376 F U N C T I O N S OF B O U N D E D C H A R A C T E R I S T I C A N D L I N D E L Ö F I A N MAPS By M A U R I C E H E I N S 1. Introduction The study of the boundary behavior of analytic functions in a modern sense begins with the fundamental and justly celebrated memoir of FatouE2]. An essential feature of that paper is the exploitation to the fullest of the then novel ideas of Lebesgue. The half century which has elapsed since Fatou's memoir has witnessed the flowering of a large and important chapter of analysis whose concern is the boundary behavior of analytic functions. Our attention will be centered on questions whose parentage may be traced back to the results of Fatou. Let us recall that the original theorem of Fatou states that a bounded analytic function/in the open unit disk possesses a radial limit p.p. and that the theorem of F. and M. Riesz[13] states that the radial limit function of/ vanishes only on a set of measure zero, if/ =£ 0. The functions of bounded Nevanlinna characteristic in the open unit disk are precisely the non-constant meromorphic functions which admit representation as quotients of bounded analytic functions. Thanks to the original Fatou theorem and the theorem of F. and M. Riesz, the conclusion of the Fatou theorem persists for functions of bounded characteristic. When we turn to the study of asymptotic values, wefindthat a marked contrast appears. For a bounded non-constant analytic function in the open unit disk, each asymptotic path tends to a point of the unit circumference and the associated asymptotic value is the Fatou radial limit at this point. On the other hand, while it is still true for a function of bounded characteristic that each asymptotic path tends to a point of the unit circumference, there exist, as examples of LohwaterC9] and Lehto[8] show, functions of bounded characteristic having more than one asymptotic value associated with a given boundary point. A recently announced theorem of GehringE4] asserts that an analytic (pole-free) function of bounded characteristic in the open unit disk admits at most two distinct finite asymptotic values associated with a given boundary point. We shall see later that the situation changes radically for unrestricted functions of bounded characteristic. In fact, there exists a quotient of two Blaschke products whose zeros cluster solely at z = 1 which has the property that the set of asymptotic values associated with F U N C T I O N S OF B O U N D E D C H A R A C T E R I S T I C 377 z = 1 has the power of the continuum.f In the construction which we give of such a function, a decisive role is played by Valiron's example[15] of a meromorphic function of finite order whose set of asymptotic values has the power of the continuum. However, the realm of pathology is small. In fact, for an arbitrary non-constant meromorphic function/ in {\z\ < 1}, the set of points of {|z| = 1} with which more than one asymptotic spot[5] is associated is countable. Examples of functions of bounded characteristic where such sets are infinite are readily constructed. We understand that a point TJ of {|z| = 1} is associated with an asymptotic spot cr of / provided that {^} = Ç]cr(ù)), a) € domain of cr. The notion of a function of bounded characteristic may be generalized to conformai maps of Riemann surfaces [6>12]. The Nevanlinna characteristic function maybe adapted to conformai maps of Riemann surfaces in such a manner that the first fundamental theorem and the basic theorems of the Nevanlinna theory associated with it persist in the general theory of conformai maps of Riemann surfaces. The notion of a map of bounded characteristic becomes an important element of the extended theory. In the present paper we shall confine our attention to Lindelöfian conformai maps ( = maps of bounded characteristic) whose domain is the open unit disk (meromorphic functions of bounded characteristic are included) and shall apply the methods developed in our papers C5] and [6] tö problems concerning the boundary behavior of such maps and the relation between their boundary behavior and their 'covering