A numerical criterion for very ample line bundles

Jean-Pierre Demailly
Université de Grenoble I,
Institut Fourier, BP 74,
Laboratoire associé au C.N.R.S. n˚188,
F-38402 Saint-Martin d’Hères

Abstract. — Let X be a projective algebraic manifold of dimension n and let L

be an ample line bundle over X. We give a numerical criterion ensuring that the
adjoint bundle KX + L is very ample. The sufficient conditions are expressed in
terms of lower bounds for the intersection numbers Lp ·Y over subvarieties Y of X.
In the case of surfaces, our criterion gives universal bounds and is only slightly
weaker than I. Reider’s criterion. When dim X ≥ 3 and codim Y ≥ 2, the lower
bounds for Lp · Y involve a numerical constant which depends on the geometry
of X. By means of an iteration process, it is finally shown that 2KX + mL is very
ample for m ≥ 12 nn . Our approach is mostly analytic and based on a combination
of Hörmander’s L2 estimates for the operator ∂, Lelong number theory and the
Aubin-Calabi-Yau theorem.

Table of contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 2
2. Singular hermitian metrics on holomorphic line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 5
3. Basic results on Lelong numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 6
4. L2 estimates and existence of holomorphic sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 8
5. Aubin-Calabi-Yau theorem and convexity inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 12
6. Mass concentration in the Monge-Ampère equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 14
7. Choice of the logarithmic singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 17
8. Upper bound for the 1-codimensional polar components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 19
9. Approximation of closed positive (1,1)-currents by divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 21
10. Self-intersection inequality for closed positive currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 25
11. Proof of the criterion in arbitrary dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 33
12. Universal bounds for very ample line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . p. 43

Key words: Very ample line bundle, numerical criterion, plurisubharmonic

function, closed positive current, Lelong number, Monge-Ampère equation, Aubin-
Calabi-Yau theorem, intersection theory.

A.M.S. Classification: 32J25, 32L10, 32C30

1
 1. Introduction

Let L be a holomorphic line bundle over a projective algebraic manifold X of

dimension n. We denote the canonical line bundle of X by KX and use an additive
notation for the group Pic(X) = H 1 (X, O⋆ ). The original motivation of this work
was to study the following tantalizing conjecture of Fujita [Fu 88]: if L ∈ Pic(X)
is ample, then KX + (n + 2)L is very ample; the constant n + 2 would then be
optimal since KX +(n+1)L = OX is not very ample when X = IPn and L = O(1).
Although such a sharp result seems at present out of reach, a consequence of our
results will be that 2KX + mL is always very ample for L ample and m larger
than some universal constant depending only on n.

Questions of this sort play a very important role in the classification theory
of projective varieties. In his pioneering work [Bo 73], Bombieri proved the
existence of pluricanonical embeddings of low degree for surfaces of general type.
More recently, for an ample line bundle L over an algebraic surface S, I. Reider
[Rei 88] obtained a sharp numerical criterion ensuring that the adjoint line bundle
KS + L is spanned or very ample; in particular, KS + 3L is always spanned
and KS + 4L very ample. Reider’s method was further developed by Catanese
[Ca 88], Sakai [Sa 88] and Beltrametti-Francia-Sommese [BFS 89], who studied
the existence of higher order embeddings via s-jets. Reider’s approach is based on
the construction of rank two vector bundles associated to some 0-cycles in special
position with respect to the linear system |KS + L| and a use of Bogomolov’s
inequality for stable vector bundles. Unfortunately, these methods do not apply
in dimension ≥ 3 and no similar general result was available. In a somewhat
different context, Fujita [Fu 87] proved that KX + (n + 2)L is always ample. This
result is obtained via Mori’s theory of extremal rays [Mo 82] and the cone theorem
of Kawamata (cf. [Ka 84], [KMM 87]), but the arguments are purely numerical
and give apparently no insight on the very ample property.

Our purpose here is to explain a completely different analytic approach

which is applicable in arbitrary dimension. Let us first recall a few usual notations
that will be used constantly in the sequel:
Z
(1.1) L1 · · · Lp · Y = c1 (L1 ) ∧ . . . ∧ c1 (Lp )
Y

denotes the intersection product of p line bundles L1 , . . . , Lp over a p-dimensional

subvariety Y ⊂ X. In case L1 = . . . = Lp we write instead Lp · Y and in case
Y = X we omit Y in the notation. Similar notations will be used for divisors.
Recall that a line bundle (or a IR-divisor) L over X is said to be numerically
effective, nef for short, if L · C ≥ 0 for every curve C ⊂ X ; in this case L is said
to be big if Ln > 0. More generally, a vector bundle E is said to be nef if the
associated line bundle OE (1) is nef over P (E ⋆ ) = projective space of hyperplanes
in E ; any vector bundle E such that some symmetric power S m E is spanned by
its global sections is nef. In this context, we shall prove:

2
 Main Theorem. — Let X be a projective n-fold and let L be a big nef
line bundle over X. Suppose that there is a number a ≥ 0 such that T X ⊗ O(aL)
is nef. Then KX + L is spanned at each point of a given subset Ξ of X (resp.
separates all points in Ξ, resp. generates s-jets at any point of Ξ ) provided that
Ln > σ0 with σ0 = nn (resp. σ0 = 2nn , resp. σ0 = (n + s)n ), and that there exists
a sequence 0 = β1 < . . . < βn ≤ 1 such that any subvariety Y ⊂ X of codimension
p = 1, 2, . . . , n − 1 intersecting Ξ satisfies
X
Ln−p · Y > (βp+1 − β1 )−1 . . . (βp+1 − βp )−1 Sjp (β) aj σp−j
0≤j≤p−1

with S0p (β) = 1, Sjp (β) = elementary symmetric function of degree j in β1 , . . . , βp

and   σ0 p/n  n
σp = 1 − 1 − n L , σp ∈ ]σ0 p/n, σ0 [ .
L

The expression “separation of points” used here includes infinitesimal

separation, that is, generation of 1-jets at each point (the constant σ0 = (n + 1)n
corresponding to s = 1 can therefore be replaced by the smaller value 2nn ). In fact,
our proof also gives sufficient conditions for the generation of jets corresponding
to arbitrary 0-dimensional subschemes (Ξ, OΞ ) of X, simply by changing the value
of σ0 ; for example, if (Ξ, OΞ ) is a local complete intersection, the constant σ0 can
be taken equal to nn h0 (Ξ, OΞ ) ; unfortunately, this value is in general far from
being optimal. Notice that the number a involved in the hypothesis on T X need
not be an integer nor even a rational number: the hypothesis then simply means
that any real divisor associated to OT X (1) + a π ⋆ L is nef over P (T ⋆ X).
As the notation is rather complicated, it is certainly worth examining the
particular case of surfaces and 3-folds. If X is a surface, we have σ0 = 4 (resp.
σ0 = 8, resp. σ0 = (2 + s)2 ), and we take β1 = 0, β2 = 1. This gives only two
conditions, namely
(1.2) L2 > σ0 , L · C > σ1
for every curve C intersecting Ξ. In that case, the proof shows that the assumption
on the existence of a is unnecessary. These bounds are not very far from those
obtained with Reider’s method, although they are not exactly as sharp. If X is
a 3-fold, we have σ0 = 27 (resp. σ0 = 54, resp. σ0 = (3 + s)3 ), and we take
β1 = 0 < β2 = β < β3 = 1. Therefore our condition is that there exists β ∈ ]0, 1[
such that
(1.3) L3 > σ0 , L2 · S > β −1 σ1 , L · C > (1 − β)−1 (σ2 + β aσ1 )
for every curve C or surface S intersecting Ξ.
In general, we measure the “amount of ampleness” of a nef line bundle L
on a subset Ξ ⊂ X by the number
(1.4) µΞ (L) = min min (Lp · Y )1/p ,
1≤p≤n dim Y =p, Y ∩Ξ6=∅

3
where Y runs over all p-dimensional subvarieties of X intersecting Ξ. The Nakai-
Moishezon criterion tells us that L is ample if and only if µX (L) > 0. An effective
version of this criterion can be easily deduced from the Main Theorem: in fact, a
suitable choice of the constants βp in terms of a, σ0 and µΞ (L) yields:

Corollary 1. — Let L be a big nef line bundle over X such that

T X ⊗ O(aL) is nef for some a ≥ 0, and let Ξ be an arbitrary subset of X. Then
the line bundle KX + mL spans (resp. separates points, resp. generates s-jets) on
Ξ as soon as
n 1 k 1 1 o


1 + 1 + . . . + k+1
m> max Bn σ0 , (Bn σ0 ) n−k (aµΞ (L)) n−k n−1 n−2
µΞ (L) 1≤k≤n−2

with σ0 = nn (resp. σ0 = 2nn , resp. σ0 = (n+s)n ) and with a constant Bn < 2.005
depending only on n (table (11.11) contains the first values of Bn ).

When L is ample, the number a always exists and we have µΞ (L) ≥ 1 for
any choice of Ξ. We thus get an explicit lower bound m0 depending only on n, a
such that KX + mL is spanned or very ample for m ≥ m0 . Unfortunately, these
lower bounds are rather far from Fujita’s expected conditions m ≥ n + 1 and
m ≥ n + 2 respectively. Observe however that the lower bound for Ln in the Main
Theorem is optimal: if X = IPn and L = O(1), then KX = O(−n − 1) so KX + nL
is not spanned, although (nL)n =
nn = σ0 . Similarly KX + (n + s)L does not
n
generate s-jets, although (n + s)L = (n + s)n = σ0 . When X ⊂ IPn+1 is the
n-dimensional quadric and L = OX (1), then KX + nL = OX is not very ample,
although (nL)n = 2nn = σ0 .
Another unsatisfactory feature is that our bounds depend on the geometry
of X through the number a, while the case of curves or surfaces suggests that they
should not. In fact, our proof uses a rather delicate self-intersection inequality for
closed positive currents, and this inequality (which is essentially optimal) depends
in a crucial way on a bound for the “negative part” of T X. It follows that new
ideas of a different nature are certainly necessary to get universal bounds for
the very ampleness of KX + L. However, an elementary argument shows that
T X ⊗ O(KX + nF ) is always nef when F is very ample (see lemma 12.1). This
observation combined with an iteration of the Main Theorem finally leads to a
universal result. Corollary 2 below extends in particular Bombieri’s result on
pluricanonical embeddings of surfaces of general type to arbitrary dimensions (at
least when KX is supposed to be ample, see 12.10 and 12.11), and can be seen as
an effective version of Matsusaka’s theorem ([Ma 72], [KoM 83]) :

Corollary 2. — If L is an ample line bundle over X, then 2KX + mL

is very ample, resp. generates s-jets, when (m − 1)µX (L) + s > 2Cn σ0 with a
constant Cn < 3 depending only on n (see table (12.3)). In particular, 2KX + mL
is very ample for m > 4Cn nn and generates higher s-jets for m > 2Cn σ0 .

Our approach is based on three rather powerful analytic tools. First, we

use Hörmander’s L2 estimates for the operator ∂ with singular plurisubharmonic

4
weights to prove a general abstract existence theorem for sections of KX + L
with prescribed jets at finitely many points; the idea is similar to that of the
Hörmander-Bombieri-Skoda theorem, but following an idea of A. Nadel [Na 89],
we consider plurisubharmonic functions with logarithmic poles associated to an
arbitrary ideal in OX,x (see Corollary 4.6). We refer to [De 90] for further results
relating ample or nef line bundles to singular hermitian metrics. The second tool
is the Aubin-Calabi-Yau theorem. This fundamental result allows us to solve the
Monge-Ampère equation (ω+ πi ∂∂ψ)n = f where ω = 2π i
c(L) is the curvature
R form
of L, and the right hand side f is an arbitrary positive (n, n)-form with X f = Ln .
We let f converge to a linear combination of Dirac measures and show that the
solution ψ produces in the limit a singular weight on L with logarithmic poles. In
order to control the poles and singularities, we use in an essential way a convexity
inequality due to Hovanski [Hov 79] and Teissier [Te 79,82], which can be seen as a
generalized version of the Hodge index theorem for surfaces. Finally we invoke in
several occasions the theory of closed positive currents and Lelong numbers (see
[Le 57,69]). In particular, the generalized Lelong numbers introduced in [De 87]
are used as a substitute of the intersection theory of algebraic cycles in our analytic
context. The self-intersection inequality 10.7 can be seen as a generalization to
currents (and in any dimension) of the classical upper bound d(d − 1)/2 for the
number of multiple points of a plane curve of degree d. It actually gives a bound
for the sum of degrees of the irreducible components in the sublevel sets of Lelong
numbers of a closed positive (1, 1)-current T with integral cohomology class {T },
in terms of an explicit polynomial in {T }.
A major part of this work has been done during the fall 1989, while
the author was visiting Bayreuth University under the support of the DFG-
Forschungsschwerpunktprogramm“Komplexe Mannigfaltigkeiten”. The author
expresses special thanks to Michael Schneider and Thomas Peternell for very
stimulating discussions.

2. Singular hermitian metrics on holomorphic line bundles

Let L be a holomorphic line bundle over a projective algebraic manifold

X and n = dim X. If L is equipped with a hermitian metric, we denote
i
by c(L) = 2π ∇2 the Chern curvature form, which is a closed real (1, 1)-form
representing the first Chern class c1 (L) ∈ H 2 (X, ZZ). It is well known that L is
ample if and only if L has a smooth hermitian metric such that c(L) is positive
definite at every point.
However, we are also interested in singular metrics, because they often give
additional information about the existence of sections of high multiples mL. By
definition, a singular metric on L is a metric which is given in any trivialization
≃
τ : L`| Ω −→ Ω × C by
(2.1) kξk = |τ (ξ)| e−ϕ(x) , x ∈ Ω, ξ ∈ Lx
1
where ϕ ∈ Lloc (Ω) is a weight function. Then the curvature of L is given P by
the (1, 1)-current c(L) = πi ∂∂ϕ on Ω. For example, to any divisor D = λj Dj

5
with coefficients λj ∈ ZZ is associated the invertible sheaf O(D) of meromorphic
functions f such that div(f ) + D ≥ 0 ; the corresponding line bundle can be
equipped with the singular metric defined by ||f || = |f |. If gj is a generator of the
Q λ
ideal of Dj on an open set Ω ⊂ X, then τ (f ) = f gj j defines a trivialization
P of
O(D) over Ω, thus our singular metric is associated to the weight ϕ = λj log |gj |.
By the Lelong-Poincaré equation, we find

i
(2.2) c O(D) = ∂∂ϕ = [D],
P π
where [D] = λj [Dj ] denotes the current of integration over D.
In the sequel, all singular metrics are supposed to have positive curvature
in the sense of currents (cf. [Le 57]); i.e., the weight functions ϕ are supposed
to be plurisubharmonic. Let us recall some results
of [De 90]: consider the real
Neron-Severi space N SIR (X) = H 2 (X, ZZ)∩H 1,1 (X) ⊗IR of algebraic cohomology
classes of degree 2, and let Γ+ ⊂ N SIR (X) (resp. Γa ⊂ Γ+ ), be the closed convex
cone generated by cohomology classes of effective (resp. ample) divisors D ; denote
by Γ◦+ (resp. Γ◦a ) the interior of Γ+ (resp. Γa ). Then, if ω is a Kähler metric on
X and ε > 0, we have the following equivalent properties:
(2.3) c1 (L) ∈ Γ+ ⇐⇒ L has a singular metric with c(L) ≥ 0 ;
(2.4) c1 (L) ∈ Γ◦+ ⇐⇒ ∃ε, L has a singular metric with c(L) ≥ εω ⇐⇒ κ(L) = n ;
(2.5) c1 (L) ∈ Γa ⇐⇒ ∀ε, L has a smooth metric with c(L) ≥ −εω ⇐⇒ L is nef ;
(2.6) c1 (L) ∈ Γ◦a ⇐⇒ ∃ε, L has a smooth metric with c(L) ≥ εω ⇐⇒ L is ample.
The notation κ(L) stands for the Kodaira dimension of L, that is by definition, the
supremum of the generic rank of the rational maps to projective space defined by
the non zero sections in H 0 (X, mL) for m ≥ 1 (if any), and κ(L) = −∞ otherwise;
alternatively, κ(L) is the smallest constant such that h0 (X, mL) ≤ O(mκ(L) ). The
only thing that will be needed here is the fact that a big nef line bundle satisfies the
equivalent properties in (2.4); we shall briefly sketch the proof of this. If L is nef,
the Hilbert polynomial of χ(X, mL) has leading coefficient Ln /n! ≥ 0, and it is
well known that hj (X, mL) = O(mn−1 ), thus h0 (X, mL) = Ln /n! mn + O(mn−1 ).
Hence L is big if and only if κ(L) = n. Let A be an ample divisor. Then
H 0 (X, mL − A) is the kernel of H 0 (X, mL) → H 0 (A, mL`| A ), and the target has
dimension ≤ Cmn−1 . When κ(L) = n we get H 0 (X, mL − A) 6= 0 for m large,
so there is an effective divisor E such that mL ≃ A + E. Now, pL + A is ample
for every p ≥ 0, so pL + A has a smooth metric with c(pL + A) ≥ εp ω, and the
isomorphism (m + p)L ≃ pL + A + E gives a metric on L such that
(2.7) c(L) = (m + p)−1 (c(pL + A) + [E]) ≥ (m + p)−1 εp ω.
Observe that the singular part (m + p)−1 [E] can be chosen as small as desired by
taking p large.

