Arxiv:1607.06300V4 [Math.CV] 29 Mar 2020 T Pc.Teagmnsfor Arguments the Space

Home , Beltrami equation, Quasicircle, Schwarzian derivative

arXiv:1607.06300v4 [math.CV] 29 Mar 2020 T pc.Teagmnsfor arguments The space. homeomorphisms mrhssof omorphisms ieacmlxBnc aiodsrcuefor structure manifold Banach complex a vide ecmu lrsaeo uhafml fcrl oemrhssi h univ the in homeomorphisms c circle we space of family homeomorphisms, a quasisymmetric such Teichmüller for of space conditions regularity tain space rirr yeblcReansraecnb nesoda h fix the as on understood acting be group Fuchsian can corresponding surface Riemann hyperbolic arbitrary cwrindrvtv;Br medn;qaiymti homeomor derivative. quasisymmetric continuous embedding; Bers derivative; Schwarzian uscnomlsl-oemrhs fteui disk unit the of self-homeomorphism quasiconformal a Möb( of self-homeomorphisms metric iemrhsswt ¨le otnosdrvtvso exponent of derivatives Hölder continuous with diffeomorphisms tde ntefaeoko h hoyo ecmulrsae.I t In Teichmüller spaces. of theory the Teichmüller of space universal framework the the in studied nti ae,w omlt h Teichmüller space the formulate we paper, this In hswr a upre yJP AEH 25287021. KAKENHI JSPS by supported was work This phrases. and words Key 2010 aaerzto foinainpeevn iemrhsso the of diffeomorphisms orientation-preserving of Parametrization n eti enmnsaeipre rmteter o h little the for theory the from imported are refinements certain and TEICHM S T T .Hr,aqaiymti efhmoopimof self-homeomorphism quasisymmetric a Here, ). ahmtc ujc Classification. Subject Mathematics nrdc h ecmulrsaeo iemrhsso h ntcircle unit the of diffeomorphisms Teichmüller of the space introduce rvd ope aahmnfl tutr o tadpoeta it local that by prove induced and it one e for the Bers structure with the manifold by Banach given complex dilatation derivative a complex Schwarzian provide the the Teichmüller of and spa universal terms extension in formal the quantitatively of diffeomorphism subspace a such a as derivatives tinuous Abstract. 0 Möb(= Möb(= LE PC FCRL IFOOPIM WITH DIFFEOMORPHISMS CIRCLE OF SPACE ULLER ¨ D S ) S ae nteqaiofra hoyo h nvra Teichmüller universal the of theory quasiconformal the on Based S hs ope iaain aiha h boundary the at vanish dilatations complex whose \ hc r h onayetnino smttclycnomlho conformal asymptotically of extension boundary the are which , ) QS. H \ y.Hr,tesbru Sym subgroup the Here, Sym. LE OTNOSDERIVATIVE CONTINUOUS OLDER ¨ nvra ecmulrsae uscnomlmp etaicoeffi Beltrami map; quasiconformal Teichmüller space; universal T 0 α r oee ntoefrteuieslTeichmüller space universal the for those on modeled are ASHK MATSUZAKI KATSUHIKO T S C hc a ergre stegopQ fquasisym- of QS group the as regarded be can which , 1+ 1. ouops-opsto fMöbius transformations of post-composition modulo α tplg ttebs point. base the at -topology Introduction rmr 06,3C2 21;Scnay3E0 58D05. 37E10, Secondary 32G15; 30C62, 30F60, Primary T fw elc h ru nainewt cer- with invariance group the replace we If . 1 T 0 α n rv ai rpriso this of properties basic prove and T ⊂ 0 α D hs;crl iemrhs;Hölder diffeomorphism; circle phism; Scnit fsmercself- symmetric of consists QS h ecmulrsaeo an of Teichmüller space The . Möb(= S stebudr xeso of extension boundary the is bdig hn we Then, mbedding. e echaracterize We ce. oooycoincides topology s with S ) α dpitlcso the of locus point ed \ ntcircle unit fisquasicon- its of i ae eutilize we case, his Diff ∈ α nas me the embed also an S -Hölder con- ra Teichmüllerersal Teichmüller sub- tcnan all contains It . (0 + 1+ pc,we space, , α ) epro- We 1). ( S fcircle of ) S a be can cients; me- 2 K. MATSUZAKI

circle diffeomorphisms; hence, Diff1+α(S) Sym. We will survey necessary results on the + ⊂ universal Teichmüller space T in Section 2 and on the little subspace T0 in Section 3. We first characterize circle diffeomorphisms with Hölder continuous derivatives in terms of their quasiconformal extension to D. This originates from the work of asymptotically conformal maps by Carleson [15]. Later, Gardiner and Sullivan [24] developed the theory of symmetric homeomorphisms of S using previous results on quasiconformal extension and Schwarzian derivatives of univalent functions in Becker and Pommerenke [12]. We will refine these results quantitatively concerning the decay order of the corresponding maps vanishing at the boundary. We verify in Section 4 that, if the complex dilatation µ of an asymptotically conformal homeomorphism of D decays in the order of O((1 z )α) as z 1, then the hyperbol- ically weighted Schwarzian derivative of the developing−| | map| of|→ the projective structure on the exterior disk D∗ determined by µ decays exactly in the same order α. This is carried out by dividing the support of µ suitably into annular regions and estimating the pre-Schwarzian derivative of the composition of conformal homeomorphisms. A different proof was previously obtained by Dyn′kin [18], but we have to prepare more precise estimates in terms of a weighted supremum norm of µ (Theorem 4.1 and Corollary 4.7). In Section 5, we mainly consider one-dimensional properties of circle diffeomorphisms 1+α S with Hölder continuous derivatives. We first provide a topology for Diff+ ( ) and see that it is a topological group (Proposition 5.2). The topology is defined in a neighborhood of the identity map by C1+α-convergence and then distributed to every point by the right 1+α S translation of the group. For the characterization of an element of Diff+ ( ), a result of Carleson [15] plays an important role, as it gives a connection between the Hölder continuity of the derivative and the quasisymmetry quotient of g. We review his theorem and supply necessary claims for our arguments. A fundamental result is that if the complex dilatation of an asymptotically conformal homeomorphism of D decays in the order of O((1 z )α), then the regularity of its boundary extension g to S is exactly C1+α. Carleson− | found| that it is at least C1+α/2. This problem was investigated further by Anderson and Hinkkanen [7] among others, and settled qualitatively by Dyn′kin [18] and Anderson, Cantón, and Fernández [6]. Combined with the aforementioned results, this can be summarized as follows (Theorem 6.7). Theorem 1.1. Let α be a constant with 0 <α < 1. The following conditions are equivalent for g QS: ∈ (1) g is a diffeomorphism of S with Hölder continuous derivative of exponent α; (2) g extends continuously to a quasiconformal self-homeomorphism of D whose complex dilatation µ(z) decays in the order of O((1 z )α) as z D tends to the boundary; −| | ∈ (3) the Schwarzian derivative ϕ(z) of the conformal homeomorphism of D∗ determined by g behaves in the order of O(( z 1)−2+α) as z D∗ tends to the boundary. | | − ∈ In Section 6, we will improve on Theorem 1.1 with a different proof, which is necessary for the arguments of Teichmüller spaces (Theorem 6.9). Our strategy is to represent a TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 3

circle diffeomorphism g by conformal welding, which was originally proposed by Anderson, Becker, and Lesley [5]. For the argument in this method, we need to know that an asymptotically conformal self-homeomorphism f of D and its inverse mapping f −1 have the complex dilatations of order O((1 z )α) at the same time. For this purpose, we extend the consequence of the Mori theorem−| | to a quasiconformal self-homeomorphism f of D with complex dilatation of order O((1 z )α). The result is that 1 f(z) is comparable to 1 z without the power of the−| maximal| dilatation K(f) (Theorem−| 6.4).| This guarantees that−| | the complex dilatation of f −1 is also of order O((1 z )α). −| | Theorem 1.2. Let f be a quasiconformal self-homeomorphism of D with f(0) = 0 whose complex dilatation µ(z) satisfies µ(z) ℓ(1 z )α almost every z D for some ℓ 0. Then, there is a constant A 1 depending| | ≤ only−| on| K(f), α, and ℓ such∈ that ≥ ≥ 1 (1 z ) 1 f(z) A(1 z ) A −| | ≤ −| | ≤ −| | for every z D. ∈ For Beltrami coefficients and Schwarzian derivatives as above, we prepare the following α D D spaces: Bel0 ( ) is the space of Beltrami coefficients µ on with a finite norm µ ∞,α = α α D∗ k k ess.sup ρD(z) µ(z) ; and B0 ( ) is the Banach space of holomorphic quadratic differentials 2| | ∗ −2+α ϕ = ϕ(z)dz on D with a finite norm ϕ ∞,α = sup ρD∗ (z) ϕ(z) . Here, ρ• denotes the hyperbolic density of each space. Then,k k Theorem 1.1 implies| that| the Teichmüller projection π, the Bers projection Φ, and the Bers embedding β for the universal Teich- müller space T also work for our spaces by restriction of the original mappings: α D Bel0 ( ) ❦❦ ❖❖ ❦❦❦ ❖❖❖ ❦❦❦ ❖❖ ❦❦❦ ❖❖❖ ❦u ❦❦ π Φ ❖❖' T α = Möb(S) Diff1+α(S) / β(T ) Bα(D∗) 0 \ + β ∩ 0 α α D α The topology on T0 is induced from Bel0 ( ) by π. If we regard T0 as the subgroup of 1+α S Diff+ ( ) consisting of normalized elements, then we can also provide it with the right 1+α S uniform topology of Diff+ ( ), which is generated by the right translations of the local 1+α α C -topology at the identity. In Section 7, we prove that these topologies on T0 are the same (Theorem 7.8). α α D α Theorem 1.3. The quotient topology on T0 induced by π : Bel0 ( ) T0 coincides with 1+α S α → the right uniform topology of Diff+ ( ) and, in particular, T0 is a topological group. α The complex structure on T0 is given by showing that the Bers embedding β as above is a homeomorphism onto its image. Moreover, we want to find that the base point change α map (right translation) of T0 is compatible with this complex structure. To this end, we will prove that the Bers projection Φ is a holomorphic split submersion. To see that Φ is continuous, we use an integral representation of the Schwarzian derivative Φ(µ), which was originally proposed by Astala and Zinsmeister [8]. Then, a careful estimate of this 4 K. MATSUZAKI

integral taking the dependence of constants into account yields the assertion on continuity. The holomorphy is a consequence from the continuity in our situation. To see that Φ is a split submersion, we construct a local holomorphic section of Φ. For the universal Teich- müller space (and Teichmüller spaces of Riemann surfaces), this was originally proved by Bers [13], and afterward certain modifications have been made to develop a standard argument. We adapt this argument to our situation to show the continuity with respect to the topology in our spaces. In Section 7, we will prove the following: α D α D∗ Theorem 1.4. The Bers projection Φ : Bel0 ( ) B0 ( ) is a holomorphic split sub- → α α D∗ mersion onto its image. This implies that the Bers embedding β : T0 B0 ( ) is α → a homeomorphism onto its image. With this complex structure of T0 identified with a α D∗ α domain of the complex Banach space B0 ( ), every base point change map of T0 is a α biholomorphic automorphism of T0 . A motivation of this work is to apply the Bers embedding of the Teichmüller space α 1+α S T0 to studies of the rigidity of the Diff+ ( )-representation of a Möbius group and the 1+α S regularity of the conjugation of a subgroup of Diff+ ( ) to a Möbius group. These arguments are developed in a continuation [33] of the present work. An overview of our project can be found in [31]. A preliminary study can be found in [30].

2. The universal Teichmuller¨ space In this section, we define the universal Teichmüller space in terms of the group of quasisymmetric self-homeomorphisms of the circle, and then introduce a topological and a complex structure on this space by using the quasiconformal theory: the Beltrami equation and the Schwarzian derivative. Basic results can be found in Lehto [28]. We denote the group of all quasiconformal self-homeomorphisms of the unit disk D by QC(D). Each quasiconformal homeomorphism f QC(D) extends continuously to the boundary S as a homeomorphism. Then, we have∈ a homomorphism q : QC(D) Homeo(S) in the group of self-homeomorphisms of the unit circle S. An orientation-→ preserving self-homeomorphism g of S is called quasisymmetric if g Im q. We denote the group Im q of all quasisymmetric self-homeomorphisms of S by QS.∈ Let Möb(D) QC(D) denote the subgroup of all conformal self-homeomorphisms of D, which are Möbius⊂ transformations of D. We define Möb(S)= q(Möb(D)) QS. ⊂ Definition. The universal Teichmüller space T is defined as the set of the cosets Möb(S) QS. We denote the coset of g QS by [g]. \ ∈ The Beltrami coefficient µ on a domain D C is a measurable function with a supremum norm µ less than 1. We denote the⊂ set of all Beltrami coefficients on D by k k∞ b Bel(D)= µ L∞(D) µ < 1 . { ∈ |k k∞ } Every quasiconformal homeomorphism f : D D′ has partial derivatives ∂f and ∂f¯ in →¯ the distribution sense and the ratio µf (z) = ∂f(z)/∂f(z) called the complex dilatation TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 5

belongs to Bel(D). The maximal dilatation of f is defined by 1+ µ K(f)= k f k∞ . 1 µ −k f k∞ Given K 1, we call f a K-quasiconformal if K(f) K. The measurable Riemann mapping theorem≥ asserts that a Beltrami coefficient uniquely≤ determines a quasiconformal homeomorphism up to post-composition of conformal homeomorphisms (see Lehto and Virtanen [29] for the history of this theorem, and Morrey [34], Ahlfors and Bers [3], and Ahlfors [1] for the proof). Applying this theorem to quasiconformal homeomorphisms of D, we see that Bel(D) can be identified with the set of the cosets Möb(D) QC(D). Then, the boundary extension q : QC(D) QS induces a surjective map π\: Bel(D) T by taking the quotient D → S → of Möb( ) ∼= Möb( ). This is called the Teichmüller projection. The topology of the universal Teichmüller space T is given as the quotient topology of the unit ball Bel(D) of the Banach space L∞(D) by the projection π so that π is continuous. There is a global continuous section for the Teichmüller projection π : Bel(D) T . This is defined by giving a canonical quasiconformal extension e : QS QC(D) for→ each quasisymmetric self-homeomorphism g of S. The extension due to Beurling→ and Ahlfors [14] can be used to obtain such a section. Douady and Earle [17] introduced another extension e : QS QC(D) having the conformal naturality such that DE → e (φ g φ )= e (φ ) e (g) e (φ ) DE 1 ◦ ◦ 2 DE 1 ◦ DE ◦ DE 2 for any φ ,φ Möb(S) and any g QS. We note that e (φ ) and e (φ ) are the 1 2 ∈ ∈ DE 1 DE 2 Möbius transformations of D extending φ1 and φ2, respectively. By taking the quotient of S D D Möb( ) ∼= Möb( ), we have a continuous map sDE : T Bel( ) such that π sDE = idT . We call this the conformally natural section. The existence→ of a global continuous◦ section implies that T is contractible. The measurable Riemann mapping theorem implies that, for every ν Bel(D), there is a unique normalized quasiconformal homeomorphism f QC(D) whose∈ complex dilatation coincides with ν. Here, the normalization is given by∈ fixing three boundary points 1, i, and 1 on S. We denote this normalized quasiconformal homeomorphism by f ν. The − D D subgroup of QC( ) consisting of all normalized elements is defined as QC∗( ). This also D defines the normalized elements of QS, which constitute the subgroup QS∗ = q(QC∗( )). Applying this normalization, we can define a group structure on Bel(D) and T as follows. For any ν1, ν2 Bel(D), we set ν1 ν2 to be the complex dilatation of the composition f ν1 f ν2 . Then,∈ Bel(D) has a group∗ structure with this operation . In other words, ◦ D D D ∗ D by the identification of Bel( ) with QC∗( ), we regard Bel( ) as a subgroup of QC( ). We denote the inverse element of ν Bel(D) by ν−1, which is the complex dilatation of (f ν)−1. The chain rule of partial differentials∈ yields a formula

