Lorentz Transformations as Hyperbolic Rotations of Spacetime

Take a stick, lay it on graph paper, and turn the paper. Every coordinate of every point changes, and yet you have not touched the stick: its length is exactly what it was. Rotation is the transformation that scrambles coordinates while protecting one number, $x^2 + y^2$ , and the whole of Euclidean geometry is the study of what that protected number allows. Ask a different question — what transformations protect $(ct)^2 - x^2$ instead? — and you get a different geometry, with different orbits, a different trigonometry, a different physics. The answer is the Lorentz boost. That is not a metaphor, it is a derivation. Write down the most general linear change of $(ct, x)$ that leaves $(ct)^2 - x^2$ alone, demand that it reduce to doing nothing when the speed is zero, and the algebra hands you a matrix with $\cosh\phi$ on the diagonal and $-\sinh\phi$ off it. No light clocks, no rods, no trains: the postulates about signals were spent once, upstream, to establish that $(ct)^2 - x^2$ is the invariant. After that the geometry runs by itself. And the difference from Euclid is a single sign. Swap the minus for a plus and every step still goes through, but the answer comes back as $\cos\theta$ and $\sin\theta$ , the orbits close into circles, and the transformation runs out of room after a full turn. Keep the minus and the orbits open into hyperbolae running forever toward two straight lines they never touch — the paths of light. The difference between a rotation and a boost, between a universe with a speed limit and one without, is the sign of one term in one quadratic form.

Key takeaways

A Lorentz boost preserves $s^2 = (ct)^2 - x^2$ exactly as a rotation preserves $x^2 + y^2$ . Demanding invariance is enough: the matrix $\begin{pmatrix}\cosh\phi & -\sinh\phi\\ -\sinh\phi & \cosh\phi\end{pmatrix}$ is forced, not assumed.
The orbits of a rotation are circles; the orbits of a boost are hyperbolae. Under a boost an event slides along its own hyperbola, and $s^2$ never budges.
$\cos^2\theta + \sin^2\theta = 1$ becomes $\cosh^2\phi - \sinh^2\phi = 1$ for one reason: the metric's minus sign. Track it into the generator and it turns $K^2 = -\mathbb{1}$ into $K^2 = +\mathbb{1}$ — an alternating power series (trigonometric) into a non-alternating one (hyperbolic).
The generator's eigenvalues say it sharpest. A rotation generator has eigenvalues $\pm i$ — no real direction survives, so circles close. A boost generator has eigenvalues $\pm 1$ — two real directions survive, and they are the light rays $x = \pm ct$ . The light cone is invariant because it is an eigenvector, whose eigenvalue $e^{\phi}$ is the Doppler factor.
Because $\Lambda(\phi) = \exp(\phi K)$ , boosts form a one-parameter group — which is why an additive boost parameter exists. That it is the rapidity, and that its additivity is the velocity-composition law, is proved in Rapidity and Hyperbolic Boosts.
Key idea: the minus sign in the metric is the entire difference between a rotation and a boost. Hyperbolae for circles, cosh for cos, an unreachable speed limit in place of a closed cycle of angles — all of it is that one sign, propagating.

The core insight

A Lorentz boost is not "a rotation, sort of". It is the rotation of a plane whose invariant is $(ct)^2 - x^2$ rather than $x^2 + y^2$ . Demand only that a linear map preserve that quadratic form and reduce to the identity at zero speed, and the hyperbolic form is forced. Every difference from Euclidean geometry — hyperbolic orbits, an unbounded "angle", two invariant null directions, an unreachable speed — is the propagation of the one minus sign that separates the two invariants.

Where this lives today

The generator $K$ is where a working physicist actually meets this material, rarely under that name. Write out the Lorentz group's six generators — three boosts $K_i$ , three rotations $J_i$ — compute their commutators, and the whole structure of special relativity compresses into a handful of brackets, the most eloquent being $[K_1, K_2] = -J_3$ : two boosts in different directions do not commute, and what is left over is a rotation. That leftover rotation is the Thomas precession, and its factor of one-half is the difference between a spin-orbit splitting that matches the alkali spectra and one twice too large: a minus sign in a metric, exponentiated into a matrix, ends up as a factor of two in the sodium D lines. The hyperbolic language is also the working currency of collider physics. Because a boost along the beam axis simply adds a constant to the rapidity, rapidity differences are untouched by it — and the longitudinal boost of the underlying parton collision is exactly what an experimenter never knows. So detectors are built in rapidity: ATLAS and CMS segment their calorimeters in pseudorapidity $\eta = -\ln\tan(\theta/2)$ , so the bins mean the same thing whatever the unknown boost was.

