The Light Cone

A flashbulb goes off. The light runs outward at the same speed in every direction, forming a sphere one light-second across after one second, two after two. Draw that growing sphere with time running up the page, and the expanding shell sweeps out a cone whose apex is the flash. Run the film backwards and a second cone opens downwards: all the light that could have converged onto the flash. Together the two are the light cone of that event. The cone is not a picture. It is a wall. Nothing that carries energy or information overtakes light, so the future cone is the exact boundary of everything the flash can ever touch — every particle it can push, every eye that can see it, every consequence it can have. Outside the cone lies elsewhere: events the flash cannot reach and which cannot reach the flash, and whose time-ordering relative to it is not even a well-posed question. Two observers in relative motion will disagree about which happened first, and both will be right. What makes the cone worth a chapter rather than a paragraph is that everybody draws the same one. A boost tips the time axis, tips the space axis, stretches the grid and squeezes it — and leaves the two $45^\circ$ null lines exactly where they were. That fixity is not luck. In the right coordinates the null directions are the eigenvectors of the boost, with eigenvalues $e^{+\varphi}$ and $e^{-\varphi}$ whose product is $1$ : the cone is invariant because a boost is a squeeze that stretches one edge by precisely the factor by which it shrinks the other. The whole causal architecture of relativity is that one eigenvalue statement.

Key takeaways

The light cone of an event is the locus of all events reachable from it, or reaching it, by a light signal. In our signature it is the null condition on the interval: $s^2 = c^2t^2 - x^2 - y^2 - z^2 = 0$ , i.e. $r = c|t|$ — an expanding light sphere drawn as a cone.
The interval is a Lorentz invariant, so every inertial observer assigns $s^2 = 0$ to the same events. The cone is therefore absolute — unlike "now", "here" or "simultaneous".
It partitions spacetime into three invariant classes: the timelike interior ( $s^2 > 0$ : causally connected, absolute time-order), the null surface ( $s^2 = 0$ : connected by light itself), and the spacelike exterior ( $s^2 < 0$ : elsewhere, causally disconnected, time-order frame-dependent).
Null coordinates $u^{\pm} = ct \pm x$ diagonalise everything: the interval becomes the product $s^2 = u^{+}u^{-}$ , and a boost of rapidity $\varphi$ becomes the pure rescaling $u^{+} \to e^{+\varphi}u^{+}$ , $u^{-} \to e^{-\varphi}u^{-}$ . At $\beta = 0.6$ the factors are exactly $2$ and $\tfrac12$ .
Key idea: the two null directions are the eigenvectors of every Lorentz boost, with eigenvalues $e^{\pm\varphi}$ whose product is $1$ . The cone is invariant not by coincidence but by an eigenvalue theorem — a boost stretches one null edge by exactly the factor it shrinks the other, so the surface $u^{+}u^{-} = 0$ maps onto itself and the invariant $s^2 = u^{+}u^{-}$ is conserved by inspection.

The core insight

Take an event and ask which other events a light signal can join it to. The answer, $c^2t^2 - r^2 = 0$ , is built from the spacetime interval — and the interval is the same number in every inertial frame. So although observers disagree about an event's time, its position, its simultaneity with anything else, and even the order of two events in the exterior region, they agree exactly and always on which side of the cone it falls. The light cone is the geometry that survives every boost: the skeleton on which causality hangs.

Where this lives today

The cone is the working tool of every field that must ask what can affect what. In numerical relativity the Courant condition — a simulation's time step must be small enough that information cannot cross more than one grid cell per step — is nothing but the demand that the numerical domain of dependence contain the physical light cone. Violate it and the code does not merely lose accuracy; it goes unstable, because it is computing an effect before its cause arrived. In quantum field theory the cone becomes an operator statement: microcausality demands that field operators at spacelike-separated points commute, $[\hat\phi(x), \hat\phi(y)] = 0$ whenever $(x-y)^2 < 0$ — exactly the requirement that no measurement here disturb one in the elsewhere. That condition is a fundamental requirement of relativistic quantum field theory, ensuring causal independence at spacelike separation and entering the assumptions underlying results such as the spin-statistics theorem. In general relativity the cones survive but stop being parallel. Mass tilts and narrows them; at a black-hole horizon the future cone has tipped so far inward that every future direction points toward the singularity — the honest definition of a horizon. And in gravitational-wave astronomy the cone became a measurement: GW170817, whose gravitational waves and gamma rays reached Earth $1.7$ seconds apart after roughly $130$ million years of flight, tested whether gravity and light ride the same cone to about one part in $10^{15}$ . They do. Minkowski's surface, sketched on a Cologne blackboard, is now an experimental result.

