To circumvent these difficulties, we shall formulate and establish extensions of the classical convergence and uniqueness theorems of BedfordTaylor for the MongeAmpère operator for domains in
${\mathbf{C}}^{n}$
to the case of Kähler manifolds with boundary.
3.1 Convergence of approximate solutions
The first of these extensions is the following convergence theorem, where the key hypothesis is the existence of uniform
${C}^{0}$
bounds for some transversal derivative of the
$\Phi \left(k\right)$
's at the boundary of the manifold
$\overline{M}$
.Let
$\overline{M}$
be a compact complex manifold with smooth boundary and
$M\subseteq \overline{M}$
be the interior of
$\overline{M}$
. Let
$\overline{M}={\cup}_{\alpha =1}^{N}{U}_{\alpha}$
be a covering of
$\overline{M}$
by a finite number of coordinate charts
${U}_{\alpha}$
.
Fix a smooth closed
$(1,1)$
form
${\Omega}_{0}$
on
$M$
, and let
${\Psi}_{\alpha}$
be smooth potentials for
${\Omega}_{0}$
on
${U}_{\alpha}$
, that is,
${\Omega}_{0}=\frac{\sqrt{1}}{2}\partial \overline{\partial}{\Psi}_{\alpha}$
on
${U}_{\alpha}$
. We define the class of
${\Omega}_{0}$
plurisubharmonic functions on
$M$
to be the class of functions
$\Phi $
on
$\overline{M}$
by
$$\begin{array}{c}PSH(M,{\Omega}_{0})=\{\Phi ;{\Psi}_{\alpha}+\Phi isplurisubharmonicon{U}_{\alpha},1\le \alpha \le N\}.\end{array}$$ 
(3.1)

Note that this condition means that
${\Psi}_{\alpha}+\Phi $
is upper semicontinuous and satisfies the submean value property. However, it is a stronger condition than the condition
${\Omega}_{0}+\frac{\sqrt{1}}{2}\partial \overline{\partial}\Phi \ge 0$
by itself, since it is a pointwise condition, while the condition
${\Omega}_{0}+\frac{\sqrt{1}}{2}\partial \overline{\partial}\Phi \ge 0$
depends only on the values of
$\Phi $
almost everywhere.
Next, let
$\phi $
be a continuous function on
$\partial \overline{M}$
, and
${\Omega}_{0}$
a real, smooth closed
$(1,1)$
form on
$M$
. Set
$d=\partial +\overline{\partial}$
,
${d}^{c}=\frac{\sqrt{1}}{4}(\overline{\partial}\partial )$
, so that
$d{d}^{c}=\frac{\sqrt{1}}{2}\partial \overline{\partial}$
.
Definition 1
Let
$\Phi :\overline{M}\to \mathbf{R}$
be an uppersemicontinuous function. We say
$\Phi $
is a solution of the Dirichlet problem with boundary values
$\phi $
if

1.
$\Phi \in PSH(M,{\Omega}_{0})$

2.
$\Phi $
is continuous at
$p$
for all
$p\in \partial \overline{M}$
and
$\Phi {}_{\partial \overline{M}}=\phi $
.

3.
$({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}=0$
on
$\overline{M}$
where
$m=dim\left(M\right)$
.
We wish to obtain a solution of the Dirichlet problem on
$\overline{M}$
from a sequence of approximate solutions. Let
$Y$
be a smooth real nowhere vanishing vector field on a neighborhood
$U\subseteq \overline{M}$
of
$\partial \overline{M}$
which is transversal to
$\partial \overline{M}$
, in the sense that for
$p\in \partial \overline{M}$
, the vector
$Y\left(p\right)$
is not tangent to
$\partial \overline{M}$
.
Theorem 3
Assume
${\Omega}_{0}^{m}=0$
. Let
${\Phi}_{k}\in PSH(M,{\Omega}_{0})\cap {C}^{\infty}\left(\overline{M}\right)$
have the properties:

1.
$\left{\Phi}_{k}\right$
is uniformly bounded on
$\overline{M}$
.

2.
$\leftY\right({\Phi}_{k}\left)\right$
is uniformly bounded on
$U$
;

3.
${\Phi}_{k}{}_{\partial \overline{M}}\to \phi $
uniformly;

4.
There is a sequence
${a}_{k}>{sup}_{x\in \partial \overline{M}}\left{\Phi}_{k}\right(x)\phi (x\left)\right$
with the property
${a}_{k}\searrow 0$
(i.e.
${a}_{k}$
is decreasing and approaching zero) and
${\sum}_{k}{a}_{k}<\infty $
.

5.
${lim}_{k\to \infty}{\int}_{M}({\Omega}_{0}+d{d}^{c}{\Phi}_{k}{)}^{m}=0$
Let
$$\Phi ={lim}_{k\to \infty}{\left[{sup}_{l\ge k}{\Phi}_{l}\right]}^{*}$$
Then
$\Phi $
is a solution of the Dirichlet problem with boundary values
$\phi $
.
Remark: Assumption 4. implies Assumption 3. On the other hand, after passing to a subsequence, Assumption 3. implies Assumption 4.Proof. We want to choose a sequence
${c}_{k}\searrow 0$
with the property in such a way that
${\Phi}_{k}+{c}_{k}$
is monotonically decreasing on
$\partial M$
. To do this, we just define
${c}_{k}=2{\sum}_{j\ge k}{a}_{j}$
. Then
${c}_{k}{c}_{k+1}=2{a}_{k}$
. Moreover, if
$x\in \partial M$
, then
$$\begin{array}{c}{\Phi}_{k}\left(x\right){\Phi}_{k+1}\left(x\right)>(\phi {a}_{k})(\phi +{a}_{k+1})\ge 2{a}_{k}={c}_{k+1}{c}_{k}\end{array}$$ 
(3.2)

Thus, replacing
${\Phi}_{k}$
by
${\Phi}_{k}+{c}_{k}$
, we may assume that
${\Phi}_{k}{}_{\partial \Omega}>{\Phi}_{k+1}{}_{\partial \Omega}$
.
Let
$${W}_{k}={\left[{sup}_{l\ge k}{\Phi}_{l}\right]}^{*}$$
Then
${W}_{k}\in PSH(M,{\Omega}_{0})$
by Theorem 5.7 of [
8]
. Moreover, ( 3.2 ) implies that
${W}_{k}={\Phi}_{k}$
in an open neighborhood of
$\partial \overline{M}$
. Thus
$$\begin{array}{c}{\int}_{M}({\Omega}_{0}+d{d}^{c}{W}_{k}{)}^{m}={\int}_{M}({\Omega}_{0}+d{d}^{c}{\Phi}_{k}{)}^{m}\end{array}$$ 
(3.3)

