Prof. Dror Varolin PDF

ICTS Lectures
Hilbert Space Techniques

in
Complex Analysis and Geometry
Dror Varolin
July 5, 2019
Contents
I Complex Geometry 6
1 Complex Manifolds and Holomorphic Vector Bundles 7
1.1 Complex manifolds and holomorphic functions . . . . . . . . . . . . . . . . . . . 7
1.2 Examples of complex manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Holomorphic vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 The definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 Transition functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Holomorphic structure of the tangent bundle . . . . . . . . . . . . . . . . 19
1.3.4 Some non-trivial examples of holomorphic line bundles . . . . . . . . . . 21
1.4 Differential forms on complex manifolds . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.1 Forms of bidgree (p, q) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.2 Exterior differential operators . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.3 Twisted differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2 Hermitian metrics 27
2.1 Hermitian metrics for complex vector bundles . . . . . . . . . . . . . . . . . . . . 27
2.2 Hermitian Riemannian metrics for complex manifolds . . . . . . . . . . . . . . . . 28
2.2.1 Structure of the metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Metric form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.3 Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Connections and Curvature 32

3.1 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.1 Basic definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.2 Induced connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.3 Connections with additional symmetry . . . . . . . . . . . . . . . . . . . 34
3.1.4 The Kähler condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.5 Induced connection on twisted forms . . . . . . . . . . . . . . . . . . . . 41
3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.2 Curvature of the Chern connection . . . . . . . . . . . . . . . . . . . . . . 44
3.2.3 Curvature of a line bundle . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.4 Curvature of determinant bundles . . . . . . . . . . . . . . . . . . . . . . 46
1
3.2.5 Symmetry of Kähler curvature . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.6 Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
II L2 Estimates for ∂¯ 55
4 L2 Estimates for ∂¯ in complex dimension 1 56
4.1 Domains in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2 L2 estimates for ∂¯ on Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.1 Regularity of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5 The Bochner-Kodaira Identity 70

5.1 The Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
¯
5.2 Two ∂-operators and their formal adjoints . . . . . . . . . . . . . . . . . . . . . . 71
5.3 The formal identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6 L2 Estimates on Compact Kähler manifolds 76

6.1 Hörmander’s Theorem on compact Kähler manifolds . . . . . . . . . . . . . . . . 76
6.2 A few useful applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7 L2 Estimates on Complete Kähler manifolds 81

7.1 Extending the ∂¯ operator to Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . 81
7.1.1 Closed, densely defined operators . . . . . . . . . . . . . . . . . . . . . . 81
7.1.2 The maximal extension of ∂¯ . . . . . . . . . . . . . . . . . . . . . . . . . 83
∗
7.1.3 The Hilbert space adjoint Tp,q . . . . . . . . . . . . . . . . . . . . . . . . 84
7.2 Complete Kähler manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2.1 Complete Riemannian manifolds and exhaustion functions . . . . . . . . . 85
7.2.2 Approximation of twisted (p, q)-forms on complete Kähler manifolds . . . 85
7.2.3 The basic estimate for complete Kähler metrics . . . . . . . . . . . . . . . 87
7.3 Hörmander’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.3.1 Hörmander’s Theorem for Complete Kähler metrics . . . . . . . . . . . . 87
7.3.2 Comparison of L2 norms from comparison of metrics . . . . . . . . . . . . 88
7.3.3 Hörmander’s Theorem for complete Kähler manifolds . . . . . . . . . . . 89
7.3.4 Weakly pseudoconvex Kähler manifolds . . . . . . . . . . . . . . . . . . . 90
7.3.5 Skoda’s estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.4 Singular Hermitian metrics: an overview . . . . . . . . . . . . . . . . . . . . . . . 92
III Twisted Methods 96

8 Donnelly-Fefferman-Ohsawa Technique 97
8.1 The twisted basic estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.2 Donnelly-Fefferman-Ohsawa estimate for twisted ∂¯ . . . . . . . . . . . . . . . . . 99
8.3 Ohsawa’s ∂¯ estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
2
8.4 Functions with self-bounded gradient . . . . . . . . . . . . . . . . . . . . . . . . 100
9 Extension Theorems 104

9.1 Extension without estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
9.1.1 Strongly pseudoconvex manifolds . . . . . . . . . . . . . . . . . . . . . . 104
9.1.2 Stein manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.1.3 Compact complex manifolds with positively curved line bundles . . . . . . 110
9.1.4 The Kodaira Embedding Theorem . . . . . . . . . . . . . . . . . . . . . . 112
9.2 L2 Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.2.1 Adjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.2.2 Statement of the L2 Extension Theorem . . . . . . . . . . . . . . . . . . . 116
9.2.3 Proof of Theorem 9.2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
10 Some applications of the L2 Extension Theorem 123

10.1 Beurling-Seip Theory of Interpolation . . . . . . . . . . . . . . . . . . . . . . . . 123
10.1.1 Seip’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.1.2 A more general setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.1.3 Weighted Bergman Inequalities . . . . . . . . . . . . . . . . . . . . . . . 127
10.2 Deformation Invariance of Plurigenera . . . . . . . . . . . . . . . . . . . . . . . . 131
10.2.1 L2 extension of twisted canonical forms . . . . . . . . . . . . . . . . . . . 134
10.2.2 Positively twisted canonical sections . . . . . . . . . . . . . . . . . . . . . 135
10.2.3 Construction of the metric . . . . . . . . . . . . . . . . . . . . . . . . . . 138
10.2.4 Examination of e−µ on the central fiber: conclusion of the proof . . . . . . 139
10.2.5 An L2/m -estimate for m-canonical sections . . . . . . . . . . . . . . . . . 140
11 Berndtsson’s Theorem on Plurisubharmonic Variation 142

11.1 Berndtsson’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
11.1.1 Variations of Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . . . . 142
11.1.2 Statement of Berndtsson’s Theorem . . . . . . . . . . . . . . . . . . . . . 146
11.1.3 The Bergman projection . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
11.1.4 Proof of Theorem 11.1.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
11.2 Application: The Suita Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . 151
11.2.1 Formulation of the Suita Conjecture . . . . . . . . . . . . . . . . . . . . . 152
11.2.2 Alternate formulation of the Suita Conjecture . . . . . . . . . . . . . . . . 153
11.2.3 Estimating the Bergman kernel . . . . . . . . . . . . . . . . . . . . . . . . 155
11.3 Application: Sharp Estimates for L2 extension . . . . . . . . . . . . . . . . . . . . 158
11.3.1 Statement of the Sharp L2 Extension Theorem . . . . . . . . . . . . . . . 158
11.3.2 Sharpness in Theorem 11.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . 159
11.3.3 Beginning of the proof of Theorem 11.3.1 . . . . . . . . . . . . . . . . . . 160
11.3.4 Dual formulation of the norm of the minimal extension . . . . . . . . . . . 160
11.3.5 Degeneration to the infinitesimal neighborhood . . . . . . . . . . . . . . . 163
11.3.6 The proof of Theorem 11.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . 164
11.4 Equivalence of Sharp Extension and Positive Variation . . . . . . . . . . . . . . . 166
3
11.4.1 Log plurisubharmonicity of the Bergman kernel . . . . . . . . . . . . . . . 166
11.4.2 Plurisubharmonic variation versus sharp L2 extension . . . . . . . . . . . . 170
4
Preface
These notes assume some basic familiarity with real analysis, elementary complex analysis in one
and several variables (such as the definition of holomorphic functions, Cauchy Integral Formula,
and so on, up to the point of the Hartogs Phenomenon), and the basic definitions in the study of
smooth manifolds, including differential forms.
5
Part I
Complex Geometry
6
Lecture 1
Complex Manifolds and Holomorphic

Vector Bundles
1.1 Complex manifolds and holomorphic functions

1.1.1 D EFINITION . A complex manifold of complex dimension n is a smooth manifold that has
an atlas whose transition functions are holomorphic. That is to say, a complex manifold is a
topological space X with an open cover {Uα ; α ∈ J} and homeomorphisms
open
ϕα : Uα → Vα ⊂ Cn , α ∈ J,
such that for each pair α, β ∈ J the homeomorphism
Φαβ := ϕα ◦ ϕ−1
β |U α∩U β : Vβ → V α
is a holomorphic diffeomorphism. The open sets Uα are called coordinate charts, and the maps ϕα
are called local coordinates. The maps Φαβ are called transition functions.
1.1.2 R EMARK . By analogy with real manifolds, the collection {(Uα , ϕα ) ; α ∈ J} is called a
holomorphic atlas. Two holomorphic atlases are said to be equivalent if their union is again a
holomorphic atlas. A maximal holomorphic atlas is an equivalence class of holomorphic atlases.
By using the inclusion of sets as a partial ordering, one sees that each maximal holomorphic at-
las contains a unique maximal element, so we can identify the equivalence class with an actual
holomorphic atlas. This maximal holomorphic atlas is huge, and in practice most of its charts are
irrelevant in any given situation. However, it is convenient to have this notion because the maximal
atlas is a unique object associated to the complex manifold, and hence characterizes its complex
structure. Perhaps the most important thing to keep in mind is that once a single holomorphic atlas
is found, this atlas uniquely determines a maximal holomorphic atlas.
For a holomorphic map F = (F 1 , ..., F n ) : Ω1 → Ω2 between two open subsets of Cn one has
the identity
(1.1) det(DF ) = | det ∂F |2 ,
7
where D is the real derivative in R2n ∼
= Cn and
n
∂F i

1 n
∂(F , ..., F ) = .
∂z j i,j=1
In particular, the holomorphic atlas with which the complex manifold is equipped has the property
that the derivatives of its transition functions all have positive determinants, i.e., every complex
manifold is orientable.
The definition of complex manifold is such that in the neighborhood of each of its points all of
the concepts of complex analysis in one and several variables make sense, since they are invariant
under holomorphic changes of coordinates. This feature facilitates the extension of the standard
objects of complex analysis from domains in Euclidean space to complex manifolds. The rigorous
definition is as follows.
1.1.3 D EFINITION . Let X and Y be complex manifolds.
1. A function f : X → C on a complex manifold X said to be holomorphic (resp. harmonic,
pluriharmonic, subharmonic, plurisubharmonic) if for each pinX and each coordinate chart
ϕ : U → V ⊂ Cn the function f ◦ ϕ−1 : V → C is holomorphic (resp. harmonic,
pluriharmonic, subharmonic, plurisubharmonic). The set of holomorphic functions on X is
denoted O(X).
2. A map F : X → Y is said to be holomorphic if for each f ∈ O(Y ), f ◦ F ∈ O(X).
3. A holomorphic map F : X → Y is said to be a holomorphic diffeomorphism if it is bi-
jective and the inverse map is holomorphic. In this case one also says that X and Y are
biholomorphic.
1.1.4 D EFINITION . A subset Z of a complex manifold X is said to be a complex submanifold if for
each p ∈ X and every coordinate chart ϕα : Uα → Vα such that p ∈ Uα there exist holomorphic
functions f1 , , ..., fk ∈ O(Vα ) such that
Z ∩ Uα = {x ∈ Uj ; f1 (ϕα (x)) = · · · = fk (ϕα (x)) = 0}
and for each p ∈ Uα ∩ Z
df1 (ϕα (p)) ∧ ... ∧ dfk (ϕα (p)) 6= 0,
i.e., df1 (ϕα (p)), ..., dfk (ϕα (p)) are linearly independent.
For points p ∈ Z, the number k is called the complex codimension of Z at p, and if X has
complex dimension n then the number n − k is called the dimension of Z at p.
A complex submanifold of complex codimension 1 is called a smooth complex hypersurface.
The implicit function theorem shows that a complex submanifold of a complex manifold is
itself a complex manifold. Our definition here is one of several standard definitions. In our defini-
tion a submanifold Z of a complex manifold X is always a closed subset of X (this would change
if one replaced ‘p ∈ X’ with ‘p ∈ Z’ in our definition), and its topological structure as an abstract
manifold is the relative topology it inherits as a subset of X. There are other standard definitions
in which either or these properties fails.
8
1.1.5 P ROPOSITION . If X and Y are complex manifolds then a map F : X → Y is holomorphic
if and only if its graph ΓF := {(x, F (x)) ; x ∈ X} ⊂ X × Y is a complex submanifold.
E XERCISES
1.1.1. Prove the formula (1.1).
1.1.2. Show that if D ⊂ C is an open set and f : D → C is an injective holomorphic function then
f 0 (z) 6= 0 for every z ∈ D.
In fact, the analogous fact holds for a holomorphic map F : Ω → Cn where Ω ⊂ Cn is an open
set: if F is injective then det DF is nowhere-zero. (The proof of this fact is elementary but not
obvious; You are encouraged to try to prove it or to look it up, though don’t spend to much time on
it if you choose to do it yourself.) Most relevantly, it is crucial that the dimensions of the domain
and of the range are equal. Show that the map
g : C 3 t 7→ (t2 , t3 ) ∈ C2
is injective and that its derivative vanishes at the origin.
1.1.3. Show that if Z is a complex submanifold of a complex manifold X and f : X → Y is a
holomorphic map then the restriction f |Z : Z → Y is a holomorphic map.
1.1.4. Prove Proposition 1.1.5.
1.2 Examples of complex manifolds

To get a feeling for complex manifolds and how they work, it is useful to have a number of exam-
ples. There are many other interesting example besides those we present here.
1.2.1 E XAMPLE . Open subsets of Cn are complex manifolds. One can take a chart to be the entire
space, and the coordinate map to be the identity. This atlas, which has only one coordinate chart,
is automatically holomorphic by definition, but things become non-trivial when one obtains the
maximal holomorphic atlas.
This set of examples is not particularly interesting; of course it contains the setting of almost all
problems in classical complex analysis, but if this were the only example of a complex manifold,
the theory would not exist.
1.2.2 E XAMPLE (Graphs). Let Ω ⊂ Cn be a domain and let f 1 , ..., f k ∈ C (Ω). Then the set
Γf := {(z, w1 , ..., wk ) ∈ Ω × Ck ; f i (z) = wi , 1 ≤ i ≤ k}
with its relative topology is a complex manifold. Indeed, a coordinate chart is given by the restric-
tion to Γf of the projection π : Ω × Ck → Ω to the first factor.
Again this example is not so interesting, because the complex structure is determined by a
single chart. Perhaps the one instructive aspect is that although Γf is a complex manifold in
general, it is a complex submanifold of Ω × Ck if and only if fi ∈ O(Ω), 1 ≤ i ≤ k.
9
1.2.3 E XAMPLE (Zero sets of non-degenerate holomorphic mappings in Cn ). For a collection of
functions f1 , ..., fk ∈ O(Cn ) define
Z = {x ∈ Cn ; fj (x) = 0 for all 1 ≤ j ≤ k}.
Assume moreover that the complex-valued differential 1-forms df1 , ..., dfk are linearly independent
at each point of Z. Then the set Z is a complex manifold of dimension n − k, as one can easily
show from the implicit function theorem applied to the mapping F = (f1 , ..., fk ) : Cn → Ck .
Closed submanifolds of Cn are called Stein manifolds. Those codimension-k closed submani-
folds of Cn that are cut out by k holomorphic functions are called complete intersections. Not all
closed submanifolds of Cn are complete intersections.
1.2.4 E XAMPLE (The extended complex plane). The set C b := C ∪ {∞}, where {∞} is any
singleton, is topologized by declaring that its open sets are
(i) open subset of C, and
(ii) {∞} ∪ (C − K) where K is a compact subset of C;
the latter are neighborhoods of ∞. With this topology, the space C b is homeomorphic to the 2-
sphere.
The open sets U1 = C and U2 = {∞} ∪ (C − {0}) and homeomorphisms ϕ1 : U1 → C and
ϕ2 : U2 → C defined by

1/w, w ∈ C − {0}
ϕ1 (z) := z and ϕ2 (w) :=
0 , w=∞
have transition functions
1 1
Φ12 (w) = and Φ21 (z) = ,
w z
so the atlas defined by this open cover by two coordinate charts is holomorphic. The resulting
complex manifold is called the extended complex plane, and also the Riemann sphere.
1.2.5 E XAMPLE (Riemann surfaces). As we already pointed out, every complex manifold is ori-
entable. One can ask if orientability and even-dimensionality guarantee the existence of a holo-
morphic atlas. In general this is not so, but remarkably in real dimension 2 every orientable surface
admits at least one holomorphic atlas (and sometimes many incompatible atlases), as we now ex-
plain.
Let S be an orientable surface, i.e., 2-dimensional real manifold. Orientability means that there
is an atlas of smooth coordinate charts such that the derivative matrices of all of the transition
functions have positive determinant. There are exactly two incompatible maximal atlases with this
property, and we shall fix one of them, referring to any coordinate chart in this atlas as positively
oriented.
We can also equip S with a Riemannian metric g. Of course, there are always many different
Riemannian metrics on any given manifold, but in two dimensions there is a remarkable class of
local normal forms for such metrics: for any point p ∈ S there is a coordinate chart U containing
p and a coordinate map ϕ : U → V ⊂ R2 such that
g = eρ (dϕ1 )2 + (dϕ2 )2

10
where ρ is a smooth function on U . In geometric language g is conformal to the Euclidean metric
in this coordinate chart.
R EMARK . Note that if S is oriented then we may assume that the coordinate chart is positively
oriented: if a given coordinate map ϕ = (ϕ1 , ϕ2 ) is not positively oriented then ψ := (ϕ2 , ϕ1 )
is positively oriented, and clearly swapping the first and second components preserves the normal
form.
Coordinate charts in which a metric g is conformal to the Euclidean metric are classically called
isothermal coordinates (for the metric g). Their existence when the metric g is real analytic goes
back to Gauss. For smooth metrics the existence theorem was proved by Korn and Lichtenstein,
and is perhaps the first result in the theory of elliptic partial differential equations. In particular,
it is non-trivial. There is a lot more to say about this theorem and results that extended it, but in
the interest of preserving focus let us mention only that a proof is possible by combining standard
Elliptic Regularity ideas with methods that we will develop later in the notes when we look at
Hörmander’s Theorem.
If we accept the theorem on existence of isothermal coordinates then the construction of a
holomorphic atlas is simple: at each point of S choose an isothermal coordinate system with
positive orientation, and let the atlas be composed of these chosen coordinate systems. Since
these coordinate functions map the metric g to a multiple of the Euclidean metric, the transition
functions preserve orientation and the derivatives of the transition functions, as maps of tangent
spaces, map circles to circles. Elementary complex analysis shows that in dimension 2 a map that
sends circles to circles and preserves orientation is holomorphic. Hence the atlas thus constructed
is a holomorphic atlas.
A 1-dimensional complex manifold is called a Riemann surface. Example 1.2.4 of the RIemann
sphere is a special case when the underlying manifold is S 2 ; its complex structure is determined
by for example the round metric. It turns out that any two smooth metrics on the Riemann sphere
are conformal, so that the complex structure of C b is the only possible maximal holomorphic atlas
one can put on C b once the orientation is fixed.
Complex conjugation, which reverses the orientation, gives a distinct (in the sense that its
maximal atlas is not compatible with the one defined on C b in Example 1.2.4) but biholomorphic
complex structure on C. The biholomorphic map is the map given by complex conjugation in each
b
coordinate chart.
1.2.6 E XAMPLE (Projective space). Let V be a complex vector space of complex dimension n + 1.
We say that x ∈ V − {0} is equivalent to y ∈ V − {0}, and write [x] = [y], if there is a complex
number λ such that y = λx. The relation [ ] is clearly an equivalence relation. We define the
projective space P(V ) to be the set of all equivalence classes of [ ]. The map
[ ] : V − {0} → P(V )
is continuous if we topologize P(V ) by declaring that U ⊂ P(V ) is open if and only if [ ]−1 (U ) is
open (in V − {0}, of course).
11
A holomorphic function F : V → C is said to be homogeneous if F (λv) = λk F (v) for some
k ∈ N and all λ ∈ C and v ∈ V . The zero set of such a homogeneous holomorphic function on
V − {0} gives rise to a well defined (and closed) subset of P(V ).
If we fix n + 1 linearly independent functions z 0 , ..., z n ∈ V ∗ (which are often called linear
coordinates) then we identify V with Cn+1 via the map
V 3 v 7→ (z 0 (v), ..., z n (v)) ∈ Cn+1 .
Any homogeneous function of degree k is given by a homogeneous polynomial of degree k in the

linear coordinates z 0 , ..., z n . With such a choice of coordinates, we thus define the open subsets
Ui := {[v] ∈ P(V ) ; z i (v) 6= 0}, i = 0, ..., n.
The map
1
ϕi : Ui 3 [v] 7→ (z 0 (v), ..., z i−1 (v), z i+1 (v), ..., z n (v)) ∈ Cn
z i (v)
is clearly one-to-one and onto. In other words
 zµ−1
 zi 1≤µ≤i
ζiµ =
 zµ
zi
µ>i
are coordinates on Ui , and ϕi = (ζi1 , . . . , ζin ).

Now suppose we want to pass from ϕi (Ui ∩ Uj ) to ϕj (Ui ∩ Uj ). Assume without loss of
generality that i < j. Then one easily computes that
1/ζji+1 µ = j






µ
ζi = ζjµ /ζji+1 1 ≤ µ ≤ i and µ > j ,




 µ+1 i+1
ζj /ζj i<µ<j
i.e.,  
1  1
ζi = ϕi ◦ ϕ−1
j (ζj ) = ζj , ...ζji , ζji+2 , ..., ζjj , |{z}
1 , ζjj+1 , ..., ζjn 
ζji+1
j th spot
Thus ϕi ◦ϕ−1
j is a holomorphic diffeomorphism on ϕj (Ui ∩Uj ), and so P(V ) is a complex manifold
of dimension n.
When V = Cn+1 we write Pn := P(Cn+1 ).
1.2.7 E XAMPLE (Zero locus of homogeneous polynomials). We use the notation of Example 1.2.6.
Suppose F1 , ..., Fk : Cn+1 → C are homogeneous holomorphic polynomials, possibly with differ-
ent degrees of homogeneity, and let
Z := {z ∈ Cn+1 ; Fj (z) = 0, 1 ≤ j ≤ k}.
12
On the set [ ]−1 (Uj ), we may write
F` = (z j )d` f`,j (ζj ), 1 ≤ ` ≤ k.
It follows that on Cn+1 ∩ [ ]−1 (Uj ) ∩ {F` = 0},
dF` = (z j )d` dfj,` (ζj ) ⊃ Z.
Thus by the Implicit Function Theorem we see that [Z] is a smooth submanifold of Pn if and only if
Z is a smooth submanifold of Cn+1 − {0}. Both manifolds have codimension k. Thus the common
zero loci of homogeneous polynomials on Cn+1 whose differentials are independent on the part of
this zero locus that lies away from the origin, define a submanifold of Pn .
Note that we must remove the origin. For example, in C2 the common zero set of z 0 = z 1 = 0
is a pair of transverse lines through the origin, which is not a submanifold. On the other hand, the
induced set in P1 is a pair of distinct points.
Not all codimension k submanifolds of Pn are cut out by exactly k homogeneous polynomials.
Nevertheless, it is a fact, which we shall prove later on, that all submanifolds of Pn are cut out by
homogeneous polynomials.
1.2.8 E XAMPLE (The blowup of a point). Let B ⊂ Cn+1 denote the unit ball and let
B̃o := {(z, `) ∈ B × Pn ; z ∈ `}.
Since the projection map B × Pn → B is clearly proper1 , so is its restriction π : B̃o → B to the
closed subset B̃o .
We claim that B̃o is a complex manifold. Indeed, away from the origin in B, every point lies on
exactly one complex line through the origin, and thus π is a biholomorphic map away from π −1 (0).
On the other hand, if we choose projective coordinates [x0 , ..., xn ] on Pn then π −1 (B) is given by
the zero sets of the holomorphic functions
xj i
f j (z, [x]) := z j − z, j ∈ {0, ..., n} − {i}
xi
in the coordinate chart Vi := {(z, [x]) ∈ B × Pn ; |z| < 1 and xi 6= 0} in B × Pn−1 . We compute,
for the case i = 0, where z 0 , ..., z n , ζ 1 , ..., ζ n , ζ j = xj /xi , are coordinates on V0 , that
df j (z, ζ) = dz j − ζ j dz o − z o dζ j , , 1 ≤ j ≤ n,
and these are clearly independent. Thus B̃o is a complex manifold. Moreover, π(z, [1, ζ]) = z, so
E := π −1 (0) = {0} × Pn is also a submanifold of B × Pn , clearly biholomorphic to Pn . But in
fact, E is a submanifold of B̃o . Indeed, if we let φ : B̃o → Pn denote the restriction to B̃o of the
projection B × Pn → Pn . Then, over the open set Ui = {[x] ∈ Pn ; xi = 1}, the map
φ−1 (Ui ) = {((λxo , ..., |{z}

λ , ..., λxn ), [x]) ; λ ∈ C, [x] ∈ Ui },
ith position
1
Recall that a map between topological spaces is said to be proper if the inverse image of compact sets is compact.
13
and the map
φ−1 (Ui ) 3 λx, [x] 7→ (λ, λx1 , ..., λxi−1 , λxi+1 , ..., λn ) ∈ C × ϕi (Ui ) = C × Cn−1 ∼
= Cn
is a coordinate chart, and E ∩ B̃o is mapped to {0} × Cn−1 by this chart map.
Now let X be any complex manifold, and let p ∈ X. Then there is a neighborhood of p in
X that is biholomorphic to B via a biholomorphism that sends p to the origin. If we replace B
by B̃o in X (via the chart map) then we obtain a new complex manifold X̃p that is biholomorphic
to X away from p, and in place of p has a smooth submanifold biholomorphic to a projective
space of dimension dimC (X) − 1, i.e., a hypersurface. We also have a proper holomorphic map
π : X̃p → X such that
π : X̃p − π −1 ({p}) → X − {p}
is biholomorphic.
1.2.9 D EFINITION . The complex manifold X̃p is called the blowup of X at p, and the holomorphic
map π : X̃p → X is called the blowdown map. The smooth hypersurface π −1 ({p}) is called the
exceptional hypersurface for π.
1.2.10 E XAMPLE (Complex Tori). Let Λ be a lattice in a complex vector space V of dimension n.
(Recall that a lattice in a real vector space is a collection of vectors that is closed under addition
and whose convex hull is the whole vector space.) We say that x ∈ V is equivalent to y ∈ V , and
write x ∼ y, if there exists λ ∈ Λ such that y = x + λ. The set of equivalence classes is denoted
V /Λ or TΛ (V ), and we have a map
π : V → TΛ (V )
sending x to its equivalence class. As with P(V ), we endow TΛ (V ) with the coarsest topology that
makes π continuous, i.e., U ⊂ TΛ (V ) is open if and only if π −1 (U ) is open in V .
We define coordinate charts on TΛ (V ) as follows. Let x ∈ V and let U be a neighborhood of
x such that
U ∩ (U + λ) = ∅ for all λ ∈ Λ − {0}.
Then π|U is a homeomorphism. We let ϕU := (π|U )−1 : πU → U be our coordinate neighborhood.
Clearly the set of all such coordinate charts covers TΛ (V ), and if any two such charts intersect on
TΛ (V ), then their coordinate chart images in V intersect after a translation. Thus the transition
functions are holomorphic.
1.2.11 D EFINITION . The manifold TΛ (V ) is called a complex torus.
It is easy to see that complex tori are all diffeomorphic to the Cartesian product of 2n circles.
However, among all of these mutually diffeomorphic manifolds there are many different complex
structures. For example, it turns out that some complex tori have no proper submanifolds of posi-
tive dimension. On the other hand, some complex tori can be realized as submanifolds of projective
space, and therefore have many submanifolds.
Tori that can be embedded in some projective space are called Abelian varieties, and have a
special role in algebraic geometry.
14
E XERCISES
1.2.1. Prove the last statement at the end of Example 1.2.2, i.e., that Γf is a complex submanifold
of Ω × Ck if and only if fi ∈ O(Ω), 1 ≤ i ≤ k.
1.2.2. Show that if L : V → W is an injective linear map then the map
P(L) : P(V ) 3 [v] 7→ [Lv] ∈ P(W )
is well-defined and injective.

1.2.3. Show that
(a) for every holomorphic map F : P(V ) → P(W ) there exists a homogeneous holomorphic
map Φ : V → W such that F ([v]) = [Φ(v)], and that
(b) every holomorphic diffeomorphism of P(V ) is of the form P(L) (c.f. Exercise 1.2.2) for
some L ∈ GL(V ).
1.2.4. Show that C
b and P1 are biholomorphic.
1.2.5. Show that the sets
R := {[x, y, z] ∈ P2 ; xy = z 2 } and E := {[x, y, z] ∈ P2 ; y 2 z = x3 + xz 2 + z 3 }
are Riemann surfaces, that R is biholomorphic to P1 , and that E is not biholomorphic to P1 .

H INT: Show that the meromorphic 1-form
dξ
ω :=
η
on U2 = {z 6= 0} of P2 with coordinates ξ = x/z and η = y/z, is holomorphic on E ∩ U2 , and extends to
a well-defined holomorphic 1-form on E.
1.3 Holomorphic vector bundles

1.3.1 The definitions
• A holomorphic vector bundle of rank r is a triple (V, X, π : V → X) such that
(i) V and X are complex manifolds,
(ii) π is a holomorphic map, and
(iii) each p ∈ X is contained in an open set U on which there are holomorphic maps
e1 , ..., er : U → V such that
πei = IdU and spanC {e1 (x), ..., er (x)} = Vx for all x ∈ U,
where Vx := π −1 (x) denotes the fiber of π over x ∈ X. Such a collection of maps

{e1 , ..., er } is called a frame for V over U .
15
Observe that if {ei } and {ẽi } are two frames defined over the same open set U , then there
are holomorphic functions gij ∈ O(U ) such that gij (p) ∈ GL(r, C) for all p ∈ U and
ẽi = gij ej .
1.3.1 R EMARK . If “holomorphic” and “complex manifold” are replaced by “smooth (or
continuous)” and “smooth manifold (or topological space)” one obtains the definition of
complex vector bundle.
• A holomorphic vector bundle of rank 1 is called a holomorphic line bundle. Holomorphic

line bundles play an especially important role in complex analytic and algebraic geometry.
• A map of holomorphic vector bundles is a holomorphic map F : V → W such that
(i) the diagram
F
V W
Id
X X
commutes, and
(ii) for each x ∈ X the map
Fx := F |Vx : Vx → Wx
is linear.
Two vector bundles are isomorphic if there are holomorphic vector bundle maps F : V → W
and G : W → V such that F G = IdV and GF = IdW .
• A section s of a holomorphic vector bundle π : V → X, i.e., a right inverse for π, is said
to be holomorphic (resp. smooth, measurable, etc.) if it is holomorphic (resp. smooth,
measurable, etc.) as a map X → V .
1.3.2 E XAMPLE (Trivial bundles). The simplest example of a holomorphic vector bundle is the
trivial bundle π : X ×Cr → X, where π denotes the projection to the first factor. If a vector bundle
V → X is isomorphic to the trivial bundle, then for any basis e1 , ..., er of Cr the isomorphism
F : X × Cr → V defines a frame
ei (x) := F (x, ei ), 1≤i≤r
over the whole of X. Conversely, a global frame for a vector bundle V → X defines an isomor-
phism F −1 , where F is given by the same formula and then extended fiberwise-linearly. That is to
say, if we fix a basis e1 , ..., er of Cn , we define the isomorphism F : X × Cr → V by
F −1 (f i (x)ei (x)) := (x, f i (x)ei ).
Thus a vector bundle is isomorphic to the trivial bundle if and only if the vector bundle has a global
frame. In particular, every (holomorphic) vector bundle is locally trivial.
16
1.3.3 E XAMPLE (New bundles from old, pullbacks, etc).
(i) Observe that if V → X and W → X are holomorphic vector bundles the so are V ∗ → X,
V ⊗ W → X and V ⊕ W → X. Thus Symk (V ) → X and Λk V → X are holomorphic
vector bundles, as are all vector bundles obtained from holomorphic vector bundles from
multi-C-linear operations. On the other hand, the complex conjugate bundle V → X, which
we will meet often in the text, is in general not a holomorphic vector bundle.
(ii) In topology (or even category theory) one has the so-called fiber product: Given morphisms
f : X → Z and g : Y → Z one defines
X ×Z Y := {(x, y) ∈ X × Y ; f (x) = g(y)}.
There are projection maps X ×Z Y → X and X ×Z Y → Y given by the restriction to

X ×Z Y of the Cartesian projections X × Y → X and X × Y → Y .
If π : V → Y is a holomorphic vector bundle and f : X → Y is a holomorphic map then
f ∗ V = V ×Y X → X
is a holomorphic vector bundle, called the pullback of V by f .
(iii) Given holomorphic vector bundles V → X and W → Y , one defines the holomorphic
vector bundle
V W = p∗X V ⊗ p∗Y W → X × Y,
where pX : X × Y → X and pY : X × Y → Y are the Cartesian projections.
1.3.4 E XAMPLE . A vector bundle map F : V → W can be identified with a holomorphic section
of the bundle W ⊗ V ∗ .
1.3.2 Transition functions

Let V → X be a holomorphic vector bundle. As already pointed out, the choice of a frame over an
open set U ⊂ X yields a vector bundle isomorphism ψ to the trivial bundle U × Cr → U , defined
by
ψ(ti ei (x)) = (x, t).
With two such frames, and the corresponding isomorphisms ψ and ψ 0 , one can form a map
ψ 0 ◦ ψ −1 : U ∩ U 0 × Cr → U ∩ U 0 × Cr .
It is clear from the definitions that
ψ 0 ◦ ψ −1 (x, t) = (x, gU 0 U (x)t)
for some holomorphic function gU 0 U : U ∩U 0 → GL(r, C). The functions gU 0 U are called transition
functions, and they contain all of the information of (the isomorphism class of) the vector bundle.
17
To be more precise, suppose we cover X by a collection of open sets Uα , α ∈ A, over each of
which one has a frame for V . Then we get a collection of holomorphic transition functions
{gαβ : Uα ∩ Uβ → GL(r, C) ; α, β ∈ A}.
It is easy to see that the conditions
gαα = Id and gαβ gβγ gγα = Id, α, β, γ ∈ A
on their domains of definition.
Conversely, if one is given an open cover {Uα ; α ∈ A} of X and a collection of holomorphic
functions
T := {gαβ : Uα ∩ Uβ → GL(r, C) ; α, β ∈ A}
that satisfy the so-called cocycle condition
gαα = Id and gαβ gβγ gγα = Id, α, β, γ ∈ A
−1
on their domains of definition. (By taking γ = α one sees that gαβ = gβα .) Then one can define
the total space !
a
V (T ) := Uα × Cr / ∼,
α∈A
where
Uα × Cr 3 (x, t) ∼ (y, s) ∈ Uβ × Cr ⇐⇒ x=y and s = gαβ (x)t,
(the cocycle condition implies that ∼ is an equivalence relation) and the map π(T ) : V (T ) → X
by π(T )([x, t]) = x. It is easy to see that V (T ) is a holomorphic vector bundle whose transition
functions are T . Moreover, if T is a collection of holomorphic transition functions for V → X
the V (T ) → X is isomorphic to V → X. Indeed, if we denote by e1α , ...erα the frame over Uα
then the map
F ([x, t]) := t1 e1α (x) + · · · + tr erα , x ∈ Uα (x)
yields a well-defined holomorphic vector bundle isomorphism.
On the other hand, if V → X and W → X are isomorphic, they need not have the same
transition functions.
1.3.5 P ROPOSITION . If
{gαβ : Uα ∩ Uβ → GL(r, C) ; α, β ∈ A} and {g̃αβ : Uα ∩ Uβ → GL(r, C) ; α, β ∈ A}
are transition functions for V → X and W → X over the same open cover (if the open covers
are locally finite then by taking finite intersections this can always be arranged) then V → X and
W → X are isomorphic if and only if there exist maps
{hα : Uα → GL(r, C) ; α ∈ A}
such that g̃αβ = h−1
α gαβ hβ .
The proof is left as an exercise.
18
1.3.3 Holomorphic structure of the tangent bundle
On a complex manifold X one has the real tangent bundle TX → X; this is a smooth real vector
bundle, and one can then define the complex vector bundle TX ⊗R C → X. The points of the total
space of TX ⊗R C are called complex tangent vectors.
Now, for a holomorphic coordinate system z = (z 1 , ..., z n ) on an open set U ⊂ X (i.e., an
element of the maximal holomorphic atlas of X) one can define the complex tangent vectors
√

∂ 1 ∂ ∂
(1.2) := − −1 i , 1 ≤ i ≤ n,
∂z i 2 ∂xi ∂y
√
where z i = xi + −1y i . These vectors depend on the local coordinate system, but their span does
not (Exercise 1.3.2). For each x ∈ U we define

1,0 ∂ ∂
TX,x := SpanC ,..., n .
∂z 1 ∂z
1,0
The elements of TX,x are called (1, 0)-vectors at x.
For each ξ ∈ TX,x the vector ξ¯ ∈ TX ⊗R C (the complex conjugate of ξ) does not lie in TX,x
1,0 1,0
(Exercise 1.3.3). Defining

0,1 1,0
TX,x := TX,x ,
we obtain the decomposition
1,0 0,1
(1.3) TX ⊗R C = TX,x ⊕ TX,x .
Define the vector bundle π : TX1,0 → X by

a 1,0
TX1,0 := 1,0
TX,x and π −1 (x) := TX,x ,
x∈X
with the vector bundle structure given by the frames (1.2). The chain rule shows that TX1,0 → X is
a holomorphic vector bundle.
From basic complex analysis one sees that if U is an open set in X, x ∈ U , f ∈ O(U ) and
1,0
ξ ∈ TX,x then
ξf = (2Re ξ) f.
1,0
Moreover, if ι : TX,x ,→ TX,x ⊗R C denotes the natural inclusion and π 1,0 : TX,x ⊗R C → TX,x
denotes the projection to the first factor in the decomposition (1.3) then the composite map
s1,0 := π 1,0 ◦ ι : TX → TX1,0
is a real isomorphism of vector bundles whose inverse is the map ξ 7→ 2Re ξ.

From these observations one can give TX → X the structure of a holomorphic vector bundle in
two separate but equivalent ways. The first and simplest way is to map TX isomorphically to TX1,0 ;
since the latter is a holomorphic vector bundle, so is the former.
19
1.3.6 D EFINITION . The vector bundle TX1,0 → X is called the holomorphic tangent bundle. The
dual vector bundle TX∗1,0 → X is called the holomorphic cotangent bundle of X.
Note that in a local coordinate system a frame for the holomorphic cotangent bundle is given
by the complex 1-forms dz 1 , ..., dz n .
The second way is to determine the complex structure on TX directly. To do so, it is first
necessary to make√explicit the real operator (of the underlying real vector space) associated to
multiplication by −1 in a complex vector space. To cut to the chase, the reader should verify
i i
√ thei operator Jo : TX → TX defined pointwise in a holomorphic coordinate
(Exercise 1.3.4) that,
system z = x + −1y by
∂ ∂ ∂ ∂
(1.4) Jo i
:= i and Jo i
= − i, 1≤i≤n
∂x ∂y ∂y ∂x
is (i) well-defined, independent of the holomorphic cooridnate system, and (ii) satisfies
√
2Re ( −1ξ) = Jo 2Re ξ, ξ ∈ TX1,0 .
1.3.7 R EMARK (Almost complex structures and almost complex manifolds). The operator Jo is a
vector bundle map Jo : TX → TX that satisfies
Jo2 = −Id.
√
Any vector bundle endomorphism J ∈ Hom(TX , TX ) such that J 2 = −Id has eigenvalues ± −1,
and is therefore only diagonalizable in TX ⊗R C. In the latter bundle every eigenvector has a
complex conjugate eigenvector, so that the eigenspaces
√ √
E+ (J) := {v ∈ TX ⊗R C ; Jv = −1v} and E− (J) := {v ∈ TX ⊗R C ; Jv = − −1v}
are isomorphic under the complex conjugation map, and together span TX ⊗R C. The endomor-
phism J is therefore in 1-1 correspondence with such a splitting, and either object is called an
almost complex strucutre. A real manifold with an almost complex structure is called an almost
complex manifold. and as a matter of notation, one also writes
E+ (J) = TX1,0 (J) and E− (J) = TX0,1 (J)
for the J-eigenspaces.

The almost complex structure Jo has an additional feature: one can find an atlas for X so that
(1.4) holds on every coordinate chart. An almost complex structure J is said to be integrable if
there is a holomorphic atlas with respect to which J is given by (1.4).
If the manifold X is a surface, i.e., of real dimension equal to 2, then every almost complex
structure is integrable (if we accept the theorem on existence of isothermal coordinates). Indeed,
if we choose any metric go on X and let g := go + J ∗ go then J is an isometry of g whose square is
−Id. If we then choose an atlas of isothermal coordinates for g then J will be an isometry of the
Euclidean metric in a coordinate chart, and hence either J or −J will satisfy (1.4).
20
In higher dimension, integrability is more subtle, as there are non-integrable almost complex
structures. The coordinate charts that make (1.4) hold, if they exist, are again solutions of a partial
differential equation. Fortunately the existence of solutions is equivalent to an algebraic condition
called involutivity on the almost complex structure, which is analogous to the Fröbenius involutivity
condition:
[Γ(X, TX1,0 (J)), Γ(X, TX1,0 (J))] ⊂ Γ(X, TX1,0 (J)),
where [·, ·] is the complex linear extension of the Lie bracket of vector fields on X.
More precisely, the algebraic condition of involutivity measures the obstruction to solving the
aforementioned partial differential equation (in fact, it’s a system of PDE). The result that proves
the existence of a holomorphic atlas when the manifold has an involutive almost complex structure
is called the Newlander-Nirenberg Theorem. As was already indicated, it is a generalization of the
theorem on the existence of isothermal coordinates.
With the methods developed in these notes it is almost possible to prove the Newlander-
Nirenberg Theorem; such a proof is due to J. J. Kohn, who wrote in his article on the subject
that the idea for the proof came from D. C. Spencer. Unfortunately we will not have sufficient time
to discuss this result.
1.3.4 Some non-trivial examples of holomorphic line bundles

1.3.8 E XAMPLE (The canonical bundle). The canonical bundle KX → X is the determinant of the
holomorphic cotangent bundle:
KX := det TX∗1,0 := TX∗1,0 ∧ ... ∧ TX∗1,0 .
| {z }
n copies
In a holomorphic local coordinate chart a frame of KX is given by the n-form dz 1 ∧ ... ∧ dz n . Thus
the transition functions for KX are the determinants of the transition functions for TX∗1,0 .(Note that
if X is a Riemann surface then of course KX = TX∗1,0 .)
The name ‘canonical’ refers to the fact that KX is essentially the only natural (often) non-trivial
line bundle defined on every complex manifold X.
1.3.9 E XAMPLE (The Tautological bundle and its dual, the hyperplane bundle). Consider the space
U := {(z, `) ∈ Cn+1 × Pn ; z ∈ `}.
This space was first introduced as the blowup of the origin in Cn+1 (where the notation C̃n+1
o was
used), with the blowdown map being the restriction to U of the projection to the first factor. Let
π : U → Pn be the restriction to U of the projection to the second factor. Then π is holomorphic,
and the preimage of a point ` ∈ Pn , thought of as a line through the origin, is the set of points
comprising the line `. In the chart
Uo = {[1, z] ; z ∈ Cn }
we have the holomorphic section
eo ([1, z]) = ((1, z), [1, z])
21
which defines a frame for π : U → Pn over Uo . More generally, in the chart
Uj = {[z 1 , ..., z j , 1, z j+1 , ..., z n ] ; z ∈ Cn }
we have the section
ej ([z 1 , ..., z j , 1, z j+1 , ..., z n ]) = ((z 1 , ..., z j , 1, z j+1 , ..., z n ), [z 1 , ..., z j , 1, z j+1 , ..., z n ]).
It is easy to see that the change of frame over Ui ∩ Uj is holomorphic. Thus U → Pn is a holo-
morphic line bundle, called the tautological line bundle (and sometimes also the universal line
bundle).
The line bundle U → Pn is not trivial. We shall see this non-triviality by showing that the dual
bundle is not trivial.
The dual line bundle to U → Pn is called the hyperplane line bundle, and denoted H → Pn . Its
fiber H` consists of the set of linear functionals on the 1-dimensional subspace ` of Cn+1 . Thus a
linear function λ : Cn+1 → C defines a section sλ : Pn → H of H → Pn+1 via the formula
hsλ (`), (z, `)i := λ(z).
It is easy to check that this section is holomorphic, and that sλ (`) = 0 if and only if ` ⊂ Kernel(λ).
(As we shall see later on, every holomorphic section of H → Pn is of the form sλ for some
λ ∈ (Cn+1 )∗ .)
By contrast, U → Pn has no global holomorphic sections other than the zero section. Indeed,
suppose U → Pn has a holomorphic section θ : Pn → U. Given any linear functional λ ∈
(Cn+1 )∗ − {0}, we have the section sλ . The duality pairing
g(`) := hsλ (`), θ(`)i
is thus a well-defined holomorphic function on Pn . Since Pn is compact, g must be constant. Since
every non-identically zero linear functional on Cn+1 has a non-trivial kernel for n ≥ 1, g = 0,
and therefore θ vanishes on the complement of the zero locus of sλ . But by the identity principle,
θ = 0. In particular, U → Pn is not trivial, and therefore neither is H → Pn .
We shall see the line bundles U → Pn and H → Pn very often in the sequel.
1.3.10 E XAMPLE (The line bundle of a hypersurface). Let X be a complex manifold and let Z be a
smooth complex hypersurface. By definition there is an open cover {Uα ; α ∈ J} and holomorphic
functions fα ∈ O(Uα ) such that
Z ∩ Uα = {fα = 0} and dfα (z) 6= 0 for every z ∈ Z ∩ Uα .
It follows that the functions
fα
∈ O(Uα ∩ Uβ )
gαβ :=
fβ
have no zeros or poles. The collection T of all of these functions obviously satisfies the cocycle
condition, and hence there is a holomorphic line bundle
LZ → X
whose transition functions are T .
22
1.3.11 D EFINITION . The line bundle LZ is called the line bundle associated to the smooth hyper-
surface Z.
We will meet the line bundle LZ when we study L2 extension.
E XERCISES
1,0
1.3.2. Show that the span TX,x of the vectors (1.2) is independent of the choice of local coordinates.
1,0
1.3.3. Show that if ξ ∈ TX,x then ξ¯ 6∈ TX,x
1,0
.
1.3.4. Show that the operator J defined

√ by (1.4) is independent of the √
holomorphic coordinate
system and that it is intertwined with −1 via the map s , i.e., s J = −1s1,0 .
1,0 1,0
1.3.5. Show that every holomorphic section of the hyperplane line bundle H → Pn of Example
1.3.9 is of the form sλ for some λ ∈ (Cn+1 )∗ .
1.3.6. Find all the holomorphic sections the line bundle TP1,0
1
→ P1 .
1.3.7. Show that the only holomorphic section of the line bundle (TP1,0
1
)∗ → P1 dual to the tangent
bundle is the zero section.
1.3.8. Show that the line bundle KPn → Pn is isomorphic to U⊗(n+1) → Pn .
1.4 Differential forms on complex manifolds

Not surprisingly, on a complex manifold one wants to consider complex-valued differential forms,
i.e., sections of the bundle
2n
M 2n ^
M r
EX := ΛrX ⊗R C = (TX ⊗R C).
r=1 r=1
The sections of this bundle form an algebra with respect to the wedge product (or more precisely,
the complex linear extension of the a priori real wedge product), and the differential d (again
extended C-linearly) acts on the sections of EX → X, mapping sections of ΛrX → X to sections
of Λr+1
X → X:
d(Γ(X, ΛrX )) ⊂ Γ(X, Λr+1
X ).
Forms in the kernel of d are called closed, and forms in the image of d are called exact.
23
1.4.1 Forms of bidgree (p, q)
The splitting TX ⊗R C = TX1,0 ⊕ TX0,1 induces a splitting
p q
M ^ ^
ΛrX ⊗R C = TX1,0 ∧ TX0,1 ,
p+q=r
and, with the notation

p q
^ ^
Λp,q
X := TX1,0 ∧ TX0,1 ,
one has projections
π p,q : ΛrX ⊗R C → Λp,q
X .
0 0 0 0
The wedge product sends an element of Λp,q p ,q p+p ,q+q
X,x and an element of ΛX,x to an element of ΛX,x .
The smooth vector bundle Λp,q
X → X is a holomorphic vector bundle if and only if q = 0.
1.4.1 D EFINITION . The sections of Λp,q
X → X are called forms of bidegree (p, q), or (p, q)-forms.
Locally every (p, q)-form is of the form

(1.5) α = fI J¯dz I ∧ dz̄ J ,
where dz I ∧ dz̄ J := dz i1 ∧ ... ∧ dz ip ∧ dz̄ j1 ∧ ... ∧ dz̄ jq .
The expression (1.5) is not unique, i.e., different choices of coefficient functions fI J¯ can result
in the same α. However the coefficient functions are uniquely determined by the form if one
imposes the assumption of skew symmetry on the coefficient functions fI J¯ of the (p, q)-form α: if
σ ∈ Sp and τ ∈ Sq are permutations then one can impose the condition
fIσ Jτ = (sgnσ)(sgnτ )fI J¯
on the coefficients of α, where a permutation ν ∈ Sr acts on an r-tuple K = (k1 , ..., kr ) by the
formula
Kν = (kν(1) , ..., kν(r) ).
Some authors use increasing indices I and J, meaning 1 ≤ i1 < i2 < ... < ip ≤ n and similarly
for J. However, because the wedge product and exterior derivative are naturally skew-symmetric
object, representing a form using skew symmetric coefficients often results in much simpler com-
putations.
1.4.2 R EMARK . Here and in the rest of the text we are using summation convention: one sums
over all the possible values of the (multi)indices of the same letter, and these appear in pairs with
one a subscript and the other a superscript; moreover, the conjugation-parity is important. This
summation convention is a bit confusing at first, but in fact it respects the various dualities that
arise in complex geometry, and after a while it passes from the status of ‘confusing’ to the status
of ‘assisting’, helping one to correctly think of the geometric nature of the quantities at hand.
The last sentence in the previous paragraph may seem cryptic, but it is traditional, and the
author is confident that anyone that sticks to this convention will eventually see its benefits, and
will find the remark about duality eventually illuminating.
24
1.4.2 Exterior differential operators
The exterior algebra of a complex manifold is equipped with two additional differential operators:
¯ and they are defined on sections of Λp,q → X by
these are the operators ∂ and ∂, X
∂α := π p+1,q dα ¯ := π p,q+1 dα,

∂α α ∈ Γ(X, Λp,q
and X ).
One has the following proposition.