The physical picture

Be pedantic about what a rotation does. Rotating the $(x, y)$ plane walks the point $(1, 0)$ around the unit circle; turn far enough and it comes back, because the circle is closed and the angle is periodic. Every point keeps its distance from the origin, and no direction at all is left pointing where it was. Now run the identical exercise in spacetime, where the quantity that must not change is $s^2 = (ct)^2 - x^2$ . Ask where an event may go if $s^2$ is preserved, and the answer is: along $(ct)^2 - x^2 = \text{const}$ , a hyperbola. A hyperbola differs from a circle in three decisive ways. It does not close, so the "angle" has no period. It runs to infinity, so events can be carried arbitrarily far from the origin with their invariant fixed. And it has asymptotes — two straight lines it approaches forever and never reaches. Those asymptotes are $x = \pm ct$ : the worldlines of light. That is the light barrier, drawn as a picture. Nothing inside can be slid across them, because a hyperbola never crosses its own asymptote — and a circle has no asymptotes, so a Euclidean world has no speed limit.

Figure

The same operation, two geometries. Left: a rotation preserves x² + y² and walks the point (1, 0) around a CLOSED circle — after 2π it is home. Right: a boost preserves x² − (ct)² and slides the point (1, 0) along an OPEN hyperbola, out toward the 45° light line it can never reach. β = 0.6 lands at (1.25, 0.75); β = 0.9 lands at (2.294, 2.065); x² − (ct)² = 1 at every step

The right-hand panel repays a long look. The point never gets closer to the light line in the sense that matters — its invariant is $1$ , always, while the light line is where the invariant is $0$ . Yet on the page it crowds the diagonal, and will keep crowding forever, because that is what a hyperbola does to its asymptote. So the speed of light is not a large speed. It is a geometric boundary — the asymptote of the orbits — and asking a boost to reach it is asking a hyperbola to meet its asymptote at a finite point.

The core equations

Everything here answers one question: what linear maps of the $(ct, x)$ plane preserve $s^2$ ? The invariant is taken as given — derived in The Invariant Spacetime Interval, while the boost's algebraic form is derived independently in The Lorentz Transformation Equations. What is new is that those two facts are one fact.

s^{2} \;=\; (ct)^{2} - x^{2} \;=\; (ct')^{2} - x'^{2} \qquad (1)

where: $s^2$ — the invariant spacetime interval; $ct$ — time carried in units of length, so both axes are metres; $x$ — the space coordinate; primes denote the boosted frame. Signature $(+,-,-,-)$ : timelike $s^2 > 0$ , spacelike $s^2 < 0$ , lightlike $s^2 = 0$ . Many general-relativity texts use $(-,+,+,+)$ , which flips the sign of every $s^2$ below and changes no physics. Physical reading: (1) states the problem — "the interval is the same for everybody" — and the rest follows. Impose that on a general linear map: four unknowns, three constraints, one surviving family labelled by one parameter.

\begin{pmatrix} ct' \\ x' \end{pmatrix} = \Lambda(\phi)\begin{pmatrix} ct \\ x \end{pmatrix}, \qquad \Lambda(\phi) = \begin{pmatrix} \cosh\phi & -\sinh\phi \\ -\sinh\phi & \cosh\phi \end{pmatrix} \qquad (2)

where: $\Lambda(\phi)$ — the boost matrix; $\phi$ — the rapidity, tied to velocity by $\tanh\phi = \beta = v/c$ (established in Rapidity and Hyperbolic Boosts, which owns it). Physical reading: beside the Euclidean rotation, whose entries are $\cos\theta$ and $\mp\sin\theta$ , the two matrices differ in exactly two places — $\cos \to \cosh$ , and one sign of a $\sin$ . Both trace to the minus sign in (1). Substituting (2) into (1) produces the factor $\cosh^2\phi - \sinh^2\phi$ multiplying the old interval, so the map preserves $s^2$ iff that combination is one.

\cosh^{2}\phi - \sinh^{2}\phi = 1 \qquad \longleftrightarrow \qquad \cos^{2}\theta + \sin^{2}\theta = 1 \qquad (3)

where: the left identity is what Minkowski geometry needs, the right what Euclidean geometry needs — the same statement about two different invariants. Written with the Lorentz factor, (3) is $\gamma^2 - \gamma^2\beta^2 = 1$ : with $\gamma = \cosh\phi$ and $\gamma\beta = \sinh\phi$ , equation (2) is the Lorentz transformation, disguise removed. Every one-parameter family starting at the identity is the exponential of a generator — the matrix describing what an infinitesimal boost does.

\Lambda(\phi) = \exp(\phi K), \qquad K = \begin{pmatrix} 0 & -1 \\ -1 & 0\end{pmatrix}, \qquad K^{2} = +\mathbb{1} \qquad (4)

where: $K$ — the boost generator; $\mathbb{1}$ — the $2\times 2$ identity. Physical reading: the rotation generator is $J = \begin{pmatrix}0 & -1\\ 1 & 0\end{pmatrix}$ , which squares to $-\mathbb{1}$ . That sign is the whole story: $K^2 = +\mathbb{1}$ makes the exponential series non-alternating, and such a series sums to $\cosh$ and $\sinh$ ; $J^2 = -\mathbb{1}$ makes it alternate, and an alternating series sums to $\cos$ and $\sin$ . Finally, the sharpest handle: what the generator leaves fixed.

