Penrose–Minkowski Conformal Diagrams

An ordinary spacetime diagram is honest about causality and hopeless about infinity. Light cones sit at $45^{\circ}$ , worldlines are easy to read — and then every interesting question ("where does that photon end up?", "what can this observer ever see?") walks straight off the edge of the page. Penrose's answer, in 1963, was to change the lens. Multiply the metric by a smooth positive function $\Omega^2$ that vanishes at infinity; the rescaled geometry is finite, and the original is recovered by dividing back out. Because $\Omega^2 > 0$ multiplies the whole quadratic form, it cancels out of the one condition that matters — $\mathrm{d}s^2 = 0$ . Light cones survive the rescaling exactly. Distances do not, and are gone forever. That trade is the entire idea: throw away every metric fact, keep every causal fact, and infinity comes back as an edge you can point at and label. Do it to Minkowski spacetime and you get a triangle whose three corners and two edges are five physically distinct destinations — and which one a curve reaches depends only on whether it is timelike, null or spacelike. Every massive particle, at $\beta = 0.5$ or $\beta = 0.999$ alike, ends at the same corner $i^+$ ; every photon ends on the edge $\mathscr{I}^+$ ; every spacelike slice runs out to $i^0$ . Infinity was never one place. This article stays in flat spacetime and in the locked signature $(+,-,-,-)$ — horizons, singularities and cosmology are general relativity, forward-linked and not derived — and proves the theorem underneath all of them: the compactified Minkowski diagram is a triangular patch of the Einstein static universe $\mathbb{R}\times S^3$ .

Key takeaways

A conformal transformation rescales the metric, $\tilde{g}_{\mu\nu} = \Omega^2 g_{\mu\nu}$ with $\Omega > 0$ . Because a positive factor cannot change whether a norm is zero, null cones are exactly preserved while lengths, proper times and areas are destroyed. Choosing $\Omega \to 0$ at infinity pulls the infinitely distant regions to a finite boundary, and every light ray still runs at exactly $\pm45^{\circ}$ .
The recipe in the signature $(+,-,-,-)$ : null coordinates $u = ct - r$ , $v = ct + r$ give $\mathrm{d}s^2 = \mathrm{d}u\,\mathrm{d}v - r^2\mathrm{d}\Omega^2$ (the base has $-\mathrm{d}u\,\mathrm{d}v$ — it is in the other signature); then $U = \arctan(u/c\ell)$ , $V = \arctan(v/c\ell)$ , $\tilde{T} = V + U$ , $\tilde{R} = V - U$ .
Infinity resolves into five pieces: timelike $i^{\pm}$ (where massive worldlines end), null $\mathscr{I}^{\pm}$ ("scri", where light ends), spatial $i^0$ (where spacelike slices end). Confusing them is the classic error.
The owned theorem. Carrying the angular part through — which the base never does — gives $\Omega^2\mathrm{d}s^2 = \mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2 - \sin^2\!\tilde{R}\,\mathrm{d}\Omega^2$ with $\Omega = (\cos\tilde{T} + \cos\tilde{R})/(c\ell)$ : the Einstein static universe $\mathbb{R}\times S^3$ , of which all of Minkowski spacetime is the triangular patch $|\tilde{T}| + \tilde{R} < \pi$ .
A boost is a flow: $u \to e^{+\varphi}u$ , $v \to e^{-\varphi}v$ (the base's boxed line has this backwards). Its orbits are the hyperbolae $uv = s^2$ , and $i^+, i^-, i^0$ are fixed points — which is why causal structure is frame-independent.
Key idea: a positive rescaling cannot move a light ray. So you may buy a finite picture of an infinite spacetime with a currency — distance — that causality does not spend.

The core insight

Every spacetime diagram carries two kinds of information: causal (what can influence what) and metric (how far, how long). They are not equally robust. Multiply the metric by any positive function and the metric information is scrambled beyond recovery — but the null cones, and therefore the whole causal order, come through untouched. The Penrose diagram spends all of the second to buy a bounded picture and keeps all of the first. Infinity stops being an excuse and becomes a boundary with five named parts.

Where this lives today

The flat-space triangle is a warm-up. The same recipe applied to Schwarzschild produces the diagram in which an event horizon is a $45^{\circ}$ line you can cross but not recross, and the singularity is a spacelike line in your future — which is why "the singularity is a moment, not a place" is a sentence a picture makes obvious and a coordinate calculation does not. Kerr, Reissner–Nordström and de Sitter each get their own, usually called Carter–Penrose diagrams after Carter's 1966 extension of the Kerr axis. All of that is general relativity and is forward-linked only. The deeper legacy is $\mathscr{I}$ itself. Because the construction makes null infinity a genuine hypersurface, gravitational radiation can be defined there rather than estimated in a gauge — numerical-relativity codes extrapolate waveforms to $\mathscr{I}^+$ before handing them to LIGO's template banks. The BMS group, the asymptotic symmetries of $\mathscr{I}$ , is now a research front of its own, tied to soft theorems and gravitational memory; in holography the same conformal boundary is where the dual field theory lives. And the flat-space version has the payoff this article made precise: Minkowski causal structure is a bounded, drawable object whose five pieces of infinity are the fixed points and edges of the Lorentz group's action.

