Journal of Petrology

This Article Abstract FREE Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Search for citing articles in: ISI Web of Science (12) Request Permissions Google Scholar Articles by O’HARA, M. J. Search for Related Content GeoRef GeoRef Citation Social Bookmarking          What's this?

Journal of Petrology Volume 41 Number 11 Pages 1545-1651 2000
© Oxford University Press 2000

Flood Basalts, Basalt Floods or Topless Bushvelds? Lunar Petrogenesis Revisited

M. J. O’HARA^,*

DEPARTMENT OF EARTH SCIENCES, CARDIFF UNIVERSITY, PO BOX 914, CARDIFF CF1 3YE, UK

Received September 2, 1999; Revised typescript accepted March 23, 2000

	ABSTRACT

TOP ABSTRACT INTRODUCTION CONTENTS AND ROUTE MAP LUNAR PETROGENESIS 1969-1999 NOTES Notes added in proof APPENDICES REFERENCES

 
There is a conspicuous dichotomy in the conventional model of lunar petrogenesis between the total intra-crustal differentiation postulated for the products of feldspathic volcanism in the lunar highlands and the near absence of differentiation postulated for the products of mare volcanism. Both the cumulate mantle model, and the selenotherm postulated to accompany genesis of alleged ‘primary’ mare magmas by remelting of those cumulates, imply supra-adiabatic thermal gradients in near-solidus materials throughout the lunar mantle 4·3–3·2 Ga ago. This should have resulted in vigorous convective motion, which has not occurred. There is no positive europium anomaly in the average lunar highland crust. That crust cannot, therefore, have formed by plagioclase flotation from a lunar magma ocean, for which there is no other requirement. There is no negative europium anomaly in the average mantle to be inherited by later mare basalts. Other rocky bodies of lunar size in the Solar System have accreted at rates that allowed incorporation of plenty of volatiles and without forming global magma oceans. Partial melting in the presence of water, followed by near-surface fractionation and volatile losses can explain the feldspathic character, high incompatible element concentrations and lack of Eu anomaly in the lunar highlands. Volcanic eruption on the Moon must have been accompanied by selective volatilization losses of sodium, sulphur and other elements similar to the process seen on Io, which can account for the major differences between terrestrial and lunar basalts. Siderophile element depletion in lunar lavas may reflect immiscible sulphide liquid and metal separation, rather than global impoverishment in such elements, and large ore bodies may have formed close to the lunar surface. Mare basalt volcanism appears to have been a protracted, low magma productivity event with few similarities to terrestrial ocean-floor, ocean-island, continental flood basalt or komatiite volcanism. At low pressure the crystallization of plagioclase well before pyroxene typifies those terrestrial mid-ocean ridge basalt, ocean-island basalt and continental flood basalt magmas. A similar sequence is demanded of the postulated lunar primary magmas. Mare basalt hand-specimen and pyroclastic glass bead compositions do not, however, display the required crystallization sequence and cannot represent the required primary melt compositions. The true erupted lava compositions which gave rise to the regolith compositions across all the maria are much more feldspathic than the majority of large hand specimens and, in common with basalts on other planets, they are close to low-pressure plagioclase-saturated cotectic residual liquids which have evolved by removal of gabbros in crustal magma chambers, or perhaps in giant lava lakes akin to topless Bushveld complexes. Any further debate could be resolved by a 100 m drill core in a few mare locations. Field provenance of samples from Mars, a planet half covered by flood basalts and products of central volcanoes, will be little better than for those from the Moon. It will be important to encourage multiple working hypotheses, rather than to rush to a consensus.

KEY WORDS: lunar; basalt; highland; magma ocean; europium

	INTRODUCTION

TOP ABSTRACT INTRODUCTION CONTENTS AND ROUTE MAP LUNAR PETROGENESIS 1969-1999 NOTES Notes added in proof APPENDICES REFERENCES

 
The background to this paper has been presented by O’Hara (2000) together with the substance of the argument in extended abstract form. The material of this paper is presented in a condensed narrative form, accompanied by detailed notes referenced throughout by numbers in parentheses, and by references. The arguments to be presented are complex, with items and groups of information utilized in, or relevant to, more than one thread.

	CONTENTS AND ROUTE MAP

TOP ABSTRACT INTRODUCTION CONTENTS AND ROUTE MAP LUNAR PETROGENESIS 1969-1999 NOTES Notes added in proof APPENDICES REFERENCES

 

View this table: [in this window] [in a new window]

 

	LUNAR PETROGENESIS 1969–1999

TOP ABSTRACT INTRODUCTION CONTENTS AND ROUTE MAP LUNAR PETROGENESIS 1969-1999 NOTES Notes added in proof APPENDICES REFERENCES

 
Petrogenetic background
Concepts in basalt petrogenesis
Terrestrial basalts 1950–1980. The lunar surface is composed of igneous rocks and their impact metamorphosed and impact weathered derivatives. Discussion of their origins cannot be divorced from a consideration of the evolution of ideas in igneous petrogenesis as a whole (0). The Apollo lunar samples were received into a scientific community which had been brought up in the almost unchallenged belief that the several types of abundant terrestrial basalts were primary magmas (0–2). This community was ready to accept that eruption of unmodified primary partial melts of the mantle was a commonplace event on the Moon. Contamination, assimilation and hybridization, which had been extensively explored as mechanisms in igneous evolution before 1950 (1), were becoming less fashionable (2).

Terrestrial basalts 1980–1999. Two assertions that had underpinned the view that primary magmas abounded on Earth were that the common basalts were of great uniformity and that this feature would not survive variable amounts of partial crystallization. Both assertions have been shown to be untrue (0, 5, 26–28) and the problems attending both the definition (6) and eruption (31–34) of primary magmas have been appreciated.

Lunar petrogenesis—stasis and dichotomy. In striking contrast to the situation in terrestrial basalt petrogenesis, the interpretation of lunar petrogenesis which was reached in 1969–1970 (3) has changed little in the ensuing 30 years and displays a striking dichotomy (4). Almost every hand-specimen sample from the maria is supposed to be close to a little differentiated primary magma in composition. By contrast, every igneous rock contributing to the highlands is supposed to be completely differentiated to the point that no trace of its undifferentiated parent magma remains. The only igneous compositions close to the lunar highland average composition are interpreted as clast-free impact melts which formed from rehomogenized highland target materials. A further dichotomy appears in the treatment of remote-sensing data (61), where lunar highland regoliths are assumed to be representative of the chemistry of 40–120 km of underlying crust, yet mare regoliths, which uniformly indicate feldspathic average basaltic compositions, are assumed to be unrepresentative of even the topmost flow units (107–114).

Role of trace element geochemistry
Equilibrium, not perfect fractional processes dominate. Interpretations of trace element geochemistry (7–10, 13), however, apparently provided strong support for a plethora of primary magmas. Erupted basalt sequences have distinctive geochemical features (7), which are incompatible (8) with those expected in products of perfect fractional crystallization (PFC), despite the evidence of a close approach to PFC seen in some large peridotite–gabbro layered intrusion complexes. Equilibrium between liquid and crystals at low and variable mass fractions of melt, on the other hand, produces effects which match the gross variation in basalts world-wide. All workers accepted partial melting of the upper mantle as the ultimate source of most basalts and the melting process could be modelled (9) as equilibrium partial melting (EPM). Equilibrium partial crystallization (EPC) is a process equally capable of explaining the geochemical effects (10) but could be rejected as an explanation for reasons which still appear sound. Appreciation that the actual process in the upper mantle might approach perfect fractional partial melting (PFM) did not undermine these views (11, 12), because the accumulated mixed liquid products of perfect fractional partial melting (APFM) share most geochemical characteristics with the liquids of EPM or EPC [the residues of PFM should, however, look very different from most natural rocks (13) because of the anticipated extreme depletion of highly incompatible elements]. Was it necessary to delve further into basalt petrogenesis (14)?

Possible role for eutectoid PFC. One of the distinctive general features of EPM and EPC at low mass fraction of liquid is wide variation in incompatible trace element concentration with minimal change in major element concentration in the liquid. Even perfect fractional crystallization can produce this effect in the special case of eutectoid crystallization (15), when the solid assemblage separating differs little in major and minor element composition from the liquid.

Imperfect fractionation processes. The differences between the products of PFC (8) or PFM (12, 13) and those of an equilibrium process evaporate rapidly if the process is imperfect in the sense that finite rather than infinitesimal increments of solid or liquid are removed (16).

Model dependence and apparent distribution coefficient. The conclusions from all modelling of trace element geochemistry are wholly dependent on the use of appropriate bulk distribution coefficients for each element. Crystal–liquid distribution coefficient, d, itself may vary with melt composition or element speciation. The apparent bulk distribution coefficient, d_Ap, i.e. that which should be utilized to obtain a satisfactory description of a process or relationship in terms of a simple equilibrium (batch) model, can be greatly modified from the anticipated values of the simple crystal–liquid distribution coefficient, d, by a number of factors. These include (a) changing mutual solubility of crystalline phases at the site and temperature of melting, and (b) zone refining, magma chromatography or polybaric crystallization effects during magma ascent (18). Apparent bulk distribution coefficient may further vary because of (c) forced precipitation of exotic phases on magma arrival and mixing, (d) magma diagenesis of the growing cumulate pile, (e) diffusive differentiation of the magma (19), and (f) trapping of melt in the cumulates (20) or residues (20, 21). Above all, the apparent distribution coefficient is heavily dependent on the choice of a relevant physical model (17, and Fig. 1).

View larger version (33K): [in this window] [in a new window]   Fig. 1. Variation of apparent distribution coefficient with crystal–liquid separation process at any specific crystal–liquid distribution coefficient. The apparent bulk distribution coefficient (concentration in solid product/concentration in liquid product), which should be used in modelling an unknown process using the simple equilibrium (batch) relationship, varies by several orders of magnitude (17). Key to processes: APFC, perfect fractional crystallization, average solid/surviving liquid; APFM, perfect fractional melting, surviving solid/average liquid; EPC–EPM, equilibrium (batch) partial crystallization or melting, a straight line of 1:1 slope in this figure; IFM, imperfect fractional melting, surviving solid/average liquid; IPC, imperfect fractional crystallization, average solid/surviving liquid; IEPC, integrated linear equilibrium crystallization, average mass fraction of surviving liquid 0·5 between limits of 0·0 and 1·0, average solid/average liquid; IPFC, integrated linear perfect fractional crystallization, average mass fraction of surviving liquid 0·5 between limits of 0·0 and 1·0, average solid/average liquid; SPC–EPC, small packet crystallization, equilibrium crystallization in each packet, 0·1 mass fraction of each packet surviving as liquid, 0·5 mass fraction of original melt surviving; SPC–PFC, small packet crystallization, perfect fractional crystallization in each packet, 0·1 mass fraction of each packet surviving as liquid, 0·5 mass fraction of original melt surviving.

 

How much melting?
Trace element requirement. The inference of small mass fractions of melt extraction in terrestrial and lunar basalt petrogenesis is derived from one interpretation of the trace element chemistry (7–13).

Major element requirement. The major element variation in terrestrial peridotite suites (22), however, requires that large mass fractions (>0·25) of partial melt have been removed. Few natural upper-mantle peridotite samples are sufficiently low in mg-number, and rich enough in incompatible trace elements, to support a single-stage EPM origin for common terrestrial basalts (see 47) even at low mass fractions of melting.

Integrated melting regimes. This contradiction between trace and major element inferences was partly resolved by recognition that partial melts must be integrated from regimes with high mass fractions of melting in the centre, but in which small mass fraction melts contributed from the periphery carry the dominant trace element signal (23). Integrated melting regimes logically invalidate many predictions about mantle source regions based on phase equilibria of a candidate primary magma composition because there is in these cases no unique ‘primary’ liquid which was ever in equilibrium with a specific residue.

Simulating and exceeding effects of EPM–EPC in magma chamber products
Integrated crystallization simulates EPC. The process of melt integration is not restricted, however, to the partial melting process. Mixing of residual liquids from variable amounts of PFC (24) yields aggregate liquids whose geochemical properties approach and transcend those of liquids from an EPC process acting on the same parent magma.

Magma recharge and discharge. Field and petrographic evidence establish the importance of magma recharge and discharge during partial crystallization processes (25). Geochemical evolution in a periodically recharged, periodically tapped, continuously crystallized (RTXC) magma chamber (26) spans the range between PFC and EPC. The process facilitates contamination, imposes low-pressure cotectic character on a body of liquid already collected close to the site of eruption, buffers output composition against short-term fluctuations in any of the inputs or the process parameters, and combines this with a simulation of the geochemical effects of EPC or EPM at low mass fractions of liquid. The ratios of recharge to discharge and crystallization are the controlling factors in this simulation. The feedstock for the chemical variations in the pseudo-equilibrium process (i.e. the bulk composition which appears to be undergoing the EPM or EPC process) is, however, the average total input to the magma chamber including contaminants, not some embarrassingly fertile upper-mantle source (27). Ponding and partial crystallization of new input magma may occur before mixing (28).

Small packet crystallization. Recognition that large bodies of magma may solidify by repetitive partial crystallization of small magma batches (29), with mingling of the residual liquid back into the larger pool (SPC), introduced additional complexities. When combined with a variety of partial crystallization–melt aggregation models in the small packets and the whole inserted as the partial crystallization process in an RTXC magma chamber startling results can be obtained (30).

Extremes of magma chamber processes. Concentrations of highly incompatible elements may be doubled in liquid products of SPC–RTXC magma chambers whereas the concentration of a highly compatible element such as Ni (d 10), although always decreasing in concentration relative to the parent magma, may achieve values as great as four times that expected in an EPC process; 400 times that expected in a PFC process (30)—and all this with greatly enhanced discrimination between incompatible elements in a process whose cumulates would everywhere be seen to be the product of local PFC!

Difficulties with primary magma hypotheses
Physical problems. The movement of primary magmas to the surface without modification of their compositions, posed as an industrial problem, would tax the ingenuity of large research teams of chemical and mechanical engineers (31–34). Assimilation of cooler mantle and crust and partial crystallization are likely to be pervasive (31). These effects are likely to be concentrated when magmas arrive at the density contrast at the crust–mantle boundary (32). It is doubtful whether any truly unmodified primary magma can ever be erupted (33). A window of eruptability occurs in relatively magnesian basalt compositions around the condition that plagioclase is about to start crystallizing at the liquidus, but this window could be broadened to picritic compositions by vesiculation (34).

Plagioclase-saturated low-pressure cotectic character. The vast majority of basalts erupted on the surface of the Earth have compositions and temperatures which conform to those of liquids in low-pressure, plagioclase-saturated equilibria (35), an observation which admits three explanations: coincidence of high- and low-pressure equilibria (36, 37); partial crystallization at low pressures with the formation of crustal gabbro complexes (38, 39), preferred in the terrestrial case; or partial melting of gabbro, troctolite or plagioclase–wehrlite at low pressures (40, 41), which has been advocated for asteroid-sized bodies with low central pressures.

Primary magmas should be picritic. The partial melting products of likely mantle peridotites in all the terrestrial planets should be picritic in composition (42) and they may be too dense to erupt unmodified through the crust of the Earth or Moon (43) unless assisted by reduction of bulk density by vesiculation, an effect which can become influential only at low pressures. Picritic compositions also form readily from common basalt at low pressure by the accumulation of dense, early-formed crystals in the magmas (44). Over the last 40 years the number of terrestrial magmas which have been identified as picritic liquids (45) has increased greatly but this identification depends on possession of the field relations and an undisputed knowledge of the average bulk compositions of the lava, both of which are lacking for lunar mare basalts. Even fully accredited picritic liquids have generally undergone substantial partial crystallization and some crystal accumulation subsequent to eruption.

Detection of superimposed melting and crystallization. A crucial point to keep in mind here, and at the end of several previous trains of thought, is that whereas it is easy to detect the geochemical effects of a PFC process superimposed upon primary magmas developed in an EPM process, it is extremely difficult (46) to detect the operation of a partial crystallization process superimposed on the product of a partial melting process, when the first process yields a liquid approximating to an EPM product of the true source and the second process yields liquids which approximate to EPC products of the EPM liquid.

Thermal inertia problems. A simplified interpretation of a suite of samples from such a two-stage magma output might opt for a single-stage partial melting process with very small, but varying mass fractions of melting, or for an excessively enriched source region, or for some compromise between the two. However, the partial melting regime is anticipated to be a large, thermally well-insulated volume (10–100 times that of the derived magma) of material in which short-term fluctuations of the melting parameters are unlikely. The environment of crustal magma chambers is much more susceptible to variations in thermal conditions from one location to another, and to short-term fluctuations in output chemistry, especially when rates of input and output are the critical factors in simulating the appearance of small mass fractions of equilibrium partial melting. The bulk of variation in basalt geochemistry within a single province is inherently more likely to be the result of varying partial crystallization processes than varying conditions or compositions in the source region.

Caution in the use of trace element geochemistry
Increased sophistication in modelling of igneous processes over the past 30 years has invalidated the facile use of trace element geochemistry as a definitive indicator of the presence or absence of low-pressure modification of erupted basalt compositions (47, and see Fig. 1). Some combination of the field relations, petrology, major element chemistry or phase equilibria of erupted basalts frequently suggests a possibility of modification by partial crystallization at some pressure within the crust or upper mantle. In these circumstances, truly sympathetic consideration should be given to the possible effects of those processes before projecting any geochemical features into the source region or the melting process.

Extra-terrestrial petrogenesis
Irony of the lunar mafic hand specimens
Lack of proper field relations, and lack of unambiguous knowledge of the average bulk compositions of the lunar mare lavas, underlies an important part of the debate addressed in this paper (48). It is ironic then that it has been necessary to argue over more than 40 years for largely unseen picritic parental magmas on the Earth [whose existence is still contested in mid-ocean ridge basalt (MORB), volumetrically the most important terrestrial domain of all], yet simultaneously to reject the picritic samples which dominate the lunar mare hand-specimen collections, as samples modified by crystal accumulation.

Solar System exploration post-Apollo
Accepted lunar petrogenesis. ‘Conventional’ lunar petrogenesis (3) is built around a Moon which was volatile depleted from birth. It postulates the generation of a global magma ocean during accretion with flotation of the anorthositic lunar highland crust from the consolidating magma ocean. It then requires the generation of mare basalts by remelting of the feldspar-depleted lunar mantle cumulates formed by consolidation of the global magma ocean. After the Apollo program and the formulation of the ‘conventional’ interpretation of lunar petrogenesis in 1970–1971, a wealth of information, which does not yet seem to have impinged on the interpretation of lunar rocks, has become available for more remote parts of the Solar System.

Volatiles plentiful in asteroids and other Moon-sized bodies. Many asteroids (50) are small rocky bodies which acquired plenty of volatiles during their accretion. There are now six satellites known in the Solar System that have a rock and metal content which would yield a roughly Moon-sized body if all volatiles were removed (51). Four of these bodies retain thick ice crusts, demonstrating that bodies of the same order of size and mass as the Moon could, in principle, accrete with plenty of volatiles and without generating global magma oceans (52, 53 and see 77).

Current Io volcanism a model for Precambrian lunar volcanism. Io, a body closely comparable with the Moon in size and mean density, was found 20 years ago to be the most volcanically active body in the Solar System (54), losing large amounts of sulphur and sodium to space. Surely its style of violent pyroclastic silicate volcanism needs to be considered in relation to that of the Moon, yet two recent authoritative compendiums on the Moon (Heiken et al., 1991; Papike et al., 1998) made no reference to Io and presented the ‘conventional’ view of lunar petrogenesis, which excludes a significant role for reduction and volatilization during eruption.

