bz i Bz̄
We write this as df “ Bf ` B̄f , with
Bf i Bf i
Bf “ dz , B̄f “ dz̄ .
Bz i Bz̄ i
Similarly, the exterior derivative d : Λk Ñ Λk`1 on higher differential forms decomposes into
types.
d “ B ` B̄.
Acting on χ P Λp,q pM q, we have

1 B
Bχ “ χ dz ` ^ dz i1 ^ ¨ ¨ ¨ ^ dz ip ^ dz̄ j1 ^ ¨ ¨ ¨ ^ dz̄ jq
p!q! Bz ` i1 ¨¨¨ip j̄1 ¨¨¨j̄q
and
1 B
B̄χ “ χ dz̄ ` ^ dz i1 ^ ¨ ¨ ¨ ^ dz ip ^ dz̄ j1 ^ ¨ ¨ ¨ ^ dz̄ jq
p!q! Bz̄ ` i1 ¨¨¨ip j̄1 ¨¨¨j̄q
so that B : Λp,q Ñ Λp`1,q and B̄ : Λp,q Ñ Λp,q`1 .
Example 1.17. For α P Λ1,1 , we write α “ αj k̄ dz j ^ dz̄ k and

Bα “ B` αj k̄ dz ` ^ dz j ^ dz̄ k
1
“ pB` αj k̄ ´ Bj α`k̄ qdz ` ^ dz j ^ dz̄ k
2
1
“ pBαq`j k̄ dz ` ^ dz j ^ dz̄ k . (1.11)
2
and the components formula is
pBαq`j k̄ “ B` αj k̄ ´ Bj α`k̄ .

15
1.3 Geometry of bundles
1.3.1 Chern connection
Let E Ñ M be a holomorphic vector bundle of rank r over a complex manifold. A hermitian metric
on E is H P ΓpE ˚ b Ē ˚ q which is represented in a local frame teα u of E by

H “ Hαβ̄ eα b eβ

with Hαβ̄ ppq a positive-define r ˆ r hermitian matrix at all points p. The hermitian condition is
Hαβ̄ “ Hβ ᾱ . Our conventions for the inner product on E given by H is

xu, vy “ ui Hik̄ v k , u, v P ΓpEq

so that xu, λvy “ λ̄xu, vy. The hermitian condition is xu, vy “ xv, uy. The norm of a section is
|u|2 “ xu, uy.
We note that xu, vy does not depend on the choice of trivialization. Let U, Ũ be two trivializations
of E with transition matrix rQs “ Qα β and denote components on Ũ with tildes, so for u P ΓpEq
and ϕ P ΓpE ˚ q we have
ũα “ Qα β uβ , ϕ̃α “ ϕα pQ´1 qα β
and the transformation law on H P ΓpE ˚ b Ē ˚ q is

H̃αβ̄ “ pQ´1 qµ α Hµν̄ pQ´1 qν β .

From here we can verify ũα H̃αβ̄ ṽ β “ uα Hαβ̄ v β . This can also be written using matrix nota-
tion:

xu, vy “ uT H v̄,
ũ “ Qu, H̃ “ pQ´1 qT HQ´1 , (1.12)

and it is straightforward to verify that ũT H̃ ṽ “ uT H v̄. In other words, though the direction of uα
as a column vector is not a well-defined quantity (depends on the choice of trivialization), its norm
|u|H is a measurable number.
The inverse of H is denoted in components as H ᾱβ , so that HH ´1 “ I becomes in components
Hαβ̄ H β̄γ “ δα γ . The inverse H ´1 produces a metric on E ‹ .

xψ, ϕy “ H ᾱβ ψβ ϕα , ψ, ϕ P ΓpE ˚ q.

Similarly as above, it can be verified that xψU , ϕU y “ xψŨ , ϕŨ y, so that xψ, ϕy takes two sections
and produces a global function on X.
The metric can be used to raise and lower indices. From uα P ΓpEq, we will write

uβ̄ “ uα Hαβ̄ ,

and uβ̄ defines a section of Ē ˚ . This is because if u ÞÑ Qu and H ÞÑ pQ´1 qT HQ´1 , then uT H ÞÑ
uT HQ´1 “ pQ´1 qT puT Hq. Said another way, given u P ΓpEq and a metric H, we obtain a dual
element u˚ P ΓpĒ ˚ q defined by
u˚ pv̄q “ xu, vyH .

16
Similarly, from uα P ΓpĒ ˚ q, then uα “ H αβ̄ uβ̄ is a section of ΓpEq. We note that

uα v α “ uᾱ vᾱ .

Definition 1.18. The Chern connection of a metric H on a holomorphic bundle E is a map

∇ : ΓpEq Ñ ΓpE b pTC M q˚ q given by

∇Bk puα eα q “ pBk uα ` uβ pBk Hβ ν̄ qH ν̄α qeα

∇Bk̄ puα eα q “ pBk̄ uα qeα . (1.13)

We often just write this in components as

∇k̄ uα “ Bk̄ uα ,
∇k uα “ Bk uα ` uβ Akβ α , Akβ α “ Bk Hβ ν̄ H ν̄α , (1.14)

or without indices as
∇ “ pB ` BHH ´1 q ` B̄.
For ∇k̄ sα to be a section, we need to verify that if pŨ , s̃α q and pU, sα q are two overlapping trivial-
izations of E with s̃α “ Qα β sβ , then

∇k̄ s̃α “ Qα β ∇k̄ sβ .

This is true because B̄Qα β “ 0. It can also be checked directly that ∇k sα is a section, namely

∇k s̃α “ Qα β ∇k sβ ,

by using the transformation law for H.

Recall that a general connection ∇ on a complex vector bundle E is a map ∇ : ΓpEq Ñ ΓpE bT ˚ M q
such that ∇pas1 ` bs2 q “ a∇s1 ` b∇s2 and ∇pf sq “ df b s ` f ∇s. We will also sometimes call ∇
a covariant derivative. The Chern connection is the most commonly used choice of connection on
holomorphic bundles, and it is characterized by the following uniqueness statement:
Lemma 1.19. Let pE, Hq be a holomorphic bundle with hermitian metric. The Chern connection
is the unique connection satisfying ∇0,1 “ B̄ and

Bk xu, vy “ x∇k u, vy ` xu, ∇k̄ vy. (1.15)

Proof. Let ∇ be a connection satisfying (1.15) with ∇0,1 “ B̄. We will solve for ∇1,0 . Our notation
for the unknown connection is
∇Bk eα “ Akα β eβ
where A are unknown coefficients to be solved. In other words

∇k uα “ Bk uα ` Akβ α uβ .

If we require (1.15), then in coordinates this becomes

Bk pHij̄ ui v j q “ Hij̄ ∇k ui v̄ j̄ ` Hij̄ ui Bk̄ v j

17
which simplies to
Bk Hij̄ ui v j “ Hrj̄ Aki r ui v j
If this is true for all sections u, v, then
Bk Hij̄ “ Aki r Hrj̄
and solving for A gives
Aki ` “ Bk Hij̄ H j̄`
or A “ BHH ´1 .

There is a formula for how the Chern connection changes when changing the metric. Let Ĥ and H
be two metrics on E. Let A “ BHH ´1 and h “ H Ĥ ´1 . Then
A “ BHH ´1
“ BphĤqĤ ´1 h´1
“ Bhh´1 ` hÂh´1
“ Bhh´1 ` hÂh´1 ` p ´ Âhh´1 q
“ ˆ ´1
 ` ∇hh (1.16)
ˆ “ Bh ` h ´ Âh in matrix notation, or using index notation for the component of
where ∇h
hα β “ Hαµ̄ Ĥ µ̄β , then
∇i hα β “ Bi hα β ` hα γ Aiγ β ´ Aiα γ hγ β . (1.17)
Here is the reason for the term with a minus sign. Let ∇ be a connection on E acting on sections
u P ΓpEq by
∇i uα “ Bi uα ` uβ Aiβ α , Aiβ α “ Bi Hβ ν̄ H ν̄α .
The induced dual connection acting on ϕ P ΓpE ˚ q is defined with a minus sign:
∇i ϕα “ Bi ϕα ´ Aiα β ϕβ .
This minus sign is introduced so that we can differentiate contracted indices using the product
rule
Bi puα ϕα q “ p∇i uα qϕα ` uα p∇i ϕα q.
The formula (1.17) follows from the rule for covariant differentiation where each upper index receives
a `A term and each lower index receives a ´A term. As another example,
∇i T α βγ “ Bi T α βγ ` T µ βγ Aiµ α ´ T α µγ Aiβ µ ´ T α βµ Aiγ µ ,
for T P ΓpE b E ˚ b E ˚ q. Sections of Ē follow the rules
∇i uᾱ “ Bi uᾱ , ∇ī uᾱ “ Bī uᾱ ` uν̄ Aiν α
¯ for u P ΓpEq. For example,
so that ∇ū “ ∇u
∇i Gαβ̄ “ Bi Gαβ̄ ´ Aiα ν Gν β̄
for G P ΓpE ˚ b Ē ˚ q. From this formula and Ai “ pBi HqH ´1 , we see that
∇i Hαβ̄ “ 0, ∇ī Hαβ̄ “ 0
when ∇ is the Chern connection of H.

18
1.3.2 Curvature
The curvature of the Chern connection defines a notion of Hessian of a metric tensor H. The
local combinations Bj Bk̄ Hαβ̄ do not transform as the section of any bundle, and taking covariant
derivatives gives zero: ∇i Hαβ̄ “ 0. The curvature tensor is a way to encode second derivatives of
the metric.
Definition 1.20. Let pE, Hq be a holomorphic vector bundle with metric. The curvature of the
Chern connection F P ΓpΛ1,1 b End Eq is given by F “ B̄pBHH ´1 q. In components

F “ Fβ α j k̄ eα b eβ dz j ^ dz̄ k ,

the definition is
Fα β j k̄ “ ´Bk̄ pBj Hαµ̄ H µ̄β q,
or
Fj k̄ “ ´Bk̄ pBj HH ´1 q,
without showing the endomorphism indices.
The action of Fj k̄ P ΓpEnd Eq on u P ΓpEq is

uα ÞÑ uβ Fβ α j k̄ .

We now verify that the formula for Fj k̄ gives a well-defined section of End E. The transformation
law H̃ “ pQ´1 qT HQ´1 implies
„ 
F̃j k̄ “ ´Bk̄ Bj ppQ´1 qT HQ´1 qpQH ´1 QT q
„  „ 
“ ´Bk̄ p´QT Bj QT q ` ´Bk̄ ppQ´1 qT Bj HH ´1 QT q
„ 
“ pQ q ´ Bk̄ pBj HH q QT
´1 T ´1
(1.18)

using B̄Q “ 0, B Q̄ “ 0. This matches with (1.7), and so Fj k̄ P ΓpEnd Eq acting on sections of E on
the right.
Remark 1.21. Let L Ñ M be a holomorphic line bundle. Then H is a 1 ˆ 1 matrix, and the
transformation law for a metric reads
1
H̃ “ H. (1.19)
|tŨ U |2

The formula for the curvature is

Fj k̄ “ ´Bk̄ Bj log H, iF “ ´iB B̄ log H P Λ1,1 pM q,

and it can also be checked directly that F̃j k̄ “ Fj k̄ .

19
Remark 1.22. The formula for change of curvature is
ˆ ´1 q.
F “ F̂ ` B̄p∇hh

where H, Ĥ are two metrics and h “ H Ĥ ´1 . This is because (1.16) implies

ˆ ´1 q.
B̄A “ B̄ Â ` B̄p∇hh

Remark 1.23. The formula for Chern curvature is consistent with the general formula for the
curvature of a connection. In general, the curvature of a connection ∇ on E is F “ dA ´ A ^ A,
F “ 12 Fj i µν dxµ ^ dxν b ej b ei , with

Fj i µν “ Bµ Aνj i ´ Bν Aµj i ´ Aµj r Aνr i ` Aµj r Aµr i .

For the Chern connection, ∇0,1 “ B̄, so Aīα β “ 0. Therefore Fα β k̄j̄ “ 0, and since the Chern
connection is unitary then Fα β kj “ 0 which can also be checked directly. Therefore F only has
mixed p1, 1q form indices, and
Fj i mn̄ “ ´Bn̄ Amj i
since mixed connection terms are zero.
Example 1.24. Fubini-Study metric on Op1q Ñ P1 . Recall that P1 is covered by two trivializations
pU0 , zq, pU1 , z̃q, with change of coordinates z̃ “ z ´1 and sections s P ΓpOp1qq transform as s̃ “ z̃s.
The Fubini-Study metric is defined as

hpzq “ p1 ` |z|2 q´1 , h̃pz̃q “ p1 ` |z̃|2 q´1 .

This transforms correctly as (1.19): h̃ “ p1{|z̃|2 qh. In other words, the norm

|s|2h “ ss̄h

gives the same result in either trivialization. The curvature is a 2-form iF P Λ1,1 with components

Fzz̄ “ ´Bz̄ Bz log h.

We compute
z̄ 1 |z|2
Fzz̄ “ Bk̄ “ ´ “ p1 ` |z|2 q´2 ą 0.
1 ` |z|2 1 ` |z|2 p1 ` |z|2 q2
Therefore iF is a closed positive p1, 1q form; this is a Kähler metric on P1 . The Fubini-Study Kähler
metric is sometimes denoted
ωF S “ iB B̄ logp1 ` |z|2 q.

1.3.3 Hermitian geometry

In this section, we focus on metrics and curvature on the holomorphic tangent bundle T 1,0 X. We
will denote a hermitian metric on T 1,0 M by g, with components gj k̄ .

g “ gj k̄ dz j b dz̄ k .

20
The collection of local matrices pU, gj k̄ q are related by

Bz ` Bz m
g̃j k̄ “ j
g`m̄ k .
Bz̃ Bz̃
Locally we view g as a matrix, e.g. in 2 dimensions
» fi
g11̄ g12̄
g“– fl .
g21̄ g22̄

The inner product on T 1,0 M is then

B B
gpV, W q “ V j gj k̄ W̄ k̄ , V “Vi i
, W “ Wi i.
Bz Bz
A Kähler metric on M is a metric satisfying Bi gj k̄ “ Bj gik̄ . Another way to express this is to
associate a p1, 1q form
ω “ igj k̄ dz j ^ dz̄ k
and require dω “ 0. Note that the factor of i is included in the definition so that ω is real: ω̄ “ ω.
Direct calculation gives
ωn
“ det gj k̄ idz 1 ^ dz̄ 1 ^ ¨ ¨ ¨ ^ idz n ^ dz̄ n , (1.20)
n!
and ω n {n! is a nowhere vanishing top form which will later be used for integration.
Remark 1.25. A hermitian metric g “ gj k̄ on T 1,0 M produces a Riemannian metric on the real
tangent bundle T M , which we also denote by g. Let us temporarily write this metric as gR . Let
z j be holomorphic coordinates (indices j, k) and wα “ pz 1 , . . . , z n , z̄ 1 , . . . , z̄ n q be full coordinates
(indices α, β). Then we define
gR pBα , Bβ q “ gαβ
where we declare
gjk “ gj̄ k̄ “ 0, gj k̄ “ gk̄j .
In other words,
gR pX α Bα , Y β Bβ q “ gj k̄ X j Y k̄ ` gj̄k X j̄ Y k

where X “ X i BzB i ` X ī Bz̄B i . From this perspective, the p1, 1q-form ω may be defined as ωpX, Y q “
gR pJX, Y q since ωpBj , Bk̄ q “ igj k̄ .
We can also write the metric g (we now stop using the notation gR ) in terms of q “ px1 , y 1 , . . . , xn , y n q
coordinates where z k “ xk ` iy k . We give the details in complex dimension 1. Let z “ x ` iy be
complex coordinates, and we would like to write
» fi
gxx gxy
g“– fl
gyx gyy

21
A hermitian metric is given by a mixed type gzz̄ ą 0, and we declare gzz “ 0 and gz̄z̄ “ 0. Changing
coordinates

gxx “ gpBz ` Bz̄ , Bz ` Bz̄ q “ 2gzz̄ ,

gxy “ gpBz ` Bz̄ , ipBz ´ Bz̄ qq “ 0,
gyy “ gpipBz ´ Bz̄ q, ipBz ´ Bz̄ qq “ 2gzz̄ . (1.21)

The corresponding Riemannian metric in px, yq coordinates is then diagonal with

» fi
2gzz̄ 0
g“– fl .
0 2gzz̄

Note: there is a way to go from a Riemannian metric on T M to a hermitian metric on T 1,0 M , but
this requires that g is compatible with the complex structure so that gpJX, JXq “ gpX, Xq and we
omit the details.
We will use a normalization factor of p!q! for the inner product defined on pp, qq forms. For ϕ, ψ P
Λp,q with
1
ϕ“ “ ϕα1 ¨¨¨αp β̄1 ¨¨¨β̄q dz α1 ^ ¨ ¨ ¨ ^ dz αp ^ dz̄ β1 ^ ¨ ¨ ¨ ^ dz̄ βq ,
p!q!
and
ϕ̄α1 ¨¨¨αp β̄1 ...β̄q “ g ā1 α1 ¨ ¨ ¨ g āp αp g β̄1 b1 ¨ ¨ ¨ g β̄q bq ϕa1 ¨¨¨ap b̄1 ¨¨¨b̄q
then we will use the convention
1
xϕ, ψy “ ϕ ψ̄ α1 ¨¨¨αp β̄1 ...β̄q . (1.22)
p!q! α1 ¨¨¨αp β̄1 ¨¨¨β̄q

The Chern connection ∇ on the bundle T 1,0 X Ñ X is defined by

¯
∇k̄ V i “ Bk̄ V i , ∇k V i “ Bk V i ` V p Γkp i , Γkp i “ Bk gp`¯g `i .

and a direct check shows that ∇g “ 0 and ∇J “ 0.

The curvature of the Chern connection on the tangent bundle will be denoted R P Λ1,1 pEnd T 1,0 M q,
so that
Rj i mn̄ “ ´Bn̄ Γmj i
since Amj i “ Γmj i . Explicitly in terms of the metric, the curvature of the Chern connection is
given by
Rj i mk̄ “ ´Bk̄ pBm gj p̄ g p̄i q.
We can lower the second index
Rj p̄mk̄ “ Rj i mk̄ gip̄
and the explicit formula in terms of the metric is

Rj p̄mk̄ “ ´Bk̄ Bm gj p̄ ` pBm gj q̄ qg q̄a pBk̄ gap̄ q

22
using the variation formula for the inverse of a matrix δA´1 “ ´A´1 pδAqA´1 . The conjugate of
this is
Rj p̄mk̄ “ ´Bm̄ Bk gpj̄ ` pBk gpā qg āq pBm̄ gqj̄ q
so
Rj p̄mk̄ “ Rpj̄km̄ , Rj i mk̄ “ g īp Rpj̄km̄ “ Rī j̄km̄ .
The curvature appears when exchanging covariant derivatives. We will write

r∇m , ∇k̄ sV i “ ∇m ∇k̄ V i ´ ∇k̄ ∇m V i

where ∇m ∇k̄ V i means: let Wk̄ i “ ∇k̄ V i be a tensor and compute the components ∇m Wk̄ i . The
commutator formula is
r∇m , ∇k̄ sV i “ V p Rp i mk̄ . (1.23)
Here is the check:

r∇m , ∇k̄ sV i “ Bm ∇k̄ V i ` Γm` i ∇k̄ V ` ´ Bk̄ ∇m V i

“ Bm Bk̄ V i ` Γm` i Bk̄ V ` ´ Bk̄ pBm V i ` Γm` i V ` q
“ ´Bk̄ Γm` i V ` “ R` i mk̄ V ` . (1.24)

Remark 1.26. A similar calculation gives that the commutator r∇m , ∇k sV i involves another
tensor, the torsion tensor, but we will not need this formula.
Taking the conjugate of (1.23) gives

r∇k , ∇m̄ sV ī “ ´V p̄ Rī p̄km̄ .

We can lower the index by introducing gaī , since ∇k gaī “ 0.

r∇k , ∇m̄ sVa “ ´Ra p km̄ Vp .

Similarly
r∇k , ∇m̄ sVā “ Rp̄ āmk̄ Vp̄ .
Let
Rj k̄ “ Rp p j k̄ “ g q̄p Rpq̄j k̄
and define the Chern-Ricci form by

iRicω “ iRj k̄ dz j ^ dz̄ k .

Viewing R P Λ1,1 pEnd T 1,0 Xq, we have iRicω “ iTr R P Λ1,1 pXq where we trace out the endomor-
phism indices and retain the 2-form indices. Using the general formula for the derivative of the
determinant of an invertible hermitian matrix A,
ˇ
d ˇˇ
det Aptq “ det Ap0q Tr rAp0q´1 Ap0qs
9
dt ˇt“0

we obtain Bj log det g “ g r̄p Bj gpr̄ , and so

Rj k̄ “ ´Bk̄ Bj log det g, (1.25)

23
and
iRicω “ ´iB B̄ log det g.
This expression can also be connected to Riemannian geometry: in the case when gij̄ is a Kähler
metric, then it turns out that Rj k̄ is the Levi-Civita Ricci tensor of the Riemannian metric g on T X.
This calculation can be found in Kähler’s original paper (see p.178 in [17], where (1.25) is described
as “very elegant”), and it is one of the main motivations for the field of Kähler geometry.
We note that when g is a general hermitian metric on T 1,0 X, the Chern-Ricci curvature Rj k̄ (1.25)
is different than the Riemannian Levi-Civita Ricci tensor.
Remark 1.27. From the point of view of Riemannian geometry, it looks like Rj k̄ “ Rp p j k̄ is
tracing the wrong index in the definition of the Ricci curvature and should be zero. This is not the
case because we are only tracing over holomorphic indices. Tracing over all real indices does indeed
give zero. Let α, β represents real coordinates xα “ pz 1 , . . . , z n , z̄ 1 , . . . , z̄ n q and a, b, j, k represents
holomorphic coordinates z i , and let Rαβγµ be the Riemannian curvature tensor. Then

g αβ Rαβj k̄ “ g ab̄ Rab̄j k̄ ` g āb Rābj k̄ “ g ab̄ pRab̄j k̄ ` Rb̄aj k̄ q “ 0,

since the Riemannian curvature tensor satisfies Rαβγµ “ ´Rβαγµ . But the trace

g ab̄ Rab̄j k̄

is not summing over all coordinate indices α, β, and need not be zero.
Example 1.28. We start with the line bundle Op1q Ñ Pn with trivializations Ui “ tZi ‰ 0u.
Recall that the transition functions pUi X Uj , tij q are tij “ Zj {Zi .
‚ Show that the collection pUi , hi q with

|Zi |2
hi “ ř 2
k |Zk |

defines a metric on Op1q Ñ Pn . We call h the Fubini-Study metric.

‚ Compute the curvature iF “ iB B̄ log h of the Fubini-Study metric, and show that in local coordi-
nates on Ui then
n
ÿ
iF “ iB B̄ logp1 ` |z|2 q, |z|2 “ |z i |2 . (1.26)
i“1

We will show below that iF :“ ωF S is a Kähler metric on the base Pn .

‚ We also sometimes refer to the following expression

ωF S “ iB B̄ logp1 ` |z|2 q, ωF S “ igj k̄ dz j ^ dz̄ k ,

as the Fubini-Study metric on Pn . We see that dωF S “ 0, and to show ωF S is a Kähler metric, we
need to verify that gj k̄ is a positive-definite matrix. A computation gives

p1 ` |z|2 qδjk ´ zk z̄j

gj k̄ “ .
p1 ` |z|2 q2

24
For ξ P Cn , we have

p1 ` |z|2 q|ξ|2 ´ |zk ξ¯k |2

gj k̄ ξ j ξ¯k “
p1 ` |z|2 q2
p1 ` |z|2 q|ξ|2 ´ |z|2 |ξ|2
ě
p1 ` |z|2 q2
2
|ξ|
“ ą 0. (1.27)
p1 ` |z|2 q2

Therefore g is a hermitian metric.

‚ Compute the Chern-Ricci curvature of ωF S . A computation of the determinant gives

det gj k̄ “ p1 ` |z|2 q´pn`1q .

From here, one can compute that

Rj k̄ “ pn ` 1qgj k̄ .
Another way to write this is iRicpωF S q “ pn ` 1qωF S . Kähler metrics satisfying iRicpωq “ λω for
λ P R are said to be Kähler-Einstein.
Example 1.29. Suppose X is a complex manifold with an embedding into projective space i :
X Ñ PN . Then i˚ ωF S is Kähler metric on X. In other words, projective manifolds are Kähler.
For an interesting converse, see the Kodaira embedding theorem: this states that a compact Kähler
manifold pX, ωq with rωs P H 2 pX, Zq admits an embedding i : X Ñ PN into some projective space
PN .

25
2 Kähler Manifolds
This section will follow the textbook by Kodaira-Morrow [20].

