Annals of Mathematics 200 (2024), 365–398

https://doi.org/10.4007/annals.2024.200.1.6

On L∞ estimates for fully non-linear

partial differential equations
By Bin Guo and Duong H. Phong

Abstract

Sharp L∞ estimates are obtained for general classes of fully non-linear

PDE’s on non-Kähler manifolds, complementing the theory developed ear-
lier by the authors in joint work with F. Tong for the Kähler case. The key
idea is still a comparison with an auxiliary Monge-Ampère equation, but
this time on a ball with Dirichlet boundary conditions, so that it always
admits a unique solution. The method applies not just to compact Hermit-
ian manifolds, but also to the Dirichlet problem, to open manifolds with a
positive lower bound on their injectivity radii, to (n − 1) form equations,
and even to non-integrable almost-complex or symplectic manifolds. It is
the first method applicable in any generality to large classes of non-linear
equations, and it usually improves on other methods when they happen to
be available for specific equations.

1. Introduction
A priori estimates are essential to partial differential equations, and of
these, L∞ estimates are of particular importance, as they are usually also
required for other estimates. In particular, the modern theory of partial differ-
ential equations can be argued to have started with the De Giorgi-Nash-Moser
theory of L∞ estimates for linear equations in divergence form. The impor-
tance of L∞ estimates for non-linear equations was recognized early on in the
case of complex Monge-Ampère equations on Kähler manifolds, where they
were first obtained by Yau [51] in his celebrated solution of the Calabi con-
jecture. A sharp version was subsequently obtained by Kolodziej [37] using
pluripotential theory. Very recently, an alternative proof of Kolodziej’s esti-
mates using only PDE methods was provided by the authors in joint work

Keywords: L∞ estimates, fully non-linear equations, auxiliary equations, test function for
comparison, almost-complex manifolds, De Giorgi-Nash-Moser theory
AMS Classification: Primary: 35J60, 35J96, 32W20, 53C55, 53C56.
Work supported in part by the National Science Foundation under grants DMS-1855947
and DMS-22-03273.
© 2024 Department of Mathematics, Princeton University.

365
366 BIN GUO and DUONG H. PHONG

with F. Tong [31]. This proof also gave new and sharp estimates for gen-
eral classes of non-linear equations on Kähler manifolds, thus providing a first
step towards a De Giorgi-Nash-Moser theory for fully non-linear equations in
complex geometry.
The goal of the present paper is to obtain sharp L∞ estimates for fully non-
linear equations in non-Kähler settings. There is a strong incentive for this,
as non-Kähler settings appear increasingly frequently in complex geometry,
symplectic geometry, and theoretical physics. (See, e.g., [19], [20], [22], [40],
[41] for surveys.) Extensions from Kähler to non-Kähler settings are typically
technically much more complicated, as new torsion terms can proliferate and
integration by parts is unwieldy. However, there are usually considerable new
conceptual difficulties as well, because of a lack of cohomological constraints.
In particular, because of this, a basic idea from [31], namely a comparison with
an auxiliary global complex Monge-Ampère equation, runs immediately into
difficulties: in the absence of suitable cohomological constraints, the auxiliary
equation may not be solvable.
In this paper, we shall show that comparisons with auxiliary Monge-
Ampère equations are still of powerful assistance, however the auxiliary equa-
tions should not be defined globally, but only on an open ball with Dirichlet
conditions. The existence and regularity of the auxiliary problem that we then
need is a classic result of Caffarelli, Kohn, Nirenberg, and Spruck [3], which
plays the same role for the Dirichlet problem as Yau’s existence theorem [51]
did in [31] for the compact Kähler case. An important point is that the de-
sired global estimates can still be obtained from the local estimates obtained
in this manner near a minimum. This requires an adaptation of the original
arguments of Blocki [1], combined with weak Harnack inequalities for superso-
lutions of linear elliptic equations [24], as well as an estimate of Kolodziej [37]
that generalizes the one-dimensional estimate of Brezis-Merle [2].
The resulting method for L∞ estimates, using only comparisons with a
local equation, turns out to be both flexible and powerful. It bypasses the
complicated torsion terms and integration by parts plaguing non-Kähler ge-
ometry mentioned above, and it applies to whole new classes of fully non-linear
equations, not just on compact Hermitian manifolds, but also to the Dirichlet
problem, to open manifolds, to (n − 1)-form equations, and to equations on
almost-complex manifolds. The extension to almost-complex manifolds, using
auxiliary real Monge-Ampère equations, is in particular of considerable inter-
est, as non-integrable almost-complex manifolds have revealed themselves to
be effective intermediaries between complex geometry and symplectic geome-
try (see, e.g., [35], [12], [47], [18]). On the other hand, if we assume the Kähler
condition, then comparisons with a global equation as in [31] are possible, and
we obtain L∞ estimates that have the great advantage of remaining uniform
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 367

over classes of Kähler metrics possibly degenerating to the boundary of the

Kähler cone. Such a uniformity is essential for many applications to algebraic
geometry (see, e.g., [17], [13], [29]). Thus the present paper and [31] address
different geometric requirements in complex geometry, and together, they are
a natural candidate for what may be viewed as a De Giorgi-Nash-Moser theory
for non-linear equations in this area.
Now we describe our results more fully. Let (X, ω) be a connected complex
manifold of complex dimension n, and let ω = jk igk̄j dz j ∧dz̄ k be a Hermitian
P
metric. Let f : Γ ⊂ Rn → R+ be a function satisfying the following conditions:
(1) Γ ⊂ Rn is a symmetric cone with

(1.1) Γn ⊂ Γ ⊂ Γ1 ,

where Γk is the cone of vectors λ with σj (λ) > 0 for 1 ≤ j ≤ k. Here σk (λ)
is the j-th symmetric polynomial in λ. In particular, Γ1 is the half-space
defined by λ1 + · · · + λn > 0, and Γn is the first octant, defined by λj > 0
for 1 ≤ j ≤ n.
(2) f (λ) is symmetric in λ = (λ1 , . . . , λn ) ∈ Γ and is homogeneous of degree one.
∂f
(3) ∂λ j
> 0 for each j = 1, . . . , n and λ ∈ Γ.
(4) There is a constant γ > 0 such that
n
Y ∂f
(1.2) ≥γ ∀λ ∈ Γ.
∂λj
j=1

Familiar examples of such non-linear operators f (λ) include the complex

1
Monge-Ampère operator f (λ[hϕ ]) = ( nj=1 λj ) n , Γ = Γn ; the complex Hessian
Q
1
operator f (λ[hϕ ]) = (σk (λ)) k , Γ = Γk , where σk (λ) is the k-th symmetric
polynomial in the eigenvalues λj [4], [16], [48]; and the p-Monge-Ampère op-
Q  n!
(n−p)!p!
erator of Harvey and Lawson [32], [33], f (λ) = I λI , where I runs
over all distinct multi-indices 1 ≤ i1 < · · · < ip ≤ n, λI = λi1 + · · · + λip , and
Γ is the cone defined by λI > 0 for all p-indices I. Clearly, conditions (1)–(4)
above are preserved on the intersection of the individual cones by finite lin-
ear combinations with positive coefficients of operators f (λ) satisfying them.
Remarkably, it has recently been shown by Harvey and Lawson [34] that the
set of operators f (λ) satisfying conditions (1)–(4) is quite large, and actually
includes all invariant Gårding-Dirichlet operators.
For any C 2 function ϕ : X → R, we set ωϕ = ω + i∂ ∂ϕ ¯ and let hϕ :
1,0 1,0
T X → T X be the endomorphism defined by ωϕ relative to ω, i.e., in local
coordinates, (hϕ )j i = g j k̄ (gϕ )k̄i , where gj̄i denotes the components of ω and
(g ij̄ ) its inverse. We denote by λ[hϕ ] the (unordered) vector of eigenvalues
of hϕ . Given a smooth function F on X, we consider the fully non-linear
368 BIN GUO and DUONG H. PHONG

partial differential equation

(1.3) f (λ[hϕ ]) = eF and λ[hϕ ] ∈ Γ.
Such equations were introduced systematically in the foundational paper by
Caffarelli, Nirenberg, and Spruck [4], and they now occur regularly in geometric
analysis; e.g., [25], [48].
Theorem 1. Assume that (X, ω) is a compact Hermitian manifold with-
out boundary, and consider equation (1.3) with the operator f (λ) satisfying
conditions (1)–(4). Then for any p > n and any C 2 solution ϕ to (1.3),
normalized to satisfy supX ϕ = 0, we have
(1.4) supX |ϕ| ≤ C
for some constant C > 0 depending only on X, ω, n, p, γ, and kenF kL1 ( log L)p .
Here the norm kenF kL1 ( log L)p is defined by
Z
nF
ke kL1 ( log L)p = (1 + |F |p )enF ω n
X
and is sometimes called the p-th Nash entropy of the function enF .
We briefly survey the previous known results related to this theorem. The
case (X, ω) Kähler was established in [31], including an even stronger version
allowing degenerations of the underlying Kähler class ω generalizing the case
of the Monge-Ampère equation obtained in [17], [13]. So we concentrate on
the case of (X, ω) non-Kähler, which is the focus of the present paper. In
this case, for the Monge-Ampère equation, L∞ estimates in the non-Kähler
case were first obtained by Tosatti-Weinkove [45], building on earlier works of
Cherrier [9], and a pointwise upper bound for enF was needed. The stronger
version requiring only weaker entropy bounds as described in Theorem 1 was
first obtained by Dinew and Kolodziej [14], [15], [16], [37], using a non-trivial
extension of pluripotential theory to the Hermitian setting. More recently
a new approach using the theory of envelopes has been developed by Guedj
and Lu [27]. The proof of Theorem 1 is arguably simpler than in all these
approaches. More importantly, it is a PDE-based proof, and can apply to
many more equations. The estimates in Theorem 1 appear to be the first L∞
estimates obtained in any generality for fully non-linear equations on compact
Hermitian manifolds.
It may be worth noting an intriguing consequence of Theorem 1, or rather
of its proof, as we shall see later in Section 3. Given a C 2 function ϕ such that
ωϕ = ω + i∂ ∂ϕ¯ > 0 defines a Hermitian metric on X, in general the following
identity does not hold unless extra conditions [14] are put on ω:
Z Z
¯ n
(ω + i∂ ∂ϕ) = ωn.
X X
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 369

So it is interestingR to estimate the lower/upper bound of the volume of (X, ωϕ ),

i.e., the integral X ωϕn . To this end, we define the relative volume enF of ωϕ
with respect to ω by enF = ωϕn /ω n . Then ϕ can be viewed as the solution of a
complex Monge-Ampère equation:
(1.5) ¯ n = enF ω n .
(ω + i∂ ∂ϕ)
Normalizing ϕ so that supX ϕ = 0, we can apply Theorem 1 and its proof and
obtain
Corollary 1. Given p > n and K > 0, suppose enF satisfies X enF |F |p ω n
R

