Bergman Polynomials On An Archipelago

Bergman polynomials on an Archipelago: Estimates, Zeros and Shape
ReconstructionI,II
Björn Gustafssona, , Mihai Putinarb, , Edward B. Saffc , Nikos Stylianopoulosd

a Department of Mathematics, The Royal Institute of Technology, S-10044, Stockholm, Sweden
b Department of Mathematics, University of California at Santa Barbara, Santa Barbara, California, 93106-3080, USA
c Center for Constructive Approximation, Department of Mathematics, Vanderbilt University, 1326 Stevenson Center, 37240
Nashville, USA
d Department of Mathematics and Statistics, University of Cyprus, P.O. Box 20537, 1678 Nicosia, Cyprus
Abstract
Growth estimates of complex orthogonal polynomials with respect to the area measure supported by a
disjoint union of planar Jordan domains (called, in short, an archipelago) are obtained by a combination of
methods of potential theory and rational approximation theory. The study of the asymptotic behavior of
the roots of these polynomials reveals a surprisingly rich geometry, which reflects three characteristics: the
relative position of an island in the archipelago, the analytic continuation picture of the Schwarz function
of every individual boundary and the singular points of the exterior Green function. By way of explicit
example, fine asymptotics are obtained for the lemniscate archipelago |z m − 1| < rm , 0 < r < 1, which
consists of m islands. The asymptotic analysis of the Christoffel functions associated to the same orthogonal
polynomials leads to a very accurate reconstruction algorithm of the shape of the archipelago, knowing only
finitely many of its power moments. This work naturally complements a 1969 study by H. Widom of Szegő
orthogonal polynomials on an archipelago and the more recent asymptotic analysis of Bergman orthogonal
polynomials unveiled by the last two authors and their collaborators.
Key words: Bergman orthogonal polynomials, disjoint Jordan domains, zeros of polynomials, shape
reconstruction, equilibrium measure, Green function, strong asymptotics, geometric tomography
2000 MSC: 42C05, 32A36, 30C40, 31A15, 94A08, 30C70, 30E05, 14H50, 65E05
Archipelago n. (pl. archipelagos or archipelagoes) an extensive group of islands.
I To Christine Chodkiewicz-Putinar, who has enriched and inspired us by adding a second viola to our quartet
II The first author was partially supported by grants from the Swedish Research Council, the Göran Gustafsson Foundation
and the European Network HCAA. The second and third authors were partially supported by the National Science Foundation,
USA, under grants DMS-0701094, DMS-0603828 and DMS-0808093. The fourth author was supported by a University of Cyprus
research grant. All authors are indebted to the Mathematical Research Institute at Oberwolfach, Germany, which provided
exceptional working conditions for our collaborative efforts.
Email addresses: gbjorn@math.kth.se (Björn Gustafsson), mputinar@math.ucsb.edu (Mihai Putinar),
Edward.B.Saff@Vanderbilt.Edu (Edward B. Saff), nikos@ucy.ac.cy (Nikos Stylianopoulos)
Preprint submitted to Elsevier November 13, 2008

1. Introduction
The study of orthogonal polynomials, resurrected recently by many groups of scientists, some departing
from the classical framework of constructive approximation to fields as far as quantum computing or number
theory, does not need an introduction. Maybe only our predilection in the present work for complex analytic
orthogonal polynomials on disconnected open sets needs some justification.
Complex orthogonal polynomials naturally came into focus quite a few decades ago in connection with
problems in rational approximation theory and conformal mapping. The major result, providing strong
asymptotics for Bergman orthogonal polynomials in a domain with analytic Jordan boundary, goes back to
1923 to a landmark article by T. Carleman [3]. About the same time S. Bernstein discovered that the analogue
of Taylor series in non-circular domains (specifically ellipses in his case) is a Fourier expansion in terms of
orthogonal polynomials that are well adapted to the boundary shape, a phenomenon later elucidated in full
generality by J.L. Walsh [36]. Then, it came as no surprise that good approximations of conformal mappings
of simply-connected planar domains bear on the Bergman orthogonal polynomials, that is those with respect
to the area measure supported by these domains. By contrast, the theory of orthogonal polynomials on the
line or on the circle has a longer and glorious history, a much wider area of applications and has attracted
an order of magnitude more attention. For history and details the reader can consult the surveys [22] and
[33] or the monographs [6, 24, 26, 30].
Bergman orthogonal polynomials provide a canonical orthonormal basis in the Bergman space of square
summable analytic functions associated to a bounded Jordan domain of the complex plane. Contrary to
the Hardy-Smirnov space, that is roughly speaking the closure of polynomials in the L2 space with respect
to the arc-length measure on a rectifiable Jordan curve, the functions belonging to the Bergman space do
not possess non-tangential values on the boundary. This makes their study much more challenging, and less
complete as of today. For instance, it is of recent date that the analogues of Blaschke products associated to
the Hardy space of the disk have been discovered: the so-called contractive divisors in the Bergman space
of the disk, see the monograph by Hedenmalm, Korenblum and Zhu [10].
It is our aim to discuss in the present work nth-root and strong estimates for Bergman orthogonal
polynomials on an archipelago, the asymptotics of their zero distribution, and a reconstruction algorithm of
the archipelago from a finite set of the associated power moments. The specific choice of the above problems
and degree of generality were dictated by the present status of the theory of complex orthogonal polynomials.
A brief description of the subjects touched in this article follows. Let G = ∪N j=1 Gj be an archipelago,
that is a finite union of mutually disjoint bounded Jordan domains of the complex plane. The Bergman
orthonormal polynomials with respect to the area measure supported on G:
Pn (z) = λn z n + · · · , λn > 0, n = 0, 1, 2, . . . ,
carry in a refined (one would be inclined to say, aristocratic) manner the information about G. For instance,
simple linear algebra provides a constructive bijection between the sequence {Pn }∞
n=0 and the power moments
(correlation matrix entries) Z
µmn (G) := z n z m dA, m, n ≥ 0, (1.1)
G
where dA stands for the area measure on C. Three major features distinguish Bergman orthogonal polyno-
mials:
(i) An extremality property: Pn /λn is the minimum L2 (G, dA)-norm monic polynomial of degree n,
P∞
(ii) the Bergman kernel K(z, ζ) = j=0 Pj (ζ)Pj (z) collects into a condensed form the (derivatives of the)
conformal mappings from the disk to every connected component Gj ,
2
nP o−1/2
n
(iii) the square root of the Christoffel function Λn (z) := j=0 |Pj (z)|2 is the extremum value
min kqkL2 (G,dA) , q(z) = 1, deg q ≤ n.
We repeatedly use the above characteristic properties, by combining them with general methods of
potential theory and function theory. An important object in our work is the multi-valued function
∗
Φ(z) = exp{gΩ (z, ∞) + igΩ (z)}, z ∈ C \ G,
∗
where gΩ (z, ∞) is the Green function of the exterior domain Ω := C \ G, with a pole at infinity, and gΩ is
any harmonic conjugate of gΩ . We designate the name Walsh function for Φ. At a critical moment in our
proofs, we rely on the pioneering work of Widom [38] that refers to Szegő’s orthogonal polynomials on G
and their intimate relation to the Walsh function Φ. Our Bergman space setting, however, departs in quite a
few essential points from the Hardy-Smirnov space scenario. Both estimates of the growth of Pn (z) and the
limiting distribution of the zero sets of {Pn }∞ n=1 depend heavily on Φ and its analytic continuation across
∂G.
While the estimates for Pn (z) are more or less expected, and only how to prove them might bring
new turns, the zero distribution picture on an archipelago is full of surprises. The uncovering of this rich
geometry began a few years ago, in the work of two of us and collaborators, on the zero distribution of
Bergman orthogonal polynomials on specific Jordan domains, cf. [11, 16, 23]. For example, for the single
Jordan region consisting of the interior of a regular m-gon, all the zeros of Pn , n = 1, 2, . . ., lie on the m
radial lines joining the center to the vertices, for m = 3 and m = 4 (see [13]), while for m ≥ 5 every boundary
point of the m-gon attracts zeros of Pn , as n → ∞ (see [2, Thm. 5]).
As a byproduct of the estimates we have obtained for Λn (z), we propose a very accurate reconstruction-
from-moments algorithm. In general, moment data can be regarded as the archetypal, indirect discrete
measurements available to an observer, of a complex structure. To give a simple indication how moments
appear in geometric tomography, consider a density function ρ(x, y) with compact support in the complex
plane. When performing parallel tomography along a fixed direction θ, one encounters the values of the
Radon transform along the fixed screen
R(ρ)(t, θ) = (ρ(x, y), δ(x cos θ + y sin θ − t))
where δ stands for Dirac’s distribution and (·, ·) is the pairing between test functions and distributions.
Computing then the moments with respect to t yields
Z Z
ak (θ) = tk R(ρ)(t, θ)dt = (x cos θ + y sin θ)k ρ(x, y)dxdy.
R R2
Denoting the power moments (with respect to the real variables) by

Z
σj,k = xj y k ρ(x, y)dxdy, i, j ≥ 0,
R2
we obtain a linear system

k µ
X ¶
k
ak (θ) = cosi θ sink−i θ σi,k−i .
i
i=0
After giving θ a number of distinct values, and noticing that the determinant of the system is non-zero, one
finds by linear algebra the values {σj,k }nj,k=0 . This procedure was used by the first two authors of this paper
3
in an image reconstruction algorithm based on a different integral transform of the original measure, see
[7] and [8]. In a forthcoming work we plan to compare, both computationally and theoretically, these two
different reconstruction-from-moments algorithms.
The paper is organized as follows: Sections 2 and 3 are devoted to necessary background information. We
introduce there the notation, conventions and recall certain facts from potential theory and function theory of
a complex variable that needed for the rest of the work. Sections 4 (Growth Estimates), 5 (Reconstruction
of the Archipelago from Moments) and 6 (Asymptotic Behavior of Zeros) contain the statements of the
main results. In Section 7 we enter into the only computational details available among all archipelagoes:
disconnected lemniscates with central symmetry. Finally, Section 8 contains proofs of previously stated
lemmata, propositions and theorems.
2. Basic concepts
2.1. General notations and definitions

The unit disk, the exterior disk and the extended complex plane are denoted, respectively,
D := {z ∈ C : |z| < 1}, ∆ := {z ∈ C : |z| > 1} ∪ {∞}, C := C ∪ {∞}.
For the area measure in the complex plane we use dA = dA(z) = dxdy, and for the arc-length measure on a
curve we use |dz|. By a measure in general, we always understand a positive Borel measure which is finite
on compact sets. The closed support of a measure µ is denoted by supp µ.
As to curves in the complex plane, we shall use the following terminology: a Jordan curve is a homeo-
morphic image of the unit circle into C. (Thus, every Jordan curve in the present work will be bounded.)
An analytic Jordan curve is the image of the unit circle under an analytic function, defined and univalent
in a neighborhood of the circle. Thus an analytic Jordan curve is by definition smooth. We shall sometimes
need to discuss also Jordan curves which are only piecewise analytic. The appropriate definitions will then
be introduced as needed.
If L is a Jordan curve, we denote by int(L) and ext(L) the bounded and unbounded, respectively,
components of C \ L. By a Jordan domain we mean the interior of a Jordan curve. If E ⊂ C is any set,
Co (E) denotes its convex hull.
The set of polynomials of degree at most n is denoted by Pn .
2.2. Bergman spaces and polynomials

The main characters in our story are the Bergman orthogonal polynomials associated to an archipelago
G := ∪N j=1 Gj , where G1 , ..., GN are Jordan domains with mutually disjoint closures. Set Γj := ∂Gj and
N
Γ := ∪j=1 Γj . For later use we introduce also the exterior domain Ω := C \ G. Note that Γ = ∂G = ∂Ω.
Let {Pn }∞
n=0 denote the sequence of Bergman orthogonal polynomials associated with G. This is defined
as the sequence of polynomials
Pn (z) = λn z n + · · · , λn > 0, n = 0, 1, 2, . . . ,
that are obtained by orthonormalizing the sequence 1, z, z 2 , . . . , with respect to the inner product
Z
hf, gi := f (z)g(z)dA.
G
4
Equivalently, the corresponding monic polynomials Pn (z)/λn , can be defined as the unique monic polynomials
of minimal L2 -norm over G:
1
k Pn kL2 (G) = mn (G, dA) := min kz n + r(z)kL2 (G) , (2.1)
λn r∈Pn−1
where kf kL2 (G) := hf, f i1/2 . Thus,

1
= mn (G, dA).
λn
Let L2a (G) denote the Bergman space associated with G and h·, ·i:
© ª
L2a (G) := f analytic on G and kf kL2 (G) < ∞ .
Note that L2a (G) is a Hilbert space that possesses a reproducing kernel which we denote by K(z, ζ). That
is, K(z, ζ) is the unique function K(z, ζ) : G × G → C such that, for all ζ ∈ G, K(·, ζ) ∈ L2a (G) and
f (ζ) = hf, K(·, ζ)i, ∀ f ∈ L2a (G). (2.2)
Furthermore, due to the reproducing property and the completeness of polynomials in L2a (G) (see Lemma 3.3
below), the kernel K(z, ζ) is given in terms of the Bergman polynomials by
∞
X
K(z, ζ) = Pj (ζ)Pj (z).
j=0
We single out the square root of the inverse of the diagonal of the reproducing kernel of G
1
Λ(z) := p , z ∈ G,
K(z, z)
and the finite sections of K(z, ζ) and Λ(z):

n
X 1
Kn (z, ζ) := Pj (ζ)Pj (z), Λn (z) := p . (2.3)
j=0
Kn (z, z)
We note that the Λn (z)’s are square roots of the so-called Christoffel functions of G.
2.3. Potential theoretic preliminaries

Let Q be a polynomial of degree n with zeros z1 , z2 , . . . , zn . The normalized counting measure of the
zeros of Q is defined by
n
1X
νQ := δzk , (2.4)
n
k=1
where δz denotes the unit point mass at the point z. In other words, for any subset A of C,
number of zeros of Q in A
νQ (A) = .
n
5
Next, given a sequence {σn } of Borel measures, we say that {σn } converges in the weak∗ sense to a
∗
measure σ, symbolically σn −→ σ, if
Z Z
f dσn −→ f dσ, n → ∞,
for every function f continuous on C.