ν2 −1 ν1(z) ν2(z) ∂f (z) ν2 ν1 ν2 (ζ)= − (ζ = f (z)). ∗ 1 ν (z)ν (z) · ∂f ν2 (z) − 2 1 6 K. MATSUZAKI

For the base point [id] of T , the inverse image of the Teichmüller projection π−1([id]) = ν Bel(D) q(f ν) = id { ∈ | } is a normal subgroup of Bel(D) as q : QC(D) QS is a homomorphism. Having T = Bel(D)/π−1([id]), we see that T has a group structure→ with the operation defined by ∗ π(ν1) π(ν2) = π(ν1 ν2). Then, π : Bel(D) T is a surjective homomorphism with −1 ∗ ∗ → π ([id]) its kernel. If we identify T with QS∗, we may regard T as a subgroup of QS and D the projection π as the restriction of q to QC∗( ). −1 Each ν Bel(D) induces the right translation rν : Bel(D) Bel(D) by µ µ ν . The projection∈ under π yields a well-defined map R : T →T by 7→ ∗ π(ν) → π(µ) π(µ ν−1)= π(µ) π(ν)−1. 7→ ∗ ∗ In this way, for every point τ T , we have the base point change map Rτ : T T ∈ −1 → sending τ to [id]. By the above formula, we see that rν and (rν) = rν−1 are continuous; hence, rν is a homeomorphism onto Bel(D). From this, we see that the base point change map Rτ is also a homeomorphism onto T . The universal Teichmüller space T has a complex structure modeled on a certain complex Banach space. This is seen as follows. For µ Bel(D), we extend µ(z) to C by ∈ setting µ(z) 0 for z D∗ = C D. By the measurable Riemann mapping theorem, ≡ ∈ − b there exists a unique quasiconformal self-homeomorphism f of C up to post-composition b µ of Möbius transformations whose complex dilatation coincides with the extended Beltrami b coefficient µ. We take the Schwarzian derivative Sf (z) of the conformal homeomorphism ∗ f(z)= f D∗ (z) on D . The ambiguity of f by Möbius transformations is offset by taking µ| µ the Schwarzian derivative because Sh◦f (z)= Sf (z) for every h Möb(C). We define the Banach space of holomorphic quadratic differentials∈ ϕ = ϕ(z)dz2 on D∗ with a finite hyperbolic supremum norm by b ∗ ∗ −2 B(D )= ϕ Hol2(D ) ϕ ∞ = sup ρD∗ (z) ϕ(z) < , ∗ { ∈ |k k z∈D | | ∞} 2 ∗ where ρD∗ (z)=2/( z 1) is the hyperbolic density on D . We note that an element ϕ of Hol (D∗) satisfies| ϕ| (−z) = O(1/z4) (z ). The Nehari–Kraus theorem asserts that 2 → ∞ ϕ ∞ 3/2 for the Schwarzian derivative ϕ(z)= Sf (z) of any conformal homeomorphism kf ofk D≤∗. Hence, we have a map Φ : Bel(D) B(D∗) by the correspondence of µ Bel(D) → D ∈ to Sfµ|D∗ , which is called the Bers projection (onto the image Φ(Bel( ))). With regard to the Teichmüller projection π : Bel(D) T and the Bers projection ∗ → Φ : Bel(D) B(D ), it can be proved that π(µ1) = π(µ2) if and only if Φ(µ1) = Φ(µ2). Therefore, we→ have a well-defined injection β : T B(D∗) that satisfies β π = Φ. This is called the Bers embedding of the universal Teichmüller→ space T . ◦ Proposition 2.1. The Bers projection Φ : Bel(D) B(D∗) is continuous. → Proof. For two arbitrary points µ, ν Bel(D), we apply the right translation rν to µ. On ∗ ∈ the quasidisk fν(D ), we use an estimate of the Schwarzian derivative of the conformal TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 7

homeomorphism f f −1 in terms of r (µ) (see Theorem II.3.2 in [28]). Then, µ ◦ ν k ν k∞ 3 µ ν Φ(µ) Φ(ν) 3 r (µ) k − k∞ , k − k∞ ≤ k ν k∞ ≤ 1 ν µ −k k∞k k∞ which implies that Φ is continuous. In fact, the Bers projection Φ is holomorphic. Once we have Φ as continuous, then the holomorphy is a consequence of the point-wise holomorphic dependance of the normalized solution fµ(z) of the Beltrami equation for µ, which is a signiﬁcant contribution to the measurable Riemann mapping theorem by Ahlfors and Bers [3]. Moreover, the following result was proved by Bers [13]. Theorem 2.2. The Bers projection Φ : Bel(D) B(D∗) is a holomorphic split submersion. → The condition needed for Φ to be a holomorphic split submersion is equivalent to the existence of a local holomorphic section for Φ at every ϕ Φ(Bel(D)) sending ϕ to an arbitrary point of Φ−1(ϕ) (see Section 1.6 of Nag [35] concerning∈ holomorphic split submersion between domains of Banach spaces). This implies that Φ is an open map, and in particular, the image Φ(Bel(D)) in B(D∗) is open (hence, it is a bounded domain). As π is a topological quotient map and Φ is continuous and open, the Bers embedding β = Φ π−1 : T B(D∗) is a homeomorphism onto the image β(T ) = Φ(Bel(D)). By identifying◦ T with→ a bounded domain β(T ) B(D∗), we provide a complex structure for ⊂ T . Then, the base point change map Rτ for every τ T is a biholomorphic automorphism of T . Indeed, for an arbitrary point ϕ β(T ), we take∈ a local holomorphic section η of Φ and ν Bel(D) with π(ν)= τ. We represent∈ R at β−1(ϕ) by ∈ τ R = β−1 Φ r η β. τ ◦ ◦ ν ◦ ◦ −1 As Φ r η is holomorphic, R is holomorphic. As the inverse R = R −1 is also ◦ ν ◦ τ τ τ holomorphic, Rτ is biholomorphic.

3. Symmetric homeomorphisms and the little Teichmuller¨ subspace A quasisymmetric homeomorphism was originally introduced as a function on R that has quasiconformal extension to the upper half-plane H. It can be characterized by the quasisymmetry quotient deﬁned as follows: Deﬁnition. An increasing homeomorphism h : R R is called a quasisymmetric function if there exists a constant M 1 such that → ≥ 1 h(x + t) h(x) − M M ≤ h(x) h(x t) ≤ − − holds for every x R and for every t > 0. The ratio in the mid-term is called the ∈ quasisymmetry quotient of h and is denoted by mh(x, t). 8 K. MATSUZAKI

For an orientation-preserving self-homeomorphism g : S S, we can take its lift g : R R with u g = g u for the universal cover u : R S given→ by u(x)= e2πix. This is uniquely→ determined◦ up◦ to an additive integer and is an→ increasing homeomorphism of Re satisfying g(x+1)e = g(x)+1. Conversely, for an increasing homeomorphism h : R R with h(x +1)= h(x)+1, we can take its projection h : S S with u h = h u. → It is knowne that g ise a quasisymmetric self-homeomorphism→ of S if and◦ only◦ if its lift g is a quasisymmetric function on R (see Theorem 4.4 in [30]). To determine whether g is quasisymmetric, it is enough to check the quasisymmetry quotient mge(x, t) for 0 x< e1 and 0 < t 1/2 (see Proposition 4.5 in [30]). For each g QS, we introduce≤ e the quasisymmetry≤ constant of g as ∈ ±1 M(g) = sup mge(x, t) . 0≤x<1, 0

This defines a topology on QS. More precisely, gn QS converge to g QS if M(gn −1 ∈ ∈ ◦ g ) 1 as n . Then, the relative topology on QS∗ QS coincides with the Teich- → →∞ ⊂ D müller topology on T ∼= QS∗, which is the quotient topology under π : Bel( ) T (see Theorem III.3.1 in Lehto [28]). → We consider a special class of quasisymmetric functions on R whose quasisymmetry quotient tends to 1 uniformly as t 0. We also consider the corresponding quasisymmetric homeomorphisms of S. → Definition. A quasisymmetric function h : R R is called symmetric if there exists a → non-negative increasing function ε(t) for t> 0 with limt→0 ε(t) = 0 such that (1 + ε(t))−1 m (x, t) 1+ ε(t) ≤ h ≤ for all x R. We call ε(t) a gauge function for symmetry. A quasisymmetric homeomorphism g ∈ QS is called symmetric if its lift g : R R is a symmetric function. We denote the subset∈ of all symmetric self-homeomorphisms→ of S by Sym QS. e ⊂ As the corresponding concept for quasiconformal maps, there are asymptotically conformal homeomorphisms whose complex dilatations vanish at the boundary. We will review the relation of these two maps. In particular, we consider a certain quantitative estimate of the complex dilatation of the quasiconformal extension in terms of the quasisymmetry quotient. This was originally studied by Carleson [15]. For a quasisymmetric function h : R R, we set → 1 1 α(x, y)= h(x + ty)dt; β(x, y)= h(x ty)dt, Z0 Z0 − and define 1 F (z)= [α(x, y)+ β(x, y)]+ i[α(x, y) β(x, y)] 2 − for z = x + iy H. Beurling and Ahlfors [14] proved that F is a quasiconformal self- homeomorphism∈ of H with an estimate of the maximal dilatation of F in terms of the quasisymmetry constant M 1 of h. We call this the Beurling–Ahlfors extension of ≥ TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 9

h. With regard to the Beurling–Ahlfors extension of symmetric functions, the following result, which was proved in Lemma 3 of [15] and improved slightly by providing an explicit computation for involved constants in Theorem 5.1 of [30], is crucial.

±1 Theorem 3.1. Let h : R R be a symmetric function such that mh(x, t) 1+ ε(t) for a gauge function ε(t).→ Let F be the Beurling–Ahlfors extension of h, which≤ is a quasiconformal self-homeomorphism of H. Then, the complex dilatation µF of F satisﬁes µ (z) 4ε(y) for every z = x + iy H. | F | ≤ ∈ In particular, this theorem shows that a symmetric function h : R R extends continuously to a quasiconformal homeomorphism F : H H with F ( )=→ whose complex → ∞ ∞ dilatation µF (z) uniformly tends to 0 as y 0 on x R. Conversely, such a quasiconformal self-homeomorphism→ ∈ F of H extends to a symmetric function on R. Lemma 2 of Carleson [15] proved this fact, giving the order of a gauge function for symmetry. We will reprove this result in the following form with a more explicit estimate for the gauge function. This estimate is useful in later arguments.

Theorem 3.2. If a K-quasiconformal homeomorphism F : H H with F ( ) = → ∞ ∞ satisﬁes µF (z) ε(y) uniformly on x R for a function ε(y) with ε(y) 0 as y 0, then its| boundary| ≤ extension h : R R∈ is a symmetric function whose→ quasisymmetry→ quotient satisﬁes me(x, t)±1 1+ ε→(t) for a gauge functioneε(t) with e h ≤ ε(t) c ε(√t)+ R√t (0 < t 1/2), ≤ ≤ where c = c(K) > 0 is a constante depending only on K 1, and R > 0 is an absolute constant. ≥

Proof. For each t (0, 1/2], we define a Beltrami coefficient µ (z) by letting µ (z)= µ (z) ∈ t t F on z H y > √t and µ (z) 0 elsewhere. Let F be the quasiconformal self- { ∈ | } t ≡ t homeomorphism of H with complex dilatation µt and with Ft( ) = , and ht the H ∞ ∞ −1 quasiconformal self-homeomorphism of such that F = ht Ft. As ht = F Ft , ◦ ◦H the complex dilatation of ht satisfies µht (Ft(z)) ε(√t) for almost every z . In particular, there is a constant c′ > 0 depending| only| ≤ on K such that the maximal dilatation∈ of h is estimated as K(h ) 1+ c′ε(√t). e t t ≤ By reflection with respect to R, we may assume that Ft is a quasiconformal self- homeomorphism of C. The restrictione of F to the strip domain z C y < √t t { ∈ | | | } is conformal. For each x R, we consider the ball of radius √t with center x and apply the Koebe distortion theorem∈ (Proposition 3.3 below) to the conformal homeomorphism Ft on this disk. Then,

′ ′ Ft (x) t Ft (x) t | | Ft(x + t) Ft(x) | | . (1 + t/√t)2 ≤ − ≤ (1 t/√t)2 − 10 K. MATSUZAKI