The physical picture

Photograph an infinite straight railway. The rails, which never meet, converge to a point on the horizon, and everything infinitely far away is compressed onto the frame's edge. A wide-angle lens does not lie about which sleeper comes before which — it lies about how far apart they are. That is the bargain here, plus one piece of magic: the lens is chosen so that light rays stay at $45^{\circ}$ everywhere, right up to the edge. The lens is a conformal factor $\Omega$ , and it works because a positive number cannot change the sign of a quantity, nor whether it is zero — so a vector is null under $\tilde{g} = \Omega^2 g$ if and only if it is null under $g$ . The light cone at every point is the same object in both geometries. Lengths are another matter: if $\Omega$ blows up somewhere, a curve of finite $\tilde{g}$ -length can have infinite $g$ -length. That is not a bug; it is the mechanism. So pick $\Omega$ to vanish precisely where physical distance diverges, and infinity — sitting at $\Omega = 0$ — lands at a finite place in the $\tilde{g}$ picture, where it can be drawn. Do this to Minkowski spacetime (spherically reduced: keep $t$ and $r \geq 0$ , suppress the two angles) and you get a triangle. Its boundary is not one thing, and a curve's causal character alone decides which piece it reaches: a timelike worldline ends at the single corner $i^+$ , a null ray on the edge $\mathscr{I}^+$ ("scri-plus"), a spacelike slice at the far corner $i^0$ . Massive and massless things never end up in the same place, however fast the massive thing goes. And the left edge, $\tilde{R} = 0$ , is not a boundary at all — it is the regular symmetry axis $r = 0$ , where the polar coordinates simply run out.

Figure

The conformal diagram of Minkowski spacetime, in the chapter's signature (+,−,−,−). The infinite (t, r) half-plane is squeezed into a finite triangle. Light rays run at exactly ±45° everywhere, right up to the edges. The two diagonal edges are null infinity 𝓘⁺ (T̃ + R̃ = 180°) and 𝓘⁻ (R̃ − T̃ = 180°); the three corners are future timelike infinity i⁺ (180°, 0°), past timelike infinity i⁻ (−180°, 0°) and spatial infinity i⁰ (0°, 180°). Which piece a curve reaches is decided by its causal character alone: EVERY timelike worldline ends at i⁺ no matter how fast, EVERY light ray ends on 𝓘⁺, EVERY spacelike slice runs out to i⁰. The left edge R̃ = 0 is the regular axis r = 0, not a boundary

The core equations

Locked signature $(+,-,-,-)$ , timelike-positive. The base is written in $(-,+,+,+)$ , so every sign below is opposite to the one printed there. Nothing has been transcribed.

\mathrm{d}s^2 = c^2\,\mathrm{d}t^2 - \mathrm{d}r^2 - r^2\,\mathrm{d}\Omega^2, \qquad \mathrm{d}\Omega^2 = \mathrm{d}\theta^2 + \sin^2\!\theta\,\mathrm{d}\phi^2 \qquad (1)

where: $r \geq 0$ ; $\mathrm{d}\Omega^2$ — the round metric on the unit 2-sphere. Physical reading: flat spacetime in spherical coordinates, owned by The Minkowski Metric; the cone and the causal trichotomy are owned by The Light Cone and Causal Structure of Spacetime. Timelike $\mathrm{d}s^2 > 0$ . The base prints $-c^2\mathrm{d}t^2 + \mathrm{d}r^2 + \ldots$ — correct in its convention, wrong in ours.

u = ct - r, \qquad v = ct + r \qquad\Longrightarrow\qquad \mathrm{d}s^2 = \mathrm{d}u\,\mathrm{d}v - r^2\,\mathrm{d}\Omega^2, \qquad r = \tfrac12(v-u) \qquad (2)

where: $u$ — retarded time, constant along outgoing light rays; $v$ — advanced time, constant along incoming ones. Physical reading: $\mathrm{d}u\,\mathrm{d}v = (c\,\mathrm{d}t - \mathrm{d}r)(c\,\mathrm{d}t + \mathrm{d}r) = c^2\mathrm{d}t^2 - \mathrm{d}r^2$ — the sign the base flips. Radial null rays are exactly $\mathrm{d}u = 0$ or $\mathrm{d}v = 0$ : the light cone has become the coordinate grid. And for radial motion $s^2 = uv$ — positive for timelike, here.

