Chronological and Causal Futures and Pasts

The causal sets are not a museum piece; they are the load-bearing definitions of modern gravitational physics. A black hole's event horizon is defined globally as the boundary $\dot J^-(\mathscr{I}^+)$ — the frontier of the causal past of future null infinity — and the boundary theorem proved above is exactly why that surface is null, achronal, and generated by geodesics with no future endpoints; this is also why a horizon cannot be located by any local measurement and why numerical relativists track apparent horizons on each slice instead of waiting for the whole future. The same causal-topological machinery underwrites the Penrose–Hawking singularity theorems, whose engine is the compactness of $\dot J^+$ of a trapped surface, and it defines the domain of dependence $D^\pm(S)$ that makes the Cauchy problem of general relativity well posed — the theorem, due to Choquet-Bruhat and Geroch, that initial data on a Cauchy surface determine a unique maximal globally hyperbolic development. In quantum gravity the conformal invariance of $I^\pm$ and $J^\pm$ is the backbone of the AdS/CFT causal-wedge and entanglement-wedge constructions, where bulk causal futures map to boundary subregions, and the causal set programme of Bombelli, Lee, Meyer and Sorkin takes the discrete order relation $x\prec y$ (a stripped-down $J^+$) as the fundamental substance of spacetime, from which both topology and conformal metric are meant to re-emerge. Every time a relativist writes "the region causally influenced by the collision" or "the past light cone of the CMB", the precise object behind the phrase is one of these four sets.

Stand at an event $p$ and set off a flash. The wavefront sweeps out the future light cone, and everything you could reach by travelling slower than light — by some timelike journey — lies strictly inside it, in $I^+(p)$. Everything you could reach only by riding a light signal itself lies exactly on the cone, in $J^+(p)\setminus I^+(p)$. The openness of $I^+(p)$ is the geometric statement that if you can reach $q$ with room to spare (a timelike curve leaves slack inside the cone), you can wiggle the endpoint a little in any direction and still get there. A point reachable only by a null geodesic has no such slack: nudge outward and you leave the causal future entirely. That single asymmetry — slack inside, no slack on the edge — is the seed of every theorem in this article. The first figure sets $I^+(p)$, $J^+(p)$, and the null boundary side by side and hangs concrete numbers on them.

The subtlety that curved and holey spacetimes force on us is that this tidy cone is only a local picture. Globally, the boundary of what $p$ can causally reach is still woven from null geodesics, but those generators can enter the interior — they can leave the boundary — wherever they cross each other at a caustic or run into a removed or singular region. Remove a single point from Minkowski space and a shadow appears: a locus lying in $\overline{I^+(p)}$ yet not in $J^+(p)$, because every causal curve that would reach it is forced through the missing event. And because the timelike/null/spacelike character of a tangent vector survives any rescaling $g\to\Omega^2 g$, the entire causal ledger is blind to conformal factors. That is what lets us compress all of infinite Minkowski space into a finite triangle and still read off $I^\pm$ and $J^\pm$ as ordinary regions bounded by 45° lines — the second figure.

The definitions are the whole game, so state them precisely. The chronological future collects the events a massive observer could actually visit — those joined to $p$ by a future-directed timelike curve, whose tangent stays strictly inside the cone.

where: $p,q$ — events (points) of the spacetime manifold $M$; a timelike curve is a piecewise-smooth curve whose tangent $T^\mu$ satisfies $g_{\mu\nu}T^\mu T^\nu<0$ everywhere (signature $-\,+\,+\,+$); future-directed means $T^\mu$ lies in the chosen future cone, which requires a global time orientation. Reading: $I^+(p)$ is the interior of the future — "strictly later, strictly reachable". The causal future relaxes the requirement to causal curves, whose tangent may be null as well as timelike. By convention $p$ itself is included, so that $J^+$ is reflexive and the operator composes cleanly.

where: a causal curve has tangent $T^\mu\neq 0$ with $g_{\mu\nu}T^\mu T^\nu\le 0$, future-directed; the singleton $\{p\}$ is adjoined so that $p\in J^+(p)$. Reading: $J^+(p)$ adds the null-reachable rim to the timelike interior. The pasts are not new objects but the time-reverses of the futures, obtained by flipping the time orientation. This duality halves every proof.

where: $I^-,J^-$ — chronological and causal pasts. Reading: $q$ is in the past of $p$ exactly when $p$ is in the future of $q$ — causal precedence read backwards. The structural relations are where the topology lives. The timelike interior sits inside the causal set, the interior is open, and although $J^+(p)$ need not be closed, it shares its closure and boundary with $I^+(p)$.