3. Basic results on Lelong numbers

These results will be needed in the sequel as an analytic analogue of some

standard facts in the intersection theory of algebraic cycles. They are developed

6
in more details in [De 82a,85,87] (cf. Lelong [Le 57,69] for an earlier presentation).
We first recall a few definitions. Let T be a closed positive current of bidimension
(p, p), that is, of bidegree (n − p, n − p), on an open set Ω ⊂ Cn . The Lelong
number of T at a point x ∈ Ω is defined by ν(T, x) = limr→0+ ν(T, x, r) where
Z
1
(3.1) ν(T, x, r) = T (z) ∧ (i∂∂|z|2 )p
(2πr2 )p B(x,r)
measures the ratio of the mass of T in the ball B(x, r) to the area of the ball
of radius r in Cp ; this ratio is an increasing function of r (cf. [Le 69]), and the
limit ν(T, x) does not depend on the choice of coordinates. In the case where T
is a current of integration [A] over an analytic subvariety A, the Lelong number
ν([A], x) coincides with the multiplicity of A at x (Thie’s theorem [Th 67]).
More generally, let ϕ be a continuous plurisubharmonicP function with an
isolated −∞ pole at x, e.g. a function of the form ϕ(z) = log 1≤j≤N |gj (z)|γj ,
γj > 0, where (g1 , . . . , gN ) is an ideal of germs of holomorphic functions in Ox
with g −1 (0) = {x}. According to [De 87], the generalized Lelong number ν(T, ϕ)
of T with respect to the weight ϕ is the limit when t tends to −∞ of
Z i p
(3.2) ν(T, ϕ, t) = T∧ ∂∂ϕ .
ϕ(z)<t π
Because of the logarithmic singularity of ϕ, the integral is not well defined a priori.
In fact, we can use Bedford and Taylor’s definition of the Monge-Ampère operator
for locally bounded plurisubharmonic functions (see § 10) and set
Z Z
p

p
(3.3) T ∧ (i∂∂ϕ) = T (z) ∧ i∂∂ max(ϕ(z), s) , s < t;
ϕ(z)<t ϕ(z)<t

observe that the right hand side is independent of s by Stokes’ formula. The
relation with ordinary Lelong numbers comes from the equality
(3.4) ν(T, x, r) = ν(T, ϕ, log r), ϕ(z) = log |z − x|,
in particular ν(T, x) = ν(T, log |• − x|). This equality is in turn a consequence of
the following general formula, applied to χ(t) = e2t and t = log r :
Z Z

p
(3.5) T ∧ i∂∂χ ◦ ϕ = χ′ (t − 0)p T ∧ i∂∂ϕ)p ,
ϕ(z)<t ϕ(z)<t

where χ is an arbitrary convex increasing function. To prove the formula, we use

a regularization and thus suppose that T , ϕ and χ are smooth, and that t is a non
critical value of ϕ. Then Stokes’ formula shows that the integrals on the left and
on the right of (3.5) are equal respectively to
Z Z

p−1
p−1
T ∧ i∂∂χ ◦ ϕ ∧ i∂(χ ◦ ϕ), T ∧ i∂∂ϕ ∧ i∂ϕ,
ϕ(z)=t ϕ(z)=t

and the differential form of bidegree (p − 1, p) appearing in the integrand of the

first integral is equal to (χ′ ◦ ϕ)p (i∂∂ϕ)p−1 ∧ i∂ϕ. The expected formula follows.
It is shown in [De 87] that ν(T, ϕ) depends only on the asympotic behaviour
of ϕ near the pole x, namely the Lelong number remains unchanged for a weight

7
ψ such that limz→x ψ(z)/ϕ(z) = 1. More generally, if lim supz→x ψ(z)/ϕ(z) = λ,
then
(3.6) ν(T, ψ) ≤ λp ν(T, ϕ).
Finally, let F is a proper holomorphic map from a neighborhood of x onto a
neighborhood of y in Cn , and let ψ be a continuous plurisubharmonic function
with an isolated pole at y. The definition of the direct image F⋆ T by adjunction
of F ⋆ easily shows that for t < t0 sufficiently small
(3.7) ν(F⋆ T, ψ, t) = ν(T, ψ ◦ F, t), ν(F⋆ T, ψ) = ν(T, ψ ◦ F ).
For any closed positive current T of bidimension (p, p) on a complex
manifold X and any positive number c, we let Ec (T ) be the set of points z ∈ X
where ν(T, z) ≥ c. By a theorem of [Siu 74], all sublevel sets Ec (T ) are closed
analytic subsets of X of dimension at most p. Moreover T can be written as a
convergent series of closed positive currents
+∞
X
(3.8) T = λk [Zk ] + R,
k=1
where [Zk ] is a current of integration over an irreducible analytic set of dimension p,
and R is a residual current with the property that dim Ec (R) < p for every
c > 0. This decomposition is locally and globally unique: the sets Zk are
precisely the p-dimensional components occurring in the sublevel sets Ec (T ), and
λk = minx∈Zk ν(T, x) is the generic Lelong number of T along Zk .
The Lelong number of a plurisubharmonic function w on X can also be
defined by
w(z)
(3.9) ν(w, x) = lim inf ,
z→x log |z − x|

where z = (z1 , . . . , zn ) are local coordinates near x, and | | denotes an arbitrary

norm on Cn . It is well known that ν(w, x) is equal to the Lelong number
ν(T, x) of the associated positive (1, 1)-current T = πi ∂∂w. Accordingly, we set
Ec (w) = Ec (T ).

4. L2 estimates and existence of holomorphic sections

We first state the basic existence theorem of Hörmander for solutions of ∂

equations, in the form that is most convenient to us.

Proposition 4.1. — Suppose that X is a Stein or compact projective

manifold equipped with a Kähler metric ω. Let L be a line bundle with a
hermitian metric associated to singular plurisubharmonic weight functions ψ such
that c(L) ≥ εω for some ε > 0. R For every q ≥ 1 and every (n, q) form v with
values in L such that ∂v = 0 and X |v|2 e−2ψ dVω < +∞, there is a (n, q − 1)-form
u with values in L such
Z that ∂u = v and Z
2 −2ψ 1
|u| e dVω ≤ |v|2 e−2ψ dVω .
X 2πqε X

8
 Here dVω stands for the Kähler volume element ω n /n!, and |u|2 e−2ψ denotes
somewhat abusively the pointwise norm of u(z) at each point z ∈ X, although ψ
is only defined on an open set in X. The operator ∂ is taken in the sense of
distribution theory.

Proof. — The result is standard when X is Stein and L is the trivial bundle
(see [AV 65] and [Hö 66]). In general, there exists a hypersurface H ⊂ X such
that X \ H is Stein and L is trivial over X \ H. We then solve the equation ∂u = v
over X \ H and observe that the solution extends to X thanks to the L2 estimate
(cf. [De 82b], lemma 6.9).
We will also use the concept of multiplier ideal sheaf introduced by A. Nadel
[Na 89]. The main idea actually goes back to the fundamental works of Bombieri
[Bo 70] and H. Skoda [Sk 72a]. Let ϕ be a plurisubharmonic function on X ; to
ϕ is associated the ideal subsheaf I(ϕ) ⊂ OX of germs of holomorphic functions
f ∈ OX,x such that |f |2 e−2ϕ is integrable with respect to the Lebesgue measure
in some local coordinates near x. The zero variety V I(ϕ) is thus the set of points
in a neighborhood of which e−2ϕ is non integrable. This zero variety is closely
related to the Lelong sublevel sets Ec (ϕ). Indeed, if ν(ϕ, x) = γ, the convexity
properties of plurisubharmonic functions show that
ϕ(z) ≤ γ log |z − x| + O(1) at x,
hence there exists a constant C > 0 such that e−2ϕ(z) ≥ C|z − x|−2γ in a
neighborhood of x. We easily infer that
(4.2) ν(ϕ, x) ≥ n + s =⇒ I(ϕ)x ⊂ Ms+1
X,x ,
where MX,x is the maximal ideal of OX,x . In the opposite direction, it is well
known that ν(ϕ, x) < 1 implies the integrability of e−2ϕ in a neighborhood of x
(cf. Skoda [Sk 72a]), that is, I(ϕ)x = OX,x . In particular, the zero variety V I(ϕ)
of I(ϕ) satisfies
(4.3) En (ϕ) ⊂ V I(ϕ) ⊂ E1 (ϕ).

Lemma 4.4 ([Na 89]). — For any plurisubharmonic function ϕ on X, the

sheaf I(ϕ) is a coherent sheaf of ideals over X.

Proof. — Since the result is local, we may assume that X is the unit
ball
R in Cn . Let E be the set of all holomorphic functions f on X such that
X
|f |2 e−2ϕ dλ < +∞. By the strong noetherian property of coherent sheaves, the
set E generates a coherent ideal sheaf J ⊂ OX . It is clear that J ⊂ I(ϕ); in
order to prove the equality, we need only check that Jx + I(ϕ)x ∩ Ms+1
X,x = I(ϕ)x
for every integer s, in view of the Krull lemma. Let f ∈ I(ϕ)x be defined in a
neighborhood V of x and θ be a cut-off function with support in V such that
θ = 1 in a neighborhood of x. We solve the equation ∂u = ∂(θf ) by means of
Hörmander’s L2 estimates 4.1, with L equal to the trivial line bundle and with
the strictly plurisubharmonic weight
ψ(z) = ϕ(z) + (n + s) log |z − x| + |z|2 .

9
 R
We get a solution u such that X |u|2 e−2ϕ |z − x|−2(n+s) dλ < ∞, thus F = θf − u
is holomorphic, F ∈ E and fx − Fx = ux ∈ I(ϕ)x ∩ Ms+1 X,x . This proves our
contention.
Now, suppose that X is a projective n-fold equipped with a Kähler metric ω.
Let L be a line bundle over X with a singular metric of curvature T = c(L) ≥ 0. All
sublevel sets Ec (T ) are algebraic subsets of X, and if ϕ is the weight representing
the metric in an open set Ω ⊂ X, then Ec (ϕ) = Ec (T ) ∩ Ω. The ideal sheaf I(ϕ)
is independent of the choice of the trivialization and so it is the restriction to Ω of
a global coherent sheaf on X which we shall still call I(ϕ) by abuse of notation.
In this context, we have the following interesting vanishing theorem, which can
be seen as a generalization of the Kawamata-Viehweg vanishing theorem [Ka 82],
[Vi 82].

Theorem 4.5 ([Na 89]). — Let L be a line bundle over X with κ(L) = n.
Assume that L is equipped with a singular metric
of weight ϕ such that c(L) ≥ εω
for some ε > 0. Then H q X, O(KX + L) ⊗ I(ϕ) = 0 for all q ≥ 1.

Proof. — Let F q be the sheaf of germs of (n, q)-forms u with values in

L and with measurable coefficients, such that both |u|2 e−2ϕ and |∂u|2 e−2ϕ are
locally integrable. The ∂ operator defines a complex of sheaves (F • , ∂) which is
a resolution of the sheaf O(KX + L) ⊗ I(ϕ): indeed, the kernel of ∂ in degree 0
consists of all germs of holomorphic n-forms with values in L which satisfy the
integrability condition; hence the coefficient function lies in I(ϕ); the exactness in
degree q ≥ 1 follows from proposition 4.1 applied on arbitrary small balls.
Each
sheaf F q is a C ∞ -module, so F • is a fine resolution. Moreover, H q Γ(X, F • ) = 0
for q ≥ 1 by proposition 4.1 applied globally on X. The theorem follows.

Corollary 4.6. — Let L be a big nef line bundle over X. Assume that
L is equipped with a singular metric of weight ϕ such that c(L) ≥ 0 and let
x1 , . . . , xN be isolated points in the zero variety V I(ϕ). Then for every ε > 0,
there is a surjective map
M

H 0 (X, KX + L) −→
−→ O(KX + L)xj ⊗ OX /I((1 − ε)ϕ) xj .
1≤j≤N

Proof. — This result can be seen as a generalization of the Hörmander-

Bombieri-Skoda theorem ([Bo 70], [Sk 72a,75]); it could be proved directly by
using Hörmander’s L2 estimates and cut-off functions. If c(L) ≥ δω for some
δ > 0, we apply theorem 4.5 to obtain the vanishing of the first H 1 group in the
long exact sequence of cohomology associated to
0 −→ I(ϕ) −→ OX −→ OX /I(ϕ) −→ 0,
twisted by O(KX + L).
The asserted surjectivity property follows immediately;
as I(ϕ) ⊂ I (1 − ε)ϕ , we see in that case that we can even take ε = 0 and drop
the nef assumption on L. If c(L) ≥ 0 merely, we try to modify the metric so as to
obtain a positive lower bound for the curvature. By (2.7) there is a singular metric

10
on L associated to a weight ψ with πi ∂∂ψ ≥ δω, δ > 0, and with a singularity of
ψ so small that e−2ψ ∈ L1loc . Replace the metric on L by the metric associated to
the weight ϕε = (1 − ε)ϕ + εψ. Then e−2ϕε = (e−2ϕ )1−ε (e−2ψ )ε is integrable on
any open set where e−2ϕ is integrable, so V I(ϕε ) ⊂ V I(ϕ)

and the points xj are
still isolated in V I(ϕε ). Moreover I(ϕε ) ⊂ I (1 − ε)ϕ , for ψ is locally bounded
above, and c(L)ε = πi ∂∂ϕε ≥ εδ ω. We are thus reduced to the first case.

Example 4.7. — Suppose that ν(ϕ, x) > n + s and that x is an isolated

point in E1 (ϕ). Then I((1 − ε)ϕ)x ⊂ Ms+1X,x for ε small enough, and x is isolated
in V I(ϕ) by (4.2), (4.3). We infer that H 0 (X, KX + L) −→ −→ Jxs (KX + L) is
surjective onto s-jets of sections at x.

Example 4.8. — Suppose that (z1 , . . . , zn ) are local coordinates centered

at x and that
ϕ(z) ≤ γ log(|z1 | + . . . + |zn−1 | + |zn |2 ) + O(1), γ > n.
Then I((1 − ε)ϕ) ⊂ (z1 , . . . , zn−1 , zn2 )
for ε small. To check this, observe that for
any δ > 0 the Parseval-Bessel formula gives
Z P 2 Z P
aα z α dλ(z) |aα |2 |z α |2 dλ(z)
2 2n
= 2 2n
;
|z|<δ (|z1 | + . . . + |zn−1 | + |zn | ) |z|<δ (|z1 | + . . . + |zn−1 | + |zn | )

the integral is divergent unless the coefficients a0 and a(0,...,0,1) vanish. Indeed,
using polar coordinates zn = reiθ and setting z ′ = (z1 , . . . , zn−1 ), we get
Z Z Z δ/2
|zn |2 dλ(z) ′ r3 dr
≥ 2π dλ(z )
2 2n
|z|<δ (|z1 | + . . . + |zn−1 | + |zn | ) |z ′ |<δ/2 0 (|z ′ | + r2 )2n
Z
dλ(z ′ )
≥C ′ 2n−2
= +∞.
|z ′ |<δ/2 |z |

Thus, if x is isolated in E1 (ϕ), we are able to prescribe the value of the section
at x and its derivative ∂/∂zn along the direction z ′ = 0.

Remark 4.9. — More generally, it is interesting to consider logarithmic

poles of the form  X 
ϕ(z) = γ log |gj (z)| + O(1),
1≤j≤N

where J = (g1 , . . . , gN ) ⊂ MX,x is an arbitrary ideal with isolated zero {x}.

