LETTERSPUBLISHED ONLINE: 25 JANUARY 2009 DOI: 10.1038/NGEO417

Timing of crystallization of the lunar magmaocean constrained by the oldest zirconA. Nemchin1*, N. Timms1, R. Pidgeon1, T. Geisler2, S. Reddy1 and C. Meyer3

TheMoon is thought to have formed through the consolidationof debris from the collision of a Mars-sized body with the Earthmore than 4,500million years ago. The primitive Moon wascovered with a thick layer of melt known as the lunar magmaocean1, the crystallization of which resulted in the Moon’ssurface as it is observed today. There is considerable debate,however, over the precise timing and duration of the processof magma ocean crystallization. Here we date a zircon fromlunar breccias to an age of 4,417±6million years. This dateprovides a precise younger age limit for the solidification ofthe lunar magma ocean. We propose a model that suggestsan exponential rate of lunar crystallization, based on acombination of this oldest known lunar zircon and the age of theMoon-forming giant impact. We conclude that the formationof the Moon’s anorthositic crust followed the solidification of80–85% of the original melt, within about 100million years ofthe collision. The existence of younger zircons2 is indicative ofthe continued solidification of a small percentage of melt for anextra 200–400million years.

Fractional crystallization of the lunar magma ocean (LMO)involved the early density-driven separation of mafic cumulatesand flotation of a plagioclase-rich lunar crust represented byferroan anorthosite1. Subsequent crystallization of ilmenite fromthe remaining portion of the LMO (ref. 1) left a residual liquidenriched in highly incompatible elements. This liquid formed theenriched reservoir referred to as KREEP (from high concentrationsof potassium, rare-earth elements and phosphorus3).

A precise determination of the timing of fractional crystallizationof the LMO has been inhibited by the susceptibility of Sm–Ndand other systems to the partial resetting during the later thermalpulses associatedwith themeteorite impacts. As a result, the Sm–Ndmineral isochrons constrained for the ferroan anorthosite samplesshow a wide spread of ages between 4,560± 70Myr (ref. 4) and4,290± 60Myr (ref. 5). The best estimate for the age of ferroananorthosites determined as 4,456± 40Myr from the combinationof mafic minerals in all analysed samples but excluding plagioclasedata that are partially disturbed6 has another inherited problem asit assumes that all samples have been formed at the same time.

Another way that has been used to constrain the timing ofthe LMO differentiation is using model ages of rocks derivedfrom different reservoirs in the lunar mantle. In particular, aKREEP-rich source is recognized as an essential part of late-stagecrystallization of the LMO, and model ages of KREEP formationhave been estimated as !4,600Myr by Rb–Sr analysis of lunarsoils7, !4,420Myr from U–Pb systematics of highlands rocks anda basalt sample8 and !4,360Myr from the Sm–Nd model ages ofKREEP samples9. An average of the model age for KREEP was

1Department of Applied Geology, Western Australian School of Mines, Curtin University of Technology, Bentley, Western Australia, 6102, Australia,2Institut für Mineralogie, Westfälische Wilhelms-Universität, Corrensstr 24, 48149, Münster, Germany, 3NASA Johnson Space Center, Houston, Texas,77058, USA. *e-mail: nemchina@kalg.curtin.edu.au.

estimated as 4,420± 70Myr (1! uncertainty)10. Recent tungstenisotope data onmetals from low- and high-titaniummare basalts aswell as two KREEP-rich samples11 suggest that the last equilibrationof the LMO, which is possible only up to a critical point whenabout 60% of the melt is solidified, occurred after 4,507Myr ago(60Myr after the formation of the Solar System). This result is inagreement with the 146Sm–142Nd model age of the LMO (ref. 11),which is based on the combined 147Sm–143Nd and 146Sm–142Ndsystems in lunar basalts and implies a 238+56

"40 Myr (ref. 12) to215+23

"21 Myr (ref. 13) time interval for lunar mantle formation.Despite the general agreement between the model ages determinedusing different isotope systems, their accuracy is limited by themodels and the timing of LMO crystallization remains looselyconstrained to the first 250Myr of lunar history.

Both isotopic resetting and model dependence problemsassociated with numerous previous attempts to place limits on thetime of LMO crystallization can be avoided by using the U–Pbsystem in zircon2,14, which is well known for its stability under avariety of extreme conditions. Growth of zircon inmelts is governedby zircon saturation, which can be achieved only in a mafic magmainitially enriched in zirconium15. Consequently, the presence ofzircon in the lunar samples is linked to the initial enrichmentof the magma in the KREEP component (that is, KREEP mustform on the Moon before zircon can appear in any rock type).Therefore, the oldest zircon defines a younger limit for the timeof KREEP formation.