This follows from the following simple lemma:
Lemma 2
Let
$W,\Phi \in PSH(M,{\Omega}_{0})$
with
${\Omega}_{0}^{m}=0$
and assume that
$W=\Phi $
on some neighborhood of
$\partial \overline{M}$
. Then
$${\int}_{M}({\Omega}_{0}+d{d}^{c}W{)}^{m}={\int}_{M}({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}$$
Proof of Lemma 2 . To see this, let
$K$
be any compact subset such that
$W=\Phi $
on
$M\backslash K$
, and let
$\Psi \in {C}_{0}^{\infty}\left(M\right)$
be a function with
$\Psi =1$
on a neighborhood of
$K$
. Then, expanding
$({\Omega}_{0}+d{d}^{c}W{)}^{m}$
, we obtain
$${\int}_{M}\Psi ({\Omega}_{0}+d{d}^{c}W{)}^{m}={\int}_{M}\Psi {\Omega}_{0}^{m}+{\int}_{M}\Psi d({\Theta}_{W})={\int}_{M\backslash K}d\Psi \wedge {\Theta}_{W}$$
Here
${\Theta}_{W}={\sum}_{k=0}^{m1}\left(\genfrac{}{}{0ex}{}{m}{k}\right){\Omega}_{0}^{k}\wedge {d}^{c}W\wedge (d{d}^{c}W{)}^{m1k}$
. Since
${\Theta}_{W}={\Theta}_{\Phi}$
on
$M\backslash K$
, the lemma is proved.
Now assumption 5. of the theorem together with (
3.3 ) implies that
$({\Omega}_{0}+d{d}^{c}{W}_{k}{)}^{m}\to 0$
weakly. On the other hand, since
${W}_{k}\to \Phi $
monotonically, the BedfordTaylor monotonicity theorem (Theorem 2.1 of [
2]
; this is stated for domains in
${\mathbf{C}}^{n}$
, but generalizes in a straightforward fashion to manifolds) we have
$({\Omega}_{0}+d{d}^{c}{W}_{k}{)}^{m}\to ({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}$
weakly. Thus we have
$({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}=0$
. To finish the proof of the theorem, we must show that
$\Phi $
is continuous at the boundary and has the right boundary values.
Let
$\epsilon >0$
. Choose
${k}_{0}$
such that
$k\ge {k}_{0}\u27f9{sup}_{x\in \partial \overline{M}}\left{\Phi}_{k}\right(x)\phi (x\left)\right<\epsilon $
. Extend
$\phi $
to a continuous function on a neighborhood
$U\subseteq \overline{M}$
of
$\partial \overline{M}$
in such a way that
$\phi $
is constant on the flow lines of
$Y$
. Then assumption 2. implies that if
$U$
is sufficiently small, then
$$\begin{array}{c}{sup}_{x\in U}\left{\Phi}_{k}\right(x)\phi (x\left)\right<2\epsilon \end{array}$$ 
(3.4)

and thus
${sup}_{x\in U}\Phi (x)\phi (x\left)\right\le 2\epsilon $
. In particular,
$\Phi {}_{\partial \Omega}=\phi $
and
$\Phi $
is continuous at all points
$p\in \partial \overline{M}$
. This proves the theorem. Q.E.D.
3.2 The domination principle for the MongeAmpère operator
The proof of the extension of the BedfordTaylor uniqueness theorem in this section follows closely the original arguments of [
1,
2]
, and especially the exposition of Blocki [
3]
.
When we consider generalized solutions of a partial differential equation, a particularly desirable property is their uniqueness. For bounded domains in
${\mathbf{C}}^{n}$
, the uniqueness of the generalized solution of the Dirichlet problem for the MongeAmpère equation in the class
$PSH\cap {L}^{\infty}$
has been established by Bedford and Taylor [
1]
. It seems that this uniqueness theorem should extend as well to bounded domains in Kähler manifolds, at least if good smooth approximations of
${\Omega}_{\Phi \left(0\right)}$
plurisubharmonic functions exist. Although there are now many powerful approximation theorems (see [
8,
14]
and references therein), we found it more convenient to extend the BedfordTaylor uniqueness theorem to a situation adapted to the problem at hand, in the spirit of the earlier extension. The key hypothesis which we will exploit is a capacity zero condition.Recall the following notion of capacity of a set introduced by Bedford and Taylor [
2]
: If
$E\subseteq U$
is a Borel subset of a bounded domain
$U\subseteq {\mathbf{C}}^{n}$
then
$$\begin{array}{c}c(E,U)=sup\left\{{\int}_{E}(d{d}^{c}v{)}^{n};v\in PSH(U),0\le v\le 1\right\}.\end{array}$$ 
(3.5)

For our purposes, we can adapt this notion to Kähler manifolds as follows: Let
$M={\cup}_{\alpha =1}^{N}{U}_{\alpha}$
be a finite cover of
$M$
by coordinate neighborhoods. Then we say
$c(E,M)<\epsilon $
if we can write
$E={\cup}_{\alpha}{E}_{\alpha}$
with
${E}_{\alpha}\subseteq {U}_{\alpha}$
a Borel subset and
$$\begin{array}{c}{\sum}_{\alpha}c({E}_{\alpha},{U}_{\alpha})<\epsilon .\end{array}$$ 
(3.6)

We say that
$c(E,M)=0$
if
$c(E,M)<\epsilon $
for every
$\epsilon >0$
.
Lemma 3
There is a constant
$C>0$
with the following property: If
$E\subseteq M$
is a Borel subset,
$\epsilon >0$
and
$\Phi \in PSH({\Omega}_{0},M)$
, then
$$\begin{array}{c}{\int}_{E}({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}\le C\epsilon (1+sup\Phi {)}^{m}\end{array}$$ 
(3.7)

if
$c(E,M)<\epsilon $
. In particular, if
$c(E,M)=0$
, then for all functions
$\Phi \in PSH({\Omega}_{0},M)$
, we have
$$\begin{array}{c}{\int}_{E}({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}=0.\end{array}$$ 
(3.8)

Proof. Fix a smooth potential
${\Psi}_{\alpha}$
on
${U}_{\alpha}$
such that
${\Omega}_{0}=d{d}^{c}{\Psi}_{\alpha}$
. Then
$$\begin{array}{ccc}{\int}_{E}({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}& \le & {\sum}_{\alpha}{\int}_{{E}_{\alpha}}({\Omega}_{0}+d{d}^{c}\Phi {)}^{m}={\sum}_{\alpha}{\int}_{{E}_{\alpha}}(d{d}^{c}({\Psi}_{\alpha}+\Phi ){)}^{m}\end{array}$$  
$$\begin{array}{ccc}& \le & {\sum}_{\alpha}(sup{\Psi}_{\alpha}+\Phi \left{)}^{m}c\right({E}_{\alpha},{U}_{\alpha}).\end{array}$$ 
(3.9)

