Relativity's Second Postulate Verified | Generated by AI

Home 2026.07

Question: Yin Wang’s substack post (“什么是’光速不变’”) argues that Einstein’s relativity relies on two logically distinct claims — (1) light speed is independent of the source’s motion, and (2) light speed is independent of the observer’s motion — and that (2) is an unverified, counter-intuitive assumption analogous to how sound/water-wave speed does change relative to a moving observer. Is he right?

Answer:

He’s right about the logical distinction, wrong about the empirical status of claim (2). This is a very old objection (it’s essentially the historical fight between Einstein’s postulate and Ritz’s “emission theory” from 1908-1913), and it was settled by experiment, not by philosophy. Let me break down where the argument holds and where it breaks.

What he gets right

The two postulates really are logically independent, and Einstein’s 1905 paper does need both:

Sound and water waves only satisfy the first, not the second — because they have a physical medium (air, water) that defines a preferred rest frame. Relative to a moving observer, wave speed does shift, exactly as Yin Wang describes (and as you’d derive from simple Galilean addition: v_relative = v_wave - v_observer).

Where he’s factually wrong

His claim is: “the observer-independence of light speed 未经实验证实” (unverified by experiment). This is false — it is arguably the more rigorously tested of the two postulates, not the less tested one. The reason it feels untested to a casual reader is that people usually only cite Michelson-Morley for “no ether,” but there’s a specific chain of experiments built to isolate exactly the observer-motion question:

Experiment What it isolates Result
Michelson-Morley (1887) Does light speed depend on the direction the observer/apparatus is moving through space (Earth’s orbital velocity)? Null — no directional dependence found
Kennedy-Thorndike (1932, modern optical-cavity versions) Does light speed depend on the magnitude of the observer’s velocity (using unequal arm lengths so Earth’s changing orbital speed through the year would show up)? Directly tests whether light speed is independent of the velocity of the apparatus in different inertial frames; modern resonator versions constrain violations to parts in 10⁻¹⁷–10⁻¹⁸
Ives-Stilwell (1938, modern storage-ring versions) Transverse Doppler / time dilation as seen by a moving observer relative to a source Confirms relativistic Doppler formula, which is observer-frame invariance under a different guise
De Sitter double-star (1913) + Brecher (1977) X-ray version Source-velocity dependence specifically (rules out c' = c + kv emission theories) De Sitter showed Ritz’s variable-speed-of-light theory would have predicted binary star orbits appearing more eccentric than observed, and this was confirmed by Brecher in 1977

So Michelson–Morley + Kennedy–Thorndike together are precisely the pair of experiments that test “does the observer’s own motion change the measured speed of light” — MM tests directional dependence, KT tests magnitude dependence. Recent GRAAL-ESRF work using accelerated 6 GeV electrons pushed the Kennedy-Thorndike bound to an accuracy better than existing KT-limits by three orders of magnitude, yielding 7.1×10⁻¹². Both null-result classes are consistent with SR to extraordinary precision, and inconsistent with any Galilean-style “medium-relative” picture.

Why the intuition fails

Your intuition (and Yin Wang’s) about sound is correct because sound has a medium. The entire content of relativity’s second postulate is the claim “light in vacuum has no medium” — i.e., there’s no preferred rest frame to be moving relative to. Michelson-Morley is literally the experiment that checks this: if there were an aether, Earth’s ~30 km/s orbital motion through it should show up as a seasonal, direction-dependent shift in interferometer fringes. It doesn’t, to 1 part in 10¹⁷ in modern rotating-resonator versions.

The math, not just the wave analogy

Under a Galilean transform, velocities add linearly:

def galilean_velocity_add(u, v):
    # observer moving at v, wave/object moving at u in ground frame
    return u - v  # naive subtraction, matches sound intuition

print(galilean_velocity_add(343, 343))  # sound: 0 relative speed if you match its speed

Under a Lorentz transform, they don’t — and light speed is a fixed point of the transformation regardless of v:

def lorentz_velocity_add(u, v, c=299792458):
    # relativistic velocity addition formula
    return (u - v) / (1 - u*v/c**2)

c = 299792458
for v in [0, 1e7, 1e8, 2.9e8]:          # observer velocities, up to near-c
    print(v, lorentz_velocity_add(c, v, c))
# -> always prints c, no matter what v is

That invariance under boost isn’t an assumption bolted onto the theory for convenience — it’s the direct algebraic consequence of deriving the Lorentz transform from “laws of physics look the same in every inertial frame” + “light speed is c in at least one frame,” and then requiring self-consistency (this is literally the derivation in Einstein’s 1905 §3, and it’s what MM/KT experimentally validate). If postulate 2 were false, you’d expect lorentz_velocity_add to return something other than c depending on v, exactly like the Galilean case does for sound — and that’s the specific thing Kennedy-Thorndike-type experiments have ruled out to 10⁻¹⁷ precision.

Bottom line

Yin Wang’s conceptual move — separating “independent of source” from “independent of observer” — is a genuinely useful pedagogical clarification that most popularizations blur. But his specific factual claim, that the observer-independence is untested, is wrong; it’s the more heavily and more precisely tested half of the two postulates, via a distinct family of experiments (MM, KT, Ives-Stilwell) built for exactly that purpose, going all the way to modern optical-resonator tests at the 10⁻¹⁷–10⁻¹⁸ level. This is the kind of claim in his newsletter series you should stress-test against primary experimental literature, not just his internal reasoning about Einstein’s paper — the paper’s internal logic and its experimental confirmation are two separate questions, and he’s conflating “I find this hard to derive cleanly from Einstein’s exposition” with “this is unverified.”

References:

Back Donate