¯ and consequently ∂¯2 = 0.
1.4.3 P ROPOSITION . d = ∂ + ∂,
1.4.4 R EMARK . Since the splitting of TX ⊗R C makes sense on an almost complex manifold, one
also has the splitting of differential forms into components of bidegree (p, q) in such manifolds.
Proposition 1.4.3 fails in a general almost complex manifold, and in fact it holds if and only if the
almost complex structure is involutive, i.e., if and only if
[Γ(X, TX1,0 (J)), Γ(X, TX1,0 (J))] ⊂ Γ(X, TX1,0 (J))
(c.f. Remark 1.3.7). In these notes all of our almost complex structures are integrable, so we will
not discuss the issue of integrability any further. red
1.4.3 Twisted differential forms

1.4.5 D EFINITION . Let X be a complex manifold and let E → X be a holomorphic vector bundle.
An E-valued (p.q)-form is a section of the vector bundle
Λp,q
X ⊗ E → X.
If the vector bundle E is not explicitly referred to then one says that such a section is a twisted
differential form.
After one tensors with a holomorphic vector bundle, the exterior operator is no longer well
defined. More precisely, if we choose a frame ξ1 , ..., ξr for E and write
u = uνIJ¯dz I ∧ dz̄ J ⊗ ξν
then the form

∂uνIJ¯ k
k
dz ∧ dz I ∧ dz̄ J ⊗ ξν
dz
does depend on the holomorphic coordinate system, and therefore does not define a global section
of the bundle E ⊗ Λp+1,q
X → X. However, remarkably, the form
∂uνIJ¯ k
k
dz̄ ∧ dz I ∧ dz̄ J ⊗ ξν
dz̄
is indeed globally defined. This fact, which is left to the exercises, immediately implies the fol-
lowing proposition.
25
1.4.6 P ROPOSITION –D EFINITION . Let X be a complex manifold and let E → X be a holomor-
phic vector bundle, and let p, q ∈ {1, ..., n}. Then there is a local operator
∂¯ : Γ(X, Λp,q p,q+1

X ⊗ E) → Γ(X, ΛX ⊗ E)
defined as follows. If u = uνIJ¯ξν ⊗ dz I ∧ dz̄ J in some local coordinates and a local frame then
ν
¯ := ∂uI J¯ dz̄ k ∧ dz I ∧ dz̄ J ⊗ ξν
∂u
dz̄ k
1.4.7 R EMARK . It is worth making one more observation; an observation that will greatly simplify
many computations. Since
Λp,q ∼ p,0 0,q ∼ p,0 ∗ n,q

X = ΛX ⊗ ΛX = ΛX ⊗ KX ⊗ ΛX ,
one can write

Λp,q ∼ n,q
X ⊗ E = ΛX ⊗ F
where F = Λp,0 ∗
X ⊗ KX ⊗ E. The crucial point here is that F is a holomorphic vector bundle, and so
the ∂¯ operator of Proposition-Definition 1.4.6 is computed locally in exactly the same way for E-
valued (p, q)-forms as for F -valued (n, q)-forms. Consequently when working with twisted forms,
¯
especially when focus on the ∂-equation, it often suffices to consider only twisted (n, q)-forms.
1.4.8 R EMARK . Note that while the exterior differential operator ∂¯ extends naturally to holomor-
phically twisted forms, the exterior operator ∂ does not have such a natural extension. The reason,
as one can see after doing Exercise 1.4.2, is that the transition functions for a holomorphic vector
¯ but generally not by ∂.
bundle are annihilated by ∂,
E XERCISES
1.4.2. Prove Proposition-Definition 1.4.6. (For the sake of developing intuition, it may be useful
to start with the case p = q = 0.)
¯
1.4.3. Show that there are no non-zero ∂-closed (n, 0)-forms on Pn .
26
Lecture 2
Hermitian metrics
The terminology ‘Hermitian metrics’ is ambiguous in complex geometry: on the one hand, it can
refer to metrics for a complex vector bundle, and on the other, for certain Riemannian metrics on
almost complex manifolds. In the case where the underlying manifold is a complex manifold, there
is a link between the two notions.
To avoid the standard ambiguity, we shall refer to the first type of metric as a Hermitian metric,
and to the second type of metric as a Hermitian Riemannian metric.
2.1 Hermitian metrics for complex vector bundles

Let M be a manifold and let E → M be a complex vector bundle.
2.1.1 D EFINITION . A Hermitian metric for E → M is a section H of the bundle E ∗ ⊗ E ∗ → M
such that for all x ∈ M and v, w ∈ Ex ,
(i) hH, v ⊗ w̄i = hH, w ⊗ v̄i (i.e., H is Hermitian symmetric), and
(ii) hH, v ⊗ v̄i > 0 for all v 6= 0 (i.e., H is positive-definite).
In other words, H defines a sesquilinear, positive definite Hermitian form on each fiber Ex of the
vector bundle E → M .
If α1 , ..., αr is a frame for E ∗ over an open set U then one can write
H = Hij̄ αi · ᾱj
for functions Hij̄ over U satisfying
Hij̄ = Hj ī and Hij̄ ai āj ≥ ε||a||2
for some positive function ε on U and all a ∈ Cr . (Here αi · ᾱj = αi ⊗ ᾱj + ᾱj ⊗αi is the symmetric
r
product.) That is to say, at each x ∈ U the matrix Hij̄ (x) i,j=1 is Hermitian and positive-definite.
The regularity of the functions Hij̄ is declared to be the regularity of H.
Although we will think of the Hermitian metric H for E as a section of E ∗ ⊗ E ∗ , we will often
treat it as a Hermitian inner product on the fibers of E. As such, we will abusively write
H(v, w) := hH, v ⊗ w̄i .
27
Exercises
2.1.1. Let U → Pn be the tautological line bundle of Example 1.3.9. Use the Euclidean metric on
Cn+1 to define a Hermitian metric on U → Pn .
2.1.2. Let X be a complex manifold and let L → X be a holomorphic line bundle. Let s1 , ..., sm
be sections of L whose common zero locus is empty. Show that L → X has a Hermitian metric h
such that
h(si , si ) ≤ 1
for all 1 ≤ i ≤ m.
2.2 Hermitian Riemannian metrics for complex manifolds

Let X be an almost complex manifold and let J be the almost complex structure defined by (1.4).
2.2.1 D EFINITION . A Riemannian metric g on X is said to be Hermitian if J ∗ g = g, i.e.,
g(Jξ, Jη) = g(ξ, η)
for all x ∈ X and all ξ, η ∈ TX,x .
2.2.1 Structure of the metric

The condition of J-invariance puts some strong restrictions on the form of the metric; condi-
tions that are best seen in the decomposition of the complexified tangent space according to the
eigenspaces of J. In terms of the splitting TX ⊗R C = TX1,0 ⊕TX0,1 , let us write (the complexification
of) our metric as
AB
g= ,
CD
i.e.,
g(v1 + w̄1 , v2 + w̄2 ) = v1 · Av2 + v1 · B w̄2 + w̄1 · Cv2 + w̄1 · Dw̄2 .
Let us analyze the matrix CA D B

.
1,0
First, if v1 , v2 ∈ TX , then
√ √
g(v1 , v2 ) = g(Jv1 , Jv2 ) = g( −1v1 , −1v2 ) = −g(v1 , v2 ),
and thus A = 0. Since the metric g is the complexification of a real metric, g(v̄, w̄) = g(v, w), and
therefore D = 0.
Again by conjugation we have g(v, w) = g(v, w), and thus
C = B.
28
Finally, since g is symmetric, C = B trans , and therefore
B † = B,
where † means transpose conjugate. Thus we have

0h
g=
h0
for some Hermitian metric h for TX1,0 . We therefore see that Riemannian Hermitian metrics for X
are in one-to-one correspondence with Hermitian metrics for TX1,0 . This fact partially explains the
ambiguity in the name ‘Hermitian metric’.
We compute that for real vectors v and w, one has
√ √
2h(s1,0 v, s1,0 w) = 2g( 12 (v − −1Jv), 21 (w + −1Jw))
1 √
= g(v, w) + g(Jv, Jw) + −1(g(v, Jw) − g(Jv, w))
2 √
= g(v, w) − −1g(Jv, w),
where in the last equality we used the Hermitian symmetry of g. Therefore
g(v, w) = 2Re h(s1,0 v, s1,0 w),
and hence one can recover the metric g on X from a metric h on TX1,0 .
2.2.2 Metric form

The negative of the imaginary part of 2h(s1,0 v, s1,0 w), i.e.,
ωg (v, w) := g(Jv, w),
is called the metric form of the Hermitian Riemannian metric g. The form ωg is skew symmetric,
since
g(Jv, w) = g(w, Jv) = g(Jw, J 2 v) = −g(Jw, v).
The formula g(v, w) = ωg (v, Jw) shows that there is a 1-1 correspondence between g, ωg and h.
In terms of frames, if α1 , ..., αn is a local frame for TX∗1,0 then a Hermitian metric for TX1,0 may
be written
h = hij̄ αi ⊗ ᾱj .
In terms of this frame, if g(v, w) := 2Re h(s1,0 v, s1,0 w) is the associated Hermitian Riemannian
metric then √
ωg = −1hij̄ αi ∧ ᾱj .
29
Indeed, the right hand is well-defined because the vector bundles TX∗1,0 ⊗ TX∗0,1 and TX∗1,0 ∧ TX∗0,1
have the same transition functions, and we compute that
√ √
−1hij̄ αi ∧ ᾱj (v, w) = −1hij̄ αi ∧ ᾱj s1,0 v + s1,0 v, s1,0 w + s1,0 w
√
i 1,0

i 1,0
j 1,0
= −1hij̄ α , s v hα , s wi − α , s w hα , s vi j 1,0
= −2Im hij̄ αi , s1,0 v hαj , s1,0 wi = −Im 2h(s1,0 v, s1,0 w),

where the second-to-last equality follows from the Hermitian symmetry of hij̄ .
2.2.3 Volume
Given a Riemannian metric g on an m-dimensional real manifold, in a local coordinate system
x = (x1 , ..., xm ) one can define the m-form
q
det g ∂x∂ i , ∂x∂ j dx1 ∧ · · · ∧ dxm .

If one changes to another coordinate system y = (y 1 , ..., y m ) then

i
∂y
det ∂x
r q
j
det g ∂y∂ i , ∂y∂ j dy 1 ∧ · · · ∧ dy m = ∂
, ∂ dx1 ∧ · · · ∧ dxm .

i det g ∂xi ∂xj
∂y
det ∂x j
If one can choose an atlas whose transition functions all have positive determinant then one gets a
globally defined form q
dVg := det g ∂x∂ i , ∂x∂ j dx1 ∧ · · · ∧ dxm

of top degree with no zeros, i.e., a volume form. The existence of such an atlas is the definition of
orientability of a manifold. In particular, if the manifold is complex then the transition functions,
being holomorphic, have this property due to (1.1). (In this case the manifold is not only orientable,
but in fact oriented, i.e., there is a preferred atlas whose transition functions have derivatives with
positive determinant.)
If the manifold is complex and the Riemannian metric is furthermore Hermitian then the situ-
ation for the volume form is even better: In particular, for a Riemannian Hermitian metric g on a
complex manifold X
1
dVg = ωgn ,
n!
where ωg is the metric form of g. Indeed, if in local complex coordinates z = (z 1 , ..., z n ) and
corresponding real coordinates ξ = (ξ 1 , ..., ξ 2n ) where ξ 2i−1 = Re z i , ξ 2i = Im z i , we write
g = gij dξ i dξ j and h = hαβ̄ dz α · dz̄ β then
det (gij )2n n 2

ij=1 = (−1) (det(hαβ̄ )) ,
30
and thus
√ n
ωn = −1 hα1 β̄1 ...hαn β̄n dz α1 ∧ dz̄ β1 ∧ ... ∧ dz α1 ∧ dz̄ β1
(−1)n(n+1)/2
= √ hα1 β̄1 ...hαn β̄n dz α1 ∧ ... ∧ dz αn ∧ dz̄ β1 ∧ ... ∧ dz̄ βn
(2 −1)n
 

X 1 ... n 1̄ ... n̄
= sgn hα1 β̄1 ...hαn β̄n 
α1 ... αn β̄1 ... β̄n
α1 ,...,β̄n
(−1)n(n+1)/2 1
· √ n
dz ∧ ... ∧ dz n ∧ dz̄ 1 ∧ ... ∧ dz̄ n
(2 −1)
dz 1 ∧ dz̄ 1 dz n ∧ dz̄ n
= n! det(hαβ̄ ) √ ∧ ... ∧ √ .
2 −1 2 −1
Let us consider a complex submanifold Y of our Hermitian manifold X. We can endow Y with
a Hermitian metric simply by restricting the metric from the ambient space.
Now choose local coordinates z 1 , ..., z n on X in such a way that the functions z 1 , ..., z k are
coordinates on the submanifold Y . The tangent spaces of Y are then defined by the vanishing of
the differentials dz k+1 , ..., dz n . In other words, in these coordinates, the Hermitian metric for Y is
given by
h|Y = hαβ̄ dz α · dz̄ β ,
where this time the summation is carried only from 1 to k. In particular, one can carry out all of the
above calculations on the submanifold and obtain the following remarkable fact, often incorrectly
referred to as Wirtinger’s Theorem.
2.2.2 T HEOREM . If X is a Hermitian manifold with Hermitian metric form ω and ι : Y ,→ X is a
k-dimensional submanifold then the associated Hermitian volume form of Y is
1 1 k
dVg|Y = ι∗ ωgk = ωg| .
k! k! Y
The original theorem of Wirtinger is the following.
2.2.3 T HEOREM (Wirtinger). If W is a 2k-dimensional real submanifold of a Hermitian manifold
(X, ω) then
ωk
Z
Area(W ) ≥ .
W k!
Moreover, if the area of W is finite then equality holds if and only if W is a complex manifold.
Theorem 2.2.3 is often important in the study of minimal surfaces. Since we will not need it,
the proof of Theorem 2.2.3 is omitted.
E XERCISES
2.2.1. Let X be a compact complex manifold and let g be a metric whose metric form ωg is a
¯
closed 2-form. Show that there are no non-zero ∂-exact (p, p − 1)-forms on X.
31
Lecture 3
Connections and Curvature
3.1 Connections
In this section we define and explore the notion of connection for a vector bundle on a manifold.
The underlying field can be R or C; when we use the term linear, we assume scalars lie in this
field.
3.1.1 Basic definition

3.1.1 D EFINITION . Let M be a manifold and let E → M be a complex vector bundle. A connection
for E → M is a linear map
∗
D : Γ(M, E) → Γ(M, TM ⊗ E),
satisfying the Leibniz rule

D(f s) = df ⊗ s + f Ds.
The Leibniz rule implies the following proposition.
3.1.2 P ROPOSITION . If D1 and D2 are two connections for a vector bundle E → M then
D1 − D2 ∈ Γ(M, C ∞ (TM
∗
⊗ End(E))).
Proposition 3.1.2 says that the space of connections on a vector bundle E → M is an affine
space modeled on the vector space Γ(M, C ∞ (TM ∗
⊗ End(E))).
R EMARK . There are several notions of connection, that have increasing generality. The type of
connection we are considering here is often called an affine connection, presumably in reference
to the affine linear structures of the fibers of a vector bundle.
Definition 3.1.1 implies that a connection is a local operator, and may be restricted to small
subsets. If one restricts to a sufficiently small subset U ⊂ M then the vector bundle E|U → U is
isomorphic to the trivial bundle, i.e., it admits a frame.
32
With the choice of a frame e1 , ..., er for E|U one has the trivial connection d defined by
d(si ei ) = dsi ⊗ ei .
Note that this connection depends on the trivialization.

By Proposition 3.1.2 any other connection D for E|U is obtained from the trivial connection
∗
by adding to the latter a section A of TM ⊗ Hom(E|U ) → U :
D|U = d + A.
The section is called the connection form. Since the trivial connection depends on the frame, so
does the connection form A.
In terms of the frame e1 , ..., er for E|U → U , one can write
Aei = Aji ⊗ ej ,
and then by the Leibniz rule
D(si ei ) = dsi ⊗ ei + si Aji ⊗ ej .
The matrix of 1-forms (Aji ) is called the connection matrix.

Exercise 3.1.2 asks the reader to compute the transformation formula satisfied by the connec-
tion matrix under change of frame.
3.1.2 Induced connections

A connection for a vector bundle E → M induces connections on all vector bundles obtained from
E via multilinear operations. Similarly, a collection of vector bundles with connections induces
natural connections on various products of these vector bundles.
Dual connection
Given a connection D for a vector bundle E → M , one defines the connection D∨ for the dual
vector bundle E ∗ → M as follows. Given local sections s for E and α for E ∗ , one has a pairing
hs, αi, which is a function on M . We require the dual connection D∨ to satisfy
(3.1) d hs, αi = hDs, αi + hs, D∨ αi .
If we fix a frame e1 , ..., er for E and denote by α1 , ..., αr its dual frame then the connection forms
A(D) and A(D∨ ) satisfy
0 = dδi j = d ei , αj = A(D)ik ek , αj + ei , A(D∨ )`j α` = A(D)ij + A(D∨ )ij .

Thus the dual connection D∨ is completely determined by (3.1).
33
Product connections
Let E1 → M and E2 → M be two vector bundles equipped with connections D1 and D2 respec-
tively, one would like to have a natural definition of connection for various products of V1 and V2 .
Given any product operation × (e.g. tensor, symmetric, or wedge product) one defines the product
connection D for E1 × E2 → M by the formula
D (s1 × s2 ) = (D1 s1 ) × s2 + s1 × (D2 s2 ) .
By induction, one can pass to any finite product of vector bundles.
3.1.3 E XAMPLE (Induced connections for determinant bundles). Let E → M be a vector bundle
of rank r and let a connection DE for E be given. Consider the complex line bundle
det E → M
whose transition functions are just the determinants of the corresponding transition functions for
E. (We have met the determinant construction in the definition of the canonical bundle.) Fix a
frame e1 , ..., er for E. Then e1 ∧ ... ∧ er is a frame for det E. Then
e1 ∧ ... ∧ DE ej ∧ ... ∧ er = e1 ∧ ... ∧ A(DE )kj ek ∧ ... ∧ er = A(DE )kj δjk e1 ∧ ... ∧ ej ∧ ... ∧ er ,
and thus !
X
Ddet E (e1 ∧ ... ∧ er ) = A(DE )jj e1 ∧ ... ∧ er ,
j
i.e., the connection matrix for Ddet E is the trace of the connection matrix for DE .
3.1.3 Connections with additional symmetry

Metric Compatibility
3.1.4 D EFINITION . Let E → M be endowed with a metric g. We say that a connection D for E is
compatible with g if
d(g(s, t)) = g(Ds, t) + g(s, Dt)
for all local sections s, t of E.
If we view the metric g as a section of E ∗ ⊗ E ∗ → M then for any (not necessarily g-
compatible) connection D for E one has
d(g(s, t)) = g(Ds, t) + g(s, Dt) + Dg(s, t)
(where, of course, the last term involves the induced connection). Thus the condition of metric
compatibility may be written Dg = 0.
In general a given metric has many compatible connections. If D1 and D2 are two connections
∗
for E that are compatible with a metric g then their difference Θ := D1 − D2 ∈ Γ(M, TM ⊗
End(E)) satisfies
g(Θs, t) + g(s, Θt) = 0,
i.e., Θ is anti-symmetric (or anti-Hermitian if g is a Hermitian metric) with respect to g.
34
3.1.5 R EMARK . If X is a complex manifold then the splitting TX∗ ⊗R C = TX∗1,0 ⊕ TX∗0,1 induces
the decomposition Θ = Θ1,0 + Θ0,1 ∈ (End(E) ⊗ TX1,0 ) ⊕ (End(E) ⊗ TX0,1 ), and the condition
Θ† = −Θ means that
(Θ1,0 )† = −Θ0,1 and (Θ0,1 )† = −Θ1,0 .
In particular, if Θ ∈ Γ(X, TX∗ ⊗ End(E)) is of type (1, 0) then Θ must vanish identically.
Symmetric connections
Given any smooth manifold M , we have the splitting
∗ ∗ ∗ ∗
TM ⊗ TM = Sym2 (TM ) ⊕ Λ2 (TM ),
Therefore every connection D : Γ(M, C ∞ (TM

∗
)) → Γ(M, C ∞ (TM
∗ ∗
⊗ TM )) for the cotangent
bundle splits as
D = DS + DΛ .
On any manifold there is a natural operator sending 1-forms to 2-forms and satisfying the
Leibniz rule with respect to the wedge-product, namely the exterior derivative. We can therefore
make the following definition.
∗
3.1.6 D EFINITION . A connection D for TM is said to be symmetric if DΛ = d.
∗
If D is a connection for TM and we write
D(dxi ) = Cjk
i
dxk ⊗ dxj
in some local coordinate system then

2 i
2 ∂fj i k j
Λ D(fi dx ) = Λ + Cjk dx ⊗ dx
∂xk
i i

Cjk − Ckj

1 ∂fj ∂fk k j
= − dx ⊗ dx + fi dxk ⊗ dxj
2 ∂xk ∂xj 2
= d(fi dxi ) + fi Cjk
i
dxk ∧ dxj .
It follows that D is symmetric if and only if its connection matrix satisfies

i i
Cjk = Ckj .
It is often useful to formulate the notion of symmetric connection for the tangent bundle. Since
the tangent and cotangent bundle are dual, we shall call a connection for TM symmetric if it is the
∗
dual of a symmetric connection for TM . If ∇ is a connection for TM dual to a given connection D
∗
for TM then
Γijk = ∇ ∂
∂ ( ∂xj ), dx
i
=− ∂
∂xj
, D ∂ (dxi ) i
= −Cjk ,
∂xk ∂xk
35
and thus we see that ∇ is symmetric if and only if its connection matrix Γijk is symmetric, i.e.,
Γijk = Γikj .
Finally observe that if ξ = ξ i ∂x∂ i and η = η i ∂x∂ i are local vector fields, then
j j

i ∂η i ∂ξ ∂ ∂
∇ξ η − ∇ η ξ = ξ i
−η i j
+ ξ i η j (Γkij − Γkji ) k ,
∂x ∂x ∂x ∂x
and thus ∇ is symmetric if and only if ∇ξ η − ∇η ξ = [ξ, η].
Levi-Civita connections
On a Riemannian manifold there is exactly one symmetric, metric-compatible connection.
3.1.7 T HEOREM (Levi-Civita). On a Riemannian manifold (M, γ) there is a unique symmetric
∗
connection ∇ compatible with γ. In terms of the dual metric g for TM , there exists a unique
connection D such that
d(g(s, s0 )) = g(Ds, s0 ) + g(s, Ds0 ) and DΛ = d.
Moreover ∇ and D are dual connections.

Proof of Levi-Civita’s Theorem. Fix local coordinates (x1 , ..., xm ) on a neighborhood U in M .
∗
Then dx1 , ..., dxm is a frame for TM on the coordinate neighborhood U . Then with α = αi dxi ,
ij i j i i k
g = g(dx , dx ) and ω j = ω jk dx , Levi-Civita’s Theorem states that there is a unique solution
ω ijk to the system of equations
g ia ω jak + g jb ω ibk = ∂k g ij and ω ijk = ω ikj .
The first set of equations expresses metric compatibility, and the second, symmetry.
For notational ease, define
ω ijk = g ja g kb ω iab and ∂ k := g k` ∂` ,
so that the equations to be solved are
(3.2) ω ijk + ω jik = ∂ k g ij and ω ijk = ω ikj .
To solve (3.2), observe that the right hand side of the first equation is symmetric in ij, so it makes
sense to write ω ijk = S ijk + Aijk , where
1 1
S ijk = (ω ijk + ω jik ) and Aijk = (ω ijk − ω jik ).
2 2
(For fixed k S ijk and Aijk are the symmetric and antisymmetric parts of ω ijk .) Thus
1
ω ijk = ∂ k g ij + Aijk ,
2
36
and the second equation in (3.2) reads as
1
(3.3) Aijk − Aikj = (∂ j g ik − ∂ k g ij ).
2
In view of the symmetry of g ji and the antisymmetry of Aijk in ij, (3.3) yields
1
−Aijk + Akji = Ajik − Ajki = (∂ i g jk − ∂ k g ij ).
2
Subtracting the latter from (3.3) and using the antisymmetry of Aijk yields
1
Aijk + (Aijk + Akij + Ajki ) = (∂ j g ik − ∂ i g jk ).
2
But
1
Aijk + Ajki + Akij = ω ijk + ω jki + ω kij − (∂ k g ij + ∂ i g jk + ∂ j g ki )
2
1 1 1
= ω ikj − ∂ j g ik + ω jik − ∂ k g ji + ω kji − ∂ i g kj
2 2 2
ikj jik kji ijk jki kij
= A + A + A = −(A + A + A ).
Thus Aijk + Ajki + Akij = 0 and therefore
ω ijk := 1
∂ k g ij + ∂ j g ik − ∂ i g jk

2
is the (obviously unique) solution of (3.2). Lowering the indices, we have

1
ω ijk = gjm ∂k g mi + gkm ∂j g mi − grj gsk g im ∂m g rs .

2
Using the relation
(3.4) 0 = ∂m (grj g rs ) = g rs ∂m grj + grj ∂m g rs
yields grj gsk g im ∂m g rs = −gsk g rs ∂m grj = −∂m gkj = −∂m gjk , and therefore
1
ω ijk = gjm ∂k g mi + gkm ∂j g mi + g im ∂m gjk .

(3.5)
2
∗
To complete the proof, we must show that the connection ∇ : Γ(M, TM ) → Γ(M, TM ⊗ TM )
dual to the connection D is symmetric and compatible with the dual metric γ. This last step is left
to the exercises (Exercise 3.1.3).
∗ ∗ ∗
3.1.8 R EMARK . The connection D : Γ(M, TM ) → Γ(M, TM ⊗ TM ) has associated to it the dual
connection ∇ : CM (TM ) → CM (TM ⊗ TM ) via the relation
∞ ∞ ∗
∇∂j , dxi + ∂j , Ddxi = 0.

37
Contracting both sides with ∂k and setting
Γijk := ∇∂k ∂j , dxi ,

we find that
i
Γijk = −ωjk .
Using the relation (3.4) and the symmetry of the metric, (3.5) yields
1
Γijk = g im (∂k gmj + ∂j gmk − ∂m gjk )
2
for the Christoffel symbols of ∇.
Complex connection
On a complex manifold X we have a splitting
TX∗ ⊗ C = TX∗1,0 ⊕ TX∗0,1 .
It follows that, for a complex vector bundle E → X, a connection D splits as
(3.6) D = D1,0 + D0,1 .
If the vector bundle E → X is furthermore holomorphic then there is a canonical choice for the
component D0,1 : Γ(X, E) → Γ(X, Λ0,1 ¯
X ⊗ E), namely the operator ∂, introduced in Paragraph
1.4.3 (in particular, in Proposition/Definition 1.4.6).
3.1.9 D EFINITION . A connection D for a holomorphic vector bundle V → X is said to be complex
if D0,1 = ∂¯ in terms of the splitting (3.6).
Chern connection
The basic result about connections for holomorphic Hermitian vector bundles is the following
analogue of Levi-Civita’s theorem.
3.1.10 T HEOREM . On a holomorphic Hermitian vector bundle there exists a unique complex con-
nection compatible with the Hermitian metric.
Proof. We begin with uniqueness. Since the difference of two connections is a 0-th order differen-
tial operator, i.e., a matrix multiplier, if D0,1 = ∂¯ then this multiplier is a matrix of (1, 0)-forms. It
follows from Remark 3.1.5 that a metric compatible complex connection is unique if it exists.
To prove existence, define D1,0 by
¯
h(D1,0 s, t) = ∂h(s, t) − h(s, ∂t)
for all local sections s, t. Then
h(∂s, ¯ = ∂h(t, s) − h(D1,0 t, s) = ∂h(s,

¯ t) = h(t, ∂s) ¯ t) − h(s, D1,0 t),
38
¯ which is clearly complex, satisfies
and therefore the connection D = D1,0 + ∂,
dh(s, t) = h(Ds, t) + h(s, Dt),
i.e., D is h-compatible.
3.1.11 D EFINITION . The unique metric-compatible complex connection for a Hermitian holomor-
phic vector bundle (E, h) is called the Chern connection for (E, h).
3.1.4 The Kähler condition

The definition of Kähler metric
The tangent bundle TX of a complex manifold X with Hermitian Riemannian metric g carries
two natural connections. One connection is the Chern connection for the holomorphic Hermitian
vector bundle (TX , g), and the other connection is the Levi-Civita connection for (TX , g). (Here
g is viewed both as a Hermitian metric on TX1,0 and as a Riemannian metric on TX , via the cor-
respondence s1,0 : TX → TX1,0 discussed in Subsection 2.) In general, these two connections are
different.
3.1.12 D EFINITION . A metric g for which the Chern and Levi-Civita connections agree is said to
be Kähler. A complex manifold admitting a Kähler metric is called a Kähler manifold.
There are several equivalent ways to define Kähler metrics. Here we shall discuss only the two
we will need.
Symplectic formulation
3.1.13 P ROPOSITION . A metric g with metric form ω is Kähler if and only if the (1, 1)-form ω
associated to g is closed.
Proof. The Chern connection D for the metric g satisfies (c.f. Exercise 3.1.4)
¯
D ∂ ( ∂z∂ i ) =: Γkij ∂z∂k = g k` ∂z∂ j gi`¯ ∂z∂k .
∂z j
Since both the Levi-Civita and Chern connections are compatible with g, the Chern connection
agrees with the Levi-Civita connection if and only if the Chern connection corresponds, under the
isomorphism induced by s1,0 : TX → TX1,0 , to a symmetric connection, i.e.,
(3.7) Ds1,0 w (s1,0 v) = Ds1,0 v (s1,0 w)

n o
whenever v, w ∈ ∂x∂ 1 , ∂y∂ 1 , . . . , ∂x∂n , ∂y∂n . But since
√
s1,0 ∂x∂ i = ∂
∂z i
and s1,0 ∂y∂ i = −1 ∂z∂ i ,
39
the relation (3.7) holds if and only if Γkij = Γkji . Therefore g is Kähler if and only if
∂gik̄ ∂gj k̄
= .
∂z j ∂z i
But the latter is equivalent to ∂ω
√ = 0, iwhichk in turn is equivalent to dω = 0 (since ω is real). To
be more explicit, since ω = gik̄ −1dz ∧ dz̄ ,
√ ∂gik̄ j i k
√ ∂gik̄ `
dω = −1
j
dz ∧ dz ∧ dz̄ + −1 ` dz̄ ∧ dz i ∧ dz̄ k
∂z ∂ z̄
√ ∂gik̄ j √ ∂g
= −1 j dz ∧ dz i ∧ dz̄ k + −1 kì̄ dz ` ∧ dz k ∧ dz̄ i
∂z ∂z
√ X ∂gik̄ ∂gj k̄

= −1 j
− i
dz j ∧ dz i ∧ dz̄ k
i<j
∂z ∂z
√ X ∂gik̄ ∂gj k̄

+ −1 j
− i
dz j ∧ dz i ∧ dz̄ k .
i<j
∂z ∂z
It follows that g is Kähler if and only if dω = 0, as desired.
Locally Euclidean formulation

A very useful formulation of the Kähler condition is contained in the following theorem.
3.1.14 T HEOREM . The metric g is Kähler if and only if every point of X lies in a coordinate chart
with coordinates z so that X
g= dz α · dz̄ α + O(|z|2 ).
α
The coordinates referred to in the theorem are called Kähler coordinates.

Proof. Observe that if two (1, 1)-forms ω1 and ω2 , defined in a neighborhood of 0 in Cn , have the
same Taylor expansion up to second order, then one has
(dω1 )0 = (dω2 )0 .
Therefore if g is locally Euclidean to second order then dωg = 0, and thus g is Kähler by Proposi-
tion 3.1.13.
Conversely, suppose ω is the (1, 1)-form associated to a Kähler metric g, and let z be local
coordinates such that
(3.8) gij̄ (0) = δij̄
(It is easy to find such, so-called normal coordinates.) Then the Taylor expansion of ω with respect
to z is √
ω = −1 δij̄ + aij̄k z k + aij̄ `¯z̄ ` + O(|z|2 ) dz i ∧ dz̄ j .

The two properties of the Taylor coefficients aij̄k , aij̄ k̄ are
40
(a) gij̄ = gj ī ⇒ aij̄ k̄ = aj īk ,
(b) dω = 0 ⇒ aij̄k = akj̄i .
Since the result we seek is regarding a second order Taylor expansion, it suffices to find a quadratic
biholomorphic local coordinate transformation. That is to say, we seek coordinates w defined by
1
z k = wk + bk`m w` wm
2
(which do not modify condition (3.8)) such that
√
−1
δij̄ + O(|w|2 ) dwi ∧ dw̄j .

ω=
2
Since w` wm = wm w` , we may assume that bijk = bikj . Then
dz k = dwk + bk`m w` dwm ,
and we have
√
− −1ω = δij̄ dwi + bi`m w` dwm ∧ dw̄j + bjrs w̄r dw̄s
+ aij̄k wk + aij̄ k̄ w̄k dwi ∧ dw̄j + O(|w|2 )

= δij̄ + (aij̄k + δ`j̄ b`ki )wk + (aij̄ k̄ + δmī bm k 2

i j
kj )w̄ + O(|w| ) dw ∧ dw̄ .
Thus, if we set bjki = −δ j m̄ aim̄k (which is symmetric in ki by (a)) then

nī
δmi bm
kj = −δmi δ
mn̄ a
j n̄k = −δ aj n̄k = −aij̄ k̄ .
Thus √
ω= −1δij̄ dwi ∧ dw̄j + O(|w|2 ),
so g is Euclidean to second order.
3.1.5 Induced connection on twisted forms

Symmetric connections and exterior derivatives
∗
Consider a differential 1-form α = αi dxi on a manifold M . For a connection ∇ for TM → M,
∂αi j
∇α = dx ⊗ dxi + αk θij
k
dxj ⊗ dxi ,
∂xj
where θ is the connection matrix of ∇. As we discussed earlier,
∂αi j
(3.9) Λ2 (∇α) = dx ∧ dxi + αk θij
k
dxj ∧ dxi .
∂xj
41
We then introduced the notion of symmetric connection; by definition, a connection ∇ is symmetric
if and only if
dα = Λ2 (∇α)
k k
for any 1-form α. Of course, the formula (3.9) shows that ∇ is symmetric if and only if θij = θji .
r ∗
Now suppose β is a differential r-form, i.e., a section of the product bundle Λ (TM ). For a
∗
connection ∇ for TM → M , the product connection ∇r acts on β = βi1 ···ir dxi1 ∧ · · · ∧ dxir (with
our convention that βI is skew-symmetric in I) by
r
!
∂βI X
∇r β = + βi ···(`) ···i θ` dxio ⊗ dxi1 ∧ · · · ∧ dxir ,
∂xio j=1 1 j r io ij
where the notation (`)j means that ` replaces ij . If we take the (r + 1)th skew-symmetric part of
∇r β (thought of as an (r + 1)-tensor) we obtain
r
!
` `
∂βI X (θio ij − θij io )
Λr+1 (∇r β) = i
+ βi1 ···(`)j ···ir dxio ∧ dxi1 ∧ · · · ∧ dxir ;
∂x o
j=1
2
a calculation that uses the skew-symmetry of the βI . (Note that

r
!
∂β I
X
Λr+1 (∇r β) = + βi ···(`) ···i θ` dxio ∧ dxi1 ∧ · · · ∧ dxir ,
∂xio j=1 1 j r io ij
but that the latter formula does not adhere to our convention of skew symmetry of the coefficients
of a differential form.) Thus we see that a connection ∇ is symmetric if and only if
d = Λr+1 ◦ ∇r
holds for any integer r with 1 ≤ r ≤ n.
Twisted exterior derivative

Let V → M be a vector bundle with connection D. We can define a twisted version of the exterior
derivative for sections of
Γ(M, C ∞ (TM∗
⊗ V ))
or V -valued 1-forms. This twisted exterior derivative should produce a V -valued 2-form.
∗
As in the previous paragraph, we fix a connection ∇ for TM . For a V -valued 1-form α we
compute that ν
∂αi µ ν
(∇ ⊗ D)α = + αi ωµj + αk θij dxj ⊗ dxi ⊗ eν
ν k
∂xj
and
∂αiν

2 µ ν
Λ ((∇ ⊗ D)α) = + αi ωµj + αk θij dxj ∧ dxi ⊗ eν ,
ν k
∂xj
42
where ω and θ are the Christoffel symbols for D and ∇ respectively. Again if the connection ∇
∗
for TM is symmetric, then the anti-symmetric part
ν
2 ∂αi µ ν
Λ ((∇ ⊗ D)α) = + αi ωµj dxj ∧ dxi ⊗ eν
∂xj
is independent of the connection ∇. Similarly, if β is a V -valued r-form then
Λr+1 (∇r ⊗ D)α)
is a V -valued r + 1-form, which is again independent of ∇ as soon as ∇ is symmetric.
3.1.15 D EFINITION . Let V → M be a vector bundle with connection D and let ∇ be a symmetric
∗
connection for TM . The operator D1 : Γ(M, C ∞ (TM
∗
⊗ V )) → Γ(M, C ∞ (Λ2 (TM∗
) ⊗ V )) defined
by
D1 α := Λ2 ((∇ ⊗ D)α)
(which is independent of ∇) is called the twisted exterior derivative associated to D. More gener-
∗
ally, let ∇r denote the induced product connection for Λr (TM ) → M . The operator
Dr := Λr+1 ◦ (∇r ⊗ D) : Γ(M, C ∞ (Λr (TM
∗
) ⊗ V )) → Γ(M, C ∞ (Λr+1 (TM
∗
) ⊗ V ))
is called the twisted rth exterior derivative (for V -valued r-forms) associated to D.
If e1 , ..., er is a frame for V and x1 , ..., xm is a local coordinate system on M , then for a section
σ ∈ Γ(M, V ⊗ Λr (TM ∗
)) given locally by σ = σIµ dxI ⊗ eµ , one has (with D = Dr )
∂(σIµ ) j
Dσ = j
dx ∧ dxI ⊗ eµ + ωνµ ∧ σIν dxI ⊗ eµ
∂x
∂(σIµ ) j
= dx ∧ dxI ⊗ eµ + (−1)r σIν dxI ∧ ωνµ ⊗ eµ .
∂xj
Informally, we write
Dσ = dσ + (−1)r σ ∧ ω.
E XERCISES
3.1.1. Prove proposition 3.1.2.
3.1.2. If two frames e1 , ..., er and ẽ1 , ..., ẽr determined two connection forms A and Ã respectively,
and if G : U → GL(r, C) is the matrix of functions relating the two frames by Gei = ẽi , show that
Ã = (dG)G−1 + GAG−1 .
∗
3.1.3. Show that if D is a symmetric connection for TM → M compatible with a metric g then the
dual connection ∇ is symmetric and compatible with the metric γ for TM dual to the metric g.
3.1.4. Let (E, h) be a Hermitian vector bundle. For a frame e1 , ..., er , write s = sα eα , t = tβ eβ
and hαβ̄ := h(eα , ēβ ). Show that the Chern connection is given by the formula

¯ α eα .
Ds = ∂sα + sγ hαδ̄ ∂hγ δ̄ + ∂s
3.1.5. Show that a Hermitian metric on a Riemann surface is automatically Kähler.
43
3.2 Curvature
3.2.1 Definition
3.2.1 D EFINITION . Let E → M be a vector bundle with connection D and, in terms of some
frame, connection matrix A. The curvatures of (E, D) are the operators
Θk := Dk+1 Dk : Γ(M, C ∞ (Λk (TM

∗
) ⊗ E)) → Γ(M, C ∞ (Λk+2 (TM
∗
) ⊗ E)),
where Dj is the twisted exterior derivative associated to the connection D as in Definition 3.1.15.
Observe that if s is an E-valued k-form and f is a function then
Θk (f s) = D(f Ds + df ∧ s) = f D ◦ Ds + df ∧ Ds − df ∧ Ds = f Θk s,
so that Θk s is indeed an E-valued (k + 2)-form. But even a little more is true.

3.2.2 P ROPOSITION . There exists an End(E)-valued 2-form Ω(D) such that
Θk s = s ∧ Ω(D)
for any k = 0, 1, ... and any E-valued k-form s.

Proof. We work in a local trivialization, where we denote by A the connection matrix. We calculate
that
Θk s = Dk+1 Dk s = Dk+1 (ds + (−1)k s ∧ A)

= d(ds + (−1)k s ∧ A) + (−1)k+1 (ds + (−1)k s ∧ A) ∧ A
= (−1)k (ds ∧ A + (−1)k s ∧ dA) + (−1)k+1 ds ∧ A − (−1)k s ∧ A ∧ A
= s ∧ (dA − A ∧ A).
Thus the k-independent local endomorphism
s 7→ s ∧ (dA − A ∧ A)
agrees with D ◦ D. Since D ◦ D is globally defined, the proposition is proved.
3.2.2 Curvature of the Chern connection

Fix a holomorphic Hermitian vector bundle (E, h) → X.
First observe that since D = D1,0 + ∂¯ and ∂¯2 = 0,
D1 ◦ D = D11,0 ◦ D1,0 + D11,0 ◦ ∂¯ + ∂¯1 ◦ D1,0 .
Thus far we have used the fact that the Chern connection is complex, but not that it is metric
compatible. Metric compatibility reads as
¯
∂h(s, t) = h(D1,0 s, t) + h(s, ∂t) ¯
and ∂h(s, ¯ t) + h(s, D1,0 t).
t) = h(∂s,
44
(Note that the first of these equations is consequence of the second via complex conjugation fol-
lowed by interchange of the roles of s and t.) Since ∂ 2 = 0,
¯ − h(D1,0 s, ∂t)
0 = h(D11,0 ◦ D1,0 s, t) + h(D1,0 s, ∂t) ¯ + h(s, ∂¯1 ∂t)
¯ = h(D1,0 ◦ D1,0 s, t),
1
and thus D11,0 ◦ D1,0 = 0. Therefore
Θ = D11,0 ◦ ∂¯ + ∂¯1 ◦ D1,0 .
In particular, the curvature maps sections to twisted (1, 1)-forms, and therefore the curvature form
of the Chern connection is a (1, 1)-form.
3.2.3 P ROPOSITION . The curvature of the Chern connection of (E, h) → X is given by the formula
¯ αµ̄ ∂hβ µ̄ ).
Ωαβ = ∂(h
3.2.3 Curvature of a line bundle

Let L → M be a complex line bundle. If D is any connection for L → M then its curvature is
a section of End(E) ⊗ Λ2M → M . Since the line bundle End(E) → X is canonically trivial (see
Exercise 3.2.2), the curvature of a line bundle is a well-defined 2-form on M .
In terms of any local connection form A(D), the curvature of D is
d(A(D)).
Indeed, if the fibers are 1-dimensional then A(D) ∧ A(D) = 0. In particular, the form yields, via
the isomorphism between End(L) and the trivial bundle, a globally-defined 2-form on M . Since
locally this 2-form is d(A(D)), the curvature form is a closed form, but in general this form is not
exact.
3.2.4 R EMARK . The de Rham cohomology class
2
[Θ(D)] ∈ HdR (M, C)
of the curvature of any connection for L does not depend on the connection. This cohomology
class, called the Chern class of the line bundle L → M , is denoted c1 (L).
Suppose now that X is a complex manifold and the line bundle L → X is holomorphic. Let h
be a Hermitian metric for L → X. If ξ is a holomorphic frame for L over an open set U ⊂ X then
one can define the function
ϕξ := − log h(ξ, ξ).
The curvature of the Chern connection of h is
Θ(h) = (∂ ∂¯ log ϕ(ξ) ) ⊗ ξ ⊗ ξ ∗ ,
45
where ξ ∗ is the frame for L∗ dual to ξ. Since ξ ⊗ ξ ∗ is nowhere-zero, one can ignore it and define
the curvature of the holomorphic line bundle to be ∂ ∂¯ log ϕ(ξ) .
The reader should check that the left hand side of the latter equality is independent of the choice
of holomorphic frame. Consequently some authors (probably only the present author1 ) uses the
following global notation for Hermitian metrics of holomorphic line bundles.
N OTATION . In the sequel, a metric for a holomorphic line bundle will often be denoted e−ϕ , and
¯
its curvature will be denoted ∂ ∂ϕ.
3.2.4 Curvature of determinant bundles

3.2.5 P ROPOSITION . Let (E, DE ) → M be a vector bundle of rank r with connection, and let
(det E, Ddet E ) → M be its determinant line bundle. Then
Ω(Ddet E ) = trace(Ω(DE )).
Holomorphic hermitian vector bundles

Let E → X be a holomorphic vector bundle of rank r with Hermitian metric h. We have already
observed that the (unique) Chern connection for the metric det h for the line bundle det E → X
has connection matrix
1
A= ∂(det h).
det h
It follows that the curvature matrix of det h is
¯ log det h) = ∂ ∂(−
Ω = dA − A ∧ A = ∂(∂ ¯ log det h).
With calculations similar to those we used in the study of the connection, one can easily see that
¯
Ω = trace∂(∂hh −1
),
a fact we already know from Proposition 3.2.5
The canonical bundle

Recall that the canonical bundle KX of a complex manifold X is the line bundle det TX∗1,0 . The
local sections of KX are (n, 0)-forms, where n = dimC X.
If g is a Hermitian Riemannian metric on X, theorem 3.2.5 tells us that the curvature of the
Chern connection for (KX , det(g)) is just the trace of the curvature of (TX∗1,0 , g).
1
In fact, many authors will write e−ϕ for a Hermitian metric of a holomorphic line bundle, but they mean that this
is the local form of the metric; I like to use this notation in a global way, but my convention has its pitfalls as well.
46
For a general Hermitian metric g the latter Chern curvature has nothing to do with the curvature
of the Levi-Civita connection for g. However if the metric g is Kähler, the curvature of the Chern
connection for (KX , det(g)) is the negative of the so-called Ricci curvature of g:
(3.10) Ricci(g) = −trace(Ω(g)).
In the next paragraph, we will remind the reader of the definition, from Riemannian Geometry, of
the Ricci curvature of a Riemannian metric. We will then show that (3.10) holds when the metric
in question is Kähler.
3.2.5 Symmetry of Kähler curvature

Since the connection matrix of the Levi-Civita connection is symmetric, one can expect some
symmetry in the curvature of this connection. If the metric is Kähler, the curvature of the Levi-
Civita connection has even more symmetry. In this section we write down some of the symmetries
of the Kähler curvature.
Curvature of the Levi-Civita connection

Let (M, g) be a Riemannian manifold, and let ∇ be its Levi-Civita connection. As we have seen,
∇ is uniquely determined by the two conditions
d(g(ξ, η))(ζ) = g(∇ζ ξ, η) + g(ξ, ∇ζ η) and ∇ξ η − ∇η ξ = [ξ, η]
For all vector fields ξ, η and ζ on M . It is customary to denote by R the curvature of the Levi-Civita
connection. That is to say,
R(ξ, η)ζ = ∇ξ ∇η ζ − ∇η ∇ξ ζ − ∇[ξ,η] ζ.
The curvature tensor is locally determined by
Rijk` := g(R( ∂x∂ k , ∂x∂ ` ) ∂x∂ j , ∂x∂ i ).
It is immediately clear that for any connection (Levi-Civita or not),
(3.11) Rijk` = −Rij`k .
Since d2 = 0, the curvature Θ(D) of any metric-compatible connection D must satisfy
g(Θ(D)ξ, η) + g(ξ, Θ(D)η) = d2 (g(ξ, η) = 0,
and the latter relation is equivalent to the identities
(3.12) Rijk` = −Rjik` .
47
The symmetry of the Levi-Civita connection implies that
R( ∂x∂ k , ∂x∂ ` ) ∂x∂ j = ∇ ∂

∂ ∇ ∂ ∂xj − ∇ ∂ ∇ ∂ ∂xj
∂
∂xk ∂x` ∂x` ∂xk
= ∇ ∂ ∇ ∂ ∂x` − ∇ ∂ ∇ ∂ ∂x∂ k
∂
∂xk ∂xj ∂x` ∂xj
= R( ∂xk , ∂xj ) ∂x` − R( ∂x` , ∂x∂ j ) ∂x∂ k
∂ ∂ ∂ ∂
+∇ ∂ (∇ ∂
∂ ∂xk −∇ ∂
∂ ∂x` )
∂xj ∂x` ∂xk
= −R( ∂x∂ j , ∂x∂ k ) ∂x∂ ` − R( ∂x∂ ` , ∂x∂ j ) ∂x∂ k .
Thus we obtain the First Bianchi Identity
(3.13) Rijk` + Rik`j + Ri`jk = 0.
As a consequence, we have
1 1
2
(Rijk` + Rik`j + Ri`jk ) = 0, 2
(Rjkì + Rjik` + Rjìk ) = 0,
1 1
2
(Rkìj + Rkjì + Rkij` ) = 0, 2
(Rìjk + R`kij + R`jki ) = 0.
and adding these four equations and using (3.11) and (3.12) yields
(3.14) Rik`j = R`jik .
Curvature of the Kähler connection

We let Greek letters run through {1, ..., n} and latin letters through {1, ..., n, 1̄, ..., n̄}, and we use
the notation
∂ ∂
(3.15) ∂α = and ∂ᾱ = , 1 ≤ α ≤ n.
∂z α ∂ z̄ α
Thus
∂ ∂
∂i = and ∂ī = for i = 1, ..., n,
∂z i ∂ z̄ i
and
∂ ∂
∂i = i
and ∂ī = i for i = 1̄, ..., n̄.
∂ z̄ ∂z
We denote by Γijk the Christoffel symbols of ∇, which are defined by the relation
∇∂j ∂k = Γjik ∂j .
Now let ∇ be the Kähler connection on a Kähler manifold (X, g). In Section 3.1.3 (and specif-
ically in Remark 3.1.8) we derived the formula for the Christoffel symbols with respect to the real
coordinates, and so those Christoffel are different from the symbols we have defined in the present
paragraph. Nevertheless, as already mentioned, since
√
s1,0 ∂x∂ i = − −1s1,0 ∂y∂ i = ∂z∂ i ,
48
one does retain the symmetries of the Levi-Civita connection when working on TX1,0 instead of TX .
Since ∇ is the Chern connection of the Hermitian metric, its connection matrix is a matrix of
(1, 0)-forms, and must therefore satisfy
(3.16) Γαiγ̄ = Γᾱiγ = 0.
Additionally, appropriate use of complex conjugation yields
(3.17) Γk̄īj̄ = Γkij
Let us introduce the following notation. Let Rij be the curvature tensor with respect to the local
frame ∂1 , ..., ∂n , ∂1̄ , ...∂n̄ for TX ⊗ C and write
Rij = Rik gkj .
The matrix entries Rij are (1, 1)-forms, and so we write
Rij = Rijk`¯dz k ∧ dz̄ ` .
(Note: this convention means dz 1̄ = dz̄ 1 , etc.)

