Timelike, Spacelike and Null Separations

A muon is created about $10\ \mathrm{km}$ up, when a cosmic-ray proton smashes into the top of the atmosphere, and it lives — by its own reckoning — about two microseconds. Even at the speed of light two microseconds buys only $600\ \mathrm{m}$ . Almost none should reach the ground. They arrive in droves. Two events, then: muon born and muon detected in a Cambridge basement. In the frame of the mountain the events are separated by $1907\ \mathrm{m}$ of altitude and $6.4$ microseconds of time. In the muon's own frame they are separated by no distance at all — the muon never goes anywhere; the mountain comes up to meet it — and by only $0.64$ microseconds. Nothing the two observers write down agrees. Yet if each of them computes the single combination \Delta s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2, they get the identical number, and its sign tells them something neither could have learned from their own coordinates: that these two events are timelike separated, that a signal slower than light joins them, and that no observer anywhere in the universe will ever say the detection came first. That sign is the subject of this article. Every pair of events in spacetime falls into exactly one of three invariant classes — timelike, null, spacelike — and the classification is the deepest thing two events possess. It survives every boost, it decides what can cause what, and it comes with two existence theorems that turn out, on inspection, to be one theorem wearing two hats.

Key takeaways

For any two events, the interval $\Delta s^2 = c^2\Delta t^2 - \Delta r^2$ is the same number in every inertial frame. Its sign sorts the pair into one of three classes, and that classification is absolute.
Timelike ( $\Delta s^2 > 0$ , $c|\Delta t| > |\Delta r|$ ): joinable by a sub-light signal; time order absolute; $\sqrt{\Delta s^2} = c\,\Delta\tau$ is what a clock carried between them reads. Null ( $\Delta s^2 = 0$ ): joined only by light. Spacelike ( $\Delta s^2 < 0$ ): no signal can pass; $\sqrt{-\Delta s^2} = L_0$ is the proper length.
Existence theorem 1. If the pair is timelike, there is a frame in which the two events happen at the same place, and it moves at $v_{\mathrm{col}} = \Delta x/\Delta t$ .
Existence theorem 2. If the pair is spacelike, there is a frame in which the two events are simultaneous, and it moves at $v_{\mathrm{sim}} = c^2\Delta t/\Delta x$ .
Key idea: those two theorems are one theorem. The two candidate boost speeds are reciprocals through $c$ — $v_{\mathrm{col}}\,v_{\mathrm{sim}} = c^2$ identically, so $\beta_{\mathrm{col}}\,\beta_{\mathrm{sim}} = 1$ — and of a positive number and its reciprocal exactly one is less than one, unless both equal one. Hence the co-location frame exists precisely when the simultaneity frame does not, timelike and spacelike are mutually exclusive and jointly exhaustive, and the single case where both boosts would need exactly $c$ is the null case sitting between them.

The core insight

The classification is not a taxonomy imposed on spacetime; it is spacetime's own answer to a physical question — can a signal get from here to there? Compare the distance the events are apart with the distance light could cover in the time they are apart. If light has time to spare, the pair is timelike and a material object can make the trip. If light exactly makes it, the pair is null. If light cannot make it, the pair is spacelike and nothing can. Because both quantities transform, but the comparison does not, every observer in the universe returns the same verdict.

Where this lives today

Every relativistic theory is built so that its equations respect the classification. In electrodynamics the retarded Green's function is supported on the past light cone: a field at an event depends only on sources timelike or null separated from it, and never on spacelike ones. In quantum field theory the same requirement is the axiom of microcausality — spacelike-separated field operators commute — and it is what forces antiparticles into existence. In particle physics the classification becomes a kinematic bookkeeping tool. The Mandelstam variables of a scattering process are intervals in momentum space, and the physical region of a scattering amplitude is delimited by exactly the same sign conditions: whether a momentum transfer is "spacelike" ( $t < 0$ , the region probed in deep inelastic scattering) or "timelike" ( $t > 0$ , the region of resonance production) is the single most important thing you can say about an experiment. The vocabulary of this article — timelike, spacelike, on-shell, off-shell — is spoken daily in places where the light cone is never drawn. And in gravitational physics the classification is the whole framework. Because the metric is dynamical, one asks not "what is the coordinate of this event" but "which events are timelike separated from which", and that question defines horizons, Cauchy surfaces and the very notion of a spacetime being predictable.

