Complex Variables and Elliptic Equations

An International Journal

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/gcov20

Exponential decay of Bergman kernels on

complete Hermitian manifolds with Ricci curvature
bounded from below

Franz Berger, Gian Maria Dall'Ara & Duong Ngoc Son

To cite this article: Franz Berger, Gian Maria Dall'Ara & Duong Ngoc Son (2020) Exponential
decay of Bergman kernels on complete Hermitian manifolds with Ricci curvature
bounded from below, Complex Variables and Elliptic Equations, 65:12, 2086-2111, DOI:
10.1080/17476933.2019.1691173

To link to this article: https://doi.org/10.1080/17476933.2019.1691173

© 2019 The Author(s). Published by Informa Published online: 20 Nov 2019.

UK Limited, trading as Taylor & Francis
Group

Submit your article to this journal Article views: 892

View related articles View Crossmark data

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=gcov20
COMPLEX VARIABLES AND ELLIPTIC EQUATIONS
2020, VOL. 65, NO. 12, 2086–2111
https://doi.org/10.1080/17476933.2019.1691173

Exponential decay of Bergman kernels on complete Hermitian

manifolds with Ricci curvature bounded from below
Franz Berger , Gian Maria Dall’Ara and Duong Ngoc Son
Fakultät für Mathematik, Universität Wien, Wien, Austria

ABSTRACT ARTICLE HISTORY

Given a smooth positive measure μ on a complete Hermitian mani- Received 24 July 2019
fold with Ricci curvature bounded from below, we prove a pointwise Accepted 6 November 2019
Agmon-type bound for the corresponding Bergman kernel, under COMMUNICATED BY
rather general conditions involving the coercivity of an associated J. Leiterer
complex Laplacian on (0, 1)-forms. Thanks to an appropriate ver-
sion of the Bochner–Kodaira–Nakano basic identity, we can give KEYWORDS
explicit geometric sufficient conditions for such coercivity to hold. Bergman kernel; estimate
Our results extend several known bounds in the literature to the AMS SUBJECT
case in which the manifold is neither assumed to be Kähler nor of CLASSIFICATIONS
‘bounded geometry’. The key ingredients of our proof are a local- 32A25; 32W05
ization formula for the complex Laplacian (of the kind used in the
theory of Schrödinger operators) and a mean value inequality for
subsolutions of the heat equation on Riemannian manifolds due to Li,
Schoen, and Tam. We also show in an appendix that the ‘twisted basic
identities’, e.g. [McNeal JD and Varolin D. L2 estimates for the ∂ oper-
ator. Bull Math Sci. 2015;5(2):179–249] are standard basic identities
with respect to conformally Kähler metrics.

1. Introduction
1.1. The problem and previous results
Bergman spaces of holomorphic functions and related Bergman kernels are classical
objects of complex analysis and geometry (see, e.g. [1–3] and the references therein). If M
is a complex manifold and μ a positive Borel measure on M, the Bergman space A2 (M, μ)
is the linear space of square-integrable holomorphic functions on M, i.e.
  
A (M, μ) := f : M → C : f is holomorphic and
2 2
|f | dμ < ∞ . (1)
M

Under mild assumptions on μ, A2 (M, μ) is closed in L2 (M, μ) and actually a reproducing

kernel Hilbert space (see Section 2.1). The Bergman kernel Kμ : M × M → C is defined by
the relation

Bμ f (p) = Kμ (p, q)f (q) dμ(q), f ∈ L2 (M, μ), p ∈ M, (2)
M

CONTACT Duong Ngoc Son [email protected]

© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium,
provided the original work is properly cited.
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2087

where Bμ is the orthogonal projection of L2 (M, μ) onto A2 (M, μ).

It has been shown (see, e.g. [4–10]) that, under various assumptions on μ, one can find
a Hermitian metric h such that the following Agmon-type pointwise decay estimate holds:

e−γ d(p,q)
|Kμ (p, q)|e−ψ(p)−ψ(q) ≤ C  (p, q ∈ M). (3)
Vol(p, 1)Vol(q, 1)

Here d(p, q) is the Riemannian distance between p and q, Vol is the Riemannian volume,
Vol(p, 1) is the volume of the ball centred at p and of radius 1, and ψ is determined by
the relation μ = e−2ψ Vol. The positive constants C and γ do not depend on p and q. We
point out that Uψ f := e−ψ f is a unitary isomorphism of L2 (M, μ) onto L2 (M, Vol), and
Uψ ◦ Bμ ◦ Uψ−1 is an orthogonal projector on L2 (M, Vol). The left-hand side of (3) is thus
the modulus of the integral kernel of this projector, and the estimate shows that this kernel
exhibits an off-diagonal exponential decay, which can be neatly expressed in terms of the
metric h. Estimates of the form (3) have had numerous applications in complex analysis
and geometry (see, e.g. [8] and [11] and the references therein).
Typically, assumptions for (3) to hold can be formulated as conditions on the ‘curvature
form’ F μ of μ, which is defined as follows: in local holomorphic coordinates, one writes
dμ = ie−2ϕ dz1 ∧ dz1 ∧ · · · ∧ dzn ∧ dzn , where ϕ is smooth and real-valued. Then one
can easily check that F μ := i∂∂ϕ is a real (1, 1)-form which does not depend on the choice
of the coordinates. Thus, F μ is globally defined and we shall call it the curvature form of μ.
We now proceed to describe some of the aforementioned results in a little more detail.

(1) In the one-dimensional case M = C, μ = e−2ψ λ, where λ is Lebesgue measure and

ψ is subharmonic, the curvature form F μ may be identified with (a multiple of) the
measure ψ λ, where  is the usual Laplacian. It was shown by Christ [4] (but see
also [9]) that if F μ is doubling and satisfies

inf F μ (D(z, 1)) > 0, where D(z, r) := {z ∈ C : |z| < r}, (4)
z∈C

then (3) holds with respect to the metric h = ρ −2 | dz|2 , with

 
ρ(z) := sup r > 0 : F μ (D(z, r)) ≤ 1 . (5)

(2) In [6], Delin considers M = Cn and μ = e−2ψ λ, where ψ is strictly plurisubhar-

monic, and proves an estimate that takes the form (3) with h the Kähler metric
i∂∂ψ, at least when certain quantitative assumptions on F μ are made. These con-
ditions are not explicitly discussed by Delin (see the comment after the statement
of Theorem 2 of [6]), but it is certainly sufficient that

ic∂∂|z|2 ≤ F μ ≤ iC∂∂|z|2 (6)

holds for some 0 < c and C < +∞, as shown in [7, Proposition 9].
(3) In [5], the second-named author deals with M = Cn and μ = e−2ψ λ, where ψ
is only assumed to be weakly plurisubharmonic. More precisely, if ψ is in the
2088 F. BERGER ET AL.

reverse-Hölder class RH∞ , and

inf sup ψ(w) > 0, (7)

z∈Cn |w−z|<1

then estimate (3) holds under the condition (for some c > 0)

F μ ≥ ic ψ ∂∂|z|2 . (8)

In this case the metric is h = ρ −2 | dz|2 , where

 
ρ(z) := sup r > 0 : sup|w−z|<r ψ(w) ≤ r−2 . (9)

The condition (8) amounts to the uniform comparability of the eigenvalues of the
complex Hessian (∂zj ∂zk ψ)j,k . Notice that (6) implies (7) and (8).
(4) In [10], Schuster and Varolin take M to be the unit ball B ⊆ Cn endowed with
the Bergman metric ω := − 2i ∂∂ log(1 − |z|2 ), and prove (3) for measures μ =
e−2ψ ωn /n!, under the condition

(n + σ )ω ≤ i∂∂ψ ≤ Cω, (10)

where σ > 12 . One can see that Fμ = i∂∂ψ − (n + 1)ω in this case (by (22) below
and the fact that ω is Kähler–Einstein with Ricci curvature = −2(n + 1)ω), and
hence (10) is equivalent to
cω ≤ Fμ ≤ Cω (11)
for some c > −1/2 and C < +∞.
The result was generalized by Asserda [12] to Kähler manifolds satisfying a certain
bounded geometry assumption.
(5) In [8], Ma and Marinescu prove a pointwise Ck estimate for the Bergman kernels in
the more general setting of Hermitian line bundles over symplectic manifolds (sat-
isfying appropriate compatibility conditions). Specializing to the present situation,
Theorem 1 in that paper requires in particular that the Hermitian manifold (M, h)
has ‘bounded geometry’, and that the measure μ = e−2ψ Vol is such that

cωh ≤ i∂∂ψ ≤ Cωh (12)

(where Vol is the Riemannian volume and ωh the fundamental form) for c > 0 and
C < +∞. Then, if k > 0 is large enough, the measure μ(k) = e−2k ψ Vol satisfies
2

|Kμ(k) (p, q)|e−k

2 ψ(p)−k2 ψ(q)
≤ Ck2n e−γ kd(p,q) , (p, q ∈ M) (13)

with C independent of p, q, and k. Notice that the absence of the volume factors
in (13) is due to the bounded geometry assumption. In fact, if the volumes of balls
with a fixed positive radius is bounded away from zero (which is the case if the sec-
tional curvature is bounded from above (by [13, Theorem 3.101]) then the volume
factors can be absorbed into the constant C.