Warning: This notation differs from the notation of the previous paragraph, because we are using
the frame (3.15), and not the real frame ∂x∂ 1 , ..., ∂x∂m used there. (Nevertheless, the first Bianchi
Identity holds in this section as well, since it is defined by the same sorts of relations.)
Because the curvature is a (1, 1)-form with skew-Hermitian symmetry, we have
(3.18) Rijαβ = Rij ᾱβ̄ = 0 and Rij ᾱβ = −Rijβ ᾱ = Rij β̄α .
Now, since the Kähler connection is Chern, we have (∇1,0 )2 = 0 and ∂¯2 = 0, which reads as
(3.19) Rαβij = Rᾱβ̄ij = 0.
Since the Kähler connection is also Levi-Civita, our work from the previous paragraph shows that
(3.20) Rij`k = Rjik` = −Rkìj .
In particular, we have
(3.21) Rαβ̄ij̄ = Rij̄αβ̄ = Riβ̄αj̄ ,
where the last equality is achieved as follows:
Rij̄αβ̄ = −Riβ̄ j̄α − Riαβ̄ j̄ = −Riβ̄ j̄α = Riβ̄αj̄ .
49
Ricci curvature of the Kähler connection
Finally, we come to the promised discussion of Ricci curvature. Recall that for a Riemannian
metric the Ricci curvature is the trace
Ricci(g)ij = g k` Rik`j .
As mentioned at the end of the previous paragraph, there is some symmetry to the Ricci curvature
of a Kähler metric. Indeed, the Ricci curvature tensor satisfies
(3.22) Ricci(g)αβ = Ricci(g)ᾱβ̄ = 0 and Ricci(g)αβ̄ = Ricci(g)β ᾱ .
In fact, we claimed in the last paragraph that one has the formula
(3.23) Ricci(g)αβ̄ = −∂α ∂β̄ log det (gµν̄ ) ,
where g is the Kähler metric in question. To see this formula, we use the above symmetries as
follows.
(3.21) (3.19) (∗)
−Ricci(g)αβ̄ := −g dc̄ Rαc̄dβ̄ = − g dc̄ Rαβ̄dc̄ = g dc̄ Rαβ̄c̄d =g δγ̄ Rαβ̄γ̄δ = ∂zα ∂z̄β log det (g) ,
where (∗) holds because the metric is Hermitian, and so only has (1, 1)-parts, and the last equality
holds because, by Proposition 3.2.5,
∂ ∂¯ log det g = −trace ∂(∂gg

¯ −1
).
3.2.6 Positivity
It is important to know when the curvature of the Chern connection for a Hermitian metric on a
holomorphic vector bundle is ‘positive’. Because the curvature Θ(h) of the Chern connection of
a metric h for a holomorphic E → X is a (1, 1)-form with values in Hom(E, E) → X, there are
many ways to measure its positivity. The strongest notion of positivity is called Nakano-positivity,
the weakest notion is called Griffiths positivity, and these two notions flank an intermediate se-
quence of conditions discovered by Demailly. We shall describe all of these in the present para-
graph.
Quadratic form on E ⊗ TX1,0

Using the metric h, one defines Hermitian forms { , }h,Θ(h) on the fibers of E ⊗ TX1,0 by letting
(3.24) {v ⊗ ξ, w ⊗ η}h,Θ(h) := h(Θ(h)ξ,η̄ v, w)

1,0
for indecomposable tensors on a given fiber Ex ⊗ TX,x and extending bilinearly to the entire fiber.
50
Definitions of positivity
3.2.6 D EFINITION . Let E → X be a holomorphic vector bundle with smooth Hermitian metric h,
and fix a smooth Hermitian metric g on X.
(i) We say that h has positive curvature in the sense of Griffiths at a point x ∈ X if there exists
c > 0 such that
{v ⊗ ξ, v ⊗ ξ}h,Θ(h) ≥ ch(v, v)g(ξ, ξ)
1,0
for all v ⊗ ξ ∈ Ex ⊗ TX,x .
(ii) We say that h has positive curvature in the sense of Nakano at a point x ∈ X if there exists
c > 0 such that
( n n
) n
X X X
vj ⊗ ξj , vk ⊗ ξk ≥c h(vj , vj )g(ξj , ξj )
j=1 k=1 h,Θ(h) j=1
1,0
for all v1 ⊗ ξ1 , ..., vn ⊗ ξn ∈ Ex ⊗ TX,x , where n = min(dimC Y, RankE).
(iii) Let m be an integer between 1 and min(dimC Y, RankE). We say that h has m-positive
curvature in the sense of Demailly at a point x ∈ Y if there exists c > 0 such that
( m m
) m
X X X
vj ⊗ ξj , vk ⊗ ξk ≥c h(vj , vj )g(ξj , ξj )
j=1 k=1 h,Θ(h) j=1
1,0
for all v1 ⊗ ξ1 , ..., vm ⊗ ξm ∈ Ex ⊗ TY,x .
One defines non-negative curvature by taking c = 0, and to define negative and non-positive
curvature one simply changes the sign of c and reverses the inequalities.
If V and M are two vector spaces then the elements of E ⊗ M are linear maps from M ∗ to
V . Any such linear map has a rank, which is equal to the dimension of its image. This rank is by
definition the rank of a tensor T ∈ V ⊗ M .
Thus a holomorphic vector bundle E → X with Hermitian metric h is m-positive at x ∈ X if
for any Hermitian metric g on X there exists a positive constant c such that
{T, T }h,Θ(h) ≥ c|T |2g,h

1,0
for every T ∈ Ex ⊗ TX,x whose rank is at most m.
Duality
The Hermitian holomorphic vector bundle (E, h) is Griffiths-positive if and only if its dual (E ∗ , h∗ )
is Griffiths-negative, but that this statement ceases to be true for the stronger notions of positivity.
More precisely, the Riesz Representation Theorem provides a C-conjugate linear isometry R :
E ∗ → E defined by
hα, vi = h(v, Rα).
51
Since
¯
∇1,0 α, v + α, ∇1,0 v = ∂ hα, vi = ∂h(v, Rα) = h(∇1,0 v, Rα) + h(v, ∂Rα)

¯
= α, ∇1,0 v + R−1 ∂Rα,

v ,
one sees that

−1 ¯
∇1,0
ξ α = R ∂ξ̄ Rα, and similarly, ∂¯ξ̄ α = R−1 ∇1,0
ξ Rα.
It follows that Θ(h∗ ) = R−1 Θ(h)R, and therefore that
Θ(h∗ )ξ1 ,ξ̄2 = −R−1 Θ(h)ξ2 ,ξ̄1 R.

1,0
If we now choose v1 , ..., vm ∈ Ex and ξ1 , ..., ξm ∈ TY,x then we find
( m m
)
X X
(Rvj ) ⊗ ξj , (Rvk ) ⊗ ξk
j=1 k=1 h∗ ,Θ(h∗ )
m m
X √∗ ∗ −1 −1
X √
= h ( −1Θ(h )ξj ,ξ̄k R vj , R vk ) = − h∗ (R−1 −1Θ(h)ξk ,ξ̄j vj , R−1 vk )
j,k=1 j,k=1
m m
X √ X
=− h( −1Θ(h)ξk ,ξ̄j vj , vk ) = − {vj ⊗ ξk , vk ⊗ ξj }h,Θ(h) .
j,k=1 j,k=1
And unless the rank of v1 ⊗ ξ1 + · · · + vm ⊗ ξm is at most 1, the last quadratic form is not the form
measuring negativity of the curvature.
Positivity of line bundles

When the rank of E → X is 1, i.e., when E → X is a line bundle, all of the notions of positivity
1,0 1,0
coincide. Indeed, when the fiber of E is 1-dimensional, any map Ex ⊗ TX,x → Ex ⊗ TX,x has rank
at most one 1.
In the rank one case one therefore drops all adjectives and speaks of positivity of the curvature.
In this case the curvature of a Hermitian metric e−ϕ then is
¯ = ∂ 2ϕ
∂ ∂ϕ dz i ∧ dz̄ j ,
∂z i ∂ z̄ j
and hence the curvature of e−ϕ is (semi-)positive if and only if the Hermitian matrix
dimC (X)
∂ 2ϕ

∂z i ∂ z̄ j i,j=1
is positive (semi-)definite.
52
Criterion for Griffiths positivity
In Chapter 11 the following proposition will be very useful.
3.2.7 P ROPOSITION . The metric h for V → X is non-positive in the sense of Griffiths if and only
if for any holomorphic section s of V → X the function
log h(s, s)
is a plurisubharmonic function on X.
Proof. We calculate that
¯ h(Ω(h)s, s) h(s, s)h(∇1,0 s, ∇1,0 s) − h(∇1,0 s, s) ∧ h(s, ∇1,0 s)

(3.25) ∂ ∂ log h(s, s) = − + .
h(s, s) h(s, s)2
The second term on the right hand side of (3.25) is non-negative because of the Cauchy-Schwarz
Inequality, so we see that if h is nonpositive in the sense of Griffiths then the right hand side of
(3.25) is non-negative, i.e., log h(s, s) is plurisubharmonic.
To see the converse, it is clearly enough to work locally, i.e., to assume the vector bundle V is
trivial (but with non-trivial metric). Under the condition of triviality, given any vector v ∈ Vx there
exists a holomorphic section sv of V → X such that sv (x) = v and ∇1 sv (x) = 0. Plugging the
section sv into (3.25) yields
h(Ω(h)(x)v, v)
(∂ ∂¯ log h(sv , sv ))(x) = − ,
h(v, v)
which shows that if ∂ ∂¯ log h(sv , sv ) is plurisubharmonic then h is non-positive in the sense of
Griffiths.
E XERCISES
3.2.2. Show that if L → M is a complex line bundle then the line bundle End(L) → M is trivial.
3.2.3. Show that if L → X is a holomorphic line bundle and h1 and h2 are two Hermitian metrics
for L then the curvature forms
Θ(h1 ) and Θ(h2 )
of their Chern connections are cohomologous.
3.2.5. Show that the Hermitian quadratic form (3.24) is well-defined.
3.2.6. Confirm the computation (3.25), and show that for any vector v ∈ Vx there exists a holo-
morphic section sv of V → X near x such that sv (x) = v and ∇1 sv (x) = 0.
53
3.2.7. Let X be a complex manifold of dimension n and let E → X be a holomorphic vector
bundle of rank r with smooth Hermitian metric h. Set
k := min(r, n).
For a local coordinate z, write

2 /2
Υij̄ := (−1)n dz 1 ∧ ... ∧ dz i−1 ∧ dz i+1 ∧ ... ∧ dz n ∧ dz̄ 1 ∧ ... ∧ dz̄ j−1 ∧ dz̄ j+1 ∧ ... ∧ dz̄ n .
(a) Prove the following test for Nakano negitivity: the Chern connection for (E, h) has Nakano-
negative curvature at a point p ∈ X if and only if for every local coordinate system z at p
and every k-tuple of holomorphic sections (f1 , ..., fk ) the (n, n)-form
k
!
X
¯
∂∂ h(fi , fj )Υij̄
i,j=1
is a negative multiple of Lebesgue measure dV (z) near p.
(b) Formulate a test for Nakano positivity.
54
Part II
L2 Estimates for ∂¯
55
Lecture 4
L2 Estimates for ∂¯ in complex dimension 1
In this section we prove Hörmander’s Theorem for complex manifolds of dimension 1. We shall
give two proofs. The first is longer, and will be given only for domains in the complex plane.
However, the weighted estimates obtained will hold for very general weights. The second proof is
rather short, and applies to general Riemann surfaces, but the estimates established hold only for
smooth weights.
4.1 Domains in C
For the rest of the section, fix an open connected set Ω in C and a smooth function ϕ ∈ C ∞ (Ω).
Classically the function ϕ is called the weight function, or simply the weight.
Hilbert spaces and operators

For the moment, we add hypotheses to Ω and ϕ: we assume that Ω ⊂⊂ C, that ∂Ω is smooth and
that ϕ ∈ C ∞ (Ω). With these hypotheses in place, let us proceed to define some Hilbert spaces.
Space of functions For smooth functions f, g : Ω → C we define the inner product
Z
(f, g)o := f ḡe−ϕ dA,
Ω
and we let Hϕo denote the Hilbert space closure of C ∞ (Ω) with respect to the norm induced
by this inner product. (Note that C ∞ (Ω) is contained in Hϕo because Ω is bounded and
ϕ ∈ C ∞ (Ω).)
Space of (0, 1)-forms For (0, 1)-forms α = f dz̄ and β = gdz̄ we define the inner product
Z
1
(α, β)1 := √ α ∧ β̄e−ϕ = (f, g)o ,
2 −1 Ω
and we let Hϕ1 denote the Hilbert space closure of C ∞ (Ω)dz̄ with respect to the norm
induced by this inner product.
56
The ∂¯ operator
Next we turn to the definition of the Hilbert space ∂¯ operator. The ∂¯ operator as we know it up to
now sends a smooth function f to the (0, 1)-form ∂f ¯ = ∂f dz̄. We extend ∂¯ to a densely defined
∂ z̄
operator (denoted with the same symbol)
∂¯ : Hϕo → Hϕ1
defined as follows: the domain of ∂¯ consists of all f ∈ Hϕo such that the (0, 1)-current ∂f¯ is
represented by integration against (an automatically unique) α ∈ Hϕ , i.e., there exists α ∈ Hϕ1
1
such that
Z Z
1 ∂ψ
(4.1) √ α ∧ ψdz = − f dA for all ψ ∈ Co∞ (Ω),
2 −1 Ω Ω ∂z
¯ := α for f ∈ Domain(∂).
and then we define ∂f ¯
4.1.1 P ROPOSITION . The densely defined operator ∂¯ : Hϕo → Hϕ1 is closed.

Proof. The operator ∂¯ is closed, i.e., Graph(∂)¯ is closed, if and only if for any fj ∈ Domain(∂) ¯
such that fj → f in Hϕo and ∂f ¯ j → α in H 1 , α = ∂f ¯ in the sense of currents, i.e., (4.1) holds.
ϕ
Let us fix such a sequence fj . Then for all g ∈ Co (Ω)
∞
Z Z
1 −ϕ 1 ¯ j ∧ gdz̄e−ϕ
(α, gdz̄)1 = √ α ∧ gdz̄e = lim √ ∂f
2 −1 Ω 2 −1 Ω
Z Z
∂ −ϕ −ϕ ∂ −ϕ
= lim fj −e ϕ (e g) e dA = f −e ϕ (e g) e−ϕ dA.
Ω ∂z Ω ∂z
But this equation means that (4.1) holds for ψ = e−ϕ g. Since g 7→ e−ϕ g is an isomorphism of
Co∞ (Ω), we are done.
The Hilbert space and formal adjoints of ∂¯

For reasons that will become clear soon, one wants to define the Hilbert space adjoint ∂¯∗ of the
¯ The domain of ∂¯∗ is
densely defined operator ∂.
Domain(∂¯∗ ) := {α ∈ Hϕ1 ; ∃Cα > 0 such that |(α, ∂g)

¯ 1 | ≤ Cα ||g||o for all g ∈ Domain(∂)},
¯
and ∂¯∗ α is the unique element of Hϕo corresponding to the linear functional
¯ α) ∈ C
`α : Co∞ (Ω) 3 g 7→ (∂g,
under the Riesz Representation Theorem. In particular,
(∂¯∗ α, g)o = (α, ∂g)

¯ 1.
It is a general fact that the Hilbert space adjoint of a densely defined (resp. closed) operator is
closed (resp. densely defined), and that the double adjoint of a closed densely defined operator is
57
the operator itself. These elementary facts make it possible to formulate an adjoint (weak) version
of the ∂¯ equation.
To establish an estimate we will need later on, it is useful to have a formula for ∂¯∗ on a dense
subspace of Hϕ1 . With such a formula as our goal, let α = f dz̄ be a smooth (0, 1)-form on Ω
and let g ∈ C ∞ (Ω). Recalling our temporary assumption that Ω is bounded, ∂Ω is smooth and
ϕ ∈ C ∞ (Ω), we compute that
Z
¯ 1 ¯ −ϕ
(α, ∂g)1 = √ α ∧ ∂ge
2 −1 Ω
Z
1
= √ α ∧ dḡe−ϕ
2 −1 Ω
Z Z
1 −ϕ 1
= √ ḡd(e α) − √ d(e−ϕ αḡ)
2 −1 Ω 2 −1 Ω
Z Z
1 ∂ −ϕ 1
f ḡe−ϕ dz̄

= √ ḡ fe dz ∧ dz̄ − √
2 −1 Ω ∂z 2 −1 ∂Ω
√
−1
Z
= −eϕ ∂z∂
(e−ϕ f ), g o + ḡe−ϕ α.

2 ∂Ω
4.1.2 D EFINITION . The operator ϑ : C ∞ (Ω)dz̄ → C ∞ (Ω) defined by

∂f ∂ϕ
ϑ(f dz̄) := −eϕ ∂z
∂
(e−ϕ f ) = − + f
∂z ∂z
¯
is called the formal adjoint of ∂.
4.1.3 R EMARK . As one can see from the calculation preceding Definition 4.1.2, the formal adjoint
of ∂¯ can equally be defined as the operator ϑ : C ∞ (Ω)dz̄ → C ∞ (Ω) that satisfies
¯ 1
(ϑα, ψ)o = (α, ∂ψ)
for all ψ ∈ Co∞ (Ω).

With the formal adjoint in hand, we have the following proposition.
4.1.4 P ROPOSITION . Let Ω ⊂ C be a bounded domain with smooth boundary and let ϕ ∈ C ∞ (Ω).
Then a form α ∈ C ∞ (Ω)dz̄ lies in Domain(∂¯∗ ) if and only if α|∂Ω ≡ 0, and in this case
∂¯∗ α = ϑα.
Proof. Let α ∈ C ∞ (Ω)dz̄. For g ∈ C ∞ (Ω) we have calculated that

√
−1
Z
(4.2) ¯ 1 = (ϑα, g) +
(α, ∂g) ḡe−ϕ α.
o
2 ∂Ω
Hence if α|∂Ω ≡ 0 then

¯ 1 | = |(ϑα, g)o | ≤ ||ϑα||o ||g||o ,
|(α, ∂g)
58
which shows that α ∈ Domain(∂¯∗ ), and the formula (α, ∂g)
¯ 1 = (ϑα, g) implies that ∂¯∗ α = ϑα.
o
Conversely, suppose α ∈ Domain(∂¯∗ ). Fix functions χj ∈ Co∞ (Ω) such that 0 ≤ χj ≤ 1 and
for each compact set K ⊂ Ω there exists J ∈ N such that if j ≥ J then χj |K ≡ 1. Then by the
definition of Domain(∂¯∗ )
¯ j g))1 | ≤ Cα ||χj g||o
|(α, ∂(χ
for some constant Cα that is independent of g or j. Therefore by Dominated Convergence
¯ 1 = lim(α, ∂(χ
(α, ∂g) ¯ j g))1 = lim(ϑα, χj g)o = (ϑα, g)
j j
for all g ∈ Hϕo . In particular, if g ∈ C ∞ (Ω) then by (4.2)

Z
ḡeϕ α = 0,
∂Ω
so α|∂Ω ≡ 0.
The formal identity and the basic estimate

We continue to assume that Ω is connected, bounded, and smoothly bounded, and that ϕ ∈ C ∞ (Ω).
Let α = f dz̄ be a smooth (0, 1)-form on Ω. We compute that

¯ = ∂¯ − ∂f ∂ϕ
∂ϑα + f
∂z ∂z
∂ 2f ∂ 2ϕ

∂ϕ ∂f
= − + + f dz̄
∂z∂ z̄ ∂z ∂ z̄ ∂z∂ z̄
∂ 2ϕ

ϕ ∂ −ϕ ∂f
= −e e dz̄ + α
∂z ∂ z̄ ∂z∂ z̄
∂
e−ϕ ∂f ¯ but since α is a (0, 1)-form ∂α ¯ = 0.

It is tempting to write the term −eϕ ∂z ∂ z̄
as ϑ∂α,
Therefore we need a better interpretation for this term if we are to extend these ideas to higher
dimensions. The insight that seems most natural is to view α not as a differential form, but rather
as a section of a complex vector bundle.
¯
Recall that the ∂-operator, when acting for sections, is well-defined only on a holomorphic
vector bundle. If we think of the form α = f dz̄ not as a form, but rather as a section of the
(trivial, so in this case, holomorphic) vector bundle TΩ∗0,1 , then we can act with ∂¯ on this section,
and produce the section of (TΩ∗0,1 )⊗2 given by
¯ = ∂f dz̄ ⊗2 .
∇α
∂ z̄
¯ even when T ∗0,1 is not trivial, as long as one has a
(Later we will see that there is a way to define ∇ Ω
Hermitian Riemannian metric on Ω. The present case corresponds to the Euclidean metric on our
domain Ω ⊂ C.)
59
Let us define an inner product on smooth sections of (TΩ∗0,1 )⊗2 by
Z
⊗2 ⊗2
(gdz̄ , hdz̄ )2 := f ḡe−ϕ dA.
Ω
With this inner product in hand, one can compute (Exercise 4.2.2) that the formal adjoint of the
¯ : Γ(Ω, T ∗0,1 ) → Γ(Ω, (T ∗0,1 )⊗2 ) is given by the formula
operator ∇ Ω Ω
(4.3) ¯ ∗ (gdz̄ ⊗2 ) = −eϕ ∂ (e−ϕ g)dz̄.

∇
∂z
Thus one obtains the following theorem.
4.1.5 T HEOREM . Suppose Ω is a bounded, smoothly bounded domain and ϕ ∈ C ∞ (Ω). Then
2
(4.4) ¯ =∇
∂ϑα ¯ + ∂ ϕα
¯ ∗ ∇α
∂z∂ z̄
for any smooth (0, 1)-form α on Ω.
Now let α be a smooth form on Ω lying in the domain of ∂¯∗ . Then α vanishes on ∂Ω, so
¯
(∂ϑα, α)1 = (∂¯∂¯∗ α, α)1 = ||∂¯∗ α||2o and ¯ ∗ ∇α,
(∇ ¯ α)1 = ||∇α||
¯ 22 .
Theorem 4.1.5 therefore implies the following key result.
4.1.6 T HEOREM (Bochner-Kodaira Identity). Suppose Ω is a bounded, smoothly bounded domain

and ϕ ∈ C ∞ (Ω). Then for any smooth (0, 1)-form α in the domain of ∂¯∗ one has the identity
∂ 2ϕ
Z
¯∗ ¯ 1
(4.5) 2 2
||∂ α||o = ||∇α||2 + √ α ∧ ᾱe−ϕ .
2 −1 Ω ∂z∂ z̄
Density of smooth forms

The domain of ∂¯∗ is a dense subspace of Hϕ1 , but we can endow it with another norm, namely
|||α|||2 := ||α||21 + ||∂¯∗ α||2o .
Let us denote by F the inner product space with norm ||| · ||| whose underlying vector space is
Domain(∂¯∗ ).
4.1.7 R EMARK . Since ∂¯∗ is also a closed operator, F is in fact a Hilbert space. We will not use
the completeness of ||| · ||| in this paragraph.
We have the following theorem.
4.1.8 T HEOREM . The smooth forms in Domain(∂¯∗ ) are dense in F .
60
We begin with some important preliminaries. Fix a form α ∈ F . First, we extend α by 0 to
C − Ω. This extended form, which we denote by α̃, can be seen as a form in L2òc (C), and we want
to say that, in some sense, α̃ is still in the domain of ∂¯∗ . The trouble is that ∂¯∗ itself, the Hilbert
¯ is intimately tied to the domain Ω. However, the formal adjoint is given by
space adjoint of ∂,
some formula, namely ∂¯∗ = L, where
∂f ∂ϕ
L(f dz̄) = − + f.
∂z ∂z
Since ϕ ∈ C ∞ (Ω), we can assume that ϕ ∈ Co∞ (C) and define the operator L as a densely defined
operator on L2 (C, e−ϕ dA). We can now make a more reasonable statement about α̃.
4.1.9 P ROPOSITION . The form α is in the domain of ∂¯∗ if and only if Lα̃ ∈ Hϕo in the sense of
distributions, and in that case
Lα̃ = ∂¯∗ α.
Proof. For a (0, 1)-form α in Hϕ1 we have the identity

Z Z
¯ 1 := √ 1 ¯ −ϕ 1 ¯ −ϕ = (Lα̃)(h)
(α, ∂h) α ∧ ∂he = √ α̃ ∧ ∂he
2 −1 Ω 2 −1 C
where the last equation on the right hand side is the meaning of Lα̃ in the sense of distributions.
Thus if Lα̃ ∈ Hϕo then α ∈ Domain(∂¯∗ ), and then ∂¯∗ α = Lα̃.
Conversely, suppose α ∈ Domain(∂¯∗ ). Then ∂¯∗ α is well-defined in Hϕo and
¯ 1 = (∂¯∗ α, h)o ,
(Lα̃)(h) = (α, ∂h)
which means that Lα̃ = ∂¯∗ α in the sense of distributions.

Our next goal is to localize the problem using partitions of unity. Let us choose a collection of
open disks {Uj } that cover Ω, and another collection of open disks {Vk } that cover ∂Ω, and are
sufficiently small that ∂Vk and ∂Ω intersect transversely. BecauseS 1 ∂Ω is compact, we can choose
a finite subset V1 , ..., VN1 that cover ∂Ω. Then K := Ω − N j=1 Vj is closed and bounded, so
compact, and thus we can choose U1 , ..., UNo that cover K. All together, {Uj , Vk } cover Ω. We
denote this cover by {Wi } when we don’t want to distinguish between the Uj and the Vk . Let us
choose a partition of unity {χν }Psubordinate to this cover, i.e., for each index i there exists ν such
that Support(χν ) ⊂⊂ Wi , and ν χν ≡ 1 on the union of the Wi .
We can now write X
α= χν α,
ν
and each χν α is compactly supported in some Wi . We need the following lemma.
4.1.10 L EMMA . If α ∈ Domain(∂¯∗ ) and χ ∈ C ∞ (Ω) then χα ∈ Domain(∂¯∗ ).
61
¯ we have
Proof. For any g ∈ Domain(∂)
Z
¯ 1 = (α, ∂(χg))
¯ 1 ∂χ −ϕ
(χα, ∂g) 1+ √ ḡ f e dA,
2 −1 Ω ∂z
where α = f dz̄. Thus since α ∈ Domain(∂¯∗ )

∂χ
¯ 1 | ≤ Cα ||χg||o +
|(χα, ∂g) || ∂χ f ||o ||g||o ≤ Cα sup |χ| + ||α||1 sup ||g||o ,
∂z
Ω Ω ∂z
which is what we wanted to show.

Thus we are reduced to studying a (0, 1)-form α in the domain of ∂¯∗ and whose support is
either a compact subset of Uj , or is the intersection of Ω and a compact subset of Vk .
In the first, interior case, we can use the usual mollifier method: let ψ ∈ Co∞ (D) and set
ψδ (z) = δ −2 ψ(z/δ). Then for δ > 0 sufficiently small αδ := α ∗ ψδ is smooth with compact
support in Uj and therefore lies in the domain of ∂¯∗ . Moreover, αδ and ∂¯∗ αδ can be made arbitrarily
close to α and ∂¯∗ α in Hϕ1 and Hϕo respectively by choosing δ > 0 sufficiently small.
In the second, boundary case, we need to be a little more careful. Let Kk be the intersection of
a cone in C with a small disk centered at the origin. If the cone angle and the radius of the disk
are sufficiently small, and if the cone’s axis is at an appropriate angle (for example, approximately
parallel to the normal direction of some boundary point of Ω in Vk ), then for each point z ∈ ∂Ω∩Vk
the truncated cone z +Kk lies in the complement of Ω, while the truncated cone z −Kk lies entirely
in Ω, except for its vertex, which is of course on the boundary. (We must make sure that the disk
Vk is sufficiently small, which we may R assume is the case, without loss of generality.) Choose a
function ψ (k+)
∈ Co (Kk ) such that C ψ
∞ (k+)
dA = 1, and let ψ (k−) (z) = ψ (k+) (−z).
(k−)
Now suppose that α ∈ Domain(∂¯∗ ) is compactly supported in Vk . Let αδ := α∗ψδ . Writing
α = f dz̄, we have Z
α (z) = dz̄ · f (z − δζ)ψ (k−) (ζ)dA(ζ).
δ
C
In particular, if z ∈ ∂Ω and ζ ∈ Support(ψ (k−) ) then −ζ ∈ Kk and thus z − δζ 6∈ Ω, which means

that f (z − δζ) = 0. In other words, αδ vanishes on ∂Ω, and is therefore in the domain of ∂¯∗ .
By standard real analysis αδ → 1Ω α =: α̃ in L2 (C), and ∂¯∗ αδ → Lα̃ in the sense of distribu-
tions. Thus by Proposition 4.1.9 ∂¯∗ αδ → ∂¯∗ α in Hϕo .
End of the proof of Theorem 4.1.8. Having written α = ν χν α, we can find smooth forms (χν α)δ
P
as just outlined, with each (χν α)δ vanishing on ∂Ω and therefore lying in the domain of ∂¯∗ , and
moreover such that
(χν α)δ → χν α and ∂¯∗ (χν α)δ → ∂¯∗ (χν α)
in Hϕ1 and Hϕo respectively. The proof is complete.
4.1.11 C OROLLARY. The identity (4.5) holds for all α in the domain of ∂¯∗ .
62
Hörmander’s Theorem on domains in C
We are now ready to state and prove Hörmander’s Theorem.
4.1.12 T HEOREM (Hörmander’s Theorem for domains in C). Let Ω ⊂ C be a domain and let
ϕ ∈ L1òc (Ω) be a function such that ϕ(z) − c|z|2 is subharmonic in Ω for some c > 0. Then for
any measurable function f on Ω such that
Z
|f |2 e−ϕ dA < +∞
Ω
there exists a locally integrable function u such that

∂u
=f
∂ z̄
in the sense of currents, and
Z Z
2 −ϕ 1
|u| e dA ≤ |f |2 e−ϕ dA.
Ω c Ω
Proof. Choose domains Ωj , j = 1, 2, ..., with smooth boundary, such that

[
Ωj ⊂⊂ Ωj+1 and Ωj = Ω.
j
Let us fix j for now. At the end of the proof we will let j → ∞.
Define the function ϕδ := ϕ ∗ (δ −2 ψ(·/δ)), where ψ is a radial function on C with compact
support and total integral 1, and δ > 0 is less than the Euclidean distance from Ωj to C − Ω. Then
ϕδ is smooth, subharmonic, and decreasing to ϕ, and we have
∂2 ∂ 2ψ z − ζ
Z
2 z−ζ dA(ζ)
(ϕδ (z) − c|z| ) = ϕ(ζ) − cψ ≥ 0.
∂z∂ z̄ C ∂ζ∂ ζ̄ δ δ δ2
On the domain Ωj we have the Hilbert spaces
Hjo := Hϕoδ and Hj1 := Hϕ1δ ,
as well as the closed, densely defined operators ∂¯ : Hjo → Hj1 and the Hilbert space adjoint
∂¯∗ : Hj1 → Hjo . Then Corollary 4.1.11 applies, and thus from (4.5) we obtain
||∂¯∗ β||2o ≥ c||β||21
for all (0, 1)-forms β in the domain of ∂¯∗ . Let
α := f dz̄.
Then for all β in the domain of ∂¯∗ we have

||α||21 ¯∗ 2
(4.6) |(α, β)1 |2 ≤ ||α||21 ||β||21 ≤ ||∂ β||o .
c
63
We claim that the estimate (4.6) provides a solution of the ∂¯ equation with estimates. To see this
claim let λ : Image(∂¯∗ ) → C be the linear functional defined by
λ(∂¯∗ β) := (β, α)1 .
The estimate (4.6) means that λ is continuous on Image(∂¯∗ ) (and hence on its closure). Setting λ
equal to 0 on the orthogonal complement Hjo Image(∂¯∗ ) makes λ a continuous linear functional
on Hjo with operator norm ||λ||2∗ ≤ c−1 ||α||21 . By the Riesz Representation Theorem there exists
uδ,j ∈ Hjo such that λ(h) = (h, uδ,j ) and ||uδ,j || = ||λ||∗ . Applying the first identity to h = ∂¯∗ β
shows that
¯ δ,j = α, ∂uδ,j
∂u i.e., = f,
∂ z̄
and the equality ||uδ,j || = ||λ||∗ , together with the monotonicity of ϕδ with respect to δ, yields
Z Z Z Z
2 −ϕδ 1 2 −ϕδ 1 2 −ϕ 1
|uδ,j | e dA ≤ |f | e dA ≤ |f | e dA ≤ |f |2 e−ϕ dA.
Ωj c Ωj c Ωj c Ω
In fact, since ϕδ is decreasing in δ, for all δo > 0 sufficiently small and all δ ≤ δo we have
Z Z Z
2 −ϕδo 2 −ϕδ 1
|uδ,j | e dA ≤ |uδ,j | e dA ≤ |f |2 e−ϕ dA.
Ωj Ωj c Ω
Thus all the {uδ,j } lie in a ball of radius c−1/2 ||f || in Hϕoδo . By Alaoglu’s Theorem there is a
sequence δk → 0 such that uδk ,j converges in Hϕoδo and
Z Z
2 −ϕδo 1
|uδk ,j | e dA ≤ |f |2 e−ϕ dA.
Ωj c Ω
On the other hand, for all k ≥ `

Z Z Z
2 −ϕδ` 2 −ϕδk 1
|uδk ,j | e dA ≤ |uδk ,j | e dA ≤ |f |2 e−ϕ dA,
Ωj Ωj c Ω
so we can again extract a convergent subsequence in Hϕoδ . Taking the diagonal subsequence, we
`
find that it converges in Hϕoδ for all `. Thus we have a limit uj such that
`
Z Z
2 −ϕδk 1
|uj | e dA ≤ |f |2 e−ϕ dA.
Ωj c Ω
By the Monotone Convergence Theorem

Z Z
2 −ϕ 1
|uj | e dA ≤ |f |2 e−ϕ dA.
Ωj c Ω
¯ j = α in the weak sense, i.e.,

Note that since uj is a limit of functions uk,j satisfying ∂u
(uk,j , ∂¯∗ ϕ)o = (α, ϕ)1
64
for any smooth (0, 1)-form α with compact support in Ωj , we can take limits to see that
¯ j=α
∂u
in the weak sense.

Let us now extend the functions uj from Ωj to Ω by setting them equal to 0 outside Ωj . Then
we have the estimates Z Z
2 −ϕ 1
|uj | e dA ≤ |f |2 e−ϕ dA.
Ω c Ω
We can now apply the same technique of Alaoglu’s Theorem to the index j, and extract a subse-
quence converging to a function u ∈ Hϕo on Ω such that
Z Z
2 −ϕ 1
|u| e dA ≤ |f |2 e−ϕ dA.
Ω c Ω
Now, for each smooth (0, 1)-form ψ with compact support in Ωjo and any j ≥ jo we have
(uj , ∂¯∗ ψ)o = (α, ψ)1 .
Thus we can let j → ∞ and obtain the equation

∂u
= f.
∂ z̄
The proof of Hörmander’s Theorem is complete.
4.2 L2 estimates for ∂¯ on Riemann surfaces

There is another proof of Theorem 4.1.12 that, though rather similar at in the most crucial aspects,
is rather more streamlined in its technicalities. In this paragraph we present this proof, as well as a
slight generalization going beyond domains in C to the setting of Riemann surfaces.
This generalization forces us to introduce the geometric setting of Hörmander’s Theorem. In-
deed, compact Riemann surfaces have no nonconstant subharmonic functions, but all Riemann
surfaces have holomorphic line bundles with metrics of positive curvature. Thus we formulate and
prove Hörmander’s Theorem for Hilbert spaces of sections of holomorphic line bundles, rather
than functions.
4.2.1 T HEOREM (Hörmander’s Theorem for Riemann surfaces). Let X be a Riemann surface with
Hermitian metric g whose metric form is ω. Let L → X be a holomorphic line bundle with smooth
metric e−ϕ such that √
−1(∂ ∂ϕ¯ + R(g)) ≥ cω,
where R(g) is the curvature of the Levi-Civita connection for g. Then for any L-valued measurable
(0, 1)-form α on X such that Z
1
√ α ∧ ᾱe−ϕ < +∞
2 −1 X
65
¯ = α in the sense of currents, and
there exists a measurable section u of L → X such that ∂u
Z Z
2 −ϕ 1 1
(4.7) |u| e ω ≤ √ α ∧ ᾱe−ϕ .
X c 2 −1 X
Proof. We use the square norms appearing on the left and right hand sides of (4.7) respectively
to define Hilbert spaces H o and H 1 for sections of L → X and L-valued (0, 1)-forms on X
respectively.
Let β be a smooth L-valued (0, 1)-form on X and let h be a smooth section of L → X with
compact support. We define the operator ϑ : Γo (X, TX∗0,1 ) → Co∞ (X), by
¯ 1,
(ϑβ, h)o = (β, ∂h)
where Γo (X, TX∗0,1 ) is the vector space of compactly supported smooth (0, 1)-forms on X. Then
Z
¯ 1 ¯ −ϕ
(β, ∂h)1 = √ β ∧ ∂he
2 −1 X
√
−1
Z Z Z
−ϕ 1 −ϕ 1
−eϕ ∂(e−ϕ β) h̄e−ϕ .

= d(e β h̄) − √ ∂(e β)h̄ = √
2 X 2 −1 X 2 −1 X
√
−1
Writing β = f dz̄ and ω = e−ψ 2
dz ∧ dz̄ in some local coordinate z on which the line bundle
L → X is trivial, we have
1 1 ∂ ∂
√ eϕ ∂(e−ϕ β) = √ eϕ (e−ϕ f )dz ∧ dz̄ = −eϕ+ψ (e−ϕ f )ω.
2 −1 2 −1 ∂z ∂z
Thus
ϕ+ψ ∂ −ϕ ψ ∂f ∂ϕ
ϑβ = −e (e f ) = e − + f .
∂z ∂z ∂z
The reader can check that the right hand side is independent of the coordinate system chosen.
4.2.2 R EMARK . The aforementioned check can be done by seeing how the right hand side trans-
forms with a change of coordinates, but there is a geometric way to see it as well. Given the section
β of L ⊗ TX∗0,1 → X, the metric e−ϕ , which identifies the fibers of L with the conjugates of their
∗
duals, allows us to define a section e−ϕ β of the line bundle L ⊗ TX∗0,1 → X. Since the latter
line bundle is anti-holomorphic, it has a well-defined ∂ operator, so we have a section ∂(e−ϕ β) of
∗
L ⊗ TX∗1,0 ⊗ TX∗0,1 → X. But the line bundle TX∗1,0 ⊗ TX∗0,1 → X has a nowhere zero section,
∗
namely the metric g, so it is trivial. We thus obtain a section eψ ∂(e−ϕ β) of L → X. Finally, using
the metric e−ϕ again, we have a section eϕ+ψ ∂(e−ϕ β) of L → X, and this section is denoted ϑβ.
In the same vein as the Remark 4.2.2, given our section β of TX∗0,1 ⊗ L → X, there exists an
L-valued (1, 0)-vector field Vβ (meaning Vβ is a section of TX1,0 ⊗ L → X) such that
β = Vβ yω,
66
where, for a vector ξ and a form α, ξyα denotes the contraction of α with ξ. In local coordinates,
if β = f dz̄ for a local section f of L then
∂
Vβ = f eψ .
∂z
¯ β as a section of
Since Vβ is a section of a holomorphic line bundle, it makes sense to compute ∂V
1,0 ∗0,1 ¯ ∗0,1
TX ⊗ TX ⊗ L. We think of ∂Vβ as a (1, 0)-vector field with values in TX ⊗ L, and as such
there is a section of (TX∗0,1 )⊗2 ⊗ L, which we denote ∇β
¯ and which we think of as a (0, 1)-form
∗0,1
with values in TX ⊗ L, such that
¯ β yω = ∇β.
∂V ¯
In local coordinates, we have

¯ dz̄) = e−ψ ∂ eψ f dz̄ ⊗2 = ∂f ∂ψ
dz̄ ⊗2 .

∇(f + f
∂ z̄ ∂ z̄ ∂ z̄
We can define an inner product structure on the space of compactly supported smooth sections of
L ⊗ (TX∗0,1 )⊗2 → X defined as follows. If ξi = hi dz̄ ⊗2 then
Z Z
−ϕ
(ξ1 , ξ2 )2 := ψ
e h1 h̄2 e = hξ1 , ξ2 ig e−ϕ ω,
X X
where h·, ·ig is the metric for (TX∗0,1 )⊗2 → X induced by the metric g for T 1,0 → X. (Locally, if
¯ ∗ defined by
ξi = hi dz̄ ⊗2 then hξ1 , ξ2 ig = h1 h̄2 e2ψ .) We can then calculate the adjoint operator ∇
¯ ∗ ξ, β)1 = (ξ, ∇β)

(∇ ¯ 2.
The reader can check that

¯ ∗ (f dz̄ ⊗2 ) = −eψ+ϕ ∂ (e−ϕ f )dz̄.
∇
∂z
Again the reader can check directly that the right hand side is an L-valued (0, 1)-form, but we
encourage a derivation along the lines of Remark 4.2.2.
We compute that

¯ = ∂ ψ ∂f ∂ϕ
∂ϑβ e − + f dz̄
∂ z̄ ∂z ∂z
∂ 2f ∂ 2ϕ

ψ ∂f ∂ψ ∂ϕ ∂f ∂ϕ ∂ψ
=e − − + + + f dz̄,
∂z∂ z̄ ∂z ∂ z̄ ∂z ∂ z̄ ∂z ∂ z̄ ∂z∂ z̄
while

¯∗¯ = −e ∂ ψ+ϕ −ϕ ∂f ∂ψ
∇ ∇β e + f dz̄
∂z ∂ z̄ ∂ z̄
∂ 2f ∂ 2ψ

ψ ∂ψ ∂f ∂f ∂ϕ ∂ϕ ∂ψ
=e − − + + − f dz̄.
∂z∂ z̄ ∂ z̄ ∂z ∂ z̄ ∂z ∂z ∂ z̄ ∂z∂ z̄
67
It follows that
∂ 2ϕ ∂ 2ψ

¯ −∇
∂ϑβ ¯ ∗ ∇β
¯ = eψ + β,
∂z∂ z̄ ∂z∂ z̄
and thus Z
||ϑβ||2o ¯ 2 ¯ + R(g))β, β e−ϕ ω.

= ||∇β||2 + Θg (∂ ∂ϕ g
X
The hypotheses of Theorem 4.2.1 therefore imply that
||∂¯∗ β||2o ≥ c||β||21
for all β ∈ Γo (X, L ⊗ TX0,1 ).

