STO¨ILOW’S THEOREM REVISITED RAMI LUISTOAND PEKKA PANKKA Abstract. Sto¨ılow’stheoremfrom1928statesthatacontinuous,light, and open mapping between surfaces is a discrete map with a discrete 7 branch set. This result implies that such mappings between orientable 1 surfaces are locally modelled by power mappings z 7→ zk and admit a 0 holomorphic factorization. 2 Thepurposeofthisexpositoryarticleistogiveaproofofthisclassical n theoremhavingthereadersinterestedindiscreteandopenmappingsin a mind. J 0 2 ] 1. Introduction V C Sto¨ılow’s classical theorem in [15] states that a light and open continuous . mapping between surfaces is a discrete map which has a discrete branch set. h t In what follows, we call this theorem as Sto¨ılow’s discreteness theorem. a Recall that a continuous mapping f: X → Y between topological spaces m is light if the pre-image f−1(y) of each point y ∈ Y is totally disconnected, [ and discrete if f−1(y) is a discrete set. A continuous mapping is open if the 1 image of each open set is an open set. The branch set B of a continuous v f mapping f: X → Y is the set of points x ∈X at which f fails to be a local 6 2 homeomorphism. 7 In [15] Sto¨ılow shows that these mappings are locally modelled by power 5 maps z 7→ zk [15, p.372]. This local description indicates a deep connec- 0 . tion between light and open mappings and holomorphic mappings between 1 surfaces. This connection was coined by Sto¨ılow [16, p.120] and Whyburn 0 7 [18, Theorem X.5.1, p.198] and [19, p.103]: For a light and open continuous 1 mapping f: S → S′ between orientable Riemann surfaces there exists a Rie- v: mann surface S˜ and a homeomorphism h: S → S˜ for which f◦h−1: S˜ → S′ Xi is a holomorphic map; the Riemann surface S˜ in this statement is naturally the Riemann surface associated to the map f. The first edition of [19], pub- r a lished in 1956, does not give this result a specific name, but already in the second edition from 1964 the result is referred as Sto¨ılow’s theorem. In this expository article we discuss the proof of Sto¨ılow’s discreteness theorem having readers interested in discrete and open mappings, such as quasiregularmappings(seee.g. [14])orThurstonmaps(seee.g. [4]),inmind. For this reason we separate Sto¨ılow’s theorem into two parts: the discrete- ness of the map and the discreteness of the branch set. Date: January 23, 2017. 2010 Mathematics Subject Classification. 30-02. P.P. has been partially supported by the Academy of Finland projects #256228 and #297258. 1 2 RAMILUISTOANDPEKKAPANKKA Theorem 1.1 (Sto¨ılow (1928)). Let Ω be a domain in C and f: Ω → C a light, open, and continuous mapping. Then f is a discrete map. Theorem 1.2 (Sto¨ılow (1928)). Let Ω be a domain in C and f: Ω → C a discrete, open, and continuous mapping. Then B is a discrete set. f It is interesting to notice that both results stem from path-lifting ar- guments. Indeed, after path-lifting results are establishes, standard appli- cations of the Jordan curve theorem together yield Theorems 1.1 and 1.2 almost as corollaries. For discrete and open mappings, we may use Rick- man’s path-lifting theorem [13] and for light and open mappings method of Floyd [6]. Floyd’s method suffices for all our purposes and we recall it in Section 3; see [10] for a more detailed discussion on path-lifting methods. Having Theorems 1.1 and 1.2 at our