4.1 Conformal Maps and Möbius Transformations
Analytic maps with nonzero derivative preserve angles; Möbius transformations permute circles, are fixed by three points and preserve the cross-ratio; and together with $z^n$, $e^z$ and the Joukowski map they carry half-planes, discs, strips and sectors onto one another — exactly as the Riemann mapping theorem promises.
An analytic function is more than a formula: it is a map of one region of the plane onto another, and wherever its derivative is not zero it bends the plane without distorting small shapes. A tiny square is turned and enlarged by the factor $|f'(z_0)|$, but it stays a square; two curves that cross at $60^\circ$ have images that cross at $60^\circ$. Bernhard Riemann made this the centre of his 1851 doctoral thesis and stated the astonishing consequence that bears his name: every simply connected domain other than $\mathbb{C}$ can be mapped in this angle-preserving way onto the unit disc. Four years later August Ferdinand Möbius studied the simplest such maps, the fractions $\frac{az+b}{cz+d}$, which send circles to circles; and in 1910 Nikolai Joukowski used the map $\frac12\left(z+\frac1z\right)$ to turn the flow of air past a circular cylinder into the flow past an aircraft wing.
The practical reason for the interest is transplantation: harmonic functions stay harmonic under composition with analytic maps (Theorem 4.4), so a problem in electrostatics, heat conduction or ideal fluid flow posed on an awkward region can be carried by a conformal map to a disc or a half-plane, solved there and carried back. For CSIR NET this section is a reliable source of marks. Part B asks for the Möbius map sending three given points to three others, for the fixed points of a map, for the image of a disc or half-plane, for the map of $\mathbb{H}$ onto $\mathbb{D}$ with a given normalisation, or for the points where a map fails to be conformal. Part C asks which pairs of domains are conformally equivalent, which maps are automorphisms of $\mathbb{D}$ or $\mathbb{H}$, and which "obvious" properties of Möbius maps are false — the counterexamples are collected in the ⚠️ box.
Notation: $\mathbb{D}=\{|z|\lt1\}$, $\mathbb{H}=\{\operatorname{Im}z\gt0\}$ is the upper half-plane, $\mathbb{C}_\infty=\mathbb{C}\cup\{\infty\}$ is the extended plane of §1.1, $\varphi_a(z)=\frac{z-a}{1-\bar az}$ for $|a|\lt1$, and $\operatorname{Aut}(D)$ is the group of automorphisms of a domain $D$ (§2.3). A smooth curve through $z_0$ is a $C^1$ path $\gamma:[0,1)\to\mathbb{C}$ with $\gamma(0)=z_0$ and $\gamma'(0)\ne0$.
Conformal Maps
A continuous map $f$ defined near $z_0$ is conformal at $z_0$ if for every pair of smooth curves $\gamma_1,\gamma_2$ through $z_0$ the image curves $f\circ\gamma_1$, $f\circ\gamma_2$ have tangent directions at $f(z_0)$ and the angle from $f\circ\gamma_1$ to $f\circ\gamma_2$ equals the angle from $\gamma_1$ to $\gamma_2$. If $f$ preserves the size of every angle but reverses its sense, $f$ is anti-conformal at $z_0$.
Proof. Since $\gamma'(0)\ne0$, $\gamma(t)\ne z_0$ for small $t\gt0$, so we may write $$\frac{f(\gamma(t))-f(z_0)}{t}=\frac{f(\gamma(t))-f(z_0)}{\gamma(t)-z_0}\cdot\frac{\gamma(t)-z_0}{t}\longrightarrow f'(z_0)\,\gamma'(0)\qquad(t\to0^+).$$ The same computation at every parameter value (the chain rule) shows that $f\circ\gamma$ is $C^1$ with $(f\circ\gamma)'=f'(\gamma)\gamma'$, and $(f\circ\gamma)'(0)=f'(z_0)\gamma'(0)\ne0$. So $f\circ\gamma$ is a smooth curve and its tangent direction is that of $f'(z_0)\gamma'(0)$. For two curves, $$\arg\frac{(f\circ\gamma_2)'(0)}{(f\circ\gamma_1)'(0)}=\arg\frac{f'(z_0)\gamma_2'(0)}{f'(z_0)\gamma_1'(0)}=\arg\frac{\gamma_2'(0)}{\gamma_1'(0)}.$$ The last statement is the definition of $f'(z_0)$ rewritten: $f(z)-f(z_0)=f'(z_0)(z-z_0)+o(|z-z_0|)$. ∎
The converse holds among maps that are differentiable in the real sense. Recall from §2.1 that a real-linear map of $\mathbb{C}=\mathbb{R}^2$ can be written $L(h)=\alpha h+\beta\bar h$, and that for a real-differentiable $f$ the derivative is $L(h)=f_z(z_0)h+f_{\bar z}(z_0)\bar h$ (Wirtinger derivatives).
Proof. Write $L(h)=\alpha h+\beta\bar h$. For the line $\gamma_\theta(t)=z_0+te^{i\theta}$ we have $f(\gamma_\theta(t))-f(z_0)=tL(e^{i\theta})+o(t)$, and $L(e^{i\theta})\ne0$ because $L$ is invertible; so the image of $\gamma_\theta$ has the tangent direction of $L(e^{i\theta})$. Conformality, applied to $\gamma_0$ and $\gamma_\theta$, says $\arg\frac{L(e^{i\theta})}{L(1)}=\theta$, i.e. $$\arg\big(e^{-i\theta}L(e^{i\theta})\big)=\arg\big(\alpha+\beta e^{-2i\theta}\big)=\arg L(1)\qquad\text{for every }\theta.$$ As $\theta$ runs over $[0,\pi]$, the point $\alpha+\beta e^{-2i\theta}$ runs round the circle with centre $\alpha$ and radius $|\beta|$, and all these points lie on one ray from $0$. A ray contains no circle of positive radius, so $\beta=0$. Then $f(z_0+h)-f(z_0)=\alpha h+o(|h|)$, which says $f'(z_0)=\alpha$, and $\alpha\ne0$ since $L$ is invertible. In the anti-conformal case the same argument with $\arg\frac{L(e^{i\theta})}{L(1)}=-\theta$ gives $\alpha=0$, so $\bar f(z_0+h)-\bar f(z_0)=\bar\beta h+o(|h|)$. ∎
So, for maps with nonzero Jacobian, conformal at $z_0$ $\iff$ complex differentiable at $z_0$ with $f'(z_0)\ne0$. The map $z\mapsto\bar z$ is the model anti-conformal map: it preserves sizes of angles and reverses their orientation. At a zero of $f'$ angles are not preserved but multiplied.
Proof. By the factorisation at a zero (§2.3), $f(z)-f(z_0)=(z-z_0)^mg(z)$ with $g$ analytic and $g(z_0)=c\ne0$. For small $t\gt0$, $$\frac{f(\gamma(t))-f(z_0)}{t^m}=\left(\frac{\gamma(t)-z_0}{t}\right)^{\!m}g(\gamma(t))\longrightarrow\gamma'(0)^m\,c\ne0 .$$ Dividing by the positive number $t^m$ does not change directions, so the unit vector of $f(\gamma(t))-f(z_0)$ tends to that of $c\gamma'(0)^m$. For two curves the angle between the images is $\arg\frac{c\gamma_2'(0)^m}{c\gamma_1'(0)^m}=m\arg\frac{\gamma_2'(0)}{\gamma_1'(0)}$. Taking curves at angle $\frac\pi m$, whose images meet at angle $\pi\ne\frac\pi m$, shows that $f$ is not conformal at $z_0$. ∎
Be aware that "conformal map" is also used, especially in examination questions, for a map that is merely analytic with $f'\ne0$ — conformal at each point but possibly not injective, like $e^z$ on $\mathbb{C}$. Read each question carefully.
Proof. Harmonicity is a local property. Fix $z_0\in D$ and a disc $\Delta\subseteq D'$ about $f(z_0)$. A disc is simply connected, so $u=\operatorname{Re}F$ on $\Delta$ for an analytic $F$ (existence of harmonic conjugates, §2.1). On the open set $f^{-1}(\Delta)\ni z_0$ we have $u\circ f=\operatorname{Re}(F\circ f)$, the real part of an analytic function, hence harmonic. (The formula for $\Delta(u\circ f)$ is a direct computation with the chain rule and the Cauchy–Riemann equations; it is not needed below.) For a conformal bijection apply the first part to $f^{-1}$ as well. ∎
This is how conformal maps solve boundary value problems: to find a harmonic function on $D$ with given boundary values, map $D$ conformally onto $\mathbb{H}$ or $\mathbb{D}$, solve there, and compose. Example 2(e) and Exercise IV.9 do exactly this.
