Causal Structure of Spacetime

In September 2011 the OPERA collaboration announced that a beam of muon neutrinos had crossed the $730\ \mathrm{km}$ from CERN to the Gran Sasso laboratory about $60$ nanoseconds sooner than light would have. The excess was a fractional $2.5\times10^{-5}$ — a rounding error by most standards. The reaction was not "how interesting"; it was alarm. Within days theorists were pointing out that if the result stood, one could in principle arrange for a message to arrive before it was sent. The result did not stand: by March 2012 the culprit was a loose fibre-optic connector feeding the master clock. But the panic is the point. A discrepancy of one part in forty thousand in a speed threatened the logical coherence of physics, and everybody knew immediately why. The reason is that relativity does not merely say light is fast. It says the light cone at each event is an absolute structure — a partition of all other events into those you can influence, those that can influence you, and those with which you can have no dealings at all — and that every inertial observer draws the same partition. Nothing carrying information may cross it. The moment something does, the sign of $\Delta t$ between cause and effect stops being invariant, and "before" becomes a matter of opinion. That is the causal structure of spacetime: not a rule imposed on the geometry but the geometry itself, read as an ordering. And Alexandrov and Zeeman proved a theorem about it that is still startling: if you know only which events can influence which — no coordinates, no clocks, no light postulate, not even continuity — the Lorentz group falls out anyway, uniquely, up to a change of scale. Causality does not follow from the geometry. Causality is the geometry.

Key takeaways

The causal structure of spacetime is the partition, around every event $P$ , of all other events into three invariant classes: the causal past $J^-(P)$ , the causal future $J^+(P)$ , and the elsewhere — events causally sealed off from $P$ .
The sole criterion is the sign of the invariant interval: $\Delta s^2 = c^2\Delta t^2 - \Delta r^2 \geq 0$ means causally connectable, $\Delta s^2 < 0$ (spacelike) means not. $\Delta s^2$ is Lorentz invariant, so the classification is too.
Timelike and null order is ABSOLUTE: if $\Delta t > 0$ and $\Delta s^2 \geq 0$ , then $\Delta t' > 0$ in every inertial frame. Spacelike order is not: there is always a boost $v = c^2\Delta t/\Delta x$ that makes the two events simultaneous, and $|v| < c$ exactly when the pair is spacelike.
A superluminal signal at speed $w > c$ runs backwards in time in any frame with $v\,w > c^2$ . But a genuine closed causal loop — a reply arriving before the question — needs the strictly stronger $v > 2wc^2/(w^2 + c^2)$ , i.e. $\varphi > 2\chi$ with $\tanh\chi = c/w$ . At $w = 2c$ that means $v > 0.8c$ , not $0.6c$ .
Key idea: the causal order alone knows everything. From the relation " $P$ can influence $Q$ " — with no coordinates and no metric — one can already tell a timelike pair from a null pair, hence reconstruct every light cone; and Alexandrov (1950) and Zeeman (1964) proved that any bijection preserving that order must be a Lorentz transformation, composed with a translation and a dilatation. The causal order does not follow from the metric. In at least two spatial dimensions, it determines it.

The core insight

Relativity gives up absolute simultaneity but keeps absolute causality, and the light cone is the line it draws between them. If two events can be joined by a signal, every observer agrees which came first; if they cannot, observers disagree, and one calls them simultaneous. Both facts are one inequality read twice: the boost that reverses the order is $v = c^2\Delta t/\Delta x$ , and it is a legal boost, $|v| < c$ , precisely when $\Delta s^2 < 0$ — precisely when no signal could have run between the events in the first place. Order is negotiable only where it does not matter.

Where this lives today

Causal structure is not a philosophical gloss; it is a design constraint every modern theory is built to satisfy. In classical electrodynamics it appears as the retarded Green's function, which vanishes outside the past light cone — the field at $P$ depends only on sources in $J^-(P)$ , and the mathematically equally good advanced solution is discarded on causal grounds alone. In quantum field theory the same demand becomes microcausality: field operators at spacelike separation must commute, $[\hat\phi(x), \hat\phi(y)] = 0$ for $(x-y)^2 < 0$ , so no measurement here can affect one in the elsewhere. This is not decoration — imposing it is what forces antiparticles into existence, and it is the origin of the spin-statistics connection. And in general relativity the causal order becomes the primary object of study, because the metric is dynamical and the cones tilt: horizons, Cauchy surfaces, global hyperbolicity and cosmic censorship are all statements about $J^{\pm}$ , not about coordinates.

