Mathematics · Book 5 · Bachelor Year 3

University Mathematics — Year 3

University Mathematics — Year 3 · Bachelor Year 3

18Conformal Maps and the Riemann Mapping Theorem

A holomorphic bijection between two domains transports all of complex analysis from one to the other: such maps — conformal, because they preserve angles — are the isomorphisms of the holomorphic world. This chapter classifies them where classification is possible (the disc, the plane: Schwarz’s lemma is the key, an inequality of astonishing power), builds the compactness theory of holomorphic families (Montel), and proves the deepest existence theorem of the subject: every simply connected proper subdomain of C\C, however jagged its boundary, is conformally equivalent to the unit disc. We close with harmonic functions and the Poisson kernel, solving the Dirichlet problem on the disc — the analytic payoff of conformal geometry. Throughout, D=D(0,1)\mathbb D = D(0,1) and H={Imz>0}\mathbb H = \{\operatorname{Im}z > 0\}.

18.1 Conformal maps; Möbius transformations

Definition 18.1

A conformal map (or biholomorphism) between open sets is a holomorphic bijection; its inverse is automatically holomorphic (Corollary 17.10). Two domains are conformally equivalent if such a map exists; Aut(Ω)\operatorname{Aut}(\Omega) denotes the group of conformal self-maps. Where f0f' \neq 0 — everywhere, for injective ff (Corollary 17.10’s proof) — the differential is multiplication by f(z)0f'(z) \neq 0: a similarity, so conformal maps preserve angles between curves, including orientation.

Example 18.2 (Möbius transformations)

For (abcd)GL2(C)\bigl(\begin{smallmatrix}a & b\\ c & d\end{smallmatrix}\bigr) \in GL_2(\C), the Möbius transformation zaz+bcz+dz \mapsto \frac{az + b}{cz + d} is conformal from C{d/c}\C\setminus\{-d/c\} onto C{a/c}\C\setminus\{a/c\} (inverse of the same type, from the inverse matrix; composition corresponds to matrix product). The Cayley map

φ(z)=ziz+i\varphi(z) = \frac{z - \iu}{z + \iu}

maps H\mathbb H conformally onto D\mathbb D: indeed zi<z+i\abs{z - \iu} < \abs{z + \iu} exactly when zz is closer to i\iu than to i-\iu, i.e. Imz>0\operatorname{Im}z > 0; the inverse is wi1+w1ww \mapsto \iu\frac{1 + w}{1 - w}. Möbius maps send the family circles-and-lines to itself (Exercise 18.1).

Example 18.3 (The Joukowski map)

Beyond Möbius, the most useful conformal map of classical applied mathematics is

J(z)=12(z+1z).J(z) = \frac12\Bigl(z + \frac1z\Bigr) .

On the exterior Ω={z>1}\Omega = \{\abs z > 1\} of the unit disc, JJ is injective: J(z)=J(w)J(z) = J(w) gives (zw)(11zw)=0(z - w)(1 - \frac1{zw}) = 0 and zw>1\abs{zw} > 1. Its derivative J(z)=12(1z2)J'(z) = \frac12(1 - z^{-2}) vanishes only at z=±1z = \pm1, on the boundary: JJ is a conformal equivalence from Ω\Omega onto its image, which is C[1,1]\C\setminus\intcc{-1}1 — the unit circle itself is folded two-to-one onto the segment (J(eiθ)=cosθJ(\eu^{\iu\theta}) = \cos\theta). Thus the exterior of a segment, a slit plane with no smooth boundary, is conformally the exterior of a disc: corners are no obstacle to conformal equivalence, only to boundary smoothness. Images of circles through ±1\pm1 but off-center are airfoil-shaped curves, and composing JJ with Möbius maps transported the flow past a cylinder — computable by hand — to the flow past a wing: for the first half of the twentieth century, this example was aerodynamics. It is also the door to Chebyshev: JJ conjugates zznz \mapsto z^n to the Chebyshev polynomial TnT_n (Problem 13.1, Part V), since J(zn)=cos(nθ)J(z^n) = \cos(n\theta) when z=eiθz = \eu^{\iu\theta}.

18.2 Schwarz’s lemma and automorphism groups

Theorem 18.4 (Schwarz lemma)

Let f ⁣:DDf \colon \mathbb D \to \mathbb D be holomorphic with f(0)=0f(0) = 0. Then

f(z)z  (zD)andf(0)1,\abs{f(z)} \leq \abs z \ \ (z \in \mathbb D) \qquad\text{and}\qquad \abs{f'(0)} \leq 1 ,

and if f(z0)=z0\abs{f(z_0)} = \abs{z_0} for one z00z_0 \neq 0, or f(0)=1\abs{f'(0)} = 1, then f(z)=eiθzf(z) = \eu^{\iu\theta}z is a rotation.

Proof. g(z)=f(z)/zg(z) = f(z)/z extends holomorphically to D\mathbb D (the singularity at 00 is removable: gg is bounded near 00, Theorem 17.4; its value at 00 is f(0)f'(0)). On z=r<1\abs z = r < 1: g1r\abs g \leq \frac1r, so by the maximum principle (Theorem 16.14) g1r\abs g \leq \frac1r on Dˉ(0,r)\bar D(0,r); let r1r \to 1: g1\abs g \leq 1 on D\mathbb D, which is both inequalities. Equality at an interior point makes g\abs g attain an interior maximum: gg constant of modulus 11.

Theorem 18.5 (Automorphisms of the disc)

For aDa \in \mathbb D, the Blaschke factor

φa(z)=za1aˉz\varphi_a(z) = \frac{z - a}{1 - \bar a z}

is an automorphism of D\mathbb D exchanging aa and 00, with φa1=φa\varphi_a^{-1} = \varphi_{-a}. Every automorphism of D\mathbb D is eiθφa\eu^{\iu\theta}\varphi_a for unique θR/2πZ\theta \in \R/2\pi\Z, aDa \in \mathbb D.

Proof. On z=1\abs z = 1: 1aˉz=zˉ1aˉz=zˉaˉz2=zˉaˉ=za\abs{1 - \bar az} = \abs{\bar z}\abs{1 - \bar az} = \abs{\bar z - \bar a\abs z^2} = \abs{\bar z - \bar a} = \abs{z - a}, so φa=1\abs{\varphi_a} = 1 there; by the maximum principle φa(D)Dˉ\varphi_a(\mathbb D) \subseteq \bar{\mathbb D}, and openness puts the image in D\mathbb D. The algebraic identity φaφa=id\varphi_{-a}\circ\varphi_a = \mathrm{id} (direct computation) shows bijectivity. Now let fAut(D)f \in \operatorname{Aut}(\mathbb D) and a=f1(0)a = f^{-1}(0): g=fφag = f\circ\varphi_{-a} is an automorphism fixing 00. Schwarz applied to gg and to g1g^{-1}: g(z)z\abs{g(z)} \leq \abs z and g1(w)w\abs{g^{-1}(w)} \leq \abs w, so g(z)=z\abs{g(z)} = \abs z: rotation, g=eiθidg = \eu^{\iu\theta}\,\mathrm{id}, i.e. f=eiθφaf = \eu^{\iu\theta}\varphi_a. Uniqueness: a=f1(0)a = f^{-1}(0) and θ\theta from ff'-type evaluation (or from f(0)=eiθaf(0) = -\eu^{\iu\theta}a and one more value).

Theorem 18.6 (Automorphisms of the plane)

Aut(C)={zaz+b:aC, bC}\operatorname{Aut}(\C) = \{z \mapsto az + b : a \in \C^*,\ b \in \C\}. Consequently C\C and D\mathbb D are not conformally equivalent.

Proof. Let fAut(C)f \in \operatorname{Aut}(\C) and consider g(z)=f(1/z)g(z) = f(1/z) on C\C^*: a holomorphic function with an isolated singularity at 00. If it were essential, Casorati–Weierstrass (Theorem 17.4) would make g(D(0,ε){0})g\bigl(D(0,\varepsilon)\setminus\{0\}\bigr) dense, while f(D(0,1))f(D(0, 1)) is open and disjoint from it (ff injective: the two sets are images of disjoint sets) — impossible for a dense set and a nonempty open set. So 00 is a pole or removable for gg, i.e. f(z)\abs{f(z)} has at most polynomial growth: ff is a polynomial (Exercise 16.4(a)). Injectivity forces degree 11: a higher-degree polynomial has either a multiple root of fcf - c somewhere (ff' vanishes) or several distinct preimages (d’Alembert–Gauss, Problem 16.1); either way injectivity fails. Finally, a conformal CD\C \to \mathbb D would be a bounded entire function: constant (Liouville) — no equivalence.

18.3 Montel’s theorem

Theorem 18.7 (Montel)

Let FH(Ω)\mathcal F \subseteq \mathcal H(\Omega) be locally bounded: every point has a neighborhood on which supfFsupf<\sup_{f\in \mathcal F}\sup\abs f < \infty. Then every sequence of F\mathcal F has a subsequence converging uniformly on all compact subsets of Ω\Omega (to a holomorphic limit).

Proof. Local equicontinuity: if fM\abs f \leq M on D(a,2r)ΩD(a, 2r) \subseteq \Omega for all fFf \in \mathcal F, the Cauchy formula gives, for z,zD(a,r)z, z' \in D(a, r),

f(z)f(z)=zz2πC2rf(w) ⁣dw(wz)(wz)zz2π2rM2πr2=2Mrzz:\abs{f(z) - f(z')} = \frac{\abs{z - z'}}{2\pi} \Bigl|\int_{C_{2r}}\frac{f(w)\,\dd w}{(w-z)(w-z')}\Bigr| \leq \frac{\abs{z - z'}\,2\pi\cdot2r\,M}{2\pi\,r^2} = \frac{2M}{r}\,\abs{z - z'} :

a uniform Lipschitz bound. Exhaust Ω\Omega by compacts KmK_m; each KmK_m is covered by finitely many such discs, so F\mathcal F is uniformly bounded and equicontinuous on KmK_m: Arzelà–Ascoli (Theorem 7.11) extracts a subsequence converging uniformly on KmK_m; diagonalize over mm. The limit is holomorphic by Theorem 16.15.

18.4 The Riemann mapping theorem

Definition 18.8

An open connected ΩC\Omega \subseteq \C is simply connected (in the homological sense, sufficient for all our purposes) if Indγ(w)=0\operatorname{Ind}_\gamma(w) = 0 for every cycle γ\gamma in Ω\Omega and every wΩw \notin \Omega — “no cycle of Ω\Omega surrounds a hole”. By the global Cauchy theorem (Theorem 17.1) and Proposition 16.5, on such Ω\Omega every holomorphic function has a primitive; hence every zero-free fH(Ω)f \in \mathcal H(\Omega) has a holomorphic logarithm (exp\exp\circ(primitive of f/ff'/f), adjusted by a constant, as (feL)=0(f\eu^{-L})' = 0) and holomorphic nn-th roots eL/n\eu^{L/n}.

Theorem 18.9 (Riemann mapping theorem)

Every simply connected open ΩC\Omega \subsetneq \C, Ω\Omega \neq \varnothing, is conformally equivalent to D\mathbb D; given z0Ωz_0 \in \Omega, there is a unique conformal f ⁣:ΩDf \colon \Omega \to \mathbb D with f(z0)=0f(z_0) = 0 and f(z0)>0f'(z_0) > 0.