Ancient feldspathic crust on at least three bodies. Io, together with part of Mars, Venus and the Earth, has preserved little or no early (>3·8 Ga) crust and we can learn little about early planetary evolution from these bodies (55). The Moon, Mercury and parts of Mars all developed and preserved an ancient, heavily cratered light-coloured and probably feldspathic crust (56). The early crust on Mars developed in the presence of water and other volatiles, and appears to be of calc-alkaline and at least partly of pyroclastic volcanic character (57). Neither the presence of volatiles nor the eruptive volcanism lends any support to the concept of formation of a global magma ocean and feldspathic crust flotation during the early history of Mars.

Global melting or anorthosite flotation unnecessary elsewhere. Some small bodies (58), particularly those of the inner asteroid belt, have undergone extensive partial melting, but none of these smaller bodies have developed anorthositic crustal materials (see also 64). Neither internal heating by short-lived isotope decay nor accretional energy is appealing as the cause of this melting because similar-sized bodies (50) have manifestly undergone no melting, and Callisto (4), a much larger body, may have evaded internal differentiation entirely. A localized external heat source is required. Heating by tidal deformation (59), which plays a significant role in Io, Europa and Ganymede today, may have been much more important in the evolution of small bodies such as Vesta and even the Moon in the early years of the Solar System. If heat sources were marginally adequate for melting to occur at all it is reasonable to expect petrogenesis in those bodies which have been melted to be dominated by partial rather than total melting (60).

Lunar highland crust
Petrological composition. Remote sensing of the composition of the lunar highland crust (61) combines with petrological data from recovered samples and lunar meteorites to show that the average highland composition is anorthositic norite, an average that contains a significant ferroan anorthosite (FAN) component. There is also a substantial petrographically identified component of magnesian gabbros and norites and a minor component of petrographically identified mare basalts which are up to 4·2 Ga in age. The relationship of Fe to Th/Ti ratio, which can be obtained from remote sensing and calibrated by sample analysis, requires a much more substantial mare component to be present in the average lunar highland composition, raising a question about the possible mare parentage of some of the magnesian gabbro–norite suite cumulates.

No positive europium anomaly. It had already been decided that the Moon was volatile-poor from birth in the light of the Apollo 11 basalt samples. A global magma ocean in the Moon was proposed in the light of the Apollo 14 highland material return because plagioclase flotation from a large body of basaltic melt seemed the most plausible way to generate a thick plagioclase-rich crust in a volatile-poor body. The widely publicized large positive Eu anomaly in the lunar highlands apparently supported this interpretation, but the remote-sensing data confirm what has always been evident from the original data (Figs 2 and 3). The positive Eu anomaly does not exist (62). This fact, well appreciated at least since 1988, is not mentioned in either of the lunar compendiums mentioned above. There is probably a small negative Eu anomaly in the average highland composition. The bulk of the rare earth elements (REE) in the lunar highlands reside in the KREEPy component, which was apparently excavated by the Imbrium impact and is localized in its vicinity. This leaves open the possibility that the highlands elsewhere do indeed have the low REE contents and large positive Eu anomaly required by the conventional model. The latest survey of Th concentrations across the whole lunar highland surface, however, shows values which are predominantly in the range where small negative or only small positive anomalies would be anticipated. The possibility that a deep-seated KREEPy component is more generally distributed but rarely excavated has also to be considered.

View larger version (42K): [in this window] [in a new window]   Fig. 2. Aluminium/silicon ratios in remotely sensed lunar surface materials suggest that average mare basalt compositions are significantly more feldspathic than the hand specimens and that there is no significant europium anomaly in the average lunar highlands. The vertical axis in the main figure is the Al/Si ratio, which plots information by site, with mare sites to the left and highland sites to the right of the figure. In the left-hand box, histograms are plotted of the frequency of readings from the Apollo 15 and 16 orbital X-ray fluorescence experiments (Adler et al., 1973). Assuming 10 µm penetration in materials with a bulk density of 2 t/m3, each individual reading represents the average composition of about 72 000 t of material, i.e. about 3·5 x 106 t of mare surface and almost 6 x 106 t of highland surface. These reported values may be heavily influenced by the composition of the <10 mm fraction in the soils, which is known to be excessively feldspathic, but the general lack of gradational mixing of mare and highland materials is clear. In the right-hand part of the figure Al/Si ratios in some analysed samples from a range of sites and occurrences are shown, with mare materials grouped to the left of the figure and highland materials towards the right. Randomly sampled mare basalts (clasts in breccias, meteorites, average compositions of lithic fragments in the regolith, small fragments recovered by automated sampling missions) have higher Al/Si ratios than low-pressure plagioclase-saturated cotectic liquids produced in experiments on mare basalts (red crosses), the rock 12038 which is precisely cotectic, and the average groundmass of three Apollo 11 lithic fragments which contain plagioclase microphenocrysts (circle, filled red). Both the randomly sampled basalts and the low-pressure cotectic liquids have systematically higher Al/Si ratios than the majority of the hand specimens, with the possible exception of low-K basalt samples from Apollo 11. The hand specimens in turn have systematically higher Al/Si ratios than the pristine pyroclastic glasses identified at many sites. The mare soils (squares, filled red) have as good a claim to represent average target rock in the top 5–10 m as do the highland soils (below). They have Al/Si ratios which are slightly enhanced relative to the randomly sampled basalts, consistent with a small percentage of observable added fragments of highland-derived materials (values calculated assuming the average hand-specimen composition for the basalt component are consistently higher than the observable amounts). Vertical bars illustrate the differences in composition between the coarse fraction (lowest), the bulk (middle) and the very fine grained fraction (top) in three representative mare soils and three selected highland soils. Mare soil and randomly sampled basalt compositions are consistent with the average composition of the erupted mare magmas being close to those of plagioclase-saturated low-pressure cotectic liquids and even slightly biased towards appearance of plagioclase before pyroxene. The hand specimens and pristine glasses cannot represent the average erupted basalt compositions. Rock samples (squares, filled blue) shown at the highland sites, except for the ferroan anorthosites (FAN), were selected on the basis that they were impact mixed materials whose compositions were in each case likely to average those of large masses of target crust. The soils developed from them (squares, filled cyan) show a more restricted range of composition, entirely within the range defined by the probable components. All data are consistent with an average lunar highland composition which will not display a substantial positive Eu anomaly. Individual clasts of FAN do, however, display substantial positive Eu anomalies. Data sources predominantly Haskin & Warren (1991), McKay et al. (1991), Taylor et al. (1991) and Papike et al. (1998), with other data from sources referenced in the notes. Al/Si is the vertical axis in the inset figure also, which plots this parameter as a function of log(Eu/Sm)N, a ratio which is slightly less than the true value of the Eu anomaly because of the slope of the chondrite-normalized REE patterns between Sm and Gd. The dataset is for all highland materials of Taylor et al. (1991) and shows a reasonable correlation between the two parameters, suggesting that samples with Al/Si <0·65 are likely to display negative, not positive Eu anomalies.

 

View larger version (24K): [in this window] [in a new window]   Fig. 3. Thorium concentration in lunar highland samples plotted as a function of the sum of concentrations of the rare earth elements, and of the magnitude of the europium anomaly in those samples, from the Apollo 15, 16 and 17 missions; all data are from Taylor et al. (1973a, 1973b), Bence et al. (1975) and Taylor & Bence (1975). The open circle shows the point corresponding to the average composition advanced by the two latter groups at 65 ppm total REE with a suggested Eu/Eu* of 1·6 [estimated from Taylor (1975, figs 5.20 and 7.1)] in the right-hand part of the figure. The square symbol shows the plot of rock 68415, interpreted as a recrystallized impact melt rock of close to average lunar highland crust composition.

 

No plagioclase flotation. Plagioclase crystallization and accumulation takes place predominantly at the floor of terrestrial basic magma bodies (63). Plagioclase formed elsewhere in the magma cannot float or sink significantly because of the greatly increased viscosity of magmas as they approach plagioclase saturation. Upward transport of suspended plagioclase by convection in a dry magma ocean should lead to resorption of plagioclase, not formation of an anorthositic crust. Terrestrial magmas are richer in alkalis and more oxidized than lunar magmas, but cumulates from the fragmented howardite–eucrite–diogenite and mesosiderite parent bodies (64) also display accumulation of feldspar with denser ferromagnesian minerals, and provide no evidence of plagioclase flotation or accumulation to form anorthosite.

No lunar magma ocean. Combining these thoughts with those of (62), an origin for the lunar highland composition by plagioclase flotation is excluded (65) and there is, therefore, no petrological support or requirement for the former existence of a global magma ocean in the Moon (66) and grave doubt (from 53, 57, 60 also) whether any body the size of Mars or less developed a magma ocean during its accretion (67). The manifest differentiation into crust and mantle in the terrestrial planets is, in the present state of knowledge, equally well accounted for by partial melting and serial volcanism.

Wet mantle yields feldspathic melt. Leading on from the pyroclastic, possibly andesitic character of early volcanism on Mars (57), the partial melting of peridotites at moderate (>0·1 GPa) pressures in the presence of excess water yields liquids poor in potential olivine which will have plagioclase as an early liquidus phase (68) on eruption and water loss. This crystallization sequence abounds in terrestrial calc-alkaline volcanics but might be complicated in the lunar case by the simultaneous loss of alkalis on eruption. We return now to the conclusions that partial melting is more probable than total melting (60), that the Moon could have accreted plenty of volatiles (51), and that some other origin than plagioclase flotation is required for the lunar highlands (65). The high content of highly incompatible trace elements in the average lunar highland composition (69) points to the early crust being a small average mass fraction partial melt of the whole lunar interior, which would then have had to form under moderate water vapour pressures (70). The small negative Eu anomaly inferred for the average lunar highland composition then suggests the presence of either amphibole or a trace of plagioclase in the residual lunar mantle assemblage. Some might view this suggestion as preposterous in the absence of any hydrous minerals in recovered lunar samples. There can, however, be few more efficient mechanisms for thoroughly dehydrating materials than repeatedly spraying them as hot igneous or impact-generated particles into a vacuum.

Evolving lunar volcanism. Combined with (61) the above leads to two complementary propositions. The early lunar highland crust may have formed by lower-temperature serial water-rich volcanism yielding feldspathic partial melts (37), which had KREEP as one of its differentiation products. Activity may have evolved into water-poor basaltic volcanism as temperatures rose and the body degassed (71). The lunar highland crust is then the accumulated, differentiated partial melt product of that evolving serial volcanism, incorporating a contribution from mare-related plutonics (72).

Feldspathic contribution from early mare component also. The mare components might have contributed further to plagioclase enrichment of the highland crust (see 93 below) if they were derived from a mantle which was close to saturation with an alumina-rich phase. Although this effect alone could have implanted a tendency towards a positive Eu anomaly into the average lunar highlands it would be much subdued if the average lunar highlands contain the pulverized residues of substantial pre-4·1 Ga mare volcanism.

Sulphur, carbon and their gases
Io pyroclastic volcanism driven by these gases. Gases in the sulphur–carbon–oxygen system have driven the pyroclastic volcanism of Io (54), probably for the past 4·5 Ga. The contents of S and C even in consolidated lunar mare basalts are higher than in terrestrial basalts and would sustain volatile fugacities much higher than the lunar surface confining pressure at magmatic temperatures (73).

Lunar pyroclastic volcanism guaranteed. Pyroclastic volcanism similar to that on Io was guaranteed (81) on the Moon, yet this fact receives only a one-sentence comment by Papike et al. (1998) without reference to Io or to the probable consequences for alkali contents of lunar basalts. There is a fuller consideration by Heiken et al. (1991), again without reference to Io, and tending to play down the role of reduction by sulphur loss and possible losses of alkalis during pyroclastic volcanism (see also 80).

Sulphide saturation and siderophile depletion. The lunar basalts were close to saturation with an immiscible sulphide melt (74) and the widely publicized siderophile element depletion in lunar rocks is also predominantly a chalcophile element depletion (75). Any cumulate gabbros underlying the lunar maria might contain sulphide-rich horizons which concentrate the chalcophile trace elements (78) as well as possible metal-rich horizons [a few fragments of both might then be expected in lunar highland breccias if the speculation in (72) is correct]. Overall chalcophile element depletion in lunar surface rocks might owe much to this. Also relevant to the sulphur–chalcophile element story is confirmation that the Moon contains a small core which can be expected to be sulphur bearing if not sulphide rich (76). The strong influence of oxygen fugacity on wetting angles of sulphide melts suggests that segregation of a sulphide liquid to form such a core would have been difficult in a reduced lunar interior but easy in an oxidized Moon (76).

Cerium anomalies and lunar oxygen fugacities
Oxygen fugacities higher than have accompanied either terrestrial calc-alkaline volcanism or Martian volcanism are required at some stage in lunar evolution by the presence of small positive Ce anomalies in many lunar samples, another facet of lunar petrogenesis which has been largely ignored, yet never satisfactorily explained within the ‘conventional’ petrogenesis (77). Similar anomalies in a variety of Antarctic meteorites have fuelled the wild?! (chess terminology) idea that the early lunar crust might have evolved beneath an ice-layer like that on Io or Ganymede (52). If volatiles were relatively abundant in the early Moon, and assuming no major entrapment of hydrogen gas in the body, separation or losses of water from the silicate fraction would be an oxidizing process (79). Subsequent losses of carbon and sulphur as oxide gases will tend to be strongly reducing and the effects of any alkali loss are currently undetermined (80). Given the many orders of magnitude range in the intrinsic oxygen fugacities of terrestrial basalts and the changes which might accompany eruption, there is no a priori reason to think that the oxygen fugacities of lunar basalts reflect those of the lunar interior or that the latter was necessarily uniform spatially or temporally.

Eruption style on a small planet
Fire-fountaining, frothing and volatilization losses. A line of thinking led to the postulate of a volatile-bearing, progressively devolatilizing lunar mantle (71). Given that liquidus vapour pressures even of the final (already much degassed) consolidated mare basalts would have exceeded surface confining pressures (73) it is logical to expect basalts erupted on small planets and asteroids to fire-fountain, to form ash- or droplet-emulsion flows which would spread out with very low effective viscosities, and to froth at their top surfaces even once relatively condensed (81). Such eruptions would maximize the surface area of the melt, maintain the surface temperature of each droplet with minimal cooling in a black-body environment, and minimize the diffusion distances to be covered by components seeking to volatilize (82). That this scenario probably affected lunar basalts is demonstrated by their high intrinsic vapour pressures at liquidus temperatures (73) and by the observation that even after eruption, flow and consolidation, some lunar basalts are highly vesicular. Even in the waning stages of mare volcanism there were enough volatiles to power the pyroclastic eruptions which produced the dark mantle glass bead deposits (83).

Effects of sodium loss. Loss of soda by volatilization has a dramatic effect on the CIPW norm of the residual melt (84), releasing alumina and silica from original albite molecule. These recombine with lime from augite, releasing hypersthene molecule to join that being created by reaction of released silica with original olivine molecule. Terrestrial basic melts subjected to major volatilization losses of sodium would be expected to be transformed into anorthite–pigeonite basalts like those found on the Moon (90).

Asteroidal basalts. The HED (howardite–eucrite–diogenite) parent body (?Vesta) has also yielded a range of anorthite + calcium-poor-pyroxene basalts and dolerites which are low-pressure plagioclase-saturated cotectic compositions (85). The geochemical features of these can be interpreted as products of either primary partial melting (40) or partial crystallization (47) processes. The partial crystallization interpretation is favoured by the presence of cumulate-textured gabbro–norite samples and abundant orthopyroxene cumulates (87), and perhaps by the lack of identified olivine-rich types which might represent the complementary residual mantle. Like lunar basalts, HED basalts are relatively sulphide-rich (88), may have undergone considerable volatilization losses during eruption and must have undergone major losses of volatiles and sodium if the original body was related to chondrite in composition (89). The mantle residues from which the magmas evolved might not then be easily recognizable. The character of the lavas is exactly what would be expected (90) if basic melts of familiar terrestrial or even more alkaline compositions had been subjected to reduction and volatile losses on eruption at the surface of a small planet.

Plagioclase saturation in lunar basalts
No negative Eu anomaly in the lunar mantle. We return now to the complementary conclusion which arises from remote sensing of the average lunar highlands composition (61, 62). The average lunar mantle composition must reflect extraction of the highland crust (91) and must have a small complementary positive Eu anomaly if the bulk Moon has chondritic ratios of the REE. This conclusion stands independent of the debate whether the lunar highland REE signal is dominated by the KREEPy component.

Imposed, not inherited, negative Eu anomalies in mare basalts. Mare basalts cannot then inherit their variable but in many cases very marked negative Eu anomalies as a primary magmatic feature (92) during partial melting of such a mantle. Extensive plagioclase fractionation during partial crystallization at low pressures is the most probable cause of these Eu anomalies.

Low-F, moderate-P primary magmas precipitate plagioclase before pyroxene. The effect of elevated pressure on plagioclase-saturated phase equilibria in the dry basalt–peridotite system is to displace the liquid compositions rapidly towards higher normative plagioclase within the first 0·2 GPa and less rapidly towards higher normative olivine (Fig. 4). The implications are profound—plagioclase should precipitate before pyroxene from ascending primary melts—and seem to have been overlooked in discussions of mare basalt petrogenesis (93). If oxygen fugacities were low and plagioclase were a residual phase in the lunar mantle during partial melting, the Eu anomalies of mare basalts might have been explained as a primary feature, but the major element compositions of pyroclastic glass beads, hand specimens and even the feldspathic basalts are too poor in plagioclase (93) to support this mechanism.

View larger version (26K): [in this window] [in a new window]   Fig. 4. a1,2. Results of phase equilibria studies on compositions in the system CaO–MgO–Al2O3–SiO2 at pressures from atmospheric to 10 GPa, redrawn and simplified from Herzberg & O’Hara (1998, fig. 1) where the two sub-projections, a1 from diopside and a2 from forsterite, are explained further. b1,2. Analogous results from experiments on natural basalts and peridotites, redrawn and simplified from Herzberg & O’Hara (1998, fig. 3) where the two sub-projections, b1 from a diopside-like component and b2 from an olivine-like component, are also explained further. Bold lines in each figure record the positions of phase equilibria boundaries at low pressures. Fine lines record the loci of liquid compositions in equilibrium with olivine, clinopyroxene, orthopyroxene and either plagioclase, spinel or garnet as the pressure increases from low (0 GPa) to high (5 GPa). Liquids precipitating, or formed by small mass fractions of partial melting of assemblages of olivine, two pyroxenes and either anorthite, spinel or garnet at pressures below 5 GPa (approximately the central pressure in the Moon) will be expected to precipitate olivine followed by plagioclase before pyroxene if erupted as unmodified primary magmas, as indicated by the dotted projection line in a1 from the olivine apex through the 1 GPa melt composition to the low-pressure boundary of plagioclase saturation. In a2 it is seen that this ‘projection’ from olivine will indeed result in the composition encountering plagioclase crystallization well before pyroxene, and that this will be true also for liquids formed anywhere along the locus up to close to 5 GPa. In natural multi-component systems there is a range of liquid compositions which may form in equilibrium with the alumina-saturated lherzolite assemblages as mg-number, etc. varies, the extent of which is indicated in the projections b1 and b2. This extra freedom does not alter the argument. Most of the change in composition along the loci between 0 and 1 GPa is accomplished in the first 0·2 GPa.

 

MORB, CFB crystallize plagioclase early. Terrestrial basaltic partial melts have formed by larger average mass fractions (0·1) of partial melting than are supposed in the lunar case. They almost certainly separated from alumina-undersaturated, feldspar-free harzburgites. Nevertheless, these melts are typically so rich in potential plagioclase that precipitation of extensive olivine and plagioclase, the common phenocryst phases in MORB and continental flood basalt (CFB), precedes the arrival of the residual melt at the low-pressure cotectic equilibria where augitic pyroxene also begins to precipitate (94, 95).

