the dirichlet problem for complex monge … dirichlet problem for complex monge-ampere` equations...
TRANSCRIPT
THE DIRICHLET PROBLEM FOR COMPLEX MONGE-AMPEREEQUATIONS AND APPLICATIONS
BO GUAN
Abstract. We are concerned with the Dirichlet problem for complex Monge-Ampere equations and their applications in complex geometry and analysis.2000 Mathematical Subject Classification: 35J65, 35J70, 53C21, 58J10, 58J32,32W20, 32U05, 32U35, 32Q15.Key words and phrases: Complex Monge-Ampere equations, Dirichlet problem,plurisubharmonic solutions, CKNS Theorem, a priori estimates, Kahler manifolds,Kahler potentials, Calabi-Yau Theorem, Donaldson conjecture, intrinsic norms,Chern-Levine-Nirenberg conjecture, pluricomplex Green functions.
1. Introduction
The complex Monge-Ampere equation is at the center of fully nonlinear elliptic
equation theory, and is of fundamental importance in complex geometry and analysis,
especially in Kahler geometry and pluripotential theory.
In a series of work, Bedford and Taylor (e.g. [4], [6]) established the foundation
of weak solutions based on the concept of positive currents and the Chern-Levine-
Nirenberg inequality [16]. The theory has developed enormously over the years with
contributions from many researchers; we refer the reader to, e.g., [31] and [32] for
background and references to the subject and pluripotential theory. In [48], S.-T.
Yau proved fundamental existence theorems of classical solutions to the complex
Monge-Ampere equation on compact Kahler manifolds, which are the centerpiece of
his proof of the Calabi conjecture.
The classical solvability of the Dirichlet problem was established by Caffarelli,
Kohn, Nirenberg and Spruck (CKNS) [13] for strongly pseudoconvex domains in
Cn. Later on the author [23] extended their results to general domains under the
assumption of existence of a C2 strictly plurisubharmonic subsolution.
The author was supported in part by NSF grants.Department of Mathematics, Ohio State University, Columbus, OH 43210. [email protected].
1
2 BO GUAN
In these notes we present the results of CKNS [13], with modifications by the
author [23] to general domains. As applications of the main theorem (Theorem 2.7)
we will discuss briefly some problems in complex analysis and geometry, most of which
involve solving the homogeneous complex Monge-Ampere equation in domains that
are not pseudoconvex. The notes are organized as follows.
In Section 2 we discuss briefly the basic properties of the complex Monge-Ampere
operator and the continuity method which reduces, at least in some important special
cases, the proof of existence results to deriving a priori estimates for the perspective
solutions. The materials in this section are all well known. At the end of the section
we state the theorem of CKNS (Theorem 2.7).
In Section 3 we present the C2 a priori estimates for plurisubharmonic solutions of
the Dirichlet problem in bounded smooth domains in Cn. This is roughly divided into
three steps: the gradient estimates, the boundary, and global estimates for second
derivatives. Most part of the proof in this section can be found in [13], [23] and [24]
(with minor modifications), except the estimate for double-normal derivative on the
boundary which was first given in a lecture by the author at NCTS, Taiwan, in 1999.
This section also contains some results that are needed in Section 6.
Section 4 concerns the complex Monge-Ampere equation on Kahler manifolds. This
is a broad subject with extensive research activities and far-reaching results. In this
section we restrict ourselves to a brief presentation of existence theorems of Yau [48]
(Theorems 4.1 and 4.3) and an analogue of Theorem 2.7 for the Dirichlet problem
(Theorem 4.2). As an application we discuss Donaldson’s conjectures [18] on geodesics
in the Mabuchi space of Kahler potentials and outline the proof of the existence of
(almost) C1,1 solutions in Chen [15].
In Section 5 we describe the intrinsic norms introduced by Chern, Levine and
Nirenberg [16] and the proof of the Chern-Levine-Nirenberg conjecture by Pengfei
Guan [27], [28]. As we only received [28] recently, we shall mainly focus on results
in [27]. Finally, in Section 6 we discuss the C1,1 regularity of pluricomplex Green
functions which are related to some special Dirichlet problems for the homogeneous
complex Monge-Ampere equation.
These notes are based in part on series of lectures that I gave for graduate students
at University of Science and Technology of China (UCTC) and Harbin Institute of
Technology (HIT) in the summer of 2006. I would like to thank Professors Xiuxiong
COMPLEX MONGE-AMPERE EQUATIONS 3
Chen, Xinan Ma, Yongqiang Fu, Boying Wu, and Departments of Mathematics of
both universities for their hospitality. Due to the limitation in space I have left out
a substantial part of materials that were covered in the lectures, and many more
which are important in the area and interesting to myself. (We should remark that
this article is not intended to be a survey.) I also wish to thank Joel Spruck for
stimulating discussions and his constant encouragement during the writing of these
notes, and over the years. Needless to say, I learned much of the subject from him.
(Of course I am solely responsible for any mistakes in these notes.) This article is
dedicated to Professor Guangchang Dong on the occasion of his 80th birthday, and
for his contributions to fully nonlinear partial differential equations.
2. The Dirichlet problem in Cn
The classical complex Monge-Ampere equation in Cn takes the form
(2.1) det∂2u
∂zj∂zk
= ψ.
This equation is elliptic for a C2 strictly plurisubharmonic solution u, i.e., the complex
Hessian matrix of u
HessC(u) =( ∂2u
∂zj∂zk
)is positive definite. In this case the function ψ on the right hand side of (2.1) is
necessarily positive; equation (2.1) becomes degenerate when ψ is only non-negative.
When ψ ≡ 0, (2.1) is referred as the homogeneous complex Monge-Ampere equation
which is well defined on complex manifolds and arises in many problems in complex
analysis and geometry.
We shall consider the Dirichlet problem for equation (2.1): Let Ω be a bounded
domain in Cn with smooth boundary ∂Ω, and ϕ ∈ C∞(∂Ω). We seek a plurisubhar-
monic solution u of equation (2.1) with boundary value u = ϕ on ∂Ω. The main goal
of this section is to explain the continuity method approach and how it reduces the
problem to a priori estimates for the perspective solutions.
4 BO GUAN
2.1. Some notation and basic properties. Let z = (z1, . . . , zn) be complex coor-
dinates in Cn, zj = xj +√−1yj. Recall that for a C2 function u defined on Cn,
(2.2)∂u
∂zj
=1
2
( ∂u
∂xj
−√−1
∂u
∂yj
),∂u
∂zj
=1
2
( ∂u
∂xj
+√−1
∂u
∂yj
).
When there is no confusion we shall write
uj = uzj=∂u
∂zj
uj = uzj=∂u
∂zj
ujk = uzj zk=
∂2u
∂zj∂zk
etc.
Note that
(2.3) uzj zk=
1
4
[(uxjxk
+ uyjyk) +
√−1(uxjyk
− uyjxk)].
Define
|∇u|2 =∑
j
|uj|2 =∑
j
ujuj =1
4
∑j
(u2xj
+ u2yj
),
∆u =∑
j
ujj =1
4
∑j
(uxjxj+ uyjyj
).
The following notation also appears in literature frequently:
∂u =∑
j
uzjdzj, ∂u =
∑j
uzjdzj,
du = ∂u+ ∂u, dcu =√−1(∂u− ∂u).
Therefore, d2 = ∂∂ + ∂∂ = ∂2 = ∂2 = 0 and
ddcu = 2√−1∂∂u = 2
√−1
∑j,k
uzj zkdzj ∧ dzk.
The complex Monge-Ampere operator (ddcu)n is defined as
(ddcu)n = 4nn! det(ujk) dV
where
dV =(√−1
2
)n
dz1 ∧ dz1 ∧ · · · ∧ dzn ∧ dzn
is the volume form of Cn; the definition of (ddcu)n extends to general locally bounded
plurisubharmonic functions ([6]).
COMPLEX MONGE-AMPERE EQUATIONS 5
We call the matrix
HessC(u) = (ujk)
the complex Hessian of u, while the real Hessian matrix of u is
HessR(u) =
[uxjxk
uxjyk
uyjxkuyjyk
].
Following the literature we shall use (ujk) to denote the inverse matrix of (ujk) when
it is invertible, i.e. ujkulk = δjl.
We next look at some important properties of the complex Monge-Ampere operator.
Lemma 2.1. Let z 7→ w be a holomorphic change of variables. Then
detuzj zk= | det(wz)|2 detuwjwk
.
It follows that the homogeneous complex Monge-Ampere equation
(2.4) det(uzj zk) = 0
is invariant under holomorphic changes of variables. This is one of the reasons the
equation is of interest in complex analysis and geometry. In particular, it is globally
well defined on complex manifolds.
Lemma 2.2. A differential operator u 7→ F (ujk) is elliptic (with respect to the real
variables x1, . . . , xn, y1, . . . , yn) if and only if the matrix( ∂F
∂uzj zk
)is positive definite. In particular, the complex Monge-Ampere operator u 7→ detujk is
elliptic in u if and only if the complex Hessian (ujk) is positive (or negative) definite.
This follows from the formula, by (2.3),
(2.5)
[∂F
∂uxjxk
∂F∂uxjyk
∂F∂uyjxk
∂F∂uyjyk
]=
1
4
∂F∂uzj zk
+ ∂F∂uzkzj
√−1
(∂F
∂uzj zk− ∂F
∂uzkzj
)−√−1
(∂F
∂uzj zk− ∂F
∂uzkzj
)∂F
∂uzj zk+ ∂F
∂uzkzj
.A function u in Cn is called (strictly) plurisubharmonic if it is (strictly) subharmonic
on every complex line (or equivalently, in each complex variable zj). A C2 function u is
strictly plurisubharmonic if and only if its complex Hessian (ujk) is positive definite.
We shall use PSH(Ω) to denote the collection of plurisubharmonic functions in a
domain Ω ⊂ Cn, while SPSH(Ω) denotes the collection of strictly plurisubharmonic
functions in Ω.
6 BO GUAN
Lemma 2.3. Let u be a strictly plurisubharmonic function and set F = log detujk.
Then
(2.6)
F jk :=∂F
∂uzj zk
= ujk,
F jk,pq :=∂2F
∂uzj zk∂uzpzq
= −ujqupk.
Consequently, log detujk is concave in ujk, i.e.
(2.7) F jk,pqξjkξpq = −ujqupkξjkξpq ≤ 0, ∀ ξjk ∈ Cn×n.
As we shall see, the concavity of log detujk is very important in the theory of
complex Monge-Ampere equations. Another fact that is often used is the concavity
of (detujk)1n . Indeed, for any increasing, concave function φ, φ((detujk)
1n ) is concave
for plurisubharmonic functions u.
We have the following maximum principle for complex Monge-Ampere operator;
see [13].
Lemma 2.4. Let Ω be a bounded domain. If u, v ∈ C2(Ω) are plurisubharmonic
functions with detujk ≥ det vjk in Ω and u ≤ v on ∂Ω, then u ≤ v in Ω.
2.2. The continuity method. Let Ω be a bounded domain in Cn with C∞ boundary
∂Ω, and consider the Dirichlet problem for complex Monge-Ampere equations
(2.8)detuzj zk
= ψ(z, u,Du) in Ω,
u = ϕ on ∂Ω
where ψ and ϕ are real-valued smooth functions, ψ > 0.
Suppose that there exists a strictly plurisubharmonic subsolution u ∈ C2,α(Ω) of
(2.8), that is,
(2.9)detuzj zk
≥ ψ(z, u,Du) in Ω,
u = ϕ on ∂Ω.
For 0 ≤ t ≤ 1, set
(2.10)detuzj zk
= ψt(z, u,Du) in Ω,
u = ϕ on ∂Ω
where
ψt(z, u,∇u) = tψ(z, u,Du) + (1− t) detuzj zk.
COMPLEX MONGE-AMPERE EQUATIONS 7
We wish to show that (2.10) has a strictly plurisubharmonic solution in C2,α(Ω) for
every 0 ≤ t ≤ 1. Define
S =t ∈ [0, 1] : (2.10) is solvable in C2,α(Ω) ∩ SPSH(Ω)
.
Obviously S is nonempty since 0 ∈ S with u being the unique solution of (2.10) when
t = 0. Our goal is to show that under suitable conditions S is both open and closed
in [0, 1], and therefore is equal to [0, 1]. This approach is known as the “continuity
method”.
