The Moon and the Magnitude-6 World: A 53-Year Global Null

Hed: Does the Moon trigger strong earthquakes? Dek: Across 4,729 independent M≥6 earthquakes from 1973 to 2026, lunar phase, syzygy, tidal phase, and tidal amplitude leave no measurable fingerprint on when the world's large earthquakes happen. The test bounds any global lunar modulation of M≥6 occurrence below about 4% — and, honestly, cannot see the smaller effects that specialized fault-resolved studies have reported. Both statements are the finding.

TerraPulse — DP-008 V1. Pre-registered at docs/scope-lunar-syzygy-seismicity.md (frozen and ratified 2026-07-02, before any data was pulled).

Abstract

We test whether the gravitational tide raised by the Moon and Sun modulates the occurrence of the world's large earthquakes. Using the full USGS global catalog of magnitude-6-and-greater events from 1973 to 2026 — 4,729 independent mainshocks after deduplication and Gardner–Knopoff aftershock removal — we compute, for each earthquake, the lunar synodic phase, the syzygy (new/full-moon) phase, the semidiurnal tidal phase, and the combined lunar-plus-solar equilibrium tidal potential at the hypocenter, all from a deterministic astronomical ephemeris. We find no non-uniformity in any of these quantities, in any depth stratum, at any magnitude threshold. The single strongest cell in a 72-test grid reaches only p = 0.009, and nothing survives a Bonferroni correction. A weak concentration present in the raw catalog (syzygy p = 0.09) vanishes after aftershock removal (p = 0.68), confirming it was clustering, not tides. The result is a clean upper bound: any global lunar/tidal enhancement of M≥6 rates is smaller than about 4%. This bound does not contradict the smaller (~1%), fault-resolved tidal-triggering signals reported for shallow thrust faults; those operate on a quantity and a fault subset our coarse global test does not isolate, and lie below our detection floor.

Why this question refuses to die

"Does the Moon cause earthquakes?" is among the oldest statistically tested questions in seismology, argued since the 1890s. The physics is not crazy: the same tide that lifts the ocean by a meter also flexes the solid Earth, and the resulting stress — a few kilopascals — is real. The debate is whether that tiny, cyclic stress is ever the last straw that tips a fault already near failure. The literature is genuinely split. Most global studies find nothing. A minority find small, specific effects: Cochran et al. (2004) reported that shallow thrust faults are measurably more likely to slip when the resolved tidal shear stress on the fault plane is high; Tanaka (2004) and Métivier et al. (2009) reported related small correlations. None of these is a large effect, and all are contested.

What has been missing is a single, pre-registered, global test at decisive scale with the aftershock problem handled honestly. That is what this paper provides — and what it does not provide, stated up front.

The data

Earthquakes. The USGS ComCat global catalog, magnitude 6.0 and above, 1973-01-01 to 2026-06-27, as ingested by TerraPulse. Raw pull: 7,706 records. After the canonical dedup (collapsing the same event re-reported across feed windows, keeping the largest magnitude per minute-and-location): 7,477 events.

Completeness. The M≥6 catalog is complete across the whole window. The annual count is flat at ~140 events per year with no long-term trend, and the frequency–magnitude distribution above M6.0 is textbook Gutenberg–Richter with b = 1.01. (Caveat, stated plainly: TerraPulse's feed catches M≥6 comprehensively but under-samples smaller events, so the full magnitude distribution has a 5.6× discontinuity at exactly M6.0. That is an ingestion boundary, not a real completeness-magnitude rollover; only the M≥6 branch is used, and only it is clean. We do not present the sub-6 distribution as a completeness curve.)

The sky. No data were fetched for the tidal driver. For each earthquake's exact time and location we compute the geocentric positions of the Moon and Sun from the astropy ephemeris (itself a fit to lunar-laser-ranging and radar). This is measured reality in the strict TerraPulse sense: the position of the Moon is a measurement, and the tidal potential is a deterministic transform of it — the same category as solar zenith angle or day-of-year, and categorically not a forecast, reanalysis, or model of a chaotic system.

What we computed, and how

For every event we derived four phase/amplitude quantities:

Lunar synodic phase φ — where the Moon is in its cycle (0° = new, 180° = full).
Syzygy phase (the doubled angle 2φ) — the physically correct test for a new-and-full moon excess. A syzygy signal peaks at both new and full moon, which is symmetric and therefore invisible to an ordinary test on φ; doubling the angle folds the two peaks together so the test can see them.
Semidiurnal tidal phase θ — where in the ~12.4-hour tidal cycle the event fell. (Flagged: this is the quantity most confounded by the solid-Earth tide's own semidiurnal frequency.)
Equilibrium tidal potential at the hypocenter — the combined lunar + solar degree-2 tide, used as a per-event amplitude.

The independence step that matters most. Aftershocks are not independent: an aftershock shares its mainshock's time, and therefore its lunar phase, so an uncorrected catalog can manufacture a fake tidal signal out of a few big sequences. We remove aftershocks and foreshocks with the standard Gardner–Knopoff (1974) space-time windows, keeping only mainshocks. This cuts the catalog from 7,477 to 4,729 independent events — a 37% reduction — and 4,729, not 7,477, is the honest sample size for every test. The undeclustered catalog is carried only as a sensitivity check.

Tests. For each phase quantity we run a Schuster test for non-uniformity (with a Rayleigh cross-check), stratified by depth (shallow ≤70 km, intermediate 70–300 km, deep >300 km) and by magnitude (M6, M6.5, M7). For the amplitude we compare each event's tidal potential against a location-matched random-time background, so the test is not circular. The pre-registered primary family is nine tests (three phase variables × three depth strata, declustered, M6), and we apply a Bonferroni-9 correction at family α = 0.01 (per-test threshold p < 0.0011). The headline claim would be a syzygy excess in shallow M6 events surviving that correction.