2.1 Hodge theory

We start with the definition of the Dolbeault cohomology groups. Let X be a compact complex
manifold and E Ñ X a holomorphic vector bundle. Let Λp,q pEq be smooth pp, qq forms with
coefficients in E, so that sections s P Λp,q pEq are of the form s “ u b η with u P ΓpEq and
η P Λp,q pXq. The Dolbeault cohomology groups are defined
kerpB̄ : Λ0,q pEq Ñ Λ0,q`1 pEqq
H q pX, Eq “
im pB̄ : Λ0,q´1 pEq Ñ Λ0,q pEqq
Letting E “ Ωp be the holomorphic bundle of pp, 0q forms, we also define
kerpB̄ : Λp,q pXq Ñ Λp,q`1 pXqq
HB̄p,q pXq :“ H q pX, Ωp q “
im pB̄ : Λp,q´1 pXq Ñ Λp,q pXqq
and the Hodge numbers are
hp,q “ dim HB̄p,q pXq.
The theme of Hodge theory is to represent the cohomology class rϕs by a unique optimal repre-
sentative ϕ0 P rϕs. The selected representative from the equivalence class is found by solving an
elliptic PDE. In this particular case, we will look for solutions ϕ0 P rϕs to the Laplace equation
∆B̄ ϕ0 “ 0.
To define the Laplacian, we must equip pE, Xq with metrics: let H be a metric on E and g a
metric on the base X. As described in earlier sections, this defines a pointwise inner product on
ϕ1 , ϕ2 P Λp,q pEq:
xϕ1 , ϕ2 yg,H : X Ñ C.
The L2 inner product on Λp,q pEq is then
ωn
ż
pϕ1 , ϕ2 qL2 “ xϕ1 , ϕ2 yg,H .
X n!
The Dolbeault operator B̄ : Λp,q pEq Ñ Λp,q`1 pEq has an L2 -adjoint denoted B̄ : : Λp,q pEq Ñ
Λp,q´1 pEq. The adjoint satisfies
pB̄ϕ1 , ϕ2 qL2 “ pϕ1 , B̄ : ϕ2 qL2 , ϕ1 P Λp,q´1 pEq, ϕ2 P Λp,q pEq.
The B̄-Laplacian ∆B̄ : Λp,q pEq Ñ Λp,q pEq is then defined by
∆B̄ “ B̄ B̄ : ` B̄ : B̄.
There is an explicit formula for the adjoint in local coordinates. Roughly speaking, B̄ is a like a curl
operator while B̄ : is like a divergence operator. We give the formula without the additional vector
bundle E Ñ X, so that we consider usual pp, qq forms with adjoint B̄ : : Λp,q pXq Ñ Λp,q´1 pXq. The
formula in local coordinates is
„ 
¯
IJ ¯
pIJ
: ´1
pB̄ ϕq “ ´pdet gq Bp pdet gqϕ , ϕ P Λp,q pXq. (2.1)

26
Here I “ i1 ¨ ¨ ¨ ip , J “ j1 ¨ ¨ ¨ jq are multi-indices, and we raise indices using the metric: e.g. if
1
ϕ “ q! ϕJ¯dz̄ J is p0, qq form, then ϕJ “ g j1 ā1 ¨ ¨ ¨ g jq āq ϕā1 ¨¨¨āq , or e.g. if ϕ “ ϕrs̄ dz r ^ dz̄ s is a
p1, 1q-form, then ϕīj “ g īr g s̄j ϕrs̄ .
Let us verify formula (2.1) in the special case of p0, 2q forms. Let θ P Λ0,2 pXq and ϕ P Λ0,1 pXq.
Then
ϕ “ ϕī dz̄ i , pB̄ϕqj̄ k̄ “ Bj̄ ϕk̄ ´ Bk̄ ϕj̄ .
Using the inner product (1.22) we have
1 j̄i k̄`
xB̄ϕ, θyg “ g g pBj̄ ϕk̄ ´ Bk̄ ϕj̄ qθī`¯ “ Bj̄ ϕk̄ θ̄j̄ k̄
2
and
Bj̄ ϕk̄ θ̄j̄ k̄ pdet gq “ ´ϕk̄ Bj̄ pθ̄j̄ k̄ pdet gqq ` Bj̄ rϕk̄ pθ̄j̄ k̄ pdet gqqs.
Let W k̄ “ pdet gq´1 Bj̄ ppdet gqθ̄j̄ k̄ q. Multiplying by idz 1 ^ dz̄ 1 ¨ ¨ ¨ ^ idz n ^ dz̄ n and using (1.20)
gives
ωn ωn ωn
xB̄ϕ, θy “ ´ϕk̄ W k̄ ` dιV .
n! n! n!
The last term involves V j̄ “ ϕk̄ θj̄ k̄ and will be explained below. Integrating this identity over X
and applying Stokes’s theorem gives
ωn
ż
pB̄ϕ, θqL2 “ ´ pϕk̄ W k̄ q .
X n!
We compare this with the definition of the adjoint:
k̄ ω n
ż
pB̄ϕ, θqL2 “ pϕk̄ pB̄ : θq q “ pϕ, B̄ : θqL2 .
X n!
Thus pB̄ : θqk “ ´W̄ k , and since
W̄ k “ pdet gq´1 Bp ppdet gqθpk q
this is the formula (2.1). We now explain why
ωn
Bj̄ rV j̄ pdet gqs idz 1 ^ dz̄ 1 ¨ ¨ ¨ ^ idz n ^ dz̄ n “ dιV .
n!
First, we recall the definition of the interior product: if V is a vector field, then ιV : Λk Ñ Λk´1
via
pιV ηqpW1 , . . . , Wk´1 q “ V i ηpBi , W1 , . . . , Wk´1 q,
and it satisfies ιV pη1 ^ η2 q “ ιV η1 ^ η2 ` p´1qk η1 ^ ιV η2 if η1 P Λk . Therefore applying ιV to (1.20)
gives
ωn
„ 
1̄ 1 2 2 2̄ 1 1 2
´ιV “ det g V pidz q ^ pidz ^ dz̄ q ^ p¨ ¨ ¨ q ` V pidz ^ dz̄ q ^ pidz q ^ p¨ ¨ ¨ q ` . . .
n!
and d “ B ` B̄ becomes
ωn
dιV “ Bk̄ rpdet gqV k̄ s pidz 1 ^ dz̄ 1 q ^ ¨ ¨ ¨ ^ pidz n ^ dz̄ n q,
n!

27
as claimed.
The B̄-Laplacian can be studied using techniques from the theory of elliptic PDE. We now give the
general definition of an elliptic operator.
Definition 2.1. Let E, F Ñ X be vector bundles trivialized by a finite cover X “ YN
i“1 Ui . An
elliptic operator of order k is a map L : ΓpEq Ñ ΓpF q such that:
‚ In each trivialization Ui Ă X, L appears as
ÿ ÿ
pLuqα “ AIα β BI uβ ` B Iα β BI uβ .
|I|“k 0ď|I|ăk

‚ Let p P X. Then for all ξ P Λ1 pX, Rq with ξp ‰ 0, then

σpL, ξqppq : Ep Ñ Fp

is an isomorphism. Here for ξ P Λ1 pX, Rq with ξ “ ξi dxi , we define σpL, ξq P ΓpHompE, F qq by

σpξqα β “ ξI AIα β , ξI “ ξi1 ¨ ¨ ¨ ξik .

In other words, for an elliptic operator the matrix rξI AI s is invertible.

Example 2.2. The B̄-Laplacian ∆B̄ : Λp,q pXq Ñ Λp,q pXq is an elliptic operator of order 2. A
calculation (illustrated below) shows that it is locally of the form

p∆ψqP Q̄ “ ´g j̄i Bi Bj̄ ψP Q̄ ` . . . , (2.2)

and therefore
σp∆, ξq “ p´g j̄i ξi ξj̄ q id “ ´|ξ|2g id.
Here we write ξ “ ξi dz i ` ξī dz̄ i for ξ P Λ1 pM, Rq, and since ξ is real then ξ¯ “ ξ and ξj̄ “ ξj .
We verify (2.2) for p0, 1q forms ϕ “ ϕī dz̄ i . By using the expression for the adjoint (2.1) and
pB̄ϕqj̄ k̄ “ Bj̄ ϕk̄ ´ Bk̄ ϕj̄ , we compute

pB̄ B̄ : ϕqᾱ “ Bᾱ pB̄ : ϕq

ˆ „ ˙
´1 q̄p
“ ´Bᾱ pdet gq Bp g pdet gqϕq̄ (2.3)

and
„ 
: ´1 pβ
pB̄ B̄ϕqᾱ “ ´pdet gq Bp pdet gqpB̄ϕq gβ ᾱ
„ 
´1 q̄p ν̄β
“ ´pdet gq Bp pg g det gqpBq̄ ϕν̄ ´ Bν̄ ϕq̄ q gβ ᾱ . (2.4)

The terms involving 2 derivatives of ϕ are

p∆ϕqᾱ “ ´g q̄p Bᾱ Bp ϕq̄ ´ g q̄p Bp pBq̄ ϕᾱ ´ Bᾱ ϕq̄ q ` . . .

and cancellation gives the result (2.2).

28
We say that an elliptic operator L is self-adjoint if

pLψ, ϕqL2 “ pψ, LϕqL2 , ψ P ΓpEq, ϕ P ΓpF q.

For example, ∆B̄ is self-adjoint.

Theorem 2.3. Let L be a self-adjoint elliptic operator on a vector bundle over a compact manifold.
There is an L2 orthogonal decomposition

ΓpEq “ ker L ‘ Im L.

Thus we can solve Lψ “ ϕ if and only if ϕ P pker LqK .

Proof. See Theorem 7.3 in the appendix of [19].

Applying this to ∆B̄ , we see that we can write any η P Λp,q pXq as

η “ h ` pB̄ B̄ : ` B̄ : B̄qβ,

where h P ker ∆B̄ , and this is usually written as

η “ h ` B̄β1 ` B̄ : β2 .

We will often use that

ker ∆B̄ “ tη : B̄η “ 0, B̄ : η “ 0u.
This can be seen by the formula

p∆B̄ η, ηqL2 “ pB̄η, B̄ηqL2 ` pB̄ : η, B̄ : ηqL2 .

As a consequence of this discussion, we obtain:

Corollary 2.4. Let X be a compact complex manifold with hermitian metric gij̄ . Every Dolbeault
cohomology class rηs P H p,q pXq admits a unique representative h P rηs with ∆B̄ h “ 0. Therefore:

dim ker ∆B̄ |Λp,q “ dim H q pX, Ωp q “ hp,q .

Proof. Write η “ h ` B̄β1 ` B̄ : β2 . Since B̄η “ 0, then

η “ h ` B̄β1 .

This is because pB̄ : β2 , B̄ : β2 q “ pβ2 , B̄ηq “ 0. It follows that rηs “ rhs for h P ker ∆B̄ . For uniqueness,
suppose η “ h1 ` B̄β1 “ h̃1 ` B̄ β̃1 . Then

0 “ ph1 ´ h̃1 q ` B̄pβ1 ´ β̃1 q

and so
0 “ ph1 ´ h̃1 , h1 ´ h˜1 qL2 ` pB̄pβ1 ´ β̃1 q, h1 ´ h˜1 qL2
and }h1 ´ h̃1 }2L2 “ 0 since B̄ : ph1 ´ h̃1 q “ 0.

29
There is a similar theory for vector bundle valued pp, qq forms, and in general

H q pX, E b Ωp q :“ HB̄p,q pEq

“ tη P Λp,q pEq : ∆B̄ η “ 0u
“ tη P Λp,q pEq : B̄η “ 0, B̄ : η “ 0u. (2.5)

Theorem 2.5. (Serre duality) Let E Ñ X be a holomorphic vector bundle over a compact complex
manifold. Then
dim HB̄p,q pEq “ dim HB̄n´p,n´q pE ˚ q,
which implies that
dim H q pX, Eq “ dim H n´q pX, KX b E ˚ q. (2.6)
p,q n´p,n´q
and h pXq “ h pXq.

Proof. We will use the Hodge star operator. This is a linear map

‹ : Λp,q Ñ Λn´q,n´p

defined by the property

ωn
ϕ ^ ‹ψ̄ “ xϕ, ψyg , ϕ, ψ P Λk . (2.7)
n!
Here xϕ, ψy is defined by zero if ϕ, ψ are of different pp, qq type. For example, if g is the Euclidean
metric on Cn ,
ωn ωn
dz 1 ^ ‹dz 1 “ xdz 1 , dz̄ 1 yg “ 0, dz̄ 1 ^ ‹dz 1 “ , (2.8)
n! n!
and so ‹dz 1 “ ´idz 1 ^ pidz 2 ^ dz̄ 2 q ^ ¨ ¨ ¨ ^ pidz n ^ dz̄ n q.
The Hodge star satisfies:
‚ ‹ψ “ ‹ψ̄
‚ ‹2 ψ p,q “ p´1qp`q ψ p,q
The first property can be verified by manipulating and taking the conjugate of (2.7). The second
property can be verified by manipulating (2.7) and using x‹ϕ, ‹ψy “ xϕ, ψy. To show ‹ preserves
the inner product, one can calculate in a similar way to (2.8) to show that ‹ takes an orthonormal
basis of Λp,q to an orthonormal basis Λn´q,n´p .
Next, we extend ‹ to vector bundle valued forms. For ϕ P Λp,q pEq, we can write ϕ “ ϕα b η p,q and
define
‹pϕα b η p,q q “ ϕα b ‹η p,q .
Equip E with a hermitian metric H. The L2 inner product can then be written as
ż
pϕ, ψqL2 “ ϕα ^ ‹ψ β Hαβ̄ .

The L2 adjoint of B̄ can be written as

pB̄ : ψqβ “ ´ ‹ H µ̄β BpHν µ̄ ‹ ψ ν q. (2.9)

30
This can be verified by Stokes’s theorem and substitution of (2.9) into the defining relation pB̄ϕ, ψq “
pϕ, B̄ : ψq. Indeed, for ϕ P Λp,q pEq and ψ P Λp,q`1 pEq, then
ż
pϕ, B̄ : ψq “ ϕα ^ ‹pB̄ : ψqβ Hαβ̄
ż ˆ ˙
α 2 β̄µ ν̄
“ ´ ϕ ^ ‹ H B̄pHµν̄ ‹ ψ̄ q Hαβ̄
ż
n´p n´q
“ p´1qp´1q p´1q ϕα ^ B̄pHαν̄ ‹ ψ̄ ν̄ q
ż
“ p´1qp´1qn´p p´1qn´q p´1qp´1qp`q B̄ϕα ^ ‹ψ̄ ν̄ Hαν̄

“ pB̄ϕ, ψq. (2.10)

Next, we define the map

# : Λp,q pEq Ñ Λn´p,n´q pE ˚ q
by
p#ψqβ “ Hβ ᾱ ‹ ψ α .
This satisfies ## “ p´1qp`q , and formula (2.9) can be written

pB̄ : ψqµ “ ´ ‹ B̄p#ψqβ H β µ̄ .

Therefore
HB̄p,q pEq “ tη P Λp,q pEq : B̄η “ 0, B̄ : η “ 0u
can be written in this notation as

HB̄p,q pEq “ tη P Λp,q pEq : B̄η “ 0, B̄#η “ 0u.

It follows that
η ÞÑ #η
is a map from H p,q pEq to H n´p,n´q pE ˚ q, and this is an isomorphism.
Note that here we used # because this would not have worked using ‹ : Λp,q pXq Ñ Λn´q,n´p pXq,
since ∆B̄ η “ 0 if and only if B̄η “ 0 and B ‹ η “ 0, and so if ∆B̄ η “ 0 then it is not necessarily true
that ∆B̄ ‹ η “ 0.

On a Kähler manifold, there are the following symmetries for the Hodge numbers.
Theorem 2.6. Let X be a compact Kähler manifold. Then
à
H k pX, Cq “ H p,q pXq, hp,q “ hq,p , hn´p,n´q “ hp,q .
p`q“k

If bk “ dim H k pX, Cq denote the Betti numbers of X, then

k
ÿ
bk “ hp,k´p . (2.11)
p“0

31
For example,

b1 “ h1,0 ` h0,1 “ 2h0,1 ,

b2 “ h2,0 ` h1,1 ` h0,2 “ 2h0,2 ` h1,1 . (2.12)

Here
kerpd : Λq pXq Ñ Λq`1 pXqq
H q pX, Cq “
im pd : Λq´1 pXq Ñ Λq pXqq
are the deRham cohomology groups, and Λk pXq denotes differential k-forms with coefficients in C.
The Laplacians are
∆d “ dd: ` d: d, ∆B “ ∆B̄ “ BB : ` B : B,
where d: , B : are the L2 adjoints of d, B. These Laplacians are elliptic operators and satisfy the
Hodge decomposition Theorem 2.3, and it follows from elliptic PDE theory that each de Rham
class rηs P H k pX, Cq admits a unique representative h P rηs with ∆d h “ 0.

Proof. This theorem follows from the Kähler Laplacian identities

1
∆B̄ “ ∆B “ ∆d . (2.13)
2
We will prove (2.13) below. We will show how (2.13) implies the result. From

H k pX, Cq “ tη P Λk : ∆d η “ 0u
p,q
H pXq “ tη P Λp,q : ∆B̄ η “ 0u, (2.14)

we can decompose η P Λk as η “ p`q“k η p,q . Since ∆B̄ preserves type, we have ∆d η “ 2∆B̄ η p,q .
ř ř
p`q“k
Therefore if η P Λk with ∆d η “ 0, then
ÿ
η ÞÑ η p,q
p`q“k

is a map from H k pX, Cq to ‘p`q“k H p,q pXq and this is an isomorphism.

Next, we prove hp,q “ hq,p . If η P H p,q pXq with ∆B̄ η “ 0, then η̄ P Λq,p pXq and (2.13) implies

∆B̄ η̄ “ ∆B̄ η̄ “ ∆B̄ η̄ “ 0.

Therefore η ÞÑ η̄ is a map from H p,q pXq to H q,p pXq and this is an isomorphism.

Example 2.7. We compute the Hodge numbers of Pn . The cell-decomposition Pn “ tptu Y C1 Y

C2 Y ¨ ¨ ¨ Y Cn implies
dim H 2k pPn , Cq “ 1, k P t0, 1, . . . u
and all odd cohomology groups are zero. Since H k pX, Cq “
À p,q
H pXq, we conclude

hk,k pPn q “ 1

and all other Hodge numbers are zero.

32
We now show ∆B̄ “ ∆B “ 12 ∆d . First, we note the following Kähler identities:

pB̄ : ϕqJ K̄ “ ´g āb ∇b ϕāJ K̄ , (2.15)

and
iB̄ : “ BΛω ´ Λω B
:
´iB “ B̄Λω ´ Λω B̄, (2.16)
where Λω : Λp,q Ñ Λp´1,q´1 is
pΛϕqJ K̄ “ ig āb ϕbāI J¯.
For example, for ω “ igj k̄ dz j ^ dz̄ k then Λω ω “ ´n.
We will discuss (2.15) in the following section on the Kodaira vanishing theorem, and we assume it
for now. Let us verify (2.16) for ϕ P Λ1,1 . Then
pBΛω ϕqj “ ∇j pig āb ϕbā q “ ig āb ∇j ϕbā .
and
pΛω Bϕqj “ ig āb pBϕqbāj ,
while
Bϕ “ B` ϕj k̄ dz ` ^ dz j ^ dz̄ k
1
“ p∇` ϕj k̄ ´ ∇j ϕ`k̄ qdz ` ^ dz j ^ dz̄ k
2
1
“ pBϕq`j k̄ dz ` ^ dz j ^ dz̄ k (2.17)
2
where B was switched for ∇ since Γkij “ Γkji for Kähler metrics, as is seen by the explicit formula
Γkij “ Bi gj p̄ g p̄k and the Kähler definition Bi gj p̄ “ Bj gip̄ . So

pΛω Bϕqj “ ig āb p´∇b ϕjā ` ∇j ϕbā q

Therefore
pBΛω ϕqj ´ pΛω Bϕqj “ ig āb ∇b ϕjā “ ipB̄ : ϕqj
which proves (2.16) for p1, 1q-forms.
Next, using the Kähler identity (2.16), we note
B B̄ : “ ´B̄ : B, (2.18)
since
iB B̄ : ` iB̄ : B “ BpBΛ ´ ΛBq ` pBΛ ´ ΛBqB “ 0.
Combining (2.16) and (2.18), we derive
∆B “ pBB : ` B : Bq
“ iBpB̄Λ ´ ΛB̄q ` ipB̄Λ ´ ΛB̄qB
“ ´iB̄pBΛ ´ ΛBq ´ ipBΛ ´ ΛBqB̄
“ B̄ B̄ : ` B̄ : B̄
“ ∆B̄ (2.19)

33
and

∆d “ pdd: ` d: dq
“ pB ` B̄qpB : ` B̄ : q ` pB : ` B̄ : qpB ` B̄q
“ ∆B ` ∆B̄ ` pB B̄ : ` B̄ : Bq ` pB̄B : ` B : B̄q
“ 2∆B̄ . (2.20)

To end this section, we state the B B̄-Lemma.

Theorem 2.8. (B B̄-Lemma) Let X be a compact Kähler manifold. Let α P Λp,q pXq. Then:
‚ If α “ dη for η P Λp`q´1 , then α “ B B̄β for β P Λp´1,q´1 .
‚ If α “ Bη for η P Λp´1,q and dα “ 0, then α “ B B̄β for β P Λp´1,q´1 .

Proof. We prove the first statement, and the second statement has a similar proof. Using the Hodge
decomposition (Theorem 2.3) for ∆B ,

α “ Bα1 ` B : α2 ` α1 , α1 P ker ∆B .

Using that α1 P ker ∆B “ ker ∆d , we have pα1 , α1 q “ pα1 , αq “ pd: α1 , ηq “ 0. Also 0 “ Bα “ BB : α2

and so pB : α, B : αq “ 0. Therefore
α “ Bα1 .
Next, we use the Hodge decomposition for ∆B̄ to write

α1 “ B̄β1 ` B̄ : β2 ` β 1 , β 1 P ker ∆B̄

Substituting gives
α “ B B̄β1 ` B B̄ : β2 .
To remove the last term, we must use the Kähler identity (2.18) which reads B B̄ : “ ´B̄ : B. Using
this and B̄α “ 0, we see that
0 “ B̄ B̄ : Bβ2 .
It follows that pB̄ : Bβ2 , B̄ : Bβ2 q “ 0.

2.2 Kodaira vanishing theorem

Let L Ñ pM, ωq be a holomorphic line bundle over a compact Kähler manifold. We say that L is
a positive line bundle if it admits a metric h such that its curvature satisfies Fj k̄ ě εgj k̄ for some
ε ą 0. Here the inequality is in the sense of positive-definite matrices, meaning

Fj k̄ v j v k ě εgj k̄ v j v k

for all v P Cn .

34
Example 2.9. The main example of a positive line bundle is Op1q Ñ Pn , as

|Zi |2
phF S qi “ ř 2
, over Ui “ tZi ‰ 0u
p |Zp |

defines a metric with curvature iF “ ´iB B̄ log hF S “ ωF S ą 0 as we computed in (1.26). A similar

computation shows that Opkq Ñ Pn is positive for any k ě 1. For this, equip Opkq with phF S qk so
that the curvature is iF “ kωF S .
Example 2.10. Let P pZ0 , . . . , Zn q be a homogeneous polynomial of degree k, and let Y “ tP “
0u Ă Pn . Then OpY q Ñ Pn is a positive line bundle by example 1.14.
Example 2.11. There is a notion of ample line bundle L Ñ M , which means that there are sections
si P H 0 pM, Lq such that ϕ : M Ñ PN with ϕpzq “ rs0 pzq, . . . , sN pzqs is an embedding. We can
cover X with trivializations Ui “ tsi ‰ 0u and equip L with metrics phi , Ui q with

|si |2
hi “ ϕ˚ hF S “ ř 2
.
k |sk |

Then ´iB B̄ log h “ ϕ˚ p´iB B̄ log hF S q ą 0. Therefore ample line bundles are positive. Kodaira’s
embedding theorem (e.g. [20]) states that positive line bundles are ample.
Theorem 2.12. Let L Ñ pM, ωq be a positive holomorphic line bundle over a compact Kähler
manifold. Then
H q pX, L b KX q “ 0
for all q ě 1.
We will show that dim H q pX, L b KX q “ dim ker ∆B̄ |Λ0,q pLbKX q “ 0. We will give the proof for
q “ 1 for simplicity. For the general calculation, see e.g. [20].
Let h be a metric on L, so that the inner product on sections u, v P ΓpLq is xu, vyh “ uvh. Let
ϕ P Λ0,1 pLq, which we write as
ϕ “ ϕk̄ dz̄ k ,
where ϕk̄ is a local section of L. The L2 inner product for u, v P ΓpLq is

ωn
ż
pu, vq “ puv̄hq
X n!

and for ϕ, ψ P Λ0,1 pLq is

ωn
ż
pϕ, ψq “ g īk pϕī ψk̄ hq .
X n!
:
We will start by computing B̄ ϕ P ΓpLq. The difference with (2.1), in addition to introducing the
line bundle L into the mix, is that we now use the assumption that g is Kähler. The formula in
this case is:
Lemma 2.13. Let pX, ωq be a compact Kähler manifold and ϕ P Λ0,q pLq.

pB̄ : ϕqK̄ “ ´g āb ∇b ϕāK̄ .