These results, despite being of the same nature, present two different points of view on
the problem of establishing exponential decay of Bergman kernels: (1)–(3) start with a
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2089

measure μ and construct a metric h with respect to which the exponential decay (3) holds,
while (4) and (5) start with a Hermitian manifold and look for conditions on the density of
μ with respect to the Riemannian volume that are sufficient for (3) to hold. Moreover, in
(1) to (3) a natural candidate for h is the Kähler metric with fundamental form F μ , but the
latter form need not be positive, and in fact (1) and (3) consider a sort of regularization of
F μ and the resulting metric is typically non-Kähler.

1.2. Our results

To state our results, we shall need to recall and fix some more notation. Let (M, h) be a
complete Hermitian manifold with a Hermitian metric h:

h = hjk dzj ⊗ dzk . (14)

The associated (1, 1)-form ωh := ihjk dzj ∧ dzk is called the fundamental form. As usual,
we refer to both h and ωh as a metric on M. As is well-known, the torsion tensor T of
the Chern connection is non-trivial if and only if the metric is Kähler: locally, T has
components

Tjk = T = h m̄ ∂j hkm̄ − ∂k hjm̄ . (15)
jk
We shall deal with the torsion 1-form, obtained by taking the trace of the torsion:
k
θ = Tjk dzj + Tjk
k
dzj , (16)

and the torsion (1, 1)-form T ◦ T defined by

p q
T ◦ T := ha hbm hqj hpk Tab T m dz
j
∧ dzk . (17)

The Riemannian metric g := 2Reh induces a distance dh and a volume Vol. We denote
by Vol(p, R) the volume of the metric ball B(p, R) := {q ∈ M : d(p, q) ≤ R} of radius R cen-
tred at p. If the Levi-Civita connection of (M, g) is denoted by ∇, then the Riemannian
curvature tensor is given by

R(Xp , Yp , Zp , Wp ) = g ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y] Z, W ,
p

where X, Y, Z, W are smooth vector fields on M and the subscript indicates evaluation at a
point p. If {ek : k = 1, 2, . . . , 2n} is a local frame of TM, then the (Riemannian) Ricci tensor
of (M, g) is defined by
2n
Ric(X, Y) := R(ek , X, Y, ek ).
k=1
We say that (M, h) has Ricci curvature bounded from below if the (Riemannian) Ricci
tensor of (M, g = 2Reh) satisfies
Ric(X, Y)  Kg(X, Y)
for some constant K > −∞.
2090 F. BERGER ET AL.

Since a Hermitian metric h induces inner products for tensors of all ranks, we can con-
sider the space L20,q (M, h, μ) of square-integrable (0, q)-forms on M, with inner product
given by

(u, v) → u, vh dμ. (18)
M
∗
We denote by ∂ h,μthe Hilbert space adjoint of (the weak extension of) ∂ with respect to
this inner product, and define the complex Laplacian associated to μ and h by
∗ ∗
h,μ := ∂∂ h,μ + ∂ h,μ ∂. (19)

This is an unbounded self-adjoint and nonnegative operator that encapsulates the interac-
tion between μ and h. In this paper, we only consider h,μ acting on (0, 1)-forms. We say
that h,μ is b2 -coercive (b > 0) if h,μ ≥ b2 in the sense of quadratic forms. We refer to
Section 2.2 for precise definitions. We are finally in a position to state our main result.

Theorem 1.1: Let (M, h) be a complete Hermitian manifold with (Levi-Civita) Ricci curva-
ture bounded from below. Assume that μ = e−2ψ Vol satisfies the following properties:

(i) h,μ is b2 -coercive for some b > 0.

(ii) trωh (i∂∂ψ) + 18 |θ |2h ≤ B < +∞.
√
Then the Bergman kernel Kμ satisfies the following estimate for every γ < 2b :

e−γ d(p,q)
|Kμ (p, q)|e−ψ(p)−ψ(q) ≤ C  , (p, q ∈ M), (20)
Vol(p, 1)Vol(q, 1)

where C depends only on γ , b, B, and the bound on the Ricci curvature.

Moreover, the coercivity condition (i) holds if the curvature form F μ satisfies

μ 2 σ
F ≥ σ b ωh + i T ◦ T. (21)
2σ − 1

for some σ > 12 . If T = 0, the conclusion still holds under the condition F μ ≥ 12 b2 ωh .

To compare this result with existing ones in the literature (e.g. [7,8,10]), it is use-
ful to reformulate (21) in terms of the Chern–Ricci form h and i∂ ∂ψ. ¯ Indeed, since
= −i∂∂ log det(h ), it follows that if μ = e −2ψ Vol, then
h jk

F μ = i∂∂ψ + 1
2 h. (22)

Hence, (21) is equivalent to


1 2 σ
i∂∂ψ + h ≥ σ b ωh + i T ◦ T. (23)
2 2σ − 1
Observe that the assumptions in Theorem 1.1 are much simplified if h is Kähler. Indeed,
if (M, h) is Kähler then T = 0 and hence (21) reduces to i∂∂ψ + 12 h ≥ b2 ωh . On the
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2091

other hand, since θ = 0 and the Chern–Ricci form is bounded from below, condition
(ii) is implied by the assumption that F μ = i∂∂ψ + 12 h ≤ Bωh < +∞. We obtain the
following Corollary which is new already in this special (Kähler) case.

Corollary 1.2: Let (M, h) be a complete Kähler manifold with Ricci curvature bounded from
below. Assume that μ satisfies
≤ F μ ≤ Bωh ,
1 2
2 b ωh (24)
√
for some b > 0 and B < +∞, then (20) holds for γ < 2b.

After a preprint of this work was made public, the authors were informed by Shoo Seto
that a result similar to Corollary 1.2, in the special case of polarized Kähler manifolds, was
obtained in collaboration with Lu, and appeared in his thesis [14].
It is worth noticing that under the assumptions of Corollary 1.2, F μ is the fundamental
form of a metric hμ that is ‘comparable’ to h, and estimate (20) also holds with respect to
hμ (with a possibly different constant γ ).
Also note that when h is the flat metric on Cn , the condition (24) is equivalent to (6),
which is considered by Lindholm [7].
For the Bergman metric on the unit ball, h = −2(n + 1)ωh , and thus (24) reduces to
i∂∂ψ ≥ (n + 1 + b2 )ωh , which is stronger than the assumption in Theorem 1.1 of [10].
If (M, h) has bounded geometry in the sense of [8], then (23) holds if ψ is replaced by
k2 ψ for k large enough, provided that i∂∂ψ ≥ ωh for some  > 0, and hence the esti-
mate hold for μ(k) := e−2k ψ Vol. In Corollary 1.3 below, we state precisely the geometric
2

conditions for (20) to hold for μ(k) .

Corollary 1.3: Let (M, h) be a complete Hermitian manifold with Ricci curvature bounded
from below. Suppose that there exist η > 1, Q ≥ 0, and P ∈ R such that

|θ |2h ≤ Q, and h − iη T ◦ T ≥ Pωh . (25)

Suppose further that ψ satisfies

1 2
b ωh ≤ i∂∂ψ ≤ Bωh , (26)
2
√
where b > 0 and B < +∞. Put μ(k) := e−2k ψ Vol, k > 0. If γ < b 2(η − 1)/η and k is
2

large enough (depending on γ , P), then the Bergman kernel Kμ(k) satisfies the following
estimate:

k2n e−γ kd(p,q)

|Kμ(k) (p, q)|e−k
2 ψ(p)−k2 ψ(q)
≤ C (p, q ∈ M), (27)
Vol(p, 1)Vol(q, 1)

where C depends only on γ , b, B, Q, and K.

√inequalities in (25) trivially hold for η > 0 arbitrary

It is clear that if h is Kähler, then the
large, so that (27) holds for any γ < 2b (if k is large enough).
2092 F. BERGER ET AL.

1.3. Structure of the paper and main ingredients of the proof

After discussing some generalities about Bergman spaces and complex Laplacians in
Section 2.1, we start to present the ingredients of the proof of Theorem 1.1.
As a first step, we establish the following exponential decay of canonical solutions of the
∂-equation, which could be of independent interest.