The physical picture

Put the first event at the origin and the second at $(c\Delta t, \Delta x)$ , with the diagram scaled so light travels at $45^\circ$ . There are exactly three places the second event can be: inside the cone, on it, or outside it. The three positions are the three classes, and the question that decides between them is childishly simple — is $c|\Delta t|$ bigger than, equal to, or smaller than $|\Delta x|$ ? Equivalently: could a signal at speed $|\Delta x/\Delta t|$ have made the journey without exceeding $c$ ? The subtlety is that $\Delta t$ and $\Delta x$ are both frame-dependent, and both change under a boost. A pair of events an observer calls " $3$ seconds apart and $5$ light-seconds apart" another calls something else entirely. What no observer can change is which side of the cone the second event lies on, because $\Delta s^2$ is invariant and its sign is the answer. The classification is more robust than any of the quantities that go into it. And each class comes with a privileged frame — a frame in which the pair looks as simple as it can possibly look. For a timelike pair, there is an observer who never has to move: she sits still and both events happen right where she is, one after the other. Her clock reads the proper time, and the interval is pure time. For a spacelike pair there is instead an observer who sees both events at the same instant, spread along her ruler; the interval is pure space, and her ruler reads the proper length. The next two derivations construct these observers, and the third shows why you can never have both.

Figure

The three classes, and their two privileged frames. Left: with the cone at 45°, the second event lands inside (timelike), on (null), or outside (spacelike). Right: the timelike pair (cΔt, Δx) = (5, 3) ls has Δs² = +16 ls² and a CO-LOCATION frame moving at β = 3/5, whose tilted ct′ axis passes through both events. The spacelike pair (3, 5) has Δs² = −16 ls² and a SIMULTANEITY frame at β = 3/5, whose tilted x′ axis passes through both. Same 3 and 5, mirrored — and the classification flips

The core equations

The whole apparatus is one quadratic form and its sign. In the chapter's locked signature $(+,-,-,-)$ :

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

where: $\Delta r^2 = \Delta x^2 + \Delta y^2 + \Delta z^2$ is the ordinary squared distance. Physical reading: $\Delta s^2$ compares $c\Delta t$ (how far light could travel in the time available) with $\Delta r$ (how far it must travel). Timelike is POSITIVE in this chapter. The convention $(-,+,+,+)$ — used by Taylor–Wheeler and MTW — flips every sign in this article; the physics is identical and every intermediate sign is not.

\begin{cases} \Delta s^2 > 0 & \textbf{timelike}: \ c|\Delta t| > |\Delta r| \\ \Delta s^2 = 0 & \textbf{null}: \ c|\Delta t| = |\Delta r| \\ \Delta s^2 < 0 & \textbf{spacelike}: \ c|\Delta t| < |\Delta r| \end{cases} \qquad (2)

where: the three exhaustive, mutually exclusive classes. Physical reading: read them as a statement about a signal. The speed a signal would need to join the events is $v_{\text{sig}} = |\Delta r|/|\Delta t|$ , and the three cases are $v_{\text{sig}} < c$ , $=c$ , $>c$ . The classes are the signal-speed question. The two classes with a nonzero interval each name a physical quantity.

\text{timelike:} \quad \Delta s^2 = +c^2\Delta\tau^2 \ \Rightarrow \ \Delta\tau = \frac{\sqrt{\Delta s^2}}{c}; \qquad \text{spacelike:} \quad \Delta s^2 = -L_0^2 \ \Rightarrow \ L_0 = \sqrt{-\Delta s^2} \qquad (3)

where: $\Delta\tau$ — the proper time, read by a clock present at both events; $L_0$ — the proper length, read by a ruler for which the events are simultaneous. Physical reading: the minus sign in $L_0 = \sqrt{-\Delta s^2}$ is not optional. For a spacelike pair $\Delta s^2$ is negative, and taking its bare square root gives an imaginary number. Both quantities are constructed to be real, and each is real for exactly one class. (Proper time and proper length are developed in Proper Time and Proper Length.) Finally the two boosts this article is about. Under a boost of speed $v$ along $x$ ,

\Delta t' = \gamma\!\left(\Delta t - \frac{v\,\Delta x}{c^2}\right), \qquad \Delta x' = \gamma\left(\Delta x - v\,\Delta t\right) \qquad (4)

where: $\gamma = (1-v^2/c^2)^{-1/2}$ . Physical reading: set the second to zero and you have asked for a frame in which the events are co-located; set the first to zero and you have asked for one in which they are simultaneous. Each request produces a unique candidate speed, and the rest of the article is about which requests can be granted.