properties'4 By confining our attention to Lindelöfian maps with domain the open unit disk we put at our disposal an extensive apparatus from real function theory. This fact permits us to expect more precise results than one could for arbitrary Lindelöfian maps. 2. Non-negative harmonic functions Parreau[11] has introduced the notions of a quasi-bounded non-negative harmonic function and of a singular non-negative harmonic function. f The author is indebted to Professor F . W. Gehring for pointing out the fact t h a t he carried out a construction along the lines of the second paragraph of § 10 of the present paper (F. W. Gehring, The asymptotic values for analytic functions with bounded characteristic, Oxford Quart. J. Math. 1958). His construction shows the existence of a function / of bounded characteristic in {\z\ < 1} having the property t h a t the set of asymptotic values of/ associated with paths terminating at z = 1 has the power of the continuum. However, he does not establish the more refined result to which the present footnote refers. J For the case of meromorphic functions of bounded characteristic, cf. Lehto [7»83. I am indebted to Professor Pfluger for raising the question of extending the Fatou theorem to Lindelöfian maps with domain the open unit disk. 378 MAURICE HEINS These may be defined as follows. A non-negative harmonic function u on a Riemann surface F is termed quasi-bounded provided t h a t it is the limit of a monotone non-decreasing sequence of bounded non-negative harmonic functions on F; u is termed singular provided t h a t the only non-negative bounded harmonic function on F which is dominated by u is zero.f A non-negative harmonic function on F admits a unique representation as a sum of a quasi-bounded non-negative harmonic function on F and a singular non-negative harmonic function on F. A positive harmonic function u on F is termed minimal (Martin [10] ) provided t h a t the positive harmonic functions on F dominated by u are proportional to u. If a minimal positive harmonic function is not bounded, it is singular. fi-maps. Let D denote a non-empty open subset of a Riemann surface F which has the property t h a t each point of fr O is a point of a continuum contained in fr O. Given a non-negative harmonic function u on O which vanishes continuously on fr Q., by /IQ(U) is meant the least harmonic majorant of the subharmonic function on F which agrees with u on CI and vanishes elsewhere on F, provided t h a t the subharmonic functionin question admits a harmonic majorant; otherwise /iQ(u) = +00. Let Qa denote the set of u for which /IQ(U) < + 00. The restriction of fin to Qa is a univalent homogeneous additive map of Qa into the set of non-negative harmonic functions on F. Further u (eQo) is quasibounded (singular) on O if and only if /ia (u) is quasi-bounded (singular) on F. If O is a region, then u (€ Qa) is minimal on D if and only if [içi(u) is minimal on F. By a /£-map we shall understand the restriction of /IQ to Qa for some admitted Q. 3 . T h e Lindelöf p r i n c i p l e Let 0 denote a conformai map (not necessarily univalent) of a Riemann surface F into a Riemann surface G. Let n(p; <fi) denote the multiplicity of <ß&tp € F. Let v^(q) denote the valence of (j) at q e G,i.e.H^p)=Qn(p; ç5). Suppose now t h a t F and G are hyperbolic and t h a t ®F and &G are their respective Green's functions. Let S(p, q) = S ^ ^ n f o ci) &F(p, r). The exact form of the Lindelöf principle asserts t h a t for each qeG, ®G^(P),q) = S(p,q) + ua(p) (peF), (3.1) where uq is a non-negative harmonic function on F. Obviously uq is f These concepts are also pertinent to non-negative harmonic functions whose domains are non-empty open subsets of a Riemann surface. FUNCTIONS OF BOUNDED CHARACTERISTIC 379 unique. We note that uq is the greatest harmonic minorant (abbreviated henceforth by 'G.H.M.') of the superharmonic function £> -> &a(^>(p), q). Let vq denote the quasi-bounded component of uq and let wq denote the singular component of uq. We have the alternatives: vq = 0 for all q e G, or vq > 0 for all q e G. In the first case we say