The Invariant Spacetime Interval

Relativity spends its first three chapters taking things away from you. Duration: gone — a moving clock ticks slow. Distance: gone — a moving ruler is short. Simultaneity: gone, and worst of all, because now two observers cannot even agree on what "at the same time" picks out. By the end of the Lorentz transformation almost nothing you took for a fact about the world has survived as one; it has all turned out to be a fact about you. So the question that turns a bag of tricks into a theory is: what is left? Is there any number two inertial observers can compute from the same two events and be guaranteed to agree on? There is exactly one — and it is neither the time between them nor the distance between them, but a lopsided combination of the two —

s^{2} = c^{2}\Delta t^{2} - \Delta x^{2} - \Delta y^{2} - \Delta z^{2}

— which looks like Pythagoras with a printing error. That minus sign is the entire theory. It is why fast clocks run slow and fast rulers shrink by exactly compensating amounts, so that this one combination comes out the same for everyone. A cosmic-ray muon is born high in the atmosphere and dies $1907\ \mathrm{m}$ lower down. In the mountain's frame that took $6.39\ \mathrm{\mu s}$ ; in the muon's own frame it took $0.639\ \mathrm{\mu s}$ , and it never moved at all. The two observers agree on nothing — not the time, not the distance. They agree that $s^2 = 36\,641\ \mathrm{m^2}$ . That number is the muon's real, frame-free separation between being born and dying, and this article is about why it exists, why it is unique, and why demanding it is enough to force the whole of special relativity into being.

Key takeaways

Between any two events, the quantity $s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2$ has the same value in every inertial frame, even though $\Delta t$ and $\Delta x$ separately do not.
Signature $(+,-,-,-)$ , timelike-positive: $s^2 > 0$ timelike, $s^2 = 0$ lightlike, $s^2 < 0$ spacelike. Half the literature uses the opposite sign; every sign flips and no physics does.
Invariance does not need the Lorentz transformation. It follows from the two postulates plus homogeneity and isotropy alone: constancy of $c$ preserves the surface $s^2 = 0$ , linearity then forces $s'^2 = a(v)\,s^2$ , and isotropy forces $a = 1$ .
The street runs both ways: demanding that a linear map preserve $s^2$ forces the Lorentz transformation. And $s^2$ is the only independent quadratic invariant of a boost — there is no second conserved combination waiting to be found.
For timelike pairs $c\Delta\tau = \sqrt{s^2}$ is the proper time; for spacelike pairs $\sqrt{-s^2}$ is the proper length.
Key idea: different observers slice spacetime into "time" and "space" differently, so they disagree about $\Delta t$ and $\Delta x$ individually — but the particular combination $c^2\Delta t^2 - \Delta x^2$ is the same for all of them. That invariant is the true geometric separation of two events, and it is the reason relativity is a geometry rather than a list of effects.

The core insight

Euclidean geometry has rotations, and rotations change every coordinate of a point while preserving $x^2 + y^2$ . Minkowski geometry has boosts, and boosts change every coordinate of an event while preserving $c^2t^2 - x^2$ . Time dilation and length contraction are not two independent oddities: they are the two halves of one conspiracy, tuned so that when a clock slows and a ruler shrinks the combination $s^2$ does not move at all.

Where this lives today

Every collider experiment is, at bottom, an interval calculator. When the LHC collides two protons the useful question is not "how much energy is in the lab frame?" but "how much is available in the centre of mass?" — and the answer is an invariant, the Mandelstam $s = (E_1 + E_2)^2/c^2 - |\mathbf{p}_1 + \mathbf{p}_2|^2$ , which is equation (1) applied to four-momenta instead of four-positions. Same minus sign, same job: picking out the one combination every observer agrees on. Particle physicists spend their lives computing invariant masses, and every one is $\sqrt{s^2}$ in disguise. The interval is also the working tool of gravitational-wave astronomy and of GPS. In general relativity $\mathrm{d}s^2 = g_{\mu\nu}\mathrm{d}x^\mu\mathrm{d}x^\nu$ with a position-dependent metric, and a passing gravitational wave is a ripple in those coefficients — LIGO measures the interval between its mirrors and watches it change. A GPS satellite's clock accumulates proper time along its orbit, and your receiver corrects for the gap between that and ground time: about $+38.6\ \mathrm{\mu s}$ per day, of which special relativity contributes $-7.2$ and gravity $+45.8$ . Without the correction, positions would drift by kilometres within a day. The minus sign in equation (1) is quite literally load-bearing.