Since the
${\Psi}_{\alpha}$
are fixed, we have
${\Psi}_{\alpha}+\Phi \le C(1+\Phi \left\right)$
and the lemma follows.
The following lemma follows immediately from the quasicontinuity theorem of BedfordTaylor [
2]
:
Lemma 4
Let
$\Phi \in PSH(M,{\Omega}_{0})$
. Then for every
$\epsilon >0$
, there is an open set
$G\subseteq M$
such that
$c(G,M)<\epsilon $
and
$\Phi $
is continuous on
$M\backslash G$
.
Proof. Since
${\Psi}_{\alpha}+\Phi $
is a plurisubharmonic function on
${U}_{\alpha}$
, the quasicontinuity theorem of BedfordTaylor ([
2]
, Theorem 3.5) implies that there is an open set
${G}_{\alpha}\subseteq {U}_{\alpha}$
such that
${\Psi}_{\alpha}+\Phi $
is continuous on
${U}_{\alpha}\backslash {G}_{\alpha}$
and
$c({G}_{\alpha},{U}_{\alpha})<\epsilon $
. Let
$G={\cup}_{\alpha}{G}_{\alpha}$
. Then
$\Phi $
is continuous on
$M\backslash G$
and, by definition,
$c(G,M)<N\epsilon $
.
We also require a notion of “nearly continuous” functions:
Definition 2
We say that a bounded function
$v:\overline{M}\to \mathbf{R}$
is “nearly continuous” if

1.
There exists a lower semicontinuous function
${v}_{0}$
on
$\overline{M}$
such that
$v={v}_{0}^{*}$
;

2.
$\{{v}_{0}<v\}$
has capacity zero, that is,
$c\left(\right\{{v}_{0}<v\},M)=0$
;

3.
$v={v}_{0}$
on
$\partial M$
.
With this notion, we shall prove the following:
Theorem 4
Assume that
$\overline{M}$
is a complex manifold of dimension
$m$
with smooth boundary, let
$M=\overline{M}\backslash \partial M$
and let
${\Omega}_{0}$
be a real closed smooth
$(1,1)$
form on
$M$
satisfying
${\Omega}_{0}^{m}=0$
.
Let
$u,v\in PSH(M,{\Omega}_{0})\cap {L}^{\infty}$
be such that
$(uv{)}_{*}\ge 0$
on
$\partial \overline{M}$
. Assume as well:

1.
$u$
is continuous;

2.
There is a decreasing sequence
${v}_{k}$
of nearly continuous functions in
$PSH(M,{\Omega}_{0})$
such that
${v}_{k}\searrow v$
;

3.
For every
$\delta >0$
there is a compact set
$K\subseteq M$
such that
${v}_{k}{}_{M\backslash K}<v{}_{M\backslash K}+\delta $
for
$k$
sufficiently large.
Then
$$\begin{array}{c}{\int}_{u<v}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{u<v}({\Omega}_{0}+d{d}^{c}u{)}^{m}.\end{array}$$ 
(3.10)

Proof. We divide the proof into several steps.
Step 1. If we replace
$u$
by
$u+\delta $
, then we have
$\{u+\delta <v\}\uparrow \{u<v\}$
as
$\delta \downarrow 0$
. Since for any positive measure
$\mu $
we have
$\mu \left({E}_{j}\right)\to \mu (\cup {E}_{j})$
whenever
${E}_{j}$
is an increasing family of measurable sets, we may replace
$u$
by
$u+\delta $
. Thus we may assume that
$$\begin{array}{c}(uv{)}_{*}\ge \delta andthat{M}^{\prime}=\{u<v\}isrelativelycompactinM.\end{array}$$ 
(3.11)

Step 2. We prove the theorem under the assumption that
$v$
is continuous (in which case the hypotheses 2. and 3. are automatic, since we can take
${v}_{k}=v$
for all
$k$
).
For
$\epsilon >0$
let
${u}_{\epsilon}=max(u+\epsilon ,v)$
. Then
${u}_{\epsilon}=u+\epsilon $
on a neighborhood of
$\partial {M}^{\prime}$
. Since
$d{d}^{c}(u+\epsilon )=d{d}^{c}u$
, we can invoke Lemma 2 and conclude that
$$\begin{array}{c}{\int}_{{M}^{\prime}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}={\int}_{{M}^{\prime}}({\Omega}_{0}+d{d}^{c}u{)}^{m}.\end{array}$$ 
(3.12)

On the other hand,
${u}_{\epsilon}\downarrow v$
on
${M}^{\prime}$
and, by the BedfordTaylor monotonicity theorem,
$$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\to ({\Omega}_{0}+d{d}^{c}v{)}^{m}on{M}^{\prime}(weakconvergenceofmeasure).$$
Since
$\{u<v\}$
is open, we obtain
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {liminf}_{\epsilon \to 0}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}.$$
This completes step 2.
Step 3. Now we treat the case where
$v$
itself is nearly continuous (in which case the hypotheses 2. and 3. are again automatic, since we can take
${v}_{k}=v$
for all
$k$
). This step will be parallel to the argument in Step 2:
For
$\epsilon >0$
let
${u}_{\epsilon}=max(u+\epsilon ,v)$
. Then
${u}_{\epsilon}\searrow v$
on the open set
$\{u<{v}_{0}\}$
(this set is open since
${v}_{0}$
is lower semicontinuous). In particular, the BedfordTaylor monotonicity theorem implies
$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\to ({\Omega}_{0}+d{d}^{c}v{)}^{m}$
weakly on
$\{u<{v}_{0}\}$
(as measures or as currents the two notions of weak convergence are equivalent). Thus we have
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}={\int}_{\{u<{v}_{0}\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {liminf}_{\epsilon \to 0}{\int}_{\{u<{v}_{0}\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}$$
$$={liminf}_{\epsilon \to 0}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}$$
The first equality follows from the assumption that
$\{{v}_{0}<v\}$
has capacity zero and the inequality from the fact that
$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\to ({\Omega}_{0}+d{d}^{c}v{)}^{m}$
weakly. Here we are making strong use of the fact that
$\{u<{v}_{0}\}$
is open.
Next we claim that for all
$\epsilon >0$
:
$$\begin{array}{c}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}={\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}.\end{array}$$ 
(3.13)