Theorem 1.4: Let (M, h) be a complete Hermitian manifold and assume that the smooth
positive measure μ is such that h,μ is b2 -coercive for some b > 0. Let u ∈ L20,1 (M, h, μ) be
supported on the geodesic ball B(p, R) and ∂-closed (i.e. assume that ∂u = 0) and put f :=
∗
∂ h,μ −1
h,μ u. √
Then for every q ∈ M and γ < 2 2b, the following bound holds:
 
2 −γ d(p,q)
|f | dμ ≤ Ce |u|2h dμ, (28)
B(q,R) B(p,R)

where C depends only on γ , b, and R.

If in addition (M, h) has Ricci curvature bounded from below by K with K ≤ 0, and μ =
e−2ψ Vol satisfies the condition

trωh (i∂∂ψ) + 18 |θ|2h ≤ B < +∞, (29)

√
then we have the pointwise bound (γ < 2 2b as above)

C
|f (q)|2 e−2ψ(q) ≤ e−γ d(p,q) |u|2h dμ, (30)
Vol(q, R) B(p,R)
√
where C depends only on γ , b, BR2 , and R −K.

Notice that the function f of the statement is the solution of the equation ∂f = u with
minimal L2 (M, μ) norm (see Section 2.2 below), that is, the so-called canonical solution.
The first half of Theorem 1.4 states that, under the sole geometric assumption of com-
∗
pleteness of (M, h), coercivity of h,μ implies the L2 exponential decay (28) of ∂ −1 h,μ u
off the support of u. Its proof occupies Section 3 and is based on a method developed by
Agmon to establish exponential decay of eigenfunctions of Schrödinger operators (see,
e.g. [15]). The key observation is that h,μ satisfies a localization formula analogous
to the simple yet very effective IMS localization formula of Schrödinger operators (see
Section 3.1).
In a second step, accomplished in Section 4, we improve the L2 decay to an L∞ decay,
exploiting a mean value inequality for nonnegative subsolutions of the heat equation on
Riemannian manifolds due to Li and Tam [16] (but see also [17]), which holds under a
lower bound on the Ricci curvature. To apply this inequality in the Hermitian context, we
need to control the difference between the Laplacian of the background Riemannian metric
and the Laplacian of the Chern connection, which may be expressed in terms of the torsion
and ultimately leads to condition (29). Thanks to this mean value inequality, we can avoid
the ‘Kerzman trick’ (as in [6] and [5]) and the ‘pluriharmonic recentering of the weight’
techniques (as in [10]). These methods are difficult to implement on manifolds without
some sort of ‘bounded geometry’ assumptions.
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2093

The analysis just sketched has a conditional nature, resting on the assumption that h,μ
is coercive (condition (1) in Theorem 1.1). This hypothesis is made more transparent by a
‘basic identity with torsion term’ (Proposition 5.2), thanks to which we can give a sufficient
condition for coercivity that involves only the geometry of the Hermitian metric and the
curvature form of the measure (inequality (21)). As evidence of the interest of basic iden-
tities involving a torsion term, we show in an Appendix that the ‘twisted basic identities’ of
the kind discussed, e.g. in Section 3 of [18], can be thought of as ‘standard’ basic identities
with respect to conformally Kähler metrics.
The last two sections of the paper (Section 6 and Section 7) contain the deduction
of Theorem 1.1 and Corollary 1.3 from Theorem 1.4, and a discussion of the interesting
example of asymptotically complex hyperbolic metrics of Bergman-type.
As a final remark, let us point out that Bergman kernels can be fruitfully defined in the
more general setting where holomorphic functions are replaced by holomorphic sections
of a holomorphic line bundle on M endowed with a Hermitian metric (see [3] for a com-
prehensive treatment of this matter). Most of our techniques work in this more general
framework, but we confine ourselves to the scalar setting for the sake of simplicity.

1.4. Further directions

While the pointwise condition (21) is easy to check and sufficient to prove some interesting
results, coercivity of h,μ is expected to hold under much weaker conditions (cf. [19]). This
is mainly due to the fact that, in loose terms, h,μ is a generalized Schrödinger operator, as
made apparent by the basic identity of Proposition 5.2. Condition (21) is morally a uniform
positive lower bound on the ‘potential’ of h,μ , while coercivity amounts to positivity of
the minimal eigenvalue: in the case of ordinary Schrödinger operators it is well-known that
the latter condition is much weaker (see, e.g. [20]). This idea has an antecedent in [4] and is
considered in [19], but, to the authors’ knowledge, has never been explored in the general
context of Hermitian manifolds (but see Theorem 3 of [21] for a Riemannian counterpart).
We also believe that a better analytical understanding of the quadratic form of h,μ
would allow an improvement of Corollary 1.2 in the same vein as the result of [10] for the
unit ball (see the comment after Corollary 1.2).

2. Preliminaries on Bergman kernels and the complex Laplacian on

Hermitian manifolds with measure
2.1. Bergman spaces and Bergman kernels
We recall that in the rather general setting of a complex manifold M equipped with a
positive Borel measure μ, one may consider the Bergman space
  
A (M, μ) := f : M → C : f is holomorphic and
2 2
|f | dμ < ∞ , (31)
M

which is a linear subspace of L2 (M, μ). While in complete generality this is not the case,
for many kind of measures the evaluation maps f → f (p) are locally uniformly bounded
linear functionals on A2 (M, μ), i.e. for every compact K ⊆ M there is C(K) < +∞ such
2094 F. BERGER ET AL.

that

|f (p)|2 ≤ C(K) |f |2 dμ ∀f ∈ A2 (M, μ), ∀p ∈ K. (32)
M
This condition is sometimes called admissibility of the measure μ (see, e.g. [22] and [23]).
In this paper we restrict our attention to smooth positive measures, that is, measures having
smooth positive density with respect to Lebesgue measure in local coordinates. It is a simple
consequence of the mean value property of holomorphic functions (in local holomorphic
coordinates) that such measures always satisfy the admissibility condition (32). In any case,
under assumption (32), the Bergman space is closed in L2 (M, μ), so that the associated
orthogonal projector
Bμ : L2 (M, μ) → A2 (M, μ), (33)
is well-defined, and in fact A2 (M, μ) is a reproducing kernel Hilbert space. Explicitly, there
is a function
Kμ : M × M → C, (34)
which we call the Bergman kernel, that satisfies the following properties:

(i) Kμ (·, q) ∈ A2 (M, μ) for every q ∈ M and

 
Kμ (·, q)2L2 (μ) = Kμ (q, q) = sup |f (q)|2 : f ∈ A2 (M, μ) and f  ≤ 1 ; (35)

(ii) Kμ (p, q) = Kμ (q, p);

(iii) Kμ is the integral kernel of Bμ :

Bμ f (p) = Kμ (p, q)f (q) dμ(q). ∀f ∈ L2 (M, μ), ∀ p ∈ M. (36)
M

Moreover, the following Cauchy–Schwarz type inequality holds:



|Kμ (p, q)| ≤ |Kμ (p, p )Kμ (p , q)| dμ(p ) ≤ Kμ (p, p) Kμ (q, q) ∀ p, q ∈ M. (37)
M
For proofs of these properties, see for instance [1,22].

2.2. The complex Laplacian h,μ and its coercivity

A Hermitian manifold is a complex manifold M endowed with a Hermitian metric h =
hjk dzj ⊗ dzk , where hjk = hkj = hjk . The associated real (1, 1)-form ωh := ihjk dzj ∧ dzk
is called the fundamental form (or Kähler form) of h. As usual, we refer to both h and ωh as
a metric on M. A Hermitian scalar product ·, ·h is induced in the usual way on cotangent
spaces: in particular, if u = uj dzj and v = vj dzj are (0, 1)-forms, then u, vh = hjk uj vk ,
where [hjk ] is the inverse matrix of [hjk ], and vk := vk . This Hermitian scalar product can
be extended to tensors of all ranks and our convention for the case of covariant tensors of
rank 2 is that
1
η1 ⊗ η2 , η3 ⊗ η4 h = η1 , η3 h η2 , η4 h , (38)
2
whenever the ηk ’s are 1-forms. The associated norms will be denoted by | · |h .
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2095

We identify differential forms with alternating tensors in such a way that η1 ∧ η2 :=

