Invariant Hyperbolae and Calibration

Hand a draughtsman a rotated axis and a pair of compasses and he calibrates it in ten seconds: swing an arc of radius one, mark where it cuts the tilted line, done. The mark is right, because a rotation cannot move a point off its circle. The compass is not a convenience — it is the geometry, in brass. Hand him a Minkowski diagram and the compass is worthless. The tilted $ct'$ axis at $\beta = 0.6$ meets the unit circle at $(ct, x) = (0.8575,\ 0.5145)$ — a perfectly definite event, which is nevertheless not one second on the moving clock: its invariant interval from the origin is $\sqrt{0.8575^2 - 0.5145^2} = 0.6860$ , not $1$ . Worse, boost to a third frame and the compass marks a different event again. A calibration that depends on who holds the compass is not a calibration. The instrument that works is a curve — the invariant hyperbola $c^2t^2 - x^2 = \text{const}$ , in the chapter's locked signature $(+,-,-,-)$ . It cuts the $\beta = 0.6$ time axis at $(1.25,\ 0.75)$ , and that event is one second of proper time, for every observer at once. But the question nobody asks is: why that curve? "Because $s^2$ is invariant" says the hyperbola works, not that nothing else does. The real answer is a theorem: the boost-invariant functions on the $(ct, x)$ plane are exactly the functions of $s^2$ , and there are no others. The level sets of $s^2$ are therefore the only curves a boost cannot slide you off — so the hyperbola is not the best available ruler. It is the only ruler there is.

Key takeaways

An invariant hyperbola is a level set of the interval, $c^2t^2 - x^2 = \text{const}$ , in the locked signature $(+,-,-,-)$ — timelike-positive. Timelike hyperbolae ( $s^2 = +c^2\tau^2$ ) calibrate clocks; spacelike ones ( $s^2 = -\ell^2$ ) calibrate rulers; the null case $s^2 = 0$ is the light cone, the shared asymptote of both.
A boost is a hyperbolic rotation, so it slides events along these curves and does nothing else: the hyperbolae are the orbits of the boost group, $(c\tau\cosh\varphi,\ c\tau\sinh\varphi)$ . Hence the calibration rule — the moving observer's $\tau$ -tick is where his tilted axis crosses the hyperbola, at $(\gamma, \gamma\beta) = (1.25,\ 0.75)$ for $\beta = 0.6$ . Its page-length is $1.4577$ , not $1$ ; that distortion is owned by Minkowski Diagrams and Skewed Axes.
The owned theorem: the hyperbola is forced, not chosen. A function is boost-invariant if and only if it is a function of $s^2$ , so the hyperbolae are the only invariant curves. A compass provably cannot calibrate: a boost sends $(1,0)$ to $(1.1276,\ 0.5211)$ , still on the hyperbola but of Euclidean radius $1.2422 \neq 1$ .
The same curve is the protractor. Its invariant arc length — measured with the metric, not with ink — is exactly the rapidity $\varphi$ . The page disagrees: one arc, $0.6931$ in physics and $0.8005$ in Euclidean tape.
Key idea: the invariant hyperbolae are the orbits of the Lorentz boost, and an orbit is the only structure a group cannot move you out of. Every tick, every "one second" agreed on by observers in relative motion, lives on one of these curves — because there is nowhere else for it to live.

The core insight

Circles are to rotations what hyperbolae are to boosts: the orbits. Because the boost-invariant functions on the $(ct,x)$ plane are exactly the functions of $s^2 = c^2t^2 - x^2$ , the level sets of $s^2$ are the unique frame-independent curves in flat spacetime. That one fact does two jobs: it makes the hyperbola the only possible ruler — its intersection with a tilted axis marks that axis's true unit — and, measured with the metric rather than a pencil, the only possible protractor, since its invariant arc length is the rapidity.

Where this lives today

The most literal survival of the invariant hyperbola is as a worldline. An observer with constant proper acceleration $a$ travels not on a parabola, as Newton would have it, but on the hyperbola $x^2 - c^2t^2 = (c^2/a)^2$ — asymptotic to a light line he can never catch, and from beyond which no signal can ever reach him. That line is a horizon, produced by nothing but acceleration in flat spacetime: the flat-space rehearsal for the event horizon of a black hole, and the foundation of the accelerated-frames chapter. The second is as an arc length. Because rapidity is the invariant length along the hyperbola, it is the coordinate in which boosts merely translate — which is why particle physicists work in rapidity rather than velocity. A longitudinal boost along a collider's beam axis shifts every rapidity by the same constant, so rapidity differences are boost-invariant: exactly what a detector needs when the parton-level collision frame is unknown. The pseudorapidity $\eta$ on every LHC event display is the massless limit of this arc length, a payoff belonging to Rapidity and Additive Velocity. And in momentum space the mass shell $E^2 - (pc)^2 = (mc^2)^2$ is the same invariant hyperbola with $(E, pc)$ for $(ct, x)$ : the uniqueness theorem applies verbatim, which is why mass is the one label surviving every change of frame.

