Möbius transformations are the fractional linear changes of variable that preserve the projective geometry of the complex line. They sit at the meeting point of complex analysis, linear algebra, [group actions](/page/Group%20Action), and the geometry of the Riemann sphere. Their defining formula is short, but it encodes several large facts: the [automorphisms of the Riemann sphere](/theorems/7862), the symmetries of many planar domains, the invariance of the cross-ratio, and the way lines and circles form one unified family. The naive class of affine maps $z \mapsto az+b$ is too small for the natural compactification of the complex plane. Once the point at infinity is included, inversion becomes as basic as translation and rotation: \begin{align*} z &\mapsto \frac{1}{z}. \end{align*} Möbius transformations are the family generated under composition by translations, dilations, rotations, and inversion; equivalently, they are the projective linear group $PGL(2,\mathbb{C})$ acting on the complex projective line $\mathbb{CP}^1$. Here $PGL(2,\mathbb{C})$ means the group of invertible $2\times 2$ complex matrices after identifying matrices that differ by a nonzero scalar multiple, and $\mathbb{CP}^1$ is the set of one-dimensional complex lines in $\mathbb{C}^2$, naturally identified with the Riemann sphere. ## Definition The main object is a fractional [linear map](/page/Linear%20Map) that is allowed to send an ordinary complex number to infinity and to have a value at infinity itself. Temporarily write $\widehat{\mathbb{C}}$ for the complex plane enlarged by one point. The determinant condition below is the algebraic test that the fractional expression is not a disguised constant map. [definition: Mobius Transformation] A Möbius transformation is a map $T: \widehat{\mathbb{C}} \to \widehat{\mathbb{C}}$ for which there exist $a,b,c,d \in \mathbb{C}$ satisfying \begin{align*} ad-bc \ne 0 \end{align*} and \begin{align*} T(z) = \frac{az+b}{cz+d} \end{align*} for every $z \in \mathbb{C}$ with $cz+d \ne 0$, extended to the exceptional finite point and to $\infty$ as a map on $\widehat{\mathbb{C}}$. [/definition] The extension rule is part of the construction: if $c \ne 0$, then $T(-d/c)=\infty$ and $T(\infty)=a/c$; if $c=0$, then $T(\infty)=\infty$. To make those assignments mathematical rather than informal, the symbol $\widehat{\mathbb{C}}$ has to denote a genuine space. The point at infinity is not a decorative extra point: it is what turns the pole of a fractional expression into an ordinary value and lets these maps be studied globally on the same compact surface as the Riemann sphere. [definition: Extended Complex Plane] The extended complex plane is the set \begin{align*} \widehat{\mathbb{C}} := \mathbb{C} \cup \{\infty\}. \end{align*} It is equipped with the topology of the Riemann sphere. [/definition] A practical difficulty appears as soon as the coefficients are written down: multiplying all of them by the same nonzero scalar changes no value of the map. The next observation records this ambiguity before we use matrices to organise composition. [remark: Scalar Ambiguity] If $\lambda \in \mathbb{C}\setminus\{0\}$, then the coefficient lists $(a,b,c,d)$ and $(\lambda a,\lambda b,\lambda c,\lambda d)$ define the same Möbius transformation. [/remark] The formula needs a bookkeeping device that remembers composition but forgets irrelevant scalar multiples. The projective matrix representation supplies exactly that device: it turns Möbius transformations into matrix classes rather than isolated fractional expressions. [definition: Projective Matrix of a Mobius Transformation] A projective matrix representing a Möbius transformation is an equivalence class $[A]$ of invertible matrices $A \in GL(2,\mathbb{C})$ under the relation $A \sim \lambda A$ for $\lambda \in \mathbb{C}\setminus\{0\}$. If $A$ has first row $(a,b)$ and second row $(c,d)$, then $[A]$ determines the map $T_A: \widehat{\mathbb{C}} \to \widehat{\mathbb{C}}$ defined by \begin{align*} T_A(z)=\frac{az+b}{cz+d} \end{align*} for $z \in \mathbb{C}$ with $cz+d \ne 