Causality and the Light Cone Constraint

Clap your hands. The sound arrives across the room as a clap — a sharp event with a beginning and an end. Flash a torch: the light arrives as a flash. This is so ordinary that it takes effort to notice it is a strong claim about the universe. The disturbance you made is over, at the receiver, at a definite moment. It did not have to be that way. In a world with two spatial dimensions it would not be: a Flatland torch would produce not a flash but an afterglow, dying away like $1/t$ , forever — and the total signal it delivered would be logarithmically infinite. So "signals stay inside the light cone" is only half of causality, and it is the half the geometry hands you free. The other half is a question about the dynamics: given that the cone is there, do the field equations actually respect it — and does the signal they deliver arrive sharp, or smeared through the cone's interior? That question is a theorem about where a Green's function is allowed to be non-zero, and the answer turns on how many spatial dimensions you have. This article proves that the light cone of the metric is exactly the characteristic cone of the field equations — so causality is not bolted onto the dynamics, it is made of them — and then proves that the sharpness of a signal is an accident of living in three.

Key takeaways

Convention lock. $(+,-,-,-)$ , $\eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ , $\Delta s^2 = c^2\Delta t^2 - \Delta x^2 - \Delta y^2 - \Delta z^2$ . Timelike is POSITIVE, spacelike negative, $u\cdot u = +c^2$ , $p\cdot p = +m^2c^2$ .
The kinematic half of causality is settled elsewhere and is not re-derived here: order reversal iff spacelike (`aq-sr-timelike-spacelike-lightlike-separations`), and the invariant $J^\pm$ /elsewhere partition plus the antitelephone (`aq-sr-causal-structure`).
The dynamical half is this article, and it is not automatic. A metric supplies a cone; it does not promise the field equations stay inside it. The heat equation shares the metric and ignores the cone completely.
The light cone IS the characteristic cone of $\Box$ . A surface across which a solution of $\Box\psi = 0$ may be non-smooth must have a null normal: $\eta^{\mu\nu}\partial_\mu\varphi\,\partial_\nu\varphi = 0$ . Only the principal symbol decides it — so masses and currents cannot shift the cone.
Causality is a support theorem. In $3+1$ , $G_{\mathrm{ret}} = \delta(t-r/c)/(4\pi r)$ : supported exactly ON the cone, identically zero inside it. A flash is a flash. In $2+1$ it is not: descent gives $c\,\Theta(ct-\rho)/\big(2\pi\sqrt{c^2t^2-\rho^2}\big)$ — non-zero throughout the interior, decaying only as $1/t$ , with a logarithmically divergent time integral.
Key idea: "nothing outside the cone" and "nothing after the front" are different statements. The first is causality and holds in every dimension. The second is sharpness — Huygens' principle in the strict sense — and it holds only in an ODD number of spatial dimensions, for massless fields. That we see stars as points and hear speech as speech is not a fact about light or sound. It is a fact about the number three.

The core insight

Relativity gives you an invariant cone at every event. That is geometry: necessary for causality, but not sufficient, because a cone constrains nothing until some equation is obliged to respect it. The obligation comes from the field equations. The wave operator has characteristics — the surfaces along which a discontinuity is permitted to live and travel — and those are precisely the null cones of the metric, because both are the zero set of $\eta^{\mu\nu}k_\mu k_\nu$ . Nothing was assumed; the cone of the geometry and the cone of the dynamics are the same object. Causality is then a statement about the support of a Green's function — and once you ask where $G_{\mathrm{ret}}$ may be non-zero, a sharper question appears that geometry can never answer: on the cone only, or throughout it? The answer is the number of spatial dimensions.

Where this lives today

In quantum field theory the same demand appears as microcausality: field operators at spacelike separation must commute, $[\hat\phi(x), \hat\phi(y)] = 0$ whenever $(x-y)\cdot(x-y) < 0$ . That is the operator version of " $G_{\mathrm{ret}}$ has no support outside the cone", and imposing it is what forces antiparticles into existence and is the origin of the spin-statistics connection. `aq-sr-reinterpretation-principle-causal-paradoxes` takes up that thread. In general relativity the characteristic cone becomes the primary object, because the metric is dynamical and the cones tilt: horizons, Cauchy surfaces and cosmic censorship are all statements about where the characteristics of the Einstein equations run. And in computational physics the Courant condition is nothing but the demand that the numerical domain of dependence contain the physical light cone. The dimensional punchline is not idle either. Waves on a membrane or in shallow water genuinely carry the tail; and even in $3+1$ , giving a field a mass destroys strict Huygens. Causality survives untouched; sharpness does not. Only massless fields in three spatial dimensions get to send a clean pulse — and that is the narrow doorway through which every signal you have ever received has come.

