Hyperbolic Geometry of Boosts

Write down the matrix that rotates the plane by an angle $\theta$ , and the matrix that boosts spacetime by a rapidity $\varphi$ . Set them side by side. They are the same object with one sign changed — cosines become hyperbolic cosines, sines become hyperbolic sines, and a single minus sign migrates from one corner to another. That is the entire difference between turning your head and changing your speed. This is not a coincidence of notation, and it is worth stating precisely, because it is the deepest structural fact in special relativity and it fits on one line. Every continuous transformation is generated by a matrix $G$ — the thing you exponentiate. For a rotation the generator satisfies $J^2 = -\mathbb{1}$ , and the exponential series separates into the two series for $\cos$ and $\sin$ . For a boost the generator satisfies $K^2 = +\mathbb{1}$ , and the very same series separates into $\cosh$ and $\sinh$ . One sign. And that sign is the metric's: the minus in $\eta = \mathrm{diag}(+1,-1,-1,-1)$ is what forces the boost generator to be symmetric rather than antisymmetric, and a symmetric generator is one whose square is $+\mathbb{1}$ . Circles become hyperbolae. That is all that happened to geometry in 1905. Everything else — time dilation, the impossibility of $c$ , the failure of velocities to add — follows from drawing a different conic.

Key takeaways

Signature $(+,-,-,-)$ : $\Delta s^2 = c^2\Delta t^2 - \Delta x^2$ ; timelike $s^2 > 0$ , spacelike $s^2 < 0$ . (The AllQFT base for this slug is mixed and writes "spacelike, $s^2 = x^2 - (ct)^2 > 0$ " — the other convention. Every sign here is re-derived.)
A boost is $\Lambda(\varphi) = \begin{pmatrix}\cosh\varphi & \sinh\varphi \\ \sinh\varphi & \cosh\varphi\end{pmatrix}$ acting on $(ct, x)$ — a hyperbolic rotation. The curves $c^2t^2 - x^2 = \text{const}$ are the invariant hyperbolae: the circles of Minkowski geometry.
Marquee result: $K = \left.\dfrac{d\Lambda}{d\varphi}\right|_0 = \begin{pmatrix}0&1\\1&0\end{pmatrix}$ satisfies $K^2 = +\mathbb{1}$ , so $\exp(\varphi K) = \mathbb{1}\cosh\varphi + K\sinh\varphi$ — Euler's formula with the sign flipped. The Euclidean generator has $J^2 = -\mathbb{1}$ and gives $\mathbb{1}\cos\theta + J\sin\theta$ . The sign of $G^2$ is the metric signature, and it is the whole difference between a rotation and a boost.
Galileo completes the trichotomy: his generator is nilpotent, $N^2 = 0$ , giving $\exp(vN) = \mathbb{1} + vN$ — a linear shear, with no speed limit.
Second owned result: boosts in different directions do not commute. $[K_1, K_2] = J_3$ — antisymmetric, a rotation generator. Boost twice, in two directions, and you come out rotated.
Key idea: exponentiate the generator. The metric fixes its symmetry, the symmetry fixes the sign of its square, and that sign decides whether you get circles or hyperbolae.

The core insight

Every one-parameter family of transformations starting at the identity is $\exp(\lambda G)$ for a single generator $G$ . You never need to guess a transformation law — differentiate at the identity, find $G$ , exponentiate. Sophus Lie built an entire theory on this decades before physics needed it, and a Lorentz boost is one of its simplest instances: a one-parameter subgroup whose canonical parameter is the rapidity.

Where this lives today

The exponential map is how modern physics writes down symmetries. You do not postulate a transformation law; you postulate a generator, exponentiate, and see what you get. Rotations come from $J$ , boosts from $K$ , gauge transformations from the generators of $SU(3)\times SU(2)\times U(1)$ — and in every case the group's structure is encoded in the commutators, precisely as $[K_1,K_2] = J_3$ encodes the fact that boosts drag rotations behind them. That commutator has a measurable consequence. An electron in an atom is repeatedly boosted in changing directions, and the rotations left behind accumulate into Thomas precession — whose factor of one half in the spin-orbit coupling was a crisis in atomic spectroscopy until Llewellyn Thomas resolved it in 1926. It is the difference between matching the fine structure of hydrogen and being wrong by a factor of two. The magnitude belongs to Thomas Precession; what this article establishes is why the effect must exist at all. In high-energy physics the hyperbolic picture is the working language. Boosts along the beam axis form the one-parameter group $\exp(\varphi K_3)$ and shift every particle's rapidity by the same constant — which is why detectors are segmented uniformly in rapidity, and why the hyperbolae Minkowski drew in 1908 as a pedagogical picture are now the geometry inside which every LHC collision is analysed.