0$, with the corresponding extended values at the pole and at $\infty$. [/definition] The projective viewpoint is the bridge from formulas to geometry. It packages the fractional expression in a way that remembers composition while still identifying scalar multiples that give the same map. The same viewpoint will reappear later when composition, inverses, and domain automorphisms are treated as consequences of one projective action rather than as unrelated coordinate tricks. [example: Translation as a Projective Matrix] The translation $T(z)=z+2$ is represented by the matrix with first row $(1,2)$ and second row $(0,1)$: \begin{align*} \begin{pmatrix}1&2\end{pmatrix} \quad\text{above}\quad \begin{pmatrix}0&1\end{pmatrix}. \end{align*} Indeed, for finite $z$, \begin{align*} \frac{1\cdot z+2}{0\cdot z+1}=z+2, \end{align*} and the determinant is $1$. This small example shows why affine maps are already Möbius transformations: the fractional formula includes them as the case $c=0$. [/example] ## Equivalent Characterisations The most concrete definition uses a fractional formula, but the same class can be recognised intrinsically. This matters because in complex analysis one often meets a map first as a holomorphic bijection, not as a displayed quotient of linear functions. The intrinsic question is now whether the fractional formula is merely one convenient way to produce symmetries of the sphere, or whether it captures all of them. A holomorphic automorphism of the Riemann sphere is first encountered as a global analytic bijection, with no coefficients displayed. The next theorem closes that gap by showing that the analytic and projective descriptions define the same class of maps. [quotetheorem:7862] This classification is the reason Möbius transformations are not just a useful supply of examples. Every holomorphic self-symmetry of the Riemann sphere is already obtained from a projective linear change of coordinates, so there are no additional global automorphisms outside the fractional linear family. The compactness of the sphere and the single point at infinity are essential here; the statement is not a classification of automorphisms of arbitrary plane domains. The point of this classification is that the fractional linear formula is unavoidable rather than merely convenient. It says that every global holomorphic symmetry of the Riemann sphere is already a Möbius transformation, so there are no additional automorphisms hiding beyond projective linear maps. The global hypothesis is essential: on smaller domains, holomorphic bijections can have very different forms, but on the compact sphere the single pole allowed by the point at infinity forces degree one. This result is the reason later geometric statements about Möbius maps can be read as statements about all holomorphic self-symmetries of the sphere. Once the global automorphisms have been identified, the next practical question is how much data determines one of them. The classification says that the map must have a fractional linear form, but the coefficients still contain redundant scale and are not the most geometric way to prescribe the transformation. When constructing a Möbius transformation, solving for four coefficients includes an artificial scaling freedom: multiplying all four coefficients by the same nonzero scalar gives the same map. This makes the raw formula a poor way to specify a transformation uniquely. Three distinct source points and three distinct target points remove exactly this ambiguity, so the natural existence-and-uniqueness question is whether those three prescribed images determine one Möbius transformation. [quotetheorem:7863] This three-point result is the rigidity principle behind most computations with Möbius transformations. For example, there is a unique transformation carrying any ordered triple of distinct points to $(0,1,\infty)$, so a problem about an arbitrary triple can often be moved to that standard configuration. The distinctness hypotheses are necessary: if two source points coincide, or two target points coincide, the