The physical picture

Stand at an event $P$ and ask a question with a definite answer: which other events can I still affect? Not "which are in my future" — that is frame-dependent — but which can receive a signal I emit. No signal outruns light, so: everything on or inside the future light cone of $P$ . Call it the causal future, $J^+(P)$ ; the events that could be among the causes of anything at $P$ make up the causal past, $J^-(P)$ . Everything else is elsewhere. Now boost. The grid shears, times and distances change, and the exterior events reshuffle their apparent order — but the three sets keep their membership, because membership is decided by the sign of $\Delta s^2$ , and that sign is invariant. Say it starkly, because the slogan "simultaneity is relative" is routinely over-read into "time order is relative", which is false. Two events you can influence have an absolute order; two you cannot have no absolute order, and no need of one. The asymmetry is no coincidence: the very boost that would reverse an order is available only when the pair is causally disconnected. Relativity is being exactly as careless with causality as it can afford to be.

Figure

Left: the causal structure of an event P — the causal future J⁺(P), causal past J⁻(P), and the elsewhere. Every observer draws the same three sets. Right: why spacelike order is negotiable. The events A and B are spacelike separated (cΔt = 3, Δx = 5, so Δs² = −16). Observer S calls A first; the boosted observer at v = 0.6c has a tilted line of simultaneity and calls them SIMULTANEOUS; a faster observer calls B first. No signal can run between A and B, so nobody is wrong

The core equations

Everything rests on one invariant and one inequality. The invariant is the interval, in the 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: $\Delta s^2 > 0$ timelike, $= 0$ null, $< 0$ spacelike. Physical reading: timelike is positive here. The other convention $(-,+,+,+)$ flips every sign below — no physics changes, every intermediate sign does. Causal precedence is then a statement purely about $\Delta s^2$ and the sign of $\Delta t$ .

P \prec Q \quad\Longleftrightarrow\quad \Delta s^2(P,Q) \geq 0 \ \ \text{and} \ \ t_Q > t_P, \qquad J^{\pm}(P) = \{Q : P \prec Q \ \text{or} \ Q \prec P\} \qquad (2)

where: $\prec$ — "can causally influence"; $J^{\pm}(P)$ — the causal future and past. Physical reading: this looks frame-dependent, since $t_Q > t_P$ is a coordinate statement. It is not, and that is the first derivation: once $\Delta s^2 \geq 0$ , no boost can change the sign of $\Delta t$ . Under a boost of speed $v$ along $x$ the time separation transforms as

\Delta t' = \gamma\left(\Delta t - \frac{v\,\Delta x}{c^2}\right), \qquad \gamma = \frac{1}{\sqrt{1-v^2/c^2}} \qquad (3)

v_{\mathrm{sim}} = \frac{c^2\,\Delta t}{\Delta x}, \qquad |v_{\mathrm{sim}}| < c \iff c\,|\Delta t| < |\Delta x| \iff \Delta s^2 < 0 \qquad (4)

where: $v_{\mathrm{sim}}$ — the simultaneity boost. Physical reading: this is the hinge of the whole subject. The boost that would erase the time-order exists as a physical boost exactly when the events are spacelike — exactly when there is no causal order to erase. Boost past $v_{\mathrm{sim}}$ and the order reverses. The causality bound $v_{\text{signal}} \leq c$ , for any transfer of energy or information, is therefore not a decree but a consequence: allow $w > c$ and equation (4) hands you a legal boost that puts the effect before its cause.

Why the causal order is absolute — and when it is not

Derivation ATimelike order is absolute and spacelike order is negotiable, from a single inequality

The two headline facts of causal structure look like separate claims: observers always agree on the order of causally connected events, and can always be made to disagree about the order of causally disconnected ones. They are one inequality read in two directions — the boost that would reverse a time order is a legal boost, $|v| < c$ , if and only if the pair is spacelike. Relativity is not carefully protecting causality; it simply cannot reach the events it would have to reorder.

Step 1. Take two events with separation $(c\Delta t, \Delta x)$ , $\Delta t > 0$ , and boost along $x$ . From equation (3), $\Delta t' = \gamma\,\Delta t\,\big(1 - \beta\,\Delta x/(c\,\Delta t)\big)$ with $\beta = v/c$ . Since $\gamma > 0$ , the sign of $\Delta t'$ is the sign of the bracket. Why: the entire question — can any observer see the effect precede the cause — has been reduced to whether one bracket can go negative.

