Geodesic Incompleteness and Singularities

A singularity is not a place. The naive picture — "a point where the curvature blows up" — collapses on the first hard look, because the would-be locus $r=0$ at the centre of a black hole is simply not part of the manifold, and no tensor can be evaluated where there are no points. Worse, the picture misfires in both directions: the Schwarzschild horizon $r=2GM$ has metric components that diverge in one chart yet is perfectly smooth spacetime you can fall straight through, while a cosmic-string cone has finite curvature everywhere yet still tears freely-falling histories short. So the concept has to be rebuilt from the ground up on something coordinate-invariant. The accepted answer, forged by Penrose, Hawking and Geroch in the 1960s, is geodesic incompleteness: a spacetime $(M,g)$ is singular when some inextendible timelike or null geodesic runs out of affine parameter at a finite value — a freely-falling observer or a photon reaches the edge of existence in finite proper time, with no manifold point to continue to. The essential fine print is maximality: one must first check the spacetime cannot be enlarged by a smooth isometric extension, or one has merely cut the manifold short with bad coordinates. Once genuine incompleteness is established, curvature returns as a classification tool — scalar, non-scalar, or quasiregular — not as the definition. And the deep payoff, the content of the singularity theorems, is that incompleteness is not a fragile artefact of perfect symmetry but a generic, unavoidable consequence of an energy condition, a causality condition, and a single trapped surface. Gravitational collapse and the Big Bang are therefore not accidents; general relativity predicts its own breakdown.

Key takeaways

A spacetime is singular when it contains an inextendible timelike or null geodesic of finite affine length — that is, when the manifold is geodesically incomplete and cannot be smoothly extended.
The definition is operational, not local: a freely-falling observer or light ray reaches the edge of spacetime at finite proper time or affine parameter, and there is no point there to continue to.
Singularities are NOT defined by curvature blow-up — the horizon $r=2GM$ is removable (Kruskal), while the Kretschmann scalar $K=R_{abcd}R^{abcd}=48G^2M^2/r^6$ divergence at $r=0$ merely classifies that incompleteness as a scalar-curvature type.
One must verify inextendibility (maximality) before declaring incompleteness physical, or a poor chart manufactures a fake edge.
The Raychaudhuri equation forces an initially-converging geodesic congruence to a conjugate point in finite parameter, $\Delta\tau\le 3/|\theta_0|$ — the engine of every singularity theorem.
Penrose 1965 and Hawking–Penrose 1970 turn that focusing into a theorem: an energy condition plus a trapped surface plus causality implies incompleteness, with no symmetry assumed.
Key idea: A spacetime is singular iff it is geodesically incomplete AND inextendible — some freely-falling observer or photon reaches finite affine parameter with no manifold point to continue to, and no smooth isometric extension repairs it. Curvature divergence classifies, it does not define.

The core insight

A spacetime is singular iff it is geodesically incomplete and inextendible: some freely-falling observer or photon reaches finite affine parameter with no manifold point to continue to, and no smooth isometric extension can repair it — curvature divergence is a classification tool, not the definition.

Where this lives today

Geodesic incompleteness is not a settled historical result but the live language of several frontiers. In numerical relativity, the codes that model binary black-hole and neutron-star mergers — the simulations that let LIGO and Virgo template their gravitational-wave detections since 2015 — run apparent-horizon finders to detect trapped surfaces on the fly, precisely because a trapped surface is the theorem's guarantee that a singularity is forming, and the region inside must be excised from the grid before the diverging curvature crashes the run. In mathematical relativity the sharpest open question is strong cosmic censorship: whether the Cauchy horizon inside a rotating (Kerr) or charged (Reissner–Nordström) black hole, beyond which predictability fails, is stabilised into a genuine inextendible singularity by infalling perturbations — a problem framed entirely as incompleteness-versus-extendibility, and one where recent work on near-extremal black holes finds the classical conjecture can be violated. And at the deepest level the theorems read as a no-go: classical general relativity provably predicts its own breakdown at the Big Bang and inside black holes, so any candidate quantum gravity — loop cosmology's "big bounce", string-theoretic resolutions, asymptotic safety — is judged partly by whether and how it replaces the incomplete geodesic with something complete.

The physical picture

Picture a freely-falling observer as a car and its geodesic as a road, with affine parameter as the odometer. A singularity is a road that simply ends after a finite number of kilometres, with no continuation — and crucially not because of a removable roadblock (a coordinate singularity, a pothole that vanishes once you repave with better charts) but because the asphalt of spacetime itself ceases to exist. The car falls off the edge of the world in finite mileage. We cannot point at the edge, because it is not a point of the manifold; we detect it only through the histories that try to reach it and fail. The whole subject is a study of what is missing, characterised by the geodesics that run into the gap. To tell a genuine edge from a fake one, we need an invariant, and the first figure shows the cleanest one: the Kretschmann scalar of the Schwarzschild geometry. It is finite at the horizon — proof that the horizon is no edge at all — and diverges only as $r\to 0$ .