Here we report the oldest zircon crystal found on the Moon sofar, which is located in thematrix of Apollo 17 clast-rich impactmeltbreccia 72215, in the thin section 72215, 195. The 0.5mmgrain lackswell-developed crystal faces and contains several brittle fractures(Fig. 1), and we thus consider it to be a relict fragment of a largergrain that was incorporated into the host breccia.

Forty-one secondary-ionmass spectrometryU–Pb analyses weremade on this grain (see Supplementary Information, Table S1,Fig. 2a). The results indicate a complex pattern of isotope resettingthat systematically varies with the microstructural features of thegrain (see Supplementary Information, Table S1; Fig. 1). Thesemicrostructural features are a combination of primary magmaticcharacteristics and a different degree of self-irradiation damagehighlighted by the variable birefringence and cathodoluminescenceemission, as well as deformation patterns revealed by crystal-lographic orientation analysis of electron backscatter diffraction(EBSD) patterns. The observed overall decrease in 207Pb/206Pb agescorrelated with an increase of the local misorientation determinedfor each sensitive high-resolution ionmicroprobe (SHRIMP) spot16(Fig. 2b), indicates that this differential resetting of the U–Pbsystem occurred as a result of impact-related plastic deformation,

NATURE GEOSCIENCE | VOL 2 | FEBRUARY 2009 | www.nature.com/naturegeoscience 133© 2009 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO417

3029

27

28

25

26

17

9

13

14

21

2423

2232

33

1519

312

4810

16

2039

38

11

12

41

37

36 356

34

17 18

403

5

>4.404.4004.3754.350<4.35

Age (Gyr)

10

100

200 µm 200 µm

i

i

200 µm 200 µm0

12°

a b

c d

Figure 1 |Microstructure of the zircon grain from lunar breccia 72215, 195. a, Cross-polarized-light photomicrograph showing sector zones and faintcompositional growth zones (inset). b, Panchromatic cathodoluminescence image with superimposed mean U–Pb ages for individual SHRIMP analyses.c, Map showing variations in EBSD pattern quality (band contrast) from poor (black) to good (white). d, Map derived from EBSD data showing variations incrystallographic orientation relative to the mean reference orientation (red cross).

an interpretation that is consistent with trace-element variationsrecorded in other deformed zircons16,17.

All 41U–Pb analyses are distributed along concordia between4,418 ± 8 and 4,331 ± 16Myr (uncertainties are 2! ; Fig. 2a).The four oldest analyses, from undeformed parts of the grain,form a coherent group on a concordia plot (Fig. 2a) with aconcordia intercept at 4,420± 15Myr and an average 207Pb/206Pbage of 4,417 ± 6Myr. We interpret this age as the age ofzircon crystallization. The five youngest analyses form a coherentgroup in the 207Pb/206Pb versus 238U/206Pb diagram (Fig. 2a),defining a concordia intercept age of 4,334 ± 10Myr for thecommon-lead uncorrected data and an average 207Pb/206Pb ageof 4,333± 7Myr for the Stacey–Kramers modern-lead correcteddata. These analyses correspond to areas of moderate luminescence,good EBSD pattern quality and low uranium and thoriumconcentrations (Fig. 1). Importantly, these analyses are also fromareas where the deformation bands intersect and/or have highmisorientation, suggesting deformation-related lead loss. It isevident that the most deformed areas of the grain have sufferedthe greatest lead loss, and we interpret the concordia intersectionage as a reflection of mobility of the U–Pb system in the grainduring an impact, although the resetting can be incomplete.The remaining intermediate ages are from areas of moderately

strained parts of the grain, and probably reflect a partial resettingof the U–Pb system.

Our results indicate that the KREEP source formed by4,417 ± 6Myr and it follows that crystallization of the LMOwas almost completed by this time. The zircon age is almost100Myr older than the age calculated from combined 142Nd–143Ndsystematics of lunar basalts and highland rocks12,13. These laterestimates, however, are based on the assumption that the separatemantle reservoirs have been formed at the same time and hadsimilar initial isotopic compositions of neodymium. This maynot be the case, even for KREEP magmas and the source ofhigh-titanium basalts. Both formed last in the LMO crystallizationsequence and largely define the slope of the combined 142Nd–143Ndisochrons. Nevertheless, the formation of the KREEP source at4,417 ± 6Myr is in agreement with the age of 4,456 ± 40Myrdetermined for the ferroan anorthosite samples6, even though theages are not completely resolved within the errors.

