The palaeoinclination of the ancient lunar magnetic field from an Apollo 17 basalt

Palaeomagnetic studies of Apollo samples indicate that the Moon generated a magnetic field for at least 2 billion years1,2. However, the geometry of the lunar magnetic field is still largely unknown because the original orientations of essentially all Apollo samples have not been well constrained. Determining the direction of the lunar magnetic field over time could elucidate the mechanism by which the lunar dynamo was powered and whether the Moon experienced true polar wander. Here we present measurements of the lunar magnetic field 3.7 billion years ago as recorded by Apollo 17 mare basalts 75035 and 75055. We find that 75035 and 75055 record a mean palaeointensity of ~50 μT. Furthermore, we could infer from the magnetization direction of 75055 and the layering of its parent boulder that the inclination of the magnetic field at the time was 34 ± 10°. Our recovered inclination is consistent with, but does not require, a selenocentric axial dipole (SAD) field geometry: a dipole in the centre of the Moon and aligned along the spin axis. Additionally, although true polar wander is not required by our data, true polar wander paths inferred from some independent studies of lunar hydrogen deposits and crustal magnetic anomalies4–6 are consistent with our measured paleoinclination. Basalt samples from the Moon gathered during the Apollo 17 mission hold information on the lunar magnetic field as it was 3.7 billion years ago. Its mean intensity was ~50 μT and its inclination 34 ± 10°. Such results suggest that the lunar dynamo was active at the time and was axially aligned and dipolar.

magnetic field. Although some such studies found support for the presence of a selenocentric axial dipole (SAD) geometry from the clustering of some palaeopoles, the palaeopoles as a whole are spread over the entire surface of the Moon 5,6 . The large scatter is likely in part a consequence of the fact that palaeopole inversions depend on numerous assumptions about the nature of the magnetization source (Supplementary Text 3). Therefore, measurement of the palaeomagnetic directions recorded by lunar samples, whose original orientations can be reconstructed and that can be demagnetized to remove confounding magnetic overprints and dated with radiometric techniques, is likely the most robust approach to reconstruct reliable lunar palaeopoles.
Although there is strong evidence for the existence of a past lunar dynamo, its physical mechanism and power source are highly uncertain. The small size of the lunar core means that the apparent surface palaeointensities before 3.5 Gyr ago (Ga hereafter) imply a very strong dynamo field (>24 mT) in the core even in the limiting case of a purely dipolar field. As a result, convective core dynamo scaling laws fail by more than an order of magnitude to generate surface intensities >50 μT over time periods lasting >30 Myr (ref. 7 ). Precession-driven dynamos, although potentially able to achieve higher palaeointensities 8 , struggle to generate magnetic fields in spherical cores for realistic viscosities 9 although the dissipative heat from precession might instead serve as a heat source for a convective dynamo 10 . Impact-driven dynamos that initiate mantle stirring of the core may be able to generate the observed palaeointensities, but can be sustained for only thousands of years 11 and require basin-forming impacts that ceased before 3.7 Ga (ref. 12 ). It has also been proposed that a basal magma ocean could generate the observed intensity and longevity of the lunar dynamo, but this requires the magma ocean to have an exceptionally high electrical conductivity 13 . Measurements of the geometry of the ancient lunar magnetic field would provide invaluable constraints for distinguishing between these potential dynamo mechanisms, as well as between dynamo and non-dynamo processes (Fig. 1, Supplementary Fig. 19 and Supplementary Text 5).
The relationship between magnetic inclination, I, and latitude, λ, of an axial dipolar magnetic field is given by tan I = 2 tan λ. (1) The relative contributions of an axial dipole and multipolar terms to a dynamo field can be estimated from the local Rossby number, Ro l ,

Nature astroNomy
which quantifies the ratio of inertial to Coriolis forces for characteristic flow scales within the core 14 . Synchronous rotation of the Moon was likely established early in lunar history, such that by 3.7 Ga the rotation rate was sufficiently slow (~5 × 10 −6 rad s −1 ) that its Ro l would have reached a value of ~2, suggesting that the field in the core was dominantly nondipolar 14 . Therefore, we should not expect the lunar dynamo to have followed equation (1). Even so, a high Ro l does not by itself preclude a dipolar surface magnetic field geometry. The multipolar state implied by the high Ro l applies to only the field-generating region in the core, whereas palaeomagnetic observations constrain only the surface field (Supplementary Text 4).
To constrain the geometry and intensity of the ancient lunar dynamo, we studied two mare basalts collected during the Apollo 17 mission: 75035 and 75055. These samples have indistinguishable

