Advances in Mathematics 228 (2011) 2366–2384

www.elsevier.com/locate/aim

Comparison of the Bergman and Szegö kernels ✩

Bo-Yong Chen a,b , Siqi Fu a,b,∗
a Department of Mathematics, Tongji University, Shanghai, 200092, China
b Department of Mathematical Sciences, Rutgers University, Camden, NJ 08102, USA

Received 9 June 2010; accepted 19 July 2011

Communicated by Charles Fefferman

Abstract
The quotient of the Szegö and Bergman kernels for a smooth bounded pseudoconvex domains in Cn
is bounded from above by a constant multiple of δ| log δ|p for any p > n, where δ is the distance to the
boundary. For a class of domains that includes those of D’Angelo finite type and those with plurisubhar-
monic defining functions, the quotient is also bounded from below by a constant multiple of δ| log δ|p for
any p < −1. Moreover, for convex domains, the quotient is bounded from above and below by constant
multiples of δ.
© 2011 Elsevier Inc. All rights reserved.

MSC: 32A25; 32W05; 32U35

¯
Keywords: Bergman kernel; Szegö kernel; ∂-operator; L2 -estimate; Diederich–Fornæss exponent; Pluricomplex Green
function

1. Introduction

The Bergman and Szegö kernels are two important reproducing kernels in complex analysis.
They are related yet distinct. Whereas the Bergman kernel K (as a measure) is biholomorphically
invariant, the Szegö kernel S is not. The former is connected to the ∂-problem and the ∂-Neumann
Laplacian and the latter the ∂¯b -problem and the Kohn Laplacian. In his book published in 1972,

✩
Research supported in part by NSF grants DMS-0805852 and DMS-1101678, a U.S.–China Collaboration in
Mathematical Research supplementary to NSF grant DMS-0500909, and by Fok Ying Tung Education Foundation grant
No. 111004 and Chinese NSF grant No. 11031008.
* Corresponding author at: Department of Mathematical Sciences, Rutgers University, Camden, NJ 08102, USA.
E-mail addresses: boychen@tongji.edu.cn (B.-Y. Chen), sfu@camden.rutgers.edu (S. Fu).

0001-8708/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2011.07.013
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2367

Stein posed the following problem: What are the relations between K and S? He further noted
that the relation between K and S was known only in very special circumstances [50, p. 20]. The
Szegö kernel was expressed in terms of the Cauchy–Fantappiè kernels for planar domains and
smooth bounded strictly pseudoconvex domains in Cn by Kerzman and Stein in [32,33].
There has been an extensive literature that connects the ∂-Neumann Laplacian to boundary
pseudo-differential operators associated with the Kohn Laplacian (cf. [26,37,42]) and mapping
properties of the Szegö projection to that of the Bergman projection (cf. [6–8,35,36,38,46]).
However, there are few results, as far as we know, that directly relate these two kernels.
In this paper, we study boundary behavior of the quotient S(z, z)/K(z, z) of the Szegö and
Bergman kernels for a smooth bounded pseudoconvex domain Ω in Cn . When Ω is strictly
pseudoconvex, boundary limiting behavior of the Bergman kernel K(z, z) was obtained by Hör-
mander [30] and asymptotic expansions for the Bergman and Szegö kernels were established by
Fefferman [24] and later by Boutet de Monvel and Sjöstrand [9]. As a result, S(z, z)/K(z, z) is
asymptotically δ(z)/n near the boundary, where δ(z) is the Euclidean distance to the boundary.
When Ω is a pseudoconvex domain of finite type in C2 or a convex domain of finite type in Cn ,
estimates of the Bergman kernel on diagonal from above and below were obtained by Catlin
[13] and J. Chen [16]. Estimates for the Bergman and Szegö kernels (on and off diagonal) and
their derivatives from above were established by McNeal [39,40], Nagel et al. [42], and Mc-
Neal and Stein [41]. It follows that on these domains, S(z, z)/K(z, z)  Cδ(z) for some positive
constant C.
Our main result can be stated as follows:

Theorem 1.1. Let Ω  Cn be a pseudoconvex domain with C 2 -smooth boundary.

(1) For any a ∈ (0, 1), there exists a constant C > 0 such that

S(z, z)   n/a
 Cδ(z)log δ(z)  .
K(z, z)

(2) If there exist a neighborhood U of bΩ, a bounded continuous plurisubharmonic function ϕ

on U ∩ Ω, and a defining function ρ of Ω satisfying i∂∂ϕ  iρ −1 ∂∂ρ on U ∩ Ω as currents,
then there exist constants a ∈ (0, 1] and C > 0 such that

S(z, z)   −1/a
 Cδ(z)log δ(z)  .
K(z, z)

The constant a in the second part of the theorem is a Diederich–Fornæss exponent [19]:
Namely, there exists a negative plurisubharmonic function ϕ on Ω such that C1 δ a (z)  −ϕ(z) 
C2 δ a (z) for some positive constants C1 and C2 . It was shown by Catlin [11,12] that any
smooth bounded pseudoconvex domain of D’Angelo finite type satisfies Property (P ). Sibony
further showed that for a smooth bounded pseudoconvex domain satisfying Property (P ), the
Diederich–Fornæss index, the supremum of the Diederich–Fornæss exponents, is one (see [48,
Theorem 2.4]). More recently, Fornæss and Herbig [25] showed that a smooth bounded domain
with a defining function that is plurisubharmonic on the boundary also has Diederich–Fornæss
index one.
For the convenience of the discussion, a bounded domain that satisfies the condition in (2) will
be called δ-regular. As we will show in Section 5, such a domain is necessarily pseudoconvex
2368 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

with a positive Diederich–Fornæss index. It is easy to see that the class of δ-regular domains
includes smooth bounded pseudoconvex domains with defining functions that are plurisubhar-
monic on bΩ, and it is a consequence of the above-mentioned work of Catlin [12] that this class
of domains also includes pseudoconvex domains of D’Angelo finite type (see Proposition 5.2
below). Therefore, in light of these and Theorem 1.1, we have:

Theorem 1.2. Let Ω be a smooth bounded pseudoconvex domain in Cn . Suppose that bΩ is

either of D’Angelo finite type or has a defining function that is plurisubharmonic on bΩ. Then
for any constant a ∈ (0, 1), there exist positive constants C1 and C2 such that
  −1/a S(z, z)   n/a
C1 δ(z)log δ(z)    C2 δ(z)log δ(z)  . (1.1)
K(z, z)

The logarithmic terms in the above theorems do not materialize when the domain is convex.
More precisely, we have:

Theorem 1.3. Let Ω  Cn be a bounded convex domain with C 2 -smooth boundary. Then there
exist positive constants C1 and C2 such that

C1 δ(z)  S(z, z)/K(z, z)  C2 δ(z). (1.2)

¯
Our analysis depends on the L2 -estimates for the ∂-operator by Hörmander [30], Demailly
[17], and Berndtsson [2]. Also essential to our analysis is Blocki’s estimate for the pluricomplex
Green function on hyperconvex domains [4].
This paper is organized as follows: In Section 2, we recall necessary background for the Hardy
spaces, the Bergman and Szegö kernels. In Section 3, we review the relevant L2 -estimates of the
¯
∂-operator by Hörmander [30], Demailly [17], and Berndtsson [2]. The first part of Theorem 1.1
is proved in Section 4 and the second part in Section 5. Further remarks are given in Section 6.

2. Preliminaries

We first recall necessary harmonic analysis background. We refer the reader to [50,51] for an
extensive treatise on the subject. Let D be a bounded domain in RN with C 2 -smooth boundary.
Let Dε = {x ∈ D | δD (x) > ε}, where δD (x) denotes the Euclidean distance to the boundary bD.
For 1 < p < ∞, the harmonic Hardy space hp (D) is the space of harmonic functions f such that

 p
f hp = lim sup f (z) dS < ∞.
p
ε→0+
bDε

The level sets bDε in the above definition can be replaced by those of any defining function of D
(see [50]). A classical result says that the non-tangential limit f ∗ (y) of f exists for almost every
point y on bD. Furthermore, f ∗ ∈ Lp (bD), f hp = f ∗ Lp (bD) , and

f (x) = P (x, y)f ∗ (y) dS(y),
bD

where P (x, y) is the Poisson kernel of D.

 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2369

Throughout the paper, we will use C, together with subscripts, to denote positive constants
which could be different in different appearances. We will need the following two simple well-
known lemmas. We provide the details for completeness.

Lemma 2.1. Let D1 ⊂ D2 be bounded domains in RN with C 2 -smooth boundaries. There exists
a positive constant C such that

f hp (D1 )  Cf hp (D2 ) (2.1)

for any f ∈ hp (D2 ).

Proof. This is a consequence of the following well-known fact: For a bounded domain D in RN
with C 2 -boundary and a fixed x0 ∈ D, if f ∈ hp (D), then f hp is comparable to inf(g(x0 ))1/p
where the infimum takes over all harmonic majorants g of |f |p on D. We provide the details
below, following the proof of Theorem 1 in [50]. Let

 p
g(x) = P2 (x, y)f ∗ (y) dS(y),
bD2

where P2 (x, y) is the Poisson kernel of D2 . Then for any x ∈ D2 ,

  p 
     p
f (x)p =  P2 (x, y)f (y) dS(y) 
∗
P2 (x, y)f ∗ (y) dS(y) = g(x).

bD2 bD2

Thus g(x) is a harmonic majorant of |f |p on D2 . Now fix a point x0 in D1 . Let G1 (x, y) be the
Green function of D1 . Let D1ε = {x ∈ D1 | G1 (x, y) > ε}. Then −∂G1 (x, y)/∂νy = Pε (x, y) is
the Poisson kernel of D1ε , where νy is the outward normal direction on bD1ε . Let πε : bD1ε → bD1
be the projection along the normal direction. Since Pε (x0 , πε−1 (y)) converges uniformly on bD1
to P1 (x0 , y) and C1 = min{P1 (x0 , y) | y ∈ bD1 } > 0, we have
 
C1
g(x0 ) = Pε (x0 , y)g(y) dS  g(y) dS
2
bD1ε bD1ε

for sufficiently small ε > 0. It follows that

  
  2 2  p
f (x)p dS  g(x) dS  g(x0 ) = P2 (x0 , y)f ∗ (y) dS
C1 C1
bD1ε bD1ε bD2

2C2  ∗ p
 f (y) dS = 2C2 f pp
C1 C1 h (D2 ) ,
bD2

where C2 = max{P2 (x0 , y) | y ∈ bD2 } < ∞. Thus (2.1) holds with C = (2C2 /C1 )1/p . 2

In what follows, we will also use f to denote the boundary values f ∗ for f ∈ hp (D).
2370 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

Lemma 2.2. Let D be a bounded domain in RN with C 2 -smooth boundary. For any harmonic
function f on D,
 
 
lim sup |f |p dS = lim sup(1 − r) f (x)p δ −r (x) dV . (2.2)
ε→0+ r→1−
bDε D

Furthermore, when the above limits are finite, then f ∈ hp (D) and
 
 
|f |p dS = lim (1 − r) f (x)p δ −r (x) dV . (2.3)
r→1−
bD D

Proof. If the limit on the left-hand side of (2.2) is finite, then f ∈ hp (D). Hence
 
lim |f |p dS = |f |p dS.
ε→0+
bDε bD

Write

λ(ε) = |f |p dS.
bDε

Then λ(ε) is continuous on [0, a] for any sufficiently small a > 0. Therefore,

 a 
p −r
lim (1 − r) |f | δ dV = lim (1 − r) ε −r λ(ε) dε = λ(0) = |f |p dS.
r→1− r→1−
D 0 bD

Now suppose the limit on the right-hand side of (2.2) is finite. For any sufficiently small 0 <
ε1 < ε2 , we assume that λ(ε) takes its minimum on [ε1 , ε2 ] at ε0 . Then
 
 
(1 − r) |f |p δ −r dV  (1 − r) |f |p δ −r dV  ε21−r − ε11−r λ(ε0 ).
D ε1 δε2

Taking lim infε1 →0+ and then lim supr→1− , we then have

 
∞ > lim sup(1 − r) f (x)p δ −r (x) dV  lim inf λ(ε0 ).
r→1− ε1 →0+
D

It follows that there exists a sequence εj → 0+ such that λ(εj ) is bounded. Let πε : bDε → bD
be the projection along the outward normal direction. Then fj (x) = f (πε−1 j
(x)) is a bounded
sequence in L (bD). By Alaoglu’s theorem, it has a subsequence that converges to some f˜ ∈
p