The two existence theorems

Derivation AThe class is a Lorentz invariant, so all observers agree on it

Before either existence theorem can mean anything, we need to know that the three classes are frame-independent — otherwise "timelike" would be an opinion, and the causal claims that rest on it would evaporate. The goal is to show that $\Delta s^2$ is unchanged by a boost, and to be careful about it, because this is the step where the base article for this entry was written in the opposite signature and every sign below had to be rebuilt from scratch.

Step 1. Boost along $x$ with speed $v$ , so $c\Delta t' = \gamma(c\Delta t - \beta\,\Delta x)$ and $\Delta x' = \gamma(\Delta x - \beta\,c\Delta t)$ , with $\beta = v/c$ and $\Delta y' = \Delta y$ , $\Delta z' = \Delta z$ . Form the primed interval $\Delta s'^2 = c^2\Delta t'^2 - \Delta x'^2 - \Delta y'^2 - \Delta z'^2$ . Why: we are testing invariance, not assuming it — the transverse pieces ride along untouched and can be set aside.

Step 2. Expand the two squares: c^2\Delta t'^2 - \Delta x'^2 = \gamma^2\Big[(c\Delta t)^2 - 2\beta\,c\Delta t\,\Delta x + \beta^2\Delta x^2\Big] - \gamma^2\Big[\Delta x^2 - 2\beta\,c\Delta t\,\Delta x + \beta^2 (c\Delta t)^2\Big]. The cross terms $-2\beta\,\gamma^2\,c\Delta t\,\Delta x$ appear in both brackets with the same sign and cancel on subtraction. Why: that cancellation is the entire mechanism. It is why the interval is quadratic and why no linear-in- $v$ effect can survive.

Step 3. Collect what is left: $\gamma^2\big[(c\Delta t)^2(1 - \beta^2) - \Delta x^2(1-\beta^2)\big] = \gamma^2(1-\beta^2)\big[(c\Delta t)^2 - \Delta x^2\big]$ . Since $\gamma^2(1-\beta^2) = 1$ by definition of $\gamma$ , this is exactly $(c\Delta t)^2 - \Delta x^2$ . Adding back $-\Delta y^2 - \Delta z^2$ gives $\Delta s'^2 = \Delta s^2$ . Why: the factor $\gamma^2(1-\beta^2) = 1$ is the Lorentz factor doing the only job it was invented for. Invariance of the interval and the formula $\gamma = 1/\sqrt{1-\beta^2}$ are the same fact.

Step 4. Conclude about the sign. Since $\Delta s'^2 = \Delta s^2$ exactly, for every boost and every pair, the sign is invariant too, and with it the classification. No boost can carry a timelike pair to a spacelike one, or an event across the light cone. Why: this is what licenses everything that follows. "Timelike" is not a description of a pair of events in a frame; it is a property of the pair. *Verified in verify_aq-sr-timelike-spacelike-null-separations.py: $\Delta s'^2 - \Delta s^2$ simplifies to zero symbolically, and over 5000 random pairs and boosts the sign of $\Delta s^2$ never changes. The convention gate asserts timelike $\Rightarrow \Delta s^2 > 0$ , spacelike $\Rightarrow \Delta s^2 < 0$ , $c\Delta\tau = +\sqrt{\Delta s^2}$ , $L_0 = \sqrt{-\Delta s^2}$ — all four of which the base had inverted.*

Go deeper · Derivation B · 2 minTimelike pairs have a co-location frame; spacelike pairs have a simultaneity frame

Derivation BTimelike pairs have a co-location frame; spacelike pairs have a simultaneity frame

Each nonzero class comes with an observer for whom the pair is as simple as possible. The goal is to construct both, and to notice that in each case the construction hands you a specific speed — a number you must then check is physically attainable. That check is where the two theorems get their teeth, and where the next card gets its punchline.

Step 1. Ask for co-location: a frame in which the two events happen at the same place, $\Delta x' = 0$ . From equation (4), $\gamma(\Delta x - v\Delta t) = 0$ , and since $\gamma \neq 0$ the unique answer is v_{\mathrm{col}} = \frac{\Delta x}{\Delta t}. Why: this is exactly the average velocity a traveller would need to be present at both events, which is the obvious answer — but it had to come out of the transformation, not out of intuition.

Step 2. Check that it is legal. $|v_{\mathrm{col}}| < c \iff |\Delta x| < c|\Delta t| \iff \Delta s^2 > 0$ : the boost is attainable exactly when the pair is timelike, and never otherwise. In that frame $\Delta x' = 0$ and equation (1) collapses to $\Delta s^2 = c^2\Delta t'^2$ , so $\Delta t' = \sqrt{\Delta s^2}/c = \Delta\tau$ : the interval is pure time, and the co-located observer's clock reads it directly. Why: the interval of a timelike pair is not merely like a time — in one particular frame it is the time, and the frame is the one that stays put.