The physical picture

Spin the axes of a Euclidean plane and every point travels on a circle about the origin, because $x^2 + y^2$ is what the rotation preserves. The circles are the orbits: the sets you can be moved around inside but never out of — which is why the compass calibrates. Now boost. The invariant is $s^2 = c^2t^2 - x^2$ , a difference of squares, so the level sets are hyperbolae in two families separated by the light cone. The timelike family, $s^2 = +c^2\tau^2 > 0$ , opens along $ct$ inside the cone; its upper branch is the set of events one proper time $\tau$ to the future of the origin, so it calibrates clocks. The spacelike family, $s^2 = -\ell^2 < 0$ , opens sideways outside the cone, its points at proper distance $\ell$ : it calibrates rulers. Between them lies the degenerate member $s^2 = 0$ — the lines $x = \pm ct$ , the light cone, the common asymptote of everything else. Signature discipline matters: in $(+,-,-,-)$ the positive $s^2$ is the timelike one, so $c\tau = +\sqrt{s^2}$ is real and positive while $\ell = \sqrt{-s^2}$ — the minus is not optional. Flip to $(-,+,+,+)$ and every one of those signs flips with it; the curves are the same, and only the bookkeeping moves. A boost, being a hyperbolic rotation, slides an event along the hyperbola it is already on — never onto another, never across the asymptote. Take $(ct, x) = (1, 0)$ , on $s^2 = 1$ , and boost it by rapidity $\varphi = 0.5$ : it moves to $(\cosh 0.5,\ \sinh 0.5) = (1.1276,\ 0.5211)$ , and $\cosh^2 0.5 - \sinh^2 0.5 = 1$ exactly — it never left the hyperbola. But its Euclidean radius is now $\sqrt{1.1276^2 + 0.5211^2} = 1.2422$ : it has flown clean off the circle. The compass has already failed at $\varphi = 0.5$ , and fails worse the faster you go. That is not a defect of the compass — it is the statement that circles are not orbits of the boost.

Figure

The orbits of the boost. LEFT: under a rotation the point (1,0) rides the unit CIRCLE — which is why a compass calibrates a rotated axis. RIGHT: under a boost of rapidity φ = 0.5 the same point rides the unit HYPERBOLA to (cosh 0.5, sinh 0.5) = (1.1276, 0.5211). Its invariant interval is unchanged — cosh²φ − sinh²φ = 1 exactly — but its Euclidean radius has grown to 1.2422, so it has left the circle entirely. Circles are the orbits of rotations; hyperbolae are the orbits of boosts; and an orbit is the only set a group cannot move you out of

The core equations

The signature is the chapter's locked one, $(+,-,-,-)$ , timelike-positive; it is stated once and used without exception.

s^2 \;=\; \eta_{\mu\nu}x^{\mu}x^{\nu} \;=\; c^2t^2 - x^2 \;=\; \text{invariant under every boost} \qquad (1)

where: $x^{\mu} = (ct, x)$ , $\eta = \mathrm{diag}(+1,-1)$ . Physical reading: $s^2 > 0$ is timelike, $s^2 < 0$ spacelike, $s^2 = 0$ null, and the convention fixes how each is cashed out: $c\tau = +\sqrt{s^2}$ , $\ell = \sqrt{-s^2}$ . The invariance is owned by Lorentz Invariance and Invariants; here it is raw material.

c^2t^2 - x^2 = +c^2\tau^2 \ \ \text{(timelike — calibrates CLOCKS)}, \qquad c^2t^2 - x^2 = -\ell^2 \ \ \text{(spacelike — calibrates RULERS)} \qquad (2)

where: $\tau$ — proper time; $\ell$ — proper length. Physical reading: two families, separated and asymptotically joined by the null cone $x = \pm ct$ . A boost preserves $s^2$ and therefore its sign, so nothing is ever carried from one family to the other — the causal partition of spacetime is not a matter of opinion (Causal Structure of Spacetime).