U = \arctan\frac{u}{c\ell}, \qquad V = \arctan\frac{v}{c\ell}, \qquad \tilde{T} = V + U, \qquad \tilde{R} = V - U \qquad (3)

where: $\ell$ — any positive length scale, present only to make the argument dimensionless; $U, V \in (-\tfrac{\pi}{2}, \tfrac{\pi}{2})$ . Physical reading: $\arctan$ is the lens — it maps each infinite null line onto a finite open interval and, being monotone, reorders nothing. The causal condition $v \geq u$ becomes $\tilde{R} \geq 0$ , and the region is the triangle $|\tilde{T}| + \tilde{R} < \pi$ . Lines of constant $U$ or $V$ stay straight at $\pm45^{\circ}$ .

\boxed{\;\Omega^2\,\mathrm{d}s^2 \;=\; \mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2 - \sin^2\!\tilde{R}\;\mathrm{d}\Omega^2, \qquad \Omega \;=\; \frac{2\cos U\,\cos V}{c\ell} \;=\; \frac{\cos\tilde{T} + \cos\tilde{R}}{c\ell}\;} \qquad (4)

where: $\Omega$ — the conformal factor, strictly positive inside and vanishing on every boundary piece. Physical reading: the article's owned result, and the line the base never writes. The right-hand side is the metric of the Einstein static universe $\mathbb{R}\times S^3$ , with $\tilde{R}\in[0,\pi]$ the polar angle on the 3-sphere. Minkowski spacetime is not merely like a triangle — it is conformal to the patch $|\tilde{T}| + \tilde{R} < \pi$ of that cylinder. The inverse map is explicit: $ct = \sin\tilde{T}/\Omega$ , $r = \sin\tilde{R}/\Omega$ .

i^{\pm}: \tilde{T} = \pm\pi,\ \tilde{R} = 0; \qquad i^{0}: \tilde{T} = 0,\ \tilde{R} = \pi; \qquad \mathscr{I}^{+}: \tilde{T} + \tilde{R} = \pi; \qquad \mathscr{I}^{-}: \tilde{R} - \tilde{T} = \pi \qquad (5)

where: $i^{\pm}$ — timelike infinity, single points; $i^0$ — spatial infinity, a point; $\mathscr{I}^{\pm}$ — null infinity, one-dimensional edges. Physical reading: $\Omega = 0$ on all five, which is precisely the statement that they sit at infinite physical distance.

Deriving the structure

Derivation AWhy a positive rescaling cannot move a light ray

This is the one fact that licenses the entire construction, and it is three lines long. If it failed, a conformal diagram would be a pretty picture of nothing. It does not fail — and the only hypothesis it needs is that $\Omega^2$ is positive, which is far weaker than it sounds and buys far more than it should.

Step 1. Let $\tilde{g}_{\mu\nu} = \Omega^2 g_{\mu\nu}$ with $\Omega(x) > 0$ smooth. For any vector $k^{\mu}$ , $\tilde{g}_{\mu\nu}k^{\mu}k^{\nu} = \Omega^2\, g_{\mu\nu}k^{\mu}k^{\nu}$ — a strictly positive multiplier cannot change the sign of a real quantity, and in particular cannot change whether it is zero: g_{\mu\nu}k^{\mu}k^{\nu} = 0 \iff \tilde{g}_{\mu\nu}k^{\mu}k^{\nu} = 0. In our signature $(+,-,-,-)$ : timelike ( $>0$ ) stays timelike, spacelike ( $<0$ ) stays spacelike, null ( $=0$ ) stays null. Why: the causal character of every vector at every point is preserved, with no assumption about what $g$ is. The base states this with the inequalities reversed, being in the other signature — and this is the one step whose conclusion does not care.

Step 2. Harvest. The light cone at each point is the same set of directions for $g$ and $\tilde{g}$ , so "event $A$ can signal event $B$ " is identical in the two geometries — the entire causal order survives (Causal Structure of Spacetime owns that order; here it is carried across). But $\Omega$ is free to blow up: a curve of finite $\tilde{g}$ -length can have infinite $g$ -length. Lengths and proper times are not preserved, and the failure is unbounded. Why: this is the trade. You may bring infinity to a finite boundary precisely because the currency you spend — distance — is not the currency causality is denominated in. Everything below is the construction of one particular $\Omega$ . *Verified in verify_aq-sr-conformal-diagrams.py: for thousands of random $\Omega > 0$ and random $k^{\mu}$ , the causal character (sign of the norm) is invariant — checked in both signatures, because this step, uniquely, does not care.*

The owned result: the Penrose diagram is a patch of the Einstein static universe

The base carries the compactification as far as the two-dimensional $(t,r)$ part and stops — one line short of the theorem. With $r^2\mathrm{d}\Omega^2$ left outside the conformal factor, nothing has actually been compactified: $r \to \infty$ on the boundary and the metric is still singular there. Carry it through and the answer is not a shapeless finite blob; it is one of the most familiar spacetimes in physics.

Go deeper · Derivation B · 2 minCompactifying Minkowski spacetime, angular part and all, and recognising what comes out

Derivation BCompactifying Minkowski spacetime, angular part and all, and recognising what comes out

Take the flat metric in our signature, run it through null coordinates and an arctangent, and — the step the base omits — carry the two-sphere along too. The conformal factor turns out to rescale $r$ into $\sin\tilde{R}$ , which is exactly what a 3-sphere's polar coordinate does, and the rescaled metric is the Einstein static universe.