that ci is a map of type-Bl of F into G, the designation 'Bl' being employed because of the close relation of such maps with Blaschke products. If uq = 0 for all q e G, we say that ci is a map of type-Blx of F into G. We observe that regardless of whether ci is of type-Bl or not, wq = 0 save for an Fff of capacity zero[5]. Suppose that F and G are now unrestricted. We say that ci is of type-Bl (Blt) at q e G provided that there exists a simply connected Jordan region ù, qe Q, c: G, such that çi-1(£2) 4= 0 and the restriction of ci to each component o) of ^ _ 1 (0) is a map of type-Bl (Bl^ of a> into O. 4. Lindelöfian maps A conformai map (p of a hyperbolic Riemann surface J7 into a hyperbolic Riemann surface G satisfies S(p,q) <+co tf(p)*q). (4.1) Interest therefore attaches to the study of conformai maps ci: F ->G, where F is hyperbolic but G is unrestricted, which satisfy (4.1). (It is to be noted that in the definition of S(p, q) we need not assume that G is hyperbolic.) Because of the genesis of this condition we term such maps Lindelöfian. In terms of the extension of the Nevanlinna theory to conformai maps of Riemann surfaces the Lindelöfian maps are precisely the conformai maps of bounded characteristic. 5. Given distinct points a,b eG, there exists a function u on G which is harmonic save at a and b, which has a normalized positive logarithmic singularity at a and a normalized negative logarithmic singularity at b, and which is bounded in the complement of some relatively compact neighborhood of {a, b}. (If G is parabolic, then u is determined up to an additive constant.) Given a Lindelöfian map <fi: F->G, it may be concluded[6] from the extended form of Nevanlinna's first fundamental theorem that uocj) admits a representation of the form uo<f>(p) = S(p,a)-S(p,b) + H(p), (5.1) where H is the difference of two non-negative harmonic functions on F. We shall see that this representation is the basis of the boundary 380 MAURICE HEINS behavior theorems of the present paper. It is to be observed that H admits a unique representation of the form P — N, where P and N are nonnegative harmonic functions on F satisfying G.H.M. min {P,iV} = 0. There is an intimate connection between the harmonic function P and the way in which ci covers a neighborhood of a. In fact, let O0 = {u > 0}, let (oQ = çi_1(O0), and let çiWo denote the restriction of ci to o)Q. If co0 4= 0, then, thanks to (5-1) and simple facts concerning greatest harmonic minorants of superharmonic functions, we are led to /n TT -MT JL \ • T> /* o\ (5.2) Äo (G.H.M. u o & J = P. If o)0 = 0, by convention we take the left-hand side of (5.2) to be zero; it is readily verified that in this case we also have P = 0. It is now easy to conclude that, if ß is an arbitrary Jordan region of G which contains a, a) = çi-^O), and ® a is the Green's function of D, then ^{G.H.M.® Q (çi w ,a)}-P = 0(1). (5.3) Here ^ refers to the restriction of ci to a) and the obvious convention prevails when o) = 0. From (5.3) we conclude that the singular component of the first member of the left side of (5.3) is independent of £1 and is simply the singular component of P. We denote it by Wa. A situation in which (5.2) may be exploited advantageously is the following. Suppose that O0 is connected and that the restriction of u to Q0 is the Green's function of Q0 with pole at a. (This is certainly the case if G is parabolic.) Suppose further that o)0 4= 0. Then P is singular if and only if the restriction of ci to each component of o)Q is a map of typeBl of that component into O0. If P is singular, then by (5.2) G.H.M. u o^Wo is singular, and ^ is readily seen to have the stated property. The converse is also readily established on noting that a /*-map carries a singular non-negative harmonic function into a singular nonnegative harmonic function. To give a specific application of this result, we consider a function/of bounded characteristic in {\z\ < 1} whose Fatou radial limits are p.p. of modulus one. If/is bounded, a known theorem of Frostman[3] asserts that for all a, |a| < 1, save for an Fa of zero capacity, is a Blaschke