To see this, let
$A$
be an open set containing
$\{u<v\}$
, which is relatively compact in
$M$
.
Then,
$p\in \partial A$
implies that
$p/\in A$
, and we have
$u\left(p\right)\ge v\left(p\right)$
, so
${u}_{\epsilon}\left(p\right)=u\left(p\right)+\epsilon >v\left(p\right)$
.
Since the set
$\{u+\epsilon >v\}$
is open (
$u$
is continuous and
$v$
is upper semicontinuous), we see that
$u+\epsilon ={u}_{\epsilon}$
in a neighborhood of
$\partial A$
. Thus Lemma 2 implies
$${\int}_{A}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}={\int}_{A}({\Omega}_{0}+d{d}^{c}(u+\epsilon ){)}^{m}={\int}_{A}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Since
$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{n}$
and
$({\Omega}_{0}+d{d}^{c}u{)}^{n}$
are positive Borel measures which are finite on compact subsets of
$M$
, they are both regular. In particular
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}& =& {inf}_{A}{\int}_{A}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\end{array}$$  
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}& =& {inf}_{A}{\int}_{A}({\Omega}_{0}+d{d}^{c}u{)}^{m}\end{array}$$ 
(3.14)

where the inf is taken over all open sets
$A$
which contain
$\{u<v\}$
. This proves ( 3.13 ).
Step 4. We treat the general case. From step 1. we can assume that
$(uv{)}_{*}\ge \delta >0$
.
Thus
$u>{v}_{k}$
on a neighborhood of
$\partial M$
for
$k$
sufficiently large. By Lemma 4 , there is an open set
$G\subseteq M$
such that
$c(G,M)<\epsilon $
(the capacity of
$G$
in
$M$
) and such that
$v$
is continuous on
$F=M\backslash G$
. Choose
$\phi $
which is continuous
$M$
such that
$\phi =v$
on
$F$
. Then
$\{u<v\}\subseteq \{u<\phi \}\cup G$
implies
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}& \le & {\int}_{\{u<\phi \}}({\Omega}_{0}+d{d}^{c}v{)}^{m}+{\int}_{G}({\Omega}_{0}+d{d}^{c}v{)}^{m}\end{array}$$  
$$\begin{array}{ccc}& \le & {\int}_{\{u<\phi \}}({\Omega}_{0}+d{d}^{c}v{)}^{m}+C\epsilon \end{array}$$ 
(3.15)

where the last inequality follows from Lemma 3 , the fact that
$c(G,M)<\epsilon $
and that the function
$v$
is bounded. Here the constant
$C$
depends only on the sup norm of
$v$
. Šince
$\{u<\phi \}$
is open and
${v}_{k}\searrow v$
, the BedfordTaylor monotonicity theorem implies
$${\int}_{\{u<\phi \}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {liminf}_{k\to \infty}{\int}_{\{u<\phi \}}({\Omega}_{0}+d{d}^{c}{v}_{k}{)}^{m}$$
Now
$$\{u<\phi \}\subseteq \{u<v\}\cup G\subseteq \{u<{v}_{k}\}\cup G$$
Thus
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {liminf}_{k\to \infty}{\int}_{\{u<{v}_{k}\}}({\Omega}_{0}+d{d}^{c}{v}_{k}{)}^{m}+2C\epsilon ,$$
since our assumptions imply that the functions
${v}_{k}$
are uniformly bounded. Next, the assumption 3. implies that the sets
$\{u<{v}_{k}\}$
are all contained in a relatively compact subset of
$M$
. Using Step 3,
$${\int}_{\{u<{v}_{k}\}}({\Omega}_{0}+d{d}^{c}{v}_{k}{)}^{m}\le {\int}_{\{u<{v}_{k}\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
and since
${v}_{k}\searrow v$
,
${\cap}_{k}\{u<{v}_{k}\}=\{u\le v\}$
, we can conclude that
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u\le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}+2C\epsilon $$
Since
$\epsilon $
is arbitrary:
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u\le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Applying this last inequality to
$u+\eta $
and
$v$
, for some
$\eta >0$
:
$${\int}_{\{u+\eta <v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u+\eta \le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Finally, taking the limit as
$\eta \to 0$
and noting that
${\cup}_{\eta >0}\{u+\eta <v\}={\cup}_{\eta >0}\{u+\eta \le v\}=\{u<v\}$
, we obtain ( 3.16 ).
Next we prove another version of the domination theorem, but this time with the roles of
$u,v$
reversed:
Theorem 5
Assume that
$\overline{M}$
is a complex manifold of dimension
$m$
with smooth boundary, let
$M=\overline{M}\backslash \partial M$
and let
${\Omega}_{0}$
be a real closed smooth
$(1,1)$
form on
$M$
with the property:
${\Omega}_{0}^{m}=0$
. Let
$u,v\in PSH(\overline{M},{\Omega}_{0})\cap {L}^{\infty}$
satisfy
$(uv{)}_{*}\ge 0$
on
$\partial \overline{M}$
. Assume as well:

1.
$v$
is continuous;

2.
There is a decreasing sequence
${u}_{k}$
of nearly continuous PSH functions on
$M$
such that
${u}_{k}\searrow u$
.
Then
$$\begin{array}{c}{\int}_{u<v}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{u<v}({\Omega}_{0}+d{d}^{c}u{)}^{m}.\end{array}$$ 
(3.16)

Proof. The proof is parallel to that of Theorem 4 , although there are some important differences. We again divide it into several steps. Štep 1. As before, we may assume that
$$(uv{)}_{*}\ge \delta andthat{M}^{\prime}=\{u<v\}isrelativelycompactinM$$
Moreover, we have
$$\{{u}_{0}<v\}isrelativelycompactinM$$
To see this, observe first that
$u\ge v+\delta /2$
in an open neighborhood of
$\partial M$
. Also,
$\{{u}_{0}u>\delta /4\}$
is open since
${u}_{0}$
is lower semicontinuous and
$u$
is upper semicontinuous.
Since
${u}_{0}=u$
on
$\partial M$
, the set
$\{{u}_{0}u>\delta /4\}$
is an open neighborhood of
$\partial M$
. Thus
${u}_{0}>v+\delta /4$
in an open neighborhood of
$\partial M$
.
Step 2. Now we treat the case where
$u$
itself is nearly continuous (in which case hypotheses 2. and 3. are automatic, since we can take
${u}_{k}=u$
for all
$k$
).
For
$\epsilon >0$
let
${u}_{\epsilon}=max(u+\epsilon ,v)$
. Then
${u}_{\epsilon}\searrow v$
on the open set
$\{u<v\}$
(this set is open since
$u$
is upper semicontinuous). As before, the BedfordTaylor monotonicity theorem implies
$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\to ({\Omega}_{0}+d{d}^{c}v{)}^{m}$
weakly and, making use of the fact that
$\{u<v\}$
is open, we get
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}& \le & {liminf}_{\epsilon \to 0}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\end{array}$$  
$$\begin{array}{ccc}& =& {liminf}_{\epsilon \to 0}{\int}_{\{{u}_{0}<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m},\end{array}$$ 
(3.17)

where
${u}_{0}$
is the function which appears in the definition of the near continuity of
$u$
. Next we claim that for all
$\epsilon >0$
:
$$\begin{array}{c}{\int}_{\{{u}_{0}<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}={\int}_{\{{u}_{0}<v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}\end{array}$$ 
(3.18)