Step 1. Null coordinates. With $u = ct - r$ , $v = ct + r$ : $\mathrm{d}u\,\mathrm{d}v = (c\,\mathrm{d}t - \mathrm{d}r)(c\,\mathrm{d}t + \mathrm{d}r) = c^2\mathrm{d}t^2 - \mathrm{d}r^2$ , so $\mathrm{d}s^2 = \mathrm{d}u\,\mathrm{d}v - r^2\mathrm{d}\Omega^2$ with $r = \tfrac12(v-u)$ . Why: in $(+,-,-,-)$ the cross term comes with a PLUS; the base prints $-\mathrm{d}u\,\mathrm{d}v$ , right in mostly-plus and wrong here. Bonus: for radial motion $s^2 = uv$ directly — the null coordinates already factorise the interval.

Step 2. Compactify and rotate. Put $u = c\ell\tan U$ , $v = c\ell\tan V$ , so $\mathrm{d}u = (c\ell/\cos^2U)\mathrm{d}U$ and likewise for $v$ ; with $\tilde{T} = V+U$ , $\tilde{R} = V-U$ , $\mathrm{d}U\,\mathrm{d}V = \tfrac14(\mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2)$ . Hence \mathrm{d}u\,\mathrm{d}v = \frac{c^2\ell^2}{4\cos^2U\cos^2V}\big(\mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2\big) \equiv \Omega^{-2}\big(\mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2\big), \qquad \Omega \equiv \frac{2\cos U\cos V}{c\ell}, and product-to-sum gives the closed form $2\cos U\cos V = \cos\tilde{R} + \cos\tilde{T}$ . Why: $\arctan$ is monotone, so no event overtakes another and the causal order is untouched; $\Omega$ was read off, not chosen.

Step 3. The step the base skips: the sphere. r = \tfrac{c\ell}{2}(\tan V - \tan U) = \tfrac{c\ell}{2}\,\frac{\sin(V-U)}{\cos U\cos V} = \frac{\sin\tilde{R}}{\Omega} \qquad\Longrightarrow\qquad r^2\mathrm{d}\Omega^2 = \Omega^{-2}\sin^2\!\tilde{R}\;\mathrm{d}\Omega^2. Why: the load-bearing line. The angular part carries exactly the same $\Omega^{-2}$ as the radial-temporal part. Without it, $r^2\mathrm{d}\Omega^2$ sits outside the conformal factor, diverges on the boundary, and the "compactified" metric is not compact at all.

Step 4. Collect, and recognise. \boxed{\;\Omega^2\mathrm{d}s^2 = \mathrm{d}\tilde{T}^2 - \big(\mathrm{d}\tilde{R}^2 + \sin^2\!\tilde{R}\,\mathrm{d}\Omega^2\big), \qquad \Omega = \frac{\cos\tilde{T} + \cos\tilde{R}}{c\ell}\;} The bracket is the round metric on a unit 3-sphere in polar coordinates $(\tilde{R}, \theta, \phi)$ , $\tilde{R}\in[0,\pi]$ . So the rescaled spacetime is the Einstein static universe $\mathbb{R}\times S^3$ , and Minkowski spacetime is conformal to its patch $|\tilde{T}| + \tilde{R} < \pi$ . Why: an infinite, open spacetime turns out to be a triangular window cut into a compact, spatially closed one, whose edges are the five pieces of infinity. This is Penrose's picture, and why the recipe transplants to curved spacetime unchanged.

Step 5. Prove which curves reach which piece — read off $u$ and $v$ separately, using that $\arctan$ is monotone with limits $\pm\pi/2$ . An inertial worldline $x = \beta ct$ has $u = ct(1-\beta)$ , $v = ct(1+\beta)$ : for $|\beta| < 1$ both run to $+\infty$ , so $U, V \to \pi/2$ and $(\tilde{T},\tilde{R}) \to (\pi, 0) = i^+$ — for every $\beta$ . A photon has $u$ constant, so $U$ stays finite while $V \to \pi/2$ : it lands on $\tilde{T} + \tilde{R} = \pi$ , which is $\mathscr{I}^+$ . A slice $t = 0$ , $r\to\infty$ has $u\to-\infty$ , $v\to+\infty$ : $(0,\pi) = i^0$ . Why: massive and massless things do not share a destination, and the difference between "both null coordinates run away" and "only one does" is exactly $|\beta| < 1$ versus $|\beta| = 1$ . Hence infinity has five pieces, not six. *Verified in verify_aq-sr-conformal-diagrams.py: Step 4 is checked symbolically by pulling $\mathrm{d}s^2$ back through $t(\tilde{T},\tilde{R})$ , $r(\tilde{T},\tilde{R})$ and confirming $\Omega^2 g = \mathrm{diag}(1, -1, -\sin^2\tilde{R}, -\sin^2\tilde{R}\sin^2\theta)$ component by component; $2\cos U\cos V = \cos\tilde{T} + \cos\tilde{R}$ , $ct = \sin\tilde{T}/\Omega$ , $r = \sin\tilde{R}/\Omega$ symbolically; and Step 5's three limits numerically, including $\beta = 0.999$ landing within $0.001^{\circ}$ of $i^+$ .*