However, in this case, we do not know what is the general rule relating the ideal
I(γ log |g|)x to the ideal J . Observe thatPI(γ log |g|)x only depends on the integral
closure J = {germs f such that |f | ≤ C |gj |}. It is almost obvious by definition
that I(ϕ) itself is always integrally closed. Let Je = (e gn ) ⊂ J be the
g1 , . . . , e
ideal generated by n generic linear combinations g
e k of g ,
1P . . . , g e
N . Then J and J
P
have the same integral closure and we have gk | ≥ C ′ |gj | with some C ′ > 0 ;
|e
indeed these ideals have the same multiplicity by a result of Serre [Se 57], and this
implies the equality of their integral closures thanks to a result of D. Rees [Ree 61].

11
 g|)x associated to Je thus coincides with I(γ log |g|)x . We see
The ideal I(γ log |e
that there is no loss of generality considering only ideals generated by exactly n
generators (as we shall do in §§ 6,7). Finally, the proof of the Briançon-Skoda
theorem [BSk 74] shows that
(4.10) g |)x ⊂ Je ⊂ J
I(γ log |g|)x = I(γ log |e when γ > n.
In fact, (4.10) is a straightforward consequence of Skoda’s division theorem
[Sk 72b], applied to the elements of I(γ log |e
g|).

5. Aubin-Calabi-Yau theorem and convexity inequalities

The above results can be applied to construct sections of a given line bundle,
provided we are able to produce singular metrics with logarithmic poles . For this,
we use in several essential ways the well-known theorem of Aubin-Yau on the
Calabi conjecture. What we need is the following existence result about solutions
of Monge-Ampère equations.

Lemma 5.1 [Yau 78], see also [Au 78]). — Let X be a compact complex
n-dimensional manifold Rwith a Rsmooth Kähler metric ω. Then for any smooth
volume form f > 0 with X f = X ω n , there exists a Kähler metric ω
e in the same
en = f .
Kähler class as ω such that ω

The method for constructing singular metrics from the Aubin-Calabi-Yau

theorem will be explained in detail in § 6. Before, we need a useful convexity
inequality due to Hovanski [Hov 79] and Teissier [Te 79,82], which is a natural
generalization of the usual Hodge index theorem for surfaces. This inequality
is reproved along similar lines in [BBS 89], where it is applied to the study of
projective n-folds of log-general type. For the sake of completeness, we include
here a different and slightly simpler proof, based on Yau’s theorem 5.1 instead of
the Hodge index theorem. Our proof also has the (relatively minor) advantage of
working over arbitrary Kähler manifolds.

Proposition 5.2. — In any dimension n :

(a) if α1 , . . . , αn are semipositive (1, 1)-forms on Cn , then
α1 ∧ α2 ∧ . . . ∧ αn ≥ (αn1 )1/n (αn2 )1/n . . . (αnn )1/n .
(b) if u1 , . . . , un are semipositive cohomology classes of type (1, 1) on a Kähler
manifold X of dimension n, then
u1 · u2 · · · un ≥ (un1 )1/n (un2 )1/n . . . (unn )1/n .
By a semipositive cohomology class of type (1, 1), we mean a class in the
closed convex cone of H 1,1 (X, IR) generated by Kähler classes. For instance,
inequality (b) can be applied to uj = c1 (Lj ) when L1 , . . . , Ln are nef line bundles
over a projective manifold.

Proof. — Observe that (a) is a pointwise inequality between (n, n)-forms

whereas (b) is an inequality of a global nature for the cup product intersection

12
form. We first show that (a) holds when only two of the forms αj are distinct,
namely that
αp ∧ β n−p ≥ (αn )p/n (β n )(n−p)/n
for all α, β ≥ 0. By a density argument, we may suppose α, β > 0. Then there is
a simultaneous orthogonal basis in which
X X
α=i λj dzj ∧ dz j , β=i dzj ∧ dz j
1≤j≤n 1≤j≤n

with λj > 0, and (a) is equivalent to

X
p!(n − p)! λj1 . . . λjp ≥ n! (λ1 . . . λn )p/n .
j1 <...<jp

As both sides are homogeneous of degree p in (λj ), we may assume λ1 . . . λn = 1.

Then our inequality follows from the inequality between the arithmetic and
geometric means of the numbers λj1 . . . λjp . Next, we show that statements (a)
and (b) are equivalent in any dimension n.
(a) =⇒ (b). By density, we may suppose
R that u1 , . . . , un are Kähler classes. Fix
a positive (n, n) form f such that X f = 1. Then lemma 5.1 implies that there is
a Kähler metric αj representing uj such that αnj = unj f . Inequality (a) combined
with an integration over X yields
Z Z
u1 · · · un = α1 ∧ . . . ∧ αn ≥ (un1 )1/n . . . (unn )1/n f.
X X
(b) =⇒ (a). The forms α1 , . . . , αn can be considered as constant (1, 1)-forms on
any complex torus X = Cn /Γ. Inequality (b) applied to the associated cohomology
classes uj ∈ H 1,1 (X, IR) is then equivalent to (a).
Finally we prove (a) by induction on n, assuming the result already
proved Pin dimension n − 1. We may suppose that αn is positive definite, say
αn = i dzj ∧ dz j in a suitable basis. Denote by u1 , . . . , un the associated
cohomology classes on the abelian variety X = Cn /ZZ[i]n . Then un has integral
periods, so some multiple of un is the first Chern class of a very ample line bundle
O(D) where D is a smooth irreducible divisor in X. Without loss of generality,
we may suppose un = c1 (O(D)). Thus
u1 · · · un−1 · un = u1`| D · · · un−1`| D
and by the induction hypothesis we get
u1 · · · un ≥ (u1n−1 n−1
)1/(n−1) . . . (un−1 )1/(n−1) .
`| D `| D
However ujn−1 = ujn−1 · un ≥ (unj )(n−1)/n (unn )1/n , since (a) and (b) are equivalent
`| D
and (a) is already proved in the case of two forms. (b) follows in dimension n, and
therefore (a) holds in Cn .

Remark 5.3. — In case αj (resp. uj ) are positive definite, the equality

holds in 5.2 (a,b) if and only if α1 , . . . , αn (resp. u1 , . . . , un ) are proportional.

13
In our inductive proof, the restriction morphism H 1,1 (X, IR) −→ H 1,1 (D, IR) is
injective for n ≥ 3 by the hard Lefschetz theorem, hence it is enough to consider
the case of αp ∧ β n−p . The equality between arithmetic and geometric means
occurs only when all numbers λj1 . . . λjp are equal, so all λj must be equal and
α = λ1 β, as desired. More generally, there is an inequality
α1 ∧ . . . ∧ αp ∧ β1 ∧ . . . ∧ βn−p ≥
(5.4) ≥ (αp1 ∧ β1 ∧ . . . ∧ βn−p )1/p . . . (αpp ∧ β1 ∧ . . . ∧ βn−p )1/p
for all (1, 1)-forms αj , βk ≥ 0. Once again, inequality (5.4) is easier to be proved
with cohomology classes rather than forms. By a density argument, we may
suppose that all forms βj are positive definite and have coefficients in Q[i]. Let
u1 , . . . , up be the cohomology classes of type (1, 1) associated to α1 , . . . , αp on
X = Cn /ZZ[i]n . The cohomology class of β1 is a rational multiple of the first
Chern class of a very ample line bundle O(Y1 ), where Y1 is a smooth irreducible
divisor in X, that of β2`| Y1 is a multiple of such a divisor Y2 in Y1 , and by induction
the cohomology class of β1 ∧ . . . ∧ βn−p is equal to a multiple of the cohomology
class of a connected p-dimensional submanifold Y ⊂ X. Then (5.4) is equivalent
to the already known inequality
u1`| Y · · · up`| Y ≥ (up1| Y )1/p . . . (upp| Y )1/p .
` `

6. Mass concentration in the Monge-Ampère equation

In this crucial section, we show how the Aubin-Calabi-Yau theorem can be

applied to construct singular metrics on ample (or more generally big and nef)
line bundles. We first suppose that L is an ample line bundle over a projective
n-fold X and that L is equipped with a smooth metric of positive curvature. We
i
consider the Kähler metric ω = 2π c(L). Any form ω e in the Kähler class of ω can
i
be written as ω e = ω + π ∂∂ψ, i.e. is the curvature form of L after multiplication of
the original metric by a smooth weight function e−ψ . By lemma 5.1, the Monge-
Ampère equation
 i n
(6.1) ω + ∂∂ψ = f
π
R
can be solved for ψ, whenever f is a smooth (n, n)-form with f > 0 and X f = Ln .
In order to produce logarithmic poles at given points x1 , . . . , xN ∈ X, the main
idea is to let f converge to a Dirac measure at xj ; then ω
e will be shown to converge
to a closed positive (1, 1)-current with non zero Lelong number at xj .
Let (z1 , . . . , zn ) be local coordinates centered at xj , defined on some
neighborhood Vj ≃ {|z| < Rj }. Let gj = (gj,1 , . . . , gj,n ) be arbitrary holomorphic
functions on Vj such that gj−1 (0) = {xj }, and let
 X 1/2
(6.2) log |gj | = log |gj,k |2 .
1≤k≤n

14
Then log |gj | has an isolated logarithmic pole at xj and ( πi ∂∂ log |gj |)n = ρj δxj ,
where ρj is the degree of the covering map gj : (Cn , xj ) −→ (Cn , 0). Indeed
∂∂ log |gj | = gj⋆ ∂∂ log |z| has rank (n − 1) on Vj \ {xj }, and formula (3.5) with
χ(t) = e2t gives
Z i n Z
1
∂∂ log |gj | = g ⋆ (i∂∂|w|2 )n
|gj (z)|<r π (2πr2 )n |gj (z)|<r j
Z
ρj
= (i∂∂|w|2 )n = ρj
(2πr2 )n |w|<r
for every r > 0 small enough. Now, let χ : IR → IR be a smooth convex increasing
function such that χ(t) = t for t ≥ 0 and χ(t) = −1/2 for t ≤ −1. We set
i

(6.3) αj,ε = ∂∂ χ(log |gj |/ε) .
π
Then αj,ε is a smooth positive (1, 1)-form, and αj,ε = πi ∂∂ log |gj | over the set of
points z ∈ Vj such that |gj (z)| > ε. It follows that αnj,ε has support in the compact
set |gj (z)| ≤ ε, and Stokes’ formula gives
Z Z  n
n i
(6.4) αj,ε = ∂∂ log |gj | = ρj .
Vj Vj π

Hence αnj,ε converges weakly to the Dirac

P measure ρj δxj as ε tends to 0. For all
positive numbers τj > 0 such that σ = ρj τjn < Ln , lemma 5.1 gives a solution
of the Monge-Ampère equation
X  σ  i
(6.5) ωεn = τjn αnj,ε + 1 − n ω n with ωε = ω + ∂∂ψε ,
L π
1≤j≤N

since the right-hand side of the first equation is > 0 and has the correct integral
value Ln over X. The solution ψε is merely determined up to a constant. If
Rγ is an narbitrary Kähler metric on X, we can normalize ψε in such a way that
ψ γ = 0.
X ε

Lemma 6.6. — There is a sequence εν converging to zero such that ψεν has
a limit ψ in L1 (X) and such that the sequence of (1, 1)-forms ωεν converges weakly
towards a closed positive current T of type (1, 1). Moreover, the cohomology class
of T is equal to c1 (L) and T = ω + πi ∂∂ψ.
R
Proof. — The integral X ωε ∧ γ n−1 = L · {γ}n−1 remains bounded, so we
can find a sequence εν converging to zero such that the subsequence ωεν converges
weakly towards a closed positive current T of bidegree (1, 1). The cohomology
class of a current is continuous with respect to the weak topology (this can be
seen by Poincaré duality). The cohomology class of T is thus equal to c1 (L). The
function ψε satisfies the equation π1 ∆ψε = trγ (ωε − ω) where ∆ is the Laplace
operator associated to γ. Our normalization of ψε implies
ψε = π G trγ (ωε − ω),

15
where G is the Green operator of ∆. As G is a compact operator from the Banach
space of bounded Borel measures into L1 (X), we infer that some subsequence
(ψεν ) of our initial subsequence converges to a limit ψ in L1 (X). By the weak
continuity of ∂∂, we get T = lim(ω + πi ∂∂ψεν ) = ω + πi ∂∂ψ.

Let Ω ⊂ X be an open coordinate patch such that L is trivial on a

neighborhood of Ω, and let e−h be the weight representing the initial hermitian
metric on L`| Ω . Then πi ∂∂h = ω and πi ∂∂(h+ψε ) = ωε , so the function ϕε = h+ψε
defines a plurisubharmonic weight on L`| Ω , as well as its limit ϕ = h + ψ. By the
continuity of G, we also infer from the proof of lemma 6.6 that the family (ψε )
is bounded in L1 (X). The usual properties of subharmonic functions then show
that there is a uniform constant C such that ϕε ≤ C on Ω. We use this and
equation (6.5) to prove that the limit ϕ has logarithmic poles at all points xj ∈ Ω,
thanks to Bedford and Taylor’s maximum principle for solutions of Monge-Ampère
equations [BT 76]:

Lemma 6.7. — Let u, v be smooth (or continuous) plurisubharmonic

functions on Ω, where Ω is a bounded open set in Cn . If
u`| ∂Ω ≥ v`| ∂Ω and (i∂∂u)n ≤ (i∂∂v)n on Ω,
then u ≥ v on Ω.

In the application of lemma 6.7, we suppose that Ω is a neighborhood of xj

and take

u = τj χ(log |gj |/ε) + log ε + C1 , v = ϕε ,
where C1 is a large constant. Then for ε > 0 small enough
u`| ∂Ω = τj log |gj | + C1 , v`| ∂Ω ≤ C,
i n i n
∂∂v = ωεn ≥ τjn αnj,ε = ∂∂u on Ω.
π π
For C1 sufficiently large, we infer u ≥ v on Ω, hence
ϕε ≤ τj log(|gj | + ε) + C2 on Ω.

Corollary 6.8. — The plurisubharmonic weight ϕ = h + ψ on L`| Ω

associated to the limit function ψ = lim ψεν satisfies πi ∂∂ϕ = T . Moreover, ϕ has
logarithmic poles at all points xj ∈ Ω and
ϕ(z) ≤ τj log |gj (z)| + O(1) at xj .

Case of a big nef line bundle. All our arguments were developed under
the assumption that L is ample, but if L is only nef and big, we can proceed in
the following way. Let A be a fixed ample line bundle with smooth curvature form
γ = c(A) > 0. As mL + A is ample for any m ≥ 1, by 5.1 there exists a smooth
1
hermitian metric on L depending on m, such that ωm = c(L)m + m c(A) > 0 and
1
n (L + m A)n n
(6.9) ωm = n
γ .
A
16
However, a priori we cannot control the asymptotic behaviour of ωm when m
tends to infinity, so we introduce the sequence of non necessarily positive (1, 1)-
′ 1
forms ωm = c(L)1 + m c(A) ∈ {ωm }, which is uniformly bounded in C ∞ (X) and
converges to c(L)1 . Then we solve the Monge-Ampère equation
X  σ 
n
(6.10) ωm,ε = τjn αnj,ε + 1 − 1 n
n
ωm
1≤j≤N
(L + m A)

with
R ωm,ε = ωm ′
+ πi ∂∂ψm,ε and some smooth function ψm,ε such that
n
X ψm,ε γ = 0 ; this is again possible by Yau’s theorem 5.1. The numerical con-
dition needed on σ to solve (6.10) is obviously satisfied for all m if we suppose
X  1 n
σ= ρj τjn < Ln < L + A .
m
The same arguments as before show that there exist a convergent subsequence
limν→+∞ ψmν ,εν = ψ in L1 (X) and a closed positive (1, 1)-current T = lim ωmν ,εν
= c(L)1 + πi ∂∂ψ ∈ c1 (L) such that Corollary 6.8 is still valid; in this case, h is
taken to be the weight function corresponding to c(L)1 . Everything thus works as
in the ample case.

