GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS JANINA KOTUS AND MARIUSZ URBANSKI Abstrat. We explore the lass of ellipti funtions whose all ritial points ontained in the Julia set are non-reurrent and whose ! -limit sets form ompat subsets of the omplex plane. In partiular, this lass omprises hyperboli, subhyperboli and paraboli ellipti maps. Let h be the Hausdor dimension of the Julia set of suh ellipti funtion f . We onstrut an atomless h-onformal measure m and we show that the h-dimensional Hausdor measure of the Julia set of f vanishes unless the Julia set is equal to the entire omplex plane C I . The h-dimensional paking measure is always positive and it is nite if and only if there are no rationally indierent periodi points. Furthermore, we prove the existene of a (unique up to a multipliative onstant) -nite f -invariant measure equivalent with m. The measure is then proved to be ergodi and onservative and we identify the set of those points whose all open neighborhoods have innite measure . In partiular we show that 1 is not among them. 1. Introdution and General Preliminaries 1.1. Introdution. First examples of ellipti (in fat }-Weierstrass) funtions with detailed desription of their Julia sets appeared in [11℄. Our paper dealing with ellipti funtions whose all ritial points ontained in the Julia set are non-reurrent and whose !-limit sets form ompat subsets of the omplex plane, basially stems from [21℄, [22℄ and [12℄. Any suh ellipti funtion will be alled non-reurrent. We study geometri properties of the Julia sets ultimately resulting in Theorem 4.1 whih says that the h-dimensional Hausdor measure of the Julia set of f vanishes unless the Julia set is equal to the entire omplex plane CI. The h-dimensional paking measure is always positive and it is nite if and only if there are no rationally indierent periodi points. We would like to emphasize that both Hausdor and paking appearing in this theorem are taken with respet to the spherial metri on CI. The fat of vanishing h-dimensional Hausdor measure of the Julia set in the ase when h < 2 (note that due to [12℄ h > 1) aused by the existene of poles, dramatially dierentiate non-reurrent ellipti funtions from the ase of analogous lass of rational funtions (see [21℄). Our main tehnial tool employed in this paper is the onept of semi-onformal, almost-onformal and onformal 2000 Mathematis Subjet Classiation. Primary 37F35. Seondary 37F10, 30D30. The researh of the rst author was supported in part by the Foundation for Polish Siene, the Polish KBN Grant No 2 P03A 009 17 and TUW Grant no 503G 112000442200. She also wishes to thank the University of North Texas where this researh was onduted. The researh of the seond author was supported in part by the NSF Grant DMS 0100078. 1 2 JANINA KOTUS AND MARIUSZ URBANSKI measures. We provide an elaborated proof of the existene, uniqueness and ontinuity of an honformal measure. Another important tool is provided by Proposition 2.21, where, expressed in an appropriate language, all non-singular points are shown to be onial. Although there are some overlaps with rational funtions (see [21℄), most of the proofs are substantially dierent, mainly beause of the existene of poles in the Julia set. Our seond major theme in this paper is the dynamis of f with respet to the onformal measure m. As the rst result in this diretion we we prove the existene of a onservative ergodi -nite measure equivalent to m. Developing this diretion, we study points of nite and innite ondensation of the measure , the onepts introdued in [22℄. After olleting some some basi fats about these points we show in Subsetion 5.2 that 1 is always a point of nite ondensation, perhaps the most interesting fat about the measure . In the next subsetion we relate points of innite ondensation with the set (f ) of rationally indierent periodi points, providing in partiular some suÆient onditions ( (f ) = ;) for the invariant measure to be nite. In the end of this setion we deal with paraboli points themselves. 1.2. General Preliminaries. Throughout the entire paper f , diams and Bs(z; r) denote respetively the derivatives, di0 ameters and open balls dened by means of the spherial metri whereas f , diam and B (z; r) are onsidered in the Eulidean sense. Denition 1.1. If H : D ! CI is an analyti map, z 2 CI, and r > 0, then by Comp(z; H (z ); H; r) we denote the onneted omponent of H 1 (B (H (z ); r)) that ontains z . Suppose now that is a ritial point of an analyti map H : D ! CI. Then there exists R = R(H; ) > 0 and A = A(H; ) 1 suh that A 1 jz jq jH (z ) H ()j Ajz jq and A 1 jz jq 1 jH 0(z )j Ajz jq 1 for every z 2 Comp(; H (); H; R) and that H (Comp(; H (); H; R)) = B (H (); R) where q = q(H; ) is the order of H at the ritial point . Moreover letting R > 0 to be suÆiently small we an require the two above inequalities to hold for every z 2 B (; (AR)1=q ) and the ball B (; (AR)1=q ) [ Comp(; H (); H; R) to be expressed as a union of the point and q open mutually disjoint sets suh that H restrited to eah of them is injetive. Koebe's Distortion Theorem, I (Eulidean version). There exists a funtion k : [0; 1) ! [1; 1) suh that for any z 2 CI; r > 0; t 2 [0; 1) and any univalent analyti funtion H : B (z; r) ! CI we have that supfjH 0(x)j : x 2 B (z; tr)g k(t) inf fjH 0(x)j : x 2 B (z; tr)g: GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 3 We put K = k(1=2). Koebe's Distortion Theorem, I (spherial version). Given a number s > 0 there exists a funtion ks : [0; 1) ! [1; 1) suh that for any z 2 CI; r > 0; t 2 [0; 1) and any univalent analyti funtion H : B (z; r) ! CI suh that the omplement CI n H (B (z; r)) ontains a ball of radius s we have supfjH (x)j : x 2 B (z; tr)g ks(t) inf fjH (x)j : x 2 B (z; tr)g: The following is straightforward onsequene of these two Koebe's Distortion Theorems. Lemma 1.2. Suppose that D CI is an open set, z 2 D and H : D ! CI is an analyti map whih has an analyti inverse Hz 1 dened on B (H (z ); 2R) for some R > 0. Then for every 0rR B (z; K 1 rjH 0(z )j 1 ) Hz 1 (B (H (z ); r)) B (z; KrjH 0 (z )j 1 ): Lemma 1.3. Suppose that D CI is an open set, z 2 D and H : D ! CI is an analyti map whih has an analyti inverse Hz 1 dened on B (H (z ); 2R) for some R > 0 avoiding a ball of some radius s. Then for every 0 r R B (z; k 1 (1=2)rjH 0(z )j 1 ) H 1 (B (H (z ); r)) B (z; k (1=2)rjH 0(z )j 1 ): s z s We shall also use the following more geometri versions of Koebe's Distortion Theorems involving moduli of annuli. Koebe's Distortion Theorem, II (Eulidean version). There exists a funtion w : (0; +1) ! [1; 1) suh that for any two open topologial disks Q1 Q2 with Mod(Q2 nQ1 ) t and any univalent analyti funtion H : Q2 ! CI suh that the omplement CIn H (Q2) ontains a ball of radius s we have supfjH 0(x)j : x 2 Q1 g w(t) inf fjH 0(x)j : x 2 Q1g: Koebe's Distortion Theorem, II (spherial version). Given a number s > 0 there exists a funtion ws : (0; +1) ! [1; 1) suh that for any two open topologial disks Q1 Q2 with Mod(Q2 n Q1) t and any univalent analyti funtion H : Q2 ! CI suh that the omplement CI n H (Q2 ) ontains a ball of radius s we have supfjH 0(x)j : x 2 Q1 g ws(t) inf fjH 0(x)j : x 2 Q1g: Lemma 1.4. Suppose that an analyti map Q Æ H : D ! CI, a radius R > 0 and a point z 2 D are suh that Comp(H (z); Q(H (z)); Q; 2R) \ Crit(Q) = ; and Comp(z; Q Æ H (z); Q Æ H; R) \ Crit(H ) 6= ; JANINA KOTUS AND MARIUSZ URBANSKI 4 If belongs to the last intersetion and diam Comp(z; Q Æ H (z); Q Æ H; R) (AR(H; ))1=q then jz j KA2 j(Q Æ H )0(z)j 1 R: 2. The Dynamis of Non-reurrent Ellipti Funtions 2.1. Preliminaries from Ellipti Funtions. As we already indiated in the introdution, throughout the entire paper f : CI ! CI denotes a non-onstant ellipti funtion. Every suh funtion is doubly periodi and meromorphi. In partiular there exist two vetors w1; w2, Im( ww ) 6= 0, suh that for every z 2 CI and n; m 2 ZZ , f (z ) = f (z + mw1 + nw2 ): The set = fmw1 + nw2 : m; n 2 ZZ g will be alled the lattie of the ellipti funtion f . This objet is independent of the hoie of its generators w1 and w2. Let R = ft1 w1 + t2 w2 : 0 t1 ; t2 1g; be the basi fundamental parallelogram of f . It follows from periodiity of f that f (CI) = f (R). Therefore f (CI) as a losed and open subset of the onneted set CI is equal to CI. This means that eah ellipti funtion is surjetive. It also follows from periodiity of f that [ f 1 (1) = R \ f 1(1) + mw1 + nw2 : 1 2 m;n2ZZ For every pole b of f let qb denote its multipliity. We dene q := maxfqb : b 2 f 1 (1)g = maxfqb : b 2 f 1 (1) \ Rg: Let BR = fz 2 CI : jz j > Rg: For every pole b of f by Bb (R) we denote the onneted omponent of f 1(BR) ontaining b. If R > 0 is large enough, say R R0, then BR ontains no ritial values of f , all sets Bb(R) are simply onneted, mutually disjoint and for z 2 Bb(R) G (z ) f (z ) = b qb (2.1) (z b) where Gb : Bb (R) ! CI is a holomorphi funtion taking values out of some neighbourhood of 0. If U BR n f1g is an open simply onneted set, then all the holomorphi inverse GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 5 1 of f are well-dened on U and for every 1 j q and all z 2 U branhes fb;U;1 1; : : : ; fb;U;q b b we have q 1 )0 (z )j jz j bqb : j(fb;U;j (2.2) Therefore qb qb 2 qb qb qb j z j 1 + j z j j z j 1 ) (z )j jz j qb j(fb;U;j (2.3) 1 + j(fb;U;j1 )(z)j2 1 + jbj2 jbj2 ; where the last omparability sign we wrote assuming in addition that jbj is large qenough, say jbj R1 > R0 . Let M be an upper bound of the ratios of j(fb;U;j1 )(z)j and jzj bqb jbj 2 with b; U; j as above. A straightforward alulation based on (2.1) shows that there exists a onstant L 1 suh that for all poles b and all R R1 we have L 1 R qb diam(Bb (R)) LR qb ; (2.4) L 1 R qb (1 + jbj2 ) 1 diams (Bb (R)) LR qb (1 + jbj2 ) 1 : We will frequently use the following fat proven in [12℄. +1 1 1 +1 1 1 1 1 1 Theorem 2.1. If f : CI ! CI is an arbitrary ellipti funtion, then HD(J (f )) > q 2+q 1 1; where q = inf fqb : b 2 inf 1 (1)g = maxfqb : b 2 R \ f 1 (1)g. 2.2. Julia Sets and Non-Reurrent Ellipti Funtions. The Fatou set F (f ) of a meromorphi funtion f : CI ! CI is dened in exatly the same manner as for rational funtions; F (f ) is the set of points z 2 CI suh that all the iterates are dened and form a normal family on a neighborhood of z. The Julia set J (f ) is the omplement of F (f ) in CI. Thus, F (f ) is open, J (f ) is losed, F (f ) is ompletely invariant while f 1(J (f )) J (f ) and f (J (f )) = J (f ) [f1g. For a general desription of the dynamis of meromorphi funtions see e.g. [5℄. We would however like to note that it easily follows from Montel's riterion of normality that if f : CI ! CI has at least one pole whih is not an omitted value then [ J (f ) = f n(1): n0 Let Crit(f ) be the set of ritial points of f i.e. Crit(f ) = fz : f 0(z) = 0g: 6 JANINA KOTUS AND MARIUSZ URBANSKI Its image, f (Crit(f )), is alled the set of ritial values of f . Sine R \ Crit(f ) is nite and sine f (Crit(f )) = f (R \ Crit(f )), the set of ritial values f (Crit(f )) is also nite. Let [ I1 (f ) = fz 2 CI : z 2 f n (1) or nlim f n (z ) = 1g !1 n0 be the set of points esaping to innity under iterates of f . We say that the ellipti funtion f : CI ! CI is non-reurrent, if the following onditions are satised: (1) If 2 Crit(f ) \ J (f ), then the !-limit set !() is a ompat subset of CI (i.e. 1 2= !()) and 2= !() (2) If 2 Crit(f ) \ F (f ) then either there exists an attrating periodi point w or a p 1 rationally indierent periodi point w suh that !() fw; f (w); : : : ; f (w)g, p is a period. From now on, unless otherwise stated, we assume throughout the entire paper that the ellipti funtion f : CI ! CI is non-reurrent. If t 0, then a measure m supported on J (f ) is said to be semi t-onformal for f : CI ! CI, if Z m(f (A)) jf jt dm (2.5) A for every Borel set A J (f ) suh that f jA is injetive and m is said to be t-onformal for f : CI ! CI, if Z m(f (A)) = jf jt dm (2.6) A for these sets A. 2.3. Loal behavior around paraboli xed points. In this setion f : CI ! CI is an arbitrary ellipti funtion of degree 2. In partiular the map f is not assumed yet to be non-reurrent. In what follows we basially summarize the results onerning loal behavior around paraboli xed points whih have been proved in [1℄, [8℄, and [9℄. Although they were formulated and proved in the ontext of paraboli rational maps that is assuming that the Julia set ontains no ritial points, nevertheless they and their proofs are of loal harater and, in partiular, extend to the lass of all ellipti funtions. Through this setion ! is a simple paraboli xed point of f , that is f (!) = ! and f 0(!) = 1. First note 1that on a suÆiently small open neighbourhood V of ! a holomorphi inverse branh f! : V ! CI of f is well dened whih sends ! to !. Moreover, V an be taken so small that on V the transformation f! 1 expresses in the form f! 1 (z ) = z a(z ! )p+1 + a2 (z ! )p+2 + a3 (z ! )p+3 + : : : (2.7) where a 6= 0 and p = p(!) is a positive integer. f! 1 (z ) ! = z ! a(z ! )p+1 + a2 (z ! )p+2 + a3 (z ! )p+3 + : : : Consider the set fz : a(z !)p 2 IR and a(z !)p > 0g. This set is the union of p rays beginning in ! and forming angles whih are integer multiples of 2=p. Denote these rays by GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 7 L1 ; L2 ; : : : ; Lp . For 1 j p, 0 < r 1 and 0 < 2 let Sj (r; ) V be the set of those points z lying in the open ball B (!; r) for whih the angle between the rays Lj and the interval whih joins the points ! and z does not exeed . Using (2.7) an easy omputation leads to the following 8 > 0 9r1 () > 0 90 < 0 81 j p f! 1 (Sj (r1 (); 0 )) Sj (1; ) (2.8) and there are > 0 and 1 > 0 suh that jf! 1(z) !j < jz !j and j(f! 1)0(z)j < 1 (2.9) for every ! 