Lp (bD) in the weak∗ topology. It follows that

 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2371


f (x) = P (x, y)f˜(y) dS(y).
bD

Hence f ∈ hp (D) and we can refer back to the first part of the proof. 2

We now review the rudiments of the Bergman and Szegö kernels. Let Ω be a bounded domain
in Cn and let A2 (Ω) be the Bergman space, the space of square integrable holomorphic functions
on Ω. The Bergman kernel KΩ (z, w) is the reproducing kernel of A2 (Ω):

f (z) = KΩ (z, w)f (w) dV , ∀f ∈ A2 (Ω), ∀z ∈ Ω.
Ω

Assume that bΩ is of class C 2 . The Hardy space H 2 (Ω) is the space of holomorphic functions
on Ω that are also in h2 (Ω). The Szegö kernel is the reproducing kernel of H 2 (Ω):

f (z) = SΩ (z, w)f (w) dS, ∀f ∈ H 2 (Ω), ∀z ∈ Ω.
bΩ

It follows from these reproducing properties that

 2 
KΩ (z, z) = sup f (z) ; f ∈ A2 (Ω), f Ω  1 (2.4)

and
 2 
SΩ (z, z) = sup f (z) ; f ∈ H 2 (Ω), f bΩ  1 . (2.5)

From (2.4), we know that the Bergman kernel has the decreasing property: if Ω1 ⊂ Ω2 , then
KΩ1 (z, z)  KΩ2 (z, z). Combining Lemma 2.1 with the extremal property (2.5), we have:

Lemma 2.3. Let Ω1 ⊂ Ω2 be bounded domains in Cn with C 2 -smooth boundaries. Then there
exists a constant C > 0 such that

SΩ2 (z, z)  CSΩ1 (z, z) (2.6)

for all z ∈ Ω1 .

¯
3. Weighted L2 -estimates for the ∂-operator

In this section, we review relevant weighted L2 -estimates for the ∂-operator of Hörmander,
Demailly, and Berndtsson. We will only state their results for (0, 1)-forms, which are what we
will need later in this paper. Let Ω be a bounded pseudoconvex domain in Cn and let ψ be a
plurisubharmonic function on Ω. Let L2 (Ω, e−ψ ) be the Hilbert space of all measurable func-
tions satisfying

f 2ψ = |f |2 e−ψ dV < ∞
Ω
2372 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

and let L2(0,1) (Ω, e−ψ ) be the space of (0, 1)-forms with coefficients in L2 (Ω, e−ψ ). Suppose
∂∂ψ  c∂∂|z|2 as currents where c is a positive continuous function. (Here and in what follows,
we will drop the letter i from the real (1, 1)-form i∂∂ψ .) Hörmander’s theorem says that for any
∂-closed (0, 1)-form f , one can solve the equation

¯ =f
∂u (3.1)

in the sense of distribution, together with the estimate

 
|u| e dV  |f |2 e−ψ /c dV ,
2 −ψ
(3.2)
Ω Ω

provided the right-hand side is finite ([30, Theorem 2.2.1 ]; see also [31, Lemma 4.4.1]). Suppose
ψ ∈ C 2 (Ω). For any (0, 1)-form f , let
  
|f |∂∂ψ = sup  f, X ; X ∈ T 0,1 (Ω), |X|∂∂ψ  1

be the norm induced by the (1, 1)-form ∂∂ψ , where ·,· denotes the pairing of a form and a
vector and |X|∂∂ψ = ∂∂ψ(X, X) is the length of X with respect to ∂∂ψ. According to Demailly’s
reformulation of Hörmander’s theorem, one can solve ∂u = f with the following estimate
 
|u|2 e−ψ dV  |f |2∂ ∂ψ
¯ e
−ψ
dV , (3.3)
Ω Ω

provided the right-hand side is finite (see [17, Theorem 4.1]). It follows that for any function u
in the orthogonal complement of the nullspace N (∂) in L2 (Ω, e−ψ ), we have
 
2 −ψ −ψ
|u| e dV  |∂u|2∂ ∂ψ ¯ e dV . (3.4)
Ω Ω

The following theorem is a slight reformulation of a result due to Berndtsson [2, Theorem 2.8].
Berndtsson’s proof uses an integration by parts formula related to the ∂∂-Bochner–Kodaira tech-
nique of Siu (see Section 3 in [49]). We provide a proof here as a simple application of (3.4).
Similar approach was used in [3] to prove an estimate of Donnelly and Fefferman [23].

Theorem 3.1. Let Ω  Cn be a bounded pseudoconvex domain. Let ρ ∈ C 2 (Ω) with ρ < 0.
Suppose that there exists a plurisubhamornic function ψ ∈ C 2 (Ω) such that

¯ + ∂ ∂ρ
Θ := (−ρ)∂ ∂ψ ¯

is positive. Let u be the solution to (3.1) that is orthogonal to N (∂) in L2 (Ω, e−ψ ). Then for any
0 < r < 1,
 
−r −ψ 1
(1 − r) |u| (−ρ) e dV 
2
|f |2Θ (−ρ)1−r e−ψ dV . (3.5)
r
Ω Ω
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2373

Proof. Let φ = −r log(−ρ) and ϕ = φ + ψ . Then ueφ ⊥ N (∂) in L2 (Ω, e−ϕ ). Applying (3.4)
to ueφ with weight e−ϕ , we have
 
|u|2 eφ−ψ dV  |∂u + u∂φ|2∂∂ϕ eφ−ψ dV .
Ω Ω

It remains to show that

1
|∂u + u∂φ|2∂∂ϕ  r|u|2 + |∂u|2Θ (−ρ). (3.6)
r
Notice that

r 1
∂∂ϕ = Θ + ∂φ ∧ ∂φ + (1 − r)∂∂ψ
−ρ r
r 1
 
Θ + ∂φ ∧ ∂φ =: Θ.
−ρ r

Thus

| ∂u + u∂φ, X |2
|∂u + u∂φ|2∂∂ϕ  |∂u + u∂φ|2Θ = sup ; X ∈ T 0,1 (Ω) . (3.7)
−ρ |X|Θ + 1r | ∂φ, X |2
r 2