35
We will verify this for ϕ P Λ0,1 pLq. The definition of the adjoint is

pB̄ : ϕ, uq “ pϕ, B̄uq, ϕ P Λ0,1 pLq, u P ΓpLq.

We start with
xϕ, B̄uyg,h “ g īk pϕī Bk̄ uhq “ ϕk Bk uh
which implies
xϕ, B̄uyg,h pdet gq “ Bk pϕk updet gqhq ´ Bk phpdet gqϕk qu
The first term integrates to zero by Stokes’s theorem (see earlier notes for more justification on
this), and so wedging by dz 1 ^ ¨ ¨ ¨ ^ dz̄ n and integrating gives

ωn
ż
pϕ, B̄uq “ ´ Bk phpdet gqϕk qpdet gq´1 h´1 hu “ pB̄ : ϕ, uq
X n!

where
B̄ : ϕ “ ´h´1 pdet gq´1 Bk phpdet gqg kī ϕī q. (2.21)
On a Kähler manifold, this is in fact

B̄ : ϕ “ ´g k̄p ∇p ϕk̄ , (2.22)

where the Chern connection of ph, gq acts on ϕ P Λ0,1 pLq by

¯
∇p ϕk̄ “ Bp ϕk̄ ` ph´1 Bp hqϕk̄ , ∇p̄ ϕk̄ “ Bp̄ ϕk̄ ´ Γp̄k̄ ` ϕ`¯.

To see (2.22), expand (2.21)

B̄ : ϕ “ ´g kī Bk ϕī ´ h´1 Bk hg kī ϕī ´ pdet gq´1 Bk ppdet gqg kī qϕī

The last term is zero. Indeed,

Bk ppdet gqg īk q “ Bk pdet gqg īk ` pdet gqBk g īk

“ pdet gqg b̄a Bk gab̄ g īk ´ pdet gqg b̄k Bk gab̄ g īa
“ 0 (2.23)

by the Kähler condition Bk gab̄ “ Ba gkb̄ . This proves (2.22). With this formula for the adjoint, we
now compute the Laplacian.
Lemma 2.14. Let pL, hq Ñ pX, ωq be a holomorphic line bundle with metric over a compact Kähler
manifold. For any ϕ P Λ0,1 pLq, we have

p∆B̄ ϕqk̄ “ ´g āb ∇b ∇ā ϕk̄ ` ϕi Rik̄ ` ϕi Fik̄ .

Proof. We start with pB̄f qk̄ “ ∇k̄ f on functions, which implies

pB̄ B̄ : ϕqk̄ “ ∇k̄ p´g āb ∇b ϕā q “ ´g āb ∇k̄ ∇b ϕā ,

36
since ∇k̄ g āb “ 0. Next,

pB̄ : B̄ϕqk̄ “ ´g āb ∇b pB̄ϕqāk̄

“ ´g āb ∇b pBā ϕk̄ ´ Bk̄ ϕā q
“ ´g āb ∇b p∇ā ϕk̄ ´ ∇k̄ ϕā q. (2.24)

This is because Γij k “ Γji k for a Kähler manifold since Γij k “ Bi gj p̄ g p̄k . Therefore

p∆B̄ ϕqk̄ “ ´g āb ∇b ∇ā ϕk̄ ` r∇b , ∇k̄ sϕb

The commutator formula for covariant derivatives on ϕa P ΓpT 1,0 X b Lq implies

r∇b , ∇k̄ sϕb “ ϕa Ra b bk̄ ` ϕb Fbk̄ .

In Kähler geometry, one can see directly from Rj i mk̄ “ ´Bk̄ pBm gj p̄ g p̄i q the symmetry

Ra b j k̄ “ Rj b ak̄ .

Since Rak̄ “ Rb b ak̄ , we conclude the formula.

We now let ϕ P Λ0,1 pLbKX q. We can apply the previous formula with ϕ P Λ0,1 pL̃q, and L̃ “ LbKX
equipped with the product metric h̃ “ h b pdet gq´1 . Using the formula Rik̄ “ ´Bi Bk̄ log det g and
F̃j k̄ “ ´Bj Bk̄ log h, we see that the curvature is

F̃j k̄ “ Fj k̄ ´ Rj k̄

Therefore cancellation occurs and we have

p∆B̄ ϕqk̄ “ ´g āb ∇b ∇ā ϕk̄ ` ϕi Fik̄ .

Suppose ϕ P ker ∆. Then

ωn ωn
ż ż
0 “ p∆ϕ, ϕq “ ´ ∇a ∇a ϕk̄ ϕ̄k̄ h ` ϕi Fik̄ ϕ̄k̄ h
X n! X n!

Integrating the first term by parts and using positivity of Fik̄ ,

n
ωn
ż ż
a k̄ ω
0ě ∇ ϕk̄ ∇a ϕ̄ h `ε gik̄ ϕi ϕ̄k̄ h ,
X n! X n!

which is
0 ě p∇ϕ, ∇ϕqL2 ` εpϕ, ϕqL2 .
It follows that ϕ “ 0. This proves that ker ∆ “ t0u. Therefore dim H 1 pX, L b KX q “ 0.
Here we used integration by parts on a Kähler manifold, which follows from the divergence theo-
rem ż
∇a V a ω n “ 0, V P ΓpT 1,0 Xq. (2.25)
X

37
In the above computation, this was used with V a “ ∇a ϕk̄ ϕ̄k̄ h P ΓpT 1,0 Xq and

ωn n
ωn
ż ż ż
a k̄ a k̄ ω
0“ ∇a p∇ ϕk̄ ϕ̄ hq “ ∇a ∇ ϕk̄ ϕ̄ h ` ∇a ϕk̄ ∇a pϕ̄k̄ hq
X n! X n! X n!

and we also then used metric compatibility ∇a h “ 0. The divergence theorem (2.25) comes
from
ωn
ż
dιV “ 0,
X n!
and the integrand is
ωn
Bi pV i det gqpidz 1 dz̄ 1 q . . . pidz n dz̄ n q “ pdet gq´1 Bi pV i det gq .
n!
We compute

pdet gq´1 Bi pV i det gq “ Bi V i ` pdet gq´1 Bi pdet gqV i “ Bi V i ` g āb Bi gāb V i .

Therefore ż
pBi V i ` Γib b V i q ω n “ 0.
X
On the other hand
∇i V i “ Bi V i ` Γib i V b .
Since Γib “ Γbi , these are equal.
This vanishing theorem can be generalized, and more generally there holds (see e.g. [20]
Theorem 2.15. Let L Ñ pX, ωq be a positive holomorphic line bundle over a compact Kähler
manifold. Then
H q pX, Ωp b Lq “ 0
for all integers p, q with p ` q ą n.

2.3 Sheaves and the Lefschetz hyperplane theorem

We will use the vanishing theorem (Theorem 2.15) to prove the Lefschetz hyperplane theorem.
First, we state some results from the theory of sheaves. For a nonempty open set U Ď X, let
OpU q denote holomorphic functions on U . Note that the only holomorphic functions defined on
the entirety of a compact manifold are constant functions. But for small open sets U Ď X there
are many holomorphic functions in OpU q.
Here we prove: if f : X Ñ C is holomorphic on a compact complex manifold X, then f is constant.
Let M “ supX |f | be attained a point p P X. After possibly replacing f by eiθ f for a constant angle
eiθ , we may assume that f ppq “ M . Consider Re f “ 21 pf ` f¯q, so that Re f ď M and Re f ppq “ M .
Let S “ tx P X : Re f pxq “ M u. Then S is non-empty and closed. It is also in fact open: if x P S,
in a local coordinate ball B1 p0q centered at x then Re f is a harmonic function:
1
Bz1 Bz̄1 Re f “ Bz1 pBz1 f q “ 0,
2

38
 ř
so that pBxi Bxi ` Byi Byi qRe f “ 0. By the maximum principle for harmonic functions on B1 p0q Ď
R2n , since pRe f qp0q “ M then Re f ” M in all of B1 p0q. Hence S is open, and S “ X. Since
Re f “ M and |f | ď M , then Im f “ 0 and f is a constant.
We will define sheaves of OpU q modules (there is also a notion of sheaves of groups, vector spaces,
etc).
Definition 2.16. A presheaf (of OX modules) F on a complex manifold X is defined by the
following information. For any non-empty open set U Ď X, we associate a nonempty OpU q-module
FpU q, and a collection of restriction maps ρU,V : FpV q Ñ FpU q defined when U Ď V satisfying

ρU,V ˝ ρV,W “ ρU,W , ρU,U “ idU , (2.26)

for U Ď V Ď W . The set FpU q is called the set of sections of F over U . We also use the notation
s|U “ ρU V psq for s P FpV q.
Definition
Ť 2.17. A sheaf F on X is a presheaf satisfying the following glueing property. Suppose
Ω “ Uµ are open sets in X. If sµ P FpUµ q are such that

ρUµ XUν ,Uµ psµ q “ ρUµ XUν ,Uν psν q (2.27)

then there exists s P FpΩq such that ρUµ ,Ω psq “ sµ . Also, if s, t P FpΩq and ρUµ ,Ω psq “ ρUµ ,Ω ptq
for all µ, then s “ t.
In other words, local sections of a sheaf can be uniquely glued together.
Example 2.18. We write OX for the sheaf of holomorphic functions on a complex manifold X.
Example 2.19. Let E Ñ M be a holomorphic bundle. We write E for the sheaf of holomorphic
sections: EpU q are holomorphic sections over U .
Example 2.20. Sheaf I0 described by holomorphic functions in C2 vanishing at the origin. If U
does not contain the origin, then this is generated by 1 so I0 pU q – OpU q. In a neighbourhood V
of the origin, this is generated by x and y: any local holomorphic function f with f p0q “ 0 can be
written as f pxq “ gpx, yqx ` hpx, yqy. Thus I0 pV q is a module of rank 2. Thus the rank jumps up
to 2. Also, at the origin, the module is not free. For example, we have the relation ´y ¨ x ` x ¨ y “ 0.
In this sense, sheaves are sometimes viewed as a generalization of vector bundles where the rank
may jump.
Example 2.21. Another example to note are the constant sheaves Z, R, C. These are sheaves of
groups, meaning that FpU q attaches a group for every open set U (rather than a module). So for
example, ZpU q are locally constant Z-valued functions on U .

Ů Stalk of a sheaf. Let x P X. The stalk Fx is the set of equivalence classes in the
Definition 2.22.
disjoint union xPU FpU q with s1 P FpU1 q and s2 P F pU2 q satisfying s1 „ s2 if s1 |V “ s2 |V for
some V Ă U1 X U2 . The stalk Fx is an Ox,X -module.
A map of sheaves ϕ : F Ñ E is a collection of homomorphisms ϕU : FpU q Ñ EpU q such that ϕU ,
ϕV commute with the restriction maps. We say

0ÑE ÑF ÑGÑ0

39
is an exact sequence of sheaves if, denoting the arrows by fi , we have that all fi are maps of sheaves
with fi`1 ˝ fi “ 0 and the associated complex of stalks

0 Ñ Ex Ñ Fx Ñ Gx Ñ 0

is exact for all x P X. Recall that exact means that the kernel of one arrow is the image of the
previous arrow.
We will use the following two results from Cech cohomology [13]. Rather than define the Cech
cohomology groups Ȟ q pX, Eq, we will just directly use the following two facts:
‚ Given an exact sequence of sheaves 0 Ñ E Ñ F Ñ G Ñ 0, there exists a long exact sequence in
cohomology
¨ ¨ ¨ Ñ Ȟ p pX, Eq Ñ Ȟ p pX, Fq Ñ Ȟ p pX, Gq Ñ Ȟ p`1 pX, Eq Ñ ¨ ¨ ¨ .

‚ Dolbeault theorem:
kerpB̄ : Λp,q pEq Ñ Λp,q`1 pEqq
Ȟ q pX, Ωp b Eq – .
im pB̄ : Λp,q´1 pEq Ñ Λp,q pEqq

We will write H q pX, Ωp b Eq as before instead of Ȟ q pX, Ωp b Eq.

Theorem 2.23. (Lefschetz hyperplane) Let Y Ď X be a smooth analytic hypersurface of a compact
Kähler manifold X such that the line bundle OpY q is positive. Then the restriction map

H q pX, ΩpX q Ñ H q pY, ΩpY q

is an isomorphism when p ` q ď n ´ 2. Thus we have equality of Hodge numbers:

hp,q pY q “ hp,q pXq, p ` q ď n ´ 2.

As a consequence of the Hodge decomposition, we obtain that

H q pX, Cq Ñ H q pY, Cq

is an isomorphism for q ď n ´ 2.
Example 2.24. Let Y “ tP “ 0u Ă Pn be a smooth complex manifold cut out by a homogeneous
polynomial P of degree m ě 1. We showed earlier that OpY q “ Opmq, which is positive. Therefore
H k pY, Cq is isomorphic to H k pPn q for k ď n ´ 2.

Proof. First, we note the exact sequence sheaves

0 Ñ Op´Y q Ñ OX Ñ OY Ñ 0.

Here Op´Y q is the dual bundle of OpY q. This means that sections s P ΓpOp´Y qq transform as
sU “ tU V sV with tU V “ fV {fU , where Y has defining function fU “ 0 over an open set U . The
transformation relation shows that the combination sU fU is a well-defined function on the manifold.
To explain the exact sequence, over an open set U where Y has defining function fU “ 0, a local
holomorphic section sU P Op´Y qU gets sent to sU fU P OU which is a holomorphic function on U
vanishing along Y .

40
Tensoring with bΩpX implies the exact sequence

0 Ñ ΩpX p´Y q Ñ ΩpX Ñ ΩpX |Y Ñ 0.

The corresponding long exact sequence in cohomology gives

H q pX, ΩpX p´Y qq Ñ H q pX, ΩpX q Ñ H q pY, ΩpX |Y q Ñ H q`1 pX, ΩpX p´Y qq

We note that

dim H q pX, ΩpX p´Y qq “ dim H n´q pX, pΩp q˚ b OpY q b KX q

“ dim H n´q pΩn´p pOpY qq
“ 0 (2.28)

by Serre duality and the vanishing theorem (Theorem 2.15) when pn ´ qq ` pn ´ pq ą n. Also
considering this with q replaced by q ` 1, we see that when p ` q ă n ´ 1 we have

0 Ñ H q pX, ΩpX q Ñ H q pY, ΩpX |Y q Ñ 0 (2.29)

and so H q pX, ΩpX q

Ñ H q pY, ΩpX |Y q is an isomorphism.
Next, we will show that H q pY, ΩpX |Y q Ñ H q pY, ΩpY q is an isomorphism. For this we use the exact
sequence of vector spaces

0 Ñ pN |Y q˚y b pΛYp´1,0 qy Ñ pΛp,0 p,0

X |Y qy Ñ pΛY qy Ñ 0. (2.30)

Here y P Y and coordinates are chosen over an open set U Ă X such that U X tz n “ 0u “ U X Y ,
and

pN |Y q˚y “ span tdz n u

pΛ1,0
X |Y qy “ span tdz 1 , . . . , dz n´1 , dz n u
pΛ1,0
Y qy “ span tdz 1 , . . . , dz n´1 u (2.31)

and we may multiply the generators by local holomorphic functions on Y . The sequence (2.30)
implies
0 Ñ Op´Y q|Y b Ωp´1
Y Ñ ΩpX |Y Ñ ΩpY Ñ 0,
as sheaves over Y . This is because
pN |Y q˚ “ Op´Y q
which can be seen as follows: if there are two sets of coordinates z, z̃ where both z n “ 0 and z̃ n “ 0
locally cut out Y , then z̃ n pzq “ z n f pzq, where f pzq is the transition function for OpY q. Note that
f pzq is non-vanishing; this is because Y “ tz̃ n pzq “ 0u smooth means that Bzn z̃ n pyq ‰ 0. Next,
n
dz̃ n “ Bz̃ i n n
Bz i dz implies dz̃ |Y “ f pzqdz |Y . This is the transformation law for the local frame dz ,
n
˚
so components of pN |Y q transform by the inverse 1{f which is why the dual Op´Y q appears.
Thus
H q pY, Ωp´1 q p q p
Y p´Y qq Ñ H pY, ΩX |Y q Ñ H pY, ΩY q Ñ H
q`1
pY, Ωp´1
Y p´Y qq.

We apply Serre duality and the vanishing theorem as before to obtain

dim H q`1 pY, Ωp´1

Y p´Y qq “ dim H
n´q´1
pY, ΩYn´p`1 b OpY qq “ 0.

41
Here we used that OpY q|Y Ñ Y is positive if OpY q Ñ X is positive. The dimension of Y is n ´ 1,
so the vanishing theorem applies if

pn ´ q ´ 1q ` pn ´ p ` 1q ą pn ´ 1q

which holds for p ` q ď n ´ 2. Therefore

0 Ñ H q pY, ΩpX |Y q Ñ H q pY, ΩpY q Ñ 0

so H q pY, ΩpX |Y q Ñ H q pY, ΩpY q is an isomorphism, which combined with (2.29) gives that

H q pX, ΩpX q Ñ H q pY, ΩpY q

is an isomorphism.

42
3 Deformations of Complex Manifolds
3.1 Families of complex manifolds
From a complex manifold given by
X “ tP “ 0u Ă Pn ,
we can consider a 1-parameter family Xt by inserting a parameter t P C in front of one of the
polynomial coefficients. A famous example [3] is
"ÿ 4
ź *
Xt “ Zk5 ´ t Zk “ 0 Ă P4 .
k“0 k“0

The total space including the parameter t is

X “ tpp, tq P Xt ˆ Cu.

We can also consider families where t P Cr , and we use the notation ∆ Ă Cr for a ball of radius 1.
The formal definition of a family of complex manifolds is:
Definition 3.1. Let X0 be a compact complex manifold. A family of deformations of X0 over ∆ Ă
Cr is given by π : X Ñ ∆ where X is a complex manifold, π is a holomorphic map, π ´1 p0q “ X0 ,
and the Jacobian of π has maximal rank.

Ť π : X Ñ ∆ and the maximal rank theorem for holomorphic submersions, we

Using the definition
may cover X “ i Ui so that local coordinates on Ui are of the form pz 1 , . . . , z n , tq, where z i are
holomorphic coordinates on Ui X Xt , and

πpz 1 , . . . , z n , tq “ t.

Change of coordinates on an overlap pz, tq, pz̃, t̃q are of the form

z̃ k “ f k pz 1 , . . . , z n , tq, t̃ “ t. (3.1)

We will now show that in a family of complex manifolds, all underlying smooth manifolds are
diffeomorphic.

43
Lemma 3.2. Let π : X Ñ ∆ be a family of compact complex manifolds.
Let t1 P ∆. From any path γ : r0, 1s Ñ ∆ with γp0q “ 0 and γp1q “ t1 , we can construct a
1-parameter family of diffeomorphisms Θs : X0 Ñ Xγpsq such that Θ0 “ id.
In particular Xt1 is diffeomorphic to X0 for all t1 P ∆.

Proof. Let γpsq be a path on ∆ from 0 to t0 . Extend the vector field γ9 arbitrarily to all of ∆. Let
tbi u be real coordinates on ∆, and write

B
γpbq
9 “ γ9 i pbq .
Bbi
We will lift up the vector field on the base γ9 P ΓpT B, Bq using a partition of unity. Cover X with
finitely many coordinates charts
ř pz 1 , . . . , z n , b1 , . . . , bn q as described earlier where πpz, bq “ b. Let
ρα be a partition of unity ( ρα “ 1, supppρα q Ă Uα ) subordinate to this cover. Define on X the
vector field
ÿ „ 
i B
V “ ρα γ9 .
α
BbiUα
It may be clearer to write V using fixed coordinates on say U0 Ă X , in which case

B ÿ Bz k B
V “ γ9 i i
` ρα γ9 i iU0 k .
BbU0 α
Bb Uα BzU0

44
 9 Next, we solve the ODE system on X
So we get a lifted vector field V P ΓpT X q with π˚ V “ γ.

d
Θε “ V, Θ0 “ id,
dε
which is well-known to produce a 1-parameter family of diffeomorphisms Θε : X Ñ X together with
d
inverses Θ´ε : X Ñ X satisfying dε Θ´ε “ ´V and Θε ˝ Θ´ε “ id.
The last thing to check is that this construction produces diffeomorphisms from fiber to fiber
Θε : X0 Ñ Xγpεq . Fix x P X0 and consider the function f pεq “ π ˝ Θε pxq ´ γpεq. Then

f p0q “ 0, f 1 pεq “ π˚ V ´ γ9 “ 0.

Therefore f pεq ” 0 and so Θε pxq P Xγpεq .

A similar argument shows Θ´ε : Xγpεq Ñ X0 : consider x P Xγpεq , f prq “ π ˝ Θ´r pxq ´ γpε ´ rq, and
prove f pεq “ 0. Since Θε ˝ Θ´ε “ id, we have that Θε : X0 Ñ Xγpεq is a family of diffeomorphisms.

Looking relative to the moving family of diffeomorphisms, we may regard Xt as the fixed differen-
tiable manifold X0 , and let the complex structure tensor Jt vary in t. More precisely: let X Ñ ∆
be a family of complex manifolds, so that pXt , Jˇt q is a complex manifold for each parameter t P ∆.
Take a path γ : p´ε, εq Ñ ∆ with γp0q “ 0, γp0q
9 ‰ 0, and so from Lemma 3.2 we obtain a family
of diffeomorphisms Θt : X0 Ñ Xγptq . Then we may consider

pX0 , Jt q, Jt “ pΘt q´1 ˇ

˚ Jγptq pΘt q˚

which is a 1-parameter family of complex structures on a fixed differentiable manifold X0 . This

defines new complex structures because Jt2 “ ´id and N pJt q “ 0 (as because N transforms as a
tensor under coordinate transformation). We now define the Nijenhuis tensor N .
On a differentiable manifold X, an almost-complex structure is a tensor J P ΓpEndTR Xq satisfying
J 2 “ ´id. The Newlander-Nirenberg theorem (see e.g. [7] for a proof) states that an almost
complex structure J comes from holomorphic coordinates tz α u with J BzBα “ i BzBα , J Bz̄Bα “ ´i Bz̄Bα
if and only if the Nijenhuis tensor
ˆ ˙
k 1 r k k r
N ij “ J i Br J j ` J r Bj J i ´ pi Ø jq . (3.2)
4

vanishes identically: N k ij “ 0. It can also be checked that the components of N k ij transform

correctly so that N is a legitimate tensor.
Remark 3.3. Here is some motivation regarding N . From an almost complex-structure J on a
smooth manifold X, we may split TC X “ T 1,0 X ‘ T 0,1 X, where T 1,0 X is the `i eigenspace of J
and T 0,1 X is the ´i eigenspace of J. The Nijenhuis tensor N “ 12 N p mn dxm ^ dxn b Bp can be
shown to satisfy N pU, V q “ ´rU, V s0,1 for U, V P ΓpT 1,0 Xq. In other words, N “ 0 if and only if
rU, V s P T 1,0 X for all U, V P T 1,0 X. Hence N measures the failure of the subbundle T 1,0 X Ă TC X
being closed under the Lie bracket.