Proof. Step 0: the family is nonempty. Pick bΩb \notin \Omega: zbz - b is zero-free on Ω\Omega, so it has a holomorphic square root hh (h2=zbh^2 = z - b). hh is injective (h(z)=h(z)h(z) = h(z') squares to z=zz = z'), and if wh(Ω)w \in h(\Omega) then wh(Ω)-w \notin h(\Omega) (h(z)=h(z)h(z) = -h(z') also squares to z=zz = z', giving w=w=0w = -w = 0, impossible as hh is zero-free). Since h(Ω)h(\Omega) is open, it contains a disc D(h(z0),ρ)D(h(z_0), \rho); then D(h(z0),ρ)h(Ω)=D(-h(z_0), \rho) \cap h(\Omega) = \varnothing, i.e. h(z)+h(z0)ρ\abs{h(z) + h(z_0)} \geq \rho for every zΩz \in \Omega. Hence

g(z)=ρ2(h(z)+h(z0))g(z) = \frac{\rho}{2\,\bigl(h(z) + h(z_0)\bigr)}

is holomorphic, injective (a Möbius map composed with the injective hh), with g12<1\abs g \leq \frac12 < 1. Composing with a Blaschke factor (Theorem 18.5) to move g(z0)g(z_0) to 00, the family

F={f ⁣:ΩD holomorphic, injective, f(z0)=0}\mathcal F = \{f \colon \Omega \to \mathbb D \text{ holomorphic, injective, } f(z_0) = 0\}

is nonempty.

Step 1: an extremal element. Let s=supFf(z0)(0,+]s = \sup_{\mathcal F}\abs{f'(z_0)} \in \intoc0{+\infty} (>0> 0: members are injective, so f(z0)0f'(z_0) \neq 0). Take fnFf_n \in \mathcal F with fn(z0)s\abs{f_n'(z_0)} \to s: the family is bounded by 11, so Montel (Theorem 18.7) extracts fnff_n \to f uniformly on compacts; ff is holomorphic, f(z0)=0f(z_0) = 0, f(z0)=s\abs{f'(z_0)} = s (Theorem 16.15 for the derivatives), in particular ff is nonconstant; ff is injective by Hurwitz (Exercise 17.8(b)), and f(Ω)Dˉf(\Omega) \subseteq \bar{\mathbb D}, hence D\subseteq \mathbb D (open mapping). So fFf \in \mathcal F attains the supremum: s<s < \infty.

Step 2: the extremal map is onto. Suppose aDf(Ω)a \in \mathbb D\setminus f(\Omega). The Blaschke transport φaf\varphi_a\circ f is zero-free on the simply connected Ω\Omega: it has a holomorphic square root FF (with F(Ω)DF(\Omega) \subseteq \mathbb D, as F2=φaf<1\abs F^2 = \abs{\varphi_a\circ f} < 1), injective (squares distinguish). Normalize: G=φF(z0)FFG = \varphi_{F(z_0)}\circ F \in \mathcal F. Undoing: f=φas2φF(z0)Gf = \varphi_{-a}\circ s_2 \circ \varphi_{-F(z_0)}\circ G where s2(w)=w2s_2(w) = w^2; the map Ψ=φas2φF(z0) ⁣:DD\Psi = \varphi_{-a}\circ s_2\circ\varphi_{-F(z_0)} \colon \mathbb D \to \mathbb D is holomorphic with Ψ(0)=f(z0)=0\Psi(0) = f(z_0) = 0 and is not a rotation (it is not injective: s2s_2 is not). Schwarz’s lemma (strict case): Ψ(0)<1\abs{\Psi'(0)} < 1, and the chain rule f=ΨGf = \Psi\circ G gives f(z0)=Ψ(0)G(z0)<G(z0)\abs{f'(z_0)} = \abs{\Psi'(0)}\,\abs{G'(z_0)} < \abs{G'(z_0)} — contradicting maximality (note GFG \in \mathcal F). Hence ff is onto: a conformal equivalence.

Step 3: normalization and uniqueness. Multiply ff by eiargf(z0)\eu^{-\iu\arg f'(z_0)} to make f(z0)>0f'(z_0) > 0 (this stays in F\mathcal F). If f1,f2f_1, f_2 both work, ψ=f2f11Aut(D)\psi = f_2\circ f_1^{-1} \in \operatorname{Aut}(\mathbb D) fixes 00 with ψ(0)=f2(z0)/f1(z0)>0\psi'(0) = f_2'(z_0)/f_1'(z_0) > 0; by Theorem 18.5 ψ\psi is a rotation eiθ\eu^{\iu\theta} with eiθ>0\eu^{\iu\theta} > 0: ψ=id\psi = \mathrm{id}.

Remark 18.10

The theorem is a pure existence statement of astonishing scope: a square, a half-plane, the complement of a slit, the region between two tangent circles, a fractal-boundary domain — all conformally identical to D\mathbb D. What it does not give: any formula (explicit maps are the exception: Exercise 18.5), boundary behavior (a deeper theory — Carathéodory’s theorem — handles it), or uniqueness of extension to C\C or multiply connected domains: the annulus {1<z<2}\{1 < \abs z < 2\} is not conformally a punctured disc, and annuli of different radius ratios are inequivalent (a genuinely harder fact).

18.5 Harmonic functions and the Poisson kernel

Proposition 18.11

Let Ω\Omega be simply connected and u ⁣:ΩRu \colon \Omega \to \R harmonic (C2\mathcal C^2 with Δu=uxx+uyy=0\Delta u = u_{xx} + u_{yy} = 0). Then u=ReFu = \operatorname{Re}F for a holomorphic FF, unique up to an imaginary constant. Consequently uu is C\mathcal C^\infty, satisfies the mean value property, and obeys the maximum principle (no strict interior extremum unless constant).

Proof. g=uxiuyg = u_x - \iu u_y satisfies the Cauchy–Riemann equations (P=uxP = u_x, Q=uyQ = -u_y: Px=uxx=uyy=QyP_x = u_{xx} = -u_{yy} = Q_y and Py=uxy=uyx=QxP_y = u_{xy} = u_{yx} = -Q_x) with continuous partials: gg is holomorphic (Proposition 16.2; the R\R-differentiability follows from C1\mathcal C^1). Let F0F_0 be a primitive (simple connectivity, Definition 18.8); then ReF0\operatorname{Re}F_0 has gradient (ux,uy)(u_x, u_y) (F0=gF_0' = g unpacks to exactly that via Cauchy–Riemann for F0F_0), so uReF0u - \operatorname{Re}F_0 is constant (Ω\Omega connected): adjust F=F0+cF = F_0 + c. The properties transfer from Theorem 16.14 and Exercise 16.10 (for the maximum principle applied to uu itself, use eF\eu^{F} as there).

Theorem 18.12 (Poisson formula; Dirichlet problem on the disc)

For 0r<10 \leq r < 1 define the Poisson kernel

Pr(θ)=nZrneinθ=1r212rcosθ+r2  >  0.P_r(\theta) = \sum_{n\in\Z}r^{\abs n}\eu^{\iu n\theta} = \frac{1 - r^2}{1 - 2r\cos\theta + r^2} \;>\; 0 .

Let g ⁣:DRg \colon \partial\mathbb D \to \R be continuous, and set, for z=reiφDz = r\eu^{\iu\varphi} \in \mathbb D,

u(z)=12π02πPr(φt)g(eit) ⁣dt.u(z) = \frac1{2\pi}\int_0^{2\pi} P_r(\varphi - t)\,g(\eu^{\iu t})\,\dd t .

Then uu is harmonic on D\mathbb D and extends continuously to Dˉ\bar{\mathbb D} with boundary values gg: the unique such harmonic function.

Proof. Kernel identities: summing two geometric series,

nZrneinθ=Re1+reiθ1reiθ=1r21reiθ2,\sum_{n\in\Z}r^{\abs n}\eu^{\iu n\theta} = \operatorname{Re}\frac{1 + r\eu^{\iu\theta}}{1 - r\eu^{\iu\theta}} = \frac{1 - r^2}{\abs{1 - r\eu^{\iu\theta}}^2},

which is the displayed quotient; positivity is clear, and 12π02πPr=1\frac1{2\pi}\int_0^{2\pi}P_r = 1 (only n=0n = 0 survives).

Harmonicity: with z=reiφz = r\eu^{\iu\varphi},

u(z)=Re[12π02πeit+zeitzg(eit) ⁣dt],u(z) = \operatorname{Re}\biggl[\frac1{2\pi}\int_0^{2\pi} \frac{\eu^{\iu t} + z}{\eu^{\iu t} - z}\, g(\eu^{\iu t})\,\dd t\biggr],

(the bracketed kernel has real part Pr(φt)P_r(\varphi - t): compute), and the bracket is holomorphic in zz on D\mathbb D (Exercise 16.7): uu is the real part of a holomorphic function, hence harmonic.

Boundary values: Pr()P_r(\cdot) is an approximate identity as r1r \to 1^-: mass 11, and for δθπ\delta \leq \abs\theta \leq \pi, Pr(θ)1r212rcosδ+r20P_r(\theta) \leq \frac{1 - r^2}{1 - 2r\cos\delta + r^2} \to 0 uniformly. The standard split (continuity of gg near eiφ0\eu^{\iu\varphi_0}, boundedness elsewhere) gives u(reiφ)g(eiφ0)u(r\eu^{\iu\varphi}) \to g(\eu^{\iu\varphi_0}) as reiφeiφ0r\eu^{\iu\varphi} \to \eu^{\iu\varphi_0}, uniformly in the boundary point: the extension is continuous. Uniqueness: the difference of two solutions is harmonic on D\mathbb D, continuous on the closure, zero on the boundary: by the maximum principle (applied to ±\pm the difference), it vanishes.

The level curves of Rez2 = x2 - y2 (red hyperbolas) and Imz2 = 2xy (blue hyperbolas) intersect at right angles away from 0: a conformal map (z z2, where z ≠ 0) preserves the orthogonality of coordinate lines. At z = 0, where the derivative vanishes, angles are doubled instead.
The level curves of Rez2=x2y2\operatorname{Re}z^2 = x^2 - y^2 (red hyperbolas) and Imz2=2xy\operatorname{Im}z^2 = 2xy (blue hyperbolas) intersect at right angles away from 00: a conformal map (zz2z \mapsto z^2, where z0z \neq 0) preserves the orthogonality of coordinate lines. At z=0z = 0, where the derivative vanishes, angles are doubled instead.

18.6 Exercises

Exercise 18.1

(a) Verify that the Cayley map φ(z)=ziz+i\varphi(z) = \frac{z - \iu}{z + \iu} is a bijection HD\mathbb H \to \mathbb D with the stated inverse, and compute the images of i\iu, 00, 11, \infty (limit). (b) Show that z1/zz \mapsto 1/z maps circles and lines to circles and lines. (Write their common equation αz2+βˉz+βzˉ+γ=0\alpha\abs z^2 + \bar\beta z + \beta\bar z + \gamma = 0, α,γR\alpha, \gamma \in \R.) Deduce the same for all Möbius maps.

Solution

Solution of Exercise 18.1.

(a) φ\varphi and ψ(w)=i1+w1w\psi(w) = \iu\frac{1+w}{1-w} compose to the identity in both orders (direct computation); φ\varphi maps H\mathbb H into D\mathbb D and ψ\psi back (Example 18.2). Values: φ(i)=0\varphi(\iu) = 0, φ(0)=1\varphi(0) = -1, φ(1)=1i1+i=i\varphi(1) = \frac{1 - \iu}{1 + \iu} = -\iu, and φ(z)1\varphi(z) \to 1 as zz \to \infty.

(b) Circles and lines are the solution sets of αz2+βˉz+βzˉ+γ=0\alpha\abs z^2 + \bar\beta z + \beta\bar z + \gamma = 0 (α,γR\alpha, \gamma \in \R, βC\beta \in \C, β2>αγ\abs\beta^2 > \alpha\gamma): α0\alpha \neq 0 circles, α=0\alpha = 0 lines. Substituting z=1/wz = 1/w and multiplying by w2\abs w^2: γw2+βw+βˉwˉ+α=0\gamma\abs w^2 + \beta w + \bar\beta\bar w + \alpha = 0 — same family. Affine maps clearly preserve the family, and every Möbius map is a composition of affine maps and one inversion (az+bcz+d=ac+bcadc1cz+d\frac{az+b}{cz+d} = \frac ac + \frac{bc - ad}{c}\cdot\frac1{cz + d} for c0c \neq 0).