The physical picture

Take the Euclidean plane. The transformations preserving $x^2 + y^2$ are rotations. Sweep $\theta$ and a point traces the circle $x^2 + y^2 = r^2$ : the circle is the orbit, the set of places the symmetry can send that point. Every point sits on exactly one orbit, and $r^2$ labels it. Now take Minkowski's plane, with $s^2 = c^2t^2 - x^2$ . The transformations preserving $s^2$ are boosts. Sweep the rapidity $\varphi$ and an event traces the hyperbola $c^2t^2 - x^2 = s^2$ . The hyperbola is the orbit; every event sits on exactly one; $s^2$ labels it. Same picture, one difference that changes everything: a circle is closed and bounded, and a hyperbola is neither. Rotate far enough and you come back where you started; boost far enough and you go on forever, sliding along a branch that approaches the light cone asymptotically and never touches it. That topological fact — the orbit is unbounded but never crosses its asymptote — is the speed limit. It is not an extra postulate bolted onto the geometry. It is what a hyperbola looks like. And the light cone is the degenerate orbit, $c^2t^2 - x^2 = 0$ : the "circle of radius zero", except a pair of lines rather than a point, along which a boost slides events without ever moving them off. The invariance of $c$ is the statement that the degenerate orbit is a genuine orbit.

Interactive Simulation

Boost as a Hyperbolic Rotation — drag β and watch the 4-momentum P slide along the invariant hyperbola E²−p²=const while the axes scissor toward the light cone (not a circle: length in the Euclidean sense is not preserved).

Figure

PUNCHLINE — one exponential, three geometries. Sweep the SAME parameter through the SAME formula exp(λG) and the sign of G² decides the shape. EUCLID: J² = −1, orbits are CIRCLES; at θ = ln 2 = 0.6931 the point is (0.7692, 0.6390), with x² + y² = 1. MINKOWSKI: K² = +1, orbits are HYPERBOLAE; at φ = ln 2 the point is (1.25, 0.75) — γ and γβ for β = 0.6 — with x² − y² = 1. GALILEO: N² = 0, orbits are straight lines, no bound and no light cone. The sign of G² is the metric

The core equations

The boost, written as a hyperbolic rotation, is the object this whole article is about.

\Lambda(\varphi) = \begin{pmatrix}\cosh\varphi & \sinh\varphi \\ \sinh\varphi & \cosh\varphi\end{pmatrix}, \qquad \Lambda^{\mathsf T}\eta\,\Lambda = \eta, \qquad \det\Lambda = \cosh^2\varphi - \sinh^2\varphi = 1 \qquad (1)

where: $\eta = \mathrm{diag}(+1,-1)$ on $(ct, x)$ ; $\beta = \tanh\varphi$ . Physical reading: this is the active boost — it carries a particle at rest to rapidity $\varphi$ , taking $(1,0)$ to $(\cosh\varphi, \sinh\varphi)$ . The passive transformation, into a frame moving at $+\beta c$ , is its inverse $\Lambda(-\varphi)$ , which is where the familiar minus signs come from. Both are Lorentz transformations; only the direction of the sweep differs. The generator is what you get by differentiating at the identity, and it is the whole story in a $2\times2$ box.

K \;\equiv\; \left.\frac{d\Lambda}{d\varphi}\right|_{\varphi=0} = \begin{pmatrix}0&1\\1&0\end{pmatrix}, \qquad\qquad K^2 = +\mathbb{1} \qquad (2)

where: $K$ is the boost generator of $\mathfrak{so}(1,3)$ restricted to the $t$ – $x$ plane. Physical reading: $K$ is symmetric, and that is why its square is $+\mathbb{1}$ . Contrast the Euclidean rotation generator $J = \begin{pmatrix}0&-1\\1&0\end{pmatrix}$ , which is antisymmetric and squares to $-\mathbb{1}$ . The generator's symmetry is dictated by the metric, as we will prove; nothing else in this article matters as much. Exponentiating a generator whose square is $\pm\mathbb{1}$ splits the exponential series into two halves, and which two halves you get is decided by that sign:

\exp(\varphi K) = \mathbb{1}\cosh\varphi + K\sinh\varphi, \qquad\qquad \exp(\theta J) = \mathbb{1}\cos\theta + J\sin\theta \qquad (3)

where: both follow from $\exp(A) = \sum A^n/n!$ by collecting even and odd powers. Physical reading: the right-hand identity is Euler's formula in matrix form, and the left-hand one is Euler's formula with the sign flipped. Since $K^2 = +\mathbb{1}$ , even powers give $\mathbb{1}$ and odd powers give $K$ , so the two series are $\cosh\varphi$ and $\sinh\varphi$ . Since $J^2 = -\mathbb{1}$ , the same collection produces alternating signs, hence $\cos$ and $\sin$ . One sign in the generator's square; two entirely different geometries. And the orbits — the curves swept out — are the conics that give the two geometries their names:

c^2t^2 - x^2 = \text{const} \quad\text{(Minkowski: hyperbolae)}, \qquad\qquad x^2 + y^2 = \text{const} \quad\text{(Euclid: circles)} \qquad (4)

where: const $> 0$ is timelike, $< 0$ spacelike, $= 0$ null. Physical reading: the invariant hyperbolae are the circles of Minkowski geometry — the events "the same interval away" from the origin, what a compass would draw if a compass measured $s^2$ . The null case is degenerate: the light cone is the pair of asymptotes, an orbit of "radius" zero that a boost slides along but never leaves.