The openness of S usually follows from the implicit function theorem for Banach
spaces, while the closedness reduces to a priori C2,α estimates.
Proposition 2.5 (Openness of S). Suppose ψu ≥ 0. Then S is open in [0, 1].
This follows from the implicit function theorem for Banach spaces and the Schauder
theory for linear elliptic equations.
Theorem 2.6 (Implicit Function Theorem). Let X, Y, Z be Banach spaces and
F : U × V → Z a continuous map and continuously differentiable with respect to
x ∈ X where U , V are open sets in X, Y respectively. Suppose that F (x0, y0) = 0 for
some x0 ∈ U , y0 ∈ V , and that the Frechet derivative
Fx(x0, y0) : X → Z
is an isomorphism. Then there exist balls Bε(x0) ⊂ U , Bδ(y0) ⊂ V and exactly one
map η : Bδ(y0) → Bε(x0) such that η(y0) = x0 and
F (η(y), y) = 0, ∀ y ∈ Bδ(y0).
The map η is continuous.
Proof of Proposition 2.5. Let
X = C2,α0 (Ω) ≡ v ∈ C2,α(Ω) : v|∂Ω = 0, Y = R, Z = Cα(Ω)
and
U = X ∩ SPSH(Ω).
Consider
F : (v, t) 7→ log detujk − logψt(z, u,Du), u = u+ v.
Suppose F (v0, t0) = 0 and let u0 = u + v0. The Frechet derivative Fv(v0, t0) is equal
to
(2.11) L = ujk0 ∂j∂k − (logψt0)Du ·D − (logψt0)u
8 BO GUAN
which is a linear elliptic operator with coefficients in Cα(Ω).
By the Schauder theory, the Dirichlet problem for the linear elliptic equation
(2.12)
Lw = f in Ω
w = 0 on ∂Ω
has a unique solution w ∈ C2,α(Ω for any f ∈ Cα(Ω). Moreover,
(2.13) |w|C2,α(Ω) ≤ C|f |Cα(Ω).
This means that L : C2,α0 (Ω) → Cα(Ω) is invertible and L−1 : Cα(Ω) → C2,α
0 (Ω) is
bounded. By Theorem 2.6, there exists δ > 0 and unique v = v(t) ∈ C2,α0 (Ω) for all
t ∈ [0, 1], |t− t0| < δ, such that F (v(t), t) = 0. This proves Proposition 2.5.
The closedness of S reduces to a priori C2,α estimates, which are uniform in t ∈[0, 1], for perspective solutions of (2.10). This can be seen as follows.
We wish to show that if tj ∈ S and tj → t0 as j → ∞ then t0 ∈ S. Let ut denote
the unique solution of (2.10) in C2,α(Ω)∩ SPSH(Ω) for t ∈ S and let vt = ut− u. We
need to show at least there is a subsequent of vtj converging in C2,α(Ω) to a function
v0 ∈ U and therefore F (v0, t0) = 0. Applying the Ascoli-Arzela Theorem, to show
that vtj has convergent subsequence in C2,α(Ω) it suffices to establish the estimate
(2.14) |utj |C2,α′(Ω) ≤ C, independent of j.
Consequently, the solvability of the Dirichlet problem (2.8) reduces to the a priori
estimate
(2.15) |u|C2,α(Ω) ≤ C
for strictly plurisubharmonic solutions.
2.3. CKNS Theorem. In the non-degenerate case (ψ > 0), a fundamental existence
theorem of classical solution for the Dirichlet problem (2.8) was proved by Caffarelli,
Kohn, Nirenberg and Spruck [13] for strongly pseudoconvex domains. It was later
extended to general domains by the author in [23] under the assumption of existence
of subsolutions, following the corresponding work [26], [22] on real Monge-Ampere
equations in general domains.
Theorem 2.7. Let Ω be a bounded smooth domain in Cn, and ϕ, ψ = ψ(z, u, p) real-
valued smooth functions, ψ(z, u, p) > 0 for all z ∈ Ω, u ∈ R and p ∈ Cn. Suppose
COMPLEX MONGE-AMPERE EQUATIONS 9
either ψ = ψ(z, u), i.e., it does not depend on the gradient of u, or for any M > 0
there exists CM > 0 such that
(2.16) −ψu, |ψzj|, |ψpj
| ≤ CMψ1− 1
n , ∀ (z, u, p) ∈ Ω× [−M,M ]× Cn.
The Dirichlet problem (2.8) admits a strictly plurisubharmonic solution u ∈ C∞(Ω),
provided that there exists a strictly plurisubharmonic subsolution u ∈ C2(Ω) satisfying
(2.9).
Remark 2.8. For strongly pseudoconvex domains, it was shown in [13] that condition
(2.16) implies the existence of strictly plurisubharmonic subsolutions. It is, however,
difficult to formulate for nonpseudoconvex domains general conditions which guaran-
tee the existence of subsolutions. Therefore, it is important in applications to find
natural conditions that one can use to construct subsolutions. As we shall see in Sec-
tions 4-6, in many interesting problems in geometry and analysis it is possible to find
such conditions. In the proof of Theorem 2.7, condition (2.16) is only used in deriving
the gradient estimates; see Subsection 3.2. We also note here that (2.16) should be
included in the assumptions of Theorem 1.1 in [23]. This was also pointed out to the
author by Zhuoliang Hou to whom the author wishes to express his gratitude.
Remark 2.9. Note that we do not assume ψu ≥ 0 in Theorem 2.7. Therefore we can
not apply the continuity method directly in the proof of Theorem 2.7. Nevertheless,
the proof of existence reduces to establishing the C2,α estimate (2.15). This can
be done following the argument in [14] for the real Monge-Ampere equation. In
the original CKNS theorem [13], the assumption ψu ≥ 0 was also used in deriving
estimates for the second derivatives. Here we modify their argument in order to drop
the assumption.
In the next section we shall derive the C2 estimate
(2.17) ‖u‖C2(Ω) ≤ C.
Once this is established, (2.8) becomes uniformly elliptic. We therefore obtain (2.15)
by the well known Evans-Krylov Theorem [19], [33], [34] (see also [13] and [12]).
Higher order estimates then follows from the classical Schauder theory for linear
uniform elliptic equations.
10 BO GUAN
3. A priori estimates
Let u ∈ C4(Ω) be a strictly pluri-subharmonic solution of (2.8) with u ≥ u. Our
goal is to derive the a priori bound (2.17). Without loss of generality we shall assume
ϕ ∈ C4(Ω).
3.1. The linearized operator. Rewrite equation (2.8) in the form
(3.1) log detujk = logψ(z, u,Du) ≡ f(z, u,Du) in Ω.
Let L be the linearized operator:
Lv = ujkvjk − fpj(z, u,Du)vj − fpj
(z, u,Du)vj.
We have
(3.2) Lu =n− fpj(z, u,Du)uj − fpj
(z, u,Du)uj.
and, for a function η ∈ C2(R),
(3.3) Lη(u) = η′Lu+ η′′ujkujuk.
Let
τ =n∑
j=1
(aj∂zj+ bj∂zj
)
be a first order linear operator with constant (complex) coefficients. Differentiating
equation (3.1) twice, we find
(3.4) L(uτ ) = fτ + fuuτ
and
(3.5) ujkujkτ τ − ujmulkujkτulmτ = (f)τ τ ,
where
(3.6)
(f)τ τ = fpjuτ τj + fpj
uτ τ j
+ fτ τ + fuuuτuτ + fuuτ τ + fpj pk(ujτuτ k + ujτukτ )
+ 2Refτuuτ + (fτpj
+ fupjuτ )ujτ
+ (fτ pk+ fupj
uτ )ujτ + fpjpkujτukτ
and
fτ = ajfzj+ bjfzj
, etc.
COMPLEX MONGE-AMPERE EQUATIONS 11
Therefore,
(3.7) Luτ τ = ujmulkujkτulmτ + (f)τ τ − fpjuτ τj − fpj
uτ τ j.
3.2. C1 estimates. We first observe that
(3.8) u ≤ u ≤ h in Ω,
where h is the harmonic extension of ϕ to Ω. Thus
maxΩ|u|+ max
∂Ω|∇u| ≤ C.
We next derive the global gradient estimate
(3.9) |∇u| ≤ C in Ω
under assumption (2.16).
Lemma 3.1. For any φ ∈ C2(Ω),
(3.10) e−φL(eφ|∇u|2) ≥ (2fu + Lφ)|∇u|2 + ∆u+ 2Re(fzk+ φk)uk.
Proof. This follows from straightforward calculations. First,
(3.11) e−φLeφ = Lφ+ ujkφjφk.
Next,
(3.12) (|∇u|2)i = (ukuk)i = ukiuk + ukuki,
(3.13) (|∇u|2)ij = (ukiuk + ukuki)j = ukijuk + ukukij + ukiukj + ukjuki.
Differentiating equation (3.1) we obtain
(3.14) uij(ukijuk + ukukij) = 2fu|∇u|2 + 2Refzkuk+ fpj
(|∇u|2)j + fpj(|∇u|2)j.
Thus, from (3.13) and (3.14),
(3.15) L(|∇u|2) = 2fu|∇u|2 + ∆u+ uijukiukj + 2Refzkuk.
We also note thatujk(|∇u|2)jφk = ujk(ululj + ululj)φk
= ujφj + uijuljulφk.
By Cauchy-Schwarz inequality,
2Reuijuljulφk| ≤ ujkuljulk + |∇u|2ujkφjφk.
12 BO GUAN
Finally, plugging these into
e−φL(eφ(u)|∇u|2) = |∇u|2(Lφ+ ujkφjφk) + 2Reujk(|∇u|2)jφk+ L(|∇u|2),
we derive (3.10).
Corollary 3.2. Under assumption (2.16), (3.9) holds.
Proof. Let φ = eλ|z|2 ≥ 1. We compute
φ−1Lφ =λ2ujkzjzk + λ( ∑
ujj − 2Refpjzj
)≥λ2ujkzjzk + λ
( ∑ujj − Aψ−
1n
)for some constant A > 0. Let µ1 ≤ . . . ≤ µn be the eigenvalues of ujk. One has
(3.16)∑
ujj =∑
µj ≥ n(µ1 · · ·µn)1n = n(detujk)
1n = nψ−
1n .
We consider two cases.
Case (i). µ1 ≤ (1 + A)1−nψ−1n . In this case, µn ≥ (1 + A)ψ−
1n and therefore,∑
ujj − Aψ−1n ≥ ψ−
1n +
n−1∑1
ujj.
Consequently, by Lemma 3.1 and assumption (2.16),
(3.17)
e−φL(eφ|∇u|2) ≥∆u− C(1 + λψ−1n )|∇u|
+((λφ− 2C)ψ−
1n + λφ
∑j<n
µj
)|∇u|2
≥(λ− 2C)|∇u|2 − Cλ|∇u|ψ−1n + (
√λ− C)|∇u|.
Here we used the fact that
λ|∇u|2∑j<n
µj + ∆u = λ|∇u|2∑j<n
µj +∑
µ−1j ≥ 2(n− 1)
√λ|∇u|.
Case (ii). µ1 ≥ (1 + A)1−nψ−1n . We have
ujkzjzk ≥ (1 + A)1−nψ−1n |z|2 ≥ (1 + A)1−nψ−
1n
if we assume |z| ≥ 1 for z ∈ Ω without loss of generality. It follows, similarly to
(3.17), that
(3.18) e−φL(eφ|∇u|2) ≥ ((1+A)1−nλ2−Aλ)|∇u|2−Cλ|∇u|ψ−1n +(
√λ−C)|∇u|.
COMPLEX MONGE-AMPERE EQUATIONS 13
From (3.17) and (3.18) we see that in both cases,
L(eφ|∇u|2) ≥ 0 for λ sufficiently large.
By the maximum principle,
maxΩ|∇u|2 ≤ max
Ω(eφ|∇u|2) = max
∂Ω(eφ|∇u|2) ≤ Cmax
∂Ω|∇u|2,
proving (3.9).
Applying Lemma 3.1 to φ = η(u) where η is a convex function, by (3.3) and (3.2)
we find
(3.19)
e−η(u) L(eη(u)|∇u|2) ≥ ∆u+ 2Refzkuk
+ 2fu + (n+ 2− fpjuj − fpj
uj)η′ + η′′uijuiuj|∇u|2
≥ 2fu + (n+ 2− fpjuj − fpj
uj)η′ + 2
√η′′|∇u|2 − 2|∇u|
∑|fzj
|.
