Spacetime Interval

In Newtonian physics space and time stand apart, and both are absolute: every observer agrees on how much time separates two events and how far apart two simultaneous points lie. Special relativity dismantles both certainties at once — moving clocks run slow, moving rods contract, and even the temporal order of two events can flip between observers. Something, though, must survive, or physics would collapse into a babel of frames that never agree. That survivor is the spacetime interval: the quantity built by folding the time gap and the space gap between two events together with one decisive minus sign, $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2 + \mathrm{d}y^2 + \mathrm{d}z^2$ . It inherits its invariance from the Lorentz transformations of the prerequisite articles: the boost that stretches one term shrinks another in exactly compensating measure, so the combination comes out the same in every inertial frame. Two consequences organize everything downstream. First, the interval is the geometry of spacetime — it plays the role Euclidean distance plays in ordinary space, except the minus sign makes the geometry indefinite, so surfaces of constant separation are hyperbolae rather than spheres. Second, its sign is causal: timelike ( $\mathrm{d}s^2<0$ ) events can be joined by a sub-light worldline, spacelike ( $\mathrm{d}s^2>0$ ) events cannot influence one another at all, and null ( $\mathrm{d}s^2=0$ ) events lie on a light ray. Promote $\eta_{\mu\nu}$ to a position-dependent $g_{\mu\nu}(x)$ and the same object becomes the carrier of the entire gravitational field — every geodesic and proper time is built on the interval.

Key takeaways

The spacetime interval $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2 + \mathrm{d}y^2 + \mathrm{d}z^2$ fuses temporal and spatial separation into one number that every inertial observer agrees on, even though $\mathrm{d}t$ and $\mathrm{d}\vec{x}$ separately are frame-dependent.
Its invariance is the geometric content of the constancy of the speed of light and follows in two lines from the Lorentz transformations.
The sign classifies causality: timelike $\mathrm{d}s^2<0$ (reachable, proper time $\mathrm{d}\tau=\sqrt{-\mathrm{d}s^2}/c$ ), spacelike $\mathrm{d}s^2>0$ (no signal, proper length $\sqrt{\mathrm{d}s^2}$ ), null $\mathrm{d}s^2=0$ (a light ray).
The straight timelike worldline maximises proper time between two events — the indefinite signature flips the Euclidean minimum into a maximum, and that is the twin paradox.
Time dilation, length contraction and the light-cone structure are not separate rules but consequences of this one invariant.
Promoted to $\mathrm{d}s^2 = g_{\mu\nu}(x)\,\mathrm{d}x^\mu\,\mathrm{d}x^\nu$ , the interval becomes the metric of curved spacetime and the foundation of general relativity.
Key idea: the spacetime interval $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2 + \mathrm{d}y^2 + \mathrm{d}z^2$ is the Lorentz-invariant measure of separation between events; its frame-independence replaces absolute space and time, and its sign sorts every pair of events into timelike, spacelike, or null.

The core insight

The spacetime interval $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2 + \mathrm{d}y^2 + \mathrm{d}z^2$ is the Lorentz-invariant measure of separation between events: its frame-independence replaces the Newtonian notions of absolute distance and absolute time, and its sign classifies causal relationships into timelike, spacelike and null.

Where this lives today

The interval is quietly load-bearing across modern physics and technology. Every satellite navigation system — GPS, Galileo, BeiDou — computes proper time along orbiting worldlines, summing the kinematic and gravitational pieces of $\mathrm{d}s^2$ into the $+38\ \mu\mathrm{s/day}$ correction without which positioning would fail within minutes; it is arguably the most widely deployed test of relativity in daily life. In gravitational-wave astronomy, LIGO and Virgo detect a passing wave as a fractional change in the interval between mirrors of order $\mathrm{d}s/s \sim 10^{-21}$ , literally measuring the metric oscillate. At the frontier, the causal structure encoded by the sign of $\mathrm{d}s^2$ underpins microcausality in quantum field theory, the holographic bounds of AdS/CFT, and research into whether spacetime is fundamental or emergent from entanglement. The little formula that sorts two events into timelike or spacelike is the seed of all of it.