45
Let X Ñ ∆ be a family of complex manifolds with central fiber pX, Jq, together with a path
γ : p´ε, εq Ñ ∆ with γp0q “ 0 and γp0q
9 ‰ 0. From this data, we would like to produce an
element
rηs P H 1 pX, T 1,0 Xq.
Given our discussion so far, from this information we can create a path of complex structures
Jt P ΓpEndT Xq on the fixed differentiable manifold X with J0 “ J satisfying the constraints
Jt2 “ ´id and N pJt q “ 0. We will now show:
9
‚ Differentiation η “ Jp0q produces an element η P Λ0,1 pT 1,0 Xq which satisfies B̄η “ 0 and hence
defines a Dolbeault cohomology class rηs P H 1 pX, T 1,0 Xq.
‚ Different choices of diffeomorphisms Θt : X Ñ Xγptq in Lemma 3.2 produce the same class
rηs P H 1 pX, T 1,0 Xq.
d
We start by differentiating Jt2 “ ´id. Take dt |t“0 and obtain

J9i k J k j ` J i k J9k j “ 0. (3.3)

d
Here J9 “ dt |t“0 Jt . We will work in coordinates x “ pz 1 , . . . , z n , z̄ 1 , . . . , z̄ n q on the fixed complex
manifold pX, Jq where tz α u are holomorphic coordinates on pX, Jq. Let α, β denote holomorphic
coordinates, so that α P t1, . . . , nu and Bα “ BzBα , Bᾱ “ Bz̄Bα . Let i, j, p denote indices for the real
coordinates x, so that i P t1, . . . , n, 1̄, . . . , n̄u and so we could have i “ α or i “ ᾱ. Summations
over i, j, k run over both unbarred coordinates z α and barred coordinates z̄ α .
With this convention, then J α β “ iδ α β , J ᾱ β̄ “ ´iδ α β and J α β̄ “ J ᾱ β “ 0. Then (3.3) with i “ α,
j “ β implies
J9α β “ 0.
Similarly J9ᾱ β̄ “ 0. Since Jptq is real, so is J,
9 and hence J9 is determined by the components

η α β̄ :“ J9α β̄ .
d
Next, we differentiate N pJt q “ 0. Taking dt |t“0 of (3.2) gives

0 “ J9r β̄ Br J α γ̄ ` J r β̄ Br J9α γ̄ ` J9α r Bγ̄ J r β̄ ` J α r Bγ̄ J9r β̄ ´ pβ̄ Ø γ̄q

which becomes „ 
9 α 9 α
0 “ 2 ´ iBβ̄ J γ̄ ` iBγ̄ J β̄ .

Since η α β̄ P ΓpT 1,0 X b Λ0,1 q, this implies

B̄η “ 0, rηs P H 1 pX, T 1,0 Xq

and we call rηs the Kodaira-Spencer class.

We now want to show that rηs P H 1 pX, T 1,0 Xq is independent of the choice of family of diffeomor-
phisms in Lemma 3.2. Suppose from the path γptq on ∆, we produce two families of diffeomor-
phisms
Θt : X Ñ Xγptq , Ψt : X Ñ Xγptq , Θ0 “ Ψ0 “ id.

46
From the complex manifold pXt , Jˇt q, we produce a family of complex structures on the fixed X0 as
before by Jt “ pΘt q´1 ˇ ˜ ´1 ˇ
˚ Jγptq pΘt q˚ and Jt “ pΨt q˚ Jγptq pΨt q˚ . These are related by

J˜t “ pft q˚ Jt pft q´1

˚

d
with ft “ Ψ´1
t ˝ Θt . Then ft be a 1-parameter family of diffeomorphisms with dt |t“0 ft “ V (for
some vector field V ) and f0 “ id.
We now compute dt d
|t“0 of J˜t . Let y α “ ftα pxq be a change of coordinates by the diffeomorphism.
The formulas for the pushfoward f˚ : Tp M Ñ Tf ppq M are

By a Bxi
pf˚ V qa pf ppqq “ ppqV i ppq, pf˚´1 V qa ppq “ pf ppqqV a pf ppqq,
Bxi By a
and so acting Jt on V P Tf ppq M gives

By a k Bx
i
rJ˜t a b V b spf ppqq “ ppqJt ppq i pf ppqqV a pf ppqq
Bxk By a
The components of Jt are then

By a Bxi
J˜ta b pyt pxqq “ tk pxqJt pxqk i b pyt pxqq.
Bx Byt

Differentiating this in time at t “ 0 along a path with y0 pxq “ x and yp0q 9 “ V gives
ˇ
d ˇ Bxi
J9̃a b ` Bk J a b V k “ Bk V a J k b ` J9a b ` J a i ˇˇ pyt pxqq
dt t“0 By b

In complex coordinates, the second term on the left is zero since the components J a b are constant.
For the last term, we differentiate in time the chain rule identity

Bxi By k B
k
pypxqq j
pxq “ j xi “ δ i j
By Bx Bx
to obtain ˇ
d ˇˇ Bxi
pyt pxqq “ ´Bj V i .
dt ˇt“0 By j
Therefore
J9̃a b “ J9a b ` Bk V a J k b ´ J a i Bb V i .
We showed earlier that considering J 2 “ ´id implies that the non-zero contributions are η α β̄ “
9 α β̄ and its conjugate. So we let a “ α, b “ β̄ and obtain
Jp0q

η̃ α β̄ “ η α β̄ ´ 2iBβ̄ V α .

It follows that
η̃ “ η ´ 2iB̄V 1,0
and so rηs “ rη̃s P H 1 pT 1,0 Xq.

47
Remark 3.4. Let pX, Jq be a complex manifold. This calculation shows that deformations
J˜t “ pft q˚ Jpft q´1
˚ created by 1-parameter families of diffeomorphisms ft produce the zero class
rηs “ 0 P H 1 pX, T Xq (since J9 “ 0). Deformations of complex structure coming from families of
diffeomorphisms are not counted by rηs P H 1 pX, T Xq.
A central question in deformation theory is the inverse problem: given η P H 1 pT 1,0 Xq, does it
9
come from a family pXptq, Jptqq of complex manifolds with Jp0q “ η? This statement is not true
in general, but it is true for Kähler Calabi-Yau manifolds (this is the Bogomolov-Tian-Todorov
theorem). For more references on this topic, see for example [20, 16, 14].

3.2 Semi-continuity theorem

We start with an illustrative example from linear algebra. Let At : Rn Ñ Rn be a family of
symmetric real n ˆ n matrices with entries continuously varying in t P R. Then there exists ε ą 0
such that for all |t| ă ε, then
dim ker At ď dim ker A0 .
The dimension of ker At may jump at t “ 0, as seen from e.g.
» fi
t 0
At “ – fl .
0 t

Returning to complex geometry, let pX, Jt q be a family of complex structures on a compact manifold
X with J0 “ J. A differential form α P Λk pXq has different decompositions into pp, qq types for
each parameter t. Write Λ1,0 0 for p1, 0q forms with respect to the initial structure J, and Λt1,0
for p1, 0q forms with respect to Jt . We can decompose α P Λ1 pXq into pp, qq components via the
formula
1 1
α “ pα ´ iJt αq ` pα ` iJt αq
2 2
:“ pαq1,0
t ` pαqt
0,1
(3.4)

Here we define JαpXq “ αpJXq. With this notation, we have that α P Λ1,0
0 if and only if Jα “ `iα.
The map ϕt : Λ1,0
0 Ñ Λ 1,0
t given by
α ÞÑ pαq1,0
t

is an isomorphism for small t, since ϕ0 “ id and ϕt varies continuously as seen from the explicit
expression above. Similarly, we obtain isomorphisms

ϕt : Λp,q p,q
0 Ñ Λt

for small t. Recall that

hp,q pXt q “ dim kert∆B̄t : Λp,q
t Ñ Λp,q
t u.

We will show that hp,q pXt q is an upper-semicontinuous function, meaning

hp,q pXt q ď hp,q pX0 q, |t| ă ε.

To view all operators on the same space, instead of ∆B̄ we will use

Lt : Λp,q p,q
0 Ñ Λ0 , Lt “ ϕ´1
t ˝ ∆B̄t ˝ ϕt ,

48
and show
dim ker Lt ď dim ker L0 , |t| ă ε.
To prove this, we will need some PDE estimates.
Let EŤÑ X be a vector bundle over a compact manifold. Cover X by finitely many trivializations
N
X “ i“1 Ui where Bi Ă Ui Ă Cn are balls of radius 1 still covering X. Let ψ P ΓpX, Eq and let ψUi
denote the vector valued function of the components of ψ in the trivialization Ui . Let 0 ă α ă 1.
Define
}ψ}C k,α :“ sup }ψUi }C k,α pBi q
i

where for a function f : B Ñ Rp the Hölder norm is

|DI f pxq ´ DI f pyq|

}f }C k,α pBq “ }f }C k pBq ` sup sup ,
|I|“k x‰y |x ´ y|α

where }f }C k pBq “ sup|I|“k supB |DI f |. We write ψ P C k,α pX, Eq if ψ is a k-times differentiable
section of E with finite } ¨ }C k,α norm. The main features of C k,α spaces for our purposes are:
‚ C k,α pX, Eq is a Banach space.
‚ Compactness: suppose tψn u P C k,α pX, Eq is a sequence of sections such that

}ψn }C k,α ď C

for uniform constant C ą 0. Let 0 ă α1 ă α. Then there exists a limiting section ψ8 P C k,α and a
1
subsequence tψnk u such that ψnk Ñ ψ8 in C k,α .
‚ The Schauder estimates. (Theorem 3.5 below)
Let us prove that C 0,α pX, Eq :“ C α is a Banach space and its compactness property. It is routine
to check that } ¨ }C α is a norm. To show completeness, we must show that if tψn u is a Cauchy
sequence, then ψn converges to ψ8 P C α . By Arzela-Ascoli applied on each coordinate ball Bi , a
subsequence ψnk converges in C 0 to a continuous limit ψ8 , and since tψn u is Cauchy then the full
sequence converges
ψn Ñ ψ8 in C 0 .
The limit section ψ8 is in C α , since for x, y P Bi with |x ´ y| “ δ, then
|ψ8 pxq ´ ψ8 pyq| |ψ8 pxq ´ ψk pxq| |ψk pxq ´ ψk pyq| |ψk pyq ´ ψ8 pyq|
ď ` `
|x ´ y|α |x ´ y|α |x ´ y|α |x ´ y|α
ď C (3.5)

for k large enough such that }ψ8 ´ ψk }C 0 ă δ. To show the sequence converges in C α , let ε ą 0.
There exists N such that if k, ` ě N then for all x ‰ y there holds
|rψk ´ ψ` spxq ´ rψk ´ ψ` spyq|
ď ε.
|x ´ y|α
Fix x, y and send ` Ñ 8 to conclude

}ψk ´ ψ8 }C α ă ε.

49
We now show compactness. Suppose tψn u P C α is a sequence with }ψn }C α ď C. By the Arzela-
Ascoli theorem applied to coordinate balls Bi , we obtain a subsequence ψnk converging in C 0 to a
continuous limit ψ8 . The limiting section is Hölder continuous ψ8 P C α by estimate (3.5). To see
convergence in 0 ă α1 ă α, we let vk “ ψnk ´ ψ8 and write
ˆ ˙α1 {α
|vk pxq ´ vk pyq| |vk pxq ´ vk pyq| 1
“ |vk pxq ´ vk pyq|pα{α q´1 .
|x ´ y|α1 |x ´ y|α

This goes to zero as k Ñ 8.

Next, we state the Schauder estimates.
Theorem 3.5. (Schauder estimates I) Let E, F Ñ X be vector bundles over a compact manifold.
Let L : ΓpX, Eq Ñ ΓpX, F q be an elliptic operator of order k. There exists C ą 0 such that for all
sections s P C k,α pX, Eq, then

}s}C k,α pXq ď Cp}s}L8 ` }Ls}C 0,α q.

Here C only depends on the constant of ellipticity and the C α norms of the coefficients of L.
For a reference, see [19], remark after Theorem 4.3 in the appendix.
We can upgrade this estimate for sections s P pker LqK .
Theorem 3.6. (Schauder estimates II) Let E, F Ñ X be complex vector bundles over a compact
complex manifold. Let L : ΓpX, Eq Ñ ΓpX, F q be an elliptic operator of order k. Let H be a metric
on E and ω a hermitian metric on X. There exists C ą 0 such that for all s P C k,α pX, Eq with

s P pker LqK

then
}s}C k,α pXq ď C}Ls}C 0,α .
Here s P pker LqK means: ps, ϕqL2 “ş 0 for all ϕ P ΓpX, Eq with Lϕ “ 0, and the L2 inner product
on ΓpX, Eq is as before: ps, ϕqL2 “ X xs, ϕyH ω n .

Proof. Suppose this estimate is false. Then there exists a sequence tsi u P pker LqK X C k,α such that

}sn }C k,α ě Mn }Lsn }C 0,α , Mn Ñ 8.

Let un “ sn {}sn }C k,α . Then tun u P pker LqK satisfies

1
}Lun }C 0,α pXq ď , }un }C k,α “ 1.
Mn
From }u}C k,α “ 1, compactness of Hölder space allows us to extract a subsequence un Ñ u8
1
converging in the C k,α norm with 0 ă α1 ă α. Since }Lun }C 0,α1 Ñ 0, the limit satisfies Lu8 “ 0.
Next, un P pker LqK and u8 P ker L implies pun , u8 qL2 “ 0 for all n, and we conclude

u8 “ 0.

50
We want to obtain a contradiction, but we cannot take a limit of }un }C k,α “ 1 since α ą α1 .
However, by the usual Schauder estimates, for n " 1 large enough we have
1
1 “ }un }C k,α ď Cp}un }L8 ` }Lun }C 0,α q ď C}un }L8 `
2
and so }un }L8 ě 1{2C for all n large. Taking a limit gives }u8 }L8 ą 0 which is a contradiction.

From the Schauder estimates, we can deduce the semi-continuity theorem.

Theorem 3.7. Let E, F Ñ X be complex vector bundles over a compact manifold. Let Lt :
ΓpX, Eq Ñ ΓpX, F q be a continuous family of elliptic operators of order k. Then there exists ε ą 0
so that
dim ker Lt ď dim ker L0 , |t| ă ε
so that dim ker Lt is upper semi-continuous.

Proof. Let u P ΓpX, Eq satisfy u P pker L0 qK . Then by the Schauder estimate,

}u}C k,α ď C}L0 u}C α

ď C}Lt u}C α ` C}pLt ´ L0 qu}C α
ď C}Lt u}C α ` Cε}u}C k,α (3.6)
1
if }Lt ´ L0 }C α ď ε. For ε ą 0 small enough, then Cε ă 2 and we obtain

}u}C k,α ď 2C}Lt u}C α .

It follows that
pker L0 qK X pker Lt q “ t0u.
From
pker Lt q Ă ΓpX, Eq “ pker L0 q ‘ pker L0 qK ,
we have ker Lt Ď ker L0 .
p,q p,q
To conclude semi-continuity of Hodge numbers, we let Lt “ ϕ´1 t ˝ ∆B̄ ˝ ϕt : Λ0 Ñ Λ0 as before.
Then
hp,q pXt q ď hp,q pX0 q, |t| ă ε
and we see that t ÞÑ dim H q pXt , Ωp q is upper semi-continuous in a family of complex mani-
folds.

51
3.3 Stability of Kähler metrics
Kodaira-Spencer’s stability theorem states that if X0 is a compact Kähler manifold, then any
deformation Xt admits a Kähler metric for small enough t [21].
After this section, we will be restricting our attention from general complex manifolds to simply
connected Calabi-Yau threefolds. We will in this section give the proof of Kodaira-Spencer’s stability
theorem assuming that X0 is a simply connected Calabi-Yau threefold. Explicitly, the only extra
hypothesis that we will use is H 0,2 pX0 q “ 0, so the proof here applies to any complex manifold
X0 satisfying this property. Our proof under this hypothesis is simpler compared compared to
Kodaira-Spencer [21], and we will follow the argument and exposition in Fu-Li-Yau [11]. The setup
in Fu-Li-Yau [11] is different as it concerns balanced metrics rather than Kähler metrics and the
central fiber X0 has nodal singularities, but the outline of the argument is readily adapted.
Let ω0 be the Kähler metric on X0 , and Θt be a family of diffeomorphisms Θt : X0 Ñ Xt .
Then
Θ˚t ω0 “ pΘ˚t ω0 q2,0 ` pΘ˚t ω0 q1,1 ` pΘ˚t ω0 q0,2
is a closed 2-form on Xt . To obtain a Kähler metric on Xt , it must be of type p1, 1q, so we let

χt “ pΘ˚t ω0 q1,1 .

But this is no longer closed.

0 “ dpΘ˚t ω0 q
„  „  „  „ 
“ BpΘ˚t ω0 q2,0 ` B̄pΘ˚t ω0 q2,0 ` BpΘ˚t ω0 q1,1 ` B̄pΘ˚t ω0 q1,1 ` BpΘ˚t ω0 q0,2 ` B̄pΘ˚t ω0 q0,2

By type considerations, we have

B̄χt “ ´BpΘ˚t ω0 q0,2 , B̄pΘ˚t ω0 q0,2 “ 0. (3.7)

So B̄χt Ñ 0 uniformly in all C k norms. The claim is that we can correct χt by

„  „ 
: : : :
ωγ “ χt ` B B̄ B γ ` B̄ B B̄ γ̄ (3.8)

where γ P Λ1,2 solves

B̄B B̄ : B : γ “ ´B̄χt . (3.9)
1,1
This will produce a real p1, 1q form ωγ P Λ pX, Rq satisfying

B̄ωγ “ B̄χt ` B̄B B̄ : B : γ “ 0.

Taking the conjugate, we conclude dωγ “ 0. To show that ωγ is a Kähler metric, we will show its
positivity ωγ ą 0. For this, we will show that the correction γ is small.
Roughly speaking, the strategy is to correct ω̃t “ χt ` Bα ` B̄ ᾱ so that α solves B̄Bα “ ´B̄ ω̃t and
so B̄ ω̃t “ 0. If B̄B were an elliptic operator, one could try to use elliptic PDE theory to estimate
}α} ď C}B B̄α} “ C}B̄χt } and show smallness of α since B̄χt Ñ 0 as t Ñ 0. To make this strategy
work, we do not use α but rather γ with α “ B̄ : B : γ. Furthermore (3.9) is not quite an elliptic PDE
for γ, but this equation can be modified to make this strategy work.

52
To make (3.9) elliptic, we consider the Kodaira-Spencer operator

Et “ B B̄ B̄ : B : ` B̄ : B : B B̄ ` B̄ : BB : B̄ ` B : B̄ B̄ : B ` B̄ : B̄ ` B : B

and solve
Et pγt q “ B̄χt .
In the definition of Et , the B, B̄ are with respect to the complex structure at t, and the adjoints
B : , B̄ : are with respect to the non-Kähler metrics χt . The operator Et is a 4th order elliptic operator
as proved by Kodaira-Spencer. The first term in Et matches with (3.9), so we will need to look
for solutions in a space where all the other terms vanish; we see that if we can find a solution to
Epγq “ B̄ω with dγ “ 0, we will solve (3.9).
By the Fredholm alternative, we can find a solution γt P pker Et qK if B̄χt K ker Et . Note that
considering pEt ϕ, ϕq and integrating by parts shows that

ker Et “ tϕ : dϕ “ 0, B̄ : B : ϕ “ 0u.

We claim that there exists µ P Λ0,1 with

B̄χt “ iB B̄µt . (3.10)

Here we use the assumption of vanishing cohomology H 0,2 pX0 q “ 0. By the semi-continuity theo-
rem,
h0,2 pXt q ď h0,2 pX0 q “ 0
and by (3.7), we have rpΘ˚t ω0 q0,2 s P H 0,2 pXt q “ 0 and so we can write pΘ˚t ω0 q0,2 “ B̄ν for ν P Λ0,1 .
By (3.7), we have B̄χt “ ´B B̄ν.
It follows from (3.10) that for β P ker Et then

xB̄χt , βy “ xµt , B̄ : B : βy “ 0.

Therefore, we can find a solution to Et pγt q “ B̄χt “ iB B̄µt . Next, we prove that such a solution is
closed.
Lemma 3.8. If Epγq “ iB B̄µ, then dγ “ 0.

Proof. We compute

0 “ Et pγq ´ iB B̄µt
„  „  „ 
: : : : : : :
“ B B̄ B̄ B γt ´ iµt ` B B̄ B̄ Bγt ` Bγt ` B̄ B B B̄γt ` BB B̄γt ` B̄γt

:“ B B̄α1 ` B : α2 ` B̄ : α3 . (3.11)

Let σ “ B : α2 ` B̄ : α3 . Then the equation above implies dσ “ 0, and

}σ}2 “ pα2 ` α3 , pB ` B̄qσq “ 0.

Setting σ “ 0, we obtain
„  „ 
: : : : :
0 “ B B̄ B̄ Bγt ` Bγt ` B̄ B B B̄γt ` BB B̄γt ` B̄γt .

53
and taking an inner product with γt implies

0 “ }B̄ : Bγ}2 ` }Bγ}2 ` }B B̄γt }2 ` }B : B̄γ|2 ` }B̄γ}2

and so Bγ “ 0 and B̄γ “ 0.

In summary, by the Fredholm alternative we have γt P pker Et qK with Et γt “ B̄χt and dγt “ 0.
Therefore (3.9) holds and the corrected ωγ (3.8) satisfies dωγ “ 0. To show ωγ is a Kähler metric,
we prove the estimate
}γt }C 3 pX,g0 q ď C|t|. (3.12)
Then since χt ą p1{2qω0 for small t, we see that ωγ ą 0 for small t.
Lemma 3.9. There exists C ą 1 independent of t such that for all t small, we can estimate

}γt }C 4,α ď C}Et γt }C α , (3.13)

for all γt P pker Et qK with dγt “ 0.

Proof. Suppose by contradiction that the estimate fails, and there is a sequence ti Ñ 0 such that

}γt }C 4,α ě Ci }Et γt }C α ,

with Ci Ñ 8. Replacing γt by γt {}γt }C 4,α , so that we have a sequence γti with

}γti }C 4,α “ 1, }Eti γti }C α Ñ 0.

The Eti Ñ E0 smoothly, and after relabeling a subsequence we have that γti Ñ γ0 in C 4,α{2 with
E0 γ0 “ 0 and dγ0 “ 0. This limit is non-trivial due to the Schauder estimates

}γt }C 4,α ď Cp}γt }L8 ` }Et γt }C 8 q

where C ą 1 is uniform in t as Et is a smoothly varying family of elliptic operators on a fixed

smooth manifold, and by compactness of t, the coefficients of Et and its ellipticity are uniformly
bounded. Taking t Ñ 8 implies }γ0 }L8 ě C ´1 .
Since E0 γ0 “ 0, we showed earlier that B̄ : B : γ0 “ 0. Since B̄γ0 “ 0, we have B̄B : γ0 “ 0 by the Kähler
identities on the Kähler central fiber X0 . Now B : γ0 P H 0,2 pX0 q and we are assuming H 0,2 pX0 q “ 0,
so B : γ0 “ B̄q. Therefore B̄ : B̄q “ 0, and so B̄q “ 0. It follows that

Bγ0 “ 0, B : γ0 “ 0.

Hence γ P ker ∆B , and in Kähler geometry ∆B “ ∆d , so

dγ0 “ 0, d: γ0 “ 0.

We will now use γt P pker Et qK , dγt “ 0, to show that γ0 “ 0, which is a contradiction. Since
Λ1,2 “ Im Et ‘ ker Et , we can write γt “ Et pβt q, and so

γt “ B B̄β1 ` B̄ : β2 ` B : β3 .

54
Since dγt “ 0, we have
0 “ pγt , B̄ : β2 ` B : β3 q “ }B̄ : β2 ` B : β3 }2 .
So γt “ dαt . We note
pγt , γ0 qL2 pX0 q “ pαt , d:0 γ0 qL2 pX0 q “ 0.
Let t Ñ 0, we conclude
pγ0 , γ0 qL2 pX0 q “ 0
which is a contradiction.

Using this lemma, we obtain

}γt }C 4,α ď C}B̄χt }C α “ C}BpΘ˚t ω0 q0,2 }C α .

The estimate (3.12) now follows from

}pΘ˚t ω0 q0,2 }C k ď Ck |t|.

This is because Θ0 “ id, ω0 P Λ1,1 , and

ˇż t ˇ żt żt
˚ 0,2
ˇ d ˚
ˇ
0,2 ˇ
|pΘt ω0 q | “ ˇ
ˇ pΘs ω0 q dsˇ ď |LVs ω0 |ds ď C ds.
0 ds 0 0

and similarly for any C k norm of pΘ˚t ω0 q0,2 .

We now prove that the forms ωγt Ñ ω as t Ñ 0.

}ωγt ´ ω}C 0 pX,g0 q ď }χt ´ ω}C 0 pX,g0 q ` C}γt }C 3 pX,g0 q ď C|t|.

Similarly, ωγt Ñ ω as t Ñ 0 in any C k norm.

Corollary 3.10. Let X Ñ ∆ be a family of complex manifolds with X0 a compact Kähler manifold.
Then hp,q pXt q “ hp,q pX0 q for all small enough t.

Proof. By the Kodaira-Spencer theorem, Xt is a Kähler manifold for all t small enough. By the
semi-continuity theorem, hp,q pXt q ď hp,q pX0 q for small t, so we suppose by contradiction that there
exists t and p, q such that hp,q pXt q ă hp,q pX0 q. Let p ` q “ k, and
ÿ ÿ ÿ
hi,j pXt q ă hi,j pX0 q “ bk “ hi,j pXt q
i`j“k i`j“k i`j“k

which is a contradiction. Here we used that X0 and X

řt are diffeomorphic so they have the same
Betti numbers bk , and the Hodge decomposition bk “ i`j“k hi,j for Kähler manifolds.

