Causal Structure of Spacetime

Strip general relativity down to the one thing that makes it a theory of cause and effect, and what remains is the light cone. A Lorentzian metric $g_{ab}$ with signature $(-,+,+,+)$ does what a Riemannian metric never can: at every event it makes the form $g_{ab}V^aV^b$ indefinite, so tangent vectors split into three kinds — timelike, null, spacelike — divided by the cone $g_{ab}V^aV^b=0$ . Assemble these cones over the manifold and you have the causal structure: the invariant skeleton on which relativity hangs past, future, "elsewhere", signal propagation, horizons, and prediction itself. Locally these are the flat 45° cones of special relativity — the equivalence principle guarantees a freely falling frame in which $g_{ab}=\eta_{ab}$ to first order. Gravity does exactly one thing: it tilts and reshapes the cones from event to event. Where the tilt is gentle you have ordinary curved-spacetime physics; where the future cone tips until every ray inside points to smaller radius, you have a horizon; where cones tip until a future-directed curve closes on itself, you have a closed timelike curve and a causally sick spacetime. The deepest single fact here is that causal structure is conformally invariant. Rescale the metric by any positive function, $g_{ab}\to\Omega^2 g_{ab}$ , and the sign of $g_{ab}V^aV^b$ cannot change, so every light cone and causal relation is untouched — only the conformal class $[g_{ab}]$ matters. That is what lets Penrose and Carter squash an infinite spacetime onto a finite diagram without distorting a single causal relationship, and it seeds the whole global machinery — $I^{\pm}$ , $J^{\pm}$ , achronal sets, Cauchy surfaces, global hyperbolicity — that the rest of this cluster unfolds.

Key takeaways

The causal structure assigns every pair of events a timelike, null, or spacelike relation, fixed entirely by the light cones of the Lorentzian metric $g_{ab}$ with signature $(-,+,+,+)$ .
At each event the metric carves the tangent space into a double cone: $g_{ab}V^aV^b<0$ is timelike, $=0$ is null, $>0$ is spacelike.
A continuous, nowhere-zero choice of which half-cone is "the future" is a time orientation; it exists exactly when the manifold is time-orientable.
Causal structure is a conformal invariant: $g_{ab}\to\Omega^2 g_{ab}$ with $\Omega>0$ leaves every cone and causal relation unchanged — the theorem behind Penrose-Carter diagrams.
From local cones one builds the global objects $I^{\pm}(p),\,J^{\pm}(p)$ , achronal boundaries ruled by null geodesics, domains of dependence, and Cauchy surfaces, ending in global hyperbolicity.
Key idea: the Lorentzian metric hangs a light cone at every event; the timelike/null/spacelike trichotomy built from these cones — and preserved under conformal rescaling — is the causal structure, from which all global notions of past, future, prediction, and horizons descend.

The core insight

The Lorentzian metric assigns to each event a light cone; the timelike/null/spacelike trichotomy of vectors and curves built from these cones — and crucially preserved under conformal rescaling — is the causal structure, from which all global notions of past, future, prediction, and horizons are derived.

Where this lives today

Causal structure is the working language of every frontier in gravity. Numerical relativity, the engine behind LIGO-Virgo-KAGRA waveform templates, is built on it: because the event horizon $\partial J^{-}(\mathscr I^{+})$ needs the entire future, the codes cannot see it while they run, so they track local causal conditions on each slice and reconstruct the true boundary afterward by shooting null generators backward from the settled remnant. Cosmology reads the observable universe off a conformal diagram — the particle horizon and the de Sitter event horizon are causal-structure statements, and the puzzles that motivated inflation are questions about which regions were ever in causal contact. The deepest open problems are causal too: cosmic censorship (are breakdowns of prediction always hidden behind horizons?) remains unproven; the information problem turns on whether an evaporating hole ever truly cuts a region off from infinity; and causal set theory takes this article's closing fact — causal order plus a volume element recovers the whole metric — to propose the discrete causal order of events as the substance of spacetime.

The physical picture

Attach to every event a double cone: the future half is every direction a light ray could leave along, the past half every direction one could arrive from. Everything you can ever influence lies inside or on your future cone; everything that could have influenced you lies inside or on your past cone; the rest is "elsewhere" — spacelike separated, causally out of reach at that instant. The whole content of relativistic causality is the orientation of these cones, and a worldline inherits its character from its tangent: timelike inside the cones (a massive particle), null on them (light), spacelike piercing them (no physical signal).