Nature astroNomy
Pb-Pb ages of 3,753 ± 9 Ma and 3,752 ± 9 Ma, respectively 15 . They were collected at the rim of the 610-m-diameter, ~500 ± 200 Myr-old Camelot crater 16 located at 20.2° N, 37.3° E. The Camelot cratering event is thought to have exposed >120 m of mare basalt flow stratigraphy. The possibility that the original orientations of these samples could be reconstructed only recently came to light following reassessment of the geology of the Taurus-Littrow valley 16 . Although they were originally interpreted as ejecta blocks, the parent boulders of these samples are now thought to represent near-in-situ wall rock as evidenced by the absence of an ejecta blanket. New interpretations suggest Camelot crater is 400 Myr older than previously thought and has undergone extensive mass wasting to expose wall rock while the ejecta blanket was ground into regolith. It is unlikely that these boulders have moved more than ~10 m from their original location, although they may have been differentially tilted or rotated with respect to the underlying bedrock 16 . The top and bottom of each panel show a schematic of the approximate field direction with respect to the Moon and the inclination as a function of latitude, respectively. In the top panels, black represents the core, grey represents solid silicates, dark red represents molten silicate, red lines represent the magnetic field, blue lines represent electrical currents and the yellow exterior region represents the solar wind for cases in which it is invoked for field generation. a, A selenocentric axial dipolar magnetic field generated by a core dynamo with a rapid planetary rotation rate (ro l < 0.12). b, A multipolar, non-axisymmetric magnetic field generated by a core dynamo with a slow planetary rotation rate (ro l > 0.12). c, An axially aligned quadrupolar dynamo generated by a deep magma ocean dynamo 13 . d, An axially aligned octupolar dynamo generated by a shallow magma ocean dynamo 13 . e, The Earth's mean axial dipole field at the location of the Moon. f, The interplanetary magnetic field. g, Antipodal field amplified by basin-forming impacts 52 . The predicted inclination is for rocks at the crater antipode. h, Fields generated by impact plasmas 53 . The predicted inclination is for rocks directly below the crater. i, Magnetic fields generated by the thermoelectric effect in lava basins that are electrically connected to subsurface magma and the solar wind 53,54 . j, Magnetic fields generated by a unipolar dynamo in lava basins 53 . The predicted inclination is for a rock at the surface close to the lava pool. k, Compression of the interplanetary magnetic field around the Moon by a cometary coma 53 . remanence-anisotropy-corrected high-coercivity components are shown in deep blue and red for 75035 and 75055, respectively. Samples of 75035 with bandsaw overprints are omitted. The mean (stars) and 95% confidence interval (ellipses) were calculated for the anisotropy-corrected high-coercivity components. a,b, Equal-area stereonet projections of mutually oriented specimen components for 75035 (a) and for 75055 (b) in lunar coordinates. c, Palaeoinclinations in lunar palaeohorizontal coordinates for the high-coercivity, origin-trending components in 75055. The mean (horizontal red line) and one standard deviation (red shaded region) are shown for anisotropy-corrected components. Uncertainties correspond to the measured maximum angular deviation (Supplementary Table 11).