The geometry, derived

Derivation AThe invariant hyperbolae are the circles of Minkowski geometry, and rapidity is their arc length

In Euclidean geometry the circle is not merely a shape you draw; it is the orbit of the rotation group, and the angle is not a label on it but the arc length measured along it. Both statements have exact Minkowski counterparts, and getting them straight is what turns a set of formulas into a geometry — what lets you draw special relativity instead of computing it.

Step 1. Establish the orbits. Apply $\Lambda(\varphi)$ to an event: $ct' = ct\cosh\varphi + x\sinh\varphi$ , $x' = ct\sinh\varphi + x\cosh\varphi$ , so $(ct')^2 - (x')^2 = (c^2t^2 - x^2)(\cosh^2\varphi - \sinh^2\varphi) = c^2t^2 - x^2$ once the cross terms cancel. A boost moves an event along $c^2t^2 - x^2 = \text{const}$ and cannot move it off. Why: this is the definition of an orbit. The hyperbolae are to boosts what circles are to rotations, and $s^2$ is the "radius squared" labelling them.

Step 2. Sweep the unit hyperbola. Start at $(1, 0)$ and apply $\Lambda(\varphi)$ : you land on $(\cosh\varphi,\ \sinh\varphi)$ , and as $\varphi$ runs over $\mathbb{R}$ this traces the entire upper branch of $c^2t^2 - x^2 = 1$ . Boosting the spacelike unit event $(0,1)$ gives $(\sinh\varphi,\ \cosh\varphi)$ , sweeping $c^2t^2 - x^2 = -1$ . Why: the boost is the hyperbola's natural parametrisation, as the rotation is the circle's. Note the signs: the timelike unit hyperbola is $s^2 = +1$ and the spacelike one $s^2 = -1$ — the base prints the opposite.

Step 3. Calibrate the tilted axes. The boosted observer's time axis is $x = \beta\,ct$ ; her space axis is $ct = \beta x$ . Where along the tilted time axis is her one-second tick? Where it crosses the unit hyperbola: substituting $x = \beta\,ct$ into $c^2t^2 - x^2 = 1$ gives $c^2t^2(1-\beta^2) = 1$ , hence $ct = \gamma = \cosh\varphi$ and $x = \gamma\beta = \sinh\varphi$ . Why: this settles the classic complaint that the tilted axes "look wrongly scaled". They are scaled by $\gamma$ , and the hyperbola is what tells you so. You calibrate a Minkowski diagram with a hyperbola, never with a ruler.

Step 4. Put a number on it. At $\beta = 0.6$ , so $\varphi = \ln 2$ , the tick on the tilted $ct'$ -axis sits at $(1.25,\ 0.75)$ in the lab, and the tick on the tilted $x'$ -axis at $(0.75,\ 1.25)$ . Check: $1.25^2 - 0.75^2 = 1$ and $0.75^2 - 1.25^2 = -1$ . Why: those two numbers are $\gamma$ and $\gamma\beta$ . The calibration tick and the Lorentz factor are the same thing seen twice.

Step 5. Identify the arc length. Parametrise the unit hyperbola as $C(\varphi) = (\cosh\varphi,\ \sinh\varphi)$ and compute its Minkowski line element: $ds^2 = d(ct)^2 - dx^2 = (\sinh^2\varphi - \cosh^2\varphi)\,d\varphi^2 = -d\varphi^2$ . It is spacelike, so $|ds| = |d\varphi|$ , and the arc length from $(1,0)$ out to $C(\varphi)$ is exactly $\varphi$ . Why: this is the theorem that names the rapidity. Rapidity is the Minkowski arc length along the unit hyperbola, precisely as angle is the Euclidean arc length along the unit circle — the identical computation with one sign changed. Rapidity is not "like" an angle. It is an angle, measured with the metric we actually live in. *Verified symbolically in verify_aq-sr-hyperbolic-geometry-of-boosts.py: $(ct')^2 - (x')^2 = (ct)^2 - x^2$ for symbolic events; $\Lambda(\varphi)\cdot(1,0) = (\cosh\varphi, \sinh\varphi)$ and $\Lambda(\varphi)\cdot(0,1) = (\sinh\varphi,\cosh\varphi)$ ; the $ct'$ -axis meets the unit hyperbola at $ct = \gamma = \cosh\varphi$ ; the $\beta = 0.6$ ticks are $(1.25, 0.75)$ and $(0.75, 1.25)$ exactly; and $ds^2 = -d\varphi^2$ on the unit hyperbola, versus $ds^2 = +d\theta^2$ on the unit circle.*