In the last inequality we used
η′′uijuiuj + |∇u|−2∆u ≥ 2√η′′.
which follows from the Cauchy-Schwarz inequality.
Corollary 3.3. Suppose that η is a convex function satisfying
α|∇u| −∑
|fzj| > 0 in Ω
where
α ≡ fu + (n+ 2− fpjuj − fpj
uj)η′ +
√2η′′.
Then
(3.20) maxΩ
eη(u)|∇u| = max∂Ω
eη(u)|∇u|.
Proof. In (3.19) replace η by 2η and apply the maximum principle.
Using Corollary 3.3 we can derive gradient estimates under some different condi-
tions than (3.9).
Corollary 3.4. Suppose for any M > 0 there exists CM > 0 such that
(3.21)ψu ≥ −CMψ,
∑|ψzj
| ≤ CMψ(1 + |p|),∑
(ψpjpj + ψpj
pj) ≤ (n+ 2)ψ,
∀ (z, u, p) ∈ Ω× [−M,M ]× Cn.
Then
(3.22) maxΩ|∇u| ≤ C(1 + max
∂Ω|∇u|).
14 BO GUAN
Proof. Let η(u) = 4C2M(M + u)2. Then η′ = 8C2
M(u + M) ≥ 0 and η′′ = 8C2M .
Therefore
α ≥ fu +√
2η′′ ≥ 3CM
and conditions of Corollary 3.3 are satisfied or |∇u| ≤ 1 where η(u)|∇u| achieves its
maximum. By (3.20) we obtain (3.22).
Remark 3.5. Functions of the form
ψ(z, u, p) = g(z, u)(1 + |p|2)β, β ≤ n+ 2
2, g > 0, log g ∈ C1
satisfies condition (3.21) but for β > n it does not satisfy (2.16).
3.3. Boundary estimates for second derivatives. In this subsection we derive a
priori second derivative estimates on the boundary
(3.23) max∂Ω
|D2u| ≤ C.
We assume
(3.24) |u|+ |Du| ≤ K in Ω.
Set
(3.25) ψ ≡ min|u|+|p|≤K, z∈Ω
ψ(z, u, p) > 0, ψ ≡ max|u|+|p|≤K, z∈Ω
ψ(z, u, p).
(a) Consider a boundary point on ∂Ω. We may assume it to be the origin of Cn
and choose coordinates z1, . . . , zn such that the positive xn axis is the interior normal
direction to ∂Ω at 0. For convenience we set
t1 = x1, t2 = y1, . . . ,
t2n−3 = xn−1, t2n−2 = yn−1,
t2n−1 = yn, t2n = xn
and
t′ = (t1, . . . , t2n−1).
Near the origin ∂Ω may be represented as a graph
(3.26) xn = ρ(t′) =1
2
∑α,β<2n
Bαβtαtβ +O(|t′|3).
COMPLEX MONGE-AMPERE EQUATIONS 15
Since (u− u)(t′, ρ(t′)) = 0, we have
(3.27) (u− u)tαtβ(0) = −(u− u)xn(0)Bαβ, α, β < 2n.
It follows that
(3.28) |utαtβ(0)| ≤ C, α, β < 2n.
(b) To estimate utαxn(0) for α ≤ 2n, we will employ a barrier function of the form
(3.29) v = (u− u) + td− Nd2
2,
where t, N are positive constants to be determined and d is the distance function to
∂Ω. We may take δ > 0 small enough so that d is smooth in Ωδ = Ω ∩Bδ(0). Recall
that u ∈ C2(Ω) is strictly plurisubharmonic. Therefore there exists ε > 0 such that
(3.30) ujk ≥ εI on Ω.
Lemma 3.6. For N sufficiently large and t, δ sufficiently small,
Lv ≤ − ε4
(1 +
∑ukk
)in Ωδ,
v ≥ 0 on ∂Ωδ.
Proof. By (3.30) we have
ujk(ujk − ujk) ≤ n− ε∑
ukk.
It follows that
(3.31) L(u− u) ≤ C0 − ε∑
ukk.
Obviously,
|Ld| ≤ C1
(1 +
∑ukk
)for some constant C1 > 0 under control. Thus
Lv ≤ C0 + C1(t+Nd) + C1(t+Nd)− ε∑
ukk −Nujkdjdk in Ωδ.
Let λ1 ≤ · · · ≤ λn be the eigenvalues of ujk. We have∑ukk =
∑λ−1
k and
(3.32) ujkdjdk ≥1
λn
since |Dd| ≡ 1. By the inequality for arithmetic and geometric means,
ε
4
∑ukk +
N
λn
≥ nε
4(Nλ−1
1 · · ·λ−1n )
1n ≥ nε
4(ψ)1/nN
1n ≡ c1N
1n .
16 BO GUAN
We now fix t > 0 sufficiently small and N large so that C1t ≤ ε4
and c1N1/n ≥ C0+ε.
Consequently,
Lv ≤ − ε4
(1 +
∑ukk
)in Ωδ
if we require δ to satisfy C1Nδ ≤ ε4
in Ωδ.
On ∂Ω ∩Bδ(0) we have v = 0. On Ω ∩ ∂Bδ(0),
v ≥ td−Nd2 ≥ (t−Nδ)d ≥ 0
if we require, in addition, Nδ ≤ t.
Remark 3.7. Lemma 3.6 was first proved in Guan [23]. A preliminary version can
be found in the author’s thesis [21]. It is parallel to results in Guan [22] for real
Monge-Ampere equations in Rn and on general Riemannian manifolds, which improve
the corresponding results of Hoffman-Rosenberg-Spruck [30], Guan-Spruck [26] and
Guan-Li [25].
Lemma 3.8. Let h ∈ C2(Ω ∩Bδ(0)). Suppose that h satisfies
h ≤ C0|z|2 on (∂Ω) ∩Bδ(0), h(0) = 0
and
−Lh ≤ C1
(1 +
∑ujj
)in Ω ∩Bδ.
Then hν(0) ≤ C, where ν is the interior unit normal to ∂Ω, and C depends on ε−1,
C0, C1, |h|C0(Ω∩Bδ(0)) and |u|C1(Ω).
Proof. By Lemma 3.6, Av +B|z|2 − h ≥ 0 on ∂(Ω ∩Bδ(0)) and
L(Av +B|z|2 − h) ≤ 0 in Ω ∩Bδ
when A B are sufficiently large. By the maximum principle,
Av +B|z|2 − h ≥ 0 in Ω ∩Bδ.
Consequently,
Avν(0)− hν(0) = Dν(Av +B|z|2 − h)(0) ≥ 0
since Av +B|z|2 − h = 0 at the origin.
(c) We first apply Lemma 3.8 to estimate utαxn(0) for α < 2n. Define
T =∂
∂tα+ ρtα
∂
∂xn
.
COMPLEX MONGE-AMPERE EQUATIONS 17
We have (see [13])
L(±T (u− ϕ)− (uyn − ϕyn)2) ≤ C(1 +
∑F kk
)in Ω ∩Bδ(0).
By (3.24),
|T (u− ϕ)|+ (uyn − ϕyn)2 ≤ C in Ω ∩Bδ(0).
From the fact that u− ϕ = 0 and that T is a tangential operator on ∂Ω, we see that
T (u− ϕ) = 0 on ∂Ω ∩Bδ(0),
and
(3.33) (uyn − ϕyn)2 ≤ C|z|2, on ∂Ω ∩Bδ(0)
by (3.26) since
(u− ϕ)yn = −(u− ϕ)xnρyn , on ∂Ω ∩Bδ(0).
Applying Lemma 3.8 to h = (uyn − ϕyn)2 ± T (u− ϕ), we obtain
(3.34) |utαxn(0)| ≤ C, α < 2n.
(d) It remains to establish the estimate
(3.35) |uxnxn(0)| ≤ C.
Since we have already derived
(3.36) |utαtβ(0)|, |utαxn(0)| ≤ C, α, β < 2n,
it suffices to prove
(3.37) 0 ≤ unn(0) = uxnxn(0) + uynyn(0) ≤ C.
Expanding detujk:
detujk(0) = aunn(0) + b
where
a = detuαβ(0)|1≤α,β≤n−1and b is bounded in view of (3.36). By equation (2.8) we only have to derive an a
priori positive lower bound for a, which is equivalent to
(3.38)∑
α,β<n
uzαzβ(0)ξαξβ ≥ c0|ξ|2, ∀ ξ ∈ Cn−1
for a uniform constant c0 > 0.
Proposition 3.9. There exists c0 = c0(ψ−1, ϕ, u, ∂Ω) > 0 such that (3.38) holds.
18 BO GUAN
Proof. Let σ ∈ C4 be a defining function of Ω, i.e,
Ω = σ < 0, ∂Ω = σ = 0, Dσ|∂Ω 6= 0.
For convenience we assume |Dσ| = 1 on ∂Ω, without losing generality. Recall that
for a point p ∈ ∂Ω, the real and complex tangent spaces of ∂Ω at p are defined by
Tp∂Ω =ξ ∈ Cn : Re
∑σzj
(p)ξj = 0
and
TCp ∂Ω =
ξ ∈ Cn :
∑σzj
(p)ξj = 0,
respectively. Consider
m0 = minp∈∂Ω
minξ∈T C
p ∂Ω,|ξ|=1ujk(p)ξj ξk
and assume that m0 is attained at p = 0 and ξ = (1, 0, · · · , 0). We choose coordinates
z = (z1, · · · , zn) as before so that the positive xn axis is in the inner normal direction
to ∂Ω at 0. We need to show
(3.39) m0 = u11(0) ≥ c0 > 0.
Near the boundary of Ω, write
u− u = hσ.
We have
(u− u)xn = hxnσ + hσxn
and
(u− u)jk = hjkσ + hσjk + 2 Rehjσk.Therefore,
(u− u)xn(0) = −h(0)
and
(3.40) (u− u)αβ(0) = h(0)σαβ(0) α, β < n.
In particular,
(3.41) u11(0) = u11(0)− (u− u)xn(0)σ11(0).
We assume u11(0) ≤ 12u11(0); otherwise we are done. By (3.41),
(3.42) (u− u)xn(0)σ11(0) ≥ 1
2u11(0).
COMPLEX MONGE-AMPERE EQUATIONS 19
It follows from (3.24) that
(3.43) σ11(0) ≥1
2Cu11(0) ≥ c1 > 0.
Let δ > 0 be small enough so that
w ≡ (|σz1|2 + |σzn|2)12 > 0 in Ω ∩Bδ(0).
Define ζ = (ζ1 . . . , ζn): ζ1 = −σzn
w,
ζj = 0, 2 ≤ j ≤ n− 1,
ζn =σz1
wand
Φ = ϕjkζj ζk − (u− ϕ)xnσjkξj ξk − u11(0).
Note that ζ ∈ TC∂Ω on ∂Ω. We have
(3.44) Φ = ujkζj ζk − u11(0) ≥ 0 on ∂Ω ∩Bδ(0)
and Φ(0) = 0.
Write G = σjkξj ξk. We obviously have
LΦ ≤ −L(uxnG) + C(1 +
∑ujj
).
Using the fact that |L(uxn)| ≤ C, we calculate
L(uxnG) =GL(uxn) + ujk(uxnkGj + uxnjGk) + uxnLG
≥ujk(uxnkGj + uxnjGk)− C(1 +
∑ujj
).
Since
uxnk = 2unk +√−1uynk
and ujkunk = δjn, by Schwarz inequality,
ujk(uxnkGj + uxnjGk) = 2(Gn +Gn) +√−1ujk(uynkGj − uynjGk)
≥ − ujkuynjuynk − C(1 +
∑ujj
).
Next,
L(uyn − φyn)2 = 2(uyn − φyn)L(uyn) + L(φ2yn
) + 2ujkuynjuynk
≥ 2ujkuynjuynk − C(1 +
∑ujj
).
20 BO GUAN
It follows that
(3.45) L(Φ− (uyn − φyn)2) ≤ C(1 +
∑ujj
)in Ω ∩Bδ(0).
Moreover, by (3.33) and (3.44),
(uyn − φyn)2 − Φ ≤ C|z|2 on ∂Ω ∩Bδ(0).
Consequently, we may apply Lemma 3.8 to
h = (uyn − φyn)2 − Φ
to obtain Φxn(0) ≥ −C, which, by (3.43), implies
(3.46) uxnxn(0) ≤ C
σ11(0)≤ C
c1.