Step 3. Now ask for simultaneity: a frame in which the events happen at the same moment, $\Delta t' = 0$ . From equation (4), $\gamma(\Delta t - v\Delta x/c^2) = 0$ , so v_{\mathrm{sim}} = \frac{c^2\,\Delta t}{\Delta x}. Why: note the $c^2$ , and note that $\Delta t$ and $\Delta x$ have swapped roles relative to Step 1. That inversion is not decorative; it is the whole of the next card.

Step 4. Check that one. $|v_{\mathrm{sim}}| < c \iff c|\Delta t| < |\Delta x| \iff \Delta s^2 < 0$ : the simultaneity boost is attainable exactly when the pair is spacelike. In that frame $\Delta t' = 0$ and equation (1) gives $\Delta s^2 = -\Delta x'^2$ , so $|\Delta x'| = \sqrt{-\Delta s^2} = L_0$ : the interval is pure space, and the proper length is what a ruler laid between the two simultaneous events reads. Why: and here is the minus sign that the base article had backwards. For a spacelike pair $\Delta s^2$ is negative, and the proper length is the square root of minus it. Get that wrong and every length in the chapter becomes imaginary. *Verified in verify_aq-sr-timelike-spacelike-null-separations.py: $\Delta x' = 0$ solves to $v = \Delta x/\Delta t$ and $\Delta t' = 0$ to $v = c^2\Delta t/\Delta x$ ; substituting each back gives $\Delta t'^2 = \Delta s^2/c^2$ and $\Delta x'^2 = -\Delta s^2$ respectively.*

The owned result: one inequality, read twice

Two theorems, two boosts, two conditions. Put the two speeds side by side, though, and something falls out that no textbook seems to say out loud.

Go deeper · Derivation C · 2 minThe two existence theorems are the same inequality — the boosts are reciprocals through c

Derivation CThe two existence theorems are the same inequality — the boosts are reciprocals through c

We have two candidate boosts, $\beta_{\mathrm{col}} = \Delta x/(c\Delta t)$ and $\beta_{\mathrm{sim}} = c\Delta t/\Delta x$ , each legal under a different condition. It looks like a coincidence that the conditions are complementary. It is not a coincidence, and this card is the reason: the two candidate speeds are not independent numbers at all. They are reciprocals, always, for every pair of events in the universe — and of a positive real number and its reciprocal, exactly one is less than one, unless both are exactly one.

Step 1. Multiply them. \beta_{\mathrm{col}}\;\beta_{\mathrm{sim}} = \frac{\Delta x}{c\,\Delta t} \cdot \frac{c\,\Delta t}{\Delta x} = 1 \qquad\Longleftrightarrow\qquad v_{\mathrm{col}}\;v_{\mathrm{sim}} = c^2. This is an identity: no assumption about the pair, no condition, no approximation. Why: the co-location boost and the simultaneity boost are conjugate through the speed of light, exactly as the object and image distances of a lens are conjugate through its focal length. Knowing one, you know the other for free.

Step 2. Apply the trichotomy of the reals. For any $\beta_{\mathrm{col}} > 0$ with $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} = 1$ : if $\beta_{\mathrm{col}} < 1$ then $\beta_{\mathrm{sim}} > 1$ ; if $\beta_{\mathrm{col}} > 1$ then $\beta_{\mathrm{sim}} < 1$ ; and $\beta_{\mathrm{col}} = 1$ forces $\beta_{\mathrm{sim}} = 1$ . There is no fourth possibility, and there is no case in which both are subluminal. Why: a two-line piece of arithmetic has just proved a theorem about spacetime. You can never have both frames. No observer can find the events both co-located and simultaneous — that would make the interval both pure time and pure space, i.e. zero, i.e. the events identical.

Step 3. Translate each branch. $\beta_{\mathrm{col}} < 1$ means $|\Delta x| < c|\Delta t|$ , which is $\Delta s^2 > 0$ — timelike, and the branch on which the co-location frame is legal and the simultaneity boost is superluminal. $\beta_{\mathrm{col}} > 1$ means $\Delta s^2 < 0$ — spacelike, and the branch on which the simultaneity frame is legal and the co-location boost is superluminal. Why: the two "separate" existence theorems have collapsed into the two halves of one dichotomy. The co-location boost exists precisely when the simultaneity boost does not, and the reason is a reciprocal.