Nature astroNomy
The parent boulder for 75055 exhibits clear layering and planar features that can be used to infer palaeohorizontality. Sample 75055 is therefore one of very few known Apollo samples whose original orientation can be unambiguously constrained. We have used this sample to measure the palaeoinclination (along with the palaeointensity) of the lunar field 3.7 Ga. Because the parent boulder of 75035 is smaller and was documented with fewer astronaut photographs, we can only infer the field paleointensity from this sample.
Electron microscopy and rock magnetic analyses indicate that the magnetic carriers in our samples are pure Fe kamacite (Supplementary Figs. 7 and 8, Supplementary Table 3 and Supplementary Text 2). Our alternating-field demagnetization of specimens of 75035 and 75055 revealed non-origin-trending low-coercivity components of natural remanent magnetization (NRM) that were removed by 0-11 mT (Extended Data Fig. 1 and Supplementary Table 11). The low-coercivity components are highly scattered within each parent sample (Fig. 2). Once the low-coercivity overprints were removed, we found that four non-bandsawn specimens of 75035 and five specimens of 75055 each contained a final origin-trending high-coercivity component that was blocked up to 40-60 mT (similar directions can be recovered up to 145 mT, but with less certainty) and is essentially unidirectional within each parent sample (Extended Data Fig. 1 and Supplementary Table  11). The Fisher mean anisotropy-corrected high-coercivity direction for these four specimens of 75035 is 032°/18° in present-day lunar coordinates (see Table 1 for definitions and descriptions of coordinate systems) with a 95% confidence interval of α 95 = 14°. The Fisher mean anisotropy-corrected high-coercivity direction  Fig. 1) are shown for impact fields, a unipolar dynamo and a thermoelectric dynamo (red lines), Earth's magnetic field (purple line), the interplanetary magnetic field (IMF, yellow line) and a selenocentric axial dipole (green line). b, The predicted magnetic inclination at Camelot crater for a magnetic field with zonal dipolar, quadrupolar and octupolar components, where G2 and G3 are the quadrupolar-to-dipolar and octupolar-to-dipolar field ratios, respectively. The thick black line is the mean palaeoinclination and the regions bounded by thin black lines represent the 95% confidence interval. both low and high degrees of multipolarity are permitted by our measured palaeoinclination. c, Equal-area stereographic projection showing the possible palaeopole locations 3.7 Ga in present-day lunar geographic coordinates. Open symbols and dashed lines are in the northern hemisphere, and closed symbols and solid lines are in the southern hemisphere. The star is the current location of Camelot crater where the samples were collected. The thick black lines are the permitted locations of the north pole calculated from our mean recovered palaeoinclination. Thin black lines mark the 95% confidence interval for our measurement. Two bands are permitted given the possibility of a reversing dynamo. The grey shaded region represents <10° dipole tilt. North palaeopole locations 3.7 Ga from independent studies are shown in purple 4 , pink 24 , red 23 , orange 5 , green 55 and blue 56 .

Nature astroNomy
for the five specimens of 75055 in present-day lunar coordinates is 334°/00° with α 95 = 10°. In lunar palaeohorizontal (LPH) coordinates, the high-coercivity component of 75055 has a palaeoinclination of I LPH = 34° with α 95 = 10°, and an estimated Fisher precision parameter of κ = 61.6 where the uncertainty is estimated from the scatter in the high-coercivity directions among individual specimens (Supplementary Table 9).
We measured mean high-coercivity palaeointensities of 50.7 ± 13.5 μT and 57.0 ± 17.3 μT (1σ standard error for four and eight specimens) for 75035 and 75055, respectively, using the anhysteretic remanent magnetization (ARM) method (Supplementary  Figs. 15 and 16 and Supplementary Table 10). The palaeointensity estimates for 75035 and 75055 are both within error of the mean value of 77 μT measured from six previously measured Apollo samples of age 3.5-4.2 Gyr (refs. 4,12,17,18 ). These results thereby provide additional evidence that >3.5 Ga, the Moon had an active dynamo generating a strong magnetic field.
We consider the implications of our palaeoinclination results based on two alternative assumptions about the palaeoinclination record for 75055; first, that it reflects only the instantaneous lunar magnetic field and, second, that it is representative of the time-averaged (over ~0.1-10 Myr) lunar field. Considering the instantaneous magnetic field, we calculated the possible locations of the virtual magnetic pole. As the palaeoazimuth is unconstrained, the possible virtual magnetic pole locations define two small circles (Fig. 3c). For our recovered palaeoinclination, the dipole tilt is permitted (but not required) to be <10°, the maximum degree of tilt expected for an axially aligned instantaneous magnetic field that is dipole-dominated like those of other inner solar system bodies 19 .
We next consider our palaeoinclination results assuming that the instantaneous and time-averaged lunar magnetic fields were similar (that is, that the recorded inclination reflects a durable feature of the lunar field geometry). We consider both a multipolar core dynamo field and implications for true polar wander. To assess different magnetic field sources, we assume that Camelot crater was at its present latitude at the time of NRM acquisition. Using equation (1) and assuming no subsequent true polar wander, the magnetic inclination of a SAD at Camelot crater would be ~36°, which is indistinguishable from our palaeoinclination estimates (Fig. 3a). Because a dipolar magnetic field is at odds with the high Ro l of the Moon, this could indicate that the lunar core is stratified [20][21][22] and/or that the dynamo mechanism was non-convective (for example, precession 8,10 ).
Although our recovered palaeoinclination is consistent with a SAD generated by a core dynamo, our results do not require such a geometry. For example, our data cannot exclude a substantial multipolar contribution from low-order, purely zonal terms (Fig. 3b). A multipolar magnetic field is consistent with the low lunar rotation rate, and would also be favoured by dynamo sources closer to the surface (for example, a basal magma ocean 13 ) given the r −(l+2) dependence of multipolar fields, where r is the distance from the top of the core, and l is the spherical harmonic degree. The observed clustering of palaeopoles near the poles and the equator would also be consistent with a predominantly quadrupolar dynamo or a persistent non-axial dipole field 23 .
Finally, we consider the implications of our palaeoinclination for true polar wander, assuming that our measured palaeoinclination corresponds to a SAD. Considering the possibility that a SAD may have reversed, our palaeoinclination measurement predicts palaeopole locations offset from the spin axis that are indistinguishable from those of several independent studies at ~3.7 Ga (refs. 4,5,23,24 ), although true polar wander is not required to explain our results ( Fig. 3c and Supplementary Figs. 17 and 18).
We were able to reconstruct the palaeohorizontal orientation of the parent block for 75055 from planar features documented in astronaut photographs. This has enabled an accurate measurement of palaeoinclination at a known latitude on the lunar surface. The recovered palaeoinclination is consistent with, but does not require, a SAD. In addition, our palaeointensity measurements have shown that both 75035 and 75055 cooled in the presence of a strong field (~50 μT), consistent with previous evidence for an epoch of intense magnetic fields ~3.5-4.2 Ga (ref. 1 ).
Most importantly, we have provided a framework in which other potentially orientable Apollo samples can be included to improve our understanding of the ancient lunar magnetic field. Contemporaneous measurements of the ancient lunar field from samples at different latitudes on the Moon will enable the magnetic field geometry to be unambiguously determined. Palaeoinclination measurements from samples of different ages would also enable the rate and extent of true polar wander and lunar secular variation to be quantified. Future sample return missions to the Moon should collect oriented samples from confirmed bedrock, which will greatly enhance our understanding of the geometry and temporal variability of the ancient lunar magnetic field and the mechanism of dynamo generation.  Table 1