K\begin{pmatrix}1\\1\end{pmatrix} = -\begin{pmatrix}1\\1\end{pmatrix}, \quad K\begin{pmatrix}1\\-1\end{pmatrix} = +\begin{pmatrix}1\\-1\end{pmatrix} \;\Longrightarrow\; \Lambda(\phi)\begin{pmatrix}1\\ \pm 1\end{pmatrix} = e^{\mp\phi}\begin{pmatrix}1\\ \pm 1\end{pmatrix} \qquad (5)

where: $(1, \pm 1)$ — the null directions, i.e. the light rays $x = \pm ct$ written as vectors $(ct, x)$ ; $e^{\mp\phi}$ — the eigenvalue. Physical reading: a boost does not move the light rays, it rescales them by reciprocal factors, so each ray is carried onto itself as a line. This is the invariance of the light cone stated as an eigenvalue problem — and the eigenvalue $e^{\phi} = \sqrt{(1+\beta)/(1-\beta)}$ is the relativistic Doppler factor, exactly $2$ at $\beta = 0.6$ . The rotation generator has eigenvalues $\pm i$ : no real direction survives a rotation, which is why circles close and a Euclidean world has no speed limit.

Figure

PUNCHLINE — the invariant hyperbolae. One event, three frames. Boosting slides it ALONG its hyperbola: (1.000, 0) → (1.250, 0.750) at β = 0.6 → (2.294, 2.065) at β = 0.9. The coordinates change wildly; s² = (ct)² − x² = 1.000 every time. The Euclidean circle (dashed) is what the orbit would be if the metric carried a plus sign — bounded, closed, no asymptote, no light barrier

Three observers write down three completely different pairs of numbers for one and the same event — $(1.000,\, 0.000)$ , $(1.250,\, 0.750)$ , $(2.294,\, 2.065)$ — and all three compute $s^2 = 1.000$ . The coordinates are opinions. The hyperbola they all sit on is the fact.

Where the geometry comes from

Derivation AThe boost matrix, forced by interval invariance alone

The claim is strong enough to state precisely before starting. We will not assume the Lorentz transformation, not use a light clock, not invoke a rod or a train. We assume three things only: that the change of coordinates is linear, that it preserves $s^2 = (ct)^2 - x^2$ , and that it is continuously connected to doing nothing at all. From those alone the boost matrix is completely determined, with one free parameter left over — and that parameter turns out to be the velocity.

Step 1. Write the most general linear map with four unknown coefficients: $ct' = A\,ct + B\,x$ , $x' = C\,ct + D\,x$ , or $M = \begin{pmatrix}A & B\\ C & D\end{pmatrix}$ . Why: a nonlinear map would bend a free particle's straight worldline, and the new frame would not be inertial — the principle of relativity itself forbids anything but a linear map.

Step 2. Impose invariance. Demanding $(ct')^2 - x'^2 = (ct)^2 - x^2$ for every event means matching the coefficients of $(ct)^2$ , $x^2$ and the cross term separately: $A^2 - C^2 = 1$ , $D^2 - B^2 = 1$ , $AB - CD = 0$ . Compactly, with $\eta = \mathrm{diag}(1,-1)$ , all three are $M^{\mathsf T}\eta\,M = \eta$ . Why: invariance must hold identically, and one quadratic form equals another identically only if their coefficients agree term by term.

Step 3. Solve the first condition. $A^2 - C^2 = 1$ is a hyperbola in the $(A, C)$ plane, and every point of a hyperbola can be written $A = \cosh\phi$ , $C = -\sinh\phi$ for some real $\phi$ — that is what the hyperbolic functions are for, exactly as $\cos$ and $\sin$ parametrise $A^2 + C^2 = 1$ . Why: the hinge of the article. Had the invariant been $(ct)^2 + x^2$ , the same step would have produced $\cos$ and $\sin$ , because a circle would have replaced the hyperbola. The functions are chosen by the shape of the constraint curve — and that shape is chosen by the sign in the metric.

Step 4. Finish. $D^2 - B^2 = 1$ likewise gives $D = \cosh\psi$ , $B = -\sinh\psi$ ; the cross condition then forces $\sinh(\phi - \psi) = 0$ , hence $\psi = \phi$ , and continuity with the identity at $\phi = 0$ fixes the signs. So $\Lambda(\phi) = \begin{pmatrix}\cosh\phi & -\sinh\phi\\ -\sinh\phi & \cosh\phi\end{pmatrix}$ , with $\det\Lambda = 1$ . Why: four unknowns minus three constraints leaves one parameter — right, since the one thing distinguishing two boosts is one number, the relative velocity.