For any finite positive Borel measure σ of compact support in C, we define its logarithmic potential by
Z
1
U σ (z) := log dσ(t).
|z − t|
In particular, if Qn is a monic polynomial of degree n, then

1
U νQn (z) = − log |Qn (z)|.
n
Let Σ ⊂ C be a compact set. Then there is a smallest number γ ∈ R ∪ {+∞} such that there exists a
probability measure µΣ on Σ with U µΣ ≤ γ in C. The (logarithmic) capacity of Σ is defined as cap (Σ) = e−γ
(interpreted as zero if γ = +∞). If cap (Σ) > 0, then µΣ is unique and is called the equilibrium measure of
Σ. For the definition of capacity of more general sets than compact sets see, e.g., [20] and [24]. A property
that holds everywhere, except on a set of capacity zero, is said to hold quasi everywhere (q.e.). For example,
it is known that U µΣ = γ, q.e. on Σ.
Let W denote the unbounded component of C \ Σ. It is known that supp(µΣ ) ⊂ ∂W , µΣ = µ∂W and
cap (Σ) = cap (∂W ). If cap (Σ) > 0, then the equilibrium potential is related to the Green function gW (z, ∞)
of W , with pole at infinity, by
1
U µΣ (z) = log − gW (z, ∞), z ∈ W. (2.5)
cap(Σ)
In our applications ∂W will be a finite disjoint union of mutually exterior Jordan curves (typically Σ = G
or Σ = Γ, W = Ω, ∂W = Γ = ∂Σ, in the notations of Subsection 2.2). Then, every point of ∂W is regular
for the Dirichlet problem in W [20, Thm 4.2.2] and therefore:
(i)
supp µΣ = ∂W, (2.6)
(ii)
1
U µΣ (z) = log , z ∈ Σ. (2.7)
cap(Σ)
If µ is a measure on a compact set Σ with cap (Σ) > 0, the balayage (or “swept measure”) of µ onto ∂Σ
is the unique measure ν on ∂Σ having the same exterior potential as µ, i.e., satisfying
Uν = Uµ in C \ Σ. (2.8)
The potential U ν of ν can be constructed as the smallest superharmonic function in C satisfying U ν ≥ U µ

in C \ Σ. Since U µ itself is competing it follows that, in addition to (2.8), U ν ≤ U µ in all C.
6
2.4. The Green function and its level curves
Returning to the archipelago, let gΩ (z, ∞) denote the Green function of Ω = C \ G with pole at infinity.
That is, gΩ (z, ∞) is harmonic in Ω \ {∞}, vanishes on the boundary Γ of G and near ∞ satisfies
1 1
gΩ (z, ∞) = log |z| + log + O( ), |z| → ∞. (2.9)
cap(Γ) |z|
We consider next what we call the Walsh function associated with Ω. This is defined as the exponential
of the complex Green function,
∗
Φ(z) := exp{gΩ (z, ∞) + igΩ (z, ∞)}, (2.10)
∗
where gΩ (z, ∞) is a (locally) harmonic conjugate of gΩ (z, ∞) in Ω. In the single-component case N = 1,
(2.10) defines a conformal mapping from Ω onto ∆. In the multiple-component case N ≥ 2, Φ is a multi-
valued analytic function in Ω. However, |Φ(z)| is single-valued. We refer to Walsh [36, §4.1] and Widom [38,
§ 4] for comprehensive accounts of the properties of Φ. We note in particular that Φ is single-valued near
∗
infinity and, since gΩ is unique apart from a constant, that it can be chosen so that Φ has near infinity a
Laurent series expansion of the form
1 α1 α2
Φ(z) = z + α0 + + 2 + ··· . (2.11)
cap(Γ) z z
We also note that Φ0 (z)/Φ(z) = 2∂gΩ (·, ∞)/∂z is single-valued and analytic in Ω, with periods
Z Z
1 Φ0 (z) 1 ∂gΩ (z, ∞)
bj := dz = ds, j = 1, 2, . . . , N. (2.12)
2πi Γj Φ(z) 2π Γj ∂n
Here Γj is oriented as the boundary of Gj and the normal derivative is directed into Ω. If Γj is not smooth
PN
the path of integration in (2.12) is understood to be moved slightly into Ω. Note that j=1 bj = 1.
Next we consider for R ≥ 1 the level curves (or equipotential loci) of the Green function,
LR := {z ∈ Ω : gΩ (z, ∞) = log R} = {z ∈ Ω : |Φ(z)| = R} (2.13)
and the open sets
ΩR := {z ∈ Ω : gΩ (z, ∞) > log R} = {z ∈ Ω : |Φ(z)| > R} = ext(LR ),
GR := C \ ΩR = int(LR ).
Note that L1 = Γ, Ω1 = Ω, G1 = G. It follows from the maximum principle that ΩR is always connected.
The Green function for ΩR is given by
gΩR (z, ∞) = gΩ (z, ∞) − log R, (2.14)
hence the capacity of LR (or G R ) is

cap(LR ) = R cap(Γ). (2.15)
Unless stated to the contrary, we hereafter assume that N ≥ 2, i.e. G consists of more than one island.
For small values of R > 1, GR consists of N components, each of which contains exactly one component of
G, while for large values of R, GR is connected (with G ⊂ GR ). Consequently, we introduce the following
sets and numbers:
7
Gj,R := the component of GR that contains Gj , j = 1, 2, . . . , N.
Lj,R := ∂Gj,R , j = 1, 2, . . . , N.
Rj := sup{R : Gj,R contains no other island than Gj }.

R0 := min{R1 , . . . , RN } = sup{R : GR has N exactly components }.
R00 := inf{R : GR is connected } = inf{R : ΩR is simply connected }.
Thus, when 1 < R < R0 , GR is the disjoint union of the domains Gj,R , j = 1, 2, . . . , N and LR consists of the
N mutually exterior analytic Jordan curves Lj,R , j = 1, 2, . . . , N , while for R > R00 , we have G1,R = G2,R =
· · · = GN,R = GR and LR is a single analytic curve.
It is well-known that gΩ (z, ∞) has exactly N − 1 critical points (multiplicities counted), i.e., points where
the gradient ∇gΩ (z, ∞), or equivalently Φ0 /Φ = 2∂gΩ (·, ∞)/∂z, vanishes. These critical points show up as
singularities of LR , which are points of self-intersection. Such singularities must appear when LR changes
topology. It follows that there are no critical points in GR0 \ G, at least one critical point on each LRj ,
j = 1, 2, . . . , N (one of them is LR0 ), at least one on LR00 and no critical point in ΩR00 . Any Lj,R that does
not contain a critical point is an analytic Jordan curve. In particular, this applies whenever 1 < R < R0 or
R00 < R < ∞.
When R ≥ R00 , Φ is the unique conformal map of ΩR onto ∆R := {w : |w| > R} that satisfies (2.11) near
infinity.
2 G3
LR
R > R00
1 LR0
LR00
LR0
0 G1 G2
−1
−1 0 1 2 3
Figure 1: Green level curves
In Figure 1 we illustrate the three different types of level curves LR0 , LR00 and LR with R > R00 , introduced
above.
Remark 2.1. The level curves in Figure 1 were computed by means of Trefethen’s MATLAB code manydisks.m
[34]. This code provides an approximation to the Green function gΩ (z, ∞) in cases when G consists of a
finite number of disks, realizing an algorithm given in [5].
8
Consider now the N Hilbert spaces L2a (Gj ) defined by the components Gj , j = 1, 2, . . . , N , and let
Gj
K (z, ζ), j = 1, 2, . . . , N , denote their respective reproducing kernels. Then, it is easy to verify that the
kernel K(z, ζ) is related to K Gj (z, ζ) as follows:
½ G
K j (z, ζ) if z, ζ ∈ Gj ,
K(z, ζ) = (2.16)
0 if z ∈ Gj , ζ ∈ Gk , j 6= k.
In view of (2.16), we can express K(z, ζ) in terms of conformal mappings ϕj : Gj → D, j = 1, 2, . . . , N .

This will help us to determine the singularities of K(·, ζ) and, in particular, whether or not this kernel has
a singularity on ∂Gj . This is so because, as it is well-known (see e.g. [6, p. 33]),
ϕ0j (z)ϕ0j (ζ)

K Gj (z, ζ) = h i2 , z, ζ ∈ Gj , j = 1, 2, . . . , N. (2.17)
π 1 − ϕj (z)ϕj (ζ)
By saying that a function analytic in Gj has a singularity on ∂Gj , we mean that there is no open neighborhood
of Gj in which the function has an analytic continuation.
3. Preliminaries
3.1. The Schwarz function of an analytic curve and extension of harmonic functions
Let Γ be a Jordan curve. Then Γ is analytic if and only if there exists an analytic function S(z), the
Schwarz function of Γ, defined in a full neighborhood of Γ and satisfying
S(z) = z̄ for z ∈ Γ;
see [4] and [25]. The map z 7→ S(z) is then the anticonformal reflection in Γ, which is an involution (i.e., is
its own inverse) on a suitably defined neighborhood of Γ. In particular, S 0 (z) 6= 0 in such a neighborhood.
If u is a harmonic function defined at one side of an analytic Jordan curve Γ and u has boundary values
zero on Γ, then u extends as a harmonic function across Γ by reflection. In terms of the Schwarz function
S(z) of Γ the extension is given by
u(z) = −u(S(z)) (3.1)
for z on the other side of Γ (and close to Γ). Conversely we have the following:
Lemma 3.1. Let Γ be a Jordan curve and let u be a (real-valued) harmonic function defined in a domain
D containing Γ such that, for some constant c > 0, there holds:
(i) u = 0 on Γ,
(ii) |u| → c as z → ∂D,
(iii) ∇u 6= 0 in D,
where ∇u denotes the gradient of u. Then Γ is an analytic curve, the Schwarz function S(z) of Γ is defined in
all D, and z 7→ S(z) maps D onto itself. Moreover, u and S(z) are related by (3.1). In particular z 7→ S(z)
maps a level line u = α of u onto the level line u = −α.
9
We note that the Schwarz function is uniquely determined by Γ, but u is not; there are many different
harmonic functions that vanish on Γ. A domain which is mapped into itself by z → 7 S(z) will be called a
domain of involution for the Schwarz reflection.
Example 3.1. Assume that, under our main assumptions, one of the components of Γ, say Γ1 , is analytic.
Then the Green function u(z) = gΩ (z, ∞) extends harmonically, by the Schwarz reflection (3.1), from Ω into
G1 . We keep the notation gΩ (z, ∞) for so extended Green function. Recall that the level lines reflect to level
lines, so that for R > 1 close enough to one, L1,R is reflected to the level line
1
L1, R1 = {z ∈ G1 : gΩ (z, ∞) = − log R} = {z ∈ G1 : |Φ(z)| = }
R
of the extended Green function (and extended Φ). Generally speaking, whenever applicable we shall keep
notations like Lj,ρ , Gj,ρ , Lρ , Ωρ etc. for values ρ < 1 in case of analytic boundaries.
As was previously remarked, u(z) = gΩ (z, ∞) has no critical points in the region G1,R1 \ G1 . It follows
then from (3.1) that if the Green function extends harmonically to a region G1 \ G 1,ρ with R11 ≤ ρ < 1, then
it has no critical points there, and the region D = G1, ρ1 \ G 1,ρ is symmetric with respect to Schwarz reflection
and is a region of the kind D discussed in Lemma 3.1.
3.2. Regular measures

The class Reg of measures of orthogonality was introduced by Stahl and Totik [27, Definition 3.1.2] and
shown to have many desirable properties. Roughly speaking, µ ∈ Reg means that in an “n-th root sense”,
the sup-norm on the support of µ and the L2 -norm generated by µ have the same asymptotic behavior (as
n → ∞) for any sequence of polynomials of respective degrees n. It is easy to see that area measure enjoys
this property.
Lemma 3.2. The area measure dA|G on G belongs to the class Reg.
Lemma 3.2 yields the following n-th root asymptotic behavior for the Bergman polynomials Pn in Ω.
Proposition 3.1. The following assertions hold:

(a)
1
lim λ1/n
n = . (3.2)
n→∞ cap(Γ)
(b) For every z ∈ C \ Co(G) and for any z ∈ Co(G) \ G not a limit point of zeros of the Pn ’s, we have
lim |Pn (z)|1/n = |Φ(z)|. (3.3)

n→∞
The convergence is uniform on compact subsets of C \ Co(G).

(c)
lim sup |Pn (z)|1/n = |Φ(z)|, z ∈ Ω, (3.4)
n→∞
locally uniformly in Ω.
10
(d)
1 Pn0 (z) Φ0 (z)
lim = , (3.5)
n→∞ n Pn (z) Φ(z)
locally uniformly in C \ Co(G).
The first three parts of the proposition follow from Theorems 3.1.1, 3.2.3 of [27] and from Theorem III.4.7
of [24], in combination with the results of [1], because Ω is regular with respect to the Dirichlet problem; see
e.g. [20, p. 92]. The last assertion (d) is immediate from (b).
Another fundamental property of Bergman polynomials, whose proof involves a simple extension of the
simply-connected case treated in Theorem 1, Section 1.3 of Gaier [6] is the following.
Lemma 3.3. Polynomials are dense in the Hilbert space L2a (G). Consequently, for fixed ζ ∈ G,
∞
X
K(z, ζ) = Pn (ζ)Pn (z), (3.6)
n=0
locally uniformly with respect to z in G.

The analytic continuation properties of K(z, ζ) play an essential role in the analysis. The following
notation will be useful in this regard. If f is an analytic function in G, we define
ρ(f ) := sup {R : f has an analytic continuation to GR } . (3.7)
Note that 1 ≤ ρ(f ) ≤ ∞. The following important lemma, which is an analogue of the Cauchy-Hadamard
formula, is due to Walsh.
Lemma 3.4. Let f ∈ L2a (G) . Then,
1
lim sup |hf, Pn i|1/n = . (3.8)
n→∞ ρ(f )
Moreover,
∞
X
f (z) = hf, Pn iPn (z),
n=0
locally uniformly in Gρ(f ) .
The result is given in Walsh [36, pp. 130–131] (see also [18, Thm 2.1]) for a single Jordan region and, as
Walsh asserts, is immediately extendable to several Jordan regions.
By applying Lemma 3.4 to f = K(·, ζ), and by using the reproducing property (2.2), in conjunction with
(2.16) and (3.6), we obtain:
Corollary 3.1. Let j be fixed, 1 ≤ j ≤ N . Then for any ζ ∈ Gj ,
1 1
lim sup |Pn (ζ)|1/n = = , (3.9)
n→∞ ρ (K(·, ζ)) min{ρ (K Gj (·, ζ)) , Rj }
where (as previously defined) Rj > 1 is the largest R such that the component Gj,R of GR containing Gj
contains no other Gk . In particular,
lim sup |Pn (ζ)|1/n = 1 (3.10)
n→∞
Gj
if and only if K (·, ζ0 ) has a singularity on ∂Gj , for some (and then for every) ζ0 ∈ Gj .
11
The last statement is based on the observation, from (2.17), that the property of K Gj (·, ζ0 ) having a singu-
larity on ∂Gj is independent of the choice of ζ0 (within Gj ). We remark also that the appearance of Rj in
(3.9) is essential since, for R > Rj , the component Gj,R contains an open set where K(·, ζ) is identically zero
(recall (2.16)) and hence not an analytic continuation of K Gj (·, ζ). Corollary 3.1 will be further elaborated
in Theorem 6.1.
Corollary 3.1 describes a basic relationship between the orthogonal polynomials {Pn (ζ)}∞ n=0 and the
kernel function K(·, ζ) which will play an essential role in deriving our zero distribution results for the
sequence {Pn }∞ n=1 .
Remark 3.1. A well-known result by Fejér asserts that the zeros of orthogonal polynomials with respect to
a compactly supported measure σ are contained in the closed convex hull of supp σ. This result was refined
by Saff [22] to the interior of the convex hull of supp σ, provided this convex hull is not a line segment.
Consequently, we see that all the zeros of the sequence {Pn }∞ n=1 are contained in the interior of convex hull
of G. This fact should be coupled with a result of Widom [37] to the effect that, on any compact subset E
of Ω and for any n ∈ N, the number of zeros of Pn on E is bounded independently of n.
3.3. Carleman estimates

We continue this section by recalling certain results due to T. Carleman and P.K. Suetin, regarding the
asymptotic behavior of the Bergman polynomials in the case where G consists of a single component (i.e.
for N = 1). In this case the Walsh function (2.10) coincides with the unique conformal map Φ : Ω → ∆
satisfying (2.11).
The first result requires the boundary Γ to be analytic (hence the conformal map Φ has an analytic and
univalent continuation across Γ inside G) and is due to Carleman [3]; see also [6, p. 12].
Theorem 3.1. Assume that Γ is an analytic Jordan curve and let ρ, 0 < ρ < 1, be the smallest index for
which Φ is conformal in Ωρ . Then,
r
n+1 1
λn = {1 + O(ρ2n )}, (3.11)
π cap(Γ)n+1
and r
n+1 0
Pn (z) = Φ (z)Φn (z){1 + An (z)}, (3.12)
π
where ½ √ n
O(
√ n)ρ , n if z ∈ Ω,
An (z) = (3.13)
O(1/ n) (ρ/r) , if z ∈ Lr , ρ < r < 1.
The second result which is due to Suetin [30, Thms 1.1 & 1.2], requires that Γ can be parameterized
with respect to the arc-length, so that the defining function has a p-th order derivative (where p is a positive
integer) in a Hölder class of order α. We express this by saying Γ is C p+α -smooth. (In particular, this
implies that Γ can have no corners.)
Theorem 3.2. Assume that Γ is C (p+1)+α -smooth, with p + α > 1/2. Then,
r µ ¶
n+1 1 1
λn = {1 + O }, (3.14)
π cap(Γ)n+1 n2p+2α
and r µ ¶
n+1 0 log n
Pn (z) = Φ (z)Φn (z){1 + O }, z ∈ Ω. (3.15)
π np+α
12
We emphasize that the above two theorems concern only the case when N = 1. We also remark that
for the case when Γ is analytic, E. Miña-Dı́az [15] has recently derived an improved version of Carleman’s
theorem for the special case when Lρ is a piecewise analytic Jordan curve without cusps.
3.4. Comparison of area and line integrals of polynomials

The following observation is due to Suetin [29]; see also [30, p. 38].
Lemma 3.5. Let G be a Jordan domain with C 1+α -smooth boundary. Then there exists a positive constant
C, with the property that, for every polynomial Qn of degree at most n, there holds
√
kQn kL2 (Γ) ≤ C n + 1 kQn kL2 (G) ,
where k · kL2 (Γ) denotes the L2 -norm on Γ with respect to |dz|.