disposal, it is a straightforward cov- ering space argument to show that light open mappings between surfaces are locally modelled by power maps. Theorem 1.3 (Sto¨ılow (1928)). Let f: Σ → Σ′ a continuous, light, and open map between surfaces. For each z ∈ Σ, there exists k ∈ N, a neighbor- hood U of z, and homeomorphisms ψ: U → D and φ: fU → D for which the diagram U f|U // fU ψ φ (cid:15)(cid:15) (cid:15)(cid:15) D z7→zk // D commutes. This local version of Sto¨ılow’s discreteness theorem yields the global fac- torization theorem. Indeed, by Theorem 1.3, a light and open mapping f: Σ → S′ induces a conformal structure on Σ making it a Riemann surface S˜; we refer to [7, Section 1.1] or [4, Lemma A.10] for a short proof. Thus we may consider f as a holomorphic map f: S˜ → S′. If the surface Σ a priori carries a conformal structure, and we consider Σ as a Riemann surface S, we may take the homemorphism h in the factorization to be the identity homeomorphism Σ → Σ. Theorem (Sto¨ılow’s factorization theorem (1938)). Let f: S → S′ be a light, open, and continous mapping between Riemann surfaces. Then there exists a Riemann surface S˜ and a homeomorphism h: S → S˜ such that f ◦h−1: S˜ → S′ is a holomorphic map. Recallthatanorientabletopologicalsurfacecarriesaconformalstructure. Indeed,byaclassicaltheoremofRad´o,every 2-manifoldcanbetriangulated (see e.g. [11, Theorem 8.3, p. 60] or [1, Section II.8, p.105]) and every triangulatedorientablesurfacehasaconformalstructure(seee.g.[1,II.2.5E, Theorem, p.127] or [3, Section 2.2, pp. 9-11]). In this way we recover the interpretation that a light open mapping between orientable surfaces is a holomorphic map between Riemann surfaces. As authors’ interest to these theorems of Sto¨ılow stems from their role in the theory of quasiregular mappings, we finish this introduction with a STO¨ILOW’S THEOREM REVISITED 3 related remark. In the quasiconformal literature Sto¨ılow’s theorem typicaly refers to the result that each quasiregular mapping S → S′ between Rie- mann surfaces factors to a holomorphic map S → S′ and a quasiconformal homeomorphism S → S. Proof of this result is analytic in its nature and based on the Beltrami equation. We refer to Astala, Iwaniec, and Martin [2, Theorem 5.5.1] for a detailed discussion. Acknowledgements. Authors would like to thank Mario Bonk for his encouragement to write this expository article and for several helpful re- marks. 2. Preliminaries A totally disconnected closed set in the plane has topological dimension zero. We recall this fact in the following form; see [9, Section II.2, p.14] for a proof. Fact 2.1. Let C be a closed and totally disconnected set in C. Then ev- ery point x ∈ C has a neighbourhood basis consisting of neighbourhoods U satisfying C ∩∂U = ∅. Let Ω be a planar domain and f: Ω → C be a continuous, light, and open mapping. A domain U ⊂ Ω is a normal domain if U is compactly contained in Ω and ∂fU = f∂U. A normal domain U is a normal domain of x if it is also a neighbourhood of x. We call U a normal neighbourhood of x, if U ∩ f−1{f(x)} = {x}. Note that for a normal domain U, the restriction f| : U → fU is a proper map, i.e. the pre-image f−1K of a compact set U K ⊂ fU is compact. For each x ∈ Ω and r > 0, we denote U(x,f,r) the component of f−1B(f(x),r) containing the point x. The following lemma shows that do- mains U(x,f,r) are normal domains of x for all r > 0 small enough; cf.