6= z 2 S1 (1; ) [ : : : [ Sp(1 ; ). The following version of Fatou's ower theorem, (see [4℄, [17℄, omp. [1℄) shows that the Julia set J (f ) approahes the xed point ! tangentially to the lines L1 ; L2; : : : ; Lp. This an be preisely formulated as follows. Lemma 2.2. (Fatou's ower theorem) For every > 0 there exists r2 () > 0 suh that J (f ) \ B (!; r2 ()) S1 (r2 (); ) [ : : : [ Sp(r2 (); ): Sine the Julia set J (f ) is fully invariant (f 1(J (f )) = J (f ) and f (J (f )) = J (f ) [ f1g, we onlude from this lemma and (2.9) that for every 0 < 2 minf1 ; r2( )g we have f! 1 (J (f ) \ B (!; 2 )) J (f ) \ B (!; 2 ): Thus all iterates f! n : J (f ) \ B (!; 2) ! J (f ) \ B (!; 2), n = 0; 1; 2; : : : are well dened. From (2.8), Lemma 2.2, and (2.9) we obtain the following 8 > 0 9r3() > 0 81 j p (2.10) f! 1 (Sj (r3 (); ) \ J (f )) Sj (r3 (); ): Put = (f; ! ) = minf2 ; r2 ( ); r3 ( )g (2.11) Then, it follows from (2.9), (2.8), and Lemma 2.2 that for every z 2 J (f ) \ B (!; ). lim f n(z) = ! (2.12) n!1 ! In fat it an be proved that this onvergene is uniform on ompat subsets of B (!; ) \ J (f ) n f! g. See (2.13) for even stronger result. By preise omputations one an prove the following. Lemma 2.3. For every > 0 suÆiently small and every z 2 J (f ) \ B (!; ) f! 1 (B (z; jz ! j)) B (f! 1 (z ); jf! 1 (z ) ! j): This lemma immediately leads to the following. 8 JANINA KOTUS AND MARIUSZ URBANSKI Lemma 2.4. For every > 0 suÆiently small, every z 2 J (f ) \ B (!; ) and every n 0 there exists a unique holomorphi inverse branh f!;zn : B (z; 2 jz of f n whih sends z to f! n (z ). ! j) ! B (f! n (z ); 2 jf! n (z ) ! j) The following three results (omp. Lemma 1 and Lemma 2 of [8℄ and Lemma 4.8 of [9℄) an be proved in exatly the same way as in [8℄ and [9℄. lim jf n(z) !jn1=p = (jajp) 1=p and n p p j(f!;zn)0(z)j; j(f!;zn)(z)j n p p n!1 !;z (2.13) uniformly on ompat subsets of B (!; ) \ J (f ) n f!g. Lemma 2.5. Let m be a semi t-onformal measure for f . Then for every R > 0 there exists a onstant C = C (t; !; R) 1 suh that for every 0 < r R m(B (!; r) n f! g) m(Bs (!; r) n f! g) ; C: rt (!) rt (!) where t (! ) = t + p(! )(t 1). If m is t-onformal, then in addition m(B (!; r) n f! g) m(Bs (!; r) n f! g) ; C 1: rt (!) rt (!) 2.4. Basi properties of non-reurrent ellipti funtions. . In this setion the ellipti funtion f : CI ! CI is assumed to be non-reurrent. A periodi point ! of f is alled paraboli if there exits q 1 suh that f q (!) = ! and (f q )0 (1) = 1. The set of all paraboli points will be denoted by (f ). Sine the set of ritial values of f is nite, it follows from Fatou's theorem that (f ) is also nite. In addition, (f ) is ontained in the Julia set J (f ). The ruial tool for our approah in this paper similarly as in [21℄ is the following version of Mane's theorem proven in [13℄. +1 +1 Theorem 2.6. Let f : CI ! CI be a non-reurrent ellipti funtion. If X J (f ) n (f ) is a losed subset of C I , then for every > 0 there exists Æ > 0 suh that for every x 2 X and every n 0, all onneted omponents of f n (B (x; Æ )) have diameters . Corollary 2.7. Let f : CI ! CI be a non-reurrent ellipti funtion. If X J (f ) [f1gn (f ) is ompat, then for every > 0 there exists Æ > 0 suh that for every x 2 X and every n 0, all onneted omponents of f 1 (Bs (x; Æ )) have Eulidean diameters . 1 (1) and given > 0. This gives us the orProof. Apply Theorem 2.6 for the set f responding number Æ1 > 0. Taking now > 0 so small that eah onneted omponent of f n (Bs (1; )) is ontained in B (b; Æ1 ) for some pole b 2 f 1 (1) onsider the set Y = X n Bs (1; ). Sine Y is a ompat subset of CI, it follows from Theorem 2.6 that there exists Æ2 > GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 9 0 suh that for every x 2 Y and every n 0 all the onneted omponents of f n(Bs(x; Æ)) have Eulidean diameters . Consider a nite over fBs(x1 ; Æ2); : : : ; Bs(xk ; Æ2 ); Bs(1; )g of X , where xj 2 Y for all j = 1; 2; : : : ; k. Taking as Æ half of the Lebesgue number of this over nishes the proof. Beause of an extremal importane of this theorem and its orollary for our onsiderations, we provide in the Appendix the proof of Theorem 2.6 adapting to the ontext of ellipti funtions original Mane's proof from [13℄ and some lemma from [19℄. We put Crit(J (f )) = Crit(f ) \ J (f ); n o 1 = (f ) = min minf(f a ; ! ) : ! 2 (f )g; dist( (f ); Crit(f )) > 0 (2.14) 2 where a 1 is so large that all paraboli points of f a are simple and the numbers (f a ; !) are dened in (2.11). We also denote for every set A CI [ O+(A) = f n (A): and n0 A = A(f ) = maxfA(f; ) : 2 Crit(f )g (2.15) We all two points z and w equivalent and we write z w if w z 2 , the lattie assoiated with the ellipti funtion f . Obviously z w implies that O+(z) = O+(w) and ! (z ) = ! (w). Sine the set Crit(fS) \ R is nite, we onlude that the sets !(Crit(f )) = S2Crit(f )\R !() and O+(Crit(f )) = 2Crit(f )\R O+() are ompat subsets of CI. A positive number < =2 is now hosen to be less than the following three numbers. minfdist(; O+(f ()) : 2 Crit(f )g minf(A()R(f; ))1=q() : 2 Crit(f )g minfj 0j : ; 0 2 Crit(f ) and 6= 0 g; where q() = q(f; ) is the order of the ritial point of f . Notie that the rst of these numbers is positive sine O+(f (Crit(f )) is a ompat subset of CI and Crit(f ) has no aumulation points in CI. Sine f ontains no reurrent ritial points, it follows from nTheorem 2.6 that there exists 0 < < 1=4 suh that if n 0 is an integer, z 2 J (f ) and f (z) 2= B ( (f ); ), then diam Comp(z; f n(z); f n; 2 ) < : (2.16) From now on x also 0 < < 1 minf; 2 g so small as required in Lemma 2.4 for every ! 2 (f ) and so small that for every z 2 J (f ) diam Comp(z; f (z); f; ) < minf; 2 g: (2.17) JANINA KOTUS AND MARIUSZ URBANSKI 10 Lemma 2.8. If n 0 is an integer, > 0, z 2 J (f ) and for every k 2 f0; 1; : : : ; ng diam Comp(f k (z); f n(z); f n k ; ) ; then eah onneted omponent Comp(f k (z ); f n (z ); f n k ; ) ontains at most one ritial point of f and the equivalene lass of eah ritial point intersets at most one of these omponents. Proof. The rst part is obvious by the hoie of . In order to prove the seond part suppose that 1 2 Crit(f ) \ Comp(f k (z ); f n (z ); f n k ; ); 2 2 Comp(f k (z ); f n (z ); f n and 1 2 , where 0 k1 < k2 n. But then f k k (2 ) = f k k (1 ) 2 Comp(f k (z ); f n (z ); f n k ; ) and therefore jf k k (2 ) 1 j < , ontrary to the hoie of . 1 2 2 1 1 2 1 2 2 k2 ; ) 2 1 Lemma 2.9. The set ! (Crit(J (f ))) is nowhere dense in J (f ). Proof. Suppose that the interior (relative to J (f )) of ! (Crit(J (f ))) is nonempty. Then there exists 2 Crit(J (nf )) suh that !() has nonempty interior. But then there would exist n 0 suh that f (!()) = J (f ) and onsequently !() = J (f ). This however is a ontradition as 2= !(). Let = 2Crit(f )\R q() 1. We shall prove the following. Lemma 2.10. If z 2 J (f ), f n(z ) 2= B ( (f ); ), then Mod Comp(z; f n(z); f n; 2 ) n Comp(z; f n (z); f n; ) log2=#(Crit(f ) \ R) n n Proof. In view of Lemma 2.8 there exists a geometri annulus R B (f (z ); 2 )nB (f (z ); ) entered at f n(z) of modulus log 2=#Crit(f ) suh that f n(R) \ Comp(z; f n(z); f n; 2 ) \ n Crit(f )) = ;. Sine overing maps inrease moduli of annuli at most by fators equal to their degrees, we onlude that Mod Comp(z; f n(z); f n; 2 ) n Comp(z; f n(z); f n; ) Mod(Rn) log2=#(Crit(f ) \ R =2Crit(f )\R q() log2 ; = #(Crit( f ) \ R) where Rn Comp( z; f n (z ); f n ; 2 ) is the onneted omponent of f n(B (f n (z ); 2 )) enlosing Comp(z; f n(z); f n; ). As an immediate onsequene of this lemma and Koebe's Distortion Theorem, II (Eulidean version) we get the following. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 11 Lemma 2.11. Suppose that z 2 J (f ) and f n (z ) 2= B ( (f ); ). If 0 k n and f k : Comp(z; f n (z); f n; 2 ) ! Comp(f k (z); f n(z); f n k ; 2 ) is univalent, then j(f k )0(y)j onst j(f k )0(x)j for all x; y 2 Comp(z; f n (z ); f n ; ), where onst is a number depending only on #(Crit(f ) \R) and . For A, B , any two subsets of a metri spae put dist(A; B ) = inf fdist(a; b) : a 2 A; b 2 B g and Dist(A; B ) = supfdist(a; b) : a 2 A; b 2 B g: We shall prove the following. Lemma 2.12. Suppose that z 2 J (f ) and f n(z ) 2= B ( (f ); ). Suppose also that Q(1) n (2) Q(2) B (f n (z ); ) are onneted sets. If Q(2) n is a onneted omponent of f (Q ) ontained n (1) (2) in Comp(z; f n (z ); f n ; 0 ) and Q(1) n is a onneted omponent of f (Q ) ontained in Qn , then diamQ(1) diamQ(1) n : diam Q(2) diam Q(2) n Let 1 n1 : : : nu n be all the integers k between 1 and n suh that Crit(f ) \ Comp(f n k (z); f n(z); f k ; 2 ) 6= ;: Fix 1 i u. If j 2 [ni ; ni+1 1℄ (we set nu+1 = n 1), then by Lemma 2.10 there exists a universal onstant T > 0 suh that (1) diam Q diamQ(1) j n (2)i (2.18) (2) T diam Qj diam Qni Sine, in view of Lemma 2.8, u #(Crit(f ) \ R), in order to onlude the proof is therefore enough to show the existene of a universal onstant E> 0 suh that for every 1 i u 1. diamQ(1) diam Q(1) ni n (2)i : (2) E diam Qni diam Qni And indeed, let be the ritial point ontained in Comp(f n ni (z); f n(z); f ni ; 2 ) and let (1) q denote its order. Sine both sets Q(2) ni and Qni are onneted, we get for i = 1; 2 that diam Q(nii) 1 diam Q(nii) supfjf 0(x)j : x 2 Q(nii) g diam Q(nii) Dist(; Q(nii) ): Proof. +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 JANINA KOTUS AND MARIUSZ URBANSKI 12 Hene, using (2.18), we obtain (1) (1) diamQ(1) diam Qni diam Qni 1 Dist(; Q(2) ) ni n i (1) diam Q(2) diam Q(2) diam Q(2) ni ni 1 Dist(; Qni ) ni diamQ(1) ni : T diam Q(2) ni We are done. +1 +1 +1 +1 +1 +1 +1 1 +1 1 2.5. Partial order in Crit(J (f )) and stratiations of losed forward-invariant subsets of J (f ). Now we introdue in Crit(J (f )) a relation < whih, in view of Lemma 2.13 below, is an ordering relation, by putting 1 < 2 () 1 2 ! (2 ): (2.19) Sine 2 3 implies !(2) = !(3), if 1 < 2, then if 1 < 2 and 2 3, then 1 < 3 Lemma 2.13. If 1 < 2 and 2 < 3 , then 1 < 3 . Proof. Indeed, we have 1 2 ! (2 ) ! (3 ). Lemma 2.14. There is no innite, linear subset of the partially ordered set (Crit(J (f )); <) Proof. Indeed, suppose on the ontrary that 1 < 2 < : : : is an innite, linearly ordered subset of Crit(J (f )). Sine the number of equivaleny lasses of relation is equal to #(Crit(J (f )) \ R) whih is nite, there exist two numbers 1 i < j suh that !(i) = !(j ). But this implies that i 2 !(j ) = !(i) and we get a ontradition. The proof is nished. The following observation is a reformulation of the ondition that J (f ) ontains no reurrent ritial points. Lemma 2.15. If 2 Crit(J (f )), then ( < ). Dene now indutively a sequene fCri(f )g of subsets of Crit(J (f )) by setting Cr0(f ) = ; and 9 8 i = < [ Cri+1 (f ) = : 2 Crit(J (f )) n Crj (f ) : if 0 < ; then 0 2 Cr0 (f ) [ : : : [ Cri (f ); j =0 (2.20) Lemma 2.16. We have (a) If 2 Cri(f ) and 0 , then 0 2 Cri(f ). GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 13 (b) The sets fCri(f )g are mutually disjoint. () 9p1 8ip+1 Cri(f ) = ; (d) Cr0 (f ) [ : : : [ Crp(f ) = Crit(J (f )) (e) Cr1 (f ) 6= ; The item (a) follows immediately from the denition of the sets Cri and the fat that two equivalent points have the same !-limit sets. By denition Cri+1(f ) \ Sij=1 Crj (f ) = ;, so disjointness in (b) is lear. As the number of equivaleny lasses of the relation is equal to #(Crit(J (f )) \ R whih is nite, (a) and (b) imply Sp (). Take p to be minimal number satisfying (b) and suppose that 2 Crit(J (f )) n j=1 Crj (f ). Sine Crp+1(f ) = ;, there exists 0 2= Spj=1 Crj (f ) suh that 0 < . Iterating this proedure we would obtain an innite 0 sequene of ritial points 1 = > 2 = > 3 > : : : . But this ontradits Lemma 2.14 proving (d). Now part (e) follows from () and (2.20). As an immediate onsequene of the denition of the sequene fCri(f )g we get the following simple lemma. Proof. Lemma 2.17. If ; 0 2 Cri (f ), then ( < 0 ). For every point z 2 J (f ) dene the set Crit(z) = f 2 Crit(J (f )) : 2 !(z)g We shall prove the following. S Lemma 2.18. If z 2 J (f ) n I1 (f ), then either z 2 n0 f n( (f )) or ! (z ) n f1g is not ontained in O+ (f (Crit(z )) [ (f ). S Proof. Suppose that z 2= n0 f n( (f )) [ I1 (f ). Then by (2.12) the set ! (z ) n f1g is not ontained in (f ). So, if we suppose that ! (z ) n f1g O+ (f (Crit(z )) [ (f ); (2.21) then, as !(z) n f1g 6= ;, we onlude that Crit(z) 6= ;. Let 1 2 Crit(z). It means that 1 2 ! (z ) and as 1 2= (f ), it follows from (2.21) that there exists 2 2 Crit(z ) suh that either 1 2 !(2) or 11 = f n (2 ) for some n1 1. Iterating this proedure wen obtain an innite sequene fj gj=1 suh that for every j 1 either j 2 !(j+1) or j = f j (j+1) for some nj 1. Consider an arbitrary blok k ; k+1; : : : ; l suh that j = f nj (j+1) for every k j l 1 and suppose that l (k 1) #(Crit(f ) \ R). Then there are two indexes k a < b l suh that a b . Then f na+na +:::+nb (a ) = f na+na +:::+nb (b ) = a and onsequently, as na + na+1 + : : : + nb 1 b a 1, a is a super-attrating periodi point of f . Sine a 2 J (f ), this is a ontradition, and in onsequene the length of the 1 +1 1 +1 1 JANINA KOTUS AND MARIUSZ URBANSKI 14 blok k ; k+1; : : : ; l is bounded above by #(Crit(f ) \ R). Hene, there exists an innite subsequene fnk gk1 suh that nk 2 !(nk +1) for every k 1. But then nk 2 !(nk ) for every k 1, or equivalently nk < nk for every k 1. This however ontradits Lemma 2.14 and we are done. Dene now for every i = 0; 1; : : : ; p Si (f ) = Cr0 (f ) [ : : : [ Cri (f ) and for every i = 0; 1; : : : ; p 1 onsider 0 2 S2Cri (f ) !() \ Crit(J (f )). Then there exists 2 Cri+1 (f ) suh that 0 2 ! () whih equivalently means that 0 < . Thus, by (2.20) we get 0 2 Si(f ). So [ ! () \ (Crit(J (f )) n Si (f )) = ; (2.22) +1 +1 +1 Therefore, sine the set mulation point in CI, S 2Cri+1 (f ) 2Cri+1 (f ) ! Æi = dist Set () CI is ompat and Crit(J (f )) n Si (f ) has no au [ 2Cri+1 (f ) ! (); Crit(J (f )) n Si (f ) > 0 (2.23) = minfÆi =2 : i = 0; 1; : : : ; p 1g: Fix a losed forward-invariant subset F of J (f ) and for every i = 0; 1; : : : ; p dene Fi (f ) = fz 2 F : dist O+(z); Crit(J (f )) n Si(f ) g: Let us now prove the following two lemmas onerning the sets Fi(f ). Lemma 2.19. F0 F1 : : : Fp = F . Proof. Sine Si+1 (f ) Si (f ), the inlusions Fi Fi+1 is obvious. it holds Jp(f ) = J (f ). We are done. Let Sine Sp(f ) = Crit(J (f )), PC(f ) = O+(Crit(J (f ))) We shall prove the following. Lemma 2.20. There exists l = l(f ) suh that for every i = 0; 1; : : : ; p [ ! () O+ (f l (Cri+1 (f ))) PC(f )i 2Cri+1 (f ) 1 GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 15 The left-hand inlusion is obvious regardless whatever l(f ) is. In order to prove the right-hand one x i 2 f0; 1; : : : ; p 1g. By thedenition ofS!