Timelike, Spacelike, and Lightlike Separations

Relativity has spent several chapters cheerfully dismantling the idea of "now". Two observers moving past each other disagree about whether two distant events happened at the same time — and worse, they can disagree about which one happened first. That sounds like the end of physics: if the order of events is negotiable, then somewhere out there is an observer who sees the bullet arrive before the trigger is pulled, and cause and effect collapse. They do not collapse, and the reason is the most important theorem in special relativity. The order of two events is negotiable exactly when nothing could have travelled between them. If a signal — anything at all, at any speed up to and including light — could have gone from the first to the second, then every inertial observer agrees on which came first. If no signal could have made the trip, observers disagree about the order, and this is harmless, because the events cannot influence each other anyway. Relativity does not abolish causal order; it abolishes it precisely and only where it does not matter. Which case you are in is decided by a single number: the sign of the invariant interval $s^2$ . Positive is timelike — the events are causally connectable, and their order is absolute. Zero is lightlike — only light can bridge them. Negative is spacelike — nothing can, they lie in each other's elsewhere, and their time order is up for grabs. Because $s^2$ is the same in every frame, so is its sign, and so is the classification. The causal skeleton of spacetime is the one thing relativity leaves standing, and this article is about why.

Key takeaways

Signature $(+,-,-,-)$ , timelike-positive. $s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2$ , with $s^2 > 0$ timelike, $s^2 = 0$ lightlike (null), $s^2 < 0$ spacelike. Roughly half the literature uses the opposite sign; every sign flips and no physics does.
Since $s^2$ is invariant, so is its sign — so the three-way classification is the same for every inertial observer, even though $\Delta t$ and $\Delta x$ are not.
The taxonomy is not a definition but a theorem about what is negotiable. Timelike or lightlike ⇒ no boost with $|\beta| < 1$ reverses the time order. Spacelike ⇒ there is always such a boost, namely $\beta_{\text{flip}} = c\Delta t/\Delta x$ .
It compresses to one line: $\beta_{\text{flip}} = 1/\beta_{\text{signal}}$ . A sub-light signal would need a super-light boost to be reordered — impossible. Only a superluminal signal has $|\beta_{\text{flip}}| < 1$ , which is exactly why superluminal signalling would destroy causality.
Outside the cone is the elsewhere: not "the far away", but the events you can neither affect nor learn about, ever.
Key idea: the sign of $s^2$ partitions every pair of events into causally connectable, light-connectable, and causally disconnected — and because the sign is boost-invariant, this partition is absolute. Simultaneity is relative; causal order is not.

The core insight

Boosts tilt the axes of spacetime like scissors closing, pivoting about the $45^\circ$ light lines. A boost can slide an event anywhere within its region, but never across the cone — that would change the sign of an invariant, a contradiction in terms. The cone is the hinge, and no boost moves the hinge. That one geometric fact keeps cause before effect in every frame.

Where this lives today

The classification is the daily working tool of every experiment that must decide whether two things could have talked to each other. In quantum optics it is load-bearing: a Bell test is only a Bell test if the two measurement events are spacelike separated, so that no signal — not even at $c$ — could carry one detector's setting to the other. Experimenters enforce this by choosing measurement bases with random number generators fired nanoseconds before the photon arrives, and by separating the stations by hundreds of metres or, in the 2017 Micius satellite experiment, by $1200\ \mathrm{km}$ . The entire force of the result — that no local hidden-variable theory reproduces quantum mechanics — rests on the interval between two clicks being negative. The same taxonomy draws the boundary of the observable universe: an event is visible to us only if it lies on or inside our past light cone, and everything spacelike-separated from here-and-now lies beyond the particle horizon, elsewhere in the strict sense of this article. And every faster-than-light claim that surfaces — the 2011 OPERA neutrino anomaly, later traced to a loose fibre-optic connector — is met with the argument of equation (4) rather than a measurement. A superluminal signal is not merely fast. It is, in some perfectly ordinary frame, a signal into the past.

The physical picture

Stand at an event and ask a question with real stakes: which other events could I possibly have caused, and which could have caused me? The answer carves spacetime into three pieces, and the boundary is drawn by light. Everything you could reach travelling slower than light lies inside your future cone; everything that could have reached you lies in the past cone; everything a light flash exactly touches lies on the cone. And everything else — too far away in space for the time available — lies outside, in the region Taylor and Wheeler named the elsewhere: not merely distant, but causally sealed. The competition is between the time light has, $c\Delta t$ , and the distance it must cover, $\Delta r$ . Time wins ( $c\Delta t > \Delta r$ ): a slower-than-light signal suffices — timelike. A tie: only light — lightlike. Distance wins: nothing bridges the gap — spacelike. The referee of that footrace is $s^2 = (c\Delta t)^2 - \Delta r^2$ , whose sign records who won.