Step 5. Name the parameter. The origin of the primed frame is the set of events with $x' = 0$ ; the second row of (2) gives $-\sinh\phi\,(ct) + \cosh\phi\,x = 0$ , so $x = (\tanh\phi)\,ct$ — a point moving at $v = c\tanh\phi$ . Hence $\tanh\phi = \beta$ , and immediately $\cosh\phi = \gamma$ , $\sinh\phi = \gamma\beta$ : equation (2) has become the Lorentz transformation. Why: physics re-enters only as a label. The geometry produced the transformation; calling its parameter a velocity is bookkeeping done afterwards. *Verified in verify_aq-sr-lorentz-boosts-hyperbolic-rotations.py: the three coefficient conditions reproduced symbolically from $M^{\mathsf T}\eta M = \eta$ ; $\Lambda^{\mathsf T}\eta\,\Lambda - \eta = 0$ identically; $\det\Lambda = 1$ ; the cross condition reduces to $\sinh(\phi-\psi) = 0$ ; $\cosh(\mathrm{arctanh}\,\beta) = \gamma$ and $\sinh(\mathrm{arctanh}\,\beta) = \gamma\beta$ symbolically and at $\beta = 0.6,\ 0.9$ .*

Go deeper · Derivation B · 2 minThe metric's minus sign, tracked through the generator

Derivation BThe metric's minus sign, tracked through the generator

The previous derivation produced hyperbolic functions but did not really explain them — they appeared because a hyperbola happened to turn up at Step 3. This one locates the sign flip at its source, by looking not at the boost but at the infinitesimal boost, the generator $K$ defined by $\Lambda(\phi) \approx \mathbb{1} + \phi K$ . The generator is where the metric leaves its fingerprint: "preserve the metric" becomes a linear condition on $K$ , and that condition has a different solution for a plus sign than for a minus sign. Everything else follows mechanically. This is the article's owned result, and it is exact.

Step 1. Differentiate the invariance condition. The finite statement is $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ ; putting $\Lambda = \mathbb{1} + \phi K$ and keeping first order in $\phi$ gives $K^{\mathsf T}\eta + \eta K = 0$ . Why: this one equation carries the whole difference between the two geometries, because it is the only place $\eta$ appears.

Step 2. Feed it the Euclidean metric, $\eta = \mathbb{1}$ . It collapses to $K^{\mathsf T} + K = 0$ : the generator must be antisymmetric. In two dimensions the only such matrix up to scale is $J = \begin{pmatrix}0 & -1\\ 1 & 0\end{pmatrix}$ , and $J^2 = -\mathbb{1}$ . Why: antisymmetry means the diagonal vanishes and the off-diagonal entries are opposite — and opposite entries multiply to a negative when the matrix is squared.

Step 3. Feed it the Minkowski metric, $\eta = \mathrm{diag}(1,-1)$ . Writing $K = \begin{pmatrix}a & b\\ c & d\end{pmatrix}$ , the condition becomes $\begin{pmatrix}2a & b - c\\ b - c & -2d\end{pmatrix} = 0$ , so $a = d = 0$ and $b = c$ : the generator must be symmetric off the diagonal, not antisymmetric. Taking $b = c = -1$ gives $K = \begin{pmatrix}0 & -1\\ -1 & 0\end{pmatrix}$ , and $K^2 = +\mathbb{1}$ . Why: here is the sign flip, caught in the act. The minus sign in $\eta$ turns the requirement $c = -b$ into $c = +b$ , and that single change in one off-diagonal entry flips the sign of the generator's square.

Step 4. Exponentiate both. Because $J^2 = -\mathbb{1}$ , the powers of $J$ cycle as $\mathbb{1}, J, -\mathbb{1}, -J, \dots$ , so the even terms alternate: $\exp(\theta J) = \mathbb{1}\cos\theta + J\sin\theta$ . Because $K^2 = +\mathbb{1}$ , the powers of $K$ cycle as $\mathbb{1}, K, \mathbb{1}, K, \dots$ with no sign changes, so nothing alternates: $\exp(\phi K) = \mathbb{1}\cosh\phi + K\sinh\phi$ . Why: an alternating series is a circular function and a non-alternating one is a hyperbolic function. The metric decides which by deciding the sign of $K^2$ , and nothing else in the calculation differs by so much as a symbol.