Incorrect low-P phase equilibria of alleged mare primary magmas. None of the lunar pyroclastic glass beads or alleged primary magma hand-specimen compositions displays the required crystallization sequence and most encounter pigeonite saturation before the entry of either augite or plagioclase (96). If these had been the parental magma compositions, and given (91), there would be no way of generating the observed negative Eu anomalies (97).

Alleged primary magmas do not display required moderate-P phase equilibria. A corollary of the issues discussed in (93) is that true primary magmas formed in the manner required in the ‘conventional’ lunar petrogenetic scheme should show simultaneous saturation with plagioclase, pyroxene and olivine at their pressure of formation (98), but none do.

Lunar primary magmas and absence of tectonic deformation
Global magma ocean cumulates unstable from birth. Deposition (solidus) temperatures in cumulates from a global magma ocean would decline towards the surface because of two effects—the declining liquidus temperature of any dry mafic magma with declining pressure, and the decreasing mg-number of a differentiating magma ocean. Liquidus thermal gradients in dry mafic and ultramafic materials are typically supra-adiabatic, an effect which underwrites the partial melting of mantle plumes in the terrestrial mantle. The thermal gradient at deposition of a thick global cumulate would greatly exceed the adiabatic gradient, promoting convective motion which would be enhanced by the density inversion implicit in a plagioclase-free cumulate sequence of declining mg-number, especially where later cumulates contained ilmenite. There is no tectonic evidence to suggest that the anticipated deformations took place.

Supra-adiabatic thermal gradient still required 1 Ga later. The two-phase (olivine + orthopyroxene in most cases) saturation which occurs in the putative mare basalt primary magma compositions at a variety of pressures and temperatures has no special petrogenetic significance (99). If petrogenetic significance is imputed, the array of co-saturation conditions would define a pressure–temperature gradient which is grossly supra-adiabatic (100), has to be supposed to have persisted over at least a billion years through 500 km depth of mantle, and yet has produced no evidence of tectonics caused by convective motion (101) even where the lunar crust is no thicker than the continental crust of the Earth. Consequently, the glass-bead and hand-specimen compositions cannot be primary partial melts of the alleged cumulate mantle (102). No problems with global supra-adiabatic thermal gradients need arise in the alternative petrogenesis proposed here.

Average compositions of mare basalts
Quench crystal sinking. The low viscosity of the alkali-poor mare basalt lavas, even in their condensed state, ensures significant sinking of crystals, including the large zoned metastable phenocrysts formed during quenching (103). Settling rates may have been more rapid if quench phenocrysts formed during the pyroclastic phase in the eruption. Quench texture throughout a lunar hand specimen is no criterion of the former existence of its bulk composition as a liquid. Some, possibly many, of the samples must be enriched in ferro-magnesian phases by accumulation of the quench phenocrysts (104). Petrographic variability at each site is more extensive than was then known at individual terrestrial sites (but see 133).

Flow thicknesses. The general lack of flow fronts in the maria points either to very fluid flows (103) or to flow thickness (105) comparable with or less than the regolith depth (5 m), in which case the average regolith composition should be close to the average lava composition (106). Estimates of cooling rates required to produce the observed petrographic textures in hand-specimen samples, on the other hand, suggest much thicker cooling units (107), in which case the regolith, being restricted to the top 5 m, may preferentially sample a phenocryst-depleted zone and not represent the average composition of the lava. Small impacts into the maria excavating to 100 m depth (108) should, however, sample the average lava composition accurately whichever estimate of flow thickness is correct, and the same should be true of materials exposed along the walls of the large rilles. There is no evidence to suggest that the regolith surface is not representative of materials to several tens, even a hundred metres deep in the maria (109).

Persistently feldspathic regolith compositions. Remote sensing of the mare surfaces (110) indicates average compositions much more feldspathic than all but a few of the large hand specimens from the mare landing sites (Fig. 2). These regoliths are expected to comprise about 95% locally derived material. The remote-sensing results are in good agreement with the limited ground truth established by returned regolith samples (111). They also agree with the feldspathic basaltic compositions of lithic fragments, breccia fragments, lunar mare meteorites and impact-generated glass groups in the regolith (112–116), which leads on to the conclusion, reinforced by (103–109), that the hand-specimen samples do not represent the average consolidated liquid compositions (117).

Composition bias in hand specimens. The hand-specimen compositions are inevitably biased towards materials excavated relatively recently from the top of the bedrock at 5 m depth, as well as towards more cohesive samples from shallower depths, because longer exposed and more friable materials have become preferentially comminuted (118). It is quite proper to accord all such samples equal weight when selecting them for investigation, but potentially misleading to accord each of them equal weight when arriving at an average mare basalt composition which ignores the regolith contribution. This entire line of reasoning (103–118) reinforces the phase equilibria, petrographic and geophysical arguments (93–101) pointing to the conclusion that hand-specimen compositions cannot be primary partial melts of the alleged cumulate mantle (102).

Low-P plagioclase-saturated cotectic average compositions. Information and samples made available in a random manner (110–116) indicate that the average mare basalt composition is close to that of a low-pressure cotectic, plagioclase-saturated basalt which could be the residual liquid of a low-pressure gabbro–norite crystallization process (120) during which enrichment in incompatible elements such as REE and titanium and negative Eu anomalies could be generated. These magmas might have been contaminated by highland crustal materials during that process.

Misfits in experimental data. Objections to this interpretation based on small misfits in the experimental data, specifically small mismatches in the mg-number of liquidus olivines and the failure of armalcolite crystallization to overlap with that of plagioclase in some results, can be discounted because of the problems in precise control of charge compositions and oxygen fugacities (119).

Petrogenesis of lunar high-titanium basalt. The petrogenesis of the high-titanium basalts of Mare Tranquillitatis and Taurus Littrow collected by the first and last manned missions of the Apollo program encapsulates the discussion about mare basalts and is pursued at greater length in Figs 5–9. These rocks are alkali-poor basalts very rich in TiO₂, relatively rich in FeO relative to MgO, relatively rich in incompatible trace elements (REE 20–100 x chondritic and higher in Apollo 11 than Apollo 17 samples), with marked negative Eu anomalies (Eu/Eu^* 0·8–0·3). These magmas have been suggested to represent either (i) very small mass fraction partial melts (1% or less by mass) of the postulated cumulate mantle, or (ii) late residual liquids (last 50–10% of the parent magma assuming a 10% initial partial melt) of differentiation involving removal of plagioclase and other minerals at low pressure.

View larger version (38K): [in this window] [in a new window]   Fig. 5. Figure 5 displays the temperature and oxygen fugacity in experiments on 12 Apollo 17 hand-specimen samples for which phase equilibria and phase composition data are reported in the Appendix. Figures 5–8 address the origin of the lunar high-titanium basalts, the first rocks returned from the Moon, which are relatively rich in incompatible elements and display marked negative Eu anomalies. Interpretation of the petrogenesis of this group encapsulates a more general problem common to many terrestrial basalts also. Does the incompatible enrichment reflect large mass fractions of partial crystallization subsequent to magma formation, or low mass fractions of partial melting in the first instance? The negative Eu anomalies demand substantial prior removal of plagioclase from the system, which presents no problem if the basalts are low-pressure plagioclase-saturated cotectic liquids (see Fig. 2) whose compositions are controlled by gabbro fractionation within the lunar crust. The conventional interpretation supposes, however, that prolonged plagioclase flotation from a lunar magma ocean had imparted a negative Eu anomaly to a mantle cumulate pile. This then underwent small mass fractions of renewed partial melting to yield the parent magmas of the mare basalts, represented by the hand-specimen compositions, a hypothesis which encounters difficulties outlined in Figs 4, 7 and 8. A significant part of this debate has concerned the accuracy and relevance of phase equilibria studies on the natural samples, and is complicated by the fact that at low oxygen fugacities precise control of oxygen fugacity and charge composition greatly influence the observed phase equilibria. Figure 5 shows the oxygen fugacity–temperature relationship of the Fe–‘FeO’ (wüstite) equilibrium at which many experiments on Apollo 11 samples and related materials were carried out in Mo containers. Circles show conditions of runs, reported in tables in the Appendix, in Mo-foil capsules, which are numbered as in the tables (unless otherwise shown, these numbers should be prefixed with a 5). Squares show conditions of runs in Fe-foil capsules, similarly numbered. The intrinsic oxygen fugacity–temperature relationships of two Apollo 17 hand specimens (Sato, 1978), and of one of these samples in contact with excess Fe metal are also indicated. An approximate boundary at which metallic iron appeared in sample 70215 in the experiments reported here is also shown. The approximate boundaries of plagioclase entry and armalcolite appearance are indicated, but plagioclase was also present in 70275 in runs 381, 389 and 403. Spinel is present in most charges at oxygen fugacities higher than those of 70017 + Fe metal. Plagioclase crystallization overlaps with that of armalcolite in a substantial number of these experiments. Conditions in these experiments are seen to be at least as relevant to the behaviour of lunar basalts as those carried out in ‘pure’ iron containers in ‘sealed’ silica glass tubes.

 

View larger version (31K): [in this window] [in a new window]   Fig. 9. A ‘simple’ plot of Al2O3 against CaO in Apollo 11 samples to illustrate proximity of low-K group basalt compositions to the low-pressure plagioclase-saturated cotectic liquids and apparent controls of compositions within the high-K and low-K groups. This is actually a very complicated projection from whatever other components happen to be in each sample, in whatever ratios they happen to be present, with no regard to how these components combine in the permissible phases; i.e. each point is projected by a different method geometrically! The figure displays differences in the amounts of two components which are associated in plagioclase and are in the main excluded from the dense ferromagnesian and oxide phases (with the exception of calcium-rich clinopyroxene) which may have sunk within the lavas. It conceals all information about relative silica saturation and titanium concentration and cannot be used uniquely to relate observed phase equilibria to the sample compositions, but in general plagioclase can be expected to crystallize before pyroxene from liquids to the CaO–Al2O3-rich side of the experimental cotectic liquids. The composition fields of the high-K and low-K hand specimens [averages of groups B1–3 and D from Rhodes & Blanchard (1980), breccias from Rhodes & Blanchard (1981)] are shaded and labelled. Regolith breccias (small filled circles) and their average (large open circle) are significantly more aluminous, extending up to the open squares representing the soil compositions (which certainly contain some highland-derived component). The plagioclase-saturated cotectic liquids developed in 10020 and 10062 (low-K group), and those in 10017 and a synthetic mixture simulating the composition of the average high-K lithic fragments (open triangles), lie within a restricted field. A field is also shown enclosing three analyses of the vitrophyric groundmasses enclosing plagioclase phenocrysts from lithic fragments in the soil (O’Hara et al., 1974) which also encloses the composition of monomict breccia 10056. This figure supports the inference from Figs 7 and 8 that the low-K Apollo 11 basalt hand specimens are close in composition to low-pressure plagioclase-saturated cotectic liquids. Also shown are two control lines (dashed). A–B passes through the cotectic liquid composition of 10017 and extends towards the compositions of plagioclase phenocrysts, An94, and reflects some strong control on the compositions of high-K hand specimens, breccias and soils. This control line is marked to show the effect of adding or subtracting percentages of plagioclase (or subtracting or adding an alumina-free but lime-containing cumulate) to produce this displacement from the 10017 cotectic liquid composition. Control line X–Y passes through the cotectic liquids in low-K basalts and extends backwards towards a Ca, Al-free cumulus (olivine plus oxide phases only).

 

View larger version (31K): [in this window] [in a new window]   Fig. 8. High-titanium basalt hand specimens, groundmasses, average lithic fragments, soils and plagioclase-saturated experimental liquid compositions compared in the isostructural equivalent weight projection FACKTS (O’Hara & Humphries, 1975) as in Fig. 7, but using a different sub-projection of the data which conceals differences in olivine and titanium-rich oxides (early crystallizing and possibly accumulating phases in these basalts) and displays essentially the differences in feldspar concentration relative to pyroxene components (all three late crystallizing phases in these materials). Composition fields are indicated in which calcium-poor pyroxene (lime green), calcium-rich pyroxene (green) or anorthite (blue) are early crystallizing phases, with olivine and oxide phases. These fields are arranged about an intersection, the low-pressure plagioclase-saturated cotectic liquid compositions, where all these phases may be simultaneously saturated at the liquidus at 1120–1150°C. All the hand-specimen compositions project conspicuously close to the cotectic liquid compositions in this projection, consistent with the hand specimens being olivine + oxide mineral cumulates into low-pressure plagioclase-saturated liquids. Data at higher pressures for rocks 10017, 70017 and 70215 and for soil 10084 are consistent with the equivalent cotectic liquid lying close to the point P at 0·5 GPa. True primary liquids, formed as required in the conventional interpretation, should have compositions close to P, should show early plagioclase crystallization and should not project where the hand-specimen compositions fall. The sub-system OL–ILM–OR–FS–CPX–AN of the projection explained in the caption to Fig. 7 is calculated with OL = (F* + A* + C* + K* - T* - S*) x 203·777 (mol. wt fayalite), ILM = T* x 151·746 (mol. wt ilmenite), OR = K* x 278·2; FS = [- F* - A* - 3(C* + K*) + T* + 2S*] x 131·93 (mol. wt ferrosilite), CPX = (C* + K* - A*) x 248·1 (mol. wt hedenbergite) and AN = (A* - K*) x 278·2 (mol. wt anorthite), where the asterisked quantities are the molar, not equivalent weight, quantities calculated initially. The components of selected sub-groups are then scaled to 100%. The projection within the sub-system in Fig. 8 is from OL, ILM and OR into the plane of total plagioclase (FELS)–calcium-rich pyroxene (HED)–calcium-poor pyroxene (FS). The central diagram, (a), compares the compositions of soils and hand-specimen rocks [details in (c)], experimental and natural plagioclase-saturated liquids [details in (d)], and the average compositions of basalt fragments in the Apollo 11 soils. (b) interprets the phase equilibria reported in a range of synthetic compositions (O’Hara et al., 1970b) which yield a phase diagram closely similar to that deduced from the natural samples.

 

View larger version (28K): [in this window] [in a new window]   Fig. 7. High-titanium basalt hand specimens, groundmasses, average lithic fragments, soils and plagioclase-saturated experimental liquid compositions compared in the isostructural equivalent weight projection FACKTS (O’Hara & Humphries, 1975), details of which are given below. Composition fields are indicated in which olivine or calcium-poor pyroxene (yellow), a titanium-rich oxide mineral (grey) or anorthite (blue) are early crystallizing phases. They are arranged about an intersection, the low-pressure plagioclase-saturated cotectic liquid composition, where all these phases and calcium-rich pyroxene may be simultaneously saturated at the liquidus at 1120–1140°C. Some hand specimens from the Apollo 11 low-K group and Apollo 17 samples 75035 and 75055 appear conspicuously close to the cotectic liquid compositions in this projection. Data at higher pressures for rocks 10017, 70017 and 70215 and for soil 10084 (O’Hara et al., 1970b; Longhi et al., 1974) are consistent with the equivalent cotectic liquid lying close to the point P at 0·5 GPa. True primary liquids, formed as required in the conventional interpretation, should have compositions close to P, should show early plagioclase crystallization and should not project where the hand-specimen compositions fall. If the interpretation offered in this paper is correct, the hand-specimen compositions should plot in the triangle linking the plagioclase-saturated cotectics to olivine and to titanium oxide phase compositions, and with a little imagination they do. Compositions are first converted to six projection components thus: F = (MgO + FeO + MnO + NiO + Cr2O3 + Fe2O3) x 71·846 (mol. wt FeO); A = (Al2O3 + Na2O + K2O) x 101·961 (mol. wt Al2O3); C = (CaO + 2Na2O - 3·333P2O5) x 56·08 (mol. wt CaO); K = 2K2O x 94·203 (mol. wt K2O); T = (TiO2 + Cr2O3 + Fe2O3) x 79·90 (mol. wt TiO2); S = [SiO2 - 2(Na2O + K2O)] x 60·085 (mol. wt SiO2); where all oxide symbols represent the number of moles present in the analyses. The C and K components may be combined and treated as CaO, thus superimposing all feldspar components at a single point, when it is not desired to separate materials on the basis of their potassium contents. The sub-system FM–SIL–T–CPX–FELS is calculated with FM = (F* + A* - C* - K*) x 71·846, SIL = (S* - 2C* - 2K*) x 60·085; CPX = (C* + K* - A*) x 248·1 (mol. wt hedenbergite) and FELS = A* x 278·2 (mol. wt anorthite), where the asterisked quantities are the molar, not equivalent weight, quantities calculated initially. The components of selected sub-groups are then scaled to 100%. The projection within the sub-system in Fig. 7 is from FM and CPX into the plane of total feldspar (FELS)–titanium oxides (T, where armalcolite, ulvöspinel and ilmenite project together)–available silica (SIL), where olivine and calcium-poor pyroxene project together. This projection conceals differences in calcium-rich clinopyroxene (not an early phase in these basalts) and displays essentially the differences in feldspar concentration between the various materials, and their ratios of oxide phases to olivine and calcium-poor pyroxene (all but the last of which could be accumulating crystal phases). The central diagram, (a), compares the compositions of hand-specimen rocks [details in (c)], experimental and natural plagioclase-saturated liquids [details in (d)], mare and massif soils and the average compositions of basalt fragments in the Apollo 11 soils. (b) interprets the phase equilibria reported in a range of synthetic compositions (O’Hara et al., 1970b) which display a more restricted field of early oxide crystallization than the natural samples because of the lack of chromium in the system.

 

View larger version (29K): [in this window] [in a new window]   Fig. 6. Schematic representation of atmospheric pressure phase equilibria anticipated in a representative high-titanium basalt hand-specimen composition as a function of oxygen fugacity and temperature, based on available experimental data and inspired prediction. Crystallization conditions observed in experiments which maintain charge composition, and are at an externally controlled or imposed oxygen fugacity, will follow curves such as those shown for the iron–wüstite equilibrium or an 80% H2–20% CO2 gas mixture (both dotted) or in the presence of iron (parallel bold curve through B–F). A bewildering array of crystallization sequences might be observed according to the precise experimental conditions achieved. Phases present in addition to liquid are: An, plagioclase; Arm, armalcolite; Fe, iron metal; Ilm, ilmenite; Ol, olivine; Px, pyroxene; Sp, spinel. Three bold boundaries mark the limits of stability of Fe metal, armalcolite and spinel, all of which are constrained principally by the oxygen fugacity. Finer boundaries illustrate the appearance of olivine, ilmenite, pyroxene and plagioclase, which are constrained principally by the temperature achieved. The sketch assumes that bulk charge composition has been maintained. The FeO content of the silicate charge declines rapidly to the left of the steep bold boundary (B–F extended) marking the appearance of metallic iron in the composition with falling oxygen fugacity. The FeO content of the charge decreases gradually to the right of this boundary as it is oxidized progressively to Fe2O3, and in both directions the mg-number of silicates will increase as FeO content decreases. Iron gain by the charges by oxidation of an Fe container will shrink the bold boundary of the armalcolite crystallization field (A–B–C–D–E–F) to lower oxygen fugacity, i.e. to the left, and lower the temperature of entry of ferro-magnesian silicates in general. Iron loss to impure Fe containers by reduction, or to Mo containers at lower oxygen fugacity, will enhance the field of armalcolite crystallization and raise temperatures of entry of the ferro-magnesian silicates. The bold boundary of spinel crystallization (G–H–C–J–E) is strongly dependent upon the oxygen fugacity because of its effect on the oxidation state of chromium. Spinel crystallization is inhibited in the presence of metallic iron as a container or as an equilibrium phase in the charge. The pyroxene crystallizing in these compositions is calcic at and to the right of the boundary marking the entry of iron, but will become increasingly sub-calcic with the possible appearance of separate pigeonite to the left of this boundary as potential olivine reacts with silica released by the reduction of FeO. All these effects are believed to be displayed in available experimental data for the Apollo 17 hand specimens. The argument in this paper is that the natural hand specimens have been formed by accumulation of the dense minerals olivine, spinel, ilmenite and armalcolite into a multiply saturated cotectic liquid and erupted basalt composition close to that present at E, and that this interpretation should not be rejected because of small discrepancies in some of the reported phase equilibria obtained by different groups in different laboratories using different techniques on different sub-samples of the rocks.