Go deeper · Derivation C · 2 minA boost is a flow on the diagram, and the base has its direction backwards

Derivation CA boost is a flow on the diagram, and the base has its direction backwards

In null coordinates a Lorentz boost stops being a matrix and becomes a pair of independent rescalings — one direction stretched, the other squeezed by the reciprocal factor. That makes its action on the compactified picture explicit, and the frame-independence of causal structure visible rather than merely true. The base boxes this result with the two signs interchanged; here it is derived.

Step 1. Boost, in the chapter's locked convention $x' = \gamma(x - vt)$ — rapidity $\varphi$ , $\beta = \tanh\varphi$ : $ct' = ct\cosh\varphi - x\sinh\varphi$ , $x' = x\cosh\varphi - ct\sinh\varphi$ , read on the equatorial $(t,x)$ plane. Take the null combinations, using $\cosh\varphi \pm \sinh\varphi = e^{\pm\varphi}$ : u' = ct'-x' = (ct-x)(\cosh\varphi + \sinh\varphi) = e^{+\varphi}u, \qquad v' = ct'+x' = (ct+x)(\cosh\varphi - \sinh\varphi) = e^{-\varphi}v. \boxed{\;u \to e^{+\varphi}\,u, \qquad v \to e^{-\varphi}\,v\;} Why: the base's boxed line says $u \to e^{-\varphi}u$ , $v \to e^{+\varphi}v$ — backwards — while its own intermediate step derives $ct'-x' = e^{+\varphi}(ct-x)$ , which is what we just got. The box contradicts the steps. The retarded coordinate is stretched, the advanced one squeezed, and the factor is the Doppler factor $k = e^{\varphi} = \sqrt{(1+\beta)/(1-\beta)}$ — at $\beta = 0.6$ , $k = 2$ exactly. The $\Lambda$ is owned by The Lorentz Transformation Equations.

Step 2. The orbits and the fixed points. The product survives, $u'v' = e^{+\varphi}e^{-\varphi}uv = uv$ , so — since $s^2 = uv$ for radial motion — the orbits of the flow are exactly the invariant hyperbolae $uv = \mathrm{const}$ (owned by Invariant Hyperbolae; here they become the streamlines of a flow on a bounded page). On the diagram $\tan U \to e^{+\varphi}\tan U$ and $\tan V \to e^{-\varphi}\tan V$ , so a point is fixed iff $\tan U$ and $\tan V$ are each $0$ or $\pm\infty$ — i.e. $U, V \in \{0, \pm\pi/2\}$ : the origin, $i^+$ , $i^-$ and $i^0$ , and the two points where the origin's light cone meets $\mathscr{I}^{\pm}$ . Everything else slides, and $\mathscr{I}^{\pm}$ map to themselves as sets while their points move along them. Why: the boost is a squeeze map — area-preserving in the $(u,v)$ plane, sliding events ALONG the hyperbolae and never across a $45^{\circ}$ line. This is the geometric content of "Lorentz transformations preserve causal structure": the three infinities are literally fixed points of the group action, and "causal order is frame-independent" has become a picture rather than an inequality. *Verified in verify_aq-sr-conformal-diagrams.py: the flow is derived symbolically from the boost matrix and asserted to be $u \to e^{+\varphi}u$ , $v \to e^{-\varphi}v$ , with an explicit assertion that the base's boxed alternative is false; $uv$ is checked invariant; $k = e^{\varphi} = 2$ at $\beta = 0.6$ ; the fixed-point set is enumerated numerically; and the worked example's chain $(1, 0.8) \to (0.65, 0.25)$ is confirmed against the Lorentz matrix independently.*

Figure

PUNCHLINE — a boost is a squeeze, and the diagram shows exactly what it costs. The event E = (ct, x) = (1, 0.8)ℓ has u = 0.2, v = 1.8, so s² = uv = 0.36ℓ². Boost to β = 0.6: the Doppler factor is k = e^φ = 2 exactly, so u is STRETCHED to 0.4 and v is SQUEEZED to 0.9 — giving E′ = (0.65, 0.25)ℓ, and u′v′ = 0.36 unchanged. On the diagram E sits at (T̃, R̃) = (72.255°, 49.635°) with conformal factor Ω = 0.9524; E′ sits at (63.789°, 20.186°) with Ω = 1.3803. The INTERVAL is invariant and the CONFORMAL FACTOR is not — Ω′∕Ω = 1.4492 — which is the trade, in one number: causality survives, distance does not