Figure

The light cone and the three regions, with three real event pairs measured from the vertex O. M — a cosmic-ray muon born at O and detected 1907 m lower down: timelike, s² = +36 641 m², and no boost reorders it. P — a photon leaving the Sun at O and reaching Earth 499 s later: lightlike, s² = 0 exactly. X — an explosion 3.00 km away, 5.00 μs after O: spacelike, s² = −6.75×10⁶ m², and a boost at β = 0.4997 makes them simultaneous. Positions schematic; the numbers are exact

Figure

PUNCHLINE — the causality theorem in one picture. Since Δt′ = γ(Δt − βΔx/c) and γ > 0, the SIGN of Δt′ is the sign of the straight line 1 − β·β_signal, plotted here normalised. Its root is β_flip = 1/β_signal. The muon (β_signal = 0.995) has its root at β = 1.005 — OUTSIDE the physical strip, so no boost ever reorders it. The spacelike pair needs β_signal = 2.0014, so its root falls at β = 0.4997, INSIDE the strip: a boost at half light-speed makes the two explosions simultaneous, and anything faster reverses them

That figure is the entire article. Each line's slope is minus the speed of the signal you would need, in units of $c$ , and its root is that speed's reciprocal. Sub-light signal $\Rightarrow$ root outside the strip $\Rightarrow$ order absolute. Superluminal signal $\Rightarrow$ root inside the strip $\Rightarrow$ order negotiable. That is why nothing may travel faster than light: not because it would be fast, but because it would make "before" a matter of opinion. The muon's events, $M$ and $O$ , sit inside the future cone — a massive particle was present at both. The photon's pair sits exactly on the cone, in every frame. The explosion $X$ sits in the elsewhere, where "order" is a fact about the observer, not about the world.

The core equations

Everything begins with the interval, whose invariance is derived in The Invariant Spacetime Interval and is here taken as given. What this article adds is that the sign is where the physics lives.

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

where: $\Delta r^2 = \Delta x^2 + \Delta y^2 + \Delta z^2$ — the spatial separation; $\Delta t$ — the coordinate time separation, both measured in one inertial frame. Signature $(+,-,-,-)$ : timelike $s^2 > 0$ , lightlike $s^2 = 0$ , spacelike $s^2 < 0$ . Physical reading: $s^2$ compares the time light has against the distance it must cover. If the interval is invariant then so is its sign, and so is everything the sign encodes.

\boxed{\;s^{2} > 0 \;\;\text{TIMELIKE} \qquad s^{2} = 0 \;\;\text{LIGHTLIKE} \qquad s^{2} < 0 \;\;\text{SPACELIKE}\;} \qquad (2)

where: the three mutually exclusive, frame-independent cases. Physical reading: timelike means a massive particle can be present at both events, and some frame has them at the same place. Spacelike means nothing can connect them, and some frame has them at the same time. Lightlike means only a massless signal bridges them, and no frame does either. These are not three names but three different physical situations. The speed a signal would need in order to be present at both events is the cleanest way to see all of this at once.

\beta_{\text{signal}} \;\equiv\; \frac{\Delta r}{c\,\Delta t}, \qquad s^{2} = c^{2}\Delta t^{2}\left(1 - \beta_{\text{signal}}^{2}\right) \qquad (3)

where: $\beta_{\text{signal}}$ — the speed, in units of $c$ , a signal must have to connect the events. Physical reading: the sign of $s^2$ is now transparent. $\beta_{\text{signal}} < 1$ gives $s^2 > 0$ (timelike); $=1$ gives $s^2 = 0$ (light); $>1$ gives $s^2 < 0$ (spacelike — it would need a superluminal signal, so nothing can). The whole taxonomy answers one question: how fast would you have to go? Finally, the boost that reorders a pair — the quantity this article owns.