In view of (3.36) and (3.46) we have an a priori upper bound for all eigenvalues
of the complex Hessian ujk(0). Since detujk ≥ ψ > 0, the eigenvalues of ujk(0)must admit a positive lower bound, i.e.,
minξ∈Cn
,|ξ|=1ujk(0)ξj ξk ≥ c0.
Therefore,
m0 = minξ∈T C
0 ∂Ω,|ξ|=1ujk(0)ξj ξk ≥ min
ξ∈Cn,|ξ|=1
ujk(0)ξj ξk ≥ c0.
The proof of Proposition 3.9 is complete.
This establishes (3.23).
3.4. Global estimates for second derivatives. In this subsection we derive the
global estimate for second derivatives
(3.47) maxΩ|D2u| ≤ C
where C depends on |u|C1(Ω) and |D2u|∂Ω. We wish to include the degenerate case
ψ ≥ 0. So we shall still assume ψ > 0 in this subsection but the estimates will not
depends on the lower bound of ψ.
Instead of L, it is more convenient to consider the linearized operator L = ujk∂j∂k.
Lemma 3.10. For φ ∈ C2(Ω),
(3.48) e−φL(eφ∆u) ≥ (Lφ)∆u+ ∆(log detujk).
COMPLEX MONGE-AMPERE EQUATIONS 21
Proof. By (3.11) and (3.5),
(3.49) e−φL(eφ∆u) = ∆u(Lφ+ ujkφjφk) + 2Reujk(∆u)jφk+ L(∆u)
and
(3.50) L∆u = uimuljuijkulmk + ∆(log detujk).
By Cauchy-Schwarz inequality,
(3.51) 2|Reuij(∆u)jφk| ≤ (∆u)−1ujk(∆u)j(∆u)k + (∆u)ujkφjφk.
On the other hand,
(3.52) (∆u)−1uij(∆u)i(∆u)j ≤ uimuljuijkulmk.
To see this we may assume uij is diagonal. Applying Cauchy-Schwarz inequality,
uij(∆u)i(∆u)j =∑
i
uii∣∣∣ ∑
k
u1/2
kku−1/2
kkukki
∣∣∣2≤ ∆u
∑i,k
uiiukk|ukki|2
≤ ∆u∑i,j,k
uiiujjuijkuijk.
This proves (3.48).
We first present a corollary of Lemma 3.10 when ψ does not depend on the gradient
Du.
Corollary 3.11. For η ∈ C2(R),
(3.53) e−η(u)L(eη(u)∆u) ≥ (ψu + nη′ + η′′uijuiuj)∆u+ fuu|∇u|2 − 2|∇fu||∇u|+ ∆f
where
∆f =∑
fzj zj, |∇fu|2 =
∑|fuzj
|2.
Consequently, if η is a convex function and ψu + nη′ > 0 then
maxΩ
eη(u)∆u = max∂Ω
eη(u)∆u
or
∆u ≤ |∇fu|2 − η′′ − fuu)|∇u|2 −∆f
fu + nη′
wherever eη(u)∆u attains an interior (local) maximum.
22 BO GUAN
Proposition 3.12. Suppose for any M > 0 there exists constant C = C(M) such
that
(3.54)
− fuu, |fzj|, |fuzj
|,∑
fzj zj,≤ Cψ−
1n ,
− fu, |fzjpk|, |fupj
| ≤ Cψ−12n ,
|fpj|, |fpjpk
| ≤ C,
fpj pk ≥ −CI
whenever |u|+ |p| ≤M . Then
(3.55) ∆u ≤ C in Ω
where C depends on |u|C1(Ω) and |D2u|∂Ω but is independent of ψ−1 (see (3.25)).
Proof. Consider G = eφ∆u where
φ = A|∇u|2 +B|z|2.
Suppose G attains its maximum at an interior point z0 ∈ Ω. Then we have (all
calculations below are at z0)
(3.56) φj∆u+ (∆u)j = 0, ∀ j
and,
(3.57) 0 ≥ e−φLG ≥ (Lφ)∆u+ ∆(f).
By (3.15) and (3.16),
(3.58)
Lφ =AL(|∇u|2) +B∑
ujj
≥A∆u+ Aujkujlulk + 2Refpjφj
+Bψ−1n + 2Afu|∇u|2 + 2ReAfzj
uj −Bfpjzj
while from (3.6) we see that
(3.59)
∆(f) ≥ 2Refpj(∆u)j+ fpj pk
(ujlulk + ujlulk) + 2Refpjpkujlulk
+ fu∆u− C∆u∑
(|fpju|+ |fpjzk|)
− C∑
(|fpju|+ |fpjzk|)|ujk|
+ fuu|∇u|2 +∑
fzj zj− C
∑|fuzj
|.
COMPLEX MONGE-AMPERE EQUATIONS 23
Using (3.56) and Cauchy-Schwarz inequality, we obtain
(3.60)
(Lφ)∆u+ ∆(f) ≥ (Aujk∆u+ fpj pk− Cδjk)(ujlulk + ujlulk)
+ fu∆u+ fuu|∇u|2 +∑
fzj zj− C
∑|fuzj
|
+ 2Afu|∇u|2 +Bψ−1n − C
∑(A|fzj
|+B|fpj|)
−∑
(|fpjzk|2 + |fpju|2 + |fpjpk
|2).
By assumption (3.54) if we choose A and B sufficiently large then (3.57) and (3.60)
yield
0 ≥ A
2(∆u)2 − Cψ−
12n ∆u+ (B − CA− C)ψ−
1n − CB
≥ A− 1
2(∆u)2 + (B − CA− C)ψ−
1n − CB
which gives (3.55) when B − CA− C ≥ 0 (and, certainly, A > 1).
By (3.55) and equation (2.8) we have
(3.61) c0I ≤ ujk ≤ C0I in Ω.
It follows that
(3.62) ujk ≥ C−10 I in Ω.
We next finish the proof of (3.47), that is,
(3.63) |uxixj|, |uxiyj
|, |uyiyj| ≤ C, ∀ 1 ≤ i, j ≤ n.
Let
τ = aj∂xj+ bj∂yj
, aj, bj ∈ R,∑
(a2j + b2j) = 1,
be a first order linear differential operator of constant real coefficients. We shall show
under assumption (3.54) that
(3.64) maxΩ
uττ ≤ max∂Ω
uττ + C,
which obviously implies (3.63) since u is plurisubharmonic.
By (3.6), (3.7) and assumption (3.54), we find
(3.65)
Luττ ≥ 2fpj pkuτjuτ k + 2Refpjpk
uτjuτk
+ fuuττ + fuuu2τ + fττ − C
∑|fuzj
|
− C∑
(|fpjzk|+ |fpju|)|uτj|
≥ − C∑
|uτj|2 − Cψ−1n .
24 BO GUAN
Next, by (3.62)
(3.66)Lu2
τ = 2ujkuτjuτ k + 2uτLuτ
≥ 2C−10
∑|uτj|2 − Cψ−
1n
and
(3.67) Leλ|z|2 ≥ C−10 |z|2λ2 + λψ−
1n − Cλ ≥ λψ−
1n
if we assume |z| ≥ 1 in Ω and λ is sufficiently large. Therefore,
(3.68) L(uττ + Au2τ + eλ|z|2) ≥ (C−1
0 A− C)∑
|uτj|2 + (λ− CA− C)ψ−1n ≥ 0
when A and λ are sufficiently large. This shows (3.64) by the maximum principle.
4. Complex Monge-Ampere equations in Kahler geometry
4.1. Kahler metrics and curvature. Let (M,J) be a complex manifold and g a
Riemannian metric on M . Recall that J2 = −id. We assume that J is compatible
with g, i.e.
(4.1) g(u, v) = g(Ju, Jv), u, v ∈ TM.
The Kahler form ωg of g is defined by
(4.2) ωg(u, v) = −g(u, Jv).
The complexified tangent bundle TCM = TM × C has a natural splitting
(4.3) TCM = T 1,0M + T 0,1M
where T 1,0M and T 0,1M are the ±√−1-eigenspaces of J . Obviously, the metric g is
extended C-linearly to TCM , and
(4.4) g(u, v) = 0 if u, v ∈ T 1,0M , or u, v ∈ T 0,1M.
In local coordinates (z1, . . . , zn) this gives
(4.5) g( ∂
∂zj
,∂
∂zk
)= 0, g
( ∂
∂zj
,∂
∂zk
)= 0
since
(4.6) J∂
∂zj
=√−1
∂
∂zj
, J∂
∂zj
= −√−1
∂
∂zj
.
COMPLEX MONGE-AMPERE EQUATIONS 25
We write
(4.7) gjk = g( ∂
∂zj
,∂
∂zk
).
The Kahler form ωg is then given by
(4.8) ωg =
√−1
2gjkdzj ∧ dzk.
The metric g is Kahler if the Kahler form ωg is closed, i.e. dωg = 0. This is
equivalent to ∇J = 0, where ∇ denote the Levi-Civita connection of g. In local
coordinates, this is characterized by
(4.9)∂gjk
∂zl
=∂glk
∂zj
, ∀ j, k, l.
Another useful fact is that a metric g is Kahler if and only if g has local potentials,
namely, in local coordinates there exists a function ϕ such that
(4.10) gjk =∂2ϕ
∂zj∂zk
.
The Christoffel symbols in local coordinates (z1, . . . , zn) are defined by
∇ ∂∂zj
∂
∂zk
= Γljk
∂
∂zl
+ Γljk
∂
∂zl
and
∇ ∂∂zj
∂
∂zk
= Γljk
∂
∂zl
+ Γljk
∂
∂zl
.
For a Kahler manifold, using ∇J = 0 and (4.6) one can see that the only possible
non-vanishing terms are Γljk and Γl
jk= Γl
jk. Moreover,
(4.11) Γljk = glm∂gjm
∂zk
.
We recall that the Riemannian curvature tensor is defined by
R(u, v)w = ∇u∇vw −∇v∇uw −∇[u,v]w.
Since ∇J = 0,
R(u, v)Jw = JR(u, v)w.
It follows that
(4.12) g(R(u, v)Jw, Jx) = g(R(u, v)w, x) ≡ R(u, v, w, x).
26 BO GUAN
Therefore R(u, v, w, x) = 0 unless w and x are of different type. By (4.11) we have
in local coordinates,
(4.13) Rijkl ≡R( ∂
∂zi
,∂
∂zj
,∂
∂zk
,∂
∂zl
)= −
∂2gij
∂zk∂zl
+ gpq ∂gpj
∂zk
∂giq
∂zl
.
The Ricci curvature is given by
(4.14) Rkl = gijRijkl = − ∂2
∂zk∂zl
log det gij,
or equivalently, the Ricci form is
(4.15) Ricg :=
√−1
2Rkldzk ∧ dzl = −
√−1
2∂∂ log det gij.
Recall that a Kahler metric g is Einstein if there exists a real constant λ such that
(4.16) Ricg = λωg.
4.2. The complex Monge-Ampere equation in Kahler manifolds. Let (Mn, ω)
be a Kahler manifold and ψ ∈ C∞(M ×R), ψ ≥ 0. We consider the complex Monge-
Ampere equation which plays fundamental roles in Kahler geometry
(4.17) (ω(φ))n = ψ(z, φ)ωn
where
ω(φ) = ω +
√−1
2∂∂φ.
If in local coordinates,
ω =
√−1
2gjkdzj ∧ dzk,
then
(4.18) ω(φ) =
√−1
2(gjk + φjk)dzj ∧ dzk
and therefore, equation (4.17) is equivalent to
(4.19) det(gij + φij) = ψ(z, φ) det gij.
As we see in Section 2 for complex Monge-Ampere equations in Cn, equation (4.17)
is elliptic for solutions in the Mabuchi space of Kahler potentials defined as
(4.20) H = φ ∈ C∞(M) : ω(φ) > 0.
The following fundamental existence result is due to S.-T. Yau [48] and is the core
of his proof of Calabi conjecture.
COMPLEX MONGE-AMPERE EQUATIONS 27
Theorem 4.1 ([48]). Let M be a compact Kahler manifold (without boundary) and
ψ ∈ C∞(M × R). Suppose that ψ > 0, ψu ≥ 0, and that there exists a function
φ ∈ C∞(M) such that
(4.21)
∫M
ψ(z, φ(z))ωn = Vol (M).