Inequality (3.6) then follows from (3.7) and the inequalities | ∂u, X |  |∂u|Θ |X|Θ and
   
2u ∂φ, X ∂u, X   2|u||X|Θ |∂u|Θ  ∂φ, X 
r2 −ρ  2
 |u|2 |X|2Θ + 2 |∂u|2Θ  ∂φ, X  . 2
−ρ r

4. Upper bound estimates

We prove the first part of Theorem 1.1 in this section. For a bounded domain Ω in Cn , the
pluricomplex Green function with a pole at w ∈ Ω is defined by

 
gΩ (z, w) = sup u(z); u ∈ P SH (Ω), u < 0, lim sup u(z) − log |z − w| < ∞ .
z→w

It is known that for any bounded hyperconvex domain Ω, the pluricomplex Green function
gΩ (·, w) : Ω → [−∞, 0) is a continuous plurisubharmonic function such that limz→bΩ gΩ (z, w)
= 0 ([18]; see also Chapter 5 in [34]).
Recall that a constant a ∈ (0, 1] is said to be a Diederich–Fornæss exponent for a bounded
pseudoconvex domain Ω if there exist a negative plurisubharmonic function ϕ on Ω and positive
constants C1 and C2 such that

C1 δ a (z)  −ϕ(z)  C2 δ a (z). (4.1)

2374 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

The supremum of all Diederich–Fornæss exponents is called the Diederich–Fornæss index of Ω.

It follows from the work of Diederich and Fornæss that any bounded pseudoconvex domain with
C 2 -smooth boundary has a positive Diederich–Fornæss index, which can be arbitrarily small [19,
20]. It was proved by Demailly [18] that bounded pseudoconvex Lipschitz domains are hyper-
convex. More recently, Harrington [27] showed that bounded pseudoconvex Lipschitz domains
have indeed positive Diederich–Fornæss indices.
The following quantitative estimate for the pluricomplex Green function due to Blocki ([4,
Theorem 5.2]; see also [29] for prior related results) plays an essential rôle in our analysis. We
provide the detail of a proof below, following Blocki’s arguments,1 because we will need to refer
back to it.

Theorem 4.1. Let Ω be a bounded pseudoconvex domain in Cn . Suppose there exists a negative
plurisubharmonic function ϕ on Ω such that

C1 δ a (z)  −ϕ(z)  C2 δ b (z), z∈Ω (4.2)

for some positive constants C1 , C2 , and a  b. Then there exists positive constants δ0 and C such
that
    − 1  n 
z ∈ Ω; gΩ (z, w)  −1 ⊂ C −1 δ b (w)log δ(w) b  δ(z)  Cδ a (w)log δ(w) a , (4.3)
a b

for any w ∈ Ω with δ(w)  δ0 .

Proof. Assume that Ω has diameter R. Let w ∈ Ω with r = δ(w)  e−2 . Let z ∈ Ω. Sup-
pose that δ = δ(z)  r/2. It follows from comparison with the pluricomplex Green function of
B(w, R) that
 
gΩ (ζ, w)  log |ζ − w|/R (4.4)

for all ζ ∈ Ω. By the maximal property of the pluricomplex Green function, we have

log(2R/r)
gΩ (ζ, w)  ϕ(ζ ) (4.5)
inf{|ϕ(ζ )|; ζ ∈ B(w, 2r )}

on Ω \ B(w, r/2) because the same inequality holds on the boundary. By (4.2),

  r
inf ϕ(ζ ); ζ ∈ B w,  C(r/2)a . (4.6)
2

Therefore,

δb 1
gΩ (z, w)  −C a
log . (4.7)
r r

1 There was a slight inaccuracy in the proof of Theorem 5.2 in [4] which can be easily corrected by an appropriate
choice of ε [5].
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2375

It follows that

  r a − 1
z ∈ Ω; gΩ (z, w)  −1 ⊂ δ(z) > or δ(z)  Cr b log(1/r) b
2
  − 1 
⊂ δ(z)  C −1 δ b (w)log δ(w) b ,
a
(4.8)

provided the last constant C is sufficiently large.

Now suppose that e−2  δ(z)  2r. It follows from (4.7) that for any 0 < ε < r/2,

εb 1
inf gΩ (ζ, w)  −C log . (4.9)
δ(ζ )=ε ra r

We also obtain from (4.7) that

rb 1
gΩ (w, z)  −C a
log , (4.10)
δ δ
by reversing the rôles of z and w. By Theorem 5.1 in [4], we have

b 1− 1 1 
log 1ε εb 1 rn 1 n 1 n
gΩ (z, w)  −C r a
log + a log log . (4.11)
log 2ε r r δn ε δ

Set
1 a 1
1 r 1+ n + b log log 1r b
ε= a 1
. (4.12)
2 δ bn log r

Since δ  2r, log log 1r  log 1r , and

1 a a
+ −  1, (4.13)
n b bn
we have
1 a a
r 1+ n + b − bn 1
ε a
1+ bn
< r 2.
2 2

Therefore,

r 1
log  log . (4.14)
2ε r
It is easy to see that

1 1 1 1
log  C log and log  log . (4.15)
ε r δ r
2376 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

Combining (4.12), (4.14), and (4.15) with (4.11), we then obtain

b
rn 1
gΩ (z, w)  −C alog .
δn r

Therefore,
    n/a 
gΩ (z, w) < −1 ⊂ z ∈ Ω; δ(z)  e−2 or δ(z)  Cδ b/a (w)log δ(w) .

Together with (4.8), we then obtain (4.3) by choosing a sufficiently small δ0 and a sufficiently
large C. 2

We also need the following localization of the Bergman kernel ([14, Lemma 4.2]; also [28,
Proposition 3.6]).

Proposition 4.2. Let Ω be a bounded pseudoconvex domain in Cn . Then there exists a positive
constant C such that for any w ∈ Ω,

KΩ (w, w)  CK{gΩ (·,w)<−1} (w, w).