(b) Find the images of the lines $x=c$ and $y=d$ under $e^z$, and check that they meet at right angles. What is the scale factor at $z=x+iy$?
(c) Show that $z\mapsto\bar z$ preserves the size of angles but is not conformal.
(d) Under $z^3$, what is the angle between the images of two rays from $0$ that meet at angle $\frac\pi6$?
(b) If $f$ is analytic at $z_0$ and $f'(z_0)=0$, then $f$ is not conformal at $z_0$.
(c) The real-linear map $L(x+iy)=2x+iy$ maps the coordinate axes to perpendicular lines, so it is conformal at $0$.
(d) If $u$ is harmonic on $\mathbb{D}$, then $v(z)=u(z^2)$ is harmonic on $\mathbb{D}$.
(e) Find a function harmonic on $\mathbb{H}$, continuous on $\overline{\mathbb{H}}\setminus\{0\}$, equal to $1$ on $(0,\infty)$ and to $0$ on $(-\infty,0)$.
Möbius Transformations
The condition $ad-bc\ne0$ excludes constants: if $ad=bc$ and $c\ne0$, then $a(cz+d)-c(az+b)=ad-bc=0$, so $\frac{az+b}{cz+d}\equiv\frac ac$; for instance $\frac{2z+4}{z+2}=2$.
- Every Möbius transformation $T$ is a bijection of $\mathbb{C}_\infty$; its inverse $T^{-1}(w)=\dfrac{dw-b}{-cw+a}$ is again Möbius.
- Composition corresponds to matrix multiplication: $M_{S\circ T}=M_SM_T$. Two matrices define the same map if and only if they are proportional. Hence the Möbius transformations form a group, isomorphic to $\mathrm{GL}_2(\mathbb{C})/\{\lambda I\}=\mathrm{PGL}_2(\mathbb{C})\cong\mathrm{SL}_2(\mathbb{C})/\{\pm I\}$.
- $T$ is analytic on $\mathbb{C}\setminus\{-\frac dc\}$ with $$T'(z)=\frac{ad-bc}{(cz+d)^2}\ne0,$$ so $T$ is conformal at every point of $\mathbb{C}$ except its pole.
- Every Möbius transformation is a composition of translations $z\mapsto z+\beta$, rotation-dilations $z\mapsto\alpha z$ ($\alpha\ne0$) and the inversion $z\mapsto\frac1z$: if $c=0$, $T(z)=\frac adz+\frac bd$; if $c\ne0$, $$T(z)=\frac ac-\frac{ad-bc}{c^2}\cdot\frac{1}{z+\frac dc}.$$
Proof. (2) Let $S(w)=\frac{pw+q}{rw+s}$. Substituting $w=T(z)$ and multiplying numerator and denominator by $cz+d$, $$S(T(z))=\frac{p(az+b)+q(cz+d)}{r(az+b)+s(cz+d)}=\frac{(pa+qc)z+(pb+qd)}{(ra+sc)z+(rb+sd)},$$ whose coefficients are the entries of $M_SM_T$; and $\det(M_SM_T)=\det M_S\det M_T\ne0$. This computation holds for all but at most two values of $z$; at the remaining points ($z=\infty$ and the pole of $T$) one checks the conventions directly. If $\frac{az+b}{cz+d}=z$ for all $z$, then $cz^2+(d-a)z-b=0$ for all $z$, so $b=c=0$ and $a=d$: the matrix is $aI$. Applying this to $M_S^{-1}M_T$ shows that $S=T$ iff $M_T$ is a multiple of $M_S$. So $M\mapsto T_M$ is a surjective homomorphism from $\mathrm{GL}_2(\mathbb{C})$ onto the Möbius maps with kernel the scalar matrices; dividing a matrix by a square root of its determinant shows that $\mathrm{SL}_2(\mathbb{C})$ already maps onto, with kernel $\{\pm I\}$.
(1) The matrix $\begin{pmatrix}d&-b\\-c&a\end{pmatrix}$ is $(ad-bc)M_T^{-1}$, so by (2) the corresponding map is a two-sided inverse of $T$. (3) The quotient rule gives $T'=\frac{a(cz+d)-c(az+b)}{(cz+d)^2}=\frac{ad-bc}{(cz+d)^2}$. (4) For $c\ne0$: $\frac{az+b}{cz+d}-\frac ac=\frac{c(az+b)-a(cz+d)}{c(cz+d)}=\frac{bc-ad}{c^2\,(z+d/c)}$. So $T$ is: translate by $\frac dc$, invert, multiply by $-\frac{ad-bc}{c^2}$, translate by $\frac ac$. ∎
A Möbius map is also continuous as a map of $\mathbb{C}_\infty$ (near its pole it tends to $\infty$, and as $z\to\infty$ it tends to $\frac ac$), so by (1) it is a homeomorphism of $\mathbb{C}_\infty$; in particular it maps the boundary of a region onto the boundary of the image region. Möbius maps are exactly the bijective analytic maps of the Riemann sphere:
(b) Let $f:\mathbb{C}_\infty\to\mathbb{C}_\infty$ be a bijection that is meromorphic on $\mathbb{C}$ and has a removable singularity or a pole at $\infty$. Then $f$ is a Möbius transformation. Conversely, every Möbius transformation is such a map.
Proof. (a) is the theorem that injective entire functions are linear (§3.1). (b) Let $p=f(\infty)$ and let $S(w)=\frac1{w-p}$ if $p\ne\infty$, $S=\mathrm{id}$ if $p=\infty$. Then $g=S\circ f$ is again meromorphic on $\mathbb{C}$, it is a bijection of $\mathbb{C}_\infty$, and $g(\infty)=S(p)=\infty$. Since only $\infty$ is mapped to $\infty$, $g$ has no poles in $\mathbb{C}$: it is entire, and it maps $\mathbb{C}$ injectively (onto $\mathbb{C}$). By (a), $g(z)=\alpha z+\beta$, so $f=S^{-1}\circ g$ is Möbius. The converse is Theorem 4.5(1),(3). ∎
Fixed points organise the whole theory. By Theorem 4.5(2), conjugating $T$ by another Möbius map $U$ — replacing $T$ by $UTU^{-1}$ — changes coordinates without changing the dynamics, and we can always move the fixed points to $0$ and $\infty$.
- $T$ has one or two fixed points in $\mathbb{C}_\infty$: $\infty$ is fixed iff $c=0$, and the finite fixed points are the roots of $cz^2+(d-a)z-b=0$. A Möbius transformation with three fixed points is the identity.
- If $T$ has two fixed points $z_1,z_2\in\mathbb{C}$, there is a multiplier $k\in\mathbb{C}\setminus\{0,1\}$ with $$\frac{T(z)-z_1}{T(z)-z_2}=k\,\frac{z-z_1}{z-z_2}\quad\text{for all }z;$$ if the fixed points are $z_1\in\mathbb{C}$ and $\infty$, then $T(z)-z_1=k(z-z_1)$.
- If $T$ has exactly one fixed point $z_0\in\mathbb{C}$, there is $h\ne0$ with $\dfrac1{T(z)-z_0}=\dfrac1{z-z_0}+h$; if the only fixed point is $\infty$, then $T(z)=z+h$.
- Put $\tau=\dfrac{(a+d)^2}{ad-bc}$ (the square of the trace after normalising $ad-bc=1$). Then $T$ has exactly one fixed point iff $\tau=4$, and in the two-fixed-point case $\tau=k+\frac1k+2$. Accordingly $T$ is parabolic ($\tau=4$), elliptic ($|k|=1$; equivalently $0\le\tau\lt4$), hyperbolic ($k\gt0$; equivalently $\tau\gt4$) or loxodromic (all other cases; equivalently $\tau\notin[0,\infty)$).