Then there exists a solution φ ∈ H of equation (4.17).
SinceM is compact (without boundary), one needs certain compatibility conditions
such as (4.21) to solve equation (4.17). Indeed, any solution φ ∈ H of equation (4.17)
must satisfy (4.21) with φ in place of φ. Note that φ is not assumed to be in H.
For the complete proof of Theorem 4.1 the reader is referred to Yau’s original
paper [48]. For ψ = ψ(z) it can also be found in several articles and monographs such
as [46], [47], and [10].
We next consider the Dirichlet problem for equation (4.17). Let Ω be a compact
domain with smooth boundary in a Kahler manifold M . Given ψ ∈ C∞(Ω × R),
ψ > 0, and ϕ ∈ C∞(∂Ω), we seek solutions φ ∈ H(Ω) = φ ∈ C∞(Ω) : ω(φ) > 0 of
the Dirichlet problem
(4.22)
(ω(φ))n = ψ(z, φ)ωn in Ω
φ = ϕ on ∂Ω.
Theorem 4.2. Suppose that there exists a subsolution u ∈ C2(Ω) satisfying ω(u) > 0
and
(4.23)
ω(u)n ≥ ψ(z, u)ωn in Ω
u = ϕ on ∂Ω.
There exists a solution φ ∈ H of the Dirichlet problem (4.22).
This is an analogue of Theorem 2.7. Its proof also reduces to a priori C2 estimates.
The C0 and boundary gradient estimates follow as before from the inequalities u ≤φ ≤ h where h is the harmonic extension of ϕ. For the global gradient estimates see
Blocki citeBlocki09 and P.-F. Guan [29]. The boundary estimates in Subsection 3.3
for second derivatives are still valid in this case as the proof is purely local and there
are always local Kahler potentials for Kahler manifolds. Finally, the global estimates
for second derivatives follows from Yau [48].
Below we shall see some applications of Theorems 4.1 and 4.2.
28 BO GUAN
4.3. The Calabi-Yau theorem. Given two Kahler forms on a Kahler manifold M
(4.24) ω =
√−1
2gjkdzj ∧ dzk, ω =
√−1
2gjkdzj ∧ dzk,
by (4.15) their Ricci forms are related by the formula
(4.25) Ric(ω)− Ric(ω) =
√−1
2∂∂ log
det gij
det gij
which is a globally defined function on M . Consequently, the cohomology class
[Ric(ω)] of Ric(ω) is independent of ω and is in fact equal to πc1(M) where c1(M) is
the first Chern class of M .
The Calabi conjecture asserts that the converse is true: any (1, 1) form on a compact
Kahler manifold M representing πc1(M) must be the Ricci form of a Kahler form on
M . This conjecture was proved by S.-T. Yau [48].
Theorem 4.3 ([48]). Let (Mn, ω) be a compact Kahler manifold. Given any (1, 1)
form Ω representing πc1(M), there exists a unique Kahler form ω ∈ [ω] such that
Ric(ω) = Ω.
The proof of Theorem 4.3 reduces to solving a complex Monge-Ampere equation
of form (4.17) on M . In fact, since both Ω and Ric(ω) represent πc1(M), there exists
a smooth function f such that
(4.26) Ω− Ric(ω) =
√−1
2∂∂f.
(This follows from the ∂∂-lemma; see e.g. [47].) Suppose that Ω is the Ricci form of
ω ∈ [ω]. By the ∂∂-lemma again there exists a smooth function ϕ such that
ω = ω +
√−1
2∂∂φ.
In local coordinates we have,
(4.27) ∂∂ logdet(gij + φij)
det gij
= ∂∂f.
Since M is compact, this implies that
(4.28) logdet(gij + φij)
det gij
− f = c
is a constant, that is
(4.29) det(gij + φij) = ef+c det gij.
COMPLEX MONGE-AMPERE EQUATIONS 29
Integrating (4.29) we see that c satisfies
(4.30)
∫M
ef+cωn = Vol(M).
Consequently, Theorem 4.3 reduces to solving equation (4.29) under the compatibility
condition (4.30).
4.4. Geodesics in the Mabuchi space of Kahler potentials. Let (M, g) be a
compact Kahler manifold. The Mabuchi space of Kahler potentials (see (4.20))
(4.31) H = φ ∈ C∞(M) : ω(φ) > 0
is an open subset of C∞(M). Therefore, its tangent space TφH at φ ∈ H is naturally
identified to C∞(M). In [39], Mabuchi introduced a Riemannian structure onH using
the L2 inner product on TφH with respect to the volume form of ω(φ):
(4.32) 〈ξ, η〉φ =
∫M
ξη (ω(φ))n, ψ ∈ TφH.
Accordingly, one can define the length of a regular curve ϕ : [0, 1] → H by
(4.33) L(ϕ) =
∫ 1
0
〈ϕ, ϕ〉12φdt.
Here and in what follows ϕ = ∂ϕ/∂t and ϕ = ∂2ϕ/∂t2. The geodesic equation takes
the form
(4.34) ϕ− |∇ϕ|2ϕ = 0.
This means in local coordinates
(4.35) ϕ− g(ϕ)jkϕzjϕzk
= 0.
Here [g(ϕ)jk] is inverse matrix of [g(ϕ)jk] = [gjk + ϕjk].
Donaldson [18] conjectured that H is geodesically convex, i.e., any two functions
in H can be connected by a smooth geodesic, and that H is a metric space.
Conjecture 4.4 (Donaldson [18]). (i) Let ϕ0, ϕ1 ∈ H. There exists a smooth geodesic
ϕ : [0, 1] → H such that
(4.36) ϕ(0) = ϕ0, ϕ(1) = ϕ1.
(ii) The function
d(ϕ0, ϕ1) = infL(ϕ)|ϕ : [0, 1] → H, ϕ(0) = ϕ0, ϕ(1) = ϕ1, ϕ0, ϕ1 ∈ H
30 BO GUAN
defines a distance on H.
This is a reformulation of the original Donaldson conjecture in [18]. Indeed, by an
observation of Donaldson [18], Mabuchi [39] and Semmes [41], part (i) of Donaldson’s
conjecture may be reduced to a Dirichlet problem for the homogeneous complex
Monge-Ampere equation in M × A where A is the cylinder [0, 1]× S1. Let
w = zn+1 = t+√−1s
be the local coordinate of A. We may view a smooth curve ϕ in H as a function in
C∞(M× [0, 1]) and therefore a rotation-invariant function (constant in s) in C∞(M×A). We have
ϕ =∂ϕ
∂t= 2
∂ϕ
∂w= 2
∂ϕ
∂w
and
ϕ =∂2ϕ
∂t2= 4
∂2ϕ
∂w∂w.
Suppose now that ϕ solves the geodesic equation (4.35). Then
(4.37)
det
ϕ1w
(g(ϕ)jk)...
ϕnw
ϕw1 · · · ϕwn ϕww
=1
4det
ϕ1
(g(ϕ)jk)...ϕn
ϕ1 · · · ϕn ϕ
=1
4det(g(ϕ)ij) · det
g(ϕ)k1ϕk
I...
g(ϕ)knϕk
ϕ1 · · · ϕn ϕ
=
1
4det(g(ϕ)ij) · (ϕ− g(ϕ)jkϕzj
ϕzk) = 0.
That is,
(4.38)(Ω +
√−1
2∂∂ϕ
)n+1
= 0 in M × A
COMPLEX MONGE-AMPERE EQUATIONS 31
where
(4.39) Ω = ω +
√−1
2∂∂|zn+1|2 =
√−1
2
( ∑j,k≤n
gjkdzj ∧ dzk + dzn+1 ∧ dzn+1
)is the lift of ω to the product M × A.
Conversely, if ϕ ∈ C∞(M × A) is a rotation-invariant solution of (4.38) such that
(4.40) ϕ(·, w) ∈ H, ∀ w ∈ A,
then one sees from (4.37) that ϕ is a geodesic in H.
The original conjecture of Donaldson [18] may be stated as follows.
Conjecture 4.5 (Donaldson [18]). Let A be a compact Riemann surface with bound-
ary and ρ ∈ C∞(M × ∂A) such that ρ(·, w) ∈ H for w ∈ ∂A. Then there exists
a unique solution ϕ of the Monge-Ampere equation (4.38) satisfying (4.40) and the
boundary condition ϕ = ρ.
The uniqueness was proved by Donaldson [18] as a consequence of the maximum
principle. In [15], X.-X. Chen obtained the following existence result.
Theorem 4.6. Let A and ρ be as in Conjecture 4.5. There exists a unique (weak)
solution ϕ ∈ C1,α(M×A), 0 < α < 1, with ∆ϕ ∈ L∞(M×A) of the Dirichlet problem
(4.41)
Ω +
√−1
2∂∂ϕ ≥ 0 in M × A(
Ω +
√−1
2∂∂ϕ
)n+1
= 0 in M × A
ϕ = ρ on M × ∂A.
In what follows we outline the proof of Theorem 4.6 for A = [0, 1] × S1. Since
equation (4.38) is degenerate, we shall consider the approximation problem
(4.42)
Ω +
√−1
2∂∂ϕ ≥ 0 in M × A(
Ω +
√−1
2∂∂ϕ
)n+1
= εΩn+1 in M × A
ϕ = ρ on M × ∂A.
In order to apply Theorem 4.2 to the Dirichlet problem (4.42) we need to construct
a subsolution. This is easily done for the product space M×A when A is the annulus
32 BO GUAN
[0, 1]× S1. Let ϕ0 = ρ|t=0 and ϕ1 = ρ|t=1. Define
ϕ = (1− t)ϕ0 + tϕ1 +K(t2 − t).
By the assumption that ρ(·, w) ∈ H for w ∈ ∂A we can choose K > 0 sufficiently
large such that ϕ satisfies
(4.43)
Ω +
√−1
2∂∂ϕ > 0 in M × A(
Ω +
√−1
2∂∂ϕ
)n+1
≥ Ωn+1 in M × A
ϕ = ρ on M × ∂A.
Consequently, by Theorem 4.2 we obtain a unique solution ϕε ∈ C∞(M×A) of (4.42)
for any ε ∈ (0, 1]. Moreover, by the maximum principle,
ϕε ≥ ϕε′ ∀ 0 < ε < ε′ ≤ 1
and therefore the limit
ϕ(z) = limε→0
ϕε(z)
exists for all z ∈M ×A. To finish the proof of Theorem 4.6 we need to establish the
estimate
(4.44) |ϕε|C2(M×A) ≤ C independent of ε.
It is straightforward to verify that
|ϕε|C1(M×A) ≤ C independent of ε
and
(4.45) maxM×A
∆ϕε ≤ C + C max∂(M×A)
∆ϕε.
On the boundary ∂(M × A) = M × ∂A we have
(4.46) ϕεξη = ϕ
ξη, ∀ ξ, η ∈ Tp∂(M × A), ∀ p ∈ ∂(M × A)
since ∂(M ×A) is given by t = 0 or t = 1, and, from the estimates in Subsection 3.3,
(4.47) |ϕεξt| ≤ C, ∀ ξ ∈ Tp∂(M × A), |ξ| = 1, ∀ p ∈ ∂(M × A)
where C is independent of ε. As in Subsection 3.3, using (4.46) and the fact that ϕ
is strictly plurisubharmonic we derive from equation (4.42)
(4.48) |ϕεtt| ≤ ϕε
ww + |ϕεss| ≤ C, independent of ε on ∂(M × A).
This completes the proof of (4.44) and (4.45).
COMPLEX MONGE-AMPERE EQUATIONS 33
Remark 4.7. If ρ is rotation invariant then so is ϕ by the uniqueness.
5. The Chern-Levine-Nirenberg conjecture
In this section we present a brief description of the work of Pengfei Guan [27], [28]
on the Chern-Levine-Nirenberg conjecture on intrinsic norms.
Let M be a compact complex manifold and boundary ∂M = Γ0 ∪ Γ1. Define
PSH1(M) = u ∈ PSH(M) : 0 < u < 1
F = C2(M) ∩ PSH1(M), F = C0(M) ∩ PSH1(M)
and
B = u ∈ F : u|Γ0 = 0, u|Γ1 = 1.In [16], Chern, Levine, and Nirenberg introduced the following nonnegative function
on the homology group H∗(M,R): for γ ∈ H∗(M,R),
(5.1) N(γ) = supu∈F
infT∈γ
|T (dcu ∧ (ddcu)k−1)|, if dimγ = 2k − 1,
|T (du ∧ dcu ∧ (ddcu)k−1)|, if dimγ = 2k.