The proof is similar to that of ( 3.13 ), using
$A$
an open set containing
$\{{u}_{0}<v\}$
,
${u}_{0}\left(p\right)\ge v\left(p\right)$
for
$p\in \partial A$
so that
${u}_{0}\left(p\right)+\epsilon >v\left(p\right)$
. We use now the continuity of
$v$
and the lower semicontinuity of
${u}_{0}$
to deduce that the set
$\{{u}_{0}+\epsilon >v\}$
is open, and
$u+\epsilon ={u}_{\epsilon}$
in a neighborhood of
$\partial A$
. As before, Lemma 2 implies
$$\begin{array}{c}{\int}_{A}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}={\int}_{A}({\Omega}_{0}+d{d}^{c}(u+\epsilon ){)}^{m}={\int}_{A}({\Omega}_{0}+d{d}^{c}u{)}^{m},\end{array}$$ 
(3.19)

and, using the fact that
$({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}$
and
$({\Omega}_{0}+d{d}^{c}u{)}^{m}$
are positive Borel measures which are finite on compact subsets of
$M$
,
$$\begin{array}{ccc}{\int}_{\{{u}_{0}<v\}}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}& =& {inf}_{A}{\int}_{A}({\Omega}_{0}+d{d}^{c}{u}_{\epsilon}{)}^{m}\end{array}$$  
$$\begin{array}{ccc}{\int}_{\{{u}_{0}<v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}& =& {inf}_{A}{\int}_{A}({\Omega}_{0}+d{d}^{c}u{)}^{m}\end{array}$$ 
(3.20)

where the inf is taken over all open sets
$A$
which contain
$\{{u}_{0}<v\}$
. This proves ( 3.18 ).
Since
$u$
and
${u}_{0}$
differ only on a set of capacity 0, we obtain the desired inequality.
Step 4. We treat the general case: From step 1. we can assume that
$(uv{)}_{*}\ge \delta >0$
. Thus
${u}_{j}>v$
on a neighborhood of
$\partial M$
for
$j$
sufficiently large. Choose an open set
$G\subseteq M$
such that
$c(G,M)<\epsilon $
(the capacity of
$G$
in
$M$
) and such that
$u$
is continuous on
$F=M\backslash G$
.
Choose
$\phi $
which is continuous
$M$
such that
$\phi =u$
on
$F$
. Now
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}& =& {lim}_{j\to \infty}{\int}_{\{{u}_{j}<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {lim}_{j\to \infty}{\int}_{\{{u}_{j}<v\}}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}\end{array}$$  
$$\begin{array}{ccc}& \le & {liminf}_{j\to \infty}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}\end{array}$$ 
(3.21)

where we have made use of step 3. and the fact that
${u}_{j}>v$
on
$\partial M$
to prove the first inequality. Now we have
$\{u<v\}\subseteq \{\phi <v\}\cup G\subseteq (K\cap \{\phi \le v\}\cup G$
where
$K\subseteq M$
is a compact set such that
$\{u<v\}\subseteq K$
so
$$\begin{array}{ccc}{\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}& \le & {\int}_{\{\phi \le v\}\cap K}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}+{\int}_{G}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}\end{array}$$  
$$\begin{array}{ccc}& \le & {\int}_{\{\phi \le v\}\cap K}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}+C\epsilon \end{array}$$ 
(3.22)

where the last inequality follows from the fact that
$c(G,M)<\epsilon $
and that the function
$v$
is bounded. Here the constant
$C$
depends only on the sup norm of
$v$
.
Since
$\{\phi \le v\}\cap K$
is compact, the BedfordTaylor monotonicity theorem implies
$${limsup}_{k\to \infty}{\int}_{\{\phi \le v\}\cap K}({\Omega}_{0}+d{d}^{c}{u}_{j}{)}^{m}\le {\int}_{\{\phi \le v\}\cap K}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Now
$$\{\phi \le v\}\cap K\subseteq \{\phi \le v\}\subseteq \{u\le v\}\cup G$$
Thus
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u\le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}+2C\epsilon $$
Since
$\epsilon $
is arbitrary:
$${\int}_{\{u<v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u\le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Applying this last inequality to
$u+\eta $
and
$v$
, for some
$\eta >0$
:
$${\int}_{\{u+\eta <v\}}({\Omega}_{0}+d{d}^{c}v{)}^{m}\le {\int}_{\{u+\eta \le v\}}({\Omega}_{0}+d{d}^{c}u{)}^{m}$$
Finally, taking the limit as
$\eta \to 0$
, we obtain ( 3.16 ).
We can now state and prove the uniqueness theorem which we need for the proof of Theorem 1:
Theorem 6
Let
$u,v$
be as in Theorem 4 , that is,

1.
$u$
is continuous;

2.
There is a decreasing sequence
${v}_{k}$
of nearly continuous PSH functions on
$M$
such that
${v}_{k}\searrow v$
;

3.
For every
$\delta >0$
there is a compact set
$K\subseteq M$
such that
${v}_{k}{}_{M\backslash K}<v{}_{M\backslash K}+\delta $
for
$k$
sufficiently large.
Assume that
$(uv{)}_{*}=(uv{)}^{*}=0$
on
$\partial M$
and that
$({\Omega}_{0}+d{d}^{c}u{)}^{m}=({\Omega}_{0}+d{d}^{c}v{)}^{m}=0$
.
Then
$u=v$
.
Proof. We may assume, after replacing
$u$
and
$v$
by
$u+C$
and
$v+C$
for some constant
$C$
, that
$u$
and
$v$
are positive. We wish to show
$u=v$
. Assume not: Let
$\psi \in {C}^{\infty}\left(M\right)$
be such that
${\Omega}_{0}+d{d}^{c}\psi >0$
. Replacing
$\psi $
by
$\psi C$
, we may assume that
$\psi <0$
on
$M$
.
Case 1.
$\{u<v\}\ne \varnothing $
. This implies
$\{u<(1\epsilon )v+\epsilon \psi \}\ne \varnothing $
for some
$\epsilon >0$
. Let
$p\in \{u<(1\epsilon )v+\epsilon \psi \}$
and let
$D$
be a disk in some coordinate neighborhood of
$p$
. Then
$D\cap \{u<(1\epsilon )v+\epsilon \psi \}$
has nonzero Lebesgue measure (in general, if
$u,v$
are psh functions such that
$u=v$
almost everywhere in a disk
$D$
, then
$u=v$
everywhere in
$D$
; this follows from local regularization). Now we have, using Theorem 4
$$\begin{array}{ccc}0={\int}_{\{u<(1\epsilon )v+\epsilon \psi \}}({\Omega}_{0}+d{d}^{c}u{)}^{m}& \ge & {\int}_{\{u<(1\epsilon )v+\epsilon \psi \}}({\Omega}_{0}+d{d}^{c}[(1\epsilon )v+\epsilon \psi ]{)}^{m}\end{array}$$  
$$\begin{array}{ccc}& \ge & {\epsilon}^{m}{\int}_{\{u<(1\epsilon )v+\epsilon \psi \}}({\Omega}_{0}+d{d}^{c}\psi {)}^{m}>0\end{array}$$ 
(3.23)