where: $\overline{\,\cdot\,}$ — topological closure in $M$; $\dot{\,\cdot\,}$ — topological boundary, $\dot A=\overline A\setminus\operatorname{int}A$. Reading: the two futures differ at most by part of their common null boundary — the interiors are identical, and $J^+$ can only add boundary points that $I^+$ omits. The definitions extend verbatim from single points to arbitrary sets by union, and the openness of $I^+$ survives even when the seed set is closed — a fact used constantly in domain-of-dependence arguments.

where: $S\subset M$ — any subset. Reading: $I^+(S)$ is open (a union of open sets) regardless of $S$; $J^+(S)$ inherits the same subtleties as $J^+(p)$. The operators satisfy a small algebra that simply restates the transitivity of "can influence". Chaining causal or chronological reachability collapses back to a single step, and — crucially — a timelike leg anywhere in the chain forces the strict (open) result.

where: the composition is "apply one future operator, then another". Reading: the mixed identity $J^+\!\circ I^+=I^+$ says timelike-then-causal (or causal-then-timelike) still lands strictly inside the cone — the timelike leg dominates. The boundary of the causal future is the global light cone. Away from the apex $p$ it is ruled by null geodesic segments, the generators, each of which either starts at $p$ or is past-inextendible within the boundary.

where: a generator is a null geodesic ruling the boundary hypersurface. Reading: generators may leave the boundary (into the interior) at caustics but can never re-enter — the light cone can crease inward but not outward. Everything so far uses only causal characters, which are conformally invariant. Hence a positive rescaling of the metric leaves all four sets untouched — the theorem that legitimizes Penrose diagrams.

where: $\Omega:M\to(0,\infty)$ — a smooth conformal factor. Reading: causal structure sees only the conformal class $[g]$, never the scale. Finally, the well-behaved regime is fixed by a compactness condition on causal diamonds. When it holds the spacetime is globally hyperbolic, $J^\pm$ are closed, and maximal geodesics between events exist.

where: $J^+(p)\cap J^-(q)$ — the causal diamond of events after $p$ and before $q$. Reading: compact diamonds guarantee predictability, Cauchy surfaces, and the existence of longest causal curves.

The four sets specialize sharply across the canonical spacetimes; the table reads from a defining relation to its physical content.

1 · Classifying events in Minkowski space (fully worked, with numbers)

Take $p$ at the origin of Minkowski space with signature $(-,+,+,+)$ and $c=1$. Classify the events $q=(3,1,0,0)$, $r=(2,2,0,0)$, and $w=(1,2,0,0)$ relative to $p$, writing the coordinates as $(t,x,y,z)$.

Step 1. The relevant invariant is the squared interval $s^2=\eta_{\mu\nu}(x-p)^\mu(x-p)^\nu=-t^2+x^2+y^2+z^2$; the sign classifies the separation and the sign of $t$ selects future versus past. This one number does all the work — no curve-hunting required in flat space, where straight lines are the extremal causal curves.

Step 2. For $q=(3,1,0,0)$: $s^2=-(3)^2+(1)^2=-9+1=\mathbf{-8}<0$, timelike, and $t=3>0$, so $q\in I^+(p)$. The proper time along the straight worldline is $\tau=\sqrt{-s^2}=\sqrt{8}=\mathbf{2.83}$. A massive observer can be present at both $p$ and $q$.

Step 3. For $r=(2,2,0,0)$: $s^2=-(2)^2+(2)^2=-4+4=\mathbf{0}$, null, with $t>0$, so $r\in J^+(p)\setminus I^+(p)$ — it sits exactly on the future light cone $\dot J^+(p)$, reachable only by a light ray.

Step 4. For $w=(1,2,0,0)$: $s^2=-(1)^2+(2)^2=-1+4=\mathbf{+3}>0$, spacelike, so $w\notin J^+(p)$ and likewise $w\notin J^-(p)$ — $w$ is causally disconnected from $p$, neither able to influence it nor be influenced by it. Result: $q$ timelike-future ($I^+$), $r$ on the null boundary ($J^+\setminus I^+$), $w$ spacelike (causally disjoint).

2 · The shadow of a deleted point (why $J^+$ is not closed)

In Minkowski space delete the single event $a=(1,1,0,0)$, leaving $M'=\mathbb{R}^{1,3}\setminus\{a\}$. Put $p$ at the origin and examine $q=(2,2,0,0)$. Show that $q\in\overline{I^+(p)}$ but $q\notin J^+(p)$ in $M'$.

Step 1. In full Minkowski, $p$, $a$, and $q$ are collinear on the null ray $(t,t,0,0)$: $\eta$-norms of $a-p$ and $q-a$ both vanish, so $a\in\dot J^+(p)$ and $q\in\dot J^+(p)$, connected to $p$ only by the unique null geodesic through $a$. Null-separated points in Minkowski admit no timelike curve and no other causal curve between them — the generator is rigid.