η1 ⊗ η2 − η2 ⊗ η1 , when η1 and η2 are 1-forms. With this definition, if u and v are (0, 1)-
forms, we have
|u ∧ v|2h + | u, vh |2 = |u|2h |v|2h . (39)
Suppose μ is a smooth positive measure on M (we point out that most of the
facts discussed below hold under much weaker regularity assumptions). We can define
L2(0,q) (M, h, μ) as the Hilbert space of square-integrable (0, q)-forms with respect to μ
and h. More explicitly, if u and v are (0, q)-forms, the scalar product on L2(0,q) (M, h, μ)

has the expression M u, vh dμ anticipated in Section 1. We restrict our attention to q ≤ 2
(recall convention (38)). Observe that L2(0,0) (M, h, μ) = L2 (M, μ) is the standard L2 -space
of C-valued functions, defined with respect to the measure μ.
We define
 
domq (∂) := u ∈ L2(0,q) (M, h, μ) : ∂u ∈ L2(0,q+1) (M, h, μ) , (40)

where the ∂ in the formula above is to be taken in the sense of distributions (or, more pre-
cisely, currents). It is clear that ∂ defines an unbounded operator mapping L2(0,q) (M, h, μ)
into L2(0,q+1) (M, h, μ), whose domain is domq (∂). This is called the weak extension of the
differential operator ∂. We skip any reference in the notation to the degree of forms on
which ∂ acts, since this should always be clear from the context. Putting all the operators
together, we get a weighted ∂-complex on (M, h):

∂ ∂ ∂
L2 (M, μ) −→ L2(0,1) (M, h, μ) −→ L2(0,2) (M, h, μ) −→ · · · (41)

Notice that the operators above are closed, so that (41) is a Hilbert complex in the sense of
[24] (closure follows immediately from the fact that convergence in L2(0,q) (M, h, μ) implies
convergence in the sense of currents). Thus, we have the dual complex
∗ ∗ ∗
∂ h,μ ∂ h,μ ∂ h,μ
L2 (M, μ) ←− L2(0,1) (M, h, μ) ←− L2(0,2) (M, h, μ) ←− · · · , (42)
∗
where every ∂ h,μ is the Hilbert space adjoint of the corresponding ∂. We decided to use
∗
the slightly cumbersome notation ∂ h,μ to stress the fact that not only the domains, but also
the ‘formulas’ of these first-order differential operators depend on the metric h and the
measure μ.
We are finally in a position to define the complex Laplacian:
(q) ∗ ∗
h,μ := ∂∂ h,μ + ∂ h,μ ∂ (1 ≤ q ≤ n − 1). (43)
(q)
The operator h,μ is self-adjoint and nonnegative when considered on the natural domain
(q)  ∗ ∗ ∗ 
dom(h,μ ) := u ∈ domq (∂) ∩ domq (∂ h,μ ) : ∂u ∈ domq+1 (∂ h,μ ), ∂ h,μ u ∈ domq−1 (∂) ,
(44)
∗
where we used the obvious notation for the domains of the ∂ h,μ ’s. One can analogously
(0) ∗ (n) ∗
define h,μ = ∂ h,μ ∂ and h,μ = ∂∂ h,μ . For the purposes of this paper, it is enough to
2096 F. BERGER ET AL.

consider the complex Laplacian for q = 1, and we will consequently drop the superscript,
(1)
putting h,μ := h,μ . As usual, a key role is played by the quadratic form
 
∗ ∗ v dμ,
Eh,μ (u, v) := ∂u, ∂vh dμ + ∂ h,μ u · ∂h,μ (45)
M M
∗
which is well-defined whenever u, v ∈ dom1 (∂) ∩ dom1 (∂ h,μ ) =: dom(Eh,μ ). Notice that
∗
∂ h,μ u is a scalar function, while ∂u is a (0, 2)-form. We adopt the convention that
Eh,μ (u) := Eh,μ (u, u). By definition,

Eh,μ (u, v) = h,μ u, vh dμ (46)
M
if u ∈ dom(h,μ ) and v ∈ dom(Eh,μ ).
Our first restriction on the metric h is justified by the following proposition.

Proposition 2.1: If the Hermitian metric h is complete, the space D(0,1) of smooth compactly
supported (0, 1)-forms is dense in dom(Eh,μ ) with respect to the graph norm. It is also a core
of h,μ , and the restriction of h,μ to D(0,1) is essentially self-adjoint.

Proof: See for instance [3] or Theorem 2.6 of [25]. The fact that we do not use the measure
induced by the Hermitian metric is of no consequence, since we may rewrite μ = e−2ψ Vol
and view Eh,μ as the quadratic form of the complex Laplacian on (M, h, Vol) for forms with
values in the trivial line bundle on M, with fibre metric given by e−2ψ . 

We say that h,μ is c-coercive (c > 0) if h,μ ≥ c in the sense of quadratic forms or,
equivalently, if

Eh,μ (u) ≥ c |u|2h dμ ∀u ∈ dom(Eh,μ ). (47)
M
In view of Proposition 2.1, it is enough that the inequality above holds for u ∈ D(0,1) . By
standard functional analysis, whenever h,μ is c-coercive there exists a bounded inverse
−1
h,μ with domain L(0,1) (M, h, μ) and range dom(h,μ ). The operator norm of h,μ is
2 −1

bounded from above by c−1 . Moreover, under assumption (47), the ∂-equation
∂f = u (48)
admits a unique solution orthogonal to the Bergman space A2 (M, μ), whenever the datum
u is in L2(0,1) (M, h, μ) and ∂-closed, i.e. ∂u = 0. This solution may be expressed as
∗
f = ∂ h,μ −1
h,μ u, (49)
and satisfies the bound  
2 −1
|f | dμ ≤ c |u|2h dμ. (50)
M M
For our purposes, the most important consequence of this formula is the well-known
Kohn’s identity for the Bergman projection:
∗
Bμ (f ) = f − ∂ h,μ −1
h,μ ∂f for all f ∈ dom0 (∂). (51)
Notice that while the terms appearing on the right-hand side of this identity depend on
the metric h, the left-hand side depends only on μ. It is this asymmetry that gives us the
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2097

freedom to choose, given μ, the most appropriate metric, e.g. one that makes h,μ coercive
(if it exists).
See, e.g. [26,27] for proofs of the well-known facts just discussed.

¯
3. L2 Exponential decay of canonical solutions of the ∂-equation
The goal of this section is to prove the first half of Theorem 1.4, that is, (3.4) below. In order
to do that, we need a localization lemma and a Caccioppoli-type inequality.

3.1. A localization formula for h,μ

Lemma 3.2 below is a localization formula for h,μ that is analogous to the very useful
IMS localization formula in the theory of Schrödinger operators. For the latter see, e.g.
Lemma 3.1 of [28] or Lemma 11.3 of [29]. Before stating and proving it, we need a few
preliminaries.
First, notice that if Lip(M, h) is the class of scalar functions χ : M → R that are Lips-
chitz with respect to the Riemannian distance, then by Rademacher’s theorem, χ is almost
everywhere differentiable and, by our convention g = 2Reh,

|∂χ |2h = 12 | dχ |2g ≤ 12 L2 , (52)

where L is the Lipschitz constant of χ.

∗
Next, we state the Leibniz rule for ∂ h,μ for future reference. For this, we employ the
notation w ∨ v for the interior product of the forms v and w (with respect to h). This is the
form defined by the condition

w ∨ v, uh = v, w̄ ∧ uh , (53)

where u is an arbitrary form. Observe that the conjugation on the right hand side makes
the interior product bilinear. The following lemma is well-known (see, e.g. [27], page 11).
We include a short proof for the reader’s convenience.

∗
Lemma 3.1: If χ ∈ Lip(M, h) ∩ L∞ (M) and v ∈ domq (∂ h,μ ) (1 ≤ q ≤ n), then χ v ∈
∗
domq (∂ h,μ ) and
∗ ∗
∂ h,μ (χ v) = χ ∂ h,μ v − ∂χ ∨ v. (54)

Proof: Let u ∈ domq−1 (∂). Then ∂(χu) = χ∂u + ∂χ ∧ u and the remark we made about
the differentiability of Lipschitz functions implies immediately that χ u ∈ domq−1 (∂).
Hence,
   
∗
u, χ ∂ h,μ vh dμ = ∂(χu), vh dμ = ∂u, χ vh dμ + u, ∂χ ∨ vh dμ,
M M M M
(55)
which gives the thesis. 
2098 F. BERGER ET AL.