Case 2.
$\{v<u\}\ne \varnothing $
. This is treated exactly in the same was as in case 1 except that we use Theorem 5 instead of Theorem 4 . Q.E.D.
4 Proof of Theorem 1
We can give now the proof of Theorem 1.First, we apply Theorem 3 to construct a generalized solution of the Dirichlet problem for the MongeAmpère equation
$$\begin{array}{c}{\Omega}_{\Phi}^{n+1}=0onM=X\times A,\Phi {}_{\partial M}=\phi ,\end{array}$$ 
(4.1)

where
$\phi :\partial M\to \mathbf{R}$
is defined by
$\phi {}_{\leftw\right=1}=0$
and
$\phi {}_{\leftw\right=e}=log\frac{{h}_{0}}{{h}_{1}}$
. Define
$\Phi \left(k\right)(z,w)=\phi (t;k)$
, where
$t=log\leftw\right$
and
$\phi (t;k)$
is defined as in ( 1.9 ). Let
$Y={\partial}_{t}$
. Then
$\Phi (k\left)\right\le C$
and
$\leftY\right(\Phi \left(k\right)\left)\right\le C$
by Lemma 1. We also have
$\left\right\Phi \left(k\right){}_{\partial \overline{M}}\phi {}_{{L}^{\infty}}\le C\frac{1}{{k}^{2}}$
by ( 1.7 ), so that
$\Phi (k{)}_{\partial \overline{M}}\to \phi $
uniformly. Furthermore, Theorem 2 implies that
${\int}_{M}{\Omega}_{\Phi \left(k\right)}^{n+1}\to 0$
as
$k\to \infty $
.
Thus, since
${\sum}_{k=1}^{\infty}\frac{1}{{k}^{2}}<\infty $
, we can apply Theorem 3, and conclude that
$$\begin{array}{c}\Phi =li{m}_{k\to \infty}[su{p}_{l\ge k}\Phi (l){]}^{*}\end{array}$$ 
(4.2)

is a generalized solution of the desired Dirichlet problem ( 4.1 ).
Consider next the
${C}^{1,1}$
geodesic
${\phi}_{t}$
joining
${\phi}_{0}$
to
${\phi}_{1}$
in the space
$\mathcal{\mathscr{H}}$
of Kähler potentials.
Let
$\stackrel{~}{\Phi}(z,w)={\phi}_{t}\left(z\right)$
with
$t=log\leftw\right$
as before. We shall show that
$\Phi =\stackrel{~}{\Phi}$
. To do this, we would like to apply Theorem 6 with
$$\begin{array}{c}u=\stackrel{~}{\Phi},({v}_{k}{)}_{0}=su{p}_{l\ge k}\Phi (l),{v}_{k}=({v}_{k}{)}_{0}^{*},v=\Phi ={lim}_{k\to \infty}{v}_{k}\end{array}$$ 
(4.3)

First, we show that
${v}_{k}$
is nearly continuous. Since
$\Phi \left(l\right)$
is smooth,
$({v}_{k}{)}_{0}$
is lower semicontinuous.
Moreover,
$\left\{\right({v}_{k}{)}_{0}<{v}_{k}\}$
has capacity 0, by Proposition 5.1 of [
2]
. It remains to show that
${v}_{k}=({v}_{k}{)}_{0}$
on
$\partial \overline{M}$
. The equation ( 1.7 ) implies
$D\Phi (l\left)\right(z,w\left)\right\le C$
, for some constant
$C$
independent of
$l$
, if
$(z,w)\in \partial \overline{M}$
and
$D$
is any derivative tangent to
$\partial \overline{M}$
. Thus, if
$\delta >0$
, there exists
$\epsilon >0$
such that if
$({z}_{0},{w}_{0})\in \partial \overline{M}$
then
$$\begin{array}{c}\Phi \left(l\right)({z}_{1},{w}_{0})+\delta >\Phi \left(l\right)({z}_{0},{w}_{0})>\Phi \left(l\right)({z}_{1},{w}_{0})\delta ,\end{array}$$ 
(4.4)

for all
$l$
if
${z}_{1}{z}_{0}<\epsilon $
. Also,
$\leftY\right(\Phi \left(l\right)\left)\right\le C$
implies that
$$\begin{array}{c}\Phi \left(l\right)({z}_{1},{w}_{1})+\delta >\Phi \left(l\right)({z}_{1},{w}_{0})>\Phi \left(l\right)({z}_{1},{w}_{1})\delta if{w}_{0}{w}_{1}<\epsilon .\end{array}$$ 
(4.5)

We have then
$\Phi \left(l\right)({z}_{1},{w}_{1})+2\delta >\Phi \left(l\right)({z}_{0},{w}_{0})>\Phi \left(l\right)({z}_{1},{w}_{1})2\delta $
, which implies that
$$\begin{array}{c}\left({v}_{k}{)}_{0}\right({z}_{1},{w}_{1})+2\delta >({v}_{k}{)}_{0}({z}_{0},{w}_{0})>\left({v}_{k}{)}_{0}\right({z}_{1},{w}_{1})2\delta ,\end{array}$$ 
(4.6)

and hence
$$\begin{array}{c}\left({v}_{k}{)}_{0}\right({z}_{1},{w}_{1})+2\delta >({v}_{k}{)}_{0}({z}_{0},{w}_{0})>{v}_{k}({z}_{1},{w}_{1})2\delta ,\end{array}$$ 
(4.7)

if
${z}_{1}{z}_{0}<\epsilon $
,
${w}_{1}{w}_{0}<\epsilon $
. In particular,
$\left({v}_{k}{)}_{0}\right({z}_{0},{w}_{0})>{v}_{k}({z}_{0},{w}_{0})2\delta $
for all
$\delta $
so
$({v}_{k}{)}_{0}={v}_{k}$
on
$\partial M$
so
${v}_{k}$
is indeed nearly continuous.
Thus the first two assumptions of Theorem
6 are satisfied.
In the next step, we will need the bound
$$\begin{array}{c}v({z}_{1},{w}_{1})+2\delta >v({z}_{0},{w}_{0})>v({z}_{1},{w}_{1})2\delta ,\end{array}$$ 
(4.8)