A combination of the KREEP minimum formation age of4,417± 6Myr and other data reflecting different stages of LMOevolution enables us to model the history of magma ocean differ-entiation and crystallization on the Moon, and two endmembermodels are presented (Fig. 3). Both models are constrained bythe new 4,417± 6Myr zircon age, defining a minimum age for

134 NATURE GEOSCIENCE | VOL 2 | FEBRUARY 2009 | www.nature.com/naturegeoscience© 2009 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO417 LETTERS

4,460

4420

4380

4340

4,300

0.51

0.53

0.55

0.57

0.59

0.90 0.94 0.98 1.02 1.06 1.10

238U/206Pb

207 P

b/20

6 Pb

Local misorientation (°) /alpha dose (!1010)

207 P

b/20

6 Pb

age

(Myr

)

Intercept: 4,420 ± 15 MyrMSWD = 0.11; probability = 0.90207Pb/206Pb age = 4,417 ± 6 MyrMSWD = 0.08; probability = 0.97

Intercept: 4,334 ± 10 MyrMSWD = 0.08; probability = 0.97207Pb/206Pb age = 4,333 ± 7 MyrMSWD = 0.05; probability = 0.99

4,300

4,340

4,380

4,420

0 10 20 30 40 50

a

b

Figure 2 |U–Pb SHRIMP data for the zircon from the breccia thin section72215, 195. a, Tera–Wasserburg concordia diagram. Data are not correctedfor the initial lead. Blue ellipses represent the four oldest analyses; redellipses represent the five youngest analyses; yellow ellipses representanalyses with intermediate U–Pb ages. MSWD: mean square weighteddeviation. Data-point error ellipses are 2! . b, Age versus ‘localmisorientation’ value determined at each SHRIMP spot from EBSD mapdata by calculating the mean misorientation between a central point and itsnearest neighbours on an 11# 11 pixel grid (13.2 µm# 13.2 µm area; ref. 16).Local misorientation data were normalized to alpha dose to account for theradiation damage. The resultant local misorientation values are interpretedto reflect lattice distortions associated with crystal-plastic deformation.Error bars are 2! .

formation of lunar KREEP at a late stage in the crystallizationof the LMO. Both are also based on the assumption that theLMO formed as a result of fast accretion following the giantimpact1 and, therefore, the age of LMO formation is similar tothe age of the Moon. The best current estimate of the age ofthe giant impact based on the Hf–W data is 62+90

"10 Myr after theformation of the Solar System11. These data place an older limitof LMO formation of 50Myr after the first condensation in thesolar nebula (4,517Myr ago). A simple model of LMO evolution(Fig. 3, solid line) suggests a sequential fractionation of olivine$orthopyroxene±olivine$olivine+clinopyroxene±plagioclase$clinopyroxene + plagioclase $ clinopyroxene + plagioclase+ ilmenite assemblages. However, the assumption of sequentialfractionation of mineral phases throughout the whole LMO isprobably an oversimplification because it is likely that: (1) a sig-nificant temperature difference would exist between the lower andupper parts of the LMO, (2) the appearance of different mineralson the liquidus is unlikely to be contemporaneous in differentparts of the magma ocean, (3) convection can prevent effectiveremoval of minerals from the liquid and (4) the formation of

4,2004,3004,4004,500Age (Myr)

Youngest limit for olivine¬pyroxene cumulates

Oldest limit for anorthosite crust formation

Mel

t rem

aini

ng (

%)

Ilmenite cumulate

KREEP

Oldest limit for plagioclase appearence

0

20

40

60

80

100

Figure 3 | LMO crystallization paths based on the available chronologicaldata. Solid line projected through the points representing (1) initialformation (100%melt—182W age; ref. 11), (2) mean time of lunar crustformation (30%melt—143Nd age; ref. 6) (3) KREEP formation (5–7%melt—age from this study), (4) time of cessation of magmatic activity inthe Serenitatis region (2.5–3.5% melt—age estimate from zircondistribution patterns2); dotted line based on (1) and (2) and the assumptionof a turbulent convection in the LMO resulting in the fast initial cooling; theyellow circle represents predicted formation of the lunar crust compatiblewith such fast cooling of the LMO. Data-point error crosses are 2! .

an insulation lid can change the cooling regime of the LMO. Amore complex model of LMO crystallization (Fig. 3, dotted line)involves rapid initial cooling of the magma ocean as a result ofvigorous turbulent convection18, which results in solidification ofa substantial proportion of LMO without significant fractionation.This was followed by fractionation limited to the relatively thin toplayer of the LMOowing tomuch slower cooling resulting from a lessvigorous convection regime, and possibly formation of a thermallyinsulating surface lid.

