Mobius Transformations

A Möbius transformation is the rational map $f(z)=(az+b)/(cz+d)$ with $ad-bc\neq0$ , read not on the plane but on the Riemann sphere $\widehat{\mathbb{C}}=\mathbb{C}\cup\{\infty\}$ , where it becomes a bijection. Among the analytic functions a complex-variables course meets, these are the simplest ones still interesting: their derivative never vanishes, they are globally one-to-one, and they carry the sphere onto itself. Everything else is the working-out of a single structural fact — that composing two such maps is the same as multiplying their coefficient matrices. That one fact makes the subject a physicist's tool rather than a curiosity. Because rescaling $\begin{pmatrix}a&b\\c&d\end{pmatrix}$ leaves the map untouched, the Möbius group is the projective group $\mathrm{PGL}(2,\mathbb{C})\cong\mathrm{PSL}(2,\mathbb{C})$ , and inverses, iterates, fixed points, and the four-way classification all reduce to $2\times2$ linear algebra. The geometry is equally clean: every Möbius map is conformal and sends the family of "circlines" — circles and straight lines, all just circles once drawn on the sphere — into itself. Both properties follow from decomposing any map into translations, a rotation-dilation, and one inversion $z\mapsto1/z$ . The deepest invariant is the cross-ratio of four points, preserved by every Möbius map, and it is what pins the map down: three points and their three images determine the whole transformation with no freedom left. In physics these maps transplant harmonic functions between domains in two-dimensional potential theory, realise the isometries of the Poincaré disk, implement the action of the Lorentz group $\mathrm{SL}(2,\mathbb{C})$ on the celestial sphere, and organise the symmetry of two-dimensional conformal field theory.

Key takeaways

A Möbius transformation is a map $w=f(z)=\dfrac{az+b}{cz+d}$ with complex coefficients and $ad-bc\neq0$ , acting bijectively on the Riemann sphere $\widehat{\mathbb{C}}=\mathbb{C}\cup\{\infty\}$ .
These are exactly the conformal automorphisms of the sphere; composition mirrors $2\times2$ matrix multiplication, so the group is $\mathrm{PGL}(2,\mathbb{C})\cong\mathrm{PSL}(2,\mathbb{C})$ .
Every map factors into translations, a rotation-dilation, and one inversion $z\mapsto1/z$ , from which angle-preservation and circline-preservation both follow.
Three points fix a unique Möbius map; the cross-ratio is the fundamental invariant and gives the explicit triple-to-triple formula.
Fixed points (one or two on the sphere) sort every map into elliptic, hyperbolic, loxodromic, or parabolic through the single number $\tau=\mathrm{tr}^2$ of its unit-determinant matrix.
Key idea: a Möbius map is its $2\times2$ coefficient matrix up to scale, so composition, inversion, fixed points, and classification are all linear algebra in $\mathrm{PGL}(2,\mathbb{C})$ , while conformality, circline-preservation, and cross-ratio invariance follow from splitting every map into translations, scalings, and one inversion.

The core insight

A Möbius map is determined by its $2\times2$ coefficient matrix up to scale, so the entire theory — composition, inversion, fixed points, classification — is linear algebra in $\mathrm{PGL}(2,\mathbb{C})$ , while the geometry (conformality, circle-preservation, cross-ratio invariance) follows from decomposing every map into translations, scalings, and one inversion.

Where this lives today

The Möbius group is the visible symmetry of an enormous amount of modern physics, because $\mathrm{PSL}(2,\mathbb{C})$ is at once the conformal group of the two-sphere and, through its double cover $\mathrm{SL}(2,\mathbb{C})$ , the spin group of the Lorentz symmetry of spacetime. In two-dimensional conformal field theory the global conformal symmetry is exactly the Möbius group: its invariance fixes two- and three-point correlators completely and reduces four-point functions to a single function of the cross-ratio (7), so the kinematics of a 2D CFT is Möbius bookkeeping. String theory inherits this — gauge-fixing the $\mathrm{SL}(2,\mathbb{C})$ redundancy of the spherical worldsheet is why one fixes three vertex-operator positions in a tree amplitude. For three-dimensional hyperbolic manifolds the fundamental groups are discrete subgroups of $\mathrm{PSL}(2,\mathbb{C})$ — Kleinian groups — whose fractal limit sets (the images in Indra's Pearls) are orbits of Möbius maps. The active frontier is celestial holography: since a Lorentz boost acts on the sphere of light directions as a Möbius transformation, four-dimensional scattering amplitudes are being recast as correlators of a conjectural two-dimensional "celestial" CFT on that sphere, turning stellar aberration into the symmetry principle of a holographic dual.

