BEHAVIOR OF THE BERGMAN KERNEL AND METRIC NEAR CONVEX BOUNDARY POINTS 2 0 NIKOLAI NIKOLOV AND PETER PFLUG 0 2 Abstract. The boundary behavior of the Bergman metric near n a convex boundary point z0 of a pseudoconvex domain D ⊂ Cn a is studied; it turns out that the Bergman metric at points z ∈ D J 0 in direction of a fixed vector X0 ∈ Cn tends to infinite, when z is 1 approaching z0, if and only if the boundary of D does not contain any analytic disc through z0 in direction of X0. ] V C For a domain D ⊂ Cn we denote by L2(D) the Hilbert space of all . h h holomorphic functions f that are square-integrable and by ||f|| the D t a L -norm of f. Let K (z) be the restriction on the diagonal of the 2 D m Bergman kernel function of D. It is well-known (cf. [5]) that [ K (z) = sup{|f(z)|2 : f ∈ L2(D), ||f|| ≤ 1}. D h D 1 v IfK (z) > 0forsomepointz ∈ D,thentheBergmanmetricB (z;X), D D 8 X ∈ Cn, is well-defined and can be given by the equality 8 0 M (z;X) D 1 B (z;X) = , D 0 K (z) D 2 0 where p / h M (z;X) = sup{|f′(z)X| : f ∈ L2(D), ||f|| = 1, f(z) = 0}. t D h D a We say that a boundary point z of a domain D ⊂ Cn is convex if m 0 there is a neighborhood U of this point such that D ∩U is convex. : v In [4], Herbort proved the following i X Theorem 1. Let z be a convex boundary point of a bounded pseudo- r 0 a convex domain D ⊂ Cn whose boundary contains no nontrivial germ of an analytic curve near z . Then 0 lim B (z;X) = ∞ D z→z0 for any X ∈ Cn \{0}. Herbort’s proof is mainly based on Ohsawa’s ∂¯-technique. The main purpose of this note is to generalize Theorem 1 using more elementary methods. 2000 Mathematics Subject Classification. Primary: 32A25. Key words and phrases. Bergman kernel, Bergman metric. 1 2 NIKOLAI NIKOLOV AND PETER PFLUG For a convex boundary point z of a domain D ⊂ Cn we denote by 0 L(z ) the set of all X ∈ Cn for which there exists a number ε > 0 0 X such that z + λX ∈ ∂D for all complex numbers λ, |λ| ≤ ε . Note 0 X that L(z ) is a complex linear space. 0 Then our result is the following one. Theorem 2. Let z be a convex boundary point of a bounded pseudo- 0 convex domain D ⊂ Cn and let X ∈ Cn. Then (a) liminfK (z)dist2(z,∂D) ∈ (0,∞]; D z→z0 (b) lim B (z;X) = ∞ if and only if X 6∈ L(z ). Moreover, this D 0 z→z0 limit is locally uniform in X 6∈ L(z ); 0 (c) If L(z ) = {0}, then (a) and (b) are still true without the as- 0 sumption that D is bounded. Proof of Theorem 2. Toprove(a)and(b)wewillusethefollowing localization theorem for the Bergman kernel and metric [2]. Theorem 3. Let D ⊂ Cn be a bounded pseudoconvex domain and let V ⊂⊂ U be open neighborhoods of a point z ∈ ∂D Then there exists a 0 constant C˜ ≥ 1 such that C˜−1K (z) ≤ K (z) ≤ K (z), D∩U D D∩U C˜−1B (z;X) ≤ B (z;X) ≤ C˜B (z;X) D∩U D D∩U for any z ∈ D ∩ V and any X ∈ Cn. (Here K (z) and B (z;·) D∩U D∩U denote the Bergman kernel and metric of the connected component of D∩U that contains z.) So, we may assume that D is convex. To prove part (a) of Theorem 2, for any z ∈ D we choose a point z˜∈ ∂D such that ||z −z˜|| = dist(z,∂D). We denote by l the complex line through z and z˜. By the Oshawa-Takegoshi extension theorem for L2–holomorphic functions [7], it follows that there exists a constant C > 0 only depending on the diameter of D (not on l) such that 1 (1) K (z) ≥ C K (z). D 1 D∩l Since D ∩l is convex, it is contained in an open half-plane Π of the l-plane with z˜∈ ∂Π. Then 1 (2) K (z) ≥ K (z) = . D∩l Π 4πdist2(z,∂Π) Now, part (a) of Theorem 2 follows from the inequalities (1), (2) and the fact that dist(z,∂Π) ≤ ||z −z˜|| = dist(z,∂D). To prove part (b) of Theorem 2, we denote by N(z ) the complex 0 affine space through z that is orthogonal to L(z ). Set E(z ) = D ∩ 0 0 0 3 N(z ). Note that E(z ) is a nonempty convex set. So, part (b) of 0 0 Theorem 2 will be a consequence of the following Theorem 4. Let z be a boundary point of a bounded convex domain 0 D ⊂ Cn. Then: (i) lim B (z;X) = ∞ locally uniformly in X 6∈ L(z ); D 0 z→z0 (ii) for any compact set K ⊂⊂ E(z ) there exists a constant C > 0 0 such that B (z;X) ≤ C||X||, z ∈ K0, X ∈ L(z ), D 0 where K0 := {z +tz : z ∈ K, 0 < t ≤ 1} is the cone generated by K. 0 Proof of Theorem 4. To prove (i) we will use the well-known fact thattheCarath´eodorymetricC (z;X)ofD doesnotexceedB (z;X). D D On the other hand, we have the following simple geometric inequality [1]: 1 C (z;X) ≥ , D 2d(z;X) where d(z;X) denotes the distance from z to the boundary of D in the X-direction, i.e. d(z;X) := sup{r : z +λX ∈ D, λ ∈ C, |λ| < r}. So, if we assume that (i) does not hold, then we may find a number a > 0 and sequences D ⊃ (z ) , z → z , Cn ⊃ (X ) , X → X 6∈ L(z ) j j j 0 j j j 0 1 such that B (z ;X ) ≤ . Hence d(z ;X ) ≥ a which implies that D j j j j 2a for |λ| ≤ a the points z +λX belong to D¯ and, in view of convexity, 0 they belong to ∂D. This means that X ∈ L(z ) – contradiction. 0 To prove part (ii) of Theorem 4, we may assume that z = 0 and 0 L := L(0) = {z ∈ Cn : z = ... = z = 0} for some k < n. Then 1 k N := N(0) = {z ∈ Cn : z = ... = z = 0}. From now on we will k+1 n write any point z ∈ Cn in the form z = (z′,z′′), z′ ∈ Ck, z′′ ∈ Cn−k. Note that L ∈ ∂D near 0, i.e. there exists an c > 0 such that (3) {0′}×∆′′ ∈ ∂D, c where ∆′′ ⊂ Cn−k is the polydisc with center at the origin and radius c c. Since K ⊂⊂ E := E(0) and since E is convex, there exists an α > 1 such that K ⊂⊂ E , where E := {z : αz ∈ E}. Note that K0 ⊂ E . α α α Using (3), the following equality 1 1 1 (z′,z′′) = (αz′,0′′)+(1− )(0′,(1− )−1z′′), α α α and the convexity of D, it follows that (4) F ×∆′′ ⊂ D, α ε 4 NIKOLAI NIKOLOV AND PETER PFLUG 1 where ε := c(1 − ) and where F is the projection of E in Ck (we α α α can identify E with F ). For δ := c(α − 1) we get in the same way α α that (5) D˜ := D ∩(Ck ×∆′δ′) ⊂ F1 ×∆′δ′. α Now, let (z,X) ∈ K0 ×L. Note that z = (z′,0′′) and X = (0′,X′′). Then, using (4) and the product properties of the Bergman kernel and metric, we have (6) MD(z;X) ≤ MFα×∆′ε′(z;X) = = M∆′ε′(0′′;X′′) KFα(z′) ≤ C1||X|| KFα(z′) for some constant C1 > 0. pOn the other side, spince K0 ⊂⊂ Ck ×∆′δ′, in virtue of Theorem 3 there exists a constant C˜ ≥ 1 such that K (z) ≥ C˜−1K (z). D D˜ Moreover, in view of (5), we have KD˜(z) ≥ KF1(z′)K∆′′(0′′) δ α and hence (7) K (z) ≥ (C )2K (z′) D 2 F1 α for some constant C > 0. Now, by (6) and (7), it follows that 2 M (z;X) C K (z′) (8) B (z;X) = D ≤ 1||X|| Fα . D KD(z) C2 sKF1(z′) α Note that z′ → α−2z′ is pa biholomorphic mapping from F1 onto Fα α and, therefore, (9) K (z′) = α−4kK (α−2z′). F1 Fα α In view of (8) and (9), in order to finish (ii) we have to find a constant C > 0 such that 3 (10) K (z′) ≤ C K (α−2z′) Fα 3 Fα for any z′ ∈ H0 with H0 := {tz′ : z′ ∈ H, 0 < t ≤ 1}, where H is the projection of K into Ck (we can identify K with H). To do this, note first that γ := dist(H,∂F ) > 0 since K ⊂⊂ E . α α Fix τ ∈ (0,1] and z′ ∈ H0, and denote by Tτ,z′ the translation that maps the origin in the point τz′. It is easy to check that (11) Tτ,z′(F¯α ∩Bγ) ⊂ Fα, 5 where B is the ball in Ck with center at the origin and radius γ. To γ prove (10), we will consider the following two cases: Case I. z′ ∈ H0 \Bγ ⊂⊂ Fα: Then 2 m (12) K (z′) ≤ 1K (α−2z′), Fα m Fα 2 where m := sup K and m := infK . 1 Fα 2 Fα H0\Bγ 2 Case II. z′ ∈ H0∩Bγ: By Theorem 3 there exists a constant C˜3 ≥ 1 2 such that C˜3KFα ≥ KFα∩Bγ on Fα ∩Bγ2. In particular, (13) C˜ K (α−2z′) ≥ K (α−2z′). 3 Fα Fα∩Bγ On the other side, by (11) with data T := T1−α−2,z′ it follows that (14) K (α−2z′) = K (z′) ≥ K (z′). Fα∩Bγ T(Fα∩Bγ) Fα m Now,(12),(13),and(14)implythat(10)holdsforC := max{ 1,C˜ }. 