Now let α be the L-valued (0, 1)-form in the hypotheses of Theorem 4.2.1. Define the linear
functional λα : ∂¯∗ (Γo (X, L ⊗ TX0,1 )) → C by
λα (∂¯∗ β) := (α, β)1 .
Then
1
||λα ||2 := sup |(α, β)1 |2 ≤ ||α||21 < +∞,
||∂¯∗ β||o =1 c
so λα is bounded on the (non-closed) subspace ∂¯∗ (Γo (X, L ⊗ TX0,1 )) of H o . By the Hahn-Banach
Theorem (or simply by extending by 0 in the orthogonal complement) we can assume that the
linear functional λα ∈ H o∗ is bounded with norm at most c−1/2 ||α||. By the Riesz Representation
Theorem there is a section u ∈ H o such that λα (v) = (u, v)o and
Z Z
2 −ϕ 1 1
|u| e ω ≤ √ α ∧ ᾱe−ϕ .
X c 2 −1 X
Restricting to ∂¯∗ (Γo (X, L ⊗ TX0,1 )) shows that
(u, ∂¯∗ β)o = (α, β)1 ,

¯ = α in H 1 . The proof of Theorem 4.2.1 is complete.
i.e., that ∂u
4.2.1 Regularity of solutions

In Theorems 4.1.12 and 4.2.1 the following question is left unanswered. Suppose that α is a smooth
form on Ω and that u is the solution of the equation ∂u¯ = α. Is u smooth? The answer to this
question is yes. Indeed, any two solutions of the ∂¯ equation differ by a holomorphic function,
and so either all of the solutions are smooth or none are smooth. But for domains in C there is
a standard solution, using Green’s Theorem, to the inhomogeneous Cauchy-Riemann equations.
This solution is given by an integral formula, and one can read off it that the solution is smooth
when the data is smooth. Hence the positive answer. These ideas are more fully developed in
Exercise 4.2.4.
68
E XERCISES
4.2.1. Prove that the Hilbert space adjoint ∂¯∗ of the ∂¯ operator, defined on page 57, is closed and
densely defined, and that (∂¯∗ )∗ = ∂¯ as densely defined operators.
¯
4.2.2. Establish Formula (4.3) for the formal adjoint of the operator ∇.
4.2.3. Let ϕ be a strictly plurisubharmonic function (i.e., such that ∆ϕ ≥ c > 0 in the sense of
currents) on a smoothly bounded domain Ω ⊂⊂ C. Show that if α is a smooth (0, 1)-form on Ω
that vanishes on ∂Ω and satisfies ∂¯∗ α = 0 then α = 0. Show that there is a smooth (0, 1)-form
α 6≡ 0 on Ω such that ∂¯∗ α ≡ 0.
4.2.4. Let Ω ⊂⊂ C be a domain with smooth boundary.
(a) Use Green’s Theorem to show that for any z ∈ Ω and any function f ∈ C ∞ (Ω) one has the
formula Z Z
1 f (ζ)dζ 1 ∂f ∂A(ζ)
f (z) = √ − .
2π −1 ∂Ω z − ζ π Ω ∂ ζ̄ z − ζ
(b) Show that if g ∈ Cok (Ω) then the function

Z
1 g(ζ)∂A(ζ)
u(z) := −
π Ω z−ζ
∂u
satisfies the equation ∂ z̄
= g in Ω.
(c) Let X be a Riemann surface and let α be a smooth (0, 1)-form on X. Show that any function
¯ = α is smooth.
u such that ∂u
69
Lecture 5
The Bochner-Kodaira Identity
We fix a Kähler manifold (X, g) and a holomorphic vector bundle E → X with Hermitian metric
h. We let ω denote the metric form of g. We denote by Θ the curvature of the Chern connection
associated to (E, h) and by R the curvature operator of the Kähler connection associated to (X, g).
These curvature operators induce operators on E-valued (p, q)-forms, whose definitions will be
apparent from the derivation. We shall denote the induced operators by the same letters.
5.1 The Hilbert spaces

Because of Remark 1.4.7 we shall work only with E-valued (n, q)-forms.
Let ϕ be an (n, q)-form with values in E. If we choose local coordinates z and a local frame
e1 , ..., er for E, then we may write
ϕ = ϕαJ¯dz 1 ∧ ... ∧ dz n ∧ dz̄ J ⊗ eα .
Let
gij := g(∂i , ∂j̄ ) and hαβ̄ = h(eα , eβ ).
As usual, g ij̄ and hαβ̄ denote the inverse matrices; they are the matrices of the metrics for the dual
bundles, with respect to the dual frame.
The metrics g and h induce a metric on E ⊗ Λn,q X . In terms of the above metrics the new metric
is given locally by
1 ¯
hϕ, ψi := ϕαJ¯ψL̄β hαβ̄ g LJ det(g ij̄ ),
q!
¯
where g LJ = g `1 j̄1 · · · g `q j̄q . It is easy to check that the right hand side is independent of all the
frames that were chosen. When we want to indicate the dependence on the metrics g and h, we
shall write hϕ, ψig,h .
¯
5.1.1 R EMARK . It is sometimes convenient to employ the notation ψᾱI := ψL̄β hβ ᾱ g J L̄ , with respect
to which one has hϕ, ψi = q!1 ϕαJ¯ψᾱJ .
70
5.1.2 D EFINITION . Let (E, h) → X be a Hermitian complex vector bundle and let dV be a volume
element, i.e., a smooth (n, n)-form with no zeros, on a Hermitian manifold (X, g).
(i) For two compactly supported E-valued (p, q)-forms ϕ and ψ, we define the inner product
Z
(ϕ, ψ) = hϕ, ψi dVg .
X
With this inner product, the space of compactly supported smooth E-valued (p, q)-forms is
an inner product space.
(i) We let L2p,q (g, h) denote the Hilbert space completion of the inner product space of smooth
compactly supported E-valued (p, q)-forms, with the inner product (·, ·).
5.1.3 R EMARK . Note that dVg = det(gij̄ )dV (z), and therefore the (n, n)-form
1 α β ¯
(5.1) {ϕ, ψ} := ϕJ¯ψL̄ hαβ̄ g LJ
q!
In the case q = 0 the (n, n)-form {ϕ, ψ} depends only on the metric h, and not on the metric g.
5.2 ¯
Two ∂-operators and their formal adjoints
A smooth E-valued (n, q)-form can be viewed in two ways. In the first, usual way, such a form is a
(n, q)-form with values in a holomorphic vector bundle. From the second perspective, a (n, q)-form
is a section of the vector bundle Λn,q
X ⊗ E → X; a complex vector bundle that is not holomorphic.
In this paragraph we will define natural ∂¯ operators for these two types of sections.
Let us begin with the first perspective. We fix a basis {eα } of local holomorphic sections of
E → X. To indicate the type of the forms we are using, it is helpful to include subscripts for our
multi-indices. Let us denote by ∇ the Chern connection associated to E-valued (p, q)-forms on
our Kähler manifold (X, g).
5.2.1 P ROPOSITION . Let ϕ = ϕαJ¯q dz 1 ∧ · · · ∧ dz n ∧ dz̄ Jq ⊗ eα be an E-valued (n, q)-form. Then
q
X
¯ = (−1)n
∂ϕ (−1)k (∇j̄k ϕ)αj̄ ...ˆj̄ dz 1 ∧ ... ∧ dz n ∧ dz̄ J ⊗ eα .
0 k ...j̄q
k=0
Proof. By definition of ∂¯ we have

¯ = (∂ϕ
∂ϕ ¯ α¯ ) ∧ dz 1 ∧ · · · ∧ dz n ∧ dz̄ Jq ⊗ eα .
Jq
Then
¯ = (−1)n ∂j̄ ϕα 1 n j Jq
∂ϕ j̄0 ...j̄q ∧ dz ∧ · · · ∧ dz ∧ dz̄ ∧ dz̄ ⊗ eα
q
X
n
= (−1) (−1)k ∂j̄k ϕαj̄ ...ˆj̄ dz 1 ∧ · · · ∧ dz n ∧ dz̄ j0 ∧ ... ∧ dz̄ jq
0 k ...j̄q
k=0
q
X
= (−1)n (−1)k (∇j̄k ϕ)αj̄ ...ˆj̄ dz 1 ∧ · · · ∧ dz n ∧ dz̄ j0 ∧ ... ∧ dz̄ jq ,
0 k ...j̄q
k=0
71
where the last equality holds since ∇0,1 = ∂¯ for the Chern connection.
¯ Thus d is defined by the relation
We denote by d the formal adjoint of ∂.
¯
(dϕ, ψ) = (ϕ, ∂ψ)
for all smooth twisted forms ϕ, and all smooth twisted forms ψ with compact support. Note that d
sends smooth E-valued (n, q)-forms to smooth E-valued (n, q − 1)-forms.
5.2.2 P ROPOSITION . Let ϕ = ϕαJ¯q dz 1 ∧ · · · ∧ dz n ∧ dz̄ Jq ⊗ eα be an E-valued (n, q)-form. Then
dϕ = (−1)n+1 g ij̄ (∇i ϕ)αj̄j̄1 ...j̄q−1 dz 1 ∧ ... ∧ dz n ∧ dz̄ j1 ∧ ... ∧ dz̄ jq−1 ⊗ eα .
Proof. One has

¯
(dϕ, ψ) = (ϕ, ∂ψ)
Z q
1 X 0
= α
ϕj̄1 ...j̄q (−1) n
(−1)k+1 (∇j 0 ψ)β0 b0 0 g Jq Jq hαβ̄
n!q! X k=1
k j1 ...jk ...jq
Z q
1 X 0 α 0
jk0 j̄k 0
= (−1) n+1
(−1)k+1 g jk j̄k ∇jk0 ϕ j̄ ...j̄ ψ β0 b0 0 g j1 j̄1 ...gd ...g jq j̄q hαβ̄
n!q! X 1 q j1 ...jk ...jq
Z k=1
1 0
= (−1)n+1 g ij̄ (∇i ϕ)αj̄j̄1 ...j̄q−1 ψjβ0 ...j 0 g Jq−1 Jq−1 hαβ̄ ,
n!(q − 1)! X 1 q−1
Where the second-to-last inequality follows from the metric compatibility of the connection.
V perspective of a V -valued (n, q)-form seen as a section of the

Now let us turn to the second
complex vector bundle KX ⊗ q TX∗0,1 ⊗ E → X. As we have pointed out, the latter vector
bundle is not holomorphic. However, because X is equipped with a metric, there is a natural
way toVmap sections of KX ⊗ q TX∗0,1 ⊗ E → X to sections of the holomorphic vector bundle
V
KX ⊗ q TX1,0 ⊗ V → X. In terms of local data, if we write ϕ = ϕαJ¯dz 1 ∧ ... ∧ dz n ∧ dz̄ J ⊗ eα then
¯ ∂ ∂ I J¯ α ∂
Fϕ := g I J ϕαJ¯dz 1 ∧ ... ∧ dz n ⊗ ∧ ... ∧ ⊗ eα =: g ϕJ¯dz 1
∧ ... ∧ dz n
⊗ ⊗ eα
∂z i1 ∂z iq ∂z I
is a section of KX ⊗ q TX1,0 ⊗ E → X, and clearly the map ϕ 7→ Fϕ is a 1-1 correspondence that
V
depends only the pointwise values of ϕ. The map F extends
Vqon 1,0 Vk naturally to a 1-1
Vq correspondence
∗0,1
Fk of KX ⊗ TX ⊗E-valued (0, k)-forms, i.e., sections of (TX )⊗KX ⊗ TX1,0 ⊗E → X,
by acting trivially on the last factor, i.e., for J = (j1 , ..., jk ) set
Fk dz̄ J ⊗ dz 1 ∧ ... ∧ dz n ∧ dz̄ I ⊗ eα = dz̄ J ⊗ F dz 1 ∧ ... ∧ dz n ∧ dz̄ I ⊗ eα .

The vector bundle KX ⊗ q TX1,0 ⊗ E → X,

V
being a holomorphic vector bundle, is equipped
¯ and thus we have a well-defined KX ⊗ Vq T 1,0 ⊗ E-valued (0, 1)-form ∂Fϕ.
with ∂, ¯
X
72
¯ : Γ(X, (Λn,q ) ⊗ E) → Γ(X, T ∗0,1 ⊗ (Λn,q ) ⊗ E) is defined by
5.2.3 D EFINITION . The operator ∇ X X X
¯
¯ := F−1 ∂Fϕ.
∇ϕ 1
In terms of local data
¯ := gI L̄ ∂ g I J¯ϕα¯ dz̄ k ⊗ dz 1 ∧ ... ∧ dz n ∧ dz̄ L ⊗ eα = (∇k̄ ϕ)α¯ dz̄ k ⊗ dz 1 ∧ ... ∧ dz n ∧ dz̄ I ⊗ eα ,

∇ϕ J I
∂ z̄ k
where ∇ denotes the Kähler-Chern connection for E-valued tensors.
We can compute the formal adjoint ∇ ¯ ∗ of ∇¯ with respect to the inner products induced on
∗0,1
sections of (Λn,q TX∗ ) ⊗ E and TX ⊗ (Λn,q TX∗ ) ⊗ E) by the metrics h and g. For smooth sections
ϕ and ψ of KX ⊗ (Λq TX∗0,1 ) ⊗ TX∗0,1 ⊗ V and KX ⊗ Λq TX∗0,1 ⊗ V respectively, with ψ compactly
supported, and we write
ϕ = ϕαJ,¯ j̄ dz̄ j ⊗ dz 1 ∧ ... ∧ dz n ∧ dz̄ J ⊗ eα
then Z Z
¯
(ϕ, ∇ψ) = g kj̄
ϕαJ,¯ j̄ (∇k̄ ψ)Jα¯dV (z) = −g kj̄ (∇k ϕ)αJ,¯ j̄ ψαJ¯dV (z),
X X
√
where dV (z) = ( 2−1 )n dz 1 ∧ dz̄ 1 ∧ ... ∧ dz n ∧ dz̄ n and we have used the fact that the Chern
connection is compatible with the metric. Thus
¯ ∗ ϕ)α¯ = −g kj̄ (∇k ϕ)α¯
(∇ J J,j̄
¯ ∗ ϕ = 0.
When viewing ϕ as a section rather than a form, one sets ∇
¯ its Laplace-Beltrami operator ∇
We can therefore associated to ∇ ¯ ∗∇
¯ on sections of the vector
n,q
bundle (ΛX ) ⊗ E → X. Our computations show that
(5.2) ¯ ∗∇
∇ ¯ = −g kj̄ ∇k ∇j̄ .
5.3 The formal identity

The following identity is crucial in establishing L2 estimates for solutions of the ∂¯ equation.
5.3.1 T HEOREM (The formal Bochner-Kodaira Identity). Let (X, g) be a Kähler manifold and let
E → X be a holomorphic vector bundle with Hermitian metric h. Denote by = ∂d ¯ + d∂¯ the
¯ Then one has the formal identity
Laplace-Beltrami operator associated to ∂.
(5.3) ¯ ∗∇
=∇ ¯ + Θg (h),
where Θg (h) is the operator on E-valued (n, q)-forms induced by the curvature of the metric h.
73
Proof. Using Propositions 5.2.1 and 5.2.2, we calculate that

¯ α n+1 ij̄ ¯ α 1 n
(d∂ϕ)j̄1 ...j̄q = (−1) g ∇i (∂ϕ)j̄ J¯dz ∧ · · · ∧ dz ∧ dz̄ ⊗ eα J

= (−1)n+1 g ij̄ ∇i (−1)n (∇j̄ ϕ)αJ¯dz 1 ∧ · · · ∧ dz n ∧ dz̄ J ⊗ eα
q
X
n
+(−1) (−1)k (∇j̄k ϕ)αj̄j̄ dz 1 ∧ · · · ∧ dz n ∧ dz̄ J ⊗ eα
1 ...j̄k ...j̄q
b
k=1
q
!
X
= (−g ij̄ ∇i ∇j̄ ϕ)αj̄1 ...j̄q + (g ij̄ ∇i ∇j̄k ϕ)αj̄1 ...(j̄)k ...j̄q dz 1 ∧ · · · ∧ dz n ∧ dz̄ J ⊗ eα ,
k=1
and
q
X
¯ α
(∂dϕ) n
(−1)k+1 (∇j̄k (dϕ))αj̄ ...b̄j
j̄1 ...j̄q = (−1) 1 k ...j̄q
k=1
q
X
= (−1)n (−1)k+1 (∇j̄k ((−1)n+1 g ij̄ ∇i ϕ))αj̄j̄
1 ...j̄k ...j̄q
b
k=1
q
X
=− (∇j̄k (g ij̄ ∇i ϕα ))j̄1 ...(j̄)k ...j̄q
k=1
q
X
=− g ij̄ (∇j̄k ∇i ϕ)αj̄1 ...(j̄)k ...j̄q ,
k=1
Thus q
¯
X
(ϕ)αJ¯q = −g ij̄ (∇i ∇j̄ ϕ)αJ¯q + g i` ([∇i , ∇j̄k ]ϕ)αj̄1 ...(`)
¯ k ...j̄q .
k=1
Finally, since [∇i , ∇j̄ ] = Θ(h)ij̄ is the curvature,

q q
i`¯ ¯
X X
g i` Θ(h)αβij̄k ϕβj̄1 ...(`)
α
g ∇i , ∇j̄k ϕj̄1 ...(`)
¯ k ...j̄q = ¯ k ...j̄q .
k=1 k=1
The right hand side of the last equation is by definition the operator Θg (h) induced in E-valued
(n, q)-forms by the curvature of h:
q
¯
X
(5.4) Θg (h)ϕ := g i` Θ(h)αβij̄k ϕβj̄1 ...(`) 1 n j1 jq
¯ k ...j̄q dz ∧ · · · ∧ dz ∧ dz̄ ∧ · · · ∧ dz̄ ⊗ eα .
k=1
The proof is complete.
74
Exercises
5.3.1. Let X be a compact complex manifold and let L → X be a holomorphic line bundle with a
smooth Hermitian metric h whose Chern connection has positive curvature. Show that if a smooth,
L-valued (n, 1)-form α satisfies α = 0 then α = 0.
75
Lecture 6
L2 Estimates on Compact Kähler manifolds
¯ Before
In this section we present our first, and simplest, result on L2 estimates for solutions of ∂.
beginning, we should say a word on the meaning of the solution of ∂¯ in spaces of not necessarily
smooth sections, such as L2p,q (g, h). In fact, the ∂¯ operator can act on currents, and in particular
on any locally integrable sections. In the space the distribution equation ∂u¯ = ϕ, ϕ ∈ L2 (g, h),
p,q
2
formulates as follows. A section u ∈ Lp,q−1 (g, h) is a weak solution of the equation ∂u¯ = ϕ if
(u, dψ) = (ϕ, ψ)

for all smooth, compactly supported ψ ∈ L2p,q (g, h). In coming paragraphs we shall go further,
discussing the structure of ∂¯ as a closed, densely defined operator on L2p,q (g, h).
6.1 Hörmander’s Theorem on compact Kähler manifolds

6.1.1 T HEOREM . Let (X, g) be a compact Kähler manifold and let F → X be a holomorphic
vector bundle with Hermitian metric h. Fix p, q with 0 ≤ p ≤ n and 1 ≤ q ≤ n. Assume that
the curvature of the Chern connection of the metric vector g (p) ⊗ det(g) ⊗ h for the vector bundle
V p 1,0 ∗
TX ⊗ KX ⊗ F satisfies
(6.1) Θg (g (p) ⊗ det g ⊗ h) ≥ c · Id
¯
for some positive constant c. Then for every F -valued ∂-closed (p, q)-form ϕ satisfying
Z
|ϕ|2g,h dVg < +∞
X
there exists a F -valued (p, q − 1)-form u satisfying

Z Z
¯ 2 1
∂u = ϕ and |ϕ|g,h dVg ≤ |ϕ|2g,h dVg .
X c X
6.1.2 R EMARK . Note that the condition (6.1) holds if the Hermitian metric g (p) ⊗ det g ⊗ h has
Nakano-positive curvature. However, as we shall see, if q > 1 then (6.1) can hold under weaker
assumptions.
76
Proof of Theorem 6.1.1. Let E := p TX1,0 ⊗KX ∗
V
⊗F . Then an F -valued (p, q)-form identifies with
an E-valued (n, q)-form. Moreover, the metric h := g (p) ⊗ det g ⊗ h identifies the corresponding
L2 structures, and the curvature assumption is equivalent to the existence of a positive constant c
such that
Θ(h) ≥ c · Id.
Let ϕ be an E-valued (n, q)-form with finite L2 norm. For any smooth, E-valued (n, q)-form
ψ we have the estimate
1 ¯ 2
(6.2) ||ψ||2 ≤ ||∂ψ|| + ||∂¯∗ ψ||2 .
c
Now consider the bilinear form (·, ·)H defined on smooth sections of Λn,q
X ⊗ E → X by
¯ 1 , ∂ψ
(ψ1 , ψ2 )H := (∂ψ ¯ 2 ) + (∂¯∗ ψ1 , ∂¯∗ ψ2 ).
The inequality (6.2) implies that || · ||H is a norm. Define H to be the Hilbert space completion
of Γ(X, Λp,q TX∗ ⊗ E) with respect to the norm || · ||H .
Let λϕ : L2p,q (g, h) → C be the linear functional defined by
λϕ (ψ) := (ψ, ϕ).
By Cauchy-Schwarz
||ϕ||2
|λϕ (ψ)|2 ≤ ||ψ||2H ,
c
and thus λϕ ∈ H ∗ . By the Riesz Representation Theorem there exists v ∈ H such that
||ϕ||2
||v||2H ≤ and (v, ψ)H = λϕ (ψ), ψ∈H.
c
The latter means that
¯ ∂ψ)
(∂v, ¯ + (∂¯∗ v, ∂¯∗ ψ) = (ϕ, ψ), ψ∈H.
In particular, the latter holds for smooth ψ.
Now, since ∂ϕ¯ = 0 we have
0 = (ϕ, ∂¯∗ ψ) = (∂v,

¯ ∂¯∂¯∗ ψ) + (∂¯∗ v, ∂¯∗ ∂¯∗ ψ) = (∂v,
¯ ∂¯∂¯∗ ψ),
which implies that ∂¯∂¯∗ ∂v

¯ = 0 in the sense of currents. Therefore ||∂¯∗ ∂v||
¯ 2 = (∂¯∂¯∗ ∂v,
¯ ∂v)
¯ = 0,
¯ ¯
and thus (∂v, ∂ψ) = 0. Hence
(∂¯∗ v, ∂¯∗ ψ) = (ϕ, ψ),
which means that u := ∂¯∗ v is a solution of ∂u¯ = ϕ. Moreover
||ϕ||2
||u||2 = ||∂¯∗ v||2 = ||v||2H ≤ .
c
The proof is therefore complete.
77
6.2 A few useful applications
Regularity and cohomology vanishing
Theorem 6.1.1 provides L2 solutions to the ∂¯ equation with L2 data, but leaves open an important
question: If the data ϕ is a smooth E-valued (p, q)-form, is the solution provided by Theorem 6.1.1
also smooth.
The answer to this question is affirmative. However, except in the case q = 1, the reason is
a bit subtle. For general q, the solution we provided is of the form u = dv, and any section of
¯
this form is orthogonal to the ∂-closed ¯
twisted forms. Since any two solutions differ by a ∂-closed
twisted form, our solution is the one of minimal norm. Consequently v is a (weak) solution of the
equation v = ϕ. The operator is an elliptic system, and general machinery, which we have not
discussed, implies the desired smoothness. However, when q > 1 there are always other solutions
¯ = ϕ that are not smooth.
of the equation ∂u
The case q = 1 is slightly different. In this case, all solutions are either smooth or not. Indeed,
any two solutions differ by a holomorphic section. So to show regularity, one has only to show
that there exists some smooth solution.PThe smoothness of the solution is a local problem: using a
partition of unity, one can write ϕ = χj ϕj , and if we can find smooth solutions with compact
support uj of the equations ∂u ¯ j = ϕj then the section u = P uj provides the needed smooth
solution. To then find the smooth solution of the localized equation, one can work in Cn . In this
case there are a number of available techniques, and at least one of them, which is based on Green’s
Theorem, is rather elementary; it is often taught in first courses in complex analysis. The approach
is to produce an integral formula for the solution, and the smoothness of the solution can be read
off from the formula. Therefore, for q = 1, any section u of F → X that satisfies ∂u ¯ = ϕ is
smooth when this is the case for ϕ.
As a corollary of this discussion, one obtains the following important proposition.
6.2.1 P ROPOSITION . Under the hypotheses of Theorem 6.1.1, the Dolbeault cohomology groups
Kernel ∂¯ : Γ(X, Λp,q p,q+1

p,q X ⊗ F ) → Γ(X, ΛX ⊗ F)
H∂¯ (X, F ) :=
∂¯ Γ(X, Λp,q−1 ⊗ F )

X
are trivial, i.e., H∂p,q

¯ (X, F ) = {0}.
Kodaira Vanishing
Let L → X be a holomorphic line bundle on a complex manifold X of complex dimension n, and
let e−ϕ be a smooth Hermitian metric for L. The positivity of the curvature ∂ ∂ϕ ¯ of e−ϕ means
precisely that the (1, 1)-form √
−1 ¯
ω := ∂ ∂ϕ
2π
is a Kähler form. (Thus, a posteriori, every complex manifold admitting a line bundle with a metric
of positive curvature is necessarily Kähler.) We therefore can, and do, use the Kähler form ω to
define our Kähler metric g on X.
78
Suppose β is a smooth, L-valued (n, q)-form. In terms of a local frame ξ for L, if we write
β = βJ¯dz 1 ∧ · · · ∧ dz n ∧ dz̄ J then we compute that
Θg (e−ϕ )β
q
¯
X
= 2π g i` gij̄k βj̄1 ...(`) 1 n j1
¯ k ...j̄q dz ∧ · · · ∧ dz ∧ dz̄ ∧ · · · ∧ dz̄
jq
k=1
= 2πqβ
By Proposition 6.2.1 one has the following theorem.
6.2.2 T HEOREM (Kodaira Vanishing Theorem). Let X be a compact Kähler manifold and let
L → X be a holomorphic line bundle admitting a smooth metric of positive curvature. Then
H∂n,q
¯ (X, L) = {0}.
Kodaira-Nakano Vanishing
Another direct corollary of Proposition 6.2.1 is the following generalization of Kodaira Vanishing.
The proof is left to the reader.
6.2.3 T HEOREM (Kodaira-Nakano Vanishing Theorem). Let X be a compact Kähler manifold and
let E → X be a holomorphic vector bundle admitting a smooth metric whose curvature is positive
in the sense of Nakano. Then
H∂n,q
¯ (X, E) = {0}.
Vanishing theorems for q > 1

Let (X, ω) be a Kähler manifold.
√ Suppose L → X is a line bundle with smooth Hermitian metric
−ϕ ¯
e whose curvature θ = −1∂ ∂ϕ is not necessarily positive. Let
λ1 (p) ≤ λ2 (p) ≤ ... ≤ λn (p)
1,0
denote the eigenvalues of θp relative to ωp , thought of as Hermitian forms on TX,p . (These eigen-
values depend on the point p, but nature of this dependence will not be important to us.) Fix a
1,0 1,0∗
correspoinding ω-orthonormal basis ξ1 (p), ..., ξn (p) of eigenvectors for TX,p , and let αi ∈ TX,p be
defined by
1,0
αi (v) = ω(ξi , v), v ∈ TX,p .
Given an L-valued (n, q)-form β, we can write β = βJ¯α1 ∧ ... ∧ αn ∧ ᾱJ ⊗ e. We compute that
q
¯
X
−ϕ
Θg (e )β = g i` θij̄k βj̄1 ...(`) 1 n J
¯ k ...j̄q α ∧ ... ∧ α ∧ ᾱ ⊗ e
k=1
q
¯
X
= δ i` λi δij̄k βj̄1 ...(`) 1 n J
¯ k ...j̄q α ∧ ... ∧ α ∧ ᾱ ⊗ e
k=1
= (λj1 + ... + λjg )βJ¯α1 ∧ ... ∧ αn ∧ ᾱJ ⊗ e.
79
√ √ P
αi ∧ ᾱi and θ = −1 i λi αi ∧ ᾱi , we have
P
Since ω = −1 i
√ X
θ ∧ ω q−1 = ( −1)q (q − 1)! (λj1 + · · · + λjq )αj1 ∧ ᾱj1 ∧ · · · ∧ αjq ∧ ᾱjq
|J|=q
and √ X
ω q = ( −1)q q! αj1 ∧ ᾱj1 ∧ · · · ∧ αjq ∧ ᾱjq .
|J|=q
By Theorem 6.1.1 we have the following theorem.
6.2.4 T HEOREM . Let (X, ω) be a compact Kähler manifold and let L → X be a holomorphic line
bundle admitting a smooth metric with curvature θ. Fix q ∈ {1, ..., n} and suppose
θ ∧ ω q−1 ≥ cω q
for some positive constant c. Then

H∂n,q
¯ (X, L) = {0}.
There are analogous vanishing theorems for vector bundles, but they are a little more elaborate,
involving the notion of m-positivity discussed in Chapter 1. We shall not formulate them here.
80
Lecture 7
L2 Estimates on Complete Kähler manifolds
7.1 Extending the ∂¯ operator to Hilbert spaces

7.1.1 Closed, densely defined operators
Let H1 and H2 be Hilbert spaces. We are interested in linear operators from subspaces of H1 into
H2 . Two important things to keep in mind about linear operators are
• each operator T : H1 → H2 comes with its own domain of definition Domain(T ), and
• T : Domain(T ) → H2 need not be continuous (with respect to the relative topology of

Domain(T ) ⊂ H1 ).
It may sometimes be possible to extend an operator to a larger domain. If this is so, and S is such
an extension, then the Graph
Graph(T ) := {(x, T x) ; x ∈ Domain(T )} ⊂ H1 × H2
of T is a subspace of the graph Graph(S) of S.
7.1.1 D EFINITION . An operator T : H1 → H2 is said to be
1. densely defined if Domain(T ) is a dense subspace of H1 , and
2. closed if Graph(T ) is a closed subspace of H1 × H2 .
7.1.2 R EMARK . If an operator is bounded on a Hilbert space (or more generally on a Banach
space), then the Closed Graph Theorem tells us that it is closed.
As the reader will recall, a Hilbert space is isomorphic to its dual. The explicit theorem yielding
the isomorphism is a rather elementary version of the Riesz Representation Theorem. The result
states that the map sending v ∈ H to the bounded linear functional λv ∈ H ∗ define by
λv w := hw, vi ,
81
is a conjugate linear isomorphism that is also an isometry:
||λv ||H ∗ = ||v||.
The next objective on our agenda is to define the adjoint of a linear operator T : H1 → H2 as
an operator T ∗ : H2 → H1 . To define the adjoint operator, we begin by defining its domain. We
let Domain(T ∗ ) consist of all η ∈ H2 such that
Lη : x 7→ hT x, ηi2
is a continuous linear functional on Domain(T ). By the continuity of Lη , one can extend Lη to
the closure of Domain(T ). We can also extend Lη to the orthogonal complement of Domain(T )
by setting
Lη (y) = 0 for all y ∈ Domain(T )⊥ .
By the Riesz representation theorem Lη is represented by a unique element T ∗ η of H1 satisfying
hT x, ηi2 = hx, T ∗ ηi1 , x ∈ Domain(T ).
The linearity of η 7→ T ∗ η is trivial to prove.
7.1.3 R EMARK . Extending Lη by zero to Domain(T )⊥ might initially seem rather arbitrary. How-
ever, if λ is any extension of Lη and λo denotes the extension by zero to Domain(T )⊥ then λ − λo
vanishes on Domain(T ). Thus, writing u = v + w ∈ Domain(T ) ⊕ Domain(T )⊥ ,
||λ||2 = sup |λ(v + w)|2 = sup |λo (v) + (λ − λo )(w)|2 = ||λo ||2 + ||λ − λo ||2 .
|v|2 =|w|2 =1 |v|2 =|w|2 =1
We see that λo is the extension of minimal norm, which does seem rather more natural.
7.1.4 P ROPOSITION . If T : H1 → H2 is densely defined (resp. closed) then T ∗ : H2 → H1 is
closed (resp. densely defined).
Proof. We denote by πi : H1 × H2 → Hi the projection to the ith factor, i = 1, 2, and by
h , i1,2 = π1∗ h , i1 + π2∗ h , i2 the inner product on H1 × H2 . Let F : H1 × H2 → H1 × H2 be
given by
F (ξ, η) = (−ξ, η).
Observe that (ξ, η) ⊥ F (Graph(T )) if and only if
hx, ξi1 = hT x, ηi2
for all x ∈ Domain(T ). Note that while the latter means that η ∈ Domain(T ∗ ), we cannot conclude
that ξ = T ∗ η unless we also know that ξ ⊥ Domain(T )⊥ .
Now suppose Domain(T ) is dense. Then Domain(T )⊥ = {0}, so (ξ, η) ⊥ F (Graph(T )) if
and only if η ∈ Domain(T ∗ ) and T ∗ η = ξ. In other words,
F (Graph(T ))⊥ = Graph(T ∗ ),
and thus T ∗ is closed.
Next suppose T is closed. Let η ⊥ Domain(T ∗ ). Then the vector (0, η) lies in Graph(T ∗ )⊥ =
F (Graph(T )) = F (Graph(T )) = F (Graph(T )). Thus η = T 0 = 0, and from Lemma 7.1.5 below
and the Riesz Representation Theorem we conclude that T ∗ is densely defined.
82
7.1.5 L EMMA . Let A be a subspace of a topological vector space H. Then A = H if and only if
every continuous linear functional on H that vanishes on A is zero.
Proof. If A = H then clearly every continuous linear functional on H that vanishes on A is zero.
Conversely, let v ∈ H − A. The linear functional on the closed subspace Cv + A of H defined by
tv + a 7→ t is continuous and vanishes on A. By the Hahn-Banach Theorem, this linear functional
extends to a continuous linear functional ` ∈ H 0 . Evidently `|A ≡ 0 and ` 6≡ 0.
7.1.6 P ROPOSITION . If T is a closed, densely defined operator, then T ∗∗ = T .
Proof. We know that T ∗ is also closed and densely defined, and we have
hx, T ∗ ηi = hT x, ηi
whenever x ∈ Domain(T ) and η ∈ Domain(T ∗ ). From the boundedness of η 7→ hT x, ηi we

therefore conclude that Domain(T ) ⊂ Domain(T ∗∗ ).
Let θ ∈ Domain(T ∗∗ ). Since T is densely defined there exist Domain(T ) 3 xj → θ. Then for
all η ∈ Domain(T ∗ ) we have
hT ∗∗ θ, ηi = hθ, T ∗ ηi = lim hxj , T ∗ ηi = lim hT xj , ηi .
Thus T xj → T ∗∗ θ in H2 , and therefore Graph(T ) 3 (xj , T xj ) → (θ, T ∗∗ θ) in H1 × H2 . Since T

is closed, (θ, T ∗∗ θ) ∈ Graph(T ), and thus θ ∈ Domain(T ) and T ∗∗ θ = T θ.
7.1.2 The maximal extension of ∂¯

Let (X, g) be a Hermitian manifold and let E → X be a holomorphic vector bundle with smooth
Hermitian metric h. By our definition of L2p,q (g, h) the subspace Γo (X, Λp,q 2
X ⊗ E) of Lp,q (g, h)
consisting of smooth compactly supported E-valued (p, q)-forms is dense.
We have the operator ∂¯ : Γo (X, Λp,q TX∗ ⊗ E) → Γo (X, Λp,q+1 TX∗ ⊗ E) defined on smooth
sections, and we extend ∂¯ to a densely-defined operator L2p,q (g, h) → L2p,q+1 (g, h) as follows.
Since every element of L2p,q (g, h) is trivially locally integrable, we can extend ∂¯ as on operator
on all of L2p,q (g, h) in the sense of currents: for ϕ ∈ L2p,q (g, h),
¯
∂ϕ(ψ) := (ϕ, dψ), ψ ∈ Γo (X, Λp,q+1 TX∗ ⊗ E).
We can construct a densely defined extension of ∂¯ if we take the domain of the extension of ∂¯
to be any subspace H of L2p,q (g, h) containing Γo (X, Λp,q TX∗ ⊗ E) such that for each ϕ ∈ H the
¯ is represented by integration against some Fϕ ∈ L2 (g, h):
E-valued (p, q + 1)-current ∂ϕ p,q+1
¯
∂ϕ(ψ) = (Fϕ , ψ) for all ψ ∈ Γo (X, Λp,q+1 ⊗ E).
X
¯ ∈ L2 (g, h) in the sense of currents, or

(We will shorten the terminology and simply say that ∂ϕ p,q+1
¯ ∈L
simply that ∂ϕ 2
(g, h).) Each such subspace H yields a different densely defined operator.
p,q+1
In this chapter, we will take the so-called maximal extension Tp,q of ∂¯ defined by the domain
¯ ∈ L2 (g, h)}.
Domain(Tp,q ) := {ϕ ∈ L2p,q ; ∂ϕ p,q+1
83
7.1.7 P ROPOSITION . The operator T = Tp,q is closed.
Proof. Let {ϕj } ⊂ Domain(T ) be a sequence that converges to some ϕ ∈ Domain(T ) such that
T ϕj → Φ in L2p,q+1 (g, h). Then for all η ∈ Γo (X, Λp,q+1
X ⊗ E) we have
(T ϕ − Φ, η) = lim(T ϕ − T ϕj , η) = lim(ϕ − ϕj , ∂¯∗ η) = 0.
Thus T ϕ = Φ.
∗
7.1.3 The Hilbert space adjoint Tp,q
The maximal extension Tp,q of ∂¯ has a closed, densely defined Hilbert space adjoint Tp,q
∗
whose
domain is
∗
Domain(Tp,q ) = {α ∈ L2p,q+1 (g, h) ; |(α, Tp,q ψ)| ≤ Cα ||ψ|| for all ψ ∈ Domain(Tp,q )}.
Below we will need the following simple proposition.

7.1.8 P ROPOSITION . If χ ∈ Co∞ (X) and ϕ ∈ Domain(Tp,q
∗
) ∩ Domain(Tp,q+1 ) then
∗
χϕ ∈ Domain(Tp,q ) ∩ Domain(Tp,q+1 ).
Moreover,
¯ ∧ ϕ and
Tp,q+1 (χϕ) = χTp,q+1 ϕ + ∂χ ∗
Tp,q (χϕ) = χT ∗ ϕ − grad0,1 χ̄yϕ,
where, for a function f , grad0,1 f is the (0, 1)-vector field defined by
g(ξ, grad0,1 f¯) = ∂f (ξ), ξ ∈ TX1,0 .
Proof. We compute that, as currents,

¯
∂(χϕ) ¯ ∧ ϕ + χTp,q+1 ϕ,
= ∂χ
so clearly χϕ ∈ Domain(Tp,q+1 ) and the formula for Tp,q+1 holds. Next, if ψ ∈ Domain(Tp,q )
then the calculation just completed shows that χψ ∈ Domain(Tp,q ). Thus
|(χϕ, Tp,q ψ)| = |(ϕ, Tp,q (χ̄ψ) − ∂¯χ̄ ∧ ψ)|

≤ Cϕ ||χ̄ψ|| + ||ϕ|| · ||∂¯χ̄ ∧ ψ||
≤ (Cϕ sup |χ| + ||ϕ|| sup |∂χ|)||ψ||,
X
∗
which shows that χϕ ∈ Domain(Tp,q ) and
(χϕ, Tp,q ψ) = (ϕ, Tp,q (χ̄ψ)) − (ϕ, ∂¯χ̄ ∧ ψ) = (χTp,q

∗
ϕ, ψ) − (grad0,1 χ̄yϕ, ψ),
∗
and the formula for Tp,q follows.
84
7.2 Complete Kähler manifolds
7.2.1 Complete Riemannian manifolds and exhaustion functions
Recall that in Riemannian geometry one has the notion of a complete (connected) Riemannian
manifold (M, g): The Riemannian metric g induces a distance function, with the distance δg (x, y)
between two points x, y ∈ M being the infimum of the lengths of any two paths connecting those
two points. (In general, a curve realizing this infimum, which is called the minimizing geodesic,
need not exist.) The underlying manifold with this distance function is a metric space, and we say
that a Riemannian manifold is complete if this induced metric space is complete.
The celebrated Hopf-Rinow Theorem says that the completeness property of the Riemannian
manifold is equivalent to the condition that for each xo ∈ M the function
ψo : M 3 x 7→ δg (x, xo ) ∈ [0, ∞)
is proper. In general the function ψo is not smooth, but it is Lipschitz with constant 1, and thus
it is almost everywhere differentiable. Moreover, |dψo |g ≤ 1. We can therefore smooth ψo to a
function ψ satisfying
|dψ|g ≤ 2 and |ψ(x) − ψo (x)| ≤ 1.
The function ψ : M → [−1, ∞) is then also proper.
It is not hard to show, on the other hand, that the Riemannian manifold (M, g) is complete
if there exists a smooth proper function ψ : M → [A, ∞) on M such that |dψ|g is uniformly
bounded.
7.2.1 R EMARK . Note that the constant A plays no role in the completeness of (M, g), but we must
have it if we insist on using the word ‘proper’. In complex analysis, it is customary to avoid this
trivial issue by introducing the notion of an exhaustion function.
We summarize the discussion in the following proposition.
7.2.2 P ROPOSITION . A Riemannian manifold (M, g) is complete if and only if there exists an
exhaustion function ψ ∈ C ∞ (M ) such that |dψ|g ≤ 1.
7.2.2 Approximation of twisted (p, q)-forms on complete Kähler manifolds

For the rest of the chapter, we employ the following convention: we fix p and q, and let
T := Tp,q−1 and S := Tp,q .
Since the operator of primary interest is T , while the operator S comes in to take care of the
compatibility condition for the data, it is useful to distinguish these two operators more clearly, as
this notation does.
On a complete Kähler manifold forms in Domain(S) ∩ Domain(T ∗ ) can be approximated by
smooth forms, with respect to the norm
(7.1) ϕ 7→ ||ϕ|| + ||T ∗ ϕ|| + ||Sϕ||.
85
7.2.3 T HEOREM . Let (X, ω) be a complete Kähler manifold and let E → X be a holomorphic
Hermitian vector bundle. Then for any E-valued (p, q)-form ϕ ∈ Domain(T ∗ ) ∩ Domain(S) there
exist smooth, compactly supported E-valued (p, q)-forms {ϕk } such that
lim ||ϕ − ϕk || + ||T ∗ ϕ − ∂¯∗ ϕk || + ||Sϕ − ∂ϕ

¯ k || = 0.
k→∞
Proof. Fix an exhaustion function ψ ∈ C ∞ (X) such that |dψ|ω ≤ 1 and a function χ ∈ C ∞ (R)
such that χ(r) = 1 for r ≤ 0 and χ(r) = 0 for r ≥ 1. Let fk (x) := χ(ψ(x) − k + 1). Then
fk ∈ Co∞ (X), and in fact fk is supported in the compact set Xk := {x ∈ X ; ψ(x) ≤ k}. For any
ϕ ∈ Domain(T ∗ ) ∩ Domain(S) Proposition 7.1.8 implies that the compactly supported form
Φk := fk ϕ
lies in Domain(S) ∩ Domain(T ∗ ) and satisfies

¯ k∧ϕ
SΦk = fk Sϕ + ∂f and T ∗ Φk = fk T ∗ ϕ + (grad0,1 fk )yϕ.
We estimate that
||SΦk − Sϕ|| ≤ ||(1 − fk )Sϕ|| + (sup |dfk |)||1X−Xk ϕ|| ≤ C(||1X−Xk ϕ|| + ||1X−Xk Sϕ||),
X
and the right hand side converges to 0 as k → ∞ because ϕ and Sϕ are in L2 . Since
√
||grad0,1 fk || = ||∂fk || = 2||dfk ||,
a similar calculation shows that ||T ∗ Φk − T ∗ ϕ|| → 0 as k → ∞. Therefore
||ϕ − Φk || + ||T ∗ ϕ − T ∗ Φk || + ||Sϕ − SΦk || → 0 as k → ∞.
Using a partition of unity {χi }, we can write Φk = Φ1k + · · · + ΦN i

k with each Φk := χi fk ϕ
supported in some coordinate chart U i , and, again by Proposition 7.1.8, lying in the domain of
T ∗ . Using mollifiers in the coordinate chart Ui we can approximate Φik by a compactly supported
smooth form ϕik that converges to Φik in the Sobolev 1-norm. Thus, with ϕk := ϕ1k + · · · + ϕN
k ,
N
X
||Φk − ϕk || + ||T Φk − ∂¯∗ ϕk || + ||SΦk − ∂ϕ
∗ ¯ k || . ||Φik − ϕk || + ||d(Φik − ϕk )|| → 0
i=1
as k → ∞. Thus by the triangle inequality ||ϕ − ϕk || + ||T ∗ ϕ − ∂¯∗ ϕk || + ||Sϕ − ∂ϕ

¯ k || → 0 as
k → ∞, and the proof is complete.
7.2.4 R EMARK . The norm (7.1) is often called the graph norm, since it is the restriction of the
norm of L2p,q ⊕ L2p,q−1 ⊕ L2p,q+1 to the graph of the operator T ∗ ⊕ S.
86
7.2.3 The basic estimate for complete Kähler metrics
We can now prove the following version of the Basic Estimate.
7.2.5 T HEOREM (The Basic Estimate: Complete Kähler metric case). Let (X, g) be a complete
Kähler manifold and let E → X be a holomorphic vector bundle with Hermitian metric h. Then
for all E-valued (p, q)-forms ϕ ∈ Domain(T ∗ ) ∩ Domain(S) such that Θg (g (p) ⊗ det g ⊗ h)ϕ ∈
L2p,q (g, h) one has the estimate
||T ∗ ϕ||2 + ||Sϕ||2 ≥ (Θg (g (p) ⊗ det g ⊗ h)ϕ, ϕ).
Proof. By Theorem 7.2.3 the smooth compactly supported E-valued forms are dense, and thus the
result follows from integration by parts applied to the formal Bochner-Kodaira Identity (Theorem
¯ 2 ≥ 0.
5.3.1) and the obvious inequality ||∇ϕ||
7.3 Hörmander’s Theorem

7.3.1 Hörmander’s Theorem for Complete Kähler metrics
We shall now use Theorem 7.2.5 to establish the following result.
7.3.1 T HEOREM (Hörmander’s Theorem; complete metric case). Let (X, g) be a complete Kähler
manifold and let V → X be a holomorphic vector bundle with Hermitian metric h. Assume that
Θg (g (p) ⊗ det g ⊗ h) ≥ cIdΛp,q

X ⊗V
for some c > 0. Then for each V -valued (p, q)-form ϕ such that
Z
¯ = 0 and
∂ϕ |ϕ|2h,g ω n < +∞
X
there exists a V -valued (p, q − 1)-form u such that

Z Z
¯ = ϕ and 2 n 1
∂u |u|h,g ω ≤ |ϕ|2h,g ω n .
X c X
Proof. Since Kernel(S) is a closed subspace of L2p,q (h, g) we can replace the latter by the former.
Restricting the Basic Estimate to Kernel(S) then gives us that for all ψ ∈ Domain(T ∗ )∩Kernel(S),
||ϕ||2 ∗ 2
(7.2) |(ϕ, ψ)|2 ≤ ||ϕ||2 ||ψ||2 ≤ ||T ϕ|| ,
c
where the last inequality follows from Theorem 7.2.5. The Functional Analysis Lemma ?? with
H1 = L2p,q (h, g) and H2 = Kernel(S) then yields a (p, q)-form u such that T u = ϕ and ||u|| ≤
c−1 ||ϕ||, as desired.
87
7.3.2 Comparison of L2 norms from comparison of metrics
Let X be a complex manifold and V → X a holomorphic vector bundle with Hermitian metric h.
Fix Hermitian metrics g and γ with metric forms ω and θ respectively, and assume γ ≥ g.
Choose an ω-orthonormal basis of (1, 0)-forms α1 , ..., αn such that
√ √
θ = λ1 −1α1 ∧ ᾱ1 + · · · + λn −1αn ∧ ᾱn .
Then λi ≥ 1 for 1 ≤ i ≤ n.
Given a V -valued (p, q)-form η, one can locally write
η = ηI J¯αI ∧ ᾱJ
with ηI J¯ local sections of V . With the notation λI = λi1 · · · λip and λJ¯ = λj1 · · · λjq , the norms of
η with respect to the metrics (h, g) and (h, γ) are respectively
1 X 1 X |ηI J¯|2h
|η|2h,g = |ηI J¯|2h and |η|2h,γ = .
p!q! p!q! λI λJ¯
|I|=p,|J|=q |I|=p,|J|=q
In particular,
|η|2h,γ ≤ |η|2h,g .
On the other hand, the volume forms for the two metrics are
√ √
dVg = ( −1)n α1 ∧ ᾱ1 ∧ · · · αn ∧ ᾱn and dVγ = λ1 · · · λn ( −1)n α1 ∧ ᾱ1 ∧ · · · αn ∧ ᾱn ,
so dVg ≤ dVγ , and hence it is in general hard to compare the L2 -norms of (p, q)-forms with respect
to these two metrics.
However, there is one exceptional but important case: the case p = n. In this case
1 X |ηI J¯|2h √
|η|2h,γ dVγ = ( −1)n α1 ∧ ᾱ1 ∧ · · · αn ∧ ᾱn
n!q! λJ¯
|I|=n,|J|=q
1 X √
≤ |ηI J¯|2h ( −1)n α1 ∧ ᾱ1 ∧ · · · αn ∧ ᾱn
n!q!
|I|=n,|J|=q
= |η|2h,g dVg .
Consequently one has the comparison of L2 -norms: if η is a V -valued (n, q)-form and γ ≥ g then
||η||2h,γ ≤ ||η||2h,g .
This monotonicity of norms is very useful if one wants to generalize Hörmander’s Theorem 7.3.1
to the setting in which the manifold X is complete Kähler but the metric g is not necessarily
complete. Indeed, of g is a Hermitian (resp. Kähler) metric and g∗ is a complete Hermitian (resp.
Kähler) metric then for every ε > 0 the metric gε := g +εg∗ is a complete Hermitian (resp. Kähler)
metric that dominates g.
There is also of course the problem that the monotonicity of L2 -norms only works when p = n,
but this matter is dealt with by another clever trick.
88
7.3.3 Hörmander’s Theorem for complete Kähler manifolds
We can now remove the hypothesis of completeness for the metric g in Theorem 7.3.1.
7.3.2 T HEOREM (Hörmander’s Theorem). Let X be a complete Kähler manifold with metric g
that is not necessarily complete, and let E → X be a holomorphic vector bundle with Hermitian
metric h. Assume that
Θg (g (p) ⊗ det g ⊗ h) ≥ cIdΛp,q
X ⊗E
for some c > 0. Then for each E-valued (p, q)-form ϕ such that
Z
|ϕ|2h,g ω n < +∞ and ∂ϕ ¯ =0
X
in the sense of distributions, there exists a E-valued (p, q − 1)-form u such that
Z Z
¯ 2 n 1
∂u = ϕ and |u|h,g ω ≤ |ϕ|2h,g ω n .
X c X
Proof. Fix a complete Kähler metric g∗ and qrite gε := g + εg∗ . If ε > 0 is sufficiently small then
Θgε (g (p) ⊗ det g ⊗ h) ≥ cε IdΛp,q

X ⊗E
with cε > 0 and lim cε = c.