Proof. (1) If $c\ne0$, then $T(\infty)=\frac ac\ne\infty$ and $T(-\frac dc)=\infty\ne-\frac dc$, so the fixed points are the solutions in $\mathbb{C}\setminus\{-\frac dc\}$ of $z(cz+d)=az+b$, i.e. of the genuine quadratic $cz^2+(d-a)z-b=0$; it has one or two roots, and $-\frac dc$ is not a root (substituting gives $\frac{ad-bc}{c}\ne0$). If $c=0$, $\infty$ is fixed and $T(z)=\frac adz+\frac bd$; there is one finite fixed point if $a\ne d$, and none if $a=d$ (then $b\ne0$ because $T\ne\mathrm{id}$). Three fixed points are therefore possible only when $c=0$ and $\left(\frac ad-1\right)z+\frac bd=0$ has two roots, i.e. $a=d$, $b=0$: $T=\mathrm{id}$.
(2) Let $U(z)=\frac{z-z_1}{z-z_2}$, so $U(z_1)=0$, $U(z_2)=\infty$. Then $V=UTU^{-1}$ fixes $0$ and $\infty$. Fixing $\infty$ forces $V(\zeta)=\alpha\zeta+\beta$, and fixing $0$ forces $\beta=0$: $V(\zeta)=k\zeta$ with $k\ne0$, and $k\ne1$ because $T\ne\mathrm{id}$. The identity $U(T(z))=kU(z)$ is the formula. For fixed points $z_1,\infty$ use $U(z)=z-z_1$. (3) With $U(z)=\frac1{z-z_0}$ (or $U=\mathrm{id}$ if $z_0=\infty$), $V=UTU^{-1}$ has $\infty$ as its only fixed point, so $V(\zeta)=\alpha\zeta+h$ with $\alpha=1$ (otherwise $\frac{h}{1-\alpha}$ would be fixed) and $h\ne0$.
(4) $\tau=\frac{(\operatorname{tr}M)^2}{\det M}$ does not change when $M$ is multiplied by a scalar or replaced by $NMN^{-1}$, because trace and determinant are similarity invariants. So $\tau(T)=\tau(V)$. For $V(\zeta)=k\zeta$, with matrix $\begin{pmatrix}k&0\\0&1\end{pmatrix}$, $\tau=\frac{(k+1)^2}{k}=k+\frac1k+2$; for $V(\zeta)=\zeta+h$, with matrix $\begin{pmatrix}1&h\\0&1\end{pmatrix}$, $\tau=4$. Since $k+\frac1k+2=4$ only for $k=1$, $\tau=4$ characterises the parabolic case. Write $k=re^{i\alpha}$; then $\operatorname{Im}\left(k+\frac1k\right)=\left(r-\frac1r\right)\sin\alpha$, so $\tau$ is real iff $|k|=1$ or $k$ is real. If $|k|=1$, $\tau=2+2\cos\alpha\in[0,4)$; if $k\gt0$, $k\ne1$, then $\tau\gt4$ (AM–GM); if $k\lt0$, $k\ne-1$, then $\tau\lt0$. This proves the equivalences. ∎
Geometrically, in the coordinate $\zeta=U(z)$ an elliptic map is a rotation $\zeta\mapsto e^{i\alpha}\zeta$: it moves points along the circles $|\zeta|=$ const, which in the $z$-plane are the Apollonius circles $\left|\frac{z-z_1}{z-z_2}\right|=$ const of §1.1. A hyperbolic map is a dilation $\zeta\mapsto k\zeta$: it moves points along the circular arcs from $z_1$ to $z_2$ (the rays from $0$ in the $\zeta$-plane). A loxodromic map does both, spiralling; a parabolic map is a translation in the coordinate $\frac{1}{z-z_0}$, moving points along circles tangent at $z_0$. Interactive 1 shows all four. Note that the multiplier depends on the order of the fixed points (swapping $z_1,z_2$ replaces $k$ by $\frac1k$), but $\tau$ does not.
For finite $z_j$ the map $z\mapsto(z,z_1,z_2,z_3)$ has coefficient matrix $\begin{pmatrix}z_2-z_3&-z_1(z_2-z_3)\\z_2-z_1&-z_3(z_2-z_1)\end{pmatrix}$ with determinant $(z_2-z_3)(z_2-z_1)(z_1-z_3)\ne0$, so it is indeed Möbius; uniqueness is part of the next theorem. Conventions differ between books (some put $z$ last, or send $z_1\mapsto1$); an examination question always fixes one, and all of them give the same theory.
- Given distinct $z_1,z_2,z_3\in\mathbb{C}_\infty$ and distinct $w_1,w_2,w_3\in\mathbb{C}_\infty$, there is exactly one Möbius transformation $T$ with $T(z_j)=w_j$ ($j=1,2,3$). It is obtained by solving $$(w,w_1,w_2,w_3)=(z,z_1,z_2,z_3)$$ for $w=T(z)$.
- For every Möbius transformation $T$, every $z$ and distinct $z_1,z_2,z_3$: $\ (Tz,Tz_1,Tz_2,Tz_3)=(z,z_1,z_2,z_3)$.
Proof. (1) Let $S_1(z)=(z,z_1,z_2,z_3)$ and $S_2(w)=(w,w_1,w_2,w_3)$. Then $T=S_2^{-1}\circ S_1$ sends $z_1\mapsto0\mapsto w_1$, and similarly for $z_2,z_3$; the displayed equation is $S_2(T(z))=S_1(z)$. If $T'$ is another such map, $T^{-1}T'$ fixes $z_1,z_2,z_3$ and is the identity by Theorem 4.7(1). (The same argument proves the uniqueness claim in Definition 4.4.) (2) The map $S_1\circ T^{-1}$ sends $Tz_1\mapsto0$, $Tz_2\mapsto1$, $Tz_3\mapsto\infty$; by uniqueness it equals $w\mapsto(w,Tz_1,Tz_2,Tz_3)$. Evaluating at $w=Tz$ gives the claim. ∎
So the Möbius group acts simply transitively on ordered triples of distinct points: three points can be sent anywhere, and then the map is fixed. A Möbius map has "three complex degrees of freedom" — four coefficients up to a common factor.
(b) Find the fixed points of $T(z)=\dfrac{3z-4}{z-1}$, its type and its normal form.
(c) The same for $S(z)=\dfrac{2z+1}{z+2}$.
(d) The same for $R(z)=\dfrac{z+1}{1-z}$; compute $R\circ R$ and $R^4$.
Circles, Cross-Ratios and Symmetry
Proof. By Theorem 4.5(4) it suffices to treat translations, rotation-dilations and the inversion $z\mapsto\frac1z$. The first two map circles to circles and lines to lines, and fix $\infty$. For the inversion, let $z\in\Gamma$ with $z\ne0,\infty$, and put $w=\frac1z$. Dividing $Az\bar z+\bar\beta z+\beta\bar z+C=0$ by $z\bar z$: $$A+\bar\beta\,\bar w+\beta\,w+C\,w\bar w=0,\quad\text{i.e.}\quad C\,w\bar w+\overline{\beta'}\,w+\beta'\,\bar w+A=0\ \text{ with }\beta'=\bar\beta .$$ This is an equation of the same kind with $A$ and $C$ exchanged and the same discriminant $|\beta'|^2-CA\gt0$, so it describes a generalised circle $\Gamma'$. Finally $0\in\Gamma$ iff $C=0$ iff $\Gamma'$ is a line, i.e. contains $\infty$; and $\infty\in\Gamma$ iff $A=0$ iff $0\in\Gamma'$. So $\frac1z$ maps $\Gamma$ onto $\Gamma'$. The last statement holds because $\infty\in T(\Gamma)$ iff $T^{-1}(\infty)\in\Gamma$. ∎
Proof. $S(\zeta)=(\zeta,z_1,z_2,z_3)$ maps the generalised circle $\Gamma$ through $z_1,z_2,z_3$ onto the generalised circle through $0,1,\infty$, which is $\mathbb{R}\cup\{\infty\}$ (Theorem 4.9). Since $S$ is a bijection of $\mathbb{C}_\infty$, $z\in\Gamma$ iff $S(z)\in\mathbb{R}\cup\{\infty\}$, and $S(z)\ne\infty$ because $z\ne z_3$. ∎
For example, $(0,2,1+i,1-i)=\frac{(0-2)(2i)}{(-1+i)(-1+i)}=\frac{-4i}{-2i}=2$ is real, and indeed the four points lie on $|z-1|=1$. Circles alone do not carry enough information to locate images of regions: we also need the Möbius-invariant version of "centre".