Exercise 18.2

Let f ⁣:DDf \colon \mathbb D \to \mathbb D be holomorphic. (a) If f(0)=0f(0) = 0 and f(a)=af(a) = a for some a0a \neq 0, show f=idf = \mathrm{id}. (b) If ff is an automorphism with two distinct fixed points in D\mathbb D, show f=idf = \mathrm{id} (conjugate by a Blaschke factor to reduce to (a)).

Solution

Solution of Exercise 18.2.

(a) Schwarz gives f(a)a\abs{f(a)} \leq \abs a with equality (both sides =a= \abs a): the equality case forces f(z)=eiθzf(z) = \eu^{\iu\theta}z, and f(a)=af(a) = a pins eiθ=1\eu^{\iu\theta} = 1.

(b) Let aba \neq b be fixed points and g=φafφaAut(D)g = \varphi_a\circ f\circ\varphi_{-a} \in \operatorname{Aut}(\mathbb D) — using Theorem 18.5 for φ±a\varphi_{\pm a}. Then g(0)=φa(f(a))=0g(0) = \varphi_a(f(a)) = 0 and g(c)=cg(c) = c for c=φa(b)0c = \varphi_a(b) \neq 0: by (a), g=idg = \mathrm{id}, so f=φaφa=idf = \varphi_{-a}\circ\varphi_a = \mathrm{id}.

Exercise 18.3 ★★

(Schwarz–Pick) For holomorphic f ⁣:DDf\colon \mathbb D \to \mathbb D, prove

f(z)1f(z)2    11z2(zD),\frac{\abs{f'(z)}}{1 - \abs{f(z)}^2} \;\leq\; \frac{1}{1 - \abs z^2} \qquad (z \in \mathbb D),

with equality (at one point, hence everywhere) iff fAut(D)f \in \operatorname{Aut}(\mathbb D). (Apply Schwarz to φf(z)fφz\varphi_{f(z)}\circ f\circ\varphi_{-z}.) Interpretation: holomorphic self-maps contract the hyperbolic metric.

Solution

Solution of Exercise 18.3.

Fix zz and set g=φf(z)fφzg = \varphi_{f(z)}\circ f\circ\varphi_{-z}: holomorphic DD\mathbb D \to \mathbb D with g(0)=0g(0) = 0, so g(0)1\abs{g'(0)} \leq 1 (Schwarz). Chain rule with φa(ζ)=1a2(1aˉζ)2\varphi_a'(\zeta) = \frac{1 - \abs a^2}{(1 - \bar a\zeta)^2}:

g(0)=φf(z)(f(z))f(z)φz(0)=11f(z)2f(z)(1z2),g'(0) = \varphi_{f(z)}'\bigl(f(z)\bigr)\cdot f'(z)\cdot \varphi_{-z}'(0) = \frac{1}{1 - \abs{f(z)}^2}\cdot f'(z)\cdot(1 - \abs z^2),

whence the Schwarz–Pick inequality. Equality at some zz makes gg a rotation, hence f=φf(z)(rotation)φzAut(D)f = \varphi_{-f(z)}\circ(\text{rotation})\circ\varphi_z \in \operatorname{Aut}(\mathbb D) — and then equality holds everywhere (compute, or reapply with roles of f,f1f, f^{-1} exchanged). Holomorphic self-maps of the disc are 11-Lipschitz for the hyperbolic metric 2 ⁣dz1z2\frac{2\abs{\dd z}}{1 - \abs z^2}; automorphisms are its isometries.

Exercise 18.4 ★★

Find explicit conformal equivalences: (a) the strip {0<Imz<π}H\{0 < \operatorname{Im}z < \pi\} \to \mathbb H; (b) the quadrant {Rez>0,Imz>0}H\{\operatorname{Re}z > 0, \operatorname{Im}z > 0\} \to \mathbb H; (c) the half-disc DH\mathbb D\cap\mathbb H \to a quadrant, then H\to \mathbb H; (d) DD\mathbb D \to \mathbb D sending 12\frac12 to 00 with positive derivative there.

Solution

Solution of Exercise 18.4.

(a) zezz \mapsto \eu^z: maps {0<Imz<π}\{0 < \operatorname{Im}z < \pi\} bijectively onto H\mathbb H (ex+iy=exeiy\eu^{x+\iu y} = \eu^x\eu^{\iu y}: modulus free, argument y(0,π)y \in \intoo0\pi), holomorphic with nonvanishing derivative and holomorphic inverse (principal log\log). (b) zz2z \mapsto z^2 doubles arguments: the open quadrant {0<argz<π2}\{0 < \arg z < \frac\pi2\} maps conformally onto H\mathbb H (inverse: principal square root). (c) z1+z1zz \mapsto \frac{1 + z}{1 - z} maps D\mathbb D onto the right half-plane and preserves the upper/lower symmetry: it sends the upper half-disc onto the first quadrant; then square, by (b), to reach H\mathbb H: z(1+z1z)2z \mapsto \bigl(\frac{1 + z}{1 - z}\bigr)^2. (d) The Blaschke factor φ1/2(z)=z121z2\varphi_{1/2}(z) = \frac{z - \frac12}{1 - \frac z2}: φ1/2(12)=0\varphi_{1/2}(\tfrac12) = 0 and φ1/2(12)=114(114)2=43>0\varphi_{1/2}'(\tfrac12) = \frac{1 - \frac14}{(1 - \frac14)^2} = \frac43 > 0.

Exercise 18.5 ★★

(a) Show that no conformal map CD\C \to \mathbb D or CH\C \to \mathbb H exists, and none DC\mathbb D \to \C. (b) Which of the following are conformally equivalent to D\mathbb D? Justify via Theorem 18.9 or an obstruction: a square; C(,0]\C\setminus\intoc{-\infty}0; D{0}\mathbb D\setminus\{0\}; {1<z<2}\{1 < \abs z < 2\}. (For the last two: a conformal image of the punctured disc would extend over the puncture by Theorem 17.4(1) — develop this.)

Solution

Solution of Exercise 18.5.

(a) A conformal CD\C \to \mathbb D (or H\mathbb H, after composing with Cayley) is a bounded entire function: constant by Liouville — not bijective. A conformal DC\mathbb D \to \C would have a conformal inverse CD\C \to \mathbb D: same contradiction.

(b) The square is convex, hence simply connected, and proper: conformally D\mathbb D (Theorem 18.9). The cut plane C(,0]\C\setminus\intoc{-\infty}0 is star-shaped about 11 (segments from 11 avoid the cut), hence simply connected, and proper: conformally D\mathbb D. The punctured disc: if g ⁣:D{0}Dg \colon \mathbb D\setminus\{0\} \to \mathbb D were conformal, gg is bounded, so 00 is removable (Theorem 17.4): gg extends to g~ ⁣:DD\tilde g \colon \mathbb D \to \mathbb D, and g~(0)\tilde g(0), being in the open image g(D{0})=Dg(\mathbb D\setminus\{0\}) = \mathbb D, is also g(w)g(w) for some w0w \neq 0; two disjoint neighborhoods of 00 and ww have images that are open and share the value g~(0)\tilde g(0), hence share other values too (open sets): gg takes some value twice on D{0}\mathbb D\setminus\{0\} — contradicting injectivity. The annulus A={1<z<2}A = \{1 < \abs z < 2\}: suppose F ⁣:DAF \colon \mathbb D \to A conformal. FF is zero-free on the simply connected D\mathbb D, so F=eLF = \eu^L for holomorphic LL (Definition 18.8). Let σ\sigma be the circle z=32\abs z = \frac32 in AA and γ=F1σ\gamma = F^{-1}\circ\sigma, a closed path in D\mathbb D; then

1=Indσ(0)=12iπFγ ⁣dww=12iπγFF=12iπγL=01 = \operatorname{Ind}_\sigma(0) = \frac1{2\iu\pi}\int_{F\circ\gamma}\frac{\dd w}{w} = \frac1{2\iu\pi}\int_\gamma\frac{F'}{F} = \frac1{2\iu\pi}\int_\gamma L' = 0

(LL' has a primitive): contradiction. Neither the punctured disc nor the annulus is a disc in disguise.

Exercise 18.6 ★★

Let F={fH(D):f(0)=1, Ref>0}\mathcal F = \{f \in \mathcal H(\mathbb D) : f(0) = 1,\ \operatorname{Re}f > 0\}. (a) Show that F\mathcal F is locally bounded. (Compose with the Cayley-type map ww1w+1w \mapsto \frac{w - 1}{w + 1} sending the right half-plane to D\mathbb D, and apply Schwarz.) (b) Deduce the Herglotz bound: f(z)1+z1z\abs{f(z)} \leq \frac{1 + \abs z}{1 - \abs z} for fFf \in \mathcal F, with equality possibilities.

Solution

Solution of Exercise 18.6.

(a) T(w)=w1w+1T(w) = \frac{w - 1}{w + 1} maps {Rew>0}\{\operatorname{Re}w > 0\} conformally onto D\mathbb D (Cayley rotated: w1<w+1\abs{w - 1} < \abs{w + 1} iff Rew>0\operatorname{Re}w > 0), with T(1)=0T(1) = 0. For fFf \in \mathcal F, g=Tf ⁣:DDg = T\circ f\colon \mathbb D \to \mathbb D is holomorphic with g(0)=0g(0) = 0: Schwarz gives g(z)z\abs{g(z)} \leq \abs z.

(b) Inverting TT: f=1+g1gf = \frac{1 + g}{1 - g}, so

f(z)1+g(z)1g(z)1+z1z:\abs{f(z)} \leq \frac{1 + \abs{g(z)}}{1 - \abs{g(z)}} \leq \frac{1 + \abs z}{1 - \abs z} :

locally bounded (uniformly on zr<1\abs z \leq r < 1). Equality at z00z_0 \neq 0 forces g(z0)=z0\abs{g(z_0)} = \abs{z_0} and alignment: gg a rotation, i.e. f(z)=1+eiθz1eiθzf(z) = \frac{1 + \eu^{\iu\theta}z}{1 - \eu^{\iu\theta}z} — the Herglotz extremals, conformal maps onto the right half-plane.

Exercise 18.7 ★★★

Where does the proof of Theorem 18.9 use each hypothesis? Trace: (i) simple connectivity (twice); (ii) ΩC\Omega \neq \C; (iii) connectedness. Then show that the theorem fails for Ω=C\Omega = \C and for the annulus, pinpointing which step of the proof breaks in each case.

Solution

Solution of Exercise 18.7.

(i) Simple connectivity enters exactly twice, through the existence of holomorphic square roots of zero-free functions (Definition 18.8): in Step 0 (the root of zbz - b) and in Step 2 (the root of φaf\varphi_a\circ f). (ii) ΩC\Omega \neq \C provides the point bb of Step 0 — without it the family F\mathcal F is empty of injective bounded maps (Liouville). (iii) Connectedness is used whenever the identity theorem or Hurwitz (Exercise 17.8) speaks: the extremal limit is “injective or constant”, and constancy is excluded by s>0s > 0; also in “zero derivative implies constant”. Failure for C\C: Step 0 impossible, and the conclusion is false (Exercise 18.5(a)). Failure for the annulus: not simply connected — the square-root construction breaks (e.g. zz itself, zero-free on AA, has no holomorphic square root: the same index computation as in Exercise 18.5(b) with 12σ ⁣dzz2iπZ\frac12\int_\sigma\frac{\dd z}z \notin 2\iu\pi\Z) — and the conclusion is false too.

Exercise 18.8 ★★

Solve the Dirichlet problem on D\mathbb D for the boundary data: (a) g(eit)=costg(\eu^{\iu t}) = \cos t; (b) g(eit)=cos2tg(\eu^{\iu t}) = \cos^2 t; (c) g=1upper semicircleg = \mathbf 1_{\text{upper semicircle}} — for (c) compute u(0)u(0) and interpret via the mean value property. (Expand gg in Fourier series and use PrP_r’s series: u(reiφ)=ncn(g)rneinφu(r\eu^{\iu\varphi}) = \sum_n c_n(g) r^{\abs n}\eu^{\iu n\varphi}.)