To illustrate the idea of the proof, we first prove the following weaker version of Theo-
rem 1.1(1):

Proposition 4.3. Let Ω  Cn be a pseudoconvex domain with C 2 -smooth boundary. Suppose

that the Diederich–Fornæss index of Ω is β. Then for any a ∈ (0, β), there exists a constant
C > 0 such that
S(z, z)   n/a
 Cδ(z)log δ(z)  . (4.16)
K(z, z)

Proof. By the definition of the Diederich–Fornæss index, there exists a negative plurisubhar-
monic function ϕ satisfying (4.1). By Theorem 4.1, there exists a positive constant C such that
    n/a 
gΩ (·, w) < −1 ⊂ δ(·) < Cδ(w)log δ(w) (4.17)

for any w ∈ Ω sufficiently closed to the boundary. Therefore, for any f ∈ H 2 (Ω),

 log δ(w)|n/a
Cδ(w)|  
 n/a
|f | dV 
2
dε |f |2 dS  Cf 2bΩ δ(w)log δ(w) .
{gΩ (·,w)<−1} 0 bΩε

(Here in the last inequality, we have used Lemma 2.1 and its proof.) It then follows from the
extremal properties (2.4) and (2.5) that
 n/a
S(w, w)  Cδ(w)log δ(w) K{g(·,w)<−1} (w, w).

Applying Proposition 4.2, we then conclude the proof of the proposition. 2

 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2377

To get from Proposition 4.3 to the first statement of Theorem 1.1, we use Lemma 2.3 to
localize the Szegö kernel and then apply the following fact: For any z0 ∈ bΩ and a ∈ (0, 1),
there exist a defining function r of Ω and a neighborhood U of z0 such that ϕ2 = −(−r)a is
strictly plurisubharmonic on U ∩ Ω ([19], Remark on p. 133). The problem is that this function
ϕ2 is not an exhaustion function of U ∩ Ω and thus one cannot directly apply Theorem 4.1. We
now show how to overcome this difficulty and prove Theorem 1.1(1).
Let χ(t) be a smooth function such that χ(t) = 0 when t  1, χ(t) > 0 is strictly increasing
and convex when t > 1. We may further assume that χ(t) = exp(−1/(t − 1)) when t ∈ (1, 5/4)
so that (χ(t))b ∈ C ∞ (R) for any positive number b. Let
 
ϕ1 (z) = −(−r)a + Mχ |z − z0 |2 /m2 .

Then ϕ1 is strictly plurisubharmonic on U ∩ Ω. Write g(z) = Mχ(|z − z0 |2 /m2 ). Let

 
 = z ∈ Ω; ϕ1 (z) = −(−r)a + g < 0 .
Ω

 and
By choosing m sufficiently small and M sufficiently large, we know that B(z0 , m) ∩ Ω ⊂ Ω
 
Ω ⊂ B(z0 , 2m). Furthermore, Ω is pseudoconvex with a C -smooth defining function
2

r̃ = r + g 1/a

(see, for example, [1, pp. 470–471]).

Evidently, ϕ1 is a plurisubharmonic exhaustion function for Ω.  However, ϕ1 does not satisfy
(4.1). In fact, it is easy to show that there exists a positive constant C1 such that

C1 |r̃|  −ϕ1  |r̃|a , 

z ∈ Ω. (4.18)

If we directly invoke Theorem 4.1 with (4.18), we obtain

    n 
gΩ (·, w) < −1 ⊂ δ(·) < Cδ a (w)log δ(w) .

Consequently, we have as in the proof of Proposition 4.3 that

 n
S(z, z)/K(z, z)  Cδ a (z)log δ(z) ,

which is even weaker than (4.16).

Instead of directly appealing to Theorem 4.1, we proceed as follows. We follow the proof of
Theorem 4.1 with Ω replaced by Ω,  and with δ(z) = δΩ (z) now denoting the Euclidean distance
 Notice that C −1 δ  |r̃|  Cδ on Ω.
to bΩ.  Assume that |w − z0 | < m/4. When ζ ∈ B(w, r/2),
|ζ − z0 |  m/2. Hence
     
ϕ1 (ζ ) = r(ζ )a  δ(ζ ) a  r a .

Applying (4.5) to the function ϕ1 , we have

log(2R/r) εa 1
inf gΩ (ζ, w)  inf ϕ1 (ζ )  −C log , (4.19)
δ(ζ )=ε inf{|ϕ1 (ζ )|; ζ ∈ B(w, 2r )} δ(ζ )=ε ra r
2378 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

which is our analogue in this case to (4.9). (Here we have a = b.) Now applying (4.5) to the
function ϕ2 = −(−r)a with the rôle of z and w reversed, we have

log(2R/δ)
gΩ (ζ, z)  ϕ2 (ζ )
inf{|ϕ2 (ζ )|; ζ ∈ B(z, 2δ )}

 \ B(z, δ/2). It follows that

on Ω

ra 1
gΩ (w, z)  −C log , (4.20)
δa δ
which plays the rôle of (4.10) in this case. Following exactly the same lines for the rest of the
proof of Theorem 4.1, we then obtain
     
 δ(z)  Cδ(w)log δ(w)n/a .
 gΩ (z, w) < −1 ⊂ z ∈ Ω;
z ∈ Ω;

From the proof of Proposition 4.3, we then have

SΩ (w, w)   n/a

 Cδ(w)log δ(w)  ,
KΩ (w, w)

when w is sufficiently close to z0 . By the localization property of the Bergman kernel (see the
proof of Theorem 1 in [43]; also [21, Proposition 1]), KΩ (w, w)  CKΩ (w, w). Together with
Lemma 2.3, we then conclude the proof of the first statement in Theorem 1.1.

5. Lower bound estimates

Recall that a continuous function ρ is said to be a defining function of a domain Ω ⊂ Cn

if Ω = {z ∈ Cn ; ρ(z) < 0} and C −1 δ  −ρ  Cδ for a constant C > 0. We also assume the
defining function ρ to be in the same smoothness class as that of the boundary bΩ. A bounded
domain Ω ⊂ Cn is δ-regular if there exist a neighborhood U of bΩ, a bounded continuous
plurisubharmonic function ϕ on U ∩ Ω, and a defining function ρ of Ω such that

∂∂ϕ  ρ −1 ∂∂ρ (5.1)

on U ∩ Ω as currents. By adding |z|2 to ϕ, we may assume that it is strictly plurisubharmonic.