-limit sets there exists li 1 l suh that for every 2 Cri+1 (f ) we have dist O+ (f i ()); 2Cri (f ) ! () < Æi =2. Thus, by (2.23), dist O+(f li ()); Crit(J (f )) n Si(f ) > Æi =2. Sine Æi =2 and sine for every z 2 O+ (f li ()) also O+ (z ) O+ (f li ()), we therefore get O+ (f l (Cri+1 (f ))) PC(f )i . So, putting l(f ) = maxfli : i = 0; 1; : : : ; p 1g the proof is ompleted. Proof. +1 2.6. Holomorphi inverse branhes. In this setion we prove the existene of suitable holomorphi inverse branhes-our basi tools in the next setion. Set [ [ Sing (f ) = f n (f ) [ Crit(J (f )) [ f 1(1) and I (f ) = f n(1): n0 n1 We start with the following. Proposition 2.21. If z 2 J (f ) n Sing (f ), then there exist a positive number (z ), an inreasing sequene of positive integers fnj gj 1 , and a point x = x(z ) 2 ! (z ) n ( (f ) [ ! (Crit(z ))) suh that x 6= 1 if z 2= I1 (f ), limj !1 f nj (z ) = x and Comp(z; f nj (z); f nj ; (z)) \ Crit(f nj ) = ; for every j 0. I, Proof. Suppose rst that z 2 I1 (f ) n Sing (f ). Sine O+ (Crit(f )) is a ompat subset of C we onlude that for all n large enough dist(f n(z); O+(Crit(f ))) 1. We are therefore done taking x = 1 and (z) = 1. So, suppose that z 2= I1(f ). This means that !(z) n f1g 6= ;. Suppose that !(z) nf1g is unbounded. Sine O+(Crit(f )) is a ompat subset of CI, there thus exists1 x 2 !(z)nn f1g suh that dist(x; O+(Crit(f ))) 2 and we are done xing a sequene fnj gj=1 suh jf j (z) xj 1 and taking (z) = 1. So, assume that !(z) = F [f1g where F 1 CI is a ompat set. Then F \ f 1 (1 ) 6= ; and x x 2 F \ f (1). Again, sine O+ (Crit(f )) is a ompat subset of CI and sine f O+(Crit(f )) O+(Crit(f )), we see that x 2= O+(Crit(f )) and we are done taking (z) = dist(x; O+(Crit(f ))). So suppose nally that !(z) is a ompat subset of CI. In view of Lemma 2.18 there exists x 2 !(z) n ( (f ) [ O+(f (Crit(z)) [ f1g). The number = dist(x; (f ) [ O+(f (Crit(z)))=2 is positive sine !(Crit(z)) is a ompat subset of CI and (f ) is nite. Then there exists an innite inreasing sequene fmj gj1 suh that and lim f mj (z) = x (2.24) [ n f n1 (2.25) j !1 B (f mj (z ); ) \ (Crit(z)) = ;: 16 JANINA KOTUS AND MARIUSZ URBANSKI Now we laim that there exists (z) suh that for every j 1 large enough Comp(z; f mj (z); f mj ; (z)) \ Crit(f mj ) = ;: (2.26) Otherwise we would nd an inreasing to innity subsequene fmji g of fmj g and a dereasing to zero sequene of positive numbers i suh that i < and Comp(z; f mji (z); f mji ; i) \ Crit(f mji ) 6= ; Let ~i 2 Comp(z; f mji (z ); f mji ; i ) \ Crit(f mji ). Then there exists i 2 Crit(f ) suh that p f i (~i ) = i for some 0 pi mji 1. Sine the set f 1 (x) is at a positive distane from (nf ) and sine i ! 0, it follows from Theorem 2.6 that limi!1 ~i = z. Sine z 2= S f (Crit(p f )), it implies that limi!1 pi = 1. But then using Lemma 2.6 again and n0 the formula f i (~i ) = i we onlude that the set of all aumulation points of the sequene fig is ontained in !(z). Hene, passing to a subsequene, we may assume that the limit = limi!1 i exists. But sine 2 ! (z ), sine ! (z ) is a ompat subset of CI and sine 1 is the only aumulation point of Crit(f ), we onlude that the sequene i is eventually onstant. Thus, dropping some nite number of initial terms, we may assume that this sequene is onstant. This means that i = for all i = 1; 2; : : : . Sine = f pi (~i), we get jf mji (z) f mji pi ()j = jf mji (z) f mji (~i)j < i: Sine limi!1 i = 0 and sine !(z) is a ompat subset of CI, we onlude that limi!1 jf mji (z) m p f ji i ()j = 0. Sine 2 Crit(z ), in view of (2.25) this implies that mji pi 0 for all i large enough. So, we get a ontradition as 0 pi mji 1 and (2.26) is proved. We are done. Sine if z 2 J (f ) n (Sing (f ) [ I1(f )), the limit points of the normal family fz nj : B (x(z ); (z )=2) ! CI onsist only of onstant funtions. Therefore we get the following. Corollary 2.22. If z 2 J (f ) n (Sing (f ) [ I1 (f )) and the sequene fnj g1 j =1 is taken from Proposition 2.21, then nj 0 lim sup j(f n)(z)j = limn!1 sup j(f n)0 (z)j = nlim !1 j(f ) (z )j = +1: n!1 In addition, if we assume only that z 2 J (f ) n Sing (f ), then limn!1 sup j(f n)0(z)j = 1: 3. Conformal Measures In this setion we deal in detail with the existene, uniqueness tand some properties of onformal measures. Let HD denote the Hausdor dimension, H and l2 denote respetively GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 17 t-dimensional Hausdor measure and 2-dimensional Lebesgue measure, both onsidered with respet to the spherial metri on CI. Throughout this setion and the entire paper we set h = HD(J (f )): We begin with the following. Lemma 3.1. If m is a t-onformal measure for f : J (f ) ! J (f ) [ f1g, then t HD(J (f )) and Ht jJ (f ) is absolutely ontinuous with respet to m. Proof. Fix z 2 J (f ) n (Sing (f ) [ I1 (f )). Let (z ) > 0, x 2 ! (z ) n f1g and the sequene fnj gj1 be taken from Proposition 2.21. It then follows from this proposition and Koebe's Distortion Theorem, I(spherial version) that fz nj (B (f nk (z ); (z )=2)) B (z; j(f nj ) (z )j 1 (z )=2): Applying again this Koebe's Distortion Theorem and onformality of the measure m, we get for all j 1 large enough m(B (z; j(f nj )0 (z )j 1 (z )=2)) j(f nj ) (z )j t m(B (f nj (z ); (z )=2)) j(f nj ) (z)j t m(B (x; (z)=4)) j(f nj )0 (z)j tm(B (x; (z)=4)) = (2(z) 1)t m(B (x; (z)=4)) j(f nj ) (z)j 1(z)=2) t; where the seond omparability sign depends on jzj and holds for all j 1 large enough so that f nj (z) is suÆiently lose to x. In partiular m(B (z; r)) limr!sup R(z) > 0; rt 0 where R(z) = (2(z) 1 )tm(B (x; (z)=4)). Therefore, putting Xk = fz 2 J (f ) n Sing (f ) : jz j k and R(z ) 1=kg S1 we have k1 Xk = J (f ) n (Sing (f ) [ I1(f )) and in view of Theorem 4.3(1) (whih is of purely geometri harater independent of our onsiderationshere), dHt =dm b(2)k on Xk . In partiular Ht m on J (f ) n (Sing (f ) [ I1(f )). Hene HD J (f ) n (Sing (f ) [ I1(f )) t. By Theorem 1 and Theorem 2 in [12℄), HD(J (f )) > HD(I1(f )). Thus HD(J (f )) = HD J (f )n (Sing (f ) [ I1(f )) t and Ht m on J (f ). We will need in the sequel the following result whih is interesting itself. Lemma 3.2. If m is a t-onformal measure for f : CI ! CI, then m(I1 (f ) n I (f )) = 0. Even more, there exists R > 0 suh that m(fz : lim inf jf n(z)j > Rg) = 0: n!1 JANINA KOTUS AND MARIUSZ URBANSKI 18 Let b be a pole of f : CI ! CI. We shall obtain rst an upper estimate on m(Bb(R)) similar to the seond inequality in (2.4). And indeed, overing BR n f1g by two simply onneted domains BR+ = fz 2 BR n f1g : Imz > 0g and BR1 = fz 2 BR n f1g : Imz < 1g we obtain qb Z qb Z X X 1 t m(Bb (R) n fbg) j(fb;BR ;j ) j dm + B j(fb;B1R ;j )jtdm: B Proof. Using now (2.3), we obtain + + j =1 R 1 j =1 R !t qb 1 1 Z jzj qbqb tdm(z) q b j z j dm ( z ) = j ( fb;B ;j R (1 + jbj2)t BR BR BR 1 + jbj2 Z q (1 + jbj2) t B jzj q tdm(z): R Looking Rat the qrst line of this formula with a pole b of maximal multipliity, we see that the integral BR jzj q t dm(z) is nite and even more: Z 1 + ) jtdm + Z 1 1 + + 1 + 1 + Z lim R!1 Similarly is nite the integral RBR jzj q q 1 R = max (Z BR+ (3.1) 1 BR+ 1 jzj q q t dm(z) = 0: t dm(z ) and it also onverges to 0 as R ! 1. jzj q (z); q t dm 1 Z BR1 jzj q (z) q t dm 1 ) Putting we therefore onlude that m(Bb (R) n fbg) 2q R (1 + jbj2 ) t 2q R jbj 2t : (3.2) Now the argument goes essentially in the same way as in [12℄. We present it here for the sake of ompleteness. We take R2 R1 dened in Setion 2.1 so large that (3.3) LR qb < R0 for all poles b 2 BR and all R R2 . Given two poles b1 ; b2 2 B2R we denote by fb ;b1 ;j : 1 B (b1 ; R0 ) ! CI all the holomorphi inverse branhes fb ;B (b ;R );j . It follows from (2.4) and (3.3) that fb ;b1 ;j B (b1 ; R0 ) Bb (2R2 R0 ) Bb (R2 ) B (b2 ; R0 ) (3.4) Set IR (f ) = fz 2 CI : 8n0 jf n(z )j > Rg: 1 2 2 2 2 1 2 1 2 0 2 1 GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 19 Sine the series Pb2f (1)nf0g jbj s onverges for all s > 2 and sine by Lemma 3.1 and Theorem 3 from [12℄, t h > q2+1q there exists R3 R2 suh that X q qM t (3.5) jbj q t 1=2: 1 +1 b2BR3 \f Consider R 4R3. Put (1) 1 I = f 1 (1) \ B(R=2) Sine R=2 + R0 R=2 + R3 < R=2 + R=2 = R, it follows from (3.4), (2.4) and (3.3) that for every l 1 the family Wl dened as n o 1 1 1 1 ;jl Æ fbl ;bl ;jl Æ : : : Æ fb ;b ;j Æ fb ;b ;j Bb (R=2) n f (1) ; where bi 2 I : 1 ji qbi ; i = 0; 1; : : : ; l, is well-dened and overs IR (f ). Applying (2.3) fbl ;b1 l 1 1 2 2 1 1 and (2.4) we may now estimate as follows. m(IR (f )) qbl qb X X X XX ::: m fbl ;b1 l ;jl Æ fbl 1 ;bl 2 1 XX ::: bl 2I jl =1 1 b1 2I j1 =1 b0 2I qb1 X X X b1 2I j1 =1 b0 2I jj fbl;b1l 1 1 Æ : : : Æ fb ;b1 ;j Æ fb ;b1 ;j Bb (R=2) 1 bl 2I jl =1 qbl 0 0 1 2 ;jl 2 1 1 2 1 0 0 1 1 ;jl Æ fbl ;bl 1 2 ;jl 1 Æ : : : Æ fb ;b1 ;j Æ fb ;b1 ;j jBb (R=2) jjt1m Bb (R=2) 2 1 2 1 0 1 qb1 X X X qbl t M lt X X = (2qR) (2qR )tM lt bl 2I jl =1 qbl XX bl 2I jl =1 0 t lt X jbj R M (2q ) b2I 0 t t R qM (2q ) ::: ::: qb1 X X X b1 2I j1 =1 b0 2I qb1 X X X b1 2I j1 =1 b0 2I 1l q+1 t q A ql X b2BR3 \f jbl j 1 (1) jbj jbl j 2t jbl 1j q+1 t q bl j q+1 t q ::: 1j jb0 j q+1 t q ::: 0 0 0 0 qbl 1 1 1t qbl 1 1t 0 qb1 1 1t q q b XX B B B j bl 1 j bl C j bl 2 j l 1 C j b0 j qb1 C lt B B B C C C ::: M 2 A jbl 1 j2 A : : : jb1 j2 A j b j l bl 2I jl =1 b1 2I j1 =1 b0 2I qbl (2qR )t jb 1j2t 0 q+1 t q jb0j q+1 t q 1l q+1 t q+ A Applying (3.5) we therefore1 get m(IR (f )) (2qR)t 2 l . Letting l ! 1 we get S therefore j (I (f )), m(IR (f )) = 0. Sine m Æ f m and sine fz : lim infn!1 jf n(z )j > Rg = 1 f R j =0 we onlude that m fz : lim infn!1 jf n(z)j > Rg = 0. We are done. JANINA KOTUS AND MARIUSZ URBANSKI 20 Developing the general sheme from [7℄ we shall now prove in several steps the existene of an h-onformal measure. In order to begin we all Y f1g [ (f ) [ Sn1 f n(Crit(J (f ))) a rossing set if Y is nite and the following two onditions are satised. (y1) 1 2 Y n. (y2) Y \ ff (x) : n 1g is a singleton for all x 2 Crit(J (f )). (y3) Y \ Crit(f ) = ;. (y4) (f ) Y . Sine f (Crit(f )) is nite, rossing sets do exist. Let V CI be an open neighbourhood of Y . We dene K (V ) = fz 2 J (f ) : f n (z ) 2= V 8(n 0)g: Obviously f (K (V )) K (V ) and sine f : CI ! CI is ontinuous and V is open, we see that K (V ) is a losed subset of CI. Sine in addition K (V ) CI n V , we onlude that K (V ) is a ompat subset of CI. Fix w 2 K (V ) and t 0. For all n 1 onsider the sets n En = f jK (V ) (w) and the number 1 X j(f n)(x)j t : (f ) = lim sup log n!1 n x2En Sine the ontinuous map f jK (V ) : K (V ) ! K (V ) has no ritial points, all the sets K (V ) are (n; Æ)-separated, where 1 Æ = inf fminfjz xj : x; z 2 f jK (V ) (y ) and x 6= z gg > 0: y2K (V ) Therefore (f ) P f jK (V ) ; t log jf j ; (3.6) where the right-hand side of this inequality is the topologial pressure of the potential t log jf j with respet to the dynamial system f jK (V ) : K (V ) ! K (V ). Denote this pressure simply by P(f; V ). We all a Borel set A CI speial if jA is injetive. Lemma 3.1 and 3.2 from [7℄ (omp. [6℄) enlarged by the reasoning started from the seond paragraph of the proof of Lemma 5.3 in [7℄ an be now formulated together as follows. Lemma 3.3. For every t 0 there exists a Borel probability measure mV;t supported on K (V ) suh that (a) mV;t (f (A)) RRA e(f )jf jt dmV;t for every speial set A CI and (b) mV;t (f (A)) = A e(f ) jf jtdmV;t for every speial set A CI n V . We will need the following tehnial lemma. Lemma 3.4. The funtion t 7! (f ) is ontinuous, (0) > 0 and 1 (0) \ (0; h℄ 6= ; if V a suÆiently small diameter. has GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 21 Continuity of the funtion (f ) follows from the fat that 0 < inf K (V )fjf jg supK (V ) fjf jg < 1. Sine periodi points of f are dense in J (f ), K (V ) 6= ; for all V suÆiently small. Also if V is suÆiently small and w 2 K (V ), then #En 2n and onsequently (0) log 2 > 0. Sine (0) > 0 and sine the funtion (f ) is ontinuous, in order to prove the last laim of our lemma, it suÆes to show that (f ) 0 for all t h. So, suppose on the ontrary that (f ) > 0 for some t h. It follows from (3.6) that P(f; V ) > 0: (3.7) Sine the proof of Lemma 4.1 and Corollary 4.2R from [7℄ go word by word in our ontext, we onlude that the Lyapunov exponent = log jf jd 0 for every Borel probability f -invariant measure supported on K (V ). It follows from (3.7) and the variational priniple for topologial pressure that there exists a Borel probability f -invariant measure supported on K (V ) suh that h(f ) t > 0. Sine 0, this implies that h (f ) > 0 and due to Ruelle's inequality > 0. Hene, applying Przytyki's-Manne volume lemma (see [18℄, omp. [14℄), we an write h (f ) t < = HD() h and this ontradition nishes the proof. Let s(V ) = minf 1 (0) \ (0; h℄g > 0: Combining Lemma 3.3 and Lemma 3.4 we get the following. Proof. Lemma 3.5. There exists a Borel probability measure mV supported on K (V ) suh that (a) mV (f (A)) RRA jf js(V ) dmV for every speial set A CI and (b) mV (f (A)) = A jf js(V )dmV for every speial set A CI n V . Sine the sequene n 7! s(Bs(Y; 1=n)) is monotonially non-dereasing, proeeding similarly as in the proof of Lemma 5.4 from [7℄ (note that in the plae where Lemma 3.3 from [7℄ is invoked, only the rst inequality in (d) is needed; in partiular mY (1) = 0, where mY is an arbitrary weak aumulation point of the sequene mBs (Y;1=n) we obtained the following. Lemma 3.6. For every s(Y ), an aumulation point of the sequene s(Bs (Y; 1=n)), s(Y ) 2 (0; h℄ and there exists a Borel probability measure mY (an appropriate week aumulation point of the sequene fmBs (Y;1=n) gn1 ) supported on J (f ) suh that (a) mY (f (A)) RRA jf js(Y )dmY for every speial set A CI and (b) mY (f (A)) = A jf jS(Y ) dmY for every speial set A CI n Y . The next fat proven in this setion is provided by the following. JANINA KOTUS AND MARIUSZ URBANSKI 22 Lemma 3.7. For every rossing set Y , m = mY is an s(Y )-onformal Smeasure for f : J (f ) ! J (f ) [ f1g, s(Y ) = h, and all atoms of m are ontained in I (f ) [ n0 f n (Crit(J (f )). Proof. Sine we already know that m(1) = 0 and sine Y \ (Sing (f ) [ I1 (f )) (f ) [ f1g, it follows from Lemma 3.6(b) and Corollary 2.22 that m(Y n (f )) = 0: (3.8) We shall show now that m( (f )) = 0. And indeed, x ! 