 

The first hypothesis is incompatible with the high-pressure phase equilibria of the hand-specimen compositions (93–100) and will also be incompatible with the high-pressure phase equilibria of the average erupted magma compositions advocated here (110–117), but less strikingly so. To validate the second hypothesis the phase equilibria of the average erupted magma compositions are required to demonstrate near-simultaneous entry at the liquidus of all the mineral phases required to be present in the hypothetical cumulates, i.e. plagioclase, pyroxene, ilmenite, probably olivine, possibly armalcolite and spinel as well. If the hand specimens represent samples of average liquid enriched in phenocrysts of dense olivine, ilmenite, armalcolite and spinel formed during quenching after eruption as advocated here then, subject to the points raised in (119), the hand-specimen samples should show near-simultaneous entry of pyroxene and plagioclase when the phenocryst phases are still in equilibrium with the liquid.

Data presented in (121), comprising with those published by O’Hara et al. (1970a, 1970b) by far the largest datasets for low-pressure crystallization of these rock types, are interpreted in Figs 5–8, and demonstrate that the phase equilibria are appropriate for the second hypothesis, making due allowance for the presence of a small amount of highland component in the soils and of some pyroclastic bead material in the Apollo 17 soils.

Cognate crustal cumulates. The presence of extensive bodies of cognate gabbro–norite and probably peridotite–pyroxenite is predicted somewhere within the crust beneath the mare surfaces (105, 122), some older members of which may be sampled in the highland breccias (72).

Comparative petrogenesis
Meantime, back on Earth ...
Poor terrestrial controls. Most of the rock materials now forming the lunar highland crust were in place, the period of heavy bombardment and large basin formation was over, and the filling of the western maria complete by 3·7 Ga ago. The terrestrial record from that period is sparse and fragmentary, providing little in the way of control over interpretations of lunar geology.

Magma production in large impacts. Large basin-forming impacts undoubtedly produce large volumes of impact melt (123), part at least of which may become clast free. In the small-basin-sized terrestrial impact basin at Sudbury (124) the whole body of norite and gabbro with its associated nickel sulphide deposits may be a pool of impact melt which cooled slowly enough to undergo extensive fractionation. Mafic magma may also form within the target body by partial melting triggered by a combination of pressure release accompanying the basin excavation and shock-implanted energy. This internally generated magma would have to rise through shattered upper-mantle and crustal rocks in the breccia lens, when it would have a high probability of becoming contaminated (125). If it arrived soon enough, it would be liable to hybridize with the purely impact-generated melt in the crater.

Bushveld complex a model for mare filling? The 2·2 Ga Bushveld complex of southern Africa has been put forward as a possible large terrestrial impact basin filling (126) and it has several of the features one might seek: lack of tectonic association with an internal plume-generated event, extremely high magma production rate, pervasive magma contamination, and mare-like dimensions. Extreme differentiation by partial crystallization of norites and gabbros in this body is complicated by many magma recharge events, accompanied by copious chromite and sulphide precipitation. Large outflows of magma, whose compositions were constrained to be multiply saturated with plagioclase, pyroxenes and other phases at low pressure, are deduced. These factors outline an environment similar to that required to explain the lunar mare basalts (122). The Bushveld lacks, however, any unambiguous evidence of origin as an impact basin and there may be some differences in timing which distinguish it from the lunar maria. R. G. Cawthorn (personal communication, 1999) has, moreover, drawn attention to the impoverishment of the cumulate section in sulphur.

Komatiite–greenstone analogies. Some of the terrestrial komatiite–greenstone sequences date from a period overlapping that of lunar mare filling (127) and it has been suggested that they might be analogous features. Expanding knowledge of both komatiite–greenstone and continental flood basalt provinces, however, has not strengthened the desired connection. The stratigraphic records in these provinces typically open with the eruption of more primitive magma types, which have a better claim than average to approximate to primary magma compositions. They then proceed through thick sequences of basalts whose compositions are controlled by partial crystallization somewhere within the crust and terminate with types which may be seriously contaminated or hybridized.

Realities of mare basin filling. Mare basalt filling took place over a period of about 2 Ga from >4·2 Ga ago (128) and was at least partly independent of the basin-forming process because large basins are known which are virtually devoid of basaltic fill. Magma supply was sustained over a long period of time. Even the last few surface flows at a single site may span as much as 250 Ma, but there are few constraints on the time required to emplace some 99% of the underlying mare fill, beyond noting that in the western maria the oldest surface basalts are almost as old as the basins themselves (129), consistent with very rapid initial filling of the basins. There is no association with plume-like motions in the underlying mantle. Individual flow volumes must have been very large but total magma production was low.

Dubious continental flood basalt analogies. From the outset, majority opinion among Apollo scientists favoured an analogy between lunar mare basalts and CFBs, partly under the mistaken belief that CFB provinces were arrays of primary magmas. Terrestrial CFB events by contrast are strongly plume related, marked by very high magma productivity and high total magma production (130) and of very short duration (2 Ma) although any more extended igneous activity in CFB may be detached from the main centre by the onset of active spreading at a divergent plate margin and passive margin subsidence.

Flood basalts and lava lakes. Whenever eruptions are sufficiently voluminous and sufficiently rapid, lava lakes may form and there could be no more convenient site for the low-pressure differentiation of basaltic magma (131). Rates of irruption in komatiite–greenstone and CFB provinces have manifestly been too low for lava lake production but do seem to have supported formation of large crustal or sub-crustal magma chambers (127).

An ocean-island basalt or mid-plate volcanism analogy? The mare basalt samples recovered by the Apollo missions must represent the final eruptions from waning volcanic cycles, whichever model for mare volcanism is adopted, and may not be typical of the average mare fill. In a non-convecting Moon which provides no recharge of fertile source material into the melting regime, these late eruptions might be expected to be thoroughly differentiated late-stage residual liquids in which volatile contents might have built up to relatively high levels (132). The late stages of terrestrial plume-related volcanism producing ocean-island basalts yield very small volumes of magma rich in volatiles and incompatible elements, often relatively rich in titanium. As usual with terrestrial magma types, there is one body of opinion which regards these as very small mass fraction partial melts of a fertile mantle source. There is another which regards them as residual liquids from advanced high-pressure fractional crystallization at 2·5 GPa because the requisite garnet pyroxenite and eclogite precipitates are carried up in explosive eruptions. The much reduced pressure gradient in the Moon would make such high-pressure evolution improbable but the possibility of an analogous evolution at 1·0 GPa, with the final stage magmas not directly related geochemically to the main mare fill by low-pressure events, should be entertained.

Back to the Moon
Limited lunar field controls. We are attempting to infer the field relations of the lunar basalts from their experimental petrology and trace element geochemistry, an unfortunate way to have to progress. Controls based on terrestrial examples where the same had been attempted were inadequate (133). Rock exposures in the walls of Hadley Rille (134) and the fallen blocks on its floor were confidently identified as basalt flows when the photographs came back from Apollo 15, but the upper 60 m is poorly exposed. The underlying layered units and the fallen blocks appear much more massive and less conspicuously jointed than typical basalts, where columnar jointing commonly develops in thick flow units. Other observations from Hadley Rille and its surroundings point to the availability of very large volumes of fluid magma in a single event, and to lava ponding and magma withdrawal (see 107, 131).

Gabbros at shallow depth in the maria? Although there is no indication of major compositional variations in the uppermost layers of the maria (109), larger craters which excavate material down to 1 km depth do in some cases excavate more mafic or more magnesian materials (135) and at least one lunar meteorite, Asuka 881757, has sampled an ancient coarse-grained mare gabbro (136), further observations consistent with the presence of gabbros at no great depth below the surface basalts (122). The inferred rock assemblage would closely resemble that seen in samples from the HED parent body (85–90).

Pyroclastic flow units. The final eruptive events in the maria were pyroclastic eruptions (83). The requirements imposed by the morphology of most mare basalt flows (105) and that of the earliest basalt flows on Mars (137) suggest that the majority of mare-filling eruptions may have been mafic pyroclastic flows. Such eruptions, especially those into low confining pressures, would have been accompanied by extensive selective volatilization from the magma (138). Coupling this with the low-pressure cotectic character of the basalts (120) the conclusion is reached that the mare basalt parent magmas are not preserved at the lunar surface (139). Little can be reliably deduced from the lava geochemistry about the detail of the lunar mantle without first unravelling the effects of modification at low pressures (33).

Pyroclastic bead compositions. It has to be conceded by all parties to the debate that there is no simple mechanism of closed system magmatic evolution which can link the compositions of the late pyroclastic glass beads to those of the hand specimens (140) and the matter is best left without further speculation until the issue of the average composition of mare basalt (120) has been resolved by the observations suggested (134).

Finale
Mars, science and politics. Supremely expensive science projects are ultimately funded from the public purse by politicians who may legitimately feel that the interests of their constituents and their own chances of re-election are better served by spending closer to home. Vigorous debate about interpretation, the democratic lifeblood of science, can easily be portrayed as reprehensible disarray and used as an argument for diverting that funding towards more ‘worthy’ objectives. It is not always clear where the balance of advantage will lie between quality of debate and total quantity of science achieved.

The first returns of systematically collected, well-located samples from Mars, stamping ground of the largest central volcanoes, largest calderas and some of the most extensive spreads of basalt in the Solar System, are almost upon us. It is to be hoped that there will be no Gadarene rush to a consensus of interpretation. An extensive, protracted and dispassionate examination of a wide range of multiple working hypotheses would be appropriate.

Basalt petrogenesis—a Solar System round-up. Evidence relating to basalt genesis from seven planetary bodies has increased greatly in the past 30 years, and that for extensive evolution of basaltic magmas between source region and vent has multiplied. Central volcanic complexes and calderas on the Earth are associated with high-level magma chambers, partial crystallization of magmas and eruption of residual liquids biased towards low-pressure cotectic compositions. High-level magma chambers cannot, however, be the site of whatever low-pressure modification has affected continental flood basalt compositions—subcrustal magma chambers are preferred. The crust of Venus is riddled with central volcanic complexes which suggest an abundance of high-level magma chambers in which partial crystallization of parental magmas might occur. The surface of Mars has the largest central volcanic complexes and some of the largest calderas known in the Solar System, again potential sites of advanced low-pressure partial crystallization. The majority of lavas erupted on the surfaces of Mars and Venus are likely to be extensively modified by partial crystallization and assimilation within the crusts and central volcanic superstructures on those planets; most Shergottite–Nakhlite–Chassignite (SNC) group meteorites derived from Mars are cumulates.

There is an abundance of central volcanic features and calderas on Io. The lavas erupting on Io are anticipated to be evolved basalts on the basis of the existence of numerous large calderas which imply extensive high-level magma chambers and the probability of refluxing of the partial melt compositions at low pressure over a period of 4·5 Ga. The parent planet of the basaltic achondrite meteorites had a crust covered with low-pressure cotectic plagioclase-saturated basic extrusives. These 4·5-Ga-old lavas have the sodium, volatile and siderophile element depletion and the high sulphur abundance of lunar mare basalts. They display a range of negligible to moderately negative Eu anomalies. Their geochemistry can now be interpreted as products of crust-forming, periodically recharged, periodically tapped magma chambers perhaps afflicted by some form of small packet crystallization. Complementary slowly cooled orthopyroxenite and gabbro cumulates are known among the meteorites and some (Moore County, Serra de Magé) have the requisite substantial positive Eu anomalies. Ancient igneous rocks from the Mesosiderite Parent Bodies have similar relationships, with one of the gabbro clasts containing the most extreme positive Eu anomaly known. Some very effective mechanism of volatile and sodium loss has to be found to arrive at these compositions from chondritic or carbonaceous chondritic starting materials.

Anomaly of the established lunar petrogenetic model. Yet the conventional interpretation of lunar petrogenesis requires that the Apollo 11, 12, 15 and 17 missions to the Moon each sampled, within a diminutive area, not one but many near-primary quenched liquid samples which had arrived unmodified through the volcanic plumbing system from depths of between 130 and 480 km, yielding a greater diversity of samples than would be expected from most comparably small sampling areas on the Earth. This is a proposition worthy of the most careful re-evaluation on the eve of planned sample returns from four localities on Mars in the next decade. Would the same conclusions about the Apollo and Luna samples have been reached if they had arrived in 1999, not 1969?

Predictions. The essence of a satisfactory hypothesis is that it should make predictions which can be tested and proved or disproved. The alternative interpretation of mare basalt petrogenesis suggested here predicts that the average compositions of the small lithic fragments in the Apollo 17 regolith will be found to be close to those of low-pressure plagioclase-saturated cotectic basalts, like those reported from the Apollo 11 and 12 sites (but perhaps not those in the Apollo 15 mare regoliths, where there may have been magma drainage when the last flow was partially crystallized, and least of all the winnowed regolith from the edge of Hadley Rille). It anticipates that diligent search among vitrophyric lithic fragments from the regolith at the Apollo 12, 15 and 17 mare sites will find evidence of small plagioclase phenocrysts in an appropriate composition groundmass, similar to those reported from the Apollo 11 site. It predicts that a few such plagioclase phenocrysts are present in small amounts in the vitrophyric groundmasses of some of the mafic hand specimens but accepts that they might be very difficult to find and even more difficult to prove conclusively absent. It predicts that a significant part of the gabbroic debris in highland breccias from Apollo 15, 16 and 17 can be linked to the evolution of earlier mare basalt magmas. It anticipates that remote sensing of the layers exposed in the walls of Hadley Rille will demonstrate compositions akin to the surface regoliths, not that of the mafic hand specimens, and that the same will be true of the average regoliths on the steep sides of the rille, which should be more mafic than the surface regoliths if the ‘conventional’ version were true. All the above can be tested with available samples or readily accessible measurements. Vents erupting the glass-bead deposits may also have carried up xenoliths of the underlying stratigraphy. High-resolution remote sensing of, or, better, a visit to the huge blocks on the floor of Hadley Rille and the materials exposed in the slumped walls of Copernicus, or a few 100 m drill cores almost anywhere in the maria, would settle the debate finally and unambiguously.

See also Notes added in proof, which follow Note 140 on p. 1627.

	NOTES

TOP ABSTRACT INTRODUCTION CONTENTS AND ROUTE MAP LUNAR PETROGENESIS 1969-1999 NOTES Notes added in proof APPENDICES REFERENCES

 
Notes are numbered as in the text.

• 0. In 1950 petrologists were less than 20 years from contemplating the derivation of tholeiitic and alkaline basalts by the total melting of conveniently provided glassy basalt layers of the appropriate compositions (e.g. Turner & Verhoogen, 1951, p. 162). They were simultaneously less than 20 years from having a new planet to explore, and closer to both events than we are today to the first Apollo landing, over 30 years ago. During the preparations for the lunar missions and the receipt of samples, the team at the Lunar Receiving Laboratory and the Principal Investigator group was dominated by geochemists. Field and conventional petrologists had little with which to work, in contrast to the wealth of superb quality data generated by the numerous geochemistry laboratories involved. Inevitably, the subsequent interpretation of the lunar samples was dominated by the views of geochemists, whose opinions, like those of most workers in the project, had been formulated mainly in the 15 years following the middle of the century. The Apollo lunar samples were delivered into a climate of petrogenetic opinion which accepted as commonplace the widespread eruption of little modified primary magmas, whose trace element geochemistry was believed to be dominated by the effects of small degrees of partial melting of appropriate source rocks. This was a scientific community which was temporarily disenchanted with the role of assimilation and contamination in igneous petrogenesis.
  Postulating basalt layers of convenient composition, waiting to be melted to yield observed basalts, is the ultimate in primary magma hypotheses. A minimalist step from this is to postulate conveniently situated mantle compositions, also waiting to be partially melted to yield observed basalts without further modification of their composition. Such a model has problems. There is a requirement to get the right mantle to the right place at the right time in a continuously and vigorously convecting body so as to match the observed correlation of magma type with plate tectonic environment. The view that basalts originated by the partial melting of peridotite (Bowen, 1928) was embodied in the standard petrological teaching text of the day (Turner & Verhoogen, 1951, 1960) and the fundamental role of partial melting of peridotite in producing basalt was accepted by all involved. Thereafter the community diverged.
  The premise that the primary magmas were erupted in abundance was also widely accepted and was particularly seductive to petrologists and geochemists because it immediately invested a sample with fundamental importance. As a primary magma, it is a stepping stone to mantle source mineralogy, pressure of formation and the mantle dynamics leading to its partial melting. Breath-taking vistas in planetology and cosmology unfold before the possessor of such a talisman. As a composition modified by low-pressure processes, that same sample has much to say about crust-forming processes of immediate interest to the petrologist, but little to say about the mantle until those near-surface processes have been unravelled and their effects stripped away. Entrants to the profession were instructed:
  ‘Probably the most satisfactory ... [criterion by which a primary magma may be recognized] ... is a pronounced tendency for the magma to appear repeatedly throughout geologic time, in great quantities and in extensive individual bodies (lava floods, batholiths, lopolithic sheets, etc.) over large sectors of the earth’s crust ... The ultimate criterion of a primary magma is its abundance in space and time.... The case for world-wide development of primary basaltic magmas is now satisfactorily established’ (Turner & Verhoogen, 1951, 1960, pp. 361–362).
  Continental flood basalt was high on the list of terrestrial rocks conforming to these expectations. Immediately before the Apollo missions, mid-ocean ridge basalts were being discovered and proclaimed as the primary magma (Engel et al., 1965), despite telling petrological evidence to the contrary (Muir & Tilley, 1964) and later identification of phase equilibria impediments (O’Hara, 1968a, 1968b; Stolper, 1980). Four quotations from Carmichael et al. (1974) convey the prevailing mood among the community during and after the Apollo program with respect to Hawaiian basalts, MORB, CFB and basalt petrogenesis generally:
  ‘At many large volcanic centres—Mauna Loa for example—a single type of lava has been generated a thousand times or more and erupted with only minor subsequent modification ... [6·0–10·0% MgO in their quoted tholeiitic analyses, still more variable in practice!] ... The distinctive individuality of such a pile ... can scarcely be due to repeated retracing of an identical course through the maze of an ingenious model.... While a degree of preeruptive fractionation can explain much of the major-element chemistry of a mafic series, it commonly fails to account for more striking characteristics of the dispersed-element pattern ... the chemistry of [a magma’s] incompatible dispersed elements ... will be determined purely by that of the source rocks and by the degree and regime of fusion’ (Carmichael et al., 1974, p. 649).
  ‘If, as seems likely from their great volume and regional chemical uniformity ... [5·4–10·2% MgO in quoted analyses!] ... over very large areas, these lavas [abyssal oceanic tholeiites] were generated as substantial melt fractions of source rocks’ (Carmichael et al., 1974, p. 651).
  ‘Continental tholeiitic flood basalts and related diabases, more than any other class of volcanic rocks, satisfy the two criteria ... for magmas generated directly by fusion—great volume and compositional homogeneity within each province. How else can one explain the uniform character of more than 100,000 km³ of Yakima basalts ... [3·8–4·4% MgO in their quoted analyses] ... erupted within 5 m.y. in the Columbia River province, northwestern United States? To derive this magma from picritic basalt of deep-seated origin by low pressure fractionation requires again and again that each successive draught of magma must rid itself cleanly ... of the same fraction of crystalline olivine along some identical course of ascent. This seems highly improbable’ (Carmichael et al., 1974, p. 654).
  ‘In general we hold the ... view that voluminously erupted mafic magmas of uniform composition ... appear at the surface with their initial chemical character—imparted by fusion of appropriate ultramafic source rocks—unimpaired or at least still recognisable’ (Carmichael et al., 1974, p. 649).
  At least four implicit assumptions underpinned the Carmichael et al. (1974) interpretation:
  
  • (i) That each erupted magma comes as a discrete batch from source to vent without intervention of staging chambers. This was an oversimplification. There was already abundant field evidence concerning the complexity of volcanic plumbing (e.g. Harker, 1908; Brown, 1956; Wager & Brown, 1968), which has been further supported by postulated sub-crustal magma chambers (Cox, 1980), by the discovery of such sub-crustal ultramafic complexes in the Ivrea zone (Quick et al., 1994), by studies of individual active volcanoes (Ryan et al., 1981) and by the presence of massive ultramafic cumulates within ocean-island volcanoes (Watts et al., 1985; Carress et al., 1995).
    