By Richberg’s approximation theorem [47, Satz 4.3], we may further assume that ϕ ∈ C ∞ (Ω).

Proposition 5.1. Let Ω  Cn be a δ-regular domain. Then Ω is hyperconvex with a positive

Diederich–Fornæss index.

Proof. Let ϕ and ρ be the functions that satisfy (5.1). Assume 0  ϕ  M for some positive
constant M. Let ψ = eϕ and K = eM . Then

∂∂ρ
1  ψ  K, ∂ψ ∧ ∂ψ  K∂∂ψ, and ∂∂ψ  + K −1 ∂ψ ∧ ∂ψ, (5.2)
ρ
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2379

on U ∩ Ω. Let

ρ̃ = ρe−ψ and r = −(−ρ̃)η .

It follows from a simple (formal) computation and (5.2) that


∂ ρ̃ ∧ ∂ ρ̃
∂∂r = η(−ρ̃) η
∂∂ − log(−ρ̃) − η
ρ̃ 2

1 ∂ρ ∧ ∂ρ ∂ ρ̃ ∧ ∂ ρ̃
 η(−ρ̃) η
∂ψ ∧ ∂ψ + −η
K ρ2 ρ̃ 2
 
∂ρ ∧ ∂ρ ∂ρ ∂ρ 1
 η(−ρ̃)η (1 − η) + η ∧ ∂ψ + η∂ψ ∧ + − η ∂ψ ∧ ∂ψ .
ρ2 ρ ρ K

We then obtain from the Schwarz inequality that ∂∂r is a positive current on U ∩ Ω, provided η
is sufficiently small. The extension of r to the whole domain Ω is standard (see [19, p. 133]). 2

Proposition 5.2. Let Ω be a smooth pseudoconvex bounded domain in Cn . If Ω has a defining

function that is plurisubharmonic on bΩ or if bΩ is of D’Angelo finite type, then Ω is δ-regular.

Proof. If Ω has a defining function ρ that is plurisubharmonic on bΩ. Then ∂∂ρ  Cρ∂∂|z|2 .
Therefore in this case, we can choose ϕ(z) = C|z|2 with a sufficiently large C.
The case when Ω is of D’Angelo finite type is a consequence of Catlin’s construction of
bounded plurisubharmonic function [12] (see [52, p. 464] for a related discussion): There exist
positive constants τ < 1, C > 0, and a smooth bounded plurisubharmonic function λ on Ω such
that

¯ 2
∂ ∂|z|
¯ C
∂ ∂λ . (5.3)
|ρ|τ

By Oka’s lemma, we can choose a defining function ρ such that ∂∂(− log(−ρ))  ∂∂|z|2 . It then
follows from a theorem of Diederich–Fornæss that for any sufficiently small η,
   
∂ ∂¯ −(−ρ)η  Cη|ρ|η ∂∂|z|2 + |ρ|−2 ∂ρ ∧ ∂ρ (5.4)

for some positive constant C ([19], Theorem 1 and

its proof; compare also [10, Lemma 2.2]).
Now we fix an η ∈ (0, τ ). Write N = |∂ρ|−1 ρz̄j ∂/∂zj . For any (1, 0)-vector X, write
XN = X, N N and XT = X − XN . By the pseudoconvexity of bΩ, we have
   
ρ −1 ∂∂ρ(X, X)  C |X|2 + |ρ|−1 |X||XN |  C |ρ|−η |X|2 + |ρ|−2+η |XN |2 .

The desirable function is then given by

 
ϕ = C λ − (−ρ)η

for any sufficiently large C > 0. 2

2380 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

We are now in position to prove the second part of Theorem 1.1. Let κ be a standard Friedrichs
mollifier. Let εj be a decreasing sequence of positive number tending to 0. Let w be a point
in Ω, sufficiently closed to the boundary bΩ. Let gj = gΩ (·, w) ∗ κεj . Then gj is a decreasing
sequence of plurisubharmonic functions on Ωj = {z ∈ Ω; δ(z) < εj } with limit gΩ (·, w). By
Oka’s lemma, Ωj is pseudoconvex. Let
 
ψ = 2ngΩ (·, w) − log −gΩ (·, w) + 1 + ϕ and ψj = 2ngj − log(−gj + 1) + ϕ,

where ϕ is the smooth bounded strictly plurisubharmonic function, obtained from the δ-regularity
assumption, such that ∂∂ϕ  ρ −1 ∂∂ρ for a defining function ρ of Ω. Clearly ψj is a plurisub-
harmonic function on Ωj . Moreover,

¯ j  ∂ log(−gj + 1) ∧ ∂¯ log(−gj + 1)
∂ ∂ψ (5.5)

¯ j + ∂ ∂ρ
and Θj = (−ρ)∂ ∂ψ ¯ is positive on Ωj . Let χ : R → [0, 1] be a C ∞ cut-off function such
that χ|(−∞,−1) = 1 and χ|(0,∞) = 0. Put

 
¯ − log(−gj ) √KΩ (·, w) .
vj = ∂χ
KΩ (w, w)

Let uj be the solution to ∂uj = vj that is in the orthogonal complement of N (∂) in

L2 (Ωj , e−ψj ). By Demailly’s estimate (3.3),
 
2 −ψj −ψj
|uj | e dV  |vj |2∂ ∂ψ
¯ e dV .
j
Ωj Ωj

It follows from (5.5) that the right-hand side is uniformly bounded from above, independent of j .
Passing to a subsequence, we may assume that uj converges to u ∈ L2 (Ω, e−ψ ) in the weak∗
topology. (We extend uj = 0 on Ω \ Ωj .) Let
  
f = χ − log −gΩ (·, w) KΩ (·, w)/KΩ (w, w)1/2 − u.

Then f is holomorphic on Ω. Since u is holomorphic in a neighborhood of w and gΩ (z, w) =

log |z − w| + O(1) near w,

u(w) = 0.

