What VLBI measures, and how an ether wind would show up in it
An animated walk from Miller's rotating interferometer to a planet-sized one, and an honest ledger of why the VLBI test is strong for the wind Miller, Allais and Galaev describe, and where it is not the same measurement.
1 · Miller's instrument: turn the arms, watch the fringe
Miller's interferometer sent light down two perpendicular arms, folded to 32 m each, and back. If the apparatus moves through an ether at speed v, the round trip along the arm pointing into the wind takes slightly longer than the round trip across it. Rotating the table swaps the arms every quarter turn, so the fringe pattern should slide back and forth twice per rotation. The size of the slide is second order: it goes as (v/c)², times the arm length over the wavelength.
Left: the interferometer turning in a wind (orange arrow). Right: the predicted fringe shift against rotation angle for Miller's arm length and wavelength. At 10 km/s the full swing is about a tenth of a fringe, which is roughly what Miller reported. The pattern repeats every half turn: that is the "second harmonic" the Shankland and Roberts critiques are about.
Because the horizontal projection of a sky-fixed wind changes as the Earth turns, the amplitude of this half-turn pattern should rise and fall with a period of one sidereal day. That sidereal curve is the periodicity Miller, Allais and Galaev all point to.
2 · The same wind, measured with a planet-sized interferometer
VLBI keeps the idea of comparing two light paths, but makes the two paths the routes from a distant radio source to two antennas on different continents. The source is a quasar: not because anything about quasars matters, but because it is far enough away that both antennas receive the same wavefront, and bright enough in broadband radio noise to correlate. Each antenna records the noise against a hydrogen-maser clock; sliding the two records against each other until they line up gives the arrival-time difference, the delay, to tens of picoseconds.
Wavefronts (blue lines) arrive from the source at upper right. The delay is the time between the front reaching antenna A and reaching antenna B, which is the projected baseline divided by the speed of the front. Drag the wind slider: in a Galilean ether the front crosses the Earth at c ± W·K, so the delay is stretched by a factor (1 + W·K/c). The numbers under the figure use a real 6,000 km baseline. Note the mismatch of scales: the effect is hundreds of nanoseconds on a delay of milliseconds, measured to picoseconds.
3 · Why Miller sees a sidereal wave and VLBI sees a dipole on the sky
These are the same physics seen from two vantage points, and nothing here depends on which body you take to be turning. Miller's apex and the VLBI sources are both fixed to the sky. Describe it as the Earth turning under a sky-fixed wind, or as the sky and its wind turning around a static Earth: either way the wind direction sweeps Miller's horizon once per sidereal day, because the sidereal day is simply the period of the sky relative to the ground. A VLBI source is fixed on the sky too, so for that source the quantity W·K never changes: the wind just multiplies that source's delay by a constant. Looking across all the sources observed in a day, the multipliers trace a dipole: largest toward the apex, zero around the great circle 90° from it, reversed on the far side.
The dots are the 263 sources actually observed in the twelve 2024 sessions. Colour and size show the per-source delay scale factor a wind at the chosen apex would impose (blue toward, orange away). The right-hand trace is what Miller's Mount Wilson horizon would see from the same wind over one sidereal day. The fit on the main page asks: is there any dipole like this in the data? The answer was a dipole of 0.3 m/s, against the 10,000 m/s shown here by default.
3½ · Does this assume a rotating Earth? Is a sidereal signal lost in sidereal-locked geometry?
Both are reasonable worries for a geostatic reading, and the answers are mechanical rather than philosophical.
The model uses only relative motion. Every term in the delay model is built from the orientation of the baseline relative to the source direction at each instant, and the velocity of each antenna relative to the frame in which the sources are fixed. The Earth-orientation numbers it takes from the IERS (precession, nutation, UT1, polar motion) are a map between the ground frame and the sky frame. In the standard picture they describe how the Earth turns; in a geostatic picture they describe how the sky turns about the Earth. The numbers are the same, and so is every delay computed from them. The analysis never asks which one is "really" moving, and could not tell if it tried.
A rigidly co-rotating ether is already in the model, so it is not what was tested. If the ether turns with the sky around a static Earth, then at each antenna the ether moves at Ω × r relative to the ground, a few hundred metres per second. That is exactly the station-velocity term the standard model carries, which in the standard picture is the antenna moving through a non-rotating frame. The two descriptions give identical arrival times at first order; this is the same equivalence Sungenis and Bennett invoke for Michelson–Gale. So the geostatic daily rotation is neither assumed nor tested here. What the fit adds on top is a translation: a wind with a fixed apex on the sky, which is what Miller claimed (RA 17.5 h, Dec +65°), and which in a geostatic picture is the universe and its ether drifting past the Earth toward that point, or the Earth drifting through them. Only the relative velocity enters, so the result, 0.3 m/s, applies to both readings.