Step 5. Compare. $\exp(\theta J) = \begin{pmatrix}\cos\theta & -\sin\theta\\ \sin\theta & \cos\theta\end{pmatrix}$ ; $\exp(\phi K) = \begin{pmatrix}\cosh\phi & -\sinh\phi\\ -\sinh\phi & \cosh\phi\end{pmatrix}$ . Both have determinant $1$ , and both form a one-parameter group, $\Lambda(\phi_1)\Lambda(\phi_2) = \Lambda(\phi_1 + \phi_2)$ . Why: the group property is why an additive boost parameter exists at all. That the parameter is the rapidity, and that its additivity is the velocity-composition law, is owned and proved in Rapidity and Hyperbolic Boosts . *Verified in verify_aq-sr-lorentz-boosts-hyperbolic-rotations.py: sympy solves $K^{\mathsf T}\eta + \eta K = 0$ and returns $\{a = 0,\ d = 0,\ b = c\}$ for the Minkowski metric and $\{a = 0,\ d = 0,\ b = -c\}$ for the Euclidean one; $K^2 = +\mathbb{1}$ , $J^2 = -\mathbb{1}$ ; both exponentials computed by matrix exponential and matched entry by entry; the group law checked symbolically.*

Go deeper · Derivation C · 2 minThe light cone is an eigenvector, and its eigenvalue is the Doppler factor

Derivation CThe light cone is an eigenvector, and its eigenvalue is the Doppler factor

There is a sharper way still to separate a boost from a rotation, the one a mathematician reaches for first: look at what the transformation leaves pointing where it was. A rotation leaves no direction fixed — turn the paper and every arrow has moved. A boost leaves two directions fixed, and they are the light rays. This derivation proves it and extracts the eigenvalue, which turns out to be a quantity astronomers measure. The payoff is a one-line explanation of why the light cone is the one structure every observer draws identically.

Step 1. Ask the eigenvalue question of the rotation generator. The characteristic polynomial of $J$ is $\lambda^2 + 1 = 0$ , so $\lambda = \pm i$ : imaginary eigenvalues, and no real eigenvectors at all. Why: the algebraic reason a circle closes. A real invariant direction would be a line the rotation left alone, and a plane rotation leaves none — so the orbits curl back and the parameter is periodic.

Step 2. Ask it of the boost generator. The characteristic polynomial of $K$ is $\lambda^2 - 1 = 0$ , so $\lambda = \pm 1$ : real, and therefore carrying real eigenvectors. Solving $Kv = -v$ gives $v = (1, 1)$ ; solving $Kv = +v$ gives $v = (1, -1)$ . Why: the same sign flip as before, now surfacing in a characteristic polynomial. $\lambda^2 + 1$ has no real roots; $\lambda^2 - 1$ has two. One minus sign in the metric, two real invariant directions in the boost.

Step 3. Identify them physically. In the convention $(ct, x)$ , the vector $(1, 1)$ points along $x = ct$ and $(1, -1)$ along $x = -ct$ : the two light rays through the origin — the light cone of $1{+}1$ dimensions. Why: the invariant directions of a boost are not an algebraic curiosity — they are the paths of light, fixed the moment a minus sign was written into $s^2$ .

Step 4. Exponentiate to a finite boost. Since $Kv_\pm = \mp v_\pm$ for $v_\pm = (1, \pm 1)$ , $\Lambda(\phi)v_\pm = e^{\mp\phi}v_\pm$ . A boost does not tilt a light ray; it rescales it, by $e^{-\phi}$ in one null direction and $e^{+\phi}$ in the other — reciprocal factors, consistent with $\det\Lambda = 1$ . Why: an eigenvector is mapped along itself, so the line is carried to itself. That is the precise sense in which the light cone is the same in every frame — not that light's coordinates are unchanged, but that each null line maps onto itself.

Step 5. Recognise the eigenvalue. With $\tanh\phi = \beta$ , $e^{\phi} = \cosh\phi + \sinh\phi = \gamma(1+\beta) = \sqrt{\dfrac{1+\beta}{1-\beta}}$ — the longitudinal relativistic Doppler factor. At $\beta = 0.6$ , $\phi = \ln 2$ and the eigenvalue is $e^{\phi} = 2$ exactly. Why: "the null directions are eigenvectors" and "light from a source receding at $0.6c$ arrives with its wavelength doubled" are the same statement. The eigenvalue of a boost on a light ray is the factor by which that light is Doppler-shifted. *Verified in verify_aq-sr-lorentz-boosts-hyperbolic-rotations.py: the characteristic polynomials $\lambda^2 - 1$ and $\lambda^2 + 1$ ; eigenvalues $\{+1,-1\}$ of $K$ and $\{+i,-i\}$ of $J$ , with $J$ carrying zero real eigenvectors; $\Lambda(\phi)(1,\pm1)^{\mathsf T} = e^{\mp\phi}(1,\pm1)^{\mathsf T}$ symbolically; $e^{\mathrm{arctanh}\beta} = \sqrt{(1+\beta)/(1-\beta)}$ ; $e^{\mathrm{arctanh}\,0.6} = 2$ to machine precision.*