Figure

The light cone at an event p splits the tangent space into three: a timelike vector points inside the cone with g(V,V) = −1, a null vector rides the cone edge with g(V,V) = 0, and a spacelike vector pierces it with g(V,V) = +1, in signature −+++

In flat spacetime every cone points the same way and opens at 45° (with $c=1$ ); gravity tilts and deforms them. The tilt is not a metaphor: in Schwarzschild coordinates the radial cone edges obey $c\,\mathrm{d}t/\mathrm{d}r=\pm(1-2GM/c^2 r)^{-1}$ , a slope that steepens without bound as $r$ falls toward $2GM/c^2$ . The cone pinches shut in those coordinates, and in a regular chart it has tipped so far inward that its whole future opens toward smaller radius — a horizon expressed in nothing but the direction of light. The second figure makes this a set of concrete slopes.

Figure

Radial light cones in Schwarzschild coordinates narrow as r approaches 2M because the null slope dt/dr = (1 − 2M/r)⁻¹ diverges: it is 1.5 at r = 6M, 2 at r = 4M, 3 at r = 3M, and → ∞ at the horizon r = 2M where the cone pinches shut; inside, the cones tip over so that every future points toward r = 0

Conformal invariance ties the whole subject together. The cone's shape — its opening — carries the causal information; its size, how fast proper time accumulates, is metric information you can rescale away without changing who can signal whom. Stretch the metric by any $\Omega^2>0$ and every cone stays put, which is why a causal diagram can be finite yet perfectly faithful. Global causal analysis then tracks which events can be joined by causal curves while the cones twist underneath them — and names the pathologies (closed timelike curves, Cauchy horizons) that appear when the twisting is severe.

The core equations

Everything else descends from classifying a tangent vector by the sign of its squared norm. That sign is coordinate-independent, so the trichotomy is a genuine geometric fact, not an artifact of the chart.

g_{ab}V^aV^b\;\begin{cases}<0 & \text{timelike}\\[2pt]=0 & \text{null}\\[2pt]>0 & \text{spacelike}\end{cases}\qquad (V\neq 0) \qquad (1)

where: $g_{ab}$ — Lorentzian metric tensor, signature $(-,+,+,+)$ ; $V^a$ — a nonzero tangent vector at $p\in M$ ; $g_{ab}V^aV^b$ — the squared norm, whose sign is the causal character. Reading: three regimes of one real number encode "who can reach whom" at a point. By the equivalence principle there is always a freely falling frame at $p$ in which the metric reduces to the flat Minkowski form and the cone opens at exactly 45°. This is the sense in which causal structure is locally that of special relativity everywhere.

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

where: $\eta_{ab}$ — flat metric in an orthonormal frame; $c$ — speed of light, setting the cone opening, usually put to $1$ . Reading: setting $\mathrm{d}s^2=0$ gives $|\mathrm{d}\vec x|/\mathrm{d}t=c$ — the light cone is the locus of rays travelling at $c$ . The heart of the subject is that this classification survives any positive rescaling of the metric. This single line makes the entire conformal-compactification program legitimate.

\tilde g_{ab}=\Omega^2 g_{ab},\ \Omega>0\ \Rightarrow\ \operatorname{sgn}\!\big(\tilde g_{ab}V^aV^b\big)=\operatorname{sgn}\!\big(g_{ab}V^aV^b\big) \qquad (3)

where: $\Omega(x)$ — a smooth strictly positive conformal factor; $\tilde g_{ab}$ — the rescaled metric. Reading: timelike stays timelike, null stays null, spacelike stays spacelike — causality depends only on the conformal class $[g_{ab}]$ , never the scale. The size a rescaling discards is proper time: along a timelike curve $\mathrm{d}\tau/\mathrm{d}\lambda=\sqrt{-g_{ab}\,\dot x^a\dot x^b}$ is genuine metric information, exactly what conformal invariance throws away while holding the cones fixed. From the pointwise cones one lifts to global reachability. The chronological and causal futures collect everything an event can influence — one strictly inside the cones, one including their null edges.