Step 2. Suppose the pair is timelike or null: $\Delta s^2 \geq 0$ , i.e. $|\Delta x| \leq c\,\Delta t$ , so $|\Delta x/(c\Delta t)| \leq 1$ . Then $|\beta\,\Delta x/(c\Delta t)| \leq |\beta| < 1$ , so the bracket is strictly positive for every allowed boost, and $\Delta t' > 0$ always. Why: the two constraints — the pair is subluminally connectable, and the observer is subluminal — multiply to something strictly less than one. The order cannot flip. This is the absoluteness of causal order, and it is a two-line proof.

Step 3. Now suppose the pair is spacelike: $\Delta s^2 < 0$ , i.e. $|\Delta x| > c\,\Delta t$ . Solve $\Delta t' = 0$ : the bracket vanishes at $\beta_{\mathrm{sim}} = c\,\Delta t/\Delta x$ , that is $v_{\mathrm{sim}} = c^2\Delta t/\Delta x$ . Its magnitude is $|v_{\mathrm{sim}}|/c = c|\Delta t|/|\Delta x| < 1$ — precisely because the pair is spacelike. Why: the simultaneity boost exists as a physical boost iff the separation is spacelike. The condition $|v_{\mathrm{sim}}| < c$ and the condition $\Delta s^2 < 0$ are literally the same inequality, $c|\Delta t| < |\Delta x|$ , divided through differently.

Step 4. Push past it. For $|\beta| > |\beta_{\mathrm{sim}}|$ (same sign) the bracket goes negative and $\Delta t' < 0$ : the order has reversed. For a spacelike pair all three regimes — $B$ after $A$ , simultaneous, $B$ before $A$ — are realised by legal observers, so the pair's "time order" is not a property of the pair at all. Why: and it costs nothing, because Step 2 already guarantees no signal joins them. Relativity relativises exactly the orderings that carry no causal content, and not one more. *Verified in verify_aq-sr-causal-structure.py: over thousands of random causal pairs and boosts, $\Delta t' > 0$ without exception; over random spacelike pairs, $v_{\mathrm{sim}} = c^2\Delta t/\Delta x$ is always subluminal, always gives $\Delta t' = 0$ , and boosting beyond it always gives $\Delta t' < 0$ ; and $|v_{\mathrm{sim}}| < c \iff \Delta s^2 < 0$ .*

The next card is where students are usually told a half-truth: that a faster-than-light signal "immediately" lets you phone your own past because $vw > c^2$ . That inequality buys a frame in which one leg runs backwards — already fatal for any Lorentz-invariant theory, but not yet a closed loop. Closing the loop takes a strictly stronger condition, and the difference is worth the algebra.

Go deeper · Derivation B · 3 minThe antitelephone: what a superluminal signal actually costs

Derivation BThe antitelephone: what a superluminal signal actually costs

Suppose some influence propagates at $w > c$ in the frame of whoever emits it. Two different disasters are available, and they are routinely conflated. The first: a single such signal, sent forward in time by Alice, runs backwards in time as seen by some legal observer. The second, far worse: Alice sends a question, Bob replies, and the reply reaches Alice before she asked. The goal is to derive both thresholds exactly and show they differ — because the slogan " $vw > c^2$ gives you an antitelephone" is, as a statement about the closed loop, false.

Step 1. The single leg. Alice, at rest in $S$ , sends a signal at speed $w$ from the origin; it arrives at $(\Delta t, \Delta x) = (\Delta t, w\Delta t)$ with $\Delta t > 0$ . In a frame moving at $v$ , equation (3) gives $\Delta t' = \gamma\,\Delta t\,(1 - vw/c^2)$ . This is negative iff $v\,w > c^2$ . Since $w > c$ , the required $v = c^2/w$ is less than $c$ , so such an observer always exists. Why: any superluminal signal, of any speed, is seen running backwards in time by somebody. That is the first and most famous half of the argument, and it is exactly what $vw > c^2$ buys — no more.