\big(ct(\varphi),\ x(\varphi)\big) = \big(c\tau\cosh\varphi,\ c\tau\sinh\varphi\big), \quad \cosh\varphi = \gamma,\ \sinh\varphi = \gamma\beta,\ \tanh\varphi = \beta; \qquad \text{unit tick of frame }\beta = \{x = \beta\,ct\} \cap \{(ct)^2 - x^2 = 1\} = (\gamma,\ \gamma\beta) \qquad (3)

where: $\varphi$ — the rapidity; the $ct'$ axis is the moving observer's worldline. Physical reading: the left half is the hyperbola as an orbit — $\varphi = 0$ gives the rest observer's $\tau$ -tick, $\varphi = \mathrm{artanh}\,\beta$ gives that of the observer moving at $\beta$ : one curve, every observer, all at once, which is exactly what a scale bar must be. The right half is the calibration rule — do not measure, intersect. At $\beta = 0.6$ it lands on $(1.25,\ 0.75)$ , with $1.25^2 - 0.75^2 = 1$ exactly; the page-distance to it is $\sqrt{(1+\beta^2)/(1-\beta^2)} = 1.4577$ , not $1$ — a distortion owned by Minkowski Diagrams and Skewed Axes and quoted, not re-derived. Hyperbolic rotations themselves are owned by Rapidity. And the fact that makes the hyperbola inevitable rather than merely convenient:

\boxed{\;f(ct, x)\ \text{is boost-invariant} \iff f = F\big(c^2t^2 - x^2\big)\;} \qquad (4)

where: $F$ is an arbitrary function of one variable. Physical reading: there are no other invariants on the plane, hence no other invariant curves. Any candidate curve that is not a level set of $s^2$ is moved by some boost — and a curve that moves is not a scale bar.

Deriving the structure

Derivation AA boost slides events along the hyperbolae, and can do nothing else

Before the hyperbola can be a ruler it must be shown to be an orbit — that the set of events one proper time $\tau$ from the origin, according to some inertial observer, is exactly one branch of one curve, with no gaps and no extras. That is what lets a single line of ink stand in for the unit ticks of infinitely many observers at once, and it must be an equality of sets, not merely an inclusion.

Step 1. Write the boost as a hyperbolic rotation: $ct' = ct\cosh\varphi - x\sinh\varphi$ , $x' = x\cosh\varphi - ct\sinh\varphi$ , with $\cosh\varphi = \gamma$ , $\sinh\varphi = \gamma\beta$ , $\tanh\varphi = \beta$ . It preserves the interval: $c^2t'^2 - x'^2 = (c^2t^2-x^2)(\cosh^2\varphi - \sinh^2\varphi) = c^2t^2 - x^2$ . Why: the form is forced by $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ and derived in The Lorentz Transformation Equations, and $\cosh^2 - \sinh^2 = 1$ does the job $\cos^2 + \sin^2 = 1$ does for a compass. That $\varphi \to -\varphi$ is the inverse makes "orbit" the right word: the boosts are a one-parameter group.

Step 2. Show the orbit fills the branch. Apply the inverse boost to the rest observer's tick $(c\tau, 0)$ : the image is $E(\varphi) = (c\tau\cosh\varphi,\ c\tau\sinh\varphi)$ . As $\varphi$ runs over $\mathbb{R}$ , $\sinh\varphi$ runs over $\mathbb{R}$ bijectively, and $\cosh\varphi = \sqrt{1+\sinh^2\varphi} > 0$ is then determined — so $\varphi \mapsto E(\varphi)$ is a bijection onto the entire upper branch. Why: this upgrades "the tick lies on the hyperbola" to "the hyperbola is nothing but the set of ticks" — every point of the branch is some observer's $\tau$ -tick, and conversely. Nothing is left over on either side.

Step 3. Read the calibration rule off, and note what the boost cannot do. Frame $\beta$ has $ct'$ axis $x = \beta\,ct$ ; substituting into $(ct)^2 - x^2 = c^2\tau^2$ gives $(ct)^2(1-\beta^2) = c^2\tau^2$ , so the tick is at $(\gamma,\ \gamma\beta)\,c\tau$ — at $\tau = 1$ , $\beta = 0.6$ , the point $(1.25,\ 0.75)$ , with $1.25^2 - 0.75^2 = 1.0000$ . And since $s^2$ is preserved and each branch is a single orbit, no boost of any rapidity moves an event onto a different hyperbola, changes the sign of its $s^2$ , or carries it across the asymptote $x = \pm ct$ . Why: nothing was measured — a line was intersected with a curve, an operation no change of frame can spoil, because "these two loci cross here" names an event. And the last sentence is the light barrier as a fact about curves: $\cosh\varphi$ and $\sinh\varphi$ diverge together, and their ratio reaches $1$ only at $\varphi = \infty$ . *Verified in verify_aq-sr-invariant-hyperbolae.py: $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ with $\eta = \mathrm{diag}(+1,-1)$ ; $c^2t'^2 - x'^2 - (c^2t^2-x^2)$ simplifies symbolically to $0$ ; $(\cosh\varphi, \sinh\varphi)$ satisfies the unit hyperbola identically; the intersection of $x = \beta\,ct$ with it is $(\gamma, \gamma\beta)$ symbolically and $(1.25, 0.75)$ at $\beta = 0.6$ ; $\mathrm{sign}(s^2)$ is boost-invariant over 5000 random samples; and the signature is asserted timelike-positive.*