7. Choice of the logarithmic singularities

Let us assume (with the notation of § 6) that each point xj is isolated in

E1 (ϕ). Then we conclude by (4.3) and Corollary 4.6 that there is a surjective map
M

(7.1) H 0 (X, KX + L) −→
−→ O(KX + L)xj ⊗ OX /I((1 − ε)ϕ) xj .
1≤j≤N

However, finding sufficient conditions ensuring that xj is isolated in E1 (ϕ) = E1 (T )

is a harder question. Therefore, we postpone this task to the next section and
explain instead how to choose the logarithmic poles log |gj | and the constants τj
to obtain specified ideals and jets of sections at each point xj .
Suppose that an ideal Jj ⊂ MX,xj is given at xj , in other words, that
we are given a 0-dimensional subscheme (Ξ, OΞ ) with Ξ = {x1 , . . . , xN } and
OΞ,xj = OX,xj /Jj . We want to find sufficient conditions for the surjectivity of the
restriction map

M
H 0 (X, KX + L) −→ H 0 Ξ, OΞ (KX + L) = O(KX + L)xj ⊗ OX,xj /Ij .
1≤j≤N

By (7.1), we need only find a germ of map gj : (X, xj ) → (Cn , 0) and a constant
τj,0 such that I(τj,0 log |g
j |) ⊂ Jj . For τj > τj,0 and ε small enough, corollary 6.8
then P implies I (1 − ε)ϕ ⊂ Jj . Thus we have to choose σ slightly larger than
n
σ0 = ρj τj,0 where ρj is the degree of the covering map gj ; this is possible only
n
if L > σ0 . Let us discuss some specific cases.
Spannedness. To obtain that KX + L spans at x ∈ X, we consider a
single point x1 = x and take J1 = MX,x , g1 (z) = (z1 , . . . , zn ), τ1,0 = n and
n
σ0 = τ1,0 = nn . Then I(τ1,0 log |g1 |) ⊂ MX,x , as desired.

17
 Separation of points. To obtain the separation of two points x1 6= x2
in X by sections of KX + L, we make the same choices as above at x1 , x2 and
n n
get σ0 = τ1,0 + τ2,0 = 2nn . If x1 , x2 are “infinitely near” in some direction
ξ ∈ T X, we choose coordinates (z1 , . . . , zn ) centered at x = x1 = x2 so that
∂/∂zn = ξ, and we set J1 = (z1 , . . . , zn−1 , zn2 ). By example 4.8, we can choose
g1 (z) = (z1 , . . . , zn−1 , zn2 ) and τ1,0 = n. Then the degree of g1 is ρ1 = 2 and we
n
find again σ0 = ρ1 τ1,0 = 2nn .
Generation of s-jets. Instead of just considering jets at one point, we
wish to look at several points simultaneously which may come into coincidence.
Such a concern appears also in the work of Beltrametti-Sommese [BSo 90], where
an extensive study of the surface case is made. The relevant definition is as follows.

Definition 7.2. — L Wesj

say that L generates s-jets on a given subset
Ξ ⊂ X if H 0 (X, L) −→ −→ xj L is onto for any choice of points x1 , . . . , xN ∈ Ξ
JP
and integers s1 , . . . , sN with (sj + 1) = s + 1. We say that L is s-jet ample if
the above property holds for Ξ = X.

With this terminology, L is 0-jet ample if and only if L is spanned and 1-jet
ample if and only if L is very ample. In order that KX + L generates s-jets on Ξ,
we take x1 , . . . , xN ∈ P
Ξ arbitrary, gj (z) = (z1 , . . . , zn ) at each xj andPτj,0 = n + sj .
Therefore σ0 = max (n + sj )n over all decompositions s + 1 = (sj + 1). In
fact, if we set tj = sj + 1, the following lemma gives σ0 = (n + s)n , that is, the
maximum is reached when only one point occurs.

Lemma 7.3. — Let t1 , . . . , tN ∈ [1, +∞[. Then

X  X n
(n − 1 + tj )n ≤ n − 1 + tj .
1≤j≤N 1≤j≤N

Proof. — The right hand side is a polynomial with nonnegative coefficients

and the coefficient of a monomial tkj involving exactly one variable is the same as
in the left hand side (however, the constant term is smaller). Thus the difference
is increasing in all variables and we need only consider the case t1 = . . . = tN = 1.
This case follows from the obvious inequality
 
n n−1
nn N = nn + n (N − 1) ≤ (n + N − 1)n .
1

Corollary 7.4. — Let L be a big nef line bundle. A sufficient condition

for spannedness (resp. separation of points, s-jet ampleness) of KX + L on a given
set Ξ is Ln > σ0 with
σ0 = nn , resp. σ0 = 2nn , resp. σ0 = (n + s)n ,
′
provided that the solution ωε of (6.5) (resp. the solution ωm,ε = ωm + πi ∂∂ψm,ε of
(6.10)), always has a subsequence converging to a current T for which all points
x ∈ Ξ ∩ E1 (T ) are isolated in E1 (T ).

18
 Case of an arbitrary 0-dimensional subscheme. Let Jj = (hj,k )1≤k≤N
be an arbitrary ideal in OX,xj with V Jj = {xj }. By remark 4.9, we can take
gj = (gj,1 , . . . , gj,n ) to be n generic elements of Ij and τj,0 = n. Indeed, property
P shows that I(γ log |gj |) ⊂ Jj for γ > n. In this case, we find
(4.10) then
σ0 = nn ρj . Unfortunately, this value is in general very far from being optimal:
for instance, we would get σ0 = nn (s + 1)n instead of (n + s)n in the case of s-jets.
If (Ξ, OΞ ) is a local complete intersection, that is, if each Jj has N = n generators,
we simply take gj = (hj,1 , . . . , hj,n ). Thus we obtain ρj = dim OX,xj /Jj and

(7.5) σ0 = nn h0 (Ξ, OΞ ).

8. Upper bound for the 1-codimensional polar components

The goal of this section is to give a rather simple derivation of numerical

conditions ensuring that codim(E1 (T ), x) ≥ 2 at a given point x. In particular, we
will obtain a criterion for very ample line bundles over surfaces. Although these
results are only formal consequences of those obtained in the next two sections,
we feel preferable to indicate first the basic ideas in a simple case.
Let L again denote an ample line bundle over a projective algebraic
manifold X and keep the same notation as in § 6. Siu’s decomposition formula
(3.8) applied to T = lim ωεν gives
+∞
X
(8.1) T = λk [Hk ] + R,
k=1

where [Hk ] is the current of integration over an irreducible hypersurface Hk and

codim Ec (R) ≥ 2 for every c > 0. As we would like E1 (T ) to have isolated points
at xj , a difficulty may come from the singular points of high multiplicities in the
hypersurfaces Hk . We thus need to find upper bounds for the coefficients λk . The
convexity inequality 5.2 can be used for this purpose to obtain a lower bound of
the mass of R :

Proposition 8.2. — We have

+∞
X

λk Ln−1 · Hk ≤ σ1 = 1 − (1 − σ/Ln )1/n Ln .
k=1

R R
Proof. — As X T ∧ ω n−1 = Ln and X [Hk ] ∧ ω n−1 = Ln−1 · Hk , we need
only prove that
Z
(8.3) R ∧ ω n−1 ≥ (1 − σ/Ln )1/n Ln .
X

Let θ be a smoothR function on X such that 0 ≤ θ ≤ 1, θ = 1 in a neighborhood of

S n
1≤k≤N Hk and X θ ω < ε0 , where ε0 > 0 is an arbitrarily small number. This

19
 S
is possible because 1≤k≤N Hk is a closed set of zero Lebesgue measure in X.
Then Z Z
n−1
(1 − θ) T ∧ ω ≥ lim inf (1 − θ) ωε ∧ ω n−1
X ε→0 X
Z
≥ lim inf (1 − θ) (ωεn )1/n (ω n )1−1/n
ε→0 X
Z
≥ 1 − σ/Ln )1/n (1 − θ) ω n
X

by the convexity
P inequality and equation (6.5). By our choice of θ we have
(1 − θ) T ≤ k>N λk [Hk ] + R , so
Z X 
λk [Hk ] + R ∧ ω n−1 ≥ 1 − σ/Ln )1/n (Ln − ε0 ).
X k>N

Since N and ε0 were arbitrary, we get the expected inequality (8.3).

If L is big and nef, the same result can be obtained by replacing ω with ωk0
and ωε with ωk,ε in the above inequalities (ε → 0, k → +∞) and by letting k0
tend to +∞ at the end. Now suppose that for any hypersurface H in X passing
through a given point x we have


(8.4) Ln−1 · H > 1 − (1 − σ0 /Ln )1/n Ln .
We can choose σ = σ0 + ε such that inequality (8.4) is still valid with σ instead
of σ0 , and then all hypersurfaces Hk passing through x have coefficients λk < 1 in
(8.1). Thus
[ [ [
[
E1 (T ) ⊂ E1 (R) ∪ Hk,sing ∪ (Hk ∩ Hl ) ∪ Hk ∩ E1−λk (R) ∪ Hk ,
k k6=l λk <1 λk ≥1

because the contribution of [Hk ] to the Lelong number of T is equal to 1 at a

regular point. As all terms in the union have codimension ≥ 2 except the
last
ones which do not contain x, condition (8.4) ensures that codim E1 (T ), x ≥ 2.
In the case of surfaces, we can therefore apply corollary 7.4 to obtain:

Corollary 8.5. — Let X be a smooth algebraic surface, and let L be a

big nef line bundle over X. Then on a given subset Ξ ⊂ X
KX + L is spanned separates points generates s-jets
L2 > 4 8 9 12 (2 + s)2
when
∀C, L·C > 2 6 5 4 2 + 3s + s2

for all curves C ⊂ X intersecting Ξ. In particular, if L is ample, KX + mL is

always globally spanned for m ≥ 3 and very ample for m ≥ 5.

Proof. — For s-jets, we have σ0 = (2 + s)2 , so we find the condition

L2 > (2 + s)2 , L · C > 1 − (1 − (2 + s)2 /L2 )1/2 L2 .

20
The last constant decreases with L2 and is thus at most equal to the value obtained
when L2 = (2 + s)2 + 1 ; its integral part is precisely 2 + 3s + s2 .
The above lower bounds on L2 are sharp but not those for L · C. Reider’s
method shows in fact that KX + mL is very ample as soon as m ≥ 4. In the

higher dimensional case, a major difficulty is to ensure that the germs E1 (T ), xj
do not contain any analytic set of dimension 1, 2, . . . , n − 2. This cannot be done
without considering “self-intersections” of T and prescribing suitable bounds for
all intermediate intersection numbers Lp · Y .

9. Approximation of closed positive (1,1)-currents by divisors

Let L be a line bundle with c1 (L) ∈ Γ+ and let T = c(L) ≥ 0 be the

curvature current of some singular metric on L. Our goal is to approximate T in the
weak topology by divisors which have roughly the same Lelong numbers as T . The
existence of weak approximations by divisors has already been proved in [Le 72]
for currents defined on a pseudoconvex open set Ω ⊂ Cn with H 2 (Ω, IR) = 0, and
in [De 82c] in the situation considered here. However, the result of [De 82c] is less
precise than what we actually need and moreover the proof contains a small gap;
a complete proof will therefore be included here.

Proposition 9.1. — For any T = c(L) ≥ 0 and any ample line bundle
F , there is a sequence of non zero sections hs ∈ H 0 (X, ps F + qs L) with ps , qs > 0,
lim qs = +∞ and lim ps /qs = 0, such that the divisors Ds = q1s div(hs ) satisfy
T = lim Ds in the weak topology and supx∈X |ν(Ds , x)− ν(T, x)| → 0 as s → +∞.

Remark 9.2. — The proof will actually show, with very slight modifica-
tions, that Prop. 9.1 also holds when X is a Stein manifold and L is an arbitrary
holomorphic line bundle. The last assertion concerning Lelong T numbers implies
that there is a sequence εs > 0 converging to 0 such that Ec (T ) = s≥1 Ec−εs (Ds ).
When D is an effective divisor, given locally as the divisor of a holomorphic func-
tion h, then Ec (D) is the set of points x ∈ X such that the derivatives h(α) (x) = 0
for all multi-indices α with |α| < c. This gives a new proof of Siu’s result
[Siu 74] that Ec (T ) is an analytic set, at least in the case of bidegree (1, 1)-currents
(in fact the case of an arbitrary bidegree is easily reduced to the (1, 1) case by a
standard argument due to P. Lelong). Proposition 9.1 is therefore already non
trivial locally.

Proof. — We first use Hörmander’s L2 estimates to construct a suitable

family of holomorphic sections and combine this with some ideas of [Le 72] in
a second step. Select a smooth metric with positive curvature on F , choose
ω = c(F ) > 0 as a Kähler metric on X and fix some large integer m (how large m
must be will be specified later). For all s ≥ 1 we define
1
ws (z) = sup log ||fj (z)||,
1≤j≤N s

21
where (f1 , . . . , fN ) is an orthonormal
R basis of the space of sections of O(mF + sL)
with finite global L2 norm X ||f ||2 dVω . Let eF and eL be non vanishing
holomorphic sections of F and L on a trivializing open set Ω, and let e−ψ = ||eF ||,
e−ϕ = ||eL || be the corresponding weights. If f is a section of O(mF + sL)
and if we still denote by f the associated complex valued function on Ω with
−mψ(z)−sϕ(z)
respect to the holomorphic frame em s
F ⊗ eL , we have ||f (z)|| = |f (z)|e ;
here ϕ is plurisubharmonic, ψ is smooth and strictly plurisubharmonic, and
T = πi ∂∂ϕ, ω = πi ∂∂ψ. In Ω, we can write
1 m
ws (z) = sup log |fj (z)| − ϕ(z) − ψ(z).
1≤j≤N s s

In particular Ts := πi ∂∂ws + T + m s ω is a closed positive current belonging to the

cohomology class c1 (L) + m s c 1 (F ).
Step 1. We check that Ts converges to T as s tends to +∞ and that Ts
satisfies the inequalities
n
ν(T, x) − ≤ ν(Ts , x) ≤ ν(T, x)
s
at every point x ∈ X. Note that Ts is defined on Ω by Ts = πi ∂∂vs,Ω with
Z
1
vs,Ω (z) = sup log |fj (z)|, |fj |2 e−2mψ−2sϕ dVω ≤ 1.
1≤j≤N s Ω

We suppose here that Ω is a coordinate open set with analytic coordinates

(z1 , . . . , zn ). Take z ∈ Ω′ ⊂⊂ Ω and r ≤ r0 = 21 d(Ω′ , ∂Ω). By the L2 estimate and
the mean value inequality for subharmonic functions, we obtain
Z
C1 C2
|fj (z)|2 ≤ 2n |fj (ζ)|2 dλ(ζ) ≤ 2n sup e2sϕ(ζ)
r |ζ−z|<r r |ζ−z|<r

with constants C1 , C2 independent of s and r (the smooth function ψ is bounded

on any compact subset of Ω). Hence we infer
1 C2
(9.3) vs,Ω (z) ≤ sup ϕ(ζ) + log 2n .
|ζ−z|<r 2s r

If we choose for example r = 1/s and use the upper semi-continuity of ϕ,

we infer lim sups→+∞ vs,Ω ≤ ϕ. Moreover, if γ = ν(ϕ, x) = ν(T, x), then
ϕ(ζ) ≤ γ log |ζ − x| + O(1) near x. By taking r = |z − x| in (9.3), we find
n  n
vs,Ω (z) ≤ sup ϕ(ζ) − log r + O(1) ≤ γ − log |z − x| + O(1),
|ζ−x|<2r s s
 n n
ν(Ts , x) = ν(vs,Ω , x) ≥ γ − ≥ ν(T, x) − .
s + s
In the opposite direction, the inequalities require deeper arguments since we
actually have to construct sections in H 0 (X, mF + sL). Assume that Ω is chosen
isomorphic to a bounded pseudoconvex open set in Cn . By the Ohsawa-Takegoshi

22
L2 extension theorem [Oh 88], for every point x ∈ Ω, there is a holomorphic
function g on Ω such that g(x) = esϕ(x) and
Z
|g(z)|2 e−2sϕ(z) dλ(z) ≤ C3 ,
Ω

where C3 depends only on n and diam (Ω). For x ∈ Ω′ , set

σ(z) = θ |z − x|/r g(z) eF (z)m ⊗ eL (z)s , r = min 1, 2−1 d(Ω′ , ∂Ω) ,
where θ : IR → [0, 1] is a cut-off function such that θ(t) = 1 for t < 1/2 and
θ(t) = 0 for t ≥ 1. We solve the global equation ∂u = v on X with v = ∂σ, after
multiplication of the metric of mF + sL with the weight


e−2nρx (z) , ρx (z) = θ |z − x|/r log |z − x| ≤ 0.
The (0, 1)-form v can be considered as a (n, 1)-form with values in the line bundle
O(−KX + mF + sL) and the resulting curvature form of this bundle is
i
Ricci(ω) + mω + sT + n ∂∂ρx .
π
Here the first two summands are smooth, i∂∂ρx is smooth on X \ {x} and ≥ 0 on
B(x, r/2), and T is a positive current. Hence by choosing m large enough, we can
suppose that this curvature form is ≥ ω, uniformly for x ∈ Ω′ . By proposition 4.1,
we get a solution u on X such that
Z Z
||u||2 e−2nρx dVω ≤ C4 |g|2 e−2mψ−2sϕ−2nρx dVω ≤ C5 ;
X r/2<|z−x|<r

to get the estimate, we observe that v has support in the corona r/2 < |z − x| < r
and that ρx is bounded there. Thanks to the logarithmic pole of ρx , we infer that
u(x) = 0. Moreover
Z Z
||σ||2 dVω ≤ |g|2 e−2mψ−2sϕ dVω ≤ C6 ,
Ω Ω′ +B(0,r/2)
R
hence f = σ − u ∈ H 0 (X, mF + sL) satisfies X ||f ||2 dVω ≤ C7 and

||f (x)|| = ||σ(x)|| = ||g(x)|| ||eF (x)||m ||eL (x)||s = ||eF (x)||m = e−mψ(x) .
P P
In our orthonormal basis (fj ), we can write f = λj fj with |λj |2 ≤ C7 .
Therefore
X p
e−mψ(x) = ||f (x)|| ≤ |λj | sup ||fj (x)|| ≤ C7 N esws (x) ,
1 1 
ws (x) ≥ log(C7 N )−1/2 ||f (x)|| ≥ − log(C7 N )1/2 + m ψ(x)
s s
where N = dim H 0 (X, mF + sL) = O(sn ). By adding ϕ + m s ψ, we get
vs,Ω ≥ ϕ − C8 s−1 log s. Thus lims→+∞ vs,Ω = ϕ everywhere, Ts = πi ∂∂vs,Ω
converges weakly to T = πi ∂∂ϕ, and
ν(Ts , x) = ν(vs,Ω , x) ≤ ν(ϕ, x) = ν(T, x).

