Count the ways a spin network can pierce a horizon and the Bekenstein-Hawking entropy comes back out
Bekenstein argued in 1972 that a black hole must carry entropy, and Hawking supplied the coefficient two years later by showing that a hole radiates at a temperature set by its surface gravity. The result is one quarter of the horizon area in Planck units. It is a thermodynamic statement about an object made of pure geometry, which means it is a statement about the microscopic structure of geometry, and it has been the standard benchmark for quantum gravity ever since. Any candidate theory has to answer the question of what exactly is being counted.
Loop quantum gravity has an answer, and the shape of it follows directly from the last few lessons. A horizon is a surface. A surface in this theory gets its area from the points where links of a spin network pierce it. So a horizon is a boundary with punctures on it, each puncture carrying a spin, each spin contributing 5.17 Planck areas or more, and each puncture carrying orientation degrees of freedom that the boundary theory keeps track of. Fix the total area, count the configurations, take the logarithm. Ashtekar, Baez, Corichi and Krasnov did exactly this in 1998, and the answer is proportional to the area.
That last sentence is the honest one, and the rest of this page is about the gap between it and the sentence people sometimes write instead. The count gives proportionality. It does not give the number 1/4, because the constant of proportionality depends on the Barbero-Immirzi parameter γ, which the classical theory leaves free. What actually happens is that γ is chosen so that the coefficient comes out right. The coefficient is an input. Proportionality to area rather than volume, and a concrete countable set of microstates, are the outputs, and they are worth having.
A horizon with the bulk spin network ending on it. Add punctures, change their spins, fill the boundary to a chosen area, and compare two numbers: the Bekenstein-Hawking entropy A/4, and the count of orientation states the punctures carry. Then drag γ until they coincide, and notice what kind of act that was.
Watch the ratio in the last box. Whatever you do to the punctures, the counted entropy stays proportional to the area, and that proportionality is the real result: a horizon made of graph endpoints has an entropy that scales with its surface rather than its volume. Now drag γ. The ratio slides, and at γ = 0.1274 it lands on 1 for any all-spin-half horizon, because √3πγ = ln 2 there. Nothing was predicted at that moment. A free parameter of the classical theory was tuned so that a count would match a result Bekenstein and Hawking obtained by other means, and that tuned value then has to be carried into every area and volume spectrum elsewhere in the theory.
The textbook definition of a black hole is global: the event horizon is the boundary of the region from which light cannot escape to future null infinity. That definition is useless here, and not only for computational reasons. Deciding where an event horizon is requires knowing the entire future of the spacetime, which is exactly the thing a canonical quantum theory built on spatial slices does not have. Ashtekar and collaborators replaced it with the isolated horizon: a null surface with no flux through it, characterised entirely by local conditions on the geometry there. An isolated horizon can be identified on a single slice, and that is what makes the quantum treatment possible at all.
The technical structure that comes out of this is worth stating even in outline. The phase space of general relativity with an isolated horizon as an inner boundary acquires a boundary term, and that term is a Chern-Simons action for a U(1) connection living on the horizon, with a level fixed by the horizon area. Chern-Simons theory on a sphere with punctures has a finite-dimensional Hilbert space, and its dimension is a combinatorial function of the punctures and their labels. So the bulk spin network fixes where the punctures are and what spins they carry, and the horizon theory tells you how many states sit on top of that data. The entropy is the logarithm of that number at fixed area.
One detail matters for the arithmetic and is easy to miss. The punctures are not free: their labels have to be compatible with the horizon being a horizon, which appears as a projection constraint on the boundary states. Different papers have imposed it differently, and the value of γ that comes out has been revised more than once as the counting was tightened. The demo below performs the count without that constraint, which is a simpler and cleaner thing to show, and lands at a different number. It is labelled as such.
The logarithm of the number of horizon states, plotted against the horizon area. The curve is a straight line, which is the result: entropy proportional to area. The dashed line is the target A/4. Drag γ to tilt the count onto it, or press the button and let a bisection find the value for you.
Two things are happening here and they are worth separating. The first is that the green curve is a straight line: the number of horizon states grows exponentially with the area, so its logarithm grows linearly, so entropy is proportional to area rather than volume. Nothing about the construction forced that, and it is a genuine result. The second is that the line has the wrong slope until you drag γ, and dragging γ is exactly what fixing the Barbero-Immirzi parameter means. The bars show that spin 1/2 dominates the count without monopolising it, which is why the spin-half-only shortcut gives 0.1274 while the full count gives a larger value. Two honest caveats: this count omits the horizon projection constraint, so it lands near 0.25 rather than the published 0.2375, and log N / A is only a chord, still creeping up towards the local slope shown beside it. Neither caveat touches the conclusion, which is that the coefficient 1/4 is an input to this calculation and not an output of it.