The physical picture

Set the origin at an event $\mathcal{E}$ — a flash, a decay, a handclap. Plot time upward, scaled by $c$ so the axis carries units of length, and one or two spatial dimensions across the page. At coordinate time $t$ a pulse of light leaving $\mathcal{E}$ occupies the sphere $r = ct$ ; as $t$ runs, the circles stack into a cone of half-angle $45^\circ$ — the future light cone. The mirror cone below, $r = c|t|$ with $t < 0$ , holds the events from which a light signal could have arrived exactly at $\mathcal{E}$ : the past light cone. Now add the physics. No causal influence propagates faster than light, and a massive particle passing through $\mathcal E$ has $|v|<c$ , so its worldline lies strictly inside the light cone. A light ray lies on it. Nothing lies outside. The future cone is therefore not merely where the light goes: it is the outer envelope of everything $\mathcal{E}$ can influence, and the past cone of everything that could have influenced $\mathcal{E}$ . The exterior consists of events spacelike separated from $\mathcal E$ : no causal signal can travel from $\mathcal E$ to such an event or from such an event to $\mathcal E$ . Minkowski called it elsewhere, and the name has stuck because no better one exists: not "now", not "far away", simply the set of events with which $\mathcal{E}$ can have no dealings. The regions look as though a drawing defines them — but the drawing is frame-dependent and the regions are not. Boost, and the time axis tips toward the cone, the space axis tips up to meet it, the grid shears, points move. Yet no point ever crosses the cone: that would flip the sign of the invariant $s^2$ .

Figure

The light cone of an event 𝓔. Left, a 2+1 dimensional slice: the flash's expanding light sphere has radius r = c|t| and sweeps a double cone. Right, the same thing in 1+1 dimensions, where the cone degenerates to the two null lines ct = ±x at 45°. The interior is timelike (s² > 0, causally connected), the surface null (s² = 0), the exterior spacelike (s² < 0) — the region Minkowski called elsewhere

The cone's shape on the page depends on how you scale the axes, and its dimensionality depends on how many spatial directions you draw: two lines in $1+1$ , a genuine circular cone in $2+1$ , a spherical hypercone in $3+1$ . None of that is physics. The physics is the single equation $c^2t^2 = r^2$ and the single fact that both sides of it transform so as to keep it true.

The core equations

Everything in this article comes from one quadratic form and its zero set. Take an event at coordinates $(ct, x, y, z)$ relative to the apex. The spacetime interval in the timelike-positive signature $(+,-,-,-)$ is

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

where: $s^2$ — the squared interval, positive for timelike, zero for null, negative for spacelike; $\eta_{\mu\nu}$ — the Minkowski metric; $\Delta x^{\mu} = (c\Delta t, \Delta x, \Delta y, \Delta z)$ — the coordinate separation four-vector. Physical reading: the one number two events possess that all inertial observers compute alike. Signature warning: the opposite convention $(-,+,+,+)$ is in wide use and flips the sign of $s^2$ throughout, so a timelike separation reads $s^2 < 0$ there. No physics depends on the choice; every intermediate sign does. This chapter is uniformly $(+,-,-,-)$ . The light cone of the apex is the null locus — the set of events at zero interval from it.

s^2 = 0 \quad\Longleftrightarrow\quad c^2t^2 = r^2 \quad\Longleftrightarrow\quad r = c\,|t|, \qquad r \equiv \sqrt{x^2+y^2+z^2} \qquad (2)

where: $r$ — radial distance from the apex; $|t|$ — elapsed coordinate time, future for $t > 0$ , past for $t < 0$ . Physical reading: at time $t$ the flash has reached exactly the sphere of radius $c|t|$ ; as $t$ runs, the sphere sweeps the cone. In $1+1$ dimensions equation (2) collapses to the two straight lines $ct = \pm x$ — the $45^\circ$ asymptotes of every invariant hyperbola. The cone's usefulness comes from what it does to the algebra. Introduce null coordinates — coordinates that run along the two null directions rather than across them.

u^{+} = ct + x, \qquad u^{-} = ct - x \qquad\Longrightarrow\qquad s^2 = u^{+}u^{-}, \qquad \mathrm{d}s^2 = \mathrm{d}u^{+}\,\mathrm{d}u^{-} \qquad (3)

where: $u^{\pm}$ — the light-cone (null) coordinates; a right-moving light ray has $u^{-}$ constant, a left-moving one has $u^{+}$ constant. Physical reading: the interval is no longer a difference of squares but a product, so the cone is simply $u^{+}u^{-} = 0$ — the union of this system's two coordinate axes. In $(+,-,-,-)$ the product carries a plus sign; in the other convention it carries a minus, a favourite hiding place for a sign error. Finally, the boost. Write it with rapidity $\varphi$ , defined by $\tanh\varphi = \beta = v/c$ , so that $\gamma = \cosh\varphi$ and $\gamma\beta = \sinh\varphi$ . Acting on $(ct, x)$ the boost matrix is