Similarly one can define N by replacing F by F .
It was shown in [16] that N and N are a seminorm on H∗(M,R). Namely,
(5.2)N(aγ) = |a|N(γ), a ∈ R,
N(γ1 + γ2) ≤ N(γ1) +N(γ2), dimγ1 = dimγ2.
(Same is true for N .) Moreover, N and N decrease under holomorphic mappings: if
f : M → P is a holomorphic mapping then
(5.3) N(f∗γ) ≤ N(γ), N(f∗γ) ≤ N(γ), γ ∈ H∗(M,R).
From now on we shall consider a special case:
M = Ω \m⋃
j=1
Ωj, Γ0 =m⋃
j=1
∂Ωj, Γ1 = ∂Ω
where Ω1, . . . ,Ωm and Ω are bounded smooth strongly pseudoconvex domains in Cn,
Ωi ∩ Ωj = ∅ (i 6= j) and Ωj ⊂ Ω, 1 ≤ j ≤ m. Define
Jk(u) =
∫M
du ∧ dcu ∧ (ddcu)k−1 ∧ ωn−k
34 BO GUAN
where
ω =
√−1
2
∑dzj ∧ dzj
is the Kahler form of Cn.
The Chern-Levine-Nirenberg conjecture [16]:
N(Γ1) = infu∈B
Jn(u).
In [27] P.-F. Guan proved the following modified version of the conjecture, formu-
lated by Bedford and Taylor [5].
Theorem 5.1 ([27]). Suppose that
(5.4)m⋃
j=1
Ωj is holomorphically convex in Ω.
Then
N(Γ1) = infu∈B
Jn(u).
More recently, P.-F. Guan [28] was able to prove that N(Γ1) = N(Γ1) for general
complex manifolds, therefore confirmed the original Chern-Levine-Nirenberg conjec-
ture.
The Chern-Levine-Nirenberg conjecture is associated with the Dirichlet problem
(5.5)
(ddcu)n = 0 in M
u|Γ0 = 0, u|Γ1 = 1.
Indeed if u ∈ C0(M) ∩ PSH(M) solves problem 5.5 then for all T ∈ [Γ0],∫T
du ∧ (ddcu)n−1 =
∫M
du ∧ dcu ∧ (ddcu)n−1.
A crucial ingredient in the proof of Theorem 5.1 is the following regularity result
for the Dirichlet problem (5.5).
Theorem 5.2 ([27]). Under condition (5.4), problem 5.5 admits a unique solution
u ∈ C1,1(M) ∩ PSH(M).
The C1,1 regularity is optimal as shown by an example of Bdeford and Faenass [3].
Because the homogeneous complex Monge-Ampere equation (ddcu)n = 0 is degener-
ate, the best regularity of its solution one can hope is C1,1 in general, even for smooth
data.
In what follows we outline the proof of Theorem 5.2 in [27].
COMPLEX MONGE-AMPERE EQUATIONS 35
Proof of Theorem 5.2 (Sketch). The first step is to construct a strictly plurisubhar-
monic function u ∈ C2(M) that satisfies
(5.6)
(ddcu)n ≥ ε0 > 0 in M
u|Γ0 = 0, u|Γ1 = 1.
Let σ1, . . . , σm, σ be smooth defining functions of Ω1, . . . ,Ωm,Ω, respectively. By
condition (5.4), there is a function ψ ∈ C∞(Ω) ∩ PSH(Ω) such thatψ < 0 in a neighborhood of
⋃mk=1 Ωk
ψ > 0 in Ω \⋃m
k=1 Uk
where
Ωk ⊂ Uk ⊂ Uk ⊂ Ω, 1 ≤ k ≤ m, U j ∩ Uk = ∅, j 6= k.
We may assume that for some λ > 0,
(5.7) ψ − λ|z|2, σ − λ|z|2, σk − λ|z|2 ∈ PSH(Ω), 1 ≤ k ≤ m.
Define
ϕk(z) = maxε2σk(z), εψ(z), z ∈ Ω.
For ε > 0 small,
ϕk =
ε2σk, in a neighborhood Vk ⊂ Uk of Ωk,
εψ, outside Vk.
Therefore, we may choose ε sufficiently small and A > 0 sufficiently large such that
the function u ∈ C0,1(Ω) defined by
u = max1 + Aσ, ϕ1, . . . , ϕm,
which is plurisubharmonic, satisfies (5.6) and
u =
1 + Aσ near ∂Ω,
ϕk in Vk.
Note that u is smooth near ∂M . Finally, by a regularization procedure one obtains a
strictly plurisubharmonic function in C∞(M), still denoted by u, satisfying (5.6); see
P.-F. Guan’s original argument in [27] for details.
Using u constructed above as a subsolution, for any 0 < ε ≤ ε0 we can apply
Theorem 2.7 to obtain a unique strictly plurisubharmonic solution uε ∈ C∞(M) of
36 BO GUAN
the Dirichlet problem
(5.8)
(ddcu)n = ε in M,
u|Γ0 = 0, u|Γ1 = 1.
By the maximum principle,
uε ≤ uε′if ε ≤ ε′.
Consequently, uε converges pointwise to a function u in M as ε tends to zero. In order
to prove that u ∈ C1,1(M) and solves (5.5) we only need to derive the estimate
(5.9) |uε|C2(M) ≤ C independent of ε.
Note that (5.9) does not follow directly from the estimates in Section 3 because
equation (5.8) becomes degenerate as ε vanishes. However, as we see in Section 3 this
only affects the estimate in (3.38). In other words, (5.9) will follow if we can recover
(3.38) for uε with c0 independent of ε.
For this we make use of the fact that uε is constant on each component of ∂M .
Indeed, consider a point, which we may assume to be the origin in Cn, on ∂M .
Under a coordinate system as in Subsection 3.3 we have, since uε is constant on the
components of ∂M ,
(5.10) uεjk(0) =
uεxn
(0)
σxn(0)σjk(0), 1 ≤ j, k ≤ n− 1
where σ = σl if 0 ∈ ∂Ωl (1 ≤ l ≤ m) and σ = σ if 0 ∈ ∂Ω. On the other hand, from
u ≤ uε ≤ h in M
where h is the harmonic function in M with h = 0 on Γ0 and h = 1 on Γ1, we see
thatuε
xn(0)
σxn(0)≥ c1
for some constant c1 > 0 independent of ε. By (5.7) we obtain
(5.11) uεjk(0)|1≤j,k≤n−1 ≥ c1λI,
therefore proving (5.9).
COMPLEX MONGE-AMPERE EQUATIONS 37
6. Regularity of pluricomplex Green functions
6.1. Pluricomplex Green function with logarithmic poles. Let Ω be a bounded
domain in Cn and ζ ∈ Ω. The pluricomplex Green function on Ω with a logarithmic
pole at ζ is defined as
gζ(z) = supv(z) : v ∈ PSH(Ω), v < 0 and v(z) ≤ log |z − ζ|+O(1)
.
This function is a higher dimensional analogue of the Green function in C, and plays
important roles in pluripotential theory. We refer the reader to [1], [17], [31] for
background and connections in complex analysis. In this section we present some
regularity results when Ω is strongly pseudoconvex.
A striking regularity theorem is due to Lempert [36] who proves that gζ(z) is smooth
in Ω \ ζ for smooth strictly convex domains Ω. This result has found important
applications; see for instance [42] where Semmes develops a theory of generalized
Riemann mappings.
In the strongly pseudoconvex case, however, E. Bedford and J.-P. Demailly [2]
found counterexamples which show that gζ in general does not belong to C2(Ω\ζ).In [23] the author proved gζ(z) ∈ C1,α(Ω \ ζ). Later on Blocki [8] improved it to
gζ(z) ∈ C1,1(Ω \ ζ), which is optimal as shown by counterexamples in [2].
A fundamental property of the pluri-complex Green function is that it is a week
solution of the following problem
(6.1)
u is pluri-subharmonic in Ω \ ζdet(uzj zk
) = 0 in Ω \ ζu = 0 on ∂Ω
u(z) = log |z − ζ|+O(1) as z → ζ.
Theorem 6.1 ([23], [8]). Let Ω be a smooth strongly pseudoconvex domain. There
exists a unique solution of (6.1) in C1,1(Ω \ ζ). Consequently, gζ ∈ C1,1(Ω \ ζ).
Proof. The uniqueness is a consequence of the minimum principle of Bedford-Taylor [4]
as in [37]. In what follows below we prove the existence of a solution in C1,α(Ω\ζ);for the proof of C1,1 regularity see [8].
Without loss of generality we assume ζ = 0 and B1 = B1(0) ⊂ Ω. Consider the
Dirichlet problem
(6.2)
det(vjk) = 1 in Ω,
v = − log |z| on ∂Ω.
38 BO GUAN
Since Ω is strongly pseudoconvex, by Theorem 2.7 problem (6.2) admits a unique
strictly pluri-subharmonic solution v ∈ C∞(Ω). Let u ≡ v + log |z| ∈ C∞(Ω \ 0).We see that u satisfies
(6.3)
u is strictly pluri-subharmonic in Ω \ 0det(ujk) ≥ 1 in Ω \ 0u ≤ 0 in Ω \ 0, u = 0 on ∂Ω
u(z) = log |z|+O(1) as z → 0.
For ε > 0 small (we shall assume ε ≤ 12) let Ωε = Ω \Bε and consider the problem
(6.4)
det(ujk) = ε in Ωε,
u = u on ∂Ωε.
Note that u is a subsolution of (6.4) when ε ≤ 1. We may apply Theorem 2.7 again
to obtain a unique strictly pluri-subharmonic solution uε ∈ C∞(Ωε) of the Dirichlet
problem (6.4). By the maximum principle,
(6.5) u ≤ uε ≤ uε′ ≤ log |z| in Ωε if ε′ ≤ ε.
Thus the limit
u(z) ≡ limε→0
uε(z)
exists for all z ∈ Ω \ 0. We need to show that u ∈ C1,α(Ω \ 0). By standard
elliptic theory it suffices to establish the following estimate
(6.6) |∇uε|2 + ∆uε ≤ C
|z|2in Ωε, independent of ε.
Lemma 6.2. There exists a constant C1 independent of ε such that
(6.7) |∇uε| ≤ C1 on ∂Ω and |∇uε| ≤ C1
εon ∂Bε.
Proof. Since u ≤ uε ≤ 0 in Ωε and uε = u = 0 on ∂Ωε, we have
|∇uε| = uεν ≤ uν = |∇u| on ∂Ω
where ν is the exterior unit normal vector to ∂Ω. This proves the first inequality in
(6.7). To prove the second one, for z ∈ B2 \B1 let
u(z) = uε(εz)− log ε,
u(z) = u(εz)− log ε = v(εz) + log |z.|
COMPLEX MONGE-AMPERE EQUATIONS 39
Note that
(6.8) det(ujk) = ε2n det(ujk) = ε2n+1.
Let h be the harmonic function on B2 \ B1 with h = log 2 on ∂B2 and h(z) = v(εz)
on ∂B1. Then u ≤ u ≤ h on B2 \ B1 by the maximum principle, since ∆u ≥ 0 in
B2 \ B1, u ≤ h on ∂B2 and u = h on ∂B1. Consequently,
|∇u| = uν ≤ hν ≤ C1 on ∂B1.
Here ν is the exterior unit normal vector to ∂B1. This implies the second inequality
in (6.7) as ∇uε(z) = 1ε∇u( z
ε).
Using Lemma 6.2 and Corollary 3.3 with η(u) = u we obtain, by (6.5)
(6.9) |∇uε| ≤ Ce−uε ≤ C
|z|on Ωε.
Lemma 6.3. There exists a constant C2 independent of ε such that
(6.10) |∇2uε| ≤ C2 on ∂Ω and |∇2uε| ≤ C2
ε2on ∂Bε.
Proof. The first estimate in (6.10) can be derived as in Section 5. We only prove the
second one here. Let u, u and h be as in the proof of Lemma 6.2. It suffices to show
that
(6.11) |∇2u| ≤ C2 on ∂B1.