I^{+}(p)=\{q:\ p\ll q\},\qquad J^{+}(p)=\{q:\ p\preceq q\} \qquad (4)

where: $p\ll q$ — a future-directed timelike curve runs from $p$ to $q$ (chronology); $p\preceq q$ — $q=p$ or a future-directed causal (timelike-or-null) curve runs from $p$ to $q$ . Reading: $I^{+}$ is open, strictly inside the cones; $J^{+}$ adds the null edge; pasts $I^{-},J^{-}$ reverse time. The frontier between "reachable" and "unreachable" is woven from light rays. Penrose's theorem states that an achronal causal boundary is a null hypersurface ruled by null geodesic generators.

\dot J^{+}(p)\setminus\{p\}\ \text{is generated by null geodesics}:\quad k^a\nabla_a k^b=0,\ \ g_{ab}k^ak^b=0 \qquad (5)

where: $\dot J^{+}(p)$ — the boundary (frontier) of the causal future; $k^a$ — the null tangent of a generator. Reading: a boundary point is never reachable chronologically, so any causal curve that does reach it can only be a null geodesic — never spacelike (not causal) nor timelike (that lies inside $I^{+}$ ). In a general spacetime $J^{+}(p)$ need not be closed, so a boundary point may not be causally reachable at all; the generator through such a point is past-inextendible instead of running back to $p$ . Finally, a globally consistent notion of "future" is a topological condition, not a given. It holds precisely when a continuous timelike vector field exists to point the way.

M\ \text{time-orientable}\ \Longleftrightarrow\ \exists\, t^a\ \text{continuous},\ g_{ab}t^at^b<0\ \text{everywhere} \qquad (6)

where: $t^a$ — a nowhere-vanishing timelike vector field selecting the future half-cone. Reading: without such a field the "future" label would have to flip around some loop, exactly as an orientation flips on a Möbius band.

Derivation AWhy a positive rescaling cannot change a single causal relation

The goal is to prove that every light cone, and hence every causal relation, is untouched when the metric is multiplied by an arbitrary smooth positive function $\Omega^2(x)$ — the theorem that licenses Penrose-Carter diagrams and conformal compactification. We work pointwise first, then promote to curves and to the global sets. The whole result rests on one elementary fact: multiplying a real number by a strictly positive number preserves its sign.

Step 1. Substitute the rescaled metric into the squared norm: $\tilde g_{ab}V^aV^b=\Omega^2(x)\,g_{ab}V^aV^b$ , since $V^a$ is unchanged and only the metric is scaled. Why: the causal character lives entirely in the number $g_{ab}V^aV^b$ .

Step 2. Because $\Omega>0$ gives $\Omega^2>0$ , multiplication cannot flip a sign: $\operatorname{sgn}(\tilde g_{ab}V^aV^b)=\operatorname{sgn}(g_{ab}V^aV^b)$ for every $V^a$ , so the timelike, null, and spacelike vectors of $g$ are exactly those of $\tilde g$ . Why: a positive factor is invisible to the sign, the only thing causality reads.

Step 3. A curve's causal character is fixed by its tangent, and every tangent keeps its character under Step 2, so the causal curves of $g$ are precisely those of $\tilde g$ . Why: causal relations are defined by the existence of such curves, so unchanged curves mean unchanged relations.

Step 4. Since $I^{\pm}(p)$ and $J^{\pm}(p)$ are defined purely by the existence of such curves, they are identical for $g$ and $\tilde g$ , and the causal structure depends only on the conformal class $[g_{ab}]$ . Why: one may then pick a representative that pulls infinity to a finite distance while keeping every cone at 45° — the Penrose-Carter construction. *The identity $\tilde g_{ab}V^aV^b=\Omega^2 g_{ab}V^aV^b$ follows by direct substitution of the rescaled metric.*

Go deeper · Derivation B · 1 minTime orientability as the existence of a global timelike vector field

Derivation BTime orientability as the existence of a global timelike vector field

The goal is to show that calling one half-cone "the future" consistently across the entire manifold — a time orientation — is equivalent to the existence of a single continuous, nowhere-vanishing timelike vector field $t^a$ . Locally there is nothing to prove: each cone has two sheets and one picks one. The content is global and genuinely topological, because the choice must stay continuous around every closed loop. We prove both directions of the equivalence.