Limiting cases

Case	Result	Meaning
Slow boost, $\phi \ll 1$	$\Lambda \approx \mathbb{1} + \phi K$ ; $ct' \approx ct - \beta x$ , $x' \approx x - \beta\,ct$	the generator is the small boost. Dropping the $\beta x$ term — what a Galilean transformation does — is the assumption that time is absolute
$\beta \to 1$	$\phi = \mathrm{arctanh}\,\beta \to \infty$	the light barrier as a range statement: $\tanh$ maps $\mathbb{R}$ into $(-1,1)$ , so no finite $\phi$ arrives
Null direction, $s^2 = 0$	$\Lambda(\phi)(1,\pm1) = e^{\mp\phi}(1,\pm1)$	the light rays are eigenvectors: rescaled, never rotated — why every observer draws the light cone identically
Metric sign flipped, $\eta \to \mathbb{1}$	$K \to J$ , so $K^2 = +\mathbb{1} \to J^2 = -\mathbb{1}$ , $\cosh \to \cos$	the counterfactual: circular orbits, a periodic "boost angle", no invariant directions and no speed limit

Common trapA boost is not a rotation of the diagram — and the two axes do not stay perpendicular

Under a Euclidean rotation the axes turn rigidly: same way, same amount, still at right angles. Under a boost they do nothing of the kind. The $ct'$ axis tilts one way and the $x'$ axis tilts the other way, both closing scissor-like on the $45^\circ$ light line. They were never supposed to be Euclidean-perpendicular — they are Minkowski-orthogonal, $A^0B^0 - A^1B^1 = 0$ , a condition met by any two directions that are mirror images across the light line. A second trap follows: there is no maximum boost. Rotating by $2\pi$ is no rotation at all, but no rapidity ever comes back round — $\phi$ runs over the whole real line, and $\cosh\phi \geq 1$ always, so time and space never simply swap the way $x$ and $y$ do at a quarter-turn. The skewed axes and the calibration that fixes the scale are the business of Minkowski Diagrams and Skewed Coordinate Axes.

Worked examples

1 · Slide one event along its hyperbola and watch the invariant refuse to move

An event occurs on the worldline of a clock at rest at the origin, one metre of light-travel time after the clock was zeroed: its coordinates are $(ct, x) = (1.000,\ 0.000)\ \mathrm{m}$ . Find its coordinates in a frame boosted by $\beta = 0.6$ , and in one boosted by $\beta = 0.9$ , and check the interval each time.

Step 1. Identify the invariant. $s^2 = (1.000)^2 - (0.000)^2 = +1.000\ \mathrm{m}^2$ — positive, so the separation is timelike, as it must be on the worldline of a clock. That sign is frozen: no boost changes $s^2$ at all.

Step 2. Get the rapidities. $\phi_1 = \mathrm{arctanh}(0.6) = \tfrac12\ln\dfrac{1.6}{0.4} = \ln 2 = 0.6931$ ; $\phi_2 = \mathrm{arctanh}(0.9) = \tfrac12\ln 19 = 1.4722$ . Then $\cosh\phi_1 = 1.2500$ , $\cosh\phi_2 = 2.2942$ — cross-check against $1/\sqrt{1-\beta^2}$ : $1/\sqrt{0.64} = 1.2500$ , $1/\sqrt{0.19} = 2.2942$ . They agree, as (3) requires.

Step 3. Apply the boost. With the event at $x = 0$ , (2) collapses to $ct' = \cosh\phi$ , $x' = -\sinh\phi$ ; the active slide along the hyperbola is the same arithmetic with the sign of $x$ reversed, putting the event at $(\cosh\phi,\ \sinh\phi)$ . At $\beta = 0.6$ : $(1.2500,\ 0.7500)\ \mathrm{m}$ . At $\beta = 0.9$ : $(2.2942,\ 2.0647)\ \mathrm{m}$ .

Step 4. Check the invariant, which is the whole point. At $\beta = 0.6$ : $(1.2500)^2 - (0.7500)^2 = 1.5625 - 0.5625 = \mathbf{1.000\ m^2}$ . At $\beta = 0.9$ : $(2.2942)^2 - (2.0647)^2 = 5.2632 - 4.2632 = \mathbf{1.000\ m^2}$ . The time coordinate has grown by $129\%$ and the space coordinate from nothing to two metres, and the interval has not moved in the fourth decimal place. Three coordinate pairs are three names for one event; $s^2 = 1.000\ \mathrm{m}^2$ is its name in every language.

2 · The eigenvalue you can see through a telescope

A source recedes at $\beta = 0.6$ and emits a green photon, $\lambda = 500\ \mathrm{nm}$ . Using only the eigenvalue property (5), find the wavelength you receive.