- Let $z_1,z_2,z_3$ be distinct points of a generalised circle $\Gamma$. Then $z$ and $z^*$ are symmetric with respect to $\Gamma$ if and only if $$(z^*,z_1,z_2,z_3)=\overline{(z,z_1,z_2,z_3)}.$$
- If $T$ is a Möbius transformation and $z,z^*$ are symmetric with respect to $\Gamma$, then $T(z),T(z^*)$ are symmetric with respect to the generalised circle $T(\Gamma)$.
Proof. (1) Directly from the formula, $\overline{(z,z_1,z_2,z_3)}=(\bar z,\bar z_1,\bar z_2,\bar z_3)$. Translations and rotations are Möbius maps (so they do not change cross-ratios, Theorem 4.8(2)) and they carry symmetric pairs to symmetric pairs, so we may assume that $\Gamma$ is $|z|=R$ or the real line.
If $\Gamma$ is $|z|=R$, then $\bar z_j=\frac{R^2}{z_j}$ for the points $z_j\in\Gamma$. Applying the Möbius map $u\mapsto\frac{R^2}{u}$, which preserves cross-ratios, $$\overline{(z,z_1,z_2,z_3)}=\left(\bar z,\frac{R^2}{z_1},\frac{R^2}{z_2},\frac{R^2}{z_3}\right)=\left(\frac{R^2}{\bar z},z_1,z_2,z_3\right)=(z^*,z_1,z_2,z_3),$$ since $z^*=\frac{R^2}{\bar z}$ (also for $z=0$, $z^*=\infty$, with the conventions for $\infty$). If $\Gamma=\mathbb{R}\cup\{\infty\}$, then $\bar z_j=z_j$ and $\overline{(z,z_1,z_2,z_3)}=(\bar z,z_1,z_2,z_3)$, with $\bar z=z^*$. So in both cases the symmetric point satisfies the equation, and it is the only solution because $\zeta\mapsto(\zeta,z_1,z_2,z_3)$ is injective.
(2) The points $Tz_1,Tz_2,Tz_3$ lie on $T(\Gamma)$, and by Theorem 4.8(2) and part (1), $$(Tz^*,Tz_1,Tz_2,Tz_3)=(z^*,z_1,z_2,z_3)=\overline{(z,z_1,z_2,z_3)}=\overline{(Tz,Tz_1,Tz_2,Tz_3)}.$$ By part (1) for $T(\Gamma)$, $Tz$ and $Tz^*$ are symmetric with respect to $T(\Gamma)$. ∎
The centre of a circle is not Möbius-invariant, but the pair (centre, $\infty$) is a symmetric pair. Hence: if $T$ maps $\Gamma$ onto a circle $\Gamma'$, then the centre of $\Gamma'$ is $T(z^*)$, where $z^*$ is the point symmetric to the pole $T^{-1}(\infty)$ with respect to $\Gamma$. This often gives the image circle at once (Exercise II.7).
- Boundary first. Each boundary circle or line goes to a generalised circle (Theorem 4.9). It is a line exactly when it passes through the pole $-\frac dc$. Find it from the images of three of its points (use convenient ones: real and imaginary intercepts, $\infty$).
- Centres via symmetry. If the image is a circle, its centre is the image of the point symmetric to the pole (Theorem 4.11).
- Which side? A Möbius map is a homeomorphism of $\mathbb{C}_\infty$, so a region bounded by $\Gamma$ goes to one of the two regions bounded by $T(\Gamma)$. Decide which by mapping one interior point. (If the pole is inside, the image is the exterior.)
- Several boundary arcs. Track the corners: angles there are preserved (Theorem 4.5(3)), unless the corner is the pole, which goes to $\infty$.
- Constructing a map. Send three boundary points to three boundary points (Theorem 4.8), or send an interior point and its symmetric point to $0$ and $\infty$ (Theorem 4.12 below).
(b) There is a Möbius transformation with $T(0)=1$, $T(1)=2$, $T(2)=3$ and $T(3)=5$.
(c) If a Möbius transformation $T$ maps a circle $\Gamma$ with centre $a$ onto a circle $\Gamma'$, then $T(a)$ is the centre of $\Gamma'$.
(d) Every Möbius transformation with exactly one fixed point is conjugate to $z\mapsto z+1$.
The Disc and the Upper Half-Plane
The two most important simply connected domains are conformally equivalent by a Möbius map. The key is the symmetry principle: the real axis is the "circle" bounding $\mathbb{H}$, and $\alpha,\bar\alpha$ are symmetric with respect to it.
Proof. For $\alpha\in\mathbb{H}$, $$|z-\bar\alpha|^2-|z-\alpha|^2=4\operatorname{Im}z\operatorname{Im}\alpha$$ (expand both sides, or note that $\mathbb{R}$ is the perpendicular bisector of $\alpha$ and $\bar\alpha$). So $g(z)=\frac{z-\alpha}{z-\bar\alpha}$ satisfies $|g|\lt1$ on $\mathbb{H}$, $|g|=1$ on $\mathbb{R}$ (and $g(\infty)=1$), and $|g|\gt1$ on the lower half-plane. These three sets partition $\mathbb{C}_\infty$ (with $g(\bar\alpha)=\infty$), and $g$ is a bijection of $\mathbb{C}_\infty$; so every $w\in\mathbb{D}$ has its preimage in $\mathbb{H}$, and $g(\mathbb{H})=\mathbb{D}$. The same holds for $e^{i\theta}g$. With $\alpha=i$ this is the Cayley transform; solving $w=\frac{z-i}{z+i}$ gives $z=\frac{i(1+w)}{1-w}$.
Conversely, let $f$ be a conformal map of $\mathbb{H}$ onto $\mathbb{D}$ and $\alpha=f^{-1}(0)\in\mathbb{H}$. Then $f\circ C^{-1}\in\operatorname{Aut}(\mathbb{D})$ is a Möbius map (automorphisms of the disc, §2.3), hence so is $f$. As a homeomorphism of $\mathbb{C}_\infty$, $f$ maps the boundary $\mathbb{R}\cup\{\infty\}$ of $\mathbb{H}$ onto the unit circle. The points $\alpha,\bar\alpha$ are symmetric with respect to $\mathbb{R}$, so by Theorem 4.11 $f(\alpha)=0$ and $f(\bar\alpha)$ are symmetric with respect to the unit circle: $f(\bar\alpha)=\infty$. A Möbius map with zero $\alpha$ and pole $\bar\alpha$ is $f(z)=\lambda\frac{z-\alpha}{z-\bar\alpha}$, and $|f(0)|=|\lambda|\frac{|\alpha|}{|\bar\alpha|}=|\lambda|$ must be $1$ because $0\in\mathbb{R}$. ∎
The same symmetry argument re-derives the Möbius automorphisms of $\mathbb{D}$: if $T$ maps $\mathbb{D}$ onto $\mathbb{D}$ and $T(a)=0$, then $T(1/\bar a)=\infty$, so $T(z)=\lambda\frac{z-a}{1-\bar az}$ with $|\lambda|=1$. The theorem of §2.3 says more — every automorphism of $\mathbb{D}$, Möbius or not a priori, is of this form (the Schwarz lemma did that work). In matrix language, writing $e^{i\theta}=u^2$ and dividing by $\sqrt{1-|a|^2}$, $$\operatorname{Aut}(\mathbb{D})=\left\{z\mapsto\frac{\alpha z+\beta}{\bar\beta z+\bar\alpha}:\ |\alpha|^2-|\beta|^2=1\right\}\cong\mathrm{SU}(1,1)/\{\pm I\}.$$
Proof. For real $a,b,c,d$: $T(z)=\frac{(az+b)(c\bar z+d)}{|cz+d|^2}$, and the imaginary part of the numerator $ac|z|^2+adz+bc\bar z+bd$ is $(ad-bc)\operatorname{Im}z$. So if $ad-bc\gt0$, $T$ maps $\mathbb{H}$ into $\mathbb{H}$; its pole $-\frac dc$ is real (or $\infty$), so $T$ is analytic on $\mathbb{H}$; and $T^{-1}$, with the real matrix $\begin{pmatrix}d&-b\\-c&a\end{pmatrix}$ of the same determinant, also maps $\mathbb{H}$ into $\mathbb{H}$. Hence $T\in\operatorname{Aut}(\mathbb{H})$.