The physical picture

Take the plane and turn it inside-out through the unit circle: a far point comes close, a near point flies out, and the origin and infinity trade places. That single move is the inversion $z\mapsto1/z$ , and if you also allow sliding (translation) and spinning-and-stretching (rotation-dilation), any Möbius map whatsoever is a composition of these pieces. Because each elementary move sends circles-and-lines to circles-and-lines and preserves the angle at which curves cross, so does every map built from them. The first figure makes the inversion concrete and shows why a straight line must become a circle.

Figure

Inversion w = 1 divided by z turns the plane inside-out across the unit circle: the point z = 2 maps to 0.5, and the vertical line Re z = 2 maps to a circle through the origin of centre 0.25 and radius 0.25 — a line is just a circle that happens to pass through infinity

Two more ideas complete the picture, and neither is visible from the finite plane. First, on the Riemann sphere there are no special points: $\infty$ is simply the north pole, a plane "line" is a sphere-circle running through it, and a Möbius map is a smooth reshuffling of the whole sphere that always maps the family of drawn circles to itself. Second, iterating a map makes points flow, and the flow is governed entirely by the map's fixed points — the one or two places it holds still. Whether orbits rotate around a fixed point, race between two fixed points, or spiral between them is read from a single number, the squared trace $\tau$ of the matrix. The second figure is that classification: four dynamical types, four values of $\tau$ .

Figure

The four dynamical types of a Möbius map, read from the squared trace τ of its unit-determinant matrix: elliptic rotates about the fixed points (τ = 2 for f = iz), parabolic slides from a single fixed point (τ = 4 for f = z + 1), hyperbolic flows between two fixed points (τ = 6.25 for f = 4z), and loxodromic spirals (τ = 2 + 1.5i for f = 2iz). Real τ in the range 0 to 4 is elliptic, τ = 4 is parabolic, real τ above 4 is hyperbolic, and complex τ is loxodromic

The core equations

The definition is a ratio of two linear expressions gated by one inequality. That $ad-bc\neq0$ is the whole difference between an interesting map and a constant: it makes $f$ invertible and, as equation (6) shows, keeps the derivative from vanishing.

w=f(z)=\frac{az+b}{cz+d},\qquad ad-bc\neq0 \qquad (1)

where: $a,b,c,d$ — complex coefficients; $ad-bc$ — the determinant, nonzero to guarantee invertibility; $z,w$ — source and image points of $\widehat{\mathbb{C}}$ . Reading: two points collide onto one image only if the determinant vanishes, so $ad-bc\neq0$ is precisely the one-to-one condition. The map becomes a bijection only once its two exceptional points are supplied by hand: the finite pole $z=-d/c$ has nowhere to go but the north pole, and $z=\infty$ lands at the ratio of leading coefficients.

f(\infty)=\frac{a}{c},\qquad f\!\left(-\frac{d}{c}\right)=\infty\quad(c\neq0) \qquad (2)

where: $\infty$ — the north pole of the Riemann sphere; $-d/c$ — the unique simple pole of $f$ . Reading: when $c=0$ there is no finite pole, $\infty$ is fixed, and the map is affine. The organising fact: attach to $f$ its coefficient matrix, and composing maps becomes multiplying matrices, up to the overall scale the map cannot see.

f\;\longleftrightarrow\;M=\begin{pmatrix}a&b\\c&d\end{pmatrix},\qquad f_2\circ f_1\;\longleftrightarrow\;M_2M_1 \qquad (3)

where: $M$ — a representative matrix, defined only up to a nonzero scalar; $\circ$ — composition of maps. Reading: because $\lambda M$ and $M$ give the same $f$ , the group is projective, $\mathrm{PGL}(2,\mathbb{C})$ , and normalising $\det M=1$ leaves only the sign ambiguity of $\mathrm{PSL}(2,\mathbb{C})$ . The inverse map is read straight off the matrix inverse — no fraction algebra needed once you trust equation (3).