The physical picture

The invariant is best met by watching two observers fail to agree about everything else. Take an event $E$ that, in your reference frame, occurs $3\ \mathrm{m}$ from the chosen spatial origin and $5\ \mathrm{m}$ of light-travel time later — $\Delta x = 3\ \mathrm{m}$ , $c\Delta t = 5\ \mathrm{m}$ . (Measuring time in metres of light-travel is the single best habit in this subject: it sets $c = 1$ and puts both axes in the same units.) Then $s^2 = 5^2 - 3^2 = 16\ \mathrm{m^2}$ — positive, so timelike, and $\sqrt{s^2} = 4\ \mathrm{m}$ is the proper time a particle present at both events would clock. Now hand the same two events to an observer moving at $\beta = 0.6$ . She finds $c\Delta t' = 4\ \mathrm{m}$ and $\Delta x' = 0$ : for her they happen in the same place, only $4\ \mathrm{m}$ of time apart. Hand them to a third observer at $\beta = -0.6$ and he reports $c\Delta t'' = 8.5\ \mathrm{m}$ , $\Delta x'' = 7.5\ \mathrm{m}$ — nearly double the time, two and a half times the distance. Three observers, three irreconcilable descriptions. And $4^2 - 0^2 = 16$ , and $8.5^2 - 7.5^2 = 72.25 - 56.25 = 16$ . Every one of them gets $16\ \mathrm{m^2}$ .

Figure

The invariant hyperbolae. The timelike event E sits on (ct)² − x² = 16 m² and the spacelike event F on x² − (ct)² = 16 m². A boost SLIDES an event along its own hyperbola — it never lifts it off. Read E in three frames and you get three different coordinate pairs, all with s² = +16 m²; read F and you get three more, all with s² = −16 m². The hyperbola is the set of all the ways one event can look

Look at what the boost actually does. It does not move the event — the event is the muon decaying, a physical happening, and no change of coordinates can undo it. What changes is the grid laid over spacetime, and as the grid tilts, the event's coordinates slide along the hyperbola it lives on. That is the geometric content of the whole theory: a boost is the transformation that slides events along hyperbolae of constant $s^2$ , exactly as a rotation slides points along circles of constant $r^2$ . The hyperbolae never cross the light lines, which is why no boost can turn a timelike separation into a spacelike one, and why causality survives.

Figure

PUNCHLINE — three observers, three totally different accounts of the SAME pair of events, one number they all agree on. The blue and red bars (cΔt and Δx) swing wildly from frame to frame; the gold line, s² = c²Δt² − Δx², never moves. Time dilation and length contraction are not two effects. They are one conspiracy, tuned to hold this bar fixed

The core equations

The definition, and the convention that goes with it. Write it once and never think about it again.

s^{2} \;=\; c^{2}\Delta t^{2} - \Delta x^{2} - \Delta y^{2} - \Delta z^{2} \qquad (1)

where: $s^2$ — the invariant interval (squared), which may be positive, negative or zero; $c$ — the invariant speed of light, $299\,792\,458\ \mathrm{m/s}$ exactly; $\Delta t, \Delta x, \Delta y, \Delta z$ — the coordinate separations of the two events in one chosen inertial frame. Signature $(+,-,-,-)$ : timelike $s^2 > 0$ , lightlike $s^2 = 0$ , spacelike $s^2 < 0$ . Physical reading: the minus signs are not decoration. Had all four terms carried the same sign, $s^2$ would be a Euclidean distance in four dimensions, time would be just another direction, and nothing would distinguish cause from effect. The single relative minus sign is what carves spacetime into a past, a future, and an unreachable elsewhere. The claim the whole article rests on is that this number does not care which frame you computed it in.

c^{2}\Delta t'^{2} - \Delta x'^{2} - \Delta y'^{2} - \Delta z'^{2} \;=\; c^{2}\Delta t^{2} - \Delta x^{2} - \Delta y^{2} - \Delta z^{2} \qquad (2)

where: primes denote a second inertial frame $S'$ in standard configuration with $S$ . Physical reading: individually, every term on the left differs from its partner on the right — that is time dilation and length contraction. The equation says the differences are exactly correlated. Note what is not being claimed: $\Delta t$ is not invariant, $\Delta x$ is not invariant, and $\Delta t^2 - \Delta x^2/c^2$ is not "the time in some true frame". Only the combination is anything at all. In differential form, which is the version that survives into general relativity when the coefficients start depending on position:

\mathrm{d}s^{2} = c^{2}\mathrm{d}t^{2} - \mathrm{d}x^{2} - \mathrm{d}y^{2} - \mathrm{d}z^{2} = \eta_{\mu\nu}\,\mathrm{d}x^{\mu}\mathrm{d}x^{\nu}, \qquad \eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1) \qquad (3)

where: $x^\mu = (ct, x, y, z)$ — the four coordinates, $\mu = 0,1,2,3$ ; $\eta_{\mu\nu}$ — the Minkowski metric; repeated indices are summed. Physical reading: this is the sentence "spacetime is flat and has signature $(+,-,-,-)$ ", written in the only language that generalises. Replace the constant $\eta_{\mu\nu}$ by a metric $g_{\mu\nu}(x)$ with non-vanishing curvature and you have gravity; everything else in the formalism stays where it is. Finally, what the number means on each side of the light cone. This is the reason it is worth computing at all.

c\,\Delta\tau = \sqrt{s^{2}} \;\;(s^2 > 0), \qquad L_0 = \sqrt{-s^{2}} \;\;(s^2 < 0) \qquad (4)

where: $\Delta\tau$ — the proper time, the reading of a clock actually present at both events; $L_0$ — the proper length, the distance in the frame where the events are simultaneous. Physical reading: $s^2$ is not an abstraction but a clock reading or a ruler reading, depending on its sign. For a timelike pair, boost to the frame where the events happen in the same place ( $\Delta x' = 0$ ): then $s^2 = c^2\Delta t'^2$ , literally the elapsed time on the co-moving clock. For a spacelike pair, boost to where they are simultaneous ( $\Delta t' = 0$ ): then $s^2 = -\Delta x'^2$ , literally the distance. Proper time and proper length are one quantity, read on the two sides of the cone.

Where the invariance comes from

Derivation AInvariance from the two postulates alone — no Lorentz transformation assumed

It is easy, and slightly circular-feeling, to prove that $s^2$ is invariant by substituting the Lorentz transformation and watching the algebra cancel. That proof is worth having and we give it as a corollary below, but it inverts the logic: it makes the interval look like an accident of some equations that fell out of the sky. The honest derivation runs the other way. Assume only the two postulates — the principle of relativity, and the constancy of $c$ — plus the homogeneity and isotropy of spacetime. Nothing about $\gamma$ , nothing about boosts. The invariance of $s^2$ drops out, for all separations, not merely for light.

Step 1. Take two events joined by a light pulse. In $S$ the pulse travels at $c$ , so $|\Delta \mathbf{x}| = c\,\Delta t$ and hence $s^2 = 0$ . By the second postulate the pulse also travels at $c$ in $S'$ , so $s'^2 = 0$ as well. The vanishing of the interval is therefore frame-independent: $s^2 = 0 \iff s'^2 = 0$ . Why: this is the entire empirical content of "the speed of light is the same for everybody", restated as a geometric fact — the light cone is a structure every observer draws identically.

Step 2. Spacetime is homogeneous — no event is special — so the map from $S$ to $S'$ must be linear; otherwise the transformation would depend on where you put your origin, which is exactly what homogeneity forbids. A linear map sends the quadratic form $s^2$ to another quadratic form. Why: linearity is not an extra postulate smuggled in — it is what "no preferred place, no preferred moment" means in equations.

Step 3. So $s'^2$ is a quadratic form vanishing exactly where $s^2$ vanishes — on the same cone. Two quadratic forms in four variables with the same non-degenerate zero set must be proportional, so $s'^2 = a\,s^2$ , with $a$ depending only on the relative velocity and, by isotropy, only on its magnitude: $a = a(|v|)$ . Why: this is the crux. The postulate constrains only the cone, and proportionality then propagates that constraint off the cone, to timelike and spacelike separations that have nothing to do with light.

Step 4. Apply the boost from $S$ to $S'$ and then back from $S'$ to $S$ . The composite is the identity, so $a(|v|)\cdot a(|-v|) = 1$ . But isotropy already told us $a$ depends only on $|v|$ , so $a(|-v|) = a(|v|)$ , giving $a(|v|)^2 = 1$ , so $a = \pm 1$ . Continuity with the identity at $v = 0$ , where $a(0) = 1$ , rules out $-1$ . Therefore $a = 1$ and $s'^2 = s^2$ . Why: the theorem is proved without ever writing a $\gamma$ . Invariance of the interval is not a property of the Lorentz transformation — it is a property of spacetime, and the Lorentz transformation is what you get when you insist on it.