\beta_{\text{flip}} \;=\; \frac{c\,\Delta t}{\Delta x} \;=\; \frac{1}{\beta_{\text{signal}}} \qquad (4)

where: $\beta_{\text{flip}}$ — the speed of the boosted frame in which the two events become simultaneous; boosting faster than this (in the same direction) reverses their order. Physical reading: this is the theorem in one symbol. Reordering requires $|\beta_{\text{flip}}| < 1$ , which by (4) requires $|\beta_{\text{signal}}| > 1$ , which is exactly the spacelike condition. The events whose order can be changed are precisely the events that nothing could have travelled between. Causality is not protected by an extra postulate; it is protected by a reciprocal.

The three-way classification, derived

Derivation AThe sign of the interval is invariant, so the classification is absolute

The invariance of $s^2$ belongs to The Invariant Spacetime Interval and is not re-proved here. What we prove is the stronger statement: that the sign is invariant, and that each sign corresponds to a concrete, constructible physical situation — a frame in which the events are co-located, or one in which they are simultaneous, or neither. That upgrade, from "a number is preserved" to "a physical classification is absolute", makes causality a theorem rather than a hope. Signature $(+,-,-,-)$ throughout.

Step 1. Since $s'^2 = s^2$ exactly, $\mathrm{sgn}(s'^2) = \mathrm{sgn}(s^2)$ : every inertial observer sorts a given pair into the same one of the three bins. Why: trivial as algebra, enormous as physics. Frames disagree about $\Delta t$ and about $\Delta r$ ; they cannot disagree about which is winning.

Step 2 (timelike $\Rightarrow$ a co-located frame exists). Suppose $s^2 > 0$ , i.e. $c\Delta t > \Delta r$ , with $\Delta t > 0$ . Boost at $\beta = \Delta r/(c\Delta t) = \beta_{\text{signal}}$ , which satisfies $|\beta| < 1$ and is therefore legal. Then $\Delta x' = \gamma(\Delta r - \Delta r) = 0$ : the events happen at the same place, and $s^2 = c^2\Delta t'^2$ , so $\Delta t' = \sqrt{s^2}/c \equiv \Delta\tau$ . Why: this is what "timelike" means operationally — you can ride along with the pair, and if you do, the interval is just the time on your wristwatch.

Step 3 (spacelike $\Rightarrow$ a simultaneous frame exists). Suppose $s^2 < 0$ , i.e. $c\Delta t < \Delta x$ . Boost at $\beta = c\Delta t/\Delta x = \beta_{\text{flip}}$ , which now satisfies $|\beta| < 1$ — legal. Then $c\Delta t' = \gamma(c\Delta t - c\Delta t) = 0$ : the events are simultaneous, $s^2 = -\Delta x'^2$ , and $\sqrt{-s^2} = |\Delta x'| \equiv L_0$ . Why: the mirror of Step 2. Notice the two boosts are reciprocals of one another — and only one of them can ever be legal.

Step 4 (lightlike $\Rightarrow$ neither). If $s^2 = 0$ then $\beta_{\text{signal}} = 1$ exactly, and both of the boosts above would require $|\beta| = 1$ : unreachable. There is no frame in which a light pulse's emission and absorption are co-located, and none in which they are simultaneous. A photon has no rest frame, and this is the reason. Why: the null cone is not a limiting case you can sneak up on; it is a genuine boundary, preserved exactly by every boost.

Step 5. Name the regions, relative to an event $O$ : the future cone's interior ( $s^2>0$ , $\Delta t > 0$ ) is what $O$ can influence; the past cone's interior is what can influence $O$ ; the cone itself is the light-connectable boundary; and all $s^2 < 0$ is the elsewhere, which $O$ can neither affect nor observe. Why: the elsewhere is the genuinely new region — in Galilean physics an instantaneous "now" slices the whole universe and the elsewhere is empty. *Verified in verify_aq-sr-timelike-spacelike-lightlike-separations.py: the two constructed boosts are checked symbolically to give $\Delta x' = 0$ and $\Delta t' = 0$ respectively; the sign of $s^2$ is confirmed constant across a sweep of $\beta \in (-0.99, 0.99)$ for all three worked pairs; and $\beta_{\text{signal}}\,\beta_{\text{flip}} = 1$ identically.*

The reordering theorem proved in the next two cards is also proved — from the other side, the transformation side, via the flip velocity — in Relativity of Simultaneity, which owns that route. The proof here is deliberately interval-first: the taxonomy of the card above is the premise, not the conclusion, and the causal payoff falls out of the invariance of the sign of $s^2$ . That is why a single reciprocal, $\beta_{\text{flip}} = 1/\beta_{\text{signal}}$ , settles both cases at once.