Conversely, let $f\in\operatorname{Aut}(\mathbb{H})$. Then $C\circ f\circ C^{-1}\in\operatorname{Aut}(\mathbb{D})$ is Möbius, hence $f$ is Möbius. As a homeomorphism of $\mathbb{C}_\infty$ mapping $\mathbb{H}$ onto $\mathbb{H}$, $f$ maps the boundary $\mathbb{R}\cup\{\infty\}$ onto itself. So $w_1=f(0)$, $w_2=f(1)$, $w_3=f(\infty)$ are distinct points of $\mathbb{R}\cup\{\infty\}$, and $f$ is the solution of $(w,w_1,w_2,w_3)=(z,0,1,\infty)=z$ (Theorem 4.8). The map $w\mapsto(w,w_1,w_2,w_3)$ has a real coefficient matrix (Definition 4.4, also when some $w_j=\infty$), so its inverse $f$ does too. Finally $\operatorname{Im}f(i)\gt0$ forces $ad-bc\gt0$ by the formula. Normalising $ad-bc=1$ leaves exactly the sign ambiguity $\pm(a,b,c,d)$, which gives $\mathrm{PSL}_2(\mathbb{R})$.
If $T(i)=i$ with $ad-bc=1$, then $ai+b=i(ci+d)=-c+di$, so $a=d$, $b=-c$ and $a^2+b^2=1$: $(a,b)=(\cos\theta,\sin\theta)$. Conversely $\rho_\theta(i)=\frac{\sin\theta+i\cos\theta}{\cos\theta-i\sin\theta}=i$. The map $y_0z+x_0$ has determinant $y_0\gt0$. ∎
Via the Cayley transform the stabiliser $\{\rho_\theta\}$ of $i$ corresponds to the rotations of $\mathbb{D}$ about $0$; $\rho_\theta$ corresponds to the rotation by $2\theta$, which is why $\rho_\pi=\mathrm{id}$. Note the role of the sign: $z\mapsto-\frac1z$ (matrix $\begin{pmatrix}0&-1\\1&0\end{pmatrix}$, determinant $1$) is an automorphism of $\mathbb{H}$, while $z\mapsto\frac1z$ (determinant $-1$) maps $\mathbb{H}$ onto the lower half-plane.
(b) Find all $f\in\operatorname{Aut}(\mathbb{H})$ with $f(i)=1+2i$, and the one with $f'(i)\gt0$.
(c) Find the conformal maps of the right half-plane $\{\operatorname{Re}z\gt0\}$ onto $\mathbb{D}$ with $1\mapsto0$.
(d) Let $f:\mathbb{H}\to\mathbb{D}$ be analytic with $f(i)=0$. How large can $|f(2i)|$ and $|f'(i)|$ be?
(e) Prove that an automorphism of $\mathbb{H}$ with two distinct fixed points in $\mathbb{H}$ is the identity.
Standard Maps and the Riemann Mapping Theorem
Powers. In polar form $z^n=r^ne^{int}$: rays $\arg z=t$ go to rays $\arg w=nt$ and arcs $|z|=r$ to arcs $|w|=r^n$. So $z^n$ maps the sector $\{0\lt\arg z\lt\frac\pi n\}$ conformally onto $\mathbb{H}$, and $\{0\lt\arg z\lt\frac{2\pi}n\}$ onto $\mathbb{C}\setminus[0,\infty)$; it is injective on an open sector at $0$ exactly when the opening is at most $\frac{2\pi}n$, and at $0$ it multiplies angles by $n$ (Theorem 4.3). More generally, for $0\lt\theta_0\le\pi$ and $\alpha\gt0$ with $\alpha\theta_0\le2\pi$, the principal power $z^\alpha=e^{\alpha\operatorname{Log}z}$ maps $\{0\lt\arg z\lt\theta_0\}$ conformally onto $\{0\lt\arg w\lt\alpha\theta_0\}$. In particular $z^2$ is injective on $\mathbb{H}$ and maps it onto $\mathbb{C}\setminus[0,\infty)$, and the principal square root maps $\mathbb{C}\setminus(-\infty,0]$ onto the right half-plane.
The exponential. $e^{x+iy}=e^xe^{iy}$ maps vertical lines to circles and horizontal lines to rays (Example 1(b)). For $0\lt h\le2\pi$ it maps the strip $\{a\lt\operatorname{Im}z\lt a+h\}$ conformally onto the sector $\{a\lt\arg w\lt a+h\}$ — in particular $\{0\lt\operatorname{Im}z\lt\pi\}$ onto $\mathbb{H}$ and $\{0\lt\operatorname{Im}z\lt2\pi\}$ onto $\mathbb{C}\setminus[0,\infty)$ — and the half-strip $\{\operatorname{Re}z\lt0,\ 0\lt\operatorname{Im}z\lt\pi\}$ onto the upper half-disc. The inverse maps are branches of $\log$ (§1.2).
- $J(z)=J(w)$ iff $w=z$ or $w=\frac1z$. $J'(z)=\frac12\left(1-\frac1{z^2}\right)$ vanishes only at $z=\pm1$, where $J(\pm1)=\pm1$ and angles are doubled.
- $J$ maps $\{|z|\gt1\}$ conformally onto $\mathbb{C}\setminus[-1,1]$, and also $\mathbb{D}\setminus\{0\}$ conformally onto $\mathbb{C}\setminus[-1,1]$.
- For $r\gt0$, $r\ne1$, $J$ maps the circle $|z|=r$ onto the ellipse $\dfrac{x^2}{A^2}+\dfrac{y^2}{B^2}=1$ with $A=\frac12\left(r+\frac1r\right)$, $B=\frac12\left|r-\frac1r\right|$ and foci $\pm1$; it maps the unit circle onto the segment $[-1,1]$, covered twice. For $t\notin\frac\pi2\mathbb{Z}$ it maps the ray $\{\arg z=t\}$ onto one branch of the hyperbola $\dfrac{x^2}{\cos^2t}-\dfrac{y^2}{\sin^2t}=1$, again with foci $\pm1$.
Proof. (1) $J(z)-J(w)=\frac12(z-w)\left(1-\frac{1}{zw}\right)$. The derivative is immediate, and $J(z)-J(1)=\frac{(z-1)^2}{2z}$ shows a zero of order $2$ (similarly at $-1$), so Theorem 4.3 applies.
(2) If $|z|,|w|\gt1$ then $|zw|\gt1$, so $zw\ne1$ and $J$ is injective on $\{|z|\gt1\}$; similarly on $\mathbb{D}\setminus\{0\}$. Given $w\in\mathbb{C}$, the solutions of $J(z)=w$ are the roots of $z^2-2wz+1=0$, two numbers $z,\frac1z$ with product $1$ (or a double root $\pm1$ when $w=\pm1$). If some root has modulus $1$, $z=e^{it}$, then $w=J(e^{it})=\cos t\in[-1,1]$. So for $w\notin[-1,1]$ exactly one root has modulus $\gt1$ and the other modulus $\lt1$, while for $w\in[-1,1]$ both roots lie on the unit circle. Hence $J$ maps each of the two regions bijectively onto $\mathbb{C}\setminus[-1,1]$; it is conformal there because $J'\ne0$ off $\pm1$.
(3) With $z=re^{it}$, $$J(re^{it})=\tfrac12\left(r+\tfrac1r\right)\cos t+i\,\tfrac12\left(r-\tfrac1r\right)\sin t .$$ For fixed $r\ne1$ this is the ellipse with semi-axes $A$, $B$, and $A^2-B^2=1$ puts the foci at $\pm1$; for $r=1$ it is $\cos t\in[-1,1]$, each value (except $\pm1$) taken twice. For fixed $t$, $\frac{x}{\cos t}=\frac12\left(r+\frac1r\right)$ and $\frac{y}{\sin t}=\frac12\left(r-\frac1r\right)$, whose squares differ by $1$; as $r$ runs over $(0,\infty)$, $x$ keeps the sign of $\cos t$ and $y$ runs over $\mathbb{R}$, which is one branch of the hyperbola, whose foci are at $\pm\sqrt{\cos^2t+\sin^2t}=\pm1$. ∎
The ellipses and hyperbolas are confocal and meet at right angles — the images of the orthogonal polar grid. Joukowski's idea was to take a circle passing through the critical point $1$ and enclosing $-1$, with centre slightly to the left of $0$ and slightly above the real axis. Its image is a closed curve, smooth except at $J(1)=1$, where the angle $\pi$ of the circle is doubled to $2\pi$: a cusp. This is the Joukowski aerofoil, with a rounded leading edge, a sharp trailing edge, and camber from the vertical offset of the centre. Since the flow of an ideal fluid past a circle is known explicitly, Theorem 4.4 (applied to the stream function) gives the flow past the aerofoil, and with it the lift. Interactive 2 draws it. Compositions give more: since $J(e^{iz})=\cos z$, the cosine is the Joukowski map after an exponential, and $\sin z=J(-ie^{iz})$.