Step 1. At each $p$ the timelike vectors form two disjoint open convex cones, the two sheets of the hyperboloid $g_{ab}V^aV^b=-1$ ; labelling one sheet "future" is a binary choice. A time orientation is a continuous such labelling over all of $M$ , so the future sheet never swaps discontinuously as $p$ moves. Why: convexity of each sheet lets us add future-directed vectors and stay future-directed, which the next step needs.

Step 2. Suppose $M$ is time-orientable. Locally pick the future-directed unit normal of any foliation; these local timelike fields lie in the chosen future cones. Patch them with a partition of unity: a positive-weighted sum of future-directed timelike vectors is again future-directed and timelike (the cone is convex), giving a global $t^a$ with $g_{ab}t^at^b<0$ . Why: the partition of unity turns a local choice into one continuous global object without leaking out of the cone.

Step 3. Conversely, given a global continuous timelike $t^a$ , declare a timelike $V^a$ future-directed exactly when $g_{ab}t^aV^b<0$ . This test is continuous and picks one sheet at every point, so it is a time orientation. Why: the inner product with a fixed timelike field is negative on one sheet and positive on the other, separating them everywhere at once.

Step 4. The two constructions are mutually inverse, so "time-orientable" and "admits a global timelike vector field" are the same condition; when it fails, the future label cannot be propagated around some loop without contradiction. Why: this is why time orientability is a genuine hypothesis in the theorems, assumed so that "past" and "future" are globally meaningful.

Go deeper · Derivation C · 2 minHow the light cones tip over at the Schwarzschild radius

Derivation CHow the light cones tip over at the Schwarzschild radius

The goal is to trace quantitatively how the radial light cones tilt as one approaches the Schwarzschild radius, and to show that a black-hole horizon is nothing but a statement about the direction of light — a place where the future cone has tipped so far that every future-directed causal curve is forced to smaller radius. We compute the cone-edge slope in Schwarzschild coordinates, watch it diverge, recognise the divergence as a coordinate artifact, and read off the causal meaning in a regular chart.

Step 1. Start from the Schwarzschild line element and set $\mathrm{d}s^2=0$ with $\mathrm{d}\vartheta=\mathrm{d}\varphi=0$ for a purely radial ray: $-(1-2GM/c^2r)c^2\mathrm{d}t^2+(1-2GM/c^2r)^{-1}\mathrm{d}r^2=0$ . Solving gives the two cone edges $c\,\mathrm{d}t/\mathrm{d}r=\pm(1-2GM/c^2r)^{-1}$ . Why: the null condition is exactly the equation of the light cone, so its two roots are the two edges of the radial cone in the $(t,r)$ plane.

Step 2. Evaluate the slope factor $(1-2GM/c^2r)^{-1}$ as $r\to r_s=2GM/c^2$ . Writing $r$ in units where $r_s=2M$ , it is $1.5$ at $r=6M$ , $2$ at $r=4M$ , $3$ at $r=3M$ , and $\to\infty$ as $r\to 2M$ : the cone edges go vertical and the cone pinches shut. Why: the diverging slope shows that Schwarzschild $t$ is a bad coordinate at the horizon, not that anything physical is singular there.

Step 3. Pass to ingoing Eddington-Finkelstein coordinates, regular at $r=2M$ . There the cones do not pinch; they tilt smoothly, and at $r=2M$ the outgoing edge is vertical (light barely holds station) while the ingoing edge still falls inward. Why: only a chart regular across the surface shows the true shape — a tilt, not a collapse.

Step 4. For $r<2M$ both edges of the future cone point toward decreasing $r$ , so every future-directed causal curve obeys $\mathrm{d}r<0$ : all futures lead to $r=0$ . This region is the black hole, and its boundary $r=2M$ is the event horizon $\partial J^{-}(\mathscr I^{+})$ . Why: the horizon is located with no appeal to force or curvature — purely from where the light cones may point, the defining move of causal-structure reasoning. *The radial null slope $c\,\mathrm{d}t/\mathrm{d}r=\pm(1-2GM/c^2r)^{-1}$ is confirmed by direct evaluation, giving slope factors $1.5,\,2,\,3\to\infty$ at $r=6M,\,4M,\,3M\to 2M$ .*