The physical picture

Take the settled results as given, because this article's strength depends on not pretending to prove them twice. `aq-sr-timelike-spacelike-lightlike-separations` shows that a boost reverses time order iff the events are spacelike separated. `aq-sr-causal-structure` shows the $J^\pm$ /elsewhere partition is observer-independent, and settles the antitelephone — including that a closed causal loop needs the strictly stronger $v > 2wc^2/(w^2+c^2)$ , not the folklore $vw > c^2$ . `aq-sr-the-light-cone` owns the cone. What none of it says is why a field obeys it. Minkowski geometry hands you a beautiful invariant cone — and so far it is decoration, lines on a diagram. Hand yourself a different field equation, say the heat equation $\partial_t T = D\nabla^2 T$ . It is well posed, it describes real physics, and its solutions are instantaneously non-zero everywhere: heat a point of an infinite rod at $t=0$ and the temperature a light-year away is non-zero at $t = 10^{-30}\ \mathrm{s}$ — exponentially tiny, but not zero. The heat equation never heard of your cone. So the light-cone constraint is a contentful demand on the dynamics, and a randomly chosen PDE fails it. The equations of relativistic physics pass — and not "because we imposed it". They are hyperbolic; a hyperbolic operator has characteristic surfaces; and for the operators of relativistic physics those surfaces coincide with the null cones of $\eta_{\mu\nu}$ , identically, because both are the zero set of the same quadratic form. Then a second question becomes visible, and it is the one this article is really about. Knowing a signal cannot escape the cone tells you nothing about where inside it the signal goes. Two very different worlds are compatible with strict causality: a sharp one, in which a pulse arrives at radius $r$ at exactly $t = r/c$ and is back to zero an instant later — supported on the cone, a shell of zero thickness; and a smeared one, in which the pulse arrives at $t = r/c$ (never earlier — causality is safe) and then keeps arriving, with a wake that comes not from any wall but from the propagator itself. Causality cannot tell these apart. Both are perfectly causal. The dimension of space tells them apart.

Interactive Simulation

Spacelike vs Timelike Ordering — a boost can flip the time-order of spacelike-separated events (simultaneity is relative) but never timelike ones (causality is preserved).

The core equations

The invariant, in the locked signature — re-derived from $\eta_{\mu\nu}$ here, not transcribed:

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

where: $\Delta s^2 > 0$ is timelike, $=0$ null, $<0$ spacelike. Physical reading: timelike is positive here, and $u\cdot u = \gamma^2(c^2 - v^2) = +c^2$ , a positive constant, so a massive worldline never leaves its local cone. The wave operator, and the object that decides everything:

\Box \equiv \eta^{\mu\nu}\partial_\mu\partial_\nu = \frac{1}{c^2}\frac{\partial^2}{\partial t^2} - \nabla^2, \qquad \text{principal symbol}\quad \sigma(k) = \eta^{\mu\nu}k_\mu k_\nu \qquad (2)

where: $\sigma(k)$ is what you get by replacing $\partial_\mu \to k_\mu$ in the highest-derivative terms. Physical reading: the principal symbol is the whole story of propagation. Lower-order terms — masses, currents, couplings — change what propagates; they cannot change where it may go. A surface $\varphi = \mathrm{const}$ is characteristic when its normal $k_\mu = \partial_\mu\varphi$ annihilates it:

\boxed{\;\eta^{\mu\nu}\,\partial_\mu\varphi\,\partial_\nu\varphi = 0 \;\Longleftrightarrow\; \frac{1}{c^2}\left(\frac{\partial\varphi}{\partial t}\right)^{\!2} = |\nabla\varphi|^2\;} \qquad (3)

where: $\varphi = 0$ is the wavefront. Physical reading: the normal to a wavefront must be a null vector — which is the equation of a light cone and nothing else. Equation (3) is the light-cone constraint, written in the language of PDEs rather than geometry. It was not imposed; it is what $\Box$ is. The retarded Green's function in three spatial dimensions, $\Box\,G_{\mathrm{ret}} = \delta^4(x-x')$ :