Step 2. Now close the loop. Let Bob be at rest in $S'$ (speed $v$ ), co-located with Alice at $t = 0$ , so his worldline is $x = vt$ . Alice transmits at $t = t_A$ at speed $w$ in her frame; the signal reaches Bob at $t_B = w\,t_A/(w - v)$ , $x_B = v\,t_B$ . Bob replies at speed $w$ in his own frame — the only fair rule, since a tachyon's speed is a property of the emitter's physics, not of Alice. Why: that symmetry is what makes the paradox a paradox. If Bob were obliged to use Alice's frame, the theory would already have a preferred frame and relativity would be dead for a duller reason.

Step 3. Transform the reply. In $S'$ it moves at $-w$ ; velocity addition gives its velocity in Alice's frame, $u = (v - w)/(1 - vw/c^2)$ . Follow it back to $x = 0$ : the arrival time is $t_C = t_B(1 - v/u)$ , and grinding through gives \frac{t_C}{t_A} = \frac{w^2\left(1 - v^2/c^2\right)}{(w - v)^2}. Why: every ingredient is elementary, but the outcome is not obvious by inspection — which is exactly why the slogan gets it wrong.

Step 4. Read off the threshold. The reply beats the question ( $t_C < t_A$ ) iff $w^2(1 - v^2/c^2) < (w-v)^2$ , which rearranges to v > \frac{2wc^2}{w^2 + c^2}. Write $\tanh\chi = c/w$ and $\tanh\varphi = v/c$ : the right-hand side is exactly $\tanh(2\chi)$ , so the condition is the rapidity inequality $\boldsymbol{\varphi > 2\chi}$ . Why: the threshold is not " $vw > c^2$ " ( $\varphi > \chi$ ) but twice that rapidity. A closed loop needs two legs, and each leg must be dragged past the reversal point — the factor of two is the second leg.

Step 5. Put in the numbers and watch the folklore fail. Take $w = 2c$ : the threshold is $v = 2(2)/(4+1) = 0.8c$ . At $v = 0.6c$ we have $vw = 1.2c^2 > c^2$ , so Step 1 fires — Bob does receive the message before Alice sent it, in Bob's frame. Yet the loop does not close: $t_C/t_A = 4(1-0.36)/(1.4)^2 = 64/49 = 1.306$ , so the reply lands 30.6 % later than the question. Push to $v = 0.9c$ and $t_C/t_A = 76/121 = 0.628$ : if Alice transmits one year after passing Bob, the reply reaches her at $0.628$ yr — $0.372$ yr, about $136$ days, before she asks. Why: this is the honest antitelephone, and it is still fatal — Alice can be told the answer and choose not to ask. No values of $v$ and $w$ rescue that, which is why $w \leq c$ is not an aesthetic preference. *Verified in verify_aq-sr-causal-structure.py: the single-leg reversal has its zero at $v = c^2/w$ ; the loop ratio is $w^2(1-v^2)/(w-v)^2$ exactly; its threshold solves to $2w/(w^2+1) = \tanh(2\operatorname{artanh}(1/w))$ ; and the three anchor cases evaluate to $64/49$ , $1$ and $76/121$ exactly.*

Figure

PUNCHLINE — the antitelephone threshold, with the folklore case shown FAILING. A reply beats the question only when v exceeds v = 2wc²/(w² + c²) — the curve. For w = 2c the threshold is v = 0.8c. At v = 0.6c (the case vw = 1.2c² ＞ c², usually quoted as sufficient) the reply arrives at t_C = 64/49 = 1.306 t_A — 30.6 % LATE. At v = 0.8c it lands exactly on t_A. Only at v = 0.9c does it close the loop: t_C = 76/121 = 0.628 t_A, i.e. 0.372 yr ≈ 136 days BEFORE Alice asked

What the causal order alone knows

The last derivation is what makes this a chapter rather than a footnote to the light cone. Strip away coordinates, clocks, the metric, even continuity, and keep only the bare relation $\prec$ . It turns out to be enough.

Go deeper · Derivation C · 2 minCausal order alone tells a timelike pair from a null pair

Derivation CCausal order alone tells a timelike pair from a null pair

Here is the question worth asking. Two events $P \prec Q$ : is the separation timelike or null? That looks like a metric question — you seem to need $\Delta s^2$ . You do not. The answer is encoded in the order relation itself, in the structure of what lies between them, and that is the key that unlocks Alexandrov–Zeeman. The tool is the causal diamond $D(P,Q) = J^+(P) \cap J^-(Q)$ : the events $P$ can influence that can in turn influence $Q$ .