Lemma 3.2 (Localization formula): If u ∈ dom(h,μ ) and χ ∈ Lip(M, h) ∩ L∞ (M),

then χu ∈ dom(Eh,μ ) and the following identity holds:
 
Eh,μ (χ u) = Re h,μ u, χ 2 uh dμ + |∂χ |2h |u|2h dμ. (56)
M M

Proof: Exactly as in the proof of Lemma 3.1 of [28], we compute in two ways the iterated
commutator [χ, [χ, h,μ ]], where χ is identified with a multiplication operator. We will
use (54) a few times without comment. All the computations below are for u and χ smooth
and compactly supported, the statement then follows appealing to Proposition 2.1. We have
 ∗  ∗ ∗
χ, ∂∂ h,μ u = χ ∂∂ h,μ u − ∂∂ h,μ (χ u)
∗ ∗
= χ ∂∂ h,μ u − ∂(χ ∂ h,μ u − ∂χ ∨ u)
∗
= ∂(∂χ ∨ u) − ∂χ ∧ ∂ h,μ u. (57)

Thus,
  ∗   ∗
χ , χ, ∂∂ h,μ u = −2∂χ ∧ ∂χ ∨ ∂ h,μ u . (58)
Analogously, we get
 ∗  ∗ ∗
χ, ∂ h,μ ∂ u = χ ∂ h,μ ∂u − ∂ h,μ ∂(χ u)
∗ ∗
= χ ∂ h,μ ∂u − ∂ h,μ (∂χ ∧ u + χ ∂u)
∗
= −∂ h,μ (∂χ ∧ u) + ∂χ ∨ ∂u, (59)

and
  ∗ 
χ , χ , ∂ h,μ ∂ u = −2∂χ ∨ (∂χ ∧ u). (60)
Putting everything together, we get, for all u ∈ dom(h,μ ),

1   ∗
− χ , [χ, h,μ ] u = ∂χ ∧ ∂χ ∨ ∂ h,μ u + ∂χ ∨ (∂χ ∧ u). (61)
2
On the other hand, we can easily see that

1 χ 2 h,μ u + h,μ (χ 2 u)
− [χ , [χ, h,μ ]]u = χ h,μ (χ u) − . (62)
2 2
Combining the two identities we get

Eh,μ (χ u) = h,μ (χ u), χuh dμ
M
 

= Re h,μ u, χ uh dμ +
2
|∂χ ∨ u|2h + |∂χ ∧ u|2h dμ. (63)
M M

Then (56) follows by observing that ∂χ ∨ u = u, ∂χh and recalling (39). 

 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2099

3.2. Caccioppoli-type inequality for h,μ -harmonic (0, 1)-forms

Proposition 3.3: Let u ∈ dom(h,μ ) be such that h,μ u = 0 on a geodesic ball B(p, R).
Then, for every R < R,
 
∗ 2
|∂ h,μ u|2 dμ ≤ |u|2h dμ. (64)
B(p,R ) (R − R )2 B(p,R)

Proof: We define

χ (q) := max{1 − (R − R )−1 d(q, B(p, R )), 0}, (65)

where d and B are the geodesic distance and balls associated to h, respectively. It is easy
to see that χ ∈ Lip(M, h) ∩ L∞ (M) and that χ (q) > 0 holds exactly on B(p, R), and that
|∂χ |2h ≤ (R − R )−2 /2.
Applying the localization formula (56) to χ u one immediately gets
 
∗
|∂ h,μ (χ u)|2 dμ ≤ Eh,μ (χ u) = |∂χ|2h |u|2h dμ. (66)
M M

Recalling the Leibniz rule (54), this gives

  
∗ 2
|∂ h,μ u|2 χ 2 dμ ≤ 4 |∂χ |2h |u|2h dμ ≤ |u|2h dμ. (67)
M M (R − R )2 M

Since χ ≡ 1 on B(p, R ), we are done. 

3.3. Coercivity implies L2 exponential decay of canonical solutions

Theorem 3.4:√ Assume that h,μ is b -coercive for some b > 0, i.e. that (47) holds.
2 Then for
every γ < 2 2b and R > 0, there exists a constant Cγ ,R,b such that if u ∈ L20,1 (M, h, μ) is
∗
supported on B(p, R) and f := ∂ h,μ −1
h,μ u, then

 
|f |2 dμ ≤ Cγ ,R,b e−γ d(p,q) |u|2h dμ (68)
B(q,R) B(p,R)

holds for every q ∈ M.

Proof: By inequality (50) in Section 2.2, under the coercivity condition (47) we have
 
2 −2
|f | dμ ≤ b |u|2h dμ. (69)
M M

In particular, (68) holds for d(p, q) ≤ 4R with Cγ ,R,b ≥ e4γ R C. Thus, without loss of
generality, we may assume that d(p, q) ≥ 4R.
2100 F. BERGER ET AL.

Since u is supported on B(p, R), we see that −1

h,μ u is h,μ -harmonic on B(q, 2R). Thus,
using (3.3), we obtain
 
2
|f | dμ ≤ 2R −2
|−1 2
h,μ u|h dμ. (70)
B(q,R) B(q,2R)

We introduce the functions

 
d̃(p ) := min d(p, p ), d(p, q) , (71)

and
 
χ (p ) := min 1, R−1 d(p , B(p, R)) (72)

for p ∈ M. Notice that d, χ ∈ Lip(M, h) ∩ L∞ (M), so that we also have χ ead ∈

Lip(M, h) ∩ L∞ (M) with a > 0. Since −1
h,μ u ∈ dom(h,μ ), we can apply (3.2) to get
 
Eh,μ (χ e ad̃
−1
h,μ u) = Re u, χ e 2 2ad̃
−1
h,μ uh dμ + |∂(χ ead̃ )|2h |−1 2
h,μ u|h dμ. (73)
M M

Observe that χ was chosen to be 0 on the support of u, and hence the first term on the right
hand side vanishes. Recalling the coercivity condition (47), we obtain
 
b χ 2 e2ad̃ |−1
h,μ u|h dμ ≤
2 |∂χ |2h e2ad̃ |−1 2
h,μ u|h dμ
M M

+a |∂ d̃|2h χ 2 e2ad̃ |−1 2
h,μ u|h dμ. (74)
M

√
The pointwise bound |∂ d|2h ≤ 1/2 suggests that we choose a < 2b and reabsorb the
rightmost term. By support considerations and the bound |∂χ |2h ≤ R−2 /2, we finally get
2  
a 1
b− √ e 2ad̃
|−1 2
h,μ u|h dμ ≤ 2 e2ad̃ |−1 2
h,μ u|h dμ. (75)
2 B(q,2R) 2R B(p,2R)

Our choice of d guarantees that this function is bounded from below by d(p, q) − 2R on
B(q, 2R), and from above by 2R on B(p, 2R). Thus,
2  
a e8aR −2ad(p,q)
b− √ |−1 2
h,μ u|h dμ ≤ e |−1 2
h,μ u|h dμ. (76)
2 B(q,2R) 2R2 B(p,2R)

To complete the proof we combine (76), and (70), and the observation that −1 is bounded
−2
 −1 2  h,μ−1 2
with operator norm at most b , so that we have B(p,2R) |h,μ u|h dμ ≤ M |h,μ u|h dμ ≤

b−4 B(p,R) |u|2h dμ. 
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2101

4. From L2 to pointwise bounds

The key ingredient in the transition to pointwise bounds from the L2 -bounds of (3.4) is
the following result by Li–Schoen and Li–Tam.

Theorem 4.1: Let (M, g) be a complete Riemannian manifold, p ∈ M and R > 0 be such
that the geodesic ball B(p, 2R) does not meet the boundary of M. Suppose that the Ricci cur-
1
vature of g is bounded below by K with K ≤ 0. Let δ ∈ (0, √ 2 ), q > 0, and λ ≥ 0. Then there
exists a constant C that depends only on δ, q, λR , and R −K such that for any nonnegative
2

smooth function f on B(p, 2R) satisfying the differential inequality

g f ≥ −λf (77)
we have

C
sup fq ≤ f q d Volg . (78)
B(p,(1−δ)R) Vol(p, R) B(p,R)

This is essentially Corollary 3.6 of [30], which follows easily from the results on
subsolutions of the heat equation on Riemannian manifolds of [16].
Here, the convention is that g is nonpositive. To apply this theorem, we will need to
compare the Riemannian Laplacian g f , where g := 2Reh, with the so-called Chern Lapla-
cian trωh (i∂∂f ) of a regular function f. The comparison is well-known and is stated and
proved in Proposition 4.2 below for convenience (cf. formula (25) in [31]).

Proposition 4.2: For a smooth function f on the Hermitian manifold (M, h), one has
g f = 2trωh (i∂∂f ) + df , θ h , (79)
where θ is the torsion 1-form defined by (16).