3 m 3 2 This completes the proofs of Theorem 4 and part (b) of Theorem 2. Remark. The approximation (5) of the domain D ∩(Ck ×∆′′) by δ the domain E1 ×∆′δ′ can be replaced by using the Oshawa-Takegoshi α theorem [7] with the data D and N. Finally, part (c) of Theorem 2 will be a consequence of the following two theorems. Theorem 5. [6] Let D ⊂ Cn. be a pseudoconvex domain and let U be an open neighborhood of a local (holomorphic) peak point z ∈ ∂D. 0 Then K (z) D lim = 1 z→z0 KD∩U(z) and B (z;X) D lim = 1 z→z0 BD∩U(z;X) locally uniformly X ∈ Cn \{0}. Theorem 6. Let z be a boundary point of a bounded convex domain 0 D ⊂ Cn. Then the following conditions are equivalent: 1. z is a (holomorphic) peak point; 0 2. z is the unique analytic curve in D¯ containing z ; 0 0 3. L(z ) = {0}. 0 Notethattheonlynontrivialimplicationis(3) =⇒ (1)itiscontained in [8]. Now, part (c) of Theorem 2 is a consequence of this implication, Theorem 5, and part (b) of Theorem 2. 6 NIKOLAI NIKOLOV AND PETER PFLUG Proof of Theorem 6. The implication (2) =⇒ (3) is trivial. The implication (1) =⇒ (2) easily follows by the maximum principle andthefactthat therearea neighborhoodU ofz andavector X ∈ Cn 0 such that (D¯ ∩U)+(0,1]X ⊂ D (cf. (11)). Denote by A0(D) the algebra of holomorphic functions on D which are continuous on D¯. Now, following [8] we shall prove the implication (3) =⇒ (1); namely, (3) implies that z is a peak point with respect to 0 A0(D). This is equivalent to the fact (cf. [3]) that the point mass at z is the unique element of the set A(z ) of all representing measures 0 0 for z with respect to A0(D), i.e. supp µ = {z } for any µ ∈ A(z ). 0 0 0 Let µ ∈ A(z ). Since D is convex, we may assume that z = 0 and 0 0 D ⊂ {z ∈ Cn : Re(z ) < 0}. Note that if a is a positive number 1 such that a inf Re(z ) > −1 (D is bounded), then the function f (z) = 1 1 z∈D exp(z +az2) belongs to A0(D) and |f (z)| < 1 for z ∈ D¯ \{z = 0}. 1 1 1 1 This easily implies (cf. [3]) that supp µ ⊂ D := ∂D∩{z = 0}. Since 1 1 L(0) = 0, the origin is a boundary point of the compact convex set D . As above, we may assume that D ⊂ {z ∈ Cn : Re(z ) ≤ 0} (z is 1 1 2 2 independent of z ) and then construct a function f ∈ A0(D) such that 1 2 |f (z)| < 1forz ∈ D \{z = 0}. Thisimpliesthatsupp µ ⊂ D ∩{z = 2 1 2 1 2 0}. Repeating this argument we conclude that supp µ = {0}, which completes the proofs of Theorems 6 and 2. Acknowledgments. This paper was written during the stay of the first author at the University of Oldenburg supported by a grant from the DAAD. He likes to thank both institutions, the DAAD and the Mathematical Department of the University of Oldenburg. References [1] E. Bedford, S. Pinchuk, Convex domains with noncompact automorphism groups, Sb. Math. 82 (1995), 3-26. [2] K. Diederich, J.E. Fornaess, G. Herbort, Boundary behavior of the Bergman metric, Proc. Symp. Pure Math. 41 (1984), 59-67. [3] T.W. Gamelin, Uniform algebras, Chelsea, New York, 1984. [4] G. Herbort, On the Bergman metric near a plurisubharmonic barrier point, Prog. Math. 188 (2000), 123-132. [5] M.Jarnicki,P.Pflug,Invariant distances and metrics in complex analysis, Wal- ter De Gruyter, Berlin, New York, 1993. [6] N. Nikolov, Localization of invariant metrics, Arch. Math. (to appear). [7] T.Ohsawa,K.Takegoshi,Extension of L2 holomorphic functions,Math.Z.195 (1987), 197-204. [8] N. Sibony, Une classe de domaines pseudoconvexes, Duke Math. J. 55 (1987), 299-319. 7 Institute of Mathematics and Informatics, Bulgarian Academy of Sciences, 1113 Sofia, Bulgaria E-mail address: [email protected] CarlvonOssietzkyUniversita¨tOldenburg,FachbereichMathematik, Postfach 2503, D-26111 Oldenburg,Germany E-mail address: [email protected]