Aether Cosmology · analysis note · built 2026-11-04 07:10 UTC

Does geodetic VLBI see the ether wind Miller and Galaev reported?

New here? Start with the animated explainer: what VLBI actually measures, why a sky-fixed wind becomes a per-source scale factor, and an honest ledger of where this test is strong and where it is not the same measurement as Miller's.

A direct test of a Galilean light-speed anisotropy using 12 months of public IVS R1 sessions (2024). The claimed signal is a sidereal-fixed wind of ~10 km/s at mountain altitude (Miller 1933) rising from ~200 m/s at ground level (Galaev 2002), i.e. a light-speed anisotropy of ~3×10⁻⁵. VLBI group delays constrain the same quantity at the level of one part in 10⁹ per session.

Result. The fitted wind is 0.33 m/s with the direction free (components +0.17, +0.19, +0.21 ± 0.06, 0.05, 0.11 m/s, J2000 axes), and -0.07 ± 0.25 m/s along Miller's apex (RA 17.5 h, Dec +65°) after inflating the error for session-to-session scatter. No station shows a height-dependent wind: every one of 20 stations between 23 m and 1,409 m elevation is consistent with 0 ± 1 m/s where Galaev's law predicts 900 to 8,300 m/s. The residuals binned by sidereal time retain a 24-hour sidereal harmonic of only 1.1 ± 0.2 ps, about 30,000 times smaller than the 39 ns swings a 10 km/s wind would produce; at the picosecond level this is indistinguishable from ordinary troposphere and clock modelling error. An injected 10 km/s wind is recovered as 10,001 ± 0.4 m/s, so the pipeline is sensitive to the claimed effect by a factor of ~10⁴.

What is being tested, precisely

In a Galilean ether model, light travels at c relative to the ether, and the ether moves at velocity W relative to Earth. For a plane wave from a source in direction K arriving at two antennas separated by baseline b, the delay becomes τ = τ₀ (1 + W·K/c), where τ₀ = −b·K/c is the ordinary geometric delay. Because W and K are both fixed on the sky, the effect is a constant per-source scale factor on the delay, a dipole pattern across the sky with amplitude |W|/c. In the Earth frame that same term appears as the sidereal-periodic signal that Miller's interferometer would see; it is one phenomenon viewed from two frames. Special relativity predicts exactly zero for this longitudinal term (the first-order velocity terms in the consensus delay model are purely aberrational and are already in τ₀).

So the test adds three free parameters (W in m/s) to a standard geodetic VLBI solution and asks whether they are non-zero. A second fit pins the direction to Miller's apex and estimates only the speed. A third fit gives every station its own wind amplitude (Miller apex direction) to test Galaev's claim that the wind depends on height above ground.

Data and method

Data: one IVS-R1 24-hour session per month of 2024, 8 to 13 stations each, from the OPAR IVS data centre (vgosDB format, X/S group delays with ionosphere correction, Calc 11 theoretical delays and partial derivatives).
Standard parameters estimated per session: station clocks (quadratic plus 1-hour piecewise-linear, with reported clock breaks), zenith wet troposphere per station (1-hour piecewise-linear) and horizontal gradients, station positions (all but one). Earth orientation is held at the a priori values (it is degenerate with a free network rotation). Observations with quality code ≥ 5 and no analyst flag are used, then 3.5σ outlier clipping.
Residual floor: weighted RMS of post-fit residuals is 63 ps (median over sessions). Production IVS solutions reach ~30 ps with more elaborate modelling; the extra 30 ps is irrelevant at the scale of the claimed effect (hundreds of ns).
Calibration convention: the stored Calc theoretical already contains polar motion; this was verified empirically (adding the stored wobble contributions degrades the fit from 0.55 ns to 5.8 ns). Dry troposphere and cable calibrations are applied per the IVS recommendation flags.
Sensitivity check: a synthetic Miller wind was added to the observed delays of session r11136 and the full pipeline re-run.

injected	recovered (free direction, projected on apex)	recovered (apex pinned)
10,000 m/s	10,001.0 ± 0.4 m/s	10,000.8 ± 0.3 m/s
200 m/s	201.0 ± 0.4 m/s	200.8 ± 0.3 m/s
2 m/s	3.0 ± 0.4 m/s	2.8 ± 0.3 m/s

The ~1 m/s offset is the session's own measured value, present with or without injection.

Result 1 — ether-wind speed, session by session

Per-session estimates. Formal errors are 0.2 to 0.6 m/s; the session-to-session scatter is about three times larger (χ²/ν = 8.5), so the combined error is inflated by that factor. Even taken at face value the largest single-session value is 3 m/s.