\Lambda(\varphi) = \begin{pmatrix} \cosh\varphi & \sinh\varphi \\ \sinh\varphi & \cosh\varphi \end{pmatrix}, \qquad \Lambda^{\mathsf{T}}\eta\,\Lambda = \eta, \qquad \det\Lambda = 1 \qquad (4)

where: $\varphi$ — the rapidity, the natural (additive) boost parameter; $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ — the defining condition of a Lorentz transformation, which is exactly the statement that $s^2$ is preserved. Physical reading: the inverse $\Lambda(-\varphi)$ gives the coordinates of the same event in a frame moving at $+v$ ; the two differ only by the sign of $\varphi$ , and nothing below depends on which you choose. Feed (3) into (4) and the interval's invariance becomes a triviality. The next section does it twice: once by brute force, once by eigenvalues.

Why the cone is invariant

Derivation AThe null condition is Lorentz invariant, so all observers share one cone

That every inertial observer draws the same light cone is the load-bearing claim of relativistic causality, and it deserves a proof rather than an appeal to the second postulate. The postulate says light travels at $c$ in every frame; what must be shown is the geometric consequence — that the surface $s^2 = 0$ is mapped onto itself by every Lorentz transformation, so an event on my cone is on yours and an event in my elsewhere is in yours. It takes three lines in ordinary coordinates and one in null coordinates, and the contrast is itself the lesson.

Step 1. Take a boost along $x$ with rapidity $\varphi$ : $ct' = ct\cosh\varphi + x\sinh\varphi$ , $x' = ct\sinh\varphi + x\cosh\varphi$ , with $y' = y$ , $z' = z$ . Form the transformed interval directly, $c^2t'^2 - x'^2 = (ct\cosh\varphi + x\sinh\varphi)^2 - (ct\sinh\varphi + x\cosh\varphi)^2$ . Why: we are not assuming invariance, we are testing it — every cross term must be tracked.

Step 2. Expand. The cross terms, $2\,ct\,x\cosh\varphi\sinh\varphi$ in each square, cancel identically on subtraction, leaving $(c^2t^2 - x^2)(\cosh^2\varphi - \sinh^2\varphi)$ . Why: that cancellation is the whole content of the boost's "hyperbolic rotation" structure — the Minkowski analogue of $\cos^2 + \sin^2$ tidying up a Euclidean rotation.

Step 3. Apply the hyperbolic Pythagoras identity $\cosh^2\varphi - \sinh^2\varphi = 1$ . Then $c^2t'^2 - x'^2 = c^2t^2 - x^2$ exactly, for every $\varphi$ and every event — not to leading order, identically. Adding back the untouched $-y^2 - z^2$ gives $s'^2 = s^2$ . Why: see what that identity is in disguise. With $\cosh\varphi = \gamma$ and $\sinh\varphi = \gamma\beta$ it reads $\gamma^2 - \gamma^2\beta^2 = 1$ , which is $\gamma = 1/\sqrt{1-\beta^2}$ rearranged. The identity $\gamma^2(1-\beta^2)=1$ is precisely the relation that allows the Lorentz boost to preserve the Minkowski interval.

Step 4. Read off the corollary. Since $s'^2 = s^2$ for all events, the sign of $s^2$ is invariant, so $s^2 = 0 \Rightarrow s'^2 = 0$ : the null surface maps to itself, and the three classes are frame-independent. The partition into future, past, elsewhere and cone is absolute. Why: this is the precise sense in which "causality is absolute while simultaneity is relative" — a boost may reorder two events in the elsewhere, but it can never move an event across the cone, because that would flip the sign of an invariant.

