Bergman polynomials on an Archipelago: Estimates, Zeros and Shape Reconstruction(cid:73),(cid:73)(cid:73) Bj¨orn Gustafssona,, Mihai Putinarb,, Edward B. Saffc, Nikos Stylianopoulosd aDepartment of Mathematics, The Royal Institute of Technology, S-10044, Stockholm, Sweden bDepartment of Mathematics, University of California at Santa Barbara, Santa Barbara, California, 93106-3080, USA cCenter for Constructive Approximation, Department of Mathematics, Vanderbilt University, 1326 Stevenson Center, 37240 Nashville, USA dDepartment 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 |zm −1| < rm,0 < r < 1, which consistsofmislands. TheasymptoticanalysisoftheChristoffelfunctionsassociatedtothesameorthogonal 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 Szeg˝o 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. (cid:73)ToChristineChodkiewicz-Putinar,whohasenrichedandinspiredusbyaddingasecondviolatoourquartet (cid:73)(cid:73)ThefirstauthorwaspartiallysupportedbygrantsfromtheSwedishResearchCouncil,theG¨oranGustafssonFoundation andtheEuropeanNetworkHCAA.ThesecondandthirdauthorswerepartiallysupportedbytheNationalScienceFoundation, USA,undergrantsDMS-0701094,DMS-0603828andDMS-0808093. ThefourthauthorwassupportedbyaUniversityofCyprus research grant. All authors are indebted to the Mathematical Research Institute at Oberwolfach, Germany, which provided exceptionalworkingconditionsforourcollaborativeefforts. Email addresses: [email protected](Bj¨ornGustafsson),[email protected](MihaiPutinar), [email protected](EdwardB.Saff),[email protected](NikosStylianopoulos) Preprint submitted to Elsevier November 13, 2008 1. Introduction The study of orthogonal polynomials, resurrected recently by many groups of scientists, some departing fromtheclassicalframeworkofconstructiveapproximationtofieldsasfarasquantumcomputingornumber theory,doesnotneedanintroduction. Maybeonlyourpredilectioninthepresentworkforcomplexanalytic 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 1923toalandmarkarticlebyT.Carleman[3]. AboutthesametimeS.Bernsteindiscoveredthattheanalogue 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 generalitybyJ.L.Walsh[36]. Then,itcameasnosurprisethatgoodapproximationsofconformalmappings ofsimply-connectedplanardomainsbearontheBergmanorthogonalpolynomials,thatisthosewithrespect 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 completeasof today. Forinstance, itisofrecentdatethatthe analoguesofBlaschkeproductsassociatedto 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 thearchipelagofromafinitesetoftheassociatedpowermoments. Thespecificchoiceoftheaboveproblems anddegreeofgeneralityweredictatedbythepresentstatusofthetheoryofcomplexorthogonalpolynomials. A brief description of the subjects touched in this article follows. Let G = ∪N G be an archipelago, j=1 j 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: P (z)=λ zn+··· , λ >0, n=0,1,2,..., n n n carryinarefined(onewouldbeinclinedtosay,aristocratic)mannertheinformationaboutG. Forinstance, simplelinearalgebraprovidesaconstructivebijectionbetweenthesequence{P }∞ andthepowermoments n n=0 (correlation matrix entries) (cid:90) µ (G):= znzmdA, m,n≥0, (1.1) mn G where dA stands for the area measure on C. Three major features distinguish Bergman orthogonal polyno- mials: (i) An extremality property: P /λ is the minimum L2(G,dA)-norm monic polynomial of degree n, n n (cid:80) (ii) the Bergman kernel K(z,ζ)= ∞ P (ζ)P (z) collects into a condensed form the (derivatives of the) j=0 j j conformal mappings from the disk to every connected component G , j 2 (cid:110) (cid:111) (cid:80) −1/2 (iii) the square root of the Christoffel function Λ (z) := n |P (z)|2 is the extremum value n j=0 j min(cid:107)q(cid:107) , q(z)=1, deg q ≤n. L2(G,dA) 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{g (z,∞)+ig∗(z)}, z ∈C\G, Ω Ω where g (z,∞) is the Green function of the exterior domain Ω := C\G, with a pole at infinity, and g∗ is Ω Ω any harmonic conjugate of g . We designate the name Walsh function for Φ. At a critical moment in our Ω proofs, we rely on the pioneering work of Widom [38] that refers to