requested map would have to identify distinct points or fail to be one-to-one on the sphere. The theorem is also a warning about limits. Three point images determine a projective coordinate change, but they do not yet describe how a fourth point behaves. Once the first three images are fixed there is no remaining freedom, so the only possible invariant for a fourth point must measure its position relative to those three anchors. The cross-ratio is exactly that numerical invariant and becomes the central invariant of Möbius geometry. [definition: Cross Ratio] Let \begin{align*} \operatorname{Conf}_4(\mathbb{C}) := \{(z_1,z_2,z_3,z_4) \in \mathbb{C}^4 : z_i \ne z_j \text{ whenever } i \ne j\}. \end{align*} The cross-ratio is the map from $\operatorname{Conf}_4(\mathbb{C})$ to $\mathbb{C}\setminus\{0,1\}$ that sends $(z_1,z_2,z_3,z_4)$ to $[z_1,z_2;z_3,z_4]$, where \begin{align*} [z_1,z_2;z_3,z_4] = \frac{(z_3-z_1)(z_4-z_2)}{(z_3-z_2)(z_4-z_1)}. \end{align*} [/definition] Because Möbius transformations naturally live on $\widehat{\mathbb{C}}$, the invariant must also allow one of the four points to be $\infty$. Rather than treating this as an informal convention each time it appears, the extended version is obtained by taking the corresponding limiting value in the finite formula. [definition: Extended Cross Ratio] Let \begin{align*} \operatorname{Conf}_4(\widehat{\mathbb{C}}) := \{(z_1,z_2,z_3,z_4) \in \widehat{\mathbb{C}}^4 : z_i \ne z_j \text{ whenever } i \ne j\}. \end{align*} The extended cross-ratio is the map from $\operatorname{Conf}_4(\widehat{\mathbb{C}})$ to $\widehat{\mathbb{C}}\setminus\{0,1,\infty\}$ that sends $(z_1,z_2,z_3,z_4)$ to $[z_1,z_2;z_3,z_4]$, where $[z_1,z_2;z_3,z_4]$ is the limiting value of \begin{align*} \frac{(u_3-u_1)(u_4-u_2)}{(u_3-u_2)(u_4-u_1)} \end{align*} over quadruples $(u_1,u_2,u_3,u_4) \in \operatorname{Conf}_4(\mathbb{C})$ tending to $(z_1,z_2,z_3,z_4)$ in $\widehat{\mathbb{C}}^4$. [/definition] On quadruples with no infinite entry, this recovers the ordinary cross-ratio. The limiting formulation matters because it lets the same symbol be used without a separate case split each time a point is $\infty$. The next question is whether this four-point number is intrinsic to the geometry of the Riemann sphere, rather than an artifact of the coordinate formula used to define it. Möbius transformations are the natural changes of coordinate on $\widehat{\mathbb{C}}$, so the essential test is whether they preserve the cross-ratio and whether the cross-ratio detects when two ordered quadruples are related by such a transformation. This criterion is needed because preserving a formula is only half of the geometric story. To use the cross-ratio as a coordinate-free invariant, one also needs the converse direction: equal cross-ratios should mean that a Möbius transformation exists carrying one ordered configuration to the other. The point of introducing the next result is to turn the limiting formula into a usable classification principle. It asks exactly which numerical invariant survives every change of spherical coordinate, and whether that invariant is strong enough to decide when two ordered quadruples are the same up to a Möbius transformation. There is a real obstruction to overcome here: the displayed formula for the cross-ratio singles out the affine coordinate $z$, while Möbius geometry treats all spherical coordinates as equally valid. The coming theorem identifies the cross-ratio as the quantity that survives this change of viewpoint, and it also explains why matching that one number is enough to build the required coordinate change between ordered configurations. Before using cross-ratio as a classification tool, the page needs a precise invariance-and-converse statement. The forward part will justify computing the same number after any Möbius change of coordinate, while the converse part will justify reading equality of cross-ratios as a certificate that two ordered four-point configurations are Möbius-equivalent. The quoted theorem uses the common notation $\mathbb{C}_\infty$ for