For a fixed point z0 ∈ ∂B1, we may assume z0 = (0, . . . , 1), i.e., the coordinates of
z0 are xj = yj = 0, 1 ≤ j ≤ n − 1, xn = 1 and yn = 0. Since u(z) = v(εz) on ∂B1
and |∇u| ≤ C1, it is trivial to obtain a bound for the pure tangential second order
derivatives at z0
(6.12) |uxixk|, |uxiyj
|, |uyjyl| ≤ C, 1 ≤ i, k ≤ n− 1, 1 ≤ j, l ≤ n.
To estimate the mixed tangential normal derivatives we need the following analogue
of Lemma 3.6.
Lemma 6.4. Let Uδ = (B2 \B1)∩Bδ(z0) and w = (u− u) + t(|z| − 1)−N(|z| − 1)2,
where t, N are positive constants. For N sufficiently large and t, δ sufficiently small,
we have
(6.13) ujkwjk ≤ − 1
64
(1 +
∑ukk
)in Uδ
and v ≥ 0 on ∂Uδ.
40 BO GUAN
Proof. We first note that this does not follow from Lemma 3.6 since ujk is not
uniformly positive definite in ε. In order to prove (6.13) we make use of a special
property of u. Since u(z) = v(εz) + log |z| and v is plurisubharmonic, we see that
ujkujk ≥ ujk(log |z|)jk ≥1
2|z|2ujk
(δjk −
zjzk
|z|2)
in Uδ.
On the other hand,
ujk((|z| − 1)2)jk = ujk(|z|2 − 2|z|)jk
=(1− 1
|z|
) ∑ukk +
1
2|z|3ujkzjzk.
≥ 1
16ujkzjzk.
The rest is obvious.
Returning to the proof of Lemma 6.3, as in Subsection 3.3 we may derive a bound
for the mixed tangential normal derivatives at z0 with the aid of Lemma 6.4,
(6.14) |uxkxn|, |uxnyj| ≤ C, 1 ≤ k ≤ n− 1, 1 ≤ j ≤ n.
It remains to establish an estimate for the pure normal second order derivative
(6.15) |uxnxn(z0)| ≤ C.
Because of (6.12) and (6.14) it suffices to prove
(6.16) unn(z0) ≤ C.
Since u− u = 0 on ∂B1,
ujk(z0) = ujk(z
0) +1
2(u− u)xn(z0)δjk, j, k < n
and therefore
(6.17)∑j,k<n
ujk(z0)ξj ξk ≥
∑j,k<n
ujk(z0)ξj ξk = |ξ|2
for any ξ = (ξ1, . . . , ξn−1) ∈ Cn−1. Finally, solving equation (6.8) for unn we see that
(6.16) follows from (6.12), (6.14) and (6.17). This completes the proof of (6.11) and
therefore that of Lemma 6.3.
Applying Corollary 3.11, by Lemma 6.3 we obtain
Corollary 6.5. There exists a constant C3 independent of ε such that
(6.18) |∆uε| ≤ C3e−2uε
in Ωε.
COMPLEX MONGE-AMPERE EQUATIONS 41
6.2. Pluricomplex Green function with logarithmic pole at infinity. For a
bounded subset E of Cn, the pluricomplex Green function of E (with pole at infinity)
is defined as
VE(z) = supv(z) : v ∈ L, v|E ≤ 0
where L denotes the collection of plurisubharmonic functions v on Cn of minimal
growth, i.e.,
v(z) ≤ log |z|+O(1) as |z| → ∞.
This function, which is also known as the Siciak-Zahariuta L-extremal function, was
first introduced by Siciak [43] (in an essentially equivalent form) and Zahariuta [49]
in the study of polynomial approximation of holomorphic functions of several vari-
ables and Bernstein-Walsh type theorems. It has proved to be a useful tool in the
pluripotential theory, and found applications in a wide range of problems in complex
analysis (see e.g. [1], [31] and [40]).
For a compact set E in Cn, the continuity of VE is equivalent to the Bernstein-
Markov inequality: For any t > 0 there exists an open set U ⊃ E such that ‖p‖U ≤etd(p)‖p‖E for any polynomial p, where d(p) is the degree of p ([49], [44], [45]). Other
criteria for continuity of VE have been discovered by various authors; see [1], [31], [40]
and references therein. This has been an important issue in applications.
As in the previous subsection we are concerned with the higher regularity of the
pluricomplex Green function, and our approach is based on its close connections with
the homogeneous complex Monge-Ampere equation. Let Ω be a bounded open set of
Cn and Ωc = Cn \ Ω. We consider the exterior Dirichlet problem
(6.19)
u is plurisubharmonic in Ωc
(ddcu)n = 0 in Ωc
u = 0 on ∂Ω
u(z) = log |z|+O(1) as |z| → ∞.
If ∂Ω ∈ C1 then VΩ is continuous and is the unique weak solution of (6.19) (see,
e.g. [17] and [31]). The uniqueness of solution to (6.19) also follows from a general
theorem of Bedford and Taylor [7]: If u, v ∈ L and (ddcu)n = (ddcv)n then u − v is
constant.
In [38] Lempert studied the problem using an interesting approach when Ω is a
smooth strictly convex domain. By constructing an analogue of the Kelvin transfor-
mation for the complex Monge-Ampere operator he demonstrated that problem (6.19)
42 BO GUAN
can be reduced to the Dirichlet problem (6.1) in the previous subsection. Conse-
quently, using his earlier results in [36] Lempert was able to prove that problem (6.19)
admits a unique solution in C∞(Ωc) for strictly convex Ω. This is a remarkable result
because in general solutions to the degenerate complex Monge-Ampere equation, even
with smooth data, may fail to be in C2 ([2], [3], [20]). The method in [38] actually
applies to strictly linearly convex domains. However, whether it can be extended to
more general cases is not clear; see for example [1]. In this subsection we treat prob-
lem (6.19) directly using elliptic PDE methods for strongly pseudoconvex domains.
More generally, let Ω1, . . . ,ΩN be bounded smooth strongly pseudoconvex domains
in Cn with Ωi ∩ Ωj = ∅ (i 6= j) and we shall consider the case
(6.20) Ω =N⋃
j=1
Ωj.
Theorem 6.6 ([24]). Let Ω be of the form (6.20) and satisfy the assumption:
(6.21) Ω is holomorphically convex in Ω0
where Ω0 is a bounded smooth strongly pseudoconvex domain in Cn. Problem (6.19)
admits a unique plurisubharmonic solution u ∈ C1,1(Ωc). Moreover, u satisfies the a
priori estimates
(6.22) −C ≤ u− log |z| ≤ C in Ωc,
(6.23) |∇u|2 + ∆u ≤ C
|z|2in Ωc
Consequently, VΩ ∈ C1,1(Ωc).
The estimate in (6.23) shows certain resemblance beyond (6.22) between VΩ and
log |z| which is the pluricomplex Green function of the closed unit ball. We also note
that condition (6.21) is necessary for Theorem 6.6, and is clearly satisfied when Ω is
a smooth strongly pseudoconvex domain, that is, when N = 1 in (6.20).
To prove Theorem 6.6 we first construct a smooth strictly plurisubharmonic func-
tion defined on Ωc which vanishes on ∂Ω using condition (6.21). Without loss of
generality let us assume that
(6.24) Br0 ⊂⊂ Ω ⊂⊂ Ω0 ⊂⊂ B1
for some 0 < r0 ≤ 12, where Br and Br denote the ball of radius r and its closure,
respectively, centered at the origin in Cn.
COMPLEX MONGE-AMPERE EQUATIONS 43
Let σ0 : Cn → R be a smooth strictly plurisubharmonic defining function of Ω0:
(6.25) Ω0 = z ∈ Cn : σ0(z) < 0 and ∇σ0 6= 0 on ∂Ω0.
By Proposition 1.1 of [27] (see Section 5) we find a strictly plurisubharmonic function
σ ∈ C∞(Ωc) such that σ = 0 on ∂Ω and σ = 1 + σ0 in a neighborhood of Ωc0. Now
fix λ0 ∈ (0, 12) sufficiently small such that 2λ0σ < 1 in B2. For ε ∈ (0, λ0), note that
1
2log
ε2 + |z|2
1 + ε2>
log 3
2> λ0σ on ∂B2
and
(6.26) logε2 + |z|2
1 + ε2≤ 0 ≤ λ0σ in B1 \ Ω.
Using the smoothing technique of P.-F. Guan [27] we can construct a strictly plurisub-
harmonic function uε ∈ C∞(Ωc) satisfying
(6.27) uε(z) =
λ0σ in B1 \ Ω12log ε2+|z|2
1+ε2 in Cn \B2
and
(6.28) uε(z) ≥ maxλ0σ,
1
2log
ε2 + |z|2
1 + ε2
in B2 \B1.
For s > 0 set
(6.29)
ψs(z) :=1
2ndet
∂2
∂zj∂zk
log(s2 + |z|2)
=1
2ndet
(s2 + |z|2)δij − zizj
(s2 + |z|2)2
=s2
2n(s2 + |z|2)n+1.
From the construction in [27] (Proposition 1.1 and Lemma 3.2 (ii)) we see that
(6.30) detuεij ≥ ψε(z) in Ωc
if 0 < ε ≤ ε0 where
ε0 := (2λ0)n2 rn+1
0 δ120 ≤ 2λ0r
n+10 ≤ 1
2nminλ0, r0
and 0 < δ0 ≤ 1 satisfies
(6.31) detσjk ≥ δ0 in B2.
44 BO GUAN
For fixed R ≥ 2 let ΣR = BR \ Ω and consider the Dirichlet problem
(6.32)
detuij = ψε in ΣR
u = uε on ∂ΣR.
Clearly uε is a smooth strictly plurisubharmonic subsolution, so we may apply The-
orem 2.7 to conclude that there exists a unique strictly plurisubharmonic solution
uε,R ∈ C∞(ΣR) of (6.32). By the maximum principle,
(6.33) uε ≤ uε,R ≤ uε,R′< log
|z|r0
in ΣR
when R′ ≥ R ≥ 2 since these inequalities hold on ∂ΣR and
(6.34) det(log |z|)ij = 0.
It follows from (6.33) that
(6.35) uε(z) := limR→∞
uε,R(z)
exists for all z ∈ Ωc.
We shall next derive the following estimates for uε,R:
(6.36) |∇uε,R|2 + ∆uε,R ≤ C
|z|2in ΣR
where C is a positive constant independent of ε and R. For convenience we write
u = uε,R and u = uε.
Consider the function
ρ(z) = α(ε, R) log|z|r0
where
α(ε, R) =log(ε2 +R2)− log(1 + ε2)
2(logR− log r0)<
logR
logR/r0< 1,
and let h = hR ∈ C1,1(ΣR) be the unique plurisuharmonic solution of the problem
(see Theorem 5.2)
(6.37)
dethij = 0 in ΣR,
h = 0 on ∂Ω
h = logR− log r0 on ∂BR.
Since
(6.38)
u = u = h < ρ on ∂Ω
u = u = ρ < h on ∂BR,
COMPLEX MONGE-AMPERE EQUATIONS 45
by the maximum principle we have
(6.39) u < u < ρ and u < u < h in ΣR.
It follows that
(6.40)α(ε, R)
R= −∂ρ
∂ν< −∂u
∂ν< −∂u
∂ν=
R
ε2 +R2<
1
Ron ∂BR
where ν denotes the interior unit normal to ∂ΣR so it is interior to ∂BR but exterior
to ∂Ω, and
(6.41) 0 < c1 ≤∂u
∂ν<∂u
∂ν<∂h
∂ν≤ c2 on ∂Ω
for some uniform positive constants c1 and c2 independent of ε and R. In fact we
may choose
c1 = λ0 min∂Ω
∂σ
∂ν, c2 = max
∂Ω
∂h2
∂ν> 0
since hR ≤ h2 for all R ≥ 2.
Note that |uν | = 2|∇u| on ∂ΣR since u is constant on each component of ∂ΣR. By
(6.40) and (6.41) we have
sup∂ΣR
eu|∇u| ≤ 1
2max
c2,
1
R
(ε2 +R2
1 + ε2
) 12≤ maxc2, 1
2.
Applying Corollary 3.3 with η(u) = u, ψ(z, u) = logψε(z), we obtain
(6.42) |∇u| ≤ C1e−u in ΣR,
where
C1 = maxc2
2,1
2, sup
ΣR
eu|∇ logψε|n+ 2
.
Since
|∇ logψε| = (n+ 1)|z|ε2 + |z|2
,
from (6.39) we see that C1 is independent ε and R.