Go deeper · Derivation B · 2 minTimelike and lightlike order is ABSOLUTE — no boost can reverse it

Derivation BTimelike and lightlike order is ABSOLUTE — no boost can reverse it

Here is the theorem that saves causality, and it is worth seeing that it is a strict inequality with nothing marginal about it. Take two events a physical signal could connect, so $s^2 \ge 0$ , with $A$ before $B$ in some frame. We prove $A$ precedes $B$ in every frame reachable by a boost with $|\beta| < 1$ — not "usually", not "for slow boosts", but in all of them, right up to the light barrier. The proof turns entirely on one bounded ratio.

Step 1. Write the time transformation for the separation between the events: $\Delta t' = \gamma\left(\Delta t - \dfrac{\beta\,\Delta x}{c}\right)$ . Take $\Delta t > 0$ (so $A$ precedes $B$ in the unprimed frame), and factor it out: $\Delta t' = \gamma\,\Delta t\left(1 - \beta\,\dfrac{\Delta x}{c\,\Delta t}\right)$ . Why: $\gamma > 0$ always, and $\Delta t > 0$ by assumption, so the entire question of the sign of $\Delta t'$ has been reduced to the sign of the bracket.

Step 2. Identify the ratio inside the bracket. It is exactly $\beta_{\text{signal}} = \Delta x/(c\Delta t)$ — the speed a signal would need. So $\Delta t' = \gamma\,\Delta t\,(1 - \beta\,\beta_{\text{signal}})$ . Why: the whole theorem now depends on a product of two speeds, and we happen to have a bound on each of them.

Step 3. Apply the two bounds. Timelike or lightlike means $|\beta_{\text{signal}}| \le 1$ . A physical boost means $|\beta| < 1$ . Therefore $|\beta\,\beta_{\text{signal}}| < 1$ strictly, so the bracket $(1 - \beta\beta_{\text{signal}})$ is strictly positive. Hence $\Delta t' > 0$ : the order is unchanged. Why: note the strictness. Even in the worst case — a light-like pair ( $\beta_{\text{signal}} = 1$ ) viewed from a frame at $\beta = 0.999999$ — the bracket is $10^{-6}$ , small but resolutely positive. The order never flips, and it never even reaches the flip.

Step 4. The consequence: if $A$ caused $B$ then they are timelike or lightlike separated (nothing else carries an influence), so every observer sees $A$ before $B$ . Cause precedes effect in all inertial frames. Why: this is what it was all for. Relativity makes simultaneity relative without making causality relative, because the two are not the same question. *Verified in verify_aq-sr-timelike-spacelike-lightlike-separations.py: for the timelike muon pair and a null pair, $\Delta t'$ is computed at 199 boosts spanning $\beta \in (-0.995, 0.995)$ and asserted strictly positive at every one; the bracket $1 - \beta\beta_{\text{signal}}$ is asserted $> 0$ symbolically under $|\beta|<1$ , $|\beta_{\text{signal}}|\le 1$ .*

Go deeper · Derivation C · 2 minSpacelike order is ALWAYS reversible — and the boost that does it is a reciprocal

Derivation CSpacelike order is ALWAYS reversible — and the boost that does it is a reciprocal

The converse is just as sharp, and it is the half students find alarming. For a spacelike pair there is not merely sometimes a frame that reverses the order — there is always one, and we can write it down. The construction also delivers this article's owned identity, $\beta_{\text{flip}} = 1/\beta_{\text{signal}}$ , from which the whole causal structure of relativity can be read at a glance. The alarm, we shall see, is misplaced — and for a reason that is not a dodge.

Step 1. Take $s^2 < 0$ , i.e. $|\Delta x| > c|\Delta t|$ , with $\Delta t > 0$ . From the previous card, $\Delta t' = \gamma\Delta t\,(1 - \beta\,\beta_{\text{signal}})$ , and now $|\beta_{\text{signal}}| = |\Delta x|/(c\Delta t) > 1$ . Why: the same algebra as before, but the bound has flipped — and that is the only thing that has changed.

Step 2. Solve for the boost that makes the bracket vanish: $\beta_{\text{flip}} = 1/\beta_{\text{signal}} = c\,\Delta t/\Delta x$ . Because $|\beta_{\text{signal}}| > 1$ , we have $|\beta_{\text{flip}}| < 1$ — a perfectly legal, physically attainable boost. In that frame $\Delta t' = 0$ : the two events are simultaneous. Why: the existence of the simultaneous frame is not an accident of a convenient example. It follows from an inequality that spacelike separation guarantees.