Limiting cases

Case	Result	Meaning
Local / Minkowski limit	$g_{ab}(p)=\eta_{ab},\ \partial_c g_{ab}(p)=0$	In a freely falling frame the cones are exactly the special-relativistic 45° cones — causal structure is locally flat by the equivalence principle.
Conformal rescaling	$\tilde g_{ab}=\Omega^2 g_{ab}\Rightarrow \operatorname{sgn}(\tilde g_{ab}V^aV^b)=\operatorname{sgn}(g_{ab}V^aV^b)$	Every $I^{\pm},J^{\pm}$ and the whole cone field are identical for conformally related metrics; only $[g_{ab}]$ matters.
Stationary / static	timelike Killing $\xi^a$ ; $g_{tt}=-(1-2GM/c^2r)$	A timelike Killing field gives a preferred time; $\xi^a$ becomes null at a Killing horizon, e.g. $r=2GM/c^2$ .
Causality violation	$\exists\,\gamma$ future timelike with $\gamma(0)=\gamma(1)$	A closed timelike curve (Gödel's universe) makes chronology fail and collapses the hierarchy at its lowest rung.
Conformally flat / FLRW	$g_{ab}=a^2(\eta)\,\eta_{ab}$	Cosmological cones share Minkowski's causal structure up to scale; horizons read straight off the conformal diagram.

Common trapA vector being timelike at p means the distant event it points at lies in J⁺(p).

It does not. Causal character is strictly pointwise — the sign of $g_{ab}V^aV^b$ at $p$ and nowhere else. To place a distant event $q$ in $J^{+}(p)$ you need an entire future-directed causal curve from $p$ to $q$ that stays inside the cones the whole way, while the cones themselves tilt underneath it. A timelike vector at $p$ says only that a curve can begin causally, nothing about whether one reaches $q$ . The mirror-image trap is reading "spacelike separation" as a global "now": the character of a relation is invariant, but there is no observer-independent slicing into simultaneous surfaces. Both errors promote a local tangent-space fact to a global claim it cannot support.

Worked examples

1 · Cone-closing at a stellar-mass black hole (fully worked)

Take a black hole of mass $M=10\,M_\odot$ . Find its Schwarzschild radius, tabulate the radial light-cone slope $c\,\mathrm{d}t/\mathrm{d}r=(1-2GM/c^2r)^{-1}$ at $r=6M,4M,3M$ , and show light emitted just outside the horizon takes infinite coordinate time to escape. Use $G=6.674\times10^{-11}$ , $c=2.998\times10^{8}\ \mathrm{m/s}$ , $M_\odot=1.989\times10^{30}\ \mathrm{kg}$ .

Step 1. Schwarzschild radius. With $M=10M_\odot=1.989\times10^{31}\ \mathrm{kg}$ , $r_s=2GM/c^2$ : the numerator $2GM=2.655\times10^{21}$ and $c^2=8.988\times10^{16}$ , so $r_s=\mathbf{2.95\times10^{4}\ m\approx 29.5\ km}$ — about $2.95$ km per solar mass, as expected.

Step 2. Cone slopes. Writing $r$ in units where $r_s=2M$ , the slope factor $(1-2M/r)^{-1}$ is $(1-\tfrac13)^{-1}=\mathbf{1.5}$ at $r=6M$ , $(1-\tfrac12)^{-1}=\mathbf{2}$ at $r=4M$ , and $(1-\tfrac23)^{-1}=\mathbf{3}$ at $r=3M$ . The edges steepen monotonically, and at $r=2M$ the factor diverges — the cone pinches shut.

Step 3. Infinite escape time. An outgoing radial ray satisfies $c\,t=\int \mathrm{d}r/(1-2M/r)=r+2M\ln(r/2M-1)+\text{const}$ , the tortoise coordinate. As the emission radius $r\to 2M$ the logarithm $\to-\infty$ , so the arrival time at a fixed distant point diverges: light from the horizon never reaches infinity in finite coordinate time.

Step 4. Causal reading. The diverging slope and the diverging escape time are one fact seen twice: at $r=2M$ the future cone's outgoing edge has gone vertical, so an outgoing photon holds station forever; cross inside and both edges point to smaller $r$ . The horizon at $r_s\approx 29.5$ km is a purely causal boundary — where light can no longer make outward headway — not a place of large curvature (tidal forces there are mild for a $10\,M_\odot$ hole).

2 · Classifying vectors and testing conformal invariance

In Minkowski space with signature $(-,+,+,+)$ and $c=1$ , classify the vectors $V^a=(2,1,1,0)$ and $k^a=(1,1,0,0)$ . Then rescale the metric by $\Omega^2=4$ and confirm the classifications are unchanged.