The proof in [29] uses the following analogue of Bernstein’s inequality (a result Suetin attributes to S.
Yu. Al’per):
kQ0n kL2 (G) ≤ C n kQn kL2 (G)
and leads to similar inequalities for Lp , 1 < p < ∞, or uniform norms.
4. Growth Estimates
The main results of this article are stated in this and the next three sections. Their proofs are given in
Section 8.
We recall the notation and definitions in Section 2, in particular the definition of the archipelago G :=
∪N
j=1 Gj consisting of the union of N Jordan domains in C, with boundaries Γj := ∂Gj . We also recall that
Ω := C \ G and note Γ := ∪N j=1 Γj = ∂G = ∂Ω.
Theorem 4.1. Assume that every curve Γj constituting Γ is C 2+α -smooth. Then there exists a positive
constant C1 such that r
n+1 1
λn ≤ C 1 , n ∈ N. (4.1)
π cap(Γ)n+1
In addition, if every Γj is analytic, j = 1, 2, . . . , N , then there exists a positive constant C2 such that
r r
n+1 1 n+1 1
C2 ≤ λn ≤ C1 , n ∈ N. (4.2)
π cap(Γ)n+1 π cap(Γ)n+1
The following estimate for the diagonal K(z, z), z ∈ G, of the reproducing kernel follows from classical
estimates for the boundary growth of the Bergman kernel of a simply-connected domain, obtained via
conformal mapping techniques. More precisely, by using the results for the hyperbolic metric presented by
Hayman in [9, pp. 682–692], which require no smoothness for the boundary curve, and recalling (2.17), it is
easy to verify the following double inequality, holding for every j, j = 1, 2, . . . , N :
1 1 1 1
2 ≤ K Gj (z, z) ≤ 2 , z ∈ Gj , (4.3)
16π dist (z, Γj ) π dist (z, Γj )
13
p
Thus K(z, z) = K Gj (z, z), z ∈ Gj , inherits the same estimates and, clearly, the function Λ(z) := 1/ K(z, z)
satisfies √ √
π dist(z, Γj ) ≤ Λ(z) ≤ 4 π dist(z, Γj ), z ∈ Gj . (4.4)
(Above and throughout this article dist(z, Γj ) stands for the Euclidean distance of z from the boundary Γj .)
It is always useful to recall that the monic orthogonal polynomials Pn (z)/λn , n = 0, 1, . . ., satisfy a
minimum distance condition with respect to the L2 -norm on G, in the sense that
1 Pn
= k kL2 (G) = min kz n + an−1 z n−1 + ... + a0 kL2 (G) . (4.5)
λn λn a0 ,a1 ,...,an−1
Similarly, the square root of the Christoffel functions Λn (z), n = 0, 1, . . ., defined by by (2.3), can be described
as
Λn (z) = min kpkL2 (G) ; (4.6)
p∈Pn ,
p(z)=1
cf. Lemma 8.1 below. Based on the above two simple

q extremal properties, we derive the following comparison
G G
between Λn (z) and the functions Λn j (z) := 1/ Kn j (z, z) associated with each individual island Gj .
Theorem 4.2. For every j = 1, 2, . . . , N and any n ∈ N,
ΛG
n (z) ≤ Λn (z),
j
z ∈ C. (4.7)
In addition, if Γj is analytic, then there exist a sequence {γn }∞
n=0 ,
with 0 < γn < 1 and lim γn = 0
n→∞
geometrically, and a number m ∈ N, m ≥ 1, such that for any n ∈ N,
1 − γn
Λmn (z) ≤ ΛG
n (z),
j
z ∈ Gj . (4.8)
2
Let Φj denote the normalized, like (2.11), exterior conformal map Φj : C \ Gj → ∆. The growth of
G
Λn j (z)
inside the island Gj is described as follows.
Theorem 4.3. Fix j, j = 1, 2, . . . , N, and assume that Γj is analytic. Then there exist positive constants
C1 , C2 and ρ < 1 such that for any n ∈ N,
µ ¶
Gj Gj n 1
0 < Λn (z) − Λ (z) ≤ C1 |Φj (z)| dist(z, Γj ) + , z ∈ Gj \ Gj,ρ . (4.9)
n
Moreover, √
2π
lim nΛG
n (z) =
j
, (4.10)
n→∞ |Φ0j (z)|
uniformly for z ∈ Γj .
Furthermore, if every curve constituting Γ is analytic then
C3 ≤ nΛn (z) ≤ C4 , z ∈ Γ, (4.11)
and
C6
C5 dist(z, Γ) δ(z) ≤ Λn (z) ≤ √ , / G,
n ∈ N, z ∈ (4.12)
n|Φ(z)|n
where C3 , C4 , C5 , C6 are positive constants and
|Φ(z)|2 − 1 1
δ(z) = p .
|Φ(z)| (n + 1)|Φ(z)| (|Φ(z)|2 − 1) + 1
2n
14
An estimate for Λn (z) on Γ which is finer than (4.11), in the sense that it coincides with (4.10) for the
case of a single island, and under weaker smoothness conditions on Γ, is presented in [32], where asymptotics
of Christoffel functions defined by more general measures on C are considered.
Finally, we derive the following exterior estimates for Bergman polynomials.
Theorem 4.4. Assume that every curve constituting Γ is analytic. Then the following hold:
(i) There exists a positive constant C, so that
C √
|Pn (z)| ≤ n|Φ(z)|n , z ∈
/ G. (4.13)
dist(z, Γ)
(ii) For every ² > 0 there exist a constant C² > 0, such that
√
|Pn (z)| ≥ C² n|Φ(z)|n , dist(z, Co(G)) ≥ ².
Note that in the region Co(G) \ G the orthogonal polynomials may have zeros (as the case of the lemnis-
cates considered in Section 7 illustrates).
5. Reconstruction of the archipelago from moments
The present section contains a brief description of a shape reconstruction algorithm. This algorithm is
motivated by the estimates established in the previous sections. The comparison of the speed of convergence
and accuracy of this approximation scheme with other known ones (see e.g. [8]) will be analyzed in a separate
work.
The algorithm is based on the following observations:
Remark 5.1.
(i) From (4.4) we see that the function Λ(z) is bounded from below and above in G by constants times
the distance of z to the boundary. Consequently, its truncation
1
Λn (z) = pPn (5.1)
k=0 |Pk (z)|2
approximates the distance function to Γ in G. Furthermore, on Γ and in Ω Λn decays to zero at certain
rates, as n → ∞. More precisely, a close inspection of the inequalities in Theorems 4.2 and 4.3, in
conjunction with (4.4), reveals the following asymptotic behavior of Λn (z) in C:
√
(a) π dist(z, Γ) ≤ Λn (z), z ∈ G;
√
(b) Λn (z) ≤ C dist(z, Γ), z ∈ G ∩ Ωρ , for some 0 < ρ < 1 and C ≥ π;
1
(c) Λn (z) ³ , z ∈ Γ;
n
1
(d) Λn (z) ³ √ , z ∈ Ω.
n|Φ(z)|n
(ii) In order to construct Λn we need to have available the finite section {P0 , P1 , . . . , Pn } of Bergman
polynomials, and this can be determined by means of the Gram-Schmidt process, requiring only the
power moments (1.1), of degree less or equal than n in each variable.
15
(iii) For any n = 1, 2, . . ., all the zeros of Pn (z) lie in the interior of the convex hull of G; see Remark 3.1.
The expression A ³ B means that C1 B ≤ A ≤ C2 B for positive constants C1 and C2 .
Consequently, Remark 5.1 supports the following algorithm for reconstructing the archipelago G, by
using a given finite set of the associated power moments
Z
i j
µij := hz , z i = z i z j dA(z), i, j = 0, 1, . . . , n.
G
Reconstruction Algorithm
1. Use an Arnoldi version of the Gram-Schmidt process, in the way indicated in [28], to construct the
Bergman polynomials {Pk }nk=0 from µij , i, j = 0, 1, . . . , n. This involves at the k-step the orthonormal-
ization of the set {P0 , P1 , . . . , Pk−1 , zPk−1 }, rather than the set of monomials {1, z, . . . , z k−1 , z k }, as in
the standard Gram-Schmidt process.
2. Plot the zeros of Pn , n = 1, 2, . . . , n.
3. Form Λn (z) as in (5.1).
4. Plot the level curves of the function Λn (x + iy) on a suitable rectangular frame for (x, y) that surrounds
the plotted zero set.
Regarding the stability of the Gram-Schmidt process in the Reconstruction Algorithm, we note a fact
that pointed out in [28]. That is the Arnoldi version of the Gram-Schmidt does not suffer from the se-
vere ill-conditioning associated with its ordinary use; see, for instance, the theoretical and numerical evi-
dence reported in [19]. This feature of the Arnoldi Gram-Schmidt has enabled us to compute accurately
Bergman polynomials for degrees as high as 160. We also note that the use of orthogonal polynomials in a
reconstruction-from-moments algorithm, was first employed in [28]. However, the algorithm of [28] is only
suitable for the single island case N = 1.
Applications of the Reconstruction Algorithm are illustrated in the following six examples. In each
example, the only information used from the associated archipelago G was its power moments. The resulting
plots indicate a remarkable fitting, even in the case of non-smooth boundaries, for which our theory, as stated
in Section 4, does not apply.
0,8
0,4
0
0 1 2 3 4
-0,4
-0,8
Figure 2: Level curves of Λ100 (x + iy), on {(x, y) : −2 ≤ x ≤ 5, −2 ≤ y ≤ 2}, with G as in Example 6.1.
16
1
0,5
-3 -2 -1 0 1 2 3
0
-0,5
-1
Figure 3: Level curves of Λ100 (x + iy), on {(x, y) : −4 ≤ x ≤ 4, −2 ≤ y ≤ 2}, with G as in Example 6.2.
1 x
0,5
0 2 4 6
y 0
-0,5
-1
Figure 4: Level curves of Λ100 (x + iy), on {(x, y) : −2 ≤ x ≤ 8, −2 ≤ y ≤ 2}, with G as in Example 6.3, case (i).
0,5
0
0 0,5 1 1,5 2 2,5 3
-0,5
-1
Figure 5: Level lines of Λ100 (x + iy), on {(x, y) : −1 ≤ x ≤ 4, −2 ≤ y ≤ 2}, with G as in Example 6.5.
0,5
0
0 0,5 1 1,5 2 2,5 3 3,5
-0,5
-1
Figure 6: Level lines of Λ100 (x + iy), on {(x, y) : −1 ≤ x ≤ 6, −2 ≤ y ≤ 2}, with G as in Example 6.4.
17
2,5 2,5
2 2
1,5 1,5
1 1
0,5 0,5
0 0
-1 0 1 2 3 -1 0 1 2 3
-0,5 -0,5
-1 -1
2,5 2,5
2 2
1,5 1,5
1 1
0,5 0,5
0 0
-1 0 1 2 3 -1 0 1 2 3
-0,5 -0,5
-1 -1
Figure 7: Level lines of Λn (x + iy), for the values of n (from left two right) 25, 50, 75 and 100, on {(x, y) : −1 ≤ x ≤ 4, −2 ≤
y ≤ 3}, with G formed by the three disjoint disks of Example 6.6.
18
6. Asymptotic behavior of zeros
6.1. General statements

The first result of this section is our general theorem on the asymptotic behavior of the zeros of the
Bergman polynomials {Pn }∞ n=1 on an archipelago of N Jordan domains. It is established under the general
assumptions made at the beginning of Section 2.2. In particular we note that, unlike the theory presented
in Section 4, no extra smoothness is required for the boundary curves Γj here. The result below, which is
valid for any N ≥ 1, requires some special attention for the single island case N = 1.
Theorem 6.1. Consider the following extension of the Green function gΩ (·, ∞) to all C:
(
gΩ (z, ∞), z ∈ Ω,
h(z) = (6.1)
− log ρ(K(·, z)), z ∈ G,
(recall (3.7)) and define

1
β = βG := ∆h, (6.2)
2π
where the Laplacian is taken in the sense of distributions. Let C denote the set of weak* cluster points of the
counting measures {νPn }∞n=1 , i.e., the set of measures σ for which there exists a subsequence Nσ ⊂ N such
∗
that νPn −→ σ, as n → ∞, n ∈ Nσ . The following assertions hold.
(i) The function h is harmonic in Ω, subharmonic in all C; hence β is a positive unit measure with support
contained in G. In addition, if N ≥ 2, then h is continuous and bounded from below. If N = 1, then
h can take the value −∞ at most at two points, and outside these points h is continuous.
(ii)
1
U β (z) = log − h(z), z ∈ C. (6.3)
cap (Γ)
and balayage of β onto Γ gives the equilibrium measure µΓ of Γ:
(
U β ≥ U µΓ in C,
(6.4)
U β = U µΓ in Ω.
(iii)
1
lim sup log |Pn (z)| = h(z), z ∈ C, (6.5)
n→∞ n
lim inf U νPn (z) = U β (z), z ∈ C. (6.6)
n→∞
Moreover, in C \ Co(G) these equalities hold with lim sup and lim inf replaced by lim.
(iv) The set of cluster points C is nonempty, and for any σ ∈ C,
(
U σ ≥ U β in C,
(6.7)
U σ = U β in the unbounded component of C \ supp β.
19
(v) The measure β is the lower envelope of C in the sense that
U β = lsc ( inf U σ ),
σ∈C
where “lsc” denotes lower semicontinuous regularization. (This means that U β is the supremum of all
lower semicontinuous functions that are ≤ inf σ∈C U σ .) In addition, if D is any component of C\supp β,
then for any σ ∈ C either U σ > U β in D or U σ = U β in D; and there exists a σ ∈ C such that the
latter holds.
(vi) If C has only one element, then this is β and
∗
νPn −→ β, n → ∞, n ∈ N, (6.8)
i.e., the full sequence converges to β.

(vii) Assume that β satisfies
(a) supp β is a nullset with respect to area measure,
(b) C \ supp β is connected.
Then β is the unique element in C; hence (6.8) holds. If (a) holds and (in place of (b))
(c) C \ supp β has at most two components,
then β ∈ C.
Remark 6.1. The measure β = βG is canonically associated to G via the Bergman kernel. Constructive
formulas for βG (or rather its potential) will be given in the proof (e.g. (8.20)–(8.22)) and will be further
elaborated in the examples of Section 6.2.
Remark 6.2. Well-known properties for any σ ∈ C follow immediately from (ii) and (iv): That is, U σ = U µΓ
in Ω, supp σ ⊂ G and balayage of σ onto Γ gives the equilibrium distribution µΓ (see e.g. [24, Thm III.4.7]).
Remark 6.3. We know of no example where β isn’t itself in C. However it remains an open question whether
it is always so.
Remark 6.4. A measure β satisfying (6.4) together with (a) and (b) in (vii) may be viewed as a potential
theoretic skeleton for µΓ (or “Madonna body”, in view of a common shape of supp β; cf. [11, 16]).
Remark 6.5. When N = 1, h may take the value −∞ at one or two points. Note that, by (6.1), h(a) = −∞
if and only if K(z, a) is an entire function of z. With G = D we have h(z) = log |z|, i.e., one pole for h. An
example with two poles is the following.
Choose a number A > 1 and let G be the image of the unit disk under the conformal map
1 A+w
ψ(w) = log ,
2 A−w
the branch chosen so that ψ(0) = 0. The inverse map is
ϕ(z) = A tanh z,
π
which is meromorphic in the entire complex plane. Here ψ maps the disk |w| < A onto the strip |Im z| < 4.
Hence G, which is the image of |w| < 1, is a subdomain of that strip (a kind of an oval).
20
The function ϕ does not attain the values ±A anywhere in the complex plane and the set ϕ|−1C (1/ϕ(ζ)),
which will play an important role in the proof of the theorem, may therefore be empty for up to two values
A2 +1
of ζ ∈ G. In fact, this occurs for ζ = ±a ∈ G, where a = 12 log A 2 −1 > 0. At these points, h(±a) = −∞,
4
K(z, ±a) = A π−1 e±2z . One also finds that β is a measure supported on the line segment [−a, a] and hence
is a Madonna body.
We call a boundary curve Γj singular if some conformal map ϕj : Gj → D does not extend analytically
to a full neighborhood of Gj , i.e., if ρ(ϕj ) = 1, or equivalently if ρ(K(·, z)) = 1, z ∈ Gj ; see (2.16) and
(2.17). Clearly, this property is independent of the choice of the conformal map ϕj . Note that a boundary
component that is not singular in the above sense still need not be fully smooth: it may be piecewise analytic
but have certain kinds of corners so that ϕj extends analytically across Γj but the extension is not univalent.
This would be the case, for instance, if Gj is a rectangle.
Corollary 6.1. For each j = 1, . . . , N , the following statements are equivalent:
(i) Γj is singular.
(ii) β|Gj = µΓ |Gj .
(iii) There is a subsequence N = Nj ⊂ N such that, with V any neighborhood of Gj not meeting the other
islands (e.g., V = Gj,Rj ),
∗
νPn |V −→ µΓ |V , n → ∞, n ∈ N . (6.9)
Clearly, under the conditions of the above corollary a certain proportion of the zeros of the Bergman
polynomials converge to the part of the equilibrium measure located on Γj . By a reasoning as in deriving
(8.24) in the proof of Theorem 6.1 below, we conclude that this proportion is
Z
dµΓ = bj ,
Γj
where bj is the period in (2.12). Thus, we easily deduce the following:

Corollary 6.2. If, for a particular j = 1, . . . , N , Γj is singular, then there is a exists a subsequence {Pn }n∈N ,
such that Pn = Qk Rk , deg(Qk ) = nk , where
nk ∗
νQk −→ µΓ |Γj , as n → ∞, n∈N (6.10)
n
and
nk
→ bj .
n
As stated in (iv) of Theorem 6.1, if σ is a weak* cluster point of the measures {νPn }∞ n=1 then: (a)
supp σ ⊂ G and (b) the balayage of σ onto Γ equals the equilibrium distribution µΓ . The following corollary
shows that the equilibrium distribution is also obtained if weak* convergence and balayage are applied in
the opposite order.
Corollary 6.3. Let Bal (νPn ) denote the measure obtained by balayage of νPn |G onto Γ while keeping νPn |C\G
unchanged. Then
∗
Bal (νPn ) −→ µΓ as n → ∞.
21
6.2. Case studies
In this subsection we make more explicit Theorem 6.1 and its corollaries, and we illustrate them by means
of a number of representative cases and examples.
Case I: Two singular boundaries.