Step 1. Recall the reversed triangle inequality, the one genuinely un-Euclidean fact about Minkowski space: for future-directed causal vectors, $\tau(P,Q) \geq \tau(P,R) + \tau(R,Q)$ whenever $P \prec R \prec Q$ , where $\tau = \sqrt{\Delta s^2}/c$ is the proper time. Going via a waypoint takes less proper time, never more. Why: this is the twin paradox as a theorem, and it is about to do work that has nothing to do with twins.

Step 2. Take $P$ and $Q$ null-separated, so $\tau(P,Q) = 0$ . If $R$ lies in the diamond, then $\tau(P,R) \geq 0$ and $\tau(R,Q) \geq 0$ , and Step 1 forces $0 \geq \tau(P,R) + \tau(R,Q)$ . Both terms must therefore vanish: $R$ is null-separated from both $P$ and $Q$ . Two future-directed null vectors whose sum is null must be parallel, so $R$ lies on the null segment from $P$ to $Q$ . Why: the diamond of a null pair has collapsed. It is not a four-dimensional region at all — it is a line.

Step 3. Notice what that means for the order. Every pair of points on a null segment is causally related (they lie on one light ray, one after the other), so the null diamond is a totally ordered chain: pick any two of its events and one precedes the other. Why: a purely order-theoretic property, stated with no metric anywhere in sight.

Step 4. Now take $P$ and $Q$ timelike-separated. The diamond is now a genuine four-dimensional region with non-empty interior. Concretely, with $P = (0,\mathbf{0})$ and $Q = (2,\mathbf{0})$ in units $c=1$ , the events $R_a = (1, 0.4, 0)$ and $R_b = (1, -0.4, 0)$ both lie in the diamond, but their separation is $\Delta s^2 = 0 - 0.8^2 < 0$ : they are spacelike, hence causally unrelated. Why: a timelike diamond always contains causally unrelated pairs; a null diamond never does. That is a difference visible to the order relation alone.

Step 5. Harvest the theorem's easy half. Let $f$ be a bijection with $P \prec Q \iff f(P) \prec f(Q)$ — a causal automorphism. By Steps 2–4 it preserves the property "the diamond between these two is totally ordered", which holds exactly for null pairs. Hence $f$ maps null pairs to null pairs: every causal automorphism carries light cones onto light cones. Why: no metric, no coordinates, no continuity assumed — the cones were reconstructed from the order alone. The remaining, hard half of Alexandrov–Zeeman is that a cone-preserving bijection of Minkowski space with at least two spatial dimensions must already be linear, hence a Lorentz transformation composed with a translation and a dilatation. *Verified in verify_aq-sr-causal-structure.py: the reversed triangle inequality holds over thousands of sampled waypoints; the null diamond solves exactly to the null segment and is totally ordered; the timelike diamond demonstrably contains the spacelike (hence causally unrelated) pair $R_a$ , $R_b$ .*

Limiting cases

Case	Result	Meaning
$\Delta s^2 > 0$ (timelike)	order absolute	$\mathrm{sign}(\Delta t)$ is invariant; a massive signal can join them; $\tau = \sqrt{\Delta s^2}/c$ is what a clock reads between them
$\Delta s^2 < 0$ (spacelike)	order relative	$v_{\mathrm{sim}} = c^2\Delta t/\Delta x$ is a legal boost; some observer says simultaneous, some say the order is reversed
signal speed $w \to c^+$	threshold $v^{*} \to c$	the antitelephone threshold $2wc^2/(w^2+c^2) \to c$ : the loop becomes unreachable, which is exactly why $w \leq c$ is safe
$w \to \infty$	$v^{*} \to 0$	an instantaneous signal opens a causal loop to an arbitrarily slow observer — the Newtonian rigid rod is a causality bomb with a long fuse
$c \to \infty$	elsewhere vanishes	the cone flattens onto $t = 0$ , every future event becomes reachable, and causal structure degenerates to Newtonian universal time

Common trapThe condition v·w greater than c² is NOT the antitelephone condition

Folklore says: a superluminal signal plus $vw > c^2$ gives you a phone to your own past. That conflates two different thresholds. $vw > c^2$ is the condition for one leg of the exchange to run backwards in the boosted frame — real, and already fatal to any Lorentz-invariant theory with faster-than-light influences. But the closed loop, in which Alice's reply arrives before Alice's question, needs the strictly stronger $v > 2wc^2/(w^2+c^2)$ — in rapidities $\varphi > 2\chi$ with $\tanh\chi = c/w$ (Benford, Book & Newcomb 1970). With $w = 2c$ that is $v > 0.8c$ ; at the frequently quoted $v = 0.6c$ the reply in fact arrives $30.6\%$ late. The conclusion — superluminal signalling destroys causality — is untouched. Only the arithmetic usually printed beside it is wrong.