2 (f ). Take a 1 so large that f a(!) = ! and (f a)0(!) = 1. It then follows from (2.13) that there exist a ompat set F! B (!; ) n f! g and a onstant C 1 suh that for every k 1 p! p! (3.9) C 1 k p ! j(f! ak ) (z )j Ck p ! and for every n 1 there exists kn 1 suh that 1 [ f! aj (F! ) and nlim (3.10) B (!; 1=n) !1 kn = 1: ( )+1 ( ) ( )+1 ( ) j =kn It follows from Lemma 3.6(b), (3.9) and the fat that the family ff! an(F! )gn1 is of bounded multipliity, that X p ! s(Y ) n p! < 1: ( )+1 ( ) n1 (Y ) > 1. Denote mjBs (Y;1=n) by mn and s(Bs(Y; 1=n)) by sn. Sine In partiular limn!1 sn = s(Y ), we see that for every n 1 large enough, say n n0 , p(! ) + 1 s > 1 + : p(! ) n for some > 0. It therefore follows from Lemma 3.6(a), (3.10) and (3.9) that for all n n0 and all l 1 1 1 p! p! X X mn f! aj (F! ) C p ! sn mn (B (!; 1=l)) j p ! sn p(!)+1 p(!) s ( )+1 ( ) j =kl Consequently p(!)+1 C p(!) s(Y ) m(B (!; 1=l)) Sine liml!1 kl = 1, we infer 1 X j =kl ( )+1 ( ) j =kl j (1+) : p(!)+1 C p(!) s(Y ) 1 X j =kl j (1+) : m( (f )) = 0: Combining this and (3.8), we see that m(Y ) = 0. Sine f ( (f )) = (f ), in order to prove s(Y )-onformality of the measure m, it therefore suÆes to show that m(f (Y n (f ))) = GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 23 0. But if y 2 Y n ( (f ) [ f1g), then due to our denition of Y , y 2= Sing (fn ) and the formula m(f (y)) = 0 immediately follows from Corollary 2.22, the formula m(f (f (y))) j(f n)(y)js(Y )m(f (y)) and the stated in Lemma 3.6 fat that s(Y ) > 0. Thus the s(Y )onformality of m is proven and in addition all the atoms of m must be ontained in J (f ) n . In view of Lemma 3.6 and Lemma 3.1, s(Y ) = h. Applying Snow Lemma 3.2 and Corollary 2.22 n we see that all atoms of m must be ontained in I (f ) [ n0 f (Crit(J (f )). The proof is omplete. 4. Hausdorff and Paking Measures Let h denote the paking measure onsidered with respet to the spherial metri on CI. We shall prove in this setion that the onformal measure m is atomless and the following main result. Theorem 4.1. Let f : CI ! CI be a non-reurrent ellipti funtion. If h = HD(J (f )) = 2, then J (f ) = C I . So suppose that h < 2. Then (a) Hhh(J (f )) = 0. (b) h(J (f )) > 0. () (J (f )) = 1 if and only if (f ) 6= ;. As an immediate onsequene of this theorem we get the following. Corollary 4.2. If (f ) = ;, then the Eulidean h-dimensional paking measure he is nite on eah bounded subset of J (f ). 4.1. Preliminaries from Geometri Measure Theory. In this setion we ollet some fats from the geometri measure theory as well as we list without proofs some more tehnial fats taken from Setion 2, Setion 3 and Setion 4 of [21℄. Given a subset A of a metri spae (X; d), a ountable family fB (xi ; ri)g1i=1 of open balls entered at the set A is said to be a paking of A if and only if for any pair i 6= j d(xi ; xj ) > ri + rj : Given t 0, the t-dimensional outer Hausdor measure Ht (A) of the set A is dened as 1 o nX rit Ht (A) = sup inf >0 where inmum is taken over all overs fB (xi ; ri)g1 with radii whih do not exeed . i=1 i=1 of the set A by open balls entered at A JANINA KOTUS AND MARIUSZ URBANSKI 24 The t-dimensional outer paking measure t (A) of the set A is dened as nX o t (A) = [Ainfi=A t(Ai ) i (Ai are arbitrary subsets of A), where 1 o nX rit : t (A) = sup sup >0 i=1 Here the seond supremum is taken over all pakings fB (xi; ri)g1i=1 of the set A by open balls entered at A with radii whih do not exeed . These two outer measures dene ountable additive measures on Borel -algebra of X . The denition of the Hausdor dimension HD(A) of A is the following HD(A) = inf ft : Ht (A) = 0g = supft : Ht (A) = 1g: Let be a Borel probability measure on X whih is positive on open sets. Dene the funtion = t ( ) : X (0; 1) ! (0; 1) by (B (x; r)) (x; r) = rt The following two theorems (see [DU5℄) are for our aims the key fats from geometri measure theory. Their proofs are an easy onsequene of Besiovi overing theorem (see [G℄). Theorem 4.3. Let X = IRd for some d 1. Then there exists a onstant b(n) depending only on n with the following properties. If A is a Borel subset of IRd and C > 0 is a positive onstant suh that (1) for all (but ountably many) x 2 A limr!sup (x; r) C 1 ; 0 then for every Borel subset E A we have Ht (E ) b(n)C (E ) and, in partiular, Ht (A) < 1. or (2) for all x 2 A limr!sup (x; r) C 1 ; 0 then for every Borel subset E A we have Ht (E ) C (E ). Theorem 4.4. Let X = IRd for some d 1. Then there exists a onstant b(n) depending only on n with the following properties. If A is a Borel subset of IRd and C > 0 is a positive onstant suh that GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 25 (1) for all x 2 A limr!inf (x; r) C 1 ; 0 then for every Borel subset E A we have t (E ) Cb(n) 1 (E ), or (2) for all x 2 A limr!inf (x; r) C 1 ; 0 then t (E ) C (E ) and, onsequently, t (A) < 1. (1') If is non{atomi then (1) holds under the weaker assumption that the hypothesis of part (1) is satised on the omplement of a ountable set. Assume now that is a Borel measure on CI nite on bounded sets. These two theorems motivated us in [21℄ to introdue the following notions. Denition 4.5. Given r > 0 and L > 0 a point x 2 CI is said to be (r; L) t:upper estimable if (x; r) L and is said to be (r; L) t:lower estimable if (x; r) L. We will frequently abbreviate the notation writing (r; L)-u.e. for (r; L) t:-upper estimable and (r; L)-l.e. for (r; L) t:-lower estimable. We also say that the point x is t-upper estimable (t-lower estimable) if it is (r; L) t:upper estimable ((r; L) t:lower estimable) for some L > 0 and all r > 0 suÆiently small. We will also need the following more tehnial notion. Denition 4.6. Given r > 0, > 0 and L > 0 the point x 2 X is said to be (r; ; L) t:strongly lower estimable, or shorter (r; ; L)-s.l.e. if (B (y; r)) Lrt for every y 2 B (x; r). We ollet now from [21℄ the tehnial fats about the notions dened above. Lemma 4.7. If z is (r; ; L)-s.l.e., then every point x 2 B (z; r=2) is (r=2; 2; 2tL)-s.l.e.. Lemma 4.8. If x is (r; ; L)-s.l.e., then for every 0 < u 1 it is (ur; =u; Lu t)-s.l.e.. Lemma 4.9. If is positive on nonempty open sets, then for every r > 0 there exists E (r) 1 suh that every point x 2 X is (r; E (r))-u.e. and (r; E (r) 1 )-l.e.. 26 JANINA KOTUS AND MARIUSZ URBANSKI Passing to onformal maps we onsider now the situation where H : U1 ! U2 is an analyti map of open subsets U1 , U2 of the omplex plane CI. We say that given t 0, the Borel measure nite on bounded sets of CI is a Eulidean semi t-onformal if and only if Z (H (A)) jH 0jt d A for every Borel subset A of U1 suh that H jA is one-to-one and is all t-onformal if the \" sign an be replaed by an \=" sign. Lemma 4.10. Let be a Eulidean semi t-onformal measure. Suppose that D CI is an open set, z 2 D and H : D ! C I is an analyti map whih has an analyti inverse Hz 1 dened on B (H (z ); 2R) for some R > 0. Then for every 0 r R K t (B (z; K 1 rjH 0(z )j 1 )) jH 0 (z )j t ((B (H (z ); r))): If, in addition, is t-onformal, then also jH 0(z)j t ((B (H (z); r))) K t (B (z; KrjH 0(z)j 1 )): Lemma 4.11. Suppose that is a Eulidean t-onformal measure. If the point H (z ) is (r; ; L)-s.l.e., where r R=2 and 1, then the point z is (K 1jH 0(z)j 1 r; K 2; L)-s.l.e.. Lemma 4.12. Suppose that is a Eulidean t-onformal measure. Let be a ritial point of an analyti map H : D ! C I . If 0 < r R(H; ) and H () is (r; L)-l.e., then is ((Ar)1=q ; A 2tL)-l.e.. Lemma 4.13. Let be a ritial point of an analyti map H : D ! CI. Let be a Eulidean semi t-onformal measure suh that () = 0. If 0 < r R(H; ) and H () is (s; L)-u.e. for 1 1 =q 2 t t=q 1 all 0 < s r, then is (A r) ; q (2A ) (2 1) L -u.e.. Note that the proof of this lemma is the same as the proof of Lemma 3.4 in [21℄. The only modiation is that the equality sign in the rst line of the rst displayed formula of this proof is to be replaed by the \" sign. Lemma 4.14. Suppose that is a Eulidean t-onformal measure. Let be a ritial point I . If 0 < r 31 R(H; ), 0 < 1 and H () is (r; ; L)-s.l.e, of an analyti map H : D ! C ~ )-s.l.e, where ~ = (2q+1KA2 )1=q and L~ = L minfK t; (A2) q q t g. then is ((A 1 r)1=q ; ~ ; L 1 GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 27 Notie now that if m 2ist a semi t-onformal measure for f : J (f ) ! J (f ) [ f1g, then the measure me = (1 + jzj ) m is Eulidean semi t-onformal, i.e. Z me (f (A)) jf 0 jt dme A for every Borel set A J (f ) suh that f jA is 1-to-1. If m is t-onformal, then so is me in the obvious sense. The measure me is alled the Eulidean version of m. Obviously me is equivalent to m and is nite on bounded subsets of CI. From now on throughout the entire paper we x a rossing set Y and we onsider an open neighbourhood V CI of Y suh that Crit(f ) \ V = ; and the losure of V is disjoint from at least one fundamental parallelogram of f . A semi t-onformal measure m is said to be almost t-onformal if Z m(f (A)) = jf 0jt dm A for every Borel set A J (f ) suh that f jA is 1-to-1 and A \ V = ;. Hene for every Borel set A suh that f jA is 1-to-1 and A \ V = ; and for every w 2 , we have Z Z 0 t jf j dme = me(f (A)) = me(f (A + w)) A+w jf 0jtdme A and the last inequality sign beomes an equality either if in addition (A + 0w) \ V = ; or if m is a t-onformal measure and we assume only that f jA is 1-to-1. Sine f is periodi with respet to the lattie , all the above statements and assumptions lead to the following. Lemma 4.15. For every w 2 , every Borel set A CI suh that A \ V = ; and every almost t-onformal measure m me (A + w) me (A): If either in addition (A + w) \ V = ; or if m is h-onformal and we assume only that f jA is 1-to-1, then this inequality beomes an equality. For every r > 0 there exists M (r) 2 (0; 1) independent of any almost t-onformal measure m suh that me (F ) M (r): (4.1) for every Borel set F C I with the diameter r. If in addition m is h-onformal, then for every R > 0 there exist onstants Q(R) and Qh (R) suh that me (Be (x; r)) Q(R)r2 Qh (R)rh (4.2) for all x 2 J (f ) and all r R. The following lemma is proven in the same way as the orresponding lemma from Setion 4 of [21℄. 28 JANINA KOTUS AND MARIUSZ URBANSKI Lemma 4.16. Suppose that me is a Eulidean t-onformal measure. Then for every R > 0 and every 0 < 1 there exists L = L(!; R; ) > 0 suh that for every 0 < r R every point ! 2 (f ) is (r; ; L)-t (! ):s.l.e. with respet to the measure me . 4.2. Conformal Measure and Holomorphi Inverse Branhes. In this subsetion we prove two tehnial propositions modeled on Proposition 6.3 and Proposition 6.4 from [21℄. Let m be an almost t-onformal measure and let me be its Eulidean version. The upper estimability and strongly lower estimability will be onsidered in this setion with respet to the measure me. When we speak about lower estimability we assume more, that the measure m is t-onformal. Sine the number of paraboli points is nite, passing to an appropriate iteration, we assume in this and the next setion without loosing generality that all paraboli points of f are simple. Fix a forward f -invariant ompat subset F of CI. Put jjf 0jjF = supfjf 0(z)j : z 2 F g: Reall that was dened in (2.11) and that > 0 is so small as required in Lemma 2.3. Proposition 4.17. Fix a forward f -invariant ompat subset F of CI. Let z 2 F , > 0 and let 0 < r jjf 0 jjF 1 1 be a real number. Suppose that at least one of the following two onditions is satised: z2Fn or [ n0 f n (Crit(J (f )) z 2 F and r > jjf 0 jjF 1 1 inf fj(f n)0 (z )j 1 : n = 1; 2; : : : g: Then there exists an integer u = u(; r; z ) 0 suh that rj(f u)0 (z )j 1 and the following four onditions are satised diam Comp(f j (z); f u(z); f u j ; rj(f u)0(z)j) (4.3) for every j = 0; 1; : : : ; u. For every > 0 there exists a ontinuous funtion t 7! Bt = Bt (; ) > 0, t 2 [0; 1), (independent of z , n, and r) and suh that if f u (z ) 2 B (!; ) for some ! 2 (f ), then f u (z ) is (rj(f u)0 (z )j; Bt ) t (! ):u.e. (4.4) and there exists a funtion Wt = Wt (; ) : (0; 1℄ ! (0; 1℄ (independent of z , n, and r) suh that if f u (z ) 2 B (!; ) for some ! 2 (f ), then for every 2 (0; 1℄ f u(z ) is (rj(f u)0 (z )j; ; Wt ( )) t (! ):s.l.e. (4.5) If f u (z ) 2 = B ( (f ); ), then formulas (4.4) and (4.5) are also true with t (! ) replaed by t: (4.6) GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 29 Suppose rst that supfrj(f j )0(z)j : j 1g > jjf 0jjF 1 and let n = n(; z; r) 0 be a minimal integer suh that rj(f n)0 (z )j > jjf 0 jjF 1: (4.7) Then n 1 (due to the assumption imposed on r) and also rj(f n)0 (z )j (4.8) If f n(z) 2= B ( (f ); ) set u = u(; r; z) = n. The items (4.4), (4.5) and (4.6) are obvious in view of our assumptions imposed on F . So suppose that f n(z) 2 B ( (f ); ), jsay f n(z) 2 B (!; ), ! 