≤ K. Then there is a constant c > 0 that depends on n, ω, p and K such that

Z Z
(1.6) ¯ n
(ω + i∂ ∂ϕ) = enF ω n ≥ c.
X X

Thus an upper bound on the p-th entropy of relative volume enF implies
a positive lower bound of the L1 -norm of enF . This is trivial if ω is Kähler,
but not in general. We remark that if X admits a closed real (1, 1)-form β
1,1
whose Bott-Chern class [β] ∈ HBC (X, C) is nef and big (i.e., X is of Fujiki
class C), then by a weak transcendental Morse inequality of Demailly [12],
equation (1.6) holds even without the assumption on X enF |F |p ω n . This is
R

based on an observation of Tosatti [44]. For the convenience of readers, we

provide a proof (see Lemma 9 in Section 3) of this fact using the arguments in
[12].
The strategy of the proof of Theorem 1 is to employ a local auxiliary
complex Monge-Ampère equations as in [31] and a localized argument similar
to that in [1]. As stressed above, this method is effective in a vast range of
different geometric situations. We discuss some of these next.
The Dirichlet problem on Hermitian manifolds. We first consider the case
of the Dirichlet problem. Thus let X be an open subset with smooth boundary
in a larger Hermitian manifold (X̃, ω). Let f (λ) be an operator satisfying
conditions (1)–(4), ρ ∈ C 0 (∂X), and consider the following Dirichlet boundary
value problem on X:
(1.7) f (λ[hϕ ]) = eF in X, λ[hϕ ] ∈ Γ, ϕ = ρ on ∂X.
Theorem 2. Given p > n, suppose ϕ ∈ C 2 (X) ∩ C 0 (X) is a solution to
the above Dirichlet problem (1.7). Then there is a constant C > 0 that depends
only on X, ω, n, p, γ and, kenF kL1 ( log L)p , such that
(1.8) supX |ϕ| ≤ kρkL∞ (∂X) + C(1 + kϕk2L1 ).
We observe that, for complex Monge-Ampère and Hessian equations, and
if ω is Kähler, then kϕkL1 can be bounded in terms of the other parameters
(see Section 4). More importantly, for general equations f (λ), there are very
370 BIN GUO and DUONG H. PHONG

few L∞ estimates, if any, that require only weak integral bounds on the right-
hand side eF such as in Theorem 2. For example, the L∞ estimates obtained
by S. Y. Li in [38] are based on the construction of subsolutions, and as is
usually the case with subsolutions, they require L∞ bounds on the right-hand
side.
Open Hermitian manifolds. Next, we consider the case of a non-compact
complex manifold (X, ω), where ω is a complete Hermitian metric on X. We
shall make the following assumption:
(A) There exists a constant r0 > 0, so that, at each point x ∈ X, there exists
a holomorphic coordinates chart (Ux , z) centered at x satisfying
1 ¯ 2 ≤ 2ω in Ux .
(1.9) {|z| < 2r0 } ⊂ Ux , ω ≤ i∂ ∂|z|
2
With assumption (1.9), we introduce two quantities associated to ϕ and eF ,
which we assume are bounded. Given a fixed p > n, we set
Z Z
(1.10) Kϕ = sup |ϕ|ω n , KF = sup enF |F |p ω n .
x∈X Ux x∈X Ux

Theorem 3. Let (X, ω) be a non-compact Hermitian manifold satisfying

assumption (A). Let ϕ ∈ L∞ (X) ∩ C 2 (X) be a solution to (1.7). Then there
is a constant C > 0 that depends on n, p, ω, γ, and Kϕ , KF such that
supX |ϕ| ≤ C + liminf z→∞ (−ϕ)+ ,
where C is a constant depending on X, ω, n, p, γ, r0 , and Kϕ , KF .
We remark that assumption (1.9) is satisfied if the metric ω is Kähler, the
Riemannian curvature is bounded, and the injectivity radius of (X, ω) at each
point is bounded from below by some positive number (see [43]). In particular,
it holds for asymptotically locally Euclidean (ALE) Kähler manifolds, or under
some asymptotic conditions on the (Kähler) metric ω near infinity, such as the
asymptotically conical (AC) and asymptotically cylindrical (ACyl) conditions,
which have been extensively studied in recent years. Typically the conditions
that have been imposed in the literature on the open complete Kähler case are
stronger than in the above theorem; see, e.g., [43]. But more importantly, we
are not aware of previous results for general fully non-linear equations on open
complete Hermitian manifolds.
General (n−1)-form equations with gradient terms. We show now how the
local comparison method developed in this paper can apply to other second-
order equations and not just functions of the eigenvalues of the standard Hes-
sian hϕ . Well-known examples of such equations arose in [21], [41] and in
the solution of the Gauduchon conjecture [42], [48], [41], and they have been
attracting considerable attention ever since [25], [23].
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 371

More specifically, let (X, ω) be a compact Hermitian manifold with ∂X = ∅

and ωh be another Hermitian metric on X. Fix a smooth (1, 0)-form θ on X,
set
(1.11) χ[ϕ] = i(ϕj θ̄k + θj ϕk̄ )dz j ∧ dz̄ k ,

and consider the endomorphism h̃ϕ of the tangent bundle T 1,0 (X) given by
1
h̃ϕ = ω −1 · ωh + ¯ + χ[ϕ]) ,

(1.12) ((∆ω ϕ)ω − i∂ ∂ϕ)
n−1
where
¯ ∧ ω n−1
i∂ ∂ϕ
∆ω ϕ = n
ωn
is the complex Laplacian of ϕ with respect to ω. We consider the fully non-
linear equation
(1.13) f (λ[h̃ϕ ]) = eF , λ[h̃ϕ ] ∈ Γ,
with ϕ normalized by supX ϕ = 0. We then have
Theorem 4. Given p > 2n, there exists a constant C > 0 that depends
on n, γ, ω, p, θ, ωh , and ke2nF kL1 ( log L)p (X,ω) such that
supX |ϕ| ≤ C.
1
Previously, this equation received most attention when f (λ) =!( nj=1 λj ) n ,
Q
χ[ϕ] = 0, and was known in this case as the (n − 1)-form Monge-Ampère equa-
tion. It was shown by Tosatti-Weinkove [46] to be solvable up to a modification
of F by a suitable additive constant. This required a priori estimates of all
orders, of which, as stressed in [46], the L∞ estimate is a major one.
In an earlier version of the present paper (arXiv:2204.12549), we had given
a new and simpler proof of the L∞ estimate of Tosatti and Weinkove using our
present method. Our proof was subsequently extended by Klemyatin, Liang,
and Wang [36] to more general operators f (λ[h̃ϕ ]) satisfying the structural
conditions (1)–(4).
The above Theorem 4 shows that L∞ bounds can still be established even
with the incorporation of gradient terms χ[ϕ] in the equation f (λ[h̃ϕ ]). Such
terms are actually required for the solution of the Gauduchon conjecture [42],
and in general, they can lead to considerable additional difficulties in a priori
estimates (see, e.g., [42], [26]). Here, we show that these terms can readily
be handled by our method, just by using as auxiliary equation the Dirichlet
problem for a real Monge-Ampère equation instead of the usual complex one.
The only cost of using a real auxiliary equation is the constraint p > 2n in the
p-Nash entropy, instead of the milder condition p > n when χ = 0 [36]. It is
not known whether p > n would suffice when χ 6= 0.
372 BIN GUO and DUONG H. PHONG

Fully non-linear equations on almost-complex manifolds. While the im-

portance of almost-complex structures has been known for a long-time, the
structures that are not integrable had been comparatively neglected. This sit-
uation changed drastically in the last twenty years or so, as basic geometric
structures, notably forms in lower dimensions or symplectic structures, have
led to non-integrable almost-complex structures. We refer to [35], [12], [47]
for more detailed discussions. Here we would like to point out that the meth-
ods of this paper readily adapt to produce L∞ estimates for general classes of
non-linear equations, even on almost-complex manifolds.
Let X be a compact manifold without boundary of real dimension 2n,
equipped with an almost-complex structure J, that is, an endomorphism of the
tangent bundle satisfying J 2 = −id. The complexified tangent bundle TC X
can be decomposed as a direct sum (1,0) X and T (0,1) X, which are the
√ of T √
eigenspaces of J with eigenvalues −1 and − −1, respectively. An almost-
Hermitian metric g on X is a Riemannian metric that is compatible with J,
i.e., g(Y, Z) = g(JY, JZ) for any vector fields Y, Z. The associated Hermitian
form ω is a real positive definite (1, 1)-form defined by ω(Y, Z) = g(JY, Z).
For a smooth function ϕ ∈ C ∞ (X), we define i∂ ∂ϕ ¯ to be

(1.14) ¯ = 1 (d(Jdϕ))(1,1) ,
i∂ ∂ϕ
2
the (1, 1)-part of the 2-form 21 d(Jdϕ). As in other sections, we write ωϕ =
¯ and let hϕ = ω −1 · ωϕ be the endomorphism of the tangent bundle
ω + i∂ ∂ϕ,
determined by the (1, 1)-form ωϕ and the metric ω.
With this notation, we can consider the same equation on an almost-
complex manifold X as in (1.3) for a given non-linear operator f : Γ → R+
satisfying conditions (1)–(4) in the introduction, and a given smooth right-hand
side eF . We then have the following general L∞ estimate for fully non-linear
equations on almost-complex manifolds:
Theorem 5. Under the above assumptions, and with ϕ normalized to
satisfy supX ϕ = 0, there exists a constant C depending on n, ω, γ, p > 2n, and
ke2nF kL1 ( log L)p (X,ωn ) such that
supX |ϕ| ≤ C.
Only in the particular case when f is the Monge-Ampère operator was an
L∞ bound for ϕ obtained before [10], and an L∞ bound for eF was required.
Potentials on almost-Kähler manifolds. Once we leave the realm of Kähler
¯
geometry, a severe impediment is the lack of a ∂ ∂-lemma. However, if the
failure of the Kähler condition is due to the non-integrability of the almost-
complex structure, we are in the realm of almost-Kähler geometry, and we can
still characterize almost-Kähler metrics with symplectic form by a potential.
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 373

More specifically, let (X, J) be an almost-complex manifold. An almost-

Kähler metric is a metric g(U, V ) that is J-compatible, i.e., g(JU, JV ) =
g(U, V ), and whose associated 2-form ωg (U, V ) = g(JU, V ) is closed. Given a
tame symplectic form Ω, that is, a closed 2-form Ω satisfying Ω(U, JU ) > 0,
we can consider the class of almost-Kähler metrics g with [ωg ] = [Ω]. It has
been pointed out by Donaldson [12] that there is a natural version of the
Calabi-Yau problem in this context, namely find an almost-Kähler metric g̃
with [ωg̃ ] = [Ω] and given volume form. He showed that a positive answer to
this question would have important consequences in symplectic geometry.
In [47], it was shown that the answer is indeed positive if one can establish
L∞ estimates for the potential ϕ defined for any metric g̃ with [ωg̃ ] = [Ω] by
ωg̃n−1 ∧ Ω
(1.15) ∆g̃ ϕ = 2n − 2n , supX ϕ = 0.
ωgn
Such estimates had been reduced first in [50] to an exponential integral estimate
for ϕ, and then to an Lp estimate in [47]. The methods of this paper can readily
reduce them to an L1 estimate:
Theorem 6. Let g̃ be an almost-Kähler metric on an almost-complex
manifold X of dimension 2n, with [ωg̃ ] = [Ω], where Ω is a tame symplectic
form. Then there is a constant C > 0 depending only on the total volume of g̃
such that
(1.16) supX |ϕ| ≤ C(1 + kϕkL1 (X,enF dVg̃ ) ),
where enF = ω̃ n /Ωn .
This estimate first appeared in our recent survey [28] in honor of S.S.
Chern. We were informed by V. Tosatti that the methods of [47] can also
be extended to derive it. Our main interest in including a brief discussion in
this paper is to illustrate how a feature of almost-Kähler metrics, namely the
lack of first derivative terms in the metric in the Laplace-Beltrami operator, is
particularly conducive to our methods.
This paper is a revised and expanded version of a paper that was first
posted as arXiv: 2204:12549. We have added new material on gradient terms in
(n−1)-form equations, on general fully non-linear equations on almost-complex
manifolds, and on the Calabi-Yau equation on tame symplectic manifolds, all
to show how to adapt our method of local auxiliary Monge-Ampère equations
to very different geometric situations.