The physical picture

Think of the interval as the Pythagorean theorem with one sign flipped. In ordinary space the squared distance $\mathrm{d}\ell^2 = \mathrm{d}x^2 + \mathrm{d}y^2$ is left untouched when you rotate your axes. In spacetime a Lorentz boost is the analogue of a rotation, and $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2$ is the invariant it preserves. The single minus sign changes the geometry utterly: where Euclidean rotations trace out circles of constant distance, Lorentz boosts trace out hyperbolae of constant interval. A boosted event slides along a hyperbola rather than a circle — which is precisely why a clock's tick and a rod's length come out different for different observers while the interval does not. Why must such an invariant exist at all? Because everyone measures the same speed of light. A flash emitted from a point satisfies $|\mathrm{d}\vec{x}| = c\,\mathrm{d}t$ , i.e. $\mathrm{d}s^2 = 0$ , in every frame. If the surface "speed = $c$ " is frame-independent, the quadratic form whose zero-set it is must be preserved — isotropy and reciprocity then pin the scale to exactly one. The constancy of $c$ therefore forces the invariance of $\mathrm{d}s^2$ , and the light cone becomes the fixed scaffold of all causality.

Figure

The light cone sorts every event by the sign of the interval: inside the cone ds² < 0 (timelike, reachable by a sub-light worldline), outside ds² > 0 (spacelike, no causal contact), and on the cone ds² = 0 (null, c²dt² = dx²) — a verdict every observer shares because ds² is invariant

The sign of the interval answers a physical question with a firm yes, no, or only-light. Timelike ( $\mathrm{d}s^2<0$ ): could I personally attend both events? Yes — a slower-than-light worldline joins them, and some frame sees them at the same place. Spacelike ( $\mathrm{d}s^2>0$ ): could one cause the other? No — too far apart for any signal, and some frame even sees them simultaneous, so their order is not absolute. Null ( $\mathrm{d}s^2=0$ ): only a light ray bridges them. Because $\mathrm{d}s^2$ is invariant, this verdict is identical for all observers, which is what keeps relativistic cause and effect consistent. For a moving clock the interval is the ticking. A clock on a timelike worldline reads proper time $\tau$ , with $c^2\,\mathrm{d}\tau^2 = -\mathrm{d}s^2 = c^2\,\mathrm{d}t^2 - \mathrm{d}\vec{x}^2$ , so the interval is not bookkeeping but the physical time a wristwatch accumulates. The next figure makes invariance visible: boosting to a new frame slides an event along a hyperbola of fixed interval, and where that hyperbola meets each frame's time axis the clock reads the same proper time.

Figure

A Lorentz boost slides an event along the hyperbola −c²t² + x² = −1 of constant interval: the rest-frame clock at ct = 1 and the moving clock where the ct′ axis meets the curve both read proper time τ = 1, so every inertial observer agrees the interval equals −1

The core equations

The defining object is the infinitesimal interval between two neighbouring events: the sum of squared spatial separations minus $c^2$ times the squared time separation. The minus sign is not decoration — it is the whole of relativity compressed into one symbol.

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

where: $\mathrm{d}s^2$ — squared interval between two infinitesimally separated events (may be negative, zero, or positive); $\mathrm{d}t$ — coordinate time difference; $\mathrm{d}x,\mathrm{d}y,\mathrm{d}z$ — spatial coordinate differences; $c$ — speed of light in vacuum, $299\,792\,458\ \mathrm{m\,s^{-1}}$ (exact, by the 2019 SI definition). A negative value means timelike, positive spacelike, zero null. Writing the coordinates as a four-vector $x^\mu=(ct,x,y,z)$ packages equation (1) as a single contraction with the Minkowski metric — the form that generalises.

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

where: $\eta_{\mu\nu}$ — Minkowski metric in the $(-,+,+,+)$ signature used by Wald, Carroll and MTW; $x^\mu$ — contravariant coordinates, indices $\mu=0,1,2,3$ ; repeated upper–lower indices are summed (Einstein convention). Replace the constant $\eta_{\mu\nu}$ by a position-dependent, symmetric rank-2 tensor $g_{\mu\nu}(x)$ and the same expression now describes a curved spacetime carrying a gravitational field — the one substitution that is the doorway from special to general relativity.

