Bergman Kernel in Convex Domains
Zbigniew B locki
Uniwersytet Jagiello´ nski, Krak´ ow, Poland http://gamma.im.uj.edu.pl/ eblocki
Several Complex Variables Symposium Tsinghua Sanya International Mathematics Forum
May 12–16, 2014
Ω ⊂ C
n, w ∈ Ω
K
Ω(w ) = sup{|f (w )|
2: f ∈ O(Ω), Z
Ω
|f |
2d λ ≤ 1}
(Bergman kernel on the diagonal)
G
w(z) = G
Ω(z, w )
= sup{u(z) : u ∈ PSH
−(Ω) : lim
z→w
u(z) − log |z − w | < ∞}
(pluricomplex Green function)
Theorem 0 Assume Ω is pseudoconvex in C
n. Then for w ∈ Ω and t ≤ 0
K
Ω(w ) ≥ 1
e
−2ntλ({G
w< t}) .
Optimal constant: “=” if Ω = B(w , r ).
Proof 1 Using Donnelly-Fefferman’s estimate for ¯ ∂ one can prove
K
Ω(w ) ≥ 1
c(n, t)λ({G
w< t}) , (1) where
c(n, t) =
1 + C
Ei (−nt)
2, Ei (a) = Z
∞a
ds se
s(B. 2005). Now use the tensor power trick: e Ω = Ω × · · · × Ω ⊂ C
nm, w = (w , . . . , w ) for m 0. Then e
K
eΩ( w ) = (K e
Ω(w ))
m, λ({G
we
< t}) = (λ({G
w< t}))
m, and by (1) for e Ω
K
Ω(w ) ≥ 1
c(nm, t)
1/mλ({G
w< t}) . But lim
m→∞
c(nm, t)
1/m= e
−2nt.
Proof 2 (Lempert) By Berndtsson’s result on log-(pluri)subharmonicity of the Bergman kernel for sections of a pseudoconvex domain it follows that log K
{Gw<t}(w ) is convex for t ∈ (−∞, 0]. Therefore
t 7−→ 2nt + log K
{Gw<t}(w )
is convex and bounded, hence non-decreasing. It follows that K
Ω(w ) ≥ e
2ntK
{Gw<t}(w ) ≥ e
2ntλ({G
w< t}) .
Berndtsson: This method can be improved to show the
Ohsawa-Takegoshi extension theorem with optimal constant.
Theorem 0 Assume Ω is pseudoconvex in C
n. Then for w ∈ Ω and t ≤ 0
K
Ω(w ) ≥ 1
e
−2ntλ({G
w< t}) .
What happens when t → −∞? For n = 1 Theorem 0 immediately gives:
Theorem (Suita conjecture) For a domain Ω ⊂ C one has
K
Ω(w ) ≥ c
Ω(w )
2/π, w ∈ Ω, (2) where c
Ω(w ) = exp lim
z→w(G
Ω(z, w ) − log |z − w |)
(logarithmic capacity of C \ Ω w.r.t. w ).
Theorem (Guan-Zhou) Equality holds in (2) iff Ω ' ∆ \ F , where ∆ is
the unit disk and F a closed polar subset.
-10 -8 -6 -4 -2
1 2 3 4 5 6 7
πK
Ωc
Ω2for Ω = {e
−5< |z| < 1} as a function of 2 log |w |
What happens with e
−2ntλ({G
w< t}) as t → −∞ for arbitrary n? For convex Ω using Lempert’s theory one can get
Proposition If Ω is bounded, smooth and strongly convex in C
nthen for w ∈ Ω
t→−∞
lim e
−2ntλ({G
w< t}) = λ(I
ΩK(w )),
where I
ΩK(w ) = {ϕ
0(0) : ϕ ∈ O(∆, Ω), ϕ(0) = w } (Kobayashi indicatrix).
Corollary If Ω ⊂ C
nis convex then K
Ω(w ) ≥ 1
λ(I
ΩK(w )) , w ∈ Ω.
For general Ω one can prove
Theorem (B.-Zwonek) If Ω is bounded and hyperconvex in C
nand w ∈ Ω then
t→−∞
lim e
−2ntλ({G
w< t}) = λ(I
ΩA(w )), where I
ΩA(w ) = {X ∈ C
n: lim
ζ→0G
w(w + ζX ) − log |ζ| ≤ 0}
(Azukawa indicatrix)
Corollary (SCV version of the Suita conjecture) If Ω ⊂ C
nis pseudoconvex and w ∈ Ω then
K
Ω(w ) ≥ 1 λ(I
ΩA(w )) .
Conjecture 1 For Ω pseudoconvex and w ∈ Ω the function t 7−→ e
−2ntλ({G
w< t})
is non-decreasing in t.
It would easily follow from the following:
Conjecture 2 For Ω pseudoconvex and w ∈ Ω the function t 7−→ log λ({G
w< t})
is convex on (−∞, 0].
Theorem (B.-Zwonek) Conjecture 1 is true for n = 1.
Proof It is be enough to prove that f
0(t) ≥ 0 where f (t) := log λ({G
w< t}) − 2t and t is a regular value of G
w. By the co-area formula
λ({G
w< t}) = Z
t−∞
Z
{Gw=s}
d σ
|∇G
w| ds and therefore
f
0(t) = Z
{Gw=t}
d σ
|∇G
w| λ({G
w< t}) − 2.