Sidereal locking is a feature, not a degeneracy. The geometric delay of a given source varies with sidereal time through b·K. The wind term varies with sidereal time through the same factor, so it cannot be told apart from geometry by its time signature alone. It is told apart by its sky signature: a per-source multiplier that follows a dipole across the sixty sources observed each day, largest toward the apex, zero on the ring 90° away, reversed beyond. No ordinary parameter has that shape. Clock errors are the same for every source in a scan; a source-position error shifts a source sideways on the sky, which changes its delay with a different time signature; a station-position error is fixed to the ground, not the sky. The injection test is the proof: a 10 km/s wind added to real observations came back as 10,001 m/s after all the geometric parameters were free to absorb it.
4 · Where the signal lives, and where the shielding argument bites
Miller and Galaev both argue that metal chambers and massive buildings slow the ether inside them, which is why their outdoor or lightly housed instruments saw a wind and the chamber experiments did not. For an interferometer whose whole light path is inside the box, that argument has teeth. The animation below shows what fraction of the VLBI light path an enclosure can touch.
The wavefront covers the projected baseline, thousands of kilometres, hundreds of kilometres above the ground, before it enters either dish. Inside the dish, feed and cables the signal travels perhaps 10 to 30 m. Even if the ether were completely at rest inside that enclosure, the delay could change by at most a few picoseconds, out of the hundreds of nanoseconds the free-space crossing would produce. Shielding at the antenna cannot hide the effect. Only an ether entrained out to thousands of kilometres altitude could, and that contradicts Galaev's own measured height law, which has the wind already at 10 km/s by 1.8 km and reaching its full value by a few hundred kilometres.
5 · The ledger: good and bad for measuring the claimed wind
Why VLBI is a strong test of the Miller / Allais / Galaev wind
- It is first order. The wind enters as W/c, not (W/c)². Galaev himself argued first-order one-way methods are 10⁴ times more sensitive than the Michelson scheme. This is one.
- The baseline is 10⁵ times Miller's arm. 6,000 km against 64 m. A 10 km/s wind moves the fringe by 700 ns of delay here, against a tenth of a fringe there.
- Direction is swept for free. The Earth turns the baseline and sixty sources cover the sky every day, so the dipole is sampled in all directions, every session, without touching the apparatus.
- Clocks are not synchronised by light. The old objection to one-way tests does not apply: clock offsets are free parameters but they are the same for every source, while the wind term differs source by source. The injection test shows the fit separates them.
- The sky-fixed geometry is exactly the geocentric claim. Miller's apex is a point on the celestial sphere; VLBI sources are points on the celestial sphere; the test is phrased in the same coordinates the claim is.
- Public data, reproducible. Anyone can rerun it from the IVS archive.
Where it is not the same measurement, and what it cannot say
- It samples the free ether, not the surface. The wavefront crosses the baseline far above the ground. If someone holds that the wind exists only in a thin surface layer and the ether is entrained with the Earth above it, VLBI tests that model, not Galaev's. Galaev's own height law says the opposite, but the point stands as a model choice.
- Miller's device was second order. A model with no first-order one-way effect and only a (v/c)² term predicts a VLBI signal of about 20 ps for 10 km/s, right at this pipeline's noise floor. That fit has not been run yet; it would need the full production modelling and many sessions.
- Models equivalent to relativity for one-way light are untouched. Lorentz ether theory, a fully entrained ether in the Earth-centred frame as in Hatch's reading of GPS, and Bennett's absolute-lab-frame model all predict this null. VLBI cannot separate them from relativity.
- Not a Miller replication. It does not reproduce his instrument, his altitude, or his second-harmonic readout. It asks whether the wind he inferred exists in the one place a wind of that size could not hide.
- Single pipeline, 60 ps floor. Production IVS solutions reach 30 ps; this one has not been independently replicated. Irrelevant at 700 ns, relevant for the second-order question.
6 · The scale of things
Everything on one logarithmic ruler. What the analysis has to model (clocks, troposphere) sits above the claimed signals; what it can resolve sits four orders of magnitude below them.
Sources for the claimed values: Miller, Rev. Mod. Phys. 5, 203 (1933); Galaev, Spacetime & Substance 2(5):211 (2001) and 3(5):207 (2002); Allais (1998) on Miller's data; Sungenis & Bennett, Galileo Was Wrong (2007), ch. 5 and 10. Code: quest haven/vlbi-ether-wind_QST-0000058.