f^{-1}(w)=\frac{dw-b}{-cw+a}\;\longleftrightarrow\;M^{-1}=\frac{1}{ad-bc}\begin{pmatrix}d&-b\\-c&a\end{pmatrix} \qquad (4)

where: $M^{-1}$ — the matrix inverse, well-defined precisely because $ad-bc\neq0$ . Reading: the overall factor $1/(ad-bc)$ is invisible to the map, so one usually writes the inverse as $\begin{pmatrix}d&-b\\-c&a\end{pmatrix}$ . Every non-affine map is a translation, an inversion, a rotation-dilation, and a final translation. Polynomial division of $\tfrac{az+b}{cz+d}$ exhibits this decomposition explicitly.

f(z)=\frac{a}{c}+\frac{bc-ad}{c^{2}}\cdot\frac{1}{\,z+\tfrac{d}{c}\,}\quad(c\neq0) \qquad (5)

where: $z\mapsto z+d/c$ — translation; $\zeta\mapsto1/\zeta$ — the single inversion; $\eta\mapsto\big((bc-ad)/c^{2}\big)\eta$ — rotation-dilation; a final shift by $a/c$ . Reading: since angle- and circline-preservation hold for each factor, they hold for $f$ — the engine behind both geometric theorems below. Conformality is a corollary of the determinant condition: the derivative is the determinant over a square.

f'(z)=\frac{ad-bc}{(cz+d)^{2}}\neq0 \qquad (6)

where: $f'$ — the complex derivative, nonzero everywhere $f$ is finite. Reading: an analytic map with $f'\neq0$ preserves angles and orientation, so $ad-bc\neq0$ literally says " $f$ is conformal". The one number every Möbius map preserves is the cross-ratio of four points — the fingerprint that makes the three-point theorem work.

(z_1,z_2;z_3,z_4)=\frac{(z_1-z_3)(z_2-z_4)}{(z_1-z_4)(z_2-z_3)} \qquad (7)

where: $z_1,\dots,z_4$ — four distinct points of $\widehat{\mathbb{C}}$ ; the bracket — their cross-ratio, unchanged by any Möbius map. Reading: because it is invariant, equating cross-ratios of source and image points yields the unique map matching three prescribed images.

\frac{(w-w_1)(w_2-w_3)}{(w-w_3)(w_2-w_1)}=\frac{(z-z_1)(z_2-z_3)}{(z-z_3)(z_2-z_1)} \qquad (8)

where: $z_k\mapsto w_k$ — the three prescribed image pairs ( $k=1,2,3$ ); $w$ — the image of a general $z$ , found by solving. Reading: three points and their images leave no freedom — the complex-analytic echo of "three points fix a projective map of a line". Finally, fixed points and classification. Solving $f(z)=z$ is a quadratic, so there are at most two, and the squared trace of the normalised matrix sorts the dynamics.

z_*=\frac{(a-d)\pm\sqrt{(a+d)^{2}-4(ad-bc)}}{2c},\qquad \tau=\frac{(a+d)^{2}}{ad-bc} \qquad (9)

where: $z_*$ — roots of $cz^2+(d-a)z-b=0$ ; $\tau$ — the squared trace of the unit-determinant representative. Reading: $\tau\in[0,4)$ is elliptic, $\tau=4$ is parabolic (a repeated fixed point), $\tau>4$ real is hyperbolic, and complex $\tau$ is loxodromic.

Derivation AComposition of Mobius maps is multiplication of their matrices

The goal is to prove the load-bearing statement of the whole theory: applying one Möbius map then a second gives a map whose coefficients are exactly the entries of the product of the two coefficient matrices, in that order. This turns every later question — inverses, iterates, fixed points, classification — from fraction-wrangling into $2\times2$ linear algebra, and explains why the Möbius group is a matrix group modulo scale.