2 (f ). Let 0 k = k(; z; r) n be the smallest integer suh that f (z ) 2 B ( (f ); ) for every j = k; k +1; : : : ; n. Consider all the numbers ri = jf i (z ) ! jj(f i)0 (z )j 1 where i = k; k + 1; : : : ; n. By (4.7) we have rn = jf n(z ) ! jj(f n)0 (z )j 1 jjf 0 jjF 1 1 r = jjf 0jjF 1 r and therefore there exists a minimal k u = u(; r; z) n suh that ru jjf 0jjF 1 r. In other words jf u(z) !j jjf 0jjF 1 rj(f u)0(z)j jjf 0jjF 1 1rj(f u)0 (z)j (4.9) IfS supfr j(f j )0(z)j : j 1g jjf 0jjF 1, then it follows from Corollary 2.22 that z 2 j j 0 f ( (f )). Dene then u(; z; r ) = k (; z; r ) to be the minimal integer j 0 suh j that f (z) 2 (f ) and put ! = f u(z). Notie that in this ase formulas (4.8) and (4.9) are also satised. Our further onsiderations are valid in both ases. First note that by (4.9) we have B (f u(z ); rj(f u)0 (z )j) B (!; (1 + jjf 0jjF 1 1 )rj(f u)0 (z )j) (4.10) and in view of Lemma 2.5 and (4.8) me B (f u (z );rj(f u)0 (z )j) C (!; (1 + jjf 0jjF 1 1) 1 )(1 + jjf 0jjF 1 1)t(!) (rj(f u)0(z)j)t(!) So, item (4.4) is proved. Also applying (4.9), Lemma 4.16, Lemma 4.7 and (4.8) we see that the point f u(z) is jjf 0jjF 1 rj(f u)0(z)j; jjf 0jjF 1 1; 2t(!) L(!; 2jjf 0jjF ; (2jjf 0jjF ) 1 1) -s.l.e. So, if jjf 0jjF 1 , then by Lemma 4.8, f u(z) is rj(f u)0 (z )j; ; (2jjf 0jjF 1 1 )t (!) L(!; 2jjf 0jjF ; (2jjf 0 jjF ) 1 ) 1 -s.l.e If instead jjf 0jjF 1 , then again it follows from (4.9), Lemma 4.16, Lemma 4.7 and (4.8) u u 0 ( ! ) 1 t that the point f (z) is rj(f ) (z)j; ; 2 L(!; 2 ; =2) -s.l.e.. So, part (4.5) is also proved. Proof. JANINA KOTUS AND MARIUSZ URBANSKI 30 In S orderj to prove (4.3) suppose rst that u = k. In partiular this is the ase if z 2 j 0 f ( (f )). Then Comp(f k 1(z); f k (z); f; rj(f u)0 (z)j) Comp(f k 1(z); f k (z); f; ) and by the hoie of k and (2.9) we have f k 1(z) 2= B ( (f ); ). Therefore (4.3) follows from the hoie of (see (2.17)) and (2.16). If u > k (so the rst ase holds), then ru 1 > jjf 0jjF 1r and by (2.16) we get jf u(z) !j jf 0(f u 1(z))j 1 r kf k 1r 1r: ru = u 1 u 1 u 1 jf (z) !j So, rj(f u)0(z)j jf u(z) !j and applying Lemma 2.4 and (2.9) u k times we onlude that for every k j u diam Comp(f j (z); f u(z); f u j ; rj(f u)0(z)j) < And now for j = k 1; k 2; : : : ; 1; 0, the same argument applies as in the ase u = k. Proposition 4.18. Fix a forward f -invariant ompat subset F of CI. Let and be both positive numbers suh that < minf1; 1 ; 1 1 g. If 0 < r < jjf 0 jjF 1 1 and z 2 F n Crit(J (f )), then there exists an integer s = s(; ; r; z ) 1 with the following three properties. j(f s)0(z)j 6= 0: (4.11) If = u(; r; z ) is well-dened, then s u(; r; z ). If either u is not dened or s < u, then there exists a ritial point 2 Crit(f ) suh that jf s(z) j rj(f s)0(z)j: (4.12) In any ase Comp z; f s(z); f s; (KA2) 12 #(Crit(f ) rj(f s)0(z)j \ Crit(f s) = ;: (4.13) Sine z 2= Crit(f ) and in view of Proposition 4.17, there exists a minimal number s = s(; ; r; z ) for whih at least one of the following two onditions is satised jf s(z) j rj(f s)0(z)j (4.14) for some 2 Crit(J (f )) or u(; r; z ) is well-dened and s(; ; r; z ) = u(; r; z ) (4.15) Sine j(f s)0(z)j 6= 0, the parts (4.11) and (4.12) are proved. In order to prove (4.13) notie rst that no matter whih of the two numbers s is, in view of Proposition 4.17 we always have rj(f s )0 (z )j 1 (4.16) Proof. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 31 Let us now argue that for every 0 j s diam Comp(f s j (z); f s(z); f j ; rj(f s)0(z)j) (4.17) Indeed, if s = u, it follows immediately from Proposition 4.17 and (4.3)s sine . Otherwise s s 0 1 jf (z) j rj(f ) (z)j < and therefore, by (2.14), f (z) 2= B ( (f ); ). Thus (4.17) follows from (2.16). Now by (4.17) and (Lemma 2.8), there exists 0 p #(Crit(f )), an inreasing sequene of integers 1 k1 < k2 < : : : < kp s and mutually distint ritial points 1 ; 2; : : : ; p of f suh that fl g = Comp(f s kl (z); f s(z); f kl ; rj(f s)0(z)j) \ Crit(f ): (4.18) for every l = 1; 2; : : : ; p and if j 2= fk1; k2; : : : ; kpg, then Comp(f s j (z); f s(z); f j ; rj(f s)0(z)j) \ Crit(f ) = ;: (4.19) Setting k0 = 0 we shall show by indution that for every 0 l p Comp(f s kl (z); f s(z); f kl ; (KA2) 12 l rj(f s)0 (z)j) \ Crit(f kl ) = ;: (4.20) Indeed, for l = 0 there is nothing to prove. So, suppose that (4.20) is true for some 0 l p 1. Then by (4.19) Comp(f s (kl 1) (z); f s(z); f kl 1; (KA2) 1 2 lrj(f s)0 (z)j) \ Crit(f kl 1 ) = ;: So, if l+1 2 Comp(f s kl (z ); f s (z ); f kl ; (KA2 ) 1 2 (l+1) rj(f s)0 (z )j) then by Lemma 1.4 applied for holomorphi maps H = f , Q = f kl 1 and the radius R = (KA2 ) 1 2 (l+1) rj(f s)0 (z )j < we get jf s kl (z) l+1j KA2j(f kl )0(f s kl (z))j 1 (KA2 ) 12 (l+1) rj(f s)0(z)j = 2 (l+1) rj(f s kl (z))0 j rj(f s kl (z))0 j whih ontradits the denition of s and proves (4.20) for l + 1. In partiular it follows from (4.20) that Comp(z; f s(z); f s; (KA2 ) 12 #(Crit(f ) rj(f s)0(z)j) \ Crit(f s) = ; The proof is nished. +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 JANINA KOTUS AND MARIUSZ URBANSKI 32 4.3. Hausdor and Conformal Measure. Let m be a Borel probability measure on CI and let me be its Eulidean version, i.e. (1 + jzj2 )t. We will need in this and the next setion the following. dme dm (z) = Lemma 4.19. If z 2 J (f ), rn & 0 and M = limn!1 rn t me (B (z; rn )), then m Bs (z; (2(1 + jz j2 )) 1 rn limn!1 sup ((2(1 + jzj2)) 1r )t 2t M n and Proof. and sine we get and m Bs(z; 2(1 + jz j2 ) 1 rn lim inf (2(1 + jzj2) 1 r )t 2 tM n!1 n Sine for every r > 0 suÆiently small B (z; 2 1 (1 + jz j2 )r) Bs(z; r) B (z; 2(1 + jz j2 )r) lim mme((BB((z;z;rr)))) = (1 + jzj2)t ; r&0 m Bs(z; (2(1 + jz j2 )) 1 rn limn!1 sup ((2(1 + jzj2)) 1 r )t n m Bs (z; 2(1 + jz j2 ) 1 rn lim inf (2(1 + jzj2) 1r )t n!1 n m(B (z; rn )) t nlim !1 2 t (1 + jz j2 ) t rnt = 2 M m(B (z; rn )) t nlim t !1 2 (1 + jz j2 ) t rt = 2 M: n We are done. Our rst goal is to show that the h-onformal measure m proven to exist in Lemma 3.7 is atomless and that H h(J (f )) = 0. We will onsider almost t-onformal measures with t 1. The notion of upper estimability introdued in Denition 4.5is onsidered with respet to the Eulidean almost t-onformal measure e. Reall that l = l(f ) 1 is the integer laimed in Lemma 2.20 and put Rl (f ) = inf fR(f j ; ) : 2 Crit(f ) and 1 j l(f )g = minfR(f j ; ) : 2 Crit(f ) \ R and 1 j l(f )g < 1 and Al (f ) = supfA(f j ; ) : 2 Crit(f ) and 1 j l(f )g = maxfA(f j ; ) : 2 Crit(f ) \ R and 1 j l(f )g where the numbers R(f j ; ) and A(f j ; ) are dened just above Denition 1.1. Sine the number of equivalene lasses of the relation is nite, looking at Lemma 2.20 and Lemma 4.15, the following lemma follows immediately from Lemma 4.13. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 33 (u) 0 is a positive onstant and t 7! Ct;i; 1 2 (0; 1), t 2 [1; 1), is a (u) ontinuous funtion suh that all points z 2 PC(f )i are (r; Ct;i;1 )-t:u.e. with respet to any Eulidean almost t-onformal measure e (with t 1) for all 0 < r Ri;(u1) , then there exists (u) a ontinuous funtion t 7! C~t;i; 1 > 0, t 2 [1; 1), suh that all ritial points 2 Cri+1 (f ) are (u) ~ (r; Ct;i;1)-t:u.e. with respet to any Eulidean almost t-onformal measure e for all 0 < r Lemma 4.20. If Ri;(u1) > Al 1 Ri;(u1) . We shall now prove the following. (u) 0 is a positive onstant and t 7! Ct;i; 2 2 (0; 1), t 2 [1; 1), is a (u) ontinuous funtion suh that all ritial points 2 Si (f ) are (r; Ct;i; 2 )-t:u.e. with respet to any Eulidean almost t-onformal measure e (with t 1) for all 0 < r Ri;(u2) , then there (u) ~i;(u2) > 0 suh that all points exist a ontinuous funtion t 7! C~t;i; 2 > 0, s 2 [1; 1), and R (u) z 2 PC(f )i are (r; C~t;i; 2 )-t:u.e. with respet to any Eulidean almost t-onformal measure e (with t 1) for all 0 < r R~ i;(u2) . We shall show that one an take n o 1 1 ; R(u) ; 1 and C~ (u) = maxfK 2 2t C (u) ; K 2t B g: R~i;(u2) = min jjf 0 jjPC( s i;2 t;i;2 t;i;2 f) Indeed, denote #(Crit(J (f ))) by #. Put = 2K (KA2)2# and then hoose > 0 so large that n o < min 1; 1 ; 1 1 minf; ; Ri;(u2) =2g : (4.21) Consider 0 < r R~i;(u2) and z 2 PC(f )i. If z 2 Crit(J (f )), then z 2 Si(f ) and we are done. Thus, we may assume that z 2= Crit(J (f )). Let s = s(; ; r; z). By the denition of , 2Krj(f s)0(z)j = (KA2) 1 2 #rj(f s)0(z)j (4.22) Suppose rst that u(; r; z) is well dened and s = u(; r; z). Then bys Proposition 4.17(4.4) or Proposition 4.17(4.6), applied with = 2K , we see that the point f (z) is (2Krj(f s)0(z)j; Bt )t:u.e.. Using (4.22), it follows from Proposition 4.18(4.13) and Lemma 4.10 that the point z is (r; K 2hBh)-h:u.e.. If either u is not dened or s < u(; r; z), then in view of Proposition 4.18(4.13), there exists a ritial point 2 Crit(J (f )) suh that jf s(z) j rj(f s)0(z)j. Sine s u, by Proposition 4.17 and (4.21) we get 2Krj(f s)0(z)j rj(f s)0(z)j < 1 minf; Ri;(u2) =2g (4.23) Lemma 4.21. If Ri;(u2) > Proof. JANINA KOTUS AND MARIUSZ URBANSKI 34 Sine z 2 PC(f )i, it implies that 2 Si (f ). Therefore using (4.23), the assumptions of Lemma 4.21, and (4.22) and then applying Proposition 4.18(4.13) and Lemma 4.10, we on(u) lude that z is (r; K 22t Ct;i; 2 )-t:u.e.. The proof is omplete. Lemma 4.22. If b 2 f 1 (1), if is a Eulidean almost t-onformal measure with t > q2b q+1b suh that (b) = 0, and if m is the h-onformal measure proven to exist in Lemma 3.7, then (Bb (R)) R2 and qb +1 qb t me (B (b; r)) r(qb +1)h 2qb 0 < r 1. It follows from Lemma 4.15 that me (fz 2 CI : R jz j < 2Rg) R2 and (fz 2 CI : R jz j < 2Rg) R2 for all R > 0 large enough. It therefore follows from (2.2) that qb (4.24) me (Bb (R) n Bb (2R) R2 R qb h : and qb (Bb (R) n Bb (2R) R2 R qb t : (4.25) Fix now r > 0 so small that R = (r=L) qb is large enough for the formula (4.24) and (4.25) to hold. Using (2.4) and0(4.25) we therefore get 1 for all Proof. +1 +1 (Bb (R)) = 1 X j 0 1 X Bb (2j R) n Bb (2j +1R) A = Bb (2j R) n Bb (2j +1 R) j =0 1 j 2 qb +1 qb +1 X t 2 t j 2 j (2 R) (2 R) qb = R qb 2 j =0 =L [ qb 2 qb +1 t qb r(qb +1)t 2qb 1 j 2 X j =0 2 j =0 qb +1 qb t qb +1 qb t r(qb+1)t 2qb ; where the last omparability sign was written sine qbq+1b t > 2. We are done with the rst part of2q our lemma. Replae now in the above formula by me and t by h, whih is greater than b qb +1 due to Theorem 2.1. Sine in this ase the \" sign an be, due to (4.24), replaed by the omparability sign \", sine the rst equality sign beomes \" (we do not rule out the possibility that me(b) > 0 yet), and sine me(B (b; r)) (Bb (R)), we are also done in this ase. We shall prove now the following. Lemma 4.23. The h-onformal measure m for f : J (f ) ! J (f ) [ f1g proven to exist in Lemma 3.7 is atomless. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 35 Using the indution on i = 0; 1; : : : ; p, it follows immediately from Lemma 4.21 (whih is an indutive step and for i = 0 the rst step of indution as S0 (f ) = ;), Lemma 4.20, and Lemma 2.19 that there exists a ontinuous funtion t 7! Ct 2 (0; 1), t 2 [1; 1), suh that if is an arbitrary almost t-onformal measure on J (f ), then e (B (x; r)) Ct rt (4.26) for all x 2 PC(f ) and all r r0 for some r0 > 0 suÆiently small. Consider now the almost tn -onformal measures mn = mBs (Y;1=n) (n is assumed to be so large that Bs (Y; 1=n) V ), where tn = S (Bs(Y; 1=n)). Letting n ! 1 and realling that m is a week limit of measures mn , formula (4.26) gives me (B (x; r)) Ch rh (4.27) for all x 2 PC(f ) and all r r0. It now follows from Lemma 4.19 that m(B (x; r) limr&sup 2 h Ch : rh 0 for all x 2 PC(f ). In partiular m(Crit(f )) = 0 and onsequently Proof. 0 [ m f n n0 2qb ; h qb +1 1 A (Crit(f )) = 0: (4.28) Fix now b 2 f 1(1). Fix t 2 . Consider all integers n 1 so large that tn t. 1 1 Sine mn (f (1)) mn(f (Bs(Y; 1=n)) = 0, it then follows from Lemma 4.22 that qb qb mn (Bb (R)) R2 qb tn R2 qb t : Hene me (b) = 0. Sine m and me are equivalent I , this gives m(b) = 0. Sine S on C S n n n0 f (b) \ Crit(f ) = ;, this implies that m n0 f (b) = 0. Invoking now (4.28) and Lemma 3.7 nishes the proof. +1 +1 Theorem 4.24. There exists a unique atomless t-onformal measure m for f : J (f ) ! J (f )[f1g. Then t = h, m is ergodi onservative and all other onformal measures are purely atomi, supported on Sing (f ) with exponents larger than h. Consequently m(Tr(f )) = 1. Proof. In view of Lemma 4.23 there exists an atomless h-onformal measure m for f : J (f ) ! J (f ) [ f1g. Suppose that is an arbitrary t-onformal measure for f and some t 0. By Lemma 3.1, t h. Fix z 2 J (f ) n (I1 (f ) [ Sing (f )). Then in view of Proposition 2.21 theren exist a point y(z) 2 J (f ) and an inreasing sequene fnk g1k=1 suh that y (z ) = limk!1 f k (z ). Dene for every l 1 Zl = fz 2 J (f ) n (I1 (f ) [ Sing (f )) : jy (z )j l and (z ) 1=lg; x l 1 and z 2 Zl . Considering for k large enough the sets fz nk (B (y; 41l )) and fz nk (B (y; 4Kl1 ))), where fz nk is the holomorphi inverse branh of f nk dened on B (y; 21l ) and sending f nk (z) JANINA KOTUS AND MARIUSZ URBANSKI 36 to z, using onformality of the measure along with Koebe's distortion theorem, we easily dedue that B (; l) 1 j(f nk ) (z )j h Bs (z; j(f nk ) (z )j 1 ) B (; l)j(f nk ) (z )j h (4.29) for all k 1 large enough, where K 1 is the onstant appearing in the Koebe's distortion theorem and asribed to the sale 1=2 and > 0 is some onstant omparable with 1. Fix now E , an arbitrary bounded Borel set ontained in Zl . Sine m is regular, for every x 2 E there exists a radius r(x) > 0he form from (4.29) suh that [ m( Bs (x; r(x)) n E ) < : (4.30) x2E Now by the Besiovi theorem (see [G℄) we an hoose a ountable subover fBs(xi; r(xi))g1i=1, r(xi ) , from the over fBs (x; r(x))gx2E of E , of multipliity bounded by some onstant C 1, independent of the over. Therefore by (4.29) and (4.30), we obtain 1 1 X X (E ) (Bs (xi ; r(xi ))) B (; l) r(xi )t i=1 i=1 1 X B (; l)B (m; l) r(xi)t hm(Bs (xi; r(xi))) i=1 1 [ B (; l)B (m; l)Ct hm( Bs(xi ; r(xi))) (4.31) i=1 CB (; l)B (m; l) ( + m(E )): In the ase when t > h, letting & 0 we obtain (Zl ) = 0. Sine J (f ) n (I1 (f ) [ Sing (f )) = S1 l=1 Zl , we therefore get J (f ) n (I1 (f ) [ Sing (f )) = 0 whih by Lemma 3.2 implies that t h (Sing (f )) = 1 and the last part of our theorem is proved . Suppose now that t = h. Sine, in view of Lemma 3.2, (I1(f ) n I (f )) = m(I1(f )) = 0, using (4.31) and letting l % 1, we onlude that jJ (f )nSing (f ) << mjJ (f )nSing (f ) . Exhanging the roles of m and we infer that the measures jJ (f )nSing (f ) and mjJ (f )nSing (f ) are equivalent. Suppose that (Sing (f )) > 0. Then there exists y 2 Crit(J (f )) [ (f ) [ f 1(1) suh that m(y) > 0. But then X j(f n())( )j h < 1; S 2y where y = n0 f n(y ) and for every 2 y , n( ) is the least integer n 0 suh that n f ( ) = y . Hene, P j(f n())( )j hÆ y = P2y n( ) ) ( )j h 2y j(f is an h-onformal measure supported on y Sing (f ). This ontradits the proven fat that the measures y jJ (f )nSing (f ) and mjJ (f )nSing (f ) are equivalent and m(J (f ) n Sing (f )) = 1. Thus and m are equivalent. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 37 Let us now prove that any h-onformal measure is ergodi. Indeed, suppose to the ontrary that f 1(G) = G for some Borel set G J (f ) with 0 < m(G) < 1. But then the two onditional measures G and J (f )nG (B \ G) (B \ J (f ) n G) G (B ) = ; J (f )nG (B ) = (G) (J (f ) n G) would be h-onformal and mutually singular; a ontradition. If now is again an arbitrary h-onformal measures, then by a simple omputation based on the denition of onformal measures we see that the Radon-Nikodyn derivative = d=dm is onstant on grand orbits of f . Therefore by ergodiity of m we onlude that is onstant m-almost everywhere. As both m and are probability measures, it implies that = 1 a.e., hene = m. Let us show now that m is onservative. We shall prove rst that every forward invariant (f (E ) E ) subset E of J (f ) is either of measure 0 or 1. Indeed, suppose to the ontrary that 0 < m(E ) < 1. Sine m(I1(f ) [ Sing (f )) = 0, it suÆes to show that m(E n (I1 (f ) [ Sing (f ))) = 0: Denote by Z the set of all points z 2 E n (I1(f ) [ Sing (f ))) suh that lim m(B (z; r) \ (Emn(B(I(1z;(rf))) [ Sing (f )))) = 1: (4.32) r!0 In view of the Lebesgue density theorem (see for example Theorem 2.9.11 in [Fe℄), m(Z ) = m(E ). Sine m(E ) > 0 we nd at least one point z 2 Z . Sine z 2 J (f ) n (I1(f ) [ Sing (f )), let x 2 J (f ), (z) > 0, and an inreasing sequene fnk g1k=1 be given by Proposition 2.21. Æ = (z )=8: Suppose that m(B (x; Æ) n E ) = 0. By onformality of m, m(f (Y )) = 0 for all Borel sets Y suh that m(Y ) = 0. Hene, 0 = m f n(B (x; Æ) n E ) m f n(B (x; Æ)) n f n(E ) (4.33) m f n(B (x; Æ)) n E m f n(B (x; Æ) m(E ) for all n 0. Sine J (f ) = Sn1 f n(1), for some p 2, the image f p 1(B (x; Æ)) ontains an open neighbourhood of 1. thus, it ontains at leastp one (in fat innitely many) opy parallelogram R and onsequently f (B (x; Æ)) = CI . In partiular of the fundamental p m f (B (x; Æ )) = 1. Then (4.33) implies that 0 1 m(E ) whih is a ontradition. Con nj (z ); 2Æ ) n E sequently m ( B ( x; Æ ) n E ) > 0. Hene for every j 1 large enough, m B ( f m B (x; Æ ) n E > 0. Therefore, as f 1 (J (f ) n E ) J (f ) n E , the standard appliation of Koebe's Distortion Theorem shows that m(B (z; r) n E ) >0 limr!sup m(B (z; r)) 0 JANINA KOTUS AND MARIUSZ URBANSKI 38 whih ontradits (4.32). Thus either m(E ) = 0 or m(E ) = 1. Now onservativity is straightforward. One needs to prove that for every Borel set B J (f ) with m(B ) > 0 one has m(G) = 0, where X G = fx 2 J (f ) : B (f n (x)) < +1g: n0 Indeed, suppose that m(G) > 0 and for all n 0 let X Gn = fx 2 J (f ) : B (f n (x)) = 0g = fx 2 J (f ) : f k (x) 2= B for all k ng: S kn Sine G = n0 Gn, there exists k 0 suh that m(Gk ) > 0. Sine all the sets Gnn are forward invariant we onlude that m(Gk ) = 1. But on the other hand all the sets f (B ), n k, are of positive measure and are disjoint from Gk . This ontradition nishes the proof of onservativity of m. Consequently m(Tr(f )) = 1. Sine, by Lemma 3.1, Hh m, we thus see that Hh(J (f ) n Tr(f )) = 0. We are done. The proof of part (a) of Theorem 4.1. Let m be the unique h-onformal atomless measure proven to exist in Theorem 4.24. Consider an arbitrary point z 2 Tr(f ). Fix a pole b 2 f 1 (1). Sine b 2= O+(Crit(f )), there exists > 0 suh that B (b; ) \ O+ (Crit(f )) = ;: (4.34) Sine z 2 Tr(f ), there exists an innite inreasing sequene fnj g1j=0 suh that lim f nj (z) = b and jf nj (z) bj < =4 (4.35) j !1 for every j 1. It nfollows from this and (4.34) that for every j 1 there exists a holomorphi inverse branh fz j : B (f nj (z); 3=4) ! CI of f nj sending f nj (z) to z. Using now Koebe's Distortion Theorem (Eulidean version) and Lemma 4.22, we onlude that me z; B K j(f nj )0 (z )j 1 2jf nj (z ) bj me B f nj (z ); 2jf nj (z ) bj j(f nj )0 (z )j h me B (b; jf nj (z) bj) j(f nj )0 (z)j h jf nj (z) bj(qb+1)h 2qb j(f nj )0(z)j h = K j(f nj )0 (z)j 1jf nj (z) bj hK hjf nj (z) bjqb(h 2) : Sine h < 2, using (4.35), this implies that limr!0r hme(B (z; r)) = 1. Hene Hh(Tr(f )) = 0 in view of Theorem 4.3. Sine by Theorem 4.24 mhe(J (f ) n Tr(f )) = 0, it follows from h Lemma 3.1 that H (J (f ) n Tr(f )) = 0. In onlusion H (J (f )) = 0 and the proof is omplete. Proposition 4.25. The onformal measure m is absolutely ontinuous with respet to the paking measure t and moreover, the Radon-Nikodym derivative dm=dt is uniformly bounded away from innity. In partiular t (J (f )) > 0. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 39 Sine J (f ) \ ! Crit(f ) n Crit(J (f )) = (f ), we onlude from Lemma 2.9 that there exists y 2 J (f ) at a positive distane, say 8, from O+(Crit(f )). Fix zn 2 Tr(f ). Then there exists ann innite sequene nj 1 of inreasing integers suh that f j (z) 2 B (y; ). Therefore B (f j (z); 4) \ O+(Crit(f )) = ; and onsequently Comp(z; f nj (z); f nj ; =2) \ Crit(f nj ) = ; Hene, it follows from Lemma 1.2 and Lemma 4.10 that me (B (z; r)) limr!inf B 0 rh for some onstant B 2 (0; 1) and all z 2 Tr(f ). Applying Lemma 4.19 we therefore get that m(Bs (z; r)) limr!inf 2hB: 0 rh Hene, by Theorem 4.4(1), the measure mjTr(f ) is absolutely ontinuous with respet to hjTr(f ) . Sine, by Theorem 4.24, m(J (f ) n Tr(f )) = 0, we are done. Proof. Lemma 4.26. If (f ) 6= ;, then h (J (f )) = +1. S n Proof. Fix ! 2 . Sine n0 f (! ) is dense in J (f ) and, by Lemma 2.9, ! (Crit(f )) is non-where dense S in J (f ), there exist an integer s > 0, a real number > 0, and a point s y 2 f (! ) n B n0 f n (Crit(f )); . Sine by Theorem 2.1, h > 1, it follows from Lemma 2.5 and Lemma 4.13 (y may happen to be a ritial point of f s!) that me (B (y; r)) = 0: (4.36) limr!inf 0 rh Consider now a transitive point z 2 J (f ), i.e. z 2 Tr(f ). Then there exists an innite inreasing sequene nj = nj (z) 1 of positive integers suh that lim jf nj (z) yj = 0 and rj = jf nj (z) yj < =7 j !1 fornevery j = 1; 2; : : : . By the hoie of y , for all j 1 there exist holomorphi inverse branhes j fz : B (f nj (z ); 6rj ) ! CI sending f nj (z ) to z . So, applying Lemma 1.2 and Lemma 4.10 with R = 3rj , we onlude from (4.36) that me (B (z; r)) limr!inf = 0: 0 rh Applying Lemma 4.19, we onlude that the same formulas remain true with me replaed by m and B (z; r) by Bs(z; r). Therefore, it follows from Theorem 4.24 (m(Tr(f )) = 1) and Theorem 4.4(1) that h(J (f )) = +1. We are done. From now on let m denote the unique atomless h-onformal measure m proven to exist in Theorem 4.24. JANINA KOTUS AND MARIUSZ URBANSKI 40 Sine the number of equivalene lasses of the relation is nite, looking at Lemma 2.20 and Lemma 4.15, the following lemma follows immediately from Lemma 4.14 Lemma 4.27. If Ci;l 1 > 0, 0 < Ri;l 1 Rl (f )=3, and 0 < 1 are three real numbers suh that all points z 2 PC(f )i are (r; ; Ci;l 1 )-h:s.l.e. with respet to the measure me , then there exists C~i;l 1 > 0 suh that all ritial points 2 Cri+1 (f ) are (r; ~ ; C~i;l 1 )-h:s.l.e. with respet to the measure me for all 0 < r Al 1 Ri;l 1 . Let us prove the following. Assume that Ci;(l2) > 0, Ri;(l2) > 0 and 0 < 1 are three real numbers suh that all ritial points 2 Si (f ) are (r; ; Ci;(l2) )-h:s.l.e. with respet to the measure me for all 0 < r Ri;(l2) . Then there exist C~i;(l2) > 0, R~i;(l2) > 0 and suh that all points z 2 PC(f )i are (r; 8K 3 A2 2#(Crit(f ) ; C~i;(l2) )-h:s.l.e. with respet to the measure me for all 0 < r R~i;(l2) . Lemma 4.28. Suppose that (f ) = ;. We shall show that this time one an take R~i;(l2) = minfjjf 0 jjF 1 1 ; Ri;(l2) ; 1g and C~i;(l2) = (8(KA2 )2# )hCi;(l2) ; where jjf 0jj = jjf 0jjPC(f )i . Indeed, denote again #(Crit(f )) by #. Take = 4K (KA2 )2# and then hoose > 0 so large that n o < min 1; 1 ; 1 1 minf; ; Ri;(l2) =2g : (4.37) Consider 0 < r R~i;(l2) and z 2 PC(f )i. If z 2 Crit(J (f )), then z 2 Si(f ) and we are done. Thus, we may assume that z 2= Crit(J (f )). Let s = s(; ; r; z). By the denition of 4Krj(f s)0(z)j = (KA2 ) 12 #rj(f s)0(z)j: (4.38) Suppose rst that u(; r; z) is well dened and s = u(; r; z). Then by Proposition 4.17(4.5) or Proposition 4.17(4.6), applied with = K , we see that the point f s (z ) is (Krk(f s)0 (z )k; =K 2 ; Wh (=K 2 )) h:s.l.e.: Using (4.38) it2 follows from Proposition 4.18(4.13) and Lemma 4.11 that the point z is (r; ; Wh(=K ))-h:s.l.e.. If either u is not dened or s u(; r; z), then in view of Proposition 4.18(4.12), there exists a ritial point 2 Crit(f ) suh that jf s(z) j rj(f s)0(z)j. Sine s u, by Proposition 4.17 and (4.37) we get 4Krj(f s)0 (z)j rj(f s)0(z)j < 1 minf; Ri;(l2) =2g: (4.39) Sine z 2 PC(f )i, it implies that 2 Si(f ). Therefore, by the assumptions of Lemma 4.28 and by (4.39) we onlude that is (2rj(f s)0(z)j; ; Ci;(l2) )-h:s.l.e.. Consequently, in view of Proof. GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 41 Lemma 4.7, the point f s(z) is (rj(f s)0(z)j; 2; 2hCi;(l2))-h:s.l.e.. So, by Lemma 4.8 this point is (Krj(f s)0(z)j; 2=K; (2K 1)hCi;(l2)) h:s.l.e. Using now formula (4.38) and Proposition 4.18(4.13) it follows from Lemma 4.11 that the point z is (r; 2K; (2K 1 )hCi;(l2)) h.s.l.e.. If z 2 Crit(J (f )), then by the denition of PC(f )i we see that z 2 Si(f ) and we are done in view of the assumption of the lemma and in view of the denitions of R~i;(l2) and C~i;(l2) ). The proof is ompleted. Lemma 4.29. If (f ) = ;, then he (F ) < 1 for every bounded Borel set F CI. Proof. Let qmin = minfqb : b 2 f 1 (1)g: Take 2 (0; 1) so small that if z 2 CI, then f jB(z;d) is 1-to-1for every d dist(z; Crit(f ) [ f 1 (1)). Using indution on i = 0; 1; : : : ; p, it follows immediately from Lemma 4.28 (whih is an indutive step and for i = 0 the rst step of indution as S0(f ) = ;), Lemma 4.27, and Lemma 2.19 that eah point z 2 PC(f ) is (r; ; G) hs.l.e. for some 2 (0; 1), G > 0, R > 0 and all r 2 (0; R). Without loss of generality we may assume R 2 (0; 1) to be so small that 1 jz bj qb jf (w)j jz bj qb (4.40) and supfjwj : w 2 PC(f )g 1R q 8R (4.41) for all b 2 f 1(1), all z 2 B (b; R) and some 1. Fix a point z 2 F n Sing (f ) and r 2 (0; R). In view of Corollary 2.22 there exists the least n 1 suh that either dist(f n(z); PC(f )) 8(K) 1rj(f n)0 (z)j or rj(f n)0(z)j 81 R: There are the following three possibilities. 10 1 K 1 rj(f n)0 (z )j < R: 8 This in partiular implies that dist(f n(z); PC(f )) 8(K) 1rj(f n)0 (z)j: 20 1 K 1 rj(f n)0 (z )j R and dist(f n (z ); PC(f )) > 8(K) 1 rj(f n)0 (z )j: 8 0 3 1 K 1 rj(f n)0 (z )j R and dist(f n(z ); PC(f )) 8(K) 1 rj(f n)0 (z )j: 8 min 42 JANINA KOTUS AND MARIUSZ URBANSKI Let us onsider the ase 10. Sine 8(K) 1rj(f n 1)0(z)j < dist(f n 1(z); PC(f )), we get 8K 1rj(f n 1)0(z)j < dist(f n 1(z); Crit(f )): (4.42) Suppose now that 8K 1rj(f n 1)0(z)j dist(f n 1(z); f 1(1)): This implies that there exists b 2 f 1(1) suh that jf n 1(z) bj < R. Hene, using (4.40), we get jf n(z)j 1jf n 1(z) bj qb 1R qb : On the other hand, using (4.41), we obtain jf n(z)j supfjwj : w 2 PC(f )g + dist(f n(z); PC(f )) supfjwj : w 2 PC(f )g + 8(K) 1rj(f n)0 (z)j supfjwj : w 2 PC(f )g + 8R 1R qb : This ontradition shows that 8K 1rj(f n 1)0 (z)j < dist(f n 1(z); f 1(1)): Alongn 1with (4.42) and the denition of , this implies that the map f restrited to the ball B (f (z ); 8K 1 rj(f n 1)0 (z )j), is univalent. It therefore follows from Koebe's 41 -theorem that f B (f n 1 (z ); 8K 1 rj(f n 1)0 (z )j) B (f n (z ); 2K 1 rj(f n)0 (z )j): (4.43) Thus,n 1there exists a nunique holomorphi inverse branh f 1 : B (f n(z ); 2K 1 rj(f n)0 (z )j) ! 1 1 0 n n 1 B (f (z ); 8K rj(f ) (z )j) of f sending f (z ) to f (z ). Sine B (f n 1 (z ); 8K 1 rj(f n 1)0 (z )j) \ PC(f ) = ; there exists a unique holomorphi inverse branh fz (n 1) : B (f n 1 (z ); 8K 1 rj(f n 1)0 (z )j) ! CI of f n 1 sending f n 1(z) to z. Therefore, the omposition fz n = fz (n 1) Æ f 1 : B (f n (z ); 2K 1 rj(f n)0 (z )j) ! CI is a well-dened holomorphi inverse branh of f n sending f n(z ) to z . As dist(f n (z ); PC(f )) < R, sine K 1 rj(f n)0 (z )j < R and sine eah point z 2 PC(f ) is (r; ; G) hs.l.e., we obtain that me B (f n (z ); K 1 rj(f n)0 (z )j) G(K 1 rj(f n)0 (z )j)h : Using now Koebe's distortion theorem, we onlude that me (B (z; r)) me fz n B (f n (z ); K 1 rj(f n)0 (z )j) (4.44) K hj(f n)0(z)j hme (B (f n(z); K 1 rj(f n)0(z)j)) K hj(f n)0(z)j hGhK hrhj(f n)0(z)jh = (GK 2 rh: GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 43 Let nus now deal with the ase 20n. In this nase the holomorphi inverse branh fz n : 1 n 0 B (f (z ); 2K rj(f ) (z )j) ! CI of f sending f (z ) to z is well-dene. Using Koebe's distortion theorem and Lemma 4.15, we get me (B (z; r)) me fz n B (f n (z ); K 1 rj(f n)0 (z )j) K hj(f n)0 (z)j hme(B (f n(z); K 1rj(f n)0 (z)j)) (4.45) K hj(f n)0 (z)j hCh 81 R (K 1rj(f n)0(z)j)h 1 = Ch 8 R K 2hrh Case 33 . Suppose rst that jf n 1(z) bj 12 K 1 rj(f n 1)0(z)j for some pole b 2 f 1(1). Then 1 B (f n 1(z ); K 1 rj(f n 1)0 (z )j B (b; K 1 rj(f n 1)0 (z )j): (4.46) 2 Sine 21 K 1rj(f n 1)0 (z)j 161 R, it follows from Lemma 4.22 that 1 (q +1)h 2qb 1 me B (b; K 1 rj(f n 1)0 (z )j) C K 1 rj(f n 1)0 (z )j b (4.47) 2 2 for some universal onstant C > 0. Sine 2K 1rj(f n 1)0 (z)j < 8K 1 1 rj(f n 1)0(z)j dist(nf n1 1(z); PC(1 f )),nwe1 0see that there nexists a unique holomorphi inverse branh fz (n 1) : B (f (z ); 2K rj(f ) (z )j) ! CI of f 1 sending f n 1 (z ) to z . Therefore, applying (4.46), (4.47) and Koebe's distortion theorem, we obtain me (B (z; r)) me fz (n 1) B (f n (z ); K 1 rj(f n 1)0 (z )j) K hj(f n 1)0(z)j hme B (f n 1(z); K 1rj(f n 1)0(z)j) K hj(f n 1)0(z)j hme B (b; 21 K 1rj(f n 1)0 (z)j) (4.48) K hC j(f n 1)0(z)j h 21 K 1 rj(f n 1)0(z)j (qb+1)h 2qb CK h K 1rj(f n 1)0 (z)j qb(h 2) rh q (h 2) 1 h CK 8 R rh : So, suppose nally that jf n 1(z) bj > 21 K 1 rj(f n 1)0(z)j: max JANINA KOTUS AND MARIUSZ URBANSKI 44 for all poles b 2 f 1(1). Sine also dist(f n 1(z); PC(f )) > 4K 1 1rj(f n 1)0 (z)j; (4.49) we onlude that the map f : CI ! CI, restrited to the ball B (f n 1(z); 12 K 1 rj(f n 1)0(z)j), is univalent. It therefore follows from Koebe's 14 -theorem that 1 1 f B (f n 1 (z ); K 1 rj(f n 1)0 (z )j) B K 1 rj(f n)0 (z )j : 2 8 Hene, there exists a unique holomorphi inverse branh f 1 : B f n(z); 18 K 1 rj(f n)0(z)j ! B (f n 1 (z ); 12 K 1 rj(f n 1 )0 (z )j) of f sending f n(z ) to f n 1 (z ). In view of (4.49) there exists a unique holomorphi inverse branh fz (n 1) : B (f n 1(z); 21 K 1rj(f n 1)0 (z)j) ! CI of f n 1 sending f n 1(z) to z. Hene, the omposition 1 fz n = fz (n 1) Æ f 1 : B f n (z ); K 1 rj(f n)0 (z )j ! CI 8 n is a well-dened holomorphi inverse branh of f sending f n(z) to z. Sine 161 K 1 rj(f n)0(z)j > 2 72 R, applying Koebe's distortion theorem and Lemma 4.15, we get 1 me (B (z; r)) me fz (n 1) B (f n (z ); K 1 rj(f n)0 (z )j) 16 h n 0 h K j(f ) (z)j me B (f n(z); (16K ) 1 rj(f n)0(z)j) K hCh(2 72 R)j(f n)0(z)j h (16K ) 1rj(f n)0(z)j h = (16) 1K 2hCh(2 72R)rh: Combining this inequality along with (4.44) (4.45) and (4.48), we onlude that he(F ) < 1. We are done. Our last lemma in this setion is this. Lemma 4.30. If (f ) = ;, then the spherial paking measure h (J (f )) is nite. Sine the paking measure h is -invariant, it follows from Lemma 4.29 and Proposition 4.25 that he J (f ) \ (B (0; 2R) n B (0; R)) R2 for all R 1. Sine in addition dPh 2 h dhe (z ) = (1 + jz j ) and sine h > 1, we get Proof. 