The owned result: the hyperbola is the only ruler there is

The card above shows the hyperbola works. It does not show that nothing else does — and until that is settled, the calibration hyperbola is a convention dressed up as geometry. It is not. It is forced.

Go deeper · Derivation B · 2 minBoost-invariant means function of s-squared, and therefore the only invariant curves are the hyperbolae

Derivation BBoost-invariant means function of s-squared, and therefore the only invariant curves are the hyperbolae

Uniqueness theorems need the right question. Not "is $s^2$ invariant?" — of course it is — but "is there anything else?" If some other function $f(ct, x)$ were boost-invariant and functionally independent of $s^2$ , its level curves would be a second, competing family of scale bars, and calibration would be ambiguous. The theorem below closes that door.

Step 1. Linearise the group. A boost of infinitesimal rapidity $\mathrm{d}\varphi$ sends $ct \mapsto ct + x\,\mathrm{d}\varphi$ and $x \mapsto x + ct\,\mathrm{d}\varphi$ , so the generator — the flow's velocity field — is $K = x\,\partial/\partial(ct) + ct\,\partial/\partial x$ , and $f$ is constant along every boost orbit iff $Kf = 0$ . Why: a one-parameter group is determined by its generator, and "constant along the flow" is literally "zero derivative along the flow". A question about all boosts at once becomes one linear PDE — and PDEs can be solved. Nothing has been assumed about $f$ beyond differentiability; in particular, nothing about hyperbolae.

Step 2. Solve by characteristics. The flow's own trajectories obey $\mathrm{d}(ct)/\mathrm{d}\varphi = x$ and $\mathrm{d}x/\mathrm{d}\varphi = ct$ , so along any trajectory \frac{\mathrm{d}}{\mathrm{d}\varphi}\big(c^2t^2 - x^2\big) \;=\; 2\,ct\cdot x \;-\; 2x\cdot ct \;=\; 0. So $s^2$ is a first integral, constant on every characteristic — and the characteristics are precisely the curves $s^2 = \text{const}$ : the hyperbolae. Why: the cancellation is the whole theorem in one line. The two terms cancel because the generator swaps $ct$ and $x$ while the invariant subtracts their squares. Change the sign pattern and the cancellation fails: the hyperbola is made of the minus sign.

Step 3. Conclude there is nothing else. Change coordinates to $(s^2,\ \varphi)$ — a genuine coordinate system on each of the four open regions the light cone cuts the plane into — and $K$ becomes simply $\partial/\partial\varphi$ . Then $Kf = 0$ reads $\partial f/\partial\varphi = 0$ , whose solutions are exactly the functions that never mention $\varphi$ : \boxed{\;K f = 0 \iff f(ct, x) = F\big(c^2t^2 - x^2\big)\;} Why: $\varphi$ is the coordinate the group moves you along; $s^2$ is the one it leaves alone. One group parameter, two dimensions, one invariant left over — no room for a second.

Step 4. Harvest the corollary. An invariant curve is a level set of an invariant function; by Step 3 every invariant function is $F(s^2)$ ; so every invariant curve is a level set of $s^2$ — the hyperbolae, and nothing else. The circle is not among them, and the failure is quantitative: $K[c^2t^2 + x^2] = 4\,ct\,x \neq 0$ , and a boost of rapidity $0.5$ carries $(1,0)$ to $(1.1276,\ 0.5211)$ — still on the hyperbola, $\cosh^2 - \sinh^2 = 1$ , but now of Euclidean radius $\mathbf{1.2422} \neq 1$ . Why: this is what "the page lies" actually means, stated as a theorem rather than a warning. The compass is not merely inaccurate — it measures a quantity, Euclidean radius, that no observer agrees on. The hyperbola template is the only instrument on the board whose output is a physical fact. *Verified in verify_aq-sr-invariant-hyperbolae.py: sympy confirms $K[F(c^2t^2-x^2)] = 0$ identically for an arbitrary $F$ ; that $\mathrm{d}(s^2)/\mathrm{d}\varphi = 0$ along the flow; that the change of variables sends $K \to \partial_\varphi$ ; and — the negative half — that $K[c^2t^2+x^2] = 4\,ct\,x \neq 0$ , with the printed $(1.1276, 0.5211)$ , radius $1.2422$ and $s^2 = 1.0000$ recomputed to the printed precision.*