[17, Lemma 5.1]. Lemma 2.2. Let Ω be a planar domain, f: Ω → C, a continuous, light, and open mapping, and let x ∈ Ω. Then there exists a radius r > 0 such x that for all r ≤ r the domain U(x,f,r) is a precompact normal domain of x x and fU(x,f,r)= B(f(x),r). Proof. Let x ∈ Ω. By Fact 2.1 there exists a precompact neighbourhood V of x for which ∂V ∩ f−1{f(x)} = ∅. Since ∂V is compact and does not contain pre-images of x, we may fix r > 0 for which B(f(x),r)∩f∂V = ∅. We set U :=U(x,f,r) and observe first that U is precompact. Indeed, f(U ∩∂V)⊂ fU ∩f∂V ⊂ B(f(x),r)∩f∂V = ∅. Thus U ∩∂V = ∅ and U ⊂ V by connectedness, so U ⊂V ⊂ Ω is compact. We claim next that ∂fU = f∂U. Since f is open, ∂fU ⊂ f∂U. On the other hand, let z ∈ U ∩ f−1B(f(x),r), and let W be the component of V ∩f−1B(f(x),r) containing z. Then W is a neighbourhood of z which intersects U. Since U is a component of f−1B(f(x),r), we conclude that W = U and z ∈ U. Hence f∂U ⊂ ∂fU. This shows that U is a normal domain and that fU is both open and closed in B(f(x),r). Thus fU = B(f(x),r). (cid:3) 4 RAMILUISTOANDPEKKAPANKKA β σn 3 f C(σ3n) β(σ3n) β(0) x 0 Figure 1. Construction of the lift of a path in Theorem 3.1. 3. Path lifting after Floyd and Sto¨ılow We now turn to path-lifting, which is one of the main tools in the forth- coming proofs. The following theorem is, essentially, due to Sto¨ılow [15, pp. 354-358] and its idea was generalized by Floyd [6] to the setting compact metric spaces. We include here a version of Floyd’s proof in the planar setting for the reader’s convenience. Theorem 3.1. Let Ω ⊂ C be a planar domain and f: Ω → C a continuous, light, and open mapping. Let U be a normal domain compactly contained in Ω. Then for any path β: [0,1] → fU and any point x ∈ U ∩f−1{β(0)} 0 there exists a path α: [0,1] → U satisfying α(0) = x and f ◦α = β. 0 For the proof of Theorem 3.1 we need the following elementary lemma; cf. [18, pp. 131,148]. Lemma 3.2. Let Ω ⊂ C be a planar domain and f: Ω → C a continuous, light, andopenmapping. Suppose U isanormal domaincompactly contained in Ω. Then for any ε > 0 there exists a constant δ > 0 having the property that, for any continuum C ⊂ fU satisfying diam(C) < δ, the components of U ∩f−1C have diameter strictly less than ε. Proof. Supposethereexists ε > 0having thepropertythat, foreach n ∈ N, 0 there exists a continuum C ⊂ fU having diameter at most 1/n and a n component C′ of U ∩f−1C having diameter at least ε . n n 0 Since both U and fU are compact we may, by passing to subsequences, assumethatthesequences (C )and(C′)converge intheHausdorffdistance n n toacontinuum C ⊂fU andtoapointC′ ∈ U, respectively. ThenfC = C′. Since f is light, this is a contradiction. Thus the claim holds true. (cid:3) Proof of Theorem 3.1. Bytheuniformcontinuity ofβ andLemma3.2,there exists an increasing sequence of integers (m ) such that, for any interval n J ⊂ [0,1] satisfying diam(J) ≤ 2−mn, each component of U ∩f−1(βJ) has diameter of at most 2−n. STO¨ILOW’S THEOREM REVISITED 5 Foreachn∈ NwedenoteΣnthebarycentricdivisionof[0,1]intointervals σn = [(k−1)2−mn,k2−mn] for k = 1,...,2mn. We say that intervals σ and k σ′ in Σn are adjacent if σ and σ′ meet at an end point, that is, σ 6= σ′ and σ∩σ′ 6= ∅. We construct next, for each n ∈ N, a collection of continua {C(σ)}σ∈Σn with the following properties: (1) C(σ) is a component of f−1β(σ) having diam(C(σ)) < 2−n, (2) if 0 ∈ σ, then x ∈ C(σ), and 0 (3) C(σ)∩C(σ′) 6= ∅ for adjacent intervals σ,σ′ ∈ Σn; see Figure 1. Let n ∈ N. We define the continua C(σn),...,C(σn), where r = 2mn, 1 r inductively as follows. Let C(σn) be a component of U∩f−1β(σn) contain- 1 1 ing x . Suppose that we have defined continua C(σn),...,C(σn) for some 0 1 k 1 ≤ k < r, and let C(σn ) be a component of U ∩f−1β(σn ) which inter- k+1 k+1 sects C(σn). Such a component exists, since for any compact set K ⊂ fU, k each component of U ∩ f−1K maps surjectively onto K. The collection {C(σ)}σ∈Σn of continua now satisfies the conditions (1)-(3). Let n ∈ N and σ ∈ Σn. For each m ≥ n, let Cn,m(σ) be the union {C(τ) |τ ∈ Σm,τ ⊂ σ}. It is straightforward to v[erify that Cn,m(σ) is connected and f(Cn,m(σ)) = β(σ). Furthermore, since β([0,1]) ⊂ fU and U is a normal domain, there exists a uniform positive lower bound for the distance between Cn,m(σ) and ∂U. Thus, for each n ∈ N and σ ∈ Σn, there exists a subsequence of (Cn,m(σ))∞ converging to a continuum Cn(σ) ⊂ U in the Hausdorff m=n distance. By a diagonal argument, we may assume that (i) for each n ∈ N and σ ∈ Σn, the sequence (Cn,m(σ))∞ converges to m=n Cn(σ), and (ii) for all n ≤ k ≤ m and intervals σ ∈ Σn and σ′ ∈ Σk satisfying σ′ ⊂ σ, we have Ck,m(σ′) ⊂ Cn,m(σ). Note that diam(Cn(σ)) ≤ 2−n, and Cn(σ)∩Cn(σ′) 6= ∅ for adjacent inter- vals σ,σ′ ∈ Σn. After these preliminaries we are now ready to define a path α: [0,1] → U asfollows. Lett ∈ [0,1]. Foreachn ∈ N,letσ andσ′ betheintervalsinΣn n n containing t; possibly with σ = σ′. Then the sequence (Cn(σ )∪Cn(σ′)) n n n n n is a decreasing sequence of continua whose diameters tend to zero. Thus we may define α(t) by {α(t)} = (Cn(σ )∪Cn(σ′)). n n n \ For each pair σ and σ′ of adjacent intervals in Σn we have α(σ∪σ′) ⊂ Cn(σ)∪Cn(σ′), whereCn(σ)∪Cn(σ′) has diameter of at most 2·2−n. Thus α is continuous. By construction, f ◦α = β and α(0) = x . This concludes the proof. (cid:3) 0 We end this section the with uniqueness for lifts into simply connected planar domains. Note that this claim clearly fails for mappings between more general surfaces, even between spheres, and in higher dimensions. 6 RAMILUISTOANDPEKKAPANKKA Proposition 3.3. Let Ω⊂ C be a simply connected planar domain, f: Ω → C a continuous open map, and let β ,β : [0,1] → Ω be lifts of the same arc 1 2 α: [0,1] → fΩ for whichβ (0) = β (0) and β (1) = β (1). Then β = β . 1 2 1 2 1 2 The proof of Proposition 3.3 is an almost immediate consequence of the following version of the Jordan curve theorem; for a proof we refer to [11, Section 4] or [12, Theorem 1.10, p.33]. Theorem (The Jordan curve theorem). Let Ω ⊂ C be a simply connected planar domain and let c: S1 → Ω be an injective continous map. Then Ω\|c| consists of two domains, exactly one of which has a compact closure in Ω. Both of these domains have the image of the curve c as their boundary. Proof of Proposition 3.3. Suppose β 6= β . Then there exists t ∈ [0,1] for 1 2 0 which β (t ) 6= β (t ). Let 1 0 2 0 a = sup{t ∈ [0,t ] |β (t) = β (t)} and b = inf{t ∈ [t ,1] |β (t) = β (t)}. 