The surface that only lets things in

In 1783 John Michell, rector of Thornhill and a competent natural philosopher, asked what would happen if a star were massive enough that the speed needed to escape it exceeded the speed of light. He worked out the size such a body would have and concluded that light emitted from its surface would fall back, so the star would be invisible. Laplace published the same idea in 1796 and quietly removed it from later editions.

The calculation is two lines. Escape from a mass $M$ at radius $R$ requires $\tfrac12 v^2 = GM/R$; set $v = c$ and

$$R = \frac{2GM}{c^2}.$$

That is exactly the Schwarzschild radius, factor of two and all — and every step used to get it is wrong.

Why the right answer is not a reason to be pleased #

Michell’s derivation assumes a particle with mass, a flat space, a Newtonian kinetic energy, and the idea of something that rises, slows, stops and falls back. Light does none of that: it has no mass, it never slows, and there is no space in which the calculation is being done. Getting the right number from those assumptions is a coincidence — two errors that happen to cancel — and it is one of the standing examples in physics of why an answer’s correctness says nothing about a derivation’s.

The honest version comes from the geometry outside a spherical mass. A clock held still at radius $r$ runs slow by

$$\sqrt{1 - \frac{r_s}{r}}, \qquad r_s = \frac{2GM}{c^2},$$

against a clock far away. As $r$ falls toward $r_s$ that factor goes to zero.

That is what the horizon is: not a place where the escape speed reaches $c$, but a place where the notion of staying still stops existing. Above it, an object can hover by burning fuel indefinitely. At it, hovering would require infinite acceleration, and below it there is no amount of thrust that keeps $r$ constant, for the same reason that no amount of effort keeps a clock at yesterday. Inside, decreasing $r$ is in the future.

The one-way property, drawn as light cones #

The clean way to see the change is on a spacetime diagram, where what is available to a body at each event is the interior of its own light cone.

In flat spacetime the light cone at every event opens the same way, and every worldline inside it can be tipped toward either future direction: which way “now” points is a choice, and every direction of space is available. Near a mass the cones tip toward it, and at the horizon they tip far enough that every direction inside the cone has decreasing radius. The tipping is a property of the geometry rather than of the drawing’s axes — which is the one thing no drawing of it can show, because a drawing has to choose axes.

Once the cones have tipped that far, moving outward is not hard, it is unavailable, in the same sense that moving into the past is unavailable in flat spacetime. That is why a horizon is described as a surface in spacetime rather than in space: what defines it is a statement about which directions are future-directed, not about where a wall is.

It also explains something the escape-speed picture gets exactly wrong. In Michell’s version light leaves the surface, rises, slows, and falls back — a fountain. In the correct version light emitted radially outward at the horizon neither escapes nor falls: it stays at $r_s$ for ever, at the speed of light, going nowhere. Below the horizon, light emitted outward moves inward.

Escape is a statement about energy, and it stops making sense #

The escape-speed reading survives one useful step further, and then breaks in an instructive place.

The Newtonian version of escape is a well, a total energy and two turning points where the kinetic energy runs out. Escape means having enough energy that no turning point exists — a statement about the shape of the curve rather than about a speed. The relativistic version keeps the shape and changes what “enough” means, because the extra term abolishes the inner part of the barrier altogether and a photon’s energy does not appear in the criterion at all.

For a massive body the Newtonian criterion generalises: there is an effective potential, and enough energy means no turning point. For light there is no such criterion, because a photon’s energy scales out — a more energetic photon is no better at escaping than a feeble one. Everything about whether light gets out is fixed by where it starts and which way it points, and by nothing about the light.

The weak-field version of this is the ordinary statement that a ray’s deflection depends on where it passes and not on what it is. Follow the trend inward and the deflection stops being small. At an impact parameter of 1.5 horizon radii a photon orbits the mass; inside that it spirals in, whatever its energy. That is capture, and it has no analogue at all in a theory where escape is a matter of going fast enough.