23
Note that ν(vs,Ω , x) = 1s min ordx (fj ) where ordx (fj ) is the vanishing order of fj
at x, so our initial lower bound for ν(Ts , x) combined with the last inequality gives
n 1
(9.4) ν(T, x) − ≤ min ordx (fj ) ≤ ν(T, x).
s s
Step 2: Construction of the divisors Ds .
Select sections (g1 , . . . , gN ) ∈ H 0 (X, m0 F ) with m0 so large that m0 F is
very ample, and set


hk,s = f1k g1 + . . . + fN k
gN ∈ H 0 X, (m0 + km)F + ksL .
For almost every N -tuple (g1 , . . . , gN ), lemma 9.5 below and the weak continuity
of ∂∂ show that
1 i 1
Dk,s = ∂∂ log |hk,s | = div(hk,s )
ks π ks
i
converges weakly to Ts = π ∂∂vs,Ω
as k tends to +∞, and that
1  1
ν(Ts , x) ≤ ν Dk,s , x ≤ ν(T, x) + .
ks ks
This, together with the first step, implies the proposition for some subsequence
Ds = Dk(s),s . We even obtain the more explicit inequality
n 1  1
ν(T, x) − ≤ ν Dk,s , x ≤ ν(T, x) + .
s ks ks

Lemma 9.5. — Let Ω be an open subset in Cn and let f1 , . . . , fN ∈

H 0 (Ω, OΩ ) be non zero functions. Let G ⊂ H 0 (Ω, OΩ ) be a finite dimensional
subspace whose elements generate all 1-jets at any point of Ω. Finally, set
v = sup log |fj | and
hk = f1k g1 + . . . + fN
k
gN , gj ∈ G \ {0}.
Then for all (g1 , . . . , gN ) in (G \ {0})N except a set of measure 0, the sequence
1 1
k log |hk | converges to v in Lloc (Ω) and
1  1
ν(v, x) ≤ ν log |hk | ≤ ν(v, x) + , ∀x ∈ X, ∀k ≥ 1.
k k

Proof. — The sequence k1 log |hk | is locally uniformly bounded above and
we have
1
lim log hk (z) = v(z)
k→+∞ k

at every point z where all absolute values |fj (z)| are distinct and all gj (z) are
nonzero. This is a set of full measure in Ω because the sets {|fj |2 = |fl |2 , j 6= l} and
{gj = 0} are real analytic and thus of zero measure (without loss of generality, we
may assume that Ω is connected and that the fj ’s are not pairwise proportional).
The well-known uniform integrability properties of plurisubharmonic functions

24
then show that k1 log |hk | converges to v in L1loc (Ω). It is easy to see that ν(v, x)
is the minimum of the vanishing orders ordx (fj ), hence
ν(log |hk |, x) = ordx (hk ) ≥ k ν(v, x).
In the opposite direction, consider the set Ek of all (N + 1)-tuples
(x, g1 , . . . , gN ) ∈ Ω × GN
for which ν(log |hk |, x) ≥ k ν(v, x) + 2. Then Ek is a constructible set in Ω × GN :
it has a locally finite stratification by analytic sets, since
[ [   \ 
Ek = x ; Dα fj (x) 6= 0 × GN ∩ (x, gl ) ; Dβ hk (x) = 0 .
s≥0 j, |α|=s |β|≤ks+1

The fiber Ek ∩({x}×GN ) over a point x ∈ Ω where ν(v, x) = min ord Px (fkj ) = s
is the
vector space of N -tuples (gj ) ∈ GN satisfying the equations Dβ fj gj (x) = 0,
|β| ≤ ks + 1. However, if ordx (fj ) = s, the linear map


(0, . . . , 0, gj , 0, . . . , 0) 7−→ Dβ (fjk gj (x)) |β|≤ks+1

has rank n + 1, because it factorizes into an injective map Jx1 gj 7→ Jxks+1 (fjk gj ). It
follows that the fiber Ek ∩ ({x} × GN ) has codimension at least n + 1. Therefore
dim Ek ≤ dim(Ω × GN ) − (n + 1) = dim GN − 1
and the projection of Ek on GN has measure S zero by Sard’s theorem. By definition
of Ek , any choice of (g1 , . . . , gN ) ∈ GN \ k≥1 pr(Ek ) produces functions hk such
that ν(log |hk |, x) ≤ k ν(v, x) + 1 on Ω.

10. Self-intersection inequality for closed positive currents

Let L be a nef line bundle over a projective algebraic manifold X and let
T = c(L) ≥ 0 be the curvature current of any singular metric on L. We want to
derive a bound for the codimension p components in the sublevel sets Ec (T ) in
terms of the p-th power {T }p of the cohomology class of T . The difficulty is that,
in general, T p does not make sense as a current. However, products of currents
can be defined in some special circumstances. Let M be an arbitrary complex
manifold and n = dimC X. Suppose given a closed positive current of bidegree
(p, p) on M and a locally bounded plurisubharmonic function ψ on M . According
to Bedford-Taylor [BT 82], the product Θ ∧ i∂∂ψ can then be defined by
(10.1) Θ ∧ i∂∂ψ = i∂∂ ψ Θ).
Here Θ is a differential form with measure coefficients, so its product by the locally
bounded Borel function ψ is a well defined current of order 0, and the derivative ∂∂
can be taken in the sense of distribution theory. The resulting current Θ ∧ i∂∂ψ
is again positive, as is easily seen by taking the weak limit with a sequence of
smooth approximations of ψ. More generally, if ψ1 , . . . , ψm are locally bounded
plurisubharmonic functions, the product Θ ∧ i∂∂ψ1 ∧ . . . ∧ i∂∂ψm is well defined by
induction on m. Various examples (cf. [Ki 84]) show that such products cannot be

25
defined in a reasonable way for arbitrary plurisubharmonic functions ψj . However,
functions with −∞ poles can be admitted if the polar set is sufficiently small.

Proposition 10.2. — Let ψ be a plurisubharmonic function on M such

that ψ is locally bounded on M \ A, where A is an analytic subset of M of
codimension ≥ p + 1 at each point. Then Θ ∧ i∂∂ψ can be defined in such a
way that Θ ∧ i∂∂ψ = limν→+∞ Θ ∧ i∂∂ψν in the weak topology of currents, for
any decreasing sequence (ψν )ν≥1 of plurisubharmonic functions converging to ψ.
Moreover, at every point x ∈ X we have
 i 
ν Θ ∧ ∂∂ψ, x ≥ ν(Θ, x) ν(ψ, x).
π

Proof. — When ψ is locally bounded everywhere, we have lim ψν Θ = ψ Θ

by the monotone convergence theorem and we can apply the continuity of ∂∂ with
respect to the weak topology to conclude that Θ ∧ i∂∂ψ = limν→+∞ Θ ∧ i∂∂ψν .
First assume that A is discrete. Since our results are local, we may suppose
that M is a ball B(0, R) ⊂ Cn and that A = {0}. For every s ≤ 0, the function
≥ ≥
ψ s = max(ψ, s) is locally bounded on M , so the product Θ ∧ i∂∂ψ s is well
≥s
defined. For |s| large, the function ψ differs from ψ only in a small neighborhood
of the origin, at which ψ may have a −∞ pole. Let γ be a (n−p−1, n−p−1)-form
with constant coefficients and set s(r) = lim inf |z|→r−0 ψ(z). By Stokes’ formula,
we see that
Z
≥
(10.3) Θ ∧ i∂∂ψ s ∧ γ
B(0,r)

does not depend on s when s < s(r), for the difference of two such integrals involves
the ∂∂ of a current with compact support in B(0, r). Taking γ = (i∂∂|z|2 )n−p−1 ,
we see that the current Θ ∧ i∂∂ψ has finite mass on B(0, r) \ {0} and we can define
h1l{0} (Θ ∧ i∂∂ψ), γi to be the limit of the integrals (10.3) as r tends to zero and
s < s(r). In this case, the weak convergence statement is easily deduced from the
locally bounded case discussed above.
In the case where codim A ≥ p + 1, we use a slicing technique to reduce the
situation to the discrete case. Set q = n − p − 1. There are linear coordinates
(z1 , . . . , zn ) centered

at any singular point of A, such that 0 is an isolated point of
A ∩ {0} × Cp+1 . Then there are small balls B ′ = B(0, r′ ) in Cq , B ′′ = B(0, r′′ )
in Cp+1 such that A ∩ (B ′ × ∂B ′′ ) = ∅, and the projection map
π : Cn −→ Cq , z = (z1 , . . . , zn ) 7−→ z ′ = (z1 , . . . , zq )
defines a finite proper mapping A ∩ (B ′ × B ′′ ) −→ B ′ . These properties are
preserved if we slightly change the direction of projection. Take sufficiently many
projections πm associated to coordinate systems (z1m , . . . , znm ), 1 ≤ m ≤ N , such
that the family of (q, q)-forms
i dz1m ∧ dz m m m
1 ∧ . . . ∧ i dzq ∧ dz q

26
defines a basis of the space of (q, q)-forms. Expressing any compactly supported
smooth (q, q)-form in such a basis, we see that we need only define
Z
(10.4) Θ ∧ i∂∂ψ ∧ f (z ′ , z ′′ ) i dz1 ∧ dz 1 ∧ . . . ∧ i dzq ∧ dz q =
′ ′′
Z n ZB ×B o
f (z ′ , •) Θ(z ′ , •) ∧ i∂∂ψ(z ′ , •) i dz1 ∧ dz 1 ∧ . . . ∧ i dzq ∧ dz q
B′ B ′′
where f is a test function with compact support in B ′ × B ′′ , and Θ(z ′ , •) denotes
the slice of Θ on the fiber {z ′ }×B ′′ of the projection π : Cn → Cq (see e.g. Federer
[Fe 69]). Here Θ(z ′ , •) is defined for almost every z ′ ∈ B ′ and is again a closed
positive current of bidegree (p, p) on B ′′ . The right hand side of (10.4) makes sense
since all fibers ({z ′ } × B ′′ ) ∩ A are discrete and the double integral is convergent
(this will be explained in a moment). The weak convergence statement can be
derived from the discrete case by (10.4) and the bounded convergence theorem.
Indeed, the boundedness condition is checked as follows: observe that the functions
ψ1 ≥ . . . ≥ ψν ≥ ψ are uniformly bounded below on some cylinder



Kδ,ε = B (1 − δ)r′ × B(r′′ ) \ B (1 − ε)r′′



disjoint from A, with ε ≪ δ ≪ 1 so small that Supp f ⊂ B (1−δ)r′ ×B (1−ε)r′′ ;
for all z ′ ∈ B((1 − δ)r′ ), the Chern-Levine-Nirenberg inequality [CLN 69] shows
that
Z Z
′ ′


+
Θ(z ′ , •) ∧ i∂∂|z ′′ |2
≥s
Θ(z , •) ∧ i∂∂ψν (z , •) ≤ Cε max ψ1 , |s|
B((1−ε)r ′′ ) Kδ,ε B((1−ε/2)r ′′ )

Proof: introduce a cut-off function χε (z ′′ ) equal to 1 near B((1 − ε)r′′ ) with

support in B((1 − ε/2)r′′ ), integrate by parts and write |i∂∂χε | ≤ Cε i∂∂|z|2 ; for
s sufficiently large (independent of ν), the left hand integral does not depend on
s and is equal by definition to the corresponding integral involving ψν ; the right
hand side, of course, has a bounded integral over B((1 − δ)r′ ) because we integrate
Θ against a smooth form. The same argument with ψ instead of ψν shows that
the right hand side of (10.4) is convergent.
It only remains to prove the final statement concerning Lelong numbers.
Assume that M = B(0, r) and x = 0. By definition
  Z i n−p−1
i i
ν Θ ∧ ∂∂ψ, x = lim Θ ∧ ∂∂ψ ∧ ∂∂ log |z| .
π r→0 |z|≤r π π
Set γ = ν(ψ, x) and


ψν (z) = max ψ(z), (γ − ε) log |z| − ν
with 0 < ε < γ (if γ = 0, there is nothing to prove). Then ψν decreases to ψ and
Z i n−p−1 Z i n−p−1
i i
Θ∧ ∂∂ψ∧ ∂∂ log |z| ≥ lim sup Θ∧ ∂∂ψν ∧ ∂∂ log |z|
|z|≤r π π ν→+∞ |z|≤r π π
by the weak convergence of Θ ∧ i∂∂ψν ; here ( πi ∂∂ log |z|)n−p−1 is not smooth on
B(0, r), but the integrals remain unchanged if we replace log |z| by χ(log |z|/r)

27
with a smooth convex function χ such that χ(t) = t for t ≥ −1 and χ(t) = 0
for t ≤ −2. Now, we have ψ(z) ≤ γ log |z| + C near 0, so ψν (z) coincides with
(γ − ε) log |z| − ν on a small ball B(0, rν ) ⊂ B(0, r) and we infer
Z i n−p−1 Z i n−p
i
Θ ∧ ∂∂ψν ∧ ∂∂ log |z| ≥ (γ − ε) Θ∧ ∂∂ log |z|
|z|≤r π π |z|≤rν π
≥ (γ − ε)ν(Θ, x).
As r ∈ ]0, R[ and ε ∈ ]0, γ[ were arbitrary, the desired inequality follows.

Corollary 10.5. — For 1 ≤ j ≤ p, let Tj = πi ∂∂ψj be closed positive

(1, 1)-currents on a complex manifold M . Suppose that there are analytic sets
A2 ⊃ . . . ⊃ Ap in M with codim Aj ≥ j at every point such that each ψj , j ≥ 2,
is locally bounded on M \ Aj . Let {Ap,k }k≥1 be the irreducible components of Ap
of codimension p exactly and let νj,k = minx∈Ap,k ν(Tj , x) be the generic Lelong
number of Tj (or ψj ) along Ap,k . Then T1 ∧ . . . ∧ Tp is well-defined and
+∞
X
T1 ∧ . . . ∧ Tp ≥ ν1,k . . . νp,k [Ap,k ].
k=1

Proof. — By induction on p, Proposition 10.2 shows that T1 ∧ . . . ∧ Tp is

well defined. Moreover, we get
ν(T1 ∧ . . . ∧ Tp , x) ≥ ν(T1 , x) . . . ν(Tp , x) ≥ ν1,k . . . νp,k
at every point x ∈ Ap,k . The desired inequality is a consequence of Siu’s
decomposition theorem (3.8).
Now, let X be a projective n-fold and let T be a closed positive (1, 1)-current
on X. By the Lebesgue decomposition theorem, we can write T = Tabc + Tsing
where Tabc has absolutely continuous coefficients with respect to the Lebesgue
measure and the coefficients of Tsing are singular measures. In general, Tabc
and Tsing are positive but non closed. We fix an arbitrary set Ξ ⊂ X and for
p = 1, 2, . . . , n, n + 1 we set


(10.6) bp = bp (T, Ξ) = inf c > 0 ; codim Ec (T ), x ≥ p, ∀x ∈ Ξ ,
with the convention that a germ has codimension > n if and only if it is empty.
Then 0 = b1 ≤ b2 ≤ . . . ≤ bn ≤ bn+1 with bn+1 = maxx∈Ξ ν(T, x) < +∞, and
for c ∈ ]bp , bp+1 ], Ec (T ) has codimension ≥ p at every point of Ξ and has at least
one component of codimension p exactly which intersects Ξ. We call b1 , b2 , . . .
the jumping values of the Lelong numbers of T over Ξ. Our goal is to prove the
following fundamental inequality for the Lelong sublevel sets Ec (T ), when T is
the curvature current c(L) of a line bundle (this restriction is unnecessary but the
general case is more involved, see [De 92] for a general proof).