0 1 2 0 1 2 Then there exists a closed loop c: S1 → Ω for which |c| = |β | |∪|β | |, 1 [a,b] 2 [a,b] andso|f◦c| = α[a,b]. BytheJordanCurveTheorem,oneofthecomponents of Ω\|c|, say U, is a precompact domain of Ω with ∂U = |c|. Thus fU is a precompact domain in C and ∂fU ⊂ |α|. This is a contra- diction since no such precompact domain exists since α is an arc. The claim follows. (cid:3) 4. Proof of Theorem 1.1 WeusenowpathliftingandtheJordanCurveTheoremtoproveTheorem 1.1, that is, to show that a continuous, light, and open planar mapping is discrete. We follow here that idea of Sto¨ılow that the discreteness follows from the finiteness of lifts of rays. Sto¨ılow calls the following proposition, together with the existence of lifts, Th´eor`eme Fondamental ([15, p. 361]). Proposition 4.1. Let Ω be a plane domain and f: Ω → C a light, open and continuous map. Let x ∈ Ω and let r > 0 be so small that U := U(x,f,r) is a normal domain contained in a simply connected neighbourhood in Ω. Suppose β: [0,1] → B(f(x),r) is a ray t 7→ tz +f(x), where x ∈ S1. Then 0 0 there exists at most finitely many lifts of β in U. Havingthispropositionatourdisposal,Theorem1.1 follows immediately. Proof of Theorem 1.1 assuming Proposition 4.1. Let x ∈ Ω be a point and 0 fix a normal domain U of x contained in some simply connected neigh- 0 0 bourhood in Ω. By Proposition 4.1, a ray in fU starting from f(x ) has 0 0 only finitely many lifts in U . Thus the set U ∩f−1{f(x )} is finite. (cid:3) 0 0 0 Proof of Proposition 4.1. Suppose a ray β: [0,1] → fU has infinitely many mutually distinct lifts β˜ : [0,1] → U, n ∈ N. By passing to a subsequence n if necessary, we may assume that β˜ (0) → x ∈ U and β˜ (1) → y ∈ U as n 0 n 0 n → ∞. There are two non-exlusive possibilities for the sequence (β˜ ): n (a) there exists a subsequence where all paths are mutually disjoint, or (b) there exists a subsequence where any two lifts intersect. STO¨ILOW’S THEOREM REVISITED 7 f β˜ β˜ N M β U(x,f,r) B(f(x),r) Figure 2. Constructinga“rectangularJordancurve”inthe proof of Proposition 4.1. Indeed, for each n there exists either infinitely many lifts β˜ ,m ≥ n, in- m tersecting β˜ or infinitely many lifts β˜ not intersecting β˜ . By a diagonal n m n argument we may fix a subsequence (β˜ ) of (β˜ ) such that for each k ∈ N, nk n either |β˜ |∩|β˜ | = ∅ for all m ≥ k or |β˜ |∩|β˜ | =6 ∅ for all m ≥ k. nk nm nk nm Since there exists infinitely many indices k for which one of these two possi- bilities hold, wemay pass toa subsequenceof (β˜ )toreceive asubsequence nk of type (a) or (b), respectively. For the rest of the proof we may assume that the original sequence (β˜ ) itself satisfies the condition in (a) or (b).1 n In case (a) we fix a radius s > 0 for which V := U(x ,f,s) and V′ := 0 U(y ,f,s) are mutually disjoint normal domains of x and y , respectively. 0 0 0 Let M,N ∈ N be indices for which β˜ (0),β˜ (0) ∈ V and β˜ (1),β˜ (1) ∈ M N M N V′. Then there exists a Jordan curve c: S1 → U such that |c| ⊂ V ∪V′∪|β˜ |∪|β˜ | M N and |c| 6⊂ V ∪V′; see Figure 2. The image |f ◦c| is a continuum contained in the ray |β| and two mutually disjoint disks, fV and fV′, located at the endpoints of β. By the Jordan Curve Theorem, the curve c bounds a precompact domain W in U, having boundary contained in |c|. Since fW is not precompact, this is a contradiction. In case (b), let γ: [0,1] → B(f(y ),r) be a ray for which γ(0) = f(y ) 0 0 and |γ|∩ |β| = f(y ); see Figure 3. For each n ∈ N, let γ˜ be a lift of γ 0 n starting from β˜ (1). The paths γ˜ are mutually disjoint, since otherwise a n n restriction of the compostition of β and γ contradicts Proposition 3.3. Now the argument of case (a) applies to the sequence (γ˜ ). (cid:3) n 1This argument does not rely on any special properties of the lifts or the geometry of the plane, only on combinatorial data on intersections – this is the infinite Ramsey theorem for two colors; see e.g. [8, Theorem 5, p. 16]. 