Proof: Let ∇ denote the Levi-Civita connection of g := 2Reh. Since h is Hermitian, the
Christoffel symbols ījk in local holomorphic coordinates reduce to

1 ¯ 
ījk = hk ∂i hj ¯ − ∂ hjī . (80)
2
It then follows that
k 1 ¯ 1 ¯
īk = hk ∂i hk ¯ − ∂ hkī = Tī ¯, (81)
2 2
¯
where Tī ¯ is the torsion (0,1)-form of the Chern connection. Thus

¯ ¯ 1 ¯ k
hjk̄ jk̄ = −ik̄k̄ hi = − hi Tik . (82)
2
¯ ) = hjk̄ ∂j ∂ f and therefore
Locally, trωh (i∂ ∂f k̄

g f = 2hjk̄ ∇df jk̄
 
= 2hjk̄ ∂j ∂k̄ f − jl̄k̄ ∂l̄ f − jlk̄ ∂l f

= 2trωh (i∂∂f ) + df , θ h . 
2102 F. BERGER ET AL.

As a consequence of Theorem 4.1, and Proposition 4.2, we have the following mean
value inequality. Recall that ψ was defined to satisfy μ = e−2ψ Vol.

Lemma 4.3: Assume that the Ricci curvature is bounded from below by K ≤ 0 on B(p, 2R)
and put
 
λ := sup trωh (i∂∂ψ) + 18 |θ |2h , (83)
B(p,2R)

where θ is the torsion 1-form. If F : B(p, 2R) → C is holomorphic, then


C
|F(p)|2 e−2ψ(p) ≤ |F|2 dμ, (84)
Vol(p, R) B(p,R)
√
where the constant C depends only on λR2 and R −K.

Proof: Let f := |F|2 e−2ψ . First, observe that by the Cauchy–Schwarz inequality

df , θh = 2| (∂F − 2F∂ψ), Fθh |e−2ψ

1
≤ 2|∂F − 2F∂ψ|2h e−2ψ + f |θ |2h . (85)
2
Next, we compute

trωh (i∂∂f ) = hjk ∂j ∂k f

= hjk (∂j F − 2∂j ψF)(∂k F − 2∂k ψF)e−2ψ − 2hjk (∂j ∂k ψ)e−2ψ


= |∂F − 2F∂ψ|2h − 2|F|2 trωh (i∂∂ψ) e−2ψ
= |∂F − 2F∂ψ|2h e−2ψ − 2trωh (i∂∂ψ)f . (86)

Putting the two estimates together and exploiting (4.2), we obtain, on B(p, 2R),

g f ≥ 2trωh (i∂∂f ) − df , θ h

1 2
≥ −4 trωh (i∂∂ψ) + |θ |h f
8
≥ −4λf . (87)

This estimate, together with the lower bound on the Ricci curvature, shows that the
hypothesis of Theorem. 4.1 are satisfied. Thus,

C
f (p) ≤ f d Vol, (88)
Vol(p, R) B(p,R)
√
where C depends on λR2 and R −K, as we wanted. 

Combining Lemma 4.3, and Theorem 3.4, we immediately obtain:

 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2103

Theorem 4.4: Let (M, h) be a complete Hermitian manifold with Ricci curvature bounded
from below by −K (K ≤ 0). Let μ = e−2ψ Vol be a smooth positive measure such that h,μ
is b2 -coercive. Suppose further that

¯ + 1 |θ |2 ≤ B < +∞.
trωh (i∂ ∂ψ) (89)
8

∗
Let u ∈ L20,1 (M, h, μ) be supported on B(p, R) and ∂-closed, and put f := ∂ h,μ −1
h,μ u. For
√
every q ∈ M and γ < 2 2b, the following bound holds:

C
|f (q)|2 e−2ψ(q) ≤ e−γ d(p,q) |u|2h dμ, (90)
Vol(q, R) B(p,R)

√
where C depends on γ , b, BR2 , and R −K.

This completes the proof of Theorem 1.4.

5. The basic identity for h,μ

We devote this section to a discussion of the basic identity for h,μ , which is essentially
[32] applied to (0, 1)-forms with compact support. We provide a simple proof of this case.
We denote by ∇ the Chern connection of h. In local holomorphic coordinates, the only
nonvanishing Christoffel symbols of ∇ are jk and  = jk , where
jk

jk = hm ∂j hkm . (91)

We shall only need the (0, 1)-part of ∇, which we denote by ∇. In particular, if u = uk dzk ,
∇u is the 2-tensor
 
∇u = ∂j uk − jk u dzj ⊗ dzk . (92)

The key to our proof of the basic inequality is an elementary pointwise identity that involves
only the metric h. In order to state it, we recall the standard notation u for the vector
field associated to the 1-form u by the metric h. Notice that if u is a (0, 1)-form, then u
is a (1, 0)-vector field, and ∇u is a 2-tensor with one covariant and one contravariant
index.

Lemma 5.1: For every (0, 1)-form u = uī dzī , the following identity holds:

tr ∇u ⊗ ∇u = 2|∇u|2h − |∂u − Tu|2h , (93)

where tr(∇u ⊗ ∇u ) = (∇u )m (∇u )m and Tu = T ī uī dzj̄ ⊗ dzk̄ .

j
j j̄k̄
2104 F. BERGER ET AL.

Proof: Notice that in local coordinates u = um ∂m where um := hmk uk , and recall that
one of the defining properties of the Chern connection is that
∇ ∂j (u ) = ∂j um ∂m . (94)

It is thus clear that the trace in the statement is

 p  q
∂j um ∂m uj = hmk ∂j uk −  up hj ∂m u − m uq . (95)
jk

Now notice that if A = Ajk dzj ⊗ dzk and A = Akj dzj ⊗ dzk , a straightforward computa-
tion gives
2|A|2h − |A − Ã|2h = Ajk Am hj hkm . (96)
If A = ∇u, by (92) we have A − A = ∂u − Tu, and the identity above becomes
 p  q
2|∇u|2h − |∂u − Tu|2h = ∂j uk −  up ∂m u − m uq hj hkm . (97)
jk

In view of (95), this is the formula we set out to prove. 

Proposition 5.2: Let (M, h) be a Hermitian manifold and μ a positive smooth measure with
curvature form F μ . Then for every u ∈ D(0,1) (M),
   
2 ∗
∂u − Tu h dμ + |∂ h,μ u|2 dμ = 2 |∇u|2h dμ + 2 F μ , u ∧ uh dμ (98)
M M M M

and, for any ν > 0,

  
2 2 μ 1
Eh,μ (u) ≥ |∇u|2h dμ + F , u ∧ uh dμ − |Tu|2h dμ (99)
1+ν M 1+ν M ν M

where Eh,μ (u) is defined by (45) and Tu is defined in Lemma 5.1.

It may be of interest to remark that the right hand side of (98) has the following explicit
epression in terms of the (1, 0)-vector field u = u ∂ (and the metric h):
 
jk m
2 h m h ∂j u ∂k u dμ + 2 ∂ ∂m ϕu um dμ.
M M

If dimM = 1, the expression above does not contain explicitly the metric, making the
analysis of h,μ much simpler.

Proof: It is enough to prove the identity for u supported on a coordinate chart with coordi-
nates zj . Let ϕ be the real-valued function such that dμ = ie−2ϕ dz1 ∧ dz1 ∧ · · · ∧ dzn ∧
dzn . Then the adjoint of ∂m with respect to dμ is δm := −∂m + 2∂m ϕ. Integrating both
sides of the identity of Lemma. 5.1, the usual commutation argument yields
   
2 2 m 2
2 |∇u|h dμ − |∂u − Tu|h dμ = |δm u | dμ − 2 ∂m ∂j ϕ um uj dμ. (100)
M M M M

To complete the proof of Equation (98), one may easily check that ∂m ∂j ϕ um uj = F μ , u ∧
∗
uh and that δm um = ∂ h,μ u. The basic inequality (99) follows immediately. 
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2105

Corollary 5.3: Let (M, h) be a complete Hermitian manifold and μ a smooth positive
measure on M. Suppose that

μ 2 σ 1
F ≥ σ b ωh + i T ◦ T, b > 0 and σ > . (101)
2σ − 1 2

Then the associated complex Laplacian h,μ is b2 -coercive. If T = 0, then the conclusion still
holds under the assumption F μ ≥ 12 b2 ωh .

Proof: Observe that iT ◦ T, u ∧ uh = |Tu|2 . The statement then follows from (99) with
ν = 2σ − 1. 