Step 4. Look at the remaining case. $\beta_{\mathrm{col}} = \beta_{\mathrm{sim}} = 1$ means $|\Delta x| = c|\Delta t|$ , that is $\Delta s^2 = 0$ — null. Both candidate boosts would require exactly $c$ , and neither is attainable: a null pair has no co-location frame and no simultaneity frame. It is the fixed point of the reciprocal map, sitting exactly between the two branches. Why: the null class is not a third species. It is the boundary, and the reciprocity says so — it is the one place where a number equals its own reciprocal.

Step 5. State the theorem. The three classes are mutually exclusive and jointly exhaustive, the co-location and simultaneity boosts are conjugate through $c^2$ , exactly one of them is physically realisable for any non-null pair, and the null pairs are precisely the fixed points where both fail. The whole classification, both existence theorems, and their complementarity are one identity, $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} = 1$ , read four ways. Why: a rapidity gloss makes it sharper still. $\varphi = \operatorname{artanh}\beta$ is real only for $\beta < 1$ , so of the two rapidities $\operatorname{artanh}\beta_{\mathrm{col}}$ and $\operatorname{artanh}\beta_{\mathrm{sim}}$ exactly one is a real boost parameter — the other is off the group entirely. *Verified in verify_aq-sr-timelike-spacelike-null-separations.py: $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} - 1$ simplifies to zero symbolically; over 20 000 random event pairs, exactly one of the two boosts is subluminal in every non-null case, the co-location boost is legal iff $\Delta s^2 > 0$ , and the simultaneity boost is legal iff $\Delta s^2 < 0$ .*

Figure

PUNCHLINE — the two boosts are reciprocals, so exactly one of them is legal. Plotting β_sim = 1/β_col, the physical region is the unit square: a point can be in the blue band (co-location legal, timelike) or the red band (simultaneity legal, spacelike), never both. The muon of the worked example sits at β_col = 0.995 with β_sim = 1.005 — its co-location boost IS its own velocity, and its simultaneity boost is superluminal by exactly the reciprocal amount. The mirror pair (5,3) and (3,5) sit at the exactly reciprocal points 0.6 and 5/3 = 1.667

Limiting cases

Case	Result	Meaning
$\Delta s^2 > 0$	timelike	co-location frame at $v = \Delta x/\Delta t$ ; interval is pure time, $\Delta\tau = \sqrt{\Delta s^2}/c$ ; order absolute
$\Delta s^2 = 0$	null	neither privileged frame exists — both boosts would need exactly $c$ ; joined only by light
$\Delta s^2 < 0$	spacelike	simultaneity frame at $v = c^2\Delta t/\Delta x$ ; interval is pure space, $L_0 = \sqrt{-\Delta s^2}$ ; order relative
$\Delta x \to 0$	pure time	$\Delta s^2 \to c^2\Delta t^2$ ; the events are already co-located and $\beta_{\mathrm{col}} \to 0$ , so $\beta_{\mathrm{sim}} \to \infty$
$\Delta t \to 0$	pure space	$\Delta s^2 \to -\Delta x^2$ ; already simultaneous, $\beta_{\mathrm{sim}} \to 0$ , so $\beta_{\mathrm{col}} \to \infty$
$c \to \infty$	all pairs timelike	$c^2\Delta t^2$ swamps $\Delta r^2$ for any $\Delta t \neq 0$ : everything becomes causally connectable and simultaneity becomes absolute — the Newtonian world
$v \to c$	$\gamma \to \infty$	but the class never changes: no amount of boosting moves an event across the cone

Common trapProper length is the square root of MINUS the interval — and the base article had it backwards

This is the sign error that propagates furthest, and the AllQFT base for this article shipped it. In the chapter's signature $(+,-,-,-)$ , a spacelike pair has $\Delta s^2 < 0$ . Its proper length is therefore L_0 = \sqrt{-\Delta s^2}, not $\sqrt{\Delta s^2}$ , which would be imaginary. Symmetrically, a timelike pair has $\Delta s^2 > 0$ and $c\,\Delta\tau = +\sqrt{\Delta s^2}$ , not $\sqrt{-\Delta s^2}$ . In the opposite convention $(-,+,+,+)$ both statements swap, which is precisely why an article that mixes conventions produces imaginary clocks and imaginary rulers without any single line looking obviously wrong. The rule is mechanical: the real root is taken of whichever sign the class makes positive, and the class is what tells you which. State your signature, then take the root.