8 RAMILUISTOANDPEKKAPANKKA U 0 γ f β˜ y 0 β U(y ,f,r ) 0 0 x 0 Figure 3. Path having infinitely many lifts with a joint starting point in the proof of Proposition 4.1. 5. Proofs of Theorems 1.2 and 1.3 Theorem 1.2 is a direct consequence of the following proposition. Proposition 5.1. Let Ω be a planar domain and let f: Ω → C be a con- tinuous, open, and discrete mapping. Let x ∈ Ω and let r > 0 be so small 0 that U := U(x ,f,r) is a normal neighbourhood of x contained in a simply 0 0 0 connected domain in Ω. Then U ∩B ⊂ {x }. 0 f 0 Note that since f is a priori both continuous and open, the branch set B f of f is actually the set of points at which f is not locally injective. Proof of Proposition 5.1. Suppose there exists b ∈ (U ∩B )\{x } and let 0 f 0 U ⊂ U be a normal neighborhood of b. Since b ∈B , the mapping f is not 0 f locally injective at b. Thus we may fix y ∈ fU for which #(U∩f−1{y }) ≥ 0 0 2. Let α: [0,1] → fU and β: [0,1] → fU \α(0,1] be arcs satisfying α(0) = 0 β(0) = y , α(1) = f(b), and β(1) = f(x ). Let also z ,z ∈ U ∩f−1{y }, 0 0 1 2 0 z 6= z . By Theorem 3.1, there exists, for i = 1,2, lifts α˜ : [0,1] → U and 1 2 i β˜ : [0,1] → U of α and β, respectively, satisfying α˜ (0) = β˜(0) = z for i 0 i i i i= 1,2. SinceU andU arenormalneighbourhoodsofx andb,respectively, 0 0 we have α˜ (1) = b, β˜(1) = x for i = 1,2. i i 0 For i= 1,2, let γ˜ : [0,1] → U be the path i α˜ (2t), t ∈ [0,1/2] i t 7→ (β˜i(2t−1), t ∈ [1/2,1]. Sinceγ˜ (0) = γ˜ (0), γ˜ (1) = γ˜ (1), andU iscontained inasimplyconnected 1 2 1 2 domain, we have by Proposition 3.3 that γ˜ = γ˜ . This is a contradiction 1 2 and the claim follows. (cid:3) We finish with a simple proof of Theorem 1.3 based on Theorem 1.2 and a covering argument. Proof of Theorem 1.3. Let z ∈ Σ. Since the property is local, we may as- sume that Σ and Σ′ are planar domains. STO¨ILOW’S THEOREM REVISITED 9 Let r > 0 be so small that U(z,f,r) is a normal neighborhood of z contained in some simply connected domain in Σ. Denote U′ = U(z,f,r)\ {z}, B′ = B(f(z),r)\{f(z)}, and D′ = D\{0}. Since f|U′: U′ → B′ is a local homeomorphism anda propermap, it is a covering map. Furthermore, since f is discrete, U′∩f−1{y} ⊂ U ∩f−1{y} is finitefor anyy ∈fU. As afinitecover of D′, U′ is atopological punctured disk and we conclude that U is a topological disk. Leth : U → Dandh : B(f(z),r) → Dbehomeomorphismswithh (z) = 1 2 1 0 and h (f(z)) = 0. Denote 2 g := h2◦f ◦(h1|U′)−1: D′ → D′. Since f|U′: U′ → B′ is a covering map, so is g. Thus the induced map g : π (D′) → π (D′) is of the form m 7→ km for some k ∈ Z\{0}. Denote ∗ 1 1 ζ := z 7→ zk,andleth′: D′ → D′betheliftofg: D′ → D′underthecovering k map ζ . Then g =ζ ◦h′ and h′ is a homeomorphism, since it is an injective k k coveringmap. Thehomeomorphismh′extendstoahomeomorphismh: D → D by the continuity of ζ and hence f| = h ◦ζ ◦h◦h =:φ−1◦ζ ◦ψ. (cid:3) k U 2 k 1 k Remark 5.2. We find it interesting that the argument