6. Proof of Theorem 1.1, and Corollary 1.3

We first prove Theorem 1.1. By the hypothesis, (4.3) may be applied uniformly on geodesic
balls of radius 1. In particular,

e2ψ(p)
|F(p)|2  |F|2 dμ ∀ F ∈ A2 (μ). (102)
Vol(p, 1) B(p,1)

where the implicit constant depends only on B and the K. By Equation (35), we have

e2ψ(p)
|Kμ (p, p)|  , (103)
Vol(p, 1)
and therefore, by the Cauchy–Schwarz inequality (37),

|Kμ (p, q)|2 e−2ψ(p)−2ψ(q) ≤ |Kμ (p, p)||Kμ (q, q)|e−2ψ(p)−2ψ(q)

1
 . (104)
Vol(p, 1)Vol(q, 1)
To prove the off-diagonal exponential decay (20), we can assume without loss of generality
that d(p, q) ≥ 4. Applying (102) to Kμ (·, q) we get:

e2ψ(p)
|Kμ (p, q)|2  |Kμ (p , q)|2 dμ(p ). (105)
Vol(p, 1) B(p,1)

Observe that if χp (p ) := max{0, 1 − d(B(p, 1), p )}, the definition of Bμ and Kohn’s iden-
tity Equation (51) give
 
 
2
|Kμ (p , q)| dμ(p ) ≤ Kμ (q, p )Kμ (p , q)χp (p ) dμ(p )
B(p,1) M

= Bμ (Kμ (·, q)χp )(q)

∗
= Kμ (q, q)χp (q) − (∂ h,μ −1
h,μ u)(q), (106)

where u = ∂(Kμ (·, q)χp ) = Kμ (·, q)∂χp is a (0, 1)-form supported on B(p, 2) (in fact
on B(p, 2) \ B(p, 1)). Since d(p, q) ≥ 4, the first term vanishes. The second one may be
2106 F. BERGER ET AL.

bounded with (4.4), which gives


∗ eψ(q)
(∂ h,μ −1
h,μ u)(q)  e −γ d(p,q)
|Kμ (·, q)|2 |∂χp |2h dμ, (107)
Vol(q, 2) B(p,2)
√
where γ < 2b. Notice that by taking the square root we lose a factor 2, with respect to
(4.4). Since |∂χp |2h ≤ 1/2, we finally estimate
 
  2 
|Kμ (p , q)| |∂χp (p )|h dμ(p )  |Kμ (q, q)|
2
|Kμ (p , p )| dμ(p )
B(p,2) B(p,2)

e2ψ(q) d Vol(p )
 , (108)
Vol(q, 1) B(p,2) Vol(p , 1)

where we used the diagonal bound (103). By the Bishop–Gromov volume compari-
son theorem [13, Theorem 4.19], Vol(p , 1) ≈ Vol(p, 2) for every p ∈ B(p, 2), where the
implicit constant depends just on the dimension of M and the lower bound on the Ricci
curvature. Thus

d Vol(p )

≈ 1. (109)
B(p,2) Vol(p , 1)
Analogously, Vol(q, 2) ≈ Vol(q, 1), and therefore

∗ e2ψ(q) −γ d(p,q)
|(∂ h,μ −1
h,μ u)(q)|  e , (110)
Vol(q, 1)
which, together with the estimates obtained above, gives the desired estimate. The last part
of the theorem follows from Corollary 5.3.
We now√ turn to the proof of Corollary 1.3. By assumption, we can take b < b such that
γ < 2b (η − 1)/η. Choose k large enough such that k2 (b2 − b2 ) + P ≥ 0 so that

k2 (b2 − b2 )ωh ≥ iη Th ◦ T h − h. (111)

Putting h(k) := k2 h, by direct calculations we get

¯ +1 h
F μ(k) = k2 i∂ ∂ψ
2
1 2 2 η
≥ k b ωh + iTh ◦ T h
2 2
1 η
= b2 ωh(k) + iTh(k) ◦ T h(k) . (112)
2 2
By Corollary 5.3, the complex Laplacian h(k) ,μ(k) is b2 -coercive with b2 = b2 (η − 1)/η
√
and Theorem 1.1 holds with γ < b 2(η − 1)/η = 2b. On the other hand,
(k)
μ(k) = e−2k ψ Volh = e−2k ψ k−2n Volh(k) = e−2ψ Volh(k)
2 2
(113)

with ψ (k) = k2 ψ + 2n log k. Therefore,

¯ (k) ) + 1 |θ |2(k) ≤ nB + 1 k−2 Q.
trωh(k) (i∂ ∂ψ (114)
8 h 8
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2107

Applying Theorem 1.1 for μ(k) and h(k) , we obtain for p, q ∈ M,

e−γ dh(k) (p,q)

|Kμ(k) (p, q)|e−k
2 ψ(p)−k2 ψ(q)
≤ C
Volh(k) (p, 1)Volh(k) (q, 1)
e−γ kdh (p,q)
= C , (115)
Volh (p, k−1 )Volh (q, k−1 )

where C does not depend on k. Finally, observe that by the assumption on the lower bound
of the Ricci tensor,
√
Volh (p, 1) ≤ Volh (p, k−1 )k2n e −K
, k≥1

and similarly for q. Plugging these into the inequality above, we finally obtain (27).

7. An example: ACH metrics of Bergman-type

In this last section, we discuss in some detail the following example. Let D be a precompact
strictly pseudoconvex domain in a complex manifold X with smooth boundary. Suppose
that D is defined by  < 0, with d = 0 on ∂D and  is smooth in a neighbourhood U of
∂D. We further assume that − log(−) is strictly plurisubharmonic on U ∩ D. In this case,
−i∂ ∂¯ log(−) defines an asymptotically complex hyperbolic (ACH) Kähler metric h on
U ∩ D.
Given any Hermitian metric h on D, we can patch, using a partition of unity, h and h to
obtain a Hermitian metric h on D such that h = h on U ∩ D. It is known that the curvature
tensor of h approaches the curvature tensor of constant holomorphic sectional curvature
−4 (see [33]). In particular, the sectional curvature is bounded from above, while Rich and
h are bounded from below. The last fact is easy to see: near the boundary ∂D, in local
coordinates

Rich = −(n + 1)ωφ − i∂ ∂¯ log J[]

where J[] is the (Levi–)Fefferman determinant:

 
 j̄
J[] = − det .
k kj̄

Notice that i∂ ∂¯ log J[], which does not depend on the local coordinates, extends smoothly
to a neighbourhood of ∂D, and is hence bounded. Moreover, since Th = 0 near the
boundary, h must have bounded torsion.
Also note that in general the metric h constructed in this way is non-Kähler and need
not have bounded geometry.
Suppose that μ is a smooth measure on D such that h,μ is b2 -coercive and with
h log(d Volh /dμ) bounded from above. Then the Bergman kernel Kμ satisfies the
2108 F. BERGER ET AL.

exponential decay estimate (20), namely

C
|Kμ (p, q)| ≤ e−γ dh (p,q) ,
η(p)η(q)

where η = d Volh /dμ, dh is the Riemannian distance of h, and γ depends on the coercivity
constant b. Observe that the volume factors have been absorbed into the constant since h
has sectional curvature bounded from above.
¯ > ωh for some  > 0, then for k large enough
Moreover, by Corollary 1.3, if i∂ ∂η

Ck2n
|Kηk2 dVol (p, q)| ≤ e−γ kd(p,q) ,
h ηk (p)ηk (q)

for some constants C and γ do not depend on k.

When h = −i∂ ∂¯ log(−) is defined on D, the coercivity is satisfied under an explicit
condition on μ and , i.e. when

i∂ ∂¯ log (dvolh /dμ) ≥ i∂ ∂¯ log((−)−n−1−b J[]).

If D ⊂ Cn and dμ = e−ϕ dλ, then the condition F μ ≥ bωφ translates into

i∂ ∂¯ log[(−)b eϕ ] ≥ 0,

in other words, when log[(−)b eϕ ] is strictly plurisubharmonic for some positive con-
stant b.

Acknowledgements
The authors would like to thank Jeffery McNeal for an interesting discussion about the subject of
this paper.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by the Austrian Science Fund (FWF) [grant numbers I01776, P28154].