Limiting cases

Case	Result	Meaning
$\Omega \to 0$	the boundary	the factor vanishes exactly where physical distance diverges — that is the compactification
$\varphi \to \infty$ ( $\beta \to 1$ )	$u \to \infty$ , $v \to 0$	the squeeze degenerates; an event is driven onto the light cone but never through it
$\|\beta\| < 1$	endpoint $i^{+}$	every massive worldline, at any speed — $\beta = 0.999$ lands within $0.001^{\circ}$ of it
$\|\beta\| = 1$	endpoint on $\mathscr{I}^{+}$	one null coordinate stays finite, so light lands on the edge, not the corner
$\ell \to \lambda\ell$	same diagram	$\ell$ is pure gauge: it reparametrises $U, V$ and changes no causal statement

Common trapThe diagram destroys distance — and the r = 0 axis is not a boundary

Never measure anything on a Penrose diagram. $\Omega$ vanishes on the boundary, so $\Omega^{-2}$ blows up there and a ruler laid on the page measures nothing physical. The worked example makes it concrete — boosting $E$ to $E'$ leaves the interval at exactly $s^2 = 0.36\,\ell^2$ while $\Omega$ moves from $0.9524$ to $1.3803$ , a factor of $1.4492$ . The interval is invariant; the page-position is not.

The left edge is not an edge. $\tilde{R} = 0$ is the regular symmetry axis $r = 0$ , where spherical polars run out — as the north pole of a globe is not the end of the Earth. Worldlines cross it, and a boost will push events through it: $(ct, x) = (1, 0.5)\ell$ boosted to $\beta = 0.6$ lands at $x' = -0.125\ell$ , out the other side. A boost along $\hat{x}$ is not spherically symmetric, so it is not a symmetry of the spherically-reduced chart at all; the triangle is the right half of the full $1{+}1$ picture, and a boost is honestly read only there.

And the base's Weyl-tensor card is out of band. $C^{\mu}{}_{\nu\rho\sigma}[\tilde{g}] = C^{\mu}{}_{\nu\rho\sigma}[g]$ is general relativity — in flat spacetime both sides vanish and it says nothing. Named, not used, as are horizons, singularities and cosmology. This article stays in flat spacetime.

Worked examples

1 · Put one event on the diagram, then boost it

Take the event $E$ at $(ct, x) = (1,\ 0.8)\,\ell$ (working in units $c = \ell = 1$ ). Locate it, then boost to $\beta = 0.6$ and locate it again.

Step 1 — null coordinates. $u = ct - x = \mathbf{0.2}$ , $v = ct + x = \mathbf{1.8}$ . Both positive, so the event lies in the origin's future light cone, and $s^2 = uv = \mathbf{0.36}\,\ell^2 > 0$ — timelike in this signature, with proper time $c\tau = \sqrt{s^2} = \mathbf{0.6}\,\ell$ .

Step 2 — compactify. $U = \arctan 0.2 = \mathbf{11.310^{\circ}}$ , $V = \arctan 1.8 = \mathbf{60.945^{\circ}}$ , so $\tilde{T} = \mathbf{72.255^{\circ}}$ , $\tilde{R} = \mathbf{49.635^{\circ}}$ , and $|\tilde{T}| + \tilde{R} = 121.891^{\circ} < 180^{\circ}$ ✓. The conformal factor is $\Omega = 2\cos U\cos V = \mathbf{0.9524}$ , and equally $\cos\tilde{T} + \cos\tilde{R} = 0.3048 + 0.6476 = 0.9524$ ✓. The inverse map returns the event: $\sin\tilde{T}/\Omega = \mathbf{1.000}$ , $\sin\tilde{R}/\Omega = 0.7619/0.9524 = \mathbf{0.800}$ ✓.

Step 3 — boost. The Doppler factor is $k = e^{\varphi} = \sqrt{(1+\beta)/(1-\beta)} = \sqrt{1.6/0.4} = \mathbf{2}$ exactly ( $\varphi = \ln 2 = 0.6931$ ). So $u' = ku = \mathbf{0.4}$ , $v' = v/k = \mathbf{0.9}$ , giving $ct' = \tfrac12(u'+v') = \mathbf{0.65}$ and $x' = \tfrac12(v'-u') = \mathbf{0.25}$ . Cross-check against the Lorentz matrix, $\gamma = 1.25$ : $1.25(1 - 0.48) = 0.65$ ✓, $1.25(0.8 - 0.6) = 0.25$ ✓. The interval: $u'v' = 0.36$ , and independently $0.65^2 - 0.25^2 = \mathbf{0.36}$ ✓.

Step 4 — the cost. $E'$ sits at $(\tilde{T}', \tilde{R}') = (\mathbf{63.789^{\circ}},\ \mathbf{20.186^{\circ}})$ with $\Omega' = \mathbf{1.3803}$ . The interval did not move; the page-position and the conformal factor did — $\Omega'/\Omega = \mathbf{1.4492}$ . That ratio is the trade: causality survived, distance did not.