Here N = 2 and ρ(ϕj ) = 1, j = 1, 2, for any two conformal maps ϕj : Gj → D. By Corollary 6.1, β
equals the equilibrium measure µΓ of G and there exists, for each island Gj , a subsequence of νPn which
converges to µΓ in a neighborhood of Gj . However, we do not know whether there necessarily exists a
common subsequence for the two islands.
Case II: One singular boundary and one analytic boundary.

Assume that Γ1 is singular and Γ2 is analytic. Then in terms of two specific conformal maps ϕj : Gj → D,
j = 1, 2: (a) ϕ1 has no analytic continuation beyond Γ1 , (b) ϕ2 extends analytically as a univalent function
to some domain containing G2 . Since Γ2 is an analytic Jordan curve, it possesses a Schwarz function, which
is given by
S2 (z) = ϕ−1
2 (1/ϕ2 (z)).
In order to formulate a particular statement we assume further that ϕ2 remains analytic and univalent
throughout G2,R0 . This implies that gΩ (·, ∞) extends by Schwarz reflection up to the level line L2, 10 ; see
R
(3.1) and the terminology in Example 3.1. Moreover, the domain
D2 := G2,R0 \ G2, 1
R0
is connected and is a domain of involution of the Schwarz reflection z 7→ S2 (z).

Set
E = G1 ∪ G2, 10 .
R
It follows that the multi-valued function

(
Φ(z) if z ∈ C \ G,
b
Φ(z) := ± ³ ´ (6.11)
1 Φ S2 (z) if z ∈ G2 \ G2, 1 .
R0
is (locally) analytic in C \ E and (locally) continuous on C \ E. It also follows from the expression (8.23) of
ρ(K(·, z)) appearing in the proof of Theorem 6.1, by taking into account (8.18) and (8.19), that

 1 if z ∈ G1 ,
ρ(K(·, z)) = exp{−gΩ (z, ∞)} if z ∈ G2 \ G2, 10 , (6.12)
 R
R0 if z ∈ G2, 10 .
R
The relations in (6.11) and (6.12) yield at once, in view of Proposition 3.1 and Corollary 3.1, the n-th
root asymptotic behavior of {Pn }∞
n=1 in C:

 1 if z ∈ G1 ,
lim sup |Pn (z)|1/n = b
|Φ(z)| if z, ∈ C \ E, (6.13)
n→∞  1
R 0 if z ∈ G 1 .
2, 0
R
In addition, these relations provide more detailed information for the potential U β of the canonical measure
β, and thus for the counting measures {νPn }∞n=1 .
22
Corollary 6.4. Under the assumption and notations of Case II, we have:
 1
 log cap (Γ)
 if z ∈ G1 ,
1
β
U (z) = log cap (Γ) − gΩ (z, ∞) if z ∈ C \ E, (6.14)

 log R0
cap (Γ) if z ∈ G2, 10 .
R
In particular,
(i) supp β = ∂E.
(ii) For any weak* cluster point σ of {νPn }, supp σ ⊂ E, and
U σ (z) = U β (z), z ∈ C \ E.
(iii) There is a subsequence N ⊂ N such that, with V any neighborhood of G1 or G 2, 10 not meeting the
R
other island,
∗
νPn |V −→ β|V , n → ∞, n ∈ N . (6.15)
Hence, every point of ∂E = Γ1 ∪ L2, 1 belongs to supp σ, for some weak* cluster point σ of {νPn }∞
n=1 .
R0
The corollary is illustrated in the following example.

Example 6.1. Bergman polynomials for G = G1 ∪ G2 , with G1 the canonical pentagon with vertices at the
fifth roots of unity and G2 = {z : |z − 7/2| < 2/3}.
The zeros of the associated Bergman polynomials Pn , for n = 80, 90 and 100 are shown in Figure 8. In
the same figure we also depict the critical line LR0 and the curve L2, 10 . Note that L2, 10 is simply the inverse
R R
image of L2,R0 with respect to the circle {z : |z − 7/2| = 2/3}.
1
0,5
0
-1 0 1 2 3 4
-0,5
-1
Figure 8: Zeros of Bergman polynomials Pn of Example 6.1, for n = 80, 90 and 100.
Case III: Two analytic boundary curves. This is the case N = 2, where both Γ1 and Γ2 are analytic curves.
Example 6.2. Bergman polynomials for the union of the disks: G1 = {z : |z + 2| < 1} and G2 := {z :
|z − 3| < 2/3}.
23
Let S1 and S2 denote the Schwarz functions defined by Γ1 and Γ2 . (Note that the Schwarz function for
the circle {z : |z − a| = r} is simply S(z) = r2 /(z − a) + a.) Clearly, the Green function gΩ extends by
Schwarz reflection to the set
D = (G1,R0 \ G1, 10 ) ∪ (G2,R0 \ G2, 10 ), (6.16)
R R
and the multi-valued function

(
Φ(z) if z ∈ C \ G,
b
Φ(z) := ± ³ ´ (6.17)
1 Φ Sj (z) if z ∈ Gj \ Gj, 1 , j = 1, 2,
R0
is (locally) analytic in C \ E and (locally) continuous on C \ E, where now
E = G1, 1 ∪ G2, 1 .
R0 R0
As in Case II, the extensions of gΩ (z, ∞) and Φ(z) lead to the expressions
½
exp{−gΩ (z, ∞)} if z ∈ G \ E,
ρ(K(·, z)) = (6.18)
R0 if z ∈ E,
½
1/n
b
|Φ(z)| if z ∈ C \ E,
lim sup |Pn (z)| = 1 (6.19)
n→∞ R0 if z ∈ E,
and in parallel with Corollary 6.4, to the conclusion
(
β
log cap1(Γ) − gΩ (z, ∞) if z ∈ C \ E,
U (z) = 0 (6.20)
log capR(Γ) if z ∈ E,
supp β = ∂E and that every point of ∂E = L1, 1 ∪ L2, 1 attracts zeros of the sequence {Pn }∞
n=1 . Further-
R0 R0
more, since the unbounded domains C \ E and Ω 10 coincide, it follows from (2.14), (2.15) and (6.20) that
R
the same is true for the potentials U β and U µ∂E in C. Hence, the canonical measure β is the equilibrium
measure of ∂E. Therefore, by applying Corollary 6.1 (ii) (with E in the place of G), we conclude that for
j = 1, 2, there is a subsequence N = Nj ⊂ N such that, with V any neighborhood of Gj, 10 not meeting the
R
other island,
∗
νPn |V −→ µ∂E |V , n → ∞, n ∈ N . (6.21)
The zeros of the associated Bergman polynomials Pn , for n = 140, 150 and 160 are shown in Figure 9.
In the same figure we also depict the critical line LR0 and the curves L1, 10 and L2, 10 . Note that Lj, 10 is
R R R
the inverse image of Lj,R0 with respect to the circle Γj , j = 1, 2.
Example 6.3. Bergman polynomials for the union of an ellipse and a disk.
In Figure 10 we plot the zeros of the Bergman polynomials Pn , for n = 80, 90 and 100 of an ellipse
(domain G1 ) and a disk (domain G2 ), in relative positions chosen to illustrate further the theory given
above. To this end, let S1 and S2 denote the Schwarz function associated with the ellipse, respectively the
circle. The three ellipses pictured in Figure 10 have all focal segment [-1,1] and canonical equation
x2 y2
2
+ 2 = 1,
a b
24
1,5
1
0,5
0
-3 -2 -1-0,5 0 1 2 3 4
-1
-1,5
with a = 5/3, b = 4/3, in (i) and a = 5/4, b = 3/4, in both (ii) and (iii).
For such ellipses the associated Schwarz function is given by
p
S1 (z) = (2a2 − 1)z − 2ab z 2 − 1,
and the focal segment [−1, 1] is reflected to the confocal ellipse x2 /A2 + y 2 /B 2 = 1, where A = 2a2 − 1 and
B = 2ab. We denote by
x2 y2
D1 = {(x, y) : 2 + 2 < 1} \ [−1, 1],
A B
the maximal domain of involution for the Schwarz reflection and by γ the outer boundary of D1 , i.e.,
x2 y2
γ = {(x, y) : 2
+ 2 = 1}.
A B
Also, if G2 is a disk centered at z = z0 , the reflection z 7→ S2 (z) is an involution on the domain D2 = C\{z0 }.
The situations illustrated in Figure 10 represent the three possible relative positions between the loop
L1,R0 of the singular level set LR0 and γ:
• Figure 10 (i) corresponds to the case that L1,R0 is interior to γ,
• Figure 10 (ii) corresponds to the case that L1,R0 intersects γ,
• Figure 10 (iii) corresponds to the case that the ellipse γ is interior to L1,R0 .
By specializing Theorem 6.1 to this example, we can conclude the following:
Case (i) is completely analogous to Example 6.2. That is, supp β = ∂E = L1, 10 ∪ L2, 10 and every point
R R
of ∂E attracts zeros of the sequence {Pn }∞
n=1 . More precisely, β = µ∂E and for any j = 1, 2, there exists a
subsequence N = Nj ⊂ N such that, with V any neighborhood of Gj, 10 not meeting the other island,
R
∗
νPn |V −→ µ∂E |V , n → ∞, n ∈ N .
In case (ii), the support of the canonical measure β consists of three parts: the inverse image L2, 10 of
R
L2,R0 with respect to the circle Γ2 , the reflection of L1,R0 ∩ D1 with respect to the ellipse Γ1 and the part
[s, 1] of the focal segment [−1, 1] of the ellipse that lies exterior to this reflection. In addition, every point of
supp β attracts zeros of the sequence {Pn }∞ n=1 .
25
2 3
1
2
10 2 4 6 8 10
0 0
-2 0 2 4 6
-1
-1
-2
-2 -3
(i) (ii)
1
0,50 2 4 6 8
0
-0,5
-1
(iii)
Finally in case (iii), supp β = [−1, 1] ∪ L2, 10 . Thus C \ supp β has exactly two components and it follows
R
from (vii) of Theorem 6.1 that there exists is a subsequence N ⊂ N such that
∗
νPn −→ β, n → ∞, n ∈ N . (6.22)
Case IV: One piecewise analytic non-singular boundary and one analytic boundary curve.
Assume that Γ2 is analytic and Γ1 is piecewise analytic and non-singular. By the latter we mean that
any conformal map ϕ1 : G1 → D has an analytic continuation to a neighborhood of G1 , but this continuation
is not univalent in any neighborhood of G1 . This occurs, for example, if Γ1 consists of circular arcs and/or
straight lines and all its interior corners are of the form π/m, m ≥ 2 an integer.
Example 6.4. Bergman polynomials for the union of the half-disk G1 = {z : |z| < 1, Re(z) > 0} and the
disk G2 = {z : |z − 3| < 2/3}.
In Figure 11 we plot the zeros of the Bergman polynomials Pn of G, for n = 80, 90 and 100. In addition
we depict:
• The critical level line LR0 of the Green function gΩ (z, ∞).
• The part of the reflection (we denote it by Γ01 ) of L1,R0 with respect to Γ1 which lies in G1 .
• The inverse image L2, 1 of L2,R0 with respect to the circle Γ2 .
R0
By considering the symmetric and inverse images of the interior points of G1 with respect to the two
arcs forming Γ1 , in conjunction with the harmonic extension of the Green function inside G1 defined by the
26
1
0,5
0
0 1 2 3 4 5
-0,5
-1
0,5
-1 0 1 2 3
0
-0,5
-1
Schwarz functions of these arcs, it is not difficult to see that the support of the canonical measure β consists
of three parts: the loop Γ01 and two (symmetric) arcs that join together each one of the points i and −i with
the nearest corner of Γ01 . In addition, every point of supp β attracts zeros of the sequence {Pn }∞
n=1 .
Example 6.5. Bergman polynomials for the union of the symmetric lens domain G1 formed by two circular
arcs meeting at −i and i with interior angles π/4 and the disk G2 = {z : |z − 5/2| < 2/3}.
In Figure 12 we plot the zeros of the Bergman polynomials Pn of G, for n = 80, 90 and 100. In addition
we depict:
• The critical level line LR0 of the Green function gΩ (z, ∞).
• The part of the reflection (we denote it by Γ01 ) of L1,R0 with respect to Γ1 which lies in G1 .
• The inverse image L2, 1 of L2,R0 with respect to the circle Γ2 .
R0
As it is expected, identical conclusions to those of Example 6.4 regarding the properties of the support of
the canonical measure β hold here.
Case V: Three analytic boundaries.
27
2
0
-1 0 1 2 3
-1
Example 6.6. Bergman polynomials for the union of the three disks G1 = {z : |z + 1| < 1/2}, G2 = {z :
|z − 2| < 1} and G3 = {z : |z − 2i| < 1/2}.
In this example we have two critical Green level lines, LR0 and LR00 , where R0 = R2 = R3 and R00 = R1 .
(See Figure 1 which depicts the present example.) On setting
E 0 = G2, 1 ∪ G3, 1 and E 00 = G1, 1 ,

R0 R0 R00
we have 
b
 |Φ(z)| if z ∈ C \ (E 0 ∪ E 00 ),
1/n 1
lim sup |Pn (z)| = R0 if z ∈ E 0 , (6.23)
n→∞  1
R00 if z ∈ E 00 ,
b
where Φ(z) is the multi-valued function defined as in (6.17), with j = 1, 2, 3. From (6.23) and (3.9) conclusions
can be drawn about the canonical measure β. In particular we note that supp β = ∂E 0 ∪ ∂E 00 = L1, R1 ∪
1
L2, R1 ∪ L3, R1 and that every point of ∂E 0 ∪ ∂E 00 attracts zeros of the sequence {Pn }∞ n=1 .
2 3
In Figure 13 we plot the zeros of the Bergman polynomials Pn of G, for n = 80, 90 and 100. In order to
illustrate the above observations regarding the zero distribution we also depict the inverse image Lj, R1 of
j
Lj,Rj with respect to the circle Γj , j = 1, 2, 3.
We end this section by noting that the critical level curves of the Green function depicted in the plots
above, were computed by a simple modification of the MATLAB code manydisks.m of Trefethen [34]. The
original code manydisks.m is designed for archipelagoes formed by circles; see also Remark 2.1.
7. An example: lemniscate islands
Let G := {z : |z m − 1| < rm }, m ≥ 2 an integer and 0 < r < 1. Then G consists of m islands