55
4 Calabi-Yau Threefolds
4.1 Parameters of threefolds
There are various inequivalent definitions of Calabi-Yau manifolds used in the literature. The
Wikipedia page for Calabi-Yau manifolds gives some of the commonly used definitions. In these
notes, we will use the following definition:
Definition 4.1. Our definition of a Kähler Calabi-Yau manifold is a simply-connected compact
complex manifold of dimension n admitting a Kähler metric ω and a nowhere vanishing holomorphic
pn, 0q form Ω.
The section Ω is a nowhere vanishing holomorphic section of the canonical bundle KX “ pdet T 1,0 Xq˚ ,
and so KX “ OX is the trivial holomorphic bundle.
Let pY, Ω, ω̂q be a Kähler Calabi-Yau threefold. Since we assume that Y is simply connected, we
have b1 pXq “ 0 and so by the Hodge decomposition (2.11) then

h1,0 “ h0,1 “ 0.

In addition to the Hodge symmetries

hp,q “ hq,p , hp,q “ h3´p,3´q ,

Calabi-Yau threefolds satisfy

h0,p “ h3,p .
This is just H p pX, KX q “ H p pX, OX q. Therefore, the only Hodge numbers to consider are

h1,1 , h2,1 .

By the Hodge decomposition (2.11), we have b2 “ h1,1 and b3 “ 2 ` 2h2,1 . The Euler characteristic
is defined by
ÿ6
χpY q “ p´1qi bi ,
i“1

which becomes in this case

χpY q “ 2ph1,1 ´ h2,1 q.
In summary, for each Calabi-Yau threefold, we associate two parameters ph1,1 , h2,1 q.
‚ h1,1 encodes the Kähler classes of X. A Kähler metric ω produces a non-zero class
2
rωs P HdR pX, Rq X H 1,1 pX, Rq.
2
ş ş
To see rωs ‰ 0 P HdR , consider X ω n . If ω n “ dα, then X ω n “ 0. Onş the other hand,
ω n “ pdet gq idz 1 ^ dz̄ 1 ^ ¨ ¨ ¨ ^ idz n ^ dz̄ n . Since pdet gq ą 0 at all points, then X ω n ą 0.
Let C be the set of all rαs P H 1,1 pX, Rq such that there exists a Kähler metric ω with ω P rαs. We
call C the Kähler cone. It turns out that C is an open convex cone in H 1,1 pX, Rq (see e.g. Tosatti’s
note [28]). Therefore the Kähler cone has real dimension h1,1 . By the B B̄-Lemma, if rωs P C,
then
rωs “ tω ` iB B̄ϕ ą 0 with ϕ P C 8 pX, Rqu

56
and a given Kähler class rωs is parametrized by functions.
‚ h2,1 encodes the infinitesimal complex structure deformations of X. We discussed earlier how
a 1-parameter family of complex structures pX, Jptqq produces an element rηs P H 1 pX, T 1,0 Xq by
9
η “ Jp0q. Note that by Serre duality

dim H 1 pT 1,0 Xq “ dim H 2 pKX b pT 1,0 Xq˚ q “ dim H 2 pΩ1 q “ h1,2 .

On a Calabi-Yau manifold, the inverse problem can be solved.

Theorem 4.2. (Bogomolov-Tian-Todorov Theorem) Let pX, Jq be a Kähler Calabi-Yau threefold.
Let rηs P H 1 pX, T 1,0 Xq. Then η can be attained by a path of complex structures pX, Jptqq such that
9
Jp0q “ J and rJp0qs “ rηs.
Textbook references for the proof of this include e.g. [14, 16]. Since dim H 1 pX, T 1,0 Xq “ dim H 2 pX, Ω1 q
on a Calabi-Yau threefold, the number h2,1 is understood as parametrizing deformations of complex
structure.

4.2 Ricci flat metrics

Let X be a complex manifold with holomorphic volume form Ω and hermitian metric ω. In local
holomorphic coordinates, then

ω “ igj k̄ dz j ^ dz̄ k , Ω “ f dz 1 ^ . . . dz n

where f pzq is a local nowhere vanishing holomorphic function. The norm of Ω induced on pdet T 1,0 Xq˚
is
f pzqf pzq
|Ω|2ω “ .
det gj k̄
The Chern-Ricci curvature Rj k̄ “ ´Bk̄ Bj log det g can also be written

Rj k̄ “ Bk̄ Bj log |Ω|2ω .

Indeed,
Bk̄ Bj log det |Ω|2ω “ Bk̄ Bj log |f |2 ´ Bk̄ Bj log det g,
and Bk̄ Bj log |f |2 “ 0 for any such function f . This is because

f Bk̄ f¯ Bk f
Bj Bk̄ log |f |2 “ Bj ¯ “ Bj̄ “ 0.
ff f

The hermitian metric ω is Chern-Ricci flat if Rj k̄ “ 0 which implies

0 “ g j k̄ Bj Bk̄ log |Ω|2ω .

The maximum principle on compact manifolds implies that |Ω|ω is constant. Here is a quick proof
if ω is Kähler. (Note: this PDE result holds without the Kähler assumption. The general proof
uses the Hopf maximum principle instead of integration by parts. The general statement is: if
aij Bi Bj f ` bi Bi f “ 0 on a compact manifold with aij positive-definite, then f is constant.)

57
First, note the identities for a real function u : X Ñ R:
1 j k̄
iB B̄u ^ ω n´1 “ pg uj k̄ qω n , uj k̄ “ Bj Bk̄ u,
n
and
1
iBu ^ B̄u ^ ω n´1 “
|Bu|2g ω n .
n
These can be checked at a point p P X with ω|p “ k idz k ^ dz̄ k . By Stokes’s theorem and
ř
dω “ 0, ż ż
plog |Ω|2ω qidB̄ log |Ω|2ω ^ ω n´1 “ ´ dplog |Ω|2ω q ^ iB̄ log |Ω|2ω ^ ω n´1 .
X X
which implies ż ż
0“ plog |Ω|2ω qpg j k̄ Bj Bk̄ log |Ω|2ω q ω n “´ |B log |Ω|2ω |2g ω n
X X
and so |B log |Ω|2ω |2g “ 0 and log |Ω|2ω is a constant.
In summary, on a compact complex manifold with trivial canonical bundle, then Rj k̄ “ 0 is equiv-
alent to |Ω|2ω “ const. To find such metrics, we fix ω an arbitrary hermitian metric, and look for
solutions to this equation of the form
g̃j k̄ “ gj k̄ ` Bj Bk̄ ϕ ą 0
or in differential form notation
ω̃ “ ω ` iB B̄ϕ ą 0.
The equation |Ω|2ω̃ pxq b
“ e , b P R, can be written
|Ω|2ω̃ log |Ω|2ω
„ 
b det g log |Ω|2ω
e “ e “ e ,
|Ω|2ω det g̃
which leads to the complex Monge-Ampère equation
„ 
detpgj k̄ ` ϕj k̄ q 2
“ elog |Ω|ω ´b .
det gj k̄
When dω “ 0, the constant b can be identified from the initial data pω, Ωq, since the equation can
also be written
pω ` iB B̄ϕqn “ e´b |Ω|2ω ω n
and integration of both sides and Stokes’s theorem gives
ż ż
n
ω “ e´b |Ω|2ω ω n .
X X

Indeed:
ż ż n
ÿ ż
pω ` iB B̄ϕqn “ ωn ` ck ω n´k ^ piB B̄ϕqk
X X k“1 X
ż ÿn ż
“ ωn ` ck dpω n´k ^ piB̄ϕq ^ piB B̄ϕqk´1 q
X k“1 X
ż
“ ωn . (4.1)
X

58
Another way to write the Kähler-Ricci flat metric equation is
2
pω ` iB B̄ϕqn “ e´b in Ω ^ Ω̄,

because of the identity

2 ωn
in Ω ^ Ω̄ “ |Ω|2ω .
n!
Yau’s theorem states:
Theorem 4.3. [31] LetşpX, ωq beş a compact Kähler manifold. Let eh be an arbitrary function and
b be the constant eb “ X eh ω n { X ω n . Then there exists a unique smooth solution u : X Ñ R
solving
pω ` iB B̄uqn “ eh´b ω n , ω ` iB B̄u ą 0
ş
and X u ω n “ 0.
References for the proof of this theorem include: Chapter 2 of Siu’s notes [24] and Chapter 3 of G.
Szekelyhidi’s book [26].
The complex Monge-Ampère equation can also be solved for a pair pu, bq on a general hermitian
manifold pX, ωq: this theorem is due to Tosatti-Weinkove [29].
As a consequence of Yau’s theorem, a Kähler Calabi-Yau manifold admits Kähler Ricci-flat metrics.
These are also called Calabi-Yau metrics ωCY and they solve

RicpωCY q “ 0, dωCY “ 0.

Furthermore, each Kähler class rωs P H 1,1 pX, Rq contains a unique Calabi-Yau representative ωCY P
rωs by solving the Monge-Ampère equation with ansatz ω ` iB B̄ϕ.

4.3 Deformations of complex structure

By the BTT theorem, on a Calabi-Yau threefold X then any element rηs P H 1 pX, T 1,0 Xq can
be attained by a family of complex manifolds. Here we note that a deformation of a Calabi-Yau
threefold also carries a Calabi-Yau structure.
Proposition 4.4. Let pX, J, ω, Ωq be a Kähler Calabi-Yau threefold. Let pX, Jptqq be a smooth path
of complex structures with Jp0q “ J and |t| ă ε. For ε ą 0 small enough, then there exists a family
pωptq, Ωptqq with pωptq, Ωptqq Ñ pω, Ωq as t Ñ 0 such that ωptq is a Kähler metric and Ωptq is a
nowhere vanishing holomorphic volume form on pX, Jptqq.

Proof. The existence of the family of Kähler metrics ωptq “ ωt with ωt Ñ ω is Kodaira-Spencer’s
stability theorem [21] which we discussed earlier. We now describe how to construct Ωptq “ Ωt . The
3-form Ω is defined on pX, Jptqq, however it does not necessarily have type p3, 0q. So we take the
p3, 0q part pΩq3,0
t , and this is nowhere vanishing for small t by continuity of the complex structures
Jptq. We now need to correct pΩq3,0t to make it holomorphic. Write

pΩq3,0
t “ σt dzt1 ^ dzt2 ^ dzt3 .

59
Note α “ σt´1 B̄t σt is a well-defined p0, 1q-form on Xt . This is because of the transformation law
σ ÞÑ tU V σ with tU V holomorphic, which implies αk̄ ÞÑ αk̄ . We also note that B̄α “ 0, since
B̄α “ ´σ ´2 B̄σ ^ B̄σ ` σ ´1 B̄ 2 σ “ 0.
We claim that we can find a smooth function ut such that
ż
B̄t ut “ ´αt , u ωtn “ 0
Xt

This is because X is assumed to be simply-connected, and so H 1 pX, OX q “ 0. By the semi-

continuity theorem, H 1 pXt , OXt q ď H 1 pX, OX q, hence H 1 pXt , OXt q “ 0. The definition of Dol-
beault cohomology states that the closed p0, 1q-form α comes from B̄ of a function. Using u, we
define
Ωt “ eut σt dzt1 ^ dzt2 ^ dzt3 .
This solves B̄Ωt “ 0 since
Bk̄ peu σq “ eu Bk̄ uσ ` eu Bk̄ σ “ eu p´σ ´1 pBk̄ σqσ ` Bk̄ σq “ 0.
Therefore Ωt defines a nowhere vanishing holomorphic volume form on pX, Jt q. Next, we show
Ωt Ñ Ω. For this, we show
|ut | ď C|t|. (4.2)
Then since σt Ñ σ smoothly,
|eu σt ´ σ| ď |σt ||eu ´ 1| ` |σt ´ σ|
ď C|eC|t| ´ 1| ` C|t|
8
ÿ 1
ď C pC|t|qk ` C|t| ď C|t|. (4.3)
k“1
k!

Hence |Ωt ´ Ω|g ď C|t|. To prove (4.2), we will use the complex Laplacian
∆gt : C 8 pX, Rq Ñ C 8 pX, Rq,
given by
∆gt f “ pgt qj k̄ Bj Bk̄ f.
ş
We proved earlier that ker ∆gt “ R ¨ 1 are constants. Therefore since Xt ut ωtn “ 0, we have that
ut P kerp∆gt qK with respect to the gt inner product for all t. By elliptic estimates,
}ut }C 2,α pX,gq ď C}∆gt ut }C α pX,gq .
One should verify that the constant C is uniform in t. We omit the proof, but it follows an outline
similar to (3.13): the usual Schauder estimates }u}C 2,α ď Cp}u}L8 ` }∆u}C α q are uniform in t, and
to remove the extra }u}L8 we assume the estimate fails as t Ñ 0 and derive a contradiction.
Therefore since Bk̄ u “ σt´1 Bk̄ σt

}ut }C 2,α pX,gq ď C}g j k̄ Bj pσt´1 Bk̄ σt q} ď C|t|

since σt tends smoothly to the holomorphic σ with B̄σ “ 0.

60
 9
Let pX, Jptq, ωptq, Ωptqq be a family of Kähler Calabi-Yau threefolds with Jp0q “ η. We use the
d
notation δ “ dt |t“0 . To end this section, we will compute the variation

χ :“ δΩ

and relate it to η via „ 

i
χᾱβγ “ ´ η µ ᾱ Ωµβγ .
2
Here everything is evaluated at holomorphic coordinates with respect to the initial pX, J, Ωq. We
will show χ P Λ2,1 pX, Jq, and since B̄η “ 0, it follows that B̄χ “ 0.
We start by differentiating
J r i J s j J t k Ωrst “ ´iΩijk (4.4)
to obtain

´ iδΩijk “ δJ r i J s j J t k Ωrst ` J r i δJ s j J t k Ωrst ` J r i J s j δJ t k Ωrst ` J r i J s j J t k δΩrst (4.5)

In holomorphic coordinates α, β, γ, then J α β “ iδ α β , J ᾱ β̄ “ ´iδ α β , and the only non-zero compo-

nents of Ω are unbarred, which implies

δΩᾱβ̄γ̄ “ 0, δΩᾱβ̄γ “ 0 (4.6)

and
´ iδΩᾱβγ “ ´δJ r ᾱ Ωrβγ ` δJ s j Ωᾱsγ ` δJ t k Ωᾱβt ` p´iqi2 δΩᾱβγ (4.7)
which implies „ 
i µ
χᾱβγ :“ δΩᾱβγ “ ´ δJ ᾱ Ωµβγ . (4.8)
2
We can also invert the formula for χ to get δJ from δΩ. At a point where gij “ δij , then

Ω “ f pzqdz 1 ^ dz 2 ^ dz 3 , |Ω|2ω “ |f pzq|2 , (4.9)

and „ 
i
Ω̄βγν ´ Ωµβγ δJ µ ᾱ “ ´i|Ω|2ω δJ ν ᾱ . (4.10)
2
Thus
i
δJ ν ᾱ “ Ω̄βγν δΩᾱβγ (4.11)
|Ω|2ω
or
i
η ν ᾱ “ Ω̄βµν χᾱβµ . (4.12)
|Ω|2ω

61
4.4 Quintic threefolds
4.4.1 Holomorphic volume form
Consider X Ă P4 given by tP “ 0u, where

P “ x50 ` x51 ` x52 ` x53 ` x54 .

In the coordinate chart U0 “ tx0 ‰ 0u with coordinates zi “ zi {x0 , X appears as

tf “ 1 ` z15 ` z25 ` z35 ` z45 “ 0u Ă C4 .

Since
Df “ r5z14 , 5z24 , 5z34 , 5z44 s
we see that Df has maximal rank except at z “ 0, which is not included in the set tf “ 0u.
Therefore X is a smooth complex manifold of complex dimension three. This manifold admits a
holomorphic volume form Ω. This can be seen by the adjunction formula, but also by the explicit
expression
dz1 ^ dz2 ^ dz3
Ω|U0 XV4 “ (4.13)
Bf {Bz4
over the set U0 “ tx0 ‰ 0u intersected with V4 “ tBf {Bz4 ‰ 0u. We now verify that Ω extends
from U0 X V to a holomorphic volume form on all of X.

‚ Extending from V4 to V3 . On the intersection V3 X V4 , by the implicit function theorem we can

write z3 “ gpz1 , z2 , z4 q. Therefore

dz1 ^ dz2 ^ dg Bg{Bz4

Ω“ “ dz1 ^ dz2 ^ dz4 .
Bf {Bz4 Bf {Bz4

Differentiating f pz1 , z2 , gpz1 , z2 , z4 q, z4 q “ 0 in z4 gives

B3 f B4 g ` B4 f “ 0.

Hence
dz1 ^ dz2 ^ dz4
Ω|U0 XV3 “ ´
Bf {Bz3
This is holomorphic and nowhere vanishing on V3 , therefore Ω extends from V4 to V3 . Similar
arguments show that Ω defines a holomorphic volume form on all of U0 . Next, we need to extend
Ω beyond U0 Ă X.

‚ Extending from U0 to U1 . On U0 we have coordinates zi “ xi {x0 , and on U1 we have coordinates

w1 “ x0 {x1 , and for i ě 2 then wi “ xi {x1 . The change of coordinates on U0 X U1 is then

z1 “ w1´1 , z2 “ w1´1 w2 , z3 “ w1´1 w3 , z4 “ w1´1 w4 .

The holomorphic volume form on U0 X U1 X V4 becomes

dw1´1 ^ dpw1´1 w2 q ^ dpw1´1 w3 q

Ω“
5z44

62
since Bf {Bz4 “ 5z44 , and therefore

dw1 ^ dw2 ^ dw3

Ω“´ .
5w44

Over U1 , tP “ 0u appears as tf˜ “ 1 ` w15 ` w25 ` w35 ` w45 “ 0u, and so

dw1 ^ dw2 ^ dw3

Ω|U1 XṼ4 “ ´ .
B f˜{Bw4

As before, Ω extends from U1 X Ṽ4 to all of U1 , and similarly Ω extends to a nowhere vanishing
holomorphic form on U2 , U3 , U4 .
Putting everything together, we see that the local expression (4.13) defines a holomorphic volume
form on all of X.

4.4.2 Hodge numbers

The Hodge numbers of the quintic threefold X are

h1,1 “ 1, h1,2 “ 101.

‚ h1,1 “ 1. This follows from the Lefschetz hyperplane theorem, which we recall states: let Y Ď X
be a complex hypersurface such that its associated line bundle OpY q Ñ X is positive. Then

hp,q pY q “ hp,q pXq, p ` q ď n ´ 2.

We computed that for Y “ tP “ 0u Ă Pn with P of degreen k, then OpY q “ Opkq. We also

computed that the Fubini-Study metric on Op1q has positive curvature, hence Opkq “ rOp1qsk is
also a positive bundle. Therefore the Lefschetz hyperplane theorem applies to the quintic Y Ă P4 ,
and
h1,1 pY q “ h1,1 pP4 q.
We computed earlier that h1,1 pPn q “ 1, and so h1,1 “ 1 for the quintic.
‚ h1,2 “ 101. We only give here a heuristic argument from the string theory literature, but a real
proof can be found in e.g. [23]. We discussed earlier that h1,2 parametrizes complex structure
deformations. We can deform the complex structure of the quintic
4
ÿ
Zi5 “ 0
i“0

by introducing parameters cI “ ci0 i1 i2 i3 i4

ÿ
cI ZI “ 0, ZI “ Z0i0 Z1i1 Z2i2 Z3i3 Z4i4 . (4.14)
|I|“5

The number of parameters are: number of ways of placing 5 objects (the powers) in 5 bins (the Zi ),
which is p5 ` 4q!{5!4! “ 126. But some of these 126 coefficients do not give genuine deformations

63
of complex structure. There are 52 degrees of freedom coming from matrices A P GLp5q which
produces a biholomorphism

A : tP pZq “ 0u Ď P4 Ñ tP pAZq “ 0u Ď P4 .

So we are left with 126 ´ 25 “ 101 parameters, which matches up with h1,2 “ 101. This is not a
proof because a priori there could be ways to deform the complex structure of a quintic which is
not by (4.14).

4.4.3 Nodal singularities

We now the quintic by a parameter t P C and consider
" 4
ÿ *
Xt “ Qt “ Zi5 ´ 5tZ0 Z1 Z2 Z3 Z4 “ 0 Ď P4 .
i“0

We note in passing that this family was used by [3] to construct one of the first examples of mirror
symmetry. We first notice that at t “ 1, this is no longer a smooth manifold as it contains singular
points. To find the singular points, we set all derivatives of Qt to zero.
ź
5Zk4 “ 5t Zi
i‰k

and so ˙4 ˙4
4
ˆź 4
ˆź
5
Zi “t Zi
i“0 i“0

If one of the Zi “ 0, then they all are, which is not a point in projective space. So we conclude
that singular points occur when t5 “ 1.
We let t “ 1 and investigate the singular quintic which we denote X. There are 125
ś singular points:
these occur when Z05 “ Z15 “ ¨ ¨ ¨ “ Z45 “ Zi . Dividing the singular points by Zi (in projective
ś
space), singular points are given by roots of unity Zk5 “ 1.

q “ rζ α0 , . . . , ζ α4 s, ζ 5 “ 1
ř
with αi P t0, 1, 2, 3, 4u. Since q P X, we must have i αi “ 0 mod 5. We can always represent q in
projective spaceřwith the first entry equal to 1, so that leaves 3 free parameters α1 , α2 , α3 with α4
determined by i αi “ 0, so there are p5qp5qp5q “ 125 singular points.
We now look locally near the point q “ r1, 1, 1, 1, 1s. In the coordinate chart pU0 , zq, q “ p1, 1, 1, 1q
and the equation of the singular quintic is
4
ÿ
f pzq “ 1 ` zi5 ´ 5z1 z2 z3 z4 “ 0.
i“1

The holomorphic function f satisfies f pqq “ 0 and Df pqq “ 0. Its holomorphic Hessian matrix

64
 B2 f
Bz i Bz j at q is » fi
20 ´5 ´5 ´5
— ffi
— ffi
—´5 20 ´5 ´5ffi
— ffi
— ffi
—´5 ´5 20 ´5ffi
– fl
´5 ´5 ´5 20
This is a non-singular matrix. Thus we have a holomorphic function f pz1 , . . . , z4 q with f pqq “ 0,
df pqq “ 0, but D2 f pqq is non-degenerate. There is a holomorphic Morse lemma (e.g. Lemma
2.11 in the book [30]) which gives the existence of holomorphic coordinates w with qpwq “ 0 such
that
ÿ 4
f pwq “ wi2 .
i“1

We give the proof in complex dimension n “ 2 following (Lemma 42 p.242 in [8]). First, shift
coordinates so that q “ 0 and f pz1 , z2 q is a holomorphic function with f p0q “ 0, Df p0q “ 0, and
Bi Bj f p0q a non-degenerate symmetric matrix. For a symmetric complex matrix A, there is a unitary
matrix U such that U AU T “ D where D is diagonal (Autonne–Takagi factorization). Since A is
non-degenerate the diagonal elements are non-zero, so we may multiply on both sides by a diagonal
complex matrix to form SAS T “ I.
So we can write S T rD2 f p0qsS “ I for a complex matrix S, and let z k “ S i k y k . By the chain
rule
Bf Bf Bz p p B2 f
“ “ S k f p , “ S p k fpq S q j .
By k Bz p By k By j By k
Written in matrix notation, this is

Dy2 f p0q “ S T rD2 f p0qsS “ I.

Hence we may assume that the power series of f is

ÿ
f “ z12 ` z22 ` aij z1i z2j .
i`ją2

Bf
Let f2 “ Bz 2
. Since B2 f2 p0q ‰ 0, by the holomorphic implicit function theorem there is a function
apz1 q with ap0q “ 0 such that f2 pz1 , apz1 qq “ 0. Differentiating this gives a1 p0q “ 0. Define new
coordinates by
z̃1 “ z1 , z̃2 “ z2 ´ apz1 q
and let f˜pz̃1 , z̃2 q “ f pz1 , z2 ´ apz1 qq. We will compute the power series of f˜. We start with 1st
derivatives:
B1 f˜ “ B1 f ´ a1 B2 f, B2 f˜ “ B2 f.
The key observation is that B2 f˜ ” 0 when z̃2 “ 0, and so

B2 f˜ “ z̃2 Qpz̃1 , z̃2 q. (4.15)

Next, we move on to second order derivatives.