Figure

The curvature invariant K = 48 G²M²/r⁶ for the Schwarzschild geometry locates the real singularity. At the horizon r = 2GM it is finite, K = 3/(4 G⁴M⁴), so the horizon is a removable coordinate artefact you can fall straight through. Only as r → 0 does K → ∞ — a genuine scalar-curvature singularity where the manifold ends.

Why is incompleteness generic rather than a quirk of one exact solution? Because gravity focuses. A bundle of neighbouring freely-falling worldlines, once even slightly converging, is squeezed relentlessly tighter by the tidal term in the Raychaudhuri equation, and — provided ordinary matter supplies a non-negative focusing term — driven to a caustic, a conjugate point where the geodesics cross, within a bounded stretch of parameter. The second figure is that mechanism: an initially-converging congruence collapsing to a focal point no later than $\Delta\tau=3/|\theta_0|$ . Conjugate points are the lever the theorems pull, because past one a geodesic can no longer be the longest curve between events — and in a well-behaved spacetime the longest curve must exist. The contradiction forces some geodesic to be incomplete.

Figure

Raychaudhuri focusing, the engine of the singularity theorems. A timelike geodesic congruence that starts converging, with negative expansion θ₀, is driven to a conjugate point where θ → −∞ within a finite lapse of proper time set by that initial convergence, provided the focusing term R_ab uᵃuᵇ ≥ 0. Past that focal point a geodesic no longer maximises proper time.

Once incompleteness is settled, the last question is how badly things break as the geodesic terminates. If a curvature scalar such as $K=R_{abcd}R^{abcd}$ diverges, tidal forces shred any infalling object regardless of its motion — the Schwarzschild centre and the Big Bang. If all scalars stay finite but curvature components in a parallel-propagated frame diverge, only certain observers are destroyed (a non-scalar, or p.p., singularity). And if everything stays finite yet geodesics still terminate, the pathology is purely geometric — the tip of a cone, locally flat but globally torn.

The core equations

A geodesic is a curve that parallel-transports its own tangent, and the affine parameter is the one for which the transport equation has no right-hand side. This is the object whose finite termination defines a singularity, so we begin there.

\frac{\mathrm{d}^2 x^\mu}{\mathrm{d}\lambda^2} + \Gamma^\mu_{\;\alpha\beta}\,\frac{\mathrm{d}x^\alpha}{\mathrm{d}\lambda}\frac{\mathrm{d}x^\beta}{\mathrm{d}\lambda} = 0 \qquad (1)

where: $\lambda$ — affine parameter (proper time $\tau$ for a timelike geodesic, an arbitrary affine parameter for a null one); $\Gamma^\mu_{\;\alpha\beta}$ — Christoffel symbols of the Levi-Civita connection; $x^\mu(\lambda)$ — the worldline of a freely-falling particle or photon. Reading: because $\lambda$ is fixed by the connection alone, up to $\lambda\to a\lambda+b$ , its total range is a coordinate-invariant fact about the curve. The definition of a singular spacetime is then a statement about that range together with maximality. It deliberately points at the geodesic, never at the missing "bad point".

(M,g)\ \textbf{singular}\iff \exists\ \text{inextendible causal geodesic},\ \lambda\in[0,\lambda_{\max}),\ \lambda_{\max}<\infty,\ (M,g)\ \text{maximal} \qquad (2)

where: $\lambda_{\max}$ — finite supremum of affine parameter, reached with no manifold endpoint; inextendible — the geodesic itself cannot be continued; maximal — no smooth isometric embedding into a larger spacetime exists. Reading: drop the maximality clause and every coordinate patch looks singular; it is the clause that separates physics from bad charts. For the Schwarzschild vacuum the Ricci tensor vanishes, so the only curvature is Weyl, and the simplest polynomial invariant that survives is the Kretschmann scalar. It is the workhorse diagnostic of the whole subject.

K \equiv R_{abcd}R^{abcd} = \frac{48\,G^2 M^2}{r^6}\qquad(\text{Schwarzschild}) \qquad (3)

where: $K$ — Kretschmann invariant, a scalar built from the full Riemann tensor; $R_{abcd}$ — Riemann curvature tensor; $M$ — mass; $G$ — Newton's constant (units $c=1$ ). Reading: $K(2GM)=3/(4G^4M^4)$ is finite while $K(r\to0)\to\infty$ , so the invariant cleanly certifies the horizon as removable and the centre as genuine. The mechanism that makes incompleteness generic is the Raychaudhuri equation, which governs the expansion $\theta$ of a congruence of neighbouring geodesics — the fractional rate of change of the little volume they span.