2. The key local estimate

Fix the Hermitian manifold (X, ω). We assume that there exists r0 > 0 so
that, at any point x0 ∈ X, there exists a local holomorphic coordinate system
374 BIN GUO and DUONG H. PHONG

Pn j
z centered at x0 , with ω(x0 ) = i j=1 dz ∧ dz̄ j and
1 ¯ 2 ¯ 2 in B(x0 , 2r0 ),
(2.1) i∂ ∂|z| ≤ ω ≤ 2i∂ ∂|z|
2
where B(x0 , 2r0 ) is the Euclidian ball of radius 2r0 . When X is compact,
with or without boundary, then such an r0 clearly exists and can be viewed
as determined by X and ω. Any dependence on r0 can be absorbed into a
dependence on X and ω. When X is not compact, the existence of such an
r0 > 0 is an assumption that we always make. Without loss of generality we
may assume r0 ≤ 1/2 throughout the paper.
Lemma 1. Let ϕ be a C 2 solution on a Hermitian manifold (X, ω) of the
equation
(2.2) f (λ[hϕ ]) = eF ,
where the operator f (λ) satisfies conditions (1)–(4) spelled out in Section 1.
Let x0 be any point in X, and assume that
(2.3) ϕ(z) ≥ ϕ(x0 ) for z ∈ B(x0 , 2r0 ).
Fix any p > n. Then we have
(2.4) −ϕ(x0 ) ≤ C,
where C is a constant depending only on (X, ω), p, γ, kenF kL1 ( log L)p , and
kϕkL1 . Here the L1 - and L1 ( log L)p -norms are on the ball B(x0 , 2r0 ) with
respect to the volume form ω n .
We will assume −ϕ(x0 ) ≥ 2, otherwise Lemma 1 already holds.
We break the proof into several steps. The first step is the most important
and requires a comparison of the solution ϕ of (2.2) with the solution of an
auxiliary Monge-Ampère equation. More precisely, set
1
(2.5) us (z) = ϕ(z) − ϕ(x0 ) + |z|2 − s
2
for each s with 0 < s < 2r02 . It is convenient to set Ω = B(x0 , 2r0 ). Then
us > 0 on ∂Ω, and thus the set Ωs defined by
(2.6) Ωs = {z ∈ Ω; us (z) < 0}
is an open and non-empty set with compact closure in Ω. Set
Z
(2.7) As = (−us )enF ω n .
Ωs
Formally, the auxiliary equation that we would like to consider is the following
Dirichlet problem for the complex Monge-Ampère equation on the ball Ω,
¯ s )n = (−us )χR+ (−us ) nF n
(2.8) (i∂ ∂ψ e ω in Ω, ψs = 0 on ∂Ω,
As
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 375

where χR+ (x) is the characteristic function of the positive real axis, and ψs
is required to be plurisubharmonic, i∂ ∂ψ ¯ s ≥ 0. Since the right-hand side
would have singularities, we choose a sequence of smooth positive functions
τk (x) : R → R+ that converges to the function x · χR+ (x) as k → ∞. Then let
the function ψs,k be defined as the solution of the following Dirichlet problem,

¯ s,k )n = τk (−us ) nF n
(2.9) (i∂ ∂ψ e ω in Ω, ψs,k = 0 on ∂Ω
As,k
¯ s,k ≥ 0, and As,k is defined by
with i∂ ∂ψ
Z
(2.10) As,k = τk (−us )enF ω n .
Ω

By the classic theorem of Caffarelli-Kohn-Nirenberg-Spruck [3], this Dirichlet

problem admits a unique solution ψs,k that is of class C ∞ (Ω̄). The maximum
of ψs,k can only be attained in ∂Ω, and thus ψs,k ≤ 0. By definition of the
constants As,k , we also have As,k → As as k → ∞, and
Z
(2.11) ¯ s,k )n = 1.
(i∂ ∂ψ
Ω

Lemma 2. Let us be a C2
solution of the fully non-linear equation (2.2)
and ψs,k be the solutions of the complex Monge-Ampère equation (2.9) as de-
fined above. Then we have
n
(2.12) −us ≤ ε(−ψs,k ) n+1 on Ω̄,
n
where ε is the constant defined by εn+1 = As,k γ −1 (n+1)
n2n
.
Proof. We have to show that the function
n
(2.13) Φ = −ε(−ψs,k ) n+1 − us
is always ≤ 0 on Ω̄. Let xmax ∈ Ω be a maximum point of Φ. If xmax ∈ Ω\Ωs ,
clearly Φ(xmax ) ≤ 0 by the definition of Ωs and the fact that ψs,k < 0 in Ω. If
xmax ∈ Ωs , then we have i∂ ∂Φ(x ¯ max ) ≤ 0 by the maximum principle.
i j̄ ∂ log f (λ[h]) 1 ∂f (λ[h])
Let G = ∂hij = f ∂hij be the linearization of the operator
log f (λ[h]). It follows from the structure conditions of f that Gij̄ is positive
definite at hϕ and
1 ∂f (λ[h]) γ
detGij̄ = n
det( )≥ .
f ∂hij f (λ)n
The assumption that f (λ) is homogeneous of degree 1 implies that
X
trG ωϕ = Gij̄ (ωϕ )j̄i = 1.
i,j
376 BIN GUO and DUONG H. PHONG

We calculate at the point xmax :

X
¯
0 ≥ (trG i∂ ∂Φ)(x max ) = Gij̄ Φj̄i
i,j
nε 1 εn − n+2
= (−ψs,k )− n+1 (trG i∂ ∂ψ
¯ s,k ) + (−ψ s,k ) n+1 |∇ψ 2
s,k |G
n+1 (n + 1)2
1 ¯ 2
−trG ωϕ + trG (ω − i∂ ∂|z| )
2
nε 1
≥ (−ψs,k )− n+1 (trG i∂ ∂ψ
¯ s,k ) − 1
n+1
n2 ε 1
≥ (−ψs,k )− n+1 (detG)1/n [det(ω −1 · i∂ ∂ψ
¯ s,k )]1/n − 1
n+1
n2 ε 1/n 1 (−us )1/n
≥ γ (−ψs,k )− n+1 1/n
− 1.
n+1 A s,k

Here in the fourth line we use (2.1) and in the fifth line we apply the arithmetic
geometric inequality. Hence at xmax , we have
(n + 1)n n n
−us ≤ As,k γ −1 2n n
(−ψs,k ) n+1 = ε(−ψs,k ) n+1 ,
n ε
by the choice of ε. This implies Φ(xmax ) ≤ 0, and the claim is proved. 
Once we have a comparison between us and ψs,k as given in Lemma 2,
we no longer need to know the differential equations satisfied by us and ψs,k .
It is most convenient to then summarize the remaining part of the proof of
Theorem 5 in a lemma of general applicability:
Lemma 3. Let 0 < s < s0 = 2r02 . Assume that we have functions us ,
us > 0 on ∂Ω, and let Ωs , As , and As,k be the corresponding notions as defined
by (2.6), (2.7) and (2.10). Assume that the inequality (2.12) holds; that is,
1 n
n+1
(2.14) −us ≤ C(n, γ) As,k (−ψs,k ) n+1 on Ω̄
for some constant C(n,
R γ), wheren ψs,k are plurisubharmonic functions on Ω,
¯ s,k ) = 1. Then for any p > n, we have
ψs,k = 0 on ∂Ω, and Ω (i∂ ∂ψ
(2.15) −ϕ(x0 ) ≤ C(n, ω, γ, p, kϕkL1 (Ω,ωn ) ).
Clearly, Theorem 1 follows immediately from Lemmas 2 and 3, so we
devote the remaining part of this section to the proof of Lemma 3.
The main property of the functions ψs,k in Lemma 3 that we need is an
exponential integral estimate, which can be stated as follows. Let Ω ⊂ Cn be
a bounded pseudoconvex domain. Then there are constants α > 0 and C > 0
that depend only on n and diam(Ω) such that
Z
(2.16) e−αψ dV ≤ C
Ω
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 377

for any ψ ∈ PSH(Ω) ∩ C 2 (Ω) ∩ C 0 (Ω) with ψ|∂Ω ≡ 0, and Ω (i∂ ∂ψ) ¯ n = 1.
R

This exponential estimate is due to Kolodziej [37], and it is a generalization

of the inequality of Brezis-Merle [2] in complex dimension one. While it has
some similarity with the α-invariant in Kähler geometry, it is a very different
estimate as it requires information on the Monge-Ampère volume of ψ. It
was first proved in [37] using pluripotential theory. Recently an elementary
and elegant proof was given in [49], which, in particular, does not rely on
pluripotential theory.
Returning to the proof of Lemma 3, we rewrite the comparison between us
and ψs,k as
(−us )(n+1)/n
1/n
≤ C(n, γ)(−ψs,k ) in Ωs .
As,k
Take a small β = β(n, α, γ) > 0 such that βC(n, γ) ≤ α, where α > 0 is as in
(2.16). We then get
Z  (−u )(n+1)/n  Z  
s n
(2.17) exp β 1/n
ω ≤ exp βC(n, γ)(−ψs,k ) ω n ≤ C,
Ωs As,k Ω

by the uniform integral estimate (2.16). We can now take the limit k → ∞
and obtain the following exponential estimate for us :
Lemma 4. For any 0 < s < s0 , the functions us satisfy the following
inequality:
Z n+1
 (−us ) n
(2.18) exp β 1/n
ω n ≤ C,
Ωs As
where β = β(n, γ, r0 ) and C = C(n, γ, r0 ) are strictly positive constants de-
pending only on n, γ, and r0 .
The next step is to show that this exponential inequality for us implies
the following estimate for As :
p−n
Lemma 5. Fix p > n, and let δ0 = pn > 0. Then we have
ÅZ ã1+δ0
(2.19) As ≤ C 0 enF ω n ,
Ωs

where C0 is a constant depending only on n, p, γ, ω, and kenF kL1 ( log L)p .

Proof. As in [31], we apply Young’s inequality. For our purposes, it is
convenient to state Young’s inequality as follows. Let η : R+ → R+ be a
positive strictly increasing function with limx→0+ η(x) = 0. Let η −1 be the
inverse function of η, so that η(η −1 (y)) = y and η −1 (η(x)) = x. Then for any
378 BIN GUO and DUONG H. PHONG

U, V ≥ 0, we have
Z U Z V
(2.20) UV ≤ η(x)dx + η −1 (y)dy.
0 0
If we make a change of variables y = η(x) in the second integral, this inequality
can be rewritten as
Z U Z η−1 (V )
(2.21) UV ≤ η(x)dx + xη 0 (x)dx.
0 0
The first integral on the right-hand side can be estimated by U η(U ), while the
second integral can be estimated by
(2.22)
Z η−1 (V ) Z η−1 (V )
0 −1
xη (x)dx ≤ η (V ) η 0 (x)dx = η −1 (V )η(η −1 (V )) = V η −1 (V ).
0 0
Thus we arrive at the following inequality, which suffices for our purposes:
(2.23) U V ≤ U η(U ) + V η −1 (V ).
We return to the proof of Lemma 5. We apply the preceding Young’s
inequality with η(x) = ( log (1 + x))p , U = enF (z) , and a general positive
1
function V . Then η −1 (y) = exp(y p ) − 1, and we find
1
enF V ≤ enF ( log (1 + enF ))p + V exp(V p ) − 1 .