Step 5. Check that a proper orthochronous boost cannot swap the future and past light cones. Along the right-moving null ray $(ct,x)=\lambda(1,1)$ , the image is $\lambda e^{+\varphi}(1,1)$ , while along the left-moving null ray $(ct,x)=\lambda(1,-1)$ , the image is $\lambda e^{-\varphi}(1,-1)$ . Since $e^{\pm\varphi}>0$ , a future-pointing null ray remains future-pointing. Why: invariance of the null surface alone guarantees only that the light cone maps onto itself; it does not by itself distinguish its future and past halves. Their preservation follows from restricting to proper orthochronous Lorentz transformations, which are continuously connected to the identity. The positive factors $e^{\pm\varphi}$ make this explicit for a boost. *These results can be confirmed directly: the boost satisfies $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ with $\det\Lambda = 1$ ; the difference $c^2t'^2 - x'^2 - (c^2t^2 - x^2)$ simplifies to zero identically; and null events map to null events on the same branch with positive scale factor $e^{\pm\varphi}$ .*

The brute-force proof works, but it hides the reason. It shows that the cross terms cancel, not why the cone is the object that survives. To see why, stop using coordinates that cut across the cone and start using coordinates that run along it.

Go deeper · Derivation B · 3 minThe owned result: the null directions are the eigenvectors of every boost

Derivation BThe owned result: the null directions are the eigenvectors of every boost

Here is the structural fact the coordinate computation conceals. A boost is a linear map on the $(ct, x)$ plane, and linear maps are understood through their eigenvectors. So ask the question nobody asks in the first course: what are the eigenvectors of $\Lambda(\varphi)$ ? They are the two null directions — for every $\varphi$ at once — with eigenvalues $e^{+\varphi}$ and $e^{-\varphi}$ . Since those multiply to $1$ , the boost is a pure squeeze: it stretches one edge of the cone by exactly the factor by which it shrinks the other. That one sentence contains the invariance of the cone, the invariance of the interval, $\det\Lambda = 1$ , and the additivity of rapidity.

Step 1. Test the two null directions. Let $n_{+} = (1, 1)$ , the future-pointing right-moving null ray $ct = +x$ , and $n_{-} = (1, -1)$ , the left-moving one. Apply the boost of equation (4): $\Lambda(\varphi)\,n_{+} = (\cosh\varphi + \sinh\varphi,\ \sinh\varphi + \cosh\varphi) = e^{+\varphi}\,(1,1) = e^{+\varphi} n_{+}$ , using $\cosh\varphi + \sinh\varphi = e^{\varphi}$ . Likewise $\Lambda(\varphi)\,n_{-} = e^{-\varphi} n_{-}$ , using $\cosh\varphi - \sinh\varphi = e^{-\varphi}$ . Why: the null vectors are not merely mapped into the cone — they are mapped onto themselves up to a scale. They are genuine eigenvectors, and the eigenvalues are exponentials of the rapidity.

Step 2. For a nontrivial boost, $\varphi\neq0$ , these are the only two eigendirections in the $(ct,x)$ plane: the eigenvalues $e^{\pm\varphi}$ are real, positive, and distinct, so the eigenspaces are one-dimensional and exhaust the two-dimensional boost plane. Every other direction in the plane is genuinely rotated. Why: a rigidity statement. A boost leaves no timelike direction fixed (there is no absolute rest) and no spacelike direction fixed (there is no absolute simultaneity), and the two lines it does fix are precisely the two the second postulate demanded. In the $(ct,x)$ plane the cone is not one invariant among many — it is the only one.

Step 3. Change to the eigenbasis, which is exactly the null-coordinate system of equation (3): $u^{\pm} = ct \pm x$ . In these coordinates the boost is diagonal, $\Lambda(\varphi) = \mathrm{diag}(e^{+\varphi}, e^{-\varphi})$ , so it acts as $u^{+} \to e^{+\varphi}u^{+}$ and $u^{-} \to e^{-\varphi}u^{-}$ . (The passive transformation — the coordinates of one event as seen from a frame moving at $+v$ — is the inverse, $\varphi \to -\varphi$ , which simply exchanges which axis stretches.) Why: a matrix that looked like four entangled trigonometric entries is really two independent numbers being multiplied by reciprocal constants. That is what a squeeze is: area-preserving, hyperbola-preserving, axis-fixing.

Step 4. Now read off the invariance for free. The interval is $s^2 = u^{+}u^{-}$ by equation (3), and under the boost $s^2 \to (e^{+\varphi}u^{+})(e^{-\varphi}u^{-}) = e^{+\varphi}e^{-\varphi}\,u^{+}u^{-} = u^{+}u^{-} = s^2$ , because the eigenvalues are reciprocals. The cone $u^{+}u^{-} = 0$ maps to itself because a product vanishes if and only if a factor does, and the factors are only rescaled. Why: the three lines of trigonometric cancellation in the previous card have collapsed into a single multiplication of $e^{+\varphi}$ by $e^{-\varphi}$ . The interval is conserved because the eigenvalue product is $1$ — and the eigenvalue product is $\det\Lambda$ .