Common trapThe classification is about EVENTS, not about objects

"The muon is timelike" is not a sentence. Timelike, null and spacelike are properties of a pair of events — two points in spacetime — never of a particle, a rod or an observer. What is true is that a massive particle's worldline is a curve every one of whose infinitesimal segments is timelike, and a light ray's is a curve whose segments are null; that is a statement about the events along the curve. The distinction matters because a single object can furnish event pairs of every class: the muon's birth and its detection are timelike separated, but its birth and a supernova going off simultaneously in Andromeda are spacelike separated, and its birth and the arrival of a photon from that supernova may be null. Classify pairs of events, never things.

Worked examples

1 · The cosmic-ray muon: classify the pair, then find its private frame

In the Frisch–Smith experiment muons were counted at the Mount Washington summit (1917 m above sea level) and 1907 m below it in Cambridge, travelling at about $0.995c$ . Classify the two events "muon passes the summit counter" and "muon passes the sea-level counter". Then find the frame in which they happen at the same place, and what a clock there would read.

Step 1 — identify. The flight path is the detector-to-detector drop, $\Delta x = \mathbf{1907\ m}$ — not the $1917\ \mathrm{m}$ summit elevation, which is a different quantity measured from a different origin. The coordinate flight time is $\Delta t = \Delta x/v = 1907/(0.995 \times 2.998\times10^{8}) = \mathbf{6.393\ \mu s}$ , matching the canonical $\approx 6.4\ \mu\mathrm{s}$ .

Step 2 — apply. Compare the two lengths. $c\,\Delta t = 2.998\times10^{8} \times 6.393\times10^{-6} = 1916.6\ \mathrm{m}$ , against $\Delta x = 1907\ \mathrm{m}$ . Light has $9.6\ \mathrm{m}$ to spare, so \Delta s^2 = (1916.6)^2 - (1907)^2 = +3.664\times10^{4}\ \mathrm{m}^2 > 0. Timelike — as it must be, since a muon (which has mass) made the trip.

Step 3 — the boundary number. By the first existence theorem the co-location frame exists and moves at v_{\mathrm{col}} = \frac{\Delta x}{\Delta t} = \frac{1907\ \mathrm{m}}{6.393\ \mu\mathrm{s}} = 2.983\times10^{8}\ \mathrm{m/s} = \mathbf{0.995\,c}, which is, of course, the muon's own velocity. In that frame the mountain moves and the muon does not, and the two events happen at the same place. What does its clock read? Exactly the interval: \Delta\tau = \frac{\sqrt{\Delta s^2}}{c} = \frac{191.4\ \mathrm{m}}{2.998\times10^{8}\ \mathrm{m/s}} = \mathbf{0.639\ \mu s}, and as a cross-check $\Delta t/\gamma = 6.393/10.01 = 0.639\ \mu\mathrm{s}$ ✓. The muon's own clock ran for two-thirds of a microsecond while ours ran for six and a half — which is why so many of them arrive.

Step 4 — and the other boost? By the owned result it must be superluminal, and by exactly the reciprocal factor: $\beta_{\mathrm{sim}} = 1/0.995 = \mathbf{1.005}$ . There is no frame in which the muon's birth and death at the two counters are simultaneous, and there never could be. Note the reciprocity is exact: $0.995 \times 1.005025 = 1.000000$ .

2 · The mirror pair: same numbers, opposite class

Two events are separated by $\Delta t = 3\ \mathrm{s}$ and $\Delta x = 5$ light-seconds. Classify them, find the privileged frame, and then do the same for the pair with $3$ and $5$ exchanged.

Step 1. Work in light-seconds, so $c = 1$ and $c\Delta t = 3$ , $\Delta x = 5$ . Then $\Delta s^2 = 3^2 - 5^2 = \mathbf{-16\ ls^2} < 0$ : spacelike. Light needs $5$ seconds to make the crossing and only has $3$ : no signal can join these events, and neither can have caused the other.

Step 2. By the second existence theorem the simultaneity frame exists, at \beta_{\mathrm{sim}} = \frac{c\,\Delta t}{\Delta x} = \frac{3}{5} = \mathbf{0.6}, a perfectly ordinary boost ( $\gamma = 1.25$ ). In it the two events are simultaneous, separated by the proper length $L_0 = \sqrt{-\Delta s^2} = \mathbf{4}$ light-seconds — note the minus sign, without which the length is imaginary. The co-location boost would need $\beta_{\mathrm{col}} = 5/3 = 1.667$ : forbidden, and reciprocal to $0.6$ , as promised.

Step 3. Now exchange the numbers: $c\Delta t = 5$ , $\Delta x = 3$ . Now $\Delta s^2 = 5^2 - 3^2 = \mathbf{+16\ ls^2} > 0$ : timelike. The same arithmetic, one transposition, and the class has flipped. The co-location boost is $\beta_{\mathrm{col}} = 3/5 = 0.6$ — legal, and the same $0.6$ — while the simultaneity boost would need $5/3$ : forbidden.