(2.24)
Specializing further to V = v(z)p for a general non-negative function v(z), we
obtain
(2.25) v(z)p enF ≤ enF ( log (1+enF ))p +v p (ev −1) ≤ enF (1+n|F |)p +Cp e2v(z) .
Now let
β (−us )(n+1)/n
(2.26) v(z) = 1/n
,
2 As
and integrate over Ωs to obtain
Å ãp Z
(−us )(n+1)p/n nF n
Z
β
p/n
e ω ≤ enF (1 + n|F |)p ω n
2 Ωs As Ωs
(2.27) ® ´
(−us )(n+1)/n
Z
+ exp β 1/n
ωn.
Ωs As
By Lemma 5, the right-hand side is bounded by a constant
C(n, p, β, kenF kL1 ( log L)p , r0 ).
Thus we have
Z p
p(n+1)
(2.28) (−us ) n enF ω n ≤ C(n, p, β, r0 )Asn .
Ωs
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 379

This implies the following upper bound for As : by Hölder’s inequality, we can
write
Z
As = (−us )enF ω n
Ωs
Z n/(p(n+1))  Z 1/q
p(n+1)/n nF n
≤ (−us ) e ω · enF ω n
Ωs Ωs
Z 1/q
≤ A1/(n+1)
s · enF ω n ,
Ωs
1 n
where q > 1 is chosen such that q + p(n+1) = 1. This is the inequality to be
proved in Lemma 5. 
We can now derive growth estimates for the function φ(s) defined by
Z
(2.29) φ(s) = enF ω n
Ωs
following the original strategy of De Giorgi [11]. We shall actually be interested
in estimates as s → 0+ , so the version relevant to our setup is the one provided
by [37].
Lemma 6. The function φ(s) is a monotone increasing function of s ∈
(0, s0 ) that satisfies lims→0+ φ(s) = 0, φ(s) > 0 for s ∈ (0, s0 ), and
(2.30) tφ(s − t) ≤ C0 (φ(s))1+δ0 ∀0 < t < s < s0 .
Proof. The monotonicity of φ is clear from the definition and the relation
Ωs0 ⊂ Ωs if 0 < s0 < s < s0 . Note that x0 ∈ Ωs for any s > 0. Being a
non-empty open set, Ωs has positive measure. So φ(s) > 0 for any s > 0.
Furthermore, ϕ(z) − ϕ(x0 ) + 21 |z|2 is strictly positive in any small punctured
neighborhood of x0 , so we have
∩s>0 Ωs = {x0 }.
This shows that φ(s) → 0 as s → 0+ .
For any t ∈ (0, s) and any z ∈ Ωs−t ⊂ Ωs , we have
0 > us−t (z) = us (z) + t, i.e., − us (z) > t > 0.
This observation shows that
Z Z Z
nF n nF n
(2.31) As = (−us )e ω ≥ (−us )e ω ≥t enF ω n .
Ωs Ωs−t Ωs−t

enF ω n
R
Note that φ(s) = Ωs for s ∈ (0, s0 ). Then equations (2.19) and (2.31)
imply that
(2.32) tφ(s − t) ≤ C0 (φ(s))1+δ0 ∀0 < t < s < s0 .
This finishes the proof of Lemma 6. 
380 BIN GUO and DUONG H. PHONG

Lemma 7. Let φ(s) be a function on (0, s0 ) satisfying all the properties

listed in the previous lemma. There there is a constant c0 > 0, depending only
on s0 , C0 , and δ0 so that
(2.33) φ(s0 ) ≥ c0 .
Proof. We apply the iteration argument in [37]. Given the initial num-
ber s0 , we inductively define a decreasing sequence of positive numbers {sj }j≥0
as follows: given sj , let
ß ™
1
sj+1 = sup 0 < s < sj | φ(s) ≤ φ(sj ) .
2
The left-continuity of φ(s) implies that sj+1 is the largest s < sj such that
1
φ(s) ≤ φ(sj ).
2
1 1
Thus we have φ(sj+1 ) ≤ 2 φ(sj ) and 2 φ(sj ) < φ(s) ≤ φ(sj ) for all s ∈ (sj+1 , sj ).
Iterating this inequality we infer that
φ(sj ) ≤ 2−j φ(s0 ) ∀j.
In particular, limj→∞ φ(sj ) = 0, and Lemma 6 implies that limj→∞ sj = 0,
which exists since {sj }j≥0 is a bounded decreasing sequence of positive num-
bers.
From (2.32), we get that for any s ∈ (sj+1 , sj ),
1
(sj − s)φ(sj ) ≤ (sj − s)φ(s) ≤ C0 φ(sj )1+δ0 ,
2
so sj − s ≤ 2C0 φ(sj )δ0 . Letting s → sj+1 , we obtain
−sj+1 + sj ≤ 2C0 φ(sj )δ0 ≤ 2C0 2−δ0 j φ(s0 )δ0 .
Taking summation over j, we conclude that
∞
X 2C0
s0 = (sj − sj+1 ) ≤ φ(s0 )δ0 .
1 − 2−δ0
j=0
Thus we have
(2.34) φ(s0 ) ≥ c0
for some constant c0 > 0 depending on n, ω, C0 , s0 . 
Proof of Lemma 1. Note that on Ωs0 , us0 = ϕ + 12 |z|2 − ϕ(x0 ) − s0 < 0,
that is,
1
(2.35) − ϕ − |z|2 > −ϕ(x0 ) − s0 > 1 on Ωs0 ,
2
and the last inequality holds from our assumption that s0 < 1/2 and ϕ(x0 ) < −2.
From (2.35) we deduce that
−ϕ − 2−1 |z|2
(2.36) log 1/2
> log (−ϕ(x0 ) − s0 )1/2 > 0 on Ωs0 .
(−ϕ(x0 ) − s0 )
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 381

Integrating (2.36) against the measure enF ω n over Ωs0 , we deduce that
(2.37) log (−ϕ(x0 ) − s0 )1/2 · φ(s0 ) = log (−ϕ(x0 ) − s0 )1/2 Ωs enF ω n
R
0
−ϕ−2−1 |z|2
< Ωs log (−ϕ(x )−s )1/2 enF ω n .
R
(2.38)
0 0 0

We now apply the same Young’s inequality (2.24) as before, but this time with
the following choice of the factor V :
−ϕ − 2−1 |z|2
(2.39) V = log =: log W.
(−ϕ(x0 ) − s0 )1/2
The inequality becomes
1
( log W ) enF ≤ enF ( log (1 + enF ))p + ( log W ) exp( log W ) p − 1 .

(2.40)
Now fix p > 1. Rough upper bounds of the two terms on the right-hand side
that suffice for our purposes are the following:
(2.41) enF ( log (1 + enF ))p ≤ enF (1 + |nF |)p
and
 1  1
( log W ) exp( log W ) p − 1 ≤ ( log W ) exp( log W + Cp ) − 1
p
1
(2.42) ≤ Cp0 ( log W )W p ≤ Cp00 (1 + |W |).
Substituting back this inequality in the initial inequality (2.37), and estimating
the integral over Ωs by the full integral over Ω, we find
log (−ϕ(x0 ) − s0 )1/2 · φ(s0 )
−ϕ(z) − 12 |z|2
Z Ç å
≤ ke nF
kL1 ( log L)p + Cp00 1+ 1 ωn
Ω (−ϕ(x0 ) − s0 ) 2
Ç å
00
kϕkL1 (Ω) + C2 (ω)
≤ kenF kL1 ( log L)p + Cp C1 (ω) + 1 .
(−ϕ(x0 ) − s0 ) 2
In particular,
log (−ϕ(x0 ) − s0 )1/2 · φ(s0 )
® ´
1 nF 00 00
kϕkL1 (Ω) + C2 (ω)
≤ max ke kL1 ( log L)p + Cp C1 (ω), Cp 1 .
2 (−ϕ(x0 ) − s0 ) 2
Since φ(s0 ) is bounded from below by Lemma 7, it follows immediately that
−ϕ(x0 ) − s0 is bounded above by a constant depending only on X, ω, n, p,
kenF kL1 ( log L)p , and kϕkL1 . The proof of Theorem 1 is complete. 
We remark that compared to the compact Kähler case in [31],
R the constant
C > 0 in Theorem 1 does not depend on the “energy” E(ϕ) = X (−ϕ)enF ω n .
This is because the function ψs,k in the auxiliary equation (2.9) is local and
382 BIN GUO and DUONG H. PHONG

plurisubharmonic in Ωs , while its counterpart in [31] is global and ω-plurisub-

harmonic. Thus we do not need to shift the function −ψs,k in (2.13) by a
constant depending on E(ϕ). The proof of Theorem 1 is in particular simpler
since it does not require an argument for bounding the energy by the entropy
as in [31] .
We also observe one feature of the auxiliary equation that sets it apart
from many other auxiliary equations in the literature such as in [6]: it depends
itself on the unknown function u. This may seem to limit its applicability,
but as we see from Lemma 1, it turns out that for our purposes, very little is
needed from the auxiliary functions ψs,k .

3. The case of Hermitian compact manifolds without boundary

Once we have Lemma 1, it is easy to establish all the theorems listed in the
introduction. In this section, we begin with the case of X Hermitian compact
without boundary and give the proof of Theorem 1.
Proof of Theorem 1. Since X is compact without boundary, the function
ϕ(z) must attain its minimum at some interior point x0 . By Lemma 1, −ϕ(x0 )
is bounded above by a constant depending only on X, ω, p, n, γ, kenF kL1 ( log L)p ,
and kϕkL1 (B(x0 ,2r0 )) . Thus the proof is complete once we have proved the
following lemma giving a bound on the L1 -norm of ϕ for which λ[hϕ ] ∈ Γ ⊂ Γ1 :
Lemma 8. For any ϕ ∈ C 2 (X) such that λ[hϕ ] ∈ Γ ⊂ Γ1 and supX ϕ = 0,
there exists a uniform constant C > 0 depending only n, ω such that
Z
(−ϕ)ω n ≤ C.
X
¯ > 0; that is, in
Proof. Note that λ[hϕ ] ∈ Γ1 means that trω (ω + i∂ ∂ϕ)
local coordinates, the following holds:
∂ 2 (−ϕ)
(3.1) g ij̄ < n.
∂zi ∂ z̄j
If ω is Kähler, the lemma follows easily from Green’s formula. In general, we
will apply local elliptic estimates to −ϕ over a finite cover of (X, ω). More
precisely, we fix three sets of open covers {U` }N 0 N 00 N
`=1 , {U` }`=1 , and {U` }`=1 , such
0 00
that U` ⊂⊂ U` ⊂⊂ U` for each `. We can also view U` as a Euclidean ball of
radius r` and U`0 , U`00 as concentric balls with radii 2r` , 3r` , respectively. The
choice of these covers is fixed and depends only on (X, ω). Assume supX ϕ =
0 = ϕ(p1 ) is achieved at some point p1 ∈ U1 ⊂ U10 .
We can apply the standard elliptic estimate (see, e.g., Theorem 8.18 in
[24]) to the equation of (−ϕ) as in (3.1) on U10 ⊂ U100 to conclude
Z
(3.2) (−ϕ)ω n ≤ C[(−ϕ)(p1 ) + 1] = C1
U10
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 383

for some constant C1 > 0 depending only on U10 ⊂ U100 and ω. If U10 ∩ U20 6= ∅,
then from (3.2) we get Z
(−ϕ)ω n ≤ C1 ,
U10 ∩U20
which by the mean-value theorem implies that there is some point p2 ∈ U10 ∩ U20
such that −ϕ(p2 ) ≤ C10 for some C10 > 0. Now we can again apply the elliptic
estimate to (−ϕ) that satisfies (3.1) in U20 ⊂ U200 to conclude that
Z
(−ϕ)ω n ≤ C[(−ϕ)(p2 ) + 1] ≤ C2
U20

for some uniform constant

R C2 > n0. We can repeat this argument finitely many
0 N
times to deduce that U 0 (−ϕ)ω ≤ C` for each `, since {U` }`=1 is an open
`
cover of (X, ω), which is assumed to be connected. From this, it is easy to see
the bound of the L1 -norm of (−ϕ). Lemma 8 is proved. 