G1 , G2 , . . . , Gm , where
Gj contains e2πji/m , j = 1, 2, . . . , m. (7.1)
Let Pn (z) = λn z n + · · · denote the (orthonormal) Bergman polynomial of degree n for the archipelago G,
and write
n = km + s, 0 ≤ s ≤ m − 1.
28
By the rotational symmetry of G and the uniqueness of the Bergman polynomials it is easy to see that
Pkm+s (z) = z s Qk,s (z m ), deg Qk,s = k. (7.2)
Then
Pkm+s (z)
pkm+s (z) := = z s qk,s (z m ) = z km+s + · · · , (7.3)
λkm+s
are the monic Bergman polynomials. Our first result concerns the asymptotic behavior of the leading
coefficient λn .
Proposition 7.1. For each s = 0, 1, . . . , m − 1 there holds
r
π 1
lim λkm+s rkm+s+1 = m−s−1 . (7.4)
k→∞ km + s + 1 r
Remark 7.1. Note that r = cap(G) = cap(Γ), where as above Γ = ∂G. Thus the sequence
r
π
λn cap(Γ)n+1 , n ∈ N,
n+1
1 1 1
has exactly m limit points, r m−1 , r m−2 , . . . , r , 1.
In Table 7.1 we illustrate Proposition 7.1 for the lemniscate depicted in Figure 14, where m = 3 and
r = 0.9. More precisely, Table 7.1 contains the computed values of the leading q coefficients λn correct to 6
n+1 π
decimal figures, for n = 38, . . . , 52, together with the computed values of λn r n+1 . As predicted by the
q
π
theory, the values of λn rn+1 n+1 alternate, as n increases, towards to the three limits
1/r2 = 1.234567 . . . , 1/r = 1.111111 . . . , 1.
The coincidence for the values of n = 38, 41, . . . , 50 is explained in the proof of Proposition 7.1.
Proposition 7.2. The following representations hold for the monic polynomials pkm+s (z):
pkm+m−1 (z) = z m−1 (z m − 1)k (7.5)
and for s = 0, 1, . . . , m − 2, we have for k sufficiently large,

¡ ¢
pkm+s (z) z m − 1 + r2m πk+1,s (−rm )
= πk+1,s (w) − πk,s (w), (7.6)
s
z r m(k+1) πk,s (−rm )
where w = (z m − 1)/rm and πn,s (w) is the monic polynomial of degree n in w that is orthogonal on the circle
|w| = 1 with respect to the weight
|dw|
2 2s . (7.7)
|rm w + 1|2− m − m
Remark 7.2. The representation formulas (7.5) and (7.6) have the same form as those found by Miña-Dı́az
[14], who studied the simpler case when r > 1, i.e. when G consists of a single island.
In our proof we utilize the following lemma that relates ’weighted’ Bergman polynomials on the unit disk
to Szegő polynomials on the unit circle. This result is somewhat implicitly contained in [14].
29
q
π
n λn λn rn+1 n+1
38 214.535664 1.000000
39 305.078943 1.263740
40 305.314216 1.124276
41 305.396681 1.000000
42 433.231373 1.261795
43 433.526043 1.123400
44 433.629077 1.000000
45 613.834469 1.260094
46 614.205506 1.122633
47 614.334958 1.000000
48 868.011830 1.258593
49 868.481244 1.121956
50 868.644692 1.000000
51 1225.297855 1.257261
52 1225.894247 1.121355
Table 7.1: Illustrating Proposition 7.1 for the lemniscate case m = 3 and r = 0.9, for n = 38, . . . , 52.
Lemma 7.1. Let tn (w) = wn +· · · be the monic polynomial orthogonal with respect to the weight |dw|/|γw +
1|τ on |w| = 1, where τ is real, τ 6= 2, 4, . . . , 2n, and |γ| < 1. Let βn (w) = wn + · · · be the monic polynomial
orthogonal with respect to the weight dA(w)/|γw + 1|τ over the unit disk |w| < 1. If tn (−γ) 6= 0, then
tn+1 (−γ)
(w + γ)βn (w) = tn+1 (w) − tn (w). (7.8)
tn (−γ)
Our next result describes the fine asymptotics for the monic Bergman polynomials.
Proposition 7.3. Let
2 2s
τ := 2 − − , s = 0, 1, . . . , m − 1. (7.9)
m m
Then for |z m − 1| ≥ r2m , z m − 1 6= −r2m , the monic Bergman polynomials satisfy for each s = 0, 1, . . . , m − 1
µ ¶τ /2
pkm+s (z) z m − 1 + r2m
lim = , (7.10)
k→∞ z s (z m − 1)k zm − 1
where the branch of the power function on the right-hand side of (7.10) is taken to equal one at infinity, and
the convergence is uniform on compact subsets.
Furthermore, for each j = 1, 2, . . . , m and z ∈ Gj with |z m − 1| < r2m , we have
(−1)k+1 k 2+τ /2 e2πij(s+1)/m z m−1 τ Γ(τ /2) sin(τ π/2)

lim pkm+s (z) = (7.11)
k→∞ rm(2k+4) 2π(1 − r2m )τ /2 (z m − 1 + r2m )2
for each s = 0, 1, . . . , m − 2, the convergence being uniform on closed subsets.
30
Observe that the lemniscate |z m −1| = r2m is the reflection of the lemniscate |z m −1| = 1 in the bounding
lemniscate of G.
Remark 7.3. From the first part of Proposition 7.3 we see that the Bergman polynomials for G have no
limit point of zeros in |z m − 1| > r2m other than at z = 0. Furthermore, from the second part of the
proposition, we deduce that, except for the subsequence (7.5), there are no limit points of the zeros of Pn (z)
in |z m −1| < r2m . Consequently, the only limit points of zeros of such Pn (z) are at z = 0 or on the lemniscate
|z m − 1| = r2m .
1.5 1.5
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
−1.5 −1.5
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
1.5
0.5
−0.5
−1
−1.5
−1.5 −1 −0.5 0 0.5 1 1.5
Figure 14: Zeros of the Bergman polynomials Pn for the lemniscate case m = 3 and r = 0.9, for n = 50, 51 and 52.
In Figure 14, we plot the zeros of the Bergman polynomials Pn , for n = 50, 51 and 52, of G := {z :
|z 3 − 1| < 0.93 }. In each plot, we depict also the defining lemniscate Γ = {z : |z 3 − 1| = 0.93 }, the reflection
{z : |z 3 − 1| = 0.96 } of {z : |z 3 − 1| = 1} in Γ and, for the cases n = 51, 52, the branch cuts for the Schwarz
z 3 − 1 + 0.96 1/3
function S(z) = ( ) of Γ.
z3 − 1
31
As a consequence of Proposition 7.3 we have the following:
Corollary 7.1. There are precisely two limit measures for the sequence {νPn }∞
n=1 ; namely
m
1 X
δz , zj = exp(2πij/m),
m j=1 j
and the equilibrium measure for the lemniscate |z m − 1| = r2m , which is given by the formula
|z|m−1
dβ = |dz|.
r2m
8. Proofs
The present section is devoted to the proofs of the results stated earlier in the article.
Proof of Lemma 3.1. That Γ is analytic is clear, since u is real analytic and ∇u 6= 0.
All of D is filled with integral curves of the gradient ∇u. These are disjoint and have no end points in D
since ∇u 6= 0. Hence they all end up on ∂D (an integral curve cannot be closed since u is single-valued and
increases along it). These integral curves are at the same time level lines of any locally defined harmonic
conjugate of u.
Given z ∈ D we want to define the reflected point S(z) using only u. Assume for example that u(z) < 0.
By the maximum principle, |u| < c in D, so actually −c < u(z) < 0. There is a unique integral curve γ of
∇u passing through z, and u increases along γ with limiting value +c as γ approaches ∂D. Thus there is a
unique point w ∈ γ at which u(w) = −u(z). In terms of this we define
S(z) = w.
The above procedure defines a function S(z) in D. To see that S(z) is analytic, note that, in some
neighborhood of γ, u has a single-valued harmonic conjugate u∗ and that γ is a level line of u∗ . The function
f = u + iu∗ is analytic in a neighborhood of γ, with f 0 6= 0; hence f can be used as a new complex coordinate
near γ, or u and u∗ are new real coordinates. In terms of these, the reflection map z 7→ S(z) just defined is
given by
u + iu∗ 7→ −u + iu∗ ,
or f (z) 7→ −f (z). This gives
S(z) = f −1 (−f (z)),
which proves that S(z) is analytic. It is also immediate that S(z) = z on Γ, so that S is indeed a Schwarz
function of Γ.
Proof of Lemma 3.2. According to Theorem 3.2.3 of [27], one criterion for dA|G to belong to the class
Reg is that
1/n
lim kPn kG = 1; (8.1)
n→∞
note that Ω is regular with respect to the Dirichlet problem [20, p. 92]. (Here and in the sequel k · k means
the sup norm on the subscripted set.)
32
The argument given in the proof of Lemma 4.3 of [18], when separately applied to each of the Jordan
regions Gj yields
1/n
lim sup kPn kG ≤ 1, j = 1, 2, . . . , N.
n→∞ j
1/n 1/n
Consequently, lim supn→∞ kPn kG ≤ 1. But lim inf n→∞ kPn kG ≥ 1, since kPn kL2 (G) = 1 for all n, and so
(8.1) follows.
8.1. The extremal problems

We use Pn to denote the space of complex polynomials of degree n. Recall that Kn (z, ζ) denotes the
n-th finite section of K(z, ζ)
n
X
Kn (z, ζ) := Pk (ζ)Pk (z),
k=0
and similarly set

n
X
KnGj (z, ζ) := Pk,j (ζ)Pk,j (z),
k=0
where
Pn,j (z) = λn,j z n + · · · , λn,j > 0, n = 0, 1, 2, . . . ,
are the sequences of the Bergman polynomials associated with Gj , j = 1, 2, . . . , N .
Lemma 8.1. For any ζ ∈ C,
|p(ζ)| p
max = Kn (ζ, ζ), n = 0, 1, . . . .
p∈Pn kpkL (G)
2
Proof. Since for any p ∈ Pn and ζ ∈ C
p(ζ) = hp, Kn (·, ζ)i,
it follows p
|p(ζ)| ≤ kpkL2 (G) kKn (·, ζ)kL2 (G) = kpkL2 (G) Kn (ζ, ζ).
Hence
|p(ζ)| p
≤ Kn (ζ, ζ)
kpkL2 (G)
with equality if p(z) = c Kn (z, ζ), for some constant c 6= 0.
Obviously
kpkL2 (Gj ) ≤ kpkL2 (G) , j = 1, 2, . . . , N,
therefore for n = 0, 1, . . .,
|p(ζ)| |p(ζ)|
max ≥ max , j = 1, 2, . . . , N,
p∈Pn kpkL2 (Gj ) p∈Pn kpkL2 (G)
or
KnGj (ζ, ζ) ≥ Kn (ζ, ζ), j = 1, 2, . . . , N, ζ ∈ C. (8.2)
33
Furthermore, since for any ζ ∈ Gj ,
KnGj (ζ, ζ) ≤ K Gj (ζ, ζ) = K(ζ, ζ), j = 1, 2, . . . , N,
it follows from (8.2) that

1 1 1
p ≥q ≥p , j = 1, 2, . . . , N. (8.3)
Kn (ζ, ζ) G
Kn j (ζ, ζ) K(ζ, ζ)
8.2. Proof of Theorem 4.1

The estimates from above require only a C 2+α -smooth boundary and are based on comparison with
corresponding estimates for the arc-length measure |dz| and the Szegő orthogonal polynomials. To this
purpose, we compare the two extremal problems
Z
m2n (G, dA) := min |z n + an−1 z n−1 · · · + a0 |2 dA(z), n = 0, 1, 2, . . . , (8.4)
a0 ,...,an−1 G
and Z
m2n (Γ, ρ|dz|) := min |z n + an−1 z n−1 · · · + a0 |2 ρ(z)|dz|, n = 0, 1, 2, . . . , (8.5)
a0 ,...,an−1 Γ
where ρ is a positive smooth function on Γ. Recall from (2.1) that

Z
2 1 Pn (z) 2
mn (G, dA) = 2 = | | dA(z), (8.6)
λn G λn
where
Pn (z) = λn z n + · · · , λn > 0, n = 0, 1, 2, . . . ,
are the Bergman polynomials of G.
The asymptotic properties of mn (Γ, ρ|dz|) have been established by Widom in [38, Thm 9.1]. In particular,
the next estimate for ρ = 1 and some constant C > 0 follows from Theorems 9.1 and 9.2 of [38]:
m2n (Γ, |dz|) ≥ C cap(Γ)2n . (8.7)
On the other hand, Suetin’s lemma (Lemma 3.5 above) applied to each island separately gives
Z Z
2 Pn (z) 2 C Pn (z) 2 C
mn (G, dA) = | | dA ≥ | | |dz| ≥ mn (Γ, |dz|)2 ,
G λ n n + 1 Γ λ n n +1
where C > 0 is a another positive constant.

Combining the above two estimates we conclude
cap (Γ)n
mn (G, dA) ≥ C √ ,
n
which yields the upper inequality in Theorem 4.1.

For estimates from below we require analyticity of the boundary. The main technical aid is provided by
a family of polynomials ωn constructed by Walsh in [35], which we thereby refer to as Walsh polynomials.
34
Lemma 8.2. Assume that each Γj , j = 1, 2, . . . , N , is analytic. Then, there exists a sequence of monic
polynomials ωn (z) = z n + · · · , n = 1, 2, . . . , with all zeros on a fixed compact subset E ⊂ G, and a constant
C such that
C
kωn kL2 (G) ≤ √ cap(Γ)n . (8.8)
n
From this we deduce the lower inequality in Theorem 4.1:
Corollary 8.1. If each Γj , j = 1, 2, . . . , N , is analytic then
√
n
C ≤ λn . (8.9)
cap(Γ)n
Proof of Lemma 8.2. Since each Γj , j = 1, 2, . . . , N , is analytic, the Green function gΩ (z, ∞) extends
harmonically across ∂G by Schwarz reflection. Choose first a number 0 < τ < 1 such that τ1 < R0 (see
Subsection 2.4 for the definition of R0 ) and such that gΩ (z, ∞) extends into each component of G, at least to
the negative level log τ . Since gΩ (z, ∞) has no critical points in GR0 \ G it follows that the extended Green
function has no critical points in D = G τ1 \ G τ = Ωτ \ Ω1/τ . The latter open set has N components, each of
which is a domain of involution for the Schwarz reflection (see Lemma 3.1).
Now choose a number ρ in the interval
τ < ρ < 1.
For any R ≥ ρ,
gΩR (z, ∞) := gΩ (z, ∞) − log R, (8.10)
is the Green function of ΩR with pole at infinity. Hence,
cap(LR ) = R cap(Γ). (8.11)
Choose the compact set E ⊂ G in the statement of the lemma to be E = Lτ . By a theorem of Walsh [35]
(see also [21, p. 515]), there exists a sequence of monic polynomials ωn (z) = z n + · · · , n = 1, 2, . . . , with
zeros approximately equidistributed with respect to the conjugate function of gΩ (z, ∞) and such that
1 C
|gΩτ (z, ∞) + log cap (Lτ ) − log |ωn (z)|| ≤ in Ωρ . (8.12)
n n
Note that gΩR (z, ∞) + log cap (LR ) is independent of R, hence in (8.12) τ can be replaced by any number
R > τ . For z ∈ LR and R ≥ ρ this gives
1 C
| log cap(LR ) − log |ωn (z)|| ≤ ,
n n
or after exponentiating and using (8.11)
|ωn (z)|
e−C ≤ ≤ eC (z ∈ LR , ρ ≤ R < ∞). (8.13)
Rn cap(Γ)n
In particular, from the maximum principle,
|ωn (z)| ≤ C Rn cap(Γ)n (z ∈ GR , ρ ≤ R < ∞), (8.14)
35
for another constant C.
Next we estimate the L2 (G)-norm of ωn . On decomposing
Z Z Z
2 2
|ωn | dA = |ωn | dA + |ωn |2 dA,
G Gρ G\Gρ
the first term can be directly estimated by means of (8.14):

Z
|ωn |2 dA ≤ C max |ωn (z)|2 ≤ Cρ2n cap(Γ)2n .
Gρ z∈Lρ
For the second term we foliate G \ Gρ by the level lines LR of gΩ (z, ∞), or |Φ(z)| = exp[gΩ (z, ∞)], and
use the coarea formula. Since ∇gΩ (z, ∞), and hence ∇|Φ(z)|, is bounded away from zero on G \ Gρ we
obtain by using once more (8.14)
Z Z 1 Z
|ωn (z)|2
|ωn |2 dA = |dz|dR
G\Gρ ρ LR |∇|Φ(z)||
Z 1 Z 1
2 2n
≤ C max |ωn (z)| dR ≤ C cap(Γ) R2n dR
ρ z∈LR ρ
1 − ρ2n+1 cap(Γ)2n
≤ C cap(Γ)2n ≤C ,
2n + 1 n
for various positive constants C. Thus altogether we have
Z
1
|ωn |2 dA ≤ C(ρ2n + )cap(Γ)2n ,
G n
and since ρ < 1, this gives (8.8).