Go deeper · Derivation C · 2 minAnd the same curve is the protractor: the rapidity is its invariant arc length

Derivation CAnd the same curve is the protractor: the rapidity is its invariant arc length

On the unit circle the angle is not merely a parameter — it is the arc length, which is why angles add. Ask the identical question of the calibration hyperbola, but measure the arc with the metric spacetime actually has rather than with the ink on the page, and the answer is as clean as it could be. This makes the hyperbola a complete instrument, ruler and protractor, and it is the geometric reason rapidity — not velocity — is the additive quantity.

Step 1. Parametrise the unit timelike hyperbola ( $c = 1$ , $s^2 = +1$ ): $r(\varphi) = (\cosh\varphi,\ \sinh\varphi)$ , which by the first card sweeps the whole upper branch exactly once.

Step 2. Take the tangent, $r'(\varphi) = (\sinh\varphi,\ \cosh\varphi)$ , and its Minkowski norm: \eta_{\mu\nu}\,r'^{\mu}r'^{\nu} \;=\; \sinh^2\varphi - \cosh^2\varphi \;=\; -1. Negative — so in $(+,-,-,-)$ the tangent is spacelike, and the calibration hyperbola is a spacelike curve. Why: it must be — every point of the curve is at the same timelike interval from the origin, so moving along it is "going sideways", and a spacelike curve carries a proper length, not a proper time. In $(-,+,+,+)$ you would get $+1$ and must flip the extraction rule: same physics, different bookkeeping.

Step 3. Extract the invariant arc element. For a spacelike curve the proper-length element is $\mathrm{d}\ell = \sqrt{-\eta_{\mu\nu}\mathrm{d}x^{\mu}\mathrm{d}x^{\nu}}$ , so $\mathrm{d}\ell = \sqrt{-(-1)}\,\mathrm{d}\varphi = \mathrm{d}\varphi$ . Why: the minus under the root is the same minus that makes $\ell = \sqrt{-s^2}$ in equation (2). The upshot: the parametrisation was already the arc length, exactly as $\theta$ already is on the unit circle.

Step 4. Integrate: \boxed{\;\ell(0 \to \varphi) \;=\; \int_0^{\varphi}\mathrm{d}\varphi' \;=\; \varphi \;=\; \mathrm{artanh}\,\beta\;} The rapidity is the invariant arc length along the calibration hyperbola. Why: the exact hyperbolic transcription of "angle = arc length on the unit circle" — and it explains geometrically the one thing everyone wants from rapidity, that it adds: sliding $\varphi_1$ and then $\varphi_2$ along one curve puts you at $\varphi_1 + \varphi_2$ , because that is what arc length does. The additivity is owned by Rapidity; what is owned here is why it is inevitable.

Step 5. Watch the page disagree. At $\beta = 0.6$ , $\varphi = \mathrm{artanh}\,0.6 = \ln 2 = \mathbf{0.6931}$ : the invariant arc from $(1,0)$ to $(1.25,\ 0.75)$ is exactly $0.6931$ . The same arc under a Euclidean tape reads \ell_{\text{page}} = \int_0^{\ln 2}\sqrt{\sinh^2\varphi + \cosh^2\varphi}\;\mathrm{d}\varphi = \int_0^{\ln 2}\sqrt{\cosh 2\varphi}\;\mathrm{d}\varphi = \mathbf{0.8005}. Why: two numbers for one arc, and only the first is physics. The page adds squares where the metric subtracts them, and the discrepancy grows without bound as $\beta \to 1$ . *Verified in verify_aq-sr-invariant-hyperbolae.py: sympy gives $\eta_{\mu\nu}r'^{\mu}r'^{\nu} = -1$ identically; the proper-length element reduces to $\mathrm{d}\varphi$ and the invariant arc integrates to exactly $\varphi$ for arbitrary $\varphi$ ; $\mathrm{artanh}(0.6) = \ln 2 = 0.693147$ ; and the Euclidean page arc $\int_0^{\ln 2}\sqrt{\cosh 2\varphi}\,\mathrm{d}\varphi = 0.80051$ — both asserted against the printed $0.6931$ and $0.8005$ .*