The singularity that is a property of the drawing #

At $r = r_s$ the Schwarzschild expression for the geometry falls apart: one term goes to zero and another to infinity. For nearly fifty years this was read as a physical catastrophe, and the surface was called the Schwarzschild singularity.

It is not one. The test that settles it is to compute a quantity that does not depend on which coordinates are used — a curvature scalar — and see whether that misbehaves. At the horizon every such scalar is finite. At $r = 0$ they diverge. So there are two very different things in this geometry, and only one is a singularity: the other is an artefact of the coordinate system, in the same way that longitude misbehaves at the North Pole without anything being wrong with the Pole.

Better coordinates remove it. Eddington had them in 1924 and did not notice what he had; Finkelstein rediscovered them in 1958; Kruskal and Szekeres produced the complete version in 1960. In those coordinates the infalling worldline crosses $r_s$ smoothly, with nothing to mark the event.

This is exactly the failure mode this collection files under a property of the drawing, mistaken for one of the world, and it is the most consequential example physics has: a coordinate artefact delayed the acceptance of the horizon’s real nature for four decades, during which the objects themselves were regarded as mathematical curiosities that nature would surely avoid.

Nothing local happens on the way through #

The corollary is worth stating on its own, because it is the single most counter-intuitive thing about the subject.

An observer falling through a horizon detects nothing there. No wall, no flash, no jolt. Locally the geometry is as flat as any other freely falling frame, and the crossing is defined by a global property — whether a signal sent outward will ever reach a distant receiver — which is not the sort of thing a local instrument can measure.

Against the faller’s own clock the horizon arrives after a finite and unremarkable interval; against a distant clock it never arrives at all. Both statements describe the same trajectory, and there is nothing on the faller’s curve to mark the crossing — no jolt, no discontinuity, no local measurement that returns a different answer than it would anywhere else. What the horizon is has to be stated in terms of where a signal can get to, and the faller has no way to run that experiment from inside.

What is felt is the tidal stretching, and its size is a surprise. The tidal acceleration across an object of length $L$ at the horizon goes as $GML/r_s^3$, and since $r_s \propto M$ that is $\propto L/M^2$. For a stellar-mass hole it is lethal well outside the horizon. For a hole of a few million solar masses it is smaller at the horizon than the tidal stretching a person experiences standing on the Earth. Bigger horizons are gentler, without limit.

Not dense, either #

The same scaling kills the other stock image. Mass over volume goes as $M/r_s^3 \propto 1/M^2$, so a horizon’s mean density falls as the square of its mass.

Run the numbers. A solar-mass horizon encloses a mean density around $2\times10^{19}$ kg/m³ — nuclear density, and genuinely extreme. A horizon of $10^8$ solar masses encloses about 1.8 kg/m³, which is denser than air and less dense than cork. A horizon of $10^9$ solar masses encloses less than the density of water.

The comparison worth carrying is with an ordinary object. Squeeze the Earth to nine millimetres and the density on the way is absurd; assemble a hundred million solar masses in a volume the size of the inner solar system and the density never exceeds that of the air in a room. Whether a horizon forms is a question about mass and radius together, and the intuition that it is a question about compression comes entirely from having only ever thought about the small cases.

Nothing about a horizon requires matter to be compressed at all, then. It requires enough mass inside a given radius, which for a large enough radius is a modest requirement — and that is the reason the phrase singularity and the phrase horizon have to be kept apart. The first is a place where the theory admits it has failed; the second is a place where nothing is happening.

The one number that is not a coordinate #

Since the radius is a label rather than a distance, it is fair to ask what about a horizon is unambiguous. The answer is its area.

$$A = 4\pi r_s^2 = \frac{16\pi G^2 M^2}{c^4}.$$

Area is defined by measurements made on the surface — lay out a grid of rulers over it and count — and those measurements are unaffected by the radial stretching that makes the radius meaningless. So the area is a genuine geometrical fact about the object in a way that its “size” is not, and it goes as the square of the mass.