Step 1. Norm of $V$ : $g_{ab}V^aV^b=-(2)^2+1^2+1^2+0^2=-4+2=\mathbf{-2}<0$ , so $V$ is timelike, with coordinate speed $|\vec V|/V^0=\sqrt2/2\approx\mathbf{0.71}<1$ — a genuine subluminal direction.

Step 2. Norm of $k$ : $g_{ab}k^ak^b=-(1)^2+1^2=\mathbf{0}$ , so $k$ is null — it rides the light cone at speed $1$ .

Step 3. Rescale by $\tilde g_{ab}=4\,\eta_{ab}$ : the norms scale by $4$ , giving $\tilde g_{ab}V^aV^b=\mathbf{-8}<0$ (still timelike) and $\tilde g_{ab}k^ak^b=\mathbf{0}$ (still null). The numbers changed but every sign is identical.

Step 4. Lesson. A conformal factor rescales proper time and length (the cone's "size") but leaves the trichotomy (its "shape") invariant. This is the calculation behind the whole conformal-diagram method: $[g_{ab}]$ alone fixes causality, so you may pick any convenient representative.

Physics of the portrait

The modern theory of causal structure is largely one person's ideas made rigorous: conformal infinity, the causal boundary, the achronal set, the diagram that fits an infinite universe on a page. Before the 1960s "the region nothing escapes" was a slogan; the geometrization of causality turned it into a theorem.

Origins · 1 minRoger Penrose and how this was discovered

Roger Penrose

1931– · conformal infinity and the global causal structure of spacetime

Roger Penrose remade general relativity by importing topology and global causal analysis into a field dominated by explicit metrics and local calculation. In the early 1960s he introduced conformal infinity — rescaling the metric by a factor that vanishes at large distance so the infinite reaches of spacetime become a finite ideal boundary, future and past null infinity $\mathscr I^{\pm}$ — and proved that, since the factor stays positive inside, it preserves every causal relation. That single move gave the event horizon its permanent definition as $\partial J^{-}(\mathscr I^{+})$ and gave the field the "Penrose diagram", which shows a spacetime's entire causal anatomy at a glance. He formalised the tools this article rests on: chronological and causal futures, achronal sets, the theorem that causal boundaries are ruled by null geodesics, and, in 1965, the closed trapped surface behind his first singularity theorem. He stated the cosmic censorship conjecture — whether the pathologies of causal structure are always safely hidden — and his case that black holes were robust predictions of the theory earned the 2020 Nobel Prize. Almost every named object in this cluster descends from his hand.

History

1908

— Minkowski geometrizes special relativity, hanging a light cone at every event

1949

— Gödel finds a rotating cosmology riddled with closed timelike curves, showing chronology can fail

1963

— Penrose introduces the conformal treatment of infinity and the boundaries 𝓘⁺ and 𝓘⁻

1970

— Geroch proves a globally hyperbolic spacetime is foliated by Cauchy surfaces

1973

— Hawking and Ellis systematize the causal-structure hierarchy in their monograph

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), Ch. 2, Box 27, Ch. 31 · Wald, General Relativity (1984), Ch. 8 and Appendix D · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), Ch. 6 · Carroll, Spacetime and Geometry (2004), Ch. 2, Ch. 5 · Penrose, Techniques of Differential Topology in Relativity (1972) · Schutz, A First Course in General Relativity (2nd ed., 2009), Ch. 1-2 · Hartle, Gravity (2003), Ch. 4-5 · Geroch, Domain of Dependence, J. Math. Phys. 11, 437 (1970). Key results in this article: the causal trichotomy follows from the indefiniteness of the Lorentzian signature via Sylvester's law of inertia; the conformal-invariance identity is $\tilde g_{ab}V^aV^b=\Omega^2 g_{ab}V^aV^b$ with sign preserved; the radial null slope is $c\,\mathrm{d}t/\mathrm{d}r=\pm(1-2GM/c^2r)^{-1}$ with $r_s=2GM/c^2$ , giving tabulated slopes $1.5,\,2,\,3$ at $r=6M,\,4M,\,3M$ ; $r_s\approx 29.5$ km for $10\,M_\odot$ ; the Minkowski norms are $g(V,V)=-2$ and $g(k,k)=0$ , invariant in sign under $\Omega^2=4$ ; and Penrose's theorem states that achronal boundaries are generated by null geodesics.