which follows by taking the limit of ( 4.7 ) as
$k\to \infty $
.
Next, we verify assumption 3: Since
${v}_{k}=({v}_{k}{)}_{0}$
on
$\partial M$
we see that
${v}_{k}$
is both upper and lower semicontinuous on
$\partial M$
and thus
${v}_{k}$
is continuous on
$\partial M$
. Moreover, by Theorem 3,
$v=\phi $
on
$\partial M$
so
$v$
is also continuous on
$\partial M$
. Since
${v}_{k}\searrow v$
we see, by Dini's theorem, that
${v}_{k}\searrow v$
uniformly on
$\partial M$
. Thus, for every
$\delta >0$
( 4.7 ) and ( 4.8 ) imply that for
$k>>0$
$${v}_{k}({z}_{1},{w}_{1})2\delta <{v}_{k}({z}_{0},{w}_{0})<v({z}_{0},{w}_{0})+\delta <v({z}_{1},{w}_{1})+3\delta $$
for all
${z}_{1}{z}_{0}<\epsilon $
and
${w}_{1}{w}_{0}<\epsilon $
.
All the conditions of Theorem
6 are satisfied. We can thus conclude that
$\Phi =\stackrel{~}{\Phi}$
.
Finally, the uniform convergence of the functions
$[su{p}_{k\ge l}\phi (t,k){]}^{*}$
follows from their upper semicontinuity and the compactness of
$X$
. This is essentially Dini's theorem, and can be proven as follows. Assume that
${u}_{n}$
is a sequence of upper semicontinuous functions, decreasing to a continuous limit
$u$
. For each
$\epsilon >0$
, the sets
$\{x\in X;{u}_{n}(x)u(x)<\epsilon \}$
form an open covering of
$X$
. Since
$X$
is compact, it admits a finite subcover, and since the sets are increasing as
$n$
increases, we must have
$X={\cap}_{n\ge {N}_{\epsilon}}\{x\in X;{u}_{n}(x)u(x)<\epsilon \}$
for some
${N}_{\epsilon}$
. Q.E.D.
5 Remarks
We conclude with a few remarks.
$\bullet $
Donaldson's conjecture says any two Kähler metrics can be connected by a smooth geodesic. One way to prove this conjecture is to establish a priori bounds on the derivatives of
$\phi (t;k)$
. This was done in Lemma 1 for
$\phi $
and
$\dot{\phi}$
. Let us now consider
$\ddot{\phi}$
:
$$\begin{array}{c}\ddot{\phi}\left(0\right)=\frac{1}{k}\sum ({\lambda}_{\alpha}{\lambda}_{\beta}{)}^{2}{s}_{i}{}_{{h}_{0}^{k}\left(k\right)}^{2}{s}_{j}{}_{{h}_{0}^{k}\left(k\right)}^{2}\end{array}$$ 
(5.1)

Note that Lemma 1 implies
$\left\ddot{\phi}\right(0\left)\right\le Ck$
, but this is not strong enough, since we need a bound which is independent of
$k$
.
Define a random variable
$Z$
whose probability distribution is given by
$$P(Z={\lambda}_{\alpha})=\left{s}_{\alpha}\right(z){}_{{h}_{0}^{k}\left(k\right)}^{2}$$
This is indeed a distribution since the total probability is
${\sum}_{\alpha}\left{s}_{\alpha}\right(z){}_{{h}_{0}^{k}\left(k\right)}^{2}=1$
. Moreover,
$\sum ({\lambda}_{\alpha}{\lambda}_{\beta}{)}^{2}{s}_{i}{}_{{h}_{0}^{k}\left(k\right)}^{2}{s}_{j}{}_{{h}_{0}^{k}\left(k\right)}^{2}$
is just the variance of
$Z$
. Thus we need to prove that the variance of
$Z$
is bounded by
$k$
. In the simplest case where
$X={\mathbf{P}}^{1}$
and the line bundle
$L=O\left(1\right)$
and the metric
$h$
is the FubiniStudy metric, then an easy computation shows that
$Z$
is just the binomial distribution with
$k$
trials, where the probability
$p$
of flipping heads is a function of
$z\in {\mathbf{P}}^{1}$
. As is well known, the variance of the binomial distribution is
$kp$
, which is the bound we need.
In the case where
$\omega $
is the FubiniStudy metric on
${\mathbf{P}}^{1}$
, the eigenvalues of the change of basis matrix are just
$0,1,...,k$
. More generally, we can show that if
$\phi $
is a radially symmetric Kähler potential on
${\mathbf{P}}^{1}$
, and if
${\lambda}_{0}\le {\lambda}_{1}\le \cdots \le {\lambda}_{k}$
are the eigenvalues of the change of basis matrix, then
${\lambda}_{j}{\lambda}_{j+1}\le C$
for some constant
$C$
, independent of
$k$
. From this one can show without difficulty that
$\ddot{\phi}\left(0\right)$
is uniformly bounded.
$\bullet $
The function
$li{m}_{l\to \infty}[su{p}_{k\ge l}\phi (t;k\left){]}^{*}\right(x)$
is equal almost everywhere to the
$limsup\phi (t;k)$
.
The convergence can also be guaranteed to take place in a Sobolev norm of positive order. Indeed, quite generally, the
${L}^{2}$
norm of
$\partial \phi $
can be bounded by
$\left\right\phi {}_{{C}^{0}}$
if
$\phi $
is
$PSH(X,{\omega}_{0})$
plurisubharmonic. Indeed,
$$\begin{array}{c}\left\right\partial \phi {}_{{\omega}_{0}}^{2}={\int}_{X}\partial \phi \wedge \overline{\partial}\phi \wedge {\omega}_{0}^{n1}={\int}_{X}\phi \partial \overline{\partial}\phi \wedge {\omega}_{0}^{n1}={\int}_{X}\phi {\omega}_{\phi}\wedge {\omega}_{0}^{n1}{\int}_{X}\phi {\omega}_{0}^{n}\end{array}$$ 
(5.2)

and the right hand side can be bounded in turn by
$$\begin{array}{c}{\int}_{X}\phi {\omega}_{\phi}\wedge {\omega}_{0}^{n1}{\int}_{X}\phi {\omega}_{0}^{n}\le \left\right\phi \left{}_{{C}^{0}}\right({\int}_{X}{\omega}_{\phi}\wedge {\omega}_{0}^{n1}+{\int}_{X}{\omega}_{0}^{n})=2\left\phi \right{}_{{C}^{0}}{\int}_{X}{\omega}_{0}^{n}.\end{array}$$ 
(5.3)