Szeg˝o’s orthogonal polynomials on G andtheirintimaterelationtotheWalshfunctionΦ. OurBergmanspacesetting,however,departsinquitea few essential points from the Hardy-Smirnov space scenario. Both estimates of the growth of P (z) and the n limiting distribution of the zero sets of {P }∞ depend heavily on Φ and its analytic continuation across n n=1 ∂G. While the estimates for P (z) are more or less expected, and only how to prove them might bring n 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 P , n = 1,2,..., lie on the m n radiallinesjoiningthecentertothevertices,form=3andm=4(see[13]),whileform≥5everyboundary point of the m-gon attracts zeros of P , as n→∞ (see [2, Thm. 5]). n As a byproduct of the estimates we have obtained for Λ (z), we propose a very accurate reconstruction- n 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(xcosθ+ysinθ−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 (cid:90) (cid:90) a (θ)= tkR(ρ)(t,θ)dt= (xcosθ+ysinθ)kρ(x,y)dxdy. k R R2 Denoting the power moments (with respect to the real variables) by (cid:90) σ = xjykρ(x,y)dxdy, i,j ≥0, j,k R2 we obtain a linear system (cid:181) (cid:182) (cid:88)k k a (θ)= cosiθsink−iθ σ . k i i,k−i i=0 After giving θ a number of distinct values, and noticing that the determinant of the system is non-zero, one findsbylinearalgebrathevalues{σ }n . Thisprocedurewasusedbythefirsttwoauthorsofthispaper j,k j,k=0 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. Thepaperisorganizedasfollows: Sections2and3aredevotedtonecessarybackgroundinformation. We introducetherethenotation,conventionsandrecallcertainfactsfrompotentialtheoryandfunctiontheoryof 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 suppµ. 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 P . n 2.2. Bergman spaces and polynomials The main characters in our story are the Bergman orthogonal polynomials associated to an archipelago G := ∪N G , where G ,...,G are Jordan domains with mutually disjoint closures. Set Γ := ∂G and j=1 j 1 N j j Γ:=∪N Γ . For later use we introduce also the exterior domain Ω:=C\G. Note that Γ=∂G=∂Ω. j=1 j Let {P }∞ denote the sequence of Bergman orthogonal polynomials associated with G. This is defined n n=0 as the sequence of polynomials P (z)=λ zn+··· , λ >0, n=0,1,2,..., n n n that are obtained by orthonormalizing the sequence 1,z,z2,..., with respect to the inner product (cid:90) (cid:104)f,g(cid:105):= f(z)g(z)dA. G 4 Equivalently,thecorrespondingmonicpolynomialsP (z)/λ ,canbedefinedastheuniquemonicpolynomials n n of minimal L2-norm over G: 1 (cid:107) P (cid:107) =m (G,dA):= min (cid:107)zn+r(z)(cid:107) , (2.1) λn n L2(G) n r∈Pn−1 L2(G) where (cid:107)f(cid:107) :=(cid:104)f,f(cid:105)1/2. Thus, L2(G) 1 =m (G,dA). λ n n Let L2(G) denote the Bergman space associated with G and (cid:104)·,·(cid:105): a (cid:169) (cid:170) L2(G):= f analytic on G and (cid:107)f(cid:107) <∞ . a L2(G) Note that L2(G) is a Hilbert space that possesses a reproducing kernel which we denote by K(z,ζ). That a is, K(z,ζ) is the unique function K(z,ζ):G×G→C such that, for all ζ ∈G, K(·,ζ)∈L2(G) and a f(ζ)=(cid:104)f,K(·,ζ)(cid:105), ∀f ∈L2(G). (2.2) a Furthermore,duetothereproducingpropertyandthecompletenessofpolynomialsinL2(G)(seeLemma3.3 a below), the kernel K(z,ζ) is given in terms of the Bergman polynomials by (cid:88)∞ K(z,ζ)= P (ζ)P (z). j j j=0 We single out the square root of the inverse of the diagonal of the reproducing kernel of G 1 Λ(z):= (cid:112) , z ∈G, K(z,z) and the finite sections of K(z,ζ) and Λ(z): (cid:88)n 1 K (z,ζ):= P (ζ)P (z), Λ (z):= (cid:112) . (2.3) n j j n K (z,z) j=0 n We note that the Λ (z)’s are square roots of the so-called Christoffel functions of G. n 2.3. Potential theoretic preliminaries Let Q be a polynomial of degree n with zeros z ,z ,...,z . The normalized counting measure of the 1 2 n zeros of Q is defined by 1 (cid:88)n ν := δ , (2.4) Q n zk k=1 where δ denotes the unit point mass at the point z. In other words, for any subset A of C, z number of zeros of Q in A ν (A)= . Q n 5 Next, given a sequence {σ } of Borel measures, we say that {σ } converges in the weak∗ sense to