Step 1. Identify the object. A light ray moving in $+x$ is the null direction $(ct, x) \propto (1, 1)$ — an eigenvector of the boost, with eigenvalue $e^{+\phi}$ under the boost carrying the emitter's description into the receiver's.

Step 2. Compute it. $\phi = \mathrm{arctanh}(0.6) = \ln 2$ , so $e^{\phi} = \mathbf{2}$ — exactly two, with no rounding. Cross-check the closed form: $\sqrt{(1+\beta)/(1-\beta)} = \sqrt{1.6/0.4} = \sqrt{4} = 2$ .

Step 3. Apply it. Every component of the null vector is multiplied by $2$ , and a wavelength is such a component: $\lambda_{\text{received}} = 2 \times 500\ \mathrm{nm} = \mathbf{1000\ nm}$ . The green photon arrives in the near infrared, invisible to the eye.

Step 4. Read the boundary. As $\beta \to 1$ , $\phi \to \infty$ and $e^{\phi} \to \infty$ : the wavelength diverges and the source reddens out of existence rather than crossing a finite threshold — "the hyperbola never reaches its asymptote", in nanometres. Note what was never used: no wave equation, no photon energy, no Doppler formula. The number $2$ fell out of a $2\times 2$ matrix's eigenvalue.

Physics of the portrait

The hyperbolic rotation is one of those ideas whose discoverers were mathematicians slightly embarrassed by how little physics they had needed. Poincaré saw it first and buried it inside an imaginary coordinate; Minkowski saw it plainly and told the world.

Origins · 1 minHyperbolic rotation and how this was discovered

Hyperbolic rotation

1906–1911 · Poincaré's imaginary coordinate · Minkowski's spacetime · Varićak's hyperbolic velocity space · Robb's naming

In the Palermo memoir Sur la dynamique de l'électron (submitted July 1905, published in the Rendiconti del Circolo Matematico di Palermo 21, 129–175, in 1906), Henri Poincaré noticed that the Lorentz transformations leave $x^2 + y^2 + z^2 - c^2t^2$ invariant and could therefore be regarded as rotations in a four-dimensional space — provided the fourth coordinate is taken to be the imaginary quantity $x^4 = ict$ . The trick works, and it hides the very thing it explains. Hermann Minkowski, lecturing at Cologne in September 1908, threw the disguise away and said the true thing — that spacetime has an indefinite metric of its own, that boosts are its rotations, and that this is a fact about the world rather than a change of variables. Two years later Vladimir Varićak and, independently, E. T. Whittaker introduced the hyperbolic angle as the natural boost parameter, and Varićak drew the deep consequence: the velocity space of special relativity is hyperbolic, a Lobachevskian plane. In 1911 Alfred Robb, in Optical Geometry of Motion, gave that angle the name it still carries — rapidity. None of the four is remembered for it: the concept outlived its authors' claims on it.

History

1906

— Poincaré's Palermo memoir "Sur la dynamique de l'électron" (submitted July 1905, published in Rendiconti del Circolo Matematico di Palermo 21, 129–175) treats the Lorentz transformations as rotations in a four-dimensional space with the imaginary coordinate x⁴ = ict.

1908

— Minkowski delivers "Raum und Zeit" at Cologne on 21 September, to the 80th Assembly of German Natural Scientists and Physicians, and states that space and time by themselves are doomed to fade away into mere shadows.

1910

— Varićak and, independently, Whittaker introduce the hyperbolic angle as the natural boost parameter; Varićak shows the velocity space of special relativity is hyperbolic (Lobachevskian).

1911

— Alfred Robb, in Optical Geometry of Motion, names the hyperbolic angle: rapidity. The term is adopted by Silberstein in 1914 and Eddington in 1924.

Going further

Texts. Taylor & Wheeler, Spacetime Physics, 2nd ed., Ch. 1 · Rindler, Introduction to Special Relativity, 2nd ed., §3.4 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.1 · Poincaré, Rendiconti del Circolo Matematico di Palermo 21 (1906) 129 · Robb, Optical Geometry of Motion (Cambridge, 1911).

Provenance. The hyperbolic form of the boost is standard (Rindler §3.4, Griffiths eq. 12.18). Two results here go beyond the usual presentation and are proved rather than asserted: the generator's symmetry, derived from $K^{\mathsf T}\eta + \eta K = 0$ — antisymmetric for $\eta = \mathbb{1}$ , symmetric off-diagonal for $\eta = \mathrm{diag}(1,-1)$ , hence $J^2 = -\mathbb{1}$ versus $K^2 = +\mathbb{1}$ — and the identification of the null directions as the eigenvectors of the boost, with eigenvalue $e^{\mp\phi}$ equal to the Doppler factor. Anchors, all recomputed in verify_aq-sr-lorentz-boosts-hyperbolic-rotations.py: $\beta = 0.6 \Rightarrow \gamma = 1.2500$ , $\phi = \ln 2$ , $e^{\phi} = 2$ exactly; $\beta = 0.9 \Rightarrow \gamma = 2.2942$ , $\gamma\beta = 2.0647$ , $\phi = \tfrac12\ln 19$ ; the slid event $(1.000,\,0.000) \to (1.2500,\,0.7500) \to (2.2942,\,2.0647)$ with $s^2 = 1.000\ \mathrm{m}^2$ throughout; and every SVG vertex encoding an invariant hyperbola. Signature: $(+,-,-,-)$ , timelike-positive.