Worked examples

1 · Did the solar flare cause the aurora?

A flare erupts on the Sun at $t = 0$ . An auroral brightening is recorded at the Earth's magnetic pole $480$ seconds later. Can the flare be its cause? And what about the coronal mass ejection that arrives days afterwards, at a typical $1000\ \mathrm{km/s}$ ?

Step 1 — identify. One Astronomical Unit is $1.495\,978\,707\times10^{11}\ \mathrm{m}$ (exact, IAU), so the Sun–Earth light time is $\mathbf{499.0\ s} = 8.32$ minutes. That is the whole boundary: any event at the Earth less than $499$ s after the flare lies outside the flare's future cone.

Step 2 — apply. For the aurora at $\Delta t = 480\ \mathrm{s}$ : \Delta s^2 = c^2\Delta t^2 - \Delta x^2 = (1.439\times10^{11})^2 - (1.496\times10^{11})^2 = -1.672\times10^{21}\ \mathrm{m}^2 < 0. Spacelike. The flare cannot have caused it — not "probably did not", cannot, on pain of a signal outrunning light. The proper distance between the two events is $\sqrt{-\Delta s^2} = 4.09\times10^{10}\ \mathrm{m} = 0.27\ \mathrm{AU}$ .

Step 3 — the boundary number, and the reversal. Because the pair is spacelike, equation (4) hands us a legal simultaneity boost: $v_{\mathrm{sim}}/c = c\Delta t/\Delta x = 480/499.0 = \mathbf{0.9619}$ . An observer crossing the solar system at $0.962c$ ( $\gamma = 3.66$ ) calls flare and aurora simultaneous; a slightly faster one says the aurora came first. Nobody is confused, because nothing travelled between them.

Step 4 — now do it properly. The coronal mass ejection travels at $1000\ \mathrm{km/s} = 3.34\times10^{-3}\,c$ and covers $1\ \mathrm{AU}$ in $\Delta t = 1.496\times10^{5}\ \mathrm{s} = 41.6\ \mathrm{h} = \mathbf{1.73\ days}$ , giving $\Delta s^2 = +2.01\times10^{27}\ \mathrm{m}^2 > 0$ : timelike, comfortably, and this is what causes the storm. (Its proper time is $149\,597.0\ \mathrm{s}$ against a coordinate time of $149\,597.9\ \mathrm{s}$ — the plasma's clock loses all of $0.83$ seconds. Time dilation is real; at $\beta = 0.0033$ it is also beside the point.)

2 · If OPERA had been right, who would have seen the neutrinos arrive before they left?

In 2011 the OPERA experiment reported neutrinos covering the $730\ \mathrm{km}$ from CERN to Gran Sasso $60\ \mathrm{ns}$ faster than light. Take the claim at face value: how fast would an observer have to move for the arrival to precede the emission?

Step 1. Light's crossing time is $\Delta x/c = 7.30\times10^{5}/2.998\times10^{8} = 2.435\ \mathrm{ms}$ . The claimed neutrino time was $60\ \mathrm{ns}$ less, a fractional excess of $60\times10^{-9}/2.435\times10^{-3} = \mathbf{2.5\times10^{-5}}$ — a superluminal $w = 1.000025\,c$ .

Step 2. The two events (emission at CERN, detection at Gran Sasso) then have $c\,\Delta t = \Delta x(1 - 2.5\times10^{-5})$ , so $\Delta s^2 < 0$ — spacelike, by a hair. By equation (4), the order is negotiable.

Step 3. The reversing boost is $v_{\mathrm{sim}}/c = c\Delta t/\Delta x = 1 - 2.5\times10^{-5} = \mathbf{0.999975}$ , corresponding to $\gamma = 143$ . Any observer moving along the beamline faster than that — an utterly ordinary speed for a cosmic-ray proton — would have recorded the neutrinos arriving at Gran Sasso before they were made at CERN.