\frac{\mathrm{d}\theta}{\mathrm{d}\tau} = -\frac{1}{3}\theta^2 - \sigma_{ab}\sigma^{ab} + \omega_{ab}\omega^{ab} - R_{ab}u^a u^b \qquad (4)

where: $\theta$ — expansion; $\sigma_{ab}$ — shear (trace-free distortion, $\sigma_{ab}\sigma^{ab}\ge0$ ); $\omega_{ab}$ — vorticity ( $=0$ for a hypersurface-orthogonal congruence); $R_{ab}u^au^b$ — the tidal focusing term; $u^a$ — unit tangent. Reading: with no rotation, every term on the right except possibly $-R_{ab}u^au^b$ is manifestly non-positive, so once matter makes the focusing term non-negative the expansion can only run downhill. That "makes the focusing term non-negative" is exactly an energy condition. Feeding Einstein's equation into $R_{ab}u^au^b$ turns geometry into a statement about the stress of matter.

R_{ab}u^au^b \ge 0 \iff \Big(T_{ab}-\tfrac{1}{2}T g_{ab}\Big)u^au^b \ge 0 \qquad(\text{Strong Energy Condition}) \qquad (5)

where: $T_{ab}$ — stress-energy tensor, $T$ its trace; $g_{ab}$ — metric; $u^a$ — any timelike unit vector; the null analogue $R_{ab}k^ak^b\ge0$ (Null Energy Condition) governs light. Reading: ordinary matter and radiation ( $\rho+3p>0$ ) satisfy these, which is why the theorems bite on realistic spacetimes. The trigger for gravitational collapse is a trapped surface — a closed 2-surface so deep in a gravitational field that even its outgoing light is converging. Both null-normal expansions are negative at once.

\theta_{(\ell)} < 0 \quad\text{and}\quad \theta_{(n)} < 0 \quad\text{everywhere on}\ \Sigma \qquad (6)

where: $\Sigma$ — a compact spacelike 2-surface; $\theta_{(\ell)},\theta_{(n)}$ — expansions of the outgoing and ingoing future-directed null-normal congruences. Reading: in flat space the outgoing wavefront of a sphere expands ( $\theta_{(\ell)}>0$ ); a trapped surface is one where gravity has dragged even that inward, and Raychaudhuri then guarantees a focal point in finite affine parameter, $\lambda\le 2/|\hat\theta_0|$ . Finally, the cosmological case reduces to a single ordinary differential equation for the scale factor — the Friedmann acceleration (Raychaudhuri) equation for a homogeneous isotropic universe.

\frac{\ddot a}{a} = -\frac{4\pi G}{3}\big(\rho+3p\big) \;\Rightarrow\; a(t)\to0\ \text{at finite past }t\ (\text{Big Bang}) \qquad (7)

where: $a(t)$ — scale factor; $\rho,p$ — energy density and pressure of the cosmic fluid; $t$ — cosmic proper time of comoving observers. Reading: for $\rho+3p>0$ the expansion decelerates, so run backwards $a$ must hit zero in finite time, where $\rho\sim a^{-3}$ or $a^{-4}$ and the curvature diverges — a past scalar singularity that comoving geodesics reach in finite proper time.

Derivation AFocusing forces a conjugate point in bounded parameter

The goal of this derivation is to extract, from the Raychaudhuri equation and a single energy condition, the one quantitative fact on which every singularity theorem stands: an initially-converging geodesic congruence must develop a conjugate point — a caustic where neighbouring geodesics cross and the expansion diverges to minus infinity — within a bounded lapse of affine parameter, whatever the detailed geometry. We take a hypersurface-orthogonal timelike congruence so vorticity vanishes, discard the terms of definite sign, integrate the inequality, and read off the bound.

Step 1. Set $\omega_{ab}=0$ (hypersurface-orthogonal, which holds for the normal congruence of any spacelike slice) so that the Raychaudhuri equation loses its only positive term and becomes $\mathrm{d}\theta/\mathrm{d}\tau = -\tfrac13\theta^2 - \sigma_{ab}\sigma^{ab} - R_{ab}u^au^b$ . Why: rotation is the one effect that can fight collapse — a spinning bundle can avoid focusing — so we must exclude it to get a clean theorem.

Step 2. Impose the strong energy condition, $R_{ab}u^au^b\ge0$ , and note $\sigma_{ab}\sigma^{ab}\ge0$ always. Both subtracted terms are then non-negative, so $\mathrm{d}\theta/\mathrm{d}\tau \le -\tfrac13\theta^2$ . Why: discarding manifestly non-positive contributions can only make $\theta$ fall faster, so any bound we derive from the reduced inequality is rigorous for the true congruence.