Solution

Solution of Exercise 18.8.

Substituting the Fourier expansion of gg into the Poisson integral and using 12πPr(φt)eint ⁣dt=rneinφ\frac1{2\pi}\int P_r(\varphi - t)\eu^{\iu nt}\dd t = r^{\abs n}\eu^{\iu n\varphi} (read off PrP_r’s series): u(reiφ)=ncn(g)rneinφu(r\eu^{\iu\varphi}) = \sum_nc_n(g)\,r^{\abs n}\eu^{\iu n\varphi}, the interchange justified by normal convergence (cng\abs{c_n} \leq \norm g_\infty, r<1r < 1).

(a) g=costg = \cos t: c±1=12c_{\pm1} = \frac12, so u=rcosφ=Rez=xu = r\cos\varphi = \operatorname{Re}z = x — indeed harmonic with the right boundary values.

(b) cos2t=12+cos2t2\cos^2t = \frac12 + \frac{\cos 2t}2: u=12+r2cos2φ2=12+Re(z2)2=12+x2y22u = \frac12 + \frac{r^2\cos2\varphi}2 = \frac12 + \frac{\operatorname{Re}(z^2)}2 = \frac12 + \frac{x^2 - y^2}{2}.

(c) g=1(0,π)g = \mathbf 1_{(0,\pi)} (upper semicircle): c0=12c_0 = \frac12 and cn=1(1)n2iπnc_n = \frac{1 - (-1)^n}{2\iu\pi n} for n0n \neq 0, so

u(reiφ)=12+2πk0r2k+1sin((2k+1)φ)2k+1,u(0)=12:u(r\eu^{\iu\varphi}) = \frac12 + \frac2\pi\sum_{k\geq0} \frac{r^{2k+1}\sin\bigl((2k+1)\varphi\bigr)}{2k + 1}, \qquad u(0) = \frac12 :

the center sees exactly the average of the boundary data — the mean value property in person.

Exercise 18.9 ★★★

(Harnack) Let u0u \geq 0 be harmonic on D\mathbb D. Prove, for z=r<1\abs z = r < 1:

1r1+ru(0)    u(z)    1+r1ru(0)\frac{1 - r}{1 + r}\,u(0) \;\leq\; u(z) \;\leq\; \frac{1 + r}{1 - r}\,u(0)

(bound the Poisson kernel between 1r1+r\frac{1-r}{1+r} and 1+r1r\frac{1+r}{1-r}; apply the representation on slightly smaller discs and pass to the limit). Deduce: a harmonic function on C\C bounded below is constant.

Solution

Solution of Exercise 18.9.

From (1r)212rcosθ+r2(1+r)2(1-r)^2 \leq 1 - 2r\cos\theta + r^2 \leq (1+r)^2:

1r1+r=1r2(1+r)2Pr(θ)1r2(1r)2=1+r1r.\frac{1-r}{1+r} = \frac{1 - r^2}{(1+r)^2} \leq P_r(\theta) \leq \frac{1 - r^2}{(1 - r)^2} = \frac{1+r}{1-r} .

For u0u \geq 0 harmonic on D\mathbb D and s<1s < 1: us(z)=u(sz)u_s(z) = u(sz) is harmonic on a neighborhood of Dˉ\bar{\mathbb D}, hence equals its Poisson integral (Theorem 18.12, uniqueness, applied to its own boundary values); sandwiching the kernel and using the mean value 12πus(eit) ⁣dt=u(0)\frac1{2\pi}\int u_s(\eu^{\iu t})\dd t = u(0):

1r1+ru(0)u(sreiφ)1+r1ru(0).\frac{1-r}{1+r}\,u(0) \leq u(s\,r\eu^{\iu\varphi}) \leq \frac{1+r}{1-r}\,u(0).

Let s1s \to 1^- at fixed reiφr\eu^{\iu\varphi} (continuity of uu): the Harnack inequalities. If uu is harmonic on C\C with umu \geq m: apply Harnack to umu - m on discs D(0,R)D(0, R), i.e. to zu(Rz)mz \mapsto u(Rz) - m: for fixed zz and r=z/R0r = \abs z/R \to 0, both bounds tend to u(0)mu(0) - m: u(z)=u(0)u(z) = u(0) — constant (a two-sided Liouville from a one-sided bound).

Exercise 18.10 ★★

Using conformal invariance of harmonicity (ufu\circ f is harmonic when uu is harmonic and ff holomorphic — prove it via Proposition 18.11 locally), solve the Dirichlet problem on the upper half-plane with boundary data 1(,0)\mathbf 1_{\intoo{-\infty}0}: show that

u(x+iy)=1πarg(x+iy)(arg(0,π) on H)u(x + \iu y) = \frac1\pi\,\arg(x + \iu y) \qquad (\arg \in \intoo0\pi \text{ on } \mathbb H)

is harmonic on H\mathbb H (imaginary part of a holomorphic logarithm) with the required boundary limits at every x0x \neq 0, and transport it to the disc by Cayley to re-derive Exercise 18.8(c).

Solution

Solution of Exercise 18.10.

Locally, u=ReFu = \operatorname{Re}F with FF holomorphic (Proposition 18.11), so uf=Re(Ff)u\circ f = \operatorname{Re}(F\circ f) is harmonic wherever defined: harmonicity is conformally invariant. On H\mathbb H: the principal logarithm gives logz=lnz+iargz\log z = \ln\abs z + \iu\arg z holomorphic on H\mathbb H, so u=1πargz=Im(1πlogz)u = \frac1\pi\arg z = \operatorname{Im}\bigl(\frac1\pi\log z\bigr) is harmonic, with boundary limits: for x>0x > 0, arg0\arg \to 0, u0u \to 0; for x<0x < 0, argπ\arg \to \pi, u1u \to 1: the data 1(,0)\mathbf 1_{\intoo{-\infty}0} at every x0x \neq 0. Transporting by the Cayley map (which sends DH\mathbb D \to \mathbb H after inversion and matches the upper semicircle to the negative axis, up to the rotation fixed by chasing three boundary points), u(Cayley)u\circ(\text{Cayley}) solves the disc problem of Exercise 18.8(c); evaluating at the center retrieves u=12u = \frac12 there, and the closed form 1πarg\frac1\pi\arg can be checked against the series by summing r2k+1sin((2k+1)φ)2k+1=12arctan2rsinφ1r2\sum\frac{r^{2k+1}\sin((2k+1)\varphi)}{2k+1} = \frac12\arctan\frac{2r\sin\varphi}{1 - r^2}-type identities — the elementary route to the same answer.

Exercise 18.11 ★★

(Fixed points and iteration in the disc) Let f ⁣:DDf\colon\mathbb D \to \mathbb D be holomorphic. (a) Show that if ff has two distinct fixed points, then f=idf = \mathrm{id} (move one to 00 by an automorphism and apply the equality case of Schwarz). (b) Suppose f(0)=0f(0) = 0 and ff is not a rotation. Show that the iterates fn0f^{\circ n} \to 0 uniformly on every compact Dˉ(0,r)\bar D(0, r), r<1r < 1 (Schwarz gives f(z)crz\abs{f(z)} \leq c_r\abs z on Dˉ(0,r)\bar D(0,r) with cr<1c_r < 1 — justify this strict constant via the maximum principle applied to f(z)/zf(z)/z). (c) Illustrate with f(z)=z2+z2f(z) = \frac{z^2 + z}2: fixed points, and the rate of convergence of the orbit of z0=12z_0 = \frac12.

Solution

Solution of Exercise 18.11.

(a) Let aba \neq b be fixed. Conjugating by φa(z)=za1aˉz\varphi_a(z) = \frac{z - a}{1 - \bar az} (an automorphism exchanging aa and 00), g=φafφa1g = \varphi_a\circ f\circ\varphi_a^{-1} fixes 00 and the point c=φa(b)0c = \varphi_a(b) \neq 0. Schwarz: g(z)z\abs{g(z)} \leq \abs z, and at z=cz = c equality holds (g(c)=cg(c) = c): the equality case forces g(z)=λzg(z) = \lambda z with λ=1\abs\lambda = 1, and λc=c\lambda c = c gives λ=1\lambda = 1: g=idg = \mathrm{id}, hence f=idf = \mathrm{id}.

(b) h(z)=f(z)/zh(z) = f(z)/z (removable singularity at 00) is holomorphic on D\mathbb D with h1\abs h \leq 1 (Schwarz); h<1\abs h < 1 everywhere, else the maximum principle (interior maximum of h\abs h) would make hh a unimodular constant, i.e. ff a rotation — excluded. On the compact Dˉ(0,r)\bar D(0,r), cr=maxh<1c_r = \max\abs h < 1, so f(z)crz\abs{f(z)} \leq c_r\abs z there; moreover ff maps Dˉ(0,r)\bar D(0,r) into itself (crzrc_r\abs z \leq r), so the bound iterates: fn(z)crnr0\abs{f^{\circ n}(z)} \leq c_r^n\,r \to 0 uniformly on Dˉ(0,r)\bar D(0, r).

(c) Fixed points of z2+z2\frac{z^2 + z}2: z2+z=2zz^2 + z = 2z iff z(z1)=0z(z - 1) = 0; only z=0z = 0 lies in D\mathbb D (z=1z = 1 is on the boundary). Not a rotation (f(0)=12f'(0) = \frac12), so orbits tend to 00; quantitatively f(z)=z2(1+z)f(z) = \frac z2(1 + z) gives f(z)34z\abs{f(z)} \leq \frac{3}{4}\abs z on z12\abs z \leq \frac12, and once the orbit is small, f(z)z2\abs{f(z)} \approx \frac{\abs z}2: asymptotically geometric with ratio f(0)=12f'(0) = \frac12. From z0=12z_0 = \frac12: z1=38z_1 = \frac38, z20.258z_2 \approx 0.258, z30.162z_3 \approx 0.162 — halving per step, as predicted by the multiplier.

Exercise 18.12 ★★

(Harmonic conjugates, concretely) Let u(x,y)=x33xy2+2yu(x, y) = x^3 - 3xy^2 + 2y. (a) Check that uu is harmonic on R2\R^2, and find all harmonic conjugates vv (i.e. u+ivu + \iu v holomorphic) by integrating the Cauchy–Riemann equations; identify f(z)=u+ivf(z) = u + \iu v as a polynomial in zz. (b) Show that on a star-shaped open set, every harmonic function admits a harmonic conjugate, unique up to an additive constant (the 11-form uy ⁣dx+ux ⁣dy-u_y\,\dd x + u_x\,\dd y is closed; Theorem 16.8’s primitive machinery, or Chapter 21’s Poincaré lemma). (c) Give the standard counterexample on C\C^*: u=lnzu = \ln\abs z has no global conjugate — relate to the angular form and the winding number (Chapter 21).

Solution

Solution of Exercise 18.12.

(a) Δu=6x6x+0=0\Delta u = 6x - 6x + 0 = 0. Cauchy–Riemann demands vy=ux=3x23y2v_y = u_x = 3x^2 - 3y^2 and vx=uy=6xy2v_x = -u_y = 6xy - 2. Integrating the first in yy: v=3x2yy3+c(x)v = 3x^2y - y^3 + c(x); plugging into the second: 6xy+c(x)=6xy26xy + c'(x) = 6xy - 2, so c(x)=2x+Cc(x) = -2x + C. Thus v=3x2yy32x+Cv = 3x^2y - y^3 - 2x + C and

f=u+iv=(x33xy2)+i(3x2yy3)+2y2ix+iC=z32iz+iC.f = u + \iu v = (x^3 - 3xy^2) + \iu(3x^2y - y^3) + 2y - 2\iu x + \iu C = z^3 - 2\iu z + \iu C .