Prerequisites. Lorentzian Metric and Signature · Tangent Spaces and Vectors · Null Geodesics and Light Cones. Leads to. Chronological and Causal Futures and Pasts · Conformal Transformations and Causal Invariance · Causality Conditions Hierarchy · Global Hyperbolicity · Penrose-Carter Conformal Diagrams.

Ask the tutor: Want to walk through how these tilting cones assemble into a full Penrose-Carter diagram, or why global hyperbolicity is the exact dividing line between spacetimes that predict and spacetimes that don't?

Why does the signature −+++ matter so much, when Riemannian geometry gets by without causality?

Because the minus sign is the entire source of causal structure. In a Riemannian metric every squared norm is positive — the form is definite — so directions cannot be distinguished. The Lorentzian minus sign makes $g_{ab}V^aV^b$ indefinite: negative, zero, or positive, which is the timelike / null / spacelike trichotomy, with the light cone the locus where the form vanishes. Flip to signature $+---$ and only the sign you call "timelike" changes; remove the indefiniteness and you lose light cones, futures, pasts, and horizons — everything this article is about.

What is the difference between the chronological future I⁺(p) and the causal future J⁺(p)?

$I^{+}(p)$ is the set of events reachable from $p$ by future-directed timelike curves — strictly inside the future cones, reachable slower than light. $J^{+}(p)$ adds $p$ and every event reachable by a causal curve (timelike or null), so it includes the light rays on the boundary — roughly, $I^{+}(p)$ plus its light-ray edge. The distinction is sharp: $I^{+}(p)$ is always open, while the frontier $\dot J^{+}(p)$ — the boundary between "can be influenced" and "cannot" — is generated by null geodesics and is where horizons live.

If causal structure ignores the scale of the metric, how can it possibly contain all of geometry?

It contains all of geometry once you add back one thing: a volume element. The conformal class $[g_{ab}]$ fixes every light cone and hence the causal order, but it forgets "sizes" — proper times and lengths. Supply a scalar volume measure and you reconstruct the full $g_{ab}$ : the cones give the metric up to a conformal factor, and the volume fixes that last factor. This "causal order plus counting" decomposition is the founding idea of causal set theory, which proposes that a discrete set of events carrying only a causal order and a counting measure is the microscopic substance of spacetime.

Can two events be spacelike separated in one frame and timelike separated in another?

No — and this invariance is exactly what makes causal structure meaningful. The causal character of the relation between two events is the sign of the interval between them, a scalar: every observer agrees on whether two events are timelike, null, or spacelike separated. What observers disagree about is the time ordering of two spacelike-separated events, and how to slice spacetime into surfaces of simultaneity. So relativity of simultaneity is real, but it never converts a spacelike relation into a timelike one; if it could, one observer could see a cause precede its effect and another see the reverse, and causality would be frame-dependent nonsense.

What actually goes wrong in a spacetime with closed timelike curves?

A closed timelike curve is a worldline that is everywhere future-directed and timelike yet returns to its own starting event — the cones have tipped so far around some loop that "always moving into the future" brings you back to your past. Gödel's rotating universe (1949) is the classic example. Once such a curve exists, the causal order $\ll$ is no longer a partial order (you have $p\ll p$ ), a consistent "before and after" breaks down, and you cannot pose initial data and evolve it. This is the lowest rung of the causality hierarchy to fail; the higher conditions (chronology, causality, strong and stable causality, global hyperbolicity) demand progressively better behaviour, and global hyperbolicity — the top rung — is the one that guarantees deterministic physics.

Why is global hyperbolicity singled out as the physically clean condition?

Because it is exactly the condition under which the equations of physics have well-posed, deterministic solutions. A spacetime is globally hyperbolic when it is causal (no closed causal curves) and every causal diamond $J^{+}(p)\cap J^{-}(q)$ is compact. Geroch's theorem then guarantees a Cauchy surface: a slice whose domain of dependence is the whole spacetime, so data on it determine the entire past and future. Every hyperbolic field equation propagates its data strictly within the light cones, so the initial-value problem is well-posed there. Drop global hyperbolicity and you can have Cauchy horizons, beyond which prediction fails; keeping it is what lets general relativity be predictive at all.