Step 3. Change variables to $u=\theta^{-1}$ . Since $\mathrm{d}u/\mathrm{d}\tau = -\theta^{-2}\,\mathrm{d}\theta/\mathrm{d}\tau$ and $\theta^2>0$ , dividing the inequality by $-\theta^2$ flips it to $\mathrm{d}(1/\theta)/\mathrm{d}\tau \ge \tfrac13$ . Why: the nonlinear $\theta^2$ term is exactly what makes the reciprocal linear — this is the trick that turns a hard differential inequality into an integrable one.

Step 4. Integrate from $\tau=0$ : $\dfrac{1}{\theta(\tau)} \ge \dfrac{1}{\theta_0} + \dfrac{\tau}{3}$ . If the congruence starts converging, $\theta_0<0$ , then $1/\theta_0<0$ and the right-hand side climbs through zero at $\tau = 3/|\theta_0|$ . Why: $1/\theta$ passing through zero from below means $\theta\to-\infty$ — the neighbouring geodesics have crossed, which is precisely a conjugate (focal) point.

Step 5. Conclude $\theta\to-\infty$ at some $\tau\le 3/|\theta_0|$ . The bound is universal: it uses no symmetry, only the energy condition and $\theta_0<0$ . Why: this finiteness is the whole game — a caustic guaranteed to arrive on a schedule is what the global arguments convert into an incomplete geodesic. The variational sting comes next. Past a conjugate point a timelike geodesic no longer maximises proper time between its endpoints; a nearby curve is longer. But in a globally hyperbolic spacetime a longest causal curve between causally related points must exist and must be an unbroken geodesic. Hold both facts and something gives — the geodesics cannot all extend far enough to reach their conjugate points, so some geodesic is incomplete. The reciprocal bound $\tau\le 3/|\theta_0|$ is the standard timelike result. The null analogue replaces $\tfrac13\to\tfrac12$ (2-D transverse space), giving $\lambda\le 2/|\hat\theta_0|$ .

Go deeper · Derivation B · 2 minPenrose 1965 — trapped surface implies null incompleteness

Derivation BPenrose 1965 — trapped surface implies null incompleteness

The goal here is to see how the local focusing bound becomes a global theorem, with no symmetry assumed, in the argument that earned Penrose the 2020 Nobel Prize. We assume a globally hyperbolic spacetime with a non-compact Cauchy surface, the null energy condition, and one closed trapped surface, then argue by contradiction: suppose every relevant null geodesic is complete, and derive an impossibility about the topology of the boundary of the future.

Step 1. Consider the boundary $\dot J^{+}(\Sigma)$ of the causal future of the trapped surface $\Sigma$ . General causal theory says this boundary is generated by null geodesic segments that leave $\Sigma$ orthogonally and have no past endpoints except on $\Sigma$ , and that a generator leaves the boundary the moment it reaches a conjugate point. Why: the boundary of a future set is always a null "achronal" surface woven from such generators — this is the structural fact that lets us apply focusing to it.

Step 2. On a trapped surface both null-normal expansions are negative, $\theta_{(\ell)}<0$ and $\theta_{(n)}<0$ . Every generator therefore starts converging. Why: this is the entire physical input — a region so strongly gravitating that even outgoing light is being pulled inward is what makes the argument go.

Step 3. Apply the null focusing bound with the NEC ( $R_{ab}k^ak^b\ge0$ ): each generator, if it extends that far, reaches a conjugate point within affine parameter $\lambda\le 2/|\hat\theta_0|$ , and leaves $\dot J^{+}(\Sigma)$ there. Under the assumption of completeness, every generator does extend far enough. Why: bounded, uniform focusing means the boundary is swept out entirely within a compact range of parameter along a compact set of directions.

Step 4. It follows that $\dot J^{+}(\Sigma)$ is compact — a closed set with every generator terminated at a nearby conjugate point. Why: compactness is the property we will collide with the global topology; it is the payoff of assuming completeness.

Step 5. But in a globally hyperbolic spacetime the integral curves of a timelike vector field map $\dot J^{+}(\Sigma)$ homeomorphically onto an open subset of the Cauchy surface $S$ . A subset of a connected, non-compact manifold cannot be simultaneously compact (Step 4) and open and non-empty. Why: this is the contradiction — it means the assumption of completeness was false, so at least one null generator is incomplete, and the spacetime is singular. The theorem is remarkable for what it does not assume: no spherical symmetry, no exact solution, no detail of the collapsing matter — a single trapped surface plus reasonable energy and causality suffices. This is what converted the black-hole singularity from a suspected artefact of Schwarzschild's perfect symmetry into an inevitable feature of realistic collapse. The Hawking–Penrose 1970 theorem replaces the Cauchy-surface hypothesis by a genericity plus a causality condition and adds the strong energy condition, covering both collapse and cosmology.