The middle term can be replaced with F (x) F (x t). This leads us to the following t − t − estimate for the quasisymmetry quotient m (x, t) of F R: Ft t| 2 2 (1 √t) Ft(x + t) Ft(x) (1 + √t) − mF (x, t)= − . 2 t 2 (1 + √t) ≤ Ft(x) Ft(x t) ≤ (1 √t) − − − ′ ±1 ′ In particular, there is an absolute constant R > 0 such that mFt (x, t) 1+ R √t for 0 < t 1/2. ≤ Next,≤ we apply the quasiconformal homeomorphism h to the points F (x t), F (x), t t − t and Ft(x + t), which are mapped to h(x t), h(x), and h(x + t), respectively. We note that the quasisymmetry quotients can be− given by the conformal moduli as follows: m (x, t)= λ(mod H(F (x t), F (x), F (x + t), )); Ft t − t t ∞ m (x, t)= λ(mod H(h(x t), h(x), h(x + t), )). h − ∞ Here, mod Q(x , x , x , x ) (0, ) stands for the conformal modulus of a quadrilateral 1 2 3 4 ∈ ∞ Q with four positively ordered vertices x1, x2, x3, x4 ∂Q, and λ : (0, ) (0, ) is the distortion function, which transforms conformal moduli∈ to quasisymmetry∞ → quotients∞ (see Section I.2.4 of [28] and Section II.6 of [29]). Moreover, the ratio of the conformal moduli are bounded by the maximal dilatation K(h ) 1+ c′ε(√t): t ≤ 1 mod H(h(x t), h(x), h(x + t), ) e − ∞ K(h ). K(h ) ≤ mod H(F (x t), F (x), F (x + t), ) ≤ t t t − t t ∞ Plugging the quasisymmetry quotients in this inequality gives m (x, t)= λ(mod H(h(x t), h(x), h(x + t), )) h − ∞ λ(K(h ) mod H(F (x t), F (x), F (x + t), )) ≤ t t − t t ∞ −1 = λ(K(ht)λ (mFt (x, t))) λ((1 + c′ε(√t))λ−1(1 + R′√t)) ≤ R −1 for all x and t (0, 1/2]. An estimatee for mh(x, t) is similarly obtained. Because λ is continuous∈ and∈ increasing with λ(1) = 1 and diﬀerentiable at 1 with a non-vanishing derivative (see e.g. [4]), we see that the last term can be represented as 1+ ε(t) for a gauge function ε(t) as in the statement of the theorem. We review the Koebe distortion theorem, which includes the one-quarter theorem (see Theorem 1.3 in [37]). Proposition 3.3. A conformal homeomorphism f of D into C satisﬁes z z f ′(0) | | f(z) f(0) f ′(0) | | ; | |(1 + z )2 ≤| − |≤| |(1 z )2 | | −| | 1 z 1+ z f ′(0) −| | f ′(z) f ′(0) | | | |(1 + z )3 ≤| |≤| |(1 z )3 | | −| | TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 11

for every z D. The first inequality in the former line in particular shows that the image f(D) contains∈ a disk with its center at f(0) and radius f ′(0) /4. | | Using the Beurling–Ahlfors extension, we can also define a quasiconformal extension of a quasisymmetric self-homeomorphism g of S to D. Actually, for the lift g : R R of g under the universal cover u : R S, we take the Beurling–Ahlfors extension F : H→ H of g. Here, we also use the extension→ of u to the holomorphic universal cover ue: H D→ 0 defined by u(z)= e2πiz. By projecting down F to a quasiconformal self-homeomorphism→ −{ } ofe D 0 by the holomorphic universal cover u and filling the puncture 0, we obtain a quasiconformal−{ } self-homeomorphism f of D. By this correspondence g f, we have a map 7→ e : QS QC(D), BA → which satisfies q e = id . ◦ BA |QS Unlike the Douady–Earle extension eDE, the Beurling–Ahlfors extension eBA does not have conformal naturality. Accordingly, it does not descend to a section T Bel(D) naturally. In order to define a section, we use the normalized quasisymmetric→ homeomorphism g QS∗ as a representative of an element [g] T . From this g, we make the quasiconformal∈ self-homeomorphism f of D as above, and∈ then take its complex dilatation µ . By this correspondence [g] µ , we have a map s : T Bel(D), which is a f 7→ f BA → section for the Teichmüller projection π : Bel(D) T . It can also be proved that sBA is continuous. → We say that a quasiconformal homeomorphism f QC(D) is asymptotically conformal ∈ if the complex dilatation µf (z) vanishes at the boundary S. This means that

lim ess.sup µf (z) z 1 t =0. t→0 {| ||| | ≥ − } We denote the subset of QC(D) consisting of all asymptotically conformal homeomorphisms by AC(D). Theorem 3.1 implies that the restriction of eBA to Sym gives e : Sym AC(D). BA → Moreover, Theorem 3.2 implies that the restriction of q to AC(D) gives q : AC(D) Sym . → We note that, for a given point z0 in D, there is a quasiconformal self-homeomorphism φ with φ(z0) = 0 and q(φ) = id S whose complex dilatation vanishes outside some compact subset in D. The composition| of such a map φ makes any asymptotically conformal self- homeomorphism of D ﬁx 0 without changing the property of vanishing at the boundary. By the above two claims, we have the following result attributed to Fehlmann [22] in Gardiner and Sullivan [24]: Corollary 3.4. A quasisymmetric homeomorphism g is in Sym if and only if g extends continuously to a quasiconformal homeomorphism in AC(D). 12 K. MATSUZAKI

By the chain rule of complex dilatations, the composition of asymptotically conformal self-homeomorphisms of D is also asymptotically conformal. Hence, AC(D) is a subgroup of QC(D). Accordingly, Corollary 3.4 shows that Sym is a subgroup of QS. Moreover, it was proved in [24] that Sym is the characteristic topological subgroup of the partial topological group QS for which the neighborhood base is given at id by using the quasisymmetry constant and is distributed at every point g QS by the right translation. In the rest of this section, we review the Teichmüller∈ space of symmetric homeomorphisms, which is already well-known in the theory of asymptotic Teichmüller spaces. This will be a prototype of our construction of the Teichmüller space of circle diffeomorphisms.

Definition. The little subspace T0 of the universal Teichmüller space T (or the Teichmül- ler space of symmetric homeomorphisms) is defined as

T = M¨ob(S) Sym T = M¨ob(S) QS . 0 \ ⊂ \

We define the subset Bel0(D) of Bel(D) consisting of all Beltrami coefficients vanishing at the boundary. As Möb(D) AC(D) can be identified with Bel0(D), Corollary 3.4 implies that the image of Bel (D) under\ the Teichmüller projection π : Bel(D) T is T . This also 0 → 0 implies that its Bers embedding β(T0) coincides with Φ(Bel0(D)) for the Bers projection ∗ Φ : Bel(D) B(D ). Under the group structure of Bel(D), Bel0(D) is a subgroup. Correspondingly,→ T is a subgroup of (T, ). In fact,∗T T is a topological subgroup as 0 ∗ 0 ⊂ T0 ∼= Sym QS∗ and Sym is a topological subgroup. It was proved∩ by Earle, Markovic, and Saric [20] that the Douady–Earle extension eDE(g) of a symmetric homeomorphism g Sym is asymptotically conformal; eDE : Sym AC(D) is a section of q : AC(D) Sym.∈ Hence, the conformally natural section → → sDE : T Bel(D) sends T0 to Bel0(D). We note that Bel0(D) is the unit ball of the Banach → ∞ D ∞ D subspace L0 ( ) L ( ) consisting of bounded measurable functions vanishing at the boundary: Bel (D⊂) = Bel(D) L∞(D). In particular, Bel (D) is contractible. Therefore, 0 ∩ 0 0 T0 is also contractible. ∗ To consider the complex structure of T0, we introduce the Banach subspace B0(D ) of B(D∗) as follows:

∗ ∗ −2 B0(D )= ϕ B(D ) lim ρD∗ (z) ϕ(z) =0 . { ∈ | |z|→1 | | }

∗ An element in B0(D ) is also called vanishing at the boundary. The following theorem, which was essentially proved by Becker and Pommerenke [12], can be found in [24].

Theorem 3.5. For the Bers projection Φ : Bel(D) B(D∗), it holds that → Φ(Bel (D)) = β(T ) B (D∗). 0 ∩ 0 ∗ By this theorem, we have β(T0)= β(T ) B0(D ). Hence, T0 is identiﬁed with a bounded ∩ ∗ contractible domain of the complex Banach space B0(D ). TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 13

4. The decay order of Schwarzian and pre-Schwarzian derivatives

We focus on the decay order of a Beltrami coefficient µ Bel0(D) vanishing at the boundary S. We define ∈ κ (t) = ess.sup µ(ζ) (0 < t 1) µ 1−t≤|ζ|<1 | | ≤ for µ Bel0(D), which satisfies κµ(t) 0 as t 0. Let α (0, 1) be a fixed constant. For a∈ Beltrami coefficient µ Bel (D),→ we define→ a new norm∈ by ∈ 0 α µ = ess.sup D ρD(ζ) µ(ζ) . k k∞,α ζ∈ | | Clearly, µ < if and only if κ (t)= O(tα) (t 0). k k∞,α ∞ µ → Definition. Let α be a constant with 0 <α< 1. The space of Beltrami coefficients µ Bel(D) with µ < is denoted by Belα(D). ∈ k k∞,α ∞ 0 As in the definition of the Bers projection, we extend a Beltrami coefficient µ Belα(D) ∈ 0 to C by setting µ(z) 0 for z D∗ and take a quasiconformal homeomorphism f : C C ≡ ∈ µ → having the complex dilatation µ. Then, f D∗ is a conformal homeomorphism (univalent b µ| b b function). Hereafter, we always give the following normalization for fµ: ′ fµ( )= ; lim fµ(z)=1. ∞ ∞ z→∞ Equivalently, the Laurent expansion of f at is ν ∞ b f (z)= z + b + 1 + . µ 0 z ··· We consider its pre-Schwarzian derivative and Schwarzian derivative on D∗, which are defined respectively as follows: ′′ fµ (z) Tfµ|D∗ (z)= ′ ; fµ(z) ′ 1 2 S ∗ (z)= T ∗ (z) T ∗ (z) . fµ|D fµ|D − 2 fµ|D It was shown in Becker and Pommerenke [12] that the condition µ Bel0(D) is equivalent to each of the conditions ∈ −1 −2 ∗ ∗ lim ρD (z) Tfµ|D∗ (z) = 0; lim ρD (z) Sfµ|D∗ (z) =0. |z|→1 | | |z|→1 | |

To estimate their decay order quantitatively in terms of κµ(t), we set

βµ(t) = max ( z 1) Tfµ|D∗ (z) , |z|=1+t | | − | | 2 σµ(t) = max ( z 1) Sfµ|D∗ (z) (0

for any ε> 0. We note that the above definitions of βµ and σµ are slightly different from those in [11]. We will improve these estimates regarding the power of t for the case where κµ(t) = O(tα) (t 0). In this case, the elimination of the constant ε was done by Dyn′kin [18]. Our improvement→ can be stated as follows. Theorem 4.1. For every α (0, 1), there is a constant C = C(α) > 0 that depends only on α such that ∈ −1 α ρ ∗ (z) T ∗ (z) C µ ( z 1) D | fµ|D | ≤ k k∞,α | | − for every µ Belα(D) and for every z D∗. Equivalently, ∈ 0 ∈ 2tα β (t) C µ µ ≤ k k∞,α t +2 for every t> 0. α D We decompose a Beltrami coefficient µ Bel0 ( ) suitably into a finite number of Bel- trami coefficients whose supports are in mutually∈ disjoint annular domains of D. Then, a computation of the pre-Schwarzian derivative of the composition of the corresponding conformal homeomorphisms establishes the estimate. These steps are given in the following two lemmata. Lemma 4.2. For every α (0, 1), there is a constant λ with 0 <λ< 1 that depends only on α such that, if a sequence∈ s ∞ of positive numbers satisfies a recurrence relation { n}n=0 1 α n sn = λ 1+ sn−1 for every n 1 and s =1, then s is increasing and diverges to + . ≥ 0 { n} ∞ Proof. The recurrence relation is equivalent to n 1 sn = λ α (1 + sn−1) α

for every n 1 and s0 = 1. For comparison with this formula, we consider another recurrence relation≥ n 1 ′ α ′ α sn = λ sn−1 ′ 1/α ′ for every n 2 by giving the initial value s1 = s1 = (2λ) . It is easy to see that sn sn ≥ ′ ≥ for every n 1, and hence, limn→∞ sn = + implies limn→∞ sn = + . Moreover, if ′ ≥ ∞ ∞ sn is increasing then so is sn . { } ′ ′ { } Let bn = sn+1/sn. Then, we have 1 1 bn = λ α (bn−1) α for every n 2 and 2 1 ≥ ′ 2 s2 λ α (2λ) α b1 = ′ = 1 . s1 (2λ) α TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 15

Taking the logarithm yields 1 1 log b = log b + log λ n α n−1 α with 1 1 1 1 log b = + log λ + log 2. 1 α2 α α2 − α This shows that if log λ log b > − , 1 1 α − then log bn are positive and uniformly bounded away from 0 for all n 1. By choosing λ < 1 such that it is sufficiently close to 1, we have such a situation.≥ For instance, λ (1−α)2/(1+α+α2) ′ can be chosen so that λ > (1/2) . This proves that sn is increasing and diverges to + . { } ∞ Lemma 4.3. For a finite sequence of real numbers 1= r >r >r > >r >r =0, −1 0 1 ··· N N+1 let A = ζ D r > ζ r be an annulus (or a disk) in D for each n = n { ∈ | n | | ≥ n+1} 1, 0,...,N. For any µ Bel(D) and each n, we define a Beltrami coefficient on C by − ∈ µ(ζ) (ζ An) b µ (ζ)= ∈ n 0 (ζ C A ). ∈ − n Let k = µ . Then, the pre-Schwarzian derivativeb of f D∗ satisfies n k nk∞ µ| N knrn T ∗ (z) 12 | fµ|D | ≤ z 2 r2 nX=−1 | | − n for every z D∗. ∈ Proof. First, we take a quasiconformal self-homeomorphism fN of C (namely, that of C fixing ) having the complex dilatation µ , and consider the push-forward µ = ∞ N N−1 (fN )∗µN−1 of µN−1 by fN , which is conformal on AN−1. Here, the push-forward f∗µ of µb Bel(D) by a conformal homeomorphism f of a domain D is defined in generale by ∈ (f −1)′(z) (f µ)(z)= µ(f −1(z)) (z f(D)). ∗ (f −1)′(z) ∈

Next, we take a quasiconformal self-homeomorphism fN−1 of C having the complex dilatation µ and the push-forward µ = (f f ) µ . Inductively, for each n 0, N−1 N−2 N−1 ◦ N ∗ N−2 ≥ let fn be a quasiconformal self-homeomorphism of C whose complex dilatation is µn and let e e µ =(f f ) µ e n−1 n ◦···◦ N ∗ n−1 e 16 K. MATSUZAKI

be the push-forward of µn−1 by fn fN . Finally, we choose a quasiconformal self- homeomorphism f of C with the complex◦···◦ dilatation µ so that f f coincides −1 −1 −1 ◦···◦ N with fµ. By the chain rule of pre-Schwarzian derivatives, we seee that

Tfµ|D∗ (z) ′ ′ = T (z)+ T (f (z))f (z)+ + T −1 (f f (z))(f f ) (z) fN fN−1 N N ··· f 0 ◦···◦ N 0 ◦···◦ N N−1 = T (z)+ T (f f (z))(f f )′(z) fN fn n+1 ◦···◦ N n+1 ◦···◦ N nX=−1 for every z D∗. Here, we∈ use the following estimates for the pre-Schwarzian derivative. For any conformal homeomorphism f of D∗ with f( )= , it was shown in Avhadiev [9] (cf. Theorem 4.2.3 in Sugawa [39]) that ∞ ∞