(b) The form ω=uy ⁣dx+ux ⁣dy\omega = -u_y\,\dd x + u_x\,\dd y is closed precisely because Δu=0\Delta u = 0 (y(uy)=uyy=uxx=x(ux)\partial_y(-u_y) = -u_{yy} = u_{xx} = \partial_x(u_x)). On a star-shaped open set the Poincaré lemma (Theorem 21.15; or the primitive construction of Theorem 16.8 applied to the holomorphic uxiuyu_x - \iu u_y) provides vv with  ⁣dv=ω\dd v = \omega, i.e. the Cauchy–Riemann system: u+ivu + \iu v is holomorphic. Two conjugates differ by a function with vanishing gradient: a constant (connectedness).

(c) For u=lnzu = \ln\abs z on C\C^*: ω=uy ⁣dx+ux ⁣dy=y ⁣dx+x ⁣dyx2+y2=ωθ\omega = -u_y\dd x + u_x\dd y = \frac{-y\,\dd x + x\,\dd y}{x^2 + y^2} = \omega_\theta, the angular form (Example 21.14), whose integral along the unit circle is 2π02\pi \neq 0: not exact, so no global conjugate exists — a conjugate would be a continuous determination of the argument, and the winding number is exactly the obstruction. Locally (on any star-shaped subdomain), v=argzv = \arg z works and u+iv=logzu + \iu v = \log z: the failure is global, not local.

18.7 Problem: the area theorem and Koebe’s quarter theorem

Problem 18.1

Weekend problem — how much must a univalent map cover?

A univalent function is a holomorphic injection. The normalized univalent functions on the disc,

S={fH(D) injective, f(z)=z+a2z2+a3z3+},\mathcal S = \bigl\{f \in \mathcal H(\mathbb D) \text{ injective},\ f(z) = z + a_2z^2 + a_3z^3 + \cdots\bigr\},

are rigidly constrained: we prove Bieberbach’s inequality a22\abs{a_2} \leq 2 and deduce the Koebe quarter theorem: the image of any fSf \in \mathcal S contains the disc D(0,14)D(0, \frac14) — the sharp universal constant of conformal geometry.

Part I — The area theorem. Let g(w)=w+b0+b1w+b2w2+g(w) = w + b_0 + \frac{b_1}w + \frac{b_2}{w^2} + \cdots be holomorphic and injective on {w>1}\{\abs w > 1\}.

  1. For ρ>1\rho > 1, let AρA_\rho be the area (Lebesgue measure) of the compact set Kρ=Cg({w>ρ})K_\rho = \C\setminus g(\{\abs w > \rho\}), the region enclosed by the smooth Jordan curve g(Cρ)g(C_\rho). Using the area formula of the Year 2 volume’s Green–Riemann theorem — the enclosed area is 12iζˉ ⁣dζ\frac1{2\iu} \oint\bar\zeta\,\dd\zeta along the positively oriented boundary — show that

    Aρ=12iCρg(w)g(w) ⁣dw=π(ρ2n1nbn2ρ2n):A_\rho = \frac{1}{2\iu}\int_{C_\rho} \overline{g(w)}\,g'(w)\,\dd w = \pi\Bigl(\rho^2 - \sum_{n\geq1}n\,\abs{b_n}^2\rho^{-2n}\Bigr) :

    substitute the Laurent series of gˉ\bar g and gg' on CρC_\rho and integrate term by term (normal convergence; only the frequency-zero products survive).

  2. Let ρ1+\rho \to 1^+ and conclude the area theorem:

    n1nbn2    1.\sum_{n\geq1}n\,\abs{b_n}^2 \;\leq\; 1 .

    In particular b11\abs{b_1} \leq 1. When is b1=1\abs{b_1} = 1?

Part II — Bieberbach’s a22\abs{a_2} \leq 2. Let f=z+a2z2+Sf = z + a_2z^2 + \cdots \in \mathcal S.

  1. Show that f(z2)/z2f(z^2)/z^2 is holomorphic and zero-free on D\mathbb D, and admits a holomorphic square root φ\varphi with φ(0)=1\varphi(0) = 1; set h(z)=zφ(z2)h(z) = z\varphi(z^2), so that h(z)2=f(z2)h(z)^2 = f(z^2). Show that hh is an odd univalent function on D\mathbb D with expansion h(z)=z+a22z3+h(z) = z + \frac{a_2}2z^3 + \cdots. (Injectivity: h(z)2=h(z)2h(z)^2 = h(z')^2 forces z2=z2z^2 = z'^2; use oddness to finish.)
  2. Apply the area theorem to g(w)=1/h(1/w)=wa22w+g(w) = 1/h(1/w) = w - \frac{a_2}{2w} + \cdots on {w>1}\{\abs w > 1\} (verify univalence and the expansion), and conclude a22\abs{a_2} \leq 2.
  3. Show that the Koebe function

    k(z)=z(1z)2=n1nznk(z) = \frac{z}{(1 - z)^2} = \sum_{n\geq1}n\,z^n

    belongs to S\mathcal S, has a2=2a_2 = 2, and maps D\mathbb D onto C(,14]\C\setminus\intoc{-\infty}{-\frac14} (write k=14[(1+z1z)21]k = \frac14\bigl[\bigl(\frac{1+z}{1 - z}\bigr)^2 - 1\bigr] and track the images): all inequalities to come are sharp.

Part III — The quarter theorem.

  1. Let fSf \in \mathcal S and cf(D)c \notin f(\mathbb D). Show that

    F(z)=cf(z)cf(z)F(z) = \frac{c\,f(z)}{c - f(z)}

    belongs to S\mathcal S, and compute its second coefficient: A2=a2+1cA_2 = a_2 + \frac1c.

  2. Apply Bieberbach to both ff and FF: conclude 1cA2+a24\abs{\frac1c} \leq \abs{A_2} + \abs{a_2} \leq 4, i.e. c14\abs c \geq \frac14. Every omitted value has modulus 14\geq \frac14: f(D)D(0,14)f(\mathbb D) \supseteq D(0, \frac14) — Koebe’s quarter theorem. Check sharpness on the Koebe function.
  3. Deduce a quantitative Riemann-map estimate: if φ ⁣:ΩD\varphi \colon \Omega \to \mathbb D is the Riemann map of Theorem 18.9 at z0z_0, then

    d(z0,Ω)4    1φ(z0)    4d(z0,Ω)\frac{d\bigl(z_0, \partial\Omega\bigr)}{4} \;\leq\; \frac1{\varphi'(z_0)} \;\leq\; 4\,d\bigl(z_0, \partial\Omega\bigr)

    — prove at least the left inequality by applying Koebe to φ1\varphi^{-1} suitably normalized, and the right by Schwarz applied to φ\varphi on the disc D(z0,d)ΩD(z_0, d) \subseteq \Omega.

Part IV — Perspective.

  1. Bieberbach conjectured (1916) ann\abs{a_n} \leq n for all nn, with equality only for rotations of the Koebe function; de Branges proved it in 1985. Verify the conjecture by hand for the Koebe function and its rotations eiθk(eiθz)\eu^{-\iu\theta}k(\eu^{\iu\theta}z). Then push question 4’s expansion one term further: writing h(z)=z+a22z3+c5z5+h(z) = z + \frac{a_2}2z^3 + c_5z^5 + \cdots, show c5=a32a228c_5 = \frac{a_3}2 - \frac{a_2^2}8 and

    g(w)=wa22w1+(3a228a32)w3+,g(w) = w - \frac{a_2}{2}\,w^{-1} + \Bigl(\frac{3a_2^2}8 - \frac{a_3}2\Bigr)w^{-3} + \cdots ,

    so the area theorem yields the refined inequality a222+33a228a3221\bigl|\frac{a_2}2\bigr|^2 + 3\bigl|\frac{3a_2^2}8 - \frac{a_3}2\bigr|^2 \leq 1. Check it on the Koebe function (a2=2a_2 = 2, a3=3a_3 = 3).

Part V — The distortion theorem. Bieberbach’s inequality, transported around the disc by automorphisms, controls ff' everywhere. Fix fSf \in \mathcal S.

  1. (Koebe transform) For z0Dz_0 \in \mathbb D let φ(z)=z+z01+zˉ0z\varphi(z) = \frac{z + z_0}{1 + \bar z_0z}, a disc automorphism (Theorem 18.5) with φ(0)=z0\varphi(0) = z_0. Show that

    F(z)=f(φ(z))f(z0)f(z0)(1z02)F(z) = \frac{f(\varphi(z)) - f(z_0)} {f'(z_0)\,\bigl(1 - \abs{z_0}^2\bigr)}

    belongs to S\mathcal S (univalence is inherited; compute φ(0)=1z02\varphi'(0) = 1 - \abs{z_0}^2 and check the normalization; recall f0f' \neq 0 for injective ff, Definition 18.1).

  2. Compute the second coefficient A2=12F(0)A_2 = \frac12F''(0) of FF:

    A2=12[(1z02)f(z0)f(z0)2zˉ0],A_2 = \frac12\Bigl[\bigl(1 - \abs{z_0}^2\bigr) \frac{f''(z_0)}{f'(z_0)} - 2\bar z_0\Bigr] ,

    and deduce from Bieberbach (question 4), for z=reiθz = r\eu^{\iu\theta}, the fundamental inequality:

    zf(z)f(z)2r21r2    4r1r2.\Bigl|\,z\,\frac{f''(z)}{f'(z)} - \frac{2r^2}{1 - r^2}\Bigr| \;\leq\; \frac{4r}{1 - r^2} .
  3. Extract the real part:

    2r24r1r2    Re(zf(z)f(z))    2r2+4r1r2.\frac{2r^2 - 4r}{1 - r^2} \;\leq\; \operatorname{Re}\Bigl(z\,\frac{f''(z)}{f'(z)}\Bigr) \;\leq\; \frac{2r^2 + 4r}{1 - r^2} .
  4. Show that  ⁣d ⁣dtlogf(teiθ)=1tRe(zf(z)f(z))\frac{\dd}{\dd t}\log\bigl| f'(t\eu^{\iu\theta})\bigr| = \frac1t \operatorname{Re}\bigl(z\frac{f''(z)}{f'(z)}\bigr) at z=teiθz = t\eu^{\iu\theta} (for a nonvanishing C1\mathcal C^1 function gg of a real variable,  ⁣d ⁣dtlogg=Re(g/g)\frac{\dd}{\dd t}\log\abs g = \operatorname{Re}(g'/g)), and integrate question 12’s bounds along the ray to obtain the distortion theorem:

    1r(1+r)3    f(z)    1+r(1r)3,z=r.\frac{1 - r}{(1 + r)^3} \;\leq\; \abs{f'(z)} \;\leq\; \frac{1 + r}{(1 - r)^3}, \qquad \abs z = r .
  5. Deduce the growth theorem:

    r(1+r)2    f(z)    r(1r)2,z=r\frac{r}{(1 + r)^2} \;\leq\; \abs{f(z)} \;\leq\; \frac{r}{(1 - r)^2}, \qquad \abs z = r

    (upper bound: integrate ff' on the segment [0,z][0, z]; lower bound: if f(z)<14\abs{f(z)} < \frac14, the segment [0,f(z)][0, f(z)] lies in f(D)f(\mathbb D) by question 7; pull it back by f1f^{-1}holomorphic by Corollary 17.10 — and bound f(z)=γf(ζ) ⁣dζ0r1t(1+t)3 ⁣dt\abs{f(z)} = \int_\gamma\abs{f'(\zeta)} \,\abs{\dd\zeta} \geq \int_0^r\frac{1 - t}{(1 + t)^3}\,\dd t, using  ⁣dζ ⁣dζ\abs{\dd\zeta} \geq \dd\abs\zeta).

  6. Verify that the Koebe function achieves equality in all four bounds, at z=rz = r for the upper ones and z=rz = -r for the lower ones: kk is simultaneously the most expanding and, at the antipode, the most contracting member of S\mathcal S.

Part VI — Extremal rigidity. In every inequality so far, equality identifies the Koebe function up to rotation. We prove it, then harvest.