Step 5. As a corollary, check it by brute force: substituting $c\Delta t' = \gamma(c\Delta t - \beta\Delta x)$ and $\Delta x' = \gamma(\Delta x - \beta c\Delta t)$ , the cross terms cancel identically and $c^2\Delta t'^2 - \Delta x'^2 = \gamma^2(1-\beta^2)(c^2\Delta t^2 - \Delta x^2) = c^2\Delta t^2 - \Delta x^2$ . Why: worth doing once, to see the whole miracle reduce to the defining identity of $\gamma$ — but remember which is the theorem and which is the check. *Two checks confirm the argument: the difference $c^2\Delta t'^2 - \Delta x'^2 - (c^2\Delta t^2 - \Delta x^2)$ simplifies to zero identically in $\beta$ ; and $a(v)a(-v) = 1$ with $a$ even forces $a = 1$ .*

Go deeper · Derivation B · 2 mins² is the ONLY quadratic invariant of a boost in 1+1 dimensions

Derivation Bs² is the ONLY quadratic invariant of a boost in 1+1 dimensions

A reasonable suspicion, on first meeting the interval, is that it might be one of several conserved combinations — that with enough cleverness one could find a second invariant, and a third, and recover some of the absolutes relativity took away. This card kills that hope. In one space and one time dimension the only quadratic form preserved by all boosts is $s^2$ , up to an overall constant. That uniqueness is what promotes $s^2$ from "a useful combination" to "the metric of spacetime", and it is the owned result of this article.

Step 1. Write the most general quadratic form in $(ct, x)$ : $Q = A\,(ct)^2 + 2B\,(ct)x + C\,x^2$ , which is $\mathbf{v}^{\mathsf T} M \mathbf{v}$ with $\mathbf{v} = (ct, x)^{\mathsf T}$ and the symmetric matrix $M = \begin{pmatrix} A & B \\ B & C\end{pmatrix}$ . Why: three free coefficients, and we are about to see how many the boost kills.

Step 2. Demand invariance under every boost: $Q(\Lambda\mathbf{v}) = Q(\mathbf{v})$ for all $\mathbf{v}$ and all $\Lambda$ , i.e. $\Lambda^{\mathsf T} M \Lambda = M$ for every boost matrix $\Lambda$ . Why: "invariant" means invariant under the whole one-parameter family, not just one particular speed — and that is a far stronger condition than it looks.

Step 3. Because the boosts form a one-parameter group through the identity, it suffices to impose the condition infinitesimally. Write $\Lambda \approx I + \varepsilon K$ with the boost generator $K = \begin{pmatrix} 0 & -1 \\ -1 & 0\end{pmatrix}$ . To first order in $\varepsilon$ , $\Lambda^{\mathsf T} M \Lambda = M + \varepsilon(K^{\mathsf T}M + MK)$ , so invariance requires $K^{\mathsf T}M + MK = 0$ , and since $K$ is symmetric, $KM + MK = 0$ . Why: a condition holding for all $\varepsilon$ must hold at first order, and at first order a matrix equation is linear — which is why the whole classification takes three lines instead of three pages.

Step 4. Multiply out. $KM = \begin{pmatrix} -B & -C \\ -A & -B\end{pmatrix}$ and $MK = \begin{pmatrix} -B & -A \\ -C & -B\end{pmatrix}$ , so $KM + MK = \begin{pmatrix} -2B & -(A+C) \\ -(A+C) & -2B \end{pmatrix} = 0$ . Reading off: $B = 0$ and $C = -A$ . Why: two of the three coefficients are destroyed outright. The cross term must vanish, and the time and space coefficients must be equal and opposite — the minus sign is not a choice we made, it is the only thing the boost permits.

Step 5. Conclude. $M = A\,\mathrm{diag}(1,-1) = A\,\eta$ , so $Q = A\,(c^2t^2 - x^2) = A\,s^2$ . Every boost-invariant quadratic form is a multiple of $s^2$ ; fixing the scale by demanding that a light-second of time gives $s^2 = c^2$ sets $A = 1$ . There is exactly one invariant, and we already have it. Why: this is why spacetime has a metric and not a family of them, and why $\eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ is forced rather than chosen. The overall sign remains free — which is precisely the freedom the other signature convention exercises. *Two checks confirm the result: solving $KM + MK = 0$ for symmetric $M$ returns $B = 0$ , $C = -A$ ; the resulting $M$ satisfies $\Lambda^{\mathsf T} M \Lambda = M$ exactly (not just infinitesimally) for symbolic $\beta$ , and a randomly chosen non-proportional $M$ fails this condition.*

Go deeper · Derivation C · 2 minThe converse — demanding invariance of s² forces the Lorentz transformation

Derivation CThe converse — demanding invariance of s² forces the Lorentz transformation

The two postulates give the interval. This card runs the implication backwards: hand someone the interval and nothing else, ask which linear maps preserve it, and the Lorentz transformation is what they must hand back. That is what makes $s^2$ a foundation rather than a consequence — either end of the street may be taken as the axiom, and you arrive at the same theory. (The full geometric development — the boost as a hyperbolic rotation, the generator, the invariant hyperbolae — is owned by Lorentz Boosts as Hyperbolic Rotations; here we need only the counting.)