Go deeper · Derivation B · 2 minThe generator, the exponential map, and the one sign that separates a boost from a rotation

Derivation BThe generator, the exponential map, and the one sign that separates a boost from a rotation

This is the marquee result of the article and, properly understood, of the whole subject. A boost and a rotation are the same construction run with a different metric, and every visible difference between them — hyperbolic versus trigonometric functions, unbounded versus periodic, a speed limit versus none — descends from a single sign: the sign of the generator's square. That sign is not free. The metric fixes it.

Step 1. Find the generator. A one-parameter family starting at the identity is determined by its derivative there. Differentiating $\Lambda(\varphi)$ at $\varphi = 0$ — with $\tfrac{d}{d\varphi}\cosh\varphi \to 0$ and $\tfrac{d}{d\varphi}\sinh\varphi \to 1$ — gives $K = \begin{pmatrix}0&1\\1&0\end{pmatrix}$ . Why: this is Lie's move. The infinitesimal transformation is the whole transformation, compressed; everything else is recovered by exponentiating.

Step 2. Square it. $K^2 = +\mathbb{1}$ . Why: this is the sentence the article exists for. Hold onto it.

Step 3. Exponentiate, and watch the series split. $\exp(\varphi K) = \sum_{n\ge0}\frac{\varphi^n}{n!}K^n$ . Because $K^2 = \mathbb{1}$ , every even power is $\mathbb{1}$ and every odd power is $K$ :

\exp(\varphi K) = \mathbb{1}\sum_{n}\frac{\varphi^{2n}}{(2n)!} \;+\; K\sum_{n}\frac{\varphi^{2n+1}}{(2n+1)!} = \boxed{\mathbb{1}\cosh\varphi + K\sinh\varphi}

which written out is exactly $\Lambda(\varphi)$ . Why: the hyperbolic functions were not assumed anywhere. They fell out, because the even and odd sums with no alternating signs are by definition $\cosh$ and $\sinh$ .

Step 4. Do it again for Euclid, and watch the sign do its work. The rotation generator $J = \begin{pmatrix}0&-1\\1&0\end{pmatrix}$ is antisymmetric, and $J^2 = -\mathbb{1}$ . Now the powers alternate: $J^{2n} = (-1)^n\mathbb{1}$ , $J^{2n+1} = (-1)^nJ$ . The identical collection gives $\exp(\theta J) = \mathbb{1}\cos\theta + J\sin\theta$ — Euler's formula, and the ordinary rotation matrix. Why: the two derivations are the same derivation. The only input that differed was one sign in $G^2$ , and out came circular functions instead of hyperbolic ones. Everything downstream — periodicity, boundedness, the speed limit — is that sign propagating.

Step 5. Show the metric chooses the sign. A generator preserving a metric $g$ must satisfy $G^{\mathsf T}g + gG = 0$ (differentiate $\Lambda^{\mathsf T}g\Lambda = g$ at the identity). Solve it. For $g = \mathbb{1}$ (Euclid) this reads $G^{\mathsf T} + G = 0$ : $G$ is antisymmetric, so $G \propto J$ and $G^2 \propto -\mathbb{1}$ . For $g = \eta = \mathrm{diag}(1,-1)$ , writing $G = \begin{pmatrix}a&b\\c&d\end{pmatrix}$ gives $\begin{pmatrix}2a & b-c\\ b-c & -2d\end{pmatrix} = 0$ , forcing $a = d = 0$ and $c = b$ : $G$ is symmetric, so $G \propto K$ and $G^2 \propto +\mathbb{1}$ . Why: $\boxed{\text{the sign of } G^2 \text{ is the signature of } g}$ . You do not choose hyperbolic functions; the minus sign in the metric chooses them for you, by forcing the generator to be symmetric. That is "spacetime is not Euclidean", as an equation.

Step 6. Complete the trichotomy. Galileo's boost, $t \to t$ , $x \to x + vt$ , has generator $N = \begin{pmatrix}0&0\\1&0\end{pmatrix}$ , which is nilpotent: $N^2 = 0$ . The series terminates: $\exp(vN) = \mathbb{1} + vN$ , a linear shear. Three signs of $G^2$ — $-1$ , $+1$ , $0$ — give three geometries: elliptic (circles), hyperbolic (a light cone), parabolic (absolute time, no speed limit). Why: Newtonian mechanics is the degenerate member of the family, the one whose metric is degenerate too. Relativity did not add complexity; it removed a degeneracy. *Verified symbolically in verify_aq-sr-hyperbolic-geometry-of-boosts.py: $K = d\Lambda/d\varphi|_0$ and $K^2 = +\mathbb{1}$ ; $\exp(\varphi K) = \mathbb{1}\cosh\varphi + K\sinh\varphi$ , by sympy's matrix exponential and independently by summing the series to 24 terms; $J^2 = -\mathbb{1}$ with $\exp(\theta J) = \mathbb{1}\cos\theta + J\sin\theta$ ; $N^2 = 0$ with $\exp(vN) = \mathbb{1} + vN$ ; and, crucially, solving $G^{\mathsf T}g + gG = 0$ returns a symmetric $G \propto K$ for $g = \eta$ and an antisymmetric $G \propto J$ for $g = \mathbb{1}$ — so the metric really does select the sign.*