ε→0
In order to exploit the monotonicity of L2 -norms with respect to the Kähler metrics, one uses
the following trick to identify E-valued (p, q)-forms with Ẽ-valued (n, q)-forms for some holo-
morphic vector bundle Ẽ. We have used this trick at the bundle level, but not at the metric level.
Let Ẽ := Λp,0 TX∗ ⊗ KX∗
⊗ E and set h̃ = g (p) ⊗ det g ⊗ h. Then h̃ is a metric for Ẽ satisfying
Θgε (h̃) ≥ cε ,
and moreover for any E-valued (p, q)-form, a.k.a. Ẽ-valued (n, q̃)-form, η
|η|2h,g = |η|2h̃,g .
We emphasize that the metric h̃ involves the metric g and not the metric gε , and so in this sense h̃
is ‘ε-static’.
Now let ϕ ∈ L2p,q (h, g) = L2n,q (h̃, g). Then by monotonicity
Z Z Z
2 2
|ϕ|h̃,gε dVgε ≤ |ϕ|h̃,g dVg = |ϕ|2h,g dVg < +∞,
X X X
so by Theorem 7.3.1 there exists uε ∈ L2n,q (h̃, gε ) such that

Z Z Z
¯ ε = ϕ and 2 1 2 1
∂u |uε |h̃,gε dVgε ≤ |ϕ|h̃,gε dVgε ≤ |ϕ|2h̃,g dVg .
X c ε X c ε X
89
Now, for 0 < ε < εo , gε ≤ gεo , and hence by monotonicity again
Z Z Z
2 2 1
|uε |h̃,gε dVgεo ≤ |uε |h̃,gε dVgε ≤ |ϕ|2h,g dVg .
X
o
X c ε X
Thus since cε → c, {uε } lies in a fixed ball inside L2n,q (h̃, gεo ). By Alaoglu’s Theorem we can
choose a subsequence uo,jo := uεjo that converges in L2n,q (h̃, gεo ). Choosing ε1 < εo , we have a
convergent subsequence
{u1,j1 ; j1 = 1, 2, ...} ⊂ {uo,jo ; jo = 1, 2, ...} ∩ L2n,q (h̃, gε1 ).
Continuing inductively, we choose εk+1 < εk and a convergent subsequence
{uk+1,jk+1 ; jk+1 = 1, 2, ...} ⊂ {uk,jk ; jk = 1, 2, ...} ∩ L2n,q (h̃, gεk+1 ).
Assuming further that εk → 0, the sequence {uj := uj,1 } converges to some

\
u∈ L2n,q (h̃, gεk ).
k≥o
Since |u|2h̃,g dVgεk is an increasing sequence (in k), by the Monotone Convergence Theorem
εk
Z Z Z Z Z
1 1
|u|2h,g dVg = |u|2h̃,g dVg = lim |u|2h̃,gε dVgεk ≤ |ϕ|2h̃,g dVg = |ϕ|2h,g dVg .
X X k→∞ X k c X c X
Finally,
(u, ∂¯∗ ψ) = lim(uj , ∂¯∗ ψ) = lim(ϕ, ψ) = (ϕ, ψ)
for any compactly supported ψ. Thus ∂u ¯ = ϕ, and the proof is complete.
7.3.4 Weakly pseudoconvex Kähler manifolds

An important class of complete Kähler manifolds in complex analysis is the class of weakly pseu-
doconvex manifolds.
7.3.3 D EFINITION . A complex manifold X is said to be weakly pseudoconvex if X has a smooth
plurisubharmonic exhaustion function.
7.3.4 T HEOREM . If X is a weakly pseudoconvex Hermitian (resp. Kähler) manifold then X has a
complete Hermitian (resp. Kähler) metric.
Proof. Fix a Hermitian form ω and a smooth plurisubharmonic exhaustion function ψ. Then
√
ω̃ = ω + −1∂ ∂e¯ 2ψ
is Hermitian, and it is Kähler if ω is Kähler. Moreover

√ √ √
ω̃ = ω + 4e2ψ −1∂ψ ∧ ∂ψ ¯ + 2e2ψ −1∂ ∂ψ¯ ≥ 4 −1∂(eψ ) ∧ ∂(e
¯ ψ ).
90
¯ = ∂ρ + ∂ρ, so that
Note that for any real-valued function ρ, dρ = ∂ρ + ∂ρ
|dρ|2ω̃ = gω̃∗ (∂ρ + ∂ρ, ∂ρ + ∂ρ) = gω̃∗ (∂ρ, ∂ρ) + gω̃∗ (∂ρ, ∂ρ) = 2gω̃∗ (∂ρ, ∂ρ) = 2|∂ρ|2ω̃ .
Thus
|d(eψ )|2ω̃ = 2|∂(eψ )|2ω̃ ≤ 12 .
Since the exponential function is increasing and proper (as a function from R to (0, ∞)), eψ is also
a smooth exhaustion function. By Proposition 7.2.2, (X, ω̃) is complete.
7.3.5 R EMARK . In the latter proof, if we can replace the exponential function by χ ◦ ψ where
χ : [0, ∞) → [0, ∞) is any non-constant convex increasing function. (Observe that a bounded in-
creasing convex function on [0, ∞) is necessarily constant, and thus in fact χ ◦ ψ is also a plurisub-
harmonic exhaustion.) We then compute that
√ √ √ √
ω̃ = ω + −1(χ0 ◦ψ)∂ ∂ψ ¯ +χ00 (ψ) −1∂ψ ∧ ∂ψ ¯ ≥ χ00 (ψ) −1∂ψ ∧ ∂ψ ¯ = −1∂(h◦ψ)∧ ∂(h◦ψ), ¯
where h is a real-valued function satisfying (h0 )2 = χ00 , i.e.,

Z xp
h(x) = C + χ00 (t)dt.
o
Note that as χ is convex, h is necessarily increasing, so in order for h ◦ ψ to be an exhaustion, it is

necessary and sufficient that h is an exhaustion of R, i.e., limx→+∞ h(x) = +∞.
Thus Hörmander’s Theorem holds on weakly pseudoconvex manifolds.
7.3.5 Skoda’s estimate

Theorem 7.3.2 has a generalization, with almost the same proof, that can be useful in certain
¯
applications. The result is still about solving the ∂-equation with L2 estimates, but the estimates are
for data that that lies in L2 with respect to different norms. The generalization was first observed
by Skoda, and then generalized by Demailly, though it is often still referred to as Hörmander’s
Theorem in the literature.
Before stating this version of Hörmander’s Theorem, let us introduce some notation. Given
a positive definite Hermitian (1, 1)-form Υ, we can define a pointwise norm on E-valued (p, q)-
forms θ by
β
|θ|2Υ,h,g := θI,
α
J¯θK,L̄ hαβ̄ g
I K̄ `1 j̄1
Υ · · · Υ`q j̄q ,
α
√ ¯
where locally, θ = θI, e ⊗ dz I ∧ dz̄ J , Υ = Υij̄ −1dz i ∧ dz̄ j , and Υk` is the inverse of the
J¯ α
matrix (Υij̄ ). In other words, this norm is much like the norm using only h and g, except that in
the (0, 1)-directions we replace the metric g by the Hermitian form Υ.
With these observations in hand, we can now state Skoda’s estimate.
91
7.3.6 T HEOREM (Demailly-Hörmander-SkodaTheorem). Let (X, g) be Kähler manifold and let
E → X be a holomorphic vector bundle with Hermitian metric h. Assume that X admits a
complete Kähler metric (which need not be g). Suppose that
Θg (g (p) ⊗ det g ⊗ h)ξ, ξ g,h ≥ hΥg ξ, ξig,h , ξ ∈ Λp,q

X ⊗ E,
where q
¯
X
Υg (ξIαJ¯dz I J
∧ dz̄ ⊗ eα ) := g i` Υij̄k ξIαj̄1 ···(`)
¯ k ···j̄q .
k=1
Then for each E-valued (p, q)-form ϕ such that

Z
¯
∂ϕ = 0 and |ϕ|2Υ,g,h dVg < +∞
X
there exists an E-valued (p, q − 1)-form u such that

Z Z
¯ = ϕ and
∂u 2
|u|g,h dVg ≤ |ϕ|2Υ,g,h dVg .
X X
Sketch of proof. The proof of Hörmander’s Theorem goes through in exactly the same way if one
proves the pointwise estimate
| hϕ, ψi |2 ≤ |ϕ|2Υ,g,h hΥg ψ, ψig,h .
The fact that such an inequality holds for a Hermitian form is a simple exercise in linear algebra:
it holds for diagonal operators by using the rescaled Cauchy-Schwarz Inequality
−1 N 2
|a · b̄| ≤ |(λ−1 1 1 N 2
1 a , ..., λN a )| · |(λ1 b , ..., λN b )| ,
and the general case is proved by diagonalizing Υ.

While the argument used when passing from complete metrics to non-complete metrics re-
quires some checking, it is not difficult to see after a careful look, as the reader can confirm.
7.4 Singular Hermitian metrics: an overview

In the statement of Hörmander’s Theorem for domains in C (Theorem 4.1.12) we assumed only
that the weight function ϕ is strictly subharmonic. Thinking of e−ϕ as a metric for the trivial line
bundle on C, one sees that Theorem 4.1.12 does not follow from Theorem 7.3.2 because the latter
theorem assumes the metric h for E → X is smooth.
For a general vector bundle there is as yet no accepted notion of a singular Hermitian metric
such that Theorem 7.3.2 holds; this question lies in an active area of research. The exceptional
situation is when the rank of E is 1, i.e., when E → X is a line bundle. The following notion was
introduced by Demailly.
92
7.4.1 D EFINITION . Let L → X be a holomorphic line bundle. A singular Hermitian metric for L
is a measurable section h ∈ Γ(X, L∗ ⊗ L∗ ) such that for any local holomorphic frame ξ of L over
U ⊂X
1. (Hermitian) h(ξ, ξ) : U → [0, ∞], and
2. (Singular) ϕξ := − log h(ξ, ξ) ∈ L1òc (U ).
The name singular Hermitian metric is slightly misleading; it should probably have been called
a possibly singular Hermitian metric, since smooth metrics satisfy the criteria of Definition 7.4.1.
¯ ξ by
The main point of the definition is that one can define the (1, 1)-current ∂ ∂ϕ
Z
¯ (η) :=
∂ ∂ϕ ξ ¯
ϕξ ∂∂η for all smooth compactly supported (n − 1, n − 1)-forms η on U
U
¯ ξ,
In the smooth case this (1, 1)-current is represented by integration against the smooth form ∂ ∂ϕ
and the latter is independent of the choice of frame ξ. This form is of course the curvature of the
Chern connection for L, h. One can therefore make the following definition.
7.4.2 D EFINITION . The current

¯ ξ
Θ(h) := ∂ ∂ϕ
is called the curvature current of the singular Hermitian metric h.
To have singular metrics with positive1 curvature current simply means that the local functions
ϕξ are plurisubharmonic. Strict positivity means these functions are strictly plurisubharmonic.
These notions, which are local, also do not depend on the choice of local frame ξ for L.
Quite often, however, positivity is not quite the correct hypothesis. For this purpose one wants
to have a class of singular Hermitian metrics whose curvature currents do not become ‘too nega-
tive’.
7.4.3 D EFINITION . Let X be a Kähler manifold with Kähler form ω. A function ψ : X →

[−∞, ∞) is said to be ω-plurisubharmonic if
√
¯ +ω
−1∂ ∂ψ
is a positive (1, 1)-current. More generally, the function ψ is said to be quasi-plurisubharmonic if

for any x ∈ X there exist an open set U containing X, a smooth function ψU,sm and a plurisub-
harmonic function ψU,psh such that
ψ|U = ψU,psh + ψU,sm .

1
In the context of currents the word ‘positive’ often means ‘non-negative’, especially when the authors are French.
The nomenclature comes from measure theory, in which a positive measure can still be supported on a proper subset.
93
In many situations, Hörmander’s Theorem 7.3.2 (again, with the rank of E equal to 1) holds
with singular Hermitian metrics in place of smooth ones. The idea of the proof is to regularize
the metrics with the right monotonicity: one seeks a sequence of smooth metrics hj that increases
pointwise to the singular metric h, such that the curvature of each of the metrics in the approxi-
mating sequence carries enough positivity to apply Hörmander’s Theorem. With such a sequence
of metrics in hand, one can deduce the singular metric version of Hörmander’s Theorem from
Theorem 7.3.2 for the smooth metrics hj and standard limit and weak-∗ compactness theorems
from real analysis. This is precisely what was done in the last part of the proof of Theorem 4.1.12.
(A key point here is that the bounds in Hörmander’s Theorem do not depend on the manifold X,
the vector bundle E, and not even on the metrics weights themselves, but only on the positivity
conditions that these metrics satisfy.)
As for finding appropriate monotonic smooth approximations to a singular Hermitian metric,
there are many interesting cases in which this can be done.
• The classical example occurs when X is a bounded domain in Cn and L → X is the trivial
line bundle. In this case we can write our metric h as e−ϕ where ϕ is a globally defined
function. If ϕ is quasi-plurisubharmonic and Ωεo := {z ∈ Ω ; Bεo (z) ⊂ Ω} is the set of
points of Ω of distance at least εo from ∂Ω then there exists C > 0 such that ϕ + C|z|2 is
plurisubharmonic on Ωεo . One can then define, for ε ∈ (0, εo ), the function ϕε : Ωεo → R
by Z
1
2
ϕε (z) := −C|z| + 2n (ϕ(ζ) + C|ζ|2 ))χ(ε−1 (z − ζ))dV (ζ),
ε Cn
R
where χ is a U (n)-invariant function in Cn satisfying Cn χdV = 1. The function ϕε de-
creases monotonically to ϕ as ε & 0, and for each ε > 0 the function z 7→ ϕε (z) + C|z|2 is
plurisubharmonic.
• One can also carry out this sort of regularization on a Stein manifold, though in that case the
argument is more subtle.
• There are some manifolds, such as projective manifolds, that carry an analytic hypersurface
whose complement is a Stein manifold. The typical example is that of projective manifolds.
In such manifolds one works on the complement of this analytic hypersurface. Often the L2
estimates will allow for extension of holomorphic sections across the analytic hypersurface.
Indeed, the L2 norm dominates the pointwise norm, so a uniform L2 bound implies locally
uniform bounds, and one can apply Riemann’s Theorem on removable singularities.
• For a general complete Kähler manifold, even a weakly pseudoconvex one, this sort of reg-
ularization is not necessarily possible. It is possible to regularize metrics by giving up some
positivity– this is a well-known and beautiful technique due to Demailly– but the loss of
positivity is often a problem since, unlike Stein or Projective manifolds, one does not have
an ambient positive line bundle to add in the extra positivity.
However, if on a complete Kähler manifold a given singular metric h has a monotonic ap-
proximation by smooth metrics without loss of positivity then for this metric h one can
establish Hörmander’s Theorem.
94
E XERCISES
95
Part III
Twisted Methods
96
Lecture 8
Donnelly-Fefferman-Ohsawa Technique
In this section we shall obtain an improvement of Hörmander’s Theorem on complex manifolds

that are equipped with certain special geometric features.
8.1 The twisted basic estimate

In Section 5 we established the Bochner-Kodaira Identity (5.3.1): If X is a Kähler manifold with
Kähler metric g and E → X is a holomorphic vector bundle with smooth Hermitian metric h then
for any smooth E-valued (n, q)-form α one has the formal identity
¯ ∗ ∇α
α = ∇ ¯ + Θg (h)α.
If α has compact support then by pairing the last identity with α and integrating by parts one has
Z Z Z Z
¯∗ 2
|∂ α| + ¯ 2
|∂α| = ¯
|∇α| +2
hΘg (h)α, αi ,
X X X X
form which the so-called basic estimate

Z Z Z
(8.1) ¯∗ 2
|∂ α| + ¯ 2
|∂α| ≥ hΘg (h)α, αi
X X X
trivially follows. If the Kähler metric g is complete, as we shall assume from here on is the case,
then (8.1) holds for all α ∈ Domain(∂) ¯ ∩ Domain(∂¯∗ ).
There are several equivalent ways to approach the twisted basic estimate. We shall take the one
developed by Siu, but reformulate it slightly. The first step is to consider a metric of the form
h = e−η h
where h is a Hermitian metric for E → X and η is a smooth function. By Propositions 5.2.2

together with the definition of the Chern connection,
∂¯h∗ α = ∂¯h∗ α + (grad0,1 η)yα and ¯

Θ(h) = Θ(h) + ∂ ∂η,
97
where
∂η ∂
grad0,1 η := g ij̄
∂z i ∂ z̄ j
is the (0, 1)-vector field obtained from ∂η via the isomorphism TX∗1,0 ∼ 0,1
= TX induced by the metric
g, and y is contraction of forms by vectors. Then
√
¯ 2 = e−η |∂α|
|∂α| ¯ 2 = | e−η ∂α|¯ 2 ,
g,h g,h g,h
D E
−η ¯ ⊗ IdE

hΘg (h)α, αig,h = e Θ(h) + ∂ ∂η g
α, α
g,h
and
2
|∂¯h∗ α|2g,h = e−η ∂¯h∗ α + (grad0,1 η)yαg,h

2
= e−η ∂¯h∗ α|2g,h + |(grad0,1 η)yαg,h + 2Re ∂¯h∗ α, (grad0,1 η)yα g,h

2
≤ e−η (1 + a) ∂¯h∗ α|2g,h + (1 + a−1 )|(grad0,1 η)yαg,h

p
−η ¯∗ 2 1 + a −η ¯

= | e (1 + a)∂h α|g,h + e ∂η ∧ ∂η ⊗ IdE g α, α ,
a g,h
where a is any positive function. The inequality follows from Cauchy-Schwarz and the estimate
2αβ ≤ C −1 α2 + Cβ 2 . From the basic estimate (8.1) we therefore have the a priori estimate
Z p 2 Z √
−η ¯ 2

e (1 + a)∂¯h α +
∗
−η
e ∂α

X Z g,h X g,h
D E
e−η Θ(h) + ∂ ∂η ¯ − 1+a ∂η ∧ ∂η
¯ ⊗ IdE α, α

≥ a g
.
X g,h
Observing that the domains of ∂¯ and ∂¯h∗ are the same as the domains of the operators
√ ∗
e−η ∂¯ and T ∗ := ∂¯h ◦ e−η (1 + a)
p
S :=
respectively, one has the following theorem.
8.1.1 T HEOREM (Twisted basic estimate). Let X be a Kähler manifold with complete Kähler
metric g and let E → X be a holomorphic vector bundle with Hermitian metric h. Then
Z Z
∗ 2
|T α|g,h + |Sα|2g,h
X Z D X E
¯ − 1+a ∂η ∧ ∂η
e−η Θ(h) + ∂ ∂η ¯ ⊗ IdE α, α

≥ a g
X g,h
√ √
for every form α ∈ Domain((∂¯ ◦ ¯
τ + A)∗h ) ∩ Domain( τ ◦ ∂).
98
8.2 Donnelly-Fefferman-Ohsawa estimate for twisted ∂¯
With Theorem 8.1.1 in hand, one can use the same method of proof of Skoda’s estimate to establish
the following theorem.
8.2.1 T HEOREM (Donnelly-Fefferman-Ohsawa Estimate). Let X be a complete Kähler manifold
with not necessarily complete Kähler metric g, let E → X be a holomorphic vector bundle with
Hermitian metric h and let η : X → R and a : X → (0, ∞) be functions with η C 2 -smooth.
Assume the operator
Ψ := e−η Θ(h) + ∂ ∂η ¯ − 1+a ∂η ∧ ∂η ¯ ⊗ IdE : Λn,q ⊗ E → Λn,q ⊗ E

a g X X
is invertible on each fiber. Then for each E-valued (n, q)-form ϕ satisfying
Z

−1
Sϕ = 0 and Ψ ϕ, ϕ g,h < +∞
X
there exists an E-valued (n, q − 1)-form u such that

Z Z
2

−1
T u = ϕ and |u|g,h ≤ Ψ ϕ, ϕ g,h .
X X
8.3 Ohsawa’s ∂¯ estimate

8.3.1 T HEOREM (Ohsawa’s ∂¯ estimate). Let X be a complete Kähler manifold with not necessarily
complete Kähler metric g, let E → X be a holomorphic vector bundle with Hermitian metric h
and let η : X → R be a C 2 -smooth function. Assume the operator
¯ − (1 + δ)∂η ∧ ∂η¯ ⊗ IdE : Λn,q ⊗ E → Λn,q ⊗ E

Φδ := Θ(h) + 2∂ ∂η g X X
is invertible for some δ > 0. Then for each E-valued (n, q)-form ϕ satisfying
Z
¯ = 0 and

−1
∂ϕ Φδ ϕ, ϕ g,h < +∞
X
there exists an E-valued (n, q − 1)-form U such that

Z Z
¯ = ϕ and 2 1 + δ
−1
∂U |U |g,h ≤ Φδ ϕ, ϕ g,h .
X δ X
8.3.2 R EMARK . In the typical application of Theorem 8.3.1 one takes η to be a strictly plurisub-
harmonic function such that
(8.2) ¯ ≥ (1 + δ)∂η ∧ ∂η,
∂ ∂η ¯
and in this case the assumption on the metric h is that the metric he−η is Nakano positive. The
latter condition holds as long as Θ(h) ≥ −∂ ∂η¯ ⊗ IdE , and therefore the curvature hypotheses can
be significantly weaker than those of Hörmander’s Theorem. Of course, one obtains a non-trivial
result only if there are non-constant functions η satisfying (8.2). We shall explore this condition a
little further in the next paragraph.
99
Proof of Theorem 8.3.1. We shall apply Theorem 8.2.1 with h := e−η h and a = 1/δ. Then
¯
Θ(h) = Θ(e−η h) = Θ(h) + ∂ ∂η
and therefore
¯ −
Ψ := e−η Θ(h) + ∂ ∂η 1+a ¯ ⊗ IdE = e−η Φδ

a
∧ ∂η
∂η g
¯ (1 + δ −1 )e−η · u) = ϕ and
p
is invertible. By Theorem 8.2.1 there exists u such that ∂(
Z Z
2

−1
|u|g,h ≤ Ψ ϕ, ϕ g,h .
X X
p
Setting U := (1 + δ −1 )e−η · u, we find that
Z Z Z Z
2 1+δ 2 1+δ
−1 1+δ
−1
|U |g,h = |u|g,h ≤ Ψ ϕ, ϕ g,h = Φδ ϕ, ϕ g,h ,
X δ X δ X δ X
and the proof is complete.
8.4 Functions with self-bounded gradient

If one wants the hypotheses of Theorem 8.2.1 to yield a gain in positivity then it makes sense to
ask for functions η and a such that
¯ − a+1 ¯ > 0.
∂ ∂η ∂η ∧ ∂η
a
Of course, any function satisfying this positivity requirement is obviously plurisubharmonic, but
a
the condition is even stronger. Indeed, since 1+a < 1, one has the estimate
¯ < ∂ ∂η,
∂η ∧ ∂η ¯
and hence the function ξ := −e−η satisfies

¯ = ∂(e−η ∂η)
∂ ∂ξ ¯ = e−η (∂ ∂η
¯ − ∂η ∧ ∂η)
¯ > 0,
i.e., ξ is a negative, strictly plurisubharmonic function.

8.4.1 D EFINITION (McNeal). A function f : X → R is said to have self-bounded gradient if
¯ ≤ ∂ ∂f,
∂f ∧ ∂f ¯
or equivalently, if −e−f is plurisubharmonic.

To complex analysts the existence of a negative plurisubharmonic is a familiar restriction that
first appears in an important result regarding Green’s functions: the latter exist on a domain X ⊂ C
(or more generally, on a non-compact Riemann surface) if and only if the Riemann surface has a
non-constant negative subharmonic function.
In higher dimensions functions with self-bounded gradient have been used by a number of
¯
authors to obtain a number of different types of estimates for ∂.
100
8.4.2 E XAMPLE . If X := {z ∈ Cn ; |z| < 1} is the unit ball and
η := − log(1 − |z|2 )
then
−e−η = |z|2 − 1
is strictly plurisubharmonic,
√ and therefore η has self-bounded gradient.
¯
Note that −1∂ ∂η is, up to a constant, the Kähler form of the Poincaré metric, the Bergman
metric, the Kobayashi metric and the Caratheodory metric. All of these metrics are so-called
Einstein metrics, i.e., their Ricci curvature is a negative constant multiple of their Kähler form.
8.4.3 E XAMPLE . Let X be a complex manifold and let f : X → Cn be a holomorphic map such
that ||f (x)|| < 1 for all x ∈ X. Then the function
η := − log(1 − ||f ||2 )
has self-bounded gradient, since it is the pullback by f of the function − log(1 − |z|2 ) of the
previous example.
8.4.4 E XAMPLE . If X = {ζ ∈ C ; 0 < |ζ| < 1} is the punctured unit disk and
η := log(log |ζ|−2 )
then
1
−e−η =
log |ζ|2
satisfies
¯ −η dζ̄ dζ ∧ dζ̄
∂ ∂(−e ) = −∂ = 2 ,
ζ̄ log |ζ| 2 |ζ| (log |ζ|2 )2
and hence η has self-bounded gradient. Note again that
√
√ √

¯ dζ̄ −1dζ ∧ dζ̄
−1∂ ∂η = −1∂ = 2
−ζ̄ log |ζ| 2 |ζ| (log |ζ|2 )2
is the Kähler form of the unique (up to a constant factor) metric for X whose Gaussian curvature
is a negative constant.
8.4.5 E XAMPLE . Let X be a complex manifold and assume there exists a non-constant holomor-
phic function T ∈ O(X) such that supX |T | ≤ 1. Let
Z := {x ∈ X ; T (x) = 0}
by the analytic hypersurface defined by T . Then the function
1
η := log log
|T |2
has self-bounded gradient on the manifold Y := X − Z. Indeed, T : Y → {0 < |ζ| < 1} is
a holomorphic map and the function η is just the pullback of the function ζ 7→ log log |ζ|12 of the
previous example.
101
8.4.6 E XAMPLE . A complex manifold is said to be hyperconvex if it has a negative plurisubhar-
monic exhaustion function. (Usually one asks for the exhaustion function to be continuous as well,
but this is not a fixed convention.) For example, every strictly pseudoconvex domain has a negative,
strictly plurisubharmonic exhaustion function.
Letting ψ be a negative plurisubharmonic function, which such domains automatically support,
one sets
η := − log(−ψ).
Clearly −e−η = ψ is plurisubharmonic, and therefore η has self-bounded gradient.
8.4.7 E XAMPLE . Let X be a bounded strictly pseudoconvex domain in Cn . One can define the
Hilbert space Z
HX := f ∈ ΓO (X, KX ) ; 2
|f | < +∞ ,
X
√ n2 /2
where |f |2 = −1 f ∧ f¯. For any orthonormal (Riesz) basis f1 , f2 , ... of HX one can define
the so-called Bergman function
∞
X |fj (z)|2
BX (z) := ,
j=1
dV (z)
where dV (z) is Lebesgue measure. The Bergman function is independent of the choice of or-
thonormal basis for HX .
Suppose ψ : Cn → R is a smooth function such that
(i) X = ψ −1 (−∞, 0),
(ii) |dψ| ≡ 0 on ∂X and
(iii) ψ is strictly plurisubharmonic on a neighborhood of the closure of X.
(A function satisfying (i) is called a defining function for X, and if (ii) also holds then it is called a
Levi defining function.) A famous result of C. Fefferman [Fe-1974] states that the Bergman kernel
has an asymptotic expression
BX (z) = F (z)(−ψ(z))−(n+1) + G(z) log(−ψ(z)), z ∈ X,
where F and G are smooth functions that do not vanish in a neighborhood of the boundary of X.
In particular, the function
1
(8.3) η(z) := log BX (z)
n+1
is strictly plurisubharmonic. Moreover,
η(z) = − log(−ψ(z)) + log F (z) + G(z)ψ(z)n+1 log(−ψ(z)) ,

102
and therefore η + log(−ψ) is a smooth function in a neighborhood of ∂X. Since the expression
for BX is only relevant near ∂X, one can choose the functions F and G so that φ := η + log(−ψ)
is smooth on a neighborhood of the closure of X.
Now, the function ηo := − log(−ψ) satisfies
¯ ¯
∂ ∂η ¯ o = ∂ ∂ψ
¯ o − ∂ηo ∧ ∂η and ¯ o = ∂ψ ∧ ∂ψ ,
∂ηo ∧ ∂η
−ψ (−ψ)2
and therefore η = ηo + φ satisfies

¯ − ∂η ∧ ∂η
∂ ∂η ¯ = (−ψ)−1 ∂ ∂ψ
¯ − 2Re ∂ηo ∧ ∂φ
¯ + ∂ ∂φ
¯ − ∂φ ∧ ∂φ
¯
¯ − (−ψ)∂ηo ∧ ∂η
≥ (−ψ)−1 ∂ ∂ψ ¯ o − (−ψ)−1 ∂φ ∧ ∂φ
¯ + ∂ ∂φ
¯ − ∂φ ∧ ∂φ
¯
¯ − ∂ψ ∧ ∂ψ
= (−ψ)−1 (∂ ∂ψ ¯ − ∂φ ∧ ∂φ)
¯ + ∂ ∂φ¯ − ∂φ ∧ ∂φ
¯
−C ¯ 2
≥ ∂ ∂|z|
(−ψ)
for some sufficiently large constant C > 0.

Although the function (8.3) does not have self bounded gradient in general (though it can,
sometimes), the estimate ∂ ∂η¯ − ∂η ∧ ∂η¯ ≥ −C ∂ ∂|z|
¯ 2 can still be very useful in obtaining L2
(−ψ)
¯
estimates for ∂.
E XERCISES
103
Lecture 9
Extension Theorems
The problem of interpolation of data is fundamental in many areas of mathematics and the hard
sciences. In our setting, one might like to interpolate holomorphic functions, or perhaps sections
of a holomorphic vector bundle, that are specified on a complex submanifold.
In this lecture we will discuss the interpolation problem from a complex submanifold (in fact,
often a hypersurface) in a complex manifold. We will restrict our attention to the case of smooth
hypersurfaces, though standard techniques can be used to extend all of our results so as to yield
extension from singular hypersurfaces.
Additionally, since we will be very interested in using singular metrics, we will confine our-
selves to the case of extension problems for holomorphic line bundles. When the metrics are
smooth, many of the techniques can be extended to the higher rank case.
9.1 Extension without estimates

Let X be a complex manifold, let S ⊂ X be a complex and let L → X be a holomorphic line
bundle. Given a holomorphic section f ∈ ΓO (S, L|S ), one wants to know if there is a holomorphic
extension of f to X, i.e., if there is a holomorphic section F ∈ ΓO (X, L) such that
F |S = f.
In general, no such extension need exist.
9.1.1 E XAMPLE . Let X = P1 , let L → X be the trivial bundle and let S = {[1, 0], [0, 1]}. Then
the function f : S → C defined by f ([1, 0]) = 0 and f ([0, 1]) = 1 is holomorphic on S (in fact,
since S is discrete any function on S is holomorphic) but there is no holomorphic extension of f
to X. Indeed, since X is compact, every holomorphic function on X is constant.
9.1.1 Strongly pseudoconvex manifolds

On the other hand, if S is a closed submanifold of Cn and f ∈ O(X) then there is always an
extension of f to Cn , i.e., a function F ∈ O(Cn ) such that F |S = f .
104
More generally, a manifold X is said to be strongly pseudoconvex if there is a smooth, strictly
plurisubharmonic exhaustion function, i.e., a function ρ : X → R such that
√
¯ > 0 and Ωc := {x ∈ X ; ρ(x) < c} ⊂⊂ X
−1∂ ∂ρ
for all c ∈ R. (The manifold Cn is strongly pseudoconvex, as one can see by taking ρ(z) = |z|2 .)
Then one has the following theorem.
9.1.2 T HEOREM . Let X be a strongly pseudoconvex manifold and let S ⊂ X be a closed subman-
ifold. Then for all f ∈ O(S) there exists F ∈ O(X) such that F |S = f .
Proof.√We fix once and for all a strictly plurisubharmonic exhaustion function ρ on X, and we let
ω := −1∂ ∂ρ ¯ be our Kähler form for X.
Let k be the dimension of S. Choose a collection {Uj ; j = 1, 2, ...} of open sets Uj ⊂⊂ X,
and local coordinates zj = (xj , yj ) such that
(i) each point of X is contained in finitely many Uj ,
(ii) the map (xj , yj ) : Uj → Dn is a holomorphic diffeomorphism,
(iii) S ∩ Uj = {yj1 = ... = yjn−k = 0} and

∞
[
(iv) S ⊂ Uj .
j=1
(Thus xj = (x1j , ..., xkj ) are local coordinates on S ∩ Uj , and n = dimC X.) Let U0 = X − S.
Define functions Fj ∈ O(Uj ), j ∈ N, by
F0 (x) = 0, Fj (xj , yj ) := f (xj ), j = 1, 2, ... and gij := Fi − Fj ∈ O(Ui ∩ Uj ).
Then gij |S∩Ui ∩Uj ≡ 0. We seek functions gi ∈ O(Ui ) such that
(9.1) gi |S∩Ui ≡ 0 and gi − gj = gij on Ui ∩ Uj .
If such gi are found then the function F ∈ O(X) given by
F := Fi − gi on Ui
is well-defined, since on Ui ∩ Uj one has Fi − gi = gij + Fj − (gij + gj ) = Fj − gj . Moreover

for x ∈ S, if x ∈ Uj then F (x) = Fj (x) − gj (x) = f (x) − 0 = f (x), and hence we have our
extension.
Let {χj } be a partition of unity subordinate to {Uj }. Define
X
g̃i := χm gik .
m
105
Then g̃i |S∩Ui ≡ 0 and
X X
g̃i − g̃j = χk (gim − gjm ) = χm gij = gij ∈ O(Ui ∩ Uj ).
m m
Of course, g̃i are not holomorphic, but since their differences are holomorphic, the (0, 1)-form α
defined by
α := ∂g̃¯ j on Uj
is globally defined on X. We seek a solution u ∈ C ∞ (X) of the equation ∂u−α¯ such that u|S ≡ 0.
If we find such a function u then the functions gi := g̃i − u satisfy (9.1) and the proof is finished.
To obtain our function u we shall use Hörmander’s Theorem with singular metrics. The idea is
to construct a weight function ψ such that e−ψ is not locally integrable at any point of S. If u has
finite L2 norm with respect to such a weight then u must vanish along S. Of course, the form α
must also be L2 with respect to the weight, so we cannot make this weight too singular along S; it
has to be just right.
Next we claim that
|α|2ω
Z
2(n−k)
dVω < +∞.
Uj |yj |
Indeed, since for each m the function gjm vanishes along Uj ∩ Um , there exist holomorphic func-
tions fmj,` such that
n−k
X
gjm = fjm,` yj` .
`=1
Thus
n−k
XX
g̃j = χm fjm,` yj` ,
m `=1
and
n−k
XX
¯ =
∂α ¯ m )fjm,` y ` .
(∂χ j
m `=1
Thus by Cauchy-Schwarz
n−k
!
Z
|α|2ω X Z X ¯ m |2
|∂χ ω
(9.2) dV ω ≤ |fjm,` |2 dVω < +∞,
Uj |yj |2(n−k) m Uj `=1
|yj |2(n−k−1)
where the finiteness follows because |y|2−2d is integrable in any relatively compact neighborhood
of 0 in Cd .
Now consider the weight
X
ψ := h ◦ ρ + χj log |yj |2(n−k) ,
j
106
where h is a smooth, convex, sufficiently rapidly increasing function. Then
¯ ◦ ρ = (h0 ◦ ρ)∂ ∂ρ
∂ ∂h ¯ + (h00 ◦ ρ)∂ρ ∧ ∂ρ
¯ ≥ (h0 ◦ ρ)ω,
and
!
X X
∂ ∂¯ χj log |yj |2(n−k) = χj ∂ ∂¯ log |yj |2(n−k)
j j
X
¯ j ) log |yj |
2(n−k) ¯ j
+ (n − k)|yj |−2 ∂χj ∧ (yj · dȳj ) + (yj · dyj ) ∧ ∂χ

+ (∂ ∂χ
j
X
¯ j
¯ j ) log |yj |2(n−k) + (n − k)|yj |−2 ∂χj ∧ (yj · dȳj ) + (yj · dyj ) ∧ ∂χ

≥ (∂ ∂χ .
j
We claim that the right hand side is locally bounded. To see this booundedness, fix j ≥ 1, and
denote by k1 , ..., kN the set of k such that Uj ∩ Uk 6= Ø. Because the local coordinates yj cut out
S, there exist smooth functions hµj`,ν such that
n−k
X
ykν` = exp(hνj`,µ )yjµ , 1 ≤ ν ≤ n − k.
µ=1
Then on Uj we have
N
X N
X
¯ k ) log |yk |2(n−k) =
(∂ ∂χ ¯ k ) log |yj |2(n−k) + a smooth function
(∂ ∂χ
` ` `
`=1 `=1
N
!
X
= ∂ ∂¯ χk` log |yj |2(n−k) + a smooth function
`=1
¯ log |yj |2(n−k) + a smooth function

= ∂ ∂(1)
= a smooth function.
The proof of boundedness of the other terms is similar, and is left to the reader. Additionally, the
above argument shows that
ψ = log |yj |2(n−k) + a smooth function
in a neighborhood of S ∩ Uj . Therefore
(a) e−ψ is not locally integrable near any point of S, and
(b) by (9.2) Z
|α|2ω e−ψ dVω < +∞.
Uj
107
Now choose h increasing so rapidly– here it is crucial that ρ is an exhaustion– that
√
Z
¯ + Ricci(ω) ≥ ω and
−1∂ ∂ψ |α|2ω e−ψ dVω < +∞.
X
¯ = α and
By Hörmander’s Theorem there is a function u ∈ C ∞ (X) such that ∂U
Z
|u|2 e−ψ dVω < +∞.
X
Because u is smooth and e−ψ is not locally integrable at any point of S, u must vanish identically
along S. The proof is therefore complete.
Note that if the function h in the proof of Theorem 9.1.2 is chosen correctly then the function
f to be extended satisfies the estimate
Z
(9.3) |f |2 e−ψ dVω < +∞,
S
and therefore we have shown the following: for every f ∈ O(S) satisfying (9.3) there exists
F ∈ O(X) such that Z
F |S = f and |F |2 e−ψ dVω < +∞.
X
An application of the Closed Graph Theorem shows that in fact there exists a constant C > 0,
independent of f , such that
Z Z
|F | e dVω ≤ C |f |2 e−ψ dVω .
2 −ψ
X S
However C does depend on S, X, ω and ψ, and hence this extension theorem is entirely too
specialized.
As we will see in Section 10.2, it is extremely useful to have a result that relies much less on
the data. In the next section we state and prove such a result.
9.1.2 Stein manifolds

Stein manifolds are fundamental objects in complex analysis and geometry. There are several
equivalent definitions, which we shall now review.
9.1.3 D EFINITION . A complex manifold X is said to be Stein if the following conditions hold.
(HC) The manifold X is holomorphically convex, i.e., for any closed discrete subset S ⊂ X there
exists a holomorphic function f ∈ O(X) such that
sup lim |f | = +∞.

S
108
(SP) The algebra O(X) separates points, i.e., if x, y ∈ X are distinct points then there exists
f ∈ O(X) such that f (x) 6= f (y).
∗1,0
(ST) The algebra O(X) separates tangents, i.e., for any x ∈ X and α ∈ TX,x there exists
f ∈ O(X) such that df (x) = α.
Note that any closed submanifold of Cn has these three properties, and in fact one can realize
each property with affine-linear functions. Thus closed submanifolds of Cn are Stein. In fact, a
celebrated theorem of R. Narasimhan concludes the converse.
9.1.4 T HEOREM (R. Narasimhan). Every Stein manifold can be embedded as a closed submanifold
of a finite-dimensional complex vector space.
Narasimhan’s theorem is rather deep and technical, and unfortunately lies outside the range of
our goals. Perhaps more relevant to our presentation is the following theorem of Grauert.
9.1.5 T HEOREM (Grauert). A complex manifold is Stein if and only if it is strongly pseudoconvex.
To prove that strongly pseudoconvex manifolds are Stein, one constructs holomorphic functions
that realize the three properties of Stein manifolds. Properties (HC) and (PS) follow immediately
from Theorem 9.1.2 applied to the 0-dimensional submanifolds S = {pj } and S = {x, y} ⊂ X
respectively. Property (TC) follows from a slight modification of the proof of Theorem 9.1.2 for
the 0-dimensional manifold S = {x} ⊂ X; we leave it to the exercises.
The other direction of Grauert’s Theorem is a consequence of the following stronger, often
very useful result (though we will not make use of it in these notes).
9.1.6 T HEOREM . Let X be a Stein manifold, K a compact subset of X, and U a neighborhood of
b O(X) . Then there is a strictly plurisubharmonic exhaustion u ∈ C ∞ (X) such that u < 0 on K
K
but u > 0 on X − U . In particular, X is strongly pseudoconvex.
Proof. Since X is holomorphically convex, we can find compact sets K1 = K, K2 , ..., such that
for all i ≥ 1 [
Ki = (K
di )
O(X) , K i ⊂ interior(K i+1 ) and Ki = X.
i≥1
Next, choose open sets Ui such that Ki ⊂ Ui ⊂ Ki+1 , and in addition U1 ⊂ U .

For each i ≥ 1, choose functions fik ∈ O(X), k = 1, ..., ki such that
sup |fik | < 1 and max |fik (z)| > 1

Ki k
for all z ∈ Ki+2 − Ui . By taking large powers of the fik if necessary, we can assume that
ki
X ki
X
2 −i
sup |fik | < 2 and inf |fik |2 > i.
Ki Ki+2 −Ui
k=1 k=1
Moreover, by property (3) of Stein manifolds, we can also make sure that at each point of Ki n of
the functions fik , 1 ≤ k ≤ ki , form a local coordinate system.
109
Now set
∞ X
X ki
u := −1 + |fik |2 .
i=1 k=1
The sum converges uniformly on each Ki , but moreover u is smooth. (The simplest way to see the
smoothness of u is to consider the function
∞ X
X ki
F (z, ζ̄) := −1 + fik (z)fik (ζ)
i=1 k=1
on the complex manifold X × X † , where X † is the complex manifold with the complex conjugate
structure of X. For the same reasons as above, F converges locally uniformly on X × X † and is
therefore holomorphic. Restricting to the real submanifold z = ζ establishes the smoothness (and
even real-analyticity) of u.) Next, u > i − 1 on X − Ui . Finally,
∞ X
ki
√ X
¯
−1∂ ∂u(ξ, ξ) = |∂fik (ξ)|2 .
i=1 k=1
Since at each point of X at least one of these terms is strictly positive for non-zero ξ, we see that
u is strictly plurisubharmonic. The proof is complete.
9.1.3 Compact complex manifolds with positively curved line bundles

In the previous section we studied complex manifolds that admit a strictly plurisubharmonic ex-
haustion function. Such manifolds are never compact, since on a compact complex manifold there
are no non-constant plurisubharmonic functions. There are, however, such functions locally, and
therefore it makes sense to ask if there is a holomorphic line bundle with a smooth Hermitian
metric whose curvature is strictly positive. Such line bundles are often called positive.
As we will see in the next paragraph, not every compact complex manifold admits a positively
curved Hermitian holomorphic line bundle. In the present paragraph we will prove that line bundles
with smooth Hermitian metrics of sufficiently positive curvature have many holomorphic sections.
The precise statement is as follows.
9.1.7 T HEOREM . Let X be a compact complex manifold and let H → X be a holomorphic √ line
¯ is positive, i.e., such that −1∂ ∂ϕ
bundle with smooth Hermitian metric e−ϕ whose curvature ∂ ∂ϕ ¯
is a Kähler form. Then for every holomorphic line bundle L → X finite subset {x1 , ..., xN } ⊂ X
and positive integers d1 , ..., dN there exists an integer
mo = mo (X, H, L, x1 , ..., xN , d1 , ..., dN )
such that for any polynomials p1 , ..., pN ∈ O(Cn ) of degrees deg(pi ) = di , any local coordinates
zi near xi , 1 ≤ 1 ≤ N , and holomorphic frames ξi for H and ηi for L near xi , 1 ≤ i ≤ N , and
any integer m ≥ mo there exists a section s ∈ ΓO (X, L⊗m ) such that
s(zi ) = pi (zi ) + O |zi |di +1 ξi⊗m ⊗ ηi , 1 ≤ i ≤ N.

110
In other words for a sufficiently positive line bundle one can specify for a holomorphic section
the Taylor polynomials of any given degree at any finite set of points.
Proof. Choose positive numbers ε1 , ..., εN such that the local coordinate patches |zi | < 3εi are
pairwise disjoint, and fix a function χ ∈ Co∞ ([0, ∞)) such that χ|[0,1] ≡ 1 and Support(χ) ⊂ [0, 2].
Consider the smooth section s̃ of H ⊗m ⊗ L → X defined by
s̃(x) := χ(|zi |2 /ε2i )p(zi )ξi⊗m ⊗ ηi on Ui := {|zi | < 2εi }, i = 1,...,N,
and s̃ = 0 away from the coordinate neighborhoods Ui . Note that s̃ is holomorphic on the open
¯ is smooth and
sets Vi := {|zi | < εi }. and therefore the H ⊗m ⊗ L-valued (0, 1)-form α := ∂s̃
supported on the union A1 ∪ ... ∪ AN of the annuli
Ai := Ui − Vi , 1 ≤ i ≤ N.
√
Fix the Kähler form ω := ¯ for X and any smooth metric e−ψ for L → X, and define the
−1∂ ∂ϕ
function
N
X
ρ := χ(|zi |/εi ) log |zi |2(di +n) .
i=1
Evidently ρ is smooth away from the points xi and ρ|Ui = (di + n) log |zi |2 is plurisubharmonic. It
follows that Z
|α|2ω e−(mϕ+ψ+ρ) dVω < +∞
X
and that there is a constant C > 0 such that
√
¯ ≥ −Cω.
−1∂ ∂ρ
By taking mo so large that

√ √
¯ + ∂ ∂ψ
mo −1∂ ∂ϕ ¯ + ∂ ∂ρ
¯ − −1∂ ∂¯ log det ω ≥ ω,
we see that for any m ≥ mo the curvature of the metric metric e−mϕ+ψ+ρ dVω for the line bundle
H ⊗m ⊗ L ⊗ KX ∗
is more than ω. By Hörmander’s Theorem there exists a section u of H ⊗m ⊗ L
such that Z Z
¯
∂u = α and 2 −(mϕ+ψ+ρ)
|u| e dVω ≤ |α|2ω e−(mϕ+ψ+ρ) dVω .
X X
¯ = α is necessarily smooth. Since e−ρ ∼ |zi |−2(di +n) on Ui , it follows
Any section u satisfying ∂u
that u vanisheds to order di + 1 at xi . Consequently the section
s := s̃ − u
is holomorphic and has the desired properties.
111
9.1.4 The Kodaira Embedding Theorem
In this paragraph we establish the following fundamental theorem of algebraic geometry.
9.1.8 T HEOREM (Kodaira Embedding Theorem). A compact complex manifold X is projective if
and only if there is a positively curved holomorphic line bundle on X.
Theorem 9.1.7 is the key ingredient in the proof of Theorem 9.1.8, as we shall soon see.
The easy direction: necessity of the existence of a positively curved line bundle
One direction of Theorem 9.1.8 is rather straight-forward. If a complex manifold M has a pos-
itively curved Hermitian holomorphic line bundle then any submanifold S of M also has such a
line bundle, namely the restriction to S of the line bundle on M . Thus to prove the easy direc-
tion of Theorem 9.1.8 we need only show that Pn has a positively curved Hermitian holomorphic
line bundle. We have already met the line bundle in question: this is the hyperplane line bundle
H → Pn of Example 1.3.9, whose sections are in 1 − 1 correspondence with linear functions on
Cn+1 (c.f. Exercise 1.3.5). Let us fix a basis z 0 , ..., z n ∈ (Cn+1 )∨ . Evidently at each point of Pn at
least one of these sections does not vanish. Using this basis we define the so-called Fubini-Study
metric e−ϕF S for H → Pn as follows. Let ` ∈ Pn be a 1-dimensional subspace of Cn+1 . Any
λ ∈ H` is a linear functional on `, and therefore we define
|λ(v)|2
|λ|2 e−ϕF S (`) = Pn i 2
for any v ∈ ` − {0}.
j=1 |z (v)|
Note that the right hand side is independent of the choice of v in ` − {0}.
Let us compute the curvature of e−ϕF S . Without loss of generality we may work in the open set
Uo and with coordinates ζ = (ζ 1 , ..., ζ n ), where ζ i = z i /z 0 . In this open set the section z o vanishes
nowhere, and therefore trivializes H. Evidently
1
|z 0 |2 e−ϕF S = ,
1 + |ζ|2
and therefore the curvature is √
−1∂ ∂¯ log(1 + |ζ|2 ),
which is a strictly positive (1, 1)-form.
Maps into projective spaces

Before understanding the structure of all maps to projective spaces, it is helpful to construct some
examples of such maps. The following set of examples, though rather abstract, turns out to be
fundamental.
9.1.9 E XAMPLE . Let X be a complex manifold and let H → X be a holomorphic line bundle.
Fix a finite-dimensional subspace W ⊂ ΓO (X, H) of holomorphic sections of H → X. For each
x ∈ X one can define
φW (x) := {s ∈ W ; s(x) = 0} ⊆ W.
112
Of course, it can happen that every section of W vanishes at x, i.e., φ|W | (x) = W , but if for a given
x ∈ X one has a section s ∈ W such that s(x) 6= 0 then φ|W | (x) is a hyperplane in W .
9.1.10 D EFINITION . The set
Bs(W ) := {x ∈ X ; s(x) = 0 for all s ∈ W }
is called the base locus of W . One says W is basepoint-free if Bs(W ) = Ø.

