The Speed of Light Is a Choice We Made

299,792,458 metres per second. The digits are exact — no error bar, no last-decimal-place uncertainty, no ongoing project to pin them down more finely. Since 1983 the metre has been defined from c, which means the most famous constant in physics is now literally a unit conversion [S3]. We didn't measure our way to that precision. We voted on it.

That is the small surprise. The larger one is that nobody has ever measured the speed of light at all — not in the sense most people picture. Every experiment from Ole Rømer's 1676 observations of Jupiter's moons forward measures a round trip: a pulse goes out, bounces off a mirror or a moon or a wheel tooth, comes back. Divide total distance by total time, call the answer c. Whether the light traveled at the same speed on the way out as on the way back is a question the round trip cannot ask.

To ask it, you need two clocks — one at the start, one at the finish — and you need them already synchronized. Which is where the trouble starts, and where Einstein noticed it.

In §1 of "On the Electrodynamics of Moving Bodies," 1905, he writes: "We have not defined a common 'time' for A and B, for the latter cannot be defined at all unless we establish by definition that the 'time' required by light to travel from A to B equals the 'time' it requires to travel from B to A" [S2]. The German word he uses is Festsetzung — a stipulation, a setting-down, a coordinative choice [S2]. He is not announcing an experimental result. He is announcing that he is going to decide the outbound and return times are equal, because no observation can decide it for him.

Hans Reichenbach formalized the gap in 1928. Synchronization, he showed, depends on a free parameter ε that can take any value strictly between 0 and 1; Einstein's familiar convention is just the special case ε = 1/2, and each inertial frame could in principle pick its own [S1]. ε = 1/2 says light is isotropic — same speed in every direction. ε = 0.9 says it tears off in one direction and crawls back. The universe is fine with all of them.

A number with nine significant figures, an internationally ratified definition, and zero empirical content.

The standard objection writes itself: fine, just walk a clock from A to B very slowly and you've sidestepped the whole problem — no light signal involved. The objection is wrong, and provably so. Take the limit as the transport velocity goes to zero; the resulting synchronization is mathematically identical to Einstein's ε = 1/2 convention [S6]. The trick fails because the time-dilation factor you'd use to argue "no drift accumulated" already presupposes that c is isotropic [S6]. Slow clock transport doesn't escape the assumption; it smuggles it in through a different door.

GPS is the next reflex. Atomic clocks synchronized across continents to nanoseconds — surely that pins down the one-way number? It does not. The GPS range equation propagates signals at constant c in an Earth-Centered Inertial frame, and that isotropy is baked into both the synchronization protocol and the pseudorange math [S5]. Any test of one-way c performed with GPS-synced clocks confirms ε = 1/2 by construction [S5]. The verification is circular before the first photon leaves the satellite.

Mössbauer rotor experiments? Round trips. Pulsar timing? Round trips, just with very large geometry. The pattern holds. Anderson, Vetharaniam and Stedman's 1998 review in Physics Reports — 88 pages, the modern canonical treatment — works through the experimental catalog one by one and shows each claimed measurement of one-way c presupposes the answer somewhere in its setup [S3]. Their deeper move is to recast synchrony choice as a gauge transformation, in the same family as the gauge freedoms of electromagnetism: ε is not a hidden physical fact awaiting better instruments, it is a coordinate convention with no observable consequences [S3].

The radical consequence is the one to sit with. An "anisotropic universe" in which light travels instantaneously in one direction and at c/2 in the reverse direction is mathematically equivalent to ours and produces identical predictions for every experiment ever performed [S3]. The two universes are not two universes. They are one universe under two coordinate labels.

This is not the unanimous view. In 1977 David Malament published a theorem widely read as proving that ε = 1/2 is the unique simultaneity relation definable from the causal structure of Minkowski spacetime — which, if accepted, would knock conventionalism over [S1]. The reply, articulated by Sarkar and Stachel in 1999 and absorbed into the AVS review, is that Malament's proof tacitly assumes simultaneity must be invariant under temporal reflection; drop that assumption and an infinite family of non-standard simultaneities walks back through the door [S4][S3]. The professional philosophy-of-physics community has not converged [S1][S4]. The argument is, by any honest reading, still live.

Most working physicists shrug. They know ε = 1/2 is a convention. They also know that any other choice makes Maxwell's equations and the Lorentz transformations uglier without changing a single prediction [S3]. Convention plus elegance plus no observable difference equals: pick the pretty one and stop talking about it.

So what is the durable thing here? Not "physicists missed something." Not "we need a better experiment." The durable thing is a distinction the conventionality argument forces you to make, between two kinds of unknown.

There is the unknown that is a gap in our knowledge — the mass of a neutrino circa 1995, the temperature at the centre of Jupiter, what killed off the megafauna. The universe has an answer; we don't yet. Patience and instruments will close the gap.

And there is the unknown that is not a gap at all, because the universe does not store the answer. "Are these two distant events simultaneous?" turns out to belong to that second category [S1][S3]. There is no fact of the matter, only a coordinate choice we agree on so the rest of the bookkeeping works out. The one-way speed of light is the most precise example we have of this second kind: a number with nine significant figures, an internationally ratified definition, and zero empirical content.

That is a stranger thing to learn than "we don't know c precisely." We know it exactly. We chose it.