That would be a footnote if it were not for what happens next. Hawking proved in 1971 that the total area of horizons can never decrease in any classical process: merge two, and the area of the result exceeds the sum: throw anything in, and the area grows. A quantity that only increases, attached to a system, in a theory that has nothing to do with heat, is an extremely suspicious object — and the suspicion was correct. It is an entropy, which is where this ladder crosses into thermodynamics.

Once inside, the centre is not a place but a time #

The most useful single sentence about the interior is that the roles of the radial and time coordinates exchange, and the consequences of taking it seriously are strange in a way that the “point of no return” phrasing entirely hides.

Outside a horizon, $r$ labels where something is and a body can hold its radius by applying thrust. Inside, decreasing $r$ is a future direction — one of the directions the light cone allows and the only ones available — so reaching $r = 0$ is not a matter of falling in a particular direction. It is as unavoidable as next Tuesday.

Two things follow that are worth stating plainly. First, no rocket, material or force can prevent the arrival, because preventing it would mean not moving into the future. That is a much stronger statement than “gravity is too strong to escape”, and it is why the singularity theorems of Penrose and Hawking are theorems about geometry rather than about the strength of matter. Second, the time available before arrival is finite and short: for a hole of $M$ solar masses it is at most about $1.5\times10^{-5} M$ seconds, and it is maximised by not struggling. Firing the engines shortens the remaining life, because the free-fall worldline is the one of greatest proper time.

What is actually measured #

Everything above is geometry. It is worth asking separately what has been observed, because the answer has a structure that the theory predicts and that is easy to miss: the evidence is strong, and it is not evidence for a horizon.

The mass measurements are the least ambiguous part. Stars near the centre of the Galaxy have been tracked through complete orbits for three decades; the one labelled S2 has a period of sixteen years and comes within about 120 astronomical units of the focus, and Kepler’s third law then requires 4.3 million solar masses inside that distance. Nothing made of stars can be packed that densely and remain stable for long, so whatever is there is a single object.

The second kind of evidence is the shadow. Light passing close enough is captured, so an object surrounded by glowing material should show a dark region of a size the theory fixes exactly — about 5.2 Schwarzschild radii across, set by the photon capture cross-section rather than by the horizon itself. Interferometry across the width of the Earth has now imaged that region for two objects, and the diameters agree with the independently measured masses.

The third is the absence of a surface. Matter falling onto a neutron star lands, accumulates, and eventually detonates in a thermonuclear flash — the type-I X-ray bursts seen from accreting neutron stars in their hundreds. Objects of the same kind but higher mass, accreting in the same way, show no such bursts and are far fainter in quiescence than a surface receiving that much matter could possibly be. The energy is going somewhere that does not radiate.

And the fourth is the ringing. Two merging horizons settle into one, and the settling emits gravitational waves at frequencies fixed by the final mass and spin alone. Those frequencies have been measured and they match.

What none of that establishes is a horizon, and the reason is a consequence of the definition rather than a limitation of the instruments. A horizon is the boundary of the region from which a signal can never reach a distant observer — never, meaning at arbitrarily late times. Deciding whether a given surface is one requires knowing the entire future of the spacetime. No finite observation can do it, in principle, and no improvement in telescopes changes that.

So what the measurements establish is that these objects have no surface down to within a few per cent of where a horizon would be, and that their exteriors match the predicted geometry to the precision available. That is a strong statement and it is not the same statement. The honest summary is that the alternatives have been pushed into a very small corner, and that the corner cannot be closed by observation alone.

The forty years, and what ended them #

The delay described above — a coordinate artefact mistaken for a catastrophe for four decades — was not merely a technical confusion, and it is worth knowing what actually changed.