\mathrm{d}s^2 = g_{\mu\nu}(x)\,\mathrm{d}x^\mu\,\mathrm{d}x^\nu \qquad (3)

where: $g_{\mu\nu}(x)$ — the metric tensor, ten independent functions of position that encode gravity. Everything measurable — proper time, proper length, redshift, orbits — is read off it. Along a timelike worldline the interval is minus the square of the proper time — literally the reading of a comoving clock, and the physical meaning behind the sign convention.

c^2\,\mathrm{d}\tau^2 = -\mathrm{d}s^2 = c^2\,\mathrm{d}t^2 - \mathrm{d}x^2 - \mathrm{d}y^2 - \mathrm{d}z^2 \qquad (4)

where: $\mathrm{d}\tau$ — proper time, the elapsed time on a clock carried between the two events; the right-hand side is positive for a timelike separation, so $\mathrm{d}\tau$ is real. Integrating $\mathrm{d}\tau$ along a worldline gives the total aged time. Setting the interval to zero recovers the light postulate exactly: the null condition is the statement that light moves at $c$ in every frame — the equation of the light cone.

\mathrm{d}s^2 = 0 \;\Longleftrightarrow\; \left(\frac{\mathrm{d}x}{\mathrm{d}t}\right)^2 + \left(\frac{\mathrm{d}y}{\mathrm{d}t}\right)^2 + \left(\frac{\mathrm{d}z}{\mathrm{d}t}\right)^2 = c^2 \qquad (5)

where: $\mathrm{d}x/\mathrm{d}t$ — a coordinate velocity component of a signal. The zero-set of $\mathrm{d}s^2$ is the cone of light rays; its invariance is why $c$ is the same for all. In flat spacetime the metric does not vary from point to point, so the interval integrates to a function of the coordinate differences alone — one may drop the differentials for a finite separation. In curved spacetime this shortcut fails; only the infinitesimal form is fundamental.

\Delta s^2 = -c^2\,\Delta t^2 + \Delta x^2 + \Delta y^2 + \Delta z^2 \qquad (6)

where: $\Delta s^2$ — squared interval between two finitely separated events joined by a straight coordinate line; $\Delta t,\Delta x,\dots$ — finite coordinate differences. Valid only in Minkowski spacetime along straight coordinate lines; elsewhere integrate $\mathrm{d}s$ along a specified path.

Derivation AInvariance of the interval under a Lorentz boost

The goal is to show, by direct substitution, that the interval $\mathrm{d}s^2$ is numerically unchanged by a Lorentz boost, so that two observers in relative motion who disagree about every $\mathrm{d}t$ and every $\mathrm{d}x$ nonetheless compute the identical separation between the same pair of events. We take a boost of velocity $v$ along $x$ with $\beta=v/c$ , $\gamma=(1-\beta^2)^{-1/2}$ , transform the differentials, and watch the cross terms cancel via the identity $\gamma^2(1-\beta^2)=1$ .

Step 1. Write the Lorentz transformation of the differentials: $\mathrm{d}t' = \gamma(\mathrm{d}t - \tfrac{v}{c^2}\mathrm{d}x)$ and $\mathrm{d}x' = \gamma(\mathrm{d}x - v\,\mathrm{d}t)$ , with the transverse pieces untouched, $\mathrm{d}y'=\mathrm{d}y$ , $\mathrm{d}z'=\mathrm{d}z$ . Why: the boost mixes time and the boost-direction coordinate but leaves the perpendicular directions alone, so only the $t$ – $x$ block can change the interval.

Step 2. Square the time term and multiply by $-c^2$ : $-c^2\,\mathrm{d}t'^2 = -c^2\gamma^2(\mathrm{d}t^2 - \tfrac{2v}{c^2}\mathrm{d}t\,\mathrm{d}x + \tfrac{v^2}{c^4}\mathrm{d}x^2)$ . Why: this isolates every contribution the transformed time coordinate makes, including the cross term to be cancelled.

Step 3. Square the space term: $\mathrm{d}x'^2 = \gamma^2(\mathrm{d}x^2 - 2v\,\mathrm{d}t\,\mathrm{d}x + v^2\,\mathrm{d}t^2)$ . Why: the boosted spatial coordinate produces its own cross term $-2v\gamma^2\,\mathrm{d}t\,\mathrm{d}x$ , equal and opposite to the one from the time term.