\boxed{\;G^{(3+1)}_{\mathrm{ret}}(t, r) \;=\; \frac{\delta\!\left(t - r/c\right)}{4\pi r}\;} \qquad (4)

where: $r = |\vec x - \vec x\,'|$ . Physical reading: read the support, not the formula. $G_{\mathrm{ret}}$ is a delta on the cone $t = r/c$ : zero outside — that is causality — and also zero strictly inside, which causality never demanded. The field at $(t,\vec x)$ is fed by the source at exactly one instant and is deaf to it otherwise. This is why the sky is not a smear. The same object in two spatial dimensions, obtained by descent:

\boxed{\;G^{(2+1)}_{\mathrm{ret}}(t, \rho) \;=\; \frac{c\;\Theta(ct - \rho)}{2\pi\sqrt{c^2t^2 - \rho^2}}\;} \qquad (5)

where: $\rho$ is distance in the plane, $\Theta$ the step function. Physical reading: the step function is causality; the missing delta is the loss of sharpness. The support is now the whole closed interior. The front still arrives at $ct = \rho$ and not a moment sooner — Flatland is perfectly causal — but the field behind it never returns to zero, decays as $1/t$ , and $\int G\,\mathrm{d}t$ diverges logarithmically.

Figure

The light cone at an event 𝓔, and the two questions that must be answered separately. LEFT — the KINEMATIC statement, already owned by aq-sr-causal-structure: the interval sign partitions spacetime into J⁺, J⁻ and the elsewhere, invariantly. The elsewhere is SEALED. RIGHT — the DYNAMICAL statement, which is this article: the cone is not decoration, it is the CHARACTERISTIC cone of the field operator. A surface of discontinuity of any solution of □ψ = 0 must have a null normal, η^μν ∂_μφ ∂_νφ = 0, so wavefronts are forced onto 45° lines by the equation itself. The heat equation has NO characteristics and its solutions leak instantly to infinity — which is exactly why the light-cone constraint is a real demand on the dynamics and not a free gift from the geometry

Two theorems: the cone is forced, and the sharpness is not

Derivation AThe characteristic cone of the wave operator is exactly the null cone of the metric

The claim is that the light-cone constraint is not an extra postulate laid on top of relativistic dynamics but a property of the dynamics themselves. We show that the surfaces across which a solution of $\Box\psi = 0$ may be non-smooth — the wavefronts, where information actually lives — are constrained by the equation itself to have null normals, and that this constraint is literally the equation of the light cone. Nothing about the metric's causal structure is assumed. It comes out.

Step 1. Ask what a wavefront is, mathematically. A propagating signal is a place where the field, or a derivative of it, jumps. Take a surface $\Sigma: \varphi(x) = 0$ and ask whether the Cauchy problem — give $\psi$ and its normal derivative on $\Sigma$ , solve off $\Sigma$ — has a unique solution. If it does, no jump can hide there. A surface where uniqueness fails is characteristic, and it is exactly where a discontinuity can sit and travel. Why: a wavefront is not "where the wave is". It is a surface on which the equation declines to say what happens next — and that freedom is what lets a sharp edge exist and move.

Step 2. Find the condition. The second-derivative term of $\Box\psi$ normal to $\Sigma$ carries the coefficient $\eta^{\mu\nu}\partial_\mu\varphi\,\partial_\nu\varphi \equiv \sigma(\partial\varphi)$ — the principal symbol on the normal covector $k_\mu = \partial_\mu\varphi$ . The Cauchy problem is solvable for the highest normal derivative precisely when that coefficient is non-zero, so $\Sigma$ is characteristic iff it vanishes. Why: the whole question has collapsed onto one scalar. Solvability is division by $\sigma(k)$ , and characteristic surfaces are where you would be dividing by zero.

Step 3. Read it. With $\eta^{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ , $\sigma(k) = k_0^2 - |\vec k|^2 = 0$ , i.e. $c^{-2}(\partial_t\varphi)^2 = |\nabla\varphi|^2$ . The normal to a characteristic surface must be null. Why: that is the light cone, reached from the analysis rather than the geometry. We asked where a discontinuity may live, and the metric's causal cone came back as the answer.

Step 4. Solve it for a point source. Try $\varphi = ct - r$ , a sphere expanding at $c$ : then $\partial_t\varphi/c = 1$ and $|\nabla\varphi| = 1$ , so $\sigma = 0$ identically — characteristic, with worldsheet the null cone $c^2t^2 = r^2$ . Now try $\varphi = ct - r/n$ , a sphere expanding at $cn$ : $\sigma = 1 - 1/n^2 \neq 0$ for $n \neq 1$ . Why: no wavefront may travel at any speed but $c$ . Not "should not" — may not. The equation admits no discontinuity surface with any other speed, so a solution has nowhere to put a faster edge.