Theorem 10.7. — Suppose that there is a semipositive line bundle G

over X and a constant a ≥ 0 such that OT X (1) + a π ⋆ G is nef; set u = a c(G) ≥ 0

28
with any smooth semipositive metric on G. Let T = c(L) ≥ 0 be the curvature
current of a nef line bundle L, let Ξ ⊂ X be an arbitrary subset and bpS= bp (T, Ξ).
Denote by {Zp,k }k≥1 the irreducible components of codimension p in c>bp Ec (T )
which intersect Ξ and let νp,k ∈ ]bp , bp+1 ] be the generic Lelong number of T
along Zp,k . Then the De Rham cohomology class ({T } + b1 {u}) · · · ({T } + bp {u})
can be represented by a closed positive current Θp of bidegree (p, p) such that
X
Θp ≥ (νp,k − b1 ) . . . (νp,k − bp ) [Zp,k ] + (Tabc + b1 u) ∧ . . . ∧ (Tabc + bp u).
k≥1
The same is true for Ξ = X if we only suppose c1 (L) ∈ Γ+ instead of L nef.
V
Here (Tabc + bj u) is computed pointwise as a (p, p)-form. It follows in
p
particular from our inequality that Tabc has locally integrable coefficients for all p.
Let ω be a Kähler metric on X. If we take the wedge product of the fundamental
inequality 10.7 by ω n−p , integrate over X and neglect Tabc in the right hand side,
we get:

Corollary 10.8. — With the notation of theorem S 10.7, the degrees with
respect to ω of the p-codimensional components Zp,k of c>bp Ec (T ) intersecting
Ξ satisfy
+∞
X Z
(νp,k − b1 ) . . . (νp,k − bp ) [Zp,k ] ∧ ω n−p
k=1 X



≤ {T } + b1 {u} · · · {T } + bp {u} · {ω}n−p .

In particular, if D is a nef divisor and if L = O(D) is equipped with the

singular metric such that T = c(L) = [D], we get a bound for the degrees of the
p-codimensional singular strata of D in terms of a polynomial of degree p in the
cohomology class {D}. The case X = IPn is of course especially simple: Since
T IPn is ample, we can take u = 0, and then the bound is simply {D}p · {ω}n−p ;
the same is true more generally as soon as T X is nef. It is natural to try to
find an interpretation of the (p, p)-form involving Tabc in the general inequality.
Unfortunately this (p, p)-form is not closed and so it does not correspond to
an intrinsic cohomology class that would have a simple counterpart in algebraic
geometry. Nevertheless, the additional Monge-Ampère mass provided by Tabc is
absolutely crucial for the purely algebraic application which we shall make in the
next section. Our intuition is that the additional (p, p)-form must be understood
as an excess of self-intersection, measuring asymptotically the amount of freedom
a divisor in the linear system H 0 (X, sL) can keep while being moved through the
fixed singular strata prescribed by sT , when s tends to infinity.

Proof of Theorem 10.7. — By the first step of the proof of theorem 9.1 and
by (9.4), there is a positive line bundle F with the following property: for every
s ≥ 0 there exist sections fi ∈ H 0 (X, F + sL), 1 ≤ i ≤ N (s), with
n 1
ν(T, x) − ≤ min ordx (fi ) ≤ ν(T, x), ∀x ∈ X.
s s
29
The main idea is to decrease the Lelong numbers by replacing each section fi by
some of its high order derivatives, or rather by some jet section. In this way,
the polar components with low generic Lelong number disappear, and we can
decrease the dimension so as to be able to take intersections of currents (thanks
to proposition 10.2 or corollary 10.5). Of course, introducing jet sections also
introduces symmetric powers of the cotangent bundle; this is the reason why the
curvature of T X plays an important role in the inequality.
First step: killing Lelong numbers in the singular metric of L.
Consider the m-jet section J m fi with values in the vector bundle Em =
m
J O(F +sL) of m-jets. First suppose that a is rational. There are exact sequences
0 −→ S m T ⋆ X ⊗ O(F + sL) −→ Em −→ Em−1 −→ 0,
and S m T X ⊗ O(maG) is nef by our assumptions. By induction on m we easily
infer that
⋆
Em ⊗ O(2F + sL + maG)
is ample (in an exact sequence of vector bundles with ample extreme terms, so is
the middle term). Hence there is a symmetric power of order q with qa ∈ IN such
that
S q Em
⋆
⊗ O(2qF + qsL + qmaG)
is generated by holomorphic sections gj . We use the pairing of S q Em and S q Em
⋆

to get sections


S q (J m fi ).gj ∈ H 0 X, O(2qF + qsL + qmaG) .
By means of these sections, for each pair (s, m) we define a new singular metric
|| ||s,m on L such that
||ξ||
||ξ||s,m = P q (J m f ).g ||1/qs
, ξ ∈ L,
i,j ||S i j

where || || denotes the original singular metric on L as well as the induced metric
on O(2qF +qsL+qmaG); here the metric of F (resp. G) is smooth and has positive
(resp. semipositive) curvature. Denote by ϕ the weight of the original metric on
L, by ϕs,m the new one, and by ψF , ψG the weights of F , G on some trivializing
open set Ω ⊂ X. Then
1 X 2 m
(10.9) ϕs,m = log S q (J m fi ).gj − ψF − aψG ,
qs i,j
s s
1
because e−ϕ appears in the numerator and exp(−2qψF − qsϕ − qmaψG ) qs in
the denominator of ||ξ||s,m . As ψF , ψG are smooth and the gj ’s do not vanish
simultaneously, we get
1 1

ν(ϕs,m , x) = min ordx (J m fi ) = min ordx (fi ) − m + .
s i s i
Hence we have the inequality
 m + n  m
(10.10) ν(T, x) − ≤ ν(ϕs,m , x) ≤ ν(T, x) − ;
s + s +

30
that is, we have been able to construct a new curvature current πi ∂∂ϕs,m on L in
which all the Lelong numbers that were ≤ m/s have been killed. Unfortunately
the curvature is no longer ≥ 0, but by (10.9) we have
i 2 m 2 m
(10.11) ∂∂ϕs,m ≥ − c(F ) − a c(G) = − ω − u,
π s s s s
where ω = c(F ) > 0. Only the term 2s ω can be made arbitrarily small. Now, for
each s, select an integer m such that bp < m/s ≤ bp + 1s . By (10.9) and (10.10), we
see that ϕs,m is locally bounded on X \ Em/s (T ), and the definition of bp implies
that Em/s (T ) has codimension ≥ p in a neighborhood of Ξ.
Second step: Construction of the p-th intersection current Θp .
By induction on p, we suppose that Θp−1 has already been constructed
(Θ1 = T satisfies the requirements for p = 1). By proposition 10.2, the wedge
product Θp−1 ∧ πi ∂∂ϕs,m is well defined in a neighborhood of Ξ. However, this
is not satisfactory when Ξ 6= X, because we need a current defined everywhere
on X. This is the reason why we have to assume L nef when Ξ 6= X. Under this
assumption, there is for each s a smooth metric on L, associated to some weight ρs
on the trivializing open set Ω, such that πi ∂∂ρs ≥ − 1s ω. We introduce the weight


ψs,m,A,B = sup ϕ , ϕs,m − A , ρs − B ,
where A, B > 0 are large constants. This weight corresponds to the singular metric
on L given by


||ξ||s,m,A,B = inf ||ξ|| , eA ||ξ||s,m , eB ||ξ||ρs .
Clearly ψs,m,A,B converges to ϕ as A, B tend to +∞, and ψs,m,A,B is locally
bounded; therefore the curvature current Ts,m,A,B = πi ∂∂ψs,m,A,B converges
weakly to T = πi ∂∂ϕ as A, B tend to +∞. Moreover, the assumed lower bound
on πi ∂∂ρs combined with (10.11) implies
2 m
Ts,m,A,B ≥ − ω − u ;
s s
this is easily seen by adding 2s ψF + m
s aψG to each term in the supremum formula
defining ψs,m,A,B . Now, the positive (p, p)-current
 2 m 
Θp,s,m,A,B = Θp−1 ∧ Ts,m,A,B + ω + u
s s
is well defined over X since ψs,m,A,B is locally bounded. Its
cohomology class is
independent of A, B and converges to {c1 (L)} · c1 (L) + bp u when s tends to +∞
(by the choice of m made at the end of the first step, we have lim m/s = bp ). Hence
the family (Θp,s,m,A,B ) is weakly compact. First extract a weak limit Θp,s,m,A
by taking some subsequence Bν → +∞. By proposition 10.2 we see that in a
neighborhood of Ξ
 2 m 
Θp,s,m,A = lim Θp,s,m,A,B = Θp−1 ∧ Ts,m,A + ω + u ,
B→+∞ s s

31
where
i
Ts,m,A = ∂∂ψs,m,A , ψs,m,A = sup(ϕ, ϕs,m − A).
π
Indeed the codimension of the set of poles of ψs,m,A is at least p in a neighborhood
of Ξ. Now, by (10.10), we have

 m + n
ν(ψs,m,A , x) ≥ min ν(ϕ, x), ν(ϕs,m , x) ≥ ν(T, x) − .
s +

Proposition 10.2 shows that

 m + n
ν(Θp,s,m,A , x) ≥ ν(Θp−1 , x) ν(T, x) − near Ξ.
s +

By induction on p, we conclude that the generic Lelong number of Θp,s,m,A along

Zp,k is at least equal to
 m + n
(νp,k − b1 ) . . . (νp,k − bp−1 ) ν(T, x) − .
s +

In fact, Zp,k meets Ξ at some point x, and therefore the inequality holds at least
on a neighborhood of x in Zp,k . Siu’s decomposition formula (3.8) yields
 m + n
Θp,s,m,A ≥ (νp,k − b1 ) . . . (νp,k − bp−1 ) ν(T, x) − [Zp,k ].
s +

Now, extract a weak limit Θp,s,m for some subsequence Aν → +∞ and then a
weak limit Θp for some subsequence mν /sν → bp with
sν → +∞. We obtain a
current Θp such that {Θp } = {Θp−1 } · c1 (L) + bp {u} and
Θp ≥ (νp,k − b1 ) . . . (νp,k − bp ) [Zp,k ].
It only remains to show by induction on p that
Θp,abc ≥ (Tabc + b1 u) ∧ . . . ∧ (Tabc + bp u).
As the coefficients of [Zp,k ] are singular with respect to the Lebesgue measure, Θp
will actually be larger than the sum. By construction, there exists a subsequence
(sν , mν , Aν , Bν ) such that
i 2 mν 
Θp = lim Θp−1 ∧ ∂∂ψsν ,mν ,Aν ,Bν + ω + u ,
π s sν

ν
ψsν ,mν ,Aν ,Bν = sup ϕ , ϕsν ,mν − Aν , ρsν − Bν .
The desired lower bound follows from lemma 10.12 below. At the beginning of the
proof, a was supposed to be rational, but this extra assumption can be removed as
above by extracting a weak limit Θp,aν → Θp with a sequence of rational numbers
decreasing to a ∈ IR+ . If Ξ = X, everything works even if we omit the term ρk − B
in the definition of ψk,m,A,B : we can start directly with ψk,m,A because its polar
set has codimension ≥ p on the whole space X. Hence the nef assumption on L is
not necessary.

Lemma 10.12. — Let Ω ⊂ Cn be an open subset and let ϕ be an arbitrary

plurisubharmonic function on Ω. Set ϕν = max(ϕ, ψν ) where ψν is a decreasing

32
sequence of plurisubharmonic functions converging to −∞, each ψν being locally
bounded in Ω (or perhaps only in the complement of an analytic subset of
codimension ≥ p). Let Θ be a closed positive current of bidegree (p − 1, p − 1).
If Θ ∧ i∂∂ϕν converges to a weak limit Θ′ , then
Θ′abc ≥ Θabc ∧ (i∂∂ϕ)abc .

Proof. — Let (ρε ) (resp. (e ρε )) be a family of regularizing kernels on Cn

2
(resp. on IR ), and let maxε (x, y) = (max⋆ ρeε )(x, y) be a regularized max function.
For ε > 0 small enough, the function
ϕν,ε = maxε (ϕ ⋆ ρε , ψν ⋆ ρε )
is plurisubharmonic and well defined on any preassigned open set Ω′ ⊂⊂ Ω. As
ϕν,ε decreases to ϕν when ε decreases to 0, proposition 10.2 shows that
lim Θ ∧ i∂∂ϕν,ε = Θ ∧ i∂∂ϕν
ε→0

in the weak topology. Let (βj ) be a sequence of test forms which is dense in the
space of test forms of bidegree (n − p, n − p) and contains strongly positive forms
with arbitrary large compact support in Ω. Select εν > 0 so small that
1
hΘ ∧ i∂∂ϕν,εν − Θ ∧ i∂∂ϕν , βj i ≤ for j ≤ ν.
ν
Then the sequence Θ∧i∂∂ϕν,εν is locally uniformly bounded in mass and converges
weakly to the same limit Θ′ as Θ ∧ i∂∂ϕν . Moreover, at every point x ∈ Ω such
that ϕ(x) > −∞, we have ϕν,εν (x) ≥ ϕ(x) > ψν ⋆ ρεν (x) + 1 for ν large, because
limν→−∞ ψν = −∞ locally uniformly. Hence ϕν,εν = ϕ ⋆ ρεν on a neighborhood
of x (which may depend on ν) and i∂∂ϕν,εν (x) = (i∂∂ϕ) ⋆ ρεν (x) for ν ≥ ν(x).
By the Lebesgue density theorem, if µ is a measure of absolutely continuous part
µabc , the sequence µ⋆ρεν (x) converges to µabc (x) at almost every point. Therefore
lim i∂∂ϕν,εν (x) = (i∂∂ϕ)abc (x) almost everywhere For any strongly positive test
form α = iα1 ∧ α1 ∧ . . . ∧ iαn−p ∧ αn−p of bidegree (n − p, n − p) on Ω, we get
Z Z
′
Θ ∧ α = lim Θ ∧ i∂∂ϕν,εν ∧ α
Ω ν→+∞ Ω
Z Z
≥ lim inf Θabc ∧ i∂∂ϕν,εν ∧ α ≥ Θabc ∧ i∂∂ϕabc ∧ α.
ν→+∞ Ω Ω

Indeed, the first inequality holds because i∂∂ϕν,εν is smooth, and the last one
results from Fatou’s lemma. This implies Θ′abc ≥ Θabc ∧(i∂∂ϕ)abc and lemma 10.12
follows.

11. Proof of the criterion in arbitrary dimension

We return here to the point where we arrived at the end of § 6, and apply
our self-intersection inequality 10.7 to complete the proof of the Main Theorem.
First suppose, with the notation of § 6, that L is an ample line bundle over X.

33
The idea is to apply inequality 10.7 to the (1, 1)-current T = lim ωεν produced
by equation (6.5), and to integrate the inequality with respect to the Kähler form
ω = c(L). Before doing this, we need to estimate the excess of intersection in
n
terms of Tabc .

Proposition 11.1. — The absolutely continuous part Tabc of T satisfies

 σ 
n
Tabc ≥ 1 − n ω n a.e. on X.
L

Proof. — The result is local, so we can work in an open set Ω which is

relatively compact in a coordinate patch of X. Let ρδ be a family of smoothing
kernels. By a well known lemma (see e.g. [BT 76], proposition 5.1), the operator
A 7→ (det A)1/n is concave on the cone of nonnegative hermitian n × n matrices,
hence we get

n 1/n  σ 1/n n 1/n
ωε ⋆ ρδ (x) ≥ (ωεn )1/n ⋆ ρδ (x) ≥ 1 − n (ω ) ⋆ ρδ (x),
L
thanks to equation (6.5). As εν tends to 0, ωεν ⋆ ρδ converges to T ⋆ ρδ in the
strong topology of C ∞ (Ω), thus

1/n  σ 1/n n 1/n
(T ⋆ ρδ )n ≥ 1− n (ω ) ⋆ ρδ on Ω.
L
Now, take the limit as δ goes to 0. By the Lebesgue density theorem T ⋆ ρδ (x)
converges almost everywhere to Tabc (x) on Ω, so we are done.