Nevertheless, both models combined with the availablechronological data suggest that ilmenite-bearing cumulates precipi-tated after about 90% of LMO crystallization, leaving a few per centof residual KREEP melt by 4,417± 6Myr ago. These data suggestthat the main volume of the LMO solidified within about 100Myr.The age distribution patterns obtained for numerous zircon grainsfrom Apollo 17 and 14 breccias2 suggest that the residual smallvolume fraction of the LMO liquid could have cooled slowly overthe subsequent 400–500Myr, probably sustained by the internalheating related to radioactive decay. These patterns indicate gradualshrinking of a semi-molten KREEP reservoir towards the centreof Procellarum KREEP terrane2, and that by about 4,250Myr agothe KREEP reservoir solidified under the area occupied by theSerenitatis basin, but continued to be active closer to the middleof Procellarum KREEP terrane near the Imbrium basin until about3,900Myr ago. Assuming that the thickness of the KREEP sourceis approximately constant throughout the Procellarum terrane, thisaccounts for an extra reduction in the residual proportion of KREEPmelt of about 50% by 4,250Myr ago.

Despite the precise fixation of the timing of the last stage of LMOcrystallization by our results, the timing of plagioclase appearancein the crystallization sequence remains imprecise. Estimates forthe appearance of plagioclase on the liquidus vary from about60% to 80% of LMO crystallization, depending on the assumedbulk Al content of the LMO (refs 19,20). Assuming sequentialcrystallization of minerals (Fig. 3, solid line) and using availablegeochronological data for the ferroan anorthosite samples, 70%of crystallization of the LMO is necessary before plagioclase canbecome a liquidus phase. In themore complexmodel (Fig. 3, dotted

NATURE GEOSCIENCE | VOL 2 | FEBRUARY 2009 | www.nature.com/naturegeoscience 135© 2009 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO417

line), the lunar crust formed after crystallization of 80–85% ofthe LMO. However, both estimates are within the uncertaintiesassociated with the relatively imprecise estimate of the age of theferroan anorthosites. The large uncertainty of this age also results inthe large range (anywhere between 20 and 100Myr) for the possibleduration of plagioclase flotation. As a result, further refinement ofthe models awaits more precise determination of the age of lunaranorthosite formation.

MethodsThe sample is a polished thin section of breccia 72215 prepared by NASA(National Aeronautics and Space Administration). The microstructure of thezircon was characterized by cathodoluminescence imaging and EBSD mappingusing a scanning electron microscope at Curtin University of Technology, Perth,Western Australia. Collection of EBSD data was processed using the proceduresdeveloped for zircon21. Slip systems were resolved from crystallographic orientationdata using simple geometric models of low-angle boundaries17.

U–Pb data were obtained using a SHRIMP at the John de Laeter Centreof Mass Spectrometry, Curtin University of Technology following the standardanalytical procedure described elsewhere14. Pb–U ratios were normalized to the564-Myr-old Sri Lankan zircon CZ3 analysed in a separate mount. Common leadwas corrected for using modern Stacey and Kramers lead, following the conclusionthat a substantial proportion of common lead in the lunar thin sections results fromthe surface contamination2. Regardless, of the selection of common lead for thecorrection, a very low proportion of 204Pb in the thin section 72215, 195 makes thecalculated ages insensitive to the uncertainty in the common lead.

Received 10 November 2008; accepted 19 December 2008;published online 25 January 2009

References1. Shearer, C. K. Thermal and magmatic evolution of the moon. In: New Views

of the Moon, Rev. Mineral. Geochem. 60, 365–518 (Mineralogica Society ofAmerica, 2006).

2. Nemchin, A. A., Pidgeon, R. T., Whitehouse, M. J., Vaughan, J. P. & Meyer, C.SIMS U–Pb study of zircon from Apollo 14 and 17 breccias: Implications forthe evolution of lunar KREEP. Geochim. Cosmochim. Acta. 72, 668–689 (2008).