Go deeper · Derivation C · 2 minTwo boosts in two directions leave you rotated

Derivation CTwo boosts in two directions leave you rotated

Along a single axis, boosts form a one-parameter group and everything commutes: $\Lambda(\varphi_1)\Lambda(\varphi_2) = \Lambda(\varphi_1+\varphi_2) = \Lambda(\varphi_2)\Lambda(\varphi_1)$ , which is why rapidities add. Use two different directions and that abelian world ends — and it ends with a striking physical consequence. The commutator of two boost generators is not a boost. It is a rotation.

Step 1. Write the boost generators in four-dimensional form. Let $E(i,j)$ be the matrix with a $1$ in row $i$ , column $j$ . Then $K_1 = E(0,1) + E(1,0)$ generates boosts along $x$ , and $K_2 = E(0,2) + E(2,0)$ boosts along $y$ . Both are symmetric, as Step 5 above requires, and both satisfy $G^{\mathsf T}\eta + \eta G = 0$ . Why: a boost mixes time with one spatial direction, so its generator lives in row/column $0$ and the matching spatial slot.

Step 2. Multiply, in both orders. $K_1K_2 = E(1,2)$ , from $E(1,0)\cdot E(0,2)$ ; while $K_2K_1 = E(2,1)$ , from $E(2,0)\cdot E(0,1)$ . Different matrices. Why: boosting along $x$ then $y$ is not the same operation as boosting along $y$ then $x$ .

Step 3. Take the commutator and look at what it is. $[K_1, K_2] = E(1,2) - E(2,1) = J_3$ , the generator of rotations in the $x$ – $y$ plane. It is antisymmetric, it has no entries in row or column $0$ , and by the criterion just established — symmetric means boost, antisymmetric means rotation — it is unambiguously $\boxed{\text{a rotation generator}}$ . Why: the failure of two boosts to commute is not a residue or an error term. It is exactly a rotation, and the algebra says so with no room for interpretation.

Step 4. Read the physics, and stop there. Two successive boosts in different directions are a single boost combined with a rotation — one that nobody applied and nobody asked for, generated purely by the geometry. This is the origin of the Thomas–Wigner rotation and of Thomas precession. Why: the algebra proves the rotation is there and identifies its generator. Its magnitude is a separate calculation, belonging to Thomas Precession, and is deliberately not attempted here.

Step 5. Note the structural moral. The commutator of two boosts is a rotation, so the boosts alone do not close into a group. You cannot have the boosts without the rotations — which is why the Lorentz group is six-dimensional and not three. Why: this is the algebraic reason rotations are compulsory in relativity rather than optional extras. *Verified symbolically in verify_aq-sr-hyperbolic-geometry-of-boosts.py: $K_1, K_2, K_3, J_3$ all satisfy $G^{\mathsf T}\eta + \eta G = 0$ ; the $K_i$ are symmetric and $J_3$ antisymmetric; $[K_1, K_2] = J_3$ exactly; the commutator is antisymmetric, is not symmetric (hence not a boost generator), has zero row and column $0$ , and exponentiates to a matrix preserving $\eta$ with $\det = 1$ . The Thomas–Wigner angle itself is asserted nowhere.*

Figure

The invariant hyperbolae are the circles of Minkowski geometry — and they are how you calibrate a spacetime diagram. Boost by β = 0.6 (φ = ln 2). The tilted ct′-axis is x = 0.6 ct; its ONE-SECOND TICK sits where it crosses the unit hyperbola (ct)² − x² = 1, at (1.25, 0.75) — exactly (γ, γβ). The tilted x′-axis is ct = 0.6 x; its one-metre tick sits at (0.75, 1.25), on the s² = −1 hyperbola. You calibrate tilted axes with a hyperbola, never with a ruler