Step 1. Take two maps $f_1(z)=\dfrac{a_1z+b_1}{c_1z+d_1}$ and $f_2(w)=\dfrac{a_2w+b_2}{c_2w+d_2}$ , and write the intermediate image as $w_1=f_1(z)$ . Why: naming the middle point lets us substitute cleanly instead of expanding a nested fraction blind.

Step 2. Substitute $w_1$ into $f_2$ : the composite is $\dfrac{a_2\frac{a_1z+b_1}{c_1z+d_1}+b_2}{c_2\frac{a_1z+b_1}{c_1z+d_1}+d_2}$ . Why: this is $f_2\circ f_1$ by definition; everything now is bookkeeping to clear the inner denominator $c_1z+d_1$ .

Step 3. Multiply numerator and denominator by $c_1z+d_1$ and collect powers of $z$ , giving $f_2\!\circ\! f_1(z)=\dfrac{(a_2a_1+b_2c_1)z+(a_2b_1+b_2d_1)}{(c_2a_1+d_2c_1)z+(c_2b_1+d_2d_1)}$ . Why: the four bracketed coefficients are precisely the entries of the matrix product $M_2M_1$ — top-left $a_2a_1+b_2c_1$ , and so on — so composition and multiplication agree term by term.

Step 4. Read off two corollaries. The determinant is multiplicative, $\det(M_2M_1)=\det M_2\,\det M_1$ , so a composite of non-degenerate maps is non-degenerate; and the same argument on $f\circ f^{-1}=\mathrm{id}$ forces $M^{-1}\propto\begin{pmatrix}d&-b\\-c&a\end{pmatrix}$ , equation (4). Why: these are the facts that make the Möbius maps a group and identify it with $\mathrm{GL}(2,\mathbb{C})$ modulo scalar matrices — that is, $\mathrm{PGL}(2,\mathbb{C})$ .

Go deeper · Derivation B · 1 minEvery Mobius map carries circles and lines to circles and lines

Derivation BEvery Mobius map carries circles and lines to circles and lines

The goal is to prove circline-preservation — the image of any circle or straight line is again a circle or straight line — without checking the general map directly. The strategy leans on the decomposition (5): translations and rotation-dilations are plane similarities that trivially preserve the family, so the theorem collapses to the single inversion $z\mapsto1/z$ , verified by one substitution into the unified equation of a circline.

Step 1. Write every circle-or-line in one complex equation, $Az\bar z+\bar\beta z+\beta\bar z+D=0$ with $A,D$ real and $\beta$ complex; $A\neq0$ is a circle, $A=0$ is a line. Why: unifying both shapes into one equation is what lets a single calculation cover the case where a circle turns into a line.

Step 2. Dispose of the easy factors. Under $z\mapsto\alpha z+\gamma$ (a translation or a rotation-dilation) a circle stays a circle and a line stays a line, since these maps are similarities of the plane. Why: by the decomposition (5) any Möbius map is built from these similarities and one inversion, so only the inversion can do anything non-obvious.

Step 3. Apply the inversion by putting $z=1/w$ , hence $\bar z=1/\bar w$ , into the circline equation: $A\dfrac{1}{w\bar w}+\bar\beta\dfrac{1}{w}+\beta\dfrac{1}{\bar w}+D=0$ . Multiply through by $w\bar w$ to clear denominators, giving $A+\bar\beta\,\bar w+\beta\,w+D\,w\bar w=0$ . Why: clearing the denominators is legitimate away from the origin, and it rearranges the equation into recognisable form.

Step 4. Read the result as $D\,w\bar w+\beta\,w+\bar\beta\,\bar w+A=0$ — the same circline equation with the roles of $A$ and $D$ swapped. It is a circle when $D\neq0$ and a line when $D=0$ . Why: since $D=0$ means the original circle passed through the origin, this is the precise mechanism by which a circle through $0$ becomes a straight line — a line is exactly a circline through the point that inversion sends to $\infty$ .