a n n ∗ measure σ, symbolically σ −→σ, if n (cid:90) (cid:90) fdσ −→ fdσ, n→∞, 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 (cid:90) 1 Uσ(z):= log dσ(t). |z−t| In particular, if Q is a monic polynomial of degree n, then n 1 UνQn(z)=−nlog|Qn(z)|. Let Σ ⊂ C be a compact set. Then there is a smallest number γ ∈ R∪{+∞} such that there exists a probabilitymeasureµΣ onΣwithUµΣ ≤γ inC. The(logarithmic) capacity ofΣisdefinedascap(Σ)=e−γ (interpreted as zero if γ =+∞). If cap(Σ)>0, then µ is unique and is called the equilibrium measure of Σ Σ. For the definition of capacity of more general sets than compact sets see, e.g., [20] and [24]. A property thatholdseverywhere, exceptonasetofcapacityzero, issaidtoholdquasi everywhere (q.e.). Forexample, it is known that UµΣ =γ, q.e. on Σ. Let W denote the unbounded component of C\Σ. It is known that supp(µ ) ⊂ ∂W, µ = µ and Σ Σ ∂W cap(Σ)=cap(∂W). Ifcap(Σ)>0,thentheequilibriumpotentialisrelatedtotheGreenfunctiong (z,∞) W of W, with pole at infinity, by 1 UµΣ(z)=log −g (z,∞), z ∈W. (2.5) cap(Σ) W Inourapplications∂W willbeafinitedisjointunionofmutuallyexteriorJordancurves(typicallyΣ=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) suppµ =∂W, (2.6) Σ (ii) 1 UµΣ(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 Uν of ν can be constructed as the smallest superharmonic function in C satisfying Uν ≥ Uµ in C\Σ. Since Uµ itself is competing it follows that, in addition to (2.8), Uν ≤Uµ in all C. 6 2.4. The Green function and its level curves Returning to the archipelago, let g (z,∞) denote the Green function of Ω=C\G with pole at infinity. Ω That is, g (z,∞) is harmonic in Ω\{∞}, vanishes on the boundary Γ of G and near ∞ satisfies Ω 1 1 g (z,∞)=log|z|+log +O( ), |z|→∞. (2.9) Ω 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{g (z,∞)+ig∗(z,∞)}, (2.10) Ω Ω where g∗(z,∞) is a (locally) harmonic conjugate of g (z,∞) in Ω. In the single-component case N = 1, Ω Ω (2.10) defines a conformal mapping from Ω onto ∆. In the multiple-component case N ≥ 2, Φ is a multi- valuedanalyticfunctioninΩ. However, |Φ(z)|issingle-valued. WerefertoWalsh[36, §4.1]andWidom[38, § 4] for comprehensive accounts of the properties of Φ. We note in particular that Φ is single-valued near infinity and, since g∗ is unique apart from a constant, that it can be chosen so that Φ has near infinity a Ω Laurent series expansion of the form 1 α α Φ(z)= z+α + 1 + 2 +··· . (2.11) cap(Γ) 0 z z2 We also note that Φ(cid:48)(z)/Φ(z)=2∂g (·,∞)/∂z is single-valued and analytic in Ω, with periods Ω (cid:90) (cid:90) 1 Φ(cid:48)(z) 1 ∂g (z,∞) b := dz = Ω ds, j =1,2,...,N. (2.12) j 2πi Φ(z) 2π ∂n Γj Γj Here Γ is oriented as the boundary of G and the normal derivative is directed into Ω. If Γ is not smooth j j (cid:80) j the path of integration in (2.12) is understood to be moved slightly into Ω. Note that N b =1. j=1 j Next we consider for R≥1 the level curves (or equipotential loci) of the Green function, L :={z ∈Ω: g (z,∞)=logR}={z ∈Ω: |Φ(z)|=R} (2.13) R Ω and the open sets Ω :={z ∈Ω: g (z,∞)>logR}={z ∈Ω: |Φ(z)|>R}=ext(L ), R Ω R G :=C\Ω =int(L ). R R R Note that L = Γ, Ω = Ω, G = G. It follows from the maximum principle that Ω is always connected. 1 1 1 R The Green function for Ω is given by R g (z,∞)=g (z,∞)−logR, (2.14) ΩR Ω hence the capacity of L (or G ) is R R cap(L )=Rcap(Γ). (2.15) R 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, G consists of N components, each of which contains exactly one component of R G, while for large values of R, G is connected (with G ⊂ G ). Consequently, we introduce the following R R sets and numbers: 7 G := the component of G that contains G , j =1,2,...,N. j,R R j L :=∂G , j =1,2,...,N. j,R j,R R :=sup{R: G contains no other island than G }. j j,R j R(cid:48) :=min{R ,...,R }=sup{R: G has N exactly components }. 