Step 4 — the moral. One number, $0.6$ , is a co-location boost for the timelike pair and a simultaneity boost for the spacelike one, and in each case its reciprocal $1.667$ is the forbidden partner. The two existence theorems are not two facts; they are one identity, $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} = 1$ , standing on its head. (And $\beta = 0.6$ carries the chapter's exact anchor: $\varphi = \operatorname{artanh}0.6 = \ln 2$ , $\gamma = 1.25$ , $\gamma\beta = 0.75$ .)

Physics of the portrait

The causal order of events is what survives when everything else about a description is thrown away — and the muon is how we know it is real, not merely consistent.

Origins · 1 minSpacelike and how this was discovered

Spacelike

1908–present · The exterior of the cone · the frame where the two events are simultaneous

The one causal class that is different in kind. A pair of events is spacelike when $\Delta s^2 < 0$ — when $|\Delta x| > c\,|\Delta t|$ , so that not even light can bridge the gap. Three consequences follow at once, and they are all the same consequence. First, there is always a frame in which the two events happen at the same time: boost with $v = c^2\Delta t/\Delta x$ , which is a legal speed precisely because $|c\,\Delta t| < |\Delta x|$ forces $|v| < c$ . Second, a boost slightly faster or slower than that one puts either event first — spacelike order is the only time order a Lorentz transformation can reverse. Third, and therefore, the pair must be causally inert: if $A$ could influence $B$ across a spacelike gap, some legitimate observer would see the effect precede the cause. Relativity does not forbid spacelike pairs — most pairs of events in the universe are one — it forbids them from mattering. The three-way classification is Minkowski's, from the 1908 Cologne lecture, and the spacelike class is the piece of it that does the real work.

History

1908

— Minkowski, in Raum und Zeit at Cologne, partitions the events around any apex into the interior of the cone, its surface, and the exterior — the timelike, null and spacelike classes.

1911

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

1941

— Rossi and Hall publish their Colorado muon measurements (Echo Lake, 3240 m, and Denver, 1616 m), showing that faster mesotrons decay more slowly, in Phys. Rev. 59, 223.

1963

— Frisch and Smith count muons at the Mount Washington summit (1917 m above sea level) and 1907 m below it in Cambridge, measuring a dilation factor of 8.8 ± 0.8 (Am. J. Phys. 31, 342).

1964

— Zeeman proves that the causal order alone — which is to say, this classification alone — determines the Lorentz group up to a dilatation, in spacetimes of at least two spatial dimensions.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §7–8 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 1 (the interval) and ch. 6 (causal structure) · Morin, Special Relativity, ch. 4 · Griffiths, Introduction to Electrodynamics, 4th ed., §12.1 · Hartle, Gravity, ch. 4. On the muon: B. Rossi and D. B. Hall, Phys. Rev. 59, 223 (1941); D. H. Frisch and J. H. Smith, Am. J. Phys. 31, 342 (1963). The S1 treatment of the same three classes, Timelike, Spacelike and Lightlike Separations, works them out kinematically; this article recasts them as the invariant partition induced by the cone.

Provenance. The base article for this entry was written in the opposite signature, $(-,+,+,+)$ , with $\Delta s^2 = -c^2\Delta t^2 + \Delta r^2$ , timelike labelled $\Delta s^2 < 0$ , spacelike $\Delta s^2 > 0$ , proper time defined by $\Delta s^2 = -c^2\Delta\tau^2$ , and proper length by $\Delta\sigma = \sqrt{\Delta s^2}$ . Every one of those five statements is false in this chapter's locked signature and has been re-derived, not transcribed: here $\Delta s^2 = +c^2\Delta t^2 - \Delta r^2$ , timelike is $\Delta s^2 > 0$ , spacelike is $\Delta s^2 < 0$ , $c\Delta\tau = +\sqrt{\Delta s^2}$ and $L_0 = \sqrt{-\Delta s^2}$ . Anchors recomputed in verify_aq-sr-timelike-spacelike-null-separations.py: the muon chain ( $\Delta x = 1907\ \mathrm{m}$ , $\beta = 0.995$ , $\Delta t = 6.393\ \mu\mathrm{s}$ , $c\Delta t = 1916.6\ \mathrm{m}$ , $\Delta s^2 = +3.664\times10^{4}\ \mathrm{m}^2$ , $\Delta\tau = 0.639\ \mu\mathrm{s}$ , $\gamma = 10.01$ , $\beta_{\mathrm{col}} = 0.995$ , $\beta_{\mathrm{sim}} = 1.005$ ); the mirror pair $(3,5)$ and $(5,3)$ ; and the identity $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} = 1$ over 20 000 random pairs.