Step 3. Push past it. For any $\beta$ with $\beta_{\text{flip}} < \beta < 1$ (taking everything positive), the bracket $1 - \beta\beta_{\text{signal}}$ turns negative, so $\Delta t' < 0$ and the order is reversed. Different observers genuinely disagree about which of the two events happened first. Why: and there is a continuum of them. The "order" of a spacelike pair is not a fact about the world; it is a fact about who is looking.

Step 4. Read the owned identity: $\beta_{\text{flip}}\,\beta_{\text{signal}} = 1$ . The boost required to reorder a pair is the reciprocal of the speed of the signal that would connect them. Sub-light signal ( $\beta_{\text{signal}} < 1$ ) $\Rightarrow$ super-light boost needed ( $\beta_{\text{flip}} > 1$ ) $\Rightarrow$ impossible. Superluminal signal ( $\beta_{\text{signal}} > 1$ ) $\Rightarrow$ sub-light boost suffices $\Rightarrow$ reordering happens. Why: this is why "nothing travels faster than light" and "cause precedes effect" are the same statement. A tachyonic telegraph would not merely be fast; it would be a machine for making some observer see the reply before the message.

Step 5. Defuse the alarm. Reversal of a spacelike pair is no causal paradox, because no influence can pass between spacelike-separated events — that is what $s^2 < 0$ means. There is no cause to precede and no effect to follow, only two unrelated happenings whose ordering is bookkeeping. Why: relativity relativises exactly the order that carries no information and protects exactly the order that does. That fit is not a coincidence — it is the same inequality, read twice. *Verified in verify_aq-sr-timelike-spacelike-lightlike-separations.py: for the spacelike pair ( $\Delta x = 3000$ m, $\Delta t = 5$ µs), $\beta_{\text{flip}} = 0.49965$ is computed, $|\beta_{\text{flip}}| < 1$ asserted, $\Delta t'$ asserted zero at $\beta_{\text{flip}}$ and negative for $\beta > \beta_{\text{flip}}$ ; at $\beta = 0.8$ , $\Delta t' = -5.009$ µs and $s'^2 = s^2$ to machine precision.*

Limiting cases

Case	Result	Meaning
$\Delta r = 0$ (same place)	$s^2 = c^2\Delta t^2 > 0$	always timelike; $\Delta\tau = \Delta t$ is the proper time
$\Delta t = 0$ (simultaneous)	$s^2 = -\Delta r^2 < 0$	always spacelike. Two distinct simultaneous events are never causally connected
$\beta_{\text{signal}} = 1$	$s^2 = 0$	the light cone; no co-moving frame, no simultaneous frame — a photon has no rest frame
$\beta_{\text{signal}} \to \infty$	$s^2 \to -\Delta r^2$ , $\beta_{\text{flip}} \to 0$	deep in the elsewhere: essentially a pure distance, and any boost reorders it
Galilean limit, $c \to \infty$	every pair timelike	the cone opens out to all of space; the elsewhere vanishes and absolute simultaneity returns

Common trapThe sign convention: half the literature writes every one of these signs backwards

This article, and this entire track, uses the timelike-positive signature $(+,-,-,-)$ :

s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2, \qquad s^2 > 0 \;\text{timelike}, \quad s^2 < 0 \;\text{spacelike}.

A large and entirely respectable fraction of the literature — including Misner, Thorne and Wheeler, and most general-relativity texts descended from it — uses the spacelike-positive convention $(-,+,+,+)$ :

\Delta s^2 = -c^2\Delta t^2 + \Delta x^2 + \Delta y^2 + \Delta z^2, \qquad \Delta s^2 < 0 \;\text{timelike}, \quad \Delta s^2 > 0 \;\text{spacelike}.

Every sign flips: proper time becomes $\sqrt{-\Delta s^2}/c$ , proper length becomes $\sqrt{\Delta s^2}$ . No physical statement changes. Both conventions agree on which events are timelike, on every measured time and distance, and on the whole causal structure. The trap is not picking the "wrong" one — there is no wrong one. The trap is mixing them mid-calculation, the single most productive source of sign errors in relativity, and it happens most often when a formula is copied from one book into a derivation begun in another. Guard yourself with the statement true in both conventions: > The sign of the interval is boost-invariant. Not "the interval is positive" — that is a convention. The invariant claim is that whichever sign a timelike pair has, it keeps in every frame. Assert " $s^2 > 0$ for timelike" as though it were a law of nature and you have mistaken bookkeeping for physics. The law of nature is the three-way partition itself, and that no boost moves an event across the cone.