B1 B2 f˜ “ z̃2 B1 Qpz̃1 , z̃2 q,

65
 B1 B1 f˜ “ B1 B1 f ´ 2a1 B1 B2 f ` pa1 q2 B2 B2 f ´ a2 B2 f
B2 B2 f˜ “ B2 B2 f.
Evaluating all these at zero, and using that (4.15) implies B1k Bf f p0q “ 0, we get the expansion
ÿ ÿ
f “ z̃12 ` bk z̃1k ` z̃22 ` aij z̃1i z̃2j
kě3 i`jě3, jě2
ˆ ÿ ˙ ˆ ÿ ˙
“ z̃12 1` bk z̃1k ` z̃22 1` aij z̃1i z̃2j (4.16)
kě1 i`jě3, jě0

We can now let

ˆ ÿ ˙1{2 ˆ ÿ ˙1{2
w1 “ z̃1 1 ` bk z̃1k , w2 “ z̃2 1 ` aij z̃1i z̃2j ,
kě1 i`jě3, jě0

using that a holomorphic function ψ : C Ñ C with ψp0q ‰ 0 has a local square root defined in a
neighborhood of the origin. In these new coordinates then, f “ w12 ` w22 .
In summary, the singular quintic X has holomorphic coordinates around each singular point where
the singularity appears as
"ÿ 4 *
zi “ 0 Ď C4 .
2

i“1

Such singularities are called nodes, or nodal singularities, or ordinary double points (ODP).

4.4.4 Examples of conifold transitions

Example 4.5. There are 2 ways to desingularize
4
"ÿ 4
ź *
X“ Zi5 ´ 5 Zi “ 0 Ď P4
i“0 i“0

which is the singular quintic discussed in the previous section.

‚ The first way is called a smoothing, which is to realize X as the central fiber of
4
"ÿ *
Xt “ Zi5 ´ p5 ` tqZ0 Z1 Z2 Z3 Z4 “ 0 Ă P4 . (4.17)
i“0

We discussed earlier how for t ‰ 0, the space Xt no longer has singular points. In the chart pU0 , zq,
we see the zero set
ÿ4
ft pzq “ 1 ` zi5 ´ 5z1 z2 z3 z4 ´ tz1 z2 z3 z4 “ 0.
i“1

Near p1, 1, 1, 1q, we apply the holomorphic Morse lemma to

ř4
1 ` i“1 zi5
gpzq “ ´ 5,
z1 z2 z3 z4

66
 wi2 (here w “ 0 corresponds to the point p1, 1, 1, 1q) so that the zero set
ř
and write gpwq “ i
tft “ 0u becomes
4
ÿ
t wi2 “ tu Ă C4 . (4.18)
i“1

wi2 “ 0 has been replaced by wi2 “ t.

ř ř
for new coordinates wi . Locally, the ODP singularities i i

‚ The second way to desingularize X is by small resolution. We will discuss how this is done in full
detail in the next section, but it results in a map

σ:XÑX

where X is a complex manifold, σ ´1 ppq “ P1 for each singular point p, and σ is a biholomorphism
outside the singular set on X. In other words, each singular point of X is replaced by P1 .
The holomorphic volume form Ω on X defines a holomorphic volume form Ω̂ “ σ ˚ Ω defined on
XzE with E “ YP1 . By Hartog’s theorem, Ω̂ extends to all of X and so X has trivial canonical
bundle. We will show later that χpXq ´ 2N “ χpXt q where N is the number of nodes. Therefore
the two threefolds have different topologies.
Note Hartog’s theorem (e.g. p.46 in [7]) states: Let X be a complex manifold. Let S Ă X be a
closed complex submanifold of complex codimension ě 2: this means there are local coordinates
where U X S is given by z1 “ ¨ ¨ ¨ “ zp “ 0 for p ě 2. Then every holomorphic function f on XzS
extends holomorphically to X. This is major difference with complex analysis in C: f pzq “ 1{z
does not extend on Czt0u, but any holomorphic f pz1 , z2 q extends on C2 zt0u.
Example 4.6. Here is another example from Greene-Morrison-Strominger [12] (see also the expo-
sition in [23]). Define a singular quintic X Ă P4 by the polynomial

P “ Z3 GpZ0 , . . . , Z4 q ` Z4 HpZ0 , . . . , Z4 q “ 0. (4.19)

where G “ Z34 ` Z24 ´ Z04 and H “ ´Z44 ´ Z14 ´ Z04 . The singular points are where DP “ 0, which
happens when
Z3 “ 0, Z4 “ 0, G “ 0, H “ 0. (4.20)
There are 16 singular points. (This is also expected by Bezout’s theorem: n homogeneous poly-
nomials of degrees di in projective space of dimension n has d1 ¨ dn intersection points generically.
16 “ p1qp1qp4qp4q singular points.) We now look at the local model for these singularities. Suppose
p P X is a singular point with p P U0 “ tZ0 ‰ 0u. In coordinates zi “ Zi {Z0 the equation for X is

z3 gpzq ` z4 hpzq “ 0. (4.21)

Since gpzq has surjective derivative, by the implicit function theorem there is a holomorphic change
of coordinates
w1 “ gpz1 , z2 , z3 , z4 q, wi`1 “ zi`1 (4.22)
with inverse z1 “ ϕpw1 , w2 , w3 , w4 q. We can repeat this for hpzq and obtain coordinates w̃i with
w̃2 “ hpw1 , w2 , w3 , w4 q. Then in these coordinates, the equation for X is a neighborhood of the
origin in
tw̃3 w̃1 ` w̃4 w̃2 “ 0u Ă C4 . (4.23)

67
This is the model for a nodal singularity. There is another change of coordinates
w̃1 ` iw̃2 w̃1 ´ iw̃2 w̃3 ` iw̃4 w3 ´ iw4
z1 “ , z2 “ , z3 “ , z4 “ . (4.24)
2 2 2 2
such that this singularity is represented by
4
ÿ
t zi2 “ 0u Ă C4 . (4.25)
i“1

One way to resolve X is to consider the family of quintics

Xt “ tZ3 G ` Z4 H “ tZ0 Z1 Z2 Z3 Z4 u Ď P4 .
which is now non-singular for t ‰ 0. As a quintic, it has h1,1 pXt q “ 1.
Another way is by small resolution. Let X Ă P4 ˆ P1 given by
U Z4 ´ V Z3 “ 0, U GpZq ` V HpZq “ 0
1 4
with rU : V s P P , rZ0 : ¨ ¨ ¨ : Z4 s P P . There is a map
σ : X Ñ X, σprZ0 : ¨ ¨ ¨ : Z4 s, rU : V sq “ rZ0 : ¨ ¨ ¨ : Z4 s.
(check lands in X). When P P SingpXq is a singular point Z3 “ Z4 “ G “ H “ 0, then σ ´1 pP q is
a full P1 . Otherwise if one of these is non-zero, there is a single point in σ ´1 pP q. So we have a map
such that σ ´1 ppi q “ Ci – P1 for pi P SingpXq and σ ´1 is a biholomorphism on XzSingpXq. There
are 16 exceptional curves Ci , and we notice that they are not linearly independent in homology.
ÿ
ai rCi s “ 0 in H2 pX, Cq.
i
1,1
This is because b2 pXq “ h pXq “ 2. To show this, we use the Lefschetz hyperplane theorem
applied to
X Ă X 1 “ tU Z4 ´ V Z3 “ 0u Ă P4 ˆ P1 .
Let P “ U GpZq ` V HpZq so X “ tP “ 0u Ď X 1 . Then OpXq Ñ X 1 is rOP4 p4q b OP1 p1qs|X 1 Ñ X 1 .
Let’s verify this in two specific charts. Take a chart U1 with coordinates u “ U {V P P1 and
z “ Z{Z0 in P4 , and chart U2 with v “ V {U and w “ Z{Z4 . In chart U1 , the defining equation of
X is in coordinates:
p1 pu, zq “ P {V Z04 “ 0,
and in chart U2 the defining equation is p2 “ P {U Z44 . The cocycle of OpXq on U1 X U2 is by
definition p1 {p2 “ U Z44 {V Z04 , which is the cocycle of OP4 p4q b OP1 p1q.
By the Lefschetz hyperplane theorem, we conclude
h1,1 pXq “ h1,1 pX 1 q.
Similarly, we can show
h1,1 pX 1 q “ h1,1 pP4 ˆ P1 q.
By the Kunneth formula, b2 pP4 ˆ P1 q “ 2. Since b2 “ h1,1 ` 2h2,0 and pullbacks of Kähler metrics
give h1,1 ě 2, we have h1,1 pP4 ˆ P1 q “ 2. So
h1,1 pXq “ 2.

68
5 Conifold Transitions: Local Model
5.1 Blowing-up a nodal singularity
5.1.1 Blow-up review
We start this section by recalling the blow-up construction. The blow-up of Cn at 0 is

X “ tpx, rusq P Cn ˆ Pn´1 : x P rusu,

where x P rus means x “ λu for some λ P C. The exceptional divisor E Ă X is the set of points
p0, rusq, so that x “ 0 and E is a copy of Pn´1 . The projection

π : X Ñ Cn , πpx, rusq “ x

satisfies π ´1 p0q “ E “ Pn´1 and π is a biholomorphism on XzE.

We work out the coordinates for n “ 2. In this case, X consists of pairs

tpx, yq, ru : vsu P C2 ˆ P1

such that
x y
“ .
u v
To obtain coordinate charts on X, we use the coordinate charts on P1 . So in this case, there are
two charts:

‚ Chart U “ tu “ 1u. Coordinates are defined by

x “ x
y “ xv
u “ 1
v “ v,

so that we only keep the two coordinate px, vq on this patch on X. The exceptional divisor (where
px, yq “ p0, 0q) appears as E X U “ tx “ 0u, and v is a free coordinate.

‚ Chart V “ tv “ 1u. Coordinates are defined by

x “ uy
y “ y
u “ u
v “ 1,

and we only keep the coordinates py, uq on this set. The exceptional divisor is E X V “ ty “ 0u
with free coordinate u.

On U X V , the change of coordinates from px, vq to py, uq on the overlap is

x “ uy, v “ u´1 .

69
Indeed, in chart U , we have v “ y{x which substituting in chart V givs v “ y{uy “ u´1 . The
change of coordinates v “ u´1 is the coordinate change on P1 . The change x “ uy is the coordinate
change on a line bundle over P1 . Recall in general a section over a line bundle over P1 transforms
as
σ̃pvq “ τ puqσpuq
where τ puq is a local function on P1 which is the transition function of the bundle. In the case
above, we have τ puq “ u. This transition function defines a line bundle

Op´1q Ñ P1 .

So X is the total space of this line bundle. The other way to define Op´1q is to cover P1 “ tru1 : u2 su
by U1 “ tu1 ‰ 0u and U2 “ tu2 ‰ 0u. Then declare the transition function on the overlap U1 X U2
to be
u1
τ12 “ ,
u2
so that local sections transform as σ1 “ τ12 σ2 . Coordinates over U1 are ζ “ u2 {u1 and coordinates
over U2 are ξ “ u1 {u2 . Then ξ “ ζ ´1 and over U2 , we do have τ12 pξq “ ξ. Thus this is the same
space as above.

5.1.2 ODP in C3
We now illustrate how the blow-up procedure can be used to resolve singularities. Consider the
space
X “ txz ´ y 2 “ 0u Ă C3 .
This space is of the form F px, y, zq “ 0 with DF “ pz, ´2y, xq, and there is a singularity preventing
it from being a submanifold at p0, 0, 0q. We can resolve this singularity

σ : X̃ Ñ X

by blowing up the origin. That is, we look at pairs

tpx, y, zq, ru : v : wsu P C3 ˆ P2

with the relation

x y z
“ “ .
u v w
The P2 will produce 3 coordinate charts that we now describe:

‚ Chart U “ tu “ 1u. Coordinates px, v, wq satisfy:

x “ x, y “ vx, z “ wx.

and u “ 1, v “ v, w “ w. The defining equation for X becomes

xpwxq ´ pvxq2 “ 0

which is x2 pw ´ v 2 q “ 0. Recall E X U “ tx “ 0u is the exceptional divisor, so we throw out the

x2 . The resolution of X is in this chart is then defined by

w ´ v2 “ 0

70
which is smooth, and called the proper transform. Therefore X̃ has two coordinates px, vq, and the
exceptional divisor in X is the curve x “ 0 with v a free coordinate.

‚ Chart V “ tv “ 1u. Coordinates py, u, wq satisfy

x “ uy, y “ y, z “ wy,

and the proper transform is

uw ´ 1 “ 0.
If g “ uw ´ 1, the only problem with Dg “ pw, uq is when pw, uq “ 0 which is not on the curve
uw ´ 1 “ 0. So X̃ is smooth with two coordinates py, uq and exceptional divisor at y “ 0.

‚ Chart W “ tw “ 1u. Coordinates pz, u, vq satisfy:

x “ uz, y “ vz, z “ z.

and u “ u, v “ v, w “ 1. The proper transform is

u ´ v 2 “ 0.

Therefore X̃ has two coordinates pz, vq, and the exceptional divisor is the curve z “ 0 with v a free
coordinate.

Next, we note that X̃ is the total space of the line bundle Op´2q Ñ P1 . To see the P1 coordinate,
convert from U to V on U X V to obtain

v “ y{x “ y{uy “ 1{u.

For the fiber,

x“y
on U X V , so nothing going on here. Converting U to W on U X W gives v “ v, but for the
fiber
x “ uz “ v 2 z.
which is the transition function for Op´2q. Recall that by definition sections of Op´2q transform
as
σ̃ “ τ 2 σ
where τ pvq “ v is the transition function on Op´1q.

5.1.3 ODP in C4
Before discussing singularities, we recall that the blow-up of Cn along a subspace Z replaces Z by
the projectivization of its normal bundle PpN Zq. We give the concrete example of the blow-up
of
Z “ tz2 “ z4 “ 0u Ă C4 .
The blow-up along Z is

X “ tppz1 , z2 , z3 , z4 q, ru : vsq P C4 ˆ P1 : pz2 , z4 q P ru : vsu,

71
in other words
z2 z4
“ .
u v
The P1 provides two charts for Z.

‚ Chart U “ tu “ 1u. Coordinates pz1 , z2 , z3 , vq are defined by

z1 “ z1 , z2 “ z2 , z3 “ z3 , z4 “ vz2 , u “ 1, v“v

with exceptional divisor E X U “ tz1 “ z2 “ z3 “ 0u.

‚ Chart W “ tv “ 1u. Coordinates pz1 , z3 , z4 , uq are defined by

z1 “ z1 , z2 “ uz4 , z3 “ z3 , z4 “ z4 , u “ u, v“1

with exceptional divisor E X V “ tz1 “ z3 “ z4 “ 0u.

Consider now the ODP singularity

V “ tz1 z2 ´ z3 z4 “ 0u Ă C4 .

Note that this singulariy is the same as t i zi2 “ 0u by a change of coordinates (4.24). We can
ř
desingularize this space by blowing-up C4 along Z “ tz2 “ z4 “ 0u and taking the proper transform
Ṽ .

‚ Chart U “ tu “ 1u. Using the relations above, the equation for V becomes

z1 z2 ´ z3 pvz2 q “ 0.

We throw out z2 “ 0 to obtain the proper transform

z1 ´ vz3 “ 0,

which is now smooth. Therefore here Ṽ has three coordinates pz2 , z3 , vq, and the relations are

z1 “ vz3 , z2 “ z2 , z3 “ z3 , z4 “ vz2

on U .

‚ Chart W “ tv “ 1u. Similarly as above, the proper transform appears as

uz1 ´ z3 “ 0,

and here Ṽ has three coordinates pz1 , z4 , uq and the relations are

z1 “ z1 , z2 “ uz4 , z3 “ uz1 , z4 “ z4 .

on W .

We note that Ṽ can be identified with the total space of Op´1q ‘ Op´1q Ñ P1 . For this, we
compute the change of coordinates on the overlap U X W . Relabel

pσ, s, ζq “ pz2 , z3 , vq, on U

72
 ps̃, σ̃, ξq “ pz1 , z4 , uq on W.
Then chasing relations gives for example,

s̃ “ z1 “ vz3 “ ζs,

and altogether
s̃ “ ζs, σ̃ “ ζσ, ξ “ ζ ´1 .
This is the change of coordinates on the total space of Op´1q ‘ Op´1q Ñ P1 with fiber coordiantes
s, σ and P1 coordinate ζ. Thus we have a resolution of singularities

σ : Ṽ Ñ V

where the origin in V is replaced by σ ´1 p0q “ P1 . This can be seen since in local coordinates on U
then σ ´1 p0q is freely parametrized by coordinate v (and coordinates z2 “ z3 “ 0), and on V then
σ ´1 is freely parametrized by coordinate u (and coordinates z1 “ z4 “ 0), and on the overlap we
noted u “ v ´1 . We note that the modulus
4
ÿ
}z}2 “ |zi |2
i“1

on V Ă C4 becomes on Ṽ the function

}z}2 “ |vz3 |2 ` |z2 |2 ` |z3 |2 ` |vz2 |2 “ p1 ` |ζ|2 qp|σ|2 ` |s|2 q

over the U chart. In the V chart, then

}z}2 “ p1 ` |ξ|2 qp|σ̃|2 ` |s̃|2 q.

We recognize this as
}z}2 “ |pσ, ζq|2hF S
where hF S is the Fubini-Study metric on the Op´1q fibers.
In summary, the total space of the bundle Op´1q ‘ Op´1q Ñ P1 is given by two trivializations
tU, pσ, s, ζqu and tV, pσ̃, s̃, ξqu with ξ “ 1{ζ and s̃ “ ζs, σ̃ “ ζσ. The set }z} “ 0 is a P1 .
Definition 5.1. A p´1, ´1q curve C in a compact threefold X is defined by C – P1 and there
exists a neighborhood of C which is biholomorphic to a neighborhood of t}z} “ 0u in the total space
Op´1q ‘ Op´1q Ñ P1 .

5.2 Smoothing a nodal singularity

We return to the nodal singularity
4
ÿ
V “t pzi q2 “ 0u Ď C4 .
i“1

We see that V is a cone, since if z P V then so is λz for λ P C. We also have the radius function
ÿ
}z}2 “ |zi |2 .
i

73
On this cone, we have the holomorphic volume form
1
Ω“ dz2 ^ dz3 ^ dz4 , on tz1 ‰ 0u
z1
and the corresponding formula on the other open sets tzi ‰ 0u. These local expressions glue on V
to give a global holomorphic volume form. Next, we write

zk “ xk ` iyk .

The equation becomes

0 “ p|x|2 ´ |y|2 q ` 2ixx, yy.
Since }z}2 “ |x|2 ` |y|2 , this constraint is equivalent to

}z}2 }z}2
|x|2 “ , |y|2 “ , xx, yy “ 0.
2 2
For each fixed r ą 0, the cross-section where }z} “ r is seen from here to be a S 2 bundle over S 3 .
These are topologically trivial, so for each r ą 0, the cross section where }z} “ r is S 3 ˆ S 2 .
The smoothing of V is given by
4
ÿ
Vt “ t pzi q2 “ tu.
i“1
ř4 2
This is smooth because the only point where the derivative of i“1 pzi q is not surjective is at the
origin, which is not included in Vt .
After rotating coordinates z, we may suppose t ą 0. The defining equation becomes

t “ |x|2 ´ |y|2 , xx, yy “ 0.

Since }z}2 “ t ` 2|y|2 , we see that

}z}2 ě t.
The point t}z}2 “ 0u Ă V has been inflated to t}z}2 “ tu Ă Vt , which is

|x|2 “ t, y “ 0.

The space of x describes an S 3 of vanishing radius as t Ñ 0. This is sometimes called the vanishing
sphere.
Lastly, we note that Vt also admits a holomorphic volume form, given by local expressions such
as
1
Ωt “ dz2 ^ dz3 ^ dz4 , on tz1 ‰ 0u.
z1

74
5.3 Candelas-de la Ossa metrics
To summarize our discussion so far, we have described two ways to resolve the singular space
ÿ
V “ t zi2 “ 0u Ă C4

The first is by small blow-up, which replaced the origin by P1 “ S 2 . We call this the small
resolution
σ : Ṽ Ñ V.
The second is by smoothing, which replaced the origin by S 3 .
ÿ
Vt “ t zi2 “ tu Ă C4 .

Candelas-de la Ossa [2] constructed a sequence of metrics on both sides. We will discuss how one
side of the sequence sends the area of a holomorphic curve P1 to zero, and the other side sends the
area of a special Lagrangian 3-sphere to zero.

5.3.1 Metrics on the small resolution

We work on the total space Op´1q ‘ Op´1q Ñ P1 with two coordinate charts pλ, u, vq and pλ̃, ũ, ṽq
satisfying the change of coordinates

λ “ λ̃´1 , u “ λ̃ũ, v “ λ̃ṽ.

We have the well-defined function

τ “ p1 ` |λ|2 qp|u|2 ` |v|2 q,

which measures the distance to the zero section P1 (coordinate λ) along the Op´1q fibers (coordi-
nates u, v) using the Fubini-Study metric. For a ą 0, the Candelas-de la Ossa ansatz is

ωco,a “ iB B̄fa pτ q ` 4a2 ωF S .

for some function fa . Note that

ż
ωco,a “ 4a2 VolpP1 , ωF S q Ñ 0, a Ñ 0.
P1

We want to solve for fa such that these metrics are Ricci-flat. We expand the ansatz as
„ 
iB f τ B̄ log τ ` 4a2 ωF S
1

and using ωF S “ iB B̄ logp1 ` |λ|2 q,

„  „ 
2 ´1 1 1 2 2
ωco,a “ pf ` τ f qiBτ ^ B̄τ ` f τ iB B̄ logp|u| ` |v| q
„ 
1 2
` pf τ ` 4a qωF S . (5.1)

75
 3
To compute the Ricci curvature, we first compute ωco,a . We note that these 3 terms each square
to zero. For example
„ 2
2 2
iB B̄ logp|u| ` |v| q “ 0. (5.2)

Indeed,
"„  „ *
1
B B̄ logp|u|2 `|v|2 q “ p|u|2 `|v|2 qpdu^dū`dv ^dv̄q ´ pūdu` v̄dvq^pudū`vdv̄q .
p|u|2 ` |v|2 q2
Squaring this is of the form pα ´ βq2 where β 2 “ 0. A direct computation then shows α2 ´ 2αβ “
2p|u|2 ` |v|2 q2 ´ 2p|u|2 ` |v|2 q2 “ 0. This verifies (5.2).
3
Going back to (5.1), computing ωco,a is of the form pa ` b ` cq3 “ 6abc since a2 “ b2 “ c2 “ 0. We
have
3
ωco,a “ 6pf 2 ` τ ´1 f 1 qpf 1 τ qpf 1 τ ` 4a2 qiBτ ^ B̄τ ^ ωF S ^ iB B̄ logp|u|2 ` |v|2 q.
Next, we have
ωF S “ p1 ` |λ|2 q´2 idλ ^ dλ̄,
and
ˆ ˙ˆ ˙
2 2 2 2 2 2
Bτ ^ B̄τ “ p|u| ` |v| qλ̄dλ ` p1 ` |λ| qpūdu ` v̄dvq p|u| ` |v| qλdλ̄ ` p1 ` |λ| qpudū ` vdv̄q

and
ˆ ˙
1
B B̄ logp|u|2 ` |v|2 q “ p|u| 2
` |v|2
qpdu ^ dū ` dv ^ dv̄ ´ pūdu ` v̄dvq ^ pudū ` vdv̄q
p|u|2 ` |v|2 q2
Using this, direct calculation gives

iBτ ^ B̄τ ^ ωF S ^ iB B̄ logp|u|2 ` |v|2 q “ cpidλ ^ dλ̄q ^ pidu ^ dūq ^ pidv ^ dv̄q.

Therefore, writing ωco,a “ ipgco,a qj k̄ dz j ^ dz̄ k , then up to a constant we have

det gco,a “ pf 2 f 1 τ ` pf 1 q2 qpf 1 τ ` 4a2 q.

Side note: at λ “ 0, in pλ, u, vq coordinates the metric is

» fi
f 1 τ ` 4a2 0 0
— ffi
gco,a p0, u, vq “ — f ` |u|2 f 2
1
f 2 uv̄ ffi .
— ffi
0
– fl
0 f 2 vū f 1 ` |v|2 f 2

Let γpτ q “ τ f 1 . The determinant in terms of γ is

det gco,a “ γ 1 γpγ ` 4a2 qτ ´1

To find a Ricci-flat metric, from Rj k̄ “ ´Bj Bk̄ log det g we look for solutions to det gco,a “ const,
which is the equation (with convenient choice of constant)
2
γ 1 γpγ ` 4a2 q “ τ.
3

76
This has solution
γ 3 ` 6a2 γ 2 “ τ 2 .
This is
τ pfa1 q3 ` 6a2 pfa1 q2 “ 1.
From here, we note that if f1 is a solution for a “ 1, then fa “ a2 f1 pτ {a3 q is a solution for arbitrary
a. We discuss the solution f for a “ 1 and obtain the other fa by rescaling.
3 2 1
We look
? for a positive solution of τ y ` 6y “ 1 with y “ f pτ q. At τ “ 0, we choose the solution
y “ 1{ 6. It turns out that there is an explicit solution ypτ q ą 0 on τ ě 0 given by
„ 
1 4
y“ z ` ´ 2 , z “ 2´1{3 pτ 2 ` pτ 4 ´ 32τ 2 q1{2 ´ 16q1{3 . (5.3)
τ z

The function f1 is then defined on r0, 8q and given by

żτ
f1 pτ q “ yptqdt.
0

By construction, it solves
pf 2 f 1 τ ` pf 1 q2 qpf 1 τ ` 4q “ 1.
The form ωco,1 is ą 0, since it is given by (5.1) with f 1 ą 0 and f 2 τ ` f 1 ą 0.