Builds on. The Spacetime Interval · Events and Worldlines · The Minkowski Metric · Lorentz Invariance and Invariants. Leads to. The Light Cone · Causal Structure of Spacetime · Proper Time and Proper Length.

Why can a timelike pair never be made simultaneous?

Because the boost you would need is superluminal, and the reciprocity identity says so in one line. Making the pair simultaneous requires $\beta_{\mathrm{sim}} = c\Delta t/\Delta x$ ; but timelike means $c|\Delta t| > |\Delta x|$ , so that ratio is greater than one. There is a more physical way to see it. A timelike pair can be joined by a signal, so one event can cause the other, so their order is a fact about the world rather than about coordinates — and "simultaneous" would destroy that order. The reciprocity identity $\beta_{\mathrm{col}}\beta_{\mathrm{sim}} = 1$ is the precise statement that these two facts are the same fact: the frame that would erase the order is exactly the one that is out of reach.

Is the null case really a separate class, or just a boundary?

It is a boundary, and the reciprocity makes that literal. The map $\beta_{\mathrm{col}} \mapsto \beta_{\mathrm{sim}} = 1/\beta_{\mathrm{col}}$ has exactly one fixed point, $\beta = 1$ , and that fixed point is the null class. On it, both privileged frames would require moving at exactly $c$ , and neither exists: a null pair can be neither co-located nor made simultaneous by any observer. It is also the only class whose interval is the same trivial number, zero, for everyone — the timelike and spacelike intervals carry information (a proper time, a proper length) while the null interval carries none. Nonetheless it is physically the most important of the three, because it is what light does.

The interval between two events on a light ray is zero, but they are clearly far apart. What does that mean?

It means the interval is not a distance, and treating it as one is the commonest source of confusion in the subject. $\Delta s^2$ measures the causal slack between two events: how much more time light has than it needs. For a null pair light needs precisely all the time available, so the slack is zero, and the fact that the events are a light-year apart in your frame — and something else in mine — is exactly what is not invariant. Photons take this literally: no proper time elapses along a light ray, which is why one cannot sensibly speak of "the photon's frame". The interval is zero not because nothing separates the events but because what separates them is exactly light.

Does the classification depend on the metric signature?

The labels do; the physics does not. In $(+,-,-,-)$ timelike is $\Delta s^2 > 0$ ; in $(-,+,+,+)$ timelike is $\Delta s^2 < 0$ . But which pairs of events are timelike — which can be joined by a sub-light signal — is a fact about the world, and no convention touches it. What conventions do touch is every intermediate sign in every derivation, which is why mixing them is so destructive: the proper time root, the proper length root, the four-velocity norm and the sign of every invariant all flip. The base article for this entry was written in the opposite convention, and translating it required rebuilding the algebra rather than relabelling it. State your signature and stay in it.

Can I tell the class of a pair without computing the interval?

Yes, and often faster. Compute the speed a signal would need: $v_{\text{sig}} = |\Delta r|/|\Delta t|$ . If it is less than $c$ , timelike; equal, null; greater, spacelike. This is not a different criterion — it is $\beta_{\mathrm{col}}$ under another name, and comparing it with $1$ is the same comparison as checking the sign of $\Delta s^2$ . The signal-speed reading is usually the quicker one at a blackboard and it makes the physics obvious, but the interval is what you want the moment you need a number rather than a verdict: only $\sqrt{\Delta s^2}$ gives you the proper time, and only $\sqrt{-\Delta s^2}$ gives you the proper length.

Why does the same β = 0.6 show up as a co-location boost in one example and a simultaneity boost in the other?

Because the two worked examples are the same pair of numbers with $c\Delta t$ and $\Delta x$ exchanged, and the two boosts are defined by exactly that exchange: $\beta_{\mathrm{col}} = \Delta x/(c\Delta t)$ and $\beta_{\mathrm{sim}} = c\Delta t/\Delta x$ . Swap the inputs and the two formulas trade places. For the timelike pair $(c\Delta t, \Delta x) = (5,3)$ the co-location boost is $3/5$ ; for the spacelike pair $(3,5)$ the simultaneity boost is $3/5$ . In both cases the other boost is $5/3$ , superluminal and forbidden. It is the cleanest possible illustration that the two existence theorems are one theorem seen from two sides.