Common trapReordering a spacelike pair is not a causality violation — but a superluminal signal would be

Students often meet the fact that boosts reverse the order of spacelike events and conclude relativity has a causality problem it then papers over. The opposite is true. No influence passes between spacelike-separated events, so when two frames disagree about which came first they disagree about the order of two unrelated happenings. Nothing depends on the answer, so none is needed. What would be fatal is a signal that did connect them. Suppose a device sent information at $\beta_{\text{signal}} = 2$ . Then by equation (4) there is a perfectly ordinary frame, moving at $\beta_{\text{flip}} = 0.5$ , in which the message is received before it is sent — and a confederate in that frame could send it back, arriving before the original was transmitted. That is not a paradox awaiting cleverness; it is a contradiction. The prohibition on superluminal signalling is therefore not an extra rule bolted on to keep relativity tidy. It is forced by equation (4): a signal faster than light is, in some frame, a signal into the past.

Worked examples

1 · Classify the muon, and prove nobody can see it decay before it was born

In the Frisch–Smith experiment, muons are counted at the summit of Mount Washington ( $1917\ \mathrm{m}$ above sea level) and $1907\ \mathrm{m}$ below it in Cambridge, Massachusetts, travelling at $\beta = 0.995$ . Take creation and detection as the two events. Classify the pair, and find the boost that would be needed to reverse their order.

Step 1 — get the coordinate separations. The flight path is the detector-to-detector drop, $\Delta x = 1907\ \mathrm{m}$ — not the $1917\ \mathrm{m}$ summit elevation, which is a different quantity measured from sea level. The flight time follows: $c\Delta t = \Delta x/\beta = 1907/0.995 = 1916.58\ \mathrm{m}$ , i.e. $\Delta t = 6.393\ \mathrm{\mu s}$ .

Step 2 — classify. $s^2 = (1916.58)^2 - (1907)^2 = \mathbf{+36\,641\ m^2} > 0$ : timelike. It had better be — a muon was physically present at both events, so a massive particle made the trip, and that is the definition.

Step 3 — the proper time comes free. $\sqrt{s^2} = 191.42\ \mathrm{m}$ , so $\Delta\tau = 0.6385\ \mathrm{\mu s}$ — comfortably under the $2.20\ \mathrm{\mu s}$ proper lifetime, which is why so many muons survive the trip.

Step 4 — the reordering boost. $\beta_{\text{flip}} = c\Delta t/\Delta x = 1916.58/1907 = \mathbf{1.00503}$ . Greater than one: no such frame exists, and none ever will. Equivalently $\beta_{\text{flip}} = 1/\beta_{\text{signal}} = 1/0.995$ — the muon travelled at $99.5\%$ of $c$ , so reordering its birth and death would need a frame at $100.5\%$ of $c$ .

Step 5 — the boundary number. Check the worst legal case: at $\beta = 0.999$ , the bracket is $1 - 0.999 \times 0.995 = 0.00600 > 0$ . Down to six parts in a thousand — but positive, and it stays positive right up to $\beta \to 1$ . No observer, however fast, sees the muon decay before it was created.

2 · Two explosions, and the observer who sees them in the wrong order

Two explosions occur $\Delta x = 3.00\ \mathrm{km}$ apart, the second $\Delta t = 5.00\ \mathrm{\mu s}$ after the first, as measured in the lab. Classify them. Is there an observer who sees them simultaneously? One who sees them in the reverse order? Could the first have caused the second?

Step 1 — put time in metres. $c\Delta t = (2.998\times10^8)(5.00\times10^{-6}) = 1498.96\ \mathrm{m}$ , against $\Delta x = 3000\ \mathrm{m}$ .

Step 2 — classify. $s^2 = (1498.96)^2 - (3000)^2 = 2\,246\,884 - 9\,000\,000 = \mathbf{-6.75\times10^{6}\ m^2} < 0$ : spacelike. The two events lie in each other's elsewhere. The proper distance between them is $\sqrt{-s^2} = 2598.7\ \mathrm{m}$ — the separation measured in the frame where they are simultaneous.

Step 3 — could the first have caused the second? Only if something covered $3000\ \mathrm{m}$ in $5.00\ \mathrm{\mu s}$ , which is $\beta_{\text{signal}} = 3000/1498.96 = \mathbf{2.0014}$ — twice the speed of light. No. The explosions are causally independent, whatever they looked like.