(In fact, the same argument gives the following useful inequality
$$\begin{array}{c}{J}_{{\omega}_{0}}\left(\phi \right)\le 2n\left\right\phi {}_{{C}^{0}},\end{array}$$ 
(5.4)

where
${J}_{{\omega}_{0}}\left(\phi \right)={V}^{1}\sqrt{1}{\sum}_{i=0}^{n1}\frac{ni}{n+1}{\int}_{X}\partial \phi \wedge \overline{\partial}\phi \wedge {\omega}_{\phi}^{i}\wedge {\omega}_{0}^{ni1}$
,
$V={\int}_{X}{\omega}_{0}^{n}$
, is the familiar AubinYau functional.) Returning to the problem at hand, we deduce that the
${H}_{\left(1\right)}(X\times [0,1\left]\right)$
Sobolev norms of the functions
$\phi (t;k)$
are uniformly bounded. The same is true for the
${H}_{\left(1\right)}\left(X\right)$
Sobolev norms of
$\phi (t;k)$
for each
$t\in [0,1]$
.
$\bullet $
The functions
$\dot{\phi}(t;k)$
also satisfy an interesting Harnack inequality of LiYau type. Let
$0\le \tau <T\le 1$
and let
$\xi ,X\in M$
. Then we claim
$$\begin{array}{c}\dot{\phi}(\xi ,\tau )\le \dot{\phi}(X,T)+\frac{1}{8}\Delta (\xi ,\tau ,X,T)\end{array}$$ 
(5.5)

where
$$\begin{array}{c}\Delta (\xi ,\tau ,X,T)={inf}_{\gamma}{\int}_{\tau}^{T}{\left(\frac{ds}{dt}\right)}^{2}dt\end{array}$$ 
(5.6)

where
$\frac{ds}{dt}$
is the velocity in space at time
$t$
and the infimum is taken over all paths from
$(\xi ,\tau )$
to
$(X,T)$
parametrized by
$\tau \le t\le T$
. To see this, let
$L=2\dot{\phi}$
. Then
$\ddot{\phi}\partial \dot{\phi}{}_{{\omega}_{\phi}}^{2}={\pi}_{N}V{}^{2}\ge 0$
, where
${\pi}_{N}V$
is the normal component of the holomorphic vector field on
$\mathbf{C}{\mathbf{P}}^{{N}_{k}}$
generated by
${\sigma}^{t}$
[
21]
, and so
$$\begin{array}{c}\frac{\partial L}{\partial t}\ge DL{}_{\phi}^{2}.\end{array}$$ 
(5.7)

As in [
17]
, we can now choose a path
$(t,s(t\left)\right)$
joining
$(\xi ,\tau )$
to
$(X,T)$
where
$\tau <T$
and
$\xi ,X\in M$
. Then
$$\begin{array}{ccc}L(X,T)L(\xi ,\tau )={\int}_{\tau}^{T}\frac{dL}{dt}dt& =& {\int}_{\tau}^{T}\left\{\frac{\partial L}{\partial t}+\frac{\partial L}{\partial s}\cdot \frac{ds}{dt}\right\}dt\end{array}$$  
$$\begin{array}{ccc}& \ge & {\int}_{\tau}^{T}\left\{DL{}^{2}+\frac{\partial L}{\partial s}\cdot \frac{ds}{dt}\right\}dt.\end{array}$$ 
(5.8)

Completing the square, we obtain
$$\begin{array}{c}L(X,T)L(\xi ,\tau )\ge \frac{1}{4}\int {\left(\frac{ds}{dt}\right)}^{2}dt.\end{array}$$ 
(5.9)

References

Bedford, E. and B.A. Taylor, “The Dirichlet problem for a complex MongeAmpre equation”, Invent. Math. 37 (1976), 144.

Bedford, E. and B.A. Taylor, “A new capacity for plurisubharmonic functions”, Acta Math. 149 (1982), 140.

Blocki, Z., “The complex MongeAmpère operator and pluripotential theory”, lecture notes available from the author's website.

Catlin, D., “The Bergman kernel and a theorem of Tian”, Analysis and geometry in several complex variables (Katata, 1997), 123, Trends Math., Birkhäuser Boston, Boston, MA, 1999.

Chen, X.X., “The space of Kähler metrics”, J. Differential Geom. 56 (2000), 189234.

Chen, X.X. and G. Tian, “Geometry of Kähler metrics and foliations by discs”, arXiv: math.DG / 0409433.

Chern, S.S., H. Levine, and L. Nirenberg, “Intrinsic norms on a complex manifold”, Global Analysis, Papers in honor of K. Kodaira, University of Tokyo Press (1969) 119139.

Demailly, J.P., “Complex analytic and differential geometry”, book available from the author's website.

Donaldson, S.K., “Remarks on gauge theory, complex geometry and
$4$
manifold topology”, Fields Medallists' lectures, 384403, World Sci. Ser. 20th Century Math., 5, World Sci. Publishing, River Edge, NJ, 1997.

Donaldson, S.K., “Symmetric spaces, Kähler geometry, and Hamiltonian dynamics”, Amer. Math. Soc. Transl. 196 (1999) 1333.

Donaldson, S.K., “Scalar curvature and projective imbeddings I”, J. Differential Geom. 59 (2001) 479522.

Donaldson, S.K., “Scalar curvature and stability of toric varieties”, J. Differential Geom. 62 (2002), 289349.

Donaldson, S.K., “Scalar curvature and projective imbeddings II”, arXiv: math.DG / 0407534.

Guedj, V. and A. Zeriahi, “Intrinsic capacities on compact Kähler manifolds”, arXiv: math.CV / 0401302.

Klimek, M., “Pluripotential theory”, London Mathematical Society monographs, New Series 6 (1991) Oxford University Press, New York.

Lelong, P., “Fonctions plurisousharmoniques et formes différentielles positives”, Gordon & Breach, ParisLondonNew York (1968).

Li, P. and S.T. Yau, “On the parabolic kernel of the Schrödinger operator”, Acta Math. 156 (1986), 153201.

Mabuchi, T., “Some symplectic geometry on compact Kähler manifolds”, Osaka J. Math. 24 (1987) 227252.

Paul, S., “Geometric analysis of Chow Mumford stability”, Adv. Math. 182 (2004), 333356.

Phong, D.H. and J. Sturm, “Stability, energy functionals, and KählerEinstein metrics”, Comm. Anal. Geometry 11 (2003) 563597, arXiv: math.DG / 0203254.

Phong, D.H. and J. Sturm, “Scalar curvature, moment maps, and the Deligne pairing”, Amer. J. Math. 126 (2004) 693712, arXiv: math.DG / 0209098.

Phong, D.H. and J. Sturm, “The Futaki invariant and the Mabuchi energy of a complete intersection”, Comm. Anal. Geometry 12 (2004) 321343, arXiv: math.DG / 0312529.

Semmes, S., “Complex MongeAmpère and symplectic manifolds”, Amer. J. Math. 114 (1992) 495550.

Tian, G., “KählerEinstein metrics with positive scalar curvature”, Invent. Math. 130 (1997) 137.

Yau, S.T., “On the Ricci curvature of a compact Kähler manifold and the complex MongeAmpère equation I”, Comm. Pure Appl. Math. 31 (1978) 339411.

Yau, S.T., “Open problems in geometry”, Proc. Symp. Pure Math. 54, AMS Providence, RI (1993) 128.

Zelditch, S., “The Szegö kernel and a theorem of Tian”, Int. Math. Res. Notices 6 (1998) 317331.

Zhang, S., “Heights and reductions of semistable varieties”, Compositio Math. 104 (1996) 77105.