Step 4 — the moral. That is what the fuss was about. A $60\ \mathrm{ns}$ discrepancy is not a small anomaly in a speed; it is the difference between a spacetime with an absolute causal order and one without. In March 2012 OPERA traced the effect to a loose fibre-optic connector in the link feeding its master clock, which had been shortening the measured flight time by about $73\ \mathrm{ns}$ . Reseated, the neutrinos travelled at $c$ .

Physics of the portrait

Causal structure has a strange biography: sketched by Minkowski, ignored for forty years, then found to contain everything.

Origins · 2 minCausal structure and how this was discovered

Causal structure

1908–1964 · Minkowski's cone · Alexandrov's chronogeometry · Zeeman's theorem

Minkowski drew the fore-cone and after-cone in Cologne in 1908, and the partition of events around an apex was there from the start. For decades it was treated as a picture — a way of depicting results derived by other means. That changed with A. D. Alexandrov, who in 1950 asked the question backwards: given nothing but the relation " $P$ can causally influence $Q$ ", with no coordinates, no clocks and no metric, how much of relativity can you rebuild? His answer, developed over two decades in a programme he called chronogeometry, was: all of it. E. C. Zeeman rediscovered the result and gave it its famous name in 1964, in a four-page paper called "Causality Implies the Lorentz Group" (J. Math. Phys. 5, 490). The theorem needs at least two spatial dimensions — in $1+1$ the causal order is far too permissive — and it delivers the Lorentz group only up to a dilatation, since rescaling all coordinates leaves every cone where it was. Those caveats aside, the content has not dulled with age: the light postulate, the linearity of the transformations, the constancy of $c$ — none of it need be assumed. Order the events by what can affect what, and Lorentz is the only symmetry left standing.

History

1908

— Minkowski, in Raum und Zeit at Cologne, draws the fore-cone and after-cone at each event and names the region outside them, which has no invariant time relation to the apex.

1950

s — A. D. Alexandrov, in part with V. V. Ovchinnikova, shows that a causal-order-preserving map of Minkowski space must be a Lorentz transformation up to a translation and a scale, founding chronogeometry.

1964

— E. C. Zeeman independently proves and publicises the result in "Causality Implies the Lorentz Group", J. Math. Phys. 5, 490 — the theorem now carries both names.

1970

— Benford, Book and Newcomb publish "The Tachyonic Antitelephone" (Phys. Rev. D 2, 263), giving the closed-loop condition for a faster-than-light signal to deliver a reply before its question.

2011

— OPERA reports muon neutrinos crossing the 730 km from CERN to Gran Sasso about 60 ns faster than light, a fractional excess of 2.5 × 10⁻⁵.

2012

— OPERA traces the anomaly to a loose fibre-optic connector in the timing chain, which had shortened the measured flight time by roughly 73 ns; a corrected run finds the neutrinos travelling at c.

2017

— GW170817: gravitational waves and gamma rays from one neutron-star merger arrive 1.7 s apart after some 130 million years, confirming that gravity and light share one cone to about a part in 10¹⁵.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §7–8 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 5–6 · Wald, General Relativity, ch. 8 · Hartle, Gravity, ch. 4. Primary sources: E. C. Zeeman, "Causality Implies the Lorentz Group", J. Math. Phys. 5, 490 (1964); A. D. Alexandrov, "On Lorentz transformations", Uspekhi Mat. Nauk 5(3), 187 (1950); G. A. Benford, D. L. Book and W. A. Newcomb, "The Tachyonic Antitelephone", Phys. Rev. D 2, 263 (1970).

Provenance. The base was already in the chapter's signature; every step has nonetheless been re-derived. One correction was needed and is carried in-article as teaching material. The base — and much standard folklore — gives the antitelephone condition as $v\,w > c^2$ . That is the condition for a single leg to run backwards in the boosted frame; the closed loop requires the strictly stronger $v > 2wc^2/(w^2 + c^2)$ , i.e. $\varphi > 2\chi$ with $\tanh\chi = c/w$ (Benford, Book & Newcomb 1970). Recomputed in verify_aq-sr-causal-structure.py: at $w = 2c$ the threshold is $v^{*} = 0.8c$ ; at $v = 0.6c$ the reply arrives at $t_C/t_A = 64/49 = 1.306$ (late, not early); at $v = 0.9c$ , at $76/121 = 0.628$ . Also verified: the Sun–Earth light time $499.0$ s; the $480$ -s aurora pair's $\Delta s^2 = -1.672\times10^{21}\ \mathrm{m}^2$ and $v_{\mathrm{sim}} = 0.9619c$ ; the CME's $1.73$ -day timelike crossing; OPERA's $2.5\times10^{-5}$ excess and $\gamma = 143$ reversing boost.