Go deeper · Derivation C · 2 minThe trace classifies the map into four dynamical types

Derivation CThe trace classifies the map into four dynamical types

The goal is to show that conjugating any Möbius map to a normal form reduces its long-time dynamics to one invariant, the squared trace $\tau=(a+d)^2/(ad-bc)$ , with four qualitatively different behaviours — rotating, translating, dilating, spiralling — in four ranges of $\tau$ . Because trace survives conjugation, and conjugation is just a change of coordinates on the sphere, $\tau$ is a genuine invariant of the map, not of the coordinates.

Step 1. Normalise $\det M=1$ by replacing $M$ with $M/\sqrt{ad-bc}$ ; the map is unchanged and now $\tau=(a+d)^2$ is just $\mathrm{tr}^2$ . Suppose first there are two distinct fixed points $p,q$ . Why: fixing the determinant removes the projective scaling freedom so that the trace becomes a well-defined number up to sign, and its square removes even the sign.

Step 2. Conjugate by the Möbius map sending $p\mapsto0$ and $q\mapsto\infty$ . In these coordinates a map fixing both $0$ and $\infty$ can only be the scaling $g(\zeta)=\lambda\zeta$ , whose matrix is $\mathrm{diag}(\sqrt\lambda,1/\sqrt\lambda)$ . Why: moving the fixed points to $0$ and $\infty$ is the cleanest coordinate choice, and it exposes the one free parameter, the multiplier $\lambda$ .

Step 3. Match traces: $\mathrm{tr}=\sqrt\lambda+1/\sqrt\lambda$ , so $\tau=\mathrm{tr}^2=\lambda+2+\lambda^{-1}$ . Because trace is conjugation-invariant, this $\tau$ equals $(a+d)^2/(ad-bc)$ of the original map. Why: this equation is the dictionary between the algebraic invariant $\tau$ and the geometric multiplier $\lambda$ that actually drives the orbits.

Step 4. Sort by $\lambda$ . If $|\lambda|=1$ with $\lambda\neq1$ the map rotates about the fixed points, orbits are circles, $\tau\in[0,4)$ real — elliptic. If $\lambda>0$ real, orbits run from the repelling to the attracting fixed point and $\tau>4$ real — hyperbolic. If $|\lambda|\neq1$ and $\lambda\notin\mathbb{R}$ , orbits spiral and $\tau$ is complex — loxodromic. Why: the modulus of $\lambda$ controls expansion and its argument controls rotation, so the two together fix the picture.

Step 5. Handle the degenerate case. If the fixed points coincide, no conjugation separates them; the normal form is the translation $\zeta\mapsto\zeta+1$ , with $\lambda=1$ and $\tau=4$ — parabolic, orbits horocycles tangent at the single fixed point. Why: $\tau=4$ is the discriminant boundary where $cz^2+(d-a)z-b=0$ has a double root, the knife-edge between elliptic and hyperbolic.

Limiting cases

Case	Result	Meaning
Affine (fixes $\infty$ )	$c=0:\ f(z)=\frac{a}{d}z+\frac{b}{d}$	rotation-dilation plus translation; no finite pole, $\infty$ is fixed
Pure inversion	$a=d=0,\ b=c=1:\ f(z)=\frac{1}{z}$	the single non-affine generator; swaps $0$ and $\infty$ , turns the plane inside-out
Identity / scalar matrix	$M=\lambda I:\ f(z)=z$	every nonzero scalar matrix is the identity map, so the group is $\mathrm{PGL}(2,\mathbb{C})$ , not $\mathrm{GL}(2,\mathbb{C})$
Degenerate (forbidden)	$ad-bc=0:\ f(z)=\frac{a}{c}$ constant	excluded; the determinant condition is exactly what rules out collapse
Disk automorphism	$f(z)=e^{i\theta}\frac{z-z_0}{1-\bar z_0 z},\ \|z_0\|<1$	the subgroup preserving $\|z\|<1$ ; the isometries of the Poincaré disk