| region | map | image |
|---|---|---|
| $\mathbb{H}$ | $\frac{z-i}{z+i}$ | $\mathbb{D}$ |
| $\mathbb{D}$ | $\frac{1+z}{1-z}$ | $\{\operatorname{Re}w\gt0\}$ |
| $\{0\lt\arg z\lt\frac\pi n\}$ | $z^n$ | $\mathbb{H}$ |
| $\{0\lt\operatorname{Im}z\lt\pi\}$ | $e^z$ | $\mathbb{H}$ |
| $\{\operatorname{Re}z\lt0,\ 0\lt\operatorname{Im}z\lt\pi\}$ | $e^z$ | upper half-disc |
| $\mathbb{C}\setminus(-\infty,0]$ | $\sqrt z$ (principal) | $\{\operatorname{Re}w\gt0\}$ |
| $\{|z|\gt1\}$ or $\mathbb{D}\setminus\{0\}$ | $\frac12\left(z+\frac1z\right)$ | $\mathbb{C}\setminus[-1,1]$ |
| $\{-\frac\pi2\lt\operatorname{Re}z\lt\frac\pi2,\ \operatorname{Im}z\gt0\}$ | $\sin z$ | $\mathbb{H}$ |
- Corners and tangencies to $0$ and $\infty$. A region bounded by two circular arcs meeting at $p$ and $q$ (a lune) becomes a sector under $\frac{z-p}{z-q}$; two circles tangent at $p$ become two parallel lines (a strip) under $\frac1{z-p}$.
- Open or close angles with powers $z^\alpha$: a sector of opening $\theta_0$ becomes $\mathbb{H}$ under $z^{\pi/\theta_0}$.
- Strips and sectors are exchanged by $e^z$ and $\log z$: a strip of width $\pi$ becomes $\mathbb{H}$; a vertical strip $\{0\lt\operatorname{Re}z\lt1\}$ becomes $\mathbb{H}$ under $e^{i\pi z}$.
- Slits and half-strips: use $\sqrt z$, the Joukowski map, $\sin z$ or $\cos z$.
- Finish with $e^{i\theta}\frac{z-\alpha}{z-\bar\alpha}$ from $\mathbb{H}$ to $\mathbb{D}$, choosing $\alpha$ and $\theta$ to meet a normalisation $f(z_0)=0$, $f'(z_0)\gt0$.
- Check one interior point and one boundary arc; check that every step is injective on the current region (opening $\le\frac{2\pi}n$ for $z^n$, width $\le2\pi$ for $e^z$).
Every example so far mapped a simply connected domain onto the disc by explicit formulas. Riemann's theorem says this is always possible, with one exception that Liouville's theorem forces.
Proof of uniqueness. If $f,g$ are two such maps, $h=f\circ g^{-1}\in\operatorname{Aut}(\mathbb{D})$ fixes $0$, so $h(w)=e^{i\theta}w$ (automorphisms of the disc, §2.3). Then $e^{i\theta}=h'(0)=\frac{f'(z_0)}{g'(z_0)}\gt0$, so $e^{i\theta}=1$ and $f=g$. ∎
Proof sketch of existence (the extremal argument of L. Fejér and F. Riesz, with P. Koebe's square-root trick).
- The family is not empty. Pick $a\notin D$. Then $z-a$ has no zero on the simply connected domain $D$, so it has an analytic square root $s$ on $D$ (§2.2). $s$ is injective, and $s(z_1)=-s(z_2)$ is impossible (squaring gives $z_1=z_2$, then $s(z_1)=0$). The open set $s(D)$ contains a disc $D(b,\varepsilon)$, so $s(D)$ misses $D(-b,\varepsilon)$, and $\frac{\varepsilon}{s(z)+b}$ maps $D$ injectively into $\mathbb{D}$. Composing with a disc automorphism, the family $\mathcal F$ of injective analytic $g:D\to\mathbb{D}$ with $g(z_0)=0$ is not empty.
- An extremal map exists. By Cauchy's estimates $|g'(z_0)|$ is bounded on $\mathcal F$. A maximising sequence has a locally uniformly convergent subsequence (Montel's theorem: uniformly bounded families of analytic functions are normal); its limit $f$ maps into $\overline{\mathbb{D}}$, hence into $\mathbb{D}$ by the maximum principle, and is injective by Hurwitz's theorem (§3.2) since it is not constant. So some $f\in\mathcal F$ maximises $|f'(z_0)|$.
- The extremal map is onto. If $f$ missed some $c\in\mathbb{D}$, then $\varphi_c\circ f$ would have no zeros, hence an analytic square root $\sqrt{\varphi_c\circ f}$, and $g=\varphi_{\beta}\circ\sqrt{\varphi_c\circ f}$ (with $\beta$ chosen so that $g(z_0)=0$) would lie in $\mathcal F$. Then $f=\Phi\circ g$, where $\Phi(w)=\varphi_{-c}\big((\varphi_{-\beta}(w))^2\big)$ maps $\mathbb{D}$ into $\mathbb{D}$, fixes $0$, and is not injective, so not a rotation. By the Schwarz lemma $|\Phi'(0)|\lt1$, and $|f'(z_0)|=|\Phi'(0)|\,|g'(z_0)|\lt|g'(z_0)|$ — contradicting maximality. Finally multiply $f$ by a unimodular constant to make $f'(z_0)\gt0$. ∎
Remarks. (i) The hypothesis $D\ne\mathbb{C}$ is necessary (Example 7(f)); $\mathbb{C}$ and $\mathbb{D}$ are homeomorphic but not conformally equivalent. (ii) Simple connectivity is necessary too: a domain conformally equivalent to $\mathbb{D}$ is homeomorphic to it, hence simply connected; for annuli there is a sharper statement — $\{r_1\lt|z|\lt R_1\}$ and $\{r_2\lt|z|\lt R_2\}$ are conformally equivalent iff $\frac{R_1}{r_1}=\frac{R_2}{r_2}$ (stated without proof). (iii) When $D$ is bounded by a Jordan curve, Carathéodory's theorem (1913) extends $f$ to a homeomorphism of $\overline D$ onto $\overline{\mathbb{D}}$. (iv) The theorem gives no formula; for polygons the Schwarz–Christoffel formula supplies one, and numerically conformal maps are computed by solving integral equations. Riemann's own 1851 argument relied on the Dirichlet principle, whose gap Weierstrass pointed out; complete proofs came only around 1900–1912 (Osgood, Koebe, Carathéodory).
- Conformal at every point does not mean injective. $e^z$ on $\mathbb{C}$, $z^2$ on an annulus around $0$, $z^3$ on a sector of opening larger than $\frac{2\pi}3$ all have $f'\ne0$ and are not injective.
- Conformality fails exactly at critical points, where angles are multiplied by the order $m$: $z^2+2z$ at $-1$, $z^3-3z$ at $\pm1$, Joukowski at $\pm1$. In real analysis $x\mapsto x^3$ is injective with derivative $0$ at $0$; in $\mathbb{C}$ an injective analytic map never has a critical point (§2.3), so a conformal bijection is conformal at every point.
- Preserving sizes of angles is not enough: $\bar z$ reverses orientation (anti-conformal, nowhere analytic). Preserving one right angle is not enough either: $2x+iy$ keeps the axes perpendicular but distorts the diagonals (Example 2(c)).
- The coefficient matrix is defined only up to a factor, and $ad-bc\ne0$ is essential: $\frac{2z+4}{z+2}=2$ is constant. Normalise $ad-bc=1$ before reading off a trace; the invariant is $\tau=\frac{(a+d)^2}{ad-bc}$, not $a+d$.
- Circles do not always go to circles — they go to generalised circles: a circle through the pole becomes a line ($\frac{z+1}{z-1}$ maps $|z|=1$ to the imaginary axis).