  • (ii) That the only physical processes of magma modification which needed to be entertained were PFC (perfect fractional crystallization), EPM (equilibrium partial melting, also known as batch melting), PFM (perfect fractional melting) and APFM (accumulated perfect fractional melting). With the discrete batch constraint removed, several plausible partial crystallization processes have been shown to produce effects on both incompatible and compatible trace elements which would make their residual liquids almost indistinguishable from small mass fraction equilibrium partial melting products (see 15, 16, 20, 21, 23, 24, 26, 29, 30 below). Many geochemical interpretations of the period, conforming to the model conveyed by Carmichael et al. (1974), called for specially fertile source regions from which the inferred small mass fraction partial melts could be derived (see 22, 23 below).
    
  • (iii) That continental flood basalts were indeed uniform in composition. The uniformity of composition of flood basalts was progressively undermined (Cox et al., 1965, 1967; Cox, 1971, 1972), and the accounts by Carmichael et al. (1974) were themselves sufficient to raise doubt.
    
  • (iv) That partial crystallization in magma chambers cannot impart uniformity of composition. Cox & Jamieson (1974) pointed out that arrival of compositions fractionating olivine at, and their diversion by, the plagioclase-saturated low-pressure cotectics provided one of the mechanisms required to impart a degree of composition uniformity. Other factors, connected with the buffering behaviour of periodically recharged magma chambers (26, 27), are comparably important.
    
  
  None of these four assumptions was justified. Nevertheless, Wilkinson (1982, 1991) has mounted a spirited defence of the Carmichael et al. (1974) position.
  At the time of the Apollo missions a major role for partial crystallization was regarded by some as essential to explain the major element compositions of basalts, coming to a head in studies by Cox et al. (1965, 1967), O’Hara (1965, 1968a, 1968b) and Cox & Hornung (1966). Partial crystallization of abundant basalt magmas was regarded as irrelevant by other petrologists (Green & Ringwood, 1967), and as unnecessary or inadmissible by trace element geochemists, coming to a head in the work of Gast (1968). All but one of these protagonists were appointed as Principal Investigators in the Apollo Lunar Science Program in 1967. Major battlelines were drawn up and positions dug between selection of the Apollo investigators and the arrival of the first samples. With most of the protagonists now involved on the same project, what more natural than that the Moon should have become a field of debate. Personalities and human nature then played a part [see Mitroff (1974) for an enlightening view of the environment and times].
  
• 1. The extent of the appeal of contamination, assimilation and hybridization in the petrogenesis of plutonic igneous sequences, predominantly those with significant acid and intermediate members, is apparent in the accounts in Turner & Verhoogen (1951) and earlier texts referenced therein. The field relations and petrography of these plutonic rocks are important parts of the evidence that many magmas undergo profound modification of their compositions within the crust. The compositions of the actual liquids involved in these processes, however, were sometimes speculative.
  
• 2. The loss of appeal of contamination, assimilation and hybridization in igneous petrogenesis generally is equally apparent from the reduced discussion of this issue in the study by Carmichael et al. (1974). This changing attitude reflected the shift of emphasis in petrological and geochemical studies away from plutonic rocks towards xenolith-poor lavas, where there was less debate about what were the liquid compositions, and towards oceanic basalts, where contamination by continental crust and sediment was clearly minimal. Justification for this substantial change of attitude could be found in the belief that erupted basalts were very uniform (0) coupled with an implicit assumption that the plumbing systems were simple, with magma batches proceeding as discrete entities from source to point of eruption. The extent of rehabilitation of processes of contamination, assimilation and hybridization in basalt petrogenesis is reviewed below (25, 27, 28).
  
• 3. A detailed petrogenetic model for the Moon was created in 1969–1974 but essentially by mid-1970. This interpretation has changed little in the following three decades (Taylor, 1975; Vaniman et al., 1991; Shearer et al., 1998; Shearer & Papike, 1999), whereas that of terrestrial basalt petrogenesis has changed dramatically—one might think we were dealing with two different planets!
  The origin of such a large yet relatively low-density, metallic iron depleted satellite as the Moon presents special problems, with current opinion favouring an oblique impact between a chemically differentiated proto-Earth and a chemically differentiated proto-planet (Lissauer, 1997). A large mass of iron-depleted mantle material is ejected as vapour, melt and fragments into Earth orbit, where some accretes to form the present Moon and the rest is either recaptured by the Earth or lost altogether from the Earth–Moon system. Ida et al. (1997) suggested that a Moon formed in this way would accrete in less than a year. Direct ejection of a disc massive enough to form the Moon may require a differentiated impactor up to twice the mass of Mars but there are residual problems concerning the angular momentum of the resulting Earth–Moon system. Such a hypothesis provides an opportunity for substantial cooling of the condensing disc materials before reaccretion. Accretion of the Moon would then be from particles and bodies whose relative speeds at impact would have been much lower than those of the bulk of the material accreted to form the two proto-planets, with adverse implications for the postulated formation of a magma ocean. Substantial further accretion of cometary and asteroidal materials to the Earth and Moon may have followed this major impact event, restoring to the upper mantle the budget of siderophile and volatile elements which might well have been severely depleted in the earlier core-separation and impact events, respectively.
  The Moon has a small iron or iron–sulphide core and is the smallest body in the Solar System known to have an ancient, heavily cratered crust enriched in plagioclase beyond the level achievable by dry basalts in low-pressure cotectic equilibria. Some type of comprehensive igneous processing of the lunar interior is demanded. This anorthositic crust composition is unrepresented among the meteorites except in some 12 fragments which are geochemically linked to the Moon itself. Relatively minor basaltic igneous activity yielding smooth plains units began early (4·2 Ga or earlier), continued vigorously for at least a billion years and less vigorously for perhaps as long again. It did not proceed to the generation of large central volcanic constructs, and there are no surface features which suggest an onset of large-scale resurfacing as a result of convection of the mantle. The Martian (55), and perhaps the Mercurian, lithosphere (56) was strong enough to support large volcano shield-building, and sufficient total volumes of magma were erupted on the Moon. The lunar lithosphere would have supported similar constructs but the rate and style of mare basalt eruption was for some reason not suitable.
  The ‘conventional’ model (e.g. Taylor, 1975, pp. 318–324, and fig. 7.2 in particular) holds that: (i) the Moon was volatile and siderophile depleted relative to the Earth and chondrite meteorites from its creation, (ii) an 500–1000 km deep magma ocean formed in the outer parts of the Moon during accretion, (iii) bottom-up fractional crystallization of this magma ocean occurred with development of an increasing negative Eu anomaly in the REE patterns of the residual liquids and of all ferromagnesian minerals later precipitated from them once plagioclase had started to crystallize; (iv) plagioclase floating to form a 60 km thick cumulate crust where it mingled with a Mg- and Cr-rich chilled surface of the original ocean and (v) mafic minerals settling to form layered magnesian to more ferriferous and titaniferous ultramafic cumulates, to which marked negative europium anomalies had been imparted, (vi) with the last residual liquids (KREEP basalts) becoming enriched in heat-producing elements and crystallizing deep in the crust or erupting to the surface, after which (vii) either reheating by this concentration of heat-producing elements (but see 101), or positive and negative plume formation as a result of the inherent gravitational instability of the upward decrease of mg-number in the ultramafic cumulates, led to the renewed partial melting of these rocks, yielding picritic and ultramafic liquids at 130–480 km depth, which (viii) erupted into the maria as thick basalt sequences, analogous to terrestrial flood basalts, of unmodified primary magmas, (ix) whose compositions, with only minor near-surface fractionation, are represented by those of the returned hand specimens, and (x) with negligible modification of all erupted mare and highland magmas by selective volatilization during eruption. The principal later modification recognizes a role for serial magnesian anorthositic norite plutonism in lunar highland petrogenesis.
  All elements of this internally consistent edifice are structural. Break one and the whole trembles, break two and the structure is unsound. This paper summarizes the case for doubting each and every element in that structure. An analogous model, which would permit most terrestrial basalt magma compositions to be interpreted as primary magmas, might be devised, were it not for the need to get the right cumulate to the right place at the right time in a convecting mantle.
  
• 4. When the current view of lunar petrogenesis was formulated, we knew nothing of the surface features and volcanic activity of Io, Mercury, Mars and Venus, and very little about the volcanic activity on the Achondrite Parent Planet. Io is very similar in size, density and presumably major element composition to the Moon. Whatever hypotheses are applicable to the Moon may be required to apply at Io, and at Europa, whose rocky portion is of slightly smaller size, with repercussions for the giant impactor hypothesis. There is very little mention of Io in publications about the Moon and still less of the Moon in publications about Io [e.g. internally within the Basaltic Volcanism Study Project (hereafter BVSP, 1981), in the Lunar Sourcebook (Heiken et al., 1991), in Planetary Materials (Papike et al., 1998) or in the latest edition of New Solar System (Beatty et al., 1999)]. Few papers about igneous processes on Io appear to have been contributed to the annual Lunar and Planetary Science conferences.
  The vast literature interpreting results and exploring speculations in the light of the ‘conventional’ model (1) should be regarded as propaganda rather than proof. Equivalent effort has not been devoted to evaluating radically different alternatives. The lunar mare basalts were widely interpreted as flood basalts (e.g. Taylor, 1975) and primary magmas from the lunar interior, an identification made by workers few of whom at the time had much direct experience of terrestrial flood basalts. Basaltic activity on the Earth, Venus, Mars, Vesta and probably Io is marked by caldera formation or other evidence of high-level magma chambers, and is probably accompanied by near-surface partial crystallization and modification of the compositions of basic magmas on their way to the surface. Eruptions of undifferentiated primary magmas are infrequent in time and small in volume on the Earth.
  Even the lunar highlands are now believed by some to be products of serial feldspathic volcanism with total fractionation at low pressure—no igneous rock with the average magma composition is identified. It is implied that the lunar highland average is made up entirely by impact mingling of plagioclase cumulates, late residual liquids of the magma ocean and early mare basalts (which are feldspathic). Only the later mare basalts of the Moon are currently held to be radically different—thick piles of predominantly little altered primary magmas from great depths.
  
• 5. During the post-Apollo period the number of different terrestrial magma types and associations which has been recognized has increased with the discovery of boninites (Crawford et al., 1981), their possible elevation to a major magma type in the genesis of some of the great layered intrusive complexes (Vogel et al., 1999), the recognition of the importance and style of ash-flow magmatism (Smith, 1979) and the exploring of the great sub-crustal peridotite–gabbro complexes (Quick et al., 1994). The correlation of terrestrial magma type with particular plate-tectonic environments has been emphasized (Pearce & Cann, 1973; Pearce, 1987; Wilson, 1989). Such correlations either demand exceptionally prescient mantle compilation four billion years ago or indicate that process, rather than the fine details of the source composition, is the controlling factor in erupted magma composition. The choice of a balance between partial melting process, partial crystallization process and assimilation effects remains but it is prudent to envisage a major role for all three.
  Terrestrial flood basalt provinces, not least through the studies of Cox and co-workers (BVSP, 1981, sections 1.2.2 and 1.2.3), are no longer regarded as a source of abundant unaltered primary magmas. In the waning phases of their activity, when partial crystallization in crustal magma chambers and probable crustal melting and contamination become most important, their eruptive products are in general far removed in chemistry from parental magmas derived from the mantle.
  BVSP (1981, section 1.2.5) provides a comprehensive review of MORB petrogenesis, concluding that low-pressure fractionation is a major factor but noting that it is often not possible to relate the trace element chemistry of one sample to another from the same site by simple crystal–liquid processes. Although this has been taken to indicate separate evolution of magma batches in small isolated magma chambers it is also a predictable consequence of the operation of some of the more sophisticated crystal–liquid separation processes (26–30).
  The parental magmas of Hawaiian tholeiites should be picritic in character if produced by partial melting of mantle peridotite at the depths commonly supposed (O’Hara, 1965, 1968a). The proposition was slow to be accepted (Carmichael et al., 1974, pp. 418, 649), not least because of the lack of picritic lavas unaffected by phenocryst accumulation. This view denied the derivative nature of Hawaiian tholeiites and relegated to a minor role the low-pressure fractionation abundantly evidenced by pervasive disrupted dunite and gabbro xenoliths (Jackson, 1968), and the petrographic and geochemical evidence of magma mixing and differentiation in Kilauean lavas (Wright, 1971, 1973; Wright & Fiske, 1971; Wright et al., 1975; Wright & Tilling, 1980). Today Kilauean tholeiites are interpreted as low-pressure differentiates of picritic parental magmas with 16% MgO or more (e.g. Clague et al., 1995), and Mauna Loa and Mauna Kea lavas from the Hawaii Scientific Drilling Project are interpreted as derived from parental liquids containing at least 15–17% MgO (Baker et al., 1996; Garcia, 1996; Rhodes, 1996; Yang et al., 1996). Large quantities of the requisite mafic cumulates have been discovered (Watts et al., 1985; Carress et al., 1993). However, current models concentrate on simple olivine or olivine plus clinopyroxene control and do not entertain more sophisticated magma chamber models (18, 19, 26–30) or possible refluxing of the erupted sequence (O’Hara, 1998). Magma chamber processes buffer the erupted, modified basalt compositions (1, 26–28).
  
• 6. The less one thinks about the definition of a primary magma, the easier it is to be sure one knows what one means! Achieving a suitable and robust definition has troubled many workers (e.g. Carmichael et al., 1974). Taking a reasonable view of modern ideas about magma formation, segregation and movement, one might expect melt to form, it is hoped in local (centimetre-scale) equilibrium with partially residual source rock, throughout a melting volume in which the temperature and melt fraction developed varies from negligible up to some maximum, and through which the depth of burial, and consequently the pressure, varies significantly. Melt extraction is likely to be continuous and the melting process one of (imperfect) fractional melting with melt segregation and mixing in some other location. The aggregated fractional partial melts are not in chemical equilibrium with their unmelted residues, even in the absence of integration of melts from different parts of a complex melting regime, which further compounds the problems. Neither the methods of conventional experimental petrology nor those of thermodynamic calculation derived from the experimental data can be used directly. Assumption of equilibration of the major element composition with a unique residue at a single mass fraction of partial melting at a unique pressure is unjustified. The abiding hope, to which most researchers clearly subscribe, is that the errors introduced by this assumption are small relative to the overall changes of melt composition as a function of pressure. This hope may be justified for elements whose bulk distribution coefficients are relatively close to unity (e.g. 0·4–2·5 or most of the major and minor components) in situations where the pressure range across which melt compositions are integrated is small relative to the rate of change of the equilibria with pressure (e.g. 0·2 GPa below 1 GPa, 0·4 GPa between 1·0 and 2·5 GPa, and 0·7 GPa at higher pressures). It is, however, readily demonstrated that this hope is not fulfilled for highly incompatible elements during melting processes. Primary magma cannot be defined as that liquid which is in equilibrium with the source rock or its residue.
  The most magnesian MORB glasses with mg-number 72 are clearly relatively primitive liquids and would be in equilibrium with olivines of mg-number 88·9 ± 1·0. This appears very similar to the values of mg-number 91 ± 1 prevalent in the residual harzburgites of ophiolites and might encourage adoption of these glasses as primary magmas. Viewed from the opposite end of the telescope, however, the fe-number of the olivines displays a mismatch of 20%, which leaves room for substantial amounts of differentiation in some models (30).
  In general, migrating melts cannot even be in local equilibrium with the residues or source mantle through which they migrate (31), hence chemical interactions between magma and mantle in response to changes in pressure and temperature must have commenced even before the magma has segregated into a recognizable body. Primary magma cannot, therefore, be defined simply as the aggregated liquid from all those regions which are undergoing partial melting. During subsequent ascent of the magma body further interactions with surrounding mantle are probably inevitable, a process in which the total mass of the liquid may initially increase when the country rocks are already hot and ‘fertile’ but will eventually decrease when the country rocks are sufficiently cool. Primary magma cannot, therefore, be defined as the composition of the liquid when its mass is at a maximum. A vague but useful concept on which all might agree is that a primary magma composition is one which exists as a liquid close to the source region of the igneous activity and is the product of the partial melting and segregation events.
  
• 7. Common terrestrial basalts display relatively small variations in mg-number, moderate variations in concentrations of highly compatible trace elements such as Ni and Cr, and relatively large variations in the concentrations and ratios of highly incompatible elements (e.g. Jamieson & Clarke, 1970; Hart & Allègre, 1980).
  
• 8. Perfect fractional crystallization cannot explain the good discrimination (variation of the ratios of almost equally incompatible elements) observed between highly incompatible elements in flows from the same province. PFC predicts large variations in highly incompatible element concentrations only as the outcome of very large reductions in mass fraction of liquid, which in turn demand large variations in the concentration of highly compatible elements (Gast, 1968). These effects are not in general observed in basalt sequences. Moreover, Ringwood (1975) evaluated the effects on Ni and Cr depletion during fractionation, arguing that they should be decoupled because they are compatible in contrasted minerals, but did not appear to be so:
  ‘Turekian ... demonstrated the strong covariance of Cr and Ni in a suite of over 100 basalts [mostly tholeiites] from all over the world (Fig 4.11) in which absolute Cr and Ni abundances varied by a factor of 50’ (Ringwood, 1975, p. 169).
  Figure 4.11 of Ringwood (1975) is a log–log plot of Ni vs Cr concentrations, of a type beloved of geochemists for sound representational reasons, but possessed of properties which require careful handling by the unwary. The visually apparent strong covariance in the plot conceals a variation in Ni/Cr ratio from 0·2 to 3·0, a variation by a factor of 10–15 at any chosen value of Ni or Cr concentration, and elbow-room for a substantial amount of decoupling if it is required! Mineral phases tend to co-precipitate in ratios determined by the phase equilibria during fractionation, thus naturally damping or eliminating the anticipated decoupling; covariation would not be a telling point in these circumstances—even if it did exist. Some factor has to account for the 50-fold variation in absolute abundances of Ni and Cr in rocks identified as basalts. It cannot be a product of the partial melting process at small to moderate mass fractions of partial melting, because the Ni and Cr in the liquid would be buffered by the compositions of the large mass of residual peridotite minerals rich in these elements during any such melting process. Partial melting cannot explain this without simultaneously postulating a matching extent of variability in the source peridotites. Partial crystallization processes must be welcomed because they offer a viable means of producing the observed spread of values from 1000 to 20 ppm Ni.
  