Go deeper · Derivation C · 2 minThe horizon is removable, the centre is not

Derivation CThe horizon is removable, the centre is not

The goal of this derivation is to work the defining distinction explicitly on the one metric everyone knows, showing side by side that the Schwarzschild horizon $r=2GM$ is a coordinate singularity a change of chart erases, while the centre $r=0$ is a true, inextendible curvature singularity. We compare three diagnostics — raw metric components, the invariant scalar, and the affine length of an infalling geodesic — and watch them disagree at the horizon and agree at the centre.

Step 1. In Schwarzschild coordinates $g_{tt}=-(1-2GM/r)$ and $g_{rr}=(1-2GM/r)^{-1}$ , so $g_{rr}\to\infty$ as $r\to2GM$ . Taken at face value this looks catastrophic. Why: it is exactly this chart-level blow-up that fooled a generation into thinking the horizon was a physical singularity — the cautionary tale that motivates the invariant definition.

Step 2. Evaluate the invariant instead: $K(2GM)=48G^2M^2/(2GM)^6 = 3/(4G^4M^4)$ , perfectly finite. No scalar built from the curvature misbehaves at the horizon. Why: a genuine singularity must show up in a coordinate-free quantity; the finiteness of $K$ is a certificate that nothing physical is wrong here.

Step 3. Exhibit the extension. In Kruskal–Szekeres coordinates $(U,V)$ the metric is $\mathrm{d}s^2 = -\tfrac{32G^3M^3}{r}e^{-r/2GM}\,\mathrm{d}U\,\mathrm{d}V + r^2\mathrm{d}\Omega^2$ , which is smooth and non-degenerate at $r=2GM$ . An infalling geodesic crosses the horizon at finite affine parameter and continues. Why: producing an explicit smooth isometric extension is the operational meaning of "removable" — the incompleteness of the Schwarzschild chart was an artefact of deleting half the manifold.

Step 4. Now push to the centre. A radial timelike geodesic released from rest at $r=2GM$ reaches $r=0$ in proper time $\tau=\int_0^{2GM}\mathrm{d}r/\sqrt{2GM/r}=\tfrac{\pi}{2}\sqrt{(2GM)^3/(2GM)}=\pi GM$ — finite. There $K=48G^2M^2/r^6\to\infty$ . Why: finite affine length that ends on divergent curvature is the archetype of a scalar-curvature singularity; the geodesic literally runs out of spacetime in finite proper time.

Step 5. No extension can exist across $r=0$ : an isometric extension would require the curvature to become a smooth tensor field there, impossible when a scalar invariant diverges. The geodesic is inextendible and the spacetime is maximal. Why: this is the maximality check made concrete — the centre passes it (genuinely singular) exactly where the horizon failed it (merely a bad chart). *The value $K(2GM)=3/(4G^4M^4)$ follows by direct substitution into $K=48G^2M^2/r^6$ ; the fall-from-horizon proper time $\tau=\pi GM$ uses the integral $\int_0^R\sqrt{r/(R-r)}\,\mathrm{d}r=\pi R/2$ .*

Limiting cases

The same metric can wear apparent pathologies that are, or are not, singularities. The reliable diagnostic is a trio: does an invariant diverge, is the spacetime extendible, and does the affine parameter reach a finite or infinite limit? Read across the table and each row is decided by that trio.

Case	Result	Meaning
Coordinate singularity (Schwarzschild horizon)	$K=\dfrac{48G^2M^2}{r^6}$ finite at $r=2GM$	The metric-component blow-up is removed by Kruskal–Szekeres coordinates; geodesics cross smoothly. NOT a singularity — the spacetime extends through the horizon.
Scalar curvature singularity (Schwarzschild centre)	$K\to\infty$ as $r\to0$	Radial geodesics reach $r=0$ at finite proper time ( $\tau=\pi GM$ from the horizon); $K$ diverges; no extension exists. A genuine, physical singularity.
Quasiregular / conical (cosmic string)	$\mathrm{d}s^2=\mathrm{d}r^2+(1-4G\mu)^2 r^2\,\mathrm{d}\phi^2$	Locally flat off-axis ( $K=0$ everywhere), every curvature scalar vanishes, yet the deficit angle makes the axis a quasiregular singularity where geodesics terminate. Incompleteness without curvature.
Conformal infinity (not a singularity)	$\lambda\to\infty$	In asymptotically flat or de Sitter regions geodesics reach $\mathscr{I}$ only at infinite affine parameter — the spacetime is complete there. Infinity is reached, not an edge.