Limiting cases

Case	Result	Meaning
$\varphi$ small	$\Lambda \approx \mathbb{1} + \varphi K$	the Galilean shear — the tangent to the hyperbola at the identity
$G^2 = -\mathbb{1}$	$\cos$ , $\sin$ ; circles	Euclid. Bounded, periodic, no cone and no speed limit
$G^2 = +\mathbb{1}$	$\cosh$ , $\sinh$ ; hyperbolae	Minkowski. Unbounded orbits that never reach the asymptote
$G^2 = 0$	$\mathbb{1} + vN$ ; straight lines	Galileo. Absolute time, no invariant speed
$\varphi$ large	$\cosh\varphi \approx \sinh\varphi \approx e^{\varphi}/2$	the orbit hugs the light cone, asymptotically and forever
Null direction	eigenvector of $K$ , eigenvalue $\pm1$	a boost rescales $ct \pm x$ by $e^{\pm\varphi}$ ; at $\varphi = \ln 2$ , by $2$ and $\tfrac12$
Two directions	$[K_1,K_2] = J_3 \ne 0$	boosts do not commute, and their failure to commute is a rotation

Common trapWhich sign convention is this boost matrix in?

$\Lambda(\varphi) = \exp(\varphi K)$ has $+\sinh\varphi$ off the diagonal. This is the active boost: it gives an object at rest the rapidity $\varphi$ , carrying $(1,0) \mapsto (\cosh\varphi, \sinh\varphi)$ . The passive transformation — re-expressing an event's coordinates in a frame moving at $+\beta c$ , and carrying the familiar $-\gamma\beta$ entries — is the inverse, $\exp(-\varphi K)$ . Both are Lorentz transformations of the same rapidity; they differ only in whether you move the object or move yourself. Every result here ( $K^2 = +\mathbb{1}$ , the exponential, the group law, the commutator) is unchanged under $\varphi \to -\varphi$ , because $(-K)^2 = K^2$ .

Common trapThe base for this slug has the spacelike sign backwards

This corpus uses $(+,-,-,-)$ : $\Delta s^2 = c^2\Delta t^2 - \Delta x^2$ , with timelike $s^2 > 0$ and spacelike $s^2 < 0$ . The AllQFT base for this slug is signature-mixed: one of its derivations preserves $s^2 = (ct)^2 - x^2$ , which is ours, while an equation card writes "spacelike, $s^2 = x^2 - (ct)^2 > 0$ ", which is the other convention's. Both cannot be right in one article, and the second is wrong here: a spacelike separation has negative $s^2$ . The timelike unit hyperbola is therefore $s^2 = +1$ and the spacelike one $s^2 = -1$ , not the reverse — and every sign here has been re-derived rather than transcribed.

Worked examples

1 · Calibrating a spacetime diagram

Draw the axes of an observer moving at $\beta = 0.6$ on a lab-frame spacetime diagram, and mark her one-second tick. Where exactly does it go?

Step 1 — identify. The rapidity is $\varphi = \operatorname{artanh}(0.6) = \ln 2$ , so $\cosh\varphi = 1.25$ and $\sinh\varphi = 0.75$ . Her time axis is her own worldline, $x = 0.6\,ct$ ; her space axis is her line of simultaneity, $ct = 0.6\,x$ . Both tilt toward the light cone by equal angles from opposite sides — which is why the cone stays fixed.

Step 2 — apply the calibration rule. The tick is not one unit of lab length along the tilted axis. It is the event her clock labels $t' = 1$ , i.e. $(ct', x') = (1, 0)$ in her coordinates. Transform it into the lab: $\Lambda(\ln 2)\cdot(1,0) = (1.25,\ 0.75)$ .

Step 3 — read the boundary number. The tick sits at lab time $1.25\ \mathrm{s}$ , having travelled $0.75$ light-seconds. Check it lies on the unit hyperbola: $1.25^2 - 0.75^2 = 1$ . ✓ Of course it does — the boost cannot move an event off its orbit, and $(1,0)$ started on $s^2 = 1$ .

Step 4 — the same for her ruler. Her one-metre tick is the event $(0,1)$ in her frame, mapping to $(0.75,\ 1.25)$ , which lies on $0.75^2 - 1.25^2 = -1$ , the spacelike unit hyperbola. ✓

Step 5 — the moral. Her tick is further from the origin on the page than the lab's tick at $(1,0)$ , and the stretch factor is $\gamma = 1.25$ . This is not a drawing artefact to be corrected. On a Minkowski diagram the compass draws hyperbolae, so distances on the page mean nothing and only the hyperbolae calibrate. Every "whose ruler is shorter" paradox comes from reading a Minkowski diagram with a Euclidean eye.

2 · Exponentiating the generator by hand

Compute $\exp(\varphi K)$ from the series alone, with no prior knowledge that the answer involves hyperbolic functions, and check it at $\varphi = \ln 2$ .

Step 1 — identify. $K = \begin{pmatrix}0&1\\1&0\end{pmatrix}$ , and $\exp(\varphi K) \equiv \sum_{n\ge0}\dfrac{\varphi^n}{n!}K^n$ . Everything hinges on the powers of $K$ .