• 9. Equilibrium partial melting (Gast, 1968) predicts large variations in both concentration and ratio of highly incompatible elements (such as the REE) at small but variable (0·001–0·05) mass fractions of liquid in the system, which would be coupled with relatively high and stable concentrations of highly compatible elements (such as Ni or Cr). Equilibrium partial melting leading to the production of partial melts within the mantle was a readily acceptable concept. The whole trace element argument was encapsulated in a review celebrating the fiftieth anniversary of the publication of Bowen’s influential text:
  ‘Partial melting is now thought to have a greater role in producing the observed chemistry of erupted rocks, relative to fractional crystallization ... The Rayleigh equation is commonly used to model fractional crystallization in a closed system (PFC), such as a magma chamber; the Berthelot–Nernst equation is used to model partial melting processes (EPM) ... Crystallization of various basaltic minerals is not effective in fractionating the incompatible elements from each other; ... In contrast to this, the melt concentrations of compatible elements may be drastically changed during crystal fractionation, so that melt concentrations of elements such as Ni, Cr, Sr (Eu) and Sc (Yb) serve as excellent indicators of the extent of crystallization of the minerals olivine, clinopyroxene, plagioclase and garnet respectively ... Such a process [of eclogite and olivine fractionation] would be expected to severely deplete the residual liquids in the heavy-REE (due to garnet crystallization) and in Ni (due to olivine crystallization)’ (Hart & Allègre, 1980, pp. 121–122, 125, 131, 133).
  
• 10. Implicit in the argument (8, 9) were the assumptions that processes only operated in chemical systems closed to contamination and hybridization, and that an approach to equilibrium partial crystallization (which explains the geochemical data equally well) in magma chambers was impossible because of the difficulty of extracting the small mass fractions of melt implied and of maintaining equilibrium between residual liquid and crystals. Equilibrium between liquid and crystals was denied by abundant petrological evidence of zoning, fractionation and failure to maintain equilibrium in plutonic rocks. Partial crystallization models more sophisticated than simple closed system perfect fractional crystallization, however, can yield solid products which show zoning and cryptic variation and liquid products with properties similar to those of the EPC process but with the residual liquid now readily available for tapping as lava flows (16, 20, 26–30).
  
• 11. See Maaløe (1982) and McKenzie (1984, 1985a, 1985b).
  
• 12. The liquid end product of perfect fractional partial melting, provided that the melt fractions are subsequently aggregated, would be very nearly the same as in the equilibrium case (Gast, 1968; Shaw, 1970; O’Hara, 1993). The final residues are, however, very different.
  
• 13. Perfect fractional melting extracts most of the incompatible trace elements in the first small mass fractions of melting, imparting spectacular discrimination (7) among the very low residual concentrations in the residual peridotites. Most peridotites exhibit higher concentrations and less discrimination than would be expected after perfect fractional extraction of even a few percent of partial melt.
  
• 14. The extremely influential conclusions of Carmichael et al. (1974) gave most weight to the trace element geochemical approach to basalt genesis. Wyllie (1979) later provided an overview of the terrestrial petrogenetic debate which gave more weight to the petrological–major element–experimental approach. Carmichael et al. (1974, p. 28) quoted Karl Popper (1972) as the greatest living exponent of the logic of scientific discovery: ‘Every good scientific theory is a prohibition: it forbids certain things to happen. The more a theory forbids, the better it is’, and, in criticism of the more sophisticated view of terrestrial basalt petrogenesis, ‘Its universal applicability is an obstacle to rigorous testing by attempted disproof and so is a weakness’. What would Popper have thought of hypotheses which postulate the existence of convenient source compositions in an inaccessible, untestable location, to explain each basalt type which is sampled?
  The more sophisticated interpretation of basalt petrogenesis is, moreover, founded on two sweeping and comprehensive prohibitions. The vast majority of erupted magmas have compositions which should not be, and are not, in equilibrium with upper-mantle mineral assemblages and compositions at the appropriate pressures (O’Hara, 1968a, 1968b; Stolper, 1980; Herzberg & O’Hara, 1998). Sample suites from upper-mantle rocks which have undergone variable degrees of partial melting should not show the extraction of commonly erupted basalt compositions, and do not (O’Hara et al., 1975b). Instead, they show the major element effects of the extraction of olivine-rich magmas.
  
• 15. The liquid composition at a true eutectic solidifies on cooling without change of composition, precipitating a fixed mineral assemblage in fixed mineral proportions which is obviously identical in composition to the bulk liquid. Add a trace amount of some incompatible element to that liquid, too small to influence the phase equilibria (at least until very substantial crystallization has occurred) and the differentiation of the residual liquid will then be marked by negligible change in major element composition accompanied by increases in the highly incompatible element concentration which are inversely proportional to the mass fraction of liquid remaining. True eutectic crystallization is most unlikely to be encountered in any natural basaltic composition but there are three situations of eutectic-like crystallization which may arise. These could lead to effects among the incompatible trace elements which might otherwise seem to demand an origin by very small and variable mass fractions of partial melting.
  Liquid compositions which are in cotectic equilibrium with three or more crystal species and are also close in composition to the liquid at a thermal maximum within that equilibrium (i.e. close to a thermal divide) may exhibit large amounts of crystallization with only minor changes in the major element composition (e.g. increases in Fe/Mg and Na/Ca) during both equilibrium and fractional crystallization. Trace elements highly compatible in one of the crystallizing phases may not have a particularly high distribution coefficient in favour of the bulk crystal assemblage if the percentage of the relevant phase is small in the eutectic-like precipitate. The eclogite thermal divide (O’Hara & Yoder, 1963, 1967) at pressures above 3·0 GPa (Milholland & Presnall, 1998) and the olivine–gabbro thermal divide at pressures below 0·8 GPa (O’Hara, 1969b) are such equilibria potentially relevant to terrestrial basalt petrogenesis. Because the locus of compositions of liquids formed by initial partial melting of peridotites passes through these two divides at 2·5 GPa and 1·0 GPa respectively (O’Hara, 1968a; Herzberg & O’Hara, 1998), it is probable that such liquid compositions do exist as parental liquids. It is not yet possible to exclude a role for the olivine–gabbro divide in lunar petrogenesis. A thermal divide also exists on the olivine + calcium-poor pyroxene + plagioclase equilibrium and might seem more relevant to lunar petrogenesis, but this involves reaction between olivine and liquid—liquids will evolve in a eutectic-like manner only if the partial crystallization process is one involving magma recharge and escape, which can lead to results not easily distinguished from simple equilibrium partial crystallization. Aspects of olivine gabbro fractionation from liquids close to the thermal divide have been little explored.
  Liquid compositions which are in cotectic equilibrium with three or more crystal species and are also close in composition to the liquid at a thermal minimum within that equilibrium may behave similarly, and without loss of the eutectic-like behaviour as differentiation proceeds. However, no case of this which is likely to be relevant to basalt origins is yet known.
  Liquid compositions which are in cotectic equilibrium with a large number of crystal species may also exhibit a close approximation to eutectoid crystallization provided the minerals include repositories for all or most of the major and minor elements in the liquid. The average (not hand-specimen) high-titanium basalt compositions from the Apollo 11 and 17 sites are close to cotectic equilibrium with olivine, calcium-poor and calcium-rich pyroxene, plagioclase, spinel, ilmenite and armalcolite. Crystallization of such an assemblage would lead to eutectic-like behaviour and is one possible factor in the trace element contrasts between the low- and high-potassium basalt groups at the Apollo 11 site (O’Hara et al., 1974).
  Nodule suites in alkali basalts (O’Hara, 1965, 1969a), kimberlites (O’Hara et al., 1975b; Cox et al., 1987) and peridotite massifs (Obata, 1980) demonstrate that eclogites and garnet pyroxenites have precipitated in proportions which would be significant in relation to the amount of residual peridotite and probable mass of partial melt which have developed. Eclogite separation is clearly implicated in alkali basalt and kimberlite genesis and a role for eclogite–liquid partitioning or garnet pyroxenite melting in the periphery of melting regimes has been argued on trace element and isotopic grounds (Johnson et al., 1990; Eggins, 1992; Hirschman & Stolper, 1995). Eclogite fractionation could not be a factor in lunar igneous petrogenesis within the outer 500 km, the depth of the postulated lunar magma ocean, because pressures are too low—but it could in principle be a factor for magmas generated at greater depths within the cooling asthenosphere, which is now encountered only at depths greater than 1000 km (Taylor, 1975, p. 291). Eclogite fractionation probably plays little part in evolution of terrestrial tholeiitic basalt eruptives at mid-ocean ridges, in voluminous continental flood basalt provinces and in the most active phase of ocean-island centres. Continental flood basalt provinces such as the Paraná, suggested to have formed by thermal conduction from a plume head into the lithospheric upper mantle (Turner et al., 1996) may, however, provide opportunity for eclogite fractionation in tholeiite petrogenesis.
  Eclogite fractionation was rejected (Gast, 1968; Ringwood, 1975) on the grounds of its predicted effects in a closed system perfect fractional crystallization process on Yb concentrations and its probable effects on Cr concentrations. This rejection, however, overlooked the low mass fraction of garnet in the precipitating solidus assemblage (O’Hara & Yoder, 1963, 1967). It also overlooked the unknown distribution coefficients of Cr between liquid, garnet and the highly aluminous sub-calcic pyroxene actually present at the solidus, and it did not of course consider more sophisticated crystallization models not then available (8).
  
• 16. Imperfect fractional processes of crystal–liquid separation, in which extraction takes place by small but finite, rather than infinitesimal, increments, have effects on highly incompatible elements in fractional melting and on highly compatible elements in fractional crystallization which rapidly approach those of the equilibrium process as increment size diverges even slightly from the infinitesimal assumption embodied in Rayleigh fractionation (O’Hara, 1993).
  
• 17. Experiment required that common basalts could not be primary magmas and petrological observations indicated that large mass fractions of partial melting were involved. Petrology and field relations indicated that basalt magmas were commonly fractionated in the crust and when erupted generally bore the hallmark of low-pressure modification by substantial mass fractions of partial crystallization (essentially the major element–sophisticated crustal process story). Trace element behaviour indicated that, of the models considered, equilibrium partial melting at very small mass fractions of melting best explained the observed geochemistry of basalts. Other aspects of basalt chemistry seemed to exclude any significant role for closed system perfect fractional crystallization (essentially the trace element–source process dominated story). The onus was on all parties to demonstrate that (a) there were no processes other than the two extreme closed system effects considered which were capable of producing the geochemical effects—processes which might be more compatible with the partial crystallization requirements, and (b) to test that there were no other feasible interpretations of the experimental petrology results which might allow low-pressure cotectic character to be a coincidence or of lesser significance. In this the geochemists may have felt under less pressure, as witnessed by the confidence expressed by Hart & Allègre (9) and in the following:
  ‘A number of more complicated models have been proposed ... but have not yet been extensively used. In many cases, our knowledge of partition coefficients is too limited to justify the use of more complex models’ (Hart & Allègre, 1980, p. 125).
  That knowledge of distribution coefficients was not too limited to prevent acceptance of sweeping conclusions which did not agree with an abundance of field, experimental and major element data. The volume of publications in the quarter-century following 1968, each reporting that some facet of the geochemistry of some suite of basalts required their origin by small mass fractions of partial melting of somewhat different mantle, is history.
  In modelling crystal–liquid separation events we are concerned with the apparent bulk distribution coefficients; i.e. with the ratio of the concentration of an element in the average extract produced divided by its concentration in the average liquid produced at the same time. Extract is specified rather than solid to cater for escaped as well as trapped liquid and the possible loss of immiscible liquids and vapours. Process can greatly influence the apparent bulk distribution coefficients of trace elements. Fractional partial melting processes (Gast, 1968) rapidly relocate highly incompatible elements into the average integrated liquid produced and deplete the residue. This causes the apparent bulk distribution coefficient (concentration in solid/concentration in liquid) to alter away from unity relative to the simple crystal–liquid equilibrium value, i.e. the distribution coefficient decreases for incompatible elements and the effect is more marked as the element becomes more incompatible. Fractional crystallization processes (Gast, 1968) have a similar effect, but acting this time on the highly compatible elements where the apparent bulk distribution coefficient increases. Processes of melt integration during partial melting (O’Hara, 1985, 1995a; Eggins, 1992) contribute disproportionately high concentrations of incompatible elements from the little melted periphery to the average residual solid whereas the large melt contributions from the centre of the regime dilute the high concentration in the average liquid. The final result is that the apparent bulk distribution coefficient for incompatible elements during integrated partial melting is changed towards the value of unity, but there is little effect on the highly compatible elements. A similar effect is observed during integrated partial crystallization (O’Hara & Fry, 1996a) but in this case it is the highly compatible elements which are most affected. The contribution of liquids which have been little fractionated and so retain their highly compatible elements dominates the budget of the integrated liquid. The contribution of solids which have been much fractionated and have their budget of the highly compatible element diluted by further crystallization dominates the budget of the average solid, with the result that the apparent bulk distribution coefficient is altered towards the value of unity, i.e. decreased. Small packet crystallization (Langmuir, 1989; O’Hara & Fry, 1996b) alters all apparent bulk distribution coefficients towards the value of unity and towards maximum or minimum values which are controlled by the parameters of the process.
  Apparent bulk distribution coefficients for trace elements can be highly process dependent (Fig. 1). The conclusion of Hart & Allègre above may be extended to state that even when the equilibrium crystal–liquid distribution coefficient for an element has been determined with high precision, there will be uncertainty, perhaps of orders of magnitude, in the appropriate bulk distribution coefficients to employ when modelling possible melting and crystallization processes. Several of the complications lead to the apparent bulk distribution coefficients being much closer to unity than the equilibrium distribution coefficient. Many discoveries and insights into igneous petrogenesis have come in the years following the Apollo program and have introduced great uncertainty into any attempt to model the geochemistry of planetary basalts in the absence of a good knowledge of their field relations. This task cannot even be undertaken reliably and unambiguously for more than a handful of terrestrial examples where field control greatly exceeds that likely to be available for any other planet for decades to come.
  
• 18. At high temperatures and pressures the pyroxenes exhibit substantial ranges of crystalline solutions towards each other and towards the alumina-bearing components which might otherwise appear as plagioclase, spinel or garnet. The mineral proportions in a slowly cooled source rock or cumulate assemblage may be a poor indication of the crystalline phases which were present and able to partition trace elements between solid and liquid at the temperature and pressure of formation. Early examples of this were the extensive solubility of potential garnet into the orthopyroxene structure at high pressure (Boyd & England, 1964) and into the clinopyroxene structure in natural and synthetic eclogites (O’Hara & Yoder, 1963, 1967). These changes in modal mineral assemblage can have dramatic effects on the calculation of bulk distribution coefficients (O’Hara & Mathews, 1981), quite apart from the effect of the changing composition of the pyroxene phase on the pyroxene–liquid partition coefficients.
  Zone refining effects during magma ascent were proposed by Harris (1957), underpinned the wall-rock reaction process (Green & Ringwood, 1967), were reviewed by Cox et al. (1979), and have found an application in understanding the concentration of the platinum-group elements into immiscible sulphide liquids from the Noril’sk flood basalts (Naldrett et al., 1996). Zone refining effects during the passage of mafic magma through the mantle would lead directly to trace element characteristics similar to those observed in basalts. The interactions between percolating magma and upper mantle may also exhibit the effects associated with chromatographic separation columns (Navon & Stolper, 1987; Asimow & Stolper, 1999), introducing geochemical complexities which it might be very difficult to invert from a knowledge of erupted liquid composition only, and which add to the complications of defining primary magmas already noted above. Maaløe (1999), however, argued that steady-state eruptions as at Kilauea volcano cannot be powered directly by a percolative plume source because overpressures generated would be too high. Episodic upward movement as disperse multiple mini-intrusions is preferred and obviously gives opportunity for partial crystallization and assimilation but lessens the scope for zone refining and chromatographic effects.
  Partial crystallization of a depressurizing magma flowing in a conduit or ascending in a convective cell, or of a pressurizing magma in a descending convective cell, can lead to some surprising effects both in the suspended crystal load and in any static cumulates formed on the side wall (Jamieson, 1970a; O’Hara et al., 1975b). (The implicit assumption is that the situation is non-adiabatic and that heat is lost from a depressurizing dry magma to the conduit walls.) Related effects must also be a factor in porous flow through peridotite whose importance will increase as the depth of melt formation increases. A liquid which is in cotectic equilibrium with two or more crystal species at the outset will, because of the changing primary liquidus phase volumes as a function of pressure, have to crystallize those minerals in proportions which may be very different from those appropriate under isobaric conditions in order to stay in the cotectic condition as the pressure changes. Let us consider a liquid in equilibrium with three crystal species at the outset. The ratio of the minerals which precipitate in the magma or on the side wall is in fact dependent on the ‘velocity’ (strictly, on the rate of change of enthalpy with pressure). Too fast a decompression and the adiabatic gradient will carry the liquid above its liquidus and nothing will precipitate; slower and only one phase will precipitate—the liquid composition will evolve but will not remain on the cotectic; slower still and a second phase may join the first, but in a proportion which will increase towards an isobaric two-crystal phase cotectic ratio as the ‘velocity’ decreases—the liquid composition will evolve still closer to the three-crystal phase cotectic and may even attain it; slower still and the liquid remains on the three-crystal phase cotectic as pressure changes but the ratio of the crystal species which must precipitate in order to stay on that cotectic will be biased away from the isobaric ratio towards excess of the first and second phases to an extent which is still ‘velocity’ controlled. Factors of this nature underlie part of the discussion of polybaric melting and crystallization by Niu (1997, 1999) and Walter (1999). Contemplation of what happens in a magma flow undergoing an increase of pressure is left as an exercise for the reader.
  
• 19. During mixing of two contrasted magma compositions, each saturated with crystal species, the resulting liquid will in general have a composition within the primary liquidus phase volume of at most one of the mineral species saturated in either of the end-member liquids, and may lie within the primary phase volume of a phase not even present in either of the end-member liquids. This can result in the forced precipitation of an exotic phase which would not have been prominent or even present in the closed system crystallization of either end-member component. This effect also has relevance to hybridization of melts separated by thermal divides, and to magma contamination (Chinner & Schairer, 1962; Gribble & O’Hara, 1967; O’Hara, 1969b, 1980; McBirney, 1979). The copious precipitation of near-monomineralic layers of chromitite in layered intrusions has been ascribed to such effects in the Bushveld, Great Dyke, Muskox and Rhum intrusions (e.g. Irvine, 1977). Precipitation of an immiscible sulphide liquid has similarly been attributed to magma mixing in such bodies (Naldrett, 1989). Such effects can dramatically alter the apparent bulk distribution coefficient for an element in the total process. Also to be taken into account are possible reactions between the top of the cumulus pile and a new magma input. These cannot in general be in equilibrium with each other.
  Partial crystallization of magma in small packets can contribute geochemical signals, which imply the separation of ‘occult’ phases, to bulk residual liquids whose major element compositions are controlled almost exclusively by the early separating phases (29). The resultant apparent bulk distribution coefficients cannot be predicted from the simple crystal–liquid values or knowledge of the parent magma composition. Oversimplified models of the relationship between the bulk crystal extract and the residual liquid developed during the crystallization of large magma bodies have been replaced by recognition that considerable diagenesis of the growing cumulate pile may be caused by liquids which differ from the average supernatant liquid in composition (Irvine, 1980; Sparks et al., 1985), an effect which may invalidate all attempts at geochemical modelling using equilibrium crystal–liquid distribution coefficients.
  Consolidation in large magma bodies is frequently accompanied by substantial assimilation of the roof and walls, which is likely to affect the incompatible trace element contents and isotopic ratios unpredictably and further complicate all modelling (27).
  Evisceration of substantial portions of the contents of magma chambers in ash-flow eruptions has established the existence of large-scale compositional zonation within acid magma bodies which cannot be explained by any crystal–liquid process yet proposed (Hildreth, 1979). The possibility of related effects being latent in basic magma chambers has been little explored but the correlation of plagioclase with olivine compositions in ophiolite gabbros is distinct from that among phenocrysts in MORB. Although there are many factors involved here, it may be unwise to assume that the minerals saturated in the magma at the roof of a magma chamber are necessarily the same in type or composition as those solidifying at its base.
  The changing composition and structure of the liquid from which a phase is separating can change the distribution coefficient dramatically. Ni, which has d 0·5 in the crystallization or melting of olivine in the forsterite–nickel–olivine system (Ringwood, 1956) must have d 2 during the production of komatiitic melts yet has d 10–15 in silica-saturated tholeiitic basalt. Ni behaviour, for this reason alone, is not a good indicator of olivine fractionation (Irvine & Kushiro, 1976; Hart & Davis, 1978; Clarke & O’Hara, 1979).
  The speciation of the trace element in question may greatly alter the bulk distribution coefficient, exemplified by the contrasting behaviour of iron in the ferrous and ferric states in basalts. Cr as Cr³⁺ is highly compatible in spinel and clinopyroxene; as Cr²⁺ it behaves like Fe²⁺ and is mildly incompatible in all common ferromagnesian phases fractionating early from basalt. Eu³⁺ is highly incompatible in all phases fractionating early from basaltic magma; Eu²⁺ is compatible in plagioclase (Morris et al., 1974; Drake & Weill, 1975; Schreiber, 1977). Platinum group element ions, PGE⁴⁺, are incompatible with respect to silicate phases but highly compatible in sulphide; PGE⁰ are highly compatible in metal relative to silicate melt. The bulk distribution coefficients of all these elements are functions of the oxygen fugacity and will be significantly affected by the differences in oxygen fugacity between typical terrestrial and typical lunar lavas. Within this range the behaviour of the Eu/Sr ratio might be a sensitive indicator of oxygen fugacities during crystal–liquid processes involving plagioclase. It may also be sensitive to those changes in oxygen fugacity in closed basaltic systems which can be induced simply by change of pressure, as well as by consequent changes in the speciation of C and S in the melt.
  