Common trapA singularity is where the curvature blows up

This is the single most common error, and it fails in both directions. The Schwarzschild horizon $r=2GM$ has a divergent metric component $g_{rr}$ in Schwarzschild coordinates, yet its curvature invariant $K=3/(4G^4M^4)$ is finite and the spacetime extends smoothly through it in Kruskal coordinates — no singularity at all. Conversely a conical (cosmic-string) spacetime is locally flat, with every curvature scalar and every parallel-propagated frame component identically zero, yet geodesics aimed at the axis terminate at finite affine parameter and cannot be continued — a genuine singularity with no curvature blow-up whatsoever. The lesson is procedural: first establish geodesic incompleteness and verify inextendibility, and only then reach for curvature scalars — and even then use invariants or parallel-propagated frame components, never raw chart components, since those diverge wherever the coordinates do.

Worked examples

1 · A solar-mass black hole, from horizon to centre (fully worked)

Take a Schwarzschild black hole of one solar mass, $M=M_\odot$ , and an observer who falls radially from rest at the horizon. Show the horizon is regular, find how long the observer survives, and confirm the centre is a true singularity. Use $G=6.674\times10^{-11}$ , $c=2.998\times10^{8}$ , $M_\odot=1.989\times10^{30}\ \mathrm{kg}$ .

Step 1. Locate the horizon. The Schwarzschild radius is $r_s=2GM/c^2 = 2(6.674\times10^{-11})(1.989\times10^{30})/(2.998\times10^8)^2 = \mathbf{2.95\times10^{3}\ m}$ , about 3 km. This is where $g_{rr}$ diverges in Schwarzschild coordinates — the apparent, removable pathology.

Step 2. Show the horizon is regular by the invariant. Restoring $c$ , $K=48G^2M^2/(c^4 r^6)$ , so at $r=r_s$ one finds $K(r_s)=\tfrac34\,(c^2/GM)^4 = \mathbf{1.6\times10^{-13}\ m^{-4}}$ — a small, finite tidal curvature. Nothing physical happens at the horizon; the geodesic sails through.

Step 3. Find the survival time. From the derivation, an observer released from rest at the horizon reaches $r=0$ in proper time $\tau=\pi GM/c^3$ . Numerically $GM/c^3 = 4.93\times10^{-6}\ \mathrm{s}$ , so $\tau=\pi(4.93\times10^{-6}) = \mathbf{1.55\times10^{-5}\ s}$ — about 15 microseconds of life beyond the horizon, no matter how you struggle.

Step 4. Confirm the centre is singular. As $r\to0$ , $K=48G^2M^2/(c^4 r^6)\to\infty$ : the tidal curvature that was $10^{-13}\ \mathrm{m^{-4}}$ at the horizon grows without bound, stretching the infaller radially and crushing them transversely (spaghettification). The geodesic ends at finite $\tau$ on divergent curvature, with no isometric extension — a scalar-curvature singularity, exactly as the theorem promises. The moral in numbers: the horizon, where the coordinates scream, is a benign 3-km sphere of finite $10^{-13}\ \mathrm{m^{-4}}$ curvature, while the centre — which no chart can reach — is where the geometry genuinely ends 15 μs later.

2 · The Big Bang is a past-incomplete edge, not a coordinate artefact

Take a spatially-flat radiation-dominated FLRW universe, $a(t)\propto t^{1/2}$ . Show that comoving observers' worldlines are past-incomplete and that the incompleteness is a genuine curvature singularity.

Step 1. Check the focusing input. For radiation $p=\rho/3$ , so $\rho+3p=2\rho>0$ and the acceleration equation $\ddot a/a = -\tfrac{4\pi G}{3}(\rho+3p)<0$ : expansion decelerates, so tracing backward $a$ must reach $0$ at a finite past time, which we set to $t=0$ . The scale factor hits zero in finite cosmic time.

Step 2. Verify the geodesics are incomplete. Comoving observers have constant spatial coordinates, and cosmic time $t$ is their proper time; their worldlines exist only for $t>0$ and reach $t=0$ (where $a=0$ ) at finite proper time in the past, with no earlier point to continue to. These timelike geodesics are past-incomplete.

Step 3. Confirm it is a curvature singularity, not a bad chart. The energy density of radiation scales as $\rho\propto a^{-4}$ , and curvature invariants track it: the Ricci scalar and Kretschmann scalar diverge as $a\to0$ (schematically $K\sim\rho^2\sim a^{-8}\to\infty$ ). Divergent invariants certify an inextendible scalar singularity — the Big Bang is a real edge of spacetime. This is the cosmological face of the same result: applied to an expanding, globally hyperbolic universe obeying the strong energy condition, Hawking's theorem guarantees this past incompleteness generically, which is why classical general relativity predicts its own breakdown at the origin and points beyond itself to quantum gravity.