3. Warren, P. H. & Wasson, J. T. The origin of KREEP. Rev. Geophys. Space Phys.17, 73–88 (1979).

4. Alibert, C., Norman, M. D. & McCulloch, M. T. An ancient Sm–Ndage for a ferroan noritic anorthosite clast from lunar breccia 67016.Geochim. Cosmochim. Acta 58, 2921–2926 (1994).

5. Borg, L. Isotopic studies of ferroan anorthosite 62236: A younglunar crustal rock from a light rare-earth-element-depleted source.Geochim. Cosmochim. Acta 63, 2679–2691 (1999).

6. Norman, M. D., Borg, L. E., Nyquist, L. E. & Bogard, D. D. Chronology,geochemistry, and petrology of a ferroan noritic anorthosite from Descartesbreccia 67215: Clues to the age, origin, structure and impact history of thelunar crust.Meteorit. Planet. Sci. 38, 645–661 (2003).

7. Papanastassiou, D. A., Wasserburg, G. J. & Burnett, D. S. Rb–Sr ages of lunarrocks from the Sea of Tranquillity. Earth Planet. Sci. Lett. 8, 1–19 (1970).

8. Tera, F. & Wasserburg, G. U–Th–Pb systematics on lunar rocks and inferencesabout lunar evolution and the age of the moon. Proc. 5th Lunar Sci. Conf.1571–1599 (1974).

9. Lugmair, G. W. & Carlson, R. W. The Sm–Nd history of KREEP. Proc. LunarPlanet. Sci. Conf. 9, 689–704 (1978).

10. Nyquist, L. E. & Shih, C.-Y. The isotopic record of lunar volcanism.Geochim. Cosmochim. Acta 56, 2213–2234 (1992).

11. Touboul, M., Kleine, T., Bourdon, B., Palme, H. & Wieler, R. Late formationand prolonged differentiation of the Moon inferred from W isotopes in lunarmetals. Nature 450, 1206–1209 (2007).

12. Nyquist, L. E., Wiesmann, H., Shih, C.-Y., Keith, J. E. & Harper, C. L.146Sm–142Nd formation interval in the lunar mantle.Geochim. Cosmochim. Acta59, 2817–2837 (1995).

13. Rankenburg, K., Brandon, A. D. & Neal, C. Neodymium isotope evidence forthe chondritic composition of the Moon. Science 312, 1369–1372 (2006).

14. Compston, W., Williams, I. S. & Meyer, C. U–Pb geochronology of zirconsfrom Lunar Breccia 73217 using a sensitive high mass-resolution ionmicroprobe, Proc. Lunar Planet. Sci. Conf. 14th, J. Geophys. Res. 89, B525–B534(1984).

15. Dickinson, J. E. & Hess, P. C. Zircon saturation in lunar basalts and granites.Earth Planet. Sci. Lett. 57, 336–344 (1982).

16. Timms, N. E., Kinny, P. D. & Reddy, S. M. Enhanced diffusion of Uraniumand Thorium linked to crystal plasticity in zircon. Geochem. Trans. 7,10 (2006).

17. Reddy, S. M. et al. Crystal-plastic deformation of zircon: A defect in theassumption of chemical robustness. Geology 34, 257–260 (2006).

18. Solomatov, V. S. in Origin of the Earth and Moon (eds Canup, R. & Righter, K.)323–338 (Univ. Arizona Press, 2000).

19. Snyder, G. A., Taylor, L. A. & Neil, C. R. A chemical model for generatingthe sources of mare basalts: Combined equilibrium and fractionalcrystallization of the lunar magmasphere. Geochim. Cosmochim. Acta 56,3809–3823 (1992).

20. Longhi, J. A model of early lunar differentiation. Proc. Lunar Planet. Sci. Conf.11, 289–315 (1980).

21. Reddy, S. M., Timms, N. E., Pantleon, W. & Trimby, P. Quantitativecharacterization of plastic deformation of zircon and geological implications.Contrib. Mineral. Petrol. 153, 625–645 (2007).

AcknowledgementsIn particular we would like to thank the astronauts of Apollo 17 for collecting the sample.The project was supported by the office of R&D department at Curtin University ofTechnology. Imaging was supported by the Australian Research Council Discovery GrantDP0664078 to S.R. and N.T.

Additional informationSupplementary Information accompanies this paper on www.nature.com/naturegeoscience.Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions. Correspondence and requests for materials should beaddressed to A.N.

136 NATURE GEOSCIENCE | VOL 2 | FEBRUARY 2009 | www.nature.com/naturegeoscience© 2009 Macmillan Publishers Limited. All rights reserved.