75035, 75055
Lunar palaeohorizontal LPH Using planar features on blocks, lunar coordinates are tilt-corrected to the palaeohorizontal. Only inclination is considered in this coordinate system.

75055
The order in which coordinate systems are listed is the order in which corrections were made to sample orientations throughout the study.

Letters
Nature astroNomy each block on the lunar surface were reconstructed using photographs taken under lunar-lighting conditions simulated at the Lunar Receiving Laboratory 25 .
The astronaut photographs indicate that the blocks were likely not undisturbed bedrock when samples 75035 and 75055 were collected but rather had undergone subsequent tilting and rotation since their formation (Extended Data Fig. 3). As a result, outcrop textural features allow the palaeohorizontal to be determined, but cannot account for the palaeoazimuthal direction. We therefore only attempt to constrain palaeohorizontal and determine the ancient field's palaeoinclination and not palaeodeclination. Planar features are observed on the parent boulders of 75035 and 75055 and on other blocks around Camelot crater (Extended Data Fig. 3). The astronauts noted that these planar features were largely defined by variations in vesicle distribution 26 and may represent the tops of lunar lava flows. Terrestrial analogues thought to be emplaced in a similar manner to mare basalt lava flows are the extensively studied pāhoehoe basalt flows of Hawaii [27][28][29] . Pāhoehoe flows grow via inflation due to injection of fresh magma beneath a solid crust. This results in a distinct distribution of vesicles, with the bottom and top of the lava flow containing horizontally aligned 'vesicular zones' that can be tens of centimetres to metres in thickness, whereas horizontal 'vesicle sheets' (~10 cm thickness) form near the centre of the flow. 75035 and 75055 are thought to have formed in the central part of a pāhoehoe-like flow given the coarse (>1 mm) crystalline nature of ilmenite within the samples 29,30 , such that the planar features observed on the parent boulder of 75055 likely correspond to vesicle sheets. The spatial distribution of features is consistent with a lava flow thickness of <10 m. Lunar lava flows have a lower viscosity than that of any known terrestrial flow 31 and are thin (<10 m) and laterally extensive 32 . It should also be noted that horizontal vesicular layers can be produced without the injection of fresh melt into the system 33 . In terrestrial basalts exhibiting vesicular layers, such as the Columbia River flood basalt flows, vesicles have been observed to form approximately every 1 m, consistent with theoretical calculations 33 . However, given the lower viscosity of lunar basalts and lunar gravity, these layers are likely to be closer together on the Moon due to faster rise times and are therefore a plausible explanation for the ~30 cm spacing between vesicle layers observed on the blocks at Camelot crater (Extended Data Fig. 3). We therefore assumed that the planar features defined by vesicles within the 75055 parent block represent the palaeohorizontal at the time of eruption. Using this assumption, we reconstructed the palaeohorizontal orientation of the block. Note also that the similar compositions of the blocks at the crater edge and their spatial distribution indicate that they are close to their formation localities such that the measured palaeointensities and palaeoinclinations can be interpreted in the context of the latitude of Camelot crater.
We used photographs of sample 75055 (in JSC coordinates) to reconstruct its original orientation in lunar geographic coordinates ( Supplementary Figs. 4  and 6). For 75035, the small number of astronaut photographs and their limited range of perspectives means that we cannot unambiguously reconstruct the palaeohorizontal for this sample (Supplementary Figs. 3 and 5). The attitude of the visible foliations on the parent boulder of 75055 were reconstructed by measuring the apparent trend and plunge of planar features observable in a series of photographs taken from different locations (Supplementary Table 2 and Extended Data Fig. 4). A great circle was fitted to the trend and plunge measurements to define the plane. The planar features have a strike and dip of 243°/36° (α 95 = 5°) in lunar coordinates. The orientation of the plane was used to tilt the boulder back to its original palaeohorizontal position. Once the palaeohorizontal orientation of the parent boulder of 75055 was established, the magnetic palaeoinclination was recovered. Because the lunar magnetic field may have undergone reversals, it is only meaningful to recover the magnitude of the palaeoinclination and not its sign.