the same sphere denoted here by $\widehat{\mathbb{C}}$, and $\mathcal{M}$ for the class of Möbius transformations. With that notation understood, the result says that the cross-ratio is not an auxiliary coordinate artifact: it is precisely the four-point quantity that Möbius transformations preserve, and it also gives the matching existence criterion for sending one ordered triple to another. This is the exact place where the four-point condition becomes more than an invariant check. A Möbius transformation is already determined by the images of three distinct points, so the unresolved question is what compatibility condition decides the possible image of a fourth point. The theorem supplies that missing condition: equality of cross-ratios is the numerical test that makes the prescribed fourth image consistent with the unique transformation determined by the first three images. The formal statement is therefore not merely saying that one expression happens to be unchanged. It packages two directions that will be used differently: invariance lets one compute the cross-ratio in whatever coordinate is convenient, while the converse turns equality of cross-ratios into an existence statement for a Möbius transformation. This is the bridge from calculation to classification. [quotetheorem:814] After the theorem, the cross-ratio can be treated as the complete invariant of ordered quadruples of distinct points on the sphere. If two quadruples have different cross-ratios, no Möbius transformation can send one to the other in the prescribed order. If their cross-ratios agree, the first three images determine the transformation and the theorem says the fourth image is forced to be the desired one. The order of the four points matters. Permuting the entries changes the displayed cross-ratio by one of the usual six related values, so the theorem is best read as a statement about ordered configurations rather than unordered sets of four points. This prevents a common misuse: equality after an arbitrary relabelling is a different question from equality for the specified correspondence of points. This also explains why normal forms such as $0$, $1$, and $\infty$ are so useful. A Möbius transformation can first move three distinct points to those standard positions; then the fourth point is measured by a single complex number, namely the cross-ratio. Later classification arguments repeatedly use this pattern: move three points by projective freedom, and let the remaining invariant record what cannot be normalized away. The standard examples below show the same rigidity in familiar pieces: translations, rotations, dilations, and inversion look elementary in coordinates, but together they already exhibit the mechanisms behind the general fractional linear map. Before turning to special domains such as disks and half-planes, it is worth keeping these examples visible, because they are the concrete generators behind the projective and conformal properties developed later. ## Standard Examples The simplest Möbius transformations are already familiar from elementary complex arithmetic. Translations, rotations, and dilations appear when $c=0$, while inversion appears when $a=d=0$. ### Affine Similarities [example: Affine Transformations] Let $a,b \in \mathbb{C}$ with $a \ne 0$, and define $T: \widehat{\mathbb{C}} \to \widehat{\mathbb{C}}$ by \begin{align*} T(z)=az+b, \qquad T(\infty)=\infty. \end{align*} This has the Möbius form with coefficient list $(a,b,0,1)$, since for every $z \in \mathbb{C}$, \begin{align*} \frac{az+b}{0z+1}=\frac{az+b}{1}=az+b. \end{align*} The determinant of this coefficient list is \begin{align*} a\cdot 1-b\cdot 0=a-0=a, \end{align*} which is nonzero by assumption, so the determinant condition in the definition of a Möbius transformation is satisfied. Because the lower-left coefficient is $c=0$, the extended value is $T(\infty)=\infty$. If $a=re^{i\theta}$ with $r=|a|>0$, then on $\mathbb{C}$ the formula factors as \begin{align*} z \mapsto rz \mapsto e^{i\theta}(rz)=az \mapsto az+b. \end{align*} Thus an affine Möbius