By the Schwarz inequality Z
{Gw=t}
d σ
|∇G
w| ≥ (σ({G
w= t}))
2Z
{Gw=t}
|∇G
w|d σ
= (σ({G
w= t}))
22π .
The isoperimetric inequality gives
(σ({G
w= t}))
2≥ 4πλ({G
w< t}) and we obtain f
0(t) ≥ 0.
Conjecture 1 for arbitrary n is equivalent to the following pluricomplex isoperimetric inequality for smooth strongly pseudoconvex Ω
Z
∂Ω
d σ
|∇G
w| ≥ 4nπλ(Ω).
Conjecture 1 also turns out to be closely related to the problem of
symmetrization of the complex Monge-Amp` ere equation.
What about corresponding upper bound in the Suita conjecture?
Not true in general:
Proposition (B.-Zwonek) Let Ω = {r < |z| < 1}. Then K
Ω( √
r ) (c
Ω( √
r ))
2≥ −2 log r π
3.
It would be interesting to find un upper bound of the Bergman kernel for domains in C in terms of logarithmic capacity which would in particular imply the ⇒ part in the well known equivalence
K
Ω> 0 ⇔ c
Ω> 0
(c
Ω2≤ πK
Ωbeing a quantitative version of ⇐).
The upper bound for the Bergman kernel holds for convex domains:
Theorem (B.-Zwonek) For a convex Ω and w ∈ Ω set F
Ω(w ) := K
Ω(w )λ(I
ΩK(w ))
1/n.
Then F
Ω(w ) ≤ 4. If Ω is in addition symmetric w.r.t. w then F
Ω(w ) ≤ 16/π
2= 1.621 . . . .
Sketch of proof Denote I := int I
ΩK(w ) and assume that w = 0. One can show that I ⊂ 2 Ω (I ⊂ 4/π Ω if Ω is symmetric). Then
K
Ω(0)λ(I ) ≤ K
I /2(0)λ(I ) = λ(I )
λ(I /2) = 4
n.
For convex domains F
Ωis a biholomorphically invariant function satisfying 1 ≤ F
Ω≤ 4. Can we find an example with F
Ω(w ) > 1? Using Jarnicki-Pflug-Zeinstra’s formula for geodesics in convex complex ellipsoids (which is based on Lempert’s theory) one can show the following
Theorem (B.-Zwonek) Define
Ω = {z ∈ C
n: |z
1| + · · · + |z
n| < 1}.
Then for w = (b, 0, . . . , 0), where 0 < b < 1, one has
K
Ω(w )λ(I
ΩK(w )) = 1 + (1 − b)
2n(1 + b)
2n− (1 − b)
2n− 4nb 4nb(1 + b)
2n= 1 + (1 − b)
2n(1 + b)
2nn−1
X
j =1
1 2j + 1
2n − 1 2j
b
2j.
0.2 0.4 0.6 0.8 1.0 1.001
1.002 1.003 1.004
F
Ω(b, 0, . . . , 0) in Ω = {|z
1| + · · · + |z
n| < 1} for n = 2, 3, . . . , 6.
Theorem (B.-Zwonek) For m ≥ 1/2 set Ω = {|z
1|
2m+ |z
2|
2< 1} and w = (b, 0), 0 < b < 1. Then
K
Ω(w )λ(I
ΩK(w )) = P m(1 − b
2) + 1 + b
22(1 − b
2)
3(m − 2)m
2(m + 1)(3m − 2)(3m − 1) , where
P =b
6m+2−m
3+ 2m
2+ m − 2 + b
2m+2−27m
3+ 54m
2− 33m + 6 + b
6m
23m
2+ 2m − 1 + 6b
4m
23m
3− 5m
2− 4m + 4
+ b
2−36m
5+ 81m
4+ 10m
3− 71m
2+ 32m − 4 + 2m
29m
3− 27m
2+ 20m − 4 .
In this domain all values of F
Ωare attained for (b, 0), 0 < b < 1.
0.2 0.4 0.6 0.8 1.0 1.002
1.004 1.006 1.008 1.010
F
Ω(b, 0) in Ω = {|z
1|
2m+ |z
2|
2< 1} for m = 4, 8, 16, 32, 64, 128.
sup
0<b<1
F
Ω(b, 0) → 1.010182 . . . as m → ∞
What is the highest value of F
Ωfor convex Ω?
What can be said the function w 7−→ − log λ(I
ΩA(w ))?
Is it plurisubharmonic?
It does not have to be C
2:
Theorem (B.-Zwonek) If Ω = {|z
1| + |z
2| < 1} and 0 < b ≤ 1/4, λ(I
ΩK((b, b)))
= π
26 30b
8− 64b
7+ 80b
6− 80b
5+ 76b
4− 16b
3− 8b
2+ 1 . λ(I
ΩK((b, b))) is not C
2at b = 1/4.
It is known (Hahn-Pflug) that for 0 < b < 1/2:
K
Ω((b, b)) = 2 8b
4− 6b
2+ 3
π
2(1 − 4b
2)
3.
0.05 0.10 0.15 0.20 0.25 1.002
1.004 1.006 1.008