Step 1. Let $\Lambda$ be a linear map of $(ct,x)$ preserving $s^2$ . In matrix language, with $\eta = \mathrm{diag}(1,-1)$ , that is exactly the condition $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ . Why: this one equation is "the interval is invariant", written so that a computer can check it.

Step 2. Count. $\Lambda$ has four entries. The condition $\Lambda^{\mathsf T}\eta\Lambda = \eta$ is an equation between symmetric $2\times2$ matrices, so it carries three independent constraints. Four unknowns minus three constraints leaves a one-parameter family. Why: this is already the punchline. The set of interval-preserving maps is one-dimensional — so a boost is specified by a single number, and that number will turn out to be the velocity.

Step 3. Solve. Writing $\Lambda = \begin{pmatrix} a & b \\ d & e\end{pmatrix}$ , the three constraints read $a^2 - d^2 = 1$ , $e^2 - b^2 = 1$ , and $ab = de$ . Taking the branch connected to the identity ( $a, e > 0$ , $\det \Lambda = +1$ ) the general solution is $a = e$ and $b = d$ , with $a^2 - b^2 = 1$ . Why: the first constraint says the columns are "unit vectors" in the Minkowski sense; $a^2 - b^2 = 1$ is a hyperbola, which is why one parameter suffices to run along it.

Step 4. Name the parameter physically. The spatial origin of $S'$ is $x' = 0$ , i.e. $d\,(ct) + e\,x = 0$ , giving $x/(ct) = -d/e \equiv \beta$ . Setting $a = e = \gamma$ and $b = d = -\gamma\beta$ , the constraint $a^2 - b^2 = 1$ becomes $\gamma^2(1-\beta^2) = 1$ , that is $\gamma = 1/\sqrt{1-\beta^2}$ . Why: nothing was assumed about light, clocks or rods. The Lorentz factor emerged as the solution of a hyperbola constraint, and $\beta$ is simply the speed of the origin of the new frame.

Step 5. Read off the result: $\Lambda(\beta) = \gamma\begin{pmatrix} 1 & -\beta \\ -\beta & 1\end{pmatrix}$ — precisely the Lorentz transformation. Why: the street runs both ways. Special relativity can be founded on "the speed of light is invariant" or on "this quadratic form is invariant", and the two are the same axiom in different clothes. *Two checks confirm the derivation: solving $\Lambda^{\mathsf T}\eta\Lambda = \eta$ for symbolic entries recovers $\gamma^2(1-\beta^2) = 1$ ; the resulting $\Lambda(\beta)$ matches the standard boost matrix and has determinant $+1$ .*

Limiting cases

Case	Result	Meaning
Same place, $\Delta x = 0$	$s^2 = c^2\Delta t^2 > 0$	timelike; $\sqrt{s^2}/c$ is the proper time on a clock sitting there
Same time, $\Delta t = 0$	$s^2 = -\Delta x^2 < 0$	spacelike; $\sqrt{-s^2}$ is a proper length. Two simultaneous events are always spacelike
Light pulse	$s^2 = 0$	$\|\Delta \mathbf{x}\| = c\Delta t$ in every frame — the second postulate, restated geometrically
$\beta \to 0$	$s'^2 \to s^2$ trivially	the identity boost; invariance is empty until the frames actually differ
Newtonian limit, $c \to \infty$	$s^2/c^2 \to \Delta t^2$	the spatial term drowns; the invariant degenerates into absolute time, and Galileo is recovered
$s^2 = 0$ but $\Delta \mathbf{x} \neq 0$	events distinct, "zero separation"	the null cone: a photon's interval between emission and absorption is zero, however far it flew

Common traps² is not a distance, and s is not always a real number

The notation invites a fatal habit: reading $s^2$ as "the square of something" and therefore as necessarily positive. It is not. $s^2$ is a single symbol for a signed quantity that is positive for timelike pairs, negative for spacelike ones, and zero on the light cone. Its "square root" is real only on one side. Write $\sqrt{s^2}$ for a spacelike pair and you have an imaginary proper time, which is not a physical statement but a signal that you should have written $\sqrt{-s^2}$ and called it a length. The related trap is to read the interval as "the distance the particle travelled". It is not. It is a property of the pair of events, computed from their coordinate separations, and it knows nothing about the path between them. A muon that flies straight from birth to decay and one that zig-zags between the same two events have the same $s^2$ — but different proper times, because proper time along a curve is an integral of $\mathrm{d}s$ , not the $s$ of the endpoints. They coincide only for straight, uniform motion, and that gap is the seed of the twin paradox.