Step 4 — the simultaneous observer. $\beta_{\text{flip}} = 1/\beta_{\text{signal}} = 1/2.0014 = \mathbf{0.4997}$ . An observer moving at just under half the speed of light, in the direction from the first explosion to the second, sees them go off at the same instant. This is a perfectly ordinary frame — the sort a fast spacecraft is in.

Step 5 — the observer who sees them backwards. Take $\beta = 0.8 > \beta_{\text{flip}}$ , so $\gamma = 5/3$ :

\Delta t' = \gamma\left(\Delta t - \frac{\beta\,\Delta x}{c}\right) = \tfrac{5}{3}\,(5.00 - 8.006)\ \mathrm{\mu s} = \mathbf{-5.009\ \mu s}.

Negative: for this observer the second explosion happened first, by five microseconds. The interval is untouched — $\Delta x' = 3001.4\ \mathrm{m}$ and $s'^2 = -6.75\times10^6\ \mathrm{m^2}$ ✓, the same number as Step 2. Three observers, three orderings of the same two events, and no contradiction anywhere, because neither explosion could ever have known about the other.

Physics of the portrait

Causal order is the concept that survived the demolition. Einstein removed absolute simultaneity in 1905 and appeared to remove absolute time order with it — an impression it took Minkowski's geometry to correct.

Origins · 2 minCausal order and how this was discovered

Causal order

1905–1914 · Einstein removes absolute simultaneity · Minkowski draws the cone · Robb rebuilds spacetime out of "before" and "after"

The three-way classification is Minkowski's, announced with the rest of spacetime geometry in his 1908 lecture "Raum und Zeit"; he supplied the light cone that makes it visible. But the deepest exploration of causal order belongs to Alfred Arthur Robb (1873–1936), a Belfast-born physicist working largely alone at Cambridge, whose Optical Geometry of Motion (1911) and A Theory of Time and Space (1914) attempted something nobody had asked for: an axiomatic reconstruction of Minkowski spacetime taking the single relation "event A can influence event B" — the causal, or "conical", order — as its sole primitive. No metric, no coordinates, no interval; just before and after, and the cone they define. Robb showed the geometry is recoverable from that relation alone up to an overall scale — causal order fixes the light cones, but it cannot fix the unit of length — a result sharpened by Zeeman (1964), who proved that in any spacetime of dimension ≥ 3 a causal-order-preserving bijection can only be a Lorentz transformation, a translation and a dilatation. Causal order is therefore not a consequence of the geometry; it very nearly is the geometry. Robb is better remembered for something smaller: it was he who, in 1911, gave the rapidity its name.

History

1905

— Einstein shows simultaneity is frame-dependent, raising the immediate question of whether temporal order is too.

1908

— Minkowski, in "Raum und Zeit", introduces the light cone and the timelike/lightlike/spacelike classification, showing that order is relative only outside the cone.

1911

— Robb, in Optical Geometry of Motion, begins the axiomatic reconstruction of spacetime from the causal ("conical") order alone.

1914

— Robb's A Theory of Time and Space completes the programme: the geometry of Minkowski spacetime is recoverable, up to an overall scale factor, from the relation "A can influence B" and nothing else.

1964

— Zeeman ("Causality Implies the Lorentz Group", J. Math. Phys. 5, 490) proves that in four dimensions — indeed in any dimension ≥ 3 — every causal-order-preserving bijection of Minkowski spacetime is a Lorentz transformation, a translation and a dilatation. The 1+1-dimensional case, the one drawn in this article's figures, is the exception: there the causal automorphisms form an infinite-dimensional group.

Going further

Texts. Taylor & Wheeler, Spacetime Physics, 2nd ed., Ch. 6 — the source of the word "elsewhere" · Rindler, Introduction to Special Relativity, 2nd ed., §§7–8 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.1.3 · Hartle, Gravity, Ch. 4. On the deep structure: A. A. Robb, A Theory of Time and Space (Cambridge, 1914); E. C. Zeeman, "Causality implies the Lorentz group", J. Math. Phys. 5, 490 (1964).