Step 4. Add the two; the cross terms cancel exactly and the survivors regroup as $-c^2\,\mathrm{d}t'^2 + \mathrm{d}x'^2 = \gamma^2[-c^2\,\mathrm{d}t^2 + v^2\,\mathrm{d}t^2 + \mathrm{d}x^2 - \tfrac{v^2}{c^2}\mathrm{d}x^2]$ . Why: relativistic invariance hinges on this cancellation — the algebraic shadow of the light cone being preserved.

Step 5. Factor as $\gamma^2(1-\beta^2)(-c^2\,\mathrm{d}t^2 + \mathrm{d}x^2)$ and use $\gamma^2(1-\beta^2)=1$ to get $-c^2\,\mathrm{d}t'^2 + \mathrm{d}x'^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2$ . Adding the untouched $\mathrm{d}y^2+\mathrm{d}z^2$ gives full invariance, $\mathrm{d}s'^2=\mathrm{d}s^2$ . Why: the defining identity of $\gamma$ is exactly the factor needed to erase the boost — no accident, since $\gamma$ was constructed to make this true. *Verified: direct algebra; matches Schutz §1.6 and Carroll §1.2. The only inputs are the cross-term cancellation and $\gamma^2(1-\beta^2)=1$ .*

Go deeper · Derivation B · 1 minThe interval forced by the constancy of light

Derivation BThe interval forced by the constancy of light

The goal is to run the logic in reverse and derive the interval rather than assume it, showing that the second postulate together with the symmetry of empty space fixes the invariant uniquely, scale factor and all. We posit a quadratic form $I=-c^2\Delta t^2+\Delta x^2+\Delta y^2+\Delta z^2$ , demand that its null cone match light propagation in every frame, and let isotropy and reciprocity whittle the transformation law down to exact invariance.

Step 1. Because a light pulse obeys $|\Delta\vec{x}|=c\,\Delta t$ in every inertial frame, the null set $I=0$ is frame-independent; linearity then forces $I' = f(v)\,I$ for some factor $f$ depending only on the relative speed. Why: two quadratic forms with the same zero-set and rank differ at most by an overall multiplier, so the problem reduces to pinning down $f$ .

Step 2. Space is isotropic, so $f$ can depend only on the magnitude of the relative velocity: $f(v)=f(|v|)=f(-v)$ . Why: there is no preferred direction, so boosting left or right by the same speed must scale the interval identically.

Step 3. Apply the boost and then its inverse (a boost by $-v$ , same $f$ by isotropy) to return to the original frame: $I''=f(v)^2\,I$ , but also $I''=I$ , hence $f(v)^2=1$ . Why: composing a transformation with its own inverse is the identity, and that closure squares $f$ down to unit modulus.

Step 4. Continuity with the identity at $v=0$ , where $f(0)=1$ , and positivity select $f(v)=+1$ , giving $I'=I$ exactly. Why: the scale factor cannot jump as $v\to0$ , so the physical branch is $f\equiv+1$ — invariance with no residual ambiguity. *Verified: standard argument following Landau–Lifshitz, The Classical Theory of Fields §2, and Schutz §1.6. The light postulate fixes the null cone; homogeneity, isotropy and reciprocity fix the scale to unity.*

Go deeper · Derivation C · 1 minProper time along a worldline and the twin paradox

Derivation CProper time along a worldline and the twin paradox

The goal is to turn the interval into a clock: to compute the total time a traveller's own watch records along an arbitrary timelike path and to prove that path matters, so that two twins reunited at the same event have aged by different amounts. We start from the timelike interval as an infinitesimal clock reading, factor out coordinate time to expose the Lorentz factor, integrate along the trip, and read off the geometric punchline — the straight worldline is the longest in proper time.

Step 1. From equation (4), the infinitesimal proper time obeys $c^2\,\mathrm{d}\tau^2 = c^2\,\mathrm{d}t^2 - \mathrm{d}\vec{x}^2$ , valid along any timelike worldline because $\mathrm{d}\tau$ is what a comoving clock reads. Why: the interval is frame-invariant, so the lab frame and the clock's rest frame must agree, and in the rest frame it is just the clock's own elapsed time.