Figure

PUNCHLINE — three ways to put a tick on the moving observer's time axis at β = 0.6, and only one of them is a tick. COMPASS (unit circle, grey): lands at (0.8575, 0.5145), whose invariant interval from the origin is only 0.6860 — not one second, and a different event in every frame. RULER (one page-unit along the axis): the same failure, differently spelled. HYPERBOLA (ct)² − x² = 1: lands at (1.25, 0.75), interval exactly 1.0000, page-distance 1.4577. The uniqueness theorem says this is not the best of three instruments — it is the ONLY instrument, because the level sets of s² are the only boost-invariant curves there are

Limiting cases

Case	Result	Meaning
$s^2 = +c^2\tau^2 > 0$	timelike branch, opens along $ct$	one proper time $\tau$ from the origin, for every observer at once — the clock calibration; $c\tau = +\sqrt{s^2}$
$s^2 = -\ell^2 < 0$	spacelike branch, opens along $x$	one proper length $\ell$ away — the ruler calibration; $\ell = \sqrt{-s^2}$
$s^2 = 0$	$x = \pm ct$	the degenerate member: the light cone, common asymptote of both families. No calibration — the interval vanishes for everyone
$\varphi \to \infty$	$(\cosh\varphi, \sinh\varphi) \to \tfrac12 e^{\varphi}(1,1)$	the tick runs toward the light line and never arrives. The light barrier, drawn
$\beta \ll 1$	$c^2t^2 - x^2 \approx c^2t^2$	hyperbolae flatten into the lines $t = \text{const}$ : universal simultaneity, and a page you can measure with a ruler
$\eta \to \delta$ (all $+$ )	circles, not hyperbolae	positive definite: orbits close, the compass works, no light barrier. The minus sign is the physics

Common trapThe hyperbola calibrates. It does not derive — and the base literature gets that backwards

The AllQFT source article boxes three results as though they were the hyperbola's: time dilation $t = \gamma\tau$ , length contraction $L = L_0/\gamma$ , and velocity addition $\beta = (\beta_1+\beta_2)/(1+\beta_1\beta_2)$ . None of the three belongs here. They are owned by Time Dilation, Length Contraction and Relativistic Velocity Addition, where they are derived from the light-clock and the Lorentz transformation — not from a picture. What the hyperbola does is read them off: locate the event by intersection, project it, compare the readings. That is genuine and valuable, and worked example 1 does exactly it. But a reading is not a derivation, and treating the hyperbola as the source of these laws inverts the logic: the interval's invariance is what makes the hyperbolae exist, and all three laws are consequences of that same invariance. The hyperbola is downstream of the physics, not upstream of it. Two further confusions. A calibration hyperbola is not a worldline — it is a locus of events at equal interval from the origin, drawn all at once like a contour line; the one exception is the uniformly accelerated (Rindler) observer. And the page-length of a tick is not a length: at $\beta = 0.6$ the primed tick is drawn $1.4577$ page-units out while being one unit, so a ruler makes the moving clock look fast when it is slow by $\gamma = 1.25$ . That distortion is owned by Minkowski Diagrams and Skewed Axes. The rule is one sentence: never measure a Minkowski diagram, always intersect it.

Worked examples

1 · Read a real muon's proper time straight off its hyperbola

Frisch and Smith timed cosmic-ray muons at the Mount Washington summit (1917 m above sea level) and 1907 m below it in Cambridge, at $\beta \approx 0.995$ . Put the arrival event on a diagram and let the hyperbola through it report how much the muon's own clock advanced.

Step 1 — identify. Departure (summit) is the origin; arrival (Cambridge) is at $x = 1907\ \mathrm{m}$ . The lab coordinate time is $t = x/(\beta c)$ , so ct = \frac{1907\ \mathrm{m}}{0.995} = \mathbf{1916.583\ m}, \qquad t = \frac{1916.583}{2.998\times10^{8}} = \mathbf{6.393\ \mu s}.

Step 2 — apply the invariant. Which hyperbola does the arrival event lie on? Compute $s^2$ from the two coordinates, in $(+,-,-,-)$ : s^2 = (1916.583)^2 - (1907)^2 = 3\,673\,290 - 3\,636\,649 = 36\,641\ \mathrm{m^2}, \qquad s = \mathbf{191.42\ m}. Positive, hence timelike — as it must be, since the muon was present at both events — and that hyperbola crosses the lab's own time axis at $ct = 191.42\ \mathrm{m}$ .