• 20. The role of trapped melt in the partial melting (dynamic melting) process was explored by Langmuir et al. (1977) and its role in both equilibrium and fractional melting and crystallization processes further explored by O’Hara (1993). The principal effects are to modify the bulk distribution coefficients of all trace elements towards the value of 1·0, reducing the impact of all crystal–liquid processes on the concentrations and ratios of both highly incompatible and compatible elements. Melt trapped in the cumulate pile during partial crystallization has similar effects on bulk distribution coefficients (O’Hara, 1993). The influence of periodic escape of partially differentiated liquid from a chamber undergoing fractional crystallization without recharge was explored in connection with the behaviour of MORB magma chambers (Cann, 1982). The trace element geochemical effects in the residual liquid composition are similar to those encountered when there is trapped liquid in the cumulates. An additional phase is being removed for which the distribution coefficient of all elements is exactly unity; apparent bulk distribution coefficients of highly incompatible trace elements are grossly modified towards higher values. This leaky fractionation, however, allows the mathematical involvement of a much higher mass fraction of apparent ‘trapped liquid’ than would be permitted by the porosity of the cumulate and is unaffected by any diagenesis of the cumulate pile. The effect of magma escape on the geochemistry and petrology of the cumulates is potentially very different from that of trapped liquid. The cumulates might be pure adcumulates with no trapped liquid yet the erupted liquids might record a partial crystallization sequence which seemingly required involvement of substantial trapped liquid.
  
• 21. Maaløe (1982) and McKenzie (1984, 1985a, 1985b) have argued that trapped melt fractions should be small during mantle melting processes, and Thompson et al. (1984) argued from observed incompatible trace element discrimination in basalts that trapped melt fractions had to be tiny and overall mass fractions of melting small throughout terrestrial basalt petrogenesis. However, a recharged mantle melting regime, such as is believed to operate wherever partial melting is the result of convective decompression of hot mantle, does permit significant trapped melt fractions to be present during more substantial mass fraction partial melting events (O’Hara, 1995b).
  
• 22. Major element chemistry of peridotite suites implies removal of >0·2 mass fraction partial melt in places (O’Hara et al., 1975b). Trace element chemistry of diopsides in residual upper-mantle peridotites suggests that melting beneath mid-ocean ridges may involve pooling of liquids which have formed by fractional partial melting up to 0·25 mass fraction over a range of pressures within the spinel peridotite and garnet peridotite stability fields (Johnson et al., 1990).
  
• 23. Integration of partial melts across realistic melting regimes in which the mass fraction of partial melt varies from zero to some maximum (McKenzie, 1984; O’Hara, 1985, 1995a; McKenzie & O’Nions, 1991) offers a reconciliation of the apparent conflict between the geochemical features favouring high mass fractions of partial melting (contributions from the centre of the regime) with those favouring low mass fractions of melting (trace element geochemical signals inherited from the periphery of the regime). It should be noted that (22), coupled with the widespread acceptance of an average melt mass fraction of 0·1 in the generation of MORB, has significant implications for the incompatible trace element behaviour (see O’Hara, 1995a, fig. 14). Melt integration when the plumbing systems which tap the melt zone are not centred with respect to the plume axis (DePaolo & Stolper, 1996) adds a further complication.
  
• 24. Integration of the residual liquids of perfect fractional crystallization from partly solidified magma bodies with temperature gradients can produce liquids whose geochemical characteristics will be closer to or identical with those of equilibrium crystallization of the same parent liquid—and can even yield liquids with higher retention of highly compatible elements while still concentrating the highly incompatible elements to the same amount (O’Hara & Fry, 1996a, table 1). Integration of liquids from a process which locally simulates equilibrium partial crystallization has even more dramatic effects.
  
• 25. Magma mixing is important in MORB (Muir & Tilley, 1964; Bonatti et al., 1974; Rhodes et al., 1979; Walker et al., 1979; Bloomer et al., 1989; Meyer et al., 1989). The Kilauea volcanic plumbing system has proved to be complex (Ryan et al., 1981) with ample opportunity for recharge and fractionation at relatively low pressures. Hawaiian basalt evolution involves much magma mixing and gabbro crystallization (Wright, 1971, 1973; Wright & Fiske, 1971). The potential importance of magma recharge during the partial crystallization of magma bodies was first recognized in the Rhum gabbros (Harker, 1908; Brown, 1956) and has assumed major importance in recent studies of the Bushveld complex (Cawthorn & Wallraven, 1998). Gabbro precipitation in ophiolites was by partial crystallization during extensive magma recharge and escape, leading to substantial major element fractionation between gabbros, dykes and basalts (Norman & Strong, 1975; Browning, 1984).
  
• 26. Modelling of the geochemical effects of a periodically recharged, periodically tapped, continuously fractionated magma chamber (O’Hara, 1977; O’Hara & Mathews, 1981) established that the liquids erupted from such a system could mimic in all geochemical respects the liquids produced during EPC of the total input (parent magma plus contaminants) to the system. These products would be indistinguishable from liquids produced by EPM of a similar source material. This apparent source material, consisting of the primitive partial melting product of the true source plus some contaminants, would of course appear to be a relatively fertile mantle source in trace element terms. All that was required to mimic the appearance of small and variable mass fractions of partial melting was a low but variable ratio of magma escape to gabbro precipitation. These conclusions applied equally to highly incompatible and highly compatible elements, allowing the observed combination of trace element behaviour to develop. The possibility of encountering ‘perched states’ (Walker et al., 1979) in the chamber means that the presence of olivine as sole liquidus crystalline phase in the erupted liquid does not preclude wehrlite or gabbro fractionation in some chamber below, another way of contributing a geochemical signal of ‘occult’ phase removal.
  
• 27. Provided the mass of the resident magma is large relative to individual recharges and discharges, short-term fluctuations in input composition, assimilants and amount of fractionation between discharges will be buffered, yielding considerable uniformity in the erupted products (O’Hara & Mathews, 1981).
  The role of assimilation in basalt genesis has been considerably rehabilitated (McBirney, 1979; O’Hara, 1980; DePaolo, 1981; O’Hara & Mathews, 1981), and in particular that of assimilation of its own earlier eruptive products by an advancing RTXC magma chamber (O’Hara, 1998). Such processes play havoc with the incompatible trace element signal while contributing to the pressure on basalts to conform to low-pressure cotectic liquid compositions. An estimated 21% bulk assimilation of crust has been inferred for the Hasvik layered intrusion, with a mass ratio of assimilation to partial crystallization of 0·27 (Tegner et al., 1999). Extensive and variable crustal contamination marks the magmas of the Bushveld complex (Cawthorn & Wallraven, 1998). Hybridization between basic and acid magmas has long been known in terrestrial environments. Invasion of the breccia lens below major impact basins by hot basaltic magma must afford a uniquely favourable opportunity for assimilation and even hybridization with any remaining impact melt.
  
• 28. Huppert & Sparks (1980a) showed that mixing of new magma input was not necessarily simple or intuitive and provided a process, supported by field observations in the Rhum magma chamber and elsewhere, by which large mass fractions of olivine can be rapidly removed from picritic magma inputs under well-stirred (i.e. quasi-equilibrium) conditions—which would also favour the kinetic reduction of the apparent distribution coefficient for highly compatible trace elements outlined by Hart & Allègre (1980). This and the RTXC magma chamber provide processes which would cleanly rid successive magma batches of their excess olivine without spectacular depletion of their Ni and Cr contents.
  
• 29. Small packet crystallization [SPC; in situ crystallization of Langmuir (1989)] can contribute geochemical signals of ‘occult’ phase separation to erupted liquids. These result from the removal, during advanced crystallization in each small packet, of phases which are not close to the liquidus of the mingled resident magma which is the source of any erupted basalts. Apparent bulk distribution coefficients of all trace elements are significantly modified and depletion of highly compatible elements in particular is greatly reduced (Langmuir, 1989; O’Hara & Fry, 1996b).
  
• 30. Combining fractional crystallization in small packets within an overall periodically recharged, periodically tapped magma chamber (RTXC–SPC–PFC; O’Hara & Fry, 1996b) can produce surprising results. RTXC–SPC–EPC, or involvement of integrated partial crystallization within the small packets has even more dramatic effects.
  
• 31. In all the terrestrial planets the adiabatic gradient with depth will be less steep than that of a dry basalt liquidus. Ascending volatile-free magmas will be superheated and should resorb entrained phenocrysts and react with their surroundings especially if proceeding by percolation or crack propagation (Shaw, 1980; Spera, 1980). Eruption of primary magma (6) requires transport without chemical interaction of a body of hot, potentially superheated, and highly reactive magma through a channel of 130–480 km length in cooler, readily reactable upper mantle with which the magma cannot be in chemical equilibrium as a result of the pressure changes alone. As a chemical engineering problem, this would be a daunting task, which only becomes more difficult on arrival at the base of the crust. Here density contrasts combined with composition contrasts inhibit escape of parental ultramafic magmas direct to the surface and favour assimilation and partial crystallization (Huppert & Sparks, 1985a) to yield residual liquids whose densities fit within a window of eruptability (33). Any hypothesis of primary magma eruption presupposes that these problems have been solved.
  Eruption of superheated ultramafic lavas on the Earth has been accompanied by extensive thermal erosion. Assimilation of several percent or more of substrate may have occurred (Williams et al., 1999) in circumstances probably less favourable to assimilation than those in the conduit. In the case of the lunar lavas there is the risk of contamination by KREEP, a material exceptionally rich in incompatible trace elements and with distinctive isotopic characteristics which might easily obscure the true characteristics of the parental magma (see also 133).
  Even workers who accept the argument that the major element compositions of most erupted basalts have been modified by extensive partial crystallization at low pressure have frequently been fortunate to find at least one flow or hand specimen from a dredge haul in the deep ocean which could be claimed as the local primary magma. On the lunar maria we are looking at the final outpourings of extremely protracted igneous cycles—the type of situation where, from terrestrial experience, advanced fractionation and contamination are most likely to become important—yet almost every large hand specimen is held to approximate in composition to a primary magma derived unmodified from depths several times greater than those from which terrestrial basalts conspicuously fail to erupt unmodified.
  The situation for volatile-containing or volatile-saturated magmas is much more complicated and was considered for the ‘closed’ system case of granitic melts by Tuttle & Bowen (1958). The potential effects of evolved volatiles or equilibrating volatile activities on the country rock also need to be taken into account.
  
• 32. Fluids will ascend by percolation or crack propagation through solids of higher density. The liquids will tend to migrate laterally on arrival at an interface with solids of lower density. Eruption of a melt then requires modification of the magma composition and density by partial crystallization, assimilation or some other process. The expansion of dry basic magma with decompression is greater than that of mantle solids, hence the density differential increases during ascent through a single rock layer.
  The density of a liquid in equilibrium with peridotite minerals changes much more rapidly than this, however, because of the rapid decrease in density as the normative olivine content decreases with decreasing pressure. Volatile-free picritic magmas are low enough in density to ascend through the terrestrial and lunar mantles but not through the terrestrial continental crust, possibly not through either the oceanic basaltic crust or the lunar highland crust where the density relationships are finely poised, and certainly not through a lens of less picritic basalt in a magma chamber.
  
• 33. The relationships between magma composition, magma density and crustal composition are a major factor controlling the ability of basaltic magmas to erupt (Huppert & Sparks, 1980b; Stolper & Walker, 1980). It is difficult to erupt volatile-free picritic and ultramafic magmas on the one hand, and more iron-rich fractionated magmas on the other, because of their high densities. The relatively magnesian basaltic magmas which have evolved from their picritic parents to the point of plagioclase saturation have minimum density and are, therefore, the most likely to erupt through crust of any composition. This observation does not preclude the occasional eruption of more primitive liquids and many researchers have claimed the discovery of examples of these (6).
  
• 34. The alleged copious outpourings of dense picritic magmas of hand-specimen composition in the lunar maria appear anomalous. Reduction of bulk density by vesiculation and frothing is the major factor which can influence the conclusion of note (33). The abundance of vesicles in consolidated lunar basalts (some are more than 25% vesicle space) and the widespread evidence for pyroclastic volcanism on the Moon (83) can circumvent the simple density considerations. Given the debate about the average condensed magma compositions and about their volatile contents at eruption (73) it is doubtful if any useful constraints can be developed at present from density considerations.
  
• 35. Bowen (1928) was the first to recognize the low-pressure cotectic character of erupted basalts and its probable significance; he appears from his text to have inclined to the interpretation that partial melting of peridotite was accompanied or accomplished by pressure release sufficient to explain this low-pressure cotectic character. Yoder & Tilley (1962) and Jamieson (1970b) established the conformity of Hawaiian basalt compositions to those of low-pressure multiply saturated cotectic liquids and further established that these compositions were not in equilibrium with olivine at elevated pressures. Low-pressure cotectic character may be explained by coincidence (36), modification of parental liquids by partial crystallization near the surface (38) or partial melting at low pressure (40).
  
• 36. The compositions of dry partial melts in equilibrium with peridotite minerals change particularly rapidly at pressures up to 0·2 GPa (6 km depth in the Earth, 40 km in the Moon), and rapidly thereafter to 1·0 GPa (30 km depth in the Earth, 200 km in the Moon), all the time in equilibrium with plagioclase, two pyroxenes and olivine, and with the composition moving towards higher normative feldspar and less rapidly towards higher normative olivine. Once spinel replaces plagioclase as the mineral in equilibrium with olivine and two pyroxenes the liquid composition changes steadily but less rapidly up to 2·5 GPa (75 km depth in the Earth, 500 km in the Moon) moving towards higher normative olivine and lower normative hypersthene contents. At higher pressures garnet replaces spinel in the equilibria and the liquid compositions continue more slowly to become richer in normative olivine but rapidly become richer in normative augite and hypersthene. There is no coincidence between the compositions of multiphase-saturated cotectic liquids at low and higher pressures in dry peridotite systems (O’Hara, 1968a; Stolper, 1980; Thompson, 1987; Herzberg & O’Hara, 1998).
  
• 37. The effect of water on peridotite-saturated equilibria is to decrease the normative olivine content of the melts but simultaneously to increase greatly the normative plagioclase contents (O’Hara, 1968a, 1972; Ford et al., 1972, 1977). No coincidence appears in the first 1·0 GPa at least. The effect of carbon dioxide is small until pressures in excess of 2·0 GPa are reached, when melts are then transposed towards higher normative olivine and augite, and towards silica-undersaturated compositions (e.g. Dalton & Presnall, 1998). There may, perhaps, be some special combination of specific pressure, H₂O, CO₂ and other volatile component concentrations at which the partial melt product of peridotite, once devolatilized, matches those of the low-pressure plagioclase-saturated liquids. This is not a satisfactory general solution to the problem.
  
• 38. Low-pressure cotectic character is an automatic consequence of modification of primary or parental magma compositions by partial crystallization at low pressure (O’Hara, 1965, 1968a, 1968b), possibly accompanied by assimilation. Low-pressure modification of parental magma compositions is the best explanation of this feature in planets of Moon-size and greater where the pressure gradient exceeds 0·005 GPa/km. The varying cumulus assemblages of countless layered gabbro bodies in the oceanic and continental crust specify the low-pressure cotectic phenocryst assemblages which their coexisting liquids might carry. Because of the large differences in composition between low- and high-pressure melts (36) extensive fractionation of much olivine, plagioclase and some pyroxene from the primary magmas has to precede eruption of commonly observed tholeiitic basalts. Large mass fractions of partial melting (0·1–0·3) were envisaged to produce an initial concentration of incompatible trace elements, followed by partial crystallization to change major element composition and variably enrich incompatible trace elements in the residual liquids. This mechanism circumvented the problem of collecting very small mass fractions of partial melt from very large volumes of mantle which is encountered by alternative models.
  
• 39. Three-quarters of the Earth is covered by ocean floor, an 8 km thick basic igneous province comprising a volume of 2·5 x 10⁹ km³ of magma formed within the past 250 Ma. This volume is comparable with that of the entire lunar crust and is nearly a hundred times the volume of visible extruded basic magma on the Moon. The view that the oceanic crustal section represents an average 0·1 mass fraction melt of underlying mantle has become well established. Geophysical data suggest that the fractionated gabbroic products of mid-ocean ridge magma chambers, rather than dykes or extrusives, make up 0·5 mass fraction of the parental magmas crossing the Moho (Christiansen & Salisbury, 1975). The vast majority of the extrusives are not primary magmas but are low-pressure plagioclase-saturated cotectic basalts which have been partially crystallized and separated from their crystals at low pressure within that magma chamber (O’Hara, 1968b; Stolper, 1980). Decisive factors pointing to extensive near-surface modification of erupted MORB include the experimental petrology data showing that olivine, which must be a residual mineral in the mantle, is not on the liquidus in the common basalt compositions even at the relatively low pressures of melt segregation from the mantle. Orthopyroxene, which must also be a residual mineral in the mantle, neither appears on the liquidus nor is saturated at the liquidus. The erupted basalts are never in equilibrium with the magnesian harzburgite or magnesian lherzolite assemblage thought to form the upper-mantle residuum. The geophysical evidence for a substantial gabbro layer beneath the basalts almost throughout the ocean basins is supported by direct sampling on oceanic scarps. There is persistent and widespread petrological evidence of eruption of basalts which are simultaneously saturated with olivine and plagioclase and close to saturation with clinopyroxene, the three major minerals of the gabbros and the minerals which should be saturated in the residual liquids from gabbro fractionation. Phenocrysts also provide evidence of mixing of contrasted magma compositions (25), a process easily envisaged in magma chambers which provide a suitable environment for sophisticated crystal–liquid separation processes. Melt inclusions in olivines and scarce hand-specimen samples provide evidence of the existence of the more picritic parental magmas which would be expected to be provided from the upper mantle (Sobolev & Shimizu, 1993).
  Thermal plumes like those of Hawaii and the Marquesas give rise to large volumes of basic magma and to substantial oceanic island volcanic constructs in which partial crystallization and formation of extensive olivine-rich cumulates is commonplace, accompanied by eruption of low-pressure cotectic basalts which have had their compositions modified by partial crystallization high in the plumbing system. Flows with MgO-rich compositions which might represent the parental magma compositions are scarce in volumetric terms or poorly exposed, although better represented in distal regions because of their greater fluidity.
  Terrestrial flood basalt provinces such as the Columbia River, Farrar, Karroo, Deccan, Siberia, Keweenawan and the Coppermine–Muskox–Mackenzie dykes large igneous province are each associated with massive intrusive as well as extrusive activity. Substantial crustal assimilation, partial crystallization in recharged magma chambers, eruption of low-pressure plagioclase-saturated cotectic basalts, and possibly still more extensive partial crystallization in sub-crustal magma chambers (Cox, 1980) mark these provinces. Lava compositions which might represent unmodified parental magmas are scarce and concentrated mainly among the early, not the late stage eruptives. The West Greenland Tertiary eruptives do, however, appear to contain a high proportion of primitive melt compositions (Lightfoot et al., 1997).
  Nevertheless, many teams even in the recent literature have interpreted the geochemistry of lava suites mainly in terms of source composition and melting process variations with little consideration of whether the major element compositions of the samples could ever be in equilibrium with harzburgite–lherzolite or the possible geochemical consequences of partial crystallization en route to the surface.
  