The corollary is an immediate consequence of the lemma and the definition of λn :
√
1 n
= mn (G, dA) ≤ kωn kL2 (G) ≤ C .
λn cap(Γ)n

G
We turn now our attention to the problem of determining the rate of convergence of Λn j as compared to
Λn . The solution will obviously depend on a set of numerical constants which reflect the global configuration
of G.
In the case of a single island N = 1 we have ΛG n ≡ Λn , hence both (4.7) and (4.8) hold trivially with
1
m = 1. For the case N ≥ 2, we assume that Γj is analytic, for some fixed j ∈ {1, 2, . . . , N }. Let X denote
the characteristic function of Gj in G and set
kX pkL2 (G)
γn := inf . (8.15)
p∈Pn kpkL2 (G)
(Note that kX pkL2 (G) = kpkL2 (Gj ) , hence 0 < γn < 1.)
By considering the Bergman polynomial Pn,j of Gj , as a competing polynomial in (8.15) and using
Carleman asymptotics (Theorem 3.1) for Pn,j in G \ Gj in conjunction with the fact |Φj (z)| > |Φ(z)|, z ∈ Ω
36
(subordinate principle for the Green function; see e.g. [20, p. 108]), we conclude that there exist constants
C > 0 and R > Rj (> 1) such that, for any n ∈ N,
1 √
≥ 1 + C n Rn .
γn
Hence for large values of n,
γn < α n ,
where 0 < α < 1. Since X has an analytic continuation up to LR0 in Ω, it follows from Walsh’s theorem of
maximal convergence [36, Thm IV.5] that for any n ∈ N, there exist a constant m ≥ 1 and a polynomial
qm(n) ∈ Pm(n) , where m(n) = mn, with the property,
kqm(n) − X kG < γn . (8.16)
Then we have:
Lemma 8.3. Assume that Γj , j ∈ {1, 2, . . . , N }, is analytic. Then for any n ∈ N
q
G 2 q
Kn j (ζ, ζ) ≤ Kn+m(n) (ζ, ζ), ζ ∈ Gj .
1 − γn
Proof. Take ζ ∈ Gj and let h ∈ Pn be an extremal polynomial for

|p(ζ)|
max .
p∈Pn kpkL2 (Gj )
Then from Lemma 8.1 q

G |(X h)(ζ)|
Kn j (ζ, ζ) = .
kX hkL2 (G)
It holds,
1
|(X h)(ζ)| ≤ |(qm(n) h)(ζ)|,
1 − γn
because from (8.16),
(1 − γn )X (ζ) ≤ |qm(n) (ζ)|.
Also
kqm(n) hkL2 (G) ≤ kX hkL2 (G) + k(X − qm(n) )hkL2 (G)

≤ kX hkL2 (G) + γn khkL2 (G) ≤ 2kX hkL2 (G) ,
where in the last inequality we made use of the defining property of γn . Finally,
|(X h)(ζ)| 2 |(qm(n) h)(ζ)|
≤
kX hkL2 (G) 1 − γn kqm(n) hkL2 (G)
2 |f (ζ)|
≤ max ,
1 − γn f ∈Pn+m(n) kf kL2 (G)
and the result follows from Lemma 8.1.

This yields Inequality (4.8) in Theorem 4.2. The other inequality (4.7) follows immediately from (8.2).
37
Keeping in mind Lemma 3.3, it is clear from its definition that the functions Λn converge uniformly on
compact subsets of G to Λ. By imposing analyticity of the boundary, we will be able to estimate jointly the
rate of convergence of Λn (z) on Γ and in a neighborhood of Γ in the interior. In view of the reduction to a
single island established in the previous subsection, we will assume in the first part of the proof that N = 1.
In order to simplify further the notation, we will simply write G = G1 , Φ = Φ1 and so forth.
Thus, we deal now with a Jordan domain G with analytic boundary Γ. The normalized external conformal
mapping Φ analytically extends to the level set Gρ , with ρ < 1. According to Theorem 3.1, the Bergman
orthogonal polynomials satisfy:
r
n+1
Pn (z) = Φ(z)n Φ0 (z){1 + An (z)}, z ∈ G \ Gρ ,
π
where An (z) = O(( ρr )n ), whenever z ∈ Γr , and ρ < r < 1. Fix a z ∈ G \ Gρ and denote t = |Φ(z)|2 . Then
n
X n
|Φ0 (z)|2 X
Kn (z, z) = |Pk (z)|2 = (k + 1)tk + Rn (z) (8.17)
π
k=0 k=0
|Φ (z)|2 1 − (n + 2)t
0 n+1
+ (n + 1)tn+2
= + Rn (z).
π (1 − t)2
Similarly,
|Φ0 (z)|2 1
K(z, z) = + R(z).
π (1 − t)2
The convergence of Rn (z) to R(z), for ρ2 < r2 ≤ t < 1, is uniformly dominated by a convergent geometric
series.
In view of (4.4) we set Λ(z) = 0 for all z ∈ Γ. Since
1 1
0 < Λn (z − Λ(z) = p −p ,
Kn (z, z) K(z, z)
we are led to the estimate
1
Λn (z) − Λ(z) ≤ C(1 − t)[ p − 1].
1 − (n + 2)tn+1 + (n + 1)tn+2
In its turn, elementary algebra yields:
1 1 (n + 2)tn+1 − (n + 1)tn+2
(1 − t)[ p − 1] = pPn p
1 − (n + 2)tn+1 + (n + 1)tn+2 n
k=0 (k + 1)t 1 + 1 − (n + 2)tn+1 + (n + 1)tn+2
n+1 1
≤ p tn+1 [1 − − t]
tn/2 1 + 2 + ... + (n + 1) n+1
1
≤ Ctn/2 (1 − t + ),
n
which implies Inequality (4.9) in Theorem 4.3, since for z near Γ:
1 − |Φ(z)|2 ³ 1 − |Φ(z)| ³ dist(z, Γ).
38
√
Using (8.17), which holds for z ∈ Γ with Rn (z) = O(n2 nρn ), we derive easily (4.10), which is the limit
of the exact form of (4.9).
We resume now our general assumption G = ∪N j=1 Gj and we turn our attention to deriving (4.11). The
lower bound emerges at once by combining (4.10) with (4.7). To obtain the upper bound we apply (4.8) to
Λk (z), for large k, with k = [k/m]m + r, where 0 ≤ r < m − 1, and [k/m] is the integral part of the fraction,
and then we use again (4.10).
In order to estimate Λn in the exterior of G we employ the Walsh polynomials: From Lemma 8.1,
kp kL2 (G)
Λn (z) = min
p∈Pn |p(z)|
and therefore,
kωn kL2 (G) 1
Λn (z) ≤ ≤ C√ ;
|ωn (z)| n|Φ(z)|n
where we made use of Lemma 8.2 and (8.13).
Finally, the lower estimate for Λn (z) for z exterior to G is directly derived from the upper estimates for
the orthogonal polynomials appearing in Theorem 4.4.

Our aim is to derive estimates for Pn (z), for z in the exterior of the archipelago. To do so, we assume
that every curve constituting Γ is analytic and we rely, once more, to the Walsh polynomials ωn .
We fix a positive integer n and consider the rational function ωPn+1
n (z)
(z) , whose poles lie in a compact subset
of G and which vanishes at infinity. With z ∈/ G, Cauchy’s formula yields:
Z
Pn (z) −1 Pn (ζ)dζ
= ,
ωn+1 (z) 2πi Γ ωn+1 (ζ)(ζ − z)
whence, from (8.13),

C |ωn+1 (z)|
|Pn (z)| ≤ kPn kL1 (Γ) ,
dist(z, Γ) cap(Γ)n+1
where k · kL1 (Γ) denotes the L1 -norm on Γ with respect to |dz|. √
Since the L1 -norm is dominated by a constant times the L2 -norm, Lemma 3.5 gives kPn kL1 (Γ) ≤ C n
and one more application of (8.13) yields
C √
|Pn (z)| ≤ n|Φ(z)|n .
dist(z, Γ)
(In the above we use C to denote positive constants, not necessarily the same in all instances.)
In order to obtain the estimates from below, we have to restrict the point z to the complement of
the convex hull Co(G). On that set, including the point at infinity, the sequence of rational functions
n
Rn (z) = Pn√(z)cap(Γ)
nωn (z)
has no zeros, and by the above estimate, it is equicontinuous on compact subsets of
U = C \ Co(G). Thus {Rn }∞ n=0 forms a normal family on U and the possible limit functions are either
identically zero, or zero free. The normalization at infinity was chosen so that, in view of (4.2) and (8.13),
inf n∈N Rn (∞) > 0. Thus, every limit point of the sequence Rn is bounded away from zero, on compact
subsets of U .
39
8.6. Distribution of Zeros
Proof of Theorem 6.1. To prove (i) we need to figure out the general structure of ρ(K(·, z)). We have
already remarked, cf. (3.9), that for ζ ∈ Gj ,
ρ(K(·, ζ)) = min{Rj , ρ(K Gj (·, ζ))}.
Recall (2.17), that is, in terms of any conformal mapping ϕj : Gj → D,
ϕ0j (z)ϕ0j (ζ)

K Gj (z, ζ) = h i2 , z, ζ ∈ Gj .
π 1 − ϕj (z)ϕj (ζ)
Conversely, if (given ζ ∈ Gj ) ϕj is chosen so that ϕj (ζ) = 0, then

Z z
π
ϕj (z) = 0 K Gj (t, ζ)dt.
ϕj (ζ) ζ
Hence, for a general ϕj ,

Z z
ϕj (z) − ϕj (ζ) π(1 − |ϕj (ζ)|2 )
= K Gj (t, ζ)dt.
1 − ϕj (z)ϕj (ζ) ϕ0j (ζ) ζ
It follows therefore that, given a ζ ∈ Gj and a simply connected region D with Gj ⊂ D ⊂ Gj,Rj , K Gj (z, ζ)
has an analytic extension to D as a function of z if and only if ϕj (z) has a meromorphic extension to D and
does not attain the value 1/ϕj (ζ) there.
We introduce a meromorphic version of the function ρ defined in (3.7) by setting, for f meromorphic in
G,
ρm (f ) := sup {R ≥ 1 : f has a meromorphic continuation to GR } . (8.18)
Next we extend each ϕj to all G by setting ϕj = 0 in G \ Gj . Clearly the so extended ϕj cannot be
meromorphic in Gj,R for any R > Rj , hence
1 ≤ ρm (ϕj ) ≤ Rj . (8.19)
(This is vacuous statement if N = 1, thus we simply set R1 = +∞ in such a case.) The largest R for which
ϕj does not take the value 1/ϕj (ζ) in Gj,R is inf{|Φ(ϕj |−1
Gj,ρ (ϕ ) (1/ϕj (ζ)))|} (≥ 1), where the infinmum is
m j
taken over all points in the preimage ϕj |−1

Gj,ρ (1/ϕj (ζ)), which is a subset of Gj,ρm (ϕj ) \ Gj . (We assign
m (ϕj )
the value +∞ for the infimum of the empty set.)

Putting things together we get, in view of (8.19),
n o
ρ(K(·, ζ)) = min ρm (ϕj ), inf{|Φ(ϕj |−1
Gj,ρ (1/ϕ j (ζ)))|} , ζ ∈ Gj , (8.20)
m (ϕj )
or, by taking the logarithm,

n o
log ρ(K(·, ζ)) = min log ρm (ϕj ), inf{gΩ (ϕj |−1
Gj,ρ (1/ϕ j (ζ)), ∞)} , ζ ∈ Gj . (8.21)
m (ϕj )
40
This may look messy, but in principle it means that we have expressed log ρ(K(·, ζ)) as the infimum of some
harmonic functions. This is the basic argument telling that log ρ(K(·, ζ)) is superharmonic as a function of
ζ in Gj .
Now, if ϕj has a singularity on Γj , then ρm (ϕj ) = 1 and ρ(K(·, ζ)) = 1, ζ ∈ Gj . In the complementary
case, i.e., if ϕj has an analytic continuation across Γj , then for any ζ ∈ Gj , ϕj |−1
Gj,ρ (ϕ ) (1/ϕj (ζ)) is either void
m j
or it defines a (possibly) multi-valued reflection map in Γj , i.e., the conjugate of a (possibly) multi-valued
Schwarz function of Γj . By our assumption that the infimum of the empty set is +∞, we only need to
concentrate on the latter case. Denoting ϕj |−1Gj,ρm (ϕj ) (1/ϕj (ζ)) by Sj,multi (ζ) we can write (8.21) somewhat
more handily as
n o
log ρ(K(·, ζ)) = min log ρm (ϕj ), inf{gΩ (Sj,multi (ζ), ∞)} , ζ ∈ Gj , (8.22)
where the infinmum is taken over all branches of Sj,multi (ζ). One step further, this reflection map gives a
multi-valued analytic extension of the Walsh function Φ into Gj :
± ³ ´
Φ̂multi (ζ) = 1 Φ Sj,multi (ζ) , ζ ∈ Gj
(where we have used hat to emphasize the analytic extension). Inserting the latter into (8.20) gives the
following, more direct, description of ρ(K(·, ζ)):
n o
ρ(K(·, ζ)) = min ρm (ϕj ), inf{1/|Φ̂multi (ζ)|} , ζ ∈ Gj , (8.23)
the infimum is taken, again, over all (local) branches.

In order to make the above considerations more rigorous we take (8.21) as our starting point. We
first treat the case N ≥ 2, which is somewhat simpler because in this case (8.19) gives an upper bound
for log ρ(K(·, ζ)) in (8.21). Let ζ ∈ Gj . Then 1/ϕj (ζ) is outside the closed unit disk, and the preimage
ϕj |−1
Gj,ρ (ϕ ) (1/ϕj (ζ)) is either empty or is a finite or infinite subset of Gj,ρm (ϕj ) \ Gj . If it is an infinite set,
m j
then all cluster points will be on the boundary of Gj,ρm (ϕj ) , where gΩ (·, ∞) is larger, than near Γj . This
means that only finitely many of the points in the preimage will be serious candidates in the competition
for the infimum in (8.21). We may also vary ζ within a small disk, compactly contained in Gj , and there
will still be only finitely many branches of the multivaled analytic function ϕj |−1 involved, when forming
the infimum. Within such a disk there will also be only finitely many branch points (where two or more
preimages coincide).
Thus, locally away from the mentioned branch points, log ρ(K(·, ζ)) is the infimum of finitely many
harmonic functions, hence is continuous and superharmonic. At the branch points log ρ(K(·, ζ)) is still
continuous, and since the set of branch points is discrete (in Gj,ρm (ϕj ) \ Gj ) they make up a removable set
for continuous superharmonic functions; see e.g. [20, Thm 3.6.1]. It follows, therefore, that log ρ(K(·, ζ)) is
superharmonic (and continuous) in all Gj .
We apply now the above inferences to h(z) = − log ρ(K(·, z)), for z ∈ G. If ρm (ϕj ) = 1, for some j,
then h(z) = 0, for z ∈ Gj , hence the transition across Γj to gΩ (z, ∞) is continuous and subharmonic. If
ρm (ϕj ) > 1 and ϕj remains univalent in a neighborhood of Gj , then it is easy to see that h(z) defines
the harmonic continuation of gΩ (z, ∞) across Γj (in fact, Γj turns out to be analytic and thus Sj,multi is
the associated ordinary single-valued Schwarz function). Finally, if ρm (ϕj ) > 1 but ϕj is not univalent in
any neighborhood of Gj then locally, away from finitely many branch points on Γj , h is still the ordinary
41
harmonic continuation of gΩ (z, ∞). At the branch points h is still continuous and the set of branch points
is too small to affect the overall subharmonicity. Hence, in all possible situations h(z) = − log ρ(K(·, z)) is
continuous and subharmonic in G.
Therefore, we have established so far that in the case N ≥ 2, h is subharmonic (and continuous) in C
and since it coincides with the Green function in Ω, β is a positive measure, with support contained in G.
Moreover, from Gauss’ theorem (see e.g. [24, p. 83]), and the singularity of the Green function at infinity,
we have for any R > 1: Z Z
1 ∂h 1 ∂gΩ (z, ∞)
β(GR ) = ds = ds = 1. (8.24)
2π LR ∂n 2π LR ∂n
Hence β is a unit measure and this completes the proof of (i), for N ≥ 2.
In order to derive (ii), we observe that the Riesz decomposition theorem for subharmonic functions
applied to h in C (see e.g. [20, p. 76]) gives,
h(z) = −U β (z) + v(z), z ∈ C,
where v is harmonic in C. Then, by considering the expansions near infinity of U β (z) and h(z) = gΩ (z, ∞),
we see that v(z) = − log cap (Γ), which yields (6.3). Relation (6.4) is an immediate consequence of (6.3) the
fact that h coincides with the Green function in Ω, in conjunction with the relations (2.5)–(2.7).
When N ≥ 2, U β is bounded from above because of (8.19):
maxj {Rj }
U β ≤ log < ∞.
cap (Γ)
Statement (iii) of the theorem is just a juxtaposition of Proposition 3.1 and Corollary 3.1 along with
(6.3).
As for (iv), C is nonempty by general compactness principles for measures and the known fact that all
counting measures νPn have support within a fixed compact set; see Remark 3.1. Let σ ∈ C. Then there is
a subsequence N = Nσ ⊂ N such that
∗
νPn −→ σ, n → ∞, n ∈ N . (8.25)
Using the lower envelope theorem [24, Thms I.6.9 ] and (6.6) we get
U σ (z) = lim inf U νPn (z) ≥ lim inf U νPn (z) = U β (z), (8.26)
n→∞ n→∞
n∈N
where the first equality holds only quasi everywhere in C. However the relation between U σ and U β persists
everywhere in C, since both members are potentials.
Let D be any component of C \ supp β. Applying the minimum principle to u = U σ − U β ≥ 0, which is
superharmonic in D, gives that either u > 0 in all D or u = 0 in all D. Since u vanishes at ∞ (recall that σ
and β are unit measures) it follows that it vanishes in the entire unbounded component of C \ supp β. From
this and the observations above follow all parts of (iv).
Turning to (v), let
U = lsc ( inf U σ ).
σ∈C
β
By (iv), U ≤ U in C. To prove the opposite inequality, choose an arbitrary point z ∈ C. Then there is
subsequence Nz ⊂ N, such that the lim inf in (6.6) is realized at z, i.e.
lim U νPn (z) = U β (z). (8.27)
n→∞
n∈Nz
42
By weak* compactness there exists a further subsequence Nz0 ⊂ Nz and a measure σ = σz ∈ C such that
∗
νPn −→ σ, n → ∞, n ∈ Nz0 . (8.28)
Then, by the principle of descent (see [24, Thm I.6.8]) and (8.27),
U β (z) = lim inf U νPn (z) ≥ U σ (z). (8.29)