product up to a constant factor of modulus one.f What we shall now see is that, if neither / nor l//is bounded, then (5.4) is a f The property of wq cited in § 3 constitutes a generalization of Frostman's theorem. FUNCTIONS OF BOUNDED CHARACTERISTIC 381 quotient of two Blaschke products up to a constant factor of modulus one, for alia, |a| < 1, save for an F^ of zero capacity. We remark that the case where l//is bounded is readily reduced to the case where/is bounded. To establish our assertion we note that with a = 0, b = oo, u(z) = —log \z\, P and N are both singular by virtue of the hypotheses on /. Hence the restriction of / to a component of / - 1 {M < *} ( o r /""KM > 1}) *s °^ type-Bl relative to {|^| < 1} (or {|^| > 1} respectively). It now follows from the property of wq cited in § 3 and (5.2) that save for an exceptional set of oc of the described type, the P and N associated with log 1-5/ /-* vanish. The assertion is readily established. This result admits generalizations to conformai maps of Riemann surfaces. We shall not pursue this question further here. 6. The Fatou theorem We shall now see that the Fatou theorem persists for Lindelöfian maps <fi: F -> G, F = {\z\ < ï}, in the sense that for almost all £ on the unit circumference, either ci tends to a point of G as z tends to £ sectorially, or else ci tends to the ideal boundary of G as z tends to £ sectorially. We consider u o <p of § 5 having fixed distinct points a,b eG. There exists a meromorphic function / i n {| z | < 1} satisfying log\f\=uo<f>. (6.1) By (5.1)/is of bounded characteristic, so that the Fatou theorem holds for/. Suppose that / h a s a sectorial limit at TJ, \ TJ | = 1. Suppose that for some S, 0 < S < \TT, it is not the case that as z tends to 7] in ^ = {|arg(l-^)|<£}, <f)(z) tends to a point of G or to the ideal boundary. There would then exist a point qeG,a, relatively compact open disk A of G containing q, and sequences (z'n) and (z"n) whose members he in Ss and which satisfy not only _. , .. „ h.mzn = iimzn = 7], but also and çHOeA, çi(4)€0-A lim $(z'n) = q. (aU n), 382 MAURICE HEINS Let g denote a meromorphic function in A for which log |gr| is the restriction of u to A. On considering / on the segments z'nz,rn we are led to conclude that g is constant. This is of course impossible. The extended Fatou theorem follows. 7. Criteria for sectorial and quasi-sectorial limits We now turn to the examination of sufficient conditions for a point G to be a sectorial (or a quasi-sectorialf ) limit of <fi. (a) Suppose that (j) has a logarithmic ramification over a eG. Then a is a sectorial limit of <fi. More formally, we suppose that the following conditions are fulfilled: (1) for each simply connected Jordan region Q, of G which contains a, ^~1(D) 4= 0, (2) there exists a function cr whose domain consists of the set of such O and which satisfies (i) or(O) is a component of çi _1 (n), (ii) Ox c Q2 implies cr^-J <= <r(Q2)> (iii) for Q, sufficiently small, (ß^Q), the restriction of ci to cr(£l), is a universal covering of Q — {a}. Thanks to these conditions, it follows that, for Ù small, ©^(Cì^Q), a) is a minimal positive harmonic function on o"(Q). From (5.3) it may be concluded that /V(Q)[@W?W «)] < + oo. (7.1) For Q. sufficiently small, the left side of (7.1) is a minimal positive harmonic function in {j z | < 1} independent of £1, say c« [J=i] <°>°.M-i>. It follows from the Julia-Carathéodory theorem that for given S, 0 < S < \TT, and for each Q, the points of the sector Ss which are sufficiently close to 7] he in cr(O). We infer that a is the sectorial limit of ci at TJ. The example f(z) = exp Jexp \rz^j\ (N < X)> shows that there exists an analytic function in {\z\ < 1} which possesses a sectorial limit save at 1 and has infinitely many logarithmic ramifications over 0 and yet does not possess the sectorial limit 0. This shows that the Lindelöfian character of ci is pertinent to the question considered. f This notion will be denned below in the present section. FUNCTIONS OF BOUNDED CHARACTERISTIC 383 Quasi-sectorial limit. We shall say that $ has the quasi-sectorial