Step 5. Harvest the corollaries. $\det\Lambda = e^{+\varphi}e^{-\varphi} = 1$ : boosts are unimodular and preserve areas in the $(ct,x)$ plane. Composing two boosts multiplies the eigenvalues, $e^{\varphi_1}e^{\varphi_2} = e^{\varphi_1+\varphi_2}$ , so rapidities add — which is why $\varphi$ and not $v$ is the group parameter. And since $\varphi = \operatorname{artanh}\beta \to \infty$ as $\beta \to 1^{-}$ , no finite rapidity corresponds to $v = c$ ; the speed of light is approached only asymptotically by a sequence of subluminal boosts: the cone is not approached, it is asymptotic. Why: three of the most quoted facts in special relativity — unimodularity, rapidity addition, the unattainability of $c$ — are one eigenvalue statement read three ways. *These results hold exactly: $\Lambda(\varphi)n_{\pm} = e^{\pm\varphi}n_{\pm}$ ; the eigenvalue product is $1$ and equals $\det\Lambda$ ; the boost in the null basis is $\mathrm{diag}(e^{\varphi}, e^{-\varphi})$ ; and $u^{+}u^{-}$ is invariant.*

Put a number on it. At $\beta = 0.6$ the rapidity is $\varphi = \operatorname{artanh}(0.6) = \tfrac12\ln\!\big(\tfrac{1.6}{0.4}\big) = \tfrac12\ln 4 = \ln 2$ , exactly — so the eigenvalues are exactly $2$ and $\tfrac12$ . A boost to $0.6c$ doubles every event's $u^{+}$ and halves its $u^{-}$ , and the products stay put.

Figure

PUNCHLINE — a boost is a squeeze on the cone. β = 0.6 gives rapidity φ = ln 2 exactly, so the eigenvalues are e^φ = 2 and e^−φ = ½. The event E = (ct, x) = (5, 3) light-seconds has u⁺ = 8, u⁻ = 2, s² = u⁺u⁻ = 16 ls². Boosting sends it to E′ = (8.5, 7.5): u⁺ doubles to 16, u⁻ halves to 1 — and 16 × 1 = 16, so E′ slides ALONG the invariant hyperbola c²t² − x² = 16 while the 45° cone stays exactly where it was

Notice what the figure does not show: the event crossing the cone. It cannot. $u^{+}$ and $u^{-}$ are each multiplied by a strictly positive number, so neither can change sign, and an event in the future interior ( $u^{+} > 0$ , $u^{-} > 0$ ) stays there under every boost, at any speed, in either direction. Absolute causality, in one line of arithmetic.

Limiting cases

Case	Result	Meaning
$s^2 = 0$	on the cone	the events are joined by a light ray; $u^{+}u^{-} = 0$ , so at least one null coordinate vanishes
$s^2 > 0$	timelike interior	causally connectable by a subluminal signal; under proper orthochronous Lorentz transformations the sign of $\Delta t$ is invariant, so future and past are preserved
$s^2 < 0$	spacelike exterior	elsewhere; no signal passes, and some observer calls the two events simultaneous
boost $\varphi \to \infty$ ( $\beta \to 1$ )	$u^{+} \to \infty$ , $u^{-} \to 0$	the transformed coordinates become asymptotically aligned with a null direction while $s^2=u^+u^-$ remains fixed; no finite boost reaches $v=c$
$\varphi \to 0$	$\Lambda \to I$	the identity boost; the eigenvalues collapse to $1$ and every direction is trivially fixed
Newtonian limit $c \to \infty$	cone opens to $t = 0$	the null lines $t = \pm x/c$ flatten onto the horizontal, elsewhere is crushed out of existence, and absolute simultaneity returns
light ray, $k^{\mu} = (\omega/c, \mathbf{k})$	$k\cdot k = 0$	photons ride the cone in spacetime and their four-momentum lies on it in momentum space

Common trapElsewhere is not a place, and the cone is not a wall in space

Two errors travel together. The first reads "elsewhere" as a region of space you are forbidden to enter, as if a sphere surrounded you beyond which you cannot go. There is no such sphere: given enough time you can reach any point in space. Elsewhere is a region of spacetime — the set of events (place and time) your present event can neither reach nor be reached by. Proxima Centauri is not elsewhere; the events at Proxima within the next 4.2465 years are. The second error takes the cone's apparent shape to mean something. Its half-angle is $45^\circ$ only because we plot $ct$ rather than $t$ ; plot $t$ in seconds against $x$ in metres and it becomes a pancake opening at $10^{-8}$ radians. Nothing physical changed. What is physical is the invariance of the surface, not its slope on your graph paper.