Next, we show that fa Ñ f0 smoothly on compact sets away from u “ v “ 0. We study the
asymptotic behavior of f1 as τ Ñ 8. For τ ě K, then the Puiseux series of (5.3) turns out to
be
2 16
y “ τ ´1{3 ´ ` 4τ ´5{3 ´ τ ´7{3 ` . . .
τ 3
Integrating gives
f1 pτ q “ c0 τ 2{3 ` c1 log τ ` c2 τ ´2{3 ` c3 τ ´4{3 ` . . .
Therefore when τ ě Ka3 , then

fa pτ q “ c0 τ 2{3 ` c1 a2 log a´3 τ ` c2 a4 τ ´2{3 ` c3 a6 τ ´4{3 ` . . . .

Therefore fa Ñ c0 f0 as a Ñ 0 on compact sets disjoint from tτ “ 0u, where

f0 “ τ 2{3 , ωco,0 “ iB B̄τ 2{3 .

At the level of metrics, the expansion is

ωco,a ´ ωco,0 “ 4a2 ωF S ´ c1 a2 iB B̄ log r ´ c2 a4 iB B̄r´2 ´ c3 a6 iB B̄r´4 ` . . .

on tr ě K0 au, where r “ τ 1{3 . Therefore ωco,a Ñ ωco,0 as a Ñ 0 uniformly on compact sets which
are disjoint from the exceptional curve P1 “ tτ “ 0u.
Remark 5.2. Let us rewrite things slightly differently. Rescale the metrics so that c0 “ 1, let
r “ τ 1{3 so that r “ p1 ` |λ|2 q1{3 p|u|2 ` |v|2 q1{3 and

ωco,0 “ iB B̄r2 , ωco,a “ a2 Sa˚´1 ωco,1

77
where
SR pλ, u, vq “ pλ, R3{2 u, R3{2 vq
˚
satisfies SR ˚
r “ Rr and SR ωco,0 “ R2 ωco,0 . Let AR “ tR ď r ď 2Ru and note SR : A1 Ñ AR . We
pullback the compact set estimate

|iB B̄pf1 ´ f0 q|ω0 ď C, t1 ď r ď 2u

to obtain
|iB B̄Sρ˚ pf1 ´ f0 q|Sρ˚ ω0 ď C, tρ´1 ď r ď 2ρ´1 u

which, using |T |Sρ˚ ω0 “ |T |ρ2 ω0 “ ρ´2 |T |ω0 for T “ rTij s, implies

sup |ωco,1 ´ ωco,0 |ω0 ď Cρ2

ρ´1 ďrď2ρ´1

and
|ωco,1 ´ ω0 |ωco,0 ď Cr´2 .
We now pullback this estimate via Sa´1 .

|Sa˚´1 pωco,1 ´ ωco,0 q|S ˚ ωco,0 ď Ca2 r´2 .

a´1

Therefore
|Sa˚´1 ωco,1 ´ a´2 ωco,0 |ωco,0 ď Cr´2 ,
and
|ωco,a ´ ωco,0 |ωco,0 ď Cr´2 a2
which shows that ωco,a Ñ ωco,0 as a Ñ 0 on sets tr ą εu.

5.3.2 Metrics on the smoothings

In this section, we construct Kähler Ricci-flat metrics on Vt . For t P C, consider as before
4
ÿ
Vt “ t pwi q2 “ tu Ď C4
i“1

|wi |2 . Consider potentials of the form

ř
with radius function τ “

ϕpzq “ f pτ q,

for f to be determined. To find Kähler Ricci-flat metrics, we solve

piB B̄ϕq3 “ iΩ ^ Ω̄.

We compute
B̄ϕ “ f 1 B̄τ,
and
iB B̄ϕ “ f 1 iB B̄τ ` f 2 iBτ ^ B̄τ.

78
Its Monge-Ampère mass is
piB B̄ϕq3 “ pf 1 q3 piB B̄τ q3 ` 3pf 1 q2 f 2 piB B̄τ q2 ^ iBτ ^ B̄τ. (5.4)
We have ÿ
Bτ ^ B̄τ “ wj w̄i dwi ^ dw̄j , B B̄τ “ dwk ^ dw̄k .
k

We now compute in coordinates pw1 , w2 , w3 q on Vt over the chart tw4 ‰ 0u. On tw4 ‰ 0u, we
have
1
Ωt “ dw1 ^ dw2 ^ dw3 .
w4
Therefore
1
iΩt ^ Ω̄t “ pidw1 ^ dw̄1 q ^ pidw2 ^ dw̄2 q ^ pidw3 ^ dw̄3 q.
|w4 |2
We write
1
dµ “ pidw1 ^ dw̄1 q ^ pidw2 ^ dw̄2 q ^ pidw3 ^ dw̄3 q
|w4 |2
for simplicity. We need to equate this to piB B̄ϕq3 (5.4). For this we will use the defining relation of
Vt , which gives
4
ÿ 1
wk dwk “ 0, dw4 “ ´ pw1 dw1 ` w2 dw2 ` w3 dw3 q.
k“1
w 4

Using this, we can compute the first term in (5.4) to be

piB B̄τ q3 “ p3!qτ dµ.
Indeed,
piB B̄τ q3 “ pi3 3!qrdw11̄22̄33̄ ` dw11̄22̄44̄ ` dw11̄33̄44̄ ` dw22̄33̄44̄ s
„ 
3 1 2 2 2
“ pi 3!q 1 ` p|w3 | ` |w2 | ` |w1 | q dw11̄22̄33̄ . (5.5)
|w4 |2
Next we compute the second term in (5.4). We start with
pidw1 ^ dw̄1 q ^ pidw2 ^ dw̄2 q ^ iBτ ^ B̄τ
“ λ11̄ ^ λ22̄ ^ p|w3 |2 λ33̄ ` |w4 |2 λ44̄ ` w4 w̄3 λ34̄ ` w3 w̄4 λ43̄ q (5.6)
with the notation λij̄ “ idzi ^ dz̄j . By the defining relation
λ11̄ ^ λ22̄ ^ iBτ ^ B̄τ
w4 pw̄3 q2 w̄4 pw3 q2
“ p|w3 |2 ` |w3 |2 ´ ´ qλ11̄ ^ λ22̄ ^ λ33̄
w̄4 w4
ˆ ˙
“ 2 |w3 |2 |w4 |2 ´ Re pw4 w̄3 q2 dµ (5.7)

We also have
λ11̄ ^ λ44̄ ^ iBτ ^ B̄τ
“ λ11̄ ^ λ44̄ ^ p|w2 |2 λ22̄ ` |w3 |2 λ33̄ ` w3 w̄2 λ23̄ ` w2 w̄3 λ32̄ q (5.8)

79
which by the defining relation becomes

λ11̄ ^ λ44̄ ^ iBτ ^ B̄τ

|w2 |2 |w3 |2 |w2 |2 |w3 |2 pw3 w̄2 q2 pw̄3 w2 q2
“ p 2
` 2
´ 2
´ qλ11̄ ^ λ22̄ ^ λ33̄
|w4 | |w4 | |w4 | |w4 |2
“ 2p|w2 |2 |w3 |2 ´ Re pw3 w̄2 q2 qdµ. (5.9)

By symmetry,

piB B̄τ q2 ^ iBτ ^ B̄τ

ˆ
“ 4 |w1 |2 |w4 |2 ` |w2 |2 |w4 |2 ` |w3 |2 |w4 |2 ` |w2 |2 |w3 |2 ` |w3 |2 |w1 |2 ` |w2 |2 |w1 |2
˙
´Re tpw1 w̄2 q2 ` pw1 w̄3 q2 ` pw1 w̄4 q2 ` pw2 w̄3 q2 ` pw2 w̄4 q2 ` pw3 w̄4 q2 u dµ

Since |t|2 “ | wi2 |2 and τ 2 “ p |wi |2 q2 , we obtain

ř ř

piB B̄τ q2 ^ iBτ ^ B̄τ “ 2pτ 2 ´ |t|2 qdµ.

Substituting this into (5.4) gives

„ 
piB B̄ϕq3 “ 3! pf 1 q3 τ ` pf 1 q2 f 2 pτ 2 ´ |t|2 q dµ.

Thus to solve piB B̄ϕq3 “ iΩ ^ Ω, we need to solve

pf 1 q3 τ ` pf 1 q2 f 2 pτ 2 ´ |t|2 q “ 1{6,

for a function f pτ q, and where t is a fixed parameter. Here is the result. When t “ 0, the solution
is proportional to
f0 pτ q “ τ 2{3
and for given t ‰ 0, the solution is proportional to
ż cosh´1 pτ {|t|q
ft pτ q “ 2´1{3 |t|2{3 psinhp2λq ´ 2λq1{3 dλ.
0

Remark 5.3. Let’s work out how to find this solution when t “ 0. The ODE is

6τ rpf 1 q3 ` τ pf 1 q2 f 2 s “ 1.

Let τ “ s2 and γpsq “ s2 f 1 ps2 q. Then we have

pγ 3 q1 “ ps6 pf 1 q3 q1 “ 6s5 f 13 ` p3qs6 f 12 p2sqf 2 “ 6s2 rf 13 ` s2 f 12 f 2 ss3 .

The equation becomes

pγ 3 q1 “ s3 ,

80
which admits the solution γ 3 “ 14 s4 . Therefore

1 4
s6 rf 1 ps2 qs3 “ s
4
which for f 1 ą 0 is the ODE
f 1 pτ q “ c0 τ ´1{3 .
The solution is f pτ q “ c1 τ 2{3 .
Remark 5.4. Let’s compute the asymptotics of f1 . Its derivative is
ˆ ˙1{3
1
f11 pτ q “ sinhp2 cosh´1 pτ qq ´ 2 cosh´1 τ ? .
τ2 ´ 1
?
We have sinhp2 cosh´1 pτ qq “ 2τ pτ ´ 1q1{2 pτ ` 1q1{2 and cosh´1 τ “ logpτ ` τ 2 ´ 1q. Then as
τ Ñ 8, the series expansion turns out to be

f 1 pτ q “ 21{3 τ ´1{3 ` a1 τ ´7{3 log τ ` a2 τ ´7{3 ` Opτ ´10{3 q

Integrating gives
f pτ q “ c0 τ 2{3 ` a1 τ ´4{3 log τ ` a2 τ ´4{3 ` Opτ ´7{3 q.
Therefore, if we let r “ τ 1{3 , then

ϕ1 “ c0 r2 ` a1 r´4 log r ` a2 r´4 ` Opr´7 q.

Remark 5.5. The metrics ωco,1 are asymptotically conical. We now describe what this means (see
[6] for more on asymptotically conical Kähler-Ricci flat metrics). Let the Kähler-Ricci flat metrics
be denoted
ωco,0 “ iB B̄ϕ0 , ωco,1 “ iB B̄ϕ1 .
We consider the map Φ : V X t|z| ą Ru Ñ V1 given by
x̄
Φpxq “ x ` .
2|x|2

We will estimate the order of Φ˚ ωco,1 ´ ωco,0 . By the asymptotics of ϕ1 and r “ }x}2{3 ,

pΦ˚ ϕ1 qpxq “ |Φpxq|4{3 ` Op|Φpxq|´p8{3q log |Φpxq|q

“ |x|4{3 ` Op|x|p4{3q´1 |x|´1 q ` Op|Φpxq|´p8{3q log |Φpxq|q
“ r2 ` Opr´1 q (5.10)

Therefore
Φ˚ ϕ1 ´ ϕ0 “ Opr´1 q.
Next, we start from
sup |Φ˚ ωco,1 ´ ωco,0 |ωco,0 ď C,
1ďrď2

81
and pull this back via Sλ : V Ñ V , Sλ pzq “ λ3{2 z to get

sup |Sλ˚ iB B̄pΦ˚ ϕ1 ´ ϕ0 q|S ˚ ωco,0 ď C.

λ
λ´1 ďrď2λ´1

The definition of Sλ is such that Sλ˚ r “ λr and Sλ˚ ωco,0 “ λ2 ωco,0 . Since

|Sλ˚ iB B̄pΦ˚ ϕ1 ´ ϕ0 q|S ˚ ωco,0 “ λ´2 λ´1 |iB B̄pΦ˚ ϕ1 ´ ϕ0 q|ωco,0

λ

we conclude
|Φ˚ ωco,1 ´ ωco,0 |ωco,0 ď Cr´3 . (5.11)
Metrics satisfying estimates of this form are said to be asymptotically conical (with rate 3).
Remark 5.6. Next, we notice that Sλ pzq “ λ3{2 z implies

St1{3 : V1 Ñ Vt .

We assume t ą 0. We have that

ωco,t “ t2{3 St˚´1{3 ωco,1
is a natural sequence of metrics, and we can check that it does agree with the sequence of explicit
metrics obtained earlier. The reason for the t2{3 out-front is that far out on the cone r " R, then
ωco,1 „ iB B̄r2 and St˚´1{3 iB B̄r2 “ t´2{3 , so that these metrics ωco,t are asymptotic to the cone iB B̄r2 .
More precisely, let
Φt “ St1{3 ˝ Φ ˝ St´1{3 : V Ñ Vt ,
then
|Φ˚t ωco,t ´ ωco,0 |ωco,0 ď C|t|r´3 (5.12)
Indeed, pulling back (5.11) gives

|St˚´1{3 Φ˚ ωco,1 ´ St˚´1{3 ωco,0 |S ˚ ωco,0 ď C|t|r´3

t´1{3

which combined with

|t´2{3 Φ˚t ωco,t ´ t´2{3 ωco,0 |t´2{3 ωco,0 “ |Φ˚t ωco,t ´ ωco,0 |ωco,0

gives the estimate (5.12). This estimate can be interpreted as ωco,t Ñ ωco,0 on compact sets away
from the cone singularity as t Ñ 0.

5.3.3 The cone metric

We return to the cone
4
ÿ
V “t zi2 “ 0u Ď C4
i“1
1
with holomorphic volume form Ω “ z1 dz2 ^ dz3 ^ dz4 . We will now equip this with a natural
Calabi-Yau metric.
Both sequence of metrics gco,a and gco,t agree at gco,0 . Recall that Op´1q ‘ Op´1q Ñ P1 with zero
section removed can be identified with V and the function τ in both previous sections agrees in this

82
identification. So we will work on V , and the Candelas-de la Ossa cone metric derived in the two
previous sections is, up to a constant,
ωco,0 “ iB B̄r2
where r “ }z}2{3 .

3
Here is the reason why the power 2/3 is natural. We are looking to solve ωco,0 “ iΩ ^ Ω̄. We can
rescale z ÞÑ λz on both sides, which gives

pλ4{3 q3 “ pλ´1 λ3 qpλ´1 λ3 q

which is consistent.
Since the cone radius is r, the natural scaling on the cone is

t ¨ pz1 , z2 , z3 , z4 q “ pt3{2 z1 , t3{2 z2 , t3{2 z3 , t3{2 z4 q

since rpt ¨ zq “ tr.

We have explained why this metric is Kähler-Ricci flat. We will now discuss why this is a cone
metric. Note that V is a cone, since if z P V then λz P V for any λ P C. In general, a cone metric
is a metric on p0, 8q ˆ Σ of the form

g “ dr b dr ` r2 gΣ . (5.13)

This type of metric is very useful in Riemann geometry. To see why, convert the Euclidean metric
gEuc on Rn to polar coordinates pr, θi q. It will be of the form (5.13) with gΣ the metric on the
sphere S n´1 . Riemannian geometry with metrics of the form (5.13) behaves like polar coordinates:
for example, the kernel of the Laplacian ∆g can be decomposed by separation of variables.
Here is why ωco,0 can be put in this form. We insert a factor of p1{2q in the definition of ωco “ 12 iB B̄r2
for convenience. Compute

ωco “ iBpr2 B̄ log rq “ 2iBr ^ B̄r ` r2 iB B̄ log r.

83
We can write this as ˆ ˙
2 i
ωco “ ´dr ^ Jdr ` r ´ dJd log r
2
where pJβqpXq “ βpJXq. This can be seen by writing dr “ Br ` B̄r and Jdr “ iBr ´ iB̄r.
Let η “ ´Jd log r. To obtain the metric, we use gco pX, Y q “ ωco pX, JY q.

gco “ dr b dr ` r2 η b η ` r2 dηp¨, J¨q.

It remains to show that

η “ ηi pθqdθi , dηp¨, J¨q “ αij pθqdθi b dθj (5.14)
where the implicit function theorem gives local coordinates pr, θi q near a given point.
‚ First, we note the following nowhere vanishing holomorphic vector field on V :
4 „ 
ÿ B B B
ξ“3 zi “ r ´ iJ r ,
i“1
Bzi Br Br

with rBr “ 32 xi BxB i ` 23 yi ByB i . This vector field generates the t¨ action on V in the sense
ˇ
d ˇˇ
ξpf q “ 2 f pt ¨ z, z̄q.
dt ˇt“1

From here, it follows that

ξplog rq “ 1,
so taking real and imaginary parts gives
„ 
B B
r plog rq “ 1, J r plog rq “ 0.
Br Br

In R2 with polar coordinates reiθ , we can think of JprBr q “ rBθ . This discussion implies ηpBr q “
´JpBr q log r “ 0 and so
η “ ηi pr, θqdθi .
The next step is to show that ηi does not depend on r, and for this we will use the Lie deriva-
tive.
‚ Next, we note
Lξ J “ 0.
The Lie derivative is
Lξ pJXq “ pLξ JqpXq ` JLξ pXq
and Lξ pXq “ rξ, Xs for vector fields X. So this amounts to showing

rξ, JBα s “ Jrξ, Bα s, (5.15)

B B
for Bα “ Bzk or Bα “ Bz̄k . Since ξ “ 3zi BzB i , this is readily verified directly.

84
‚ We now show that
LrBr η “ 0.
By Cartan’s formula, LrBr “ dιrBr η ` ιrBr dη. Since ιrBr η “ 0, we need to show
dηprBr , ¨q “ 0. (5.16)
Indeed, the invariant formula for dη gives
dηprBr , Xq “ rBr ηpXq ´ XηprBr q ´ ηprrBr , Xsq
“ ´rBr JX log r ` JrrBr , Xs log r
“ ´rBr JX log r ` rrBr , JXs log r
“ ´JXrBr log r “ 0. (5.17)
Here we used the real part of rξ, JXs “ Jrξ, Xs (5.15).
‚ Altogether, since ˇ
d ˇˇ
0 “ LrBr η “ ηi ptr, θqdθi
dt ˇt“1
it follows that in a local chart we can write η “ ηi pθqdθi and ηi does not depend on r. In other
words, η “ p˚ ηΣ where p : V Ñ Σ is ppr, θq “ θ. Similarly, (5.16) implies
dηp¨, J¨q “ αij pr, θqdθi b dθj ,
and LrBr dη “ 0 and LrBr J “ 0 implies that αij pθq does not dependent on r. We have proved (5.14),
and obtain the cone formula g “ dr2 ` r2 gΣ .

5.4 Special Lagrangian cycles

In this section, we show that the vanishing cycle is special Lagrangian with respect to the Candelas-
de la Ossa metric.

5.4.1 Definition of special Lagrangian cycles

We start with a review of calibrated cycles [15]. We say that ϕ P Λk pM q on pM, gq is a calibration
if for all k-dimensional submanifolds L then
ˇ ˇ
ˇϕ|L ˇ ď 1
g

pointwise. The norm is defined as: write ϕ|L “ f pxqdvolL and then take |f pxq|. Here dvolL is
the volume form of the induced metric g|L . A submanifold L is calibrated with respect to ϕ if
ϕ|L “ dvolL .

Let rLs P Hk pM, Zq be a fixed homology class. Any other representative can be written as L1 “
L ´ BΣ. We consider the functional on rLs defined by
ż ż
EpL1 q “ dvolL1 ` dϕ, L1 “ L ´ BΣ. (5.18)
L1 Σ

In the standard definition of a calibration, we require dϕ “ 0 and the bulk term does not appear so
that this is the area functional. The more general definition is here for potential future application
to non-Kähler geometry.

85
Proposition 5.7. Let L be a calibrated submanifold with respect to ϕ. Then L minimizes the
functional EpLq in a given homology class.

Proof. First, we note ż ż

EpLq “ dvolL “ ϕ
L L

When dϕ “ 0, this is the topological number rϕs ¨ rLs. We will show that this is the lower bound
of the functional. By Stokes’s theorem,
ż „ż ż 
1
EpL q “ dvolL `
1 ϕ´ ϕ .
L1 L L1

Using the calibration property, ϕ|L1 ď |ϕ|g dvol|L1 ď dvol|L1 , so

ż „ż ż  ż
1
EpL q ě ϕ` ϕ´ ϕ “ ϕ.
L1 L L1 L

Let Ω be a holomorphic volume form. Let ω be a hermitian metric, conformally rescaled so that
|Ω|ω “ 23{2 . A special Lagrangian cycle is a submanifold calibrated with respect to Re Ω. The
calibration argument implies that special Lagrangian cycles minimize the area functional
ż
EpLq “ dvolL
L

in a given homology class rLs. We now show that Re Ω is indeed a calibration.

Lemma 5.8. Re Ω is a calibration. Furthermore, if L is a calibrated submanifold so that Re Ω|L “
dvolL , then ω|L “ 0.

Proof. We closely follow Harvey-Lawson’s proof [15]. Let L Ď M . Suppose v1 , v2 , v3 are orthonor-
mal vectors spanning Tp L. Harvey-Lawson’s identity is

|Ωpv1 , v2 , v3 q|2 “ |v1 ^ v2 ^ v3 ^ Jv1 ^ Jv2 ^ Jv3 |. (5.19)

ˇ µ ˇ
(The norm of a top form µ is ˇ dvol ˇ.) Assuming this, then

|Ωpv1 , v2 , v3 q|2 ď |v1 ||Jv1 | ¨ ¨ ¨ |v3 ||Jv3 | “ 1.

In fact, equality in Hadamard’s inequality is achieved if and only if the vectors are orthogonal. This
implies xvi , Jvk yg “ 0, which translates to ωpvi , vk q “ 0 and so ω|L “ 0. The inequality above is

|Re Ωpv1 , v2 , v3 q|2 ` |Im Ωpv1 , v2 , v3 q|2 ď 1

Therefore, since the vi are orthonormal, then |Re Ω|L |g ď 1 and Re Ω is a calibration. If equality
Re Ω|L “ dvolL holds, then ω|L “ 0 and Im Ω|L “ 0.

86
We now prove Harvey-Lawson’s identity (5.19). We start with the left-hand side. Let e1 , e2 , e3 , Je1 , Je2 , Je3
be an orthonormal basis for Tp M . Since Ω|p is an element of the 1 dimensional vector space Λ3,0 p
it can be written Ω|p “ f ppqε1 ^ ε2 ^ ε3 with
1
εk “ pek ´ iJek q.
2
To find f ppq P C, we take the norm and use that |Ω|ω “ c. For suitable normalization (c “ 23{2 as
will be computed below), we conclude Ω|p “ eiθppq ε1 ^ ε2 ^ ε3 .
Next, we expand vi in the basis εk , εk
¯
vk “ A` k ε` ` A` k ε` .
¯
Since vk “ vk , using uniqueness of expansion of a basis we obtain that A` k “ A` k . Since Ω|p “
eiθ ε1 ^ ε2 ^ ε3 , we have

Ωpv1 , v2 , v3 q “ ΩpA` 1 ε` , Am 2 εm , An 3 εn q “ eiθ det A.

Next, we compute the right-hand side of (5.19). Let e4 “ Je1 , e5 “ Je2 , e6 “ Je3 so that tei u is
an oriented basis. Define a linear map M by its action on this basis

M pei q :“ vi , M pJei q :“ Jvi .