Step 5 — three curves, three destinations. A rocket at $\beta = 0.999$ has $u = 0.001\,ct$ , $v = 1.999\,ct$ : both run to $+\infty$ , so it lands at $i^+$ — at $ct = 10^{8}\ell$ it already sits at $(\mathbf{179.9994^{\circ}},\ \mathbf{0.0006^{\circ}})$ , the same corner as a rocket at rest. A photon has $u = 0$ always, so $U = 0^{\circ}$ while $V \to 90^{\circ}$ : it runs up $\tilde{T} = \tilde{R}$ at $45^{\circ}$ and stops at $(90^{\circ}, 90^{\circ})$ on the edge $\mathscr{I}^+$ (by $ct = 10\,\ell$ , already $(\mathbf{87.138^{\circ}}, \mathbf{87.138^{\circ}})$ ). A ruler at $t = 0$ reaches $i^0$ , hitting $\tilde{R} = \mathbf{168.579^{\circ}}$ by $r = 10\,\ell$ .

Physics of the portrait

Origins · 2 minRoger Penrose and how this was discovered

Roger Penrose

1931– · Oxford — the conformal treatment of infinity, the singularity theorem, and a Nobel Prize fifty-seven years later

Born in Colchester on 8 August 1931 and trained as a pure mathematician (a Cambridge PhD in algebraic geometry, 1957), Penrose came to relativity with a geometer's instincts and a habit of drawing. In 1963, in a two-page Physical Review Letters note — Asymptotic Properties of Fields and Space-Times, PRL 10, 66 — he introduced the idea this article is built on: attach points at infinity by a conformal rescaling, so infinity becomes an ordinary null hypersurface boundary one can do geometry on. The technique was laid out in his 1963 Les Houches lectures, published as Conformal Treatment of Infinity (in Relativity, Groups and Topology, eds. C. and B. DeWitt, Gordon & Breach, 1964, 565–584; reprinted in Gen. Rel. Grav. 43, 901, 2011), and developed in full in Zero Rest-Mass Fields Including Gravitation, Proc. R. Soc. A 284, 159 (1965). The reason it mattered was not the pictures. Asymptotic flatness — "this spacetime looks like Minkowski far away" — had been a claim about how fast components fall off in some coordinate system, which is not a geometric statement at all. Penrose replaced it with one: a spacetime is asymptotically flat if it can be conformally embedded with a smooth null boundary $\mathscr{I}$ . Gravitational radiation suddenly had a place to be measured — at $\mathscr{I}^+$ — and the strange asymptotic symmetry group that Bondi, van der Burg and Metzner (Proc. R. Soc. A 269, 21, 1962) and Sachs (Proc. R. Soc. A 270, 103, 1962) had stumbled on acquired a home: it is the symmetry group of $\mathscr{I}$ . The same geometric habit produced, in 1965, the singularity theorem (PRL 14, 57) — a trapped surface forces a singularity, so black holes are a robust prediction of general relativity, not an artefact of Schwarzschild's perfect symmetry. That is the work cited when he took half the Nobel Prize in Physics in 2020, Genzel and Ghez sharing the other half for the observations.

History

1908

— Minkowski's Raum und Zeit (Cologne, 21 September) makes the spacetime diagram, with its 45° light cones, the standard picture of special relativity. It has no room for infinity.

1917

— Einstein introduces the static universe ℝ × S³ in his cosmological paper — which will turn out, forty-six years later, to be the natural home of the compactified Minkowski diagram.

1962

— Bondi, van der Burg and Metzner (Proc. R. Soc. A 269, 21) and Sachs (Proc. R. Soc. A 270, 103) find the asymptotic symmetry group of gravitational radiation, and cannot say what it is the symmetry group of.

1963

— Penrose introduces conformal infinity: Asymptotic Properties of Fields and Space-Times, Phys. Rev. Lett. 10, 66. Infinity becomes a null hypersurface boundary attached by a conformal rescaling.

1964

— Penrose's Les Houches lectures Conformal Treatment of Infinity appear in Relativity, Groups and Topology (Gordon & Breach, 565–584), reprinted in Gen. Rel. Grav. 43, 901 (2011).

1965

— Penrose develops the method in full (Zero Rest-Mass Fields Including Gravitation, Proc. R. Soc. A 284, 159) and proves the singularity theorem (Phys. Rev. Lett. 14, 57): a trapped surface forces a singularity.

1966

— Carter's complete analytic extension of the Kerr symmetry axis (Phys. Rev. 141, 1242) supplies the black-hole diagrams that now carry both names.

2020

— Penrose receives half the Nobel Prize in Physics, for the discovery that black hole formation is a robust prediction of the general theory of relativity.

Going further

Texts. Hawking & Ellis, The Large Scale Structure of Space-Time (1973), §5.1 · Wald, General Relativity (1984), §11.1 · Hartle, Gravity (2003), ch. 7, 12. Primary: Penrose, Phys. Rev. Lett. 10, 66 (1963); Conformal Treatment of Infinity, in Relativity, Groups and Topology (1964), 565; Proc. R. Soc. A 284, 159 (1965).