Step 2. Factor out $\mathrm{d}t$ : $\mathrm{d}\tau = \mathrm{d}t\,\sqrt{1 - c^{-2}(\mathrm{d}\vec{x}/\mathrm{d}t)^2} = \mathrm{d}t\,\sqrt{1-v^2/c^2} = \mathrm{d}t/\gamma$ , with $v$ the instantaneous coordinate speed. Why: this exhibits the moving clock running slow — the integrand is at most one, so motion always costs proper time.

Step 3. Integrate over the journey: $\tau_{\text{traveller}} = \int_{t_1}^{t_2}\mathrm{d}t/\gamma(t) < t_2-t_1 = \tau_{\text{stay-at-home}}$ , the strict inequality holding whenever $v>0$ . Why: the stay-at-home twin has $\gamma=1$ throughout, so any detour with $\gamma>1$ accumulates strictly less proper time.

Step 4. Interpret geometrically: the straight timelike worldline maximises proper time between two fixed events — the mirror image of the Euclidean fact that a straight line minimises distance, caused solely by the minus sign. Why: the travelling twin's path is a longer coordinate detour but a shorter proper-time path, so they return younger — the paradox dissolves because only one twin stays inertial throughout. *Verified: standard result; Hartle Ch. 5 and Taylor & Wheeler, Spacetime Physics. The inequality $\tau<\Delta t$ follows directly from $\gamma\ge1$ .*

Limiting cases

The sign of the interval sorts every pair of events; several limits recover familiar quantities.

Case	Result	Meaning
Timelike separation	$\mathrm{d}s^2 < 0 \;\Longleftrightarrow\; c^2\,\mathrm{d}t^2 > \mathrm{d}\vec{x}^2$	Joinable by a sub-light signal; a frame exists where the events share a place; proper time $\mathrm{d}\tau=\sqrt{-\mathrm{d}s^2}/c$ . This is the interior of the light cone.
Spacelike separation	$\mathrm{d}s^2 > 0 \;\Longleftrightarrow\; c^2\,\mathrm{d}t^2 < \mathrm{d}\vec{x}^2$	No causal contact possible; a frame exists where the events are simultaneous; proper length $\mathrm{d}\ell=\sqrt{\mathrm{d}s^2}$ . Their time-order is frame-dependent.
Null (lightlike) separation	$\mathrm{d}s^2 = 0 \;\Longleftrightarrow\; c^2\,\mathrm{d}t^2 = \mathrm{d}\vec{x}^2$	Joinable only by a light ray; the cone itself. Both proper time and proper length vanish — light logs no interval along its path.
Newtonian limit	$\|\mathrm{d}\vec{x}\|/\mathrm{d}t \ll c \;\Rightarrow\; \mathrm{d}\tau \approx \mathrm{d}t$	The $-c^2\mathrm{d}t^2$ term dominates and proper time collapses onto coordinate time — absolute Newtonian time re-emerges as a slow-motion approximation.
Curved spacetime	$\mathrm{d}s^2 = g_{\mu\nu}\,\mathrm{d}x^\mu\mathrm{d}x^\nu$	The constant $\eta_{\mu\nu}$ becomes $g_{\mu\nu}(x)$ ; the finite form (6) fails and one integrates $\mathrm{d}s$ along a path. Gravity lives here.

Common trapThe interval is not a Euclidean distance, and the minus sign is not optional

It is tempting to read $\mathrm{d}s^2 = -c^2\,\mathrm{d}t^2 + \mathrm{d}x^2 + \dots$ as just a four-dimensional Pythagoras, but the leading minus sign is the entire physics. Because of it, $\mathrm{d}s^2$ can be negative or zero even between distinct events, the geometry is indefinite (pseudo-Riemannian, not Riemannian), and the triangle inequality reverses for timelike paths — which is why the straight worldline maximises rather than minimises proper time. Treat $\mathrm{d}s^2$ as a positive-definite distance and you destroy causality: the timelike/spacelike distinction, and the whole light-cone structure, evaporates. Two further traps: the rival particle-physics signature $(+,-,-,-)$ flips every sign, so a timelike interval becomes positive — always state your convention, since $c^2\mathrm{d}\tau^2 = -\mathrm{d}s^2$ in one and $+\mathrm{d}s^2$ in the other; and coordinate-dependence of $g_{\mu\nu}$ (as in the spherical form) does not imply curvature — flat spacetime in funny coordinates still has a varying metric.