  1. Suppose fSf \in \mathcal S has a2=2\abs{a_2} = 2. Chase equality through questions 2–4: the area theorem forces g(w)=w+b0+eiα/wg(w) = w + b_0 + \eu^{\iu\alpha}/w; oddness of hh makes gg odd, so b0=0b_0 = 0; invert to find hh, then ff, and conclude that

    f(z)=eiθk(eiθz)with eiθ=eiα:f(z) = \eu^{-\iu\theta}k\bigl(\eu^{\iu\theta}z\bigr) \quad\text{with } \eu^{\iu\theta} = -\eu^{\iu\alpha} :

    the rotations of the Koebe function are the only members of S\mathcal S with a2=2\abs{a_2} = 2.

  2. Show that if fSf \in \mathcal S omits a value cc with c=14\abs c = \frac14 exactly, then ff is a rotation of the Koebe function, and identify c=eiθ/4c = -\eu^{-\iu\theta}/4 (trace equality through question 7’s chain 4=1/c=A2a2A2+a244 = \abs{1/c} = \abs{A_2 - a_2} \leq \abs{A_2} + \abs{a_2} \leq 4): the quarter theorem’s constant is attained only by the extremal family.
  3. (Coefficients on the cheap) Combine the growth theorem with the Cauchy estimates (Theorem 16.10) on the circle z=11n\abs z = 1 - \frac1n to prove

    an    en2(n2).\abs{a_n} \;\leq\; \eu\,n^2 \qquad (n \geq 2) .

    (De Branges, 1985: ann\abs{a_n} \leq n; the factor en\eu n is the price of elementary tools.)

  4. (Covering of subdiscs) Show that for every 0<r<10 < r < 1,

    f(D(0,r))D(0,r(1+r)2),f\bigl(D(0, r)\bigr) \supseteq D\Bigl(0, \frac{r}{(1 + r)^2}\Bigr),

    sharp for the Koebe function, and recover the quarter theorem as r1r \to 1^-. (Boundary points of the open image f(D(0,r))f(D(0,r)) lie on f(D(0,r))f(\partial D(0,r)), hence have modulus r(1+r)2\geq \frac{r}{(1+r)^2} by question 14; a segment from 00 to a missed point of smaller modulus would have to cross that boundary.)

  5. (Koebe at every point) Let ff be univalent on D\mathbb D, not necessarily normalized, and z0Dz_0 \in \mathbb D. Prove

    14(1z02)f(z0)    d(f(z0),f(D))    (1z02)f(z0)\tfrac14\bigl(1 - \abs{z_0}^2\bigr)\abs{f'(z_0)} \;\leq\; d\bigl(f(z_0), \partial f(\mathbb D)\bigr) \;\leq\; \bigl(1 - \abs{z_0}^2\bigr)\abs{f'(z_0)}

    (left: quarter theorem applied to the Koebe transform of question 10; right: Schwarz (Theorem 18.4) applied to ψ1g^\psi^{-1}\circ\hat g, where g^(w)=f1(f(z0)+dw)\hat g(w) = f^{-1}\bigl(f(z_0) + dw\bigr), dd the distance, and ψ\psi a disc automorphism sending 00 to z0z_0). Why is f(D)\partial f(\mathbb D) nonempty?

  6. Check question 20 on f=kf = k at z0=r(0,1)z_0 = r \in \intoo01: compute d(k(r),k(D))=(1+r)24(1r)2d\bigl(k(r), \partial k(\mathbb D)\bigr) = \frac{(1+r)^2}{4(1-r)^2} and verify that the left inequality is an equality: the Koebe function saturates its own theorem at every point of (0,1)\intoo01.
  7. (The moral) In ten lines: what single principle underlies the area theorem, and how do Bieberbach, the quarter theorem, distortion, growth, and covering all flow from it? Compare with the Schwarz–Pick world of Theorem 18.4: in both, one interior inequality rigidifies the whole geometry, and the extremals are unique up to rotation.

Part VII — Compactness, inverses, and a reality check.

  1. Show that the class S\mathcal S is compact for locally uniform convergence: it is locally bounded by the growth theorem, hence normal (Theorem 18.7); and a locally uniform limit of members of S\mathcal S is again in S\mathcal S (the normalizations pass to the limit by Weierstrass convergence of derivatives; injectivity survives by Hurwitz, Exercise 17.8, the limit being nonconstant). Why does this matter for extremal problems like Bieberbach’s?
  2. For fSf \in \mathcal S, let g=f1g = f^{-1}, defined near 00. Show g(w)=wa2w2+O(w3)g(w) = w - a_2w^2 + O(w^3), so the inverse’s second coefficient obeys the same sharp bound A2=a22\abs{A_2} = \abs{a_2} \leq 2, with equality exactly for the rotated Koebe functions.
  3. Determine for which aCa \in \C the polynomial f(z)=z+az2f(z) = z + az^2 belongs to S\mathcal S: show ff is injective on D\mathbb D iff a12\abs a \leq \frac12 (factor f(z1)f(z2)f(z_1) - f(z_2)). Conclude: for degree-two polynomials the true coefficient bound is 12\frac12, four times smaller than Bieberbach’s 22 — the extremals of S\mathcal S are genuinely transcendental objects, and no polynomial comes close.
Solution

Solution of Problem 18.1.

1. gg is injective and holomorphic; on CρC_\rho (ρ>1\rho > 1) it is smooth, and the enclosed area is 12ig(Cρ)ζˉ ⁣dζ\frac1{2\iu}\oint_{g(C_\rho)}\bar\zeta\,\dd\zeta (the Year 2 Green–Riemann area formula, applied with positive orientation). Substituting ζ=g(w)\zeta = g(w), w=ρeiθw = \rho\eu^{\iu\theta}:

Aρ=12iCρg(w)g(w) ⁣dw=ρ202πg(ρeiθ)g(ρeiθ)eiθ ⁣dθ.A_\rho = \frac1{2\iu}\int_{C_\rho}\overline{g(w)}\,g'(w)\dd w = \frac\rho2\int_0^{2\pi}\overline{g(\rho\eu^{\iu\theta})} \,g'(\rho\eu^{\iu\theta})\,\eu^{\iu\theta}\,\dd\theta .

Insert gˉ=ρeiθ+bˉ0+mbˉmρmeimθ\bar g = \rho\eu^{-\iu\theta} + \bar b_0 + \sum_m\bar b_m\rho^{-m}\eu^{\iu m\theta} and g=1nnbnρn1ei(n+1)θg' = 1 - \sum_nnb_n\rho^{-n-1}\eu^{-\iu(n+1)\theta}: after multiplying by eiθ\eu^{\iu\theta}, only frequency-zero products survive the θ\theta-integration (normal convergence justifies term-by-term work): the pair (ρeiθ)1(\rho\eu^{-\iu\theta})\cdot1 contributes 2πρ2\pi\rho, and each pair bˉnρneinθ(nbnρn1ei(n+1)θ)\bar b_n\rho^{-n}\eu^{\iu n\theta}\cdot(-nb_n\rho^{-n-1}\eu^{-\iu(n+1)\theta}) contributes 2πnbn2ρ2n1-2\pi n\abs{b_n}^2\rho^{-2n-1}:

Aρ=πρ2πn1nbn2ρ2n.A_\rho = \pi\rho^2 - \pi\sum_{n\geq1}n\abs{b_n}^2\rho^{-2n}.

2. Areas are nonnegative: nNnbn2ρ2nρ2\sum_{n\leq N}n\abs{b_n}^2\rho^{-2n} \leq \rho^2 for every NN; let ρ1+\rho \to 1^+ then NN \to \infty: n1nbn21\sum_{n\geq1}n\abs{b_n}^2 \leq 1. Equality in b11\abs{b_1} \leq 1 forces all other bn=0b_n = 0: g(w)=w+b0+eiα/wg(w) = w + b_0 + \eu^{\iu\alpha}/w, which maps onto the complement of a segment of length 44 (a Joukowski-type map): the extremals.

3. f(z)/z=1+a2z+f(z)/z = 1 + a_2z + \cdots is holomorphic and zero-free on D\mathbb D (ff vanishes only at 00, simply: injectivity), hence so is f(z2)/z2f(z^2)/z^2, which has a holomorphic square root φ\varphi with φ(0)=1\varphi(0) = 1 (Definition 18.8; D\mathbb D is convex). Then h(z)=zφ(z2)h(z) = z\varphi(z^2) satisfies h2=f(z2)h^2 = f(z^2), h(z)=z(1+a22z2+)h(z) = z\bigl(1 + \frac{a_2}2z^2 + \cdots\bigr) (binomial series for the root), and h(z)=zφ(z2)h(z) = z\varphi(z^2) is odd by construction (φ(z2)\varphi(z^2) is even). Injectivity: h(z)=h(z)h(z) = h(z') gives f(z2)=f(z2)f(z^2) = f(z'^2), so z2=z2z^2 = z'^2, i.e. z=±zz' = \pm z; if z=zz' = -z then oddness gives h(z)=h(z)h(z) = -h(z), so h(z)=0h(z) = 0, forcing z=0z = 0 (φ\varphi zero-free): z=z=0z = z' = 0.

4. g(w)=1/h(1/w)g(w) = 1/h(1/w): for w>1\abs w > 1, 1/wD{0}1/w \in \mathbb D\setminus\{0\} and h0h \neq 0 there: well defined, injective (composition of injections), with expansion

g(w)=11w(1+a22w2+)=w(1a22w2+)=wa22w1+:g(w) = \frac{1}{\frac1w\bigl(1 + \frac{a_2}{2w^2} + \cdots\bigr)} = w\Bigl(1 - \frac{a_2}{2w^2} + \cdots\Bigr) = w - \frac{a_2}{2}\,w^{-1} + \cdots :

of the Part I form with b1=a22b_1 = -\frac{a_2}2. The area theorem gives a221\abs{\frac{a_2}2} \leq 1: a22\abs{a_2} \leq 2.

5. k(z)=z(1z)2=zm0(m+1)zm=n1nznk(z) = \frac z{(1-z)^2} = z\sum_{m\geq0}(m + 1)z^m = \sum_{n\geq1}nz^n: coefficients an=na_n = n, so a2=2a_2 = 2. Univalence and image: k=14[w21]k = \frac14\bigl[w^2 - 1\bigr] with w=1+z1zw = \frac{1 + z}{1 - z}, a conformal map of D\mathbb D onto the right half-plane; w2w^2 maps that half-plane conformally onto C(,0]\C\setminus\intoc{-\infty}0; then 14\frac{\cdot - 1}4 gives C(,14]\C\setminus\intoc{-\infty}{-\frac14}: injective at each stage, image as claimed.

6. F=cfcfF = \frac{cf}{c - f}: since cf(D)c \notin f(\mathbb D), the denominator never vanishes: FF is holomorphic, and injective (wcwcww \mapsto \frac{cw}{c - w} is Möbius, injective off w=cw = c). Expansion: with f=z+a2z2+f = z + a_2z^2 + \cdots,

F=f11f/c=(z+a2z2)(1+zc)+O(z3)=z+(a2+1c)z2+O(z3):F = f\cdot\frac1{1 - f/c} = \bigl(z + a_2z^2\bigr)\Bigl(1 + \frac zc\Bigr) + O(z^3) = z + \Bigl(a_2 + \frac1c\Bigr)z^2 + O(z^3):

FSF \in \mathcal S with A2=a2+1cA_2 = a_2 + \frac1c.

7. Bieberbach twice: a22\abs{a_2} \leq 2 and a2+1c2\abs{a_2 + \frac1c} \leq 2, so 1c4\abs{\frac1c} \leq 4: c14\abs c \geq \frac14. Every omitted value lies outside D(0,14)D(0,\frac14), i.e. f(D)D(0,14)f(\mathbb D) \supseteq D(0, \frac14). Sharp: the Koebe function omits 14-\frac14 (question 5).