- Interior does not always go to interior. $\frac1z$ maps $\mathbb{D}$ onto $\{|w|\gt1\}\cup\{\infty\}$; if the pole is inside a disc, its image is the exterior of the image circle. Always test a point.
- Centres are not preserved (Example 5(c)); symmetric pairs are. The centre of the image circle is the image of the point symmetric to the pole.
- Cross-ratio conventions differ. Permuting the four points changes $\lambda$ into one of $\lambda,\ 1-\lambda,\ \frac1\lambda,\ \frac1{1-\lambda},\ \frac{\lambda}{\lambda-1},\ \frac{\lambda-1}{\lambda}$ (Exercise II.8). Use the convention the question states.
- Fixed points: a Möbius map $\ne\mathrm{id}$ has one or two, never zero and never three. A map with two fixed points is not automatically elliptic: $\frac{2z+1}{z+2}$ is hyperbolic, $\frac{z+1}{1-z}$ elliptic, $2iz+1$ loxodromic.
- Real coefficients are not enough for $\operatorname{Aut}(\mathbb{H})$: $\frac1z$ has real coefficients but determinant $-1$ and maps $\mathbb{H}$ onto the lower half-plane; $-\frac1z$ is an automorphism.
- $z^2$ is injective on $\mathbb{H}$ but is not an automorphism of $\mathbb{H}$: $z^2(\mathbb{H})=\mathbb{C}\setminus[0,\infty)$. Likewise $e^z$ maps the strip of width $2\pi$ onto a slit plane, not onto a half-plane.
- Joukowski is two-to-one on $\mathbb{C}\setminus\{0\}$ ($J(z)=J(1/z)$); the unit circle goes onto the segment $[-1,1]$, traversed twice, and $|z|=r$ goes to an ellipse, not a circle.
- The Riemann mapping theorem needs both hypotheses. $\mathbb{C}$ is simply connected but not equivalent to $\mathbb{D}$; $\mathbb{C}\setminus\{0\}$ and $\mathbb{D}\setminus\{0\}$ are not equivalent although both are punctured planes topologically; the annuli $1\lt|z|\lt2$ and $1\lt|z|\lt3$ are homeomorphic but not conformally equivalent.
- Uniqueness needs the full normalisation: $f(z_0)=0$ alone leaves a rotation free; $f(z_0)=0$, $f'(z_0)\gt0$ fixes $f$.
Exercises
- 1 Find the points where $f(z)=z^3-3z$ is not conformal, and the factor by which angles are multiplied there.
- 2 Find the rotation angle and the scale factor of $e^z$ at $z=1+\frac{i\pi}4$.
- 3 Prove that the curves $x^2-y^2=1$ and $xy=\frac12$ meet at right angles at each of their intersection points.
- 4 Write $T(z)=\dfrac{2z+3}{z+1}$ as a composition of translations, rotation-dilations and the inversion.
- 5 Let $S(z)=\dfrac{z+2}{z-1}$ and $T(z)=\dfrac{2z}{z+1}$. Compute $S\circ T$ directly and by multiplying matrices.
- 6 Find the inverse of $T(z)=\dfrac{3z-i}{iz+2}$.
- 7 Show that $\frac{az+b}{cz+d}$ is constant (where defined) if $ad-bc=0$, and that the Möbius group is not abelian.
- 8 Prove that a Möbius transformation $T\ne\mathrm{id}$ satisfies $T\circ T=\mathrm{id}$ if and only if $a+d=0$. Give three examples.
- 1 Find the fixed points and the type of (a) $\dfrac{z-1}{z+1}$, (b) $\dfrac{5z-3}{z+1}$; for (b) give the multiplier.
- 2 Find the fixed points and the type of (c) $\dfrac{z}{z+1}$, (d) $2iz+1$, with the normal form in each case.
- 3 Find the Möbius map with $T(0)=1$, $T(1)=\infty$, $T(\infty)=0$. Show that $T\circ T\circ T=\mathrm{id}$, find its fixed points and its type.
- 4 Find the Möbius map with $T(-1)=0$, $T(0)=i$, $T(1)=\infty$, and the image of $\mathbb{D}$.
- 5 Use cross-ratios to decide whether (a) $1+i,1,i,-1$ and (b) $0,2,1+i,1-i$ lie on a circle or a line.
- 6 Find the point symmetric to $2+i$ with respect to the circle $|z-1|=2$.
- 7 Find the image of the circle $|z|=2$ and of the disc $|z|\lt2$ under $T(z)=\dfrac{z+2}{z-1}$.
- 8 Let $\lambda=(z,z_1,z_2,z_3)$. Show that $(z,z_3,z_2,z_1)=\frac1\lambda$ and $(z,z_2,z_1,z_3)=1-\lambda$, and deduce that permuting the four points produces only the six values $\lambda,1-\lambda,\frac1\lambda,\frac1{1-\lambda},\frac{\lambda}{\lambda-1},\frac{\lambda-1}{\lambda}$.
- 9 Prove that a Möbius map with $T(\mathbb{R}\cup\{\infty\})=\mathbb{R}\cup\{\infty\}$ has real coefficients (after multiplying them by a common factor), and that it maps $\mathbb{H}$ onto $\mathbb{H}$ or onto the lower half-plane according as $ad-bc\gt0$ or $\lt0$.
- 1 Find the conformal map $f$ of $\mathbb{H}$ onto $\mathbb{D}$ with $f(2i)=0$ and $f(\infty)=1$.
- 2 Find all Möbius maps of $\mathbb{D}$ onto $\mathbb{H}$ with $0\mapsto i$; which of them sends $1$ to $\infty$?
- 3 Find the automorphism $f$ of $\mathbb{D}$ with $f(\frac12)=0$ and $f(1)=1$.
- 4 Which of the following are automorphisms of $\mathbb{H}$? (a) $\frac{2z+1}{z+1}$; (b) $\frac{z+1}{2z+1}$; (c) $-\frac1z$; (d) $\frac{iz+1}{z+i}$.
- 5 Find the automorphism $f$ of $\mathbb{H}$ with $f(i)=1+i$ and $f(1)=\infty$.
- 6 Find the conformal map $f$ of the disc $|z|\lt3$ onto $\mathbb{D}$ with $f(1)=0$ and $f'(1)\gt0$, and compute $f'(1)$.
- 7 Let $f:\mathbb{H}\to\mathbb{H}$ be analytic with $f(i)=i$. Prove that $|f'(i)|\le1$, with equality only for the maps $\rho_\theta$.
- 8 Show that $\rho_\theta\circ\rho_\varphi=\rho_{\theta+\varphi}$, and that $\theta\mapsto\rho_\theta$ induces an isomorphism of $\mathbb{R}/\pi\mathbb{Z}$ onto the stabiliser of $i$ in $\operatorname{Aut}(\mathbb{H})$.
- 9 Find the Möbius map of $\mathbb{D}$ onto the right half-plane with $0\mapsto1$ and $1\mapsto0$.
- 1 Find the images of the sector $\{0\lt\arg z\lt\frac\pi4\}$ and of $\{|z|\lt2,\ 0\lt\arg z\lt\frac\pi4\}$ under $z^4$.
- 2 Find the images under $e^z$ of the half-strip $\{\operatorname{Re}z\lt0,\ 0\lt\operatorname{Im}z\lt\pi\}$ and of the rectangle $\{0\lt\operatorname{Re}z\lt1,\ 0\lt\operatorname{Im}z\lt\frac\pi2\}$.
- 3 Find a conformal map of the sector $\{0\lt\arg z\lt\frac\pi3\}$ onto $\mathbb{D}$.
- 4 Show that $e^{i\pi z}$ maps the strip $\{0\lt\operatorname{Re}z\lt1\}$ conformally onto $\mathbb{H}$.
- 5 Find the image of the circle $|z|=2$ under the Joukowski map $J$.
- 6 Prove that $J$ maps $\mathbb{H}$ conformally onto $\mathbb{C}\setminus\big((-\infty,-1]\cup[1,\infty)\big)$.
- 7 Map the region between the circles $|z-1|=1$ and $|z-2|=2$ (inside the larger, outside the smaller) conformally onto a strip, and then onto $\mathbb{D}$.
- 8 Prove that $\mathbb{H}$, $\mathbb{D}$, $\mathbb{C}\setminus(-\infty,0]$ and $\{0\lt\operatorname{Im}z\lt\pi\}$ are pairwise conformally equivalent, and that none of them is conformally equivalent to $\mathbb{C}$.