1 X h J f \ B ; n+1 n B ; n n=0 1 X 2hn h J f \ B ; n+1 n B ; n e n=0 1 1 X 2hn 2n X (2 2h)n < 1: n=0 n=0 h J (f ) \ (CI n B (0; 1)) = ( ) ( (0 2 ) (0 2 )) 2 ( ) ( (0 2 ) 2 2 = 2 (0 2 )) GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 45 We are done. The proof of Theorem 4.1 is therefore omplete. 5. Invariant Measures In this setion we deal with -nite invariant measures equivalent to the onformal measure m. We prove their existene, ergodiity, onservativity and we detet the points around whih these measures are nite or innite. This allows us to provide suÆient onditions for their niteness. 5.1. -nite invariant measures equivalent to the onformal measure m. In order to prove Theorem 5.2 below we apply a general suÆient ondition for the existene of -nite absolutely ontinuous invariant measure proven in [15℄. In order to formulate this ondition suppose that X is a -ompat metri spae, is a Borel probability measure on X , positive on open sets, and that a measurable map f : X ! X is given with respet to whih measure is quasi-invariant, i.e. Æ f 1 << . Moreover we assume the existene of a ountable partition = fAn : n 0g of subsets of X whih are all -ompat and of positive measure S . We also assume that (X n n0 An ) = 0, and if additionally for all m; n 1 there exists k 0 suh that (f k (Am ) \ An ) > 0; then the partition is alled irreduible. Martens' result omprising Proposition 2.6 and Theorem 2.9 of [15℄ reads the following. Theorem 5.1. Suppose that = fAn : n 0g is an irreduible partition for T : X ! X . Suppose that T is onservative and ergodi with respet to the measure . If for every n 1 there exists Kn 1 suh that for all k 0 and all Borel subsets A of An (A) (f k (A)) (A) Kn 1 Kn ; k (An ) (f (An )) (An ) then T has a -nite T -invariant measure absolutely ontinuous with respet to . In addition, is equivalent with , onservative and ergodi, and unique up to a multipliative onstant. Moreover, for every Borel set A X Pn k k=0 (f (A)) : (A) = nlim P !1 nk=0 m(f k (A0 )) The rst result of this setion is the following. Theorem 5.2. There exists a -nite f -invariant measure absolutely ontinuous with respet to h-onformal measure m. In addition, is equivalent with m and ergodi. JANINA KOTUS AND MARIUSZ URBANSKI 46 Let 2 CI be a periodi point of f with some period p 3. We put P3 (f ) = O+ (f (Crit(f ))) [ f; f ( ); : : : ; f p 1( )g: Sine O+(f (Crit(f )) is a forward-invariant nowhere-dense subset of J (f ) and sine the honformal measure m is positive on nonempty open subsets of J (f ), it follows from ergodiity and onservativity of m (see Theorem 4.24) that m(O+(f (Crit(f )))) = 0. Sine m has no atoms (see Theorem 4.24) we therefore obtain that m(P3 (f )) = 0. We shall now onstrut the partition of the set J (f ) n P3 (f ). We shall hek next that it satises the assumptions of Theorem 5.1. We rst dene the family of balls 1 B z; dist(z; P3 (f )) : 2 z 2CInP (f ) This family obviously overs CI n P3(f ). Sine CI n P3(f ) is an open set, it is a Lindelof spae, and therefore we an hoose a ountable subover of CI n P3 (f ), whih we denote by 1 1 B zi ; dist zi ; P3 (f ) : 2 i=1 We indutively dene a partition A = fAig1i=0 of CI n P3 (f ) as follows. Let 1 A0 = B z0 ; dist(z0 ; P3 (f )) : 2 Assume that we have dened the set A1 ; : : : ; An suh that 1 Aj B zj ; dist(zj ; P3 (f )) 2 and IntAj 6= ;: Then An+1 we dene as [ n 1 An+1 = B zn+1 ; dist(zn+1 ; P3 (f )) n Aj : 2 Proof. 3 j =1 The set An+1 is disjoint with the sets A1 ; : : : ; An and [ n 1 1 An+1 B zn+1 ; dist(zn+1 ; P3 (f )) n B zj ; dist(zj ; P3 (f )) : 2 2 j =1 Thus either An+1 = ; or IntAn+1 6= ; and we remove all the empty sets. We shall now hek that the partition is irreduible. And indeed, it follows from the onstrution of the sets fAig1i=0 and ontinuity of the measure m that it suÆes to demonstrate that if z 2 CI, r > 0 and K CI is a ompat set, then there exists n 1 suh that 0 f n B (z; r) n [ k0 f k (1) 1 [ k AKn f k0 (1): GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 47 Sine the set of repelling periodi points is dense in the Julia ([2℄, omp. [5℄), there thus exists a periodi point x 2 B (z; r), say ofq period q 1. Sine x is repelling there exists s > 0 so S qj small that B (x; s) B (z; r) and f (B (x; s)) B (x; s). Sine j1 f (B (x; s)) CI, sine K is a ompat subset of CI and sine ff qj (B (x; s))g1 j =1 is an inreasing family of open sets, there thus exists k 1 suh that f qk (B (x; s)) K . Let us hek now the distortion assumption of Theorem 5.1. And indeed, in view of Koebe's n distortion theorem there exists a onstant K 1 suh that if f : B zi; dist( z I i ; P3 (f )) ! C 1 n is a holomorphi branh of f , then for every k 0 and all x; y 2 Ak B zi ; 2 dist(zi; P3(f )) we have j(f n)0(y)j K: (5.1) j(f n)0(x)j We therefore obtain for all Borel sets A; B Ak with m(B ) > 0 and all n 0 that R n 0h n 0h m(f n (A)) A j(f ) j dm supAk fj(f ) j gm(A) K h m(A) : R = n 0h m(f n (B )) inf Ak fj(f n)0jhgm(B ) m(B ) A j(f ) j dm and similarly m(f n (A)) m(A) K h : n m(f (B )) m(B ) Sine by Theorem 4.24 the measure is onservative ergodi, all the assumptions of Theorem 5.1 have been heked and we are done. The following lemma easily follows from Theorem 5.1. Lemma 5.3. For every n 0 we have 0 < (An) < 1. We say that the f -invariant measure produed in Theorem 5.2 is of nite ondensation at x 2 J (f ) if and only if there exists an open neighborhood V of x suh that (V ) < 1. Otherwise is said to be of innite ondensation at x. We respetively say that x is a point of nite or innite ondensation of . We end this subsetion with the following obvious results. Lemma 5.4. If x is a point of innite ondensation of , then eah point of the losure ff n(x) : n 0g is also of innite ondensation of . Lemma 5.5. The set of points of innite ondensation of measure is ontained in the union O+(Crit(f )) [ [ f1g. fAn : n 0g Proof. If z 2 = O+ (Crit(f )) [ [ f1g, then by loal niteness of the family Sk there exist an open neighborhood V of z and an integer k 0 suh that m V n j=0 Aj = 0. JANINA KOTUS AND MARIUSZ URBANSKI 48 Hene, in view of Lemma 5.3 and Theorem 5.2 ( m) we get (A) Pkj=0 (Aj ) < 1. The proof is nished. 5.2. 1 is a Point of Finite Condensation of . The goal of this subsetion is to prove that 1 is a point of a nite ondensation of the measure . We start with the following. Lemma 5.6. For every R > 1 large enough there exists a onstant C1 (R) > 0 suh that m(Bb (R)) C1 (R)diamhs (Bb (R)). Proof. For every k 0 let Ak;R = fz 2 C I : 2k R jz j < 2k+1Rg. As in the proof of Lemma 3.2 let BR+ = fz 2 BR n f1g : Imz > 0g BR1 = fz 2 BR n f1g : Imz < 1g and BR+ = fz 2 BR n f1g : Imz > 0g: We also put A+k;R = Ak;R \ BR+ and Ak;R = Ak;R \ BR . Using formula (2.3) we an write for all b 2 f 1(1), all j 2 f1; : : : ; qb g and all k 0 that m(fb;B1 and similarly Thus (A+k;R)) = R ;j Z + m(fb;B1 q A+ k;R j(fb;B1R ;j ) jhdm (1 + jbj2) h(2k R) bqb hm(A+k;R) 1 + q b ( Ak;R )) (1 + jbj2 ) h(2k R) qb ;j R + 1 h m(Ak;R ) 1 (A )) = m(f 1 (A+ )) + m(f 1 (A )) (1 + jbj2 ) m(fb;R;j k;R b;R;j k;R b;R;j k;R Summing now over all j 2 f1; : : : ; qb g, we get qb h q (2k R) bqb m(Ak;R;b ) (1 + jbj2 ) h (2k R) qb h m(Ak;R ) where Ak;R;b = Bb(R) \ f 1(Ak;R). Therefore, putting S = Pw2(1 + jwj2) Theorem 2.1 h > 1), we obtain X m(f 1 (Ak;R )) = m(Ak;R;b ) = b2f 1 (1) X X b2R\f 1 1 h (1) w2 X q 1 + jb + wj2 h(2k R) bqb m(Ak;R) b2R\f 1 (1) q 1 S kR q h m(Ak;R) (2 ) q (2k R) bqb 1 hX w2 1 h m(Ak;R ): (5.2) h < 1 (sine by m(Ak;R;b+w ) (1) w2 X X b2R\f 1 1 m(Ak;R ) (1 + jb + wj2) h GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 49 Hene m(Ak;R) (2k R) q q hS 1m(f 1(Ak;R))1 where q = maxfqb : b 2 R \ f 1 (1)g. Combining this and (5.2), we get for every b 2 f (1) that m(Ak;R;b ) (1 + jbj2 ) h(2k R)(1 q )h (2k R)( q 1)h S 1 m(f 1 (Ak;R )) (1 + jbj2 ) hS 1m(f 1 (Ak;R)) (1 + jbj2) hm(f 1(Ak;R)) Summing now over all k 0 we get m(Bb(R)) (1 + jbj2) hm(f 1(BR )) (1 + jbj2) h. Combining in turn this with (2.4) we get h m(Bb (R)) Lh R q diamhs (Bb (R)) (5.3) The proof is omplete. 1 1 1 Lemma 5.7. Fix R > 2 suÆiently large. Re-numerating the elements of the partition fAj g1j=0, we may assume that A0 BR and diams(A0 ) = 1. For every b 2 f 1(1) and every n 0 let A(n) = f n(A0 ) \ Bn , where Bn is a onneted omponent of f n (BR ). Then there exists a onstant C2 > 0 suh that m(Bn ) C2 (R)m(A(n) ). It follows from the onstrution of the partition fAngn0 that m(A(n) ) diamhs(A(n) ) (5.4) Sine dist(01; A0) R > 2 and sine diam(A0 ) = 1 using (2.3), and (2.4), we get for every pole b 2 f (1) that q diams(A0;b) (1 + jbj2) 1 dist(0; A0) bqb diams(A0) R qbqb R qb = 1; (5.5) diams(Bb (R)) 2 1 q b (1 + jbj ) R 1 where A0;b = f (A0) 1\ Bb(R). Sine !(Crit(f )) is a ompat subset of the omplex plane CI, dist(! (Crit(f )); f (1)) > 0. Therefore there exists r > 0 suh that for all R > 1 large enough Bb(R) B (b; r) and B (b; 2r) \ O+(Crit(f )) = ;. Sine Bn = f (n 1)(Bb (R)) for an appropriate holomorphi inverse branh f (n 1) : B (b; 2r) ! CI of f (n 1) , it follows from Proof. 1 1 1 1 Koebes's distortion theorem and (5.5) diams(A(n) ) = diams(f (n 1)(A0;b)) diams(Ao;b) 1 diams(Bn) diams(f (n 1) (Bb(R)) diams(Bb(R)) and that diamhs(Bn) = diamhs(f (n 1) (Bb(R))) diamhs(Bb(R)) : m(Bn ) m(Bb (R)) m(f (n 1) (Bb (R))) Combining the last two formulas and (5.4) we get h (B (R)) ! diam s b ( n ) h m(A ) diams (Bn ) m(Bn ) m(Bn ) m(Bb (R)) The proof is omplete. JANINA KOTUS AND MARIUSZ URBANSKI 50 We are ready now to prove the main result of this setion. Theorem 5.8. 1 is the point of nite ondensation of the measure . Proof. Take R > 0 so large as required in Lemma 5.7. It follows from this lemma that m(f k (BR )) C2 (R)m(f 1 (A0 )) for every k 0. Thus, applying Theorem 5.1 , we get Pn 1 k=0 m(f (BR )) C (R) < 1: (BR ) = nlim P 2 !1 nk=0 m(f 1 (A0 )) We are done. 5.3. All Points of Finite and Innite Condensation. We say that z 2 J (f ) n is geometrially good if m(Bn ) diamh (Bn ) (5.6) for every set B of suÆiently small diameter ontaining x, every n 0 and every onneted n omponent Bn of f (B ). The diretion of the inequality above makes that heking geometrial goodness one an assume the sets B to be balls entered at x. The most general suÆient ondition for nite ondensation is the following. Lemma 5.9. If z 2 J (f ) n is geometrially good, then z is a point of nite ondensation of measure . Proof. Sine z 2= , Staking > 0 suÆiently small, z 2= B ( ; ). Set B = B (z; ). Sine m(B ) > 0 and m( n0 An ) = 1, there exists i 0 suh that m(B \ Ai) > 0. Sine B \ Ai \ J (f ) has a non-empty interior relative to J (f ), there exists an open ball F B \ Ai having nonempty intersetionn with J (f ). Of ourse m(F ) > 0. For every n 0 let Bn nbe a onneted omponent of f (B ) and let Fn Bn be some onneted omponent of f (F ) ontained in Bn. Using Koebe's Distortion Theorem, I (Eulidean version) and the fat that the point z is geometrially good, we get ! ! diam( Fn ) h diam( Fn ) h h h m(Fn ) diam (Fn ) = diam(Bn) diam (Bn) m(Bn ) diam(Bn) Applying now Lemma 2.12 to the onneted sets F and B we obtain ! diam( F) h m(Fn ) m(Bn ) diam(B ) : Thus n X k=0 m(f k (B )) n X k=0 m(f k (F )) n X k=0 m(f k (Ai)): GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 51 Hene, using Lemma 5.3, we get (B ) (Ai) < 1 and therefore z is a point of nite ondensation of . In order to make use of this lemma we need to provide suÆient onditions for points to be geometrially good. This is done below. Lemma 5.10. If is h-upper estimable at every point z 2 J (f ) with the same estimability onstant, then every point z 2 J (f ) is geometrially good. The proof of this lemma follows by a straightforward indutive argument inorporating Koebe's Distortion Theorem, Lemma 4.13, niteness of the equivalene lasses of the relation on the set of ritial points of f , Lemma 2.8, and equivalently (2.16). Proof. Theorem 5.11. The set of points of innite ondensation of is ontained in the set of paraboli points (f ). Proof. The proof of Lemma 4.23 shows that eah point z 2 J (f ) is upper estimable with respet to the Eulidean h-onformal measure me and so, also with respet to the measure m. Therefore the proof of Theorem 5.11 is ompleted by applying Lemma 5.10 and Lemma 5.9. Corollary 5.12. If = ;, then there exists an f -invariant probability measure equivalent to m. Invariant measure - Paraboli points. From what we have shown in the previous setion, it is lear that in order to loalize the points of innite ondensation of we have to look at the paraboli points. Proeeding in exatly the same way as in Setion 6 of [22℄, we an prove the following. x6. Proposition 5.13. If ! 2 !) . only if h p2(!p()+1 n O+(Crit(f )), then is of innite ondensation at ! if and Corollary 5.14. If maxfqb : b 2 R \ f then the invariant measure is nite. 1 ) 2 p(! ) (1)g sup p(!) + 1 : ! 2 ; ( JANINA KOTUS AND MARIUSZ URBANSKI 52 !) , then has innite ondensation at ! . Proposition 5.15. If ! 2 and h p2(p!()+1 Theorem 5.16. If 2 J (f ) is a ritial point of f of order q , ! = f () 2 , and h then ! is of innite ondensation of measure . 6. 2qp(!) p(!)+1 , Appendix The goal of this appendix is to provide a proof of Theorem 2.6. We rst prove a version of Przytyki's lemma from [19℄ for the sake of ompleteness and then we prove a version of Mane's theorem from [13℄. We deided to provide a full proof of this theorem sine its original Mane's proof ontains some minor misprints and it would be very diÆult to indiate in whih plaes and in whih way one needs to modify it. Lemma 6.1. For every integer K 0 and every 0 < < 1 the following holds. For every > 0 and every > 0 there exists Æ0 = Æ0 (K; ; ; ) > 0 suh that for every disk B (x; Æ ) with Æ Æ0 and every x 2 CI in the distane at least apart from the set of paraboli points and attrating points, for every n 0 and every onneted omponent W = Compf n(B (x; Æ )) suh that fjnW has at most K ritial points ounted with multipliities, for every omponent W 0 = Comp(f n (B 0 )) in W , for the dis B 0 = B (x; Æ ) we have diamW 0 diamW 0 ! 0 for n ! 1 uniformly (i.e. independently of the hoies of B and W 0). 