2 −1 z 1 ρ ∗ (z) T (z) | | − z T (z) 3 ( z > 1). D | f | ≤ 2 | f | ≤ | | In addition, if f extends to a quasiconformal self-homeomorphism of C of complex dilatation µ with µ k, then the majorant principle as described in Section II.3.5 of Lehto k k∞ ≤ b ∗ ∗ [28] yields that Tf (z) 3kρD (z). Moreover, for any simply connected domain Ω C containing and| for| any ≤ conformal homeomorphism f of Ω∗ with f( ) = , we⊂ see ∞ ∗ ∞ ∞ ∗ b that Tf (ω) 6ρΩ∗ (ω) for ω Ω , where ρΩ∗ (ω) is the hyperbolic density on Ω . This stems| from| the ≤ chain rule of∈ pre-Schwarzian derivatives and the invariance of a hyperbolic metric (see Theorem 1 in Osgood [36]). Again, if this extends to a quasiconformal homeomorphism of C with µ k, then T (ω) 6kρ ∗ (ω). k k∞ ≤ | f | ≤ Ω The conformal homeomorphism f of the disk Ω∗ = z > r into C with b N N N f ( )= satisﬁes {| | } ∪ {∞} N ∞ ∞ b 6k r T (z) N N . | fN | ≤ z 2 r2 | | − N ∗ C The conformal homeomorphism fn of the quasidisk Ωn into with fn( ) = for ∗ ∞ ∞ 1 n N 1, where Ωn is the image of the disk z >rn under fn+1 fN , satisﬁes− ≤ ≤ − {| | }∪{∞}b ◦···◦ T (ω) 6k ρ ∗ (ω) | fn | ≤ n Ωn ∗ ∗ ∗ for every ω Ωn in terms of the hyperbolic density ρΩn (ω) of Ωn. Hence, by replacing ω with f ∈ f (z), we obtain n+1 ◦···◦ N T (f f (z))(f f )′(z) | fn n+1 ◦···◦ N n+1 ◦···◦ N | ′ 12knrn 6k ρ ∗ (f f (z)) (f f ) (z) = . ≤ n Ωn n+1 ◦···◦ N | n+1 ◦···◦ N | z 2 r2 | | − n TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 17

This gives the desired inequality N knrn T ∗ (z) 12 | fµ|D | ≤ z 2 r2 nX=−1 | | − n for every z D∗. ∈ Proof of Theorem 4.1. For any µ Belα(D), let ℓ = µ < . Then, ∈ 0 k k∞,α ∞ α κµ(t) = sup µ(ζ) ℓt (0 < t 1). 1−t≤|ζ|<1 | | ≤ ≤ ∗ −1 D ∗ Fixing z , we will estimate ρD (z) Tfµ|D∗ (z) in terms of ℓ. In the case of z 2, we can ∈ | | −1 | | ≥ ∗ easily obtain the desired estimate. Indeed, by the inequality ρD (z) Tfµ|D∗ (z) 3 µ ∞ as in the proof of Lemma 4.3 and by µ µ , we obtain | | ≤ k k k k∞ ≤k k∞,α −1 α ρ ∗ (z) T ∗ (z) 3 µ 3 µ ( z 1) . D | fµ|D | ≤ k k∞,α ≤ k k∞,α | | − Hence, we may assume that 1 < z < 2. Let τ = z 1 (0, 1). | | | | − ∈ We choose t0 = τ and inductively deﬁne a sequence tn n≥1 of positive numbers by a recurrence relation { } τ α n α ℓtn = λ ℓτ τ + tn−1 · · for some constant λ with 0 <λ< 1. If we set sn = tn/τ, this is equivalent to

1 α n sn = λ 1+ sn−1

with the initial condition s0 = 1. Then, by Lemma 4.2, we can ﬁnd the constant λ = λ(α) so that the sequence s , and hence t are increasing and diverge to + . In particular, { n} { n} ∞ there is the smallest non-negative integer N 0 such that tN+1 1. By using the positive numbers t N , we≥ set r = 1 t . We≥ also set r = 1 and { n}n=0 n − n −1 rN+1 = 0. Then, as in Lemma 4.3, we divide D into the annuli (or the disk) A = ζ D r > ζ r (n = 1, 0,...,N) n { ∈ | n | | ≥ n+1} − α α and deﬁne kn = µn ∞ for µn = µ 1An . Because κµ(t) ℓt , we see that kn ℓtn+1 . k k · ≤ α ≤ We note that for n = N, this is valid as µ ∞ ℓ ℓtn+1 . Now, the application of Lemma 4.3 yields k k ≤ ≤ N N −1 2 knrn τ α ρ ∗ (z) T ∗ (z) 6( z 1) 6 ℓt . D | fµ|D | ≤ | | − z 2 r2 ≤ τ + t · n+1 nX=−1 | | − n nX=−1 n Here, the recurrence relation for t shows that the last sum is taken for λn+1 ℓτ α. Thus, { n} · −1 6ℓ α ρ ∗ (z) T ∗ (z) τ , D | fµ|D | ≤ 1 λ − where λ depends only on α. By taking C =6/(1 λ), we obtain the desired inequality. − 18 K. MATSUZAKI

Next, we consider the relation between Tf and Sf for a conformal homeomorphism f of D∗. It is known that there is some absolute constant A> 0 such that −2 −1 ∗ ρ ∗ (z) S (z) Aρ ∗ (z) z T (z) (z D ). D | f | ≤ D | f | ∈ (see Lemma 6.1 in Becker [10]). This in particular implies the following: Proposition 4.4. If β (t)= O(tα) then σ (t)= O(tα) (t 0). µ µ → Remark. Lemmata 4.2 and 4.3 can be easily modified so that they are suitable for estimation of Schwarzian derivatives. Hence, inequalities α −2 ′ α ′ 4t ρ ∗ (z) S ∗ (z) C µ ( z 1) ; σ (t) C µ D | fµ|D | ≤ k k∞,α | | − µ ≤ k k∞,α (t + 2)2 for some C′ = C′(α) > 0 can also be derived directly from these modifications in the same α way as in the proof of Theorem 4.1. The condition σµ(t)= O(t )(t 0) is equivalent to −2+α → sup D∗ ρ ∗ (z) S ∗ (z) < . z∈ D | fµ|D | ∞ α α Finally, we will show that σµ(t) = O(t ) implies that κµ′ (t) = O(t ) (t 0) for some µ′ Bel(D) with π(µ) = π(µ′). This is a consequence of the next lemma,→ which can be∈ found in Theorem 5.4 of Becker [10]. We note that the condition π(µ) = π(µ′) is ′ equivalent to f D∗ = f ′ D∗ for µ,µ Bel(D). µ| µ | ∈ Lemma 4.5. Let f be a conformal homeomorphism of D∗ having a quasiconformal ex- ∗ tension to C such that ϕ = Sf belongs to B0(D ). We set (z∗ z)f ′(z∗) b F (z)= f(z∗) − − 1+(z∗ z)f ′′(z∗)/(2f ′(z∗)) − for z D, where z∗ = 1/z¯ is the reflection of z with respect to S. Then, there is some ∈ t > 0 such that f extends to a quasiconformal self-homeomorphism of C that coincides with F on the annulus 1 t< z < 1 having the complex dilatation { − | | } b ¯ ∂F (z) −2 ∗ ∗ 2 ∗ µ (z)= = 2ρ ∗ (z )(zz ) ϕ(z ). F ∂F (z) − D Theorem 4.1, Proposition 4.4, and Lemma 4.5 conclude the equivalence of all the conditions above. Theorem 4.6. The following conditions are equivalent for µ Bel(D) and α (0, 1): α ′ ∈ ′ ∈ (1) κµ′ (t)= O(t ) (t 0) for some µ Bel(D) with π(µ)= π(µ ); α → ∈ (2) βµ(t)= O(t ) (t 0); (3) σ (t)= O(tα) (t → 0). µ → The above results can also be proved when we exchange the role of D and D∗. We will briefly mention this fact. For any Beltrami coefficient µ Bel(D), we define its reflection by ∈ µ∗(z)= µ(z∗)(zz∗)2 Bel(D∗). ∈ TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 19

This coincides with the complex dilatation of the reﬂection of f µ : D D with respect to S. If µ Belα(D) and µ = ℓ< , then µ∗ satisﬁes → ∈ 0 k k∞,α ∞ z 2 1 α µ∗(z) = µ(z∗) ℓ | | − ℓ ( z 1)α (z D∗); | | | | ≤ 2 z 2 ≤ | | − ∈ | | ∗ α κµ∗ (t) = sup µ (z) ℓt (0

¯ ∗ βµ (t) = max (1 ζ ) Tf ∗|D (ζ) (0 < t 1). |ζ|=1−t −| | | µ | ≤ We can modify Lemma 4.3 appropriately by using the corresponding estimates of pre- Schwarzian derivatives on D and any simply connected domain Ω C: ⊂ T (ζ) 3ρD(ζ) (ζ D); T (ω) 4ρ (ω) (ω Ω). | f | ≤ ∈ | f | ≤ Ω ∈ Concerning the relation between Tf and Sf for a conformal homeomorphism f of D, there is some absolute constant A′ > 0 such that ρ−2(ζ) S (ζ) A′ρ−1(ζ) T (ζ) (ζ D). D | f | ≤ D | f | ∈ (see pp.117–119 of [11] and Sections 4.2 and 5.3 of [39]). Thus, the statement corresponding to Proposition 4.4 holds true also in this case. Moreover, the interior version of Lemma 4.5 is given in Theorem 3 of [11]. Therefore, the statements that correspond to Theorems 4.1 and 4.6 are also valid in this case; in particular, we record the following claim as a corollary for later use. Corollary 4.7. For every α (0, 1), there is a constant C′ = C′(α) > 0 depending only ∈ α on α such that β¯ ∗ (t) C′ µ tα for every µ Bel (D) and for every t (0, 1]. µ ≤ k k∞,α ∈ 0 ∈ 5. Holder¨ continuity of derivatives and quasisymmetry quotients We define a class of orientation-preserving diffeomorphisms of the circle with Hölder continuous derivatives, which is of importance in our theory of Teichmüller spaces. In this section, we investigate the topology of the space of such circle diffeomorphisms. In particular, we relate this topology to the quasisymmetry quotients and the dilatations of their quasiconformal extensions. Definition. An orientation-preserving diffeomorphism g : S S belongs to the class 1+α S → Diff+ ( ) for exponent α (0, 1) if its derivative is α-Hölder continuous. This means that the lift g : R R of g∈under the universal cover R S satisfies → → g′(x) g′(y) c x y α (x, y R) e | − | ≤ | − | ∈ for some c 0. ≥ e e 20 K. MATSUZAKI

1+α S 1+α We provide the right uniform topology for Diff+ ( ). This is induced by the C - 1+α S modulus p1+α, which measures the difference between an element g Diff+ ( ) and the identity as follows: ∈ ′ p1+α(g) = sup g(ξ) ξ + sup g (x) 1 + cα(g), ξ∈S | − | 0≤x<1 | − | where e g′(x) g′(y) c (g) = sup | − |. α x y α 0<|x−y|≤1/2 e | − e| Then, g converge to g in Diff1+α(S) by definition if p (g g−1) 0 as n . n + 1+α n ◦ → →∞ 1+α S 1+α Remark. The right uniform topology on Diff+ ( ) as above is different from the C - topology given in Herman [25]. 1+α S We first verify that the neighborhood base at id Diff+ ( ) is compatible with the 1+α S ∈ group structure. In other words, Diff+ ( ) is a partial topological group in the sense of Gardiner and Sullivan [24].

1+α Proposition 5.1. The C -modulus p1+α satisﬁes the following:

(1) If p1+α(gn) 0 and p1+α(hn) 0 as n then p1+α(gn hn) 0; (2) If p (g ) → 0 as n then→p (g−→∞1) 0. ◦ → 1+α n → →∞ 1+α n → ′ Proof. (1) It is obvious that g h id and (g^h )′(x) = g ′(h (x))h (x) 1 n ◦ n → n ◦ n n n n → uniformly. Concerning the convergence of cα, we have e f f (g^h )′(x) (g^h )′(y) | n ◦ n − n ◦ n | x y α | − | ′ ′ ′ ′ ′ ′ ′ ′ gn (hn(x))hn (x) gn (hn(y))hn (x) gn (hn(y))hn (x) gn (hn(y))hn (y) | − α | + | − α | ≤ f f x y f f f f x y f f e | − e| ′ e | − e| c (g ) h (x) h (y) α h (x) α n | n − n | | n | + g ′(h (y)) c (h ). ≤ x y α | n n | α n f | −f| f ′ ′ e f As cα(gn),cα(hn) 0 and gn (x), hn (x) 1 uniformly, we see that cα(gn hn) 0 as n . → → ◦ → →∞(2) It is obvious that g−1e idf and (g−1)′(x)=1/(g ′(g−1(x)) 1 uniformly. Con- n → n n n → cerning the convergence of cα, we have g e g (g−1)′(x) (g−1)′(y) g ′(g−1(x)) g ′(g−1(y)) | n − n | = | n n − n n | x y α x y α g ′(g−1(x)) g ′(g−1(y)) g | − |g | −e | g| n n e|| gn n | −1 −1 α cα(gn) gg(x) g (gy) e| n − en | . ≤ x y α g ′(g−1(x)) g ′(g−1(y)) | − | | n gn ||gn n | e g e g TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 21

′ −1 ′ −1 As cα(gn) 0 and gn (x), (gn ) (x) 1 uniformly, we see that cα(gn ) 0 as n . → → → → ∞ e g 1+α S In fact, we see more: Diff+ ( ) is a topological group. 1+α S Proposition 5.2. With respect to the right uniform topology, Diff+ ( ) is a topological group. Proof. According to Lemma 1.1 in [24], we have only to show that the adjoint map is −1 continuous at id; if p1+α(gn) 0 as n , then p1+α(h gn h ) 0 for every h Diff1+α(S). We have that h→ g h−1→ ∞id and ◦ ◦ → ∈ + ◦ n ◦ → ′ ^−1 ^ −1 ′ h (gn h (x)) ′ −1 (h gn h ) (x)= ◦ gn (h (x)) 1 ◦ ◦ e h′(h−1(x)) → e g uniformly. Furthermore, e g (h ^g h−1)′(x) (h ^g h−1)′(y) | ◦ n ◦ − ◦ n ◦ | x y α | − | ′ ^−1 ′ ^−1 h (gn h (x)) ′ −1 h (gn h (y)) ′ −1 −α = ◦ gn (h (x)) ◦ gn (h (y)) x y , h′(h−1(x)) − h′(h−1(y)) ·| − | e e e g e g which is uniformly e g asymptotic to e g g ′(h−1(x)) g ′(h−1(y)) c (g ) h−1(x) h−1(y) α | n − n | α n | − | . x y α ≤ x y α e g | − e| g g| − | g Because c (g ) 0, we see that c (h g h−1) 0 as n . α n → α ◦ n ◦ → →∞ 1+α S As every circle diffeomorphism is symmetric, Diff+ ( ) is a subgroup of Sym. We will 1+α S characterize an element g of Diff+ ( ) in terms of the quasisymmetry quotient of g. This was shown in Lemma 5 in Carleson [15] (see also Section 9 of Gardiner and Sullivan [24]). The following statement and a detailed proof can be found in Theorem 7.1 of [30] and its corollary. Theorem 5.3. For a fixed α (0, 1), we assume that there is some b 0 such that the lift g of g Sym satisfies ∈ ≥ ∈ α −1 α (1 + bt ) me(x, t) 1+ bt e ≤ g ≤ 1+α S for every x [0, 1) and every t (0, 1/2]. Then, g belongs to Diff+ ( ), and cα(g) depends only∈ on b and tends to 0∈uniformly as b 0. Moreover, g′(x) is uniformly bounded from above and away from 0 by constants depending→ only on b with α fixed, which tend to 1 as b 0. e → 1+α S Conversely, every element g Diff+ ( ) (α (0, 1)) belongs to Sym with a gauge function for symmetry of order O∈(tα). More precisely,∈ we have the following: 22 K. MATSUZAKI

Proposition 5.4. For g Diff1+α(S), there is a constant b 0 such that ∈ + ≥ α −1 α (1 + bt ) me(x, t) 1+ bt ≤ g ≤ for every x [0, 1) and every t (0, 1/2], where b can be taken depending only on c = c (g) when∈ c 1 and tends to ∈0 as c 0. α ≤ → For the proof, we need a simple claim. Proposition 5.5. Every g Diff1+α(S) satisfies ∈ + 1 c (g) < 1 c (g)(1/2)α g′(x) 1+ c (g)(1/2)α < 1+ c (g). − α − α ≤ ≤ α α 1 ′ ′ Proof. As 0 g (x)dx = 1, there exists somee x0 [0, 1] such that g (x0) 1. Likewise, ′ ′ ′ ∈ ≥ ′ there existsR some x0 [0, 1] such that g (x0) 1. The Hölder continuity of g implies that e ∈ ≤ e g′(x) g′(x ) c (g) x x α c (g)(1/2)α | − 0 | ≤e α | − 0| ≤ α e R ′ for every x withe x xe0 1/2, and the same is true for x0. Then, using the periodicity g∈′(x +1)= g| ′(x−), we| ≤have the assertion.