The proof of Lemma 1 already contains a proof of Corollary 1. More

precisely, by definition ϕ is a solution on X of the complex Monge-Ampère
equation
 ω n 1/n
ϕ
f (λ[hϕ ]) = = eF ,
ωn
which does satisfy conditions (1)–(4) described in the introduction. Let x0 be
a point in X where ϕ attains its minimum. Then Lemmas 2–6 apply, implying
that there exists a constant c0 > 0 depending only on n, p, ω, and K such that
Z
φ(s0 ) = enF ω n ≥ c0 .
Ωs0

¯ n.
R
It is clear that φ(s0 ) ≤ X (ω + i∂ ∂ϕ) 

We now provide a proof of inequality (1.6), without the extra assumption

on the p-th entropy of enF , but under an additional condition that X admits
a smooth closed real (1, 1)-form β whose Bott-Chern class is nef and big. The
proof of Lemma 9 is motivated by an observation of Tosatti [44] using the
Morse inequality of Demailly [12].
Lemma 9. Under the assumption above, there exists a constant c > 0 that
depends on n, ω, and β such that
Z
¯ n≥c
(ω + i∂ ∂ϕ)
X

for any C 2 -function ϕ ∈ PSH(X, ω).

Proof of Lemma 9. By a weak transcendental Morse inequality of De-
mailly [12], one can show (see, e.g., Theorem 3.1 in [44]) that for any C 2
384 BIN GUO and DUONG H. PHONG

function u, the following holds:

Z Z Z
(3.3) 0< βn = βun ≤ βun ,
X X X(βu ,0)
¯ and X(βu , 0) denotes the subset of X where the (1, 1)-
where βu = β + i∂ ∂u,
form βu is nonnegative definite.
Since β is a fixed (1, 1)-form and ω is positive definite, there exists a
constant Λ > 0 such that β ≤ Λω on X. Applying (3.3) to u = Λϕ for any
ϕ ∈ C 2 ∩ PSH(X, ω) gives
Z Z Z
0< n
β ≤ ¯ n
(β + i∂ ∂Λϕ) ≤ ¯ n
(Λω + Λi∂ ∂ϕ)
X X(βu ,0) X(βu ,0)
Z
≤ Λn ¯ n,
(ω + i∂ ∂ϕ)
X
¯
since 0 ≤ β + i∂ ∂Λϕ ¯
≤ Λω + i∂ ∂Λϕ over the set X(βu , 0). 

4. The case of Hermitian manifolds with boundary

Now we consider the case of the Dirichlet problem for equation (1.7) for an
operator f (λ) satisfying conditions (1)–(4), and we give the proof of Theorem 2.
Proof of Theorem 2. In the case of the Dirichlet problem, we cannot nor-
malize the solution to satisfy supX ϕ = 0, and we need to establish an upper
bound for ϕ. However, this can readily be done by comparing ϕ to a solution of
the Dirichlet problem for the Laplace-Beltrami equation on X. More precisely,
let h be the solution of the equation
(4.1) ∆ω h = n + 1 in X, h = 0 on ∂X,
¯ ∧ ω n−1 )ω −n is the complex Laplacian defined by the Her-
where ∆ω h = n(i∂ ∂h
mitian metric ω. In the Kähler case, ∆ω coincides with the Laplace-Beltrami
equation, and the unique solvability of the corresponding Dirichlet problem is
well known. It has been shown by Z. Lu [39] that the unique solvability of
the Dirichlet problem still holds for general Hermitian metrics ω on compact
manifolds X with smooth boundary. (As Lu himself noted, the argument is
similar to the argument for Theorem 6.8, pp. 100–101, in the classic book of D.
Gilbarg and N.S. Trudinger, “Elliptic Partial Differential Equations of Second
Order,” Springer-Verlag, Berlin-Heidelberg-New York, 1983, Third printing
1998 edition.) Thus the function h exists and is unique, and a dependence on
it can be absorbed into a dependence on (X, ω).
Now λ[hϕ ] ∈ Γ ⊂ Γ1 implies that ∆ω ϕ ≥ −n. Thus ∆(ϕ + h) ≥ 1 on X,
which implies that ϕ + h must attain its maximum on the boundary. Hence
(4.2) supX (ϕ + h) ≤ sup∂X (ϕ + h) = sup∂X ρ.
This gives an upper bound for ϕ by a constant depending only on X, ω, and ρ.
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 385

We can now replace ϕ by ϕ − supX ϕ, and thus ϕ is a non-positive solution

to (1.7). Suppose inf X ϕ = ϕ(x0 ) for some x0 ∈ X. If x0 ∈ ∂X, we are done.
Otherwise we assume x0 ∈ X and
(4.3) ϕ(x0 ) ≤ inf ρ − 1.
∂X

Take a local coordinates system (B(x0 , 2r0 ), z) centered at x0 , and (2.1) holds.
If B(x0 , 2r0 ) ⊂⊂ X, we can directly apply Theorem 1 and Lemma 6, which
gives the desired bound for − inf X ϕ, modulo an L1 (X, ω) bound of (−ϕ).
If B(x0 , 2r0 ) ∩ ∂X 6= ∅, we let s0 = 2r02 , s ∈ (0, s0 ) and consider the
function
1
(4.4) us (z) = ϕ(z) − ϕ(x0 ) + |z|2 − s ∀z ∈ Ω = B(x0 , 2r0 ) ∩ X.
2
We observe that on ∂B(x0 , 2r0 ) ∩ X, us ≥ 2r02 − s > 0. On the other half of
∂Ω, ∀z ∈ B(x0 , 2r0 ) ∩ ∂X, by (4.3) we have
1
us (z) = ρ(z) − ϕ(x0 ) + |z|2 − s ≥ 1 − s > 0.
2
Thus we have us |∂Ω > 0. We can extend the function us to a smooth function
on B(x0 , 2r0 ) satisfying us > 0 on B(x0 , 2r0 )\Ω. Since only the region when
us < 0 is under consideration, the estimates in Section 2 are independent of
the extension of us . We also extend F to be a smooth function on B(x0 , 2r0 ),
and it will be clear from (4.5) that the extension is again irrelevant.
Denote the sub-level set of us , Ωs = {z ∈ Ω|us (z) < 0}. By the discus-
sions above, Ωs ⊂⊂ X. As in Section 2, we consider the following (auxiliary)
Dirichlet boundary problem (with the extended functions us and F above),

¯ s,k )n = τk (−us ) nF n
(4.5) (i∂ ∂ψ e ω in B(x0 , 2r0 ),
As,k
and ψs,k ≡ 0 on ∂B(x0 , 2r0 ). Here τk and As,k are the same as in Section 2.
We note that the calculations in Section 2 are performed essentially in Ωs ,
which is strictly contained in X. Therefore, we still obtain Theorem 1 and
Lemma 6, and hence a lower bound of ϕ(x0 ) = inf X ϕ, depending on an L1
bound of (−ϕ). We remark that because ∂X 6= ∅, the a priori L1 -norm of ϕ
as in Lemma 8 in general does not hold.
We now focus on complex Monge-Ampère and Hessian equations with ω
being Kähler, i.e., dω = 0. We rewrite equation (1.3) as (1 ≤ k ≤ n):
(4.6) ¯ k ∧ ω n−k = eF̃ ω n in X, ϕ = ρ on ∂X.
(ω + i∂ ∂ϕ)
Here we denote F̃ = kF to simplify notation. We will write ωϕ = ω+i∂ ∂ϕ ¯ ∈ Γk .
When k = n, (4.6) reduces to the Monge-Ampère equation. Since ϕ is always
bounded above depending on ω and ρ, we may assume that supX ϕ ≤ 0, and
that inf X ϕ = ϕ(x0 ) ≤ inf ∂X ρ − 1 for some x0 ∈ X ◦ ; otherwise, we are done.
386 BIN GUO and DUONG H. PHONG

For simplicity of notation, we consider the function

ϕ̃ = min(0, ϕ − inf ρ),

∂X

which is Lipschitz in X and ϕ̃|∂X ≡ 0. We calculate

Z
log (−ϕ̃ + 1)(ωϕk − ω k ) ∧ ω n−k
ZX
= log (−ϕ̃ + 1)i∂ ∂ϕ¯ ∧ (ω k−1 + ω k−2 ∧ ω + · · · + ω k−1 ) ∧ ω n−k
ϕ ϕ
ZX
1 ¯ ∧ (ω k−1 + ω k−2 ∧ ω + · · · + ω k−1 ) ∧ ω n−k
= i∂ ϕ̃ ∧ ∂ϕ ϕ ϕ
X −ϕ̃ + 1
Z
(4.7) + log (−ϕ̃ + 1)i∂ϕ¯ ∧ (ω k−1 + ω k−2 ∧ ω + · · · + ω k−1 ) ∧ ω n−k
ϕ ϕ
Z∂X  
(4.8) − log (−ϕ̃ + 1)i∂¯ϕ̃ ∧ ∂ (ωϕk−1 + ωϕk−2 ∧ ω + · · · + ω k−1 ) ∧ ω n−k
ZX p
4 2
≥ ∇ −ϕ̃ + 1 ω n ,
n X ω

where the integral in (4.7) vanishes since log (−ϕ̃ + 1)|∂X ≡ 0, and in (4.8)
is zero since we assume ω is Kähler. The last inequality holds because in the
third line we have ∂ϕ¯ = ∂¯ϕ̃ on the support of i∂ ϕ̃.
On the other hand, the left-hand-side of the equation above satisfies
Z Z
k k n−k
log (−ϕ̃ + 1)(ωϕ − ω ) ∧ ω = log (−ϕ̃ + 1)(eF̃ − 1)ω n
X X
Z
≤ log (−ϕ̃ + 1)eF̃ ω n
X

since log (−ϕ̃ + 1) ≥ 0. We can now apply Young’s inequality as in (2.24) with
p = 1, v = 41 log (−ϕ̃ + 1) to write

log (−ϕ̃ + 1)eF̃ ≤ 4 eF̃ (1 + |F̃ |) + C −ϕ̃ + 1

p

and hence
Z Z Z p
log (−ϕ̃ + 1)(ωϕk − ω k ) ∧ ω n−k ≤ 4 eF̃ (1 + |F̃ |)ω n + C −ϕ̃ + 1ω n
X X X
Z p
≤C +C −ϕ̃ + 1ω n .
X

Combining the above, we obtain

Z Z p
2
n
( −ϕ̃ + 1 − 1)ω n .
p
∇( −ϕ̃ + 1 − 1) ω ≤ C + C
X ω X
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 387

√
Observing that the function ( −ϕ̃ + 1 − 1) is equal to 0 on ∂X, we can apply
the Poincaré inequality on (X, ω) to get (CP below is the Poincaré constant)
Z p Z
2
2 n
∇( −ϕ̃ + 1 − 1) ω n
p
( −ϕ̃ + 1 − 1) ω ≤ CP
X X ω
Z p
≤C +C ( −ϕ̃ + 1 − 1)ω n
X
Z p 1/2
≤C +C ( −ϕ̃ + 1 − 1)2 ω n ,
X
√
−ϕ̃ + 1 − 1)2 ω n ≤ C, hence
R
from which we easily get X(
Z
(−ϕ̃)ω n ≤ C.
X

Since ϕ̃ differs from ϕ by a uniform constant that depends only on inf ∂X ρ,

this proves the L1 (X, ω n ) bound of the function |ϕ|. 

5. The case of non-compact Hermitian manifolds

We now turn to the proof of Theorem 3, which is analogous to that of
Theorem 2, so we only point out the very minor differences.

Proof of Theorem 3. First note that we are allowing a dependence on the

(local) L1 -norm of ϕ in the constant C in Theorem 3. With such an allowance,
the upper bound of ϕ is local since ∆ω ϕ ≥ −n, and it follows from the standard
elliptic estimates (e.g., Theorem 8.18 in [24]).
Next we derive the lower bound for ϕ. Let liminf z→∞ ϕ > −B for some
positive constant B, which can be viewed as part of the data. Then there is
a compact set K ⊂⊂ X with ϕ ≥ −B − 1 for z ∈ X \ K. If ϕ ≥ −B on K,
we are done. Otherwise, the infimum of ϕ over X must take place at some
interior point x0 of K. We can now apply Theorem 1 and obtain the desired
bound for −ϕ(x0 ) = −inf X ϕ. 