According to the notation used in § 10, we consider an arbitrary subset

Ξ ⊂ X and introduce the jumping values


bp = inf c > 0 ; codim Ec (T ), x ≥ p, ∀x ∈ Ξ .
By proposition 11.1 and inequality 5.2(a), we have
 σ j/n n
j
(11.2) Tabc ∧ ω n−j ≥ 1 − n ω .
L
Now, suppose that OT X (1) + a π ⋆ L is nef for some constant a ≥ 0. We can then
apply theorem 10.7 with u = a ω and {Θp } = (1 + b1a) · · · (1 + bp a){ω p } ; by taking
the wedge product of Θp with ω n−p , we get
Z X Z
(1 + b1 a) . . . (1 + bp a) ωn ≥ (νp,k − b1 ) . . . (νp,k − bp ) [Zp,k ] ∧ ω n−p
X k≥1 X
Z
+ (Tabc + b1 a ω) ∧ . . . ∧ (Tabc + bp a ω) ∧ ω n−p .
X
p−j
Combining this inequality with (11.2) for Tabc yields
X
n
(1 + b1 a) . . . (1 + bp a) L ≥ (νp,k − b1 ) . . . (νp,k − bp ) Ln−p · Zp,k
k≥1
X  σ (p−j)/n n
+ Sjp (b) aj 1 − n L ,
L
0≤j≤p

34
where Sjp (b), 1 ≤ j ≤ p, denotes the elementary symmetric function of degree j in
Q P
b1 , . . . , bp and S0p (b) = 1. As (1 + bj a) = Sjp (b)aj , we get
X
(νp,k − b1 ) . . . (νp,k − bp ) Ln−p · Zp,k ≤
k≥1 X p   σ (p−j)/n  n
(11.3) Sj (b) aj 1 − 1 − n L .
L
0≤j≤p
If L is only supposed to be big and nef, we follow essentially the same arguments
1
and replace ω in all our inequalities by ωm = c(L)m + m c(A) with A ample
n
(see section 6). Note that all (n, n)-forms ωm were defined to be proportional to
γ n = c(A)n , so inequality 11.1 becomes in the limit
 σ  Ln n  σ  Ln
n
Tabc ≥ 1− n γ = 1− n ωn .
L A n L (L + m 1
A)n m
The intersection inequality (11.3) is the expected generalization of proposition 8.2
in arbitrary codimension. In this inequality, νp,k is the generic Lelong S
number of T
along Zp,k , and Zp,k runs over all p-codimensional components Y of c>bp Ec (T )
intersecting Ξ ; by definition of bj we have maxk νp,k = bp+1 . Hence we obtain:

Theorem 11.4. — Let L be a big nef line bundle such that T X ⊗O(aL) is
nef, and let T ∈ c1 (L) be the positive curvature current obtained by concentrating
the Monge-Ampère mass Ln into a finite sum of Dirac measures with total mass σ,
plus some smooth positive density spread over X (equations (6.5) and (6.10)).
Then the jumping values bp of the Lelong number of T over an arbitrary subset
Ξ ⊂ X satisfy the inductive inequalities
1 X
(11.5) (bp+1 − b1 ) . . . (bp+1 − bp ) ≤ n−p
Sjp (b) aj σp−j ,
minY L ·Y
0≤j≤p−1
n j/n

n
where σj = 1 − (1 − σ/L ) L , and where Y runs over all p-codimensional
subvarieties of X intersecting Ξ.

Observe that σj is increasing in j ; in particular σj < σn = σ for

j ≤ n − 1. Moreover,
the convexity of the exponential function shows that
t 7→ 1t 1 − (1 − σ/Ln )t Ln is decreasing, thus σj > σp j/p for j < p ; in particular
σj > σj/n for j ≤ n − 1. We are now in a position to prove the following general
result, which contains the Main Theorem as a special case:

Theorem 11.6. — On a projective n-fold X, let gj : (X, xj ) → (Cn , 0) be

germs of finite holomorphic maps with covering degree

ρj . Let J1 ⊂ OX,x1 , . . . ,
JN ⊂ OX,xN be the associated ideals I τj,0 log |gj | xj and let
X 
n
σ0 = ρj τj,0 , σp = 1 − (1 − σ0 /Ln )p/n Ln , 1 ≤ p ≤ n − 1,
where L is a big nef line bundle such that Ln > σ0 . Suppose that OT X(1) + a π ⋆ L
is nef over P (T ⋆ X) and that there is a sequence 0 = β1 < . . . < βn ≤ 1 with
X
Ln−p · Y > (βp+1 − β1 )−1 . . . (βp+1 − βp )−1 Sjp (β) aj σp−j
0≤j≤p−1

35
for every subvariety Y ⊂ X of codimension p = 1, 2, . . . , n − 1 passing through one
of the points xj . Then there is a surjective map
M

H 0 (X, KX + L) −→
−→ O(KX + L)xj ⊗ OX,xj /Jj .
1≤j≤N

Proof. — Select τj > τj,0 so that Ln−p · Y still satisfies the above lower
bound with the corresponding value σ > σ0 . Then apply theorem 11.4 with
Ξ = {x1 , . . . , xN }. Inequality (11.5) shows

inductively that bp < βp for p ≥ 2,
so bn < 1 and we get codim E1 (T ), xj = n at each point xj . Thanks to (4.3),
corollaries 4.6 and 6.8 imply the desired surjectivity property.

Proof of Corollary 1. — This is only a matter of straightforward calcula-

tions, but adjusting the constants βp to get optimal exponents for σ0 , a and µΞ (L)
in the lower bound of m requires some care. By the convexity argument already
explained, we have σp−j ≤ (p − j)σ1 ≤ p(1 − 1/p)j σ1 . As β1 = 0, we find
X X  p − 1 j
Sjp (β) aj σp−j ≤ pσ1 Sjp (β) a
p
0≤j≤p−1 0≤j≤p−1
 p−1   p−1 
= pσ1 1 + β2 a . . . 1 + βp a .
p p
When we replace L by mL, the constant a is replaced by a/m, and by definition of
µ = µΞ (L) we have (mL)k · Y ≥ (mµ)k . Hence theorem 11.6 yields the sufficient
condition 
Y p−1 a   p−1 a 
(11.7) (mµ)n−p ≥ (βp+1 − βj )−1 pσ1 1 + β2 . . . 1 + βp
p m p m
j≤p
with 0 = β1 < . . . < βn < 1. When p = 1 we get (mµ)n−1 ≥ β2−1 σ1 , and when
p = n − 1 the inequality implies mµ > (n − 1)σ1 > (1 − 1/n)σ0 . We suppose in
fact mµ ≥ λσ0 where λ > 1 − 1/n is a constant which will be adjusted later to
an
Q optimal value; in particular mµ ≥ λnn . We will choose βj /βj+1 so small that
−1 p
j≤p (βp+1 − βj ) ≥ Up βp+1 with a constant Up slightly larger than 1. We are
thus led to define βp inductively by the formula
p pσ1  p−1 a   p−1 a 
(11.8) βp+1 = Up 1+β 2 . . . 1+β p , 1 ≤ p ≤ n−1,
(mµ)n−p p m p m
and m has to be taken so large that βn < 1 ; suppose that this is the case. The
first step is to determine admissible constants 1 < U1 < . . . < Un−1 . For j ≤ p,
(11.8) implies
1  βj j−1 1  βp p−1 j −1 1
j−1
− n−j
≤ =⇒ βj ≤ mµ j−1
j − 1 mµ p − 1 mµ n−1
by taking p = n in the first inequality. In general, for j ≤ p we get
 β p−1 p−j

j j − 1  βj p−j j − 1  j − 1  j−1 −(p − j) 1 + n−j
j−1
≤ ≤ (mµ) ,
βp p − 1 mµ p−1 n−1
1

βj (j − 1) j−1 1 1 γj
−(n − 1) j−1 − p−1
(11.9) ≤ 1 1 1 (mµ) ≤
βp − γ p
(p − 1) p−1 (n − 1) j−1 p−1

36
with
j −1 1 1 1


j−1 n −(n − 1) j−1 − n−1
γj = (λn ) .
n−1
The sequence (γj ) is strictly increasing and satisfies γj /γj+1 < 1/n. Thus we can
take Y
−1
Up = 1 − γj /γp+1 .
2≤j≤p

Let k be the largest integer

in {1, . . . , n − 1} such that βk a/m ≤ 1 and let
t = max βk a/m, γk /γk+1 ∈ [γk /γk+1 , 1]. Inequality (11.9) implies
 p − 1 γj 
  
 1+ t for 2 ≤ j ≤ k,
p−1 a p γ
1 + βj ≤ p − 1 γ k  a
p m 
 +
k −1
t βj for k + 1 ≤ j ≤ p.
p γj m
The product of all factors (. . . t±1 ) over j = 2, . . . , p is a polynomial with positive
coefficients in IR[t, t−1 ], hence is a convex function of t. Therefore, the product is
at most equal to the maximum of its values for t = 1 or t = γk /γk+1 :
(
Y  p − 1 γj  Y  p − 1 γk 
max 1+ + ,
p γk p γj
2≤j≤k k+1≤j≤p
)
Y  p − 1 γj  Y  p − 1 γk+1 
1+ + .
p γk+1 p γj
2≤j≤k k+1≤j≤p

We have (p − 1)/p + γk /γj ≤ 1 − 1/p + 1/n ≤ 1 for j > k, moreover the sequence
γj /γj+1 is increasing. If we introduce the increasing sequence
Y  p − 1 γj 
Vp = 1+ ,
p γp
2≤j≤p

it is then easy to check that the above maximum is bounded by Vp for p ≥ k.

Therefore (11.8) gives
pσ1  a p−k
p
βp+1 ≤ Up Vp n−p
βk+1 . . . βp for p ≥ k.
(mµ) m
As Up Vp p ≤ Wn = Un−1 Vn−1 (n − 1), by induction these inequalities yield
 W σ 1/k 1
n 1 + 1 + . . . + k+1
1
βp+1 ≤ n−k
(aµ) p p−1 for p ≥ k,
(mµ)
where the exponent of aµ is understood to be 0 if p = k. Finally, we have
(mL)n ≥ (mµ)n ≥ (λσ0 )n by definition of µ and by our initial hypothesis
mµ ≥ λσ0 , hence σ0 /(mL)n ≤ λ−n σ01−n ≤ λ−n n−n(n−1) . It follows again from a
convexity argument that

1/n  n n(n−1)
σ1 /σ0 ≤ 1 − 1 − λ−n n−n(n−1) λ n = Tn .

37
Hence Wn σ1 ≤ Bn σ0 with Bn = Wn Tn = (n−1)Un−1 Vn−1 Tn . Therefore, sufficient
conditions in order that βn < 1 are:
(11.10) mµ > Bn σ0 for k = n − 1,
1


k + 1 + ...+ 1
(11.10k ) (mµ)n−k > Bn σ0 (aµ) n−1 n−2 k+1

for k ∈ {1, . . . , n − 2}. These conditions are equivalent to the inequality stated in
corollary 1. Observe that our constant Bn depends on λ. The initial hypothesis
mµ ≥ λσ0 will be automatically satisfied if we adjust λ so that Bn (λ) = λ;
this is always possible because Bn (λ) is decreasing in λ and Bn (λ) > 1 − 1/n.
With this choice, a numerical calculation shows that Bn < 2.005 for all n and
limn→+∞ Bn = 2. For small values of n, we find (with rounding by above):
n 2 3 4 5 6 7 8 9 10 11 12
(11.11)
Bn 0.625 1.019 1.309 1.485 1.603 1.687 1.748 1.794 1.830 1.858 1.879

and Bn < 2 for n ≤ 46.

12. Universal bounds for very ample line bundles

Let X be an ample line bundle over a projective n-fold X. In order to

find universal conditions for KX + L to be very ample, our main theorem would
require a universal value a depending only on n = dimC X such that T X ⊗ O(aL)
is always nef. However, this is clearly impossible as the example of curves already
show: if X is a curve of genus g and L has degree 1, then T X ⊗ O(aL) is nef if
and only if a ≥ 2g − 2. In general, it is an interesting unsolved question to know
whether such a value a can be found explicitly in terms of geometric invariants
of X (Chern classes, . . .). Here, these difficulties can be avoided by means of the
following simple lemma.

Lemma 12.1. — Let F be a very ample line bundle over X. Then the
vector bundle T X ⊗ O(KX + nF ) is nef and generated by global sections.

Proof. — By the very ample assumption, the 1-jet bundle J 1 F is generated

by its sections. Consider the exact sequence
0 −→ T ⋆ X ⊗ F −→ J 1 F −→ F −→ 0
rank(J 1 F ) = n + 1 and det(J 1 F ) = KX + (n + 1)F . The n-th exterior
where V
n
power (J 1 F ) is also generated by sections and there is a surjective morphism
Vn 1
(J F ) = (J 1 F )⋆ ⊗ det(J 1 F ) −→ (T X ⊗ F ⋆ ) ⊗ det(J 1 F ) = T X ⊗ O(KX + nF ).
Hence T X ⊗ O(KX + nF ) is generated by sections and, in particular, it is nef.

The next idea consists in the following iteration trick: lemma 12.1 suggests
that a universal lower bound for the nefness of T X ⊗ O(aL′ ) can be achieved with
L′ = KX + L if L is sufficiently ample. Then it follows from the Main Theorem

38
that KX + L′ = 2KX + L is very ample under suitable numerical conditions.

Lemma 12.1 applied with F = 2K
X + L shows that T X ⊗ O (2n + 1)KX + nL is
nef, and thus T X ⊗ O (2n + 1)L′′ is nef with L′′ = KX + 21 L ≤ L′ . Hence we see
that the Main Theorem can be iterated. The special value a = 2n + 1 will play an
important role.

Lemma 12.2.
— Let L′ be an ample line bundle over X. Suppose that

T X ⊗ O (2n + 1)L′ is nef. Then KX + L′ is very ample, resp. generates s-jets,
as soon as µX (L′ ) > Cn σ0 with the corresponding value of σ0 and with a constant
Cn < 3 depending only on n.


Proof. — If µ = µX (L′ ), a = 2n + 1 and σ0 = 2nn resp. σ0 = (n + s)n ,
then the first arguments in the proof of corollary 1 give the sufficient condition
Y  p−1   p−1 
µn−p > (βp+1 − βj )−1 pσ1 1 + β2 (2n + 1) . . . 1 + βp (2n + 1)
p p
2≤j≤p

with 0 = β1 < . . . < βn = 1. We suppose µ ≥ λσ0 (in particular µ ≥ 2λnn ) and

choose
1/(p−1)
βp = αn−p+1 (2nn )n−p , 2≤p≤n−1
with suitable constants λ, α to be determined later. In analogy with the proof of
corollary 1, we introduce the constants
Y  βj −1 Y  p−1 
Up′ = 1− , Vp′ = 1+ βj (2n + 1) ,
βp+1 p
2≤j≤p 2≤j≤p


1/n
Tn′ = 1 − 1 − λ−n (2nn )−(n−1) λn (2nn )n−1 .

We have σ1 ≤ Tn′ σ0 and our conditions become

 ′ ′ ′ n−p
n−p ′ p ′ ′ pUp Vp Tn α (2nn )n−p−1 σ0 for p ≤ n − 2,
µ > Up βp+1 pTn σ0 Vp = ′ ′
(n − 1)Un−1 Vn−1 Tn′ σ0 for p = n − 1.
As σ0 ≥ 2nn , a sufficient condition is
 ′ ′
µ > max (n − 1)Un−1 Vn−1 Tn′ , α(pUp′ Vp′ Tn′ )1/(n−p) σ .
1≤p≤n−2 0

We adjust λ and α so that

′ ′
(n − 1)Un−1 Vn−1 Tn′ = α max (pUp′ Vp′ Tn′ )1/(n−p) = λ,
1≤p≤n−2

and we take this common value to be our constant Cn . A numerical calculation

gives Cn < 3 for all n and limn→+∞ Cn = 3. The first values are given by

n 2 3 4 5 6 7 8 9 10 11 12
(12.3)
Cn 0.563 0.737 0.995 1.201 1.370 1.510 1.629 1.730 1.817 1.893 1.959

Hence Lemma 12.2 is proved.

39
 Lemma 12.4. — Let F be a line bundle which generates s-jets at every
point. Then F p · Y ≥ sp for every p-dimensional subvariety Y ⊂ X.