Proof of Propositione 5.4.e The mean value theorem asserts that there are ξ+ and ξ− such that g(x + t) g(x)= tg′(ξ ) (x<ξ < x + t); − + + g(x) g(x t)= tg′(ξ ) (x t<ξ < x). e − e− e − − − This gives e e e ′ ′ ′ ′ g (ξ+) g (ξ−) −1 g (ξ−) g (ξ+) e e mg(x, t)=1+ ′ − ; mg(x, t) =1+ ′ − . g (ξ−) g (ξ+) e e e e Here, we can see that e e g′(ξ ) g′(ξ ) c (g) ξ ξ α c (g)(2t)α | + − − | ≤ α | + − −| ≤ α by the Hölder continuity of g′. Proposition 5.5 gives the lower estimate of g′. Moreover, e e ′ as g is a diffeomorphism, there is some c0 > 0 depending on g such that g (x) c0. Therefore, e e ≥ α ±1 2 cα(g) α e e mg(x, t) 1+ α t . ≤ max 1 cα(g)(1/2) ,c0 α { − } α We set the coefficient of t as b. If cα(g) 1, then 1 cα(g)(1/2) > 0 and b depend only on c = c (g). Moreover, b 0 as c 0.≤ − α → → 1+α S ±1 α Now we see that g Diff+ ( ) if and only if mge(x, t) = 1+O(t )(t 0). Hereafter, we use a constant ∈ → ǫ mge(x, t) 1 bα(g) = sup max α − . 0≤x<1, 0

to be quantitative as c (g) 0 (and so on) if there is a majorant of b (g) in terms of α → α cα(g). 1+α S Corollary 5.6. For g Diff+ ( ), we have that cα(g) 0 if and only if bα(g) 0 quantitatively. Moreover,⊂ under the extra assumption that →g is normalized so that it→ fixes S the three points on (g QS∗), p1+α(g) 0 if and only if bα(g) 0 or cα(g) 0 quantitatively. ∈ → → → Proof. The first statement directly follows from Theorem 5.3 and Proposition 5.4. For the second statement, we have only to show that bα(g) 0 or cα(g) 0 implies p1+α(g) 0 quantitatively under the normalization. Theorem→ 5.3 or Proposition→ 5.5 verifies that→g′ converge to 1 uniformly. Moreover, as M(g) 1+ bα(g) 1 and g are normalized, g converge to id uniformly. Hence, we obtain p ≤ (g) 0. → e 1+α → 1+α S Finally, in this section, we prepare the investigation of Diff+ ( ) by the quasiconformal D 1+α S extension to . This will be completed in the next section. We recall that, as Diff+ ( ) Sym, there is a quasiconformal extension that is asymptotically conformal. We look at⊂ the decay order of its complex dilatation close to the boundary. 1+α S Theorem 5.7. For every g Diff+ ( ), there exists a quasiconformal extension f D ∈ α D ∈ AC( ) of g whose complex dilatation µ belongs to Bel0 ( ). Here, µ ∞,α tends to 0 quantitatively as b (g) 0 or c (g) 0. k k α → α → ±1 α Proof. By Proposition 5.4, the lift g : R R of g satisfies me(x, t) 1+ b (g)t for → g ≤ α a finite constant bα(g) 0. Then, by Theorem 3.1, the complex dilatation µF (z) of the ≥ α Beurling–Ahlfors extension F (z) ofeg satisfies µF (z) 4bα(g)y for every z = x + iy H. The projection f : D 0 D 0 of |F under| ≤ the holomorphic universal cover∈ u : H D 0 (z ζ =−{e2πiz}) → is definede −{ } as e (g) after filling 0. → −{ } 7→ BA The complex dilatation µ of f = eBA(g) satisfies µ(ζ) = µ (z) = µ ((log ζ)/(2πi)) | | | F | | F | α for every ζ D. As Im [(log ζ)/(2πi)] = log ζ /(2π), the condition µF (z) 4bα(g)y yields ∈ − | | | | ≤ 4b (g) µ(ζ) α ( log ζ )α. | | ≤ (2π)α − | | As log ζ is comparable to 1 ζ near ζ = 1, we can find a continuous increasing function− d| :| [0, 1) [1, ) with−| lim | d(t)| =| 1 such that → ∞ t→0 4b (g) µ(ζ) α d( µ )(1 ζ )α | | ≤ (2π)α k k∞ −| | for every ζ D. Moreover, if c (g) 0, then b (g) 0 by Proposition 5.4, and hence, ∈ α → α → M(g) 1 which implies that µ ∞ 0 (see Theorem I.5.2 in [28]). Therefore, we see that µ→ 0 quantitativelyk ask b (→g) 0 or c (g) 0. k k∞,α → α → α → 24 K. MATSUZAKI

6. Quasiconformal characterization of circle diffeomorphisms We will establish the relationships among the following three indices quantitatively: the exponent of Hölder continuity of the derivative of a circle diffeomorphism g; the decay order of the complex dilatation of quasiconformal extension of g; and the decay order of the Schwarzian derivative of the corresponding conformal homeomorphism. We have seen the equivalence of the last two quantities (Theorem 4.6) and the implication of the second one from the first (Theorem 5.7). The new addition is the converse of the statement of Theorem 5.7. In Theorem 3.2 and Corollary 3.4, we have seen that an asymptotically conformal homeomorphism f AC(D) extends to a symmetric homeomorphism g Sym and provided a certain estimate∈ of the ∈ gauge function for symmetry in terms of the decay order of µf . The order of the gauge function and the Hölder continuity of the derivative are related to each other as shown in Theorem 5.3 and Proposition 5.4. However, the order of the gauge function is reduced to α/2 from the decay order α of µf according to Theorem 3.2. Moreover, in the course of transforming the situation from H to D, we need a certain normalization on g Sym to obtain a quantitative estimate. A summary of these situations is the following:∈ Lemma 6.1. For a K-quasiconformal self-homeomorphism f of D with complex dilatation µ Belα(D), its boundary extension g belongs to Diff1+α/2(S). In addition, under the ∈ 0 + normalization such as f(0) = 0 or g QS∗, the derivative of g is uniformly bounded from above and away from 0. More precisely,∈ there is a constant D = D(α,K,ℓ) 1 depending only on α, K and ℓ with µ ℓ such that ≥ k f k∞,α ≤ 1 g′(x) D D ≤ ≤ for every x R. ∈ e Proof. We assume that f fixes 0. In this case, f lifts to the quasiconformal self-homeomorphism F of H under the holomorphic universal cover u : H D 0 (z ζ = e2πiz). The complex dilatation of F satisfies → −{ } 7→ µ (z) = µ(ζ) (1 ζ )αℓ (2π)αℓyα (z = x + iy H). | F | | | ≤ −| | ≤ ∈ Then, Theorem 3.2 is applied for ε(y) = (2π)αℓyα to verify that the quasisymmetry quotient of g : R R, which is the boundary extension of F as well as the lift of g, satisfies → e ±1 1/2 1/2 α/2 e me(x, t) 1+ cε(t )+ Rt 1+ bt g ≤ ≤ for every x [0, 1) and every t (0, 1/2], where b = b(K,ℓ) > 0 is a constant depending ∈ ∈ e 1+α/2 S only on K and ℓ. Then, Theorem 5.3 asserts that g belongs to Diff+ ( ). Moreover, the derivative g′(x) is estimated in terms of α and b by the same theorem. For a general f not necessarily fixing 0, we take φ Möb(D) such that φ f(0) = 0. The complex dilatatione of φ f is the same as that of f∈. Then, we can apply the◦ previous ◦ 1+α/2 argument to φ f; we obtain φ g Diff (S), where the same symbol φ Möb(S) ◦ ◦ ∈ + ∈ TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 25

denotes the boundary extension of φ Möb(D). This in particular shows that g itself 1+α/2 S ∈ belongs to Diff+ ( ). Moreover, if g is normalized, Proposition 6.2 below shows that f(0) r for some r = r(K) [0, 1). Then, φ satisfies | | ≤ ∈ 1 r 1+ r − φ′(z) (z D). 1+ r ≤| | ≤ 1 r ∈ − From the uniform boundedness of (φ]g)′(x) by the previous argument, we also see that g′(x) is uniformly bounded from above◦ and away from 0.

e We often compare the condition f(0) = 0 with our normalization fixing 1, i, and 1 for f = f µ QC(D). The following proposition ensures that their differences are small.− ∈ Proposition 6.2. There is a constant r = r(K) [0, 1) depending only on K such that every K-quasiconformal homeomorphism f QC(∈ D) fixing 1, i, and 1 satisfies f(0) r. ∈ − | | ≤ Proof. We assume that f QC(D) extends to the quasiconformal self-homeomorphism of ∈ C by reflection with respect to S. The distortion theorem for cross ratio due to Teichmüller (see Section III.D of [1] and [27]) implies that for any four distinct points z , z , z , z C, b 1 2 3 4 the hyperbolic distance between the cross ratios [z , z , z , z ] and [f(z ), f(z ), f(z ), f∈(z )] 1 2 3 4 1 2 3 4b in C 0, 1 is bounded by log K. We choose z1 = 0 and z2 = . If we choose two distinct−{ points} from 1, i, 1 for z and z , we see that f(0) = f(∞)∗ cannot be close to S { − } 3 4 ∞ except in some neighborhoods of z3 and z4 within a distance depending only on K. By considering all such choices from 1, i, 1 , we obtain the assertion. { − } The full converse of Theorem 5.7 should be a statement that if the complex dilatation D α D 1+α S µf of f AC( ) is in Bel0 ( ), then the boundary extension g of f belongs to Diff+ ( ) for the same∈ α. We will prove this, which is the improvement of the weaker consequence 1+α/2 S g Diff+ ( ) in Lemma 6.1. We also do this quantitatively. The claim on the derivative ∈ 1+α of g in this lemma is still necessary for the estimation of the C -modulus p1+α(g) as well as for Theorem 6.4 below. We need distortion estimates of quasiconformal self-homeomorphisms of D, which are variants of the Mori theorem. The first one is its direct consequence. Proposition 6.3. Let f be a K-quasiconformal self-homeomorphism of D with f(0) = 0. Then, 1 (1 z )K 1 f(z) 16(1 z )1/K 16K −| | ≤ −| | ≤ −| | is satisfied for every z D. ∈ Proof. The Mori theorem (see Section III.C of [1] and Theorem II.3.2 of [29]) assert that f(w) f(z) 16 w z 1/K | − | ≤ | − | for any w and z in D. We choose w = z/ z S for every z D. Then, the upper inequality follows from 1 f(z) f(w) |f(|z ∈) . Considering f∈−1, we obtain the other inequality. −| |≤| − | 26 K. MATSUZAKI

We can remove the powers 1/K and K in the inequalities of Proposition 6.3 if the α D complex dilatation belongs to our class Bel0 ( ). The following result veriﬁes this, which will be crucial in our arguments. Theorem 6.4. Let f µ be a normalized K-quasiconformal self-homeomorphism of D with α D µ Bel0 ( ) and µ ∞,α ℓ. Then, there is a constant A = A(α,K,ℓ) 1 depending only∈ on α, K, andkℓ ksuch≤ that ≥ 1 (1 z ) 1 f µ(z) A(1 z ) A −| | ≤ −| | ≤ −| | for every z D. ∈ Proof. For the moment, we prove the inequalities for f AC(D) with f(0) = 0, whose complex dilatation µ satisﬁes the same assumption as in∈ the statement. Let t = min (2ℓ)−2/α, 1/4 > 0. 0 { } It is easy to show the inequalities for z D with 1 z t0. Indeed, using Proposition 6.3, we have ∈ −| | ≥ tK tK 1 0 (1 z ) 0 (1 z )K 1 f(z) 1 t−1(1 z ). 16K −| | ≤ 16K ≤ 16K −| | ≤ −| | ≤ ≤ 0 −| |

Thus, we may assume that 1 z < t0 hereafter. Let t = 1 z < t for a−| given| point z D. We define a Beltrami coefficient µ (ζ) −| | 0 ∈ t by setting µ (ζ) = µ(ζ) on ζ D ζ 1 √t and µ (ζ) 0 elsewhere. Let f be t { ∈ | | | ≤ − } t ≡ t the quasiconformal self-homeomorphism of D with the complex dilatation µt and with ft(0) = 0. Let ht be the quasiconformal self-homeomorphism of D such that f = ht ft. As µ(ζ) ℓ(1 ζ )α, we see that µ (w) ℓtα/2 < 1/2 for w D, which implies that◦ | | ≤ −| | | ht | ≤ ∈ the maximal dilatation Kt of ht satisfies 1 1 ℓtα/2 1+ ℓtα/2 − 1 2ℓtα/2; K 1+4ℓtα/2 < 3. K ≥ 1+ ℓtα/2 ≥ − t ≤ 1 ℓtα/2 ≤ t − First, we apply a distortion theorem to the conformal homeomorphism ft(ζ) restricted to ζ > 1 √t. In fact, we may assume that f is a conformal homeomorphism of | | − t an annulus 1 √t < ζ < 1/(1 √t) by the reflection principle. Moreover, ft is also an K-quasiconformal{ − | | self-homeomorphism− } of D whose complex dilatation satisfies

µft ∞,α ℓ independently of t. Then, we see from Lemma 6.1 that there is a constant k k ≤ ′ −1 ′ D = D(α,K,ℓ) 1 independent of t such that the derivative ft satisﬁes D ft(ξ) D for every ξ ≥S. ≤| | ≤ The Koebe∈ distortion theorem (Proposition 3.3) in the disk ∆(ξ, √t) of radius √t and center ξ = z/ z yields an upper estimate | |

′ t/√t 1 ft(z) ft(z) ft(ξ) ( f (ξ) √t) 4Dt −| |≤| − | ≤ | t | (1 t/√t)2 ≤ − TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 27

if t< 1/4. A lower estimate is more complicated. Proposition 3.3 shows that 1 t/√t 4 f ′(z) f ′(ξ) − | t |≥| t |(1 + t/√t)3 ≥ 27D with t< 1/4. We consider the reﬂection z∗ of z with respect to S. The Koebe distortion theorem applied after sending z to ξ by a conformal self-homeomorphism of the disk ∆(ξ, √t) (see Corollary 1.5 in [37]) gives