Results: nothing, cleanly

There is no signal anywhere.

Primary tests (declustered, M6):

Quantity	All depths	Shallow	Predicted if tides matter
Synodic phase φ	p = 0.71	p = 0.63	concentration, p < 0.0011
Syzygy 2φ	p = 0.68	p = 0.43	excess at new/full moon
Semidiurnal phase θ	p = 0.62	p = 0.84	concentration at tidal extremum
Tidal amplitude	mean pctile 0.500 (p = 0.99)	0.505	events at high tide

Every one of the three pre-registered directional predictions fails: no syzygy excess (the tiny, non-significant resultant actually points toward quadrature, not syzygy); no shallow-to-deep ordering of concentration; and the high-tidal-amplitude rate ratio is 1.02 (p = 0.28), a coin-flip nudge, not a trend.

The whole grid. Across all 72 Schuster cells (both catalogs × 3 phase variables × 4 depth bins × 3 magnitude thresholds), 11 reach nominal p < 0.05 against ~3.6 expected by chance — but the cells are heavily non-independent (nested magnitude thresholds, the raw catalog a superset of the declustered one, depth bins partitioning the same events), the excess sits in the raw and small-N high-magnitude bins, the smallest p anywhere is 0.009, and nothing survives Bonferroni-9. The nine-test primary family contains exactly one nominal p < 0.05 (expected 0.45). This is what a null looks like when you slice the data many ways.

Declustering earned its place. In the raw catalog the syzygy test showed a weak concentration (all-M6 2φ, p = 0.09). After aftershock removal it disappears (p = 0.68). The one whiff of a pattern in the naive catalog was aftershock clustering — exactly the artifact the pre-registration warned about, caught by the control it specified.

How strong an effect could we have seen?

This is where honesty about a null earns its keep. With 4,729 independent events, the Schuster detection floor at α = 0.01 is an occurrence enhancement of about 3.7% overall and 4.3% in the shallow stratum. (The pre-registration's stated floor of 1.5–2% was too optimistic: its arithmetic understated the required sample size — a 1% floor at α = 0.01 needs ~66,000 events, not the ~10,000 it assumed. We report the corrected number.)

So the honest reading is a bound, not a refutation: any global lunar/tidal modulation of M≥6 occurrence larger than ~4% is ruled out. Effects at the ~1% level — the scale of the fault-resolved signals in the literature — are simply below what 53 years of M≥6 data can resolve, and remain neither confirmed nor refuted here.

Discussion: a null that does not overreach

Our result sits comfortably beside, not against, the small positive findings in the literature. Cochran et al. (2004) found triggering of shallow thrust faults by resolved tidal shear stress on the fault plane — a quantity that requires each earthquake's focal mechanism to compute, and a mechanism subset (thrusts) whose signal is diluted when pooled with the strike-slip and normal events that respond with opposite tidal phase. Our V1 deliberately does neither: it uses the coarse hypocentral equilibrium potential, not resolved fault-plane stress, and it pools all mechanisms. A real few-percent effect confined to shallow thrusts would be smeared below our 4% floor by exactly these two choices. Our null is therefore a statement about large, global, mechanism-blind lunar modulation — which does not exist at M≥6 — and is silent about the smaller, mechanism-specific effect that specialized studies target.

Three alternative readings of the null we take seriously: (i) tidal triggering is real but confined to resolved shear stress and shallow thrusts, below our floor; (ii) mechanism mixing cancels a real signal in the pooled test; (iii) there is genuinely no effect at M≥6 and the smaller-magnitude positive reports do not extrapolate to the largest events. Our data cannot distinguish these, and we do not pretend otherwise.

A faint, non-significant clustering near the semidiurnal tidal minimum recurs across a few of the larger-magnitude bins. We report it rather than bury it, but it survives no correction and sits in the channel we pre-flagged as confounded by the solid-Earth tide; it is a curiosity, not a finding.

Limitations

Coarse tidal quantity. V1 uses the hypocentral equilibrium potential, not the resolved shear/normal stress on the fault plane, where the strongest published effects live. This is the single biggest reason the test is blind to the ~1% literature effect.
Fixed depths. 31% of mainshocks carry USGS default depths (10, 33, or 35 km) assigned when depth is unconstrained. This does not corrupt the coarse shallow-versus-deep split (defaults are shallow), but it forbids any finer depth inference.
Ingestion completeness boundary. The sub-M6 catalog is incomplete (the 5.6× discontinuity at M6.0); only M≥6 is used, where completeness is solid.
Power. The ~4% floor is the headline limitation. This paper cannot address the 1% regime.

Conclusion and the V2 hook

Across 53 years and 4,729 independent M≥6 earthquakes, the Moon leaves no global fingerprint on when the largest earthquakes occur, to a resolution of about 4%. That is a real, decisive, and citable bound on the large-effect version of a 130-year-old question — and an honest admission that the small-effect version is beyond the reach of a coarse global test.

The natural sequel (V2) is the version that can reach the ~1% regime: merge GCMT focal mechanisms, compute the resolved tidal shear stress on each fault plane, and repeat the test on shallow thrusts alone. That is the design that found a signal for Cochran et al.; putting it to the full 53-year M≥6 catalog is the paper that could actually confirm or refute them.

Measured reality only. The earthquakes are measured; the tide is the measured position of the Moon and Sun, transformed by gravity. No model output enters this paper.