of this corollary to- gether with the Cˇernavski˘ı-V¨aisa¨l¨a theorem (see [5] and [17]) on the branch set of discrete and open mappings gives another proof for thediscreteness of the branch set in the case of discrete and open mappings between surfaces. References [1] L. V. Ahlfors and L. Sario. Riemann surfaces. Princeton Mathematical Series, No. 26. Princeton UniversityPress, Princeton, N.J., 1960. [2] K.Astala,T.Iwaniec,andG.Martin.Ellipticpartialdifferentialequationsandquasi- conformalmappingsintheplane,volume48ofPrincetonMathematicalSeries.Prince- ton University Press, Princeton, NJ, 2009. [3] L. Bers. Riemann surfaces: Lectures by Lipman Bers. NewYork University,1958. [4] M. Bonk and D. Meyer. ExpandingThurston Maps. ArXiv e-prints, Sept. 2010. [5] A. V. Cˇernavski˘ı. Finite-to-one open mappings of manifolds. Mat. Sb. (N.S.), 65 (107):357–369, 1964. [6] E.E.Floyd.Somecharacterizationsofinteriormaps.Ann. of Math. (2),51:571–575, 1950. [7] O.Forster.LecturesonRiemannsurfaces,volume81ofGraduateTextsinMathemat- ics. Springer-Verlag, New York-Berlin, 1981. Translated from the German by Bruce Gilligan. [8] R.L.Graham,B.L.Rothschild,andJ.H.Spencer.Ramseytheory.Wiley-Interscience SeriesinDiscreteMathematicsandOptimization.JohnWiley&Sons,Inc.,NewYork, second edition, 1990. A Wiley-IntersciencePublication. [9] W. Hurewicz and H. Wallman. Dimension Theory. Princeton Mathematical Series, v.4. Princeton UniversityPress, Princeton, N.J., 1941. [10] R. Luisto. A characterization theorem for BLD-mappings between metric spaces. J. Geom. Anal., to appear. [11] E. E. Moise. Geometric topology in dimensions 2 and 3. Springer-Verlag, New York- Heidelberg, 1977. Graduate Textsin Mathematics, Vol. 47. [12] C. Pommerenke. Univalent functions. Vandenhoeck & Ruprecht, G¨ottingen, 1975. With a chapter on quadratic differentials by Gerd Jensen, Studia Mathemat- ica/Mathematische Lehrbu¨cher, Band XXV. 10 RAMILUISTOANDPEKKAPANKKA [13] S. Rickman. Path lifting for discrete open mappings. Duke Math. J., 40:187–191, 1973. [14] S. Rickman. Quasiregular mappings, volume 26 of Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)]. Springer- Verlag, Berlin, 1993. [15] S.Sto¨ılow.Surlestransformationscontinuesetlatopologiedesfonctionsanalytiques. Ann. Sci. E´c. Norm. Sup´er. (3), 45:347–382, 1928. [16] S. Sto¨ılow. Lec¸ons sur les principes topologiques de la th´eorie des fonctions analy- tiques. Profess´ees `a la Sorbonne et `a l’universit´e de Cernauti. X + 148 p. Paris, Gauthier-Villars (Collection de monographies sur la th´eorie des fonctions) (1938)., 1938. [17] J. V¨ais¨al¨a. Discrete open mappings on manifolds. Ann. Acad. Sci. Fenn. Ser. A I No.,392:10, 1966. [18] G.T.Whyburn.Analytictopology.AmericanMathematicalSocietyColloquiumPub- lications, Vol. XXVIII.American Mathematical Society,Providence, R.I.,1942. [19] G.T.Whyburn.Topological analysis.Second,revised edition.Princeton Mathemati- cal Series, No. 23. Princeton UniversityPress, Princeton, N.J., 1964. Department of Mathematics and Statistics, 520 Portola Plaza CA 90095- 1555, UCLA, USA and Department of Mathematics and Statistics, P.O. Box 35, FI-40014 University of Jyva¨skyla¨, Finland E-mail address: [email protected] DepartmentofMathematicsandStatistics,P.O.Box68(GustafHa¨llstro¨min katu 2b), FI-00014 University of Helsinki, Finland E-mail address: [email protected]