Worked examples

1 · Which events at Proxima Centauri are in your elsewhere right now?

Proxima Centauri, the nearest star, has a Gaia DR3 parallax of $768.0665\ \mathrm{mas}$ . Take the event "you finish reading this sentence" as the apex. Which events on Proxima's worldline lie inside your light cone, and which are causally sealed off? If an event there is spacelike-separated from you, at what speed must an observer travel to call it simultaneous with your reading?

Step 1 — identify. Convert the parallax to a distance: $d = 1/\pi_{\mathrm{arcsec}} = 1/0.7680665 = 1.30197\ \mathrm{pc}$ , and with $1\ \mathrm{pc} = 3.26156\ \mathrm{ly}$ that is $d = \mathbf{4.2465\ ly}$ . Proxima's worldline runs vertically in your frame at $x = 4.2465$ light-years, and your cone cuts it where $|c\Delta t| = \Delta x$ — at $\Delta t = \pm 4.2465\ \mathrm{yr}$ .

Step 2 — apply. Take a specific event: "Proxima flares four years from now", $\Delta t = 4\ \mathrm{yr}$ , $\Delta x = 4.2465\ \mathrm{ly}$ . In units of years and light-years, so that $c = 1$ , $s^2 = c^2\Delta t^2 - \Delta x^2 = 16 - 18.032 = \mathbf{-2.032\ \mathrm{ly}^2}$ . Negative: the flare is spacelike separated from your reading. It lies in your elsewhere, and its proper distance — the separation in the frame where the two are simultaneous — is $L_0 = \sqrt{-s^2} = 1.426\ \mathrm{ly}$ .

Step 3 — the boundary number. The cone cuts Proxima's worldline at $\Delta t = \pm 4.2465\ \mathrm{yr}$ , so every event on Proxima between $4.2465$ years in your past and $4.2465$ years in your future lies in your elsewhere: a slab of that star's history $\mathbf{8.4929}$ years wide with which you can have no causal dealings at all. Anything older is in your past cone (you can see it); anything further ahead is in your future cone (a message would arrive in time).

Step 4 — the reversal boost. For a spacelike pair there is always a frame in which the two events are simultaneous, moving at $v = c^2\Delta t/\Delta x$ . Here $v/c = (c\,\Delta t)/\Delta x = 4/4.2465 = \mathbf{0.9420}$ , safely below $1$ — as it must be, since $|c\Delta t| < |\Delta x|$ is the spacelike condition. An observer sailing past at $0.942c$ ( $\gamma = 2.979$ ) says the flare and your reading happen at the same moment; one going the other way says the flare came first. Nobody can say the flare caused your reading — and that is why the disagreement is harmless.

2 · The squeeze, in numbers

An event $E$ has coordinates $(ct, x) = (5, 3)$ light-seconds. Classify it, then find its coordinates in a frame boosted to $\beta = 0.6$ , using null coordinates only.

Step 1. Classify. $s^2 = 5^2 - 3^2 = 16\ \mathrm{ls}^2 > 0$ : timelike, future cone. A clock carried from the apex to $E$ reads $\tau = \sqrt{s^2}/c = 4\ \mathrm{s}$ , and the required speed is $v = x/t = 0.6c$ ; a $0.6c$ traveller leaving the apex reaches $E$ after $5\,\mathrm{s}$ of coordinate time and $4\,\mathrm{s}$ of proper time. a $0.6c$ traveller leaving the apex reaches $E$ after $5\,\mathrm{s}$ of coordinate time and $4\,\mathrm{s}$ of proper time.

Step 2. Go null. $u^{+} = ct + x = 8$ , $u^{-} = ct - x = 2$ . Check equation (3): $u^{+}u^{-} = 8 \times 2 = 16 = s^2$ . ✓

Step 3. Boost. At $\beta = 0.6$ , $\varphi = \ln 2$ , so multiply: $u'^{+} = 2 \times 8 = 16$ and $u'^{-} = \tfrac12 \times 2 = 1$ . That is the entire calculation — two multiplications, no matrices, no $\gamma$ .

Step 4. Come back. Invert (3): $ct' = \tfrac12(u'^{+} + u'^{-}) = 8.5$ , $x' = \tfrac12(u'^{+} - u'^{-}) = 7.5$ . Cross-check against the matrix: $ct' = \gamma\,ct + \gamma\beta\,x = 1.25(5) + 0.75(3) = 8.5$ ✓, $x' = \gamma\beta\,ct + \gamma\,x = 0.75(5) + 1.25(3) = 7.5$ ✓. And $s'^2 = 72.25 - 56.25 = 16$ ✓ — as $16 \times 1 = 8 \times 2$ guaranteed before any arithmetic was done.