∗ 2 ′ 2(t/√t) 2t ft(z) ft(z ) (1 (t/√t) )( f (z) √t) . | − | ≥ − | t | · 4(1 (t/√t)2) ≥ 27D − ∗ As ft(z ) is the reﬂection of ft(z) with respect to S, 1 ft(z) is nearly a half of ft(z) ∗ −| ∗| | − ft(z ) if it is small; for example, 1 ft(z) 9 ft(z) ft(z ) /20 if 1 ft(z) 2/11. This in| particular shows that −| | ≥ | − | −| | ≤ 9 2t t 2 1 f (z) = . −| t | ≥ 20 · 27D 30D ≤ 11

Next, we apply Proposition 6.3 to the quasiconformal self-homeomorphism ht of D. It implies that

α/2 1 h (w) 16(1 w )1/Kt 16(1 w )1−2ℓt ; −| t | ≤ −| | ≤ −| | α/2 1 Kt 1 1+4ℓt 1 ht(w) (1 w ) (1 w ) −| | ≥ 16Kt −| | ≥ 163 −| | for every w D. Then, by setting w = f (z), we have ∈ t 1 α/2 α/2 (t/(30D))1+4ℓt 1 f(z) 16(4Dt)1−2ℓt . 163 ≤ −| | ≤ Dividing these inequalities by t =1 z and taking the logarithm, we obtain −| | 1 f(z) 3 log(50D)+4ℓtα/2 log(t/(30D)) log −| | − ≤ 1 z −| | log(64D) 2ℓtα/2 log(4Dt). ≤ − This shows that the middle term is bounded from above and below independently of t, and hence (1 f(z) )/(1 z ) is also bounded from above and away from 0. Thus, we can ﬁnd a constant−| A|′ = A−|′(α,K,ℓ| ) 1 such that ≥ 1 (1 z ) 1 f(z) A′(1 z ) A′ −| | ≤ −| | ≤ −| | for the case of 1 z < t0 as well as for the previous case 1 z t0. Now we consider−| the| normalized quasiconformal homeomorphism−| | ≥ f µ QC(D). Propo- sition 6.2 asserts that there is r = r(K) [0, 1) such that f µ(0) r. We∈ take a M¨obius ∈ | | ≤ 28 K. MATSUZAKI

transformation φ Möb(D) such that φ f µ(0) = 0. Then, f = φ f µ satisfies the above inequalities. Moreover,∈ f µ(0) r implies◦ that ◦ | | ≤ 1 r 1+ r − φ′(z) (z D). 1+ r ≤| | ≤ 1 r ∈ − Because (min φ′(z) )(1 f(z) ) 1 f µ(z) (max φ′(z) )(1 f(z) ), z∈D | | −| | ≤ −| | ≤ z∈D | | −| | we can choose A = A′(1+ r)/(1 r) for the required inequalities, which depends only on α, K, and ℓ. − This theorem has several consequences. α D −1 Proposition 6.5. For any µ and ν in Bel0 ( ), the composition µ ν also belongs to α D α D D ∗ Bel0 ( ). Hence, Bel0 ( ) is a subgroup of Bel( ). Proof. We apply Theorem 6.4 to ζ = f ν(z) in the formula µ(z) ν(z) ∂f ν (z) µ ν−1(ζ)= − . ∗ 1 ν(z)µ(z) · ∂f ν (z) − α α α Then, ρD(ζ) (2A) ρD(z), from which we have ≤ (2A)α µ ν−1 µ ν . k ∗ k∞,α ≤ 1 µ ν k − k∞,α −k k∞k k∞ The statement follows from this inequality. Corollary 6.6. If ν Belα(D) then ν−1 Belα(D). More precisely, every ν Belα(D) ∈ 0 ∈ 0 ∈ 0 with ν ℓ and ν k < 1 satisfies ν−1 A ν for a constant A = k k∞,α ≤ k k∞ ≤ k k∞,α ≤ k k∞,α A(α,k,ℓ) 1. ≥ e e Proof.e As a special case of the above inequality by setting µ = 0, we have ν−1 (2A)α ν . k k∞,α ≤ k k∞,α Then, setting A = (2A)α gives the statement, as A depends only on α, k, and ℓ by Theorem 6.4. e Now we explain the converse of Theorem 5.7 as well as other equivalent conditions for g QS to belong to Diff1+α(S). We supply the following notation. ∈ + Definition. For a bounded holomorphic quadratic differential ϕ = ϕ(z)dz2 B(D∗), we define a new norm by ∈ −2+α ϕ ∞,α = sup ρD∗ (z) ϕ(z) . ∗ k k z∈D | | The Banach space of holomorphic quadratic differentials with this norm finite is given by Bα(D∗)= ϕ B(D∗) ϕ < B (D∗). 0 { ∈ |k k∞,α ∞} ⊂ 0 TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 29

Theorem 6.7. Let α be a constant with 0 <α< 1. For a quasisymmetric homeomorphism g QS, the following conditions are equivalent: ∈ 1+α S (1) g belongs to Diff+ ( ); α D (2) there is µ Bel0 ( ) such that π(µ) = [g] T ; (3) β([g]) β(∈T ) is in Bα(D∗). ∈ ∈ 0 Proof. The implication (1) (2) is a reformulation of Theorem 5.7. This was essentially proved by Carleson [15].⇒ The equivalence (2) (3) has been reviewed in Theorem 4.6, where previous contributions to this equivalence⇔ are also mentioned. We note that (1) (3) was also proved in Tam and Wan [40] by using the harmonic extension of diffeomorphisms⇒ of S. On the contrary, the converse (2) (1) was given in Dyn′kin [18] based on his results on the pseudoanalytic extension of differentiab⇒ le functions and independently in Anderson, Cantón, and Fernández [6], who relied on a certain approximation theorem of quasiconformal maps on the disk by polynomials. Theorem 6.9 below proves (2) (1) in complex analytic methods and provides necessary results for our theorems on the⇒ Teichmüller space. For later purposes, we prepare the proposition that follows next. We will use it for α D ∗ both µ Bel0 ( ) and its reflection µ . According to the different assumptions that we will impose∈ on them, we address both cases separately. Proposition 6.8. (1) Let f be a conformal homeomorphism of D∗ with f( )= and ′ ∞ ∞ limz→∞ f (z)=1 whose quasiconformal extension to D has the complex dilatation µ in Belα(D) with µ ℓ. Then, there is a constant B = B(α,ℓ) 1 such that 0 k k∞,α ≤ ≥ 1 f ′(z) B B ≤| | ≤ for every z D∗. (2) Let f be a conformal homeomorphism of D with e−s f ′(0) es ∈ D∗ ∗ ≤| α D| ≤ whose quasiconformal extension to has the complex dilatation µ for µ Bel0 ( ) with µ ℓ. Then, there is a constant B′ = B′(α,ℓ,s) 1 such that ∈ k k∞,α ≤ ≥ 1 f ′(z) B′ B′ ≤| | ≤ for every z D. ∈ Proof. (1) By Theorem 4.1, there is a constant L = L(α,ℓ) 0 such that βµ(t) 2Ltα/(t + 2). Because ≥ ≤ ′′ f (z) βµ(t) f ′(z) ≤ t

for t = z 1, the integration along the radial segment connecting (1 + t)ξ and ξ for any ξ S gives| | − ∈ (1+t)ξ d t 2tα−1 log f ′(z) dz L dt. Z dz | | ≤ Z t +2 ξ 0

30 K. MATSUZAKI

The right side term is bounded by Ltα/α, which implies that log f ′ extends continuously to S (see Theorem 4.1 in Pommerenke and Warschawski [38]). Moreover, by taking the limit as t , we obtain →∞ L ∞ L 2L log f ′(ξ) +2L tα−2dt = + | | ≤ α Z α 1 α 1 − for every ξ S. Then, the maximal principle yields that log f ′(z) 2L/(α(1 α)) for every z D∈∗. Hence, by taking B = exp(2L/(α(1 α))),| we obtain| ≤ the assertion.− ∈ ′ − ′ ′ α (2) By Corollary 4.7, there is a constant L = L (α,ℓ) 0 such that β¯µ∗ (t) L t . Because ≥ ≤ ′′ f (z) β¯µ∗ (t) f ′(z) ≤ t

for t =1 z , the integration along the radial segment connecting (1 t)ξ and ξ for any ξ S gives−| | − ∈ ξ d t L′tα log f ′(z) dz L′ tα−1dt = . Z dz | | ≤ Z α (1−t)ξ 0 ′ S Similarly to the above, log f extends continuously to . By taking t = 1, we obtain L′ log f ′(ξ) log f ′(0) | − | ≤ α for every ξ S. Then, the maximal principle yields that log f ′(z) log f ′(0) L′/α for every z ∈ D. As s log f ′(0) s, we have that log| f ′(z) −L′/α + s;| hence ≤ by taking B′ =∈ exp(L′/α−+≤s), we| obtain| ≤ the assertion. | | || ≤ α D 1+α S Theorem 6.9. If µ Bel0 ( ), then g QS with π(µ) = [g] belongs to Diff+ ( ). Moreover, if g is normalized∈ (g QS ), then∈ p (g) 0 quantitatively as µ 0. ∈ ∗ 1+α → k k∞,α → Proof. We may assume that the normalized quasiconformal self-homeomorphism f µ of D with the complex dilatation µ extends to g QS . We represent this g by conformal ∈ ∗ welding. The quasiconformal homeomorphism of C extended by the reflection of f µ with respect to S is also denoted by f µ. Let f be the normalized quasiconformal self- µ b homeomorphism of C whose complex dilatation is µ on D and 0 on D∗, which satisfies ′ fµ( ) = and limz→∞ fµ(z) = 1. We define the quasiconformal self-homeomorphism ∞ µ −∞1 b fµ (f ) of C by f, which is conformal on D with f(D) = fµ(D) and whose complex ◦ ∗ ∗ −1 −1 dilatation on D is (µ ) , the inverse of the reflection of µ. Then, g = f fµ on S. We note that (µ∗)−b1 =(µ−1)∗, where µ−1 belongs to Belα(D) and µ−1 can◦ be estimated 0 k k∞,α in terms of µ ∞,α by Corollary 6.6. We will estimatek k the modulus of continuity of the derivative of g : S S at e2πix S ¯ ∗ −1 → ∈ in terms of βµ and β(µ ) . This is based on an argument given by Anderson, Becker, and α Lesley [5]. By Theorem 4.1, we see that βµ(t) Lt for some constant L 0 tending ≤ ¯ ∗ −1 ≥ ′ α to 0 uniformly as µ ∞,α 0. By Corollary 4.7, we also have that β(µ ) (t) L t for ′ k k → −1≤ some constant L 0 with the same property as L; if µ ∞,α 0, then µ ∞,α 0, and hence L′ 0≥ uniformly. k k → k k → → TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 31

Now we consider the derivative of the lift g : R R at x R represented by → ∈ g(e2πi(x+s)) g(e2πix) g′(x) = lim −e = g′(e2πix) , s→0 e2πi(x+s) e2πix | |

− where g′(e2πix) is thee directional derivative along the tangent of S at e2πix. We see that g is continuously diﬀerentiable and ′ 2πix ′ 2πix ′ 2πix g (e )=(fµ) (e )/f (g(e )). ′ ∗ Indeed, as in the proof of Proposition 6.8, if µ ∞,α < , then (fµ) (z) (z D ) has a non-vanishing continuous extension to S = ∂ Dk∗.k This is∞ also true for f ′(z) (z∈ D). As g is normalized, Lemma 6.1 asserts that g′(x) D for a constant D 1 uniformly∈ bounded ≤ ≥ when µ ∞,α 0. Thek modulusk → of continuity of g′ is deﬁnede by ′ ′ ′ I(te; g ) = sup g (x) g (y) |x−y|≤t | − | for every t (0, 1/2]. We note thate e e ∈ I(t; g′) c (g) = sup . α tα 0 0, then | − | ≤ | − | ≥ e g′e(y) g′(y) e e g′(x) g′(y) D 1 D log . | − | ≤ − g′(x) ≤ g′(x) e e ′ ′ ′ ′ This yields I(t; g ) eDI(t; loge g ). The case where g (y) g (x) > 0 deduces the same estimate. Moreover,≤ e ≥ e e e e e I(t; log g′) I(t; log f ′ (e2πi •) )+ I(t; log f ′(g(e2πi •)) ) ≤ | µ | | | I(t; log f ′ (e2πi •) )+ I(Dt; log f ′(e2πi •) ). e ≤ | µ | | | ′ ′ ′ ′ Here, we note that (log fµ) (z) = Tfµ|D∗ (z) and (log f ) (z) = Tf|D (z). Taking a path of integration including the circular arc γ joining e2πix(1 + t) and e2πiy(1 + t) in D∗ for e2πix, e2πiy S with x y t, we obtain ∈ | − | ≤ log f ′ (e2πix) log f ′ (e2πiy) log f ′ (e2πix) log f ′ (e2πiy) | | µ | − | µ ||≤| µ − µ | e2πix(1+t) e2πiy

Tfµ|D∗ (z) dz + Tfµ|D∗ (z) dz + Tfµ|D∗ (z) dz ≤ Ze2πix | || | Zγ | || | Ze2πiy(1+t) | || | t βµ(t) 2 dt +2π(1 + t)βµ(t). ≤ Z0 t This implies that, if β (t) Ltα, then µ ≤ I(t; log f ′ (e2πi •) ) (2/α +3π)Ltα | µ | ≤ 32 K. MATSUZAKI

for every t (0, 1/2]. The same holds for f ′; thus, ∈ I(Dt; log f ′(e2πi •) ) (2/α +3π)L′Dαtα. | | ≤ ′ α 1+α S Hence, I(t; g )= O(t ), which means that g belongs to Diff+ ( ) by definition. Under the normalization g QS∗, we have seen that D is uniformly bounded as ′ ∈ ′ α µ ∞,α e0. Because L, L 0 as µ ∞,α 0, this shows that I(t; g )/t tends to 0k uniformly,k → which means that→ c (g) k k0. Then,→ by Corollary 5.6, this implies that α → p1+α(g) 0. Thus, p1+α(g) 0 as µ ∞,α 0. All these sequences convergee quantitatively. → → k k → Condition (2) of Theorem 6.7 says that there exists some Beltrami coefficient µ α D 1+α S∈ Bel0 ( ) whose Teichmüller projection π(µ) coincides with [g] for a given g Diff+ ( ). 1+α S ∈ D Alternatively, this means that g Diff+ ( ) has some quasiconformal extension to ∈ α D whose complex dilatation belongs to Bel0 ( ). We will show here that the Douady–Earle extension actually gives such an extension provided that Theorem 6.7 is known. 1+α S Theorem 6.10. For every g Diff+ ( ), the image sDE([g]) under the conformally α ∈D natural section belongs to Bel0 ( ). Let σ : Bel(D) Bel(D) be defined by the correspondence of µ to s (π(µ)) for the → DE conformally natural section sDE. We call this the conformally natural projection on Bel(D). A crucial property of this projection is the following, which was proved by Theorem 1 in Cui [16]. Lemma 6.11. Let µ =(σ(µ−1))−1 for any µ Bel(D). Then, ∈ 2 e 2 2 2 µ(z) µ(w) C1(1 w ) | | 4 dxdy | | ≤ −| | ZD 1 wz¯ | − | for every w D, where Ce = C (k) > 0 is a constant depending only on k with µ k. ∈ 1 1 k k∞ ≤ We also need the following claim, which can be found in Lemma 3.10 of Zhu [43]. Lemma 6.12. If µ Belα(D) (α (0, 1)), then ∈ 0 ∈ 2 µ(z) 2 2α−2 | | 4 dxdy C2(1 w ) ZD 1 wz¯ ≤ −| | | − | for every w D, where C = C (k) > 0 is a constant depending only on k with µ ∈ 2 2 k k∞,α ≤ k. e e e 1+α S α D Proof of Theorem 6.10. For g Diff+ ( ), we choose ν Bel0 ( ) such that π(ν) = [g] −1 ∈ α D ∈ −1 by Theorem 6.7. Then, ν also belongs to Bel0 ( ) by Corollary 6.6. For µ = ν , we −1 −1 α D apply Lemmata 6.11 and 6.12 to show thatµ ˜ =(σ(µ )) belongs to Bel0 ( ). Again by −1 α D Corollary 6.6, this shows that σ(ν)= σ(µ ) Bel0 ( ). As σ(ν)= sDE(π(ν)) = sDE([g]), we have the assertion. ∈ TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 33

α D We can also show that the restriction of the conformally natural projection σ to Bel0 ( ) is continuous with respect to the topology induced by the norm . The detailed k·k∞,α proof has been given in [32]. To see this, we use the relation between the norm ∞,α 1+α S k·k and the right uniform topology on Diﬀ+ ( ), which will be shown in Theorem 7.8 in the next section.