Worked examples

1 · Classify a pair of events and read off the proper time

Two events have separations $\Delta t = 5\ \mathrm{s}$ and $\Delta x = 6\times10^{8}\ \mathrm{m}$ (with $\Delta y=\Delta z=0$ ); take $c = 3.00\times10^{8}\ \mathrm{m\,s^{-1}}$ . Are they timelike, spacelike or null? If a clock can attend both, how much proper time does it record?

Step 1. Form the light-travel distance $c\,\Delta t = (3.00\times10^{8})(5) = 1.5\times10^{9}\ \mathrm{m}$ , so $c^2\,\Delta t^2 = 2.25\times10^{18}\ \mathrm{m^2}$ , while $\Delta x^2 = (6\times10^{8})^2 = 3.6\times10^{17}\ \mathrm{m^2}$ — the time term is the larger, so light has time to cross the gap.

Step 2. The interval is $\Delta s^2 = -c^2\Delta t^2 + \Delta x^2 = -2.25\times10^{18} + 0.36\times10^{18} = \mathbf{-1.89\times10^{18}\ m^2 < 0}$ , so the events are timelike separated — a sub-light clock can be present at both.

Step 3. The proper time is $\Delta\tau = \sqrt{-\Delta s^2}/c = \sqrt{1.89\times10^{18}}/(3.00\times10^{8}) = 1.375\times10^{9}/3.00\times10^{8} = \mathbf{4.58\ s}$ .

Step 4. Cross-check with the Lorentz factor. The clock's speed is $v = \Delta x/\Delta t = 1.2\times10^{8}\ \mathrm{m\,s^{-1}}$ , so $\beta = 0.40$ and $\gamma = 1/\sqrt{1-0.16} = 1.091$ . Then $\Delta\tau = \Delta t/\gamma = 5/1.091 = 4.58\ \mathrm{s}$ — identical, as invariance guarantees. Timelike; the traveller ages 4.58 s while 5 s of coordinate time elapse.

2 · The interval behind GPS timekeeping

A GPS satellite orbits at radius $r_s = 2.66\times10^{7}\ \mathrm{m}$ with speed $v \approx 3.87\times10^{3}\ \mathrm{m\,s^{-1}}$ ; a ground clock sits at $r_g = 6.37\times10^{6}\ \mathrm{m}$ . Using the interval, estimate the kinematic rate offset of the satellite clock and combine it with the gravitational term for the net daily drift.

Step 1. Along the satellite worldline the interval gives $\mathrm{d}\tau/\mathrm{d}t = \sqrt{1-v^2/c^2} \approx 1 - \tfrac12(v/c)^2$ . With $\beta = v/c = 3.87\times10^{3}/3.00\times10^{8} = 1.29\times10^{-5}$ , the fractional kinematic slowdown is $-\tfrac12\beta^2 = -\tfrac12(1.29\times10^{-5})^2 \approx -8.3\times10^{-11}$ .

Step 2. Per day ( $86\,400\ \mathrm{s}$ ) this is $-8.3\times10^{-11} \times 86\,400 \approx \mathbf{-7.2\ \mu s/day}$ : motion makes the orbiting clock run slow relative to the ground.

Step 3. The curved-spacetime part of the interval (the $g_{tt}$ term, developed in the metric articles) contributes a gravitational speed-up because the satellite sits higher in Earth's potential: $\big(GM/c^2\big)\big(1/r_g - 1/r_s\big)\times 86\,400 \approx \mathbf{+45.7\ \mu s/day}$ .

Step 4. Add the two: $-7.2 + 45.7 = +38.5\ \mu\mathrm{s/day}$ , so the satellite clock gains about $+38\ \mu\mathrm{s}$ each day. Uncorrected, this would drift the position solution by roughly $c\times38\ \mu\mathrm{s} \approx 11\ \mathrm{km}$ per day, which is why both interval corrections are engineered into every receiver. Net: about +38 μs/day, dominated by the gravitational term.