By using the canonical identification of a hyperplane H ⊂ W with the line in W ∨ consisting
of all linear functionals that vanish on H one has a map
φW : X − Bs(W ) → P(W ∨ ).
Thus we have a large collection of maps into projective space. In words, this map sends a point x
to the line in P(W ∨ ) determined by the kernel of linear function ExW : W → Hx of evaluation of
sections at the point x.
Remarkably, every holomorphic map into a projective space is almost (but not quite) of the
form described in Example 9.1.9. The precise result is as follows.
9.1.11 P ROPOSITION . Let F : X → P(V ) be a holomorphic map. Then there exist
(a) a holomorphic line bundle HF → X,
(b) a finite-dimensional subspace WF ⊂ ΓO (X, HF ), and
(c) an injective linear map PF : WF∨ → V
such that
F = ℘F ◦ φWF ,
where ℘F : P(WF∨ ) 3 [w] 7→ [PF w] ∈ P(V ) is the injective map induced by PF .
In other words, up to inclusion by projective subspaces, every holomorphic map to a projective
space is of the form given in Example 9.1.9.
Proof of Proposition 9.1.11. Let F : X → P(V ) be given. The image F (X) is contained in a
minimal projective subspace P(Vo ) ⊂ P(V ); this projective subspace corresponds to the smallest
linear subspace Vo ⊂ V all 1-dimensional subspaces ` ⊂ V such that ` ∈ F (X) (when ` is viewed
as a point of P(V ). Let us write
Fo : X → P(Vo )
for the map obtained after the range of F is changed from P(V ) to P(Vo ), i.e., F = ιo ◦ Fo , where
ιo : P(Vo ) ,→ P(V ) is the inclusion.
The subspace Vo is the intersection of all the hyperplanes in V that are the zero loci of sections
s ∈ V ∨ = ΓO (P(V ), H) (c.f. Exercise 1.3.5) vanishing identically on F (X). The set of such
sections s is a subspace SF of V ∨ , and the quotient space
V ∨ /SF
113
is naturally isomorphic to the dual space Vo∨ = ΓO (P(Vo ), H). Moreover, every section so ∈ Vo∨
that vanishes on F (X) ⊂ P(Vo ) vanishes identically. It follows that the map
Fo∗ : ΓO (P(Vo ), H) → ΓO (X, F ∗ H)
is injective1 .
We let
LF := Fo∗ H, WF := Fo∗ (ΓO (P(Vo ), H)) and PF := Fo∗ ◦ πSF ,
where πSF : V ∨ → V ∨ /SF ∼
= Vo∨ is the quotient map. With these objects, all of the claimed
properties are established.
9.1.12 R EMARK . For the reader that prefers n-tuples to vectors and matrices to linear functionals,
the map F has a useful description. Suppose F (X) lies in a proper projective subspace P(Vo ) ⊂
P(V ). Choosing a basis z 0 , ..., z n for V ∨ such that z 0 , ..., z k is a basis for W , we get projective
coordinates [z 0 , ..., z n ] for P(V ) and [z 0 , ..., z k ] for P(W ∨ ). In terms of these coordinates we may
write F as
F = [F 0 , ..., F k , 0, ..., 0].
The map G : P(W ∨ ) → P(V ) is given by
G([z 0 , ..., z k ]) = [z 0 , ..., z k , 0, ..., 0]
and the hyperplane in W ∨ is corresponding to CF (x) ∈ V is, with respect to the basis z 0 , ..., z n ,
the set of all vectors c = (co , ..., ck ) ∈ Ck+1 ∼
= W ∨ satisfying
c · (F 0 (x), ..., F k (x)) = 0.
Note also that F i ∈ ΓO (X, F ∗ H).
9.1.13 P ROPOSITION . Let H → X be a holomorphic line bundle and let W ⊂ ΓO (X, H) be
basepoint free.
1. The map φW : X → P(W ∨ ) is injective if and only if for any pair of distinct points x, y ∈ X
there exists a section s ∈ W such that s(x) = 0 and s(y) 6= 0.
2. The map φW : X → P(W ∨ ) is an immersion if and only if for any x ∈ X, α ∈ TX∗1,0 and
v ∈ Hx there exists s ∈ W such that s(x) = 0 and ds(x) = α ⊗ v.
Proof. If s(x) = 0 6= s(y) then φW (x) 3 s 6∈ φW (y), so φW separates x and y. Conversely if
φW (x) = φW (y) then every section that vanishes at y also vanishes at x. Thus 1 is proved. We
leave 2 as an exercise.
Proof of Theorem 9.1.8. Let L → X be a positively curved Hermitian holomorphic line bundle.
By Proposition 9.1.13 it suffices to show that there exists a holomorphic line bundle H → X such
that W = ΓO (X, H) is basepoint-free, separates points (i.e. satisfies Item 1) and separates tangents
(i.e. satisfies Item 2). But by Theorem 9.1.7 the line bundle H = L⊗m has these properties as soon
as m is large enough.
1
Recall that if E → Y is a holomorphic vector bundle and G : X → Y is a holomorphic map then the pullback
∗
G E → X of E by G is by definition the restriction of the Cartesian projection X × E → X to the submanifold
G∗ E := {(x, v) ∈ X × E ; v ∈ EG(x) } of X × E.
114
9.2 L2 Extension
Let us finally begin the journey to our universal extension theorem.
9.2.1 Adjunction
In order to prove our universal extension theorem, we need to have an appropriate normalization
of the data. Part of this preparation involves the so-called adjunction construction, which we now
explain.
Let X be a complex manifold and let Z ⊂ X be smooth complex hypersurface, i.e., a complex
submanifold of codimension 1. Then there exists an open cover {Uj } of X, and holomorphic
functions Tj ∈ O(Uj ) such that
Z ∩ Uj = {Tj = 0} and dTj (x) 6= 0 for all x ∈ Uj ∩ Z.
Denote by LZ → X the holomorphic line bundle associated to Z, i.e., the line bundle defined by
the transition functions
Ti
gij = ,
Tj
which have no zeros on Ui ∩Uj . Since Ti = gij Tj , the functions Tj define a section T ∈ ΓO (X, LZ ),
and evidently the zero locus of T is precisely Z, counting multiplicity. Note also that since
dTi = gij dTj + dgij Tj ,
along Z the sections dTj provide a well-defined holomorphic section dT of TX∗1,0 ⊗ LZ |Z . More-

over, since the LZ -valued holomorphic 1-forms dTj annihilate the tangent spaces of Z, dT is a
holomorphic section of the rank-1 subbundle
∗
NX/Z ⊗ LZ → Z
of TX∗1,0 ⊗ LZ |Z . The line bundle NX/Z ∗

, whose fibers consist of (1, 0)-forms that annihilate the
(1, 0)-tangent spaces of Z, is called the conormal bundle of Z in X.
∗
Since Z is smooth the section dT is nowhere zero, and therefore the line bundle NX/Z ⊗ LZ is
trivial. In particular, we see that for a smooth hypersurface Z the line bundle LZ → X restricts to
Z as the dual of the conormal bundle, i.e., the line bundle
∗
NX/Z := (HX/Z )∗ .
This line bundle, which is naturally isomorphic to the quotient (TX1,0 |Z )/TZ1,0 , is called the normal
bundle. In other words we have proved the following proposition.
9.2.1 P ROPOSITION (Adjunction Formula). The holomorphic line bundle LZ → X of a smooth
hypersurface Z is an extension to X of the normal bundle of Z.
By taking determinants, one can see that the Adjunction Formula is equivalent to the formula
KZ = (KX ⊗ LZ )|Z ,
and the identification sends an (n − 1, 0)-form f on Z to the LZ -valued n form f ∧ dT , defined
along Z.
115
9.2.2 Statement of the L2 Extension Theorem
9.2.2 T HEOREM . Let X be a Stein manifold and let Z ⊂ X be a smooth complex hypersurface.
Assume there exists a section T ∈ ΓO (X, LZ ) and a singular Hermitian metric e−λ for LZ → X
such that
(9.4) Z = {T = 0} and sup |T |2 e−λ ≤ 1.

X
Let g be a Kähler metric on X and let L → X be a holomorphic line bundle with singular
Hermitian metric e−ϕ , and assume there exists a constant δ ∈ (0, 1] such that
(9.5) ¯ + Ricci(g) ≥ (1 + tδ)∂ ∂λ

∂ ∂ϕ ¯ for all t ∈ [0, 1].
Then for each f ∈ ΓO (Z, L) such that
|f |2 e−ϕ dAg
Z
< +∞
Z |dT |2g e−λ
there exists F ∈ ΓO (X, L) such that
|f |2 e−ϕ dAg
Z Z
2 −ϕ 24π
F |Z = f and |F | e dVg ≤ ,
X δ Z |dT |2g e−λ
ω n−1
where dAg = |
(n−1)! Z
is the area form associated to the submanifold Z.
In fact, there are many other versions of the L2 extension theorem, and all of them can be
established by slight modifications of the proof of Theorem 9.2.2 that we shall give below. We
give a brief discussion of some of these, and of the problems that arise in their consideration.
1. One version that can be established involves replacing the line bundle L → X with a vector
bundle of higher rank. The trouble with such a version (and the reason we did not state it
here) is that there is at present no manageable definition of a singular Hermitian metric for
holomorphic vector bundles of higher rank. The difficulty is that there seems to be no such
definition in which the curvature can be defined as a current; this problem makes it difficult
to see how to handle condition (10.6) in the higher rank case.
2. In another version of Theorem 9.2.2 that is very useful, one relaxes the hypothesis that X is
Stein. Here again the trouble is that on a general (even complete) Kähler manifold there is
no way to approximate singular Hermitian metrics by smooth ones without losing too much
positivity. There are, however, some Kähler manifolds where this approximation is in some
sense possible.
9.2.3 D EFINITION . A complex manifold X is said to be essentially Stein if there is a (possibly
singular) complex hypersurface V such that the manifold X −V is Stein. A submanifold Z ⊂
X is said to be an essentially Stein submanifold if the hypersurface V whose complement is
Stein can be chosen so that Z 6⊂ V .
116
The most interesting examples of an essentially Stein manifold are projective manifolds, and
so-called projective families, i.e., spaces X for which there are
(a) a proper holomorphic submersion f : X → B onto some complex manifold B (i.e.,

X → B is a holomorphic family), and
(b) a holomorphic line bundle A → X with a smooth Hermitian metric of positive cur-
vature.
In particular, condition (a) implies that the fibers of f , all of which are complex submanifolds
of X , are pairwise diffeomorphic, and condition (b) implies, via the Kodaira Embedding
Theorem, that each fiber is a projective manifold.
If one has L2 estimates for sections of holomorphic line bundles (or for that matter, vec-
tor bundles) on a Stein complement of a hypersurface then those sections can be extended
across the hypersurface. Indeed, a priori the sections could have only poles or removable
singularities along the divisor, but the finiteness of the L2 norm forbids poles.
3. Yet another version of Theorem 9.2.2 considers different L2 norms for the sections of the line
bundle L → Z, or of the line bundle L → X, or perhaps both. One version of such a result
was worked out by Jeff McNeal and the author; the interested reader can go to [MV-2007]
or [MV-2015] for more information.
4. One also wants to extend holomorphic sections from submanifolds of higher codimension.
Extension theorems of this sort do exist, but they require specifying different sorts of data.
Again, proofs are not so different from what is presented here, but the notion of positivity
that is required in place of condition (10.6) becomes more complicated to state, and harder
to understand geometrically. There are short cuts that can be taken which yield useful and
interesting results– see, for example, [BL-2016]– but the most general case, which was first
treated by Ohsawa in [O-2001], is still not in its final form, in the author’s opinion.
9.2.3 Proof of Theorem 9.2.2

We fix once and for all the section f ∈ ΓO (Z, L|Z ) to be extended, i.e., which satisfies
|f |2 e−ϕ dAg
Z
2 −λ
< +∞.
Z |dT |g e
Since X is Stein, it is strongly pseudoconvex. Using the strictly plurisubharmonic exhaustion

function, one finds domains
[
Ω1 ⊂⊂ Ω2 ⊂⊂ ... ⊂⊂ X such that Ωj = X and Z t ∂Ωj for all j.
j
We shall work on these domains first, and then then take limits of our results.
117
The domain Ωj is trivially a strongly pseudoconvex manifold. By Exercise ?? there exists
Fo ∈ ΓO (X, L) such that
Fo |Z = L.
Therefore Z
|Fo |2 e−ϕ dVg < +∞,
Ωj
but at present we have no estimate on this L2 norm; in particular, one cannot let j → ∞ and
conclude, or even expect, that finiteness persists. After all, there are many extensions of f and
most of them will not have finite L2 norm.
We must therefore correct Fo somehow away from Z ∩ Ωj . To do so, let t ∈ (0, 1) (we will
eventually send t to 0) and let χ ∈ Co∞ ([0, ∞)) be a positive function supported in [0, 1] and
satisfying
0 ≤ χ ≤ 1, χ ≡ 1on [0, t] and |χ0 | ≤ 1 + ct,
where c > 1. (Eventually we will let c → 1 and then t → 0.) We define χε := χ(|T |2 e−λ /ε2 ).
Then for ε > 0 sufficiently small the section
F̃ε := χε Fo
of L → Ωj is smooth, and holomorphic in a tubular neighborhood of Z ∩ Ωj .

We wish to correct F̃ε to be a holomorphic extension of f on Ωj that has better estimates than
¯ = αε , where
Fo . To find such a correction we will solve the equation ∂u
¯ ε Fo .
αε := ∂χ
Note that αε is supported on the annular tube {ε2 t ≤ |T |2 e−λ ≤ ε2 }.

To solve this equation we will use the twisted estimates of Donnelly-Fefferman-Ohsawa, i.e.,
Theorem 8.2.1, which we state again for the situation at hand (i.e., a vector bundle of rank 1 and a
singular Hermitian metric), and for ease of reading.
9.2.4 T HEOREM (Donnelly-Fefferman-Ohsawa Estimate; singular rank 1 case). With the notation
of Theorem 9.2.2, let Ωj be as above. Let e−ψ be a singular Hermitian metric for L|Ωj and let
η : X → R and a : X → (0, ∞) be functions with η C 2 -smooth, such that

−η ¯ ¯ 1+a ¯
Ψ := e ∂ ∂ψ + Ricci(g) + ∂ ∂η − ∂η ∧ ∂η ≥ Θ ≥ 0.
a
Then for each L|Ωj -valued (0, 1)-form ϕ such that
Z
¯
∂ϕ = 0 and |ϕ|2Θ e−ψ dVg < +∞
Ωj
there exists a measurable section U of L|Ωj such that

p Z Z
¯
∂ −η
e (1 + a)U = ϕ and 2 −ψ
|U | e dVg ≤ |ϕ|2Θ e−ψ dVg .
Ωj Ωj
118
9.2.5 R EMARK . As we have suggested earlier, Theorem 9.2.4 follows from Theorem 8.2.1 on
Stein manifolds by approximation of singular metrics with smooth ones. In the present situation
there is also the issue that the (1, 1)-form Θ is not assumed to be strictly positive. The issue can be
dealt with by perturbing the metric e−ψ to be strictly positively curved; since one is working on a
domain Ωj ⊂⊂ X one can add a small multiple of the plurisubharmonic exhaustion to ψ.
To make use of Theorem 9.2.4 we must choose a singular Hermitian metric e−ψ , a smooth
function η and a positive function a such that Ψ ≥ Θ for some non-negative Θ for which the
integral Z
|αε |2Θ e−ψ dVg
Ωj
is finite. Moreover, in order to make sure that the correction
Fε := χε Fo − e−η (1 + a)U
remains an extension, the metric e−ψ must be singular along Z ∩ Ωj . We therefore choose
ψ := ϕ + log |T |2 e−λ .
The next task is to choose η and a. To simplify the estimates to come, it is convenient to
introduce the auxiliary functions
v := log |T |2 e−λ : X → [−∞, 0) and s := γ − δ log(ev + ε2 ) : X → [1, ∞),
where δ is as in Theorem 9.2.2 and γ := 1 + δ log(1 + ε2 ) ∈ (1, ∞).

With v and s in hand, we set
η = − log(2 + log(2es−1 − 1)).
At first glance, the choice of η is bewildering. We do not have a great explanation for this choice,
which comes out of necessity for obtaining good estimates later on. Perhaps the most important
clue is that this function is very similar to the self-bounded gradient function of Example 8.4.4 on
the punctured unit disk; the idea is that a tubular neighborhood of the hypersurface Z looks very
much like the product of Z with the disk, and if we are creating a singularity near Z, a function of
this form will give us some gain in positivity.
Finally the function a is chosen only to simplify the expression for Ψ so that it becomes
manageable. For this reason it is convenient to define H(x) := 2 + log(2ex−1 − 1) (so that
η = − log H(s)). Then
H 0 (s) H 0 (s)2

¯ 1+a ¯ ¯ 1 00 ¯
∂ ∂η − ∂η ∧ ∂η = (−∂ ∂s) − H (s)a + ∂s ∧ ∂s,
a H(s) aH(s) H(s)
and so we choose
−(H 0 (s))2
a = 00 ,
H (s)H(s)
119
which yields
0
¯ − 1+a ¯ = H (s) (−∂ ∂s).
¯
∂ ∂η ∂η ∧ ∂η
a H(s)
Note that for x ≥ 1,
2ex−1 1 −2ex−1
H(x) ≥ 1, H 0 (x) = x−1
= 1 + x−1 and H 00 (x) = ,
2e −1 2e −1 (2ex−1 − 1)2
so that a is a positive function– one of our requirements. Note also that e−η and a are smooth, a
fact we shall use later.
We compute that
v¯
−∂ ∂s¯ = δ∂ ∂¯ log(e + ε ) = ∂ δe ∂v
v 2
ev + ε2
δev 2 v/2 ¯ v/2
= v ∂ ¯ + δ 4ε ∂(e ) ∧ ∂(e )
∂v
(e + ε2 ) ((ev/2 )2 + ε2 )2
ev 2 v/2 ¯ v/2
¯ + δ 4ε ∂(e ) ∧ ∂(e )
¯ log |T |2 − ∂ ∂λ

=δ v ∂ ∂
(e + ε2 ) ((ev/2 )2 + ε2 )2
δev 2 v/2 ¯ v/2
=− v ∂ ¯ + δ 4ε ∂(e ) ∧ ∂(e ) .
∂λ
(e + ε2 ) ((ev/2 )2 + ε2 )2
In the last equality we have used the fact that ∂ ∂¯ log |T |2 is supported on Z, where ev = |T |2 e−λ
vanishes.
Now, the function
2es−1 1
[1, ∞) 3 s 7→ 2 + log(2es−1 − 1) − = 1 + log(2e s−1
− 1) −
2es−1 − 1 2es−1 − 1
2
s−1 s−1 (2es−1)
has derivative 2e2es−1 −1 + (2e2e
s−1 −1)2 = (2es−1 −1)2
> 0, and therefore it takes its minimum at s = 1,
where it vanishes. Consequently
2es−1
H 0 (s) ev 2es−1 −1
· v 2
≤ ,
H(s) e + ε 2 + log(2es−1 − 1)
and we find that
0
¯ + Ricci(g) + ∂ ∂η
∂ ∂ψ ¯ − 1 + a ∂η ∧ ∂η ¯ + Ricci(g) + H (s) (−∂ ∂s)
¯ = ∂ ∂ψ ¯
a H(s)
H 0 (s) ev 0 v

¯ + Ricci(g)) + H (s) · e ¯ + Ricci(g) − δ∂ ∂λ
¯

= 1− · v (∂ ∂ψ ∂ ∂ψ
H(s) e + ε2 H(s) ev + ε2
¯ v/2 )
H 0 (s) 4ε2 ∂(ev/2 ) ∧ ∂(e
+δ
H(s) ((ev/2 )2 + ε2 )2
¯ v/2 )
H 0 (s) 4ε2 ∂(ev/2 ) ∧ ∂(e
≥δ ,
H(s) ((ev/2 )2 + ε2 )2
120
where the last inequality follows from the definition of ψ and the curvature hypothesis (10.6) of
Theorem 9.2.2. Since H 0 (s) ≥ 1,
2 v/2 ¯ v/2
Ψ=e −η ¯ + Ricci(g) + ∂ ∂η
∂ ∂ψ ¯ − 1 + a ¯ ≥ δ 4ε ∂(e ) ∧ ∂(e ) .
∂η ∧ ∂η
a ((ev/2 )2 + ε2 )2
We take Θ to be the right-hand side of the inequality, and prepare to estimate the L2 norm of αε .
Now,
¯ −2 ev )Fo = χ0 (ε−2 ev )Fo ε−2 ∂e
αε = ∂χ(ε ¯ v/2 ),
¯ v = χ0 (ε−2 ev )Fo 2ε−2 ev/2 ∂(e
so
1 0 −2 v 2 1
|αε |2Θ e−ψ = 6
|χ (ε e )| |Fo |2 e−ψ ev (ev + ε2 )2 = 6 |χ0 (ε−2 ev )|2 |Fo |2 e−ϕ (ev + ε2 )2 .
δε δε
Consequently
Z Z
1
|αε |2Θ e−ψ dVg = 6 |χ0 (ε−2 ev )|2 |Fo |2 e−ϕ (ev + ε2 )2 dVg
Ωj δε Ωj
Z
1
= 6 |χ0 (ε−2 ev )|2 |Fo |2 e−ϕ (ev + ε2 )2 dVg
δε {ev <ε2 }
4(1 + ct)2
Z
≤ |Fo |2 e−ϕ dVg .
δε2 v 2
{e <ε }
In particular, note that
|f |2 e−ϕ dAg
Z Z
4π
lim sup |αε |2Θ e−ψ dVg ≤ (1 + ct)2 .
ε→0 Ωj δ Z |dT |2g e−λ
By Theorem 9.2.4 there exists a section Uε of L → Ωj such that
|f |2 e−ϕ dAg
Z Z
¯
p
−η 2 −ψ 8π 2
∂( e (1 + a)Uε ) = αε and |Uε | e dVg ≤ (1 + o(1)) (1 + ct) 2 −λ
Ωj δ Z |dT | e
as ε ∼ 0. p
Since αε is smooth, the function e−η (1 + a)Uε is smooth, and therefore so is Uε . Since
e−ψ = e−ϕ /(|T |2 e−λ ) is not locally integrable on Z, the section Uε must vanish at all points of Z.
Consequently the section p
Fε := χε Fo − e−η (1 + a)Uε
is an extension of f to Ωj . Now,
Z Z
2 −ϕ
|Fε | e dVg = (1 + o(1)) ev−η (1 + a)|Uε |2 e−ψ dVg , ε ∼ 0.
Ωj Ωj
121
−H 0 (s)2
But since s = γ − δ log(ev + ε2 ), e−η = H(s) and a = H 00 (s)H(s)
,
H 0 (s)2 − H 00 (s)H(s)
ev−η (1 + a) = e(γ−s)/δ (γ−s)/δ s−1 s−1

= e 2e + 2 + log(2e − 1) .
−H 00 (s)
A straight-forward calculus exercise shows that, for s ≥ 1, the right hand side is bounded above
by 6eγ−1 = 6(1 + ε2 )−δ .
We therefore have, for every ε > 0, a section Fε,c,t ∈ ΓO (Ωj , L) such that
24π(1 + ct)2 |f |2 e−ϕ dAg

Z Z
2 −ϕ
Fε,c,t |Ωj ∩Z = f |Ωj ∩Z and |Fε,c,t | e dVg ≤ (1 + o(1)) 2 −λ
.
Ωj δ Z |dT |g e
We can now use Alaoglu’s Theorem to take subsequential limits in ε, t and then j, and thereby
obtain a limit F ∈ ΓO (X, L) satisfying
|f |2 e−ϕ dAg
Z Z
2 −ϕ 24π
F |Z = f and |F | e dVg ≤ .
X δ Z |dT |2g e−λ
The proof of Theorem 9.2.2 is therefore complete.
E XERCISES
122
Lecture 10
Some applications of the L2 Extension

Theorem
10.1 Beurling-Seip Theory of Interpolation

In the 1950’s A. Beurling began to study interpolation and sampling problems on Hardy spaces;
he did not publish his work until 1986; it appeared in his final published paper.
A few years after the appearance of Beurling’s paper, K. Seip began to consider the analogous
problem for Bergman spaces. Seip’s motivation for consideration of the problem seems at least in
part to have been linked to the following problem in mathematical solid state physics. Consider
the so-called Fock Space
Z
2 −|z|2 2
F (C) := F ∈ O(C) ; |F (z)| e dA(z) < +∞ = L2 (C, e−|·| dA) ∩ O(C).
C
To a lattice Λ ⊂ C (i.e., a free Abelian subgroup of maximal rank) one associates the little Fock
space ( )
2 −|λ|2
X
f(Λ) := f : Λ → C ; |f (λ)| e < +∞ .
λ∈Λ
The lattice Λ is said to be interpolating or sampling if the restriction map
RΛ : F 3 F 7→ F |Λ ∈ fΛ
is, respectively, surjective or injective. The problem was to find a lattice that is both interpolating
and sampling, or to show there is no such lattice.
Mathematical physicists already knew that the quantity
#Λ ∩ Dr (z)
D(Λ) := lim sup sup ,
r→∞ z∈C r2
called the asymptotic density of Λ, is ≤ 1 if Λ is interpolating, and ≥ 1 if Λ is sampling, and hence
that if a lattice is both interpolating and sampling then it has asymptotic density 1.
123
10.1.1 Seip’s Theorem
Seip showed that if Λ is an interpolating lattice then D(Λ) < 1 and that if Λ is a sampling lattice
then D(Λ) > 1, thus supplying the negative solution to the aforementioned problem. In fact, Seip
showed more. He considered locally finite subsets Γ (for which, unlike lattices, the numbers
#(Γ ∩ D(z, r)) #(Γ ∩ D(z, r))
D+ (Γ) := lim sup sup and D− (Γ) := lim inf inf
r→∞ z∈C r2 r→∞ z∈C r2
called the asymptotic upper and lower densities, are possibly distinct).
The notions of interpolation and sampling sets have to be modified slightly in order to get the
proper result. In this setting a locally finite set Γ is interpolating if the restriction map, in addition
to being surjective, is bounded. The set Γ is said to be sampling if the restriction map, in addition
to being injective, is bounded with closed image.
, which can be provided with the same definition, in terms of the restriction map, of interpola-
tion and sampling sets. and established, partly in joint work with Wallsten, the following theorem.
10.1.1 T HEOREM . A locally finite set Γ is
1. interpolating if and only if Γ is uniformly separated and D+ (Γ) < 1, and
2. sampling if and only if Γ is a finite union of uniformly separated sequences Γ = Γ1 ∪ ... ∪ ΓN

such that D− (Γ1 ) > 1.
A locally finite set Γ is said to be uniformly separated if its separation radius

|γ − µ|
ρ(Γ) := ; γ, µ ∈ Γ, γ 6= µ
2
is positive. The necessity of uniform separation is a relatively simple but instructive result which
we shall explain momentarily.
10.1.2 A more general setting

Let W be a smooth hypersurface in Cn . In interpolation theory one considers the Hilbert spaces
Z
H (C , ϕ) := f ∈ O(C ) ;
n n 2 −ϕ n
|f | e ω < +∞
Cn
and Z
2 −ϕ n−1
H(W, ϕ) := f ∈ O(W ) ; |f | e ω < +∞ ,
W
√
where ω = −1∂ ∂|z| ¯ 2 is the Kähler form of the Euclidean metric. One can then define the
restriction operator
RW : H (Cn , ϕ) → H(W, ϕ)
as the operator sending a function F to its restriction to W .
124
10.1.2 D EFINITION . The hypersurface W is then said to be an interpolation set if RW is bounded
and surjective, and W is said to be a sampling set if RW is bounded, injective and has closed
image.
We will be primarily interested in the surjectivity of RW ; establishing such surjectivity is of
course an L2 extension result, and as such one might imagine a connection with Theorem 9.2.2.
We emphasize, however, the fact that the L2 norm defining the space H(W, ϕ) is not the norm
appearing in Theorem 9.2.2.
In what follows, we shall always assume that our weight ϕ is smooth and satisfies the bound
√ √
(10.1) 0 ≤ −1∂ ∂ϕ ¯ ≤ M −1∂ ∂|z| ¯ 2
for some constant M > 0. The us of this curvature hypothesis, which is clearly satisfied by the
Bargmann-Fock weights ϕ(z) = c|z|2 , c > 0, will become clear in due course.
The most obvious analogue of uniform separation is the notion of uniform flatness; it is defined
as follows.
10.1.3 D EFINITION . A smooth complex hypersurface W ⊂ Cn is said to be uniformly flat if there
exists ε > 0 such that the set
Uε (W ) := {z ∈ Cn ; dist(z, W ) < ε}
is a tubular neighborhood of W , i.e., if z1 , z2 ∈ W and vi ∈ Cn is perpendicular to TW,zi , i = 1, 2,

and z1 + v1 = z2 + v2 then max(|v1 |, |v2 |) > ε.
(Equivalently, any two Euclidean disks or radius < ε and perpendicular to W at their centers do
not intersect.)
10.1.4 D EFINITION . Let T ∈ O(W ) be a global holomorphic function whose zero locus is W ,
counting multiplicity1 . For r > 0 define the function
Z
T 1 1
λr (z) := log |T |2 dV = 1B (0) ∗ log |T |2 .
Vol(Br (z)) Br (z) Vol(Br (0)) r
(i) The function SrW : W → R+ defined by

T
SrW (z) := |dT |2ω e−λr
is called the separation function of W .

(ii) The (1, 1)-current
√
−1 ¯ T 1
ΥW
r := ∂ ∂λr = 1B (0) ∗ [W ]
2π Vol(Br (0)) r
is called the total mass current of W .
1
Recall that in general T is a section of the line bundle LW → Cn associated to W ; here it is a function because
every line bundle on Cn is trivial.
125
(Here [W ] is the current of integration over W , ∗ denotes convolution, and the second equality
follows from the Poincaré-Lelong Formula.)
Note that
T T
¯ T
|T |2 e−λr , |dT |2ω e−λr |W and ∂ ∂λ r
do not depend on the function T used to cut out W .

V. Pingali and the author have proved the following theorem.
10.1.5 T HEOREM . If W is uniformly flat then for each r > 0 there is a positive constant Cr such
that
SrW ≥ Cr
on W .
In fact, if n = 1 the converse of Theorem 10.1.5 is also true. It is at present not known if the
converse is true in general.
10.1.6 D EFINITION . The asymptotic upper and lower densities of W with respect to the weight ϕ
are the numbers
√
¯
Dϕ+ (W ) := lim sup sup a1 ; a > 0 and −1∂ ∂ϕ(v, v) ≥ aΥW n

r (v, v) for all v ∈ C
r→∞ z∈W
and
√

1 ¯
Dϕ− (W ) := lim inf inf ; a > 0 and −1∂ ∂ϕ(v, v) < aΥW
r (v, v) for some v ∈ C
n
r→∞ z∈W a
In other words, Dϕ+ (W ) is the supremum of all numbers 1/a such that ϕ − aλTr is plurisubhar-
monic for all r >> 0, and Dϕ− (W ) is the infimum of all numbers 1/a such that ϕ − aλTr is not
plurisubharmonic for all r >> 0.
From Theorems 9.2.2 and 10.1.5 we immediately obtain the following theorem.
10.1.7 T HEOREM . Let ϕ ∈ Cn be a smooth function and let W ⊂ Cn be a smooth hypersurface.
If Dϕ+ (W ) < 1 then for every f ∈ O(W ) such that
Z
dAW
|f |2 e−ϕ W < +∞
W Sr
there exists F ∈ H (Cn , ϕ) such that F |W = f . In particular, if W is also uniformly flat then W
is an interpolation hypersurface.
In particular, we recover the positive direction of the interpolation part of Seip’s Theorem. The
positive direction of the sampling theorem is also true; it was proved by Ortega Cerdá, Schuster
and the author [OSV-2006].
Remarkably, except in the 1-dimensional case, uniform flatness is not a necessary condition
for interpolation. A number of examples have been produced by Pingali and the author [PV-2016,
PV-2019].
126
There is very little done on interpolation theory in higher dimensions on manifolds other than
n
C . In dimension 1 there is a fair amount of work; we refer the reader to [V-2018, V-2016] for
references and further reading.
In the next paragraph we discuss some of the ideas behind the necessity of uniform separation.
This discussion is provided for the interested reader, and although very useful tools are presented,
most of the material is slightly less naturally aligned with the general topics covered in this lecture
series.
10.1.3 Weighted Bergman Inequalities

In preparation for a number of estimates we will √
need below, we now state and prove a very useful
lemma on the existence of bounded solutions of −1∂ ∂¯ when the forcing term is bounded. More
precisely, we have the following result, which was proved by Berndtsson and Ortega-Cerdà in the
1-dimensional case, and generalized by Lindholm to higher dimensions for the case where one has
only C 0 -estimates. However, a modification of the proof of Berndtsson and Ortega-Cerdà easily
gives the present form.
10.1.8 L EMMA . There exists a constant C > 0 with the following property. Let ω be a C 2 -smooth,
closed (1, 1)-form on a neighborhood of the closed unit ball B such that
√ √
−M −1∂ ∂|z|¯ 2 ≤ ω ≤ M −1∂ ∂|z| ¯ 2
for some positive constant M . Then there exist a function ψ ∈ C 2 (B) such that
√
¯ = ω and sup(|ψ| + |dψ|) ≤ CM.
−1∂ ∂ψ
B
Proof. We can assume that in fact ω has compact support in B(0, 2) by multiplication with an
appropriate cut-off function.
Let us first present the proof in the case n = 1. In this case, one simply takes
Z
ψ(z) := log |z − ζ|2 ω(ζ).
B(0,2)
√
Note that ω = h 2−1 dz∧dz̄ for some real-valued function h. A standard argument using integration-
by-parts shows that
1 ∂ 2ψ
= h.
π ∂z∂ z̄
The function ψ is clearly bounded by the constant
Z

M sup log |ζ − z|2 dA
z∈B(0,1) B(0,2)
while the derivative is controlled by

Z
dA(ζ)
M sup .
z∈B(0,1) B(0,2) |z − ζ|
127
Thus we have the stated result. √
In higher dimensions, write ω = ωij̄ 2−1 dz i ∧ dz̄ j . Then as in the 1-dimensional case, the
function Z
ψ(z) := ω11̄ (ζ, z 2 , ..., z n ) log |z 1 − ζ|2 dA(ζ)
B(0,2)
then satisfies
1 ∂ 2ψ
= ω11̄ .
π ∂z 1 ∂ z̄ 1
From the condition dω = 0, we see that, when i and j are both different from 1,
1 ∂ 2ψ ∂ 2 ω11̄
Z
= (ζ, z 2 , ..., z n ) log |z 1 − ζ|2 dA(ζ)
π ∂z i ∂ z̄ j B(0,2) ∂z i ∂ z̄ j
∂ 2 ωij̄
Z
= (ζ, z 2 , ..., z n ) log |z 1 − ζ|2 dA(ζ)
B(0,2) ∂ζ∂ ζ̄
= ωij̄ (z).
As before, ψ is bounded in C 1 -norm by CM . The proof is complete.

One important application of Lemma 10.1.8 is the following result.
√ √
10.1.9 P ROPOSITION . Let ω = −1∂ ∂|z| ¯ 2 and ϕ ∈ P SH(2B) such that −1∂ ∂ϕ ¯ ≤ M ω for
n
some positive constant M on 2B, where B is the unit ball in C . Then there is a constant C > 0
such that for all f ∈ O(2B) satisfying
Z
|f |2 e−ϕ ω n < +∞,
B
Z
2 −ϕ(0)
(10.2) |f (0)| e ≤C |f |2 e−ϕ ω n
B
and
Z r
2r −rϕ 2 −ϕ n
(10.3) |d(|f | e )|(0) ≤ C |f | e ω .
B
In particular, if f ∈ H (Cn , ϕ) then weighted point evaluation is bounded with norm independent
of the point. More generally, there exists C > 0 such that for all x ∈ Cn ,
Z 1/2
−ϕ(x)
+ d(|f |2 e−ϕ )(x) ≤ CM 2 −ϕ
n
(10.4) (|f (x)|e |f | e ω < +∞.
Cn
and thus Z 1/2

2 −ϕ n
||f ||C 1 (Cn ) := sup(|f | + |df |) ≤ CM |f | e ω < +∞.
Cn Cn
128
Proof. The estimate (10.4) follows from the estimate (10.2) and (10.3) because the Euclidean form
ω is translation-invariant.
√ √ To prove (10.2), let ψ be a bounded plurisubharmonic function on B such
¯ ¯
that −1∂ ∂ψ = −1∂ ∂ϕ. Such a function exists by Lemma 10.1.8. Let h = ψ − ϕ. Then h is
pluriharmonic, and therefore twice the real part of a holomorphic function H. Let F = f eH−H(0) .
We have
Z Z Z
2 −ϕ(0) 2 −ϕ(0) 2 −ϕ(0) n 2 −ϕ ψ−ψ(0) n
|f (0)| e = |F (0)| e ≤ − |F | e ω = − |f | e e ω ≤C |f |2 e−ϕ ω n .
B B
Next, observe that
d(|f |2r e−rϕ ) = r |f |2(r−1) (∂f f¯ + f ∂¯f¯)e−rϕ − |f |2r e−rϕ dϕ

= r(|f |2 e−ϕ )r−1 (f¯(∂f − f ∂ϕ) + f (∂¯f¯ − f¯∂ϕ))e

¯ −ϕ
.
We therefore have
2
|d(|f |2r e−rϕ )(0)| . ||f ||2r−1 ∂f (0)e−ϕ(0) − f (0)e−ϕ(0) ∂ϕ(0) .

Now, ϕ + h = ψ is bounded in C 1 -norm, and thus we have
∂f (0)e−ϕ(0) − f (0)e−ϕ(0) ∂ϕ(0)2 . ∂f (0)e−ϕ(0) + f (0)e−ϕ(0) ∂h(0)2 + |f (0)|2 e−ϕ(0) |∂ψ(0)|

= |d(f eH )(0)|2 e−ϕ(0) + C||f ||2 .
By the Cauchy estimates and Lemma 10.1.8, we have

Z Z
2 −ϕ(0) H 2 −ϕ(0) n
H
|d(f e )(0)| e . |f e | e ω . |f |2 e−ϕ ω n .
B B
Therefore (10.3) holds. The proof is complete.

As a consequence of Proposition 10.1.9, we now prove that a locally finite subset Γ ⊂ C is
(i) a finite union of uniformly separated locally finite sets if and only if RΓ : F(C) → fΓ is
bounded, and
(ii) uniformly separated if RΓ : F(C) → fΓ is surjective.
Proof of (i). Let F ∈ F(C). If Γ is a finite union of uniformly separated locally finite sets then for
any r > 0 there is an integer N such that no more than N of the disks {Dr (γ) ; γ ∈ Γ} intersect.
By Proposition 10.1.9
Z
2 −|γ|2 2 −|z|2 2
X X
|F (γ)| e ≤C |F (z)| e dA(z) ≤ CN |F (z)|2 e−|z| dA(z),
γ∈Γ γ∈Γ C
so RΓ is bounded.
129
Conversely, suppose RΓ is bounded. Let z ∈ C. Choose a function F ∈ F(C) such that
2
|F (z)|2 e−|z| = 1. Then there exists a constant µ, independent of z, such that
Z
2
|F (ζ)|2 e−|ζ| dA(ζ) ≤ µ.
C
We shall prove this fact below (Lemma 10.1.10). Let r ∈ (0, 1); shortly we will further restrict r.
For every γ ∈ Γ ∩ Dr (z) one has

2 −|γ|2 2 −|γ|2 2 −|z|2
|F (γ)| e − 1 = |F (γ)| e − |F (z)| e

Z 1
d 2
= |F (tz + (1 − t)γ)|2 e−|tz+(1−t)γ| dt
0 dt
2
≤ r sup d(|F |2 e−|·| )
Dr (z)
2
≤ r sup d(|F |2 e−|·| ) ≤ rCµ,
D1 (z)
where C is independent of r, γ or z. The last inequality follows from (10.3) of Proposition 10.1.9.
1
Now choose r < 2Cµ . Then
2 1
|F (γ)|2 e−|γ| ≥
2
for all γ ∈ Dr (z), and so
2 2
X X
#(Γ ∩ Dr (z)) ≤ 2 |F (γ)|2 e−|γ| ≤ 2 |F (γ)|2 e−|γ| ≤ 2µ||RΓ ||2
γ∈Γ∩Dr (z) γ∈Γ
Thus the number of points of Γ ∩ Dr (z) is finite, and bounded above by a number that does not
depend on z. Thus (i) is proved.
Proof of (ii). Suppose RΓ is surjective. Let EΓ : f(Γ) → F(C) be the operator that assigns to
g ∈ f(Γ) the extension whose norm is minimal among all R−1 Γ {g}. We claim that EΓ is bounded.
By the Closed Graph Theorem it suffices to show that the graph of EΓ is closed. To that end,
let gj → g in f(Γ) and let F = lim EΓ (gj ) in F(C). The sub-mean value property shows that
convergence in L2 implies locally uniform convergence, and consequently F is an extension of g.
We wish to show that F is the extension of g having minimal norm. The latter condition holds if
and only if F is orthogonal to all functions in F(C) that vanish along Γ. Let G be such a function.
Then Z Z
−|·|2 2
F Ḡe dA = lim EΓ (gj )Ḡe−|·| dA = lim 0 = 0,
C C
and thus F = EΓ (g). Therefore EΓ is bounded.

Fix a point γo ∈ Γ and consider the function f : Γ → C defined by
2 /2
f (γo ) = e|γo | and f (µ) = 0 for all µ ∈ Γ − {γo }.
130
Then
2
X
|f (γ)|2 e−|γ| = 1.
γ∈Γ
Let F := EΓ (f ). Then F |Γ = f and ||F || ≤ ||EΓ || is independent of γo .

Now let γ ∈ Γ − {γo }. Then

1 |F (γ)|2 e−|γ|2 − |F (γ )|2 e−|γo |2
o
=

|γ − γo | |γ − γo |

Z 1
1 d 2 −|tγ +(1−t)γ| 2
= |F (tγo + (1 − t)γ)| e o
dt
|γ − γo | 0 dt
2
≤ sup d(|F |2 e−|·| ) ≤ C||F ||2 ≤ C||EΓ ||2 .
C

¯ ≥ 4c∂ ∂|
10.1.10 L EMMA . Let ψ ∈ C 2 (C) be a smooth weight function such that ∂ ∂ψ ¯ · |2 for some
c > 0. Then there exists a constant µ with the following property: for every z ∈ C there exists
F ∈ O(C) such that
Z
2 −ψ(z)
|F (z)| e = 1 and |F |2 e−ψ dA ≤ µ.
C
√
Proof. Let T (ζ) := c(ζ − z) and let λ(ζ) = log(1 + c|ζ − z|2 ). Then
¯ c ¯ 2.
{T = 0} = {z}, |T |2 e−λ ≤ 1 and ∂ ∂λ(ζ) = ∂ ∂|ζ|
(1 + c|ζ − z|2 )2
Then
¯
∂ ∂ϕ(ζ) ¯
− (1 + 21 )∂ ∂λ(ζ) ¯ 2,
≥ c∂ ∂|ζ|
so by Theorem 9.2.2 there exists F ∈ O(C) such that
Z
1
F (z) = e 2 ψ(z)
and |F |2 e−ψ dA ≤ 48π,
C
and the proof is complete.

10.1.11 R EMARK . Theorem 9.2.2 was used in the proof of Lemma 10.1.10 primarily for the sake
of convenience. One can also use Hörmander’s Theorem, though doing so requires some ingenuity.
But the point is that the twisted methods need not be used here.
10.2 Deformation Invariance of Plurigenera

Plurigenera, the invariance problem and Siu’s extension theorem
10.2.1 D EFINITION . Let Y be a complex manifold. For each m ∈ N≥1 the number
Pm (Y ) := dimC ΓO (Y, KY⊗m )
is called the mth plurigenus of Y .
131
For a holomorphic family π : X → D with fibers Xt := π −1 (t), we define the function
µm : D 3 t 7→ Pm (Xt ) ∈ N.
One says that the m-genera are invariant for this family if the function µm is constant.
Our objective in the present section is to prove that, for a projective family, all the m-genera
are invariant, i.e., the function µm is constant, or equivalently, continuous. We shall prove the
continuity of µm by showing that (i) µm is upper semi-continuous and (ii) µm is lower semi-
continuous. These properties are local, so it’s enough to show them at 0 ∈ D.
The upper semi-continuity holds more generally for any holomorphic family. Let us, then, fix a
holomorphic family π : X → D, and some Hermitian metric ω on X . To prove that µm is upper
semi-continuous at 0, which means that
lim sup µm (t) ≤ µm (0).

t→0
We begin with the following lemma.
10.2.2 L EMMA . Let π : X → D be a holomorphic family and let L → X be a holomorphic line

bundle with smooth Hermitian metric e−ϕ . Let tj ∈ D(0, ε) − {0} be a sequence of complex num-
bers converging to 0 in the unit disk. Let sj ∈ ΓO (Xtj , KXtj ⊗ L) be a sequence of pluricanonical
sections satisfying Z
|stj |2 e−ϕ = 1.
Xtj
Then there is a section so ∈ ΓO (Xo , KXo ⊗ L) such that

Z
|so |2 e−ϕ = 1,
Xo
and a subsequence {tjk }, such that for any coordinate chart U ⊂ X on which ξ is a frame for
L|U , satisfying
∼
=O
π(U ) = D(0, ε) and (z, π) : U −→B × D(0, ε)
for some coordinate unit ball B ⊂ Xo and holomorphic map z : U → B, if we write
stj (x) = fj (z)dz 1 ∧ ... ∧ dz n ⊗ ξ and so (x) = fo (z)dz 1 ∧ ... ∧ dz n ⊗ ξ
then
lim fjk (z) = fo (z)
k→∞
uniformly on the set |z| ≤ 1/2.
10.2.3 R EMARK . Note that by the implicit function theorem for holomorphic submersions, each
point of Xo is contained inside the sort of product coordinate chart used in the statement of the
lemma.
132
Proof of Lemma 10.2.2. In the coordinate chart, let us write |ξ|2 e−ϕ = e−ϕ(z,t) . Then
|stj |2 e−ϕ = |fj (z)|2 e−ϕ(z,tj ) dV (z).
Note that since the metric e−ϕ is globally defined and smooth, the function e−ϕ(z,t) is uniformly
bounded above and below by positive constants on |z| ≤ 43 , and also uniformly in |t| ≤ ε. There-
fore we have
Z Z
2
|fj (z)| ≤ Cδ 2 ϕ(z,tj )
|fj | e dV (z) ≤ Cδ |stj |2 e−ϕ = Cδ .
Bδ (z) Xtj
for some delta sufficently small, and all |z| ≤ 32 . By Montel’s Theorem, the {fj } contain a subse-
quence that converges uniformly on |z| ≤ 1/2. By a diagonal argument we obtain a subsequence
{sjk } such that the restriction to each of these charts charts converges locally uniformly. The re-
sulting sections automatically glue together to produce a section of KXo ⊗ L, and the proof is
finished.
10.2.4 R EMARK . The above proof also works when the fibers of X → π are non-compact, i.e.,
when the map π is a submersion that might not be proper. (In the absence of properness, we don’t
know that the fibers are diffeomorphic.)
10.2.5 L EMMA . Let tj → 0 be a sequence of complex numbers in the unit disk, and suppose
dimC (ΓO (Xtj , KXtj ⊗ L) ≥ N for some N ∈ N. Then dimC (ΓO (Xo , KXo ⊗ L) ≥ N .
Proof. The hypothesis assumes that for each tj there is a subspace of ΓO (Xtj , KXtj ⊗L) of dimen-
sion at least N . For each j, we fix such a subspace VjN ⊂ ΓO (Xtj , KXtj ⊗ L), and an orthonormal
(j) (j)
basis{s1 , ..., sN } ⊂ VjN . After passing to a subsequence of the tj if needed, we find by Lemma
10.2.2 sections {s1 , ..., sN } ⊂ ΓO (Xo , KXtj ⊗ L), each of unit norm, such that
Z Z
(j) 2 −ϕ
lim |si | e = lim |si |2 e−ϕ .
Xtj Xo
From the identity

√ √ √
ha, bi = ||a + b||2 − ||a − b||2 + −1(||a − −1b||2 − ||a + −1b||2 ),
which holds in any Hilbert space, we see (from the proof, rather than the statement of Lemma
10.2.2) that the limit sections {s1 , ..., sN } are also orthonormal. Thus there are at least N indepen-
dent sections in ΓO (Xo , KXo ⊗ L). This completes the proof.
As an immediate corollary, we obtain the following proposition.
10.2.6 P ROPOSITION . Let π : X → D be a holomorphic family of n-dimensional complex mani-
folds. Then for any m ∈ N the function µm is upper semi-continuous.
Thus to prove the constancy of µm , it suffices to prove that every pluricanonical section on the
central fiber Xo of a holomorphic family extends to a pluricanonical section on the family X.
133
10.2.7 T HEOREM . Let π : X → D be a projective family of fiber dimension n, let L → X be a
⊗m
holomorphic with singular Hermitian metric e−κ , and let s ∈ ΓO (Xo , KXo
⊗ L) satisfy
Z
|s|2 ω −n(m−1) e−κ < +∞.
Xo
⊗m
Then there is a section S ∈ ΓO (X, KX ⊗ L) such that
Z
⊗m
S|Xo = s ∧ (dπ) and |S|2 ω −(n+1)(m−1) e−κ < +∞.
X
10.2.8 R EMARK . The constancy of µm follows by taking L = 0. One can also define twisted
plurigenera, which evidently are also invariant in families.
10.2.9 R EMARK . It is known that, for general holomorphic families of compact complex mani-
folds, plurigenera are not invariant in families. At the time of writing of these notes, the case of
the deformation invariance of plurigenera in Kähler families was still open.
We shall first prove the result when π : X → D is uniformly projective. At the end of the
proof we will see that the extension has uniform estimates, and we will be able to establish the
non-uniform case.
10.2.1 L2 extension of twisted canonical forms

Theorem 9.2.2 is the key tool in the proof of Theorem 10.2.7. We state a version of Theorem 9.2.2
for line bundle-valued holomorphic forms of top degree, or, twisted canonical forms.
10.2.10 T HEOREM . Let X be a Stein manifold and let Z ⊂ X be a smooth complex hypersurface.
Assume there exists a section T ∈ ΓO (X, LZ ) and a singular Hermitian metric e−λ for LZ → X
such that
(10.5) Z = {T = 0} and sup |T |2 e−λ ≤ 1.