6. The case of (n − 1)-form type non-linear equations

We come now to the (n − 1) form type non-linear equation and the proof
of Theorem 4.
Proof of Theorem 4. We first fix a constant Cθ = Cθ (ωh , θ) such that
iθ ∧ θ̄ ≤ Cθ2 ωh . Recall that ϕ is a C 2 solution to the (n − 1)-form non-linear
equation (1.13). For notational convenience, we denote
1 ¯ + χ[ϕ]
ω̃ = ωh + ((∆ω ϕ)ω − i∂ ∂ϕ)
n−1
388 BIN GUO and DUONG H. PHONG

for the the (1, 1)-form on the left-side of (1.13), and h̃ϕ = ω −1 · ω̃. Write
∂ log f (λ[h̃ϕ ])
Gij̄ = ∂ h̃ij̄
to be linearized operator of log f (λ[·]), which is positive
definite by (3) of the assumptions on f . We will denote trG α = Gij̄ αij̄ for any
(1, 1)-form α. Write g = (gij̄ ) to be the metric associated to the background
Hermitian metric ω and g ij̄ its inverse.
As in [46] and [36], we define a tensor Θij̄ by
1  
Θij̄ = (trG ω)g ij̄ − Gij̄ .
n−1
Lemma 10. The tensor Θij̄ is positive definite at any point and satisfies
γ
(6.1) det(Θij̄ ) ≥ nF det(g ij̄ ) on X.
e
Proof. Given any x0 ∈ X, we can choose local holomorphic coordinates
around x0 such that gij̄ |x0 = δij and ω̃|x0 = λi δij is diagonal. Note that
1 ∂f (λ) 1 ∂f (λ)
λ = (λ1 , . . . , λn ) ∈ Γ and Gij̄ |x0 = f ∂λi δij = µi δij for µi = f ∂λi > 0.
Therefore, at x0 , we have
1 X 
Θij̄ = µk δij ,
n−1
k6=i

which is clearly positive definite. Furthermore, we have

n X n Y 1/(n−1)
1 Y Y
det(Θij̄ )|x0 = µ k ≥ µ k
(n − 1)n
i=1 k6=i i=1 k6=i
n n
Y 1 Y ∂f γ
= µi = ≥ n,
fn ∂λi f
i=1 i=1

where the inequality follows from the arithmetic-geometric (AG) inequality

and the last equality follows from the structural condition (1.2) on f . The
proof of the lemma is complete since x0 ∈ X is an arbitrary point in X. 

Next, we take a minimum point x0 of ϕ, i.e., ϕ(x0 ) = inf X ϕ. Let

(z1 , . . . , zn ) be a local holomorphic coordinate system centered at x0 , such
that on the Euclidean ball Ω = B(x0 , 2r0 ) ⊂ Cn defined by these coordinates,
the following holds:
1 ¯ 2 ¯ 2.
(6.2) i∂ ∂|z| ≤ ω ≤ 2i∂ ∂|z|
2
We fix a small constant 0 > 0, which will be determined later. Let zi = x2i−1 +
√
−1x2i . Hence {xa }2n 0 2
a=1 is a local real coordinates system. Let s0 =  r0 . For
any s ∈ (0, s0 ), we define a function us similar to that in (2.5):
(6.3) us (z) := ϕ(z) − ϕ(x0 ) + 0 |z|2 − s ∀z ∈ Ω := B(x0 , 2r0 ).
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 389

Note that when z ∈ Ω\B(x0 , r0 ), us (z) ≥ 0 r02 − s > 0. Therefore the sub-level
set of us , Ωs := {z ∈ Ω| us (z) < 0} is relatively compact in Ω and is contained
in B(x0 , r0 ). By definition, Ωs is a non-empty open set, since x0 ∈ Ωs . We
consider the following real Monge-Ampère equation:
 ∂ 2 ψ  τ (−u )
s,k k s 2nF
det = e (detgij̄ )2 in Ω.
∂xa ∂xb As,k
Note that ψs,k is convex  such that ψs,k =R 0 on ∂Ω. Here As,k > 0 is chosen
∂ 2 ψs,k
so that Ω det ∂xi ∂xj dx = 1; i.e., As,k = B(x0 ,2r0 ) τk (−us )e2nF (detgij̄ )2 dx →
R

2nF (detg )2 dx =: A . Note that A ≤ C(ω)s ke2nF k

R
Ωs (−us )e ij̄ s s 0 L1 (X,ω n ) , so we
may assume As,k is uniformly bounded, by taking k large enough. Denote βn
as the volume of the unit ball in R2n .
Lemma 11. There exists a constant C0 = C0 (n) > 0 such that
(6.4) − inf B(x0 ,2r0 ) ψs,k ≤ C0 r0 , supB(x0 ,r0 ) |∇ψs,k | ≤ C0 .
 2 
R ∂ ψs,k
Proof. By the definition of ψs,k , B(x0 ,2r0 ) det ∂xi ∂x j dx = 1, it follows

from the standard ABP maximum principle (see, e.g., [24]) that
4r0 h
Z  ∂ 2 ψ  i1/2n 4r
s,k 0
− inf ψs,k ≤ − inf ψs,k + det dx = .
B(x0 ,2r0 ) ∂B(x0 ,2r0 ) βn B(x0 ,2r0 ) ∂x i ∂x j β n
To see the second inequality, for any fixed point x ∈ B(x0 , r0 ), denote Y =
Dψs,k (x)
|Dψs,k (x)| as the unit vector in the direction of Dψs,k (x). (If Dψs,k (x) = 0,
there is nothing to prove, so here we assume Dψs,k (x) 6= 0.) Then clearly
|Dψs,k (x)| = hDψs,k (x), Y i. Consider the half line L : 0 ≤ t 7→ x + tY that
intersects ∂B(x0 , r0 ) and ∂B(x0 , 2r0 ) at L(t1 ), L(t2 ), respectively. We have
t2 − t1 ≥ r0 and 0 > ψs,k (L(t1 )) ≥ − β4n r0 and ψs,k (L(t2 )) = 0. Then by the
convexity of the function t 7→ ψs,k (L(t)), we have
ψs,k (L(t2 )) − ψs,k (L(t1 )) 4
|Dψs,k (x)| = ψs,k (L(t))0 |t=0 ≤ ≤ .
t2 − t1 βn
Taking supremum over all x ∈ B(x0 , r0 ) finishes the proof of the lemma. 
It holds that
h  ∂ 2 ψ i1/2 τk (−us )1/2 nF
s,k
(6.5) det(ψs,k )ij̄ ≥ 2−n det = 2−n 1/2
e det(gij̄ ).
∂xa ∂xb A s,k
The key is now to establish the analogue of Lemma 2, but where us is now
constructed from the above form-type non-linear equation, instead of solutions
of the non-linear scalar equation f (λ):
Lemma 12. We have the comparison estimate on Ω,
2n
(6.6) −us ≤ ε(−ψs,k + Λ) 2n+1 ,
 2n+1
(2n+1)2n 10C0 Cθ nε
where ε is defined by ε1+2n = A ,
24n n4n γ 2 s,k
and Λ = 2n+1 .
390 BIN GUO and DUONG H. PHONG

Proof. We have to show that the test function

2n
Φ = −ε(−ψs,k + Λ) 2n+1 − us
is ≤ 0 on Ω. This is clear if max Ψ is achieved at Ω\Ωs . So assume maxΩ Φ =
Φ(xmax ) for some xmax ∈ Ωs . By the maximum principle we have i∂ ∂Φ| ¯ x ≤0
max
and dΦ|xmax = 0. Note that at xmax ∈ Ωs ⊂ B(x0 , r0 ), dΦ|xmax = 0 implies
that at xmax we have
2nε 1
ϕi = (−ψs,k + Λ)− 2n+1 (ψs,k )i − 0 z̄i .
2n + 1
Hence at xmax it follows that
Å ã
ij̄ ij̄ 2nε 1
− 2n+1 0
|2Re(G ϕi θ̄j )| = 2 ReG θ̄j (−ψs,k + Λ) (ψs,k )i −  z̄i
2n + 1
2nε − 2n+1
1
(6.7) ≤ 2C0 Cθ Λ trG ωh + 20 r0 Cθ trG ωh
2n + 1
2
≤ trG ωh
5
if we choose the constant 0 > 0 small enough such that
100 1
ωh ≥ (trω ωh )ω and 20 r0 Cθ ≤ .
n−1 5
In (6.7) we have applied (6.4) to bound |Dψs,k | at xmax ∈ B(x0 , r0 ). We define
a linear operator Lv = Θij̄ vij̄ . The fact that Θij̄ is positive definite implies
that LΨ|xmax ≤ 0. We calculate at xmax as follows:
0 ≥ LΨ|xmax
2nε 1
= (−ψs,k + Λ)− 2n+1 Θij̄ (ψs,k )ij̄
2n + 1
2εn 2n+2
+ 2
(−ψs,k + Λ)− 2n+1 Θij̄ (ψs,k )i (ψs,k )j̄
(2n + 1)
0
− 1 + trG ωh − (trG ω · trω ωCn − trG ωCn ) + 2Re(Gij̄ ϕi θ̄j )
n−1
2nε 1
≥ (−ψs,k + Λ)− 2n+1 Θij̄ (ψs,k )ij̄
2n + 1
20 2
− 1 + trG [ωh − (trω ωh )ω] − trG ωh
n−1 5
4n2 ε 1
≥ (−ψs,k + Λ)− 2n+1 (detΘij̄ )1/n [det(ψs,k )ij̄ ]1/n − 1
2n + 1
2n2 εγ 1/n 1 (−us )1/2n
≥ (−ψs,k + Λ)− 2n+1 1/2n
− 1.
2n + 1 A s,k

The term 2Re(Gij̄ ϕi θ̄j ) with first order derivatives of ϕ is the only reason that
we consider the real Monge-Ampère equation of ψs,k , instead of complex MA
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 391

equations as in the sections above. Here the first equality follows from the
definition of Θij̄ and the formula
−Θij̄ ϕij̄ = −Lϕ = −1 + trG ωh + 2Re(Gij̄ ϕi θ̄j ),
which is an easy consequence of the definition of ω̃. In the fourth line, we
applied the AG inequality. The last line follows from Lemma 10 and (6.5). By
the choice of ε, this implies that Φ(xmax ) ≤ 0. Hence supΩ Φ ≤ 0. 
This means that for the form type non-linear equation, we now have the
analogue of Lemma 2. In fact, in the current case, the function −ψs,k + Λ is
uniformly bounded, hence the integral estimate (2.16) still holds. We can then
proceed in the same way as in Section 2 and apply Lemma 3 to obtain the
lower bound of ϕ(x0 ) = inf X ϕ. We remark that the uniform L1 (X, ω n ) bound
of ϕ still holds in this setting. This is because the assumption that λ[h̃ϕ ] ∈ Γ1
implies that ∆ω ϕ + 2Re(g ij̄ ϕi θ̄j ) ≥ −trω ωh ≥ −C. A slight modification of
Lemma 8 then yields the desired L1 bound of (−ϕ) given our normalization
supX ϕ = 0. The proof of Theorem 4 is complete.