8. Let d=d(z0,Ω)d = d(z_0, \partial\Omega) and ψ=φ1 ⁣:DΩ\psi = \varphi^{-1} \colon \mathbb D \to \Omega, ψ(0)=z0\psi(0) = z_0, ψ(0)=1/φ(z0)>0\psi'(0) = 1/\varphi'(z_0) > 0. The normalization f~(z)=ψ(z)z0ψ(0)\tilde f(z) = \frac{\psi(z) - z_0}{\psi'(0)} lies in S\mathcal S, so its image contains D(0,14)D(0, \frac14); scaling back, Ω=ψ(D)D(z0,ψ(0)4)\Omega = \psi(\mathbb D) \supseteq D\bigl(z_0, \tfrac{\psi'(0)}4\bigr), so dψ(0)4=14φ(z0)d \geq \frac{\psi'(0)}4 = \frac1{4\varphi'(z_0)}: the left inequality. For the right: φ(z0+d)\varphi\circ(z_0 + d\,\cdot) maps D\mathbb D into D\mathbb D with 000 \mapsto 0; Schwarz bounds its derivative at 00: dφ(z0)1d\,\varphi'(z_0) \leq 1, i.e. 1φ(z0)d\frac1{\varphi'(z_0)} \geq d — together

d1φ(z0)4d.d \leq \frac1{\varphi'(z_0)} \leq 4d .

(The left inequality as displayed in the statement is the same chain rearranged.)

9. For the Koebe function an=na_n = n: equality throughout the conjecture; its rotations eiθk(eiθz)=nei(n1)θzn\eu^{-\iu\theta}k(\eu^{\iu\theta}z) = \sum n\eu^{\iu(n-1)\theta}z^n have an=n\abs{a_n} = n too. Pushing question 4: h=z+a22z3+c5z5+h = z + \frac{a_2}2z^3 + c_5z^5 + \cdots with h2=f(z2)h^2 = f(z^2); comparing z6z^6-coefficients: 2c5+a224=a32c_5 + \frac{a_2^2}4 = a_3, so c5=a32a228c_5 = \frac{a_3}2 - \frac{a_2^2}8. Then

g(w)=1h(1/w)=w(1a22w2+(a224c5)1w4+)=wa22w1+(3a228a32)w3+,g(w) = \frac1{h(1/w)} = w\Bigl(1 - \frac{a_2}{2w^2} + \Bigl(\frac{a_2^2}4 - c_5\Bigr)\frac1{w^4} + \cdots\Bigr) = w - \frac{a_2}2w^{-1} + \Bigl(\frac{3a_2^2}8 - \frac{a_3}2\Bigr)w^{-3} + \cdots,

and the area theorem (nbn21\sum n\abs{b_n}^2 \leq 1) yields

a222+33a228a3221.\Bigl|\frac{a_2}2\Bigr|^2 + 3\,\Bigl|\frac{3a_2^2}8 - \frac{a_3}2\Bigr|^2 \leq 1 .

Koebe check (a2=2a_2 = 2, a3=3a_3 = 3): a222=1\abs{\frac{a_2}2}^2 = 1 and 34832=0\frac{3\cdot4}8 - \frac32 = 0: total exactly 11 — extremal, as it must be.

10. φ\varphi is a disc automorphism with φ(0)=z0\varphi(0) = z_0, so fφf\circ\varphi is univalent on D\mathbb D (composition of injections), and f(z0)0f'(z_0) \neq 0 (Definition 18.1): FF is well defined and univalent. F(0)=0F(0) = 0. Quotient rule:

φ(z)=(1+zˉ0z)(z+z0)zˉ0(1+zˉ0z)2=1z02(1+zˉ0z)2,φ(0)=1z02,\varphi'(z) = \frac{(1 + \bar z_0z) - (z + z_0)\bar z_0} {(1 + \bar z_0z)^2} = \frac{1 - \abs{z_0}^2}{(1 + \bar z_0z)^2}, \qquad \varphi'(0) = 1 - \abs{z_0}^2 ,

so (fφ)(0)=f(z0)(1z02)(f\circ\varphi)'(0) = f'(z_0)(1 - \abs{z_0}^2) and F(0)=1F'(0) = 1: FSF \in \mathcal S.

11. With G=fφG = f\circ\varphi: G=f(φ)φ2+f(φ)φG'' = f''(\varphi)\,\varphi'^2 + f'(\varphi)\,\varphi'', and φ(z)=2zˉ0(1z02)(1+zˉ0z)3\varphi''(z) = -2\bar z_0(1 - \abs{z_0}^2)(1 + \bar z_0z)^{-3} gives φ(0)=2zˉ0(1z02)\varphi''(0) = -2\bar z_0(1 - \abs{z_0}^2). Hence

A2=G(0)2f(z0)(1z02)=12[(1z02)f(z0)f(z0)2zˉ0].A_2 = \frac{G''(0)}{2f'(z_0)(1 - \abs{z_0}^2)} = \frac12\Bigl[\bigl(1 - \abs{z_0}^2\bigr) \frac{f''(z_0)}{f'(z_0)} - 2\bar z_0\Bigr] .

Bieberbach for FF (question 4): A22\abs{A_2} \leq 2, i.e. (1z02)f(z0)f(z0)2zˉ04\bigl|(1 - \abs{z_0}^2)\frac{f''(z_0)}{f'(z_0)} - 2\bar z_0\bigr| \leq 4. Multiply through by z0/(1z02)z_0/(1 - \abs{z_0}^2), whose modulus is r/(1r2)r/(1 - r^2) for z0=reiθz_0 = r\eu^{\iu\theta}, and use z0zˉ0=r2z_0\bar z_0 = r^2:

z0f(z0)f(z0)2r21r24r1r2.\Bigl|\,z_0\frac{f''(z_0)}{f'(z_0)} - \frac{2r^2}{1 - r^2}\Bigr| \leq \frac{4r}{1 - r^2} .

12. A complex number within distance ρ\rho of the real point cc has real part in [cρ,c+ρ]\intcc{c - \rho}{c + \rho}: apply this to c=2r21r2c = \frac{2r^2}{1-r^2}, ρ=4r1r2\rho = \frac{4r}{1-r^2}.

13. For nonvanishing C1\mathcal C^1 gg: logg=12log(ggˉ)\log\abs g = \frac12\log(g\bar g), so  ⁣d ⁣dtlogg=ggˉ+ggˉ2g2=Regg\frac{\dd}{\dd t}\log\abs g = \frac{g'\bar g + g\bar g'}{2\abs g^2} = \operatorname{Re}\frac{g'}g. With g(t)=f(teiθ)g(t) = f'(t\eu^{\iu\theta}) (zero-free: univalence), g(t)=eiθf(teiθ)g'(t) = \eu^{\iu\theta}f''(t\eu^{\iu\theta}), so at z=teiθz = t\eu^{\iu\theta}:

 ⁣d ⁣dtlogf(teiθ)=Re(eiθff)=1tRe(zff)[2t41t2, 2t+41t2]\frac{\dd}{\dd t}\log\bigl|f'(t\eu^{\iu\theta})\bigr| = \operatorname{Re}\Bigl(\eu^{\iu\theta} \frac{f''}{f'}\Bigr) = \frac1t\operatorname{Re}\Bigl(z\frac{f''}{f'}\Bigr) \in \Bigl[\frac{2t - 4}{1 - t^2},\ \frac{2t + 4}{1 - t^2}\Bigr]

by question 12 at radius tt. Since  ⁣d ⁣dtlog1+t(1t)3=11+t+31t=2t+41t2\frac{\dd}{\dd t}\log\frac{1+t}{(1-t)^3} = \frac1{1+t} + \frac3{1-t} = \frac{2t+4}{1-t^2} and  ⁣d ⁣dtlog1t(1+t)3=11t31+t=2t41t2\frac{\dd}{\dd t}\log\frac{1-t}{(1+t)^3} = \frac{-1}{1-t} - \frac3{1+t} = \frac{2t-4}{1-t^2}, integrating from 00 to rr (all three functions vanish at t=0t = 0, f(0)=1f'(0) = 1) gives

log1r(1+r)3logf(z)log1+r(1r)3:\log\frac{1-r}{(1+r)^3} \leq \log\abs{f'(z)} \leq \log\frac{1+r}{(1-r)^3} :

the distortion theorem, after exponentiating.

14. Upper: along the segment [0,z][0, z],

f(z)=0rf(teiθ)eiθ ⁣dt0r1+t(1t)3 ⁣dt=r(1r)2\abs{f(z)} = \Bigl|\int_0^rf'(t\eu^{\iu\theta}) \eu^{\iu\theta}\dd t\Bigr| \leq \int_0^r\frac{1+t}{(1-t)^3}\dd t = \frac{r}{(1-r)^2}

( ⁣d ⁣dtt(1t)2=1+t(1t)3\frac{\dd}{\dd t}\frac t{(1-t)^2} = \frac{1+t}{(1-t)^3}). Lower: note r(1+r)2<14\frac r{(1+r)^2} < \frac14 for r<1r < 1 (it says (1r)2>0(1-r)^2 > 0), so if f(z)14\abs{f(z)} \geq \frac14 there is nothing to prove. Otherwise the segment [0,f(z)][0, f(z)] lies in D(0,14)f(D)D(0, \frac14) \subseteq f(\mathbb D) (question 7), and γ=f1[0,f(z)]\gamma = f^{-1}\circ[0, f(z)] is a C1\mathcal C^1 path from 00 to zz in D\mathbb D (f1f^{-1} holomorphic, Corollary 17.10). Substituting w=f(ζ)w = f(\zeta),

f(z)=[0,f(z)] ⁣dw=γf(ζ) ⁣dζ011ρ(s)(1+ρ(s))3γ(s) ⁣ds\abs{f(z)} = \int_{[0,f(z)]}\abs{\dd w} = \int_\gamma\abs{f'(\zeta)}\,\abs{\dd\zeta} \geq \int_0^1\frac{1 - \rho(s)}{(1 + \rho(s))^3} \,\abs{\gamma'(s)}\,\dd s

with ρ=γ\rho = \abs\gamma, using the lower distortion bound at radius ρ(s)\rho(s). Since γρ\abs{\gamma'} \geq \rho' (wherever defined; ρ\rho is Lipschitz) and the integrand factor is positive,

f(z)011ρ(s)(1+ρ(s))3ρ(s) ⁣ds=0r1u(1+u)3 ⁣du=r(1+r)2\abs{f(z)} \geq \int_0^1 \frac{1 - \rho(s)}{(1 + \rho(s))^3}\,\rho'(s)\,\dd s = \int_0^{r}\frac{1 - u}{(1 + u)^3}\,\dd u = \frac{r}{(1 + r)^2}

(ρ(0)=0\rho(0) = 0, ρ(1)=r\rho(1) = r; the substitution uses only an antiderivative,  ⁣d ⁣duu(1+u)2=1u(1+u)3\frac{\dd}{\dd u}\frac u{(1+u)^2} = \frac{1-u}{(1+u)^3}, not monotonicity).

15. k(z)= ⁣d ⁣dzz(1z)2=1+z(1z)3k'(z) = \frac{\dd}{\dd z}\,z(1 - z)^{-2} = \frac{1 + z}{(1 - z)^3}. At z=rz = r: k(r)=1+r(1r)3k'(r) = \frac{1+r}{(1-r)^3} and k(r)=r(1r)2k(r) = \frac r{(1-r)^2} — both upper bounds attained. At z=rz = -r: k(r)=1r(1+r)3k'(-r) = \frac{1-r}{(1+r)^3} and k(r)=r(1+r)2\abs{k(-r)} = \frac r{(1+r)^2} — both lower bounds attained. The Koebe function stretches its positive axis maximally toward the far boundary and compresses the antipodal ray maximally toward the tip of its slit.