X
Let L → X be a holomorphic line bundle with singular Hermitian metric e−ϕ , and assume there
exists a constant δ ∈ (0, 1] such that
(10.6) ¯ ≥ tδ∂ ∂λ
∂ ∂ϕ ¯ for all t ∈ [0, 1].
Then for each f ∈ ΓO (Z, KZ ⊗ L) such that

Z
|f |2 e−ϕ < +∞
Z
there exists F ∈ ΓO (X, KX ⊗ LZ ⊗ L) such that

Z Z
2 −ϕ−λ 24π
F |Z = f ∧ dT and |F | e ≤ |f |2 e−ϕ .
X δ Z
134
In particular, we shall make use of the following special case, in which Z is cut out by a
bounded holomorphic function.
10.2.11 T HEOREM . Let X be a Stein manifold and T : X → D a holomorphic function such that
dT (x) 6= 0 for all x ∈ Z := {T = 0}.
¯ ≥ 0.
Let L → X be a holomorphic line bundle with singular Hermitian metric e−ϕ such that ∂ ∂ϕ
Then for each f ∈ ΓO (Z, KZ ⊗ L) such that
Z
|f |2 e−ϕ < +∞
Z
there exists F ∈ ΓO (X, KX ⊗ LZ ⊗ L) such that

Z Z
2 −ϕ−λ
F |Z = f ∧ dT and |F | e ≤ 24π |f |2 e−ϕ .
X Z
10.2.2 Positively twisted canonical sections

We fix a smooth metric e−γ for L and a holomorphic line bundle A → X (which we may assume
admits a smooth Hermitian metric with positive curvature) with the following property:
(p) ⊗p
(GG) For each 0 ≤ p ≤ m − 1 there are sections {σ̃j 1 ≤ j ≤ Np } ⊂ ΓO (X , KX ⊗ A) that
⊗p
generate the sheaf of germs of holomorphic sections of KX ⊗ A → X .
√X → D is uniformly projective. Let us also fix a smooth

Property (GG) may be assumed because
metric e for A → X such that ω := −1∂ ∂ϕ
−ϕ ¯ is a Kähler metric. Again by uniform projectivity,
we may assume that e extends to a neighborhood of X in some larger projective family, and
−ϕ
thus Z
ω n+1 < +∞.
X
Since κ − γ is locally the sum of a smooth function and a plurisubharmonic function, the submean
value property tells us that eκ−γ is locally bounded above. Thus by uniform projectivity we may
assume that
sup eκ−γ < +∞.
X
(p) ⊗p
We let σj ∈ ΓO (Xo , KXo
⊗ A|Xo ) be defined by
(p) (p)
σ̃j |Xo = σj ∧ (dπ)⊗p .
Then we have the following key proposition.
10.2.12 P ROPOSITION . There exist a constant C > 0 and sections
(km+p) ⊗km+p
{σ̃j ∈ ΓO (X , KX ⊗ L⊗k ⊗ A) ; 1 ≤ j ≤ Np }0≤p≤m−1,k=0,1,2,...
with the following properties.
135
(mk+p) (p)
(a) σ̃j |Xo = s⊗k ⊗ σj ∧ (dπ)⊗(km+p)
(b) If k ≥ 1,
PNo (mk) 2 −γ
j=1 |σ̃j |e
Z
PNm−1 (mk−1) 2 ≤ C.
X j=1 |σ̃j |
(c) For 1 ≤ p ≤ m − 1,
PNp (mk+p) 2
|σ̃j |
Z
j=1
PNp−1 (mk+p−1) 2
≤ C.
X j=1 |σ̃j |
Proof. To simplify the notation a little, let us denote by σ̃ (mk+p) the Np -tuples
(mk+p) (mk+p) (mk+p) (mk+p)
σ̃ (mk+p) = (σ̃1 , ..., σ̃Np ) and σ (mk+p) = (σ1 , ..., σNp ),
and write
Np Np
X X
(mk+p) 2 (mk+p) 2 (mk+p) 2
||σ̃ || := |σ̃ | and ||σ || := |σ (mk+p) |2 .
j=1 j=1
Let
||σ̃ (0) ||2 ω (n+1)(m−1) ||σ̃ (p+1) ||2

C
b := sup , ; 0≤p≤m−2 .
X ||σ̃ (m−1) ||2 ||σ̃ (p) ||2 ω n+1
We proceed using double induction on k and p.
(p)
(k = 0) As far as extension there is nothing to prove; the sections σj have extensions by assump-
tion. Note that
||σ̃ (p+1) ||2
Z Z
(p) 2
≤C
b ω n+1 .
X ||σ̃ || X
(k ≥ 1) Assume the result has been proved for k − 1.

(0) ⊗mk
((p = 0)): Consider the sections s⊗k ⊗ σj of KX o
⊗ (L⊗k ⊗ A)|Xo , and define the metric
ψk,0 := log ||σ̃ (km−1) ||2

√
⊗mk−1
for KX ⊗ L⊗k ⊗ A. Observe that −1∂ ∂ψ ¯ k,0 ≥ 0 and
(0)
|σj |2
Z Z
(0)
k
|s ⊗ σj |2 e−(ψk,0 +κ) = |s|2 e−κ < +∞.
Xo Xo ||σ (m−1) ||2
By the L2 Extension Theorem 10.2.11 there exist sections

(km) ⊗mk
σ̃j ∈ ΓO (X , KX ⊗ L⊗k ⊗ A), 1 ≤ j ≤ No
such that
(km) (0)
σ̃j |Xo = s⊗k ⊗ σj ∧ (dπ)⊗km , 1 ≤ j ≤ No ,
136
and
(0)
|σj |2 −κ
Z Z
(km)
|σ̃j |2 e−(ψk,0 +κ) ≤ 48π |s| 2
e .
X Xo ||σ (m−1) ||2
Summing, we obtain
||σ̃ (km) ||2 e−γ ||σ̃ (km) ||2 e−κ

Z Z
(km−1) 2
≤ sup eκ−γ (km−1) ||2
X ||σ̃ || X X ||σ̃
||σ (0) ||2
Z
≤ 48π sup eκ−γ |s|2 (m−1) 2 e−κ
X Xo ||σ ||
Z
b sup eκ−γ
≤ 48π C |s|2 ω −n(m−1) e−κ .
X Xo
(km+p−1)
((1 ≤ p ≤ m − 1)): Assume that we have obtained the sections σ̃j , 1 ≤ j ≤ Np−1 . Con-
sider the non-negatively curved singular metric
ψk,p−1 := log ||σ̃ (mk+p−1) ||2

⊗km+p−1
for KX ⊗ L⊗k ⊗ A. We have
(p)
|σj |2
Z Z
(p)
k
|s ⊗ σj |2 e−ψk,p−1 ≤ PNp−1 (p−1) 2
< +∞.
Xo Xo j=1 |σj |
By the L2 Extension Theorem 10.2.11 there exist sections

(km+p) ⊗mk+p
σ̃j ∈ ΓO (X , KX ⊗ L⊗k ⊗ A), 1 ≤ j ≤ Np
such that
(km+p) (p)
σ̃j |Xo = s⊗k ⊗ σj ∧ (dπ)⊗km+p , 1 ≤ j ≤ Np ,
and
(p)
|σj |2
Z Z
(km+p) 2 −ψk,p−1
|σ̃j |e ≤ 48π .
X Xo ||σ (p−1) ||2
Summing, we obtain
||σ̃ (km+p) ||2
Z Z
≤ 48π C
b ωn.
X ||σ̃ (km+p−1) ||2 Xo
Let Z Z Z
b × max n+1 κ−γ 2 −n(m−1) −κ n
C=C ω , 48π sup e |s| ω e , 48π ω .
X X Xo Xo
The proof is finished.
137
10.2.3 Construction of the metric
Fix a smooth metric e−ψ for A → X . Consider the functions
λN := log ||σ̃ (km+p) ||2 ω −(n+1)(mk+p) e−(kγ+ψ) ,

where N − p = km.
Observe that by Proposition 10.2.12 and the concavity of the logarithm, we have the bound
Z
1
R
n+1
(λN − λN −1 )ω n+1 ≤ log C.
X
ω X
(This inequality if the reverse of Jensen’s inequality; it can be obtained by considering the convex
function − log.) It follows that the function
1
Λk = λmk
k
satisfies the integral bound Z
Λk ω n+1 ≤ mC 0
X
0
for some uniform constant C .
Now, locally, the functions λkm+p are a sum of a smooth function and a subharmonic function.
By applying the sub-mean value property for plurisubharmonic functions and making use of the
uniform projectivity of the family, we find that
Z
Λk (x) ≤ C Λk ≤ mCC 0 , x ∈ X .
X
It follows that the function

Λ(x) := lim sup lim sup Λk (y)
y→x k→∞
exists, and is locally the sum of a plurisubharmonic function and a continuous function on X .
Consider the singular Hermitian metric e−µ for KX ⊗m
⊗ L → X defined by
e−µ = e−Λ ω −(n+1)m e−γ .
This singular metric is given by the formula
e−µ(x) = lim inf lim inf e−µk ,
y→x k→∞
where
e−µk = e−Λk ω −(n+1)m e−γ .
The curvature of e−µk is thus
No
√ √ X (mk) 2 1√
¯ k =
−1∂ ∂µ −1∂ ∂¯ log |σ̃j | − ¯
−1∂ ∂ψ
j=1
k
1√ ¯
≥− −1∂ ∂ψ
k
138
10.2.13 P ROPOSITION . The curvature of e−µ is non-negative in the sense of currents.
Proof. It suffices to work locally. Then we have that the function

1
µk + ψ
k
is plurisubharmonic. But
1
lim sup lim sup µk + ψ = lim sup lim sup µk = µ.
y→x k→∞ k y→x k→∞
Thus µ is plurisubharmonic, as desired.
10.2.4 Examination of e−µ on the central fiber: conclusion of the proof

Since
No √ !
1 X (0) −1dπ ∧ dπ̄
Λk |Xo = log(|s|2 ω −(nm) e−γ ) + log |σj |2 e−ψ + log ,
k j=1
ω
we obtain the inequality

1
e−µ |Xo ≤ .
|s|2
Therefore
Z (m−1)µ+κ
Z Z m−1 Z 1
κ m m
2 − 2/m − m n 2 −n(m−1) −κ
|s| e m ≤ |s| e ≤ ω |s| ω e < +∞,
Xo Xo Xo Xo
where the second inequality is a consequence of Hölder’s Inequality.

An application of the L2 extension theorem concludes the proof of Theorem 10.2.7 in the
uniform case.
To pass to the general case, note that the L2 Extension Theorem 10.2.11 yields a holomorphic
⊗m
section S ∈ ΓO (X , KX ⊗ L) satisfying S|Xo = s ∧ dπ ⊗m and
Z (m−1)µ+κ
Z m−1 Z 1
m m
(10.7) |S|2 e− m ≤ 48π ωn |s|2 ω −n(m−1) e−κ .
X Xo Xo
We can therefore use Montel’s Theorem to pass from the uniformly projective case to the general
case of projective families. This completes the proof of Theorem 10.2.7.
139
10.2.5 An L2/m -estimate for m-canonical sections
Let us call a holomorphic family π : X → D uniformly projective if there exists a projective family
π̃ : X̃ → D1+r (0) such that X ⊂⊂ X̃ and π̃|X = π. Then in fact we have actually proved the
following theorem.
10.2.14 T HEOREM . Let π : X → D be a uniformly projective family, and let L → X be a
holomorphic line bundle with singular Hermitian metric e−κ whose curvature is a positive (1, 1)-
⊗m
current. Then for any holomorphic section s ∈ ΓO (Xo , KX o
⊗ L) satisfying
Z
1
|s|2/m e− m κ < +∞
Xo
⊗m
there exists a holomorphic section S ∈ ΓO (X , KX ⊗ L) such that
Z Z
1 1
2/m − m κ
S|Xo = s ⊗ dπ ⊗m
and |S| e ≤ 48π |s|2/m e− m κ .
X Xo
Proof. First suppose that Ω is a pseudoconvex domain in a Stein manifold X̃, that L → X is a
holomorphic line bundle with smooth Hermitian metric e−κ , and that π : X̃ → D is a holomorphic
submersion such that π restricts to the closure of Ω as a proper map. Assume that we have a
⊗m
section s ∈ ΓO (X̃o , KX̃ ⊗ L) that we wish to extend to Ω with L2/m -estimates, and which we
o
can normalize to satisfy Z
|s|2/m e−κ/m = 1.
Ωo
−1
(Of course, Ωo := π (0) ∩ Ω.)
A look at the proof of Theorem 10.2.7 shows that if there is a metric e−µ for KΩ⊗m ⊗ L → Ω
satisfying √
e−µ |Ωo ≤ |s|−2 and ¯ ≥0
−1∂ ∂µ
⊗m
there there is a holomorphic section S1 ∈ ΓO (Ω, KX̃ ⊗ L) satisfying
Z (m−1)µ+κ
S1 |Ωo = s ∧ dπ ⊗m
and C1 := |S1 |2 e− m < +∞.
Ω
(Conversely, if such a section S1 exists, one can take e−µ = (|S|2 e−κ )−1 , so extension is equivalent
to the existence of such a metric.) Moreover, in the proof of Theorem 10.2.7 we constructed one
such metric e−µ , which in the present setting can be assumed to satisfy
Z
e(µ−κ)/m < +∞.
Ω
Now, by Hölder’s Inequality,

Z Z (m−1)µ+κ 1/m (m−1)(µ−κ)
Z m−1
κ 1/m m
2/m − m 2 − (µ−κ)/m
|S1 | e = |S1 | e m e m2 ≤ C1 e =: A1 .
Ω Ω Ω
140
Without loss of generality we may assume A1 > 48π. Then we have shown that given any metric
e−µ as above, there exists a section S1 such that
Z
κ
⊗m
S1 |Ωo = s ∧ dπ and |S1 |2/m e− m ≤ A1
Ω
for some constant A1 > 48π.

Let us now define the metric
(m − 1) log |S1 |2 + κ
µ1 := .
m
Then Z Z
2 −µ1
|s| e = |s|2/m e−κ/m = 1,
Ωo Ωo
and thus by the L2 Extension Theorem 10.2.11 there exists a section S2 ∈ ΓO (Ω, KΩ⊗m ⊗ L) such
that Z
⊗m
S2 |Ωo = s ⊗ dπ and |S2 |2 e−µ1 ≤ 48π.
Ω
Thus by Hölder’s Inequality
Z Z 1/m Z (m−1)/m)
κ µ1
2/m − m 2 −µ1 (m−1)/m
|S2 | e ≤ |S2 | e e m−1 ≤ (48π)1/m A1 =: A2 .
Ω Ω Ω
Continuing by induction, we obtain sections Sj ∈ ΓO (Ω, KΩ⊗m ⊗ L) such that

Z
κ
⊗m
Sj |Ωo = s ⊗ dπ and |Sj |2/m e− m ≤ Aj ,
Ω
(m−1)/m
where Aj = (48π)1/m Aj−1 < Aj−1 . Since Aj is decreasing and larger than 48π, Aj converges
to some A, and taking limits of the inductive relation yields
A = (48π)1/m A(m−1)/m .
Thus A = 48π. We have thus constructed, for every ε > 0, a sequence of sections Sj , j ≥ jo such
that Z
|Sj |2/m e−κ/m ≤ A + ε.
Ω
Using arguments that are by now familiar, we can apply Montel’s Theorem to let ε → 0, pass to a
singular metric e−κ , and let Ω → X̃. Next, we can take X̃ to be the Stein manifold obtained from
a uniform projective family X → D by removing a hyperplane section, and use the L2 estimates
to extend S across the hyperplane section. Finally, we can again use Montel to pass to the case of
any projective family. The proof is therefore complete.
E XERCISES
141
Lecture 11
Berndtsson’s Theorem on
Plurisubharmonic Variation
11.1 Berndtsson’s Theorem

11.1.1 Variations of Hilbert spaces
Let (X, ω) be a Kähler manifold and let L → X be a holomorphic line bundle. We equip L with a
smooth Hermitian metric e−ϕ , and thus we can define the Hilbert space
Z
H (ϕ) := f ∈ ΓO (X, L) ; 2 −ϕ
|f | e dVω < +∞ ,
X
where dVω is the volume form on X induced by the metric ω. Our goal is to study what happens
to the Hilbert space H (ϕ) as we vary the metric e−ϕ in the sense of the next definition.
11.1.1 D EFINITION . Let Ω ⊂ Cn be an open connected set. A family of metrics for L → X
parameterized by Ω is a metric e−ϕ for the line bundle p∗ L → X × Ω, where p : X × Ω → X is
the projection to the first factor.
For each t ∈ Ω there is a natural isomorphism of line bundles
ιt : L → p∗ L|X×{t} ,
and we write
e−ϕt := ι∗t e−ϕ
for the metric induced on L by this isomorphism. We can then define the Hilbert spaces
Z
H (ϕt ) = f ∈ ΓO (X, L) ; ||f ||t :=
2 2 −ϕt
|f | e dVω < +∞ .
X
We therefore obtain a fibration H (ϕ) → Ω whose fiber H (ϕ)t over t is the Hilbert space H (ϕt ).
We want to know when this fibration is (locally) trivial.
142
√
11.1.1. Let X = Cn with its Euclidean metric −1∂ ∂|z| ¯ 2 and let ϕ(z, t) = |t|2 |z|2 . Show that for
t 6= 0 the Hilbert spaces H (ϕ)t are all equal, as subsets of O(Cn ). What is H (ϕ)o ?
To avoid the situation that occurs in Exercise 11.1.1, namely that the Hilbert spaces change
as vector spaces when we change the base parameter t, we assume from now on that X is a
bounded pseudoconvex domain in some larger Stein manifold X̃, that L is the restriction to X of
a holomorphic line bundle on X̃, and that the family of smooth metrics e−ϕ also extends to X̃.
Under these assumptions, the vector spaces H (ϕ)t , seen as subsets of ΓO (X, L), are indepen-
dent of t. Of course, their norms vary. In fact, for each t, s ∈ Ω the function e−ϕs /e−ϕt is bounded
above by a positive constant Cs,t (and therefore below by 1/Ct,s ), and thus
Z Z Z
1 2 −ϕt 2 −ϕs
|f | e dVω ≤ |f | e dVω ≤ Cs,t |f |2 e−ϕt dVω ,
Ct,s X X X
so H (ϕ)t and H (ϕ)s are quasi-isometric as Hilbert spaces (i.e. they have equivalent norms).
It follows that their dual vector spaces are also quasi-isometric. (Note, also, that under these
hypotheses these Hilbert spaces are always infinite-dimensional.) But more important for us is the
fact that in this case, the subspace
H (ϕ)t ⊂ ΓO (X, L)
is independent of t.
We can now define the trivial bundle
H (ϕ) := H (ϕ)o × Ω → Ω
and define a Hilbert metric on H (ϕ) by endowing the fiber H (ϕ)o × {t} with the norm || · ||t ;
thus the fiber is H (ϕ)t as a Hilbert space.
11.1.2 D EFINITION . Let L → X be a holomorphic line bundle and let e−ϕ be a family of Hermitian
metrics for L.
(S) A section of H (ϕ) → Ω is a section f ∈ Γ(X × Ω, p∗ L) such that ι∗t f ∈ H (ϕ)t for each
t ∈ Ω. The section f is said to be holomorphic if f ∈ ΓO (X × Ω, p∗ L). In this case, we write
f ∈ ΓO (Ω, H (ϕ)).
(Thus all sections are holomorphic on the fibers, and a holomorphic section means it is
holomorphic in the base variable as well.)
(B∗ ) Let H (ϕ)∗ → Ω denote the dual bundle, i.e., the trivial bundle H (ϕ)∗o × Ω → Ω with the
Hilbert norms
|hξ, f i|
||ξ||t∗ := sup
f ∈H (ϕ)t −{0} ||f ||t
on the fibers H (ϕ)∗t .
143
(S∗ ) A section of H (ϕ)∗ → Ω is a map ξ : H (ϕ) → C such that
ξt := ξ|H (ϕ)t ∈ H (ϕ)∗t .
The section ξ of H (ϕ)∗ → Ω is said to be holomorphic if for each f ∈ H (ϕ)o the function
Ω 3 t 7→ hξt , f i ∈ C
is holomorphic. The set of holomorphic sections is denoted ΓO (Ω, H (ϕ)).
11.1.3 R EMARK . Note that since H (ϕ)t = H (ϕ)o as subspaces of ΓO (X, L), every f ∈ H (ϕ)o
induces a constant (and thus, holomorphic) section ff of H (ϕ) → Ω defined by
ι∗t ff = f, t ∈ Ω.
We shall abusively denote this section by f , rather than ff .
11.1.2. Show that ξ ∈ ΓO (Ω, H (ϕ)∗ ) if and only if the function
Ω 3 t 7→ hξt , ι∗t fi ∈ C
is holomorphic for each holomorphic section f ∈ ΓO (Ω, H (ϕ)).
11.1.4 R EMARK . Fix a Hilbert basis {f1 , f2 , ...} for H (ϕ)o . Then for any section g of H (ϕ) → Ω
one has the Fourier series X
ι∗t g = ci (t)fi
i
Note that
∂ ∗ X ∂ci (t)
∂ ¯
fi = ι∗t

ιt g = ∂ t̄
y∂g .
∂ t̄ i
∂ t̄
The last equality holds because f is holomorphic on the fibers of p. Thus
!
X ∂ci (t)
¯ =
ι∗t ∂g fi ⊗ dt̄,
i
∂ t̄
where the ∂¯ operator on the left is the one on p∗ L → X × Ω. This fact parses well with the
¯
definition of the ∂-operator for a holomorphic vector bundle.
11.1.5 E XAMPLE . Let g ∈ ΓO (Ω, H (ϕ)). Then, with
ξtg (f) := (ι∗t f, ι∗t g)t ,
ξ g is a section of H (ϕ)∗ → Ω, but this section is almost never holomorphic.
144
11.1.6 E XAMPLE . Let x ∈ X and let ξ˜x be defined by
ξ˜t (f) := ι∗t f(x).
Then ξ˜x : H (ϕ)t → Lx is a linear map, and it is bounded provided we equip the complex line Lx
with the metric e−ϕt (x) . Indeed, the boundedness is the inequality
|f (x)|2 e−ϕt (x) ≤ Cx ||f ||2t ,
which is often called Bergman’s Inequality, and is proved as follows: first, when X is a domain in
Cn and the weight ϕ is 0, then the inequality is standard (and could be proved, for instance, from
Cauchy’s Theorem). Then the inequality holds locally for any smooth metric e−ϕ by the obvious
estimate C −1 dVo ≤ e−ϕ dVω ≤ CdVo combined with the unweighted Bergman inequality, where
dVo is the Euclidean volume form in the local coordinate. Finally, the L2 norm on a small ball is
obviously controlled by the L2 norm over all of X.
11.1.3. Fill in the details of the proof of Bergman’s Inequality.
To obtain a linear functional from ξ˜x we need a non-zero vector for L∗x , and we choose a vector
e whose norm with respect to the dual metric eϕt is 1: |e|2 eϕt (x) = 1. (This vector is of course
unique up to a unimodular factor.) We then set
ξ x := e ⊗ ξ˜x .
The section ξ x is now a bounded linear functional on each fiber H (ϕ)t , and the holomorphic
dependence on t is clear because for each f ∈ H (ϕ)o ξ x f = f (x) is independent of t.
By the Riesz Representation Theorem, for each x ∈ X there is a section Ktx ∈ H (ϕ)t such
that
(f, Ktx )t = f (x), f ∈ H (ϕ)t .
Writing
Kt (x, z) := Ktx (z),
we see that Kt (x, ·) ∈ Γ X † , L† for each x ∈ X. (Here, for a complex manifold Y the notation

Y † means the manifold with the complex conjugate structure.) Moreover, since
Z
f, Kt (x, ·) = f (z)Kt (x, z)e−ϕt (z) dVω (z) = f (x),
t X
we see that the section x 7→ Kt (x, z) of L → X is holomorphic for each z ∈ X. It follows that
Kt ∈ ΓO X × X † , L L† ,

where for (holomorphic) line bundles π1 : L1 → X1 and π2 : L2 → X2 ,
L1 L2 := π1∗ L1 ⊗ π2∗ L2 .
The section Kt is called the Bergman kernel of H (ϕ)t , and we shall meet it again soon.
145
11.1.4. Show that if {gj } is any orthonormal basis for H (ϕ)t then
X
Kt (x, y) = gj (x) ⊗ gj (y).
j
Show also that
Kt (x, x)e−ϕt (x) = sup |f (x)|2 e−ϕt (x) ; f ∈ H (ϕ)t and ||f ||t = 1 .

(Hint: Given x ∈ X, if Sx := {f ∈ H (ϕ)t ; f (x) = 0}, show that Sx⊥ has dimension 0 or 1.)
11.1.2 Statement of Berndtsson’s Theorem

We are now ready to state Berndtsson’s Theorem.
11.1.7 T HEOREM (Berndtsson’s Theorem on Plurisubharmonic Variation). Let X be a relatively
compact, pseudoconvex domain in a Stein Kähler manifold(X̃, ω) and let L → X̃ be a holomor-
phic line bundle. Let Ω ⊂⊂ Cn be an open connected set. Let e−ϕ be a family of Hermitian metrics
for L → X̃ (in the sense of Definition 11.1.1). If the Hermitian (1, 1)-form
¯ + p∗ ω
∂ ∂ϕ
is non-negative (resp. positive) then the curvature of Chern connection for the holomorphic vector
bundle H (ϕ) → Ω, with its L2 metric || · ||2t , is non-negative (resp. positive) in the sense of
Nakano.
Note that the result is local with respect to the base Ω, so that from now on we can (and will)
assume that Ω is the unit ball.
We shall not prove Theorem 11.1.7 in its full generality, but rather assume that the base domain
Ω is 1-dimensional, i.e., the unit disk. In this situation Griffiths and Nakano positivity are the same.
In view of Proposition 3.2.7, the case of Theorem 11.1.7 in which Ω ⊂ C is equivalent to the
following theorem.
11.1.8 T HEOREM . Assume that for the metric e−ϕ for p∗ L → X × D we have
¯ + p∗ Ricci(ω) ≥ 0 (resp. > 0),
∂ ∂ϕ
where D denotes the unit disk in C. Let ξ be a holomorphic section of H (ϕ) → D that is not
identically zero. Then the function
D 3 t 7→ log ||ξt ||2t∗
is subharmonic (resp. strictly subharmonic).

Though we will not explain precisely why, the proof of Theorem 11.1.7 is not more complicated
than that of Theorem 11.1.8; it just requires a little more notation and discussion. We have chosen
to skip the more general proof because we will only look at applications of Theorem 11.1.8.
146
11.1.3 The Bergman projection
As we will see very soon, the proof of Berndtsson’s Theorem requires a more systematic under-
¯
standing of solutions of the ∂-equation with minimal norm.
Fixing our X as in the previous paragraph, consider the larger Hilbert space L2 (e−ϕt ) defined
to be the Hilbert space closure of the vector space of all the smooth sections of L → X with
compact support, the closure being taken with respect to the norm
Z 1/2
2 −ϕt
||f ||t := |f | e dVω .
X
Using the sub-mean value property and a Montel type argument, one shows that H (ϕ)t is a closed
subspace of L2 (e−ϕt ). Consequently there is an orthogonal projection
Pt : L2 (e−ϕt ) → H (ϕ)t ,
called the Bergman projection.
There is an important link between the Bergman projection and the minimal solution of the
¯ ¯ = α. Then u1 − u2 is holo-
∂-equation. Indeed, suppose u1 , u2 are solutions of the equation ∂u
morphic. It follows that the solution of minimal norm is orthogonal to the holomorphic subspace.
Consequently, given solution u of ∂u¯ = α, the solution
umin := u − Pt (u) ∈ L2 (e−ϕt )

is the solution of minimal norm.
In particular, the norm of this solution is no larger than the norm of the solution provided
by Skoda’s version of Hörmander’s Theorem when the latter applies. Consequently we have the
following result.
11.1.9 P ROPOSITION . Fix t ∈ D. If the curvature of the metric e−ϕt for L → X satisfies
¯ t + Ricci(ω) ≥ Θ
∂ ∂ϕ
for some non-negative (1, 1)-form Θ on X then
Z Z
2 −ϕt
|u − Pt (u)| e dVω ≤ ¯ 2 e−ϕt dVω
|∂u|Θ
X X
for all u such that the right hand side is finite.

11.1.10 R EMARK . Note that the orthogonal projection operator Pt : L2 (e−ϕt ) → H (ϕ)t is an
integral operator. Indeed, if f ∈ H (ϕ)t then Pt f = f . Thus for each x ∈ X and f ∈ H (ϕ)t ,
Pt f (x) = hξ x , f i, where ξ x was defined in Example 11.1.6. It follows that
Z
Pt f (x) = f (z)Kt (x, z)e−ϕt (z) dVω (z), f ∈ L2 (e−ϕt ), x ∈ X,
X
where Kt ∈ ΓO X × X † , L L† is the Bergman kernel. Indeed, if f = Pt f + ft⊥ is the orthog-

onal decomposition of f in H (ϕ)t ⊕ H (ϕ)⊥

t then the holomorphicity of the Bergman kernel Kt
⊥ x
implies that (ft , Kt )t = 0.
147
11.1.4 Proof of Theorem 11.1.8
An estimate for the Laplacian of −log ||f||2t
For simplicity, let us use the notation
ft := ι∗t f
for a section f of H (ϕ) → D.
11.1.11 L EMMA . If f ∈ ΓO (D, H (ϕ)) is holomorphic up to the boundary then

∂ 2 ∂ft ∂ϕt
||f||t = − Pt ft , ft .
∂t ∂t ∂t t

Z Z Z
∂ 2 −ϕt ∂ft ∂ϕt −ϕt ∂ft ∂ϕt
|ft | e dVω = − ft , ft e dVω = − Pt ft , ft e−ϕt dVω ,
∂t X X ∂t ∂t X ∂t ∂t
where the last equality holds because for any smooth section u of L → X the section u − Pt u is
orthogonal to H (ϕ)t .
∂ on sections of H (ϕ) → D by
11.1.12 D EFINITION . We define the operator ∇1,0
∂t

∗ 1,0 1,0 ∂ft ∂ϕt
ιt (∇ ∂ f) = (∇ ∂ f)t := − Pt ft .
∂t ∂t ∂t ∂t
11.1.13 L EMMA . Let f ∈ ΓO (D, H (ϕ)) be holomorphic up to the boundary. Then
∂2
||ft ||2t
∂t∂ t̄
2
∂ 2 ϕt 2 −ϕt
Z
1,0 2 ∂ft ∂ϕt ∂ft ∂ϕt

=− |ft | e dVω + (∇ ∂ f)t + − ft − Pt − ft
X ∂t∂ t̄ ∂t t ∂t ∂t ∂t ∂t t
Z 2
2
∂ ϕt 2 −ϕt 2 ∂ϕt ∂ϕt
|ft | e dVω + (∇1,0

=− ∂ f)t + Pt ft − ft .

X ∂t∂ t̄ ∂t t ∂t ∂t t
Proof. In the proof of Lemma 11.1.11 we established the equality
Z Z
∂ 2 −ϕt ∂ft ∂ϕt
|ft | e dVω = − ft , ft e−ϕt dVω .
∂t X X ∂t ∂t
Starting from here, we compute that
∂2 ∂ft ∂ϕt 2 −ϕ
Z 2 Z
2 ∂ ϕt −ϕt
||ft ||t = − ft , ft e dVω + − f e t dVω ,
∂t∂ t̄ X ∂t∂ t̄ X ∂t
∂t
Now, by Pythagoras’ Theorem, for any smooth section u one has
Z Z Z
2 −ϕt 2
|u| e dVω = |Pt u| dVω + |u − Pt u|2 dVω ,
X X X
∂ft ∂ϕt
so with u = ∂t
− ∂t
f the result follows.
148
11.1.14 L EMMA . Let f ∈ H (ϕ)o and assume, moreover, that
√
¯ t + Ricci(ω) ≥ Θ
−1∂ ∂ϕ
for some positive Hermitian form ω on X. Then

2 Z 2
∂ϕt ∂ϕ t ∂ϕ t
∂¯X |f |2 e−ϕt dVω .

∂t f − P t f ≤
∂t
t X
∂t Θ
Proof. Since ∂¯X ∂ϕ f = − ∂¯X ∂ϕ

t
t

∂t ∂t
f , the result follows immediately from Proposition 11.1.9.
Next we calculate that

∂2 ∂ 2 ||ft ||2t ∂||ft ||2t 2

1
−||ft ||2t log ||ft ||2t = − + ∂t .

∂t∂ t̄ ∂t∂ t̄ ||ft ||2t
By Lemma 11.1.11
2 |((∇1,0 f) , f ) |2
1 ∂||ft ||2t

∂ t t t
∂t
=
||ft ||2t ∂t ||ft ||2t
√
¯ t ≥ ω, then by Lemmas 11.1.13 and 11.1.14
and if we also assume that −1∂ ∂ϕ
2
∂ 2 ||ft ||2t ∂ 2 ϕt 2 −ϕt
Z
1,0 2 ∂ϕt ∂ϕt

− = |ft | e dVω − (∇ ∂ f)t − ft − P t ft
∂t∂ t̄ X ∂t∂ t̄ ∂t t ∂t ∂t t
!
2
∂ 2 ϕt ¯
Z
∂ϕt 2 −ϕt 1,0 2

≥ − ∂X |ft | e dV ω − ∇ ∂ ft .
X ∂t∂ t̄ ∂t Θ ∂t t
We have therefore proved the following theorem.

11.1.15 T HEOREM
√ . Let f ∈ ΓO (D, H (ϕ)) be holomorphic up to the boundary and assume, more-
¯ t + Ricci(ω) ≥ Θ for some positive Hermitian form ω on X. Then
over, that −1∂ ∂ϕ
2 !
∂2 ∂ 2 ϕt ¯
Z
1 1 ∂ϕt
log ≥ − ∂X |ft |2 e−ϕt dVω
∂t∂ t̄ ||ft ||2t ||ft ||2t X ∂t∂ t̄ ∂t Θ
 1,0

2
1  |((∇ ∂ f) t , f t )t | 2
(11.1) + ∂t
− (∇1,0 ∂ f)t

2 2
||ft ||t ||ft ||t ∂t t
End of the proof of Theorem 11.1.8

√
Let us assume first that Θ := −1(∂ ∂ϕ ¯ t + Ricci(ω)) is a Kähler form for each t ∈ D. We claim
¯ + p Ricci(ω) is non-negative X × D then
that if ∂ ∂ϕ ∗
2
∂ 2 ϕt ¯

∂ϕt
(11.2) − ∂X ≥ 0.
∂t∂ t̄ ∂t Θ
149
Indeed, with respect to the product structure of X × D we compute that
¯ = ∂ 2 ϕt ∂ϕt ∂ϕt
∂ ∂ϕ dt ∧ dt̄ + ∂X ∧ dt − ∂¯X ¯ t,
∧ dt + ∂ ∂ϕ
∂t∂ t̄ ∂ t̄ ∂t
and therefore
√
( −1(∂ ∂ϕ ¯ + p∗ Ricci(ω)))n+1
(n + 1)!
2 n
Θn−1

∂ ϕt Θ ∂ϕt ∂ϕ t
= ∧ dt ∧ dt̄ − ∂X ∧ ∂¯X ∧
∂t∂ t̄ n! ∂ t̄ ∂t (n − 1)!
!
2
∂ 2 ϕt ¯ Θn

∂ϕt
= − ∂X ∧ dt ∧ dt̄,
∂t∂ t̄ ∂t n!Θ
which establishes (11.2) in this case. In the general case, since X is a bounded pseudoconvex
domain in a Stein manifold, X has a negative strictly plurisubharmonic function u ∈ C ∞ (X).
Replacing e−ϕt by e−(ϕt +εu) (which does not change the underlying vector space of the Hilbert
space), we have
2
∂ 2 ϕt ¯

∂ϕt
− ∂X ≥ 0,
∂t∂ t̄ ∂t Θ+ε√−1∂ ∂u¯
and letting ε → 0 yields the desired positivity.

Let us fix a holomorphic section ξ of H (ϕ)∗ → D and assume that
√
¯ t + Ricci(ω)) =: Θ ≥ 0.
−1(∂ ∂ϕ
First, we note that by Theorem 11.1.15, if f ∈ ΓO (D, H (ϕ)) satisfies
(∇1,0
∂ f)t = 0
∂t
for some t ∈ D, then for that same t we have

! 2 !
∂2 |hξτ , fτ i|2 ∂ 2 ϕt ¯
Z
1 ∂ϕt
log ≥ − ∂X |ft |2 e−ϕt dVω .
∂τ ∂ τ̄ ||fτ ||2 ||ft ||2t X ∂t∂ t̄ ∂t Θ
τ =t
Next, observe that by the Riesz representation Theorem, for each fixed t ∈ D there exists a section
f ∈ H (ϕ)t such that
||f ||t = ||ξt ||t∗ and hξt , gi = (f, g)t for all g ∈ H (ϕ)t .
In particular,
|hξt , f i|2
||ξt ||2t∗ = ||f ||2t = hξt , f i = |hξt , f i| = ,
||f ||2t
and moreover
|hξt , gi|2
||ξt ||2t∗ ≥ for all g ∈ H (ϕ)t .
||g||2t
150
We emphasize that t is fixed, and we assume from here on that for this particular t, the linear
functional ξt is not zero.
Now define the section f ∈ ΓO (D, H (ϕ)) by
fτ := f − (τ − t)(∇1,0
∂ f )t .
∂τ
Then ft = f and

1,0 1,0 1,0 1,0 1,0
(∇ ∂ f)t = lim (∇ ∂ f )τ − (∇ ∂ f )t − (τ − t)(∇ ∂ (∇ ∂ f )t )τ = 0.
∂t τ →t ∂t ∂t ∂t ∂t
Therefore  
1  |((∇1,0
∂ f)t , ft )t |
2 2
∂t
− (∇1,0
∂ f)t = 0,

||ft ||2t ||ft ||2t ∂t t
and from Theorem 11.1.15 we obtain

2 !
∂2 2
Z
1 1 ∂ ϕt ¯ ∂ϕt
|ft |2 e−ϕt .

log ≥ − ∂X
∂τ ∂ τ̄ ||fτ ||2t τ =t ||ft ||2t X ∂t∂ t̄ ∂t ω
Since ξ is a holomorphic section,

|hξτ , fτ i|2
τ 7→ log
||fτ ||2
has non-negative τ -Laplacian at τ = t.
Now, from the definition of the dual norm, the function
|hξτ , fτ i|2
Φ : τ 7→ log ||ξτ ||2τ ∗ − log
||fτ ||2
achieves a local minimum of 0 at τ = t, and therefore
2
∂ Φ
≥ 0.
∂τ ∂ τ̄ τ =t
But this means
!
2
2 2
∂ ∂ |hξτ , fτ i|
log ||ξτ ||2τ ∗ ≥ log ≥ 0.
∂τ ∂ τ̄ τ =t ∂τ ∂ τ̄ ||fτ ||2
τ =t
Thus we have shown that t 7→ log ||ξt ||2t∗ is subharmonic away from the set {t ∈ D ; ξt = 0}. Since
ξ is holomorphic, this set is a closed discrete subset, and thus log ||ξτ ||2τ ∗ continues across the zero
set of ξ to a subharmonic function on D. The proof of Theorem 11.1.8 is therefore complete.
11.2 Application: The Suita Conjecture

Suita’s Conjecture was proved by Błocki [B-2013] when X is a domain in C, and in general by
Guan and Zhou [GZ-2015]. We shall give a different proof here, that uses Theorem 11.1.8. Our
proof below is similar to the proof of Berndtsson and Lempert [BL-2016], but it is slightly different.
151
11.2.1 Formulation of the Suita Conjecture
Let X be a Riemann surface and assume X admits a non-constant bounded subharmonic function.
(Such Riemann surfaces are called hyperbolic, or sometimes potential theoretically-hyperbolic.) It
is well known that such a Riemann surface admits a Green’s function G : X × X → [−∞, 0), i.e.,
a function uniquely characterized by the following properties:
(i) If we write Gx (y) = G(y, x), then for each x ∈ X,
√
−1 ¯
∂ ∂Gx = δx ,
π
and
(ii) if H : X × X → [−∞, 0) is another function with property (i), then G ≥ H.

Using the Green’s Function, one can construct a conformal metric for X as follows:
√
−1 ¯ Gx )(y).
ωF (x) := lim ∂(eGx )(y) ∧ ∂(e
y→x 2
The metric ωF is called the fundamental metric.
The metric ωF can be computed from the Green’s Function at a point x as follows. Choose any
holomorphic function f ∈ O(X) that vanishes to order 1 at x and is non-zero everywhere else.
(Since X is an open Riemann surface, and thus Stein, such a function f exists.) From the definition
of Green’s function, the function
hx := Gx − log |f |
is harmonic. Then √
−1
ωF (x) = e 2hx (x)
df (x) ∧ df¯(x).
2
Now, the curvature of ωF is
¯ x (x).
ΩF := −2∂ ∂h
Locally hx (y) = G(x, y) − log |x − y| up to a harmonic function, so by the harmonicity of the
Green’s function we find that
∂ 2G ∂ 2G

¯ ∂G ∂G 1
∂ ∂hx (y) = ∂ dx̄ + + dȳ = dx ∧ dȳ + dy ∧ dx̄.
∂ x̄ ∂ ȳ x̄ − ȳ ∂y∂ x̄ ∂y∂ x̄
Therefore
∂ 2G

ΩF (x) = 4 lim dx ∧ dx̄.
y→x ∂x∂ ȳ

ζ−z
11.2.1 E XAMPLE . Let X be the unit disk. Then G(z, ζ) = log 1− ζ̄z
, and we have
√
−1dz ∧ dz̄
ωF (z) = .
2(1 − |z|2 )2
152
Thus in the unit disk the fundamental metric agrees with the Poincaré metric (normalized so that
its Gaussian curvature is identically −4).
Note that hz (ζ) = − 12 log |1 − ζ z̄|2 and
∂ 2 hz (ζ) 1
= .
∂ζ∂ z̄ 2(1 − ζ z̄)2
Consequently ΩF (z) = −4ωF (z).
Suita’s conjecture can be stated as follows:
11.2.2 C ONJECTURE . [Su-1971] Let X be a hyperbolic Riemann surface. Then the Gaussian
curvature of the fundamental metric of X is at most −4.
11.2.2 Alternate formulation of the Suita Conjecture

Before turning to the proof of the Suita Conjecture, we shall reformulate it in a manner that is more
amenable to our methods. To this end, we want to consider Example 11.1.6 in a special setting.
On our Riemann surface X, consider the canonical line bundle KX → X, which in this case is just
the cotangent bundle. We can define the Hilbert space
Z √
−1
HX := α ∈ ΓO (X, KX ) ;
2
α ∧ ᾱ <, +∞ .
X 2
This Hilbert space is a closed subspace of the larger Hilbert space L2X of measurable (1, 0)-forms
θ such that Z √
−1
θ ∧ θ̄ < +∞,
X 2
and we have the Bergman projection
PX : L2X → HX2 .
11.2.1. Prove that the Bergman projection PX is bounded.
The integral kernel BX of PX , defined by the relation
Z
PX α(z) = α(w) ∧ BX (z, w),
w∈X
is called the Bergman kernel of X. As in Example 11.1.6,

†
BX ∈ ΓO (X × X † , KX KX ).
If the Riemann surface X is a (hyperbolic) plane domain, Schiffer [Sc-1946] observed that the
Bergman kernel BX (z, ζ) of X is given by the formula
√
−1 ∂ 2 G(z, ζ)
BX (z, ζ) = dz ⊗ dζ̄.
π ∂z∂ ζ̄
Schiffer’s result extends to other hyperbolic Riemann surfaces.
153
11.2.3 T HEOREM (Schiffer[Sc-1946]). Let X be a hyperbolic Riemann surface and let G be the
Green’s function of X. Then one has the formula
1 ¯
(11.3) BX (z, ζ) =∂z ∂ζ G(z, ζ).
π
Proof. Let o ∈ X. Choose a local coordinate ζ near o and ε > 0 so small that the disk D2ε :=
{|ζ| < 2ε} is a coordinate chart. Fix Θ ∈ HX2 and write Θ = f (ζ)dζ in D2ε . Then
√
Z Z
1 ¯ 1
−1Θ(ζ) ∧ ∂z ∂ζ G(o, ζ) = √ dζ (Θ(ζ) ⊗ ∂z G(o, ζ))
X−Dε π −1π X−Dε

−1
Z
∂G(o, ζ)
= √ f (ζ) dζ dz,
−1π ∂Dε ∂z
where we have used the fact that G|∂X = 0. (Here ∂X denotes the part of the boundary that has
non-zero capacity.) But one has
∂G(o, ζ) 1 ∂h(o, ζ)
= + ,
∂z −2ζ ∂z
and we have
√
Z
1
−1Θ(ζ) ∧ ∂z ∂¯ζ G(o, ζ)
X−Dε π
Z
1 f (ζ) ∂h(z, ζ)
= √ −2 dζ dz
2 −1π |ζ|=ε ζ ∂z
= f (0)dz + O(ε) = Θ(o) + O(ε).
Letting ε → 0, we see that
√
Z
1 ¯
Θ(o) = −1Θ(ζ) ∧ ∂z ∂ζ G(o, ζ).
X π
Since the Bergman kernel is characterized by its holomorphicity and its reproducing property, the
proof is complete.
Schiffer’s Theorem immediately gives the following corollary.
11.2.4 C OROLLARY. If X is a hyperbolic Riemann surface then the curvature of the fundamental
metric is −4πKX (z, z).
11.2.2. Show that the Bergman kernel for the Bergman projection
PD : L2 (D) → L2 (D) ∩ O(D)
for the classical Bergman space is
√
−1dz ⊗ dw̄
K(z, w) = .
2π(1 − z w̄)2
(Hint: Use Exercise 11.1.4.)
154
We therefore see that Suita’s Conjecture is equivalent to the estimate
(11.4) πBX (x, x) ≥ ωF (x).
Note that, in view of Example 11.2.1 and Exercise 11.2.2, equality holds in (11.4) for X = D.
We shall prove the Suita Conjecture by establishing (11.4).
11.2.3 Estimating the Bergman kernel

For the rest of this paragraph we fix the point x at which we will estimate BX (x, x).
In Theorem 11.1.8 we take our Stein manifold to be the hyperbolic Riemann surface X, and
we fix any smooth metric ω on X. Note that ω is a Kähler form, and also an area form, so that
dVω = ω. Next we let L = KX , and we shall need a family of metrics e−ϕt .
The family of metrics is constructed as follows. We equip L with the dual metric e−ϕo := ω −1 .
Then
¯ o = −Ricci(ω).
∂ ∂ϕ
Thus if we let ψ : X × L → (−∞, 0) be a subharmonic function, where L := {τ ∈ C ; Re τ < 0}
is the left half plane, then the family of metrics
e−Φ := e−(ϕ(·)+ψ(·,τ ))
satisfies
¯ + p∗ Ricci(ω) = ∂ ∂ψ
∂ ∂Φ ¯ ≥ 0,
where p : X × L → X is projection to the first factor. Thus for any choice of such ω and ψ we
have spaces H (Φ)t and the vector bundle H (Φ) → D for which Theorem 11.1.8 holds. We fix
ω = ωF and ψ(y, τ ) := Re τ + p · max(2Gx (y) − Re τ, 0)
where p > 0. At the end of the proof, we will let p → ∞.