• 40. Partial melting of peridotite mantle assemblages at very low pressures may occur in small bodies with low central pressures (Stolper, 1977), such as the prospective parent body for the howardite–eucrite–diogenite group of meteorites (Vesta; central pressure 0·2 GPa). The pressure gradient here is 0·0008 GPa/km, which would yield pressures of 0·04 GPa at 50 km depth. The problem with this interpretation is that the shift in the compositions of liquids in equilibrium with peridotite towards plagioclase is most rapid in the first 0·2 GPa of increasing pressure (36). The basaltic eucrites are very precisely cotectic for plagioclase and pyroxene at 0·1 MPa (Stolper, 1977) and probably owe their present compositions to phase equilibria control within the outer 10 km of the body. Had they come from deeper, they should have plagioclase on the liquidus detectably before pyroxene.
  
• 41. Partial melting of peridotite assemblages at very low pressure could be entertained as an explanation of persistent low-pressure cotectic character in basalts if partial melting is invariably accompanied by, perhaps even caused by, local pressure release (35). However, at the ambient temperatures required, the pressure differentials at the depth to the upper-mantle source regions will be much greater than the strength of the rocks could sustain other than transiently and is unlikely to explain the observations for an Earth-sized, or even Moon-sized body. It might be appealed to in bodies of asteroid size.
  
• 42. O’Hara (1965, 1968a) exploited published experimental data (Yoder & Tilley, 1962; O’Hara & Schairer, 1963; O’Hara & Yoder, 1963, 1967; Green & Ringwood, 1967) to show that true primary magmas produced by partial melting of dry upper-mantle peridotite at high pressure would be picritic rather than basaltic in character (36). Erupted basalts are not in equilibrium with upper-mantle mineralogies at upper-mantle pressures. A succession of reviews of relevant experimental data (O’Hara, 1968a, 1968b; Stolper, 1980; Thompson, 1987; Herzberg & O’Hara, 1998) have demonstrated that the majority of erupted MORB, tholeiitic OIB and CFB cannot be equilibrated with residual upper mantle. Few of the compositions advanced as primary pass the Roeder & Emslie (1970) constraints on mg-number of melts in equilibrium with plausible upper-mantle olivines. The arguments have been explained with clarity by Cox et al. (1979, Chapter 9). The conclusion has been substantiated in countless experimental studies in many laboratories on a wide range of magma compositions ever since it was first recognized, and is now independently supported by studies of the thermodynamics of the basalt–peridotite system (Ghiorso, 1994; Ghiorso & Sack, 1995; Ghiorso et al., 1995). Composition variation in natural peridotite suites requires the removal of a picritic, not basaltic, composition (22).
  
• 43. See (31). The solubility of gases in silicate magmas increases very rapidly with the first small increments of pressure above the confining pressure exerted at the surface of a small body such as the Moon. Solubility thereafter increases more slowly as the pressure rises. It follows that vesiculation is a solution to the problem of erupting otherwise dense picritic magmas only if the initial volatile content is high, or the magma has already been transported to shallow depth by some other mechanism.
  
• 44. Bowen (1928) observed that many occurrences of picritic compositions among flows and minor intrusions can be demonstrated to be the products of local gravitational accumulation of early formed olivine crystals within basic magma of more ‘normal’ composition. Examples have multiplied since and the phenomenon can be observed in the field in many places.
  
• 45. The first lavas to have reasonable primary credentials came from flood basalt provinces and were the Ubekandt and Baffin picrites (Drever, 1956; Drever & Johnston, 1957; Clarke, 1970), the picrites of the Deccan (West, 1958) and those of the Karroo (Cox et al., 1965). Among ocean-island basalts picritic flows have been found in Iceland (Jakobsson et al., 1978) and picritic distal flows have been identified more recently from Mauna Loa (Baker et al., 1996; Garcia, 1996; Rhodes, 1996; Yang et al., 1996). Truly ultramafic komatiite lavas were described for the first time (Viljoen & Viljoen, 1969a, 1969b) from the Archaean.
  
• 46. A wealth of evidence favours important roles for both partial melting and partial crystallization as major factors in basalt petrogenesis. Superimposition of their geochemical effects is obvious in the major element chemistry and strongly supported by field observations. Can this duplicity be detected and distinguished in the trace element behaviour? When the melting process is one of EPM and the ensuing crystallization process PFC, this is relatively easy (Allègre et al., 1977; Minster et al., 1977; Allègre & Minster, 1978; Minster & Allègre, 1978).
  However, many partial crystallization processes can produce residual liquids more akin to those of an equilibrium process. It is extremely difficult to detect and distinguish the existence of, or the relative contributions of, an EPC process superimposed on an EPM liquid product in the trace element behaviour (O’Hara, 1994). The results for incompatible elements will be adequately described by a single-stage EPM process with a putative mass fraction of liquid which is the product of the mass fractions of liquid in the two real processes; e.g. incompatible trace elements in a real 0·1 mass fraction melt which is later subjected to 0·5 mass fraction EPC can be adequately described by a single-stage 0·05 mass fraction EPM process. The evidence of more sophisticated evolution is not recoverable from the incompatible trace element geochemistry. The compatible trace elements, which might in principle be utilized to identify the two-stage nature of the process, are by their nature subject to great uncertainty about the apparent bulk distribution coefficients to be employed.
  
• 47. Developments in geochemical modelling have proceeded far beyond the simple choices between perfect fractional melting, perfect fractional crystallization and equilibrium crystallization or melting which dominated petrochemical thinking when the first Apollo samples were returned. The subtleties (18–28) are far reaching. It would be an exhausting enterprise to prove conclusively that a particular set of geochemical data could or could not be the product of some plausible mantle composition via some complex partial melting event followed by some less-than-simple eruption and partial crystallization process—the more so when the nature of the source mantle, the pressure of partial melting, the associated volatile fugacities and the alkali contents of the melts must also be regarded as undetermined variables.
  
• 48. This is the central question in mare basalt petrogenesis. If the interpretation of mare basalt compositions advocated in this paper is correct, the typical cross-section of a mare basalt flow unit will show enrichment of ferro-magnesian minerals near the base and relative enrichment of feldspathic components and vesicles towards the top. The lack of detailed field relationships is keenly felt.
  
• 49. Intentionally spare.
  
• 50. Most of our knowledge of meteorites, asteroids, the satellites of Jupiter and Saturn, and the other three major terrestrial planets has been acquired long after the development of views and interpretations about lunar petrogenesis in the early months of the Apollo program. We have the luxury of reviewing this information before reconsidering the information for the Moon. Comparisons are restricted to bodies which have a substantial proportion of silicate materials in their present make-up. When attempting comparisons between bodies of similar size, several additional factors need to be kept in mind.
  The bodies have not accreted at the same distance from the Sun. They will not have had the same rates of accretion or access to the same pool of materials (although much depends upon the balance between a ‘uniform’ cometary input relative to local sources). Promptness of accretion after the birth of the Solar System bears on the capacity to incorporate short-lived radioactive elements and enjoy their brief but major heat output. Speed of accretion may have a major influence on the internal temperatures attained during the process and whether or not magma oceans formed. Style of accretion, i.e. the contrast between accretion from a multitude of small bodies as against one or more major impact events, greatly influences the thermal budgets and geochemical evolution. Promptness, speed and style may all be functions of distance from the Sun (and Jupiter) as also may be post-accretional effects, particularly those involved in the loss of volatiles from any atmosphere developed. Present orbits of bodies may bear little relationship to orbits during critical phases of their evolution. Tidal heating may have been a major influence in the evolution of many bodies at some stage in their evolution (Vesta, Ganymede, Europa, Io and the Moon are all possible victims). Heat production from the decay of long-lived radioactive elements must today be about one-third what it was soon after the formation of the Solar System and the amounts of short-lived radioactive elements incorporated into the larger bodies are unknown but might have varied from one region of space to another and could have dominated the thermal budgets of smaller bodies, even if the larger may have taken longer to accrete.
  There is no logical basis, therefore, to argue that a particular body should have developed in a particular way just because other bodies of similar size have done so. Nevertheless, a pattern of behaviour which appears to be size dependent does emerge. Volatiles appear to have been abundantly available even in the smaller bodies formed, but their retention is restricted to those bodies which have either not undergone prolonged igneous processes or were large enough to retain a substantial atmosphere. Prevalence and duration of igneous activity increases with size except where grossly modified by tidal factors. Early development of highly feldspathic crusts is not a feature of the asteroidal-sized bodies but is in larger bodies (Moon, Mercury and Mars). Io, Venus and Earth have all been subject to too much tectonic and igneous activity to preserve such crusts if they ever existed.
  Spectral reflectance data for asteroids have allowed important conclusions to be reached about the processing of materials in small bodies (Gaffey & McCord, 1977; Beatty et al., 1999). Carbonaceous chondrites and irons appear to dominate. Chondritic materials, prominent in the flux of meteorites striking the Earth in recent time, are scarce in the main asteroid belt but relatively common among objects in Earth-crossing orbits. Irons are prominent among asteroidal spectral types, which may in part be due to their greater resistance to fragmentation. Their silicate mantles may have been preferentially comminuted by successive collisions over the life of the Solar System. The majority of bodies in the asteroid belt are volatile rich. A minority have undergone post-accretionary heating and differentiation; these bodies are concentrated in the inner asteroid belt. All asteroids larger than 450 km present diameter appear to have undergone some heating but the two largest asteroids, Ceres and Pallas, appear to be composed of an assemblage dominated by olivine and magnetite, do not appear to have undergone igneous activity, and are very different from the slightly smaller Vesta.
  Mittlefehldt et al. (1998) reviewed the petrogenesis of the irons, most of which can be viewed as products of iron melt separation and subsequent fractional crystallization from an iron–nickel melt, taking place within 50–60 different planetary bodies. This is conventionally interpreted as the result of separation of a molten iron–nickel core during asteroid differentiation, followed by differentiation during cooling of that core while the silicate mantle evolved independently. The scale of the exsolution textures in the metal phase is a function of the cooling rate. Radii of the various parent bodies calculated, with several assumptions, from the cooling rates range from 3 to 207 km (Mittlefehldt et al., 1998, table 4).
  Metallic iron is in equilibrium with melts containing a slight excess of oxygen over that required to create stoichiometric FeO at low pressure (Bowen & Schairer, 1935). The metal phase is in reaction relationship with the silicate melt during cooling and fractional crystallization. An erupted silicate magma which undergoes reduction by carbon and sulphur loss (Brett, 1976) may precipitate an iron-, nickel- and sulphur-containing phase, probably as an immiscible liquid, which would settle and might then fractionate independently. The silicate portion of the melt will not subsequently precipitate metallic iron and will fractionate towards higher oxygen activities by the precipitation of phases containing only FeO. The inherent oxygen fugacities of the eucrites, the silicate portion of the mesosiderites and the lunar basalts are close to that at which iron metal would be in equilibrium with the melt. Their present states might have been achieved by partial crystallization following reduction and loss of a metal phase.
  Other silicate magmas represented among the asteroids have evolved at still lower oxygen fugacities where much of the FeO has been removed from the silicate magma. The assumption inherent in Mittlefehldt et al. (1998), that these oxygen fugacities have been imposed by circumstances attending the accretion of the bodies and not during eruption of magma at the surface, may not always be correct and some of the irons may represent near-surface igneous precipitates. The hallmark of these should be carbon and sulphur concentrations lower than would be anticipated from closed system processing of chondritic starting materials.
  The mesosiderites are a complex group of meteorites which have evolved by igneous and impact processes. They are a mixture of metal phase (>30%) with silicate clasts similar to the howardite–eucrite–diogenite suite. Eleven percent of large clasts are monogenetic basalts and 5% quenched vitrophyres; 9% primary orthopyroxenites; the basalts are comparable with eucrites with relatively low mg-number and have flat REE patterns with no marked Eu anomalies at concentrations 5–8 times chondritic. A high proportion of the clasts are gabbroic cumulates which may have extreme positive Eu anomalies (Rubin & Mittlefehldt, 1992, 1993; Mittlefehldt et al., 1998). The high proportion of cumulate or plutonic basic igneous clasts may be a significant pointer to the prevalence of high-level partial crystallization events in the evolution of small bodies.
  Few angrite fragments are known (Mittlefehldt & Lindstrom, 1990; Mittlefehldt et al., 1998). They are distinguished by high Fe/Mn, high FeO, and very low (Na/Sm)_N. Fassaitic pyroxenes are accompanied by calcic olivines and spinel or anorthite in extremely silica-undersaturated assemblages. Volcanic outgassing is favoured over nebular fractionation. The samples are strongly fractionated and may be from two parent bodies. Mittlefehldt et al. (1998) have reviewed ideas on the petrogenesis of this group and Longhi (1999) has discussed the development of angrite magmas in relation to the ol–cpx–plag–sp reaction point in the low sodium, high iron oxide system, a relationship which is also encountered in the critically undersaturated portion of the iron-free system CaO–MgO–Al₂O₃–SiO₂ at low pressure (O’Hara & Biggar, 1969). The petrogenesis of this group is very different from that of the two preceding groups, but there is again no suggestion of production of feldspar-rich cumulates by crystal flotation.
  The aubrites are igneous orthopyroxenites with traces of olivine, diopside, calcic plagioclase and a significant amount of sulphide. They are extremely reduced with almost no iron oxide in the pyroxenes and such low oxygen fugacities that both calcium and chromium behave partly as chalcophile elements. Basaltic members of this suite are not known, nor are they linked to a particular group of irons, but links to the strongly reduced enstatite chondrites are suspected. This group is important in indicating one extreme of the wide range of oxygen fugacities attending igneous processes in different locations within the Solar System.
  None of the above groups of meteorites has been reported to have a significant positive cerium anomaly, indicative of evolution under oxidizing conditions, regardless of where they were recovered.
  
• 51. Predominantly silicate bodies of lunar size and density (Moon, 1735 km radius, 3·344 g/cm³) have formed elsewhere in the Solar System with an apparent surfeit of sulphur (Io, 1816 km radius, 3·55 g/cm³) and water (Europa, 1536 km radius, 2·99 g/cm³). The mechanics of accretion of a body of this size do not preclude acquisition of volatiles (BVSP, 1981, table 4.2.1). There are in all five other bodies in the Solar System—Titan, Callisto, Ganymede, Europa and Io—whose combinations of size and mean density suggest that they may have a silicate and metal portion whose size is comparable with that of the Moon. All but the last of these contain significant amounts of ice. This group has a non-volatile (i.e. rocky) portion intermediate in size between the asteroids and Mercury, the smallest of the terrestrial planets. All are satellites and details of their features have been given by Beatty et al. (1999). Triton, Pluto and Charon are also ice-rich bodies which have similar mean densities to Ganymede and Callisto but are much less massive and would have rocky cores much smaller than that of the Moon but still significantly larger than Ceres, Pallas or Vesta.
  Titan, the major satellite of Saturn, is unique among satellites in having a relatively dense atmosphere and this atmosphere is methane rich. Callisto, the outermost Galilean satellite of Jupiter, has a very ancient heavily cratered surface and was thought to have escaped internal differentiation, implying that neither the energy of accretion nor that from radioactive decay was sufficient to even partially melt the interior, still less produce a silicate magma ocean. Recent results from Galileo indicate that this body may be partially differentiated with a conductive layer under the surface (Showman & Malhotra, 1999). Ganymede has an ancient but much tectonized surface, is composed of about 40% ice, and has differentiated into a thick ice crust overlying a silicate mantle and a small, still molten iron core. This difference in behaviour between two otherwise rather similar bodies is attributed to a tidal contribution to the internal heating of Ganymede. Europa, which is subject to somewhat larger energy input from tidal forces, retains an H₂O layer 100–200 km thick with an ice crust of perhaps as little as 1 km in places (Hoppa et al., 1999), the surface of which is little cratered and relatively young. It is resurfaced and deformed in ways which suggest that the ice-rich crust may overlie a water-rich ocean beneath which there may be silicate volcanic activity. Internal differentiation has produced an iron or iron–sulphide core.
  
• 52. Thick ice covers the silicate crusts of Europa, Ganymede and Titan (Callisto may be little differentiated) and conceal whether those crusts are feldspathic like the ancient lunar highlands. The silicate crusts on Europa and Ganymede are here predicted to be similar in some respects to that of the Moon, but not volatile depleted of course.
  
• 53. Small but still substantial bodies (Ceres, Pallas) appear to be composed of carbonaceous chondrite and are, therefore, liberally provided with the potentially volatile hydroxyl, sulphur and carbon compounds. If other comparable-sized bodies, such as Vesta and the parent bodies of the mesosiderites, which have undergone igneous processes are derived from the same geochemical pool as other asteroids, mechanisms must exist which lead to massive volatilization losses, including losses of the alkalis. It is simplest to assume that the volatilization losses are associated with the igneous processes. Even a much larger body such as Mars, which clearly acquired a substantial budget of volatiles during accretion, has lost a substantial part of the atmosphere and possible hydrosphere originally produced and long sustained by outgassing. Venus, a much larger body with an abundance of volatile components, with sustained and widespread volcanic activity and a dense atmosphere, appears to have lost water equivalent to a global ocean 75 m in depth.
  Plenty of water was available to accrete to bodies smaller than the largest asteroids and it did accrete in large amounts to bodies whose rocky portions are no larger than the Moon. Present volatile contents of these bodies and the Moon can be related to their surface temperatures through a combination of proximity to the Sun or Jupiter and exposure to tidally driven volcanism. Where conditions of tidal heating have been sufficient to sustain some long-term volcanic activity in Europa, the amount of ice retained is much less. Where tidally generated partial melting is rampant in Io, ice and water have disappeared but there is still plenty of sulphur, and probably carbon, too, although this is as yet unproven, to provide volatiles to drive pyroclastic volcanism today. Io may originally have had a substantial budget of water and other volatiles. Escape velocities from the surfaces of these bodies are very low. This water would have been lost if the conditions of accretion had been such as to generate a silicate magma ocean during their formation.
  Formation of silicate magma oceans was not, therefore, an inevitable accompaniment to accretion of bodies of lunar size. The existence of this class of bodies establishes two fundamentally important points. They accreted plenty of volatiles, and most did not generate global silicate magma oceans during their formation. Indeed, Kopal (1977) demolished the concept of a substantial magma ocean at any time in lunar history on astronomical grounds.
  
• 54. Showman & Malhotra (1999) have summarized much of the new wealth of information relating to all four Galilean satellites and even as this paper is being revised for publication relevant new data are arriving from the Galileo spacecraft (EOS 2000, 81, p. 1), including observation of magma erupting on Io along a 25 km fissure to heights of 1·5 km. The style of volcanism observed today on Io is particularly relevant to consideration of ancient lunar volcanism.
  Io, the innermost Galilean satellite, has only slightly greater size, density and mass than the Moon and is arguably of closely similar bulk composition, although its location in a different part of the Solar System could support any alternative composition consistent with its size and density. It is subject to major input of tidal energy and is the most volcanically active body in the Solar System. A small iron or