Step 2. Delete $a$. The only future-directed causal curve from $p$ to $q$ was that generator, and it passed through the now-missing $a$; there is no detour, because any timelike deviation would land strictly inside the cone and could never climb back onto the null boundary to reach $q$. Hence $q\notin J^+(p)$ in $M'$.

Step 3. Yet $q$ is still a limit of chronological-future points: the events $(2,\,2-\varepsilon,0,0)$ have $s^2=-4+(2-\varepsilon)^2=-4\varepsilon+\varepsilon^2<0$ for small $\varepsilon>0$, so they lie in $I^+(p)$ and avoid $a$, and they converge to $q$ as $\varepsilon\to 0$. Therefore $q\in\overline{I^+(p)}=\overline{J^+(p)}$. Result: $q\in\overline{I^+(p)}\setminus J^+(p)$, so $J^+(p)$ is not closed and $J^+(p)\subsetneq\overline{I^+(p)}$ — the canonical counterexample.

3 · Conformal invariance on explicit vectors

Let $\tilde g=\Omega^2\eta$ with constant $\Omega=2$ (a stand-in for a conformal factor such as the FLRW scale factor). Check that the causal character of the vectors $T=(2,1,0,0)$ and $N=(1,1,0,0)$ at a point is unchanged.

Step 1. Under $\eta$ with signature $(-,+,+,+)$: for $T$, $\eta(T,T)=-(2)^2+(1)^2=\mathbf{-3}<0$ (timelike); for $N$, $\eta(N,N)=-(1)^2+(1)^2=\mathbf{0}$ (null). These are the characters that fix which curves count as causal.

Step 2. Apply the scaling identity $\tilde g(V,V)=\Omega^2\,\eta(V,V)$ with $\Omega^2=4$: $\tilde g(T,T)=4\times(-3)=\mathbf{-12}<0$ and $\tilde g(N,N)=4\times 0=\mathbf{0}$. The signs are identical — timelike stays timelike, null stays null.

Step 3. Because every tangent vector along every curve behaves this way, the set of timelike and causal curves — and hence $I^\pm$ and $J^\pm$ — is untouched by the rescaling, even though proper times and lengths are stretched by $\Omega$. Result: $\tilde g(T,T)=-12$, $\tilde g(N,N)=0$; causal structure is conformally invariant, exactly the property Penrose diagrams exploit.

The theory of causal futures and pasts is, more than almost any other corner of relativity, the creation of one mind wielding topology where everyone else reached for coordinates. Before the 1960s "causal structure" meant the Minkowski light cone; after Penrose it meant a rigorous global calculus of achronal boundaries, conformal infinity, and ideal points.

Texts. Hawking & Ellis, The Large Scale Structure of Space-Time (1973), §6.2–6.4 · Wald, General Relativity (1984), Chapter 8 · Penrose, Techniques of Differential Topology in Relativity (1972), §2–3 · Misner, Thorne & Wheeler, Gravitation (1973), §34 · Carroll, Spacetime and Geometry (2004), Appendix on causal structure · O'Neill, Semi-Riemannian Geometry (1983), Chapter 14 · Geroch, Kronheimer & Penrose, "Ideal points in space-time", Proc. R. Soc. A 327, 545 (1972). The openness of $I^\pm$ and the identity $\overline{I^\pm}=\overline{J^\pm}$ are established in Derivation A. The boundary theorem (7) — achronality and null-geodesic generation of $\dot J^+(p)$ — is reproduced in Derivation B. Conformal invariance (8) follows from the pointwise scaling identity $\tilde g_{\mu\nu}=\Omega^2 g_{\mu\nu}$, worked out in Derivation C. The Minkowski checks use signature $(-,+,+,+)$ with $c=1$: $q=(3,1,0,0)\Rightarrow s^2=-8$ (timelike, $\tau=\sqrt8$); $r=(2,2,0,0)\Rightarrow s^2=0$ (null); $w=(1,2,0,0)\Rightarrow s^2=+3$ (spacelike). The deleted-point shadow and the $\Omega=2$ conformal check are self-verifying.

Prerequisites. Causal Structure of Spacetime · Null Geodesics and Light Cones · Conformal Transformations and Causal Invariance. Leads to. Achronal Sets and Edges · Domain of Dependence and Cauchy Horizons · Global Hyperbolicity · Penrose–Carter Conformal Diagrams · Causality Conditions Hierarchy.

Ask the tutor: Ask me to walk through why $\dot J^+(p)$ must be null geodesic by geodesic, or to build the deleted-point shadow region explicitly so you can see $J^+$ fail to be closed.