Step 5. Note what the argument did not use. It used only the principal symbol. Add a mass, $(\Box + m^2c^2/\hbar^2)\psi = 0$ ; add a current, $\Box A^\nu = \mu_0 J^\nu$ ; add any lower-order terms you like — the symbol is untouched, so the characteristics are untouched. Why: this is why causality is robust. Klein–Gordon, Maxwell in Lorenz gauge and Dirac (whose symbol squares to $\Box$ ) all share one principal symbol and therefore one cone. Interactions change what propagates, never how fast. Contrast the heat operator, whose principal part is $-D\nabla^2$ : its characteristic condition $|\nabla\varphi|^2 = 0$ is met only by surfaces of constant $t$ , so it has no propagating fronts at all. *Verified in verify_aq-sr-causality-light-cone-constraint.py: the principal symbol is built from $\eta = \mathrm{diag}(+1,-1,-1,-1)$ and shown to vanish on $\varphi = ct-r$ identically and to equal $1 - 1/n^2$ for every other radial speed; the heat operator is shown to have no timelike characteristic; mass and source terms are shown to leave the symbol, hence the cone, unchanged.*

Go deeper · Derivation B · 3 minThe retarded Green's function is supported ON the cone in three spatial dimensions, and INSIDE it in two

Derivation BThe retarded Green's function is supported ON the cone in three spatial dimensions, and INSIDE it in two

Causality says the field cannot be non-zero outside the cone. It says nothing about the interior — and the interior is where the interesting question lives. We compute $G_{\mathrm{ret}}$ in three spatial dimensions and find a delta riding the cone: zero before the front, zero after it. Then we obtain the two-dimensional Green's function by descent, and find that delta smeared into a tail which decays as $1/t$ and never stops. Sharp signals are therefore not a consequence of relativity. They are a consequence of the number three.

Step 1. Solve $\Box G = \delta^4(x)$ in $3+1$ . Fourier-transform in space, solve the driven oscillator for each $\vec k$ with retarded boundary conditions, transform back; the standard result is $G^{(3+1)}_{\mathrm{ret}} = \delta(t - r/c)/(4\pi r)$ . Why: the only feature that matters is the argument of the delta. The field at distance $r$ is fed by the source at exactly one instant, $t_{\mathrm{ret}} = t - r/c$ . That is the whole content of "propagation at $c$ ".

Step 2. Read the support. $\mathrm{supp}\,G^{(3+1)}_{\mathrm{ret}} = \{ct = r\}$ — the cone itself, a set of measure zero. Causality would have been satisfied by anything supported in the solid cone $\{ct \geq r\}$ ; the actual answer is far stronger — zero on the inside too. Why: this is the strict Huygens principle, and it is a gift nobody asked for. It is why a point source looks like a point, why an echo is a copy of the shout rather than a smear of it, and why a second pulse can be sent down a fibre at all.

Step 3. Build the two-dimensional world honestly, by descent. A source that is a point in the plane $(x,y)$ is, seen in three dimensions, a uniform line of sources along $z$ . So

G^{(2+1)}_{\mathrm{ret}}(t, \rho) = \int_{-\infty}^{\infty} \frac{\delta\!\left(t - \sqrt{\rho^2 + z^2}/c\right)}{4\pi\sqrt{\rho^2 + z^2}}\; \mathrm{d}z, \qquad \rho^2 = x^2 + y^2.

Why: this is Hadamard's method of descent, and it is not a trick — a Flatlander is exactly a three-dimensional observer who has agreed to ignore one direction. Whatever comes out is what a two-dimensional universe genuinely does.

Step 4. Do the integral. Substitute $r = \sqrt{\rho^2+z^2}$ , so $\mathrm{d}z = r\,\mathrm{d}r/\sqrt{r^2-\rho^2}$ , with the two signs of $z$ contributing equally:

G^{(2+1)}_{\mathrm{ret}} = 2\!\int_{\rho}^{\infty}\! \frac{\delta(t - r/c)}{4\pi r}\cdot \frac{r\,\mathrm{d}r}{\sqrt{r^2 - \rho^2}} = \frac{1}{2\pi}\!\int_\rho^\infty\! \frac{\delta(t - r/c)}{\sqrt{r^2-\rho^2}}\,\mathrm{d}r = \boxed{\frac{c\,\Theta(ct-\rho)}{2\pi\sqrt{c^2t^2 - \rho^2}}}

using $\delta(t-r/c) = c\,\delta(r-ct)$ , which fires at $r = ct$ and only if $ct > \rho$ . Why: the delta was integrated away. The line source keeps feeding the observer: its nearest point reaches her at $ct = \rho$ , and ever more distant points keep arriving, forever. The delta became a step, and the front acquired a tail.

