The Profligate Render: Past-Incompleteness, Eternal Inflation, and What Ray Tracing Actually Costs

What the theorem actually says

Borde, Guth and Vilenkin (2003) proved that a spacetime whose average expansion rate along a past-directed null or timelike geodesic is positive cannot have that geodesic extend infinitely into the past. The geodesic is incomplete: followed backwards, it ends after finite affine parameter. The proof is a kinematic argument about the Hubble average, not a dynamical one, which is why it holds without assuming an energy condition and without assuming Einstein's equations. That generality is the whole reason it is quoted so often.

It is also the reason it proves less than it is usually made to carry. Past-incompleteness is a statement about geodesics terminating, not about a moment of creation. A boundary can be a genuine origin, a breakdown of the classical description where quantum gravity takes over, a junction to a prior contracting phase, or an artefact of extending an averaged condition into a regime where the averaging fails. Vilenkin has said as much repeatedly: the theorem does not establish a beginning of existence, it establishes that the classical description cannot be extended arbitrarily far back. Any model that violates the averaged-expansion hypothesis — emergent, cyclic, or bounce scenarios with a contracting or static past epoch — sits outside the theorem's reach and is not refuted by it.

So the entry in the ledger is precise: a theorem about the limits of a description, frequently promoted to a theorem about the limits of being. The promotion is where the argument stops being mathematics.

Eternal inflation as the volume's overproduction thesis, restated

Eternal inflation is the second piece, and its standing is different in kind. It is a plausible consequence of a broad class of inflationary potentials rather than a theorem: once inflation begins, the exponential growth of inflating volume outruns the rate at which regions exit, so inflation never globally ends and bubbles nucleate indefinitely. It is not established. There is no observed bubble collision, no confirmed signature in the microwave background, and — the deeper problem — no agreed measure with which to convert an infinite ensemble into probabilities. The measure problem is not a technicality; without a measure, 'most bubbles are sterile' is not yet a quantitative claim.

Held at that standing, it fits the profligacy theme this volume has been assembling since Chapter 41. Three generations of fermions when one would suffice for chemistry. A cosmic web of filaments and voids in which the habitable fraction is negligible. Zeros of a zeta function whose statistics match an operator nobody has produced. Eternal inflation, if it holds, is the extreme case: structure generated at a rate that has nothing to do with what any observer requires.

The philosophical pincer the post describes — a designer who would not need the waste, a chance hypothesis that still inherits the finely structured laws which make any of it possible — is a real pincer, and it is not a physics result. Both horns are readings of the same equations, and the equations are silent about which reading to prefer. This volume's rule applies unchanged: the excess is the datum, and the interpretation of the excess is a separate, unpriced claim.

GPU ray tracing: the proxy, and exactly where it breaks

Now the rendering analogy, because it is the most seductive version of the profligacy story and it fails in a way worth documenting. Real-time graphics is an engineering discipline built entirely around not computing what will not be seen. Frustum culling discards geometry outside the view. Occlusion culling discards what is hidden behind other geometry. Level-of-detail systems swap in coarser meshes with distance. Path tracing is run backwards — rays are cast from the camera, not from the lights — because forward light transport wastes almost every photon. Importance sampling concentrates samples where the integrand carries weight, and denoisers reconstruct a plausible image from far fewer samples than convergence would need. The entire stack is a monument to computational parsimony under observation.

The temptation is obvious and it has been sold in popular books for years: perhaps the universe does the same, rendering only what is measured, with quantum indeterminacy as the unrendered state and measurement as the draw call. It reads as an explanation for wave-function collapse and it costs nothing to say.

It also gets the direction of the evidence backwards, and this is the chapter's actual content. A culling engine is recognisable precisely because it is cheap where nobody is looking. Cosmology shows the opposite everywhere we have checked: the microwave background is isotropic to one part in a hundred thousand across regions that never exchanged signals with each other or with us; the large-scale structure is statistically homogeneous in directions no observer occupies; the physics of a quasar twelve billion light years away, reconstructed from spectra, is the same physics as here, at full fidelity, for an audience that did not exist when the light left. Whatever the universe is doing, it is not conserving effort in unobserved regions. The correct rendering metaphor is not a game engine. It is an offline renderer computing every pixel of a frame nobody has requested.

There is one narrow place where the analogy earns something real, and Chapter 40 has already located it. The parsimony is on the observer's side, not the world's. Retinal compression, efficient coding, predictive processing, and the metabolic budget of cortex are a genuine level-of-detail system — lossy, fitness-tuned, discarding almost everything the environment offers. Perception is the culling engine. The universe is the uncompressed asset. That inversion is the useful sentence in the whole analogy, and it is also the reason the metaphor cannot explain measurement: a graphics pipeline's decisions depend on where the camera is, and no experiment has ever shown physical law depending on where an observer is.

The ledger entry

Three items, three different columns, and the chapter's only real claim is that they do not belong in the same one. Past-incompleteness of averaged-expanding spacetimes: theorem, tight hypotheses, widely over-read. Eternal inflation: unconfirmed mechanism with an unsolved measure problem, and the strongest available statement of the overproduction thesis if it survives. Render-on-demand cosmology: proxy with the evidence pointing the wrong way, retained here as a worked example of a metaphor whose failure is more informative than its success would have been.

What survives the sorting is the volume's recurring shape. The description ends before the thing described does — that is the theorem. The thing described generates far more than any observer needs — that is the profligacy. And the smallness of what reaches us is a fact about us. None of that requires a decision about whether the excess means anything, and this chapter does not make one.