3 · A cosmic string: incompleteness with zero curvature

A straight cosmic string of tension $\mu$ produces the metric $\mathrm{d}s^2=-\mathrm{d}t^2+\mathrm{d}z^2+\mathrm{d}r^2+(1-4G\mu)^2 r^2\mathrm{d}\phi^2$ . Find the deficit angle and explain why the axis is singular despite flat space everywhere off it. Take a grand-unified string with $G\mu=10^{-6}$ (units $c=1$ ).

Step 1. Read off the geometry. Off the axis the metric is flat — the coordinate $\tilde\phi=(1-4G\mu)\phi$ makes it manifestly Minkowski — so every curvature component, and hence $K$ , vanishes identically for $r>0$ . The spacetime is locally flat everywhere except the axis.

Step 2. Find the deficit angle. Because $\phi$ still runs $0\to2\pi$ while $\tilde\phi$ runs $0\to2\pi(1-4G\mu)$ , a wedge of angle $\Delta\phi=8\pi G\mu$ is missing — space is a cone. For $G\mu=10^{-6}$ , $\Delta\phi=8\pi\times10^{-6}=2.5\times10^{-5}\ \mathrm{rad}=\mathbf{5.2\ arcsec}$ . This is a real, observable double-imaging angle for lensing by such a string.

Step 3. Locate the singularity. At the cone tip $r=0$ the curvature is a delta-function (a conical defect); geodesics aimed at the axis terminate there at finite affine parameter and the spacetime cannot be smoothly extended across it. A quasiregular (conical) singularity — geodesic incompleteness with all curvature scalars finite (indeed zero), the decisive counterexample to "singularity = curvature blow-up".

Physics of the portrait

The definition of a singularity, and the theorems that make singularities inevitable, are the achievement of one person more than any other — a mathematician who imported global, topological reasoning into a field that had lived on exact solutions.

Origins · 1 minRoger Penrose and how this was discovered

Roger Penrose

b. 1931, Nobel Prize in Physics 2020 · global methods and the singularity theorems of general relativity

Roger Penrose reshaped general relativity by asking questions exact solutions could not answer. Trained as a pure mathematician at Cambridge, he brought topology and differential geometry to a subject that, into the early 1960s, still equated "singularity" with the curvature blow-up of the perfectly symmetric Schwarzschild and Friedmann solutions — leaving open the widespread hope that real, lumpy, rotating collapse would miss the singularity and bounce. In 1965, in a three-page Physical Review Letters paper, Penrose settled it. He introduced the closed trapped surface as the invariant signature of runaway collapse, recast the problem in terms of the global causal structure of spacetime rather than any coordinate, and proved that a trapped surface plus the null energy condition plus a non-compact Cauchy surface forces null geodesic incompleteness. The proof's engine was the focusing of null congruences via the Raychaudhuri equation, turned through the topology of the boundary of the future into an unavoidable contradiction. With Stephen Hawking he extended the argument in 1970 to cosmology, establishing the Big Bang as a generic past singularity. Penrose also supplied much of the field's modern vocabulary — conformal (Penrose) diagrams, cosmic censorship, the classification of causal structure — and in 2020 received the Nobel Prize "for the discovery that black hole formation is a robust prediction of the general theory of relativity". What Einstein feared might be an artefact, Penrose proved to be a theorem.

History

1955

— Raychaudhuri derives the focusing equation for geodesic congruences

1965

— Penrose proves the first singularity theorem using a closed trapped surface

1968

— Geroch reframes the question "what is a singularity" around geodesic incompleteness

1970

— Hawking and Penrose prove the combined theorem covering collapse and cosmology

2020

— Penrose receives the Nobel Prize for the black-hole singularity prediction

Going further

Texts. Hawking & Ellis, The Large Scale Structure of Space-Time (1973), Ch. 8 · Wald, General Relativity (1984), Ch. 9 · Misner, Thorne & Wheeler, Gravitation (1973), Ch. 31–34 · Carroll, Spacetime and Geometry (2004), Ch. 5 & 9 · Penrose, Gravitational Collapse and Space-Time Singularities, Phys. Rev. Lett. 14, 57 (1965) · Hawking & Penrose, The Singularities of Gravitational Collapse and Cosmology, Proc. R. Soc. A 314, 529 (1970) · Geroch, What is a Singularity in General Relativity?, Ann. Phys. 48, 526 (1968). The Schwarzschild Kretschmann scalar is $K=48G^2M^2/r^6$ , from which $K(2GM)=3/(4G^4M^4)$ follows by substitution. The timelike focusing bound is $\Delta\tau\le 3/|\theta_0|$ , with the null form $\lambda\le2/|\hat\theta_0|$ . The fall-from-horizon proper time $\tau=\pi GM/c^3\approx1.55\times10^{-5}\ \mathrm{s}$ for $M_\odot$ uses $\int_0^R\!\sqrt{r/(R-r)}\,\mathrm{d}r=\pi R/2$ ; the FLRW acceleration equation is $\ddot a/a=-\tfrac{4\pi G}{3}(\rho+3p)$ and the cosmic-string deficit angle is $\Delta\phi=8\pi G\mu$ .