Step 5. Characterise the tail. Behind the front the field is non-zero for all time, decaying only as $G \to 1/(2\pi t)$ . Worse, integrate it:

\int_{\rho/c}^{T} G^{(2+1)}_{\mathrm{ret}}\,\mathrm{d}t = \frac{1}{2\pi}\,\mathrm{arccosh}\!\left(\frac{cT}{\rho}\right) \;\sim\; \frac{1}{2\pi}\ln\!\frac{2cT}{\rho} \;\xrightarrow{\;T\to\infty\;}\; \infty.

Why: the total signal delivered by a single Flatland flash is logarithmically infinite. Not merely long — unbounded. There is no time after which the flash is over, and no finite total to the reverberation. And none of this troubles causality, which Flatland respects perfectly.

Step 6. Name the general fact. This is Hadamard's theorem: strict Huygens — $G_{\mathrm{ret}}$ supported on the cone rather than in it — holds in an odd number of spatial dimensions $\geq 3$ and fails in every even number. The extreme case is $1+1$ : $G^{(1+1)}_{\mathrm{ret}} = \tfrac{c}{2}\Theta(ct - |x|)$ , a pure step with no decay at all — the signal arrives and simply stays. Why: this separates causality from sharpness, and it is the article's owned result. Relativity buys the cone in every dimension. It does not buy the trailing edge. That, we owe to three. *Verified in verify_aq-sr-causality-light-cone-constraint.py: the descent integral is done symbolically and reproduces equation (5) exactly, and again numerically with a regularised delta (agreement to 2 parts in $10^7$ ); the $1/t$ asymptotic, the $\mathrm{arccosh}$ time-integral and the $1+1$ step are confirmed; and every tail number printed below is recomputed at mpmath dps=50, with a per-mille perturbation of the near-front parameter shown to MOVE the mpmath answer while the naive float64 answer stays frozen — the artefact guard.*

Figure

PUNCHLINE — causality does NOT imply sharpness. Both panels are perfectly causal: the field is exactly zero before the front at ct = ρ. TOP, three spatial dimensions: G = δ(t − r/c)/4πr is a spike ON the cone and IDENTICALLY ZERO afterwards. A flash is a flash. BOTTOM, two spatial dimensions: G = c/(2π√(c²t² − ρ²)) switches on at the front and then NEVER RETURNS TO ZERO, decaying only as 1/t. At ρ = 1.00 m the tail is still 0.577 of its scale c/2πρ after 6.67 ns, 0.100 after 33.4 ns, and 3.34 × 10⁻⁹ after a full second — and the time-integrated signal DIVERGES logarithmically. Relativity gives you the cone in every dimension. The trailing edge is a gift of the number three

Limiting cases

Case	Result	Meaning
$3+1$ dimensions	$G_{\mathrm{ret}} \propto \delta(t-r/c)/r$	supported on the cone; strict Huygens; sharp
$2+1$ dimensions	$G_{\mathrm{ret}} \propto \Theta(ct-\rho)/\sqrt{c^2t^2-\rho^2}$	supported in the cone; a $1/t$ tail; still causal
$1+1$ dimensions	$G_{\mathrm{ret}} = \tfrac{c}{2}\Theta(ct-\vert x\vert)$	a pure step — the signal arrives and stays
odd spatial dims $\geq 3$ / even	Huygens holds / fails	Hadamard: sharpness is a parity effect
massive field	cone unchanged; tail even in $3+1$	symbol untouched: causality survives, sharpness does not
heat operator	no timelike characteristics	infinite propagation speed; the cone is meaningless

Common trapThe base for this article is in the OPPOSITE signature, and every sign in it inverts

The AllQFT base is written throughout in $(-,+,+,+)$ : it prints $\Delta s^2 = -c^2\Delta t^2 + \Delta x^2 + \Delta y^2 + \Delta z^2$ , $\eta_{\mu\nu} = \mathrm{diag}(-1,+1,+1,+1)$ , $u^\mu u_\mu = -c^2$ , and states "timelike $\Leftrightarrow \Delta s^2 < 0$ ". In the convention locked across the shipped relativity corpus here, every one of those inverts: $\Delta s^2 = +c^2\Delta t^2 - \Delta x^2$ , $\eta = \mathrm{diag}(+1,-1,-1,-1)$ , $u\cdot u = +c^2$ , timelike $\Leftrightarrow \Delta s^2 > 0$ . No physics differs and nobody is wrong — but a reader who takes a sign from one and an equation from the other will conclude that massive particles are spacelike. The cheapest defence is one line you can run in your head: a particle at rest has $\Delta x = 0$ , so $\Delta s^2 = c^2\Delta t^2 > 0$ — and a particle at rest is certainly timelike. If your convention makes that negative, you are in the other one.