Then
|v1 ^ Jv1 ^ . . . v3 ^ Jv3 | “ |M pe1 q ^ M pe2 q ^ ¨ ¨ ¨ ^ M pe6 q| “ | det M |
To compute det M , we compute it in the basis εi , εi . We compute
1 1 i
M pεk q “ pM pek q ´ iM pJek qq “ vk ´ Jvk .
2 2 2
Using our earlier matrix Ai k and Jεk “ iεk , Jεk “ ´iεk , this is
1 n i
M pεk q “ pA k εn ` An̄ k εn q ´ piAn k εn ´ iAn̄ k εn q “ An k εn .
2 2
Similarly M pεk q “ Āεk , and so in this basis
» fi
A 0
M “– fl .
0 Ā

Therefore det M “ | det A|2 , which proves the identity.

A useful formula for calibrated cycles are the special Lagrangian equations. The submanifold L is
special Lagrangian if it solves the equations

ω|L “ 0, Im Ω|L “ 0. (5.20)

These equations imply that L is a calibrated cycle.

87
Proposition 5.9. If L Ď pM, ω, Ωq solves the special Lagrangian equations (5.20), then L is a
calibrated cycle. This means we can orient L such that Re Ω|L “ dvolL and L minimizes the area
functional in its homology class.

Proof. The key identity is that if ω|L “ 0, then

eiθ |Ω|ω
Ω|L “ dvol. (5.21)
23{2
Therefore the conditions ω|L “ 0, |Ω|ω “ 23{2 and ImΩ|L “ 0 imply Re Ω|L “ ˘dvol|L .
To prove (5.21), fix a point p P Tp L and an orthonormal basis e1 , e2 , e3 of Tp L with dual basis ek .
Since ωpei , ej q “ gpei , Jej q, the condition ω|L “ 0 implies that

te1 , e2 , e3 , Je1 , Je2 , Je3 u

is an orthonormal basis of Tp M . Let tek u the dual basis, and

εk “ ek ` iJek . (5.22)

ek ^ Jek , which in the basis

ř
One can check by the definition ωpei , ej q “ gpei , Jej q that ω|p “ k
(5.22) is
iÿ k
ω|p “ ε ^ ε̄k .
2 k

The p3, 0q form is

Ω|p “ f ppqε1 ^ ε2 ^ ε3
for f ppq P C, since εk span Λ1,0 1 2 3
p pM q and so any p3, 0q form is a multiple of ε ^ ε ^ ε . To identify
f ppq, we use the formula for the norm

ω3
iΩ ^ Ω “ |Ω|2ω .
3!
We see that
|Ω|2ω 11̄ 22̄ 33̄
i|f |2 ε1231̄2̄3̄ “ iε iε iε .
23
Therefore |f |2 “ |Ω|2 {23 and
eiθppq |Ω|ω 1
Ω|p “ ε ^ ε2 ^ ε3
23{2
Thus
eiθ |Ω|ω 1
Ω|L “ e ^ e2 ^ e3 ,
23{2
which is (5.21).

88
5.4.2 Examples on the smoothing
We give two examples of special Lagrangian cycles on

Vt “ tz12 ` z22 ` z32 ` z42 “ tu Ď C4

for t ą 0, with 3-form

1
Ωt “ dz1 ^ dz2 ^ dz3 .
z4
The Calabi-Yau metric on Vt is the Candelas-de la Ossa metric ωco,t .
‚ Vanishing cycle L “ S 3 . For this, we use the Candelas-de la Ossa metric
ÿ
ωco,t “ iB B̄ft pτ q, τ “ |zi |2

A special Lagrangian 3-sphere is given by

L “ t|z1 |2 ` |z2 |2 ` |z3 |2 ` |z4 |2 “ tu Ď Vt .

If zk “ xk ` iyk , the constraint for Vt implies t “ |x|2 ´ |y|2 , xx, yy “ 0, and the constraint for L is
|x|2 ` |y|2 “ t. Therefore y “ 0 and |x|2 “ t, so L is topologically a 3-sphere. Since y “ 0, we see
that pIm Ωt q|L “ 0. Since τ is constant, we see that ωco,t |L “ 0.
‚ Special Lagrangian discs. Here we will use the metric ωEuc “ k idz k ^ dz̄ k restricted to Vt . It
ř
is not Calabi-Yau, so this example involves a more general setup. Consider the two discs L` , L´
given by
L˘ “ tx21 ` x22 ` x23 ď tu Ď Vt X ty “ 0u.
with b
z4 “ ˘ t ´ x21 ´ x22 ´ x23 .

These share the same boundary BL which is a 2-cycle S 2 which lies on the holomorphic surface
z4 “ 0. Since y “ 0, we have pIm Ωt q|L “ 0 and ωEuc |L “ 0.

89
6 Conifold Transitions: Global Geometry
6.1 Overview
Let X̂ be a compact Calabi-Yau threefold. A conifold transition X̂ Ñ X0 ù Xt is defined as
follows:
Step 1: Find holomorphic curves Ci Ă X̂ with a neighborhood biholomorphic to the open set
t}z} ă 1u in the total space Op´1q ‘ Op´1q Ñ P1 , with Ci “ t}z} “ 0u. Such curves Ci are called
p´1, ´1q curves.
Step 2: Contract each Ci to a point pi and obtain singular
ř4 space X0 . A change of variables shows
that each pi admits a neighborhood biholomorphic to t i“1 zi2 “ 0u Ď C4 (§5.1.2).
Step 3: Realize X0 as the central fiber of a deformationřπ : X Ñ ∆ such that Xt is smooth for
t ‰ 0 and locally near pi we see the local smoothing t zi2 “ tu. For this to exist, we need to
satisfy Friedman’s condition on the initial curves, and the global smoothing result of Friedman-
Tian-Kawamata is stated below. The smoothing has the effect of replacing each pi P X0 with
3-sphere S 3 (§5.2).
Theorem 6.1. [9, 27, 18] Let X̂ be a Calabi-Yau threefold with p´1, ´1q curves Ci Ă X̂. Let
µ : X̂ Ñ X0 be the blow-down map sending Ci to nodal singularities pi P X0 . If the curves are
linearly dependent in homology, so that
ÿ
λi rCi s “ 0 P H 4 pX̂, Cq
i

with λi ‰ 0, then X0 admits a smoothing Xt . This means there is a complex space X with a proper
flat map π : X Ñ ∆ such that π ´1 p0q “ X0 and π ´1 ptq “ Xt is a smooth complex manifold for
t ‰ 0.
ř
Let Ci be an initial configuration of p´1, ´1q curves on X̂ satisfying Friedman’s condition λi rCi s “
0. The theorem above guarantees the existence of a conifold transition X̂ Ñ X0 ù Xt . Here is a
summary of known results on the geometry and topology of such a transition:
‚ Xt is a compact complex manifold with trivial canonical bundle. [9]
‚ It is conjectured [1] that all Calabi-Yau threefolds can be connected to each other by conifold
transitions.
‚ However, a conifold transition may produce an Xt which is non-Kähler. (Examples at the end
of §6.2) This suggests that these limiting non-Kähler objects should be included in the web of
Calabi-Yau threefolds.
‚ Though non-Kähler in general, Xt is expected to satisfy the ddbar lemma, but as far as I can
tell, this is still unknown in full generality [10].
‚ The topological change is
N “ k ` c,
where N is the number of nodes, k is the decrease of b2 , and 2c is the increase of b3 . (§6.2)
‚ [11] There is a sequence of metrics pX̂, ga q with dωa2 “ 0 such that ga Ñ g0 uniformly on compact
sets for a limiting metric g0 on X0 locally modeled near the singularities by a scaling of gco,0 ,

90
and
VolpCi , ga q Ñ 0, a Ñ 0.
On the smoothing side, for small t there is a sequence of metrics pXt , gt q with dωt2 “ 0 and gt Ñ g0
uniformly on compact sets with

VolpLi,t , gt q Ñ 0, t Ñ 0,

where Lt,i “ t}z}2 “ |t|u Ď Vt Ď Xt are the vanishing 3-spheres. This result is due to Fu-Li-Yau
[11]. Near the singularities, the FLY metrics are modeled by the Candelas-de la Ossa metrics:

|gt ´ λi gco,t |gco,t ď C|t|2{3 , on X X tr ă 1u.

Here r : X Ñ p0, 8q is a function which agrees with r “ }z}2{3 near the singular points and
r´1 p0q “ SingpX0 q. Each ith component of Xt X tr ă 1u is biholomorphic to Vt X tr ă 1u, and the
metric gt is close to a λi -scaled version of gco,t for λi ą 0. For higher order derivatives, the estimate
is
|∇kgco,t pgt ´ λgco,t q|gco,t ď Ck |t|2{3 r´k , on X X tr ă 1u.
Let us give a few more details of the Fu-Li-Yau construction. There are two steps:

(A) The construction of a balanced metric ω0 on X0 satisfying:

- On X0 ztr ă 1u, we have ω0 “ ωCY , where ωCY is a Calabi-Yau metric on X̂.
- On X0 X tr ă εu, we have ω0 “ Rωco,0 . Here ε ą 0 and R ą 1 are chosen parameters.
- On X0 X tε ď r ď 1u, we have ω02 “ iB B̄β for some β P Λ1,1 pX̂, Rq.
Here we go freely between X0 zSingpX0 q and X̂z Y Ci since these are biholomorphic.

(B) The construction of a balanced metric ωt on Xt :

This starts by constructing an approximate metric ω̂t :
- On Xt X tr ă εu, we have ωt “ Rωco,t .
- On compact subsets K Ă X zSingpX q, we have that ω̂t converges to ω0 smoothly uniformly as
t Ñ 0.
This metric ω̂t is not balanced, and is globally corrected by

ωt2 “ ω̂t2 ` θt ` θ̄t .

The correction θ comes from solving Et pγt q “ B̄ ω̂t2 (Et is the Kodaira-Spencer operator) and
θt “ B B̄ : B : γt , ´B̄θt “ Et pγt q. Fu-Li-Yau show that the correction is θt is small: it satisfies |θt |ω̂t ď
C|t|2{3 .
‚ [5] The vanishing cycles Lt,i Ď Xt can be represented by special Lagrangian 3-spheres with respect
to the global geometry gt . Thus from the point of view of submanifold geometry, the transition
exchanges holomorphic 2-cycles with special Lagrangian 3-cycles.
‚ [4] The compact manifolds Xt for small t admits a pair of metrics pgt , ht q solving

g j k̄ Hpj k̄ “ 0, g j k̄ F p qj k̄ “ 0

91
where H is the 3-form field strength H “ ipB ´ B̄qω and F is the 2-form field strength F “
B̄ph´1 Bhq. In particular, Xt admits balanced metrics and has stable tangent bundle. (The equation
g j k̄ Hpj k̄ “ 0 is equivalent to dω 2 “ 0.) Note that Calabi-Yau metrics gCY solve these equations for
h “ g “ gCY , so these are generalizations of Calabi-Yau metrics on the non-Kähler spaces reached
by conifold transitions.
‚ It is conjectured by S.-T. Yau that the pair pg, hq can be further deformed to solve the Strominger
system:
dp|Ω|ω ω 2 q “ 0, g j k̄ F p qj k̄ “ 0
iB B̄ω “ α1 pTr Rω ^ Rω ´ Tr Fh ^ Fh q. (6.1)
2
The equation dp|Ω|ω ω q “ 0 can be solved by conformally changing the Fu-Li-Yau metric on Xt ,
but whether (6.1) is solvable through conifold transitions is unknown.

6.2 Topological change

In this section, we follow the exposition given in Rossi’s survey [23].
Let X̂ Ñ X ù X be a conifold transition with N nodes. We denote the N holomorphic curves
which are being contracted by Ci Ă X̂, and rCi s P H2 pX̂, Rq their homology class. A transition
decreases b2 and increases b3 , and we define the jumps in homology rank by
b2 pXt q “ b2 pX̂q ´ k, b3 pXt q “ b3 pX̂q ` 2c.
With this notation, the fundamental identity for topology change to be explained in this section
is:
N “ k ` c. (6.2)
and k is also equal to
k “ dim subspace in H2 pX̂, Rq generated by rCi s. (6.3)
As a consequence of N “ k ` c, we see that if a transition exists then there must be a linear
dependence relation between the N curves rCi s. Viewed another way, we can deduce the change in
b3 by the initial position of the 2-cycles Ci by c “ N ´ k. Also, c ě 1 is the number of independent
vanishing 3-cycles in Xt .
We will need the following properties:
‚ There is a neighborhood Ûi of each p´1, ´1q-curve Ci in X̂ which is diffeomorphic to S 2 ˆ B 4 .
This means that though the bundle Op´1q ‘ Op´1q Ñ P1 is not holomorphically trivial, it is in
fact a trivial rank 4 real vector bundle. For the diffeomorphism to S 2 ˆ B 4 , see [23].
‚ There is a neighborhood Ut,i of each vanishing cycle Li in Xt which is diffeomorphic to S 3 ˆ B 3 .
We discussed earlier how the equations of Vt can be written as t “ |x|2 ´ |y|2 , xx, yy “ 0. Let
p “ x{pt ` }y}2 q1{2 , so that pp, xq P R4 ˆ R4 satisfy }p} “ 1, xp, yy “ 0. This is diffeomorphic to
the total space of T S 3 , which is known to be a trivial bundle over S 3 . (A construction of a global
basis of sections of T S 3 can be constructed using quaternions.)
‚ Let Ûi Ď Û i be two nested neighborhoods of a p´1, ´1q-curve in X̂ as above. Then the annulus
 “ Ûi X Û i is homotopic to S 2 ˆS 3 . This is because V̂ zÛi is diffeomorphic to the cone V0 zt}z} ă 1u,
whose cross-section is S 3 ˆ S 2 . Similarly, At “ Ui,t X U i,t Ď Xt is homotopic to S 2 ˆ S 3 .

92
Let Û “ YN 2 4
i“1 Ûi Ď X̂ be an open set containing all curves Ci with Û “ \S ˆ B , and Ut “
Yi“1 Ut,i Ď Xt be an open set containing all vanishing cycles with Ut “ \S ˆ B 3 . We now
N 3

compute the topology of Xt based on the knowledge of X̂. We start with:

Lemma 6.2. b1 pX̂q “ b1 pXt q.

Proof. Recall the Meyer-Vietoris sequence: if X “ U Y V then

¨ ¨ ¨ Ñ Hi pU X V q Ñ Hi pU q ‘ Hi pV q Ñ Hi pXq Ñ Hi´1 pU X V q Ñ Hi´1 pU q ‘ Hi´1 pV q . . .

We apply this to X̂ “ pX̂zÛ q Y Û , so that

H1 p\S 3 ˆ S 2 q Ñ H1 pX̂zÛ q ‘ H1 p\S 2 q Ñ H1 pX̂q Ñ H0 p\S 3 ˆ S 2 q Ñ H0 pX̂zÛ q ‘ H0 p\S 2 q

while on X, we have

H1 p\S 3 ˆ S 2 q Ñ H1 pXzU q ‘ H1 p\tptuq Ñ H1 pXq Ñ H0 p\S 3 ˆ S 2 q Ñ H0 pXzU q ‘ H0 p\tptuq

and on Xt we have

H1 p\S 3 ˆ S 2 q Ñ H1 pXt zUt q ‘ H1 p\S 3 q Ñ H1 pXt q Ñ H0 p\S 3 ˆ S 2 q Ñ H0 pXt zUt q ‘ H0 p\S 3 q

To connect these diagrams, we use the contraction maps X̂ Ñ X and Xt Ñ X, and that X̂zÛ and
Xt zUt are both diffeomorphic to XzU (as they are a smooth family of complex structures). The
5-lemma implies that H1 pX̂q “ H1 pXq “ H1 pXt q. Recall the 5-lemma: given a diagram between
two exact sequence
A /B /C /D /E

f g h i j
    
A1 / B1 / C1 / D1 / E1
If g, i are isomorphisms, f surjective and j injective, then h is an isomorphism.

Since we are assuming that X̂ is a simply connected Calabi-Yau threefold, it follows that b1 pXt q “ 0
on the other side of a conifold transition.
Lemma 6.3. b2 pX̂q “ b2 pXt q ´ κ, where κ is the dimension of the subspace in H2 pX̂, Rq spanned
by the rCi s.

Proof. Let C “ YCi Ă X̂ be degenerating 2-cycles and L “ YLi Ă Xt be the vanishing 3-cycles.
The long exact sequence for relative homology gives
ι˚ ι˚
¨ ¨ ¨ Ñ H2 pCqÑH2 pX̂q Ñ H2 pX̂, Cq Ñ H1 pCqÑ ¨ ¨ ¨ (6.4)

Recall that relative homology HpX̂, Cq means: homology of space where we identify all points in
C to be a single point (or cycles α P Zn pX̂q such that Bα P Zn´1 pCq). But we will not need to have
intuition for this definition here; the only properties we will use is the exact sequence above, and
the Lefschetz duality
Hi pX̂, Cq – Hcn´i pX̂zCq

93
which holds for topological manifolds and implies in our setup that

Hi pX̂, Cq – Hi pXt , Lq, (6.5)

since X̂zC – Xt zL. The long exact sequence for relative homology gives
N
ğ ι˚ ϕ
rCi sÑH2 pX̂qÑH2 pX̂, Cq Ñ 0.
i“1

By the rank-nullity theorem and (6.5),

dim H2 pX̂q “ dim H2 pX̂, Cq ` dim ker ϕ

“ dim H2 pXt , Lq ` dim im ι˚ . (6.6)

If κ is the dimension of the subspace in H2 pX̂q generated by rCi s, we conclude

κ “ b2 pX̂q ´ dim H2 pXt , Lq. (6.7)

Since H2 pXt , Lq – H2 pXt q, we obtain

κ “ b2 pX̂q ´ b2 pXt q. (6.8)

This can be seen by the exact sequence

H2 pLq Ñ H2 pXt q Ñ H2 pXt , Lq Ñ H1 pLq

which for L “ \N 3
i“1 S implies H2 pXt q – H2 pXt , Lq.

Lemma 6.4. N “ k ` c
ř6 k
Proof. Recall the Euler characteristic χ “ k“1 p´1q bk . The excision property for the Euler
characteristic is: for U Ď X open,

χpXzU q “ χpXq ´ χpU q.

In our setup, this gives

χpX̂zÛ q “ χpX̂q ´ N χpS 2 ˆ B 4 q
and
χpXt zUt q “ χpXt q ´ N χpS 3 ˆ B 3 q.
Since X̂zÛ and Xt zUt are diffeomorphic, and χpS 2 q “ 2 and χpS 3 q “ 0, plus χpS 2 ˆ B 4 q “ χpS 2 q
by deformation retraction, it follows that

χpX̂q ´ 2N “ χpXt q.

We proved that X̂ and Xt have b1 “ 0, so χ “ 2 ` 2b2 ´ b3 and we obtain

1
N “ pb2 pX̂q ´ b2 pXt qq ` pb3 pXt qq ´ b3 pX̂qq.
2

94
Remark 6.5. If the initial manifold has b2 pX̂q “ 1 (if for example X̂ is a quintic threefold in
P4 ), then two homologically linearly dependent curves C1 , C2 produce a conifold transition where
the resulting manifold has b2 pXt q “ 0. A compact Kähler manifold cannot have b2 “ 0, since ω
represents a non-zero cohomology class in H 2 . We see that conifold transitions possibly take us out
of Kähler geometry.
Example 6.6. Here is another example of a non-Kähler transition from [2]. Start with a small
resolution X̂ of a singular quintic P “ Z3 GpZ0 , . . . , Z4 q ` Z4 HpZ0 , . . . , Z4 q “ 0. This has h1,1 “ 2,
and there are 16 p´1, ´1q curves Ci Ď X̂ from the small resolution, but they are all linearly
dependent. Suppose they are generated by C1 . Take 2 curves C1 , C2 ; they satisfy Friedman’s
condition so we can produce a transition X̂ Ñ X ù Xt . Now the other 14 curves are trivial in
homology rCi s “ rBDi s. Furthermore by [Remark 3.2.8, Lemma 3.3.1] in McDuff-Salamon [22], the
p´1, ´1q curves deform along the family so that Ci,t are holomorphic curves. If Xt admits a Kähler
metric ωt , then ż ż
0ă ωt “ dωt “ 0,
Ci,t Di,t

which is a contradiction. So Xt is non-Kähler, but b1 pXt q “ 1.

95
References
[1] P. Candelas, P. Green, and T. Hubsch, Rolling among Calabi-Yau vacua, Nuclear Physics B
330(1) (1990), 49-102.
[2] P. Candelas and X. de la Ossa, Comments on conifolds, Nuclear Physics B 342, no. 1 (1990),
246-268.
[3] P. Candelas, X. de la Ossa, P. Green, and L. Parkes, A pair of Calabi-Yau manifolds as an
exactly soluble superconformal theory, Nuclear Physics B, Vol 359, Issue 1 (1991), 21-74.
[4] T.C. Collins, S. Picard and S.-T. Yau, Stability of the tangent bundle through conifold transi-
tions, arXiv:2102.11170.
[5] T.C. Collins, S. Gukov, S. Picard and S.-T. Yau, Special Lagrangian cycles and Calabi-Yau
transitions, arXiv:2111.10355.
[6] R. Conlon and H.-J. Hein, Asymptotically conical Calabi-Yau manifolds, I, Duke Math. J. 162
(2013), 2855-2902.
[7] J.P. Demailly, Complex Analytic and Differential Geometry. Book available on the author’s
website.
[8] S.K. Donaldson, Riemann Surfaces, Oxford University Press 2011.
[9] R. Friedman, On threefolds with trivial canonical bundles, Complex geometry and Lie theory,
Sundance, UT, 1989, Proc. Sympos. Pure Math. 53, Amer. Math. Soc. (1991), 103– 134.
[10] R. Friedman, The B B̄-lemma for general Clemens manifolds, Pure and Applied Mathematics
Quarterly 15(4) (2019), 1001-1028.
[11] J.Fu, J. Li, and S.-T. Yau, Balanced metrics on non-Kahler Calabi-Yau threefolds, J. Differen-
tial Geom. 90(1), 81-129, (2012).
[12] B.R. Greene, D.R. Morrison, and A. Strominger. Black hole condensation and the unification
of string vacua. Nuclear Physics B 451, no. 1-2 (1995), 109-120.
[13] P. Griffiths and J. Harris, Principles of algebraic geometry, John Wiley and Sons, 2014.
[14] M. Gross, D. Joyce and D Huybrechts, Calabi-Yau manifolds and related geometries, Springer
2001.
[15] R. Harvey and H.B. Lawson, Calibrated geometries, Acta Mathematica (1982), 148, pp.47-157.
[16] D. Huybrechts, Complex geometry: an introduction, Springer 2005.
[17] E. Kahler, Uber eine bemerkenswerte Hermitesche Metrik, Abh. Math. Sem. Univ. Hamburg,
9 (1933), 173–186. DOI:10.1007/BF02940642
[18] Y. Kawamata, Unobstructed deformations – a remark on a paper of Z. Ran, J. Alg. Geom. 1
(1992), 183–190.
[19] K. Kodaira, Complex manifolds and deformation of complex structures, Grundlehren der math-
matischen Wissenschaften 1986.
[20] K. Kodaira and J. Morrow, Complex manifolds, Holt, Rinehart and Winston Inc, 1971.

96
[21] K. Kodaira and D.C. Spencer, On Deformations of Complex Analytic Structures, III. Stability
Theorems for Complex Structures, Annals of Mathematics 71(1) (1960), p.43-76.
[22] D. McDuff and D. Salamon, J-holomorphic Curves and Symplectic Topology, American Math-
ematical Society 2012.
[23] M. Rossi, Geometric transitions, J.Geom.Phys. 56 (2006), 1940-1983.
[24] Y.-T. Siu, Lectures on Hermitian-Einstein metrics for stable bundles and Kähler-Einstein met-
rics, DMV Seminar, Band 8 Birkhauser 1987. Notes available on the author’s website.
[25] J. Song and B. Weinkove, An introduction to the Kahler-Ricci flow, in An introduction to the
Kahler-Ricci flow, 89-188, Lecture Notes in Math., 2086, Springer 2013.
[26] G. Szekelyhidi, An Introduction to Extremal Kahler Metrics, AMS Graduate Studies in Math-
ematics 152 (2014).
[27] G. Tian, Smoothing threefold with trivial canonical bundle and ordinary double points, Essays
on Mirror Manifolds Internat. Press, Hong Kong (1992), 458–479.
[28] V. Tosatti, KAWA lecture notes on the Kahler-Ricci flow, Annales de la Faculte des sciences
de Toulouse: Mathematiques 27 (2018), no. 2, 285-376.
[29] V. Tosatti and B. Weinkove, The complex Monge–Ampere equation on compact Hermitian
manifolds, J. Amer. Math. Soc., 23 (2010), 1187-1195.
[30] C. Voisin, Hodge Theory and Complex Algebraic Geometry II, Cambridge University Press
2003.
[31] S.-T. Yau, On the Ricci curvature of a compact Kahler manifold and the complex
Monge–Ampere equation. I, Comm. Pure Appl. Math. 31 (1978) 339-411.

97