Physics of the portrait

The light cone has no single discoverer — it is what you get when you take the constancy of $c$ seriously and then draw a picture. But it has a birthday, a room, and a sentence.

Origins · 2 minLight cone and how this was discovered

Light cone

1908–present · Minkowski's Cologne lecture · Einstein's conversion · the GW170817 test

On 21 September 1908, at the 80th Assembly of German Natural Scientists and Physicians in Cologne, Hermann Minkowski gave a lecture titled Raum und Zeit — Space and Time. It opened with a promise quoted ever since: "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." What followed was not new physics but a new object: a four-dimensional geometry in which Einstein's 1905 kinematics became statements about shape. Minkowski drew the double cone at each event — its halves the Vorkegel and Nachkegel, fore-cone and after-cone — and observed that the region outside them bears no invariant temporal relation to the apex at all. Einstein, who had been Minkowski's student at the ETH in Zürich, was at first unimpressed by what he took for mathematical dressing; within a few years he had changed his mind entirely, and the four-dimensional formalism became the language of general relativity. Minkowski saw none of it: he died of appendicitis in Göttingen on 12 January 1909, less than four months after Cologne, aged 44. The cone is his monument, and it is still being tested — in August 2017 gravitational waves and gamma rays from one neutron-star merger, after roughly $130$ million years of flight, arrived $1.7$ seconds apart: two massless messengers riding the same cone, agreeing to about one part in $10^{15}$ .

History

1905

— Einstein's Zur Elektrodynamik bewegter Körper makes the speed of light the same in every inertial frame, fixing a preferred surface through every event without yet drawing it.

1907

— Minkowski, in Göttingen, begins recasting relativistic electrodynamics in four-dimensional form.

1908

— Minkowski delivers Raum und Zeit in Cologne on 21 September, introducing the spacetime diagram, the fore-cone and after-cone, and the invariant partition of events around an apex.

1909

— Minkowski dies of appendicitis in Göttingen on 12 January, aged 44; Raum und Zeit is published posthumously.

1911

— Max von Laue publishes Das Relativitätsprinzip (Vieweg, Braunschweig), an early monograph built on Minkowski's four-dimensional formalism, which helps make the geometric language standard.

1964

— E. C. Zeeman proves that the causal order alone — which events lie in whose cone — determines the Lorentz group up to a dilatation, in spacetimes of at least two spatial dimensions.

2017

— GW170817: gravitational waves and gamma rays from one neutron-star merger arrive 1.7 s apart after some 130 million years of flight, bounding any difference in their speeds at the level of one part in 10¹⁵.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §7–8 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 5–6 · Morin, Special Relativity: For the Enthusiastic Beginner, ch. 4 · Misner, Thorne & Wheeler, Gravitation, ch. 1 · Hartle, Gravity, ch. 4. Primary source: Minkowski, Raum und Zeit (Cologne, 21 September 1908), translated in The Principle of Relativity (Dover). On the converse direction, E. C. Zeeman, "Causality Implies the Lorentz Group", J. Math. Phys. 5, 490 (1964).

Signature note. The opposite convention $(-,+,+,+)$ gives $s^2 = -c^2t^2 + r^2$ , labels timelike separations $s^2 < 0$ , factors the interval as $s^2 = -u^{-}u^{+}$ , and sends $u^{+} \to e^{-\varphi}u^{+}$ under an active boost. This chapter uses the locked signature $(+,-,-,-)$ throughout: $s^2 = +c^2t^2 - r^2$ , timelike is $s^2 > 0$ , the factorisation is $s^2 = +u^{+}u^{-}$ , and the active boost is $u^{\pm} \to e^{\pm\varphi}u^{\pm}$ . Key anchor values: $\beta = 0.6 \Rightarrow \varphi = \ln 2$ exactly, $\gamma = 5/4$ , $\gamma\beta = 3/4$ , $e^{\pm\varphi} = 2, \tfrac12$ ; the event $(5,3)$ with $u^{\pm} = (8,2)$ boosts to $(8.5, 7.5)$ with $u^{\pm} = (16,1)$ at fixed $s^2 = 16$ ; Proxima's Gaia DR3 parallax $768.0665\ \mathrm{mas} \Rightarrow 4.2465\ \mathrm{ly}$ , giving $s^2 = -2.032\ \mathrm{ly}^2$ , a simultaneity boost of $0.9420c$ , and an elsewhere slab $8.4929\ \mathrm{yr}$ wide.