Proof. — Fix an arbitrary point x ∈ Y . Then consider the singular metric

on F given by
|ξ|2
||ξ||2 = P ,
|uj (z)|2
where (u1 , . . . , uN ) is a basis of H 0 (X, F ⊗Msx ). By our assumption, these sections
have an isolated common zero of P order s at x. Hence F possesses a singular metric
such that the weight ϕ = 12 log |uj |2 is plurisubharmonic and has an isolated
logarithmic pole of Lelong number s at x. By the comparison inequality (3.6) with
ψ(z) = log |z − x|, we get
Z i p
p
F ·Y ≥ [Y ] ∧ ∂∂ϕ ≥ sp ν([Y ], ψ) = sp ν(Y, x) ≥ sp .
B(x,ε) π

Proof of Corollary 2. — As L is ample, there exists an integer q (possibly

very large) such that

 KX + qL
is ample,
(12.5) T X ⊗ O (2n + 1)(KX + qL) is nef,

µX (KX + qL) > Cn σ0 .
By lemma 12.2 applied to L′ = KX + qL, we find that F = KX + L′ = 2KX + qL
is very ample and generates s-jets. In particular KX + q2 L is an ample Q-divisor,
and for any p-dimensional subvariety Y ⊂ X we have

p 1 p
KX + (q − 1)L · Y = F + (q/2 − 1)L · Y
2
X p
= 2k−p (q/2 − 1)k F p−k · Lk · Y.
k
0≤k≤p

By the convexity inequality 5.2 (b) and lemma 12.4 we get

k
F p−k · Lk · Y ≥ (F p · Y )1−k/p (Lp · Y )k/p ≥ sp−k µX (L) .

p
p
Hence KX + (q − 1)L · Y ≥ (q/2 − 1)µX (L) + s/2 and

1

µX KX + (q − 1)L ≥ (q − 2)µX (L) + s .
2
Moreover, lemma 12.1 applied to F shows that


T X ⊗ O(KX + nF ) = T X ⊗ O (2n + 1)KX + nqL
is nef. As nq/(2n + 1) ≤ q/2 ≤ q − 1 for q ≥ 2, we find that all properties (12.5)
except perhaps the last one remain valid with q − 1 in place of q :

 KX + (q − 1)L
is ample,
(12.6) T X ⊗ O (2n + 1)(K
X + (q − 1)L)
is nef,

µX KX + (q − 1)L ≥ 21 (q − 2)µX (L) + s .

40
By induction we conclude that (12.6) is still true for the smallest integer q − 1 = m
such that
(q − 2)µX (L) + s = (m − 1)µX (L) + s > 2Cn σ0 .
For this value of m lemma 12.2 implies that 2KX + mL is very ample, resp.
generates s-jets.

Remark 12.7. — If G is a nef line bundle, the proof of corollary 2 can be

applied without modification to show that 2KX + mL + G is very ample, resp.
generates s-jets, for (m−1)µX (L)+s > 2Cn σ0 : indeed, adding G can only increase
the numbers µX (KX + qL + G) occurring in the induction.

Remark 12.8. — The condition (m − 1)µX (L) + s > Cn σ0 is never satisfied

for m = 1. However, it is still possible to obtain a sufficient condition in order that
2KX +L generates s-jets. Indeed, the last step of the iteration shows that 2KX +2L
generates s′ -jets and that µX (KX + L) ≥ s′ /2 if µX (L) + s′ > 2Cn (n + s′ )n .
Choose s′ > 2Cn (n + s)n . Then µX (KX + L) > Cn (n + s)n and we can perform
another iteration to conclude that 2KX + L generates s-jets. Of course, the
corresponding lower bound for µX (L) is extremely large, of the order of magnitude
2
of (2Cn )n+1 (n + s)n .

Remark 12.9. — A numerical computation of 4Cn nn in corollary 2 gives

the following bounds for 2KX + mL to be very ample when L is ample:
n 2 3 4 5 6
m ≥ 10 80 1019 15010 255537

We now list a few immediate consequences of our results, in connection with

some classical questions of algebraic geometry.

Corollary 12.10. — Let X be a projective n-fold of general type with

KX ample. Then mKX is very ample for m ≥ 12 nn .

Corollary 12.11. — Let X be a Fano n-fold, that is, a n-fold such that
−KX is ample. Then −mKX is very ample for m ≥ 12 nn.

Corollaries 12.10 and 12.11 follow easily from corollary 2 applied to

L = ±KX : then we obtain that 2KX + mL is very ample for m > 4Cn nn , and
a numerical check shows that 4Cn nn + 2 < 12 nn for all n. Hence we get pluri-
canonical embeddings Φ : X → IPN such that Φ⋆ O(1) = ±mKX with m = 12 nn .
The image Y = Φ(X) has degree
Z Z

n
n
deg(Y ) = c1 O(1) = c1 ± mKX = mn |KX n
|.
Y X
It can be easily reproved from this that there are only finitely many deformation
types of Fano n-folds, as well as of n-folds of general type with KX ample,

41
 n
corresponding to a given discriminant |KX | ; such results were already known by
the fundamental finiteness theorem of [Ma 72] and [KoM 72]. In the Fano case, it
is conjectured that there is a universal bound (−KX )n ≤ An : if such a universal
bound could be proved, it would become possible to obtain an explicit upper bound
for the number of deformation types of Fano n-folds in any dimension n.(⋆ )

Finally, let L be an ample line bundle over an arbitrary projective n-fold X.

It follows from Mori’s theory that KX + (n + 1)L is always nef (see [Fu 87]). If
KX + tL is nef for some integer t ≥ 0, Fujita conjectures that m(KX + tL) is
spanned for every positive integer m > n + 1 − t. Although such a sharp result
seems very hard to prove, our results allow us to prove that some explicit multiple
of KX + (t + ε)L is very ample for every ε > 0 rational.

Corollary 12.12. — If L is an ample line bundle such
that KX + tL is

nef for some integer t ≥ 0, the line bundle m KX + (t + ε)L is very ample for
every ε > 0 and every integer m > 0 such that mε ∈ IN and mε > 8Cn nn − 2t − 1.

Proof. — First suppose that m = 2p is even. Then either pε or pε − 1/2 is

an integer. Apply corollary 2 to the ample line bundle


L′ = (p − 1) KX + tL + (pε + t)L, resp.


L′ = (p − 1) KX + tL + (pε + t − 1/2)L.
′

find µX (L ) ≥ (pε + t)µX
In the first case, we (L) ≥ pε + t, hence 2KX + 2L′ =
′
2p KX + (t + ε)L is very ample when µX (L ) + 1 ≥ pε + t + 1 > 4Cn nn ; in the
second case, we get the condition pε + t + 1/2 > 4Cn nn and we apply remark 12.7
to conclude that 2KX + 2L′ + L is very ample. When m = 2p + 1 is odd, we argue
in the same way with



L′ = (p − 1) KX + tL + (2p + 1)ε/2 + t L, resp.



L′ = (p − 1) KX + tL + (2p + 1)ε/2 + t − 1/2 L,



and conclude that 2KX + 2L′ + KX + tL or 2KX + 2L′ + KX + (t + 1)L is
very ample when (2p + 1)ε/2 + t + 1/2 > 4Cn nn .

(⋆ ) Added after proof : such bounds have been obtained recently by J. Kollár-
Y. Miyaoka-S. Mori and independently by F. Campana.

42
 References

[AV 65] A. Andreotti and E. Vesentini. — Carleman estimates for the Laplace-
Beltrami equation in complex manifolds, Publ. Math. I.H.E.S., 25 (), 81–130.
[Au 78] Aubin, T. — Equations du type Monge-Ampère sur les variétés kählériennes
compactes, C.R. Acad. Sci. Paris Ser. A 283 (), 119-121 ; Bull. Sci. Math,
102 (), 63-95.
[BT 76] E. Bedford and B.A. Taylor. — The Dirichlet problem for the complex
Monge-Ampère equation, Invent. Math., 37 (), 1–44.
[BT 82] E. Bedford and B.A. Taylor. — A new capacity for plurisubharmonic
functions, Acta Math., 149 (), 1–41.
[BBS 89] M. Beltrametti, A. Biancofiore and A.J. Sommese. — Projective
n-folds of log-general type, I, Trans. Amer. Math. Soc., 314 (), 825–849.
[BFS 89] M. Beltrametti, P. Francia and A.J. Sommese. — On Reider’s
method and higher order embeddings, Duke Math. J., 58 (), 425–439.
[BSk 74] J. Briançon et H. Skoda. — Sur la clôture intégrale d’un idéal de germes de
fonctions holomorphes en un point de Cn , C. R. Acad. Sc. Paris, sér. A, 278
(), 949–951.
[BSo 90] M. Beltrametti and A.J. Sommese. — On k-jet ample line bundles and
Bogomolov’s inequality, Manuscript in preparation, personal communication to
the author (February ).
[Bo 70] E. Bombieri. — Algebraic values of meromorphic maps, Invent. Math., 10
(), 267–287 and Addendum, Invent. Math. 11 (), 163–166.
[Bo 73] E. Bombieri. — Canonical models of surfaces of general type, Publ. Math. IHES,
42 (), 171–219.
[Ca 88] F. Catanese. — Footnotes to a theorem of I. Reider, Proc. Intern. Conf. on
Algebraic Geometry (L’Aquila, June ), Lecture Notes in Math., Vol. 1417,
Springer-Verlag, Berlin, , 67-74.
[CLN 69] S.S. Chern, H.I. Levine and L. Nirenberg. — Intrinsic norms on a
complex manifold, Global Analysis (papers in honor of K.Kodaira), p.119-139,
Univ. of Tokyo Press, Tokyo, .
[De 82a] J.-P. Demailly. — Sur les nombres de Lelong associés à l’image directe d’un
courant positif fermé, Ann. Inst. Fourier (Grenoble), 32 (), 37–66.

[De 82b] J.-P. Demailly. — Estimations L2 pour l’opérateur ∂ d’un fibré vectoriel holo-
morphe semi-positif au dessus d’une variété kählérienne complète, Ann. Sci. Ec.
Norm. Sup., 15 (), 457–511.
[De 82c] J.-P. Demailly. — Courants positifs extrêmaux et conjecture de Hodge, Invent.
Math., 69 (), 347–374.
[De 85] J.-P. Demailly. — Mesures de Monge-Ampère et caractérisation géométrique
des variétés algébriques affines, Mém. Soc. Math. France (N.S.), 19 (),
1–124.
[De 87] J.-P. Demailly. — Nombres de Lelong généralisés, théorèmes d’intégralité et
d’analyticité, Acta Math., 159 (), 153–169.
[De 89] J.-P. Demailly. — Une généralisation du théorème d’annulation de Kawamata-
Viehweg, C. R. Acad. Sci. Paris Sér. I Math., 309 (), 123–126.
[De 90] J.-P. Demailly. — Singular hermitian metrics on positive line bundles, Proc.
Conf. Complex algebraic varieties (Bayreuth, April 2-6, 1990), edited by K. Hulek,

43
 T. Peternell, M. Schneider, F. Schreyer, Lecture Notes in Math., Vol. 1507,
Springer-Verlag, Berlin, .
[De 92] J.-P. Demailly. — Regularization of closed positive currents and Intersection
Theory, J. Alg. Geom., 1 (), 361-409.
[Fe 69] H. Federer. — Geometric measure theory, Springer Verlag, Vol. 153, Berlin,
Heidelberg, New-York, .
[Fu 87] T. Fujita. — On polarized manifolds whose adjoint bundles are not semipositive,
Algebraic Geometry, Sendai, , Adv. Stud. in Pure Math., Vol. 10, North
Holland, Amsterdam, , 167–178.
[Fu 88] T. Fujita. — Problem list, Conference held at the Taniguchi Foundation, Katata,
Japan, August .
[Hö 66] L. Hörmander. — An introduction to Complex Analysis in several variables,
, 2nd edition, North-Holland Math. Libr., Vol. 7, Amsterdam, .
[Hov 79] A.G. Hovanski. — Geometry of convex bodies and algebraic geometry, Uspehi
Mat. Nauk, 34 (4) (), 160–161.
[Ka 82] Y. Kawamata. — A generalization of Kodaira-Ramanujam’s vanishing theorem,
Math. Ann., 261 (), 43–46.
[Ka 84] Y. Kawamata. — The cone of curves of algebraic varieties, Ann. of Math., 119
(), 603–633.
[KMM 87] Y. Kawamata, K. Matsuda and K. Matsuki. — Introduction to the
minimal problem, Algebraic Geometry, Sendai, , Adv. Stud. in Pure Math.,
Vol. 10, North Holland, Amsterdam, , 283–360.
[Ki 84] C.O. Kiselman. — Sur la définition de l’opérateur de Monge-Ampère complexe,
Analyse Complexe, Proceedings of the Journées Fermat (SMF), Toulouse, ,
Lecture Notes in Math., Vol. 1094, Springer-Verlag, Berlin (), 139–150.

[KoM 83] J. Kollár and T. Matsusaka. — Riemann-Roch type inequalities, Amer.

J. of Math., 105 (), 229–252.
[Le 57] P. Lelong. — Intégration sur un ensemble analytique complexe, Bull. Soc.
Math. France, 85 (), 239–262.
[Le 69] P. Lelong. — Plurisubharmonic functions and positive differential forms, Gordon
and Breach, New-York, and Dunod, Paris, .
[Le 72] P. Lelong. — Eléments extrêmaux sur le cône des courants positifs fermés,
Séminaire P. Lelong (Analyse), année 1971/72, Lecture Notes in Math., Vol. 332,
Springer-Verlag, Berlin (), 112–131.
[Ma 72] T. Matsusaka. — Polarized varieties with a given Hilbert polynomial, Amer.
J. of Math., 94 (), 1027-1077.
[Mo 82] S. Mori. — Threefolds whose canonical bundles are not numerically effective, Ann.
of Math., 116 (), 133–176.
[Na 89] A.M. Nadel. — Multiplier ideal sheaves and Kähler-Einstein metrics of positive
scalar curvature, Proc. Nat. Acad. Sci. U.S.A., 86 (), 7299–7300 and
Annals of Math., 132 (), 549-596.
[Oh 88] T. Ohsawa. — On the extension of L2 holomorphic functions, II, Publ. RIMS,
Kyoto Univ., 24 (), 265–275.
[Ree 61] D. Rees. — Transforms of ideals, and a theorem on multiplicities of ideals, Proc.
Cambridge Phil. Soc., 57 (), 8–17.
[Rei 88] I. Reider. — Vector bundles of rank 2 and linear systems on algebraic surfaces,
Ann. of Math., 127 (), 309–316.

44
[Sa 88] F. Sakai. — Reider-Serrano’s method on normal surfaces, Proc. Intern. Conf.
on Algebraic Geometry (L’Aquila, June ), Lecture Notes in Math., Vol. 1417,
Springer-Verlag, Berlin, , 301-319.
[Se 57] J.P. Serre. — Algèbre locale, Multiplicités, Cours au Collège de France (),
Lecture Notes in Math. Vol. 11, Springer-Verlag, Berlin, .
[Siu 74] Y.T. Siu. — Analyticity of sets associated to Lelong numbers and the extension
of closed positive currents, Invent. Math., 27 (), 53–156.
[Sk 72a] H. Skoda. — Sous-ensembles analytiques d’ordre fini ou infini dans Cn , Bull. Soc.
Math. France, 100 (), 353–408.
[Sk 72b] H. Skoda. — Applications des techniques L2 à la théorie des idéaux d’une algèbre
de fonctions holomorphes avec poids, Ann. Scient. Ec. Norm. Sup. 4e Série, 5
(), 545–579.
[Sk 75] H. Skoda. — Estimations L2 pour l’opérateur ∂ et applications arithmétiques,
Séminaire P. Lelong (Analyse), année 1975/76, Lecture Notes in Math., Vol. 538,
Springer-Verlag, Berlin (), 314–323.
[Sk 82] H. Skoda. — Prolongement des courants positifs fermés de masse finie, Invent.
Math., 66 (), 361–376.
[Te 79] B. Teissier. — Du théorème de l’index de Hodge aux inégalités isopérimétriques,
C. R. Acad. Sc. Paris, sér. A, 288 ( Jan ), 287–289.
[Te 82] B. Teissier. — Bonnesen-type inequalities in algebraic geometry, Sem. on Diff.
Geom. edited by S.T. Yau, , Princeton Univ. Press, 85–105.
[Th 67] P. Thie. — The Lelong number of a point of a complex analytic set, Math.
Annalen, 172 (), 269–312.
[Vi 82] E. Viehweg. — Vanishing theorems, J. Reine Angew. Math., 335 (), 1–8.
[Yau 78] S.T. Yau. — On the Ricci curvature of a complex Kähler manifold and the
complex Monge-Ampère equation I, Comm. Pure and Appl. Math., 31 (),
339-411.

(December 11, 1990, revised on September 11, 1991)

45