1 Proof. Suppose on the ontrary that there exist a sequene fxn gn=1 of points in the distane at least apart from the set of paraboli points and attrating points, a sequene Æn & 0, a sequene of omponents Wn = Compfk kn (Bn) with kn ! 1, as n ! 10 suh that the number of ritial points of eah map f n on Wn is bounded by K and Wn, 0 the sequene assoiate to Wn as in the statement of the lemma, suh that limn!1 diam(Wn) = 0.k Then for eah n there exists L = L(n) : 0 L K suh that there is no ritial value of fjWnn in L+1 L P (n) := B (xn ; Æn( + (1 ) )) n B (xn ; Æn ( + (1 ) )): K +1 K +1 Without loosing generality we may assume that all the omponents Wn0 interset the funda- mental region R. Put Wn(1) := Compf Wn(2) := Compf kn B kn B (xn; Æn( + (1 (xn ; Æn( + (1 L(n) ))) K+1 L(n) + 1 ))) ) K+1 ) GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS the omponents ontaining Wn0 , 53 Pn := Wn(2) n Wn(1) and for every 0 m kn, i = 1; 2, (i) = f kn m (W (i) ); P kn m (P ) = W (2) n W (1) : Wn;m n;m := f n n n;m n;m Let, for eah n, the number m = m(n) kn be the least integer suh that (1) min(; diamWn;m inf dist(1; 2)); ; 2Crit(f ); 6= g So for every 0 t < m(n) the set Pn;t is a topologial annulus. That is so beause at eah step bak by f 1 from Pn;t 1 to Pn;t there is at most one branh point for f 1 from (i) (i) Wn;t 1 to Wn;t ; i = 1; 2. Now, all the annuli Pn;m(n) 1 's have moduli bounded below by 2 K (1 ) K1+1 . Sine in addition all the omponents Wn0 interset the fundamental region <, it follows from Montel's Theorem that there exists a topologial (maybe not geometri) annulus P ontained in all Pn;m(n) 1 's for a subsequenes ns, whih bounds a topologial disk D. So D Wn(2)s ;m(ns ) 1 . Hene f m(ns ) 1 (D) B (x; Æn ). Passing to yet another subsequene we may assume that the sequene xn onverges to a point y 2 CI at the distane at least apart from the set of paraboli points and attrating points. Sine Æn ! 0, we have also m(n) ! 1. Thus D annot interset the Julia set J (f ). If the were ontained in a preimage of a Siegel disk or a Herman ring, the limit of diameters of iterate f m(ns ) 1 (D) would be positive. Thus D is either ontained in the basin of attration to an attrating periodi orbit or a paraboli periodi orbit. In either ase the limit of the sets f m(ns) 1 (D) would be ontained in either an an attrating periodi orbit or a paraboli periodi orbit. Sine this limit would oinide with y, we get a ontradition. The proof is omplete. 1 2 1 2 Remark 6.2. Obviously this lemma remains true (with the proof required only minor modiations) if instead of the disk B (x; Æ ) one takes the square entered at x and with edges of length Æ . This is the version we will need in the next theorem. Theorem 6.3. Let f : CI ! CI be an ellipti funtion. If a point x 2 J (f ) n (f ) is not ontained in the ! -limit set of a reurrent ritial point, then for all > 0 there exists a neighbourhood U of x suh that: (a) For all n 0, every onneted omponent of f n(U ) has diameter ; (b) There exists N > 0 suh that for all n 0 and every onneted omponent V of n f (U ), the degree of fjnV is N ; The ore of the theorem is (a), from whih the property (b) will easily follow. Given an open set U CI denote j (U; n) the set of onneted omponents of f n(U ). Observe that V 2 (U; n) implies f (V ) 2 (U; n j ) for all 0 j n. If V 2 (U; n) dene (V; n) = #fx 2 V ; (f n)0(x) = 0g ounted with algebrai multipliity. A square is the set S of the form S = fz 2 CI : j<(z p)j < Æ; j=(z p)j < Æ g. The point p is the enter and Æ Proof. JANINA KOTUS AND MARIUSZ URBANSKI 54 its radius. Given a square S with enter p and radius Æ, then, given k > 0, denote by S k the square with enter p and radius kÆ. If S is a square with radius Æ, denote by L(S ) the family of squares ontained in S S and having radius Æ=4. Denote by L (S ) the family of squares S and having radius Æ=4. Denote by L (S ) the family of squares S0 with S0 2 L(S ). S Suppose that x is not a paraboli point or is ontained in the !-limit set of reurrent ritial point. Then there exists Æ0 > 0 suh that (1) There is no ritial point of f suh that there exists 0 n1 n2 satisfying jf n () j < Æ0 and jf n () j < Æ0 (2) jx pj > 10Æ0 for every paraboli or attrating periodi point p. Given > 0 take 1 > 0 satisfying (3) 0 < 1 < minf=10; Æ0=10g (4) if U is an open onneted set with diam(U ) 21 then diam(W ) Æ0 for all W 2 (U; 1) Let N0 be the number of equivalene lasses of the relation between ritial points of f . Take N1 > 2 suh that (5) If S is a square and V 2 (S; n) satises (V; n) N0 + 1 then the number of onneted omponents of f n(S ) ontained in V is N1 . Finally, take Æ given by (6) Æ = minfÆ0=10; 1=10; Æ(2N0; 20N ; 23 ; Æ0)g where Æ(2N0; 20N ; 32 ; Æ0) is given by Lemma 6.1. Let S0 be the square of enter x and radius Æ. Suppose that Theorem 6.3 fails for U = S0. Then there existsp n > 0 and V 2 (S0; n) with diamV 101. On the other hand, by (1), diamS0 = 2 2Æ < 3Æ < 1 : Hene there exists an integer n0 0 suh that there exists V0 2 (S0 ; n0 ) satisfying (7) diam(f n(n i)(S0 ) \ f i(V0)) 1 for all 1 i n0, and (8) diam(f (S0 ) \ V0 ) > 1 Sine diamS0 < 1 it follows that n0 > 0. Now, starting with S0 we shall onstrut a sequene of squares S0; S1; S2; : : : and stritly positive integers n0 n1 n2 : : : satisfying (9) Sj+1 2 L (Sj ) (10) there exists Vj 2 (Sj ; nj ) suh that diam(f ( nj i) (Sj ) \ f i(Vj )) 1 for all i i nj and diam(f nj (Sj ) \ Vj ) > 1 : 3 2 3 2 3 2 1 2 2 3 1 1 1 3 2 0 0 3 2 1 GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 55 From (7) and (8), it follows that S0 satises (10). If we onstrut suh a sequene of squares and integers, then Theorem 6.3 will be proved by ontradition beause the ondition n0 n1 n2 : : : nm : : : > 0 implies that nj = ni for all i for a ertain i. But (a) implies that the radius of Sj is ( 83 )j Æ; in partiular diam(Sj ) ! 0 when j ! +1. But by (10), 1 < diam(f nj (Sj ) \ Vj ) = diam(f ni (Sj ) \ Vj ); Vj 2 (Sj ; nj ) = (Sj ; ni ) Taking j ! +1; and realling that i is ontained and limj!1 diamSj = 0, we onlude that the inequality above annot hold. The sequene fSj g and fnj g will be onstruted by indution starting with S0. Suppose Si and ni onstruted for 0 i j . To nd Sj+1 and nj+1 we begin by observing that from (a) it follows that if p 2 S 2 L (Sj ), then, by +1 3 jX +1 3 j p jX p X i jp xj diam(Si) = ( 8 ) diam(S0 ) = 2 2 ( 8 )iÆ 4 2Æ: i=0 i=0 i=0 Hene, if a point q satises dist(q; S ) Æ0, we have p jq xj 4 2Æ + Æ0 2Æ0: By (2), this means that (11) dist(q; S ) > Æ0 for all S 2 L(Sj ) and all paraboli or attrating periodi point q. For the indution step (i.e. the onstrution of Sj+1 and nj+1), we shall use the following lemma. Lemma 6.4. If U CI is an open set and V 2 (U; n) satises diamf i(V ) Æ0 ; 0 i n 3 2 3 2 then (V; n) N0 : . If ( V; n) N0 + 1, there exists N0 + 1 dierent points xi , 1 i N0 + 1, in Vj n j suh that (fm )0(xj ) = 0. This means that for eah 1 i N0 + 1 there exist 1 mi < n, suh that f i (xi) is a ritial point. Realling that N0 is the number of the equivalene lasses of the equivalene relation , it follows that there exists two dierent points in the set fxi; 1 i N0 + 1g, that we shall denote by x1 ; x2 , and two ritial points 1 and 2 in the same equivalene lass of the equivalene relation , suh that f mi (x1 ) = 2 and f m (x2 ) = 2 . Assume without loss of generality that 0 m1 m2 . Then by the hoie of Æ0 , m1 < m2 and jf m m (1 ); 1)j = jf m m (2); 1)j = jf m (x1 ) f m (x2 )j diamf m (Vj ) Æ0 and jf n m (1); xj = jf n m (f m (x1 )) xj = jf n(x1 ) xj Æ0 ontraditing property (1) of Æ0. Proof 2 2 1 2 1 1 2 1 1 2 2 JANINA KOTUS AND MARIUSZ URBANSKI 56 Now, to nd Sj+1 and nj+1 we rst laim that there exists S 2 L(Sj ) that for some 0 < n nj has V 2 (S; n) with diamV 10N . Suppose that the laim is false. Then, for all 1 i nj ; diam(f i(Vj )) diam f (nj i) (Sj ) \ f i(Vj ) + supfdiam(W); W 2 (S; nj i); S 2 L(Sj)g 1 + 10N1 21 1 From this inequality applied to i = 1 and property (4), we have diam(Vj ) Æ0 Moreover, sine 21 Æ0 ( by (3)), diam(f i(Vj )) Æ0 for all 1 i nj , hene for all 0 i nj . By Lemma 6.4. This proves that (Vj ; nj ) N0. Then, sine Vj 2 (Sj ; nj ) it follows from (5), (11) and Lemma 6.4 that W 2 (Sj ; nj ); W Vj ! diam(W ) 1 10N1 Moreover, by the way N1 was hosen, we have #fW 2 (Sj ; nj ); W Vj g N1 and we are assuming that S 2 L(Sj ); U 2 (S; nj ) ! diam(U ) 10N1 : Now observe that Vj is the union of sets U 2 (S; nj ); U0 V00 j ; S 2 L(Sj ) and the sets W 2 (Sj ; nj ); W Vj . Moreover, for any two sets W , W in this family there exist W 0 = W0 ; W1 ; : : : ; Wk = W 00 in (Sj ; nj ) and ontained in Vj suh that for all 0 i < k there exists Si 2 L(Sj ) and Ui 2 (Si; nj ) suh that U i \ W i 6= ; U i \ W i+1 6= ;. Then 1 1 1 diam(Vj ) N1 10N + 10N = 5 1 1 ontraditing the last inequality in ondition (10). This ompletes the proof of the laim. Now we an take S 2 L(Sj ) suh that diam(V ) 10N for some V 2 (S; n); 0 n nj . Take V~ 2 (S ; n) ontaining V . Suppose that (V~ ; n) N0 . Then by Lemma6.1 and ondition (6) diam(V ) 20N1 1 sine V 2 ((S ) ; n) and is ontained in V~ . This ontradits the fat that diam(V ) 10N1 1 1 2 3 1 3 2 3 2 2 3 1 GEOMETRY AND ERGODIC THEORY OF NON-RECURRENT ELLIPTIC FUNCTIONS 57 and proves (V~ ; n) N0 + 1. From Lemma 6.4, it follows that diam(f i(V~ )) > Æ0 for some 0 i n. Now we dene Sj+1 = S . Then f i(V~ ) 2 (S ; n i) and diam(f i(V~ )) > Æ0 101 . Moreover diamSj +1 2Æ < 1 . Then there exists 0 nj +1 n i nj i and Vj +1 2 (Sj +1 ; nj +1) suh that diam(f nj (Sj+1) \ Vj+1) > 1 and diam(f nj +i(Sj+1) \ f i(Vj+1)) 1: Observe that nj+1 > 0 sine diam Sj+1 < 2Æ < 1 . This ompletes the onstrution of the sequene fSj g and fnj g and the proof of part (a) of Theorem 6.3. Property (b) of Theorem 6.3 follows from (a) and Lemma 6.4. Our destination in this appendix is the following. Proof of Theorem 2.6 If X is a ompat subset of the omplex plane CI the theorem immediately follows from Theorem 6.3 and ompatness of X . So suppose that X J (f ) n (f ) is a losed subset of CI. Let = dist( (f ); f 1(1)) > 0. In view of (2.2) and (2.4) 1 there exists R > 0 so large that if jf (z)j R=2, then for some b 2 f (1),z 2 Bb(R=2) jf 0(z)j 2 and diam(Bb (R=2)) =2: (6.1) Consider now the ompat set Y = X [ (CIn B ( (f ); =2)) n BR and the orresponding number 0 < Æ minf; R=2g asribed to Y and the number minf; R=2g. In order to omplete the proof nit suÆes to show that if x 2 BR , then the diameter of eah onneted omponent Cn(x) of f (B (x; Æ)) does not exeed for every n> k0. And indeed, x w 2 f n(x) \ Cn(x) and let 1 k n be the least integer suh that f (w) 2= BR provided it exists. Otherwise, set k = n. We shall show by mathematial indution that diam f n j (Cn(x)) Æ minf; R=2g (6.2) for every 0 j k. For j = 0 this formulanisjtrue sine f n(Cn(x)) = B (x; Æn).j So, suppose that it is true for some 0 j k 1. Sine f (w) 2 BR and sine diam (f (Cn(x))) R=2, we onlude that f n j (Cn(x)) BR=2 : (6.3) It therefore follows from the rst part of formula (6.1) that diam f n j+1j (Cn(x)) 21 diam f n j (Cn(x)) Æ: This proves formula (6.2). It follows from (6.3) and the seond part of formula (6.1) that f n k (Cn (x)) CI n B ( (f ); =2). Sine we also know that f n k (w) 2= BR , we onlude that f n k (w) 2 Y , we see that diam(Cn (x)) minf; R=2g . We are done. 3 2 3 2 3 2 +1 +1 3 2 58 JANINA KOTUS AND MARIUSZ URBANSKI Referenes [1℄ J. Aaronson, M. Denker, M. Urbanski, Ergodi theory for Markov bered systems and paraboli rational maps, Transations of A.M.S. 337 (1993), 495-548. [2℄ I. N. Baker, J.Kotus, Y. Lu, Iterates of meromorphi funtions I, Ergod. Th. and Dynam. Sys. 11 (1991), 241-248. [3℄ I. N. Baker, J.Kotus, Y. Lu, Iterates of meromorphi funtions IV, Results Math. 22 (1992), 651-656. [4℄ A.F. Beardon, Iteration of Rational Maps, Springer-Verlag, New York, 1991. [5℄ W. Bergweiler. Iteration of meromorphi funtions. Bull. Amer. Math. So., 29:2 (1993), 151-188. [6℄ M.Denker, M. Urbanski, On the existene of onformal measures, Trans. A.M.S. 328 (1991), 563-587. [7℄ M. Denker, M. Urbanski, On Sullivan's onformal measures for rational maps of the Riemann sphere, Nonlinearity 4 (1991), 365-384. [8℄ M. Denker, M. Urbanski, Geometri measures for paraboli rational maps, Ergod. Th. and Dynam. Sys. 12 (1992), 53-66. [9℄ M. Denker, M. Urbanski, The Capaity of paraboli Julia sets, Math. Zeitsh. 211, (1992), 73-86 [10℄ M. Guzman: Dierentiation of integrals in IRn . Let. Notes in Math. 481, Springer Verlag. [11℄ J. Hawkins, L. Koss, Ergodi properties and Julia sets of Weierstrass ellipti funtions, Preprint 2001. [12℄ J.Kotus, M. Urbanski, Hausdor dimension and Hausdor measures of Ellipti Funtions, to appear Bull. London Math. So. [13℄ R. Mane, On a theorem of Fatou, Bol. So. Bras. Mat., 24 (1933), 1-11. [14℄ R. Mane, The Hausdor dimension of invariant probabilities of rational maps, Dynamial Systems, Valparaiso 1986, Let. Notes in Math. 1331, Springer-Verlag (1988), 86-117. [15℄ M. Martens, The existene of -nite invariant measures,Appliations to real one-dimensional dynamis, Preprint. [16℄ R.L. Mauldin, M. Urbanski, Dimensions and measures in innite iterated funtion systems. Pro. London Math. So. 73:3 (1996), 105-154. [17℄ J. Milnor, Dynamis in one Complex Variable, Introdutory letures, Vieweg 2000. [18℄ F. Przytyki, Hausdor dimension of harmoni measure on the boundary of an attrative basin for a holomorphi map, Invent. Math. 80 (1985), 169-171. [19℄ F. Przytyki, Iterations of holomorphi Collet-Ekmann maps: onformal and invariant measures. Trans. AMS 350.2 (1998), 717-742. [20℄ P.J. Rippon, G.M. Stallard, Iteration of a lass of hyperboli meromorphi funtions, Pro. Amer. Math. So. 127 (1999), 3251-3258. [21℄ M. Urbanski, Rational funtions with no reurrent ritial points, Ergod. Th. and Dynam. Sys. 14 (1994), 391-414. [22℄ M. Urbanski, Geometry and ergodi theory of onformal non-reurrent dynamis, Ergod. Th. and Dynam. Sys. 17 (1997), 1449-1476. Janina Kotus, Faulty of Mathematis and Information Sienes Warsaw University of Tehnology Warsaw 00-661, Poland E-mail: janinakimpan.gov.pl ski, Department of Mathematis, University of North Texas, P.O. Box Mariusz Urban 311430, Denton, TX 76203-1430, USA E-mail:urbanskiunt.edu Web: http://www.math.unt.edu/urbanski

© Copyright 2021 DropDoc