Oppenheimer and Snyder worked out the collapse of a pressureless sphere in 1939 and got the modern answer complete: the star’s own clock records a finite fall through the horizon, while a distant observer sees the collapse slow, redden and freeze. The paper says both things and does not treat them as a paradox. It was almost entirely ignored, in part because it appeared in September 1939 and in part because the community’s leading figures thought the conclusion absurd. Einstein published a paper the same year arguing that such configurations could not form.

The objection was not silly. Oppenheimer and Snyder had assumed exact spherical symmetry and zero pressure, and it was entirely reasonable to suspect that a realistic collapse — lumpy, rotating, with pressure — would bounce, or fragment, or shed enough mass to avoid the fate. The symmetry was doing the work, and nobody knew whether it could be removed.

Three things ended the argument within about five years. Finkelstein’s coordinates in 1958 showed that nothing happens at the horizon, so the surface stopped being a barrier the theory had to explain away. Quasars, identified in 1963 as objects radiating more than a galaxy from a region smaller than a solar system, made gravitational collapse an observational problem rather than a thought experiment. And Penrose’s theorem of 1965 proved that once a trapped surface has formed, a singularity follows — with no assumption of symmetry anywhere in the argument. That last result is what removed the escape route, and it is why it is a theorem about the causal structure of spacetime rather than about the behaviour of matter.

The name arrived after the physics, as it usually does. “Black hole” entered general use following a lecture by Wheeler in 1967, some twenty-eight years after the object it names was first calculated, and Wheeler had spent much of the intervening period arguing that it could not exist.

What it costs, and where the model stops #

The “radius” is not a radius. Nothing in this geometry lets a tape measure be run from the centre outward: there is no static centre to run it from, and the radial ruler stretch diverges on the way. What $r$ actually means is circumference divided by $2\pi$, which is a perfectly good label and is not a distance to anywhere.

Everything here assumes the mass is not rotating. Real collapsed objects rotate, and rotation changes the structure substantially: the horizon is not spherical, there is a region outside it in which standing still is already impossible, and the horizon radius for a maximally rotating object is half the Schwarzschild value. The static solution is the hydrogen atom of the subject, not the general case.

It assumes there is nothing else in the universe. Schwarzschild’s solution is exactly spherically symmetric and asymptotically flat, with one mass and nothing else. That is an idealisation of the same kind as an isolated point charge, and it is a good one for the same reason.

Everything that went in is gone from the description. The stationary solutions are fixed by three numbers — mass, angular momentum and charge — and nothing else survives outside the horizon. Two horizons of the same mass made from hydrogen and from encyclopedias are identical in every measurable respect, which is a far stronger uniqueness statement than anything else in physics offers, and which is the seed of the difficulty the entropy essay is about: information that has no way to come back out is information a unitary theory has lost.

It is a classical statement, and the classical statement is not the last word. A horizon has a temperature and an entropy, which nothing on this page can see, and a black hole radiates and eventually evaporates. Everything derived here is the leading behaviour with all quantum effects switched off.

The first sign that the classical picture is incomplete is thermal. A horizon has a temperature, which for a solar mass is $6\times10^{-8}$ K — far too small to matter for any large object, which is why the classical treatment survives, and not zero, which is why it is not the end. The lifetime that goes with it is the other half of the trouble: the temperature rises as the mass falls, so the process runs away rather than settling.

The ladder from here #

Later rungs on this anchor: proper time against coordinate time on the way in, and the exponential fading that replaces the “frozen” image; the rotating solution, with its ergosphere and its extractable rotational energy; the photon sphere at 1.5 Schwarzschild radii, where light orbits and which sets the size of the shadow an imaging array would see; the interior, where $r$ is a time coordinate; and the singularity theorems, which say that under stated energy conditions the breakdown at the centre is not an artefact of symmetry.

The neighbouring ladder is thermodynamics, which arrives here by a route nobody expected: the area of the surface derived on this page turns out to be an entropy, and the second law forced that conclusion before any quantum calculation supported it.