n→∞
0
n∈Nz
Since z ∈ C was arbitrary,

U β ≥ inf U σ in C,
σ∈C
by which U β ≥ U follows in all C.

To finish the proof of (v), we let again D be a component of C \ supp β. By choosing above z ∈ D we get
a measure σ = σz ∈ C with U σ (z) = U β (z) (since equality necessarily holds in (8.29)). Thus U σ = U β in all
D because, as we have already proved, the other alternative would be U σ > U β in all D.
Regarding (vi), if C consists of only one point, say σ, then U β = U σ by (v), and from the unicity theorem
for logarithmic potentials (see [24, Thm II.2.1]) we must have β = σ. Clearly, the full sequence must converge
to β, because otherwise one could extract a subsequence converging to something else, which would be a
different element in C.
The assertions in (vii) are easy consequences of (iv) and (v): Since, for any σ ∈ C, U σ = U β in the
unbounded component of C \ supp β we get in the case of (a) plus (b) that (for any σ ∈ C) U σ = U β , almost
everywhere with respect to the area measure in C. This and the unicity theorem yield β = σ ∈ C. In the
case of (a) plus (c), there exists (by (v)) at least one σ ∈ C satisfying U σ = U β in the bounded component
of C \ supp β, and for this σ we have the same conclusion: U σ = U β almost everywhere in C and, as above,
β = σ ∈ C.
So far we have assumed that N ≥ 2. Let us indicate the modifications needed for N = 1. Equation (8.21)
may be written
n o
log ρ(K(·, ζ)) = lim min M, log ρm (ϕj ), inf{gΩ (ϕj |−1 Gj,ρ (ϕ ) (1/ϕ j (ζ)), ∞)} , (8.30)
M →+∞ m j
that is, by introducing an auxiliary upper bound M , which finally tends to infinity. Before passing to the
limit we can work with the corresponding quantities
1
hM = sup{h, −M }, βM = ∆hM
2π
(etc.) as before. Since a decreasing sequence of subharmonic functions is subharmonic, h = lim hM will be
M →∞
again subharmonic. It is however not clear that it will be continuous, only upper semicontinuity is automatic.
If ρm (ϕj ) < ∞, then the bound M is not needed and everything will be as in the case N ≥ 2. So assume
ρm (ϕj ) = ∞. This means that ϕj is meromorphic in the entire complex plane and hence (8.21) reads
log ρ(K(·, ζ)) = inf{gΩ (ϕj |−1

C (1/ϕj (ζ)), ∞)}, ζ ∈ Gj . (8.31)
Problems concerning the lower boundedness and continuity of h could conceivably occur at points ζ ∈ G
at which the inverse image above is either empty or is an infinite set. The first case can, by Picard’s theorem,
occur for at most two values of ζ ∈ G. At such points the infimum in (8.31) is +∞, and hence h(ζ) = −∞.
43
In particular, h will not be bounded from below, but it will still be subharmonic and upper semicontinuous.
Moreover, it will be continuous at all other points, which is enough for the reasoning in the proof (above) of
(iv), where we used the continuity of h (or U β ).
The second conceivable problem, that ϕ1 |−1 C (1/ϕ1 (ζ)) is an infinite set, presents no actual difficulty
because the only cluster points can be at infinity, hence all but finitely many branches of ϕ1 |−1C (1/ϕ1 (ζ))
will be ruled out when taking the infimum in (8.31).
Proof of Corollary 6.1. As already remarked, the boundary curve Γj is singular if and only if ρm (ϕj ) = 1,
which by the proof of the theorem (e.g., Equation (8.21)) occurs if and only if h = 0 in Gj . This, in view of
(6.3), is equivalent to
1
U β (z) = log , z ∈ Gj .
cap (Γ)
Also from (6.3),
1
U β (z) = log − gΩ (z, ∞), z ∈ Gj,Rj \ Gj .
cap (Γ)
It follows that U β is harmonic in Gj,Rj \ Γj , thus supp β ⊂ Γj . It also follows that the logarithmic potentials
of β and µΓ coincide in the domain Gj,Rj , hence the equation β|Gj = µΓ |Gj holds as a result of the unicity
theorem (see e.g. [24, p. 97]). This proves the equivalence of (i) and (ii).
By assertion (v) of the theorem, there exists a σ ∈ C such that U σ = U β in Gj (= D). The equation
persists on Γj , because of the continuity of logarithmic potentials in the fine topology and in view of (6.7), it
also holds in any neighborhood of Gj not meeting the other islands. Thus, from the unicity theorem σ = β,
in such a neighborhood. As σ is a cluster point of {νPn }, we conclude that (iii) follows from (ii).
If (iii) holds then by selecting a further subsequence we conclude σ|V = µΓ |V , for some σ ∈ C. Then
U σ = U µΓ in V , which in view of (6.4) and (6.7) yields the relation U β = U µΓ in V . Therefore β|Gj =
µΓ |Gj .
Proof of Corollary 6.3. Set µn = Bal (νPn ). Then
supp µn ⊂ C \ G, (8.32)
U νPn = U µn in Ω. (8.33)
∗
Let µ be any weak* cluster point of {µn } and let N ⊂ N be a subsequence with µn −→ µ, n ∈ N . By
∗
refining N we may assume also that νPn −→ σ, n ∈ N , for some measure σ. Then in view of (8.33) we have
σ µ
U = U in Ω.
On the other hand, U σ = U µΓ in Ω by Theorem 6.1, thus U µ = U µΓ in Ω. But U µΓ is harmonic in
Ω \ {∞} and supp µ ⊂ C \ G by (8.32), hence supp µ ⊂ Γ. Now Carleson’s unicity theorem [24, p. 123], shows
∗
that µ = µΓ . Since µ was an arbitrary cluster point of µn it follows that µn −→ µΓ for the full sequence.
Proof of Corollary 6.4. The expression for U β follows immediately after uploading (6.12) into Theo-
rem 6.1 (ii). From this expression and the unicity theorem for logarithmic potentials we gather that supp β
must be contained in ∂E. To show that eventually supp β = ∂E we can argue as in [16, pp. 215–216]. That
is, by assuming that a point z0 ∈ ∂E does not belong to supp β, hence the potential U β is harmonic in a
small disk centered at z0 , we arrive to a contradiction by comparing the resulting harmonic extension of U β
with the one given in (6.14).
In view of the connectedness of the complement of E and the fact that the support of β is contained
in E the equality U σ (z) = U β (z), for z ∈ C \ E, is immediate from Theorem 6.1 (iv). Hence supp σ ⊂ E.
44
Furthermore, since the boundary of the domain C \ E in the fine topology coincides with its boundary in the
Euclidean topology (see e.g. [24, Cor. I.5.6]), we conclude that the equality between the potentials persists
in C \ E.
The last assertion in the corollary can be deduced from Theorem 6.1 (iv)–(v), because this guarantees
the existence of a cluster point σ of the sequence νPn such that U σ = U β on both sides of Γ1 . More precisely,
U σ = U β in V \ Γ1 , where V is a neighborhood of G1 not meeting the other islands, and therefore σ = β in
such a neighborhood. Similarly we argue for L2, 10 .
R
8.7. The lemniscate example
Proof of Lemma 7.1. Let (γw + 1)τ /2 denote the analytic branch in D = {w : |w| < 1} that equals 1 at
w = 0. Then applying Green’s formula we have, for j = 0, 1, . . . , n − 1,
Z Z
j dA(w ) βn (w) j−τ /2
0 = βn (w)(γw + 1) τ
= τ /2
(γw + 1) dA(w )
|γw + 1| (γw + 1)
D D
Z Z j
βn (w) j+1 (γw + 1)
= (γw + 1) w|dw | = βn (w)(γ̄ + w) |dw |,
|γw + 1|τ |γw + 1|τ
|w|=1 |w|=1
where we have ignored nonzero constants, and in the last equality, we used that (γw + 1) = (γ̄/w + 1) for
|w| = 1. Consequently, βn (w)(γ̄ + w) is a monic polynomial of degree n + 1 that vanishes at w = −γ̄ and is
orthogonal to all polynomials of degree less than n with respect to |dw |/|γw + 1|τ . The same is true of the
right-hand side of (7.8) and hence the difference of these two polynomials (which is of degree ≤ n) must be
a multiple of tn (w) that vanishes at −γ̄. Since tn (−γ̄) 6= 0, the difference of the left and right-hand sides of
(7.8) must be identically zero.
Remark 8.1. It is essential that the cases τ = 2, 4, . . . , 2n be excluded in Lemma 7.1. Indeed for τ = 2j,
where j is a positive integer, it is well-known (cf. [31], §11.2) that tn (w) = wn−j (w + γ̄)j for n ≥ j, so that
tn (−γ̄) = 0 in this case. There appears, however, to be no simple formula2 for the polynomials βn (w) for
such values of τ . We shall show in Lemma 8.4 that if τ is not an even integer, then tn (−γ̄) 6= 0 for all n
sufficiently large.
Proof of Proposition 7.2 . Here we use the minimality property of the monic Bergman polynomials
pkm+s (z) = z s qk,s (z m ). More precisely, qk,s solves the extremal problem
Z
Ik,s := min{ |z s q(z m )|2 dA : q(t) = tk + · · · ∈ Pk }. (8.34)
G
Clearly, Z Z
s m 2
|z q(z )| dA = m |z s q(z m )|2 dA,
G Gm
2 For the weight dA/|γw + 1|2 , we have

1 γ̄
β1 (w) = w + + .
γ ln(1 − |γ|2 )
45
and the change of variables w = (z m − 1)/rm , which maps Gm conformally onto the unit disk D in the
w-plane, yields Z Z
s m 2 r2m |q(rm w + 1)|2
|z q(z )| dA(z ) = 2 dA(w ),
m |rm w + 1|τ
Gm D
where
2 2s
τ := 2 − − . (8.35)
m m
Consequently,
Z
r2m |q(rm w + 1)|2
Ik,s = min{ dA(w ) : q(t) = tk + · · · ∈ Pk }, (8.36)
m |rm w + 1|τ
D
and, moreover, r−mk qk,s (rm w + 1) is the monic (in w) orthogonal polynomial with respect to the weight
dA(w )/|rm w + 1|τ on D. Applying Lemma 7.1 then yields formulas (7.5) and (7.6), provided that πk,s (−rm )
is not zero. In the next lemma we show that this condition is indeed satisfied for k sufficiently large.
Lemma 8.4. Let πk,s (w) be as in Proposition 7.2 and τ be given by (8.35). Then, for each s = 0, 1, . . . , m−2,
we have
k τ /2 h 1 ³τ ´ b ³ 1 í
s
(−1)k mk πk,s (−rm ) = sin(τ π/2) Γ + +O 2 (8.37)
r π 2 k k
as k → ∞, where bs is a constant independent of k.
Proof. As in [14], we utilize the results of [12] for Szegő polynomials with respect to an analytic weight
on |w| = 1. For the weight |w + rm |−τ = 1/|rm w + 1|τ , we have, imitating the notation of [12], the following
formulas for the exterior and interior Szegő functions De,τ (w) and Di,τ (w), respectively,
³ w + rm ´τ /2
De,τ (w) = , Di,τ (w) = (1 + rm w)−τ /2 , (8.38)
w
where the branches of the square roots are chosen so that De,τ (∞) = Di,τ (0) = 1. The scattering function
Sτ (w) is given by
³ w + rm ´τ /2
Sτ (w) = De,τ (w)Di,τ (w) = (1 + rm w)−τ /2 for rm < |w| < r−m . (8.39)
w
As shown in [12] (see Equations (16), (25), and (39)), we have for |w| < η, where rm < η < 1,
I k
1 t Sτ (t)
Di,τ (w)πk,s (w) = dt + O(η 3k ), as k → ∞. (8.40)
2πi t−w
|t|=1
For w = −rm , we can deform the unit circle in the integral in (8.40) so that the integration takes place along
each side of the branch cut of De,τ (w) joining −rm to 0 to obtain
 
I k Z Z
t Sτ (t)   xk Sτ (x)
Ik := m
dt =  +  dx , (8.41)
t+r x + rm
|t|=1 [−r m ,0] [0,−r m ]
46
where we utilize the limiting values from below for Sτ in integrating from −rm to 0 and the limiting values
of Sτ from above in integrating from 0 to −rm . Thus we get (cf. (8.39))
Z0
xk (1 + rm x)−τ /2
Ik = 2i sin(τ π/2) dx ,
|x|τ /2 (x + rm )1−τ /2
−r m
and on making the change of variable x = −rm (1 + cos θ)/2 we find that
Zπ
irmk
Ik = k−1 sin(τ π/2)(−1)k e−kp(θ) q(θ)d θ, (8.42)
2
0
where p(θ) := − log(1 + cos θ) and

h r2m i−τ /2 ³ sin θ ´τ −1
q(θ) := 1 − (1 + cos θ) (1 + cos θ)1−τ θτ −1 . (8.43)
2 θ
We now apply Laplace’s method to deduce the asymptotic behavior of the integral in (8.42). Since
∞
X 1
p(θ) = − log 2 + pj θj+2 = − log 2 + θ2 + · · ·
j=0
4
and
∞
X
q(θ) = qj θj+τ −1 = (1 − r2m )−τ /2 21−τ θτ −1 + q2 θτ +1 + · · · ,
j=0
(note that q1 = 0) we obtain from [17, Ch. 3, Thm 8.1], that, as k → ∞,