transformation fixes the point at infinity and acts on the finite plane as a dilation by $r$, a rotation by $\theta$, and then a translation by $b$. [/example] Affine maps show that Möbius transformations include the ordinary similarity geometry of the complex plane. The new behaviour begins when $c \ne 0$, because then a finite point is sent to $\infty$. ### Inversion and Circles [example: Inversion] The map $T: \widehat{\mathbb{C}} \to \widehat{\mathbb{C}}$ given by \begin{align*} T(z)=\frac{1}{z}, \qquad T(0)=\infty, \qquad T(\infty)=0 \end{align*} has Möbius form with coefficient list $(0,1,1,0)$, because for $z \ne 0$, \begin{align*} \frac{0z+1}{1z+0}=\frac{1}{z}. \end{align*} The determinant is \begin{align*} 0\cdot 0-1\cdot 1=-1 \ne 0, \end{align*} so the determinant condition in the definition of a Möbius transformation is satisfied. Since $c=1$ and $d=0$, the exceptional finite point is $-d/c=0$, so the extended values are $T(0)=\infty$ and $T(\infty)=a/c=0$. Now let $L=\{z \in \mathbb{C}: \operatorname{Re}(z)=1\}$. Every point of $L$ has the form $z=1+iy$ with $y \in \mathbb{R}$, and \begin{align*} T(1+iy)=\frac{1}{1+iy}. \end{align*} Multiplying numerator and denominator by the conjugate $1-iy$ gives \begin{align*} \frac{1}{1+iy}=\frac{1-iy}{(1+iy)(1-iy)}. \end{align*} The denominator expands as \begin{align*} (1+iy)(1-iy)=1-iy+iy-i^2y^2=1+y^2. \end{align*} Therefore \begin{align*} T(1+iy)=\frac{1-iy}{1+y^2}=\frac{1}{1+y^2}-i\frac{y}{1+y^2}. \end{align*} If $w=u+iv=T(1+iy)$, then \begin{align*} u=\frac{1}{1+y^2}, \qquad v=-\frac{y}{1+y^2}. \end{align*} Hence \begin{align*} u-\frac{1}{2}=\frac{1}{1+y^2}-\frac{1}{2}=\frac{2-(1+y^2)}{2(1+y^2)}=\frac{1-y^2}{2(1+y^2)}. \end{align*} Squaring the two coordinates gives \begin{align*} \left(u-\frac{1}{2}\right)^2=\frac{(1-y^2)^2}{4(1+y^2)^2} \end{align*} and \begin{align*} v^2=\frac{y^2}{(1+y^2)^2}=\frac{4y^2}{4(1+y^2)^2}. \end{align*} Adding these expressions, \begin{align*} \left(u-\frac{1}{2}\right)^2+v^2=\frac{(1-y^2)^2+4y^2}{4(1+y^2)^2}. \end{align*} The numerator expands to \begin{align*} (1-y^2)^2+4y^2=1-2y^2+y^4+4y^2=1+2y^2+y^4=(1+y^2)^2. \end{align*} Thus \begin{align*} \left(u-\frac{1}{2}\right)^2+v^2=\frac{(1+y^2)^2}{4(1+y^2)^2}=\frac{1}{4}. \end{align*} So every point of the image $T(L)$ lies on the circle with centre $1/2$ and radius $1/2$. Conversely, if $w=u+iv$ satisfies \begin{align*} \left(u-\frac{1}{2}\right)^2+v^2=\frac{1}{4}, \end{align*} then expanding gives \begin{align*} u^2-u+\frac{1}{4}+v^2=\frac{1}{4}. \end{align*} Hence \begin{align*} u^2+v^2=u. \end{align*} The point $w=0$ also satisfies the circle equation and is the image of $\infty$, so it is not obtained from finite points of $L$. For $w \ne 0$, compute the inverse image under inversion: \begin{align*} \frac{1}{w}=\frac{\bar{w}}{|w|^2}=\frac{u-iv}{u^2+v^2}. \end{align*} Using $u^2+v^2=u$ gives \begin{align*} \frac{1}{w}=\frac{u-iv}{u}=1-i\frac{v}{u}, \end{align*} whose real part is $1$. Therefore every nonzero point of the circle lies in $T(L)$, while the missing point $0$ is $T(\infty)$. Inversion sends the line $L \cup \{\infty\}$ to the circle $\left|w-\frac{1}{2}\right|=\frac{1}{2}$, and on finite points it sends the Euclidean line $L$ to that circle with the point $0$ omitted. [/example] The inversion example exposes a defect in the ordinary Euclidean classification: a transformation as simple as $z \mapsto 1/z$ can turn a line into a circle once the point at infinity is included. To state an image theorem without separating many exceptional cases, lines must be folded into the same class as circles by treating each line as passing through $\infty$ on the sphere. [definition: Generalised Circle] A generalised circle in $\widehat{\mathbb{C}}$ is either a Euclidean circle in $\mathbb{C}$ or a set of the form $L \cup \{\infty\}$, where $L \subsetneq \mathbb{C}$ is a Euclidean line. [/definition] The generalised-circle language is needed because Möbius transformations can send a finite point of a circle to $\infty$. Once lines through $\infty$ are admitted, the image of every member