Provenance. ⚠ The AllQFT base for this article was written in the opposite signature $(-,+,+,+)$ throughout — twenty separate sign statements, including "timelike means $\Delta s^2 < 0$ " — and every one has been re-derived here rather than transcribed, to match the track's locked $(+,-,-,-)$ . The classification and the "elsewhere" follow Taylor & Wheeler Ch. 6; the invariance of $s^2$ is imported from The Invariant Spacetime Interval, not re-proved, and the reordering theorem is also owned, by the transformation route, by Relativity of Simultaneity. The identity $\beta_{\text{flip}} = 1/\beta_{\text{signal}}$ and the strict positivity of the bracket are derived and machine-checked above. Every number is recomputed in verify_aq-sr-timelike-spacelike-lightlike-separations.py: the Frisch–Smith muon ( $\Delta x = 1907\ \mathrm{m}$ flight path, summit $1917\ \mathrm{m}$ above sea level, $s^2 = +36\,641\ \mathrm{m^2}$ , $\beta_{\text{flip}} = 1.00503$ ); the null pair ( $\Delta t = 499\ \mathrm{s}$ , $s^2 = 0$ ); and the spacelike pair ( $s^2 = -6.753\times10^6\ \mathrm{m^2}$ , $\beta_{\text{signal}} = 2.0014$ , $\beta_{\text{flip}} = 0.49965$ , $\Delta t' = -5.009\ \mathrm{\mu s}$ at $\beta = 0.8$ ).

Builds on. The Invariant Spacetime Interval · The Two Postulates of Special Relativity · The Lorentz Transformation Equations · Relativity of Simultaneity. Leads to. Proper Time · Proper Length · Spacetime Diagrams · Events and Worldlines.

If two observers disagree about which event happened first, isn't one of them simply wrong?

No — and this is the hardest sentence in the subject to accept. For a spacelike pair there is no fact of the matter about which came first, just as there is none about whether two events are "really" simultaneous. Both observers correctly report what their own synchronized clocks record, and no experiment adjudicates, because no signal connects the events. What relativity denies is not that each observer has an answer, but that the answers must agree — and it denies this only where agreement is unnecessary. For timelike and lightlike pairs, where an influence could have passed, all observers agree, always.

Can a boost move an event across the light cone if I boost hard enough?

Never, and the reason is structural rather than numerical. Crossing the cone would mean changing the sign of $s^2$ — and you cannot change the sign of a thing that does not change. Geometrically, a boost slides an event along the hyperbola $c^2t^2 - x^2 = \text{const}$ it sits on, and those hyperbolae approach the light lines asymptotically without ever touching them. The cone is the hinge, and the hinge does not move.

Two events are simultaneous in my frame. What is their separation type?

Spacelike, always — provided they are genuinely distinct. With $\Delta t = 0$ , equation (1) gives $s^2 = -\Delta r^2 < 0$ for any $\Delta r \neq 0$ . A small result with a large consequence: your whole three-dimensional "now" slice consists entirely of events that are mutually causally disconnected. Nothing in your present can influence anything else in your present — which is why the relativity of simultaneity is safe. The slice you disagree about is one on which nothing can happen anyway.

Does a photon experience zero time? Does it have a rest frame?

The interval between a photon's emission and absorption is exactly zero in every frame, and it is tempting to summarise this as "the photon experiences no time". Resist. Proper time is the reading of a clock present at both events, and such a clock needs a frame in which the events are co-located — which for a null pair requires $|\beta| = 1$ , and no such boost exists. A photon has no rest frame, so there is no clock to read, and "what the photon experiences" is not a question the theory answers. The honest statement is $s^2 = 0$ .

My textbook says timelike means Δs² < 0. Is one of us wrong?

Neither. Your book uses the $(-,+,+,+)$ signature, where $\Delta s^2 = -c^2\Delta t^2 + \Delta r^2$ and every sign here is reversed; ours is timelike-positive $(+,-,-,-)$ . The two agree on every measurable quantity — which pairs are timelike, every proper time and proper length, the whole causal structure — and differ only in bookkeeping. Pick one, declare it, never mix them inside a calculation. And memorise the convention-free statement: the sign of the interval is boost-invariant. That is the physics; "timelike is positive" is the paperwork.

What exactly is the "elsewhere", and why does it have no Newtonian counterpart?

It is the set of events spacelike-separated from you: too far away in space for light to have covered the distance in the time available, either way. You cannot have influenced them, they cannot have influenced you, and no experiment establishes which came first. In Newtonian physics this region is empty — signals may travel arbitrarily fast, so every event is either in your past or your future, "now" is a single unambiguous slice through all of space, and every pair of events has a definite order. The elsewhere is what appears when the speed of light becomes finite and universal: the geometric price of causality.