Next fix a tangent vector e ∈ TX,x = L∗x such that |e|2ωF = 1, and consider the holomorphic
section ξ x of H (Φ)∗ → D defined in Example 11.1.6:
ΓO (X, KX ) ⊃ H (Φ)t 3 f 7→ ξ x f := he, f (x)i .
Observe that
Kt (x, x)
||ξtx ||2t∗ = sup |e ⊗ f (x)|2 = .
f ∈H (Φ)t , ||f ||t =1 ωF (x, x)
In particular, when t = 0, and thus ψ = 0, we see that Ko = BX , and thus we wish to prove that
1
||ξox ||2o∗ ≥ .
π
Since ψ depends only on Re τ , Berndtsson’s Theorem 11.1.8 yields the following theorem.
155
11.2.5 T HEOREM . Let ξ x be the section of H (Φ)∗ → L defined in Example 11.1.6. Then the
function
L 3 τ 7→ log ||ξτx ||2τ ∗
is convex.
Let us set
Xt := {y ∈ X ; 2Gx (y) < t}.
Notice that
(a) if t << 0 then Xt is simply connected. In the limit, Xt converges to a point, so once we
get past the critical points of Gx , we have a simply connected Riemann surface, which is
biholomorphic to the disk by the Uniformization Theorem.
(b) For τ ∈ L the function ψ(·, τ ) vanishes on XRe τ , and equals Re τ + p · (2Gx (y) − Re τ ) on
X − XRe τ . In particular, for τ = o we see that Ko , the Bergman kernel for H (Φ)o , is just
BX .
11.2.6 P ROPOSITION . The function
(11.5) (−∞, 0) 3 t 7→ log ||ξtx ||2t∗
is bounded, and therefore it is increasing.

Proof. Observe that for f ∈ H (Φ)t satisfying ||f ||t = 1, Bergman’s inequality implies that
|f (x)|2
Z
x 2 −t
|hξt , f i| = ≤ Ce |f |2 ≤ C||f ||2t = C,
ωF (x) Xt
where C is independent of t and f . Thus the function (11.5) is bounded. Since the function (11.5)
is also convex and defined in (−∞, 0), it must be increasing.
We have therefore proved that
(11.6) ||ξox ||2o∗ ≥ lim ||ξtx ||2t∗ .

t→−∞
(Note that the limit on the right hand side exists.) To complete the proof of (11.4), we need to show
that the limit on the right hand side is π1 .
To motivate why the latter might be true, observe that since Gx (y) = log |x − y| + hx (y) for
some harmonic a function hx , which in particular is smooth across x. If t << 0 then Xt is very
close to the disk of area et centered at x, and on X −Xt the weight e−ϕt is equal to e(p−1)t e−ϕo −2pGx ,
which should be very small when p >> 0 because 2Gx > t. We therefore expect Kt to be very
close to e−t BXt .
Let us fix f ∈ H (Φ)t ) such that
√
−1
||f ||t = 1 and f (x) ∧ f (x) = Kt (x, x).
2
156
(C.f. Exercise 11.1.4.) We compute that
Z Z Z
2 −Φt −t
1= |f | e ωF = e 2
|f | + e (p−1)t
|f |2 e−2Gx .
X Xt X−Xt
Now, if s is sufficiently close to 0 then

Z Z Z
2 −2pGx 2 −2pGx
e (p−1)t
|f | e =e (p−1)t
|f | e +e (p−1)t
|f |2 e−2pGx
X−Xt X−Xs Xs −Xt
2
|f |
Z
≤ ε + sup · e(p−1)t e2pGx ωF .
Xs ωF Xs −Xt
R
Letting V(t) := Xs −Xt
ωF ≤ Cet , we find that
Z Z s
−t
e(p−1)t
e2pGx
ωF = e e−p(u−t) dV(u).
Xs −Xt t
11.2.7 L EMMA . Let ν : (−∞, 0) → R+ be an increasing function such that ν(t) ≤ et for all
t < 0. Then for p > 1,
Z 0
−t 2
(11.7) lim inf e e−p(s−t) dν(s) ≤ .
t→−∞ t p−1
Proof. Let Z 0
−t
f (t) := e e−p(s−t) dν(s).
t
Observe that for R ≥ 1
1 0 1 0 0 (p−1)t −ps 1 0 s (p−1)t −ps

Z Z Z Z Z
f (t)dt = e e dν(s)dt = e e dtdν(s)
R −R R −R t R −R −R
Z 0 Z 0
1 −s 1 −R −s
≤ e dν(s) = ν(0) − e ν(R) + ν(s)e ds
R(p − 1) −R R(p − 1) −R
R+1 2
≤ ≤ .
R(p − 1) p−1
But if (11.7) fails then there exists Ro > 0 such that f (t) ≥ (2 + δ)/(p − 1) for all t < Ro , and
then we have
1 0 R − Ro 2 + δ
Z
f (t)dt ≥ ,
R −R R p−1
and we get a contradiction as soon as R is sufficiently large.
We have thus shown that for any ε > 0 there exists p sufficiently large and t << 0 to guarantee
that Z
−t
1−ε≤e |f |2 ≤ 1.
Xt
157
Consequently
e−t BXt ≤ (1 − ε)Kt .
But since Xt is smoothly close to the disk of radius et/2 centered at x, we see from Example 11.2.1
and Exercise 11.2.2 that for t << 0,
e−t BXt (x) 1−ε
≥ .
ωF (x) π
Therefore we have shown that for any ε > 0 there exists p >> 0 such that ,
(1 − ε)2
lim ||ξtx ||2t∗ ≥ .
t→−∞ π
Since the right hand side of (11.6) does not depend on p, we have proved that
πBX (x, x)
≥ 1,
ωF (x)
and thus the Suita conjecture is confirmed.
11.3 Application: Sharp Estimates for L2 extension

In this section we are going to give another proof of the L2 extension theorem. We will focus on
the special case of Theorem 9.2.2, in which the line bundle LZ associated to Z is trivial. In fact,
the method we present here, which is due to Berndtsson and Lempert [BL-2016], can be extended
to prove a more general version than this case, but so far it is not known how to obtain the full
statement of Theorem 9.2.2 from this approach.
11.3.1 Statement of the Sharp L2 Extension Theorem

11.3.1 T HEOREM . Let (X, ω) be a Stein manifold and let T : X → D be a holomorphic function
such that dT (x) 6= 0 for all x ∈ Z := T −1 (0). Let L → X be a holomorphic line bundle with
singular Hermitian metric e−ϕ such that
¯ + Ricci(ω) ≥ 0.
∂ ∂ϕ
Then for any f ∈ ΓO (Z, L) such that

Z
dAω
|f |2 e−ϕ < +∞
Z |dT |2ω
there exists F ∈ ΓO (X, L) such that

Z Z
2 −ϕ dAω
F |Z = f and |F | e dVω ≤ π |f |2 e−ϕ .
X Z |dT |2ω
158
Theorem 11.3.1 looks very much like Theorem 10.2.10 (the latter is the version for L-valued
holomorphic). The major difference is the constant π, which is better than the constant 24π that
we previously established. As we will show in the next paragraph, the constant π is sharp.
11.3.2 R EMARK . For many applications (for example, the invariance of plurigenera) this sharp
constant does not matter, but there are some interesting applications where the sharpness is impor-
tant. Interestingly, in the applications known to the author, the fact that the constant is sharp is less
important than the fact that it is the area of the unit disk.
11.3.2 Sharpness in Theorem 11.3.1

The constant π in Theorem 11.3.1 is sharp in the sense that there is no smaller constant for which
Theorem 11.3.1 holds in its full generality. There are, however, specific cases (and in fact, many
of them) in which the constant is far better than π.
11.3.3 E XAMPLE . Let X = D be the unit disk and let Z = {p} be a point in D. We take L to be
the trivial bundle with the trivial metric, so that
A (ϕ) = L2 (D) ∩ O(D) =: B
is the classical Bergman space. Letting
z−p
T := ∈ O(D),
|p| + 1
we see that
1
sup |T | = 1 and |dT (p)|2 = .
D (|p| + 1)2
Thus we want to know the best constant C such that for each a ∈ C there exists F ∈ O(D) such
that Z
F (p) = a and |F |2 ≤ C · |a|2 (|p| + 1)2
D
Clearly the problem of finding such a function is linear, so we may as well assume that a = 1, and
then it’s clear that the optimal constant is
Cmin (p) = min ||F ||2 ; F ∈ B and F (p) = 1 .

11.3.1. Show that if Fo ∈ B satisfies ||Fo ||2 = Cmin (p) then

Z
Fo GdA = 0
D
for every G ∈ B satisfying G(p) = 0. Use Exercise 11.1.4 to show that consequently
1
Cmin (p) =
K(p, p)
where K is the Bergman kernel of the Bergman projection P : L2 (D) → B.
159
Exercises 11.3.1 and 11.2.2 show that
Cmin (p) = π(1 − |p|)2 .
Thus we see, on the one hand, that the optimal constant in Theorem 11.3.1 cannot be less than π.
On the other hand, we also see that even though the optimal constant is π, there are certain spaces
X with hypersurfaces Z for which the norm of the operator of minimal extension is much smaller
than π, and might even go to 0.
11.3.3 Beginning of the proof of Theorem 11.3.1

The proof of Theorem 11.3.1 begins with a couple of reductions. First, we may assume that X
is actually a bounded pseudoconvex domain in some larger Stein manifold and that Z meets the
boundary of X transversely. When objects extend to the larger manifold, we shall avoid referring
to this manifold by saying that the objects extend up to the boundary of X. Next we can assume
that metric ω and the line bundle L → X extends holomorphically to this larger domain, and that
the metric e−ϕ for this extension of L is smooth. If the result is proved under these two assump-
tions, then it follows in general by weak-∗ compactness theorems and techniques of approximating
singular Hermitian metrics on Stein manifolds.
Let us fix the section f ∈ ΓO (Z, L) to be extended; we may assume, perhaps after shrinking
X, that f is holomorphic up to the boundary of Z. By Theorem 9.1.2 there exists a section F ∈
ΓO (X, L) such that Z
F |Z = f and |F |2 e−ϕ dVω < +∞.
X
Indeed, one extends f to F on a larger ambient manifold, and compactness implies that F |X is
square-integrable.
Since there exists some extension of f with finite L2 norm, there is an extension of minimal L2
norm; let us call the minimal extension Fo .
11.3.2. Show that the extension of minimal norm is unique.
Thus our goal is to prove that
Z Z
dAω
|Fo | e dVω ≤ π |f |2 e−ϕ
2 −ϕ
.
X Z |dT |2ω
Since the minimal extension is unique (Exercise 11.3.2), its norm can be thought of as a norm
of the section f to be extended. This observation suggests that there is a way to describe this norm
without reference to the minimal extension itself, and indeed this is the case.
11.3.4 Dual formulation of the norm of the minimal extension

We introduce the notation
Z
A (ϕ) := F ∈ ΓO (X, L) ; ||F || :=
2 2 −ϕ
|F | e dVω < +∞ .
X
160
(The Hilbert space A (ϕ) looks a lot like the spaces H (ϕ), but we use different notation to em-
phasize the fact that the metric e−ϕ , and hence the space A (ϕ), is not varying.)
11.3.4 P ROPOSITION . Let

Jϕ (Z) := {g ∈ A (ϕ) ; g|Z ≡ 0}
and let
Ann(Jϕ (Z)) := {ξ ∈ A (ϕ)∗ ; hξ, gi = 0 for all g ∈ Jϕ (Z)} .
Then for each f ∈ ΓO (Z, L) that is smooth up to the boundary (or more generally has some
extension in A (ϕ)) the minimal extension Fo of f satisfies
( )
|hξ, F i|2
Z
2 −ϕ
(11.8) |Fo | e dVω = sup ; ξ ∈ Ann(Jϕ (Z)) ,
X ||ξ||2∗
where F ∈ A (ϕ) is any extension of f .
Proof. First note that if ξ ∈ Ann(Jϕ (Z)) and F1 , F2 are extensions of f then F2 − F1 ∈ Jϕ (Z)
and thus
hξ, F2 i = hξ, F1 i + hξ, F2 − F1 i = hξ, F1 i .
Thus the right hand side of (11.8) is independent of the choice of extension. Next observe that
Fo ⊥ Jϕ (Z). Indeed, if G ∈ Jϕ (Z) then Fo + εG is an extension of f for every ε ∈ R, and we
have
||Fo + εG||2 = ||Fo ||2 + ε2Re (Fo , G) + O(|ε|2 ),
and since the right hand side achieves its minimum at ε = 0, we have

d
2Re (Fo , G) = ||Fo + εG||2 = 0.
dε ε=0
√
Similarly 0 = 2Re (Fo , − −1G) = 2Im (Fo , G) = 0, so (Fo , G) = 0 as claimed.
It follows that the linear functional ξo ∈ A (ϕ)∗ defined by
hξo , Gi := (G, Fo )
lies in Ann(Jϕ (Z)), and therefore its norm is

|hξ, Fo i|
sup ; ξ ∈ Ann(Jϕ (Z)) .
||ξ||∗

To estimate the right hand side of (11.8) we may obviously work with a dense subspace that is
well-suited for estimation.
161
11.3.5 L EMMA . The set of linear functionals in A (ϕ)∗ of the form
Z
dAω
ξg : A (ϕ) 3 F 7→ F ḡe−ϕ , g ∈ Γo (Z, L),
Z |dT |2ω
(where Γo means smooth sections with compact support) forms a dense subspace of Ann(Jϕ (Z)).
Proof. Clearly ξg ∈ Ann(Jϕ (Z)). Moreover, if hξg , F i = 0 for all g ∈ Γo (Z, L) then F |Z ≡ 0.
Thus the lemma follows from Exercise 11.3.3.
11.3.3. Let A be a subspace of a topological vector space H. Then A = H if and only if every
continuous linear functional on H that vanishes on A is zero.
Using (11.8), it is evident that to get an estimate for the extension of minimal norm, we have
to obtain estimates for
|hξg , Fo i|2
||ξg ||2∗
for all g ∈ Γo (Z, L). Let us begin with the numerator. We introduce the notation
LZ2 := L2 (e−ϕ /|dT |2ω ) and E (ϕ) := LZ2 ∩ ΓO (Z, L).
Letting
P : LZ2 → E (ϕ)
denote the Bergman projection, we have the following proposition.
11.3.6 P ROPOSITION . For each g ∈ Γo (Z, L)
Z Z
2 2 −ϕ dAω 2 −ϕ dAω
|hξg , Fo i| ≤ |f | e |P g| e .
Z |dT |2ω Z |dT |2ω
Z Z
−ϕ dAω dAω
hξg , f i = f ḡe = f P ge−ϕ ,
Z |dT |2ω Z |dT |2ω
and the result follows from the Cauchy-Schwarz Inequality.
Consequently, to prove Theorem 11.3.1, it suffices to show that
Z
(11.9) |P g|2 e−ϕ ≤ π||ξg ||2∗ for all g ∈ Γo (Z, L).
Z
To achieve (11.9), the strategy is to degenerate the domain X to Z through a good family of
domains Xt parameterized by negative real numbers t, i.e.,
Xo = X and lim Xt = Z.
t→−∞
We now define this good degeneration.
162
11.3.5 Degeneration to the infinitesimal neighborhood
Degeneration to the infinitesimal neighborhood: first version
Letting
L := {τ ∈ C ; Re τ < 0}
denote the left half plane, define
Xt := {x ∈ X ; log |T (x)|2 < t}, t ≤ 0.
Then Xo = X and, for t << 0, Xt is a neighborhood of Z. (Again, recall that we have assumed T
to be holomorphic up to the boundary of X.) One can define the Hilbert spaces
Z
−Re τ 2 −ϕ
H(ϕ)τ := F ∈ ΓO (XRe τ , L) ; e |F | e dVω < +∞ .
XRe τ
If we could apply Berndtsson’s Theorem to the bundle of Hilbert spaces H(ϕ) → L then we
could conclude that log ||ξg ||2τ ∗ is subharmonic in τ . As we shall see in a moment, this piece of
information is particularly useful in the proof of Theorem 11.3.1.
Unfortunately, it is not possible to apply Berndtsson’s Theorem here, because the Hilbert spaces
H(ϕ)τ might not have the same underlying vector space; in fact, each of them is defined over a
different complex manifold, so it is hard to decide if H(ϕ) → L is a vector bundle.
To get around this problem, we now modify the Hilbert spaces slightly.
Degeneration to the infinitesimal neighborhood: better version

Define the plurisubharmonic function
ψτ : X × L 3 (x, τ ) 7→ max(log |T (x)|2 − Re τ, 0).
Note that (i) ψτ depends only on Re τ , and (ii) the support of ψτ is X − XRe τ .
For a positive number p >> 0, consider the family of positively curved Hermitian metrics
e−ϕτ := e−(ϕ+pψτ ) , τ ∈ L.
We define the Hilbert spaces
H (ϕ)τ := ΓO (X, L) ∩ L2 (e−ϕτ −Re τ ).
Note that for each F ∈ H (ϕ)τ one has
Z Z
2 −ϕτ −Re τ
lim |F | e dVω = e |F |2 e−ϕ dVω ,
p→∞ X XRe τ
so that the Hilbert spaces H (ϕ)τ are in some sense approximations of H(ϕ)τ . However, since all
metrics extend up to the boundary, the Hilbert spaces H (ϕ)τ all have the same underlying vector
space (seen as a subset of ΓO (X, L)) and only their norms vary. Thus Berndtsson’s Theorem 11.1.8
applies to H (ϕ) → L.
163
11.3.6 The proof of Theorem 11.3.1
11.3.7 L EMMA . Let g ∈ Γo (Z, L). Then
sup ||ξg ||2τ ∗ < +∞.

τ ∈L
Proof. Let H ∈ H (ϕ)τ with ||H||2τ = 1. By the sub-mean value property and the smoothness of
all metrics up to the boundary, for each x ∈ Support(g) ⊂ Z we have
Z Z
2 −ϕ(x) −Re τ 2 −ϕ −Re τ
|H(x)| e ≤ Cg e |H| e ≤ Cg e |H|2 e−ϕτ = Cg .
XRe τ X
Thus Z 2
−ϕ dA ω
||ξg ||2τ ∗ = sup H ḡe 2
≤ Cg ,
||H||τ =1 Z |dT | ω
as desired.
11.3.8 L EMMA . Let g ∈ Γo (Z, L). Then the function
λg : (−∞, 0] 3 t 7→ log ||ξg ||2t∗
is non-decreasing. In particular,
||ξg ||2o∗ ≥ ||ξg ||2t∗
for all t < 0.
Proof. By Berndtsson’s Theorem, the function τ 7→ log ||ξg ||2τ ∗ is subharmonic on L. On the
other hand, ||ξg ||2τ ∗ depends only on Re τ , and thus λg is convex on (−∞, 0). If λg decreases
anywhere on (−∞, 0) then by convexity limt→−∞ λg = +∞. But by Lemma 11.3.7 λg is bounded
above.
11.3.9 T HEOREM . Let g ∈ Γo (Z, L). Then for each δ > 0 there exists p sufficiently large so that
Z
1
2
lim ||ξg ||t∗ ≥ |P g|2 e−ϕ − δ.
t→−∞ 2π Z
Theorem 11.3.9 will be proved using Lemma 11.2.7 and the following two lemmas.
11.3.10 L EMMA . Let F be a smooth function on X. Then

Z Z
−t dAω
lim sup e FdVω = π F .
t→−∞ Xt Z |dT |2ω
The proof is left to Exercise 11.3.4.
11.3.4. Prove Lemma 11.3.10.
164
11.3.11 L EMMA . Let g ∈ Γo (Z, L). Then for any ε > 0 there exists G ∈ ΓO (X, L) that is
holomorphic up to the boundary, such that
Z
dAω
|G − P g|2 e−ϕ 2
< ε2 .
Z |dT |ω
Proof. By L2 approximation theorems (see, e.g., [H-1990, Theorem 5.6.2]) we can find a section
that is holomorphic on a neighborhood of Z and approximate P g uniformly on any compact subset
of Z. If we take the compact subset to be sufficiently large then we also approximation in L2 .
Finally, since the ambient manifold is Stein, the approximation on a neighborhood of Z can be
extended to a neighborhood of X, and therefore the extension will have finite L2 norm on X.
Proof of Theorem 11.3.9. Fix ε > 0 and let G be as in Lemma 11.3.11. Then
Z 2 4
1 −ϕ dAω 2 ||P g||
2

||ξg ||t∗ ≥ GP ge ≥ (1 − ε) .
||G||t Z |dT |2ω ||G||2t
Now, Z Z
−t 2 −ϕ 2
||G||2t =e |G| e dVω + e (p−1)t
|G|2 e−(ϕ+p log |T | ) dVω .
Xt X−Xt
By Lemma 11.3.10, Z
−t
e |G|2 e−ϕ dVω ≤ (1 + ε)||P g||2
Xt
for t << 0. Turning to the second integral, we have

Z Z
2 −(ϕ+p log |T |2 ) 2 −ϕ 2
e(p−1)t
|G| e dVω ≤ sup |G| e e(p−1)t
e−p(log |T | ) dVω .
X−Xt X X−Xt
But Z Z 0
−p log |T |2 −t
e (p−1)t
e dVω ≤ Ce e−p(s−t) dν(s),
X−Xt t
dVω . By Exercise 11.3.4, Since ν(t) ≤ C 0 et . Lemma 11.2.7 therefore implies

R
where ν(t) = Xt
that Z
2 Co
e (p−1)t
|G|2 e−(ϕ+p log |T | ) ≤ ,
X−Xt p−1
for a sequence of ‘t’s tending to −∞. Thus for sufficiently large p and sufficiently negative t we
obtain
||G||2t ≤ (1 + ε)||P g||2 + ε.
Choosing ε > 0 small enough yields the desired estimate.
By combining Lemma 11.3.8 and Theorem 11.3.9 we obtain (11.9), and thus the proof of
Theorem 11.3.1 is complete.
165
11.4 Equivalence of Sharp Extension and Positive Variation
The L2 extension theorem 11.3.1 has a long history, into which we will not delve very deeply. In
brief, the first result was due to Ohsawa and Takegoshi [OT-1987]. Shortly after the appearance of
the original result of Ohsawa and Takegoshi, some quite spectacular applications were discovered
by Demailly, Diederich, Ohsawa, Siu and others. Manivel [Ma-1993] established a geometric
version [Ma-1993], and three new (and mostly rather similar) proofs of the extension theorem
appeared almost simultaneously by Berndtsson [B-1996], McNeal [Mc-1996] and Siu [S-1996].
But the true cementing of the L2 extension theorem as a result lying at the foundation of complex
analytic geometry finally occurred around 1998, when Siu established the deformation invariance
of plurigenera [S-1998]. (See also [S-2002].) Since that time, there have been and continue to be
numerous extensions, applications, and new proofs, of the L2 extension theorem.
The question of the sharp constant came into focus first in [O-2001], where Ohsawa pointed out
that the L2 extension theorem could be used to establish a conjecture of Suita, on the comparison
of the curvature of the hyperbolic metric and of the logarithmic capacity of a Riemann surface.
Ohsawa noted that the extension theorem gives a comparison of the two metrics (via the link
with the Bergman kernel that we explained in Chapter 11.2), and his proof made it clear that the
conjectured comparison holds if and only if the sharp constant is π.
The sharp extension theorem was established first by Błocki [B-2013] in a very particular
setting of hyperplanes in domains, though experts knew by that point that this setting already
contains all of the key difficulties of the general proof of L2 extension. Shortly afterwards, a very
general version of the sharp extension theorem was proved by Guan and Zhou [GZ-2015]. The
method of proof is not fundamentally different from that of Błocki, but the work is much more
general, and should still see applications beyond the sizable number of applications appearing in
the paper. The method of Błocki-Guan-Zhou can be used to establish the sharp constant
π(1 + δ)
δ
in Theorem 9.2.2; incidentally this constant works for all δ > 0, and does not require the assump-
tion δ ≤ 1.
11.4.1 Log plurisubharmonicity of the Bergman kernel

One particular application in [GZ-2015], which is at the heart of the present chapter, concerns a
new proof of the following theorem of Berndtsson [B-2006].
11.4.1 T HEOREM . [B-2006] Let D ⊂ Ck × Cn be a pseudoconvex domain and let φ ∈ PSH(D).
For t ∈ Ck let
t
Dt := {z ∈ Cn ; (t, z) ∈ D}, φt := φ|Dt , and At := L2 (Dt , e−φ ) ∩ O(Dt ).
Denote by Kt (z, w) the kernel of the Bergman projection Pt : L2 (Dt , e−φ ) → At , i.e,
t
Z
t
Pt f (z) = f (w)Kt (z, w)e−φ dV (z).
Dt
166
Then the function
(11.10) D 3 (t, z) 7→ log Kt (z, z)
is plurisubharmonic.
A special case of Theorem 11.4.1, in which the domain D is a product Ω × Z with Ω ⊂ Ck and
X ⊂ Cn , follows from Berndtsson’s Theorem 11.1.8, as we now show.
To prove the plurisubharmonicity of (11.10), it suffices (and is necessary) to show that (11.10)
is subharmonic in each variable ti separately, for any fixed x, because in view of Exercise 11.1.4
(11.10) is already plurisubharmonic in z. Thus we may assume the base Ω is 1-dimensional.
As we showed in Example 11.1.6, the linear functional ξtx of point evaluation at x multiplied
by some vector e ∈ L∗x , which is given by
Z
x −φt (y)
hξt , f i = e ⊗ f (y)Kt (x, y)e dV (y) ,
X
is a holomorphic section of H (φ) → Ω. On the other hand, by Exercise 11.1.4
||ξtx ||2t∗ = sup |hξtx , f i|2 = sup |f (x)|2 e−ϕt (x) = hKt (x, x), e ⊗ ei ,
||f ||=1 ||f ||=1
and therefore by Theorem 11.1.8 the function
t 7→ log ||ξtx ||2t∗ = log hKt (x, x), e ⊗ ei
is subharmonic. Thus Theorem 11.4.1 is proved in the special case of product domains.
Quite remarkably, Theorem 11.4.1 also follows from the L2 extension theorem with optimal
constant, as was shown by Guan and Zhou [GZ-2015]. In fact, there is a somewhat more general
result, with the same proof as that of Guan-Zhou.
Before we state the more general theorem, define the volume forms dνt on Xt by the relation
Z Z Z
ΦdVω := Φdνt dA(t).
X t∈D Xt
Since the definition is clearly local, we compute dνt in terms of ω, in local coordinates, and the
submersion T : X → D. The function T , being a submersion, is locally a coordinate function,
and X is locally a product X = Xo × D where D is a small disk in the complex plane. We use
coordinates x = (x1 , ..., xn ) on Xo and use the coordinate T on D. Then
√ √ √ √
−1 −1 −1 −1
ω = ωoō 2
dT ∧ dT̄ + ωoj̄ 2
dT ∧ dx̄j + ωiō 2
dxi ∧ dT̄ + ωij̄ 2
dxi ∧ dx̄j ,
where the indices i and j vary in {1, ..., n}. Since ω is a metric, the matrix (ωij̄ ) is an invertible
n × n Hermitian matrix, and we write its inverse as (ω j̄i ). From the formula

c v
= c − (A−1 v, v) det A,

det †
v A
167
√ (ξ)|2
we have dVω = det(ω ij̄ )dV (x) ωoō − ω j̄i ωoj̄ ωiō 2−1 dT ∧ dT̄ . Now, |dT |2ω := supξ6=0 |dT
ω(ξ,ξ)
. To
∂ i ∂
compute the supremum, we can use the vectors ξ = ∂T + η ∂xi without loss of generality, so that
|dT (ξ)|2 1
= .
ω(ξ, ξ) ωoō + ωij̄ η η̄ + 2Re (ωiō η i )
i j
The minimum of the denominator, as a function of η, is ωoō − ω j̄i ωoj̄ ωiō , and thus
dAω
dνt = .
|dT |2ω
11.4.2 T HEOREM . Let (X, ω) be a Stein manifold. Let T : X → D be a holomorphic submersion,

and denote by Xt := T −1 (t) the fiber, which is itself a Stein manifold. Fix a holomorphic line
bundle L → X with Hermitian metric e−ϕ such that
¯ + Ricci(ω) ≥ 0.
∂ ∂ϕ
Define Z
2 −ϕ dAω
Lt (ϕ) := f ∈ Γ(Xt , L|Xt ) ; |f | e < +∞
Xt |dT |2ω
and
At (ϕ) := Lt (ϕ) ∩ ΓO (Xt , L|Xt ),
and let Kt be the Bergman kernel of the Bergman projection Pt : Lt (ϕ) → At (ϕ). Then for any
holomorphic frame e for L∗ → X in a neighborhood U in X, the function

U 3 x 7→ log KT (x) (x, x), ex ⊗ ex
is plurisubharmonic.
11.4.1. Show that Theorem 11.4.1 is a special case of Theorem 11.4.2.
Proof of Theorem 11.4.2. As was to be shown in Exercise 11.1.4,
(11.11) Kt (x, x)e−ϕt (x) = sup{|f (x)|2 e−ϕt (x) ; f ∈ At with ||f || = 1}.
By Bergman’s Inequality the point evaluation functional is bounded, and hence by weak-∗ com-
pactness there exists a section gx,t ∈ At that realizes the supremum, in the sense that
|gx,t (x)|2 e−ϕt (x)

Kt (x, x)e−ϕt (x) = R 2 e−ϕt (y) dAω
.
Xt
|g x,t (y)| |dT | 2
ω
Now fix to ∈ D and ε > 0 such that Dε (to ) := {t ∈ C ; |t − to | < ε} ⊂⊂ D and let
T − to
X(ε) := T −1 (Dε (to )) and τ := ∈ O(X).
ε
168
Then by Theorem 11.3.1 there exists a holomorphic section G ∈ ΓO (X(ε), L) such that G|Xt =
gx,t and
Z Z Z
2 −ϕ 2 −ϕt dAω dAω
|G| e dVω ≤ π |gx,t | e 2
= πε2
|gx,t |2 e−ϕt .
X(ε) Xto |dτ |ω Xto |dT |2ω
Thus
Z
−ϕto (x) 2 −ϕto (x) 1
log Kto (x, x)e ≤ log |G(x)| e − log 2 |G|2 e−ϕ dVω
πε X(ε)
Z Z
2 −ϕto (x) 1 2 −ϕ dAω
= log |G(x)| e − log 2 |G| e dA(t)
πε t∈Dε (to ) Xt |dT |2ω
Z Z
2 −ϕto (x) 1 2 −ϕ dAω
≤ log |G(x)| e − 2 log |G| e dA(t),
where the equality is just Fubini’s Theorem and the second inequality follows from the concavity
of the logarithm.
Fix a frame e for L∗ over an open set U ⊂ X of the form U = T −1 (Dε (to )) with U so small
that it is biholomorphically a product. After multiplying e by a constant, we may assume without
loss of generality that ex ⊗ ex = e−ϕto (x) . Then the function hG, ei is holomorphic on U , and we
have
log hKto (x, x), ex ⊗ ex i = log Kto (x, x)e−ϕto (x)

Z Z
2 −ϕto (x) 1 2 −ϕ dAω
≤ log |G(x)| e − 2 log |G| e dA(t)
Z Z
2 1 2 −ϕ dAω
= log |hG, ei (x)| − 2 log |G| e dA(t)
πε Dε (to ) Xt |dT |2ω
Z Z
1 2 2 −ϕ dAω
≤ log |hG, ei| − log |G| e dA(t)
πε2 Dε (to ) Xt |dT |2ω
Z
1
= log |e|2 eϕt dA(t)
πε2 Dε (to )
Z Z
1 2 −ϕt 2 −ϕ dAω
+ 2 log(|G| e ) − log |G| e dA(t)
πε Dε (to ) Xt |dT |2ω
Z Z
1 1
≤ 2 ϕt
log |e| e dA(t) + 2 log(|Kt (x, x)|2 e−ϕt )dA(t)
πε2 Dε (to ) πε Dε (to )
Z
1
= log hKt (x, x), ex ⊗ ex i dA(t),
πε2 Dε (to )
where the second inequality is an application of the sub-mean value property for the plurisub-
harmonic function log |hG, ei (x)|2 and the third and last inequality comes from (11.11). We have
therefore shown that the function log hKt (x, x), ex ⊗ ex i is subharmonic in t, and since it is clearly
plurisubharmonic in x, the proof is complete.
169
11.4.3 R EMARK . In Chapter 11.2 our proof of the Suita conjecture was made longer only because
Berndtsson’s Theorem implies the log-subharmonicity of the Bergman kernels Kt , rather than
e−t BXt . With Theorem 11.4.2 in hand, a much shorter proof is possible. The interested reader
should consult [BL-2016], which treats the case of domains. It is a good exercise to adapt the
proof of the Suita Conjecture in [BL-2016] to the setting of general Riemann surfaces.
11.4.2 Plurisubharmonic variation versus sharp L2 extension

We have seen that Berndtsson’s Theorem 11.1.8 on (pluri)subharmonic variation directly implies
(i) Theorem 11.4.1 in the special case of a product domain D = Ω × X, and
(ii) Theorem 11.3.1 on L2 extension with sharp constants.
We have also seen that Theorem 11.3.1 implies Theorem 11.4.1 in its full generality. We shall now
show that Theorem 11.3.1 implies a much stronger version of Berndtsson’s Theorem, in which
the product manifold X × D is replaced by a more general Stein manifold, and the Hilbert spaces
H (ϕ)t are no longer required to be quasi-isometric, which is to say, that H (ϕ) → D need not be
a vector bundle. We shall now make things more precise.
Let (X, ω) be a Kähler manifold of complex dimension m + n and let T : X → Dm a holomor-
phic submersion of X in the unit polydisk in Cm . Denote by Xt := T −1 (t) the (n-dimensional)
fibers of T . Fix a holomorphic line bundle L → X with singular Hermitian metric e−ϕ , and define
the Hilbert spaces
Z
2 −ϕ dAω
H (ϕ)t := f ∈ ΓO (Xt , L) ; ||f ||t :=
2
|f | e 2
< +∞ , t ∈ Dm .
Xt |dT |ω
Then we have a map H (ϕ) → Dm whose fibers are Hilbert spaces, and which we call the Hilbert
fibration associated to (X, ω; L, e−ϕ ; T ). Note that we do not require local triviality, or even equiv-
alence of the fibers, from a Hilbert fibration.
11.4.4 D EFINITION . (S) A section f of a Hilbert fibration H (ϕ) → Dm is a section Ff ∈
Γ(X, L) such that Ff |Xt ∈ H (ϕ)t for each t ∈ Dm . We write
ft := Ff |Xt .
The section f of H (ϕ) → Ω is said to be holomorphic if Ff ∈ ΓO (X, L). We denote by
Γ(Ω, H (ϕ)) and ΓO (Ω, H (ϕ))
the set of sections, and holomorphic sections, of H (ϕ) → Dm respectively. (Thus all
sections are holomorphic on the fibers, and a holomorphic section means it is holomorphic
in the total space.)
(B∗ ) Let H (ϕ)∗ → Dm denote the bundle of dual spaces, i.e., the fiber H (ϕ)∗t of this bundle
over t ∈ Dm is the dual Hilbert space of H (ϕ)t , with its usual Hilbert norm
|hξ, f i|
||ξ||t∗ := sup .
f ∈H (ϕ)t −{0} ||f ||t
170
(S∗ ) A section of H (ϕ)∗ → Ω is a map ξ : H (ϕ) → C such that
ξt := ξ|H (ϕ)t ∈ H (ϕ)∗t .
The section ξ of H (ϕ)∗ → Ω is said to be holomorphic if for each f ∈ ΓO (Dm , H (ϕ)) the
function
Dm 3 t 7→ hξt , f i ∈ C
is holomorphic. The set of holomorphic sections is denoted ΓO (Ω, H (ϕ)∗ ).
Let us start with the case m = 1. We can now state the following generalization of Theorem
11.1.8.
11.4.5 T HEOREM . Let (X, ω) be a Stein Kähler manifold and let T : X → D be a holomorphic
submersion. Let L → X be a holomorphic line bundle with singular Hermitian metric e−ϕ . Let
H (ϕ) → D be the Hilbert fibration associated to (X, ω; L, e−ϕ ; T ). Assume that
¯ + Ricci(ω) ≥ 0.
∂ ∂ϕ
Then for any holomorphic section ξ ∈ ΓO (D, H (ϕ)∗ ) the function
D 3 t 7→ log ||ξt ||2t∗
is subharmonic.
Proof. The proof is very similar to that of Theorem 11.4.2. First,
|hξt , f i|2
(11.12) ||ξt ||2t∗ = sup R dAω
f ∈H (ϕ)t Xt
|f |2 e−ϕ |dT |2
ω
Now fix to ∈ D and ε > 0 such that Dε (to ) := {t ∈ C ; |t − to | < ε} ⊂⊂ D and let
T − to
X(ε) := T −1 (Dε (to )) and τ := ∈ O(X).
ε
Then by Theorem 11.3.1 there exists a holomorphic section F ∈ ΓO (X(ε), L) such that F |Xto = f
and Z Z Z
2 −ϕ 2 −ϕ dAω dAω
|F | e dVω ≤ π |f | e 2
= πε2
|f |2 e−ϕ .
X(ε) Xto |dτ |ω Xto |dT |2ω
Fixing f ∈ H (ϕ)t , we let Ff ∈ ΓO (X(ε), L) denote the unique extension of minimal norm. Then
|hξto , f i|2 1 |hξto , f i|2

R
2 e−ϕ dVω
≤ 2
R
2 e−ϕ dV
.
Xt
|f | |dT | 2 πε X(ε)
|F f | ω
o ω
171
Choosing the section fo that realizes the supremum (11.12) and writing Fo := Ffo , we have
Z
2 dVω
2
log ||ξto ||to ∗ = log |hξto , fo i| − log |fo |2 e−ϕ
X |dT |2ω
Z to
dVω
= log |hξto , Fo i|2 − log |fo |2 e−ϕ
X |dT |2ω
to Z
2 1 2 −ϕ
≤ log |hξto , Fo i| − log |Fo | e dVω
πε2 X(ε)
Z Z
2 1 2 −ϕ dAω
= log |hξto , Fo i| − log |Fo | e dA(t)
πε2 t∈Dε (to ) Xt |dT |2ω
Z Z
2 1 2 −ϕ dAω
≤ log |hξto , Fo i| − 2 log |Fo | e dA(t)
Z Z
1 2 2 −ϕ dAω
≤ 2 log |hξt , Fo i| − log |Fo | e dA(t)
Z
1
≤ 2 log ||ξt ||2t∗ dA(t),
πε Dε (to )
where the second inequality follows from concavity of the logarithm, the third inequality follows
from the holomorphicity of t 7→ hξt , Fo i (i.e., the fact that ξt is a holomorphic section), and the last
inequality follows from (11.12). The proof of Theorem 11.4.5 is complete.
E XERCISES
172
Bibliography
[B-1996] Berndtsson, B., The extension theorem of Ohsawa-Takegoshi and the theorem of
Donnelly-Fefferman. Ann. Inst. Fourier (Grenoble) 46 (1996), no. 4, 1083–1094.
[B-2006] Berndtsson, B., Subharmonicity properties of the Bergman kernel and some other func-
tions associated to pseudoconvex domains. Ann. Inst. Fourier (Grenoble) 56 (2006), no. 6,
1633–1662.
[B-2009] Berndtsson, B., Curvature of vector bundles associated to holomorphic fibrations. Ann.
of Math. (2) 169 (2009), no. 2, 531 – 560.
[BL-2016] Berndtsson, B.; Lempert, L., A proof of the Ohsawa-Takegoshi theorem with sharp
estimates. J. Math. Soc. Japan 68 (2016), no. 4, 1461–1472.
[B-2013] Błocki, Z, Suita conjecture and the Ohsawa-Takegoshi extension theorem. Invent. Math.
193 (2013), no. 1, 149 – 158.
[B-1970] Bombieri, E., Algebraic values of meromorphic maps. Invent. Math. 10 (1970), 267–
287.
[B-1970a] Bombieri, E., Addendum to my paper: “Algebraic values of meromorphic maps” (In-
vent. Math. 10 (1970), 267–287). Invent. Math. 11 (1970), 163–166.
[D-1982] Demailly, J.-P., Estimations L2 pour l’opérateur ∂¯ d’un fibré vectoriel holomorphe semi-
positif au-dessus d’une varié té kählérienne complète. (French) Ann. Sci. École Norm. Sup.
(4) 15 (1982), no. 3, 457–511.
[Fe-1974] Fefferman, C., The Bergman kernel and biholomorphic mappings of pseudoconvex do-
mains. Invent. Math. (1974) no. 26, 1–65.
[GZ-2015] Guan, Q.; Zhou, X., A solution of an L2 extension problem with an optimal estimate
and applications. Ann. of Math. (2) 181 (2015), no. 3, 1139–1208.
[H-1990] Hörmander, L., An introduction to complex analysis in several variables. Third edition.
North-Holland Mathematical Library, 7. North-Holland Publishing Co., Amsterdam, 1990.
[Ma-1993] Manivel, L., Un théorème de prolongement L2 de sections holomorphes d’un fibré

hermitien. Math. Z. 212 (1993), no. 1, 107–122.
173
[Mc-1996] McNeal, J., On large values of L2 holomorphic functions. Math. Res. Lett. 3 (1996),
no. 2, 247–259.
[MV-2007] McNeal, J.; Varolin, D., Analytic Inversion of Adjunction: L2 extension theorems with
gain. Ann. Inst. Fourier (Grenoble) 57 (2007), no. 3, 703–718.
[MV-2015] McNeal, J.; Varolin, D., L2 estimates for the ∂¯ operator. Bull. Math. Sci. 5 (2015), no.
2, 179–249.
[OT-1987] Ohsawa-, T.; Takegoshi, K., On the extension of L2 holomorphic functions. Math. Z.
195 (1987), no. 2, 197–204.
[O-2001] Ohsawa, T., On the extension of L2 holomorphic functions. V. Effects of generalization.
Nagoya Math. J. 161 (2001), 1–21.
[O-2001e] Ohsawa, T., Erratum to: ”On the extension of L2 holomorphic functions. V. Effects of
generalization”. Nagoya Math. J. 163 (2001), 229.
[PV-2016] Pingali, V.; Varolin, D., Bargmann-Fock extension from singular hypersurfaces. J.
Reine Angew. Math. 717 (2016), 227–249.
[PV-2019] Pingali, V.; Varolin, D., Non-uniformly flat affine algebraic hypersurfaces. To appear
in Nagoya Math J.
[OSV-2006] Ortega Cerdá, J.; Schuster, A.; Varolin, D., Interpolation and sampling hypersurfaces
for the Bargmann-Fock space in higher dimensions. Math. Ann. 335 (2006), no. 1, 79–107.
[Sc-1946] Schiffer, M., The kernel function of an orthonormal system. Duke Math. J. 13, (1946).
529–540.
[S-1996] Siu, Y.-T. The Fujita conjecture and the extension theorem of Ohsawa-Takegoshi. Ge-
ometric complex analysis (Hayama, 1995), 577 – 592, World Sci. Publ., River Edge, NJ,
1996.
[S-1998] Siu, Y.-T. Invariance of plurigenera. Invent. Math. 134 (1998), no. 3, 661 – 673.
[S-2002] Siu, Y.-T. Extension of twisted pluricanonical sections with plurisubharmonic weight
and invariance of semipositively twisted plurigenera for manifolds not necessarily of general
type. Complex geometry (Göttingen, 2000), 223 – 277, Springer, Berlin, 2002.
[Su-1971] Suita, N., Capacities and kernels on Riemann surfaces. Arch. Rational Mech. Anal. 46
(1972), 212 – 217.
[V-2018] Varolin, D., Bergman interpolation on finite Riemann surfaces. Part I: Asymptotically
flat case. J. Anal. Math. 136 (2018), no. 1, 103–149.
[V-2016] Varolin, D., Bergman interpolation on finite Riemann surfaces. Part II: Poincar-
hyperbolic case. Math. Ann. 366 (2016), no. 3-4, 1137–1193.
174

Prof. Dror Varolin PDF

Uploaded by

Prof. Dror Varolin PDF

Uploaded by

ICTS Lectures

Hilbert Space Techniques

3 Connections and Curvature 32

5 The Bochner-Kodaira Identity 70

6 L2 Estimates on Compact Kähler manifolds 76

7 L2 Estimates on Complete Kähler manifolds 81

III Twisted Methods 96

9 Extension Theorems 104

10 Some applications of the L2 Extension Theorem 123

11 Berndtsson’s Theorem on Plurisubharmonic Variation 142

Complex Manifolds and Holomorphic

1.1 Complex manifolds and holomorphic functions

such that for each pair α, β ∈ J the homeomorphism

(1.1) det(DF ) = | det ∂F |2 ,

1.2 Examples of complex manifolds

V 3 v 7→ (z 0 (v), ..., z n (v)) ∈ Cn+1 .

Any homogeneous function of degree k is given by a homogeneous polynomial of degree k in the

Ui := {[v] ∈ P(V ) ; z i (v) 6= 0}, i = 0, ..., n.

are coordinates on Ui , and ϕi = (ζi1 , . . . , ζin ).

Z := {z ∈ Cn+1 ; Fj (z) = 0, 1 ≤ j ≤ k}.

F` = (z j )d` f`,j (ζj ), 1 ≤ ` ≤ k.

It follows that on Cn+1 ∩ [ ]−1 (Uj ) ∩ {F` = 0},

dF` = (z j )d` dfj,` (ζj ) ⊃ Z.

B̃o := {(z, `) ∈ B × Pn ; z ∈ `}.

φ−1 (Ui ) = {((λxo , ..., |{z}

1.2.11 D EFINITION . The manifold TΛ (V ) is called a complex torus.

P(L) : P(V ) 3 [v] 7→ [Lv] ∈ P(W )

is well-defined and injective.

1.2.5. Show that the sets

R := {[x, y, z] ∈ P2 ; xy = z 2 } and E := {[x, y, z] ∈ P2 ; y 2 z = x3 + xz 2 + z 3 }

are Riemann surfaces, that R is biholomorphic to P1 , and that E is not biholomorphic to P1 .

1.3 Holomorphic vector bundles

where Vx := π −1 (x) denotes the fiber of π over x ∈ X. Such a collection of maps

• A holomorphic vector bundle of rank 1 is called a holomorphic line bundle. Holomorphic

ei (x) := F (x, ei ), 1≤i≤r

F −1 (f i (x)ei (x)) := (x, f i (x)ei ).

X ×Z Y := {(x, y) ∈ X × Y ; f (x) = g(y)}.

There are projection maps X ×Z Y → X and X ×Z Y → Y given by the restriction to

is a holomorphic vector bundle, called the pullback of V by f .

1.3.2 Transition functions

It is clear from the definitions that

ψ 0 ◦ ψ −1 (x, t) = (x, gU 0 U (x)t)

The proof is left as an exercise.

(Exercise 1.3.3). Defining

Define the vector bundle π : TX1,0 → X by

s1,0 := π 1,0 ◦ ι : TX → TX1,0

is a real isomorphism of vector bundles whose inverse is the map ξ 7→ 2Re ξ.

E+ (J) = TX1,0 (J) and E− (J) = TX0,1 (J)

for the J-eigenspaces.

1.3.4 Some non-trivial examples of holomorphic line bundles

We will meet the line bundle LZ when we study L2 extension. 

1.3.4. Show that the operator J defined

1.3.8. Show that the line bundle KPn → Pn is isomorphic to U⊗(n+1) → Pn .

1.4 Differential forms on complex manifolds

and, with the notation

Locally every (p, q)-form is of the form

∂α := π p+1,q dα ¯ := π p,q+1 dα,

One has the following proposition.

[Γ(X, TX1,0 (J)), Γ(X, TX1,0 (J))] ⊂ Γ(X, TX1,0 (J))

1.4.3 Twisted differential forms

then the form

∂¯ : Γ(X, Λp,q p,q+1

Λp,q ∼ p,0 0,q ∼ p,0 ∗ n,q

one can write

2.1 Hermitian metrics for complex vector bundles

2.2 Hermitian Riemannian metrics for complex manifolds

2.2.1 D EFINITION . A Riemannian metric g on X is said to be Hermitian if J ∗ g = g, i.e.,

g(Jξ, Jη) = g(ξ, η)

for all x ∈ X and all ξ, η ∈ TX,x . 

2.2.1 Structure of the metric

where † means transpose conjugate. Thus we have

where in the last equality we used the Hermitian symmetry of g. Therefore

g(v, w) = 2Re h(s1,0 v, s1,0 w),

2.2.2 Metric form

ωg (v, w) := g(Jv, w),

= −2Im hij̄ αi , s1,0 v hαj , s1,0 wi = −Im 2h(s1,0 v, s1,0 w),

We will meet the line bundle LZ when we study L2 extension.

for all x ∈ X and all ξ, η ∈ TX,x .