Physics of the portrait

The interval is the moment optics-of-space became geometry: the number Lorentz and Einstein computed as an invariant, Poincaré glimpsed as a rotation, and Minkowski finally saw as the metric of a four-dimensional world.

Origins · 1 minHermann Minkowski and how this was discovered

Hermann Minkowski

1864–1909 · the mathematician who fused space and time into spacetime

Hermann Minkowski was a German mathematician of extraordinary range — a prodigy who shared the Paris Academy's Grand Prix at eighteen, a pioneer of the "geometry of numbers", and, at the ETH Zürich Polytechnic, one of Albert Einstein's own mathematics teachers. His decisive contribution to relativity came late and fast. Einstein's 1905 theory had already made time and space frame-dependent, and Poincaré had noticed that the Lorentz transformations behave like rotations preserving the quadratic form $-c^2t^2+x^2+y^2+z^2$ . But it was Minkowski, in his 1908 Cologne address Raum und Zeit, who drew the full geometric conclusion and announced it with a flourish never bettered: "Henceforth space by itself, and time by itself, are doomed to fade away into mere shadows, and only a kind of union of the two will preserve an independent reality." He introduced the four-dimensional continuum, the worldline, the light cone, and — as the bedrock beneath all of it — the invariant interval, the object that replaces separate notions of duration and distance. Minkowski died suddenly of appendicitis in January 1909, months after the lecture. Einstein at first dismissed the four-dimensional dressing as "superfluous learnedness", then found it indispensable: general relativity is unthinkable without Minkowski's insight that gravity is the geometry of the interval, $\mathrm{d}s^2 = g_{\mu\nu}\mathrm{d}x^\mu\mathrm{d}x^\nu$ .

History

1905

— Einstein's special relativity makes durations and simultaneity frame-dependent

1906

— Poincaré treats the Lorentz group as rotations preserving −c²t² + x² + y² + z²

1908

— Minkowski's Cologne address Raum und Zeit introduces four-dimensional spacetime and the invariant interval

1909

— Minkowski dies suddenly; his geometric reformulation is published posthumously

1915

— Einstein promotes the interval to ds² = g_{μν} dx^μ dx^ν, the metric of curved spacetime, completing general relativity

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §1.5–1.7, §2 · Wald, General Relativity (1984), §1.1–1.3 · Carroll, Spacetime and Geometry (2004), §1.1–1.3 · Hartle, Gravity (2003), Ch. 4–5 · Schutz, A First Course in General Relativity (2nd ed., 2009), Ch. 1–2 · Taylor & Wheeler, Spacetime Physics (2nd ed., 1992) for the interval-first pedagogy.

Provenance. All equations are standard and appear in MTW §1.5–2, Wald §1.1–1.3, Carroll §1.1–1.3, Hartle Ch. 4–5, and Schutz Ch. 1–2. A verifier should: (1) reproduce the boost-invariance algebra, confirming the cross-term cancellation and $\gamma^2(1-\beta^2)=1$ (Schutz §1.6); (2) check the proper-time integral $\tau=\int\mathrm{d}t/\gamma$ and its maximisation on the inertial worldline (Hartle Ch. 5); (3) confirm $c = 299\,792\,458\ \mathrm{m\,s^{-1}}$ exact (2019 SI); (4) verify the GPS net rate $\approx +38\ \mu\mathrm{s/day}$ (Ashby, Living Reviews in Relativity, 2003) as $\approx -7$ kinematic plus $\approx +46\ \mu\mathrm{s/day}$ gravitational. Worked-example numbers were recomputed independently. No fabricated numbers are used.

Builds on: Postulates of Special Relativity · Lorentz Transformations · Relativity of Simultaneity Leads to: Minkowski Metric and Signature · Proper Time · Proper Length · Causal Structure and Light Cones · Minkowski Spacetime

Ask the tutor: Want to see how the same interval, with $\eta_{\mu\nu}$ swapped for the Schwarzschild $g_{\mu\nu}$ , turns into light bending and Mercury's precession — ask me to walk you from flat spacetime into curvature.

Is the interval a distance or a time?