Worked examples

1 · What a torch would look like in Flatland

A point flash at the origin of a two-dimensional world at $t = 0$ ; an observer at $\rho = 1.00\ \mathrm{m}$ . When does the light arrive, when does it stop, and how much total signal does she receive?

Step 1 — identify. Equation (5), with $\rho = 1.00\ \mathrm{m}$ and $c = 2.99792458\times10^8\ \mathrm{m\,s^{-1}}$ . Measured against its natural scale $c/(2\pi\rho)$ , the answer is the dimensionless tail profile $f(\tau) = 1/\sqrt{\tau^2-1}$ , $\tau \equiv ct/\rho$ .

Step 2 — the front. The signal arrives at $t = \rho/c = \mathbf{3.336\ ns}$ , and not one instant sooner; before that the field is exactly zero. Flatland is perfectly causal.

Step 3 — apply, and watch it fail to end.

time after the flash	$\tau = ct/\rho$	tail $f(\tau)$
$3.336\ \mathrm{ns}$ (the front)	$1$	$\infty$ (integrable)
$6.671\ \mathrm{ns}$	$2$	$0.5774$
$33.36\ \mathrm{ns}$	$10$	$0.1005$
$1.000\ \mathrm{\mu s}$	$299.79$	$3.336\times10^{-3}$
$1.000\ \mathrm{ms}$	$2.9979\times10^5$	$3.336\times10^{-6}$
$1.000\ \mathrm{s}$	$2.9979\times10^8$	$\mathbf{3.336\times10^{-9}}$

A full second after a flash one metre away, the field is still not zero — and never will be.

Step 4 — the boundary number, and it is the punchline. Ask for the total signal, $\int G\,\mathrm{d}t$ — what any integrating detector reports. From Step 5 of the derivation it is $\mathrm{arccosh}(\tau)/2\pi$ :

\int_{\mathrm{front}}^{1\,\mathrm{\mu s}} = \frac{6.396}{2\pi} = \mathbf{1.018}, \qquad \int_{\mathrm{front}}^{1\,\mathrm{s}} = \frac{20.21}{2\pi} = \mathbf{3.217}.

The integrated signal more than triples between one microsecond and one second, and it diverges logarithmically. There is no total, and no moment at which the flash is over. A Flatlander cannot send a second bit until the first has finished arriving — and it never finishes.

Step 5 — the contrast. In three spatial dimensions the same detector reads a delta spike at $3.336\ \mathrm{ns}$ and exactly zero at $3.337\ \mathrm{ns}$ , forever after. The difference between a universe with telephones and a universe without is one spatial dimension — and it is invisible to causality, which both worlds obey perfectly.

2 · Do gravity and light share the same cone?

The characteristic cone is a property of an operator, and electromagnetism and linearised gravity are different operators. Do they have the same cone? In August 2017 the question became experimental.

Step 1 — identify. GW170817: gravitational waves from a neutron-star merger, followed $1.7\ \mathrm{s}$ later by a gamma-ray burst from the same event, after a flight of about $130$ million light years. A shared cone means equal speeds, and the $1.7\ \mathrm{s}$ is astrophysics (the gamma rays are emitted slightly later). Different cones mean the delay is a speed difference accumulated over the whole journey.

Step 2 — apply. The flight time is $t = 130\times10^6\ \mathrm{yr} \times 3.156\times10^7\ \mathrm{s\,yr^{-1}} = 4.10\times10^{15}\ \mathrm{s}$ . A fractional speed difference $\delta = |c_{\mathrm{gw}} - c|/c$ accumulates a delay $\Delta t \approx \delta\,t$ , so

\delta \;\lesssim\; \frac{1.7\ \mathrm{s}}{4.10\times10^{15}\ \mathrm{s}} \;=\; \mathbf{4.1\times10^{-16}}.