7. The case of non-integrable almost-complex structures

In this section, we generalize the comparison approach using real Monge-
Ampère equation as the auxiliary equation. Since J is not integrable, there is
no natural choice of local holomorphic coordinates on X. Instead, we will use
local real coordinates and the real Monge-Ampère equation as the auxiliary
equation. Let r0 > 0 be a fixed number smaller than 1/10 of the injectivity
radius of (X, ω), viewed as a Riemannian manifold. Suppose x0 ∈ X is a point
where ϕ achieves its minimum. We choose the normal coordinates (x1 , . . . , xm )
centered at x0 , where m = 2n. Without loss of generality, we may assume
B(x0 , 2r0 ) = {x||x| < 2r0 } is an Euclidean ball and in this ball
1
(7.1) δab ≤ gab ≤ 2δab , a, b = 1, 2, . . . , m.
2
Let η > 0 be a small number to be determined. For any s ∈ (0, s + 0 = ηr02 ],
we consider the function us (x) defined on B(x0 , 2r0 ):
(7.2) us (x) = ϕ(x) − ϕ(x0 ) + η|x|2 − s.
It is clear that us ≥ −s on B(x0 , 2r0 ). Define the sublevel set Ωs = {x ∈
B(x0 , 2r0 )|us (x) < 0}. Note that x0 ∈ Ωs , so Ωs is a non-empty open set. By
the choice of s, we have us (x) > ηr02 − s > 0 for any x ∈ B(x0 , 2r0 )\B(x0 , r0 ),
hence Ωs ⊂ B(x0 , r0 ). We solve the following real Monge-Ampère equation on
B(x0 , 2r0 ),
 ∂ 2 ψ  τ (−u )
s,k k s 2nF
(7.3) det = e ,
∂xa ∂xb As,k
392 BIN GUO and DUONG H. PHONG

where ψs,k is convex such

 2 that  ψs,k = 0 on ∂B(x0 , 2rR0 ). Here As,k > 0 is chosen
∂ ψs,k
so that B(x0 ,2r0 ) det ∂xa ∂xb dx = 1; i.e., As,k = B(x0 ,2r0 ) τk (−us )e2nF dx →
R

2nF dx =: A .
R
Ωs (−us )e s
We observe that the estimates (6.4) in Lemma 11 hold for this ψs,k ; in
particular, supB(x0 ,r0 ) |Dψs,k | ≤ C0 for some C0 = C(n).
Define a test function Φ on B(x0 , 2r0 ) by
2n
(7.4) Φ(x) = −ε(−ψs,k (x) + Λ) 2n+1 − us (x),
where
 4n + 2  2n
2n+1 1/(2n+1)
ε= γ −2/(2n+1) As,k , with Λ to be determined.
n2
We claim that Φ ≤ 0 in B(x0 , 2r0 ). We may assume that the maximum of
Φ is obtained at xmax ∈ Ωs ; otherwise the claim follows easily. By maximum
principle, at xmax , DΦ = 0 and D2 Φ ≤ 0. Define L to be the linearized
operator of ϕ 7→ log f (λ[hϕ ]). Take a local unitary frame (with respect to ω)
{e1 , . . . , en } of T (1,0) X at xmax , and let {θ1 , . . . , θn } be the co-frame. Write
gij̄ = g(ei , ēj ). Then
√
ω = gij̄ −1θi ∧ θ̄j , and gij̄ = δij at xmax .
¯ is still Hermitian in the almost-complex structure setting.
We remark that i∂ ∂ϕ
By a unitary change ¯
√ of eii , wej may assume that ωϕ = ω + i∂ ∂ϕ is diagonal at
xmax and ωϕ = g̃ij̄ −1θ ∧ θ̄ , g̃ij̄ = gij̄ + ϕij̄ , where for a function ϕ,
ϕij̄ := ei ēj ϕ − [ei , ēj ](1,1) ϕ.
That is, ϕij̄ is decomposed as the sum of a second order derivative and a linear
term in the first order derivative Dϕ, such that the latter is bounded by C1 |Dϕ|
for a uniform constant C1 = C1 (ω, n) > 0. We now determine the constant
Λ = ( 20nC 0 C1 2n+1 2n+1
2n+1 ) ε = C(n, ω, γ)As,k . With this notation, LΦ = Gij̄ Φij̄ at
xmax . Then we calculate at xmax as follows:
0 ≥ LΦ
2nε 1
= (−ψs,k + Λ)− 2n+1 Gij̄ (ψs,k )ij̄
2n + 1
2nε 2n+2
+ (−ψs,k + Λ)− 2n+1 |∂ψs,k |2G − Gij̄ ϕij̄ − 4ηtrG g
(2n + 1)2
2nε 1
≥ (−ψs,k + Λ)− 2n+1 Gij̄ Di2j̄ ψs,k r
2n + 1
2nC1 ε 1
− (−ψs,k + Λ)− 2n+1 trG g|Dψs,k | + trG g − 1 − 4ηtrG g
2n + 1
2n2 ε 1 γ 1/n
≥ (−ψs,k + Λ)− 2n+1 F (det[Di2j̄ ψs,k ])1/n − 1
2n + 1 e
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 393

2n2 ε 1 γ 1/n
≥ (−ψs,k + Λ)− 2n+1 F {4−n (det[Dx2a xb ψs,k ])1/2 }1/n − 1
2n + 1 e
2
n ε 1 γ 1/n (−us )e2nF 1/2n
≥ (−ψs,k + Λ)− 2n+1 F ( ) − 1.
4n + 2 e As,k
We have choosen η = 1/10, and the third inequality follows from the choice
of Λ and η, and the arithmetic geometric inequality. The fourth inequality
follows from the comparison of the determinants of complex and real Hessians.
This proves the claim that Φ ≤ 0. We can then proceed in the same way as
in Section 2 and apply Lemma 3 to obtain the lower bound of ϕ(x0 ) = inf X ϕ.
We remark that the uniform L1 (X, ω n ) bound of ϕ still holds in this setting
by Lemma 13 below. The proof of Theorem 5 is complete.
Lemma 13. Suppose λ[hϕ ] ∈ Γ ⊂ Γ1 . If we normalize supX ϕ = 0, then
Z
(−ϕ)ω n ≤ C(n, ω).
X
ω ∧i∂ ∂ϕ n−1 ¯
Proof. First, λ[hϕ ] ∈ Γ1 implies that ∆C C
ω ϕ ≥ −n, where ∆ω ϕ = n ωn
is the canonical Laplacian of ϕ. Second, it follows from standard calculations
(e.g., [10]) that ∆R C
ω ϕ = 2∆ω ϕ + τ (dϕ) for some “torsion vector field” τ of
(X, ω), where ∆R ω ϕ is the usual Laplacian of ω viewed as a Riemannian metric.
Hence ∆R ω ϕ − τ (dϕ) ≥ −n. We can then apply the standard elliptic estimates
(Theorem 8.18 in [24]) over a local covering of X as in the proof of Lemma 8
to get the L1 estimate of −ϕ. 

8. The case of tame symplectic forms

The proof of Theorem 6 is analogous to that of Theorem 5, so we only
point out the main differences. Recall that m = 2n is the real dimension. We
denote the Riemannian metric associated to Ω by g, i.e.,
1
g(U, V ) = (Ω(U, JV ) + Ω(V, JU )).
2
Around a point x0 ∈ X in which ϕ achieves the minimum, we take a local real
coordinates system (x1 , . . . , xm ) under which the metric g satisfies (7.1). The
almost-Kähler condition implies that the Calabi-Yau equation ω̃ n = enF Ωn is
equivalent to the real equation det(g̃ab ) = e2nF det(gab ), where as before we
use the letters a, b, . . . to denote the indices {1, 2, . . . , m}. A straightforward
computation using the condition that ω̃ is almost-Kähler shows that
1 ∂J a b ∂J d b
g̃ ab Γ̃cab = − g̃ cd J b a − g̃ ab J c d ,
2 ∂xd ∂xa
2
which implies that the Laplace-Beltrami operator ∆g̃ u = g̃ ab ∂x∂a ∂x
u
b
∂u
− g̃ ab Γ̃cab ∂x c
is independent of the first order derivatives of g̃.
394 BIN GUO and DUONG H. PHONG

We define the function us as in (7.2) and solve the auxiliary real Monge-
Ampère equation (7.3). For parameters ε and Λ, we consider the test function
Φ as in (7.4). The observation that first order derivatives of the unknown
metric g̃ is not involved in ∆g̃ enables us to apply the maximum principle to
the linear operator LΦ := ∆g̃ Φ, and with appropriate choice of the constants
ε, Λ, we can derive that Φ ≤ 0. The rest of the proof is similar to other sections,
so we omit the details.

9. Further developments
We conclude by mentioning several recent developments of interest. The
first is an extension of the present method for L∞ estimates to parabolic equa-
tions [7]. The second is an extension of the theory to Hölder estimates [8],
and the third is an application of the method to new bounds for diameters
and Green’s functions in Kähler geometry [30], resulting in a positive answer
for some long standing questions on the Kähler-Ricci flow [29]. In a different
direction, building on the progress on Green’s functions in [30], the existence of
Hermitian-Einstein metrics on stable reflexive sheaves on normal Kähler spaces
in the sense of Grauert has been very recently established in [5].
Acknowledgements. The authors would like to thank Zhiqin Lu for very
helpful discussions on the Laplacian of Hermitian metrics. They are also grate-
ful to the referees for their helpful suggestions and comments.

References
[1] Z. Blocki, On the uniform estimate in the Calabi-Yau theorem, II, Sci. China
Math. 54 no. 7 (2011), 1375–1377. MR 2817572. Zbl 1239.32032. https://doi.
org/10.1007/s11425-011-4197-6.
[2] H. Brezis and F. Merle, Uniform estimates and blow-up behavior for solutions
of −∆u = V (x)eu in two dimensions, Comm. Partial Differential Equations 16
no. 8-9 (1991), 1223–1253. MR 1132783. Zbl 0746.35006. https://doi.org/10.
1080/03605309108820797.
[3] L. Caffarelli, J. J. Kohn, L. Nirenberg, and J. Spruck, The Dirich-
let problem for nonlinear second-order elliptic equations. II. Complex Monge-
Ampère, and uniformly elliptic, equations, Comm. Pure Appl. Math. 38 no. 2
(1985), 209–252. MR 0780073. Zbl 0598.35048. https://doi.org/10.1002/cpa.
3160380206.
[4] L. Caffarelli, L. Nirenberg, and J. Spruck, The Dirichlet problem for
nonlinear second-order elliptic equations. III. Functions of the eigenvalues of the
Hessian, Acta Math. 155 no. 3-4 (1985), 261–301. MR 0806416. Zbl 0654.35031.
https://doi.org/10.1007/BF02392544.
[5] J. Cao, P. Graf, P. Naumann, T. Peternell, M. Paun, and X. Wu,
Hermite–Einstein metrics in the singular setting, 2023. arXiv 2303.08773.
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 395