Common trapInvariant does not mean constant, and it does not mean absolute time

" $s^2$ is the same in all frames" is often mangled into " $s^2$ is the real time between the events, and $\Delta t$ is an illusion". Both halves are wrong. $\Delta t$ is not an illusion: it is exactly what a network of synchronized clocks at rest in your frame will read, and if you are designing a particle beam it is the number you must use. And $s^2$ is not "the real time": it is the time read by a clock present at both events, which is a different physical thing, and which exists at all only when the separation is timelike. The correct statement is narrower. Different frames slice a single four-dimensional structure into "time" and "space" along different planes: they disagree about the slices, and agree about the structure. $s^2$ is a fact about the structure; $\Delta t$ is a fact about the slicing. Both are real; only one is shared.

Worked examples

1 · The interval a cosmic-ray muon lives through

In the Frisch–Smith experiment, muons produced in the upper atmosphere are counted at the summit of Mount Washington ( $1917\ \mathrm{m}$ above sea level) and $1907\ \mathrm{m}$ below it in Cambridge, Massachusetts. They travel at $\beta = 0.995$ . Take the muon's creation and its arrival at the lower detector as two events. Compute $s^2$ , and get the proper time from it.

Step 1 — identify the events and the frame. In the mountain's frame the two events are separated by $\Delta x = 1907\ \mathrm{m}$ (the drop between the detectors — not the summit's $1917\ \mathrm{m}$ elevation, which is a different quantity) and by a time $\Delta t = \Delta x/(\beta c)$ .

Step 2 — put time into metres. $c\Delta t = \Delta x/\beta = 1907/0.995 = 1916.58\ \mathrm{m}$ . In ordinary units that is $\Delta t = 6.393\ \mathrm{\mu s}$ — the flight time.

Step 3 — the interval. $s^2 = (c\Delta t)^2 - (\Delta x)^2 = 1916.58^2 - 1907^2 = \mathbf{36\,641\ m^2}$ . Positive, so the pair is timelike — as it had better be, since a muon was present at both events. Note how nearly it cancels: the two squares differ by less than one percent of either, which is exactly what "the muon is moving at $99.5\%$ of $c$ " means geometrically.

Step 4 — read the proper time off the interval. $\sqrt{s^2} = \sqrt{36\,641} = 191.42\ \mathrm{m}$ , so $\Delta\tau = 191.42/c = \mathbf{0.6385\ \mu s}$ — the muon's own elapsed lifetime between birth and detection.

Step 5 — cross-check against time dilation, which we never used. $\gamma = 1/\sqrt{1 - 0.995^2} = 10.013$ , and $\Delta t/\gamma = 6.393/10.013 = 0.6385\ \mathrm{\mu s}$ ✓. The two routes agree to every digit and are genuinely independent — the first used only the invariance of $s^2$ and never mentioned dilation. Against a proper lifetime of $2.20\ \mathrm{\mu s}$ , a flight of $0.64\ \mathrm{\mu s}$ of its time is a comfortable fraction of a lifetime, which is why the muons arrive.

2 · One event, three frames, one number

An event $E$ has coordinates $(c\Delta t, \Delta x) = (5.0,\ 3.0)\ \mathrm{m}$ relative to the origin in frame $S$ . Find its coordinates in frames moving at $\beta = 0.6$ and $\beta = -0.6$ , and verify the interval.

Step 1 — the invariant in $S$ . $s^2 = 5^2 - 3^2 = 16\ \mathrm{m^2}$ . Timelike. Proper time $c\Delta\tau = 4.0\ \mathrm{m}$ .

Step 2 — boost to $\beta = 0.6$ , where $\gamma = 1.25$ . $c\Delta t' = \gamma(c\Delta t - \beta\Delta x) = 1.25(5 - 1.8) = 4.0\ \mathrm{m}$ and $\Delta x' = \gamma(\Delta x - \beta\, c\Delta t) = 1.25(3 - 3) = \mathbf{0}$ . This observer sees the two events happen at the same place, and $4.0\ \mathrm{m}$ apart in time. She is the co-moving observer — and no accident: the event lies at $\Delta x/c\Delta t = 3/5 = 0.6$ , so a frame moving at exactly $0.6c$ rides along with it.