Common trapForgetting the point at infinity breaks the whole structure

Two failures come from treating $f(z)=(az+b)/(cz+d)$ as a map of the finite plane $\mathbb{C}$ alone. First, the determinant condition is not decoration: if $ad-bc=0$ the numerator is a multiple of the denominator and $f$ collapses to the constant $a/c$ — not a transformation at all, which is why every theorem here carries $ad-bc\neq0$ . Second, the map is only a bijection on the sphere $\widehat{\mathbb{C}}$ : the finite pole $z=-d/c$ must map to $\infty$ , and $z=\infty$ must map to $a/c$ (or, when $c=0$ , $\infty$ is fixed). Drop these assignments and the "map" is neither onto nor one-to-one, the group law (3) fails, and the fixed-point count is wrong — an affine map like $f(z)=2z$ looks like it has only the fixed point $0$ until you remember it also fixes $\infty$ . The point at infinity is not a technicality; it makes lines and circles the same object and makes the group a group.

Worked examples

1 · Building a map from three prescribed points (fully worked)

Find the Möbius map sending $0\mapsto1$ , $1\mapsto i$ , and $\infty\mapsto-1$ , and identify the curve the real axis is carried to.

Step 1. Set up the three-point equation (8) with $(z_1,z_2,z_3)=(0,1,\infty)$ and $(w_1,w_2,w_3)=(1,i,-1)$ . The right-hand side simplifies because $z_3=\infty$ : the factors $(z_2-z_3)/(z-z_3)\to1$ , leaving $\dfrac{z-z_1}{z_2-z_1}=\dfrac{z-0}{1-0}=z$ .

Step 2. The left-hand side is $\dfrac{(w-1)(i-(-1))}{(w-(-1))(i-1)}=\dfrac{(w-1)(i+1)}{(w+1)(i-1)}$ . The constant $\dfrac{i+1}{i-1}$ simplifies: multiply top and bottom by $\overline{i-1}=-i-1$ to get $\dfrac{-2i}{2}=-i$ . So the equation is $-i\,\dfrac{w-1}{w+1}=z$ .

Step 3. Solve for $w$ . Divide by $-i$ (note $1/(-i)=i$ ) to get $\dfrac{w-1}{w+1}=iz$ ; cross-multiplying, $w-1=iz(w+1)$ , so $w(1-iz)=1+iz$ and $\boxed{\,w=\dfrac{1+iz}{1-iz}\,}$ , with matrix $\begin{pmatrix}i&1\\-i&1\end{pmatrix}$ , determinant $2i\neq0$ .

Step 4. Check all three: $f(0)=1/1=\mathbf{1}$ ; $f(1)=\dfrac{1+i}{1-i}=\dfrac{(1+i)^2}{2}=\mathbf{i}$ ; and $f(\infty)$ is the ratio of leading coefficients $i/(-i)=\mathbf{-1}$ . All three targets hit.

Step 5. Where does the real axis go? For real $z=x$ , $|1+ix|=|1-ix|=\sqrt{1+x^2}$ , so $|w|=1$ exactly. The real line maps onto the unit circle (and the upper half-plane onto the disk interior) — the standard transform that makes Möbius maps the workhorse of two-dimensional potential theory.

2 · Fixed points and dynamical type (fully worked)

Classify $f(z)=\dfrac{3z-1}{z+1}$ : find its fixed points, its squared trace, and its dynamical type.

Step 1. Read off $a=3,\ b=-1,\ c=1,\ d=1$ , so the determinant is $ad-bc=3\cdot1-(-1)\cdot1=\mathbf{4}\neq0$ — a legitimate Möbius map, and $f'(z)=4/(z+1)^2$ by equation (6).

Step 2. Fixed points solve $cz^2+(d-a)z-b=0$ , i.e. $z^2+(1-3)z-(-1)=z^2-2z+1=(z-1)^2=0$ . The discriminant vanishes, so there is a single, doubled fixed point at $z=1$ — the signature of a parabolic map.

Step 3. Compute the invariant: $\tau=\dfrac{(a+d)^2}{ad-bc}=\dfrac{(3+1)^2}{4}=\dfrac{16}{4}=\mathbf{4}$ . This sits exactly on the parabolic boundary, confirming Step 2 independently.

Step 4. Interpret. A parabolic map is conjugate to a translation $\zeta\mapsto\zeta+\beta$ ; conjugating by $h(z)=1/(z-1)$ (sending $1\mapsto\infty$ ) turns $f$ into a pure shift. Iterating $f$ therefore slides every point along horocycles tangent at $z=1$ , converging on that single fixed point with neither rotation nor net dilation — the parabolic panel of the second figure.