- 9 Transplant Example 2(e) to the disc: find a function harmonic on $\mathbb{D}$ with boundary values $1$ on the upper semicircle and $0$ on the lower semicircle, and its value at $0$.
E1. CSIR Part B GATE MA Let $T$ be the Möbius transformation with $T(0)=i$, $T(1)=1$ and $T(\infty)=-i$. Then $T(-1)$ equals (A) $1$; (B) $-1$; (C) $i$; (D) $0$.
E2. CSIR Part B The image of the disc $\{|z-1|\lt1\}$ under $w=\dfrac{z-2}{z}$ is (A) $\{\operatorname{Re}w\lt0\}$; (B) $\{\operatorname{Re}w\gt0\}$; (C) $\{|w|\lt1\}$; (D) $\{\operatorname{Im}w\gt0\}$.
E3. CSIR Part B GATE MA Let $f$ be analytic on the upper half-plane with $|f(z)|\lt1$ there and $f(i)=0$. The largest possible value of $|f(2i)|$ is (A) $\frac12$; (B) $\frac13$; (C) $\frac23$; (D) $1$.
E4. CSIR Part C Let $T(z)=\dfrac{az+b}{cz+d}$ with $ad-bc\ne0$, regarded as a map of $\mathbb{C}_\infty$. Which of the following statements are true? (A) $T$ has at least one fixed point in $\mathbb{C}_\infty$. (B) If $T$ has three distinct fixed points, then $T$ is the identity. (C) If $a,b,c,d$ are real, then $T$ maps the upper half-plane onto itself. (D) $T$ maps every circle in $\mathbb{C}$ onto a circle in $\mathbb{C}$.
E5. CSIR Part C Let $J(z)=\frac12\left(z+\frac1z\right)$. Which of the following are true? (A) $e^z$ maps the strip $\{0\lt\operatorname{Im}z\lt\pi\}$ conformally onto the upper half-plane. (B) $z^2$ maps the upper half-plane onto the upper half-plane. (C) $J$ is injective on $\{|z|\gt1\}$. (D) $J$ maps the circle $|z|=3$ onto a circle.
E6. CSIR Part C GATE MA Which of the following pairs of domains are conformally equivalent? (A) $\mathbb{D}$ and $\mathbb{C}$. (B) $\mathbb{D}$ and $\mathbb{C}\setminus(-\infty,0]$. (C) $\mathbb{C}\setminus\{0\}$ and $\mathbb{D}\setminus\{0\}$. (D) The strip $\{0\lt\operatorname{Re}z\lt1\}$ and $\mathbb{D}$.
Suggested plan — 3 lectures × 60 min.
- Conformality and Möbius algebra. Open with the chapter teaser (small squares stay square). Definition 4.1 and Theorem 4.1 with the chain-rule picture; Theorem 4.2 (the circle $\alpha+\beta e^{-2i\theta}$ argument is short and memorable); Theorem 4.3 with $z^2$ at $0$; Definition 4.2 and the warning about the two meanings of "conformal"; Theorem 4.4 and Example 2(e). Examples 1–2 on the board. Definition 4.3 and Theorem 4.5 with the matrix dictionary; Theorem 4.6 in five minutes.
- Fixed points, cross-ratio, circles. Theorem 4.7 — project Interactive 1, options 4–6, to show the four types moving; Example 3. Definition 4.4, Theorem 4.8 (three points), and the mini check-lab under Card II. Theorems 4.9–4.10 with the inversion computation; Definition 4.6 and Theorem 4.11 — draw a circle, a pair of symmetric points and their images; the strategy box and Examples 4–5.
- Disc, half-plane, standard maps, Riemann. Theorem 4.12 via the perpendicular bisector and the symmetry argument; Theorem 4.13 with $\operatorname{Im}T=\frac{(ad-bc)\operatorname{Im}z}{|cz+d|^2}$; Example 6. Powers, exponential and Theorem 4.14 with Interactive 2 (end on the aerofoil); the composition strategy and Example 7. Theorem 4.15: statement, uniqueness in full, existence as the three-step sketch. Exam corner E4–E6 as a class vote.
Discussion questions. (1) Why must a map that is conformal at every point of a domain be analytic, while a map that preserves only the sizes of angles can be $\bar z$? (2) Möbius maps have three complex parameters, and so do triples of points. Which other "three-point" facts follow from this count? (3) $\mathbb{C}$ and $\mathbb{D}$ are homeomorphic. Which feature of analytic maps prevents a conformal equivalence, and why does that feature not prevent $\mathbb{D}\cong\mathbb{H}$?
Misconceptions to watch for. Taking "$f'\ne0$ everywhere" to mean injective; forgetting that a circle through the pole becomes a line; assuming interior goes to interior or centre goes to centre; mixing cross-ratio conventions; reading the type from $a+d$ without normalising $ad-bc=1$; accepting $\frac1z$ as an automorphism of $\mathbb{H}$; believing $z^2$ maps $\mathbb{H}$ onto $\mathbb{H}$; applying the Riemann mapping theorem to $\mathbb{C}$ or to annuli.
Suggested homework. I(1, 3, 6, 8), II(1, 3, 5, 7, 8), III(2, 4, 5, 7), IV(2, 5, 7, 9); Exam corner E1, E3, E5.
Research question. A coaxial cable whose inner conductor has drifted off-centre is an eccentric annulus: the region between $|z|=1$ (outer conductor) and $|z-d|=r$ (inner conductor), $0\lt d\lt1-r$. Find a Möbius map that turns it into a concentric annulus $\rho\lt|w|\lt1$, and use it to compute the electrostatic potential and the capacitance per unit length. How much does the eccentricity change the capacitance?
Why it matters. The potential between two conductors is the harmonic function with constant values on each conductor, and conformal maps transplant harmonic functions (Theorem 4.4). Eccentricity of cables and of shaft–bearing gaps is a real manufacturing tolerance: the same computation describes the capacitance of an off-centre coaxial line in telecommunications, the field around a misaligned high-voltage conductor, and (with stream functions instead of potentials) the flow in an eccentric journal bearing.
What to do.
- Look for $\varphi_a(z)=\frac{z-a}{1-az}$ with real $a\in(-1,1)$; it maps $|z|=1$ onto itself. Show that $\varphi_a$ maps the inner circle onto a circle centred at $0$ iff $a$ and $\frac1a$ are symmetric with respect to the inner circle (Theorem 4.11), i.e. $(a-d)\left(\frac1a-d\right)=r^2$, and solve this quadratic.
- Put $\rho=|\varphi_a(d+r)|$. Show that $u(z)=\dfrac{\log|\varphi_a(z)|}{\log\rho}$ is harmonic in the eccentric annulus, equal to $0$ on the outer and $1$ on the inner circle.
- The capacitance per unit length of a concentric cable with radius ratio $\rho$ is $\frac{2\pi\varepsilon_0}{\log(1/\rho)}$. Compute it for $d=0.4$, $r=0.3$ and compare with the centred cable ($d=0$).
- Verify numerically the classical closed form $\log\frac1\rho=\operatorname{arccosh}\dfrac{1+r^2-d^2}{2r}$ for several $(d,r)$, and plot the capacitance against $d$.
Tools: Python (NumPy) or a spreadsheet; optionally a finite-difference Laplace solver on a grid to confirm $u$ independently.
Hand in: The derivation of $a$ and $\rho$, a table of $\rho$ and capacitance for $d=0,0.1,\dots,0.6$ (with $r=0.3$), a plot of the equipotentials $u=$ const (the preimages of circles $|w|=$ const — which family of circles from §1.1 are they?), and one paragraph interpreting the result physically.
Going further. Read about the method of images in electrostatics and its link with symmetric points; Schwarz–Christoffel maps for polygonal conductors; and numerical conformal mapping (for example the "Zipper" algorithm) used in engineering and in computing the shapes of flow regions.
🧪 Self-check quiz
Q1. At which point does $f(z)=z^2-4z$ fail to be conformal?
Q2. The fixed points of $T(z)=\dfrac{2z+1}{z+2}$ are
Q3. Under $w=\dfrac{z-i}{z+i}$ the real axis (with $\infty$) is mapped onto
Q4. Which of the following is an automorphism of the upper half-plane?
Q5. The Joukowski map $J(z)=\frac12\left(z+\frac1z\right)$ maps the circle $|z|=2$ onto