Step 2 — compute the powers. $K^2 = \mathbb{1}$ , hence $K^3 = K$ , $K^4 = \mathbb{1}$ , and so on: every even power is $\mathbb{1}$ and every odd power is $K$ . There are only two distinct matrices in the entire series.

Step 3 — split the sum. \exp(\varphi K) = \mathbb{1}\left(1 + \frac{\varphi^2}{2!} + \frac{\varphi^4}{4!} + \cdots\right) + K\left(\varphi + \frac{\varphi^3}{3!} + \frac{\varphi^5}{5!} + \cdots\right). The first bracket is the definition of $\cosh\varphi$ , the second of $\sinh\varphi$ . Hence $\exp(\varphi K) = \mathbb{1}\cosh\varphi + K\sinh\varphi = \begin{pmatrix}\cosh\varphi & \sinh\varphi\\ \sinh\varphi & \cosh\varphi\end{pmatrix}$ — the Lorentz boost, from nothing but a $2\times2$ matrix and the exponential series.

Step 4 — evaluate at the anchor. At $\varphi = \ln 2$ : $e^{\varphi} = 2$ , $e^{-\varphi} = \tfrac12$ , so $\cosh\varphi = \tfrac54$ and $\sinh\varphi = \tfrac34$ . Therefore \exp\big((\ln 2)K\big) = \begin{pmatrix}5/4 & 3/4 \\ 3/4 & 5/4\end{pmatrix}, with $\det = 25/16 - 9/16 = 1$ . ✓ And $\beta = \tanh(\ln 2) = (3/4)/(5/4) = 3/5$ , as advertised.

Step 5 — the boundary number. Now run the same series with $J$ instead of $K$ . The powers alternate ( $J^2 = -\mathbb{1}$ , $J^3 = -J$ , $J^4 = +\mathbb{1}$ ), so the brackets become $\cos\theta$ and $\sin\theta$ , and you get a rotation. At $\theta = \ln 2 = 0.693147$ it carries $(1,0)$ to $(0.7692,\ 0.6390)$ — a point on the unit circle, since $0.7692^2 + 0.6390^2 = 1$ . Same parameter, same series, same construction; one sign different in $G^2$ , and the orbit is a circle instead of a hyperbola. That sign is the metric, and the metric is the physics.

Physics of the portrait

The mathematics of one-parameter groups — generators, exponentials, commutators, the whole apparatus this article runs on — was built by a Norwegian who died six years before special relativity was published and who never gave a thought to the propagation of light.

Origins · 1 minSophus Lie and how this was discovered

Sophus Lie

1842–1899 · Nordfjordeid · Christiania · Leipzig

Marius Sophus Lie set out to do for differential equations what Galois had done for polynomials: to understand their solvability through the symmetries they possess. To do it he had to invent the symmetries — the theory of continuous transformation groups, now called Lie groups, and their infinitesimal generators, now called Lie algebras. The central idea is the one this article uses: a continuous symmetry is generated by an infinitesimal one, and the finite transformation is recovered by exponentiating the generator. In 1886 he succeeded Felix Klein in the chair of geometry at Leipzig, and over nine years with Friedrich Engel produced the three-volume Theorie der Transformationsgruppen (1888–1893). Lie died in Kristiania in 1899, six years before Einstein's paper; he did not work on relativity and never saw it. But a Lorentz boost is a one-parameter subgroup $\exp(\varphi K)$ of precisely the kind he classified; rapidity is its canonical parameter; and $[K_1, K_2] = J_3$ is a bracket in the Lie algebra he defined. He built the machinery. Physics arrived six years late and found it waiting.

History

1757

— Vincenzo Riccati defines the hyperbolic functions from the geometry of the unit hyperbola.

1872

— Felix Klein's Erlangen Programme: a geometry is the invariant theory of a transformation group.

1888

— Lie and Engel begin publishing Theorie der Transformationsgruppen: continuous groups and their infinitesimal generators.

1899

— Sophus Lie dies in Kristiania, six years before special relativity is published.

1908

— Minkowski's Cologne lecture recasts relativity as a geometry with an indefinite metric, in which boosts are hyperbolic rotations.

1910

— Vladimir Varičak reads special relativity as Bolyai-Lobachevsky geometry.

1926

— Llewellyn Thomas explains the factor of one half in spin-orbit coupling by the rotation that two non-collinear boosts leave behind.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§2, 9 · Naber, The Geometry of Minkowski Spacetime, ch. 1 · Hall, Lie Groups, Lie Algebras, and Representations, ch. 1–3 · Lie & Engel, Theorie der Transformationsgruppen (Teubner, 1888–1893) · Thomas, "The Kinematics of an Electron with an Axis", Phil. Mag. 3, 1 (1927).