Physics of the portrait

The transformation carries the name of a Leipzig astronomer who reached it through projective geometry, not analysis — one thread in a nineteenth-century weave that also runs through Riemann's sphere, Cayley's matrices, Klein's programme, and Poincaré's automorphic functions.

Origins · 1 minAugust Ferdinand Möbius and how this was discovered

August Ferdinand Möbius

1790–1868 · German mathematician and astronomer, projective and analytic geometry

Möbius spent his whole career at Leipzig, where he had gone to learn Gauss's orbit computations and stayed on as observer and then director of the observatory. His mathematics grew out of astronomy but reached furthest in geometry. The 1827 treatise Der barycentrische Calcül introduced homogeneous "barycentric" coordinates — a point as a weighted average of reference points — and with them the collineations and fractional-linear maps that now bear his name, studied as the transformations preserving the cross-ratio. He is the same Möbius of the one-sided band and of the Möbius function of number theory: a geometer whose instinct was to ask which quantities survive a transformation. The maps were not his alone — Euler and Lagrange had met special cases, and their full group-theoretic meaning arrived later — but it was Möbius who isolated the fractional-linear transformation as an object worth naming. The modern reading, that these are the conformal automorphisms of the Riemann sphere and the elements of $\mathrm{PSL}(2,\mathbb{C})$ , completes a picture whose first clean strokes were his.

History

1827

— Möbius develops the fractional-linear transformations in Der barycentrische Calcül

1851

— Riemann's doctoral thesis founds the theory of conformal maps and the Riemann sphere

1858

— Cayley's memoir on matrices supplies the 2×2 representation of the maps

1872

— Klein's Erlangen Program recasts geometry as the invariants of a transformation group

1882

— Poincaré uses Fuchsian and Kleinian (Möbius) groups for automorphic functions and hyperbolic geometry

Going further

Texts. Ahlfors, Complex Analysis, 3rd ed., Ch. 3 (Linear Transformations) · Needham, Visual Complex Analysis, Ch. 3 (Möbius Transformations) · Stone & Goldbart, Mathematics for Physics, Ch. on complex analysis and conformal mapping · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 11 (Conformal Mapping) · Hassani, Mathematical Physics, chapters on complex analysis · Nakahara, Geometry, Topology and Physics, for the $\mathrm{PSL}(2,\mathbb{C})$ and Riemann-surface context.

Provenance. The definition, the matrix homomorphism (3), the decomposition (5), the conformality relation (6), cross-ratio invariance (7), the three-point/uniqueness theorem (8), circline preservation, and the trace classification (9) are all standard and appear in Ahlfors Ch. 3, Needham Ch. 3, Stone & Goldbart, Arfken–Weber–Harris Ch. 11, and Byron & Fuller. A verifier should recompute: the composite coefficients as entries of $M_2M_1$ ; $f'(z)=(ad-bc)/(cz+d)^2$ ; the inversion substitution turning a circline equation into another; the worked map $w=(1+iz)/(1-iz)$ hitting $1,i,-1$ and carrying $\mathbb{R}$ to the unit circle; and $f(z)=(3z-1)/(z+1)$ giving the double fixed point $z=1$ with $\tau=4$ . No nonstandard numerical constants are asserted.

Prerequisites. Complex Numbers and the Complex Plane · Functions of a Complex Variable and Limits · Analytic Functions and the Cauchy–Riemann Equations · Elementary Complex Functions. Leads to. Harmonic Functions and Conjugates · Liouville, Maximum Modulus, and Morera · Zeros, Poles, and Isolated Singularities.

Ask the tutor: Ask me to classify any specific Möbius map from its matrix, or to build the map that sends three points of your choosing to three targets and watch the cross-ratio do the work.

Why do physicists insist on the point at infinity instead of just working in the plane?