Zπ h ³ τ ´ (1 − r2m )−τ /2 ³ 1 í
a2,τ
e−kp(θ) q(θ)d θ = 2k Γ + + O , (8.44)
2 k τ /2 k τ /2+1 k τ /2+2
0
where a2,τ is a constant independent of k. From (8.40)–8.44) (taking η such that η 3 < rm < η) we deduce
(8.37).
As an immediate consequence of the preceding lemma we obtain that
πk+1,s (−rm ) h τ ³ 1 í
m
= −r 1 − + O as k → ∞. (8.45)
πk,s (−rm ) 2k k2
Proof of Proposition 7.3 For s = m − 1 the assertion is obvious from (7.5). For |z m − 1| > r2m and
s = 0, 1, . . . , m − 2, we appeal to the well-known fact regarding exterior asymptotics of Szegő polynomials
(see e.g. [12], Proposition 1) that for |w| > rm we have
πk,s (w) ³ w + rm ´τ /2
lim k
= De,τ (w) = , (8.46)
k→∞ w w
where the convergence is locally uniform and takes place with a geometric rate. Thus from (8.45) and the
representation (7.6) we deduce (7.10) for |z m − 1| > r2m .
47
For |z m −1| ≤ r2m , we begin with the asymptotic analysis of πk,s (w), for s = 0, 1, . . . , m−2 and |w| ≤ rm .
Assume at first that w ∈ / [−rm , 0], and consider the integral in the representation (8.40). For each ² > 0
sufficiently small, we can write
 
I k I Z Z
1 t Sτ (t) 1   tk Sτ (t)
Jk (w) := dt =  + +  dt, (8.47)
2πi t−w 2πi t−w
|t|=1 |t−w|=² [−r m ,0] [0,−r m ]
where integration along both sides of the branch cut from −rm to 0 is as in the proof of Lemma 8.4. From
Cauchy’s formula and the representation of Sτ (t) along each side of the branch cut, we deduce that
Z0
k 1 xk (1 + rm x)−τ /2 (x + rm )τ /2
Jk (w) = w Sτ (w) + sin(τ π/2) dx ,
π |x|τ /2 (x − w)
−r m
which, upon performing the change of variable x = −rm (1 + cos θ)/2, yields
Zπ
1 rm(k+1)
Jk (w) = w Sτ (w) + sin(τ π/2)(−1)k+1 k+1
k
e−kp(θ) q̂(θ)d θ, (8.48)
π 2
0
where p(θ) = − log(1 + cos θ) and

2m
[1 − r 2 (1 + cos θ)]−τ /2 ( sinθ θ )τ +1 θτ +1
q̂(θ) := m .
[ r2 (1 + cos θ) + w](1 + cos θ)τ
Since
∞
X (1 − r2m )−τ /2 τ +1
q̂(θ) = q̂j θj+(τ +2)−1 = θ + q̂3 θτ +3 + · · ·
j=0
(rm + w)2τ
(note that q̂1 = 0), Laplace’s method yields

Zπ h τ Γ( τ )(1 − r2m )−τ /2 ³ 1 í
1 b̂s (w)
e−kp(θ) q̂(θ)d θ = 2k 2
+ + O 3+τ /2 ,
rm +w k 1+τ /2 k 2+τ /2 k
0
as k → ∞, where b̂s (w) is a constant independent of k. Thus, from (8.48) and (8.40), we obtain
k 1+τ /2 (−1)k+1 sin(τ π/2)τ Γ(τ /2) h b̂s (w) ³ 1 í

Di,τ (w)πk,s (w) m(k+1)
= 2m τ /2 m
1+ +O 2 , (8.49)
r 2π(1 − r ) (w + r ) k k
as k → ∞, provided |w| < rm and η 3 < rm < η, while for |w| = rm , w 6= −rm , we obtain
πk,s (w) ³ 1 ´
Di,τ (w) = S τ (w) + O , (8.50)
wk k 1+τ /2
as k → ∞, where we take rm < η < 1.
Combining (8.45) with (8.49) and (8.50), we deduce from the representation (7.6) that (7.11) holds for
|z m −1| < r2m , z m ∈
/ [1−r2m , 1], and that (7.10) holds for |z m −1| = r2m , except for the m roots (1−r2m )1/m .
48
In deriving (7.11) we used the fact that (z m )τ /2 z s = z m−1 e2πij(s+1)/m for z ∈ Gj (recall (7.1)). Finally, by
a slight modification of the above analysis, it is easy to see that (8.49) is valid also for w ∈ (−rm , 0] and so
(7.11) holds for all z satisfying |z m − 1| < r2m .
Proof of Proposition 7.1. We use the obvious fact that
Z
λ−2
km+s = |pkm+s (z)|2 dA(z). (8.51)
G
For s = m − 1, we have from (7.5),

Z Z
−2 m−1 m k 2
λkm+m−1 = |z (z − 1) | dA(z) = m |z m−1 (z m − 1)k |2 dA(z )
G Gm
2m(k+1) Z 2m(k+1)
r πr
= |w|2k dA(w) = ,
m m(k + 1)
D
where, as in the proof of Lemma 8.4, we have made the change of variables w = (z m − 1)/rm . Thus
r
m(k + 1)
λkm+m−1 = . (8.52)
πr2m(k+1)
Now suppose that 0 ≤ s < m − 1. Then, on utilizing the formula (7.6) we deduce that, for k sufficiently
large,
Z Z
r2m |qk,s (rm w + 1)|2
λ−2
km+s = m |z s
q k,s (z m 2
)| dA(z ) = dA(w )
m |rm w + 1|τ
Gm D
Z |π πk+1 (−r m ) 2
r2m(k+1) k+1 (w) − πk (−r m ) πk (w)|
= dA(w ), (8.53)
m |w + rm |2 |rm w + 1|τ
D
where for simplicity of notation we have written πk = πk,s . On using the orthogonality property of the πk ’s
we can simplify the last integral in (8.53) to obtain
Z
−πk+1 (−rm )r2mk+m |πk (w)|2
λ−2
km+s = |dw |. (8.54)
πk (−rm )2m(k − τ2 + 1) |rm w + 1|τ
|w|=1
Finally we note that the integral on the right-hand side of (8.54) equals µ−2
k,s , where µk,s is the leading
coefficient of the orthonormal polynomial with respect to the weight |dw |/|rm w + 1|τ on the unit circle. As
is well-known (see e.g. [12], Corollary 2)
1
|µ2k,s − | = O(η 2k ) as k → ∞,
2π
where rm < η < 1. Combining this fact with (8.54) and (8.45) yields (7.4).
49
References
[1] A. Ambroladze, On exceptional sets of asymptotic relations for general orthogonal polynomials, J. Ap-
prox. Theory 82 (1995), no. 2, 257–273.
[2] V. V. Andrievskii and H.-P. Blatt, Erdős-Turán type theorems on quasiconformal curves and arcs, J.
Approx. Theory 97 (1999), no. 2, 334–365.
[3] T. Carleman, Über die Approximation analytisher Funktionen durch lineare Aggregate von vorgegebenen
Potenzen, Ark. Mat., Astr. Fys. 17 (1923), no. 9, 215–244.
[4] P. J. Davis, The Schwarz function and its applications, The Mathematical Association of America,
Buffalo, N. Y., 1974, The Carus Mathematical Monographs, No. 17.
[5] M. D. Finn, S. M. Cox, and H. M. Byrne, Topological chaos in inviscid and viscous mixers, J. Fluid
Mech. 493 (2003), 345–361.
[6] D. Gaier, Lectures on complex approximation, Birkhäuser Boston Inc., Boston, MA, 1987, Translated
from the German by Renate McLaughlin.
[7] G. Golub, B. Gustafsson, C. He, P. Milanfar, M. Putinar, and J. Varah, Shape reconstruction from
moments: theory, algorithms, and applications, SPIE Proccedins (F. T. Luk, ed.), Advanced Signal
Processing, Algorithms, Architecture, and Implementations X, vol. 4116, 2000, pp. 406–416.
[8] B. Gustafsson, C. He, P. Milanfar, and M. Putinar, Reconstructing planar domains from their moments,
Inverse Problems 16 (2000), no. 4, 1053–1070.
[9] W. K. Hayman, Subharmonic functions. Vol. 2, London Mathematical Society Monographs, vol. 20,
Academic Press, London, 1989.
[10] H. Hedenmalm, B. Korenblum, and K. Zhu, Theory of Bergman spaces, Graduate Texts in Mathematics,
vol. 199, Springer-Verlag, New York, 2000.
[11] A. L. Levin, E. B. Saff, and N. S. Stylianopoulos, Zero distribution of Bergman orthogonal polynomials
for certain planar domains, Constr. Approx. 19 (2003), no. 3, 411–435.
[12] A. Martı́nez-Finkelshtein, K. T.-R. McLaughlin, and E. B. Saff, Szegő orthogonal polynomials with
respect to an analytic weight: canonical representation and strong asymptotics, Constr. Approx. 24
(2006), no. 3, 319–363.
[13] V. Maymeskul and E. B. Saff, Zeros of polynomials orthogonal over regular N -gons, J. Approx. Theory
122 (2003), no. 1, 129–140.
[14] E. Miña-Dı́az, Asymptotics for Faber polynomials and polynomials orthogonal over regions in the complex
plane, Ph.D. thesis, Vanderbilt University, August 2006.
[15] , An asymptotic integral representation for Carleman orthogonal polynomials, Int Math Res
Notices 2008 (2008), article ID rnn066, 35 pages.
[16] E. Miña-Dı́az, E. B. Saff, and N. S. Stylianopoulos, Zero distributions for polynomials orthogonal with
weights over certain planar regions, Comput. Methods Funct. Theory 5 (2005), no. 1, 185–221.
50
[17] F. W. J. Olver, Asymptotics and special functions, AKP Classics, A K Peters Ltd., Wellesley, MA, 1997,
Reprint of the 1974 original (Academic Press, New York).
[18] N. Papamichael, E. B. Saff, and J. Gong, Asymptotic behaviour of zeros of Bieberbach polynomials, J.
Comput. Appl. Math. 34 (1991), no. 3, 325–342.
[19] N. Papamichael and M. K. Warby, Stability and convergence properties of Bergman kernel methods for
numerical conformal mapping, Numer. Math. 48 (1986), no. 6, 639–669.
[20] T. Ransford, Potential theory in the complex plane, London Mathematical Society Student Texts, vol. 28,
Cambridge University Press, Cambridge, 1995.
[21] E. B. Saff, Polynomials of interpolation and approximation to meromorphic functions, Trans. Amer.
Math. Soc. 143 (1969), 509–522.
[22] , Orthogonal polynomials from a complex perspective, Orthogonal polynomials (Columbus, OH,
1989), Kluwer Acad. Publ., Dordrecht, 1990, pp. 363–393.
[23] E. B. Saff and N. S. Stylianopoulos, Asymptotics for polynomial zeros: Beware of predictions from plots,
Comput. Methods Funct. Theory 8 (2008), no. 2, 185–221.
[24] E. B. Saff and V. Totik, Logarithmic potentials with external fields, Springer-Verlag, Berlin, 1997.
[25] H. S. Shapiro, The Schwarz function and its generalization to higher dimensions, University of Arkansas
Lecture Notes in the Mathematical Sciences, 9, John Wiley & Sons, New York, 1992.
[26] H. Stahl and V. Totik, nth root asymptotic behavior of orthonormal polynomials, Orthogonal polynomials
(Columbus, OH, 1989), Kluwer Acad. Publ., Dordrecht, 1990, pp. 395–417.
[27] , General orthogonal polynomials, Cambridge University Press, Cambridge, 1992.
[28] N. S. Stylianopoulos, The use of orthogonal Bergman polynomials for recovering planar domains from
their moments, preprint.
[29] P. K. Suetin, Order comparison of various norms of polynomials in a complex region, Ural. Gos. Univ.
Mat. Zap. 5 (1966), no. tetrad 4, 91–100, (in Russian).
[30] , Polynomials orthogonal over a region and Bieberbach polynomials, American Mathematical
Society, Providence, R.I., 1974, Translated from the Russian by R. P. Boas.
[31] G. Szegő, Orthogonal polynomials, fourth ed., Colloquium Publications, Vol. XXIII, American Mathe-
matical Society, Providence, R.I., 1975.
[32] V. Totik, Christoffel functions on curves and domains, preprint.
[33] , Orthogonal polynomials, Surv. Approx. Theory 1 (2005), 70–125.
[34] L. N. Trefethen, Ten-digits algorithms, Report 05/13, Oxford University Computing Laboratory, 2005.
[35] J. L. Walsh, A sequence of rational functions with application to approximation by bounded analytic
functions, Duke Math. J. 30 (1963), 177–189.
51
[36] , Interpolation and approximation by rational functions in the complex domain, fourth ed., Col-
loquium Publications, Vol. XX, American Mathematical Society, Providence, R.I., 1965.
[37] H. Widom, Polynomials associated with measures in the complex plane, J. Math. Mech. 16 (1967),
997–1013.
[38] , Extremal polynomials associated with a system of curves in the complex plane, Advances in
Math. 3 (1969), 127–232 (1969).
52

Bergman Polynomials On An Archipelago

Uploaded by

Copyright:

Available Formats

Bergman Polynomials On An Archipelago

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bergman Polynomials On An Archipelago

Uploaded by

Copyright:

Available Formats

Bergman polynomials on an Archipelago: Estimates, Zeros and Shape

Björn Gustafssona, , Mihai Putinarb, , Edward B. Saffc , Nikos Stylianopoulosd

Archipelago n. (pl. archipelagos or archipelagoes) an extensive group of islands.

Preprint submitted to Elsevier November 13, 2008

R(ρ)(t, θ) = (ρ(x, y), δ(x cos θ + y sin θ − t))

Denoting the power moments (with respect to the real variables) by

we obtain a linear system

2.1. General notations and definitions

D := {z ∈ C : |z| < 1}, ∆ := {z ∈ C : |z| > 1} ∪ {∞}, C := C ∪ {∞}.

2.2. Bergman spaces and polynomials

where kf kL2 (G) := hf, f i1/2 . Thus,

f (ζ) = hf, K(·, ζ)i, ∀ f ∈ L2a (G). (2.2)

and the finite sections of K(z, ζ) and Λ(z):

2.3. Potential theoretic preliminaries

for every function f continuous on C.

In particular, if Qn is a monic polynomial of degree n, then

The potential U ν of ν can be constructed as the smallest superharmonic function in C satisfying U ν ≥ U µ

LR := {z ∈ Ω : gΩ (z, ∞) = log R} = {z ∈ Ω : |Φ(z)| = R} (2.13)

and the open sets

ΩR := {z ∈ Ω : gΩ (z, ∞) > log R} = {z ∈ Ω : |Φ(z)| > R} = ext(LR ),

gΩR (z, ∞) = gΩ (z, ∞) − log R, (2.14)

hence the capacity of LR (or G R ) is

Rj := sup{R : Gj,R contains no other island than Gj }.

Figure 1: Green level curves

In view of (2.16), we can express K(z, ζ) in terms of conformal mappings ϕj : Gj → D, j = 1, 2, . . . , N .

ϕ0j (z)ϕ0j (ζ)

3.2. Regular measures

Proposition 3.1. The following assertions hold:

lim |Pn (z)|1/n = |Φ(z)|. (3.3)

The convergence is uniform on compact subsets of C \ Co(G).

locally uniformly with respect to z in G.

3.3. Carleman estimates

3.4. Comparison of area and line integrals of polynomials

where k · kL2 (Γ) denotes the L2 -norm on Γ with respect to |dz|.

cf. Lemma 8.1 below. Based on the above two simple

5. Reconstruction of the archipelago from moments

2. Plot the zeros of Pn , n = 1, 2, . . . , n.

3. Form Λn (z) as in (5.1).

6.1. General statements

(recall (3.7)) and define

i.e., the full sequence converges to β.

where bj is the period in (2.12). Thus, we easily deduce the following:

Case I: Two singular boundaries.

Case II: One singular boundary and one analytic boundary.

is connected and is a domain of involution of the Schwarz reflection z 7→ S2 (z).

It follows that the multi-valued function

(ii) For any weak* cluster point σ of {νPn }, supp σ ⊂ E, and

The corollary is illustrated in the following example.

and the multi-valued function

is (locally) analytic in C \ E and (locally) continuous on C \ E, where now

Case V: Three analytic boundaries.

E 0 = G2, 1 ∪ G3, 1 and E 00 = G1, 1 ,

7. An example: lemniscate islands

Let G := {z : |z m − 1| < rm }, m ≥ 2 an integer and 0 < r < 1. Then G consists of m islands

Pkm+s (z) = z s Qk,s (z m ), deg Qk,s = k. (7.2)

1/r2 = 1.234567 . . . , 1/r = 1.111111 . . . , 1.

pkm+m−1 (z) = z m−1 (z m − 1)k (7.5)

and for s = 0, 1, . . . , m − 2, we have for k sufficiently large,

(−1)k+1 k 2+τ /2 e2πij(s+1)/m z m−1 τ Γ(τ /2) sin(τ π/2)