session	stations	obs used	WRMS (ps)	W along Miller apex (m/s)	\|W\| free (m/s)	free-fit apex
20240102-r11136	8	2562	62	+0.79 ± 0.30	1.29	11.9 h, -66°
20240205-r11141	10	3615	51	+0.56 ± 0.24	0.41	6.7 h, -32°
20240304-r11145	10	2754	69	+0.15 ± 0.33	1.22	16.8 h, -55°
20240402-r11149	10	2970	41	-0.31 ± 0.25	0.81	14.7 h, -8°
20240506-r11154	12	5306	58	+0.35 ± 0.27	0.78	0.3 h, -41°
20240603-r11158	11	3163	79	+0.68 ± 0.35	1.18	13.2 h, -70°
20240701-r11162	10	3117	67	-1.77 ± 0.31	1.40	18.5 h, +15°
20240805-r11167	9	2809	49	-1.41 ± 0.34	1.26	18.7 h, -7°
20240902-r11171	9	2042	57	-1.70 ± 0.48	3.06	11.9 h, +80°
20241007-r11176	9	1741	65	+0.83 ± 0.37	1.48	12.2 h, -79°
20241104-r11180	10	3559	69	-0.49 ± 0.24	0.90	12.0 h, +56°
20241202-r11184	9	2276	65	+0.57 ± 0.32	0.65	5.6 h, +24°

For comparison: Miller's Mount Wilson value is 10,000 m/s; Galaev's radio value at 42 m is 1,414 m/s; the classical stationary-ether prediction (Earth's orbital motion through an ether at rest in the solar system) is 30,000 m/s. All three are excluded by more than 10³ standard deviations in every single session.

Result 2 — height dependence

Top: the height law Galaev fits through his own and Miller's values. Bottom: the per-station wind amplitude from VLBI, Miller apex direction fixed, combined over the sessions in which each station took part. Note the y-axis of the bottom panel spans ±6 m/s.

station	height (m)	sessions	measured (m/s)	Galaev prediction (m/s)
FORTLEZA	23	7	-0.45 ± 3.67	868
AGGO	37	5	-0.93 ± 0.92	1,130
HOBART12	41	10	-0.27 ± 0.81	1,190
NYALE13S	54	6	+0.40 ± 0.52	1,380
NYALE13N	54	8	-0.25 ± 0.53	1,380
ONSALA60	59	4	-0.97 ± 0.61	1,459
MEDICINA	67	2	-1.21 ± 0.84	1,562
SVETLOE	86	1	+0.59 ± 0.49	1,790
WARK12M	128	1	+1.36 ± 1.14	2,226
NOTO	143	2	+0.11 ± 0.71	2,369
ISHIOKA	164	4	-0.02 ± 0.82	2,554
KATH12M	189	9	+0.57 ± 0.69	2,762
SEJONG	195	9	+0.81 ± 1.16	2,804
YARRA12M	248	6	-0.04 ± 1.05	3,206
MATERA	543	5	-0.54 ± 0.48	4,932
WETTZELL	669	12	-0.12 ± 0.38	5,531
BADARY	822	1	+0.42 ± 0.53	6,192
ZELENCHK	1175	1	+0.60 ± 0.46	7,539
KOKEE	1177	9	+0.66 ± 0.79	7,544
HART15M	1409	12	-0.52 ± 0.65	8,332

Result 3 — sidereal periodicity in the residuals

Post-fit residuals (standard model, no ether term) binned by local sidereal time at each baseline's midpoint and stacked over all 12 sessions, compared with what a 10 km/s wind at Miller's apex would have added to the same observations. The fitted 24-hour sidereal harmonic in the residuals has amplitude 1.1 ± 0.2 ps. It is formally non-zero but sits at the level of known modelling residuals (troposphere, clocks, station loading), 30,000× below the prediction.

What this does and does not rule out

Ruled out: a Galilean (first-order) anisotropy of the speed of light of the size reported by Miller, Michelson–Pease–Pearson (6 km/s), or Galaev at any station height, and the classical stationary-ether expectation of 30 km/s. Under Galaev's own model the quasar wavefront crosses the baseline-projected distance hundreds of kilometres up, in what he calls the free ether, so VLBI should see the full wind, not the surface-attenuated one. The result is a null at the 1 m/s level, four orders of magnitude below the smallest claimed value.

Not tested: models that are empirically equivalent to special relativity for one-way propagation (Lorentz ether theory; an ether fully entrained with the Earth in the ECI frame as in Hatch's reading of GPS; Bennett's ALFA model, which makes the lab frame absolute). Those models predict a VLBI null and this analysis cannot distinguish them from relativity. A purely second-order effect of size (v/c)² ≈ 10⁻⁹ would appear at the ~20 ps level, which is at this pipeline's noise floor; separating it would need the full production-grade modelling and many more sessions.

Known limitations of this analysis: single analyst pipeline, no independent replication yet; Earth orientation held fixed (absorbed into station positions, harmless here); residual floor 2× above production solutions; one R1 session per month rather than the full year.

Reproducibility

Code and per-session results are in the quest repository haven/vlbi-ether-wind_QST-0000058 (Gitea, scouts-git). The per-session JSON files are also served here: 20240102 · 20240205 · 20240304 · 20240402 · 20240506 · 20240603 · 20240701 · 20240805 · 20240902 · 20241007 · 20241104 · 20241202 · summary. Run scripts/fetch_sessions.sh, then python -m vlbiwind.run_all, then scripts/build_report.py.

Source texts reviewed for the predictions: D. C. Miller, Rev. Mod. Phys. 5, 203 (1933); Yu. M. Galaev, Spacetime & Substance 2(5):211 (2001) and 3(5):207 (2002); R. Sungenis & R. Bennett, Galileo Was Wrong (2007), ch. 5 and 10.