Neither, and that is the point — it is a new kind of separation with no analogue in pre-relativistic physics. Its dimensions are those of length-squared, but its sign carries information no ordinary distance has. For a timelike separation, $\sqrt{-\mathrm{d}s^2}/c$ is a proper time — a clock's reading. For a spacelike separation, $\sqrt{\mathrm{d}s^2}$ is a proper length — the distance in the frame where the events are simultaneous. So the single object $\mathrm{d}s^2$ carries both a "time" and a "length" interpretation, selected by whether the separation lies inside or outside the light cone.

Why is there a minus sign in front of the time term?

The minus sign encodes the finite, invariant speed of light. A light ray must satisfy "speed $= c$ " in every frame, so the surface $-c^2\mathrm{d}t^2 + \mathrm{d}\vec{x}^2 = 0$ has to be frame-independent. Only an indefinite quadratic form — one term with the opposite sign to the others — has a non-trivial cone as its zero-set; a positive-definite form (all plus signs) has only the origin as its zero and could never describe light propagating outward. The minus sign is therefore not a convention you could drop; it is the algebraic fingerprint of causality. Flip it to a plus and you get Euclidean four-space, where nothing propagates and past and future are indistinguishable.

How is this interval related to time dilation and length contraction?

They are all the same statement seen from different angles. Because $\mathrm{d}s^2$ is invariant while $\mathrm{d}t$ and $\mathrm{d}\vec{x}$ are not, holding the interval fixed forces $\mathrm{d}t$ and $\mathrm{d}\vec{x}$ to trade off against each other as you change frames — that trade-off is time dilation and length contraction. For a moving clock, $c^2\mathrm{d}\tau^2 = c^2\mathrm{d}t^2 - \mathrm{d}\vec{x}^2$ gives $\mathrm{d}\tau = \mathrm{d}t/\gamma$ : the clock runs slow. For a moving rod, the same invariance applied to a spacelike interval yields the contracted length $L = L_0/\gamma$ . Rather than three separate rules to memorise, there is one invariant and its bookkeeping — which is why the interval-first approach (Taylor & Wheeler) is so powerful pedagogically.

What actually changes when spacetime is curved?

The form of the interval survives untouched; only the metric coefficients change. In flat spacetime the metric is the constant $\eta_{\mu\nu}=\mathrm{diag}(-1,+1,+1,+1)$ , so the interval integrates trivially to the finite formula (6). In curved spacetime you replace $\eta_{\mu\nu}$ by a position-dependent $g_{\mu\nu}(x)$ : the finite shortcut fails — you must integrate $\mathrm{d}s$ along a path — and the metric can no longer be made constant everywhere by any coordinate choice, and that irreducible variation is curvature. Crucially, a metric that merely looks position-dependent, like the spherical form $-c^2\mathrm{d}t^2 + \mathrm{d}r^2 + r^2\mathrm{d}\theta^2 + r^2\sin^2\theta\,\mathrm{d}\phi^2$ , can still be flat; genuine curvature is detected by the Riemann tensor, not by coordinate dependence.

Can the order of two events really flip between observers?

Yes, but only for spacelike-separated events, and precisely because no signal can connect them. If $\mathrm{d}s^2>0$ , there exist frames in which A precedes B, frames in which B precedes A, and one in which they are simultaneous — and this ambiguity is harmless because neither event can influence the other, so "which came first" has no causal meaning. For timelike or null separations ( $\mathrm{d}s^2\le0$ ), the order is absolute: every observer agrees which event is in the other's past, so cause always precedes effect. The invariance of $\mathrm{d}s^2$ is what draws this line — reorderable pairs are always the causally disconnected ones.

Why does light travel a zero interval, and what parametrises its path?

For light $|\mathrm{d}\vec{x}| = c\,\mathrm{d}t$ , so $\mathrm{d}s^2 = -c^2\mathrm{d}t^2 + \mathrm{d}\vec{x}^2 = 0$ identically — a photon's worldline is null, logging zero interval no matter how far it travels. Since $c^2\mathrm{d}\tau^2 = -\mathrm{d}s^2$ , this means $\mathrm{d}\tau = 0$ : a photon has no proper time, and phrases like "from the photon's point of view" have no rigorous meaning because there is no rest frame for light. Physicists parametrise null worldlines by an affine parameter $\lambda$ instead — a smooth marker along the ray that plays the bookkeeping role proper time plays for massive particles.