limit a ( e G) at TJ provided that lim4>[r](l-rei0)] = a (\6\ < In), (7.2) save for a set of 6 of zero capacity. We have: (b)V ITa(2)>c8t[|±£], where c is a positive constant, then ci possesses the quasi-sectorial limit aatrj. In fact, it follows from (5.1) that «0(3» > c^^-{S{z,b) + N{z)). Now N does not dominate a positive multiple of ?H[(T] + Z)I(TJ — Z)] in {\z\ < 1}. It follows from the theorem of Ahlfors and Heins[1] that ]miu[(^(T] — T]reie)] = +oo, |0| < \TT, r->0 save for a set of 6 of zero capacity. Hence (j) has the quasi-sectorial limit a at T). Examples of this phenomenon are readily constructed. We cite the example given by LehtoC8] /(Z) = exp{i±|J6(Z) (H<1), (7.3) where 6 is a convergent Blaschke product whose zeros are real and cluster solely at z = 1. It is to be observed that the existence of a quasi-sectorial limit of a Lindelöfian map which is not actually a sectorial limit imposes very severe restrictions. In fact, if (7.2) holds even p.p. for \6\ < \n, but a is not the sectorial limit of ci at TJ, then G is conformally equivalent to the extended plane, and for each b ( 4= a) e G, for some sector Sd = {T,(l-reM)\\d\ <£<|7r}, ^_1({6}) n Ss clusters at TJ. 8. What information can be drawn from the non-vanishing of Waì We shall see: (i) / / G is compact, Wa > 0, and v^(b) < +oo for some b(e G) 4= a, then a is a radial limit of (j). 384 MAURICE HEINS (ii) / / G is not compact and Wa > 0, then a is a radial limit of çi.f A special instance of (i) is to be found in Lehto[8]. We shall revert to this result of Lehto below. Apart from the use of auxiliary harmonic functions on G with assigned singularities and boundary behavior, and the study of the composition of such functions with (j), the proof rests principally on the important extension due to Saks[14] of the de la Vallée Poussin decomposition theorem to the case of an unrestricted function of bounded variation of one real variable. Assertion (i) may be established as follows. We start with the representation (5.1) and note that our hypothesis on 6 renders S(z,b) innocuous near the unit circumference. Let ^jr*^]^ be a Poisson-Stieltjes representation for H(z), where ju, is a function of bounded variation on each finite interval and satisfies /t(0) = i M 0 + )+/t(0-)], /i(d + 27T) = /i(6) + [fi(27r)-/i(0)]. • We assert that fi'(d) = 4-00 for some 6. If this were not the case, then by the de la Vallée Poussin-Saks decomposition theorem, we would be led to conclude that H is of the form Q — TJ, where Q is a quasi-bounded nonnegative harmonic function and TJ is a non-negative harmonic function in {\z\ < 1}. But Wa ^ P < Q so that Wa = 0. This is impossible. Hence for some TJ, \TJ\ = 1, iimH(rT)) = +00. We conclude that limçi(n/) = a. r->l r->l Part (ii) is similarly treated; the non-compactness of G serves as a substitute for the finite valence hypothesis. The only change that is called for is to replace u by a harmonic function onff-{a} which has a normalized positive logarithmic singularity at a and is bounded above outside of each neighborhood of a, and to note that (5.1) persists save for the absence of — S(p, b). The theorem of Lehto to which reference has been made asserts that a meromorphic function of bounded characteristic in {\z\ < 1}, whose image is dense in the extended plane, omits at most one value outside of the closure of the set of its radial limits. An examination of the proof given by Lehto shows that it suffices, as far as Lehto's theorem is f Can * radial ' be replaced by ' sectorial' ? The fact t h a t we are concerned with PoissonStieltjes integrals a t points where the generating function has an infinite derivative renders the 'radial' easy; however, the argument used for the existence of a sectorial limit of a Poisson-Stieltjes integral a t a point where the generating function has a finite derivative does not appear to be available in the infinite case. FUNCTIONS OF BOUNDED CHARACTERISTIC 385 concerned, to appeal to the original de la Vallée Poussin decomposition for continuous functions of bounded variation rather than to the general decomposition theorem. The result