Builds on. Events and Worldlines · Spacetime Diagrams · The Spacetime Interval · Timelike, Spacelike and Null Separations. Leads to. Causal Structure of Spacetime · The Minkowski Metric · Lorentz Invariance and Invariants · Hyperbolic Geometry of Boosts.

If the cone's 45° slope is just an artefact of plotting ct instead of t, what is actually invariant about it?

The surface, not the slope. Different observers draw the cone at different angles on their paper, and even one observer can rescale the axes and change its opening angle at will. What nobody can change is which events lie on it. The set of events at zero interval from the apex is fixed once and for all, because $s^2$ is the same number in every inertial frame. The $45^\circ$ picture is a convenient normalisation — what you get when you measure time in metres of light-travel — and it has the pleasant property that the cone then looks the same in every frame's diagram, which is what makes spacetime diagrams worth drawing.

Can a boost tip an event from inside the cone to outside it, if I boost hard enough?

No, and the null coordinates show why in one line. A boost multiplies $u^{+}$ by $e^{+\varphi}$ and $u^{-}$ by $e^{-\varphi}$ , and both factors are strictly positive for every real rapidity — there is no $\varphi$ , however large, for which $e^{\varphi}$ is zero or negative. So the signs of $u^{+}$ and $u^{-}$ are individually preserved, and the sign of their product $s^2 = u^{+}u^{-}$ survives as a matter of arithmetic. An event in the future interior has $u^{+} > 0$ and $u^{-} > 0$ , and it will still have them at $\beta = 0.999999$ . What boosting does do is drive the event out along its hyperbola without bound — the geometric face of the fact that $c$ is a limit and not a speed.

If two events are spacelike separated and observers disagree about their order, hasn't causality already broken?

Only if something could have travelled between them, and by hypothesis nothing could. Causality is the claim that a cause precedes its effect, and the only pairs that can stand in a cause–effect relation are those a signal can connect: the timelike and null pairs. For those, the sign of $\Delta t$ is invariant — a boost preserves $u^{+}, u^{-} > 0$ , hence $c\Delta t = \tfrac12(u^{+} + u^{-}) > 0$ — so every observer agrees on which came first. The frame-dependence of order in the exterior is therefore harmless: it applies only where "which caused which" has no possible answer. Relativity gives up absolute simultaneity, not absolute causality, and the cone is the line it draws between them.

Why does the light cone matter for a photon's four-momentum, not just for its path in spacetime?

Because the same quadratic form does double duty. A plane wave has a wave four-vector $k^{\mu} = (\omega/c, \mathbf{k})$ , and for light in vacuum $\omega = c|\mathbf{k}|$ , which says precisely that $k\cdot k = (\omega/c)^2 - |\mathbf{k}|^2 = 0$ : the photon's four-momentum lies on the light cone of momentum space. That is the content of "the photon is massless" — mass is the invariant length of the four-momentum, and for a photon that length is zero. Massless particles are confined to the cone in two independent senses, and the two are the same equation with different letters.

Is the cone a consequence of the Lorentz transformations, or the other way round?

Astonishingly, the arrow runs both ways. Zeeman proved in 1964 that in a spacetime with at least two spatial dimensions, any bijection preserving the causal order must automatically be a Lorentz transformation composed with a translation and an overall scale change. You may not assume linearity, continuity, or anything about light: the causal order alone forces the group. (Alexandrov had proved essentially the same thing in the early 1950s.) So the Lorentz group is a consequence of the cone. Curiously the theorem fails in $1+1$ dimensions, and our null coordinates show why: there the causal order says only that $u^{+}$ and $u^{-}$ both increase, so any pair of increasing functions $u^{+} \mapsto f(u^{+})$ , $u^{-} \mapsto g(u^{-})$ preserves it — an infinite-dimensional group, of which the boost squeeze is the tiny linear part. It takes a second spatial dimension to lock the null coordinates together and make the geometry rigid.

Where does the light cone go in general relativity?

It stays, but it stops being rigid. In curved spacetime the metric varies from place to place, so the cone at each event is defined by the local metric and can tilt and narrow relative to the coordinate grid. Near a massive body the cones tip inward; at a black hole's event horizon they have tipped so far that the entire future cone points toward the singularity — the cleanest definition of a horizon, since you cannot avoid the interior for the same reason you cannot avoid next Tuesday. Zoom in far enough on any single event, though, and the cone is exactly the one described here: the equivalence principle says spacetime is Minkowski's in a small enough neighbourhood, and this article's flat-space cone is the tangent-space structure general relativity carries at every point.