Builds on. The Invariant Spacetime Interval · The Lorentz Transformation Equations · The Lorentz Factor Gamma · The Two Postulates of Special Relativity. Leads to. Rapidity and Hyperbolic Boosts · Minkowski Diagrams and Skewed Coordinate Axes · Invariant Hyperbolae · Proper Time.

If a boost is "just a rotation", why can I not boost by 90° and turn time into space?

Because there is no such thing as a boost by $90^\circ$ . The Euclidean angle lives on a circle and returns after $2\pi$ ; the rapidity lives on the whole real line and never returns. A quarter-turn swaps $x$ and $y$ because $\cos 90^\circ = 0$ — the diagonal entries vanish. The corresponding boost statement would need $\cosh\phi = 0$ , and $\cosh\phi \geq 1$ for every real $\phi$ : the diagonal entries of a boost never vanish, so time and space never swap roles. The analogy is exact at the level of "preserve a quadratic form", and it breaks exactly where the metric's minus sign puts it.

Where exactly does the minus sign enter? I keep losing track of it.

It enters once, in $s^2 = (ct)^2 - x^2$ , then propagates through four places in lockstep. It makes the constraint curve a hyperbola instead of a circle, which selects $\cosh/\sinh$ over $\cos/\sin$ . It flips $K^{\mathsf T}\eta + \eta K = 0$ from " $K$ antisymmetric" to " $K$ symmetric off-diagonal", flipping $K^2$ from $-\mathbb{1}$ to $+\mathbb{1}$ , which flips the exponential's series from alternating to non-alternating. And it flips the characteristic polynomial from $\lambda^2 + 1$ (no real roots) to $\lambda^2 - 1$ (two), giving the boost two invariant real directions — the light rays — and giving spacetime a speed limit.

Is Poincaré's ict trick wrong, then? It seems to make everything Euclidean.

It is not wrong; it is misleading, which in teaching is worse. Setting $x^4 = ict$ turns $-c^2t^2$ into $+(x^4)^2$ , so the Minkowski form becomes a sum of squares and boosts become rotations through an imaginary angle. Every formula you derive that way is correct. But the trick disguises the physically decisive fact — that the metric is indefinite: some non-zero vectors have zero length (the null ones), others have negative squared length, and the entire causal structure of the world lives in that distinction. Minkowski's contribution in 1908 was to stop pretending.

The two axes visibly close in on each other on a diagram. Doesn't that mean the boost distorts angles, so it cannot really be a rotation?

It preserves Minkowski angles and distorts Euclidean ones. "Perpendicular" is a statement about a metric, and here two directions are orthogonal when $A^0B^0 - A^1B^1 = 0$ . The $ct'$ axis has direction $(\cosh\phi, \sinh\phi)$ and the $x'$ axis $(\sinh\phi, \cosh\phi)$ , giving $\cosh\phi\sinh\phi - \sinh\phi\cosh\phi = 0$ : orthogonal, exactly. They look non-perpendicular because you are reading a Minkowski figure with Euclidean eyes. What is distorted is the paper, not the geometry.

If the light rays are eigenvectors and get rescaled by e^φ, does light itself "change" under a boost?

The ray does not change — the line $x = ct$ maps onto the line $x = ct$ , which is what invariance of the light cone means and why the speed of light is the same in every frame. What is rescaled is a particular vector along that line: a photon's four-momentum, say. Rescaling a null vector by $e^{\phi}$ multiplies its wavelength by $e^{\phi}$ and its frequency by $e^{-\phi}$ — the Doppler shift. The geometry of the ray is invariant; the physics carried along it is Doppler-shifted; both facts are the single eigenvalue equation (5).

Why does the derivation need "connected to the identity"? What is that ruling out?

Time reversal and spatial reflection. The condition $M^{\mathsf T}\eta M = \eta$ is satisfied not only by the boosts but also by $\mathrm{diag}(-1, 1)$ , which reverses time, and by $\mathrm{diag}(1, -1)$ , which reflects space. Both preserve $s^2$ ; neither is a boost, and neither can be reached by continuously turning a velocity up from zero. Insisting on reachability from doing nothing selects the proper orthochronous part of the Lorentz group — determinant $+1$ , time direction preserved — the piece that describes a physical change of inertial frame.