Builds on. The Spacetime Interval · The Light Cone · Timelike, Spacelike and Null Separations · Lorentz Invariance and Invariants · Spacetime Diagrams. Leads to. Proper Time and Proper Length · Four-Vectors · The Minkowski Metric.

If simultaneity is relative, why is causal order not?

Because the two apply to different pairs of events. The boost that reorders a pair is $v = c^2\Delta t/\Delta x$ , and it is legal — $|v| < c$ — only when $c|\Delta t| < |\Delta x|$ , which is exactly the condition for the pair to be spacelike, hence causally unconnectable. For a timelike or null pair the reordering boost would need $|v| \geq c$ , and no observer can have it. Relativity relativises the order of precisely those pairs between which no signal can pass, and leaves untouched every pair that could stand in a cause–effect relation. The same inequality, read in opposite directions.

Does quantum entanglement violate the causal structure?

No, and the reason is a theorem rather than a hope. Measuring one member of an entangled pair does produce instantaneous correlations with a spacelike-separated partner. But the marginal distribution at each wing — everything an experimenter there can observe without comparing notes — is completely unaffected by what was done at the other. Nothing that could carry a message crosses the cone; the correlations become visible only once the records are brought together by an ordinary subluminal channel. Quantum field theory enforces this structurally: spacelike-separated operators commute, so no local observable at $x$ depends on an operation at $y$ when $(x-y)^2 < 0$ . Entanglement is correlation without signalling.

What about things that go faster than light — phase velocity, spot beams, expanding universe?

All real, none causal. A wave's phase velocity can exceed $c$ in a dispersive medium, but a phase carries no information; the signal front never outruns $c$ . A laser spot swept across the Moon can cross its surface faster than $c$ , but nothing travels along the surface — each photon flew independently from Earth, and the spot is a pattern, not an object. Distant galaxies recede faster than $c$ , but that is the growth of space between them, not motion through it. The invariant test is always the same: is there a timelike or null curve from one event to the other? If not, no influence passed, whatever appears to be moving.

Why is the antitelephone condition not just vw greater than c²?

Because a phone call has two legs. The inequality $vw > c^2$ guarantees that one superluminal transmission runs backwards in time in some legal frame — Bob receives Alice's question, in Bob's own frame, before Alice sent it. That is already a catastrophe for a Lorentz-invariant theory, and it is what the inequality is really telling you. But for the loop to close — for the reply to arrive back at Alice before she asked — Bob's reply must also be dragged past its own reversal threshold. Working it through gives $t_C/t_A = w^2(1 - v^2/c^2)/(w-v)^2$ , and the loop closes only when $v > 2wc^2/(w^2+c^2)$ , which in rapidity is $\varphi > 2\chi$ : exactly twice the single-leg condition. With $w = 2c$ , folklore's $v = 0.6c$ leaves the reply arriving $30.6\%$ late; you need $v > 0.8c$ .

Is the causal order really enough to determine the Lorentz group? That sounds too strong.

It is that strong, with two honest caveats. The theorem (Alexandrov 1950; Zeeman 1964) says any bijection of Minkowski space preserving the causal order in both directions is a Lorentz transformation composed with a translation and a dilatation. Caveat one is the dilatation: scaling every coordinate by a common factor leaves every light cone where it was, so the order simply cannot see the overall scale. Caveat two is dimension: the theorem needs at least two spatial dimensions, because in $1+1$ any increasing map of each null coordinate separately preserves the order and the automorphism group is infinite-dimensional. Within those caveats nothing else is assumed — not linearity, not continuity, not the constancy of $c$ — and the Lorentz group comes out anyway.

Could a theory simply pick a preferred frame and let superluminal signals be harmless?

It could, and some have tried; the cost is that relativity becomes false as a symmetry of nature rather than merely inconvenient. With a preferred frame, "before" means "before in that frame", the causal order is fixed once and for all, and a tachyon running backwards in some other frame is an artefact of bad coordinates — no paradox. But the preferred frame must be undetectable by every experiment ever performed, and respected exactly by every interaction including ones not yet discovered. The alternative — no preferred frame, no superluminal influence, causal structure absolute — costs nothing and has survived every test from Michelson–Morley to the $10^{-15}$ agreement of gravitational waves with light in GW170817.