ORCID
Franz Berger http://orcid.org/0000-0002-5845-9055
Gian Maria Dall’Ara http://orcid.org/0000-0003-1168-3815
Duong Ngoc Son http://orcid.org/0000-0002-8437-8056

References
[1] Bergman S. The kernel function and conformal mapping. Revised ed. Providence (RI): Amer-
ican Mathematical Society; 1970. (Mathematical Surveys; V).
[2] Krantz SG. Geometric analysis of the Bergman kernel and metric. Vol. 268. New York: Springer;
2013. (Graduate Texts in Mathematics).
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2109

[3] Ma X. Marinescu G. Holomorphic Morse inequalities and Bergman kernels. Vol. 254. Basel:
Birkhäuser; 2007. (Progress in Mathematics).
[4] Christ M. On the ∂ equation in weighted L2 norms in C1 . J Geom Anal. 1991;1(3):193–230.
[5] Maria Dall’Ara G. Pointwise estimates of weighted Bergman kernels in several complex
variables. Adv Math. 2015;285:1706–1740.
[6] Delin H. Pointwise estimates for the weighted Bergman projection kernel in Cn , using a
weighted L2 estimate for the ∂ equation. Ann Inst Fourier (Grenoble). 1998;48(4):967–997.
[7] Lindholm N. Sampling in weighted Lp spaces of entire functions in Cn and estimates of the
Bergman kernel. J Funct Anal. 2001;182(2):390–426.
[8] Ma X, Marinescu G. Exponential estimate for the asymptotics of Bergman kernels. Math Ann.
2015;362(34):1327–1347.
[9] Marzo J, Ortega-Cerdà J. Pointwise estimates for the Bergman kernel of the weighted Fock
space. J Geom Anal. 2009;19(4):890–910.
[10] Schuster AP, Varolin D. New estimates for the minimal L2 solution of ∂ and applications
to geometric function theory in weighted Bergman spaces. J Reine Angew Math. 2014;691:
173–201.
[11] Lu Z, Zelditch S. Szegő kernels and Poincaré series. J Anal Math. 2016;130:167–184.
[12] Asserda S. Pointwise estimate for the bergman kernel of holomorphic line bundles. Palestine J
Math. 2017;6(1):6–14.
[13] Gallot S, Hulin D, Lafontaine J. Riemannian geometry. 3rd ed.. Berlin: Springer; 2004.
[14] Seto S. On the asymptotic expansion of the bergman kernel [PhD thesis]. Irvine: University of
California; 2015.
[15] Agmon S. Lectures on exponential decay of solutions of second-order elliptic equations:
bounds on eigenfunctions of N-body Schrödinger operators. Mathematical Notes. Vol. 29.
Princeton (NJ): Princeton University Press; 1982.
[16] Li P, Tam L-F. The heat equation and harmonic maps of complete manifolds. Invent Math.
1991;105(1):1–46.
[17] Li P, Schoen R. Lp and mean value properties of subharmonic functions on Riemannian
manifolds. Acta Math. 1984;153(3–4):279–301.
[18] McNeal JD, Varolin D. L2 estimates for the ∂ operator. Bull Math Sci. 2015;5(2):179–249.
[19] Maria Dall’Ara G. Coercivity of weighted Kohn Laplacians: the case of model monomial
weights in C2 . Trans Amer Math Soc. 2017;369(7):4763–4786.
[20] Maz’ya V, Shubin M. Discreteness of spectrum and positivity criteria for Schrödinger operators.
Ann Math. 2005;162(2):919–942.
[21] Devyver B. A Gaussian estimate for the heat kernel on differential forms and application to the
Riesz transform. Math Ann. 2014;358(12):25–68.
[22] Pasternak-Winiarski Z. On the dependence of the reproducing kernel on the weight of
integration. J Funct Anal. 1990;94(1):110–134.
[23] Zeytuncu YE. Lp regularity of weighted Bergman projections. Trans Amer Math Soc.
2013;365(6):2959–2976.
[24] Brüning J, Lesch M. Hilbert complexes. J Funct Anal. 1992;108(1):88–132.
[25] Ohsawa T. L2 approaches in several complex variables. Tokyo: Springer Monographs in
Mathematics; 2015.
[26] Berger F. Essential spectra of tensor product Hilbert complexes and the ∂-Neumann problem
on product manifolds. J Funct Anal. 2016;271(6):1434–1461.
[27] Straube EJ. Lectures on the L2 -Sobolev theory of the ∂-Neumann problem. ESI Lectures in
Mathematics and Physics. Zürich: European Mathematical Society (EMS); 2010.
[28] Simon B. Semiclassical analysis of low lying eigenvalues. I. Nondegenerate minima: asymptotic
expansions. Ann Inst H. Poincaré Sect. A (NS). 1983;38(3):295–308.
[29] Teschl G. Mathematical methods in quantum mechanics. 2nd ed., Graduate Studies in Mathe-
matics. Vol. 157. Providence (RI): American Mathematical Society; 2014. With applications to
Schrödinger operators.
[30] Ji L, Li P, Schoen R, et al. Handbook of geometric analysis. No. 1. Advanced Lectures in
Mathematics (ALM). Vol. 7. Somerville (MA): International Press; 2008.
2110 F. BERGER ET AL.

[31] Gauduchon P. La 1-forme de torsion d’une variété hermitienne compacte. Math Ann.
1984;267(4):495–518.
[32] Griffiths PA. The extension problem in complex analysis. II. Embeddings with positive normal
bundle. Amer J Math. 1966;88:366–446.
[33] Klembeck PF. Kähler metrics of negative curvature, the Bergmann metric near the boundary,
and the Kobayashi metric on smooth bounded strictly pseudoconvex sets. Indiana Univ Math
J. 1978;27(2):275–282.

Appendix
We now show that the ‘twisted basic identities’ of the kind discussed in, e.g. [18] are particular
instances of Proposition 5.2 when the metric is conformally Kähler.
To see this, let (M, h) be a Kähler manifold and τ : M → (0, +∞) a smooth ‘twisting factor’.
One may easily verify that the diagram below is an isomorphism of Hilbert complexes, i.e. that the
vertical arrows are unitary isomorphisms and the two squares commute:

Here the vertical arrows are multiplication operators by the indicated functions. Therefore, the
twisted complex in the top row is isomorphic to the weighted ∂-complex in the second row. As a
consequence, putting h := τ −1 h and μ := τ −1 μ, we have
 
∗
Eh,μ (u) = τ |∂ h,μ u|2 dμ + τ |∂u|2h dμ.
M M
We now compute the left-hand side using Proposition 5.2. Let η := − log τ and observe that the
Christoffel symbols of the Chern connection of h are jk
i =  i + η δ i , where  i are the Christoffel
jk j k jk
h h
symbols of the Chern connection of h and ηj is a shorthand for ∂j η. Hence, we have ∇ j̄ uk̄ = ∇ j̄ uk̄ −
ηj̄ uk̄ . Representing covariant derivatives with respect to ∇ h by indices preceeded by a vertical bar |,
we have
1 ¯ 2 2  
h 2 h 2 j̄ k̄
|∇ u|h = |∇ u|h + |∂η|h |u|h − Re uk̄|j̄ η u .
2
i = η δ i − η δ i , we have T i u = η u − η u and
Moreover, since the torsion of h is given by Tjk j k k j jk i j k k j

i
|Tjk ¯ 2 |u|2 − | u, ∂η
ui |2h = |∂η| ¯ h |2 .
h h

Equation (98) yields

 
h
Eh,μ (u) = 2 |∇ u|2h dμ + 2 F μ , u ∧ uh dμ
M M
 
+ 2Re ¯ Tu dμ −
∂u, |Tu|2h dμ
h
M M
   
h i
=2 τ |∇ u|2h dμ + 2τ F μ + ∂∂τ , u ∧ u dμ
M M 2 h
   
− 2Re τ uk̄|j̄ ηj̄ uk̄ dμ + 2Re ¯ Tuh dμ.
τ ∂u,
M M
 COMPLEX VARIABLES AND ELLIPTIC EQUATIONS 2111

We will now use the identity

¯ Tuh = u ηj̄ uk̄ − u ηj̄ uk̄ ,
∂u, k̄|j̄ j̄|k̄

and the fact that the metric h is Kähler. This allows to integrate by parts in the following way:
   
− 2Re j̄ k̄
τ uk̄|j̄ η u dμ + 2Re ¯ Tuh dμ
τ ∂u,
M M
  
= −2Re e−η uj̄|k̄ ηj̄ uk̄ dμ
M
  
= 2Re uj̄ ηj̄ uk̄ e−η−2ψ dVol
|k̄
  
=2 ηjk̄ uj uk̄ e−η dμ − 2Re ¯ h (∂¯ ∗ u)e−η dμ − 2
u, ∂η ¯ 2 e−η dμ
| u, ∂η|
μ,h h
M M M
 
= −2 τjk̄ uj uk̄ dμ + 2Re ¯ h (∂¯ ∗ u) dμ.
u, ∂τ μ,h
M M

Notice that we used the elementary identity ∂¯μ,h

∗ u = −(uk̄ e−2ψ ) . Putting everything together, we
|k̄
find
     
¯ 2 ¯ ∗ 2 h 2 μ i ¯
τ |∂u|h dμ + τ |∂μ,h u| dμ = 2 τ |∇ u|h dμ + 2 τ F − ∂ ∂τ , u ∧ u dμ
M M M M 2 h

+ 2Re (∂¯μ,h∗ ¯ h dμ.
u) u, ∂τ
M
This is Theorem 3.1 of [18] for  = M.