16. a2=2\abs{a_2} = 2 means b1=1\abs{b_1} = 1 for g(w)=1/h(1/w)=wa22w1+g(w) = 1/h(1/w) = w - \frac{a_2}2w^{-1} + \cdots (question 4); the area theorem nbn21\sum n\abs{b_n}^2 \leq 1 then kills every other coefficient: g(w)=w+b0+eiα/wg(w) = w + b_0 + \eu^{\iu\alpha}/w with eiα=a22\eu^{\iu\alpha} = -\frac{a_2}2. Since hh is odd, g(w)=1/h(1/w)=1/h(1/w)=g(w)g(-w) = 1/h(-1/w) = -1/h(1/w) = -g(w): gg is odd, so b0=0b_0 = 0. Inverting, h(z)=1/g(1/z)=z1+eiαz2h(z) = 1/g(1/z) = \frac{z}{1 + \eu^{\iu\alpha}z^2}, and f(z2)=h(z)2f(z^2) = h(z)^2 gives

f(w)=w(1+eiαw)2=w(1eiθw)2=eiθk(eiθw),eiθ=eiα.f(w) = \frac{w}{(1 + \eu^{\iu\alpha}w)^2} = \frac{w}{(1 - \eu^{\iu\theta}w)^2} = \eu^{-\iu\theta}k\bigl(\eu^{\iu\theta}w\bigr), \qquad \eu^{\iu\theta} = -\eu^{\iu\alpha} .

Conversely each rotation has a2=2eiθa_2 = 2\eu^{\iu\theta} of modulus 22: the extremals of Bieberbach are exactly the rotated Koebe functions.

17. If cc is omitted with c=14\abs c = \frac14: question 6 gives FSF \in \mathcal S with A2=a2+1cA_2 = a_2 + \frac1c, so 1c=A2a2\frac1c = A_2 - a_2 and

4=1c=A2a2A2+a22+2=4:4 = \Bigl|\frac1c\Bigr| = \abs{A_2 - a_2} \leq \abs{A_2} + \abs{a_2} \leq 2 + 2 = 4 :

equality throughout, in particular a2=2\abs{a_2} = 2. By question 16, f=eiθk(eiθ)f = \eu^{-\iu\theta}k(\eu^{\iu\theta}\cdot), whose omitted set is eiθ(,14]\eu^{-\iu\theta} \intoc{-\infty}{-\frac14}: the unique omitted value of modulus 14\frac14 is c=eiθ/4c = -\eu^{-\iu\theta}/4. (Check: a2=2eiθa_2 = 2\eu^{\iu\theta} and 1c=4eiθ\frac1c = -4\eu^{\iu\theta}, so A2=2eiθ=a2A_2 = -2\eu^{\iu\theta} = -a_2: the triangle inequality is saturated by anti-alignment, as it must be.)

18. The Cauchy estimates on the circle z=r\abs z = r (Theorem 16.10), combined with the growth theorem, give

anrnsupz=rfr1n(1r)2.\abs{a_n} \leq r^{-n}\sup_{\abs z = r}\abs f \leq r^{1-n}\,(1 - r)^{-2} .

Choose r=11nr = 1 - \frac1n (n2n \geq 2): r1n=(1+1n1)n1<er^{1-n} = \bigl(1 + \frac1{n-1}\bigr)^{n-1} < \eu (increasing sequence with limit e\eu) and (1r)2=n2(1 - r)^{-2} = n^2: an<en2\abs{a_n} < \eu\,n^2.

19. U=f(D(0,r))U = f(D(0, r)) is open (Corollary 17.10) and contains 00. Boundary: if pUp \in \partial U, write p=limf(zk)p = \lim f(z_k) with zkD(0,r)z_k \in D(0, r); a subsequence gives zkzDˉ(0,r)z_k \to z_\infty \in \bar D(0, r), and f(z)=pUf(z_\infty) = p \notin U forces zD(0,r)z_\infty \in \partial D(0, r): Uf(D(0,r))\partial U \subseteq f(\partial D(0, r)), so every boundary point of UU has modulus r(1+r)2\geq \frac r{(1+r)^2} (question 14). Now let w<r(1+r)2\abs w < \frac r{(1+r)^2} and suppose wUw \notin U. The segment [0,w][0, w] is connected, meets UU (at 00) and its complement (at ww), so it meets U\partial U; but all its points have modulus w<r(1+r)2\leq \abs w < \frac r{(1+r)^2}: contradiction. Hence D(0,r(1+r)2)UD\bigl(0, \frac r{(1+r)^2}\bigr) \subseteq U. Sharpness: k(r)=r(1+r)2k(-r) = -\frac r{(1+r)^2} is the image of a boundary point of D(0,r)D(0,r), and kk is injective, so k(r)k(D(0,r))k(-r) \notin k(D(0, r)): the radius cannot be increased. As r1r \to 1^-, r(1+r)214\frac r{(1+r)^2} \to \frac14: the quarter theorem for the full disc.

20. First, f(D)\partial f(\mathbb D) \neq \emptyset: otherwise f(D)f(\mathbb D), open, closed, nonempty, would be all of C\C, and f1 ⁣:CDf^{-1} \colon \C \to \mathbb D would be a bounded nonconstant entire function, against Corollary 16.12. Left inequality: the Koebe transform FF of question 10 lies in S\mathcal S and F(D)=f(D)f(z0)f(z0)(1z02)F(\mathbb D) = \frac{f(\mathbb D) - f(z_0)}{f'(z_0)(1-\abs{z_0}^2)}; by question 7 it contains D(0,14)D(0, \frac14), so

f(D)f(z0)+D(0, 14f(z0)(1z02)),f(\mathbb D) \supseteq f(z_0) + D\Bigl(0,\ \tfrac14\abs{f'(z_0)}\bigl(1 - \abs{z_0}^2\bigr)\Bigr) ,

and every point of f(D)\partial f(\mathbb D) — disjoint from the open set f(D)f(\mathbb D) — is at distance 14(1z02)f(z0)\geq \frac14(1-\abs{z_0}^2)\abs{f'(z_0)} from f(z0)f(z_0). Right inequality: let d=d(f(z0),f(D))<d = d(f(z_0), \partial f(\mathbb D)) < \infty. The disc D(f(z0),d)D(f(z_0), d) lies in f(D)f(\mathbb D): a segment from f(z0)f(z_0) to any of its points stays at distance <d< d from f(z0)f(z_0), so never meets f(D)\partial f(\mathbb D), and the connectedness argument of question 19 keeps it in f(D)f(\mathbb D). Then g^(w)=f1(f(z0)+dw)\hat g(w) = f^{-1}(f(z_0) + dw) maps D\mathbb D into D\mathbb D with g^(0)=z0\hat g(0) = z_0, and χ=ψ1g^\chi = \psi^{-1}\circ\hat g, with ψ(z)=z+z01+zˉ0z\psi(z) = \frac{z + z_0}{1 + \bar z_0z}, fixes 00: Schwarz (Theorem 18.4) gives χ(0)1\abs{\chi'(0)} \leq 1. Since χ(0)=g^(0)ψ(0)=df(z0)(1z02)\chi'(0) = \frac{\hat g'(0)}{\psi'(0)} = \frac{d}{f'(z_0)\,(1 - \abs{z_0}^2)}, this is d(1z02)f(z0)d \leq (1 - \abs{z_0}^2)\abs{f'(z_0)}.

21. k(D)=C(,14]k(\mathbb D) = \C\setminus \intoc{-\infty}{-\frac14}, so k(D)=(,14]\partial k(\mathbb D) = \intoc{-\infty}{-\frac14}, and for the positive real point k(r)=r(1r)2k(r) = \frac r{(1-r)^2} the nearest boundary point is 14-\frac14:

d=k(r)+14=4r+(1r)24(1r)2=(1+r)24(1r)2.d = k(r) + \frac14 = \frac{4r + (1-r)^2}{4(1-r)^2} = \frac{(1+r)^2}{4(1-r)^2} .

Left member of question 20: 14(1r2)k(r)=14(1r2)(1+r)(1r)3=(1+r)24(1r)2=d\frac14(1 - r^2)k'(r) = \frac14\,\frac{(1-r^2)(1+r)}{(1-r)^3} = \frac{(1+r)^2}{4(1-r)^2} = d: equality. (The right member equals 4d4d: the full factor 44 separates the two sides, and the Koebe function sits exactly at the bottom.)

22. The single principle is Parseval: univalence forbids overlap, so the area of the complement of the image of {w>ρ}\{\abs w > \rho\}, expanded in Fourier modes on circles, is nonnegative — the area theorem is an L2L^2 identity with a sign. Everything else is that inequality transported: a square root (question 3) turns it into a22\abs{a_2} \leq 2; a Möbius reflection off an omitted value (question 6) turns a22\abs{a_2} \leq 2 into the quarter theorem; disc automorphisms (question 10) spread a22\abs{a_2} \leq 2 over the whole disc as the distortion theorem; radial integration converts distortion into growth, and growth into covering. At every stage the case of equality propagates too, always landing on the rotated Koebe functions — one extremal family for the whole theory, just as the rotations are the unique extremals of Schwarz’s lemma. One interior inequality, plus rigidity of its equality case, governs the entire geometry: conformal mapping is the art of exploiting such inequalities.

23. The growth theorem bounds fr(1r)2\abs f \leq \frac{r}{(1-r)^2} uniformly on each Dˉ(0,r)\bar D(0, r), r<1r < 1, for all fSf \in \mathcal S at once: locally bounded, so S\mathcal S is a normal family (Theorem 18.7). If fnSff_n \in \mathcal S \to f locally uniformly: ff is holomorphic with fnff_n' \to f' locally uniformly (Weierstrass), so f(0)=0f(0) = 0, f(0)=1f'(0) = 1 — in particular ff is nonconstant — and Hurwitz (Exercise 17.8) makes the limit of injective maps injective: fSf \in \mathcal S. A continuous functional (such as fa2=f(0)2f \mapsto \abs{a_2} = \frac{\abs{f''(0)}}2) on a compact class attains its supremum: extremal functions exist before one knows what they are — the starting point of every variational attack on coefficient problems, Bieberbach’s included.

24. Write g(w)=w+A2w2+O(w3)g(w) = w + A_2w^2 + O(w^3) and compose:

z=g(f(z))=f(z)+A2f(z)2+O(z3)=z+(a2+A2)z2+O(z3),z = g(f(z)) = f(z) + A_2f(z)^2 + O(z^3) = z + (a_2 + A_2)z^2 + O(z^3),

so A2=a2A_2 = -a_2 and A22\abs{A_2} \leq 2 (Bieberbach), with equality iff a2=2\abs{a_2} = 2, i.e. iff ff is a rotated Koebe function (question 16) — and then gg is the corresponding inverse, defined on the slit plane.

25. f(z1)f(z2)=(z1z2)(1+a(z1+z2))f(z_1) - f(z_2) = (z_1 - z_2)\bigl(1 + a(z_1 + z_2)\bigr). If a12\abs a \leq \frac12: for z1z2z_1 \neq z_2 in D\mathbb D, a(z1+z2)<2a1\abs{a(z_1 + z_2)} < 2\abs a \leq 1 (strict: z1+z2<2\abs{z_1 + z_2} < 2), so the second factor cannot vanish: injective, and fSf \in \mathcal S (normalizations are built in). If a>12\abs a > \frac12: the point s=1as = -\frac1a has s<2\abs s < 2, so z1,2=s2±εz_{1,2} = \frac s2 \pm \varepsilon lie in D\mathbb D for small ε>0\varepsilon > 0, are distinct, and z1+z2=sz_1 + z_2 = s kills the factor: f(z1)=f(z2)f(z_1) = f(z_2), not injective. So S\mathcal S contains z+az2z + az^2 exactly for a12\abs a \leq \frac12. Bieberbach’s bound a22\abs{a_2} \leq 2 is thus wildly unsaturated by polynomials of degree 22 — the Koebe function’s coefficients an=na_n = n come from an infinite series conspiring along the omitted ray, a behavior no polynomial (which belongs to S\mathcal S only with tiny coefficients) can imitate.