[6] X. Chen and J. Cheng, On the constant scalar curvature Kähler metrics (I)—A
priori estimates, J. Amer. Math. Soc. 34 no. 4 (2021), 909–936. MR 4301557.
Zbl 1472.14042. https://doi.org/10.1090/jams/967.
[7] X. Chen and J. Cheng, The L∞ estimates for parabolic complex Monge-
Ampère and Hessian equations, Pure Appl. Math. Q. 19 no. 6 (2023), 2869–2913.
MR 4703037. Zbl 07801221. https://doi.org/10.4310/PAMQ.2023.v19.n6.a10.
[8] J. R. Cheng and Y. Xu, Regularization of m-subharmonic functions and Hölder
continuity, 2022, to appear in Math. Res. Lett. arXiv 2208.14539.
[9] P. Cherrier, Équations de Monge-Ampère sur les variétés hermitiennes com-
pactes, Bull. Sci. Math. (2) 111 no. 4 (1987), 343–385. MR 0921559. Zbl 0629.
58028.
[10] J. Chu, V. Tosatti, and B. Weinkove, The Monge-Ampère equation for non-
integrable almost complex structures, J. Eur. Math. Soc. (JEMS ) 21 no. 7 (2019),
1949–1984. MR 3959856. Zbl 1452.32035. https://doi.org/10.4171/JEMS/878.
[11] E. De Giorgi, Sulla differenziabilità e l’analiticità delle estremali degli integrali
multipli regolari, Mem. Accad. Sci. Torino. Cl. Sci. Fis. Mat. Nat. (3) 3 (1957),
25–43. MR 0093649. Zbl 0084.31901.
[12] J.-P. Demailly, Holomorphic Morse inequalities and asymptotic cohomology
groups: a tribute to Bernhard Riemann, Milan J. Math. 78 no. 1 (2010), 265–
277. MR 2684780. Zbl 1205.32017. https://doi.org/10.1007/s00032-010-0118-3.
[13] J.-P. Demailly and N. Pali, Degenerate complex Monge-Ampère equations
over compact Kähler manifolds, Internat. J. Math. 21 no. 3 (2010), 357–405.
MR 2647006. Zbl 1191.53029. https://doi.org/10.1142/S0129167X10006070.
[14] S. Dinew, Lectures on pluripotential theory on compact Hermitian manifolds,
in Complex Non-Kähler Geometry, Lecture Notes in Math. 2246, Springer,
Cham, 2019, pp. 1–56. MR 3971999. Zbl 1460.32075. https://doi.org/10.1007/
978-3-030-25883-2 1.
[15] S. Dinew and S. Kolodziej, Pluripotential estimates on compact Hermitian
manifolds, in Advances in Geometric Analysis, Adv. Lect. Math. (ALM ) 21, Int.
Press, Somerville, MA, 2012, pp. 69–86. MR 3077248. Zbl 1317.32066.
[16] S. Dinew and S. Kolodziej, A priori estimates for complex Hessian equations,
Anal. PDE 7 no. 1 (2014), 227–244. MR 3219505. Zbl 1297.32020. https://doi.
org/10.2140/apde.2014.7.227.
[17] P. Eyssidieux, V. Guedj, and A. Zeriahi, Singular Kähler-Einstein metrics,
J. Amer. Math. Soc. 22 no. 3 (2009), 607–639. MR 2505296. Zbl 1215.32017.
https://doi.org/10.1090/S0894-0347-09-00629-8.
[18] T. Fei, D. H. Phong, S. Picard, and X. Zhang, Geometric flows for the type
IIA string, Camb. J. Math. 9 no. 3 (2021), 693–807. MR 4400736. Zbl 1486.
53108. https://doi.org/10.4310/CJM.2021.v9.n3.a3.
[19] A. Fino and A. Raffero, Coupled SU(3)-structures and supersymmetry, Sym-
metry 7 no. 2 (2015), 625–650. MR 3366331. Zbl 1373.81414. https://doi.org/
10.3390/sym7020625.
[20] J.-X. Fu and S.-T. Yau, The theory of superstring with flux on non-Kähler
manifolds and the complex Monge-Ampère equation, J. Differential Geom. 78
396 BIN GUO and DUONG H. PHONG

no. 3 (2008), 369–428. MR 2396248. Zbl 1141.53036. https://doi.org/10.4310/

jdg/1207834550.
[21] J. Fu, Z. Wang, and D. Wu, Form-type Calabi-Yau equations, Math. Res.
Lett. 17 no. 5 (2010), 887–903. MR 2727616. Zbl 1238.32016. https://doi.org/
10.4310/MRL.2010.v17.n5.a7.
[22] M. Garcia-Fernandez, Lectures on the Strominger system, in Travaux
Mathématiques. Vol. XXIV, Trav. Math. 24, Fac. Sci. Technol. Commun. Univ.
Luxemb., Luxembourg, 2016, pp. 7–61. MR 3643933. Zbl 1379.35265.
[23] M. George, B. Guan, and C. Qiu, Fully nonlinear elliptic equations on
Hermitian manifolds for symmetric functions of partial Laplacians, J. Geom.
Anal. 32 no. 6 (2022), Paper No. 183, 27. MR 4411746. Zbl 1489.35087.
https://doi.org/10.1007/s12220-022-00918-y.
[24] D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations of
Second Order, Classics in Math., Springer-Verlag, Berlin, 2001, Reprint of the
1998 edition. MR 1814364. Zbl 1042.35002.
[25] B. Guan and P. Guan, Convex hypersurfaces of prescribed curvatures, Ann.
of Math. (2) 156 no. 2 (2002), 655–673. MR 1933079. Zbl 1025.53028. https:
//doi.org/10.2307/3597202.
[26] B. Guan and X. Nie, Fully nonlinear elliptic equations with gradient terms on
Hermitian manifolds, Int. Math. Res. Not. IMRN no. 16 (2023), 14006–14042.
MR 4631427. Zbl 07794873. https://doi.org/10.1093/imrn/rnac219.
[27] V. Guedj and C. H. Lu, Quasi-plurisubharmonic envelopes 3: Solving
Monge-Ampère equations on hermitian manifolds, J. Reine Angew. Math.
800 (2023), 259–298. MR 4609828. Zbl 1525.32024. https://doi.org/10.1515/
crelle-2023-0030.
[28] B. Guo and D. H. Phong, Auxiliary Monge-Ampère equations in geometric
analysis, ICCM Not. 11 no. 1 (2023), 98–135. MR 4614271. Zbl 1519.35198.
https://doi.org/10.4310/ICCM.2023.v11.n1.a10.
[29] B. Guo, D. H. Phong, J. Song, and J. Sturm, Diameter estimates in Kähler
geometry, 2022, First published 22 February 2024. https://doi.org/10.1002/cpa.
22196.
[30] B. Guo, D. H. Phong, and J. Sturm, Green’s functions and complex Monge-
Ampère equations, 2022, to appear in J. Differential Geom. arXiv 2202.04715.
[31] B. Guo, D. H. Phong, and F. Tong, On L∞ estimates for complex Monge-
Ampère equations, Ann. of Math. (2) 198 no. 1 (2023), 393–418. MR 4593734.
Zbl 1525.35121. https://doi.org/10.4007/annals.2023.198.1.4.
[32] F. R. Harvey and H. B. Lawson, Jr., Dirichlet duality and the nonlin-
ear Dirichlet problem, Comm. Pure Appl. Math. 62 no. 3 (2009), 396–443.
MR 2487853. Zbl 1173.35062. https://doi.org/10.1002/cpa.20265.
[33] F. R. Harvey and H. B. Lawson, Jr., Dirichlet duality and the nonlinear
Dirichlet problem on Riemannian manifolds, J. Differential Geom. 88 no. 3
(2011), 395–482. MR 2844439. Zbl 1235.53042. https://doi.org/10.4310/jdg/
1321366356.
 L∞ ESTIMATES FOR FULLY NON-LINEAR PDE’S 397

[34] F. R. Harvey and H. B. Lawson, Jr., Determinant majorization and the work
of Guo-Phong-Tong and Abja-OLive, Calc. Var. Partial Differential Equations
62 no. 5 (2023), Paper No. 153, 28. MR 4589710. Zbl 1520.15005. https://doi.
org/10.1007/s00526-023-02485-8.
[35] N. Hitchin, The geometry of three-forms in six dimensions, J. Differential Geom.
55 no. 3 (2000), 547–576. MR 1863733. Zbl 1036.53042. https://doi.org/10.
4310/jdg/1090341263.
[36] N. Klemyatin and C. Liang, S.and Wang, On uniform estimates for (n − 1)-
form fully nonlinear partial differential equations on compact Hermitian mani-
folds, 2022. arXiv 2211.13798.
[37] S. Kolodziej, The complex Monge-Ampère equation, Acta Math. 180
no. 1 (1998), 69–117. MR 1618325. Zbl 0913.35043. https://doi.org/10.1007/
BF02392879.
[38] S.-Y. Li, On the Dirichlet problems for symmetric function equations of the
eigenvalues of the complex Hessian, Asian J. Math. 8 no. 1 (2004), 87–106.
MR 2128299. Zbl 1068.32024. https://doi.org/10.4310/AJM.2004.v8.n1.a8.
[39] Z. Lu, Hermitian Dirichlet problem, preprint available at the author’s website.
[40] D. H. Phong, Geometric partial differential equations from unified string theo-
ries, in Proceedings of the International Consortium of Chinese Mathematicians
2018, Int. Press, Boston, MA, 2020, pp. 67–87. MR 4251137. Zbl 1462.35004.
[41] D. Popovici, Non-Kähler mirror symmetry of the Iwasawa manifold, Int. Math.
Res. Not. IMRN no. 23 (2020), 9471–9538. MR 4182802. Zbl 1460.32056. https:
//doi.org/10.1093/imrn/rny256.
[42] G. Székelyhidi, V. Tosatti, and B. Weinkove, Gauduchon metrics with
prescribed volume form, Acta Math. 219 no. 1 (2017), 181–211. MR 3765661.
Zbl 1396.32010. https://doi.org/10.4310/ACTA.2017.v219.n1.a6.
[43] G. Tian and S.-T. Yau, Complete Kähler manifolds with zero Ricci curvature.
I, J. Amer. Math. Soc. 3 no. 3 (1990), 579–609. MR 1040196. Zbl 0719.53041.
https://doi.org/10.2307/1990928.
[44] V. Tosatti, The Calabi-Yau theorem and Kähler currents, Adv. Theor. Math.
Phys. 20 no. 2 (2016), 381–404. MR 3541848. Zbl 1349.14136. https://doi.org/
10.4310/ATMP.2016.v20.n2.a4.
[45] V. Tosatti and B. Weinkove, The complex Monge-Ampère equa-
tion on compact Hermitian manifolds, J. Amer. Math. Soc. 23 no. 4
(2010), 1187–1195. MR 2669712. Zbl 1208.53075. https://doi.org/10.1090/
S0894-0347-2010-00673-X.
[46] V. Tosatti and B. Weinkove, The Monge-Ampère equation for (n − 1)-
plurisubharmonic functions on a compact Kähler manifold, J. Amer. Math. Soc.
30 no. 2 (2017), 311–346. MR 3600038. Zbl 1366.32015. https://doi.org/10.
1090/jams/875.
[47] V. Tosatti, B. Weinkove, and S.-T. Yau, Taming symplectic forms and
the Calabi-Yau equation, Proc. Lond. Math. Soc. (3) 97 no. 2 (2008), 401–424.
MR 2439667. Zbl 1153.53054. https://doi.org/10.1112/plms/pdn008.
398 BIN GUO and DUONG H. PHONG

[48] N. S. Trudinger and X.-J. Wang, Hessian measures. II, Ann. of Math. (2)
150 no. 2 (1999), 579–604. MR 1726702. Zbl 0947.35055. https://doi.org/10.
2307/121089.
[49] J. Wang, X.-J. Wang, and B. Zhou, Moser-Trudinger inequality for the com-
plex Monge-Ampère equation, J. Funct. Anal. 279 no. 12 (2020), 108765, 20.
MR 4155289. Zbl 1458.32037. https://doi.org/10.1016/j.jfa.2020.108765.
[50] B. Weinkove, The Calabi-Yau equation on almost-Kähler four-manifolds, J.
Differential Geom. 76 no. 2 (2007), 317–349. MR 2330417. Zbl 1123.32015.
https://doi.org/10.4310/jdg/1180135681.
[51] S. T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex
Monge-Ampère equation. I, Comm. Pure Appl. Math. 31 no. 3 (1978), 339–411.
MR 0480350. Zbl 0369.53059. https://doi.org/10.1002/cpa.3160310304.
(Received: February 27, 2023)
(Revised: February 14, 2024)
Department of Mathematics & Computer Science, Rutgers University,
Newark, NJ
E-mail : [email protected]

Department of Mathematics, Columbia University, New York, NY

E-mail : [email protected]