Step 3 — the boundary number, and why it is the right thing to have measured. Gravity and light share one characteristic cone to about four parts in $10^{16}$ — consistent with the order- $10^{-15}$ agreement `aq-sr-causal-structure` records for the same event. And by Step 5 of the first derivation the cone is fixed by the principal symbol alone — so this is not a test of "the speed of gravitational waves" as an incidental parameter. It is a test of whether the light cone is universal, shared by every field, or whether each field secretly carries its own. If gravity had its own cone, there would be no such thing as the elsewhere.

Physics of the portrait

Origins · 2 minCausality and how this was discovered

Causality

1690–1923 · Huygens' sharp wavefront · Minkowski's cone · Hadamard's characteristics — and the discovery that sharpness is a parity effect

Causality entered physics as an assumption too obvious to state, and left the nineteenth century as a theorem with a hypothesis. Christiaan Huygens published the Traité de la Lumière in 1690 with the construction taught ever since: every point of a wavefront is a source of secondary wavelets, and the new front is their envelope. It works. It is also, in the form Huygens gave it, false in most dimensions — the envelope construction quietly assumes the wavelets contribute only on their own fronts and not behind them, which is exactly the strict Huygens principle, and exactly what fails in two dimensions. Huygens never knew, because he lived in three. Hermann Minkowski supplied the geometry, introducing the fore-cone and aft-cone construction in his Cologne lecture of 1908; but a cone drawn on a diagram constrains nothing until some equation is obliged to respect it. That obligation was made precise by Jacques Hadamard, whose Lectures on Cauchy's Problem in Linear Partial Differential Equations (Yale, 1923) placed characteristics at the centre: the surfaces on which a hyperbolic equation permits a discontinuity, which for the wave operator are the null cones and can be nothing else. Hadamard introduced the method of descent, and found the sting — his lectures treat equations with an odd and an even number of independent variables in separate chapters, because the difference between them is precisely whether a signal has a trailing edge. The light cone is the geometry; the characteristic cone is the dynamics; that they coincide is why causality is a law rather than a picture. That they coincide sharply is a bonus, and we owe it to a parity.

History

1690

— Huygens publishes the Traite de la Lumiere, with an envelope construction for wavefronts that silently assumes sharp propagation.

1908

— Minkowski, in his Cologne lecture Raum und Zeit of 21 September, introduces the fore-cone and aft-cone construction at each event. The term "light cone" is not his, and the cone structure already appears in his Grundgleichungen paper earlier that year.

1920

— Eddington, in Space, Time and Gravitation, uses the word "elsewhere" for the spacelike region: events neither past nor future but simply elsewhere.

1923

— Hadamard's Lectures on Cauchy's Problem in Linear Partial Differential Equations (Yale University Press) founds the theory of characteristics and introduces the method of descent. It treats equations with an odd and an even number of independent variables in separate chapters, because sharp propagation holds only in the odd case.

2017

— GW170817: gravitational waves and gamma rays from one neutron-star merger arrive 1.7 s apart after some 130 million years, showing that gravity and light share a single characteristic cone to about a part in 10 to the 16.

Going further

Texts. Jackson, Classical Electrodynamics, 3rd ed., §6.4, §12.11 (retarded Green's functions and their support) · Wald, General Relativity, ch. 8 · Courant & Hilbert, Methods of Mathematical Physics II, ch. VI (characteristics, and Huygens' principle by dimension) · Hassani, Mathematical Physics, ch. 22 (Green's functions in $n$ dimensions; the method of descent). Primary source: J. Hadamard, Lectures on Cauchy's Problem in Linear Partial Differential Equations, Yale University Press, New Haven (1923).

Provenance. Signature $(+,-,-,-)$ , timelike positive, $u\cdot u = +c^2$ . Base defect corrected: the AllQFT base is written entirely in $(-,+,+,+)$ — eq[0] is labelled "metric signature $-+++$ ", eq[1] prints $\eta = \mathrm{diag}(-1,+1,+1,+1)$ , eq[4] prints $u^\mu u_\mu = -c^2$ , eq[6] gives the reality of proper time as $-\Delta s^2 > 0$ , and DERIV[2] identifies the cone's interior with $\Delta s^2 < 0$ . All inverted here, each re-derived from $\eta$ rather than transcribed. Second base defect: the base's DERIV[1] gives the antitelephone condition as $uv > c^2$ "whenever" a causal loop closes — that is the single-leg condition; `aq-sr-causal-structure` owns the correction ( $v > 2wc^2/(w^2+c^2)$ ), which this article recalls rather than restates. The base's retarded-Green's-function card (eq[5]) is signature-independent, correct, and is the seed of this article's owned result. verify_aq-sr-causality-light-cone-constraint.py asserts every number printed above.