To complete the proof of (6.36) we next estimate the second derivatives of u on
∂ΣR. Let us first consider ∂BR. Suppose R ≥ 4 and let
u(ξ) = u(Rξ)− a, ξ ∈ B1 \B 12
where
a = u|∂BR=
1
2log
ε2 +R2
1 + ε2.
46 BO GUAN
By (6.34) and (6.39) we have
(6.43) − log 2 ≤ u ≤ u ≤ ρ ≤ 0 in B1 \B 12
where, with ε = ε/R,
u(ξ) = u(Rξ)− a =1
2log
ε2 + |ξ|2
1 + ε2,
ρ(ξ) = ρ(Rξ)− a = α(ε, R) log |ξ|.Next,
(6.44) uξi= Ruzi
, uξi= Ruzi
and
uξiξj= R2uzizj
.
Therefore,
(6.45) det uξiξj= ψε(ξ) :=
ε2
2n(ε2 + |ξ|2)n+1in B1 \B 1
2.
By (6.40), (6.42), (6.43) and (6.44) we have
(6.46) |∇u| ≤ C in B1 \B 12
and
(6.47) α(ε0, R) ≤ α(ε, R) < −∂u∂ν
<1
1 + ε2on ∂B1.
Since u = 0 on ∂B1, we may write
u(ξ) = (|ξ|2 − 1)Λ(ξ), in B1 \B 12
where Λ is a nonnegative function. We have
(6.48)∂u
∂ν= −2Λ on ∂B1.
For an arbitrary point p ∈ ∂B1, we choose coordinates of Cn
ξ1 = t1 + it2, . . . , ξn = t2n−1 + it2n
such that p is given by ξ1 = 1, ξ2 = . . . = ξn = 0. We have at p,
(6.49) utitj = 2Λδij, ∀ 2 ≤ i, j ≤ 2n.
Therefore, by (6.47) and (6.48),
(6.50) |utitj(p)| ≤ 1, ∀ 2 ≤ i, j ≤ 2n
COMPLEX MONGE-AMPERE EQUATIONS 47
and
(6.51)1
2α(ε0, R) ≤ uξj ξj
(p) = Λ ≤ 1
2, ∀ 2 ≤ j ≤ n.
Furthermore, by the mixed tangential-normal derivative estimates in Subsection 3.3
we obtain
(6.52) |ut1tj(p)| ≤ C, ∀ 2 ≤ j ≤ 2n.
Finally, solving uξ1ξ1 from Equation (6.45) and using (6.50)-(6.52) we derive
(6.53) uξ1ξ1(p) =det uξiξj
uξ2ξ2 · · · uξnξn
+∑i≥2
|uξ1ξi|2
uξiξi
≤ C.
In both (6.52) and (6.53) the constant C is independent of ε. Since |ut2t2(p)| ≤ 1 by
(6.50), we see that (6.53) implies
(6.54) |ut1t1(p)| ≤ C.
Combining this with (6.50) and (6.52), we have proved
(6.55) |utitj | ≤ C on ∂B1, ∀ 1 ≤ i, j ≤ 2n.
By rescaling we obtain a priori estimates for all (real) second derivatives of u on ∂BR:
(6.56) |D2u| ≤ C
R2on ∂BR
where C is independent of ε and R.
The estimate for second derivatives on ∂Ω is similar to that (on Γ0) in Section 5:
(6.57) |D2u| ≤ C on ∂Ω independent of ε and R.
Now we apply Corollary 3.11 with η(u) = 2u, f(z, u) = logψε(z) which satisfies
∆ logψε = −(n+ 1)( n
ε2 + |z|2− |z|2
(ε2 + |z|2)2
)≥ −n(n+ 1)
ε2 + |z|2.
Since C−1|z|−1 ≤ e−u ≤ e−u ≤ C|z|−1 in ΣR, we have
−e2u∆ logψε ≤ C in ΣR,
and, by (6.56) and (6.57),
e2u∆u ≤ C on ∂ΣR.
By Corollary 3.11,
(6.58) ∆u ≤ C2
|z|2
where C2 is independent ε and R. Combining with (6.42) this proves (6.36).
48 BO GUAN
References
[1] E. Bedford, Survey of pluri-potential theory, in Several Complex Variables: Proc. Mittag-LefflerInst. 1987-88, Math Notes 38, pp. 48-97, Princeton Univ. Press, Princeton 1993.
[2] E. Bedford and J.-P. Demailly, Two counterexamples concerning the pluri-complex Green func-tion in Cn, Indiana Univ. Math. J. 37 (1988), 865-867.
[3] E. Bedford and J. E. Fornaess, Counterexamples to regularity for the complex Monge-Ampereequation, Invent. Math. 50 (1979), 129-134.
[4] E. Bedford and B. A. Taylor, The Dirichlet problem for a complex Monge-Ampere equation,Invent. Math. 37 (1976) 1-44.
[5] E. Bedford and B. A. Taylor, Variational properties of the complex Monge-Ampere equation,II. Intrinsic norms, Amer. J. Math., 101 (1979), 1131-1166.
[6] E. Bedford and B. A. Taylor, A new capacity for plurisubharmonic functions, Acta Math. 149(1982), 1-40.
[7] E. Bedford and B. A. Taylor, Uniqueness for the complex Monge-Ampere equation for functionsof logarithmic growth, Indiana Univ. Math. J. 38 (1989), 456-469.
[8] Z. Blocki, The C1,1 regularity of the pluricomplex Green function, Michigan Math. J. 47 (2000),211-215
[9] Z. Blocki, Regularity of the pluricomplex Green function with several poles, Indiana Univ. Math.J. 50 (2001), 335-351.
[10] Z. Blocki, The complex Monge-Apere equation on compact Kahler manifolds, preprint, 2007.[11] Z. Blocki, A gradient estimate in the Calabi-Yau theorem, Math. Annalen 344 (2009), 317–327.[12] L. A. Caffarelli and X. Cabre, Fully nonlinear elliptic equations, AMS Colloquium Publications,
43. AMS, Providence, RI, 1995.[13] L. A. Caffarelli, J. J. Kohn, L. Nirenberg and J. Spruck, The Dirichlet problem for nonlinear
second-order elliptic equations II. Complex Monge-Ampere and uniformly elliptic equations,Comm. Pure Applied Math. 38 (1985), 209–252.
[14] L. A. Caffarelli, L. Nirenberg and J. Spruck, The Dirichlet problem for nonlinear second-orderelliptic equations I. Monge-Ampere equations, Comm. Pure Applied Math. 37 (1984), 369–402.
[15] X.-X. Chen, The space of Kahler metrics, J. Differential Geom. 56 (2000), 189-234.[16] S. S. Chern, H. I. Levine, L. Nirenberg, Intrinsic norms on a complex manifold, 1969 Global
Analysis (Papers in Honor of K. Kodaira) pp. 119–139, Univ. Tokyo Press, Tokyo.[17] J.-P. Demailly, Mesures de Monge-Ampere et mesures pluriharmoniques, Math. Z. 194 (1987),
519-564.[18] S. K. Donaldson, Symmetric spaces, Kahler geometry and Hamiltonian dynamics, Northern
California Symplectic Geometry Seminar, 13–33, Amer. Math. Soc. Transl. Ser. 2, 196, Amer.Math. Soc., Providence, RI, 1999.
[19] L. C. Evans, Classical solutions of fully nonlinear, convex, second order elliptic equations,Comm. Pure Applied Math., 35 (1982), 333–363.
[20] T. W Gamelin and N. Sibony, Subharmonicity for uniform algebras, J. Funct. Anal. 35 (1980),64-108.
[21] B. Guan, Boundary value problem for prescribed Gauss curvature and Monge-Ampere equations,Thesis, Univeristy of Massachusetts Amherst, 1992.
[22] B. Guan, The Dirichlet problem for Monge-Ampere equations in non-convex domains and space-like hypersurfaces of constant Gauss curvature, Trans. Amer. Math. Soc. 350 (1998), 4955-4971.
[23] B. Guan, The Dirichlet problem for complex Monge-Ampere equations and regularity of thepluri-complex Green function, Comm. Anal. Geom. 6 (1998), 687-703. A correction, 8 (2000),213–218.
COMPLEX MONGE-AMPERE EQUATIONS 49
[24] B. Guan, On the regularity of pluricomplex Green functions, IMRN 2007, rnm106, 19pp.[25] B. Guan and Y. Y. Li, Monge-Ampere equations on Riemannian manifolds, J. Differential
Equations 132 (1996), 126-139.[26] B. Guan and J. Spruck, Boundary value problem on Sn for surfaces of constant Gauss curvature,
Annals of Math. 138 (1993), 601-624.[27] P.-F. Guan, Extremal functions related to intrinsic norms, Annals of Math. 156 (2002), 197-211.[28] P.-F. Guan, Remarks on the homogeneous complex Momnge-Ampere equation related to the
Chern-Levine-Nirenberg conjecture, preprint 2008.[29] P.-F. Guan, A gradient estimate for complex Monge-Ampere equation, preprint.[30] D. Hoffman, H. Rosenberg and J. Spruck, Boundary value problems for surfaces of constant
Gauss curvature, Comm. Pure Applied Math., 45 (1992), 1051-1062.[31] M. Klimek, Pluripotential Theory, Oxford University Press, New York, 1991.[32] S. Kolodziej, The complex Monge-Ampere equation and pluripotential theory, Mem. Amer.
Math. Soc. 178 (2005), no. 840.[33] N. V. Krylov, Boundedly inhomogeneous elliptic and parabolic equations, (Russian) Izv. Akad.
Nauk SSSR Ser. Mat. 46 (1982), 487–523, 670. English translation: Math. USSR Izv. 22 (1984),no. 1, 67–98
[34] N. V. Krylov, Boundedly inhomogeneous elliptic and parabolic equations in a domain (Russian)Izv. Akad. Nauk SSSR Ser. Mat. 47 (1983), 75–108.
[35] N. V. Krylov, Smoothness of the payoff function for a controllable diffusion process in a domain(Russian) Izv. Akad. Nauk SSSR Ser. Mat. 53 (1989), 66–96; translation in Math. USSR-Izv.34 (1990), 65–95.
[36] L. Lempert, La metrique de Kobayashi et la representation des domains sur la boule, Bull. Sci.Mat. France 109 (1981), 427-474.
[37] L. Lempert, Solving the degenerate complex Monge-Ampere equation with one concentratedsingularity, Math. Ann. 263 (1983), 515-532.
[38] L. Lempert, Symmetries and other transformations of the complex Monge-Ampere equation,Duke Math. J. 52 (1985), 869-885.
[39] T. Mabuchi, Some symplectic geometry on compact Kahler manifolds. I, Osaka J. Math. 24(1987), 227–252.
[40] W. Plesniak, Siciak’s extremal function in complex and real analysis Ann. Polon. Math. 80(2003), 37–46.
[41] S. Semmes, Complex Monge-Ampere and symplectic manifolds, Amer. J. Math. 114 (1992),495–550.
[42] S. Semmes, A generalization of Riemann mappings and geometric structures on a space ofdomains in Cn, Memoirs Amer. Math. Soc., no. 472, 1992.
[43] J. Siciak, On some extremal functions and their applications in the theory of analytic functionsof several complex variables, Trans. Amer. Math. Soc. 105 (1962), 322–357.
[44] J. Siciak, Extremal plurisubharmonic functions in Cn, Ann. Polon. Math. 39 (1981), 175–211.[45] J. Siciak, Extremal plurisubharmonic functions and capacities in Cn, Sophia Kokyuroku in
Math. No. 14, Sophia University, Tokyo, 1982.[46] Y.-T. Siu, Lectures on Hermitian-Einstein metrics for stable bundles and Kahler-Einstein met-
rics, DMV Seminar, 8. Birkhuser Verlag, Basel, 1987. 171 pp.[47] G. Tian, Canonical metrics in Kahler geometry. Notes taken by Meike Akveld. Lectures in
Mathematics ETH Zrich. Birkhuser Verlag, Basel, 2000. vi+101 pp.[48] S.-T. Yau, On the Ricci curvature of a compact Kahler manifold and the complex Monge-Ampere
equation. I. Comm. Pure Appl. Math. 31 (1978), no. 3, 339–411.
50 BO GUAN
[49] V. P. Zahariuta, Extremal plurisubharmonic functions, orthogonal polynomials and Bernstein-Walsh theorem for analytic functions of several variables, Ann. Polon. Math. 33 (1976), 137-148.