1 N R R(cid:48)(cid:48) :=inf{R: G is connected }=inf{R: Ω is simply connected }. R R Thus, when 1<R<R(cid:48), G is the disjoint union of the domains G , j =1,2,...,N and L consists of the R j,R R N mutually exterior analytic Jordan curves L , j =1,2,...,N, while for R>R(cid:48)(cid:48), we have G =G = j,R 1,R 2,R ···=G =G and L is a single analytic curve. N,R R R Itiswell-knownthatg (z,∞)hasexactlyN−1criticalpoints(multiplicitiescounted),i.e.,pointswhere Ω the gradient ∇g (z,∞), or equivalently Φ(cid:48)/Φ=2∂g (·,∞)/∂z, vanishes. These critical points show up as Ω Ω singularities of L , which are points of self-intersection. Such singularities must appear when L changes R R topology. It follows that there are no critical points in G \G, at least one critical point on each L , R(cid:48) Rj j =1,2,...,N (one of them is L ), at least one on L and no critical point in Ω . Any L that does R(cid:48) R(cid:48)(cid:48) R(cid:48)(cid:48) j,R not contain a critical point is an analytic Jordan curve. In particular, this applies whenever 1<R <R(cid:48) or R(cid:48)(cid:48) <R<∞. When R≥R(cid:48)(cid:48), Φ is the uniqueconformal map of Ω onto∆ :={w :|w|>R} thatsatisfies (2.11) near R R infinity. 2 G3 L R 1 R>R00 LR0 L R00 L R0 0 G1 G2 −1 −1 0 1 2 3 Figure1: Greenlevelcurves InFigure1weillustratethethreedifferenttypesoflevelcurvesL ,L andL withR>R(cid:48)(cid:48),introduced R(cid:48) R(cid:48)(cid:48) R above. Remark 2.1. ThelevelcurvesinFigure1werecomputedbymeansofTrefethen’sMATLABcodemanydisks.m [34]. This code provides an approximation to the Green function g (z,∞) in cases when G consists of a Ω finite number of disks, realizing an algorithm given in [5]. 8 Consider now the N Hilbert spaces L2(G ) defined by the components G , j = 1,2,...,N, and let a j j KGj(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 KGj(z,ζ) as follows: (cid:189) K(z,ζ)= KGj(z,ζ) if z, ζ ∈Gj, (2.16) 0 if z ∈G , ζ ∈G , j (cid:54)=k. j k In view of (2.16), we can express K(z,ζ) in terms of conformal mappings ϕ : G → D, j = 1,2,...,N. j j This will help us to determine the singularities of K(·,ζ) and, in particular, whether or not this kernel has a singularity on ∂G . This is so because, as it is well-known (see e.g. [6, p. 33]), j ϕ(cid:48)(z)ϕ(cid:48)(ζ) KGj(z,ζ)= (cid:104) j j (cid:105)2, z,ζ ∈Gj, j =1,2,...,N. (2.17) π 1−ϕ (z)ϕ (ζ) j j BysayingthatafunctionanalyticinG hasasingularityon∂G ,wemeanthatthereisnoopenneighborhood j j of G in which the function has an analytic continuation. j 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)=z¯ for z ∈Γ; see [4] and [25]. The map z (cid:55)→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(cid:48)(z)(cid:54)=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(cid:54)=0 in D, where∇udenotesthegradientofu. ThenΓisananalyticcurve,theSchwarzfunctionS(z)ofΓisdefinedin all D, and z (cid:55)→S(z) maps D onto itself. Moreover, u and S(z) are related by (3.1). In particular z (cid:55)→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 (cid:55)→ 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 Γ , is analytic. 1 ThentheGreenfunctionu(z)=g (z,∞)extendsharmonically, bytheSchwarzreflection(3.1), fromΩinto Ω G . Wekeepthenotationg (z,∞)forsoextendedGreenfunction. Recallthatthelevellinesreflecttolevel 1 Ω lines, so that for R>1 close enough to one, L is reflected to the level line 1,R 1 L ={z ∈G :g (z,∞)=−logR}={z ∈G :|Φ(z)|= } 1,R1 1 Ω 1 R of the extended Green function (and extended Φ). Generally speaking, whenever applicable we shall keep notations like L , G , L , Ω etc. for values ρ<1 in case of analytic boundaries. j,ρ j,ρ ρ ρ As was previously remarked, u(z) = g (z,∞) has no critical points in the region G \G . It follows Ω 1,R1 1 then from (3.1) that if the Green function extends harmonically to a region G \G with 1 ≤ρ<1, then 1 1,ρ R1 ithasnocriticalpointsthere,andtheregionD =G \G issymmetricwithrespecttoSchwarzreflection 1,1 1,ρ ρ 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| on G belongs to the class Reg. G Lemma 3.2 yields the following n-th root asymptotic behavior for the Bergman polynomials P in Ω. n Proposition 3.1. The following assertions hold: (a) 1 lim λ1/n = . (3.2) n→∞ 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 P ’s, we have n lim |P (z)|1/n =|Φ(z)|. (3.3) n n→∞ The convergence is uniform on compact subsets of C\Co(G). (c) limsup|P (z)|1/n =|Φ(z)|, z ∈Ω, (3.4) n n→∞ locally uniformly in Ω. 10