Palaeomagnetic analysis.
All demagnetization experiments were conducted on a 2G Enterprises Superconducting Rock Magnetometer 755 at the Massachusetts Institute of Technology Paleomagnetism Laboratory. The magnetometer has a sensitivity of <1 × 10 −12 A m 2 (ref. 34 ). Fifteen specimens of 75035 and nine specimens of 75055 were demagnetized using an alternating field because thermal demagnetization can cause thermochemical alteration of mare basalts even under controlled-atmosphere conditions ( Supplementary Fig. 9) 35,36 . The NRM was removed by three-axis alternating-field demagnetization in steps of 0.5 mT up to 25 mT, steps of 1 mT up to 95 mT, and then steps of 1.5 mT up to 145 mT. The magnetic moment was measured after each alternating-field step and the three orthogonal measurements were then averaged to correct for gyroscopic remanent magnetization following the Zijderveld-Dunlop method 35,37 . A subset of specimens (75055,127Aa, 75055,127Ab, 75055,127Ac, 75055,127Ae and 75055,127b) were demagnetized up to 420 mT in steps of 7.5 mT to ensure that high-coercivity components were entirely removed.
NRM components were characterized using principal component analysis. Origin-trending components were identified when the maximum angular deviation (MAD) exceeded the angular deviation. The final fits for such origin-trending components were anchored to the origin 38,39 . These fits were used to calculate a Fisher mean direction and associated 95% confidence interval (α 95 ) for each specimen in PmagPy 3.0 40 . Six specimens of 75035 exhibited resolvable origin-trending high-coercivity components with a MAD < 30°, but two of these were disregarded (75035Ad and 75035Ah) because they were partially overprinted  Table  1). Five specimens of 75055 exhibited resolvable origin-trending high-coercivity components with a MAD < 30°. Component directions were plotted in Stereonet 41,42 , rotated to present-day lunar coordinates and, for 75055, tilt-corrected using the reported palaeohorizontal estimates (see 'Reconstructing sample orientation relative to the lunar surface'). Recovered component directions were also corrected for anisotropy of ARM (AARM) (see 'Rock magnetic analysis').
Palaeointensity estimates were calculated using the ARM palaeointensity method 37,43,44 . NRM demagnetization curves were compared to alternating-field demagnetization of an ARM with a 100 μT d.c. bias field applied with a 260 mT alternating field. The thermoremanent magnetization (TRM)-equivalent palaeointensity recorded by the samples is given by where b is the d.c. bias field and f ′ = 1.34 is the TRM/ARM calibration factor experimentally determined using lunar basalts 45 . There are two main uncertainties associated with these palaeointensity estimates. First, the least-squares slope obtained from ΔNRM ΔARM has an uncertainty associated with scatter in the data around the best-fit line. Second, the calibration factor f ′ is estimated to have a 2σ uncertainty of a factor of 5 (ref. 1 ). The uncertainty in f ′ originates from its dependence on ferromagnetic mineralogy including grain size, grain morphology and grain distribution. Cooling-rate corrections are not applied because the heating time (~1 h) for ARM/TRM calibration experiments is similar to the several-day cooling timescales of 75035 and 75055 (refs. 46,47 ).