Scope. This article owns the dynamical half of causality: the characteristic cone, the support of $G_{\mathrm{ret}}$ , and the dimension-dependence of sharpness. The kinematic half — order reversal iff spacelike, the invariance of the causal partition, the antitelephone and its two thresholds — belongs to `aq-sr-timelike-spacelike-lightlike-separations`, `aq-sr-the-light-cone` and `aq-sr-causal-structure`, is recalled by name, and is not re-derived.

Builds on. `aq-sr-spacetime-interval` · `aq-sr-minkowski-metric` · `aq-sr-causal-structure` · `aq-sr-the-light-cone`. Leads to. `aq-sr-tachyons-superluminal-particles` · `aq-sr-reinterpretation-principle-causal-paradoxes`.

If relativity already proves nothing outside the light cone, what is left for this article to prove?

Relativity proves that the cone is invariant — that all observers agree which events are causally connectable. That is geometry: necessary for causality but not sufficient, because a cone constrains nothing until some equation is obliged to respect it. The heat equation is the standing counterexample: perfectly good physics, and its solutions are non-zero everywhere the instant the source switches on. So the real content of "the light-cone constraint" is a claim about field equations — that theirs is a hyperbolic operator whose characteristics coincide with the metric's null cones — and that has to be proved. This article proves it, then asks the question geometry cannot even pose: given that the signal stays in the cone, where in the cone does it go?

Is the two-dimensional tail a violation of causality?

No, and that it is not is the whole point. The two-dimensional Green's function carries a step function $\Theta(ct-\rho)$ : it is exactly zero before the front, so nothing ever reaches an observer earlier than light would. Flatland obeys the light-cone constraint perfectly. What it lacks is not causality but sharpness. These are logically independent properties, and conflating them is the commonest confusion in the area: causality is about the boundary of the support, sharpness about the interior. Relativity fixes the first in every dimension and says nothing about the second.

Does a massive field also propagate on the light cone?

Its front does, exactly, and that is the part causality cares about. A mass term changes only the lower-order part of the equation, so the principal symbol — and therefore the characteristics — are identical to the massless case. The leading edge of a Klein–Gordon wave packet travels at exactly $c$ , which is startling until you notice it is the front velocity, not the group velocity, and it carries no energy. What the mass does destroy is strict Huygens: $G_{\mathrm{ret}}$ for a massive field is non-zero throughout the cone's interior even in three spatial dimensions. A massive field is causal but not sharp.

Why should the number of spatial dimensions matter at all?

It feels as though it should not, because the cone looks the same in every dimension — and it does. The dimension enters not through the cone but through the measure: how much of the source you are integrating over. The descent argument makes it plain. A point source in two dimensions is, seen from three, an infinite line of sources along the suppressed axis. Its nearest point reaches you at $ct = \rho$ ; then the more distant points keep arriving, forever, each a little later and a little weaker. The tail is not mysterious — it is the rest of the line still being heard. Sharpness in three dimensions is the statement that a point really is a point, with nothing hiding behind it.

What exactly is a characteristic surface, in plain terms?

A surface on which the equation refuses to tell you what happens next. Give a PDE data on a surface and ask it to propagate that data off the surface: usually it can, uniquely, and then nothing surprising can hide there. On a characteristic surface the procedure breaks down — the coefficient you would need to divide by is exactly zero — and the equation permits the solution to be non-smooth across it. That freedom is what a wavefront is: a moving edge the equation allows to exist and travel. For $\Box$ the breakdown condition is $\eta^{\mu\nu}\partial_\mu\varphi\,\partial_\nu\varphi = 0$ , whose solutions are the light cones. So "wavefronts travel at $c$ " is not a physical input — it is a fact about where the wave operator is singular.

Superluminal phase velocities exist. Why is that not a violation of the support theorem?

Because a phase velocity is not the velocity of anything carrying the field's support. A monochromatic wave in a waveguide genuinely has phase fronts moving faster than $c$ — but it has existed for all time, so it has no beginning, no front, and no event at which it was "sent". To send a signal you must modulate it, and the modulation travels at the front velocity, fixed by the characteristics and exactly $c$ . This is Sommerfeld and Brillouin's result, and it is our theorem seen from the frequency side: dispersion lives in the lower-order terms, so no amount of it can move the support of the propagator. The same disposes of scissors and swept laser spots — neither has a propagator running between its successive positions.