Because without it a Möbius map is simply not a bijection, and the entire clean structure evaporates. The finite pole $z=-d/c$ has no image in $\mathbb{C}$ , and no finite point maps to $a/c$ in the wrong direction — the two exceptional points are exactly $\infty$ and $-d/c$ . Adjoining a single point $\infty$ to form the Riemann sphere $\widehat{\mathbb{C}}$ repairs both gaps at once: the map becomes a smooth bijection of the sphere, straight lines become circles through the north pole, and the group law (composition equals matrix multiplication) holds without exceptions. The point at infinity is what makes the theory geometric rather than a collection of special cases.

In what sense is a straight line the same as a circle?

On the Riemann sphere, exactly the same. Stereographic projection sends every plane circle to a circle on the sphere and every straight line to a sphere-circle through the north pole. So a "line" is just a circle running through $\infty$ ; there is no intrinsic difference. This is why the umbrella word circline is useful, and why "Möbius maps preserve circlines" is cleaner than the plane statement where circles can turn into lines. The distinction between the two shapes is an artefact of removing one point from the sphere.

How can three points possibly determine the entire map?

Count the freedom. A Möbius map has four coefficients $a,b,c,d$ , but only their three independent ratios matter since an overall scale is invisible — three complex parameters. Prescribing three images imposes three complex conditions, matching exactly, and the cross-ratio construction shows a solution always exists. Uniqueness is the elegant part: if two maps agreed on three points, their composite $g\circ h^{-1}$ would fix three points, but a non-identity Möbius map solves a quadratic and fixes at most two. The only escape is $g\circ h^{-1}=\mathrm{id}$ , so $g=h$ . Three points, no freedom left.

What is the difference between PGL(2,C) and PSL(2,C), and does it matter here?

Start from $\mathrm{GL}(2,\mathbb{C})$ , the invertible $2\times2$ complex matrices. Every nonzero scalar matrix acts as the identity map, so quotienting by scalars gives $\mathrm{PGL}(2,\mathbb{C})$ — the Möbius group. Alternatively, normalise to $\det=1$ , landing in $\mathrm{SL}(2,\mathbb{C})$ ; that still leaves the ambiguity $M$ versus $-M$ (same determinant, same map), so quotienting by $\{\pm I\}$ gives $\mathrm{PSL}(2,\mathbb{C})$ . Over $\mathbb{C}$ these coincide, $\mathrm{PGL}(2,\mathbb{C})\cong\mathrm{PSL}(2,\mathbb{C})$ , because every complex number has a square root to absorb into the normalisation. The distinction matters when you track the double cover in physics — that $\pm I$ is the same sign ambiguity that makes $\mathrm{SL}(2,\mathbb{C})$ the spin group of the Lorentz group.

Why does the trace, and not some other combination, classify the dynamics?

Because conjugation is a change of coordinates on the sphere, and the honest invariants of a map are what survive conjugation. For matrices, conjugation preserves the characteristic polynomial, hence trace and determinant. Once you normalise $\det=1$ , only the trace is left, and its square $\tau$ removes the harmless sign — so $\tau$ is essentially the unique conjugation-invariant number a Möbius map carries. It fixes the multiplier $\lambda$ through $\tau=\lambda+2+\lambda^{-1}$ , and $\lambda$ governs whether orbits rotate ( $|\lambda|=1$ ), dilate ( $\lambda>0$ real), or spiral (complex $\lambda$ ). The trace classifies the dynamics because it is the one number the dynamics cannot hide.

Where do Mobius transformations actually show up in a physics problem I might solve?

Most concretely in two-dimensional potential theory. Laplace's equation is conformally invariant in the plane, so a harmonic potential stays harmonic under a Möbius map, and you exploit this to move a hard boundary to an easy one. The map $w=(z-i)/(z+i)$ carries the upper half-plane to the unit disk, converting a grounded-plane electrostatics or ideal-flow problem into a disk problem with a known Poisson-kernel solution, then maps the answer back — field lines and equipotentials keep their right-angle crossings. Beyond electrostatics, the disk automorphisms are the rigid motions of the Poincaré model of hyperbolic geometry, and $\mathrm{SL}(2,\mathbb{C})$ on the celestial sphere is how a Lorentz boost rearranges incoming starlight — stellar aberration is a Möbius flow on the sky.