Provenance. Signature $(+,-,-,-)$ ; the AllQFT base for this slug is signature-mixed, its spacelike sign is inverted relative to ours, and its three derivations belong to other slugs — the content here is authored fresh. All results recomputed in verify_aq-sr-hyperbolic-geometry-of-boosts.py (72 checks): $\Lambda^{\mathsf T}\eta\Lambda = \eta$ and $\det\Lambda = 1$ ; the invariance of every hyperbola; the $\beta = 0.6$ calibration ticks $(1.25, 0.75)$ and $(0.75, 1.25)$ ; $ds^2 = -d\varphi^2$ on the unit hyperbola; $K^2 = +\mathbb{1}$ and $\exp(\varphi K) = \mathbb{1}\cosh\varphi + K\sinh\varphi$ , both by matrix exponential and by direct series summation; $J^2 = -\mathbb{1}$ , $N^2 = 0$ ; the generator's symmetry from $G^{\mathsf T}g + gG = 0$ ; and $[K_1, K_2] = J_3$ .

Scope. The magnitude of the Thomas–Wigner rotation and the Rindler geometry of the hyperbolic worldline are forward pointers, derived nowhere here — see Thomas Precession and Relativistic Acceleration. See also: aq-sr-lorentz-boosts-hyperbolic-rotations.

Builds on. The Minkowski Metric · The Spacetime Interval · Lorentz Invariance. Leads to. Rapidity · Four-Vectors · Spacetime Diagrams.

Is a boost literally a rotation through an imaginary angle?

You can make that work formally — substitute $\theta = i\varphi$ into $\cos$ and $\sin$ and you get $\cosh$ and $\sinh$ — and Minkowski himself used the device, writing $x^4 = ict$ to make the metric look Euclidean. But it is a trick, and it hides the real statement, which needs no imaginary numbers: the generator is symmetric rather than antisymmetric, so its square is $+\mathbb{1}$ , and the series splits into hyperbolic rather than circular functions. The $ict$ formalism fell out of use because it obscures the geometry, and because it breaks down in curved spacetime, where no change of variable makes the metric Euclidean.

If the invariant hyperbolae are the circles of Minkowski geometry, what plays the role of π?

Nothing does, and the absence is instructive. $\pi$ exists because the circle is closed: go round by $2\pi$ and you return, so the rotation group is compact and its parameter periodic. The hyperbola is open, the boost group non-compact, its parameter running over all of $\mathbb{R}$ — there is no rapidity at which you come back to where you started. That is not cosmetic: the compactness of the rotation group is why angular momentum is quantised in integer steps, and the non-compactness of the boost group is why the finite-dimensional representations of the Lorentz group are not unitary.

Why must the boost generator be symmetric? Where does that come from?

From the metric, and nothing else. Any generator preserving $g$ satisfies $G^{\mathsf T}g + gG = 0$ , obtained by differentiating $\Lambda^{\mathsf T}g\Lambda = g$ at the identity. If $g = \mathbb{1}$ this says $G^{\mathsf T} = -G$ : antisymmetric. If $g = \eta$ , the $-1$ flips the sign of the mixed entries and the condition instead forces the time-space block to be symmetric. The metric's minus sign is the whole input. Change it and you change the generator's symmetry, hence the sign of $G^2$ , hence the entire geometry.

Do rapidities still add if the boosts are in different directions?

No. Additivity is a statement about a single one-parameter subgroup, hence a single direction. In different directions the generators do not commute — $[K_1,K_2] = J_3 \ne 0$ — so $\exp(\varphi_1 K_1)\exp(\varphi_2 K_2) \ne \exp(\varphi_1K_1 + \varphi_2K_2)$ , and the product is not even a pure boost: it is a boost times a rotation. Treating rapidity as a vector that adds componentwise is a common error, and the discrepancy is the Thomas–Wigner rotation.

What happens to the light cone under a boost?

It is fixed — but as a set, not pointwise. The null directions $ct \pm x$ are the eigenvectors of $K$ , with eigenvalues $\pm1$ , so a boost rescales them: $ct + x \to e^{\varphi}(ct+x)$ and $ct - x \to e^{-\varphi}(ct-x)$ . At $\varphi = \ln 2$ the factors are exactly $2$ and $\tfrac12$ . Events slide along the cone, and light stays light in every frame; the eigenvalues multiply to $1$ , which is why $\det\Lambda = 1$ and why $s^2 = (ct+x)(ct-x)$ is invariant by inspection.

Is Galileo's transformation really the same construction with a different sign?

Yes, and the degeneracy is the point. Its generator $N$ satisfies $N^2 = 0$ , so the series has only two terms and $\exp(vN) = \mathbb{1} + vN$ — a linear shear that tilts the space axis and leaves the time axis untouched, which is what "absolute time" means geometrically. The three cases $G^2 = -\mathbb{1}$ , $+\mathbb{1}$ , $0$ are the elliptic, hyperbolic and parabolic members of one family, and Newton's mechanics is the parabolic one: its metric is degenerate, its light cone has opened out flat, and it has no invariant speed to be limited by.