of the family remains inside the family. [quotetheorem:813] This theorem turns the inversion computation into a structural rule. It absorbs the apparent exceptions caused by poles and by the point at infinity: a line is not a failure of circle preservation, but a circle through infinity on the sphere. As a result, one can track boundaries of disks, half-planes, and their images without redoing separate algebra for vertical lines, ordinary circles, and circles passing through the pole. A second model example identifies the upper half-plane with the unit disk. This calculation is the standard way to transfer boundary and harmonic function problems between two domains with different visual shapes but the same conformal geometry. The generalised-circle theorem explains why the boundary line of the half-plane should become a circle; the remaining computation determines which side of that boundary is carried into the disk. This example will later be used to transport automorphism statements between the disk model and the half-plane model. [example: Cayley Transform] Let $\mathbb{H}=\{z \in \mathbb{C}: \operatorname{Im}(z)>0\}$ and define $C: \widehat{\mathbb{C}} \to \widehat{\mathbb{C}}$ by \begin{align*} C(z)=\frac{z-i}{z+i}. \end{align*} This has Möbius form with coefficient list $(1,-i,1,i)$. Its determinant is \begin{align*} 1\cdot i-(-i)\cdot 1=i+i=2i\ne 0, \end{align*} so the determinant condition in the definition of a Möbius transformation is satisfied. Now take $z=x+iy$ with $y>0$. Then \begin{align*} z-i=x+i(y-1) \end{align*} and \begin{align*} z+i=x+i(y+1). \end{align*} Using $|a+ib|^2=a^2+b^2$ for real $a,b$, we get \begin{align*} |z-i|^2=x^2+(y-1)^2 \end{align*} and \begin{align*} |z+i|^2=x^2+(y+1)^2. \end{align*} Therefore \begin{align*} |C(z)|^2=\left|\frac{z-i}{z+i}\right|^2=\frac{|z-i|^2}{|z+i|^2}=\frac{x^2+(y-1)^2}{x^2+(y+1)^2}. \end{align*} The denominator exceeds the numerator because \begin{align*} x^2+(y+1)^2-\bigl(x^2+(y-1)^2\bigr)=(y+1)^2-(y-1)^2=4y>0. \end{align*} Hence \begin{align*} 0\le \frac{x^2+(y-1)^2}{x^2+(y+1)^2}<1, \end{align*} so $|C(z)|<1$. Thus $C$ maps $\mathbb{H}$ into the unit disk $\mathbb{D}=\{w \in \mathbb{C}: |w|<1\}$. For a boundary point $x\in\mathbb{R}$, we have \begin{align*} |x-i|^2=x^2+1 \end{align*} and \begin{align*} |x+i|^2=x^2+1. \end{align*} Thus \begin{align*} |C(x)|^2=\frac{|x-i|^2}{|x+i|^2}=\frac{x^2+1}{x^2+1}=1, \end{align*} so the real boundary is sent to the unit circle. To solve for the inverse, set $w=C(z)$: \begin{align*} w=\frac{z-i}{z+i}. \end{align*} Multiplying by $z+i$ gives \begin{align*} w(z+i)=z-i. \end{align*} Expanding both sides gives \begin{align*} wz+wi=z-i. \end{align*} Moving the $z$-terms to one side and the constant terms to the other gives \begin{align*} wz-z=-i-wi. \end{align*} Factoring gives \begin{align*} z(w-1)=-i(1+w). \end{align*} If $w\ne 1$, then \begin{align*} z=\frac{-i(1+w)}{w-1}=i\frac{1+w}{1-w}. \end{align*} So \begin{align*} C^{-1}(w)=i\frac{1+w}{1-w}. \end{align*} Now let $w=u+iv$ with $|w|<1$. Then \begin{align*} 1+w=1+u+iv \end{align*} and \begin{align*} 1-w=1-u-iv. \end{align*} Multiplying numerator and denominator by the conjugate $1-u+iv$ gives \begin{align*} \frac{1+w}{1-w}=\frac{(1+u+iv)(1-u+iv)}{(1-u-iv)(1-u+iv)}. \end{align*} The denominator is \begin{align*} (1-u-iv)(1-u+iv)=(1-u)^2+v^2. \end{align*} The numerator expands as \begin{align*} (1+u+iv)(1-u+iv)=(1-u^2-v^2)+2iv. \end{align*} Therefore \begin{align*} C^{-1}(w)=i\frac{(1-u^2-v^2)+2iv}{(1-u)^2+v^2}. \end{align*} Multiplying by $i$ in the numerator gives \begin{align*} C^{-1}(w)=\frac{-2v+i(1-u^2-v^2)}{(1-u)^2+v^2}. \end{align*} Hence \begin{align*} \operatorname{Im}(C^{-1}(w))=\frac{1-u^2-v^2}{(1-u)^2+v^2}. \end{align*} Since $|w|<1$ means $u^2+v^2<1$, the numerator satisfies $1-u^2-v^2>0$. The denominator satisfies \begin{align*} (1-u)^2+v^2=|1-w|^2>0, \end{align*} because $|w|<1$ implies $w\ne 1$. Therefore \begin{align*} \operatorname{Im}(C^{-1}(w))>0. \end{align*} Thus $C^{-1}$ sends $\mathbb{D}$ into $\mathbb{H}$, while the earlier calculation showed that $C$ sends $\mathbb{H}$ into $\mathbb{D}$. The Cayley transform therefore identifies the upper half-plane and the unit disk by mutually inverse fractional linear maps. [/example] The determinant hypothesis cannot be weakened. If it fails, the fractional expression loses projective information and no longer gives a bijection of the sphere. [example: Degenerate Coefficients] Take the coefficient list $(2,4,1,2)$, so $a=2$, $b=4$, $c=1$, and $d=2$. Its determinant is \begin{align*} ad-bc=(2)(2)-(4)(1)=4-4=0. \end{align*} For $z \ne -2$, the denominator is nonzero because $z+2 \ne 0$, and the numerator factors as \begin{align*} 2z+4=2(z+2). \end{align*} Therefore \begin{align*} \frac{2z+4}{z+2}=\frac{2(z+2)}{z+2}=2. \end{align*} Thus every finite point except $-2$ is sent to the same value $2$. In particular, \begin{align*} \frac{2\cdot 0+4}{0+2}=2 \end{align*} and \begin{align*} \frac{2\cdot 1+4}{1+2}=2. \end{align*} Since $0 \ne 1$, any extension to $\widehat{\mathbb{C}}$ would still fail to be injective. The zero determinant has collapsed the fractional expression to a constant on its ordinary domain, so the nonzero determinant condition is what prevents a Möbius formula from degenerating in this way. [/example] ## Projective Group Structure ### Composition and Inverses Because the formula is fractional, it is not immediately apparent that Möbius transformations form a group. The matrix representation makes this structural property natural. [quotetheorem:7864] This group statement is more than a closure check for a class of formulas. It says that composition of Möbius transformations is governed by multiplication of projective matrix classes, so algebraic operations in $PGL(2,\mathbb{C})$ become geometric operations on $\widehat{\mathbb{C}}$. The determinant hypothesis is necessary here: without invertibility, the fractional expression can collapse information and fail to define a bijective symmetry of the sphere. The result is the organizing fact behind the rest of the properties section, because inverses, iterates, conjugacies, and domain automorphisms can all be treated inside one group rather than by ad hoc manipulation of separate formulas. For a transformation to be useful as a coordinate change, its inverse must remain in the same class. The following formula gives the inverse directly from the same four coefficients. [quotetheorem:7865] The inverse formula makes the group structure usable in coordinates. It shows explicitly how to recover the original point from its image, and it also shows why the exceptional values at the pole and at infinity cannot be ignored: the finite pole of the original map becomes the value at infinity, while the point sent to the finite value $a/c$ must be handled on the sphere rather than only in the affine plane. This is what lets later arguments move freely between a transformation and its inverse when comparing images of circles, disks, and half-planes. To connect the projective formula with angle-preserving analysis, one needs the ordinary complex derivative away from the pole. The derivative also shows why the determinant condition is the analytic nondegeneracy condition. [quotetheorem:7866] The derivative formula records the local analytic content of a Möbius transformation away from its pole. Since $ad-bc\ne 0$, the derivative is never zero at any finite point where the formula is holomorphic, so the map preserves angles and is locally conformal there. The limitation is important: the displayed derivative is an affine-coordinate statement and does not apply at the pole or directly at $\infty$ without changing charts on the Riemann sphere. This nonvanishing derivative is the analytic reason Möbius transformations act as conformal equivalences of domains such as disks and half-planes. ### Disk and Half-Plane Automorphisms The unit disk is a standard bounded model domain in one-variable complex analysis, so identifying all of its analytic symmetries is a basic test case for the theory. The classification below shows that those symmetries are still Möbius transformations. [quotetheorem:7867] The