Step 3 — the boundary number. The intercept is the muon's proper time, because $c\tau = +\sqrt{s^2}$ in this signature: $\tau = 191.42/(2.998\times10^{8}) = \mathbf{0.6385\ \mu s}$ . Cross-check: $\gamma = 1/\sqrt{1-0.995^2} = 10.0125$ , and $t/\gamma = 6.393/10.0125 = 0.6385\ \mu\mathrm{s}$ . The routes agree to every digit — but the hyperbola route never mentioned $\gamma$ . It read the proper time off the geometry of a single event.

Step 4 — the physics that hangs on it. The muon's rest lifetime is $\tau_0 = 2.197\ \mu\mathrm{s}$ . Survival over the proper time $0.6385\ \mu\mathrm{s}$ is $e^{-0.6385/2.197} = \mathbf{0.748}$ — three quarters arrive. Survival over the coordinate time $6.393\ \mu\mathrm{s}$ , which is what you compute if you have never heard of the interval, is $e^{-6.393/2.197} = \mathbf{0.0545}$ — one in twenty. The two differ by a factor of $\mathbf{13.7}$ , and the mountain settles it: Frisch and Smith counted $563$ muons per hour at the summit and $412$ per hour in Cambridge, a ratio of $\mathbf{0.732}$ — against $0.748$ from the hyperbola and about $0.048$ (roughly $27$ per hour) classically. The clock that matters is the one on the invariant curve.

Physics of the portrait

The hyperbola is older than physics, and its journey to the centre of spacetime geometry took a little over two thousand years.

Origins · 1 minHyperbola and how this was discovered

Hyperbola

c. 200 BC – 1966 · Apollonius names the curve · Riccati and Lambert give it functions · Minkowski makes it the ruler of spacetime · Taylor and Wheeler put it in every undergraduate's hands

Conic sections were obtained by Menaechmus around 350 BC, reportedly while attacking the duplication of the cube, and given their systematic theory — and the names that survive — in the eight books of the Conics of Apollonius of Perga (c. 262–190 BC). Ellipse, parabola and hyperbola come to us from that treatise, from the Greek for "falling short", "alongside" and "exceeding": a hyperbola exceeds, its eccentricity is greater than one, and it does not close. (Historians enter a caveat: an Arabic translation of Diocles's On Burning Mirrors, recovered in the 1970s, suggests the names may predate Apollonius.) For twenty centuries the curve was a shape. It acquired functions in 1757, when Vincenzo Riccati introduced the hyperbolic sine and cosine at Bologna, and a workable notation from Johann Heinrich Lambert in the 1760s. It became an instrument on 21 September 1908, when Hermann Minkowski, lecturing at Cologne, drew the invariant hyperbola on a spacetime diagram and calibrated a moving observer's axes with it; it reached undergraduates in force with Taylor and Wheeler's Spacetime Physics (1966). The identification is not decorative — the curve's defining property, that it exceeds, running to infinity without meeting its asymptotes, is precisely the statement that $c$ is unattainable.

(The curve's pre-history — Menaechmus around 350 BC, Apollonius around 200 BC — is in the portrait above; the timeline picks it up where it becomes physics.)

Origins · 1 minhow this was discovered

History

1757

— Vincenzo Riccati introduces the hyperbolic functions sinh and cosh at Bologna; Johann Heinrich Lambert develops and popularises them through the 1760s.

1905

— Einstein's Zur Elektrodynamik bewegter Körper (Annalen der Physik 17, 891) derives the Lorentz transformation from two postulates, without identifying the invariant quadratic form it preserves.

1908

— Minkowski's Raum und Zeit, delivered at Cologne on 21 September, introduces the spacetime diagram and the calibration hyperbola.

1966

— Taylor and Wheeler's Spacetime Physics builds special relativity on the invariant interval and the diagram, making the calibration hyperbola standard undergraduate equipment.

Going further

Texts. Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 1 · Rindler, Introduction to Special Relativity, 2nd ed., §§7–9. Primary source: Minkowski, Raum und Zeit (Cologne, 21 September 1908), Physikalische Zeitschrift 10 (1909) 104.