Builds on: Causal Structure of Spacetime · Global Hyperbolicity · Null Geodesics and Light Cones · Domain of Dependence and Cauchy Horizons. Leads to: Penrose–Carter Conformal Diagrams · Cosmic Censorship · Conformal Infinity and Asymptotic Structure.

Ask the tutor: Ask me to walk through why a single trapped surface — through Raychaudhuri focusing and the topology of the future's boundary — is enough to guarantee an incomplete geodesic, with no symmetry assumed.

Isn't a singularity just the point at the centre of a black hole?

No — and the word "point" is exactly the trap. The centre $r=0$ is not a point of the spacetime manifold at all; it is a missing locus, the place where the curvature would be infinite if it belonged to the manifold, but it does not, so you cannot evaluate any field there. That is precisely why the definition never names the "bad point". Instead it describes the geodesics that try to reach the gap and fail: they run out of affine parameter at a finite value with nowhere to continue. The singularity is characterised by what is absent, detected through the incomplete histories that fall toward it, not by a coordinate you can label.

Why insist on inextendibility before calling a spacetime singular?

Because otherwise you can manufacture a fake singularity for free. Take perfectly ordinary Minkowski space and simply delete a region; now geodesics heading into the deleted zone terminate at finite parameter, and the "spacetime" is incomplete — yet nothing physical is wrong, you just cut a hole. The Schwarzschild chart does something subtler but similar: it omits the interior, so infalling geodesics appear to stop at the horizon. Only when no smooth isometric extension exists — when you have already enlarged the spacetime as far as it will go — is an incomplete geodesic evidence of a genuine edge. Maximality is the clause that separates real physics from bad bookkeeping.

How can a trapped surface, defined by light-cones, tell you a singularity is coming?

A trapped surface is a closed 2-surface on which even the outgoing light is converging — gravity is so strong that both families of light rays leaving it are being focused inward. Feed that negative initial expansion into the null Raychaudhuri equation with the null energy condition, and focusing guarantees each light ray hits a conjugate point within a bounded affine parameter. Penrose then showed, using only the global causal structure and a non-compact Cauchy surface, that these focal points cannot all be reached by complete geodesics without a topological contradiction. So the purely local condition "both light expansions negative here and now" propagates, via focusing and topology, into the global conclusion "some geodesic is incomplete". It is a remarkably economical bridge from a snapshot to a fate.

Do the singularity theorems tell you what the singularity actually looks like?

No, and this is a genuine and often-underappreciated limitation. The theorems prove only that some causal geodesic is incomplete; they say nothing about whether the curvature diverges, whether the singularity is spacelike or timelike, how strong it is, or whether it is hidden behind a horizon. The Schwarzschild and FLRW singularities happen to be scalar-curvature type, but that is extra information from those specific solutions, not from the theorems. A conical singularity is incomplete with zero curvature; a Cauchy horizon may be incomplete in a "weak" sense that still permits extension of the metric with low differentiability. The theorems are existence statements about incompleteness, deliberately silent on the character of the edge.

What is the difference between a scalar and a non-scalar curvature singularity?

It is about which observers get destroyed. At a scalar curvature singularity a coordinate-invariant polynomial in the Riemann tensor — like the Kretschmann scalar $K$ — diverges along the incomplete geodesic, so tidal forces blow up in every frame and any infalling object is torn apart regardless of how it moves; Schwarzschild's centre and the Big Bang are of this kind. At a non-scalar (parallel-propagated, or p.p.) singularity every such scalar stays finite, yet curvature components measured in a parallel-transported orthonormal frame diverge, so only observers on particular geodesics experience unbounded tidal forces. Both are genuine singularities because the geodesics are incomplete; they differ only in how the breakdown is distributed across the possible states of motion.

Do the singularity theorems mean physics really ends at the Big Bang and inside black holes?

They mean classical general relativity ends there — which is arguably their most important message. The theorems are airtight given their hypotheses, and those hypotheses (an energy condition, causality, a trapped surface or cosmological focusing) hold for realistic matter, so within the classical theory the incompleteness is unavoidable. But an incomplete geodesic is exactly where the theory declares its own inadequacy: curvatures approach the Planck scale, and quantum-gravitational effects, entirely outside general relativity's remit, must take over. So the theorems are best read not as "spacetime truly stops" but as a rigorous proof that general relativity cannot be the whole story — it predicts, from the inside, the boundary of its own validity and hands the problem to quantum gravity.