disk classification says that every conformal self-symmetry of the unit disk is built from two visible choices: a point $a$ in the disk that is moved to the origin, and a unit complex number $\lambda$ that records the remaining rotation. Both parameters are necessary. The condition $|a|<1$ keeps the pole outside the closed unit disk, and $|\lambda|=1$ preserves boundary size rather than shrinking or expanding the disk. Typical examples include rotations $z\mapsto \lambda z$ when $a=0$, and the nontrivial maps that exchange $a$ with $0$. This theorem is the model-domain analogue of the sphere classification: once it is known, automorphisms of conformally equivalent domains can be obtained by conjugating with a map such as the Cayley transform. The upper half-plane is the other standard model for one-dimensional hyperbolic geometry, and its boundary is the real line rather than a circle. The Cayley transform suggests that its analytic symmetries should again be Möbius transformations, but now the coefficients must be real so that the real boundary is preserved. [quotetheorem:7868] The real-coefficient condition says that the boundary line $\mathbb{R}\cup\{\infty\}$ is preserved: real inputs have real extended outputs, except at the pole where the value is $\infty$. The sign of the determinant chooses the side of that boundary. Positive determinant preserves the orientation of the real boundary and sends a point such as $i$ into the upper half-plane; negative determinant would reverse the side and carry the upper half-plane to the lower half-plane. Thus the theorem is the half-plane analogue of the disk classification, with the boundary encoded by real projective matrices. ## Beyond and Connected Topics Möbius transformations are best understood as projective linear maps in dimension one. In projective geometry, points of the complex projective line are one-dimensional subspaces of $\mathbb{C}^2$, and an invertible linear map of $\mathbb{C}^2$ sends lines through the origin to lines through the origin. Passing from vectors to lines removes scalar ambiguity and produces the fractional formula. They are also central examples in complex analysis. In the language of Riemann surfaces, a Möbius transformation is holomorphic on $\widehat{\mathbb{C}}$: near $\infty$ this means holomorphic after changing to the local coordinate $\zeta=1/z$. Möbius transformations are already rational functions of the affine coordinate, but not every rational function should be counted as a Möbius transformation. The obstruction is degree: a general rational map may wrap the sphere over itself several times and may have critical points, while a Möbius transformation is a one-to-one projective coordinate change. The next result identifies exactly the borderline case where a rational map is still just a Möbius transformation. [quotetheorem:7869] Higher-degree rational maps behave differently: they are generally not injective, have branching, and do not preserve cross-ratios. From the viewpoint of [topology](/page/Topology), Möbius transformations are homeomorphisms of the sphere. From the viewpoint of [group actions](/page/Group%20Action), they form projective symmetry groups acting on points and configurations. From the viewpoint of differential geometry, they are conformal diffeomorphisms of the Riemann sphere. From the viewpoint of linear fractional transformations, they are the complex case with the point at infinity included as part of the domain and codomain. In hyperbolic geometry, special Möbius transformations preserve the unit disk or upper half-plane and act as isometries for the corresponding hyperbolic metrics. This is why the same formulas appear in the study of Fuchsian groups, modular transformations, and automorphic functions. For a broader course-note setting around these complex-analytic tools, see [Cambridge IB Complex Analysis](/page/Cambridge%20IB%20Complex%20Analysis). ## References Ahlfors, *Complex Analysis* (1979). Conway, *Functions of One Complex Variable I* (1978). Needham, *Visual Complex Analysis* (1997).