Provenance. The base was already in the locked signature $(+,-,-,-)$ and no sign error was found — but its structure was wrong: two of its three derivation cards box a neighbour's result (time dilation, owned by Time Dilation; length contraction, owned by Length Contraction), and the third boxes velocity addition (owned by Relativistic Velocity Addition). All three are retained here only as readings off the calibration curve, with the owner named. In their place this article proves what the base never states: the uniqueness theorem $Kf = 0 \iff f = F(s^2)$ and the arc-length identification of rapidity. The page-length $\sqrt{(1+\beta^2)/(1-\beta^2)}$ is owned by Minkowski Diagrams and Skewed Axes and quoted, not re-derived. Every printed number is recomputed in verify_aq-sr-invariant-hyperbolae.py — the $\beta = 0.6$ anchors, the compass tick and its interval $0.6860$ , the $\varphi = 0.5$ boost image and its radius $1.2422$ , the invariant arc $0.6931$ against the page arc $0.8005$ , and the Frisch–Smith muon chain ending at $\tau = 0.6385\ \mu\mathrm{s}$ and a survival fraction $0.748$ . Signature: $(+,-,-,-)$ , timelike-positive.

Builds on. Spacetime Diagrams · The Spacetime Interval · Lorentz Invariance and Invariants · The Minkowski Metric · Hyperbolic Geometry of Boosts · Minkowski Diagrams and Skewed Axes. Leads to. Proper Time and Proper Length · Rapidity · Causal Structure of Spacetime · The Light Cone · Simultaneity Lines and Tilted Axes.

Why the hyperbola rather than an ellipse, or a straight line at the right distance?

Because it is the only invariant curve there is — a theorem, not a preference. A calibration curve must have the property that no boost can move a point off it; otherwise the "unit tick" it marks is a different event for a different observer, which is exactly what a scale bar is forbidden to be. Invariant curves are the level sets of boost-invariant functions, and every boost-invariant function on the $(ct,x)$ plane is a function of $s^2$ and nothing else. An ellipse is not among them; nor is the circle, whose failure is quantified above; nor is any straight line except the two null ones.

The moving observer's tick is drawn 1.458 times further out than mine. Doesn't that mean his second is longer?

No — and this is the commonest way of reading a Minkowski diagram backwards. The event $(1.25,\ 0.75)$ is one unit of his proper time from the origin, since $1.25^2 - 0.75^2 = 1$ exactly. But it sits at $ct = 1.25$ of your coordinate time: while his clock advanced by $1$ , yours advanced by $1.25$ , so his clock is slow by exactly $\gamma = 1.25$ , as Time Dilation requires. The page-distance $1.4577$ is a Euclidean measurement of a Minkowskian object and has no physical meaning — like the area of Greenland on a Mercator map.

Is the invariant hyperbola a worldline? Does anything actually travel along it?

Ordinarily no: it is a locus of events at equal interval from the origin, drawn all at once like a contour line. But there is one exact exception. An observer with constant proper acceleration $a$ traces the invariant hyperbola $x^2 - c^2t^2 = (c^2/a)^2$ as his actual worldline. He approaches a light line asymptotically without reaching it, and that line becomes a horizon — a boundary beyond which signals sent to him never arrive, produced by nothing more exotic than acceleration in flat spacetime. That is the Rindler observer, forward-linked to the accelerated-frames chapter.

Why two families of hyperbolae rather than one?

Because $s^2$ is indefinite — positive, negative or zero — and that is the signature doing the only thing signatures do. The timelike family lives inside the light cone, one proper time from the origin: it calibrates clocks. The spacelike family lives outside, one proper length away: it calibrates rulers. A boost preserves $s^2$ and hence its sign, so nothing crosses between them. A positive-definite geometry has only one sign available, hence one family — circles — and hence no light cone, no barrier, no causal structure. The second family is the minus sign made visible.

What exactly does the theorem Kf = 0 ⟺ f = F(s²) rule out?

A second, independent invariant — and with it, any alternative scale bar. Suppose someone proposed calibrating with the level curves of a function $g(ct,x)$ claimed to be frame-independent. Either $g$ is secretly a function of $s^2$ , in which case its level sets are the same hyperbolae and nothing new has been proposed; or $g$ is not boost-invariant, in which case some boost moves a point off its level set and the "calibration" is frame-dependent. There is no third option. The counting: two dimensions, a one-parameter group, exactly one function's worth of information left alone — and that leftover is $s^2$ .

Rapidity is the arc length along the hyperbola. Why does that make it the natural velocity variable?

Because arc length adds. Slide $\varphi_1$ along a curve and then a further $\varphi_2$ and you have gone $\varphi_1 + \varphi_2$ — true of any curve, and the whole reason angles add on a circle. Boosts are slides along the hyperbola, and the invariant length of the slide is exactly the rapidity, so successive boosts add their rapidities. Velocities, being $\tanh$ of rapidities, do not — and the discrepancy is the velocity-addition law. Rapidity owns the additivity; this article contributes the reason it is inevitable: you are measuring an arc.