By Theorem 3.1, for any 0 < r < 1, we have

 
1
(1 − r) |uj |2 (−ρ)−r e−ψj dV  |vj |2Θj (−ρ)1−r e−ψj dV
r
Ωj Ωj

C |KΩ (·, w)|2
 (−ρ)−r dV . (5.6)
r KΩ (w, w)
¯
supp ∂χ(·)
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2381

By Theorem 4.1,
   
¯
supp ∂χ(·) ⊂ −e  gj (·, w)  −1 ⊂ gΩ (·, w)  −1
    −1/a 
⊂ C −1 ρ(w)log −ρ(w)   |ρ| ,

where a is a Diederich–Fornæss exponent for Ω. Therefore, passing to the limit, we have


C | log(−ρ(w))|r/a
(1 − r) |u|2 (−ρ)−r e−ψ dV  · . (5.7)
r |ρ(w)|r
Ω

Notice that f is a holomorphic function on Ω such that f (w) = KΩ (w, w)1/2 and f = −u near
bΩ. Since e−ψ  e−ϕ  C > 0, by Lemma 2.2 (and its proof), f ∈ H 2 (Ω). Combining with
(5.7), we have
 
|f |2 dS  C lim (1 − r) |f |2 (−ρ)−r e−ψ dV
r→1−
bΩ Ω

= C lim (1 − r) |u|2 (−ρ)−r e−ψ dV
r→1−
Ω

| log(−ρ(w))|1/a
C .
|ρ(w)|

It then follows from the extremal property (2.5) of the Szegö kernel that

|ρ(w)|
SΩ (w, w)  C · KΩ (w, w).
| log(−ρ(w))|1/a

This concludes the proof of Theorem 1.1.

The proof of Theorem 1.3 is similar that of Theorem 1.1. The only difference is that, instead
of Theorem 4.1, we use the following well-known estimate for the pluricomplex Green function
on convex domains:
   
gΩ (·, w)  −1 ⊂ C −1 δ(w)  δ(·)  Cδ(w) (5.8)

(see [4], Theorem 5.4).

6. Further remarks

(1) When Ω is a smooth bounded domain of the form

 
m
 2
Ω = z∈C ; n  
fj (z) < 1 , (6.1)
k=1

where fj are holomorphic functions, the pluricomplex Green function gΩ (z, w) satisfies the esti-
mate (5.8). Therefore, the logarithmic terms do not materialize in Theorem 1.1 for such domains.
2382 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

Similar estimates also hold for smooth bounded h-extendible domains (see [53,22] for properties
of these domains). It would be interesting to know whether this is indeed the case for all smooth
bounded pseudoconvex domains in Cn . We refer the reader to the forthcoming paper [15] for
more discussions on these and other results.
(2) It follows from the Ohsawa–Takegoshi extension theorem [45] that for any bounded pseu-
doconvex Lipschitz domain Ω in Cn , K(z, z)  Cδ −2 (z) for some constant C > 0. Ohsawa [44]
conjectured that the analogue estimate S(z, z)  Cδ −1 (z) holds. In light of Theorem 1.3 and
the above remark, Ohsawa’s conjecture holds for convex domains, h-extendible domains, and
domains of form (6.1). Moreover, Theorem 1.1 implies that for a bounded δ-regular domain Ω
with C 2 -smooth boundary,
 −1/a
S(z, z)  Cδ −1 (z)log δ(z) ,

where a is any Diederich–Fornæss exponent of Ω.

Acknowledgments

Part of this paper was written while the first author visited Rutgers University–Camden and
the second author visited the Ewin Schrödinger Institute and the Australia National University.
The authors thank these institutions for hospitalities. The authors also thank Professors Zbigniew
Blocki, Alexander Nagel, Takeo Ohsawa, and Malgorzata Zajecka for helpful discussions and
communications.

References

[1] S. Bell, Differential of the Bergman kernel and pseudo-local estimates, Math. Z. 192 (1986) 467–472.
¯
[2] B. Berndtsson, Weighted estimates for the ∂-equation, in: J.D. McNeal (Ed.), Complex Analysis and Complex
Geometry, de Gruyter, 2001, pp. 43–57.
[3] B. Berndtsson, Ph. Charpentier, A Sobolev mapping property of the Bergman kernel, Math. Z. 235 (2000) 1–10.
[4] Z. Blocki, The Bergman metric and the pluricomplex Green function, Trans. Amer. Math. Soc. 357 (2004) 2613–
2625.
[5] Z. Blocki, Personal communications, December, 2010.
[6] H.P. Boas, The Szegö projection: Sobolev estimates in regular domains, Trans. Amer. Math. Soc. 300 (1987) 109–
132.
[7] H.P. Boas, E.J. Straube, Complete Hartogs domains in C2 have regular Bergman and Szegö projections, Math.
Z. 201 (3) (1989) 441–454.
[8] A. Bonami, Ph. Charpentier, Comparing the Bergman and Szegö projections, Math. Z. 204 (1990) 225–233.
[9] L. Boutet de Monvel, J. Sjöstrand, Sur la singularité des noyaux de Bergman et de Szego, in: Journées: Équations
aux Dérivées Partielles de Rennes, 1975, Asterisque 34–35 (1976) 123–164.
[10] J. Cao, M.-C. Shaw, L. Wang, Estimates for the ∂-Neumann problem and nonexistence of C 2 Levi-flat hypersurfaces
in CP n , Math. Z. 248 (2004) 183–221.
[11] D. Catlin, Global regularity of the ∂-Neumann problem, in: Y.-T. Siu (Ed.), Complex Analysis of Several Variables,
in: Proc. Sympos. Pure Math., vol. 41, Amer. Math. Soc., 1984, pp. 39–49.
¯
[12] D. Catlin, Subelliptic estimates for the ∂-Neumann problem on pseudoconvex domains, Ann. of Math. 126 (1987)
131–191.
[13] D. Catlin, Estimates of invariant metrics on pseudoconvex domains of dimension two, Math. Z. 200 (1989) 429–466.
[14] B.Y. Chen, Completeness of the Bergman metric on non-smooth pseudoconvex domains, Ann. Polon. Math. 71
(1999) 241–251.
[15] B.Y. Chen, S. Fu, The reproducing kernels and the finite type conditions, in preparation.
[16] J.-H. Chen, Estimates of the invariant metrics on convex domains, PhD thesis, Purdue University, 1989.
[17] J.-P. Demailly, Estimations L2 pour l’opérateur ∂¯ d’un fibré vectoriel holomorphe semi-positif au-dessus d’une
variété kählérienne complète, Ann. Sci. École Norm. Sup. 15 (1982) 457–511.
 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384 2383