Provenance. The base is written entirely in the mostly-plus signature $(-,+,+,+)$ — flat metric, null-coordinate identity ( $-\mathrm{d}u\,\mathrm{d}v$ ) and compactified metric all carry the opposite sign, and its causal-character inequalities are reversed. Nothing was transcribed; every equation was re-derived in $(+,-,-,-)$ . Two further defects are fixed in-article: (i) its boost derivation boxes the result backwards ( $u \to e^{-\varphi}u$ , $v \to e^{+\varphi}v$ ) while its own steps derive $ct'-x' = e^{+\varphi}(ct-x)$ — the steps are right, the box is wrong; (ii) its compactification stops at the 2-D part and leaves $r^2\mathrm{d}\Omega^2$ outside the conformal factor, so nothing is in fact compactified — completing it is this article's owned result. Its Weyl-tensor card is GR payload, named only. Every printed number is recomputed in verify_aq-sr-conformal-diagrams.py. Signature: $(+,-,-,-)$ , timelike-positive.

Builds on. Spacetime Diagrams · The Light Cone · Causal Structure of Spacetime · The Minkowski Metric · Timelike, Spacelike and Null Separations · Invariant Hyperbolae. Leads to. Lorentz Invariance · Rapidity · Hyperbolic Geometry of Boosts.

If the diagram destroys distances, what is it actually good for?

Questions that are about causality and about infinity — the two things ordinary diagrams handle worst. "Can this observer ever see that event?" is a question about whether a $45^{\circ}$ line connects them, answered exactly, including in the limit. "Does this signal escape to infinity?" becomes "does the ray reach $\mathscr{I}^+$ ?" — a question you can point at. None of that needs a length. If your question involves metres or seconds, the diagram is the wrong tool.

Is Ω unique? What happens if I pick a different one?

It is not unique, and nothing happens. Any $\Omega$ that is smooth, strictly positive inside, and vanishing appropriately at infinity gives a valid diagram; the $\arctan$ recipe is convenient, not canonical. Different choices give differently shaped boundaries and identical causal content, and the split into $i^{\pm}$ , $i^0$ , $\mathscr{I}^{\pm}$ is invariant. The scale $\ell$ is likewise pure gauge. What is forced, once you demand a smooth null boundary, is $\Omega \to 0$ there with $\mathrm{d}\Omega \neq 0$ — asymptotic flatness.

Why do 𝓘⁺ and i⁺ differ at all? Isn't a very fast rocket nearly a photon?

Nearly — and the "nearly" is the whole difference. Along an inertial worldline with $|\beta| < 1$ , both null coordinates run to infinity: $u = ct(1-\beta)$ grows slowly at $\beta = 0.999$ , but without bound, so $\arctan$ drives $U$ all the way to $90^{\circ}$ . Along a light ray $u$ is exactly constant and $U$ never moves. A difference of kind, not degree — the same fact as " $\gamma \to \infty$ but never arrives". The diagram makes it visible: a corner, an edge, and nothing in between.

Is Minkowski spacetime really the same as a piece of the Einstein static universe?

Conformally yes — exactly, not approximately. The theorem gives $\Omega^2\mathrm{d}s^2 = \mathrm{d}\tilde{T}^2 - \mathrm{d}\tilde{R}^2 - \sin^2\!\tilde{R}\,\mathrm{d}\Omega^2$ , the Einstein static metric on $\mathbb{R}\times S^3$ , with Minkowski the patch $|\tilde{T}| + \tilde{R} < \pi$ . They share light cones there, hence causal structure — but they are not isometric: one is infinite and open, the other a bounded region of a closed universe, and the map between them is the $\Omega$ that annihilates all distances.

Why compactify u and v rather than t and r directly?

Because you want the light cones to stay straight. Applying $\arctan$ to $t$ and $r$ separately gives a bounded picture too, but the null condition $\mathrm{d}r = \pm c\,\mathrm{d}t$ would not map to $\mathrm{d}\tilde{R} = \pm\mathrm{d}\tilde{T}$ : the cones would bend, and reading causality off the page would need a computation at every point — the thing the diagram exists to avoid. Compactifying $u$ and $v$ works because the light cones are the surfaces $u, v = \mathrm{const}$ , and a monotone reparametrisation of each leaves them coordinate surfaces.

Does the boost really fix i⁺, i⁻ and i⁰? A boost changes what "at rest" means.

It fixes them, by the same squeeze: $\tan U \to e^{+\varphi}\tan U$ , $\tan V \to e^{-\varphi}\tan V$ , and a point is fixed exactly when $\tan U$ and $\tan V$ are each $0$ or $\pm\infty$ — at $i^+$ both $+\infty$ , at $i^-$ both $-\infty$ , at $i^0$ one of each. All three stay put, however large $\varphi$ gets. No tension with a boost changing the notion of rest: the worldlines ending at $i^+$ differ from frame to frame, but the point they end at does not.