of Lehto may be extended to Lindelöfian maps with domain {\z\ < 1} and may be given $ slightly more refined formulation. Let R denote the closure of the set of radial Kmits(€Ö) of ci. By(5.2)ifge G — R, then either q belongs to the complement of the closure of the image of (j) or else ci is of type-Bl at q. Let B denote the set of points in G — R at which (j) is of type-Bl. / / G is not compact, then $ is of type-Bl± at each point of B. This is a consequence of (ii) of the present section. Thus, in particular, on each component of B, v^ is constant (finite or not). If G is compact, two possibilities may occur. The first m that v$ is identically infinite on G. There are examples of such ^ which are not of type-Blx at several points of B (e.g./(z)//( — z) for/of (7.3); thisfunction is not of type-Bli at 0 and oo which are both points of B). The second is that for some point q0 e G, we have ^(g 0 ) <• + oo. Thanks to (i) of the present section, it follows that if qQeG — B, then <p is of type-Blx at each point of B; if q0e B, then $ is of type-Blx at each point of B — {q0} and in this case if <f> is not of type-Bl± at q0, v^ is infinite at each point ofG — {q0). For this latter phenomenon cf. (7.3). 9. From this point on we shall be concerned with applying the methods described in this paper to problems concerning functions of bounded characteristic. We first consider a refinement of the theorem of Gehring which we quoted earlier. We note that an asymptotic spot[5i of a function of bounded characteristic in {\z\ < 1} has the property that precisely one point of {\z\ = 1} adheres to all cr(o)), co e domain of or. Let A(TJ) denote the number of distinct asymptotic spots cr off over finite points which have the property that {TJ} = D or(a>); let B(TJ) similarly denote the number of distinct asymptotic spots cr of / over oo which have the property that {TJ} = PI cr(o)). The following theorem holds: For an analytic function f of bounded characteristic in {\z\ < 1}, A(TJ) < 2 and B(TJ) < 3, \TJ\ = 1. There exists an f for which A(l) = 2 and J5(l) = 3. We note that, if A(TJ) ^ 3, then thanks to the classical results of Lindelöf which are employed in the proof of the Denjoy-CarlemanAhlfors theorem and to (5.2), u(z) being taken as log \z\, we would infer 25 TP 386 MAURICE HEINS the existence of two distinct asymptotic spots cr1 and cr2 over oo satisfying {TJ} = fi crk(o)), k = 1,2, and such that for some a) in their common domain not only is it the case that o"x(w) n cr2(o)) == 0, but also where ck> 0, ß = 1,2. This is impossible since /e-maps carry two admitted positive harmonic functions with disjoint domains into two positive harmonic functions the minimum of which has a vanishing greatest harmonic minorant. Hence A(TJ) < 2. The proof that B(TJ) < 3 is similar; it differs only in preliminary details. To construct an/with -4(1) = 2, B(l) = 3, we may proceed as follows. Let F(z) = t^siatdt, gz > — 1. Now the positive and negative real Jo axes are asymptotic paths of F and the corresponding asymptotic values are finite and distinct; the positive imaginary axis is an asymptotic path of F and the associated asymptotic value is oo. The function g{z) = F ( ^ (\Z\<1) is of bounded characteristic. Next we observe that there exists a function of bounded characteristic in {\z\ < 1}, say h, which has the following two properties: (1) B(l) = 2, (2) h tends to zero as z -> 1 in A = {|z —2 -1 | < 2 -1 }. Such a function h may be constructed as follows. We observe that a positive harmonic function p in {\z\ < 1} which does not dominate c8i[(l + z)/(l —z)] for any positive constant c may be approximated uniformly on the intersection of A with the open unit disk by a function of the form — log \b\, where 6 is a Blaschke product whose zeros cluster solely at z = 1. If we choose p tending to infinity as z -> 1 (in the open unit disk), let w denote an analytic function satisfying log \w\ = \p, and let b be as above with p 4-log |6| = 0(1) on the intersection of A with the open unit disk, then wb is an admissible h. It suffices now to t a k e / = g+wb. 10. The striking contrast