Take the simplest version of the count, in which every puncture carries the smallest spin. A horizon of area A then needs N = A / (4√3 π γ) punctures, each contributing a factor of 2 to the number of states, so the entropy is N ln 2. Setting that equal to A/4 gives γ = ln 2 / (π√3), about 0.1274. Every symbol in that derivation is fixed by the theory except γ, and γ is the one being solved for. The calculation is a measurement of γ against a known answer, not a derivation of the known answer.
The fuller count lets every spin appear and weights each puncture by its 2j+1 orientations, which obeys a simple recursion in the area and is what the demo above solves. Allowing the higher spins raises the number of states at fixed area, so the γ needed to bring the density down to 1/4 is larger. The demo lands near 0.2438, because it omits the horizon projection constraint. Imposing that constraint is what produces the value quoted in the literature, γ ≈ 0.2375, found by Domagala and Lewandowski and by Meissner in 2004 after earlier papers had reported a different number. The history is instructive: this is a hard combinatorial problem whose answer has moved as the formulation was sharpened, and the SU(2) treatment of the boundary rather than the U(1) one moves it again.
So what should a reader take from this? Three things, in order of how solid they are. The entropy is proportional to the area rather than the volume, and that did not have to happen: it happened because the horizon degrees of freedom live on the boundary surface, which is a structural feature of the theory and not an assumption about black holes. The microstates are concrete, countable and finite, which is more than most approaches offer. And the construction is not restricted to special cases: it works for rotating and charged horizons and for cosmological horizons, where several string-theoretic counts apply only to extremal or near-extremal configurations. That generality is a real point in its favour.
Against that, the standard objection is not that γ has to be fitted but that fitting it here is suspiciously specific. A single real number, fixed once by matching one thermodynamic coefficient, then propagates into the area gap, every volume eigenvalue, and the critical density at the bounce. There is something appealing about that rigidity, since it means the theory has no further dials. There is also something uncomfortable about it, since it means one comparison with a semiclassical result carries all the calibration. Whether the same γ that fits Schwarzschild also fits everything else, at every order, is a live question, and there are results suggesting the required value depends on the species of matter you allow, which would be bad.
If the area spectrum is discrete then the hole cannot shrink smoothly. It steps between eigenvalues, and each step emits a definite energy. Watch the resulting lines sit on top of the smooth Hawking curve, then watch them disappear as soon as you ask for anything realistic.
Drag the mass slider first, and notice that nothing moves. Both the step energy and the Hawking temperature fall like one over the square root of the area, so in these units a Planck-mass hole and a solar-mass hole have their lines in exactly the same places. That scale invariance is what made the Bekenstein and Mukhanov idea interesting. Then drag the other two sliders and watch it die. Allowing a transition to move more than one puncture fills in line after line, and any finite resolving power smears the survivors into the smooth curve underneath. Three further caveats, all of which cut the same way: the line heights here are read off the blackbody envelope rather than computed, because the theory does not yet supply transition amplitudes; the envelope itself omits greybody factors; and for anything heavier than about 4.5e22 kilograms the temperature is below the microwave background, so the hole is growing rather than shining. The discrete line spectrum is a suggestion the theory makes, not a measurement anyone is close to making.
Bekenstein and Mukhanov noticed that a quantised horizon area implies a quantised emission spectrum, and the arithmetic has an attractive feature. The energy released when the horizon drops by one area quantum and the Hawking temperature both scale like the inverse square root of the area, so their ratio is a pure number, independent of the mass of the hole. A stellar-mass horizon and a Planck-mass horizon would put their lines in the same places, measured in units of their own temperatures. If black hole radiation were ever observed at all, this structure would not be hidden by the size of the object.
Everything else about the situation is worse. The area spectrum is not a simple ladder: the number of ways to change the horizon by roughly a given amount grows quickly once transitions may move more than one puncture, so the lines crowd. The theory does not currently supply the transition amplitudes that would say how bright each line is, because that is a question for the Hamiltonian constraint and the Hamiltonian constraint is the module’s standing open problem. The smooth envelope itself is only the naive blackbody, with the greybody factors from propagation through the exterior geometry left out. And the practical numbers are hopeless: a hole light enough to be hotter than the microwave background is lighter than a small asteroid, no such object is known to exist, and the photons a stellar-mass hole emits are individually far below any detection threshold.
The right description of the line spectrum is therefore suggestive rather than predictive. It is the kind of thing a theory of quantum geometry ought to imply, it follows from the area spectrum with very little extra input, and it is nowhere near testable. That combination is the recurring situation for this whole programme, and it is what the final lesson is about: given a granular spacetime, what could actually reach an instrument, and what has already been ruled out by the observations we have.