[18] J.-P. Demailly, Mesures de Monge–Ampere et mesures plurisousharmoniques, Math. Z. 194 (1987) 519–564.
[19] K. Diederich, J.E. Fornæss, Pseudoconvex domains: bounded strictly plurisubharmonic exhaustion functions, In-
vent. Math. 39 (1977) 129–141.
[20] K. Diederich, J.E. Fornæss, Pseudoconvex domains: an example with nontrivial Nebenhülle, Math. Ann. 225 (1977)
275–292.
[21] K. Diederich, J.E. Fornæss, G. Herbort, Boundary behavior of the Bergman metric, in: Y.-T. Siu (Ed.), Complex
Analysis of Several Variables, in: Proc. Sympos. Pure Math., vol. 41, Amer. Math. Soc., 1984, pp. 59–67.
[22] K. Diederich, G. Herbort, Pseudoconvex domains of semiregular type, in: Contributions to Complex Analysis and
Analytic Geometry, in: Aspects Math., vol. 26, Vieweg, Braunschweig, 1994, pp. 127–161.
[23] H. Donnelly, Ch. Fefferman, L2 -cohomology and index theorem for the Bergman metric, Ann. of Math. (2) 118
(1983) 593–618.
[24] C. Fefferman, The Bergman kernel and biholomorphic mappings of pseudoconvex domains, Invent. Math. 26 (1974)
1–65.
[25] J.E. Fornæss, A.-K. Herbig, A note on plurisubharmonic defining functions in Cn , Math. Ann. 342 (2008) 749–
772.
¯
[26] P.C. Greiner, E.M. Stein, Estimate for the ∂-Neumann Problem, Princeton University Press, Princeton, NJ, 1977.
[27] P.S. Harrington, The order of plurisubharmonicity on pseudoconvex domains with Lipschitz boundaries, Math. Res.
Lett. 14 (2007) 485–490.
[28] G. Herbort, The Bergman metric on hyperconvex domains, Math. Z. 232 (1999) 183–196.
[29] G. Herbort, The pluricomplex Green function on pseudoconvex domains with a smooth boundary, Internat. J.
Math. 11 (2000) 509–522.
¯
[30] L. Hörmander, L2 -estimates and existence theorems for the ∂-equation, Acta Math. 113 (1965) 89–152.
[31] L. Hörmander, An Introduction to Complex Analysis of Several Variables, Elsevier Science Publishers, Amsterdam,
The Netherlands, 1991.
[32] N. Kerzman, E.M. Stein, The Szegö kernel in terms of Cauchy–Fantappiè kernels, Duke Math. J. 45 (2) (1978)
197–224.
[33] N. Kerzman, E.M. Stein, The Cauchy kernel, the Szegö kernel, and the Riemann mapping function, Math.
Ann. 236 (1) (1978) 85–93.
[34] M. Klimek, Pluripotential Theory, Oxford University Press, 1991.
[35] K.D. Koenig, An analogue of the Kerzman–Stein formula for the Bergman and Szegö projections, J. Geom.
Anal. 19 (1) (2009) 81–86.
[36] K.D. Koenig, L. Lanzani, Bergman versus Szegö via conformal mapping, Indiana Univ. Math. J. 58 (2) (2009)
969–997.
[37] J.J. Kohn, Superlogarithmic estimates on pseudoconvex domains and CR manifolds, Ann. of Math. 156 (2002)
213–248.
[38] M. Machedon, Szegö kernels on pseudoconvex domains with one degenerate eigenvalue, Ann. of Math. 128 (1988)
619–640.
[39] J. McNeal, Boundary behavior of the Bergman kernel function in C2 , Duke Math. J. 58 (1989) 499–512.
[40] J. McNeal, Estimates on the Bergman kernels of convex domains, Adv. Math. 109 (1994) 108–139.
[41] J.D. McNeal, E.M. Stein, The Szegö projection on convex domains, Math. Z. 224 (1997) 519–553.
[42] A. Nagel, J.P. Rosay, E.M. Stein, S. Wainger, Estimates for the Bergman and Szegö kernels in C2 , Ann. of Math. 129
(1989) 113–149.
[43] T. Ohsawa, A remark on the completeness of the Bergman metric, Proc. Japan Acad. Ser. A 57 (1981) 238–240.
[44] T. Ohsawa, Personal communication, 2003.
[45] T. Ohsawa, K. Takegoshi, On the extension of L2 holomorphic functions, Math. Z. 195 (1987) 197–204.
[46] D.H. Phong, E.M. Stein, Estimates for the Bergman and Szegö projections on strongly pseudoconvex domains,
Duke Math. J. 44 (1977) 695–704.
[47] R. Richberg, Stetige streng pseudokonvexe funktionen, Math. Ann. 175 (1968) 257–286.
[48] N. Sibony, Some aspects of weakly pseudoconvex domains, in: E. Bedford, J. D’Angelo, R. Greene, S. Krantz
(Eds.), Several Complex Variables and Complex Geometry, in: Proc. Sympos. Pure Math., vol. 52, Amer. Math.
Soc., 1991, pp. 199–231.
[49] Y.-T. Siu, Complex-analyticity of harmonic maps, vanishing and Lefschetz theorems, J. Differential Geom. 17 (1)
(1982) 55–138.
[50] E.M. Stein, Boundary Behavior of Holomorphic Functions of Several Complex Variables, Princeton University
Press, Princeton, NJ, 1972.
2384 B.-Y. Chen, S. Fu / Advances in Mathematics 228 (2011) 2366–2384

[51] E.M. Stein, Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals, Princeton Uni-
versity Press, Princeton, NJ, 1993.
¯
[52] E.J. Straube, Plurisubharmonic functions and subelliptic estimates for the ∂-Neumann problem on nonsmooth do-
mains, Math. Res. Lett. 4 (1997) 459–467.
[53] J. Yu, Peak functions on weakly pseudoconvex domains, Indiana Univ. Math. J. 43 (4) (1994) 1271–1295.