Rock magnetic analysis.
We used several rock magnetic techniques to assess the fidelity and magnetic recording properties of samples 75035 and 75055 (Supplementary Text 1). We conducted palaeointensity fidelity tests, isothermal remanent magnetization (IRM) acquisition and Curie balance analysis, and quantified the degree of AARM.
To assess the magnetic recording fidelity of samples 75035 and 75055, we gave the samples ARMs in d.c. bias fields ranging from 5 μT to 100 μT and quantified how accurately we could recover the palaeointensity of this field using the ARM palaeointensity method. The ARMs were applied with an alternating field of 260 mT and were alternating-field demagnetized following the same protocol that was used to demagnetize the NRM. The demagnetization curves were compared to the demagnetization curve of the 100 μT ARM. For each applied ARM, the difference, D ′ = L−I L , between the recovered palaeointensity (I) and the predicted equivalent TRM (L) and the error, E = W L , based on the 95% confidence for the retrieved palaeointensity (W) were calculated 35,48 . Acceptance criteria for a reliable palaeomagnetic recorder are defined by −50% < D ′ < 100% and E < 50% (Supplementary Table 4 and Supplementary Fig. 11).
We conducted IRM acquisition to assess the coercivity spectrum of magnetic carriers in each sample. Specimens were given a saturation IRM of 400 mT and were then alternating-field demagnetized. The derivative of the demagnetization curve with respect to the applied field, δIRM/δB, was calculated to infer the coercivity distribution ( Supplementary Fig. 10).
The samples' magnetic mineralogy and susceptibility to thermochemical alteration were assessed using a Curie balance at the Fort Hoofddijk Paleomagnetic Laboratory. A small (3.52 mg) fragment of specimen 75055Ac was exposed to fields of 100-300 mT during a heating cycle from room temperature to 800 °C and then back to room temperature in air. The magnetization was measured every second over the course of the experiment, which took ~12 h. Raw data were smoothed using a Savitzky-Golay filter 49 .
The AARM of seven specimens of 75035 and six specimens of 75055 was measured at the Fort Hoofddijk Paleomagnetic Laboratory using a 2G Enterprises Superconducting Rock Magnetometer. Specimens were given an ARM of 50 μT in an alternating field of 150 mT. Samples were given an ARM in nine distinct orientations, E mag , N mag , T mag , NE mag , NW mag , TE mag , TW mag , TN mag and TS mag , with each specimen mounted in a different orientation in laboratory coordinates to remove any directional measurement bias (Supplementary Table 5). After measuring the magnitude and orientation of the acquired ARM, specimens were alternating-field demagnetized up to 150 mT in three orthogonal directions. The order of the ARM and demagnetization steps is shown in Supplementary Table 6. Specimens were mounted in custom-made glass cubes with three-dimensional-printed sample holders (holder moment <7 × 10 −11 A m 2 ) to maintain exact orientations throughout. This enabled the estimation of the three principal axes defining the anisotropy ellipsoid, where p 1 is the maximum, p 2 is the intermediate and p 3 is the minimum principal axis of anisotropy 50 . After defining the axes, we calculated the degree of foliation F = p2 p3 , lineation L = p1 p2 and anisotropy P = p1 p3 for individual specimens 51 . Anisotropy corrections improved the clustering of directions for 75035, with the Fisher precision parameter, κ, increasing from 11.9 to 19.1, whereas for 75055 κ decreased slightly from 49.2 to 42.9 (Fig. 2).

Data availability
The palaeomagnetic data that support the findings of this study are available from the Magnetic Information Consortium (MagIC) database at http://www2.earthref.

Letters
Nature astroNomy org/MagIC/17123. All other data requests and correspondence should be directed to C.I.O.N. Source data are provided with this paper.