Step 3 — check. $s'^2 = 4^2 - 0^2 = 16\ \mathrm{m^2}$ ✓. And note the payoff: for this frame the interval is manifestly the proper time, because there is no spatial part left to subtract. Every timelike interval is the proper time of the frame that co-moves with it.

Step 4 — boost to $\beta = -0.6$ , the other way. $c\Delta t'' = 1.25(5 + 1.8) = 8.5\ \mathrm{m}$ , $\Delta x'' = 1.25(3 + 3) = 7.5\ \mathrm{m}$ . Wildly different: nearly double the time, two and a half times the distance.

Step 5 — check again. $s''^2 = 8.5^2 - 7.5^2 = 72.25 - 56.25 = 16\ \mathrm{m^2}$ ✓. Three observers; three irreconcilable accounts of when and where; one shared number. The three coordinate pairs $(5,3)$ , $(4,0)$ and $(8.5,7.5)$ all lie on the same hyperbola $(c\Delta t)^2 - \Delta x^2 = 16$ , which is precisely the statement that a boost slides an event along its hyperbola and never off it.

Physics of the portrait

The interval is one of the rare cases where the mathematician arrived after the physicist and understood the physics better. Einstein had the transformation in 1905, and with it every observable consequence, but he presented relativity as a theory of measuring rods and clocks.

Origins · 2 minInvariant interval and how this was discovered

Invariant interval

1905–1908 · Poincaré's four-dimensional rotations · Minkowski's geometry · the metric that outlived them both

The invariant arrived in two steps, from two mathematicians. In his Palermo memoir "Sur la dynamique de l'électron" (received July 1905, published 1906), Henri Poincaré observed that the Lorentz transformations leave $x^2 + y^2 + z^2 - c^2t^2$ unchanged and can therefore be regarded as rotations in a four-dimensional space — a point he made vivid by introducing the coordinate $x^4 = ict$ , which turns the awkward minus sign into an ordinary Euclidean plus at the cost of an imaginary time axis. Brilliant repackaging, and it stopped just short of the physical claim. That claim was Hermann Minkowski's, in his Cologne lecture "Raum und Zeit" of 21 September 1908: space and time are not two things that transform into each other but one thing, a four-dimensional manifold whose geometry is fixed by the interval — and the whole of relativity is the study of that geometry. The $ict$ trick was eventually abandoned, since it does not survive the passage to curved spacetime, and what replaced it is exactly equation (1), minus signs worn honestly on the outside. Minkowski did not live to see the theory become geometry in every textbook: he died on 12 January 1909 of a ruptured appendix, four months after Cologne, aged 44.

History

1905

— Einstein derives the Lorentz transformation from the two postulates, obtaining every consequence of the interval without ever writing it down.

1905

— Poincaré's Comptes Rendus note "Sur la dynamique de l'électron" (140, 1504, 5 June) names the Lorentz transformation and states that these transformations form a group.

1906

— Poincaré's Palermo memoir of the same title (Rendiconti del Circolo Matematico di Palermo 21, 129–175; received 23 July 1905, published 1906) shows the transformations preserve

x^2+y^2+z^2-c^2t^2

and are rotations of a four-dimensional space, introducing the coordinate

x^4 = ict

1908

— Minkowski's lecture "Raum und Zeit", delivered on 21 September at Cologne to the 80th Assembly of German Natural Scientists and Physicians, recasts relativity as the geometry of a four-dimensional spacetime with an indefinite metric.

1909

— "Raum und Zeit" is published in Physikalische Zeitschrift 10, 104, months after Minkowski's death in January.

1915

— Einstein's field equations promote the constant metric of the interval to a dynamical field, making

\mathrm{d}s^2 = g_{\mu\nu}\mathrm{d}x^\mu\mathrm{d}x^\nu

the foundation of gravitation.