between the analytic and meromorphic cases, in so far as asymptotic values of functions of bounded characteristic are concerned, is brought out by the following example of a function f of bounded characteristic in {\z\ < 1} which is the quotient of two Blaschke products whose zeros cluster solely at 1, and which has the property that the FUNCTIONS OF BOUNDED CHARACTERISTIC 387 set of asymptotic values of f associated with z = 1 has the power of the continuum. The construction is based on Vahron's exampleC15] of an even meromorphic function g in the finite plane, which satisfies T(r; g) = 0(r), and has the property that there is a non-countable subset (£ of [0, n] such that, for 6 e @, lim g(reie) exists and is finite, and 6 -> Mm g(reie) is unir~»»oo valent on @. Let a,ß e @, where a < ß < n and a condensation point of (£ lies in (à, ß), and let h denote the restriction of g to {a < arg z < /?}. Thanks to the fact that T(r; g) = 0(r), A is a Lindelöfian meromorphic function. There exists a finite point w distinct from iim g(reict) and limg(re^) which is a condensation point of {iimg(rei6)\a < d < ß, d e $}. It follows that there exists an open circular disk A centered at w such that one of the components £1 of A-^A) has the following properties: (1) every finite frontier point of £1 Mes in {a < arg z < ß) and fr £l is regular analytic at each such point, (2) the set of asymptotic values in A of h restricted to £i is not countable (and consequently has the power of the continuum since it is an analytic set). Let p denote the radius of A. Let T denote the component of fr O (relative to the finite plane) which separates 0 from O, and let Q* denote the component of the complement of T with respect to the finite plane which contains £1. Then Ü* is simply connected and its frontier (relative to the extended plane) is a closed Jordan curve. Let $ denote a univalent conformai map of {|z| < 1} onto O* which 'carries' 1 into oo. We introduce fi = P~1\go(p-w] and note that/! is of bounded characteristic and further has the property that it possesses a limit of modulus one at each point of {|z| = 1} other than 1. Further, the set of distinct asymptotic values of f± associated with z = 1 has the power of the continuum. It follows from § 5 that there exist complex numbers oc, e, \a\ < 1, |e| = 1 such that is a quotient of Blaschke products. Clearly / satisfies all the imposed requirements. 25-2 388 MAURICE HEINS REFERENCES [1] Ahlfors, L. and Heins, M. Questions of regularity connected with the Phragmen-Lindelöf principle. Ann. Math. (2), 50, 341-346 (1949). [2] Fatou, P . Séries trigonométriques et séries de Taylor. Acta Math. 30,335-400 (1906). [3] Frostman, O. Potentiel d'équilibre et capacité des ensembles. Thesis, Lund, 1935. [4] Gehring, F . W. Asymptotic values for analytic functions of bounded characteristic. Bull. Amer. math. Soc. 63, 368, abstract 6841 (1957). [5] Heins, M. On the Lindelöf principle. Ann. Math. (2), 61, 440-471 (1955). [6] Heins, M. Lindelöfian maps. Ann. Math. (2), 62, 418-446 (1955). [7] Lehto, O. Value distribution and boundary behaviour of a function of bounded characteristic and the Riemann surface of its inverse function. Ann. Acad. Sei. Fenn. (Ser. A.I.), no. 177 (1954). [8] Lehto, O. Boundary theorems for analytic functions. Ann. Acad. Sci. Fenn. (Ser. A), no. 196 (1955). [9] Lohwater, A. J . The boundary values of a class of meromorphic functions. Duke Math. J. 19, 243-252 (1952). [10] Martin, R. S. Minimal positive harmonic functions. Trans. Amer. math. Soc. 49, 137-172 (1941). [11] Parreau, M. Sur les moyennes des fonctions harmoniques et analytiques et la classification des surfaces de Riemann. Thesis, Paris, 1952. [12] Parreau, M. Fonction caractéristique d'une application conforme. Ann. Fac. Sci. Toulouse, (4), 19, 175-189 (1956). [13] Riesz, F . and M. Ueber die Randwerte analytischer Funktionen. Fourth Scandinavian Math. Congress. Stockholm, 1916. [14] Saks, S. Theory of the Integral. Warsaw-Lwów, 1937. [15] Valiron, G. Sur les valeurs asymptotiques de quelques fonctions méromorphes. B.C. Cire. mat. Palermo, 49, 1-7 (1925).