7. The Teichmuller¨ space of circle diffeomorphisms We are ready to realize the Teichmüller space of circle diffeomorphisms with Hölder continuous derivatives as a subspace of the universal Teichmüller space. Then, we will give an application of the structure of this space at the end of this section. Definition. For a constant α with 0 <α< 1, the Teichmüller space of circle diffeomorphisms with α-Hölder continuous derivatives is defined by T α = Möb(S) Diff1+α(S). 0 \ + Theorem 6.7 implies that the Teichmüller projection π : Bel(D) T gives → α D α π(Bel0 ( )) = T0 , and the Bers embedding β : T B(D∗) gives → β(T α)= β(T ) Bα(D∗), 0 ∩ 0 α D D D∗ which coincides with Φ(Bel0 ( )) for the Bers projection Φ : Bel( ) B( ). Here, we α D∗ α D∗ → see that β(T ) B0 ( ) is an open subset of the Banach space B0 ( ). Indeed, this follows ∩ ∗ from the fact that β(T ) is open in B(D ), and the norm inequality ϕ ∞ ϕ ∞,α for α D∗ k k ≤ k k ϕ B0 ( ). ∈We restrict π, Φ, and β to the spaces as above and consider the continuity and openness α α D of these maps. We provide T0 with the quotient topology from Bel0 ( ) by π, which is so defined that π is continuous. Then, from the facts listed in the proof below, we are able to prove the following: α α D∗ Theorem 7.1. The Bers embedding β : T0 B0 ( ) is a homeomorphism onto the α D∗ α → image β(T ) B0 ( ). Hence, T0 is equipped with the complex structure modeled on the ∩ α D∗ complex Banach space B0 ( ). Proof. For the proof of this theorem, it suffices to show the following claims: α D α D∗ (1) Φ : Bel0 ( ) B0 ( ) is continuous; (2) Φ : Belα(D) → Φ(Belα(D)) = β(T ) Bα(D∗) has a local continuous section. 0 → 0 ∩ 0 These claims are proved in Lemma 7.3 and Lemma 7.5 below, respectively. α D We begin by showing a basic fact of the group Bel0 ( ). This is analogous to Proposition 5.1 in Yanagishita [42]. α D α D α D Proposition 7.2. The right translation rν : Bel0 ( ) Bel0 ( ) for any ν Bel0 ( ) defined by µ µ ν−1 is a homeomorphism with respect→ to . ∈ 7→ ∗ k·k∞,α 34 K. MATSUZAKI

Proof. We have the following formula for ζ = f ν(z):

µ1(z) ν(z) µ2(z) ν(z) rν(µ1)(ζ) rν(µ2)(ζ) = − − | − | 1 ν(z)µ1(z) − 1 ν(z)µ2(z) − − µ (z) µ (z) (1 ν(z) 2) = | 1 − 2 | −| | . 1 ν(z)µ (z) 1 ν(z)µ (z) | − 1 || − 2 | Here, the last term is bounded by µ (z) µ (z) (1 ν(z) 2) | 1 − 2 | −| | 2 2 2 2 ( 1 ν(z)µ1(z) µ1(z) ν(z) )( 1 ν(z)µ2(z) µ2(z) ν(z) ) q | − | −| − | | − | −| − | µ (z) µ (z) = | 1 − 2 | . (1 µ (z) 2)(1 µ (z) 2) −| 1 | −| 2 | p ν α α α By applying Theorem 6.4 to f , we have ρD(ζ) (2A) ρD(z) for some A 1. Hence, ≤ ≥ (2A)α rν(µ1) rν(µ2) ∞,α µ1 µ2 ∞,α. k − k ≤ (1 µ 2 )(1 µ 2 )k − k −k 1k∞ −k 2k∞ p α −1 −1 −1 D This shows that rν is continuous. As (rν) = rν and ν Bel0 ( ), we also see that −1 ∈ the inverse (rν) is continuous. α D We note that the right translation rν for ν Bel0 ( ) projects down to the base point α α ∈ change map Rπ(ν) : T0 T0 ; then, as rν is a homeomorphism, so is Rπ(ν). Their holomorphy will be discussed→ later in Corollary 7.7. α D α D∗ The continuity of Φ : Bel0 ( ) B0 ( ) can be proved as a special case of the assertion that follows. In contrast to the original→ case (Proposition 2.1), we need to introduce here a certain representation of a Schwarzian derivative using Beltrami coeﬃcients and estimate it by the results which we have obtained.

′ α D ′ D Lemma 7.3. Let ν Bel0 ( ) possibly with α = α (0, 1). Then, every µ Bel( ) satisﬁes ∈ 6 ∈ ∈ Φ(µ) Φ(ν) C µ ν , k − k∞,α ≤ k − k∞,α where C = C(ν,α,k) > 0 is a constant depending only on ν, α, and k with µ ∞ k. ′ k k ≤ The dependence on ν is further given by α , ν ∞, and ν ∞,α′ . The term on the right side is assumed to be when µ ν / Belα(Dk).k k k ∞ − ∈ 0 The following integral representation of Schwarzian derivatives can be found in Lemma 3.1 and Proposition 3.2 of Yanagishita [42], which is obtained by generalizing the arguments in Astala and Zinsmeister [8].

Proposition 7.4. For Beltrami coeﬃcients µ and ν in Bel(D), let fµ and fν be the normalized quasiconformal self-homeomorphisms of C that are conformal on D∗. Let b TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 35

∗ ∗ Ω= fν(D) and Ω = fν(D ). Then, 1/2 ∗ −1 −1 2 3ρΩ (ζ) µ(fν (w)) ν(fν (w)) dudv S −1 (ζ) | − | | fµ◦fν |Ω∗ | ≤ √π Z (1 µ(f −1(w)) 2)(1 ν(f −1(w)) 2) w ζ 4 Ω −| ν | −| ν | | − | holds for every ζ Ω∗. ∈

To consider the norm of the Schwarzian derivative Φ(µ)= Sfµ|D∗ , we need an estimate D∗ α D of the derivative of the conformal homeomorphism fµ of defined by µ Bel0 ( ). We use Proposition 6.8 for this purpose. ∈ Proof of Lemma 7.3. By the definition of the norm, µ(z) ν(z) ρ−α(z) µ ν | − | ≤ D k − k∞,α −α for every z D. By Theorem 6.4, there is a constant a = a(α, ν) 1 such that ρD (z) −α ν ∈ ≥ ≤ aρD (f (z)). Let f = f (f ν)−1. This is a conformal homeomorphism of D extending to a quasi- ν ◦ conformal homeomorphism of C whose complex dilatation on D∗ coincides with (ν∗)−1. We can choose fν so that f(0) = 0 maintaining the normalization fν( ) = and ′ b D D ∞ ∞ limz→∞ fν(z) = 1. We note that f( )= fν( ). If the normalization of fν appeals to the Schwarz lemma and the Koebe one-quarter ∗ theorem (Proposition 3.3) on D , we see that fν(D) is not strictly contained in D but is instead contained in the disk z < 4 . Hence, there is some x S such that 1 {| | } 1 ∈ ≤ fν(x1) 4. Furthermore, Proposition 6.2 asserts that there is some r [0, 1) depending | | ≤ −1 ν −1 ∈ only on ν ∞ = ν ∞ such that (f ) (0) r. We take z D with z = (1+ r)/2 k k k k | ν −1| ≤ ∈ | | arbitrarily and consider the cross ratio [(f ) (0), x1, , z]. By the distortion theorem for cross ratio due to Teichmüller (see Section III.D of [1]∞ and [27]), the hyperbolic distance on C 0, 1 between [(f ν)−1(0), x , , z] and −{ } 1 ∞ [f ((f ν)−1(0)), f (x ), f ( ), f (z)] = [0, f (x ), , f (z)] ν ν 1 ν ∞ ν ν 1 ∞ ν is bounded by log K, where K = (1+ ν ∞)/(1 ν ∞). This implies that there is a constant ρ = ρ( ν ) > 0 such thatk kf (z) −kρ kfor z = (1+ r)/2, and hence k k∞ | ν | ≥ | | f(D)= fν(D) contains the disk of center at 0 and radius ρ. By the Schwarz lemma applied to the conformal homeomorphism f of D, we see that there is a constant s = s(ρ) > 0 depending only on ρ and hence on ν ∞ such that e−s f ′(0) 4. It follows from Proposition 6.8 that there is a constantk Bk = B(ν) > 0 such≤| that f|′( ≤z) 1/B for every z D. Hence, there is a constant b = b(ν,α) 1 such −α |ν | ≥ −α ∈ ≥ that ρD (f (z)) bρΩ (fν (z)). For w = f (z)≤ Ω, this inequality and ρ−α(z) aρ−α(f ν(z)) yield that ν ∈ D ≤ D µ(f −1(w)) ν(f −1(w)) abρ−α(w) µ ν . | ν − ν | ≤ Ω k − k∞,α By substituting this inequality into the integral in Proposition 7.4, we will estimate ρ−2α(w) 1/2 Ω dudv . Z w ζ 4 Ω | − | 36 K. MATSUZAKI

We follow an estimation procedure similar to that in Section 3.4 of Nag [35]. Let ηΩ(w) be the Euclidean distance from w Ω to ∂Ω and ηΩ∗ (ζ) the Euclidean distance from ζ Ω∗ to ∂Ω. We see, as a consequence∈ of the Koebe one-quarter theorem (Proposition ∈ 3.3), that both ρΩ(w)ηΩ(w) and ρΩ∗ (ζ)ηΩ∗ (ζ) are bounded below by 1/2. We have ρ−2α(w) 4η2α(w) 4 w ζ 2α Ω ≤ Ω ≤ | − | for every w Ω and every ζ Ω∗. Hence, the integral can be estimated as ∈ ∈ ρ−2α(w) dudv Ω dudv 4 Z w ζ 4 ≤ Z w ζ 4−2α Ω | − | Ω | − | dudv 4 4−2α ≤ Z ∗ w ζ |w−ζ|≥ηΩ (ζ) | − | 8π 1 16π 2−2α = ρΩ∗ (ζ) . 2 2α · η ∗ (ζ)2−2α ≤ 1 α − Ω − Plugging this estimate in the inequality of Proposition 7.4, we have

12ab µ ν ∞,α −2 −1 −α ρ ∗ (ζ) S ∗ (ζ) k − k ρ ∗ (ζ). Ω | fµ◦fν |Ω | ≤ (1 α)(1 µ 2 )(1 ν 2 ) Ω − −k k∞ −k k∞ p For ζ = f (z) with z D∗, the left side term is equal to ν ∈ −2 ρ ∗ (z) S ∗ (z) S ∗ (z) . D | fµ|D − fν |D | For the right side term, we apply Proposition 6.8 again to the quasiconformal homeomor- ∗ ′ ′ phism fν of C that is conformal on D . Then, there is a constant b = b (ν,α) 1 such −α ′ −α ≥ that ρ ∗ (f (z)) b ρ ∗ (z). Therefore, the above inequality turns out to be Ω ν b ≤ D ′ −2 12abb µ ν ∞,α −α ∗ ∗ ρD (z) Sf |D∗ (z) Sf |D∗ (z) k − k ρD (z). | µ − ν | ≤ (1 α)(1 µ 2 )(1 ν 2 ) − −k k∞ −k k∞ p This implies that 12abb′ Φ(µ) Φ(ν) ∞,α µ ν ∞,α. k − k ≤ (1 α)(1 µ 2 )(1 ν 2 )k − k − −k k∞ −k k∞ p We can choose the multiplier of the term on the right side as the constant C. α D α D∗ The existence of a local continuous section for Φ : Bel0 ( ) β(T ) B0 ( ) is verified similarly to the original Bers projection Φ, for which the local→ holomor∩ phic section was defined by using the quasiconformal reflection proposed by Ahlfors [2]. This was improved later by Earle and Nag [21]. α D α D∗ Lemma 7.5. The Bers projection Φ : Bel0 ( ) β(T ) B0 ( ) has a local continuous section at every ψ β(T ) Bα(D∗). → ∩ ∈ ∩ 0 TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 37

Proof. By Theorem 6.10, we can take ν Belα(D) in the image of the conformally natural ∈ 0 projection such that Φ(ν) = Sfν |D∗ = ψ and fν D is a bi-Lipschitz diffeomorphism with respect to the hyperbolic metric (Theorem 2 of| [17]). The quasiconformal reflection λ : D D∗ S −1 ∗ fν( ) fν ( ) with respect to the quasicircle fν( ) is defined by λ(ζ) = fν(fν (ζ) ), where →z∗ denotes the reflection of z with respect to S. We follow the arguments in Section II.4.2 of [28] and Section 14.3–4 of [23]. We have a constant ε = ε(k) > 0 depending only on k with ν k such that if ϕ B(D∗) k k∞ ≤ ∈ satisfies ϕ ∞ <ε, then there is a quasiconformal self-homeomorphism f of C conformal D∗k k on fν( ) such that S b ∗ = ψ + ϕ (see also Theorem III.4.2 in [28]). In this case, the f◦fν |D b b b Beltrami coefficient µf of f is given by 2 b b ¯ Sf (λ(ζ))(ζ λ(ζ)) ∂λ(ζ) b − µf (ζ)= 2+ S b(λ(ζ))(ζ λ(ζ))2∂λ(ζ) f − for ζ f (D). Here, by the bi-Lipschitz property, we see that ∈ ν 2 2 ¯ −2 (ζ λ(ζ)) ∂λ(ζ) (ζ λ(ζ)) ∂λ(ζ) cρ D∗ (λ(ζ)) | − |≤| − | ≤ fν ( ) for some constant c = c(k) > 0. Then, by replacing ε> 0 so that ε 1/c if necessary, we have ≤ 2 S b(λ(ζ))(ζ λ(ζ)) ∂λ(ζ) | f − | = ϕ(f −1(λ(ζ)))((f −1)′(λ(ζ)))2(ζ λ(ζ))2∂λ(ζ) | ν ν − | ∗ ′ ∗ −2 −2 ∗ 1 ∗ −2 ∗ c ϕ(z ) f (z ) ρ D∗ (fν(z )) ϕ(z ) ρD∗ (z ) < 1 ≤ | || ν | fν ( ) ≤ ε | |

for ζ = fν (z); hence

2 1 ∗ −2 ∗ µ b(f (z)) S b(λ(ζ))(ζ λ(ζ)) ∂λ¯ (ζ) ϕ(z ) ρ ∗ (z ) | f ν |≤| f − | ≤ ε | | D for every z D. ∈ α D∗ Now we take ϕ B0 ( ) such that ϕ ∞ ϕ ∞,α < ε. Then, we can apply the above argument to∈ obtain k k ≤ k k

1 ∗ −2 ∗ ϕ ∞,α −α ∗ −2α −α µ b(f (z)) ϕ(z ) ρ ∗ (z ) k k ρ ∗ (z ) < z ρD (z). | f ν | ≤ ε | | D ≤ ε D | | We use this estimate when z is bounded away from 0. When z is small, for example, if | | −α | | α z < 1/√2, then µ b(f (z)) < 1 < 4ρ (z). Thus, we see that µ b f Bel (D). | | | f ν | D f ◦ ν ∈ 0 α D Denoting the complex dilatation of f fν by µϕ, we will show that µϕ Bel0 ( ). The formula for the complex dilatation of composed◦ quasiconformal homeomorphisms∈ is b −2iθ b e µf (fν (z)) + ν(z) µϕ(z)= (z D), −2iθ b ∈ 1+ e µf (fν(z))ν(z) 38 K. MATSUZAKI

where θ = arg ∂fν (z). Then, similarly to the proof of Proposition 7.2, we have

b b′ µf fν (z) µf fν(z) µϕ(z) µϕ′ (z) | ◦ − ◦ | 2 2 | − | ≤ b b′ (1 µf ∞)(1 µf ∞) q −k k −k k ′ α D∗ ′ ′ for any ϕ and ϕ in B0 ( ) with ϕ ∞,α, ϕ ∞,α <ε, where f and f are the corresponding quasiconformal homeomorphisms.k k In particular,k k setting ϕ′ = 0 yields b b b µf fν(z) µϕ(z) ν(z) | ◦ | . | − | ≤ b 2 (1 µf ∞) q −k k α As both µ b f and ν belong to Bel (D), so does µ . f ◦ ν 0 ϕ Because Φ(µ )= S b = ψ + ϕ, we have a local section η of Φ on the neighborhood ϕ f◦fν |D∗ U(ψ,ε)= ψ + ϕ ϕ <ε β(T ) Bα(D∗) { |k k∞,α } ⊂ ∩ 0 by the correspondence η : ψ + ϕ µ . By the above inequalities for µ and µ b f , we 7→ ϕ ϕ f ◦ ν see that ∗ ′ ∗ −2 ∗ µ (z) µ ′ (z) C ϕ(z ) ϕ (z ) ρ ∗ (z ) (z D) | ϕ − ϕ | ≤ | − | D ∈ for some constant C > 0. This implies that η is continuous.

We have obtained the continuity of the Bers projection Φ and its local section restricted α D α D∗ to Bel0 ( ) and β(T ) B0 ( ), respectively, with respect to the norm ∞,α. We note ∩ α D∗ k·k that the local section at ψ β(T ) B0 ( ) can be chosen so that ψ is sent to an arbitrary −1 ∈ ∩ −1 point in the ﬁber Φ (ψ) by post-composition of the right translation rλ for λ π ([id]). These maps are given in the same form as the original ones for Bel(D) and∈ B(D∗). Moreover, we know that these maps are holomorphic on Bel(D) and B(D∗) with respect to the norm ∞ (Theorem 2.2). Once we are in this situation, to see that the new maps are in factk·k holomorphic is a matter of general argument. Indeed, Φ and its local section are holomorphic as mappings to C if we ﬁx the complex variable z of functions ϕ(z) Bα(D∗) and µ(z) Belα(D). Then, the norm inequality and the ∈ 0 ∈ 0 k·k∞ ≤k·k∞,α continuity under ∞,α justify the claim (see Lemma 3.4 in Earle [19] and Lemma V.5.1 in Lehto [28]). k·k α D α D∗ Theorem 7.6. The Bers projection Φ : Bel0 ( ) β(T ) B0 ( ) is a holomorphic split submersion. → ∩ Moreover, we have seen in Proposition 7.2 that the right translation and, hence, the base point change map are homeomorphisms. By the same reasoning as above, we also see that they are biholomorphic. α D α D α D Corollary 7.7. The right translation rν : Bel0 ( ) Bel0 ( ) for ν Bel0 ( ) and the base point change map R : T α T α for τ T α are→ biholomorphic. ∈ τ 0 → 0 ∈ 0 TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 39

α The Teichmüller space T0 is equipped with the Kobayashi metric as an invariant metric of a complex manifold. By Theorem 1.1 in Yanagishita [41], which generalizes the result α of Hu, Jiang, and Wang [26], we see that the Kobayashi distance on T0 coincides with α the restriction of the Teichmüller distance on T to T0 ; hence, the infinitesimal Kobayashi α metric on each tangent space of T0 coincides with its restriction of the infinitesimal Teich- müller metric on the tangent space of T . α Finally, in this section, we investigate the topology on T0 , which has been defined as α D the quotient topology induced from the norm ∞,α on Bel0 ( ) by the Teichmüller 1+α S k·k projection π. However, as Diff+ ( ) is equipped with the right uniform topology, we can α also introduce another topology on T0 . This topology is the relative topology under the α 1+α S identification of T0 with the subgroup Diff+ ( ) QS∗ of all normalized elements. We α ∩ also call this the right uniform topology on T0 . Concerning the relation between these α two topologies on T0 , we have the following: α Theorem 7.8. The right uniform topology on T0 coincides with the quotient topology α D induced from Bel0 ( ). α Proof. Suppose that [gn] [g] as n in the quotient topology on T0 for gn,g Diff1+α(S) QS . Then, there→ are µ and→ ∞µ in Belα(D) with π(µ ) = [g ] and π(µ) = [g∈] + ∩ ∗ n 0 n n such that µn µ with respect to ∞,α. As the right translation rµ is a homeomor- α→D k·k phism of Bel0 ( ) by Proposition 7.2, the condition µn µ is equivalent to the condition −1 α D → rµ(µn)= µn µ 0 in Bel0 ( ). Then, by Theorem 6.9, the normalized representatives −∗1 →1+α S γn = gn g Diff+ ( ) QS∗ with [γn]= π(rµ(µn)) satisfy p1+α(γn) 0 as n . ◦ ∈ ∩ 1+α S → →∞ This means that γn converge to id in Diff+ ( ) QS∗. Hence, [gn] converge to [g] in the α ∩ right uniform topology on T0 . α Conversely, suppose that [gn] [g] as n in the right uniform topology on T0 1+α S → →− ∞1 for gn,g Diff+ ( ) QS∗. Then, γn = gn g converge to id, that is, p1+α(γn) 0. In particular,∈ c (γ ) ∩ 0. Then, by Theorem◦ 5.7, we have quasiconformal extensions→ α n → fn : D D of γn, whose complex dilatations νn satisfy νn ∞,α 0 as n . Hence, → −1 k α k → → ∞ [γn] = [gn] [g] [id] in the quotient topology on T0 , and thus [gn] [g] by the ∗ → −1 α → continuity of the base point change map R[g] of T0 . Combined with Proposition 5.2, this implies the following: Corollary 7.9. (T α, ) is a topological group. 0 ∗

References [1] L. Ahlfors, Lectures on quasiconformal mappings, Van Nostrand, 1966. [2] L. Ahlfors, Quasiconformal reﬂections, Acta Math. 109 (1963), 291–301. [3] L. Ahlfors and L. Bers, Riemann’s mapping theorem for variable metrics, Ann. of Math. 72 (1960), 385–404. [4] G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen, Distortion functions for plane quasiconformal mappings, Israel J. Math. 62 (1988), 1–16. 40 K. MATSUZAKI

[5] J. M. Anderson, J. Becker, and F. D. Lesley, On the boundary correspondence of asymptotically conformal automorphisms, J. London Math. Soc. 38 (1988), 453–462. [6] J. M. Anderson, A. Cantón, and J. L. Fernández, On smoothness of symmetric mappings, Complex Var. Theory Appl. 37 (1998), 161–169. [7] J. M. Anderson and A. Hinkkanen, Quasiconformal self-mappings with smooth boundary values, Bull. London Math. Soc. 26 (1994), 549–556. [8] K. Astala and M. Zinsmeister, Teichmüller spaces and BMOA, Math. Ann. 289 (1991), 613–625. [9] F. G. Avhadiev, Conditions for the univalence of analytic functions (Russian), Izv. Vys ˘s. U ˘cebn. Zaved. Matematika (1970), no. 11 (102), 3–13. [10] J. Becker, Conformal mappings with quasiconformal extensions, Aspects of contemporary complex analysis, Academic Press, 1980, pp. 37–77. [11] J. Becker, On asymptotically conformal extension of univalent functions, Complex Variables 9 (1987), 109-120. [12] J. Becker and C. Pommerenke, Uber¨ die quasikonforme Fortsetzung schlichter Funktionen, Math. Z. 161 (1978), 69–80. [13] L. Bers, A non-standard integral equation with applications to quasiconformal mappings, Acta Math. 116 (1966), 113–134. [14] A. Beurling and L. Ahlfors, The boundary correspondence under quasiconformal mappings, Acta Math. 96 (1956), 125–142. [15] L. Carleson, On mappings, conformal at the boundary, J. Anal. Math. 19 (1967), 1–13. [16] G. Cui, Integrably asymptotic affine homeomorphisms of the circle and Teichmüller spaces, Sci. China Ser. A 43 (2000), 267–279. [17] A. Douady and C. J. Earle, Conformally natural extension of homeomorphisms of the circle, Acta Math. 157 (1986), 23–48. [18] E. Dyn′kin, Estimates for asymptotically conformal mappings, Ann. Acad. Sci. Fenn. 22 (1997), 275–304. [19] C. J. Earle, On quasiconformal extensions of the Beurling–Ahlfors type, Contributions to analysis, pp. 99–105, Academic Press, 1974. [20] C. J. Earle, V. Markovic, and D. Saric, Barycentric extension and the Bers embedding for asymptotic Teichmüller space, Complex manifolds and hyperbolic geometry, Contemporary Math. vol. 311, pp. 87–105, Amer. Math. Soc., 2002. [21] C. J. Earle and S. Nag, Conformally natural reflection in Jordan curve with applications to Teich- müller spaces, Holomorphic functions and moduli II, pp. 179–194, Springer, 1988. [22] R. Fehlmann, Uber¨ extremale quasikonforme Abbildungen, Comment Math. Helv. 56 (1981), 558–580. [23] F. P. Gardiner and N. Lakic, Quasiconformal Teichmüller theory, Mathematical Surveys and Mono- graphs vol. 76, Amer. Math. Soc., 2000. [24] F. Gardiner and D. Sullivan, Symmetric structure on a closed curve, Amer. J. Math. 114 (1992), 683–736. [25] M. R. Herman, Sur la conjugaison différentiable des difféomorphismes du cercle àdes rotations, Publ. Math. IHES 49 (1979), 5–233. [26] J. Hu, Y. Jiang, and Z. Wang, Kobayashi’s and Teichmüller’s metrics on the Teichmüller space of symmetric circle homeomorphisms, Acta Math. Sinica (Engl. Ser.) 27 (2011), 617–624. [27] I. Kra, On Teichmüller’s theorem on the quasi-invariance of cross ratios, Israel. J. Math. 30 (1978), 152–158. [28] O. Lehto, Univalent functions and Teichmüller spaces, Graduate Texts in Mathematics vol. 109, Springer, 1986. [29] O. Lehto and K. Virtanen, Quasiconformal mappings in the plane, Springer, 1973. TEICHMULLER¨ SPACE AND DIFFEOMORPHISMS 41

[30] K. Matsuzaki, The universal Teichmüller space and diffeomorphisms of the circle with Hölder continuous derivatives, Handbook of group actions (Vol. I), Advanced Lectures in Mathematics vol. 31, pp. 333–372, Higher Education Press and International Press, 2015. [31] K. Matsuzaki, Circle diffeomorphisms, rigidity of symmetric conjugation and affine foliation of the universal Teichmüller space, Geometry, dynamics, and foliations 2013, Advanced Studies in Pure Mathematics vol. 72, pp. 145–180, Mathematical Society of Japan, 2017. [32] K. Matsuzaki, Continuity of the barycentric extension of circle diffeomorphisms of Hölder continuous derivative, Trans. London Math. Soc. 4 (2017), 129–147. [33] K. Matsuzaki, Rigidity of groups of circle diffeomorphisms and Teichmüller spaces, J. Anal. Math. 140 (2020). [34] C. B. Morrey, On the solutions of quasi-linear elliptic partial differential equations, Trans. Amer. Math. Soc. 43 (1938), 126–166. [35] S. Nag, The complex analytic theory of Teichmüller spaces, John Wiley & Sons, 1988. [36] B. Osgood, Some properties of f ′′/f ′ and the Poincarémetric, Indiana Univ. Math. J. 31 (1982), 449–461. [37] C. Pommerenke, Boundary behaviour of conformal maps, Springer, 1992. [38] C. Pommerenke and S. E. Warschawski, On the quantitative boundary behavior of conformal maps, Comment. Math. Helv. 57 (1982), 107–129. [39] T. Sugawa, The universal Teichmüller space and related topics, Proceedings of the international workshop on quasiconformal mappings and their applications, pp. 261–289, Narosa Publishing House, 2007. [40] L. Tam and T. Wan, Quasi-conformal harmonic diffeomorphism and the universal Teichmüller space, J. Diff. Geom. 42 (1995), 368–410. [41] M. Yanagishita, Teichmüller distance and Kobayashi distance on subspaces of the universal Teich- müller space, Kodai Math. J. 36 (2013), 209–227. [42] M. Yanagishita, Introduction of a complex structure on the p-integrable Teichmüller space, Ann. Acad. Sci. Fenn. Math. 39 (2014), 947–971. [43] K. Zhu, Operator theory in function spaces, 2nd ed., Mathematical Surveys and Monographs vol. 138, Amer. Math. Soc., 2007.

Department of Mathematics, School of Education, Waseda University, Shinjuku, Tokyo 169-8050, Japan E-mail address: [email protected]