Going further

Texts. Taylor & Wheeler, Spacetime Physics, 2nd ed., Ch. 1 — the finest treatment of the interval in print · Rindler, Introduction to Special Relativity, 2nd ed., §7 · Landau & Lifshitz, The Classical Theory of Fields, §2 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.1.3. The uniqueness theorem — that $s^2$ is the only quadratic invariant of the boost in 1+1 D — and the converse, that $\Lambda^{\mathsf T}\eta\Lambda = \eta$ forces the Lorentz transformation, are both derived above. Key anchor values: the event $(c\Delta t, \Delta x) = (5,3)\ \mathrm{m}$ with $s^2 = 16\ \mathrm{m^2}$ maps to $(4,0)$ at $\beta = 0.6$ and $(8.5, 7.5)$ at $\beta = -0.6$ ; its spacelike partner $(3,5)$ maps to $(0,4)$ and $(7.5,8.5)$ with $s^2 = -16\ \mathrm{m^2}$ ; and the Frisch–Smith muon has $\Delta x = 1907\ \mathrm{m}$ , $\beta = 0.995$ , $c\Delta t = 1916.58\ \mathrm{m}$ , $s^2 = 36\,641\ \mathrm{m^2}$ , $\Delta\tau = 0.6385\ \mathrm{\mu s} = \Delta t/\gamma$ with $\gamma = 10.013$ . The GPS clock budget is $-7.2\ \mathrm{\mu s/day}$ (special relativity) and $+45.8\ \mathrm{\mu s/day}$ (gravity), giving a net $+38.6\ \mathrm{\mu s/day}$ .

Builds on. The Two Postulates of Special Relativity · The Lorentz Transformation Equations · The Lorentz Factor Gamma · Relativity of Simultaneity. Leads to. Timelike, Spacelike, and Lightlike Separations · Proper Time · Proper Length · Lorentz Boosts as Hyperbolic Rotations.

Why is there a minus sign? Would a plus sign not be simpler?

It would — and it would be a theory of four-dimensional Euclidean space, in which nothing distinguishes time from the three spatial directions. There would be no light cone, no causal order, no "before" and "after" that all observers agree on: you could rotate time into space freely, and any event into any other. The minus sign is what makes the invariant indefinite, which carves spacetime into timelike, lightlike and spacelike regions, which gives us causality. And it is not a choice — the uniqueness derivation above shows the boost forces $C = -A$ .

Does invariance of s² mean the interval is "the real time" and Δt is an illusion?

No — this is the commonest overcorrection in the subject. $\Delta t$ is exactly what a network of synchronized clocks at rest in your frame reads, and it is the number you use to design an accelerator or schedule a satellite. It is entirely real; it is simply not shared. The right picture is not "one true time hiding behind an illusion" but "one four-dimensional structure, sliced differently by different observers". Each slicing is correct for its observer; the structure is common to all.

If two events have s² = 0, are they the same event?

No — and this is the sharpest way to feel how un-Euclidean the geometry is. In Euclidean space, zero distance means the same point. In Minkowski spacetime, $s^2 = 0$ means the two events are joined by a light ray, and they can be a billion light-years and a billion years apart. Emission of a photon at one end of the galaxy and its absorption at the other are separated by $s^2 = 0$ exactly, in every frame. The vanishing of the interval is the statement that a light signal connects them, not that they coincide.

The other sign convention gives s² = −c²Δt² + Δx². Is one of them wrong?

Neither is wrong, and no physical statement differs between them. With $(-,+,+,+)$ every sign in every formula flips: timelike becomes $s^2 < 0$ , proper time becomes $\sqrt{-s^2}/c$ , proper length becomes $\sqrt{s^2}$ . Particle physics tends to use the timelike-positive convention adopted here, because it makes $E^2 - p^2c^2 = m^2c^4$ come out with a positive mass-squared; much of the general-relativity literature, including Misner, Thorne and Wheeler, uses the other. The one thing you must never do is mix them mid-calculation, which is the single most productive source of sign errors in the subject. The invariant statement, true in both conventions, is that the sign of $s^2$ is boost-invariant — never that " $s^2$ is positive".

Is s² the length of the path the particle took?

No, and conflating the two is how the twin paradox gets confusing. $s^2$ is computed from the coordinate differences of two events, and it knows nothing about what happened in between: a particle that flies straight from $A$ to $B$ and one that takes a wild detour and still arrives at $B$ have the same $s^2$ , because $s^2$ depends only on the endpoints. What differs is the proper time along the worldline, which is the integral $\int\!\mathrm{d}s/c$ along the actual path. For straight, uniform motion the integral equals the endpoint interval, and the two notions coincide; for anything else they do not — and the straight worldline is always the one of maximum proper time. Events and Worldlines takes that up properly.

Can I use the interval if the two events are not connected by anything physical?

Yes, and you should. The interval is defined for any two events whatever — a supernova in Andromeda and a sneeze in Basingstoke — and it is invariant regardless. Its sign tells you whether a physical connection between them is possible: timelike means a massive particle could have made the trip, lightlike means only a photon could, and spacelike means nothing could. So the interval is not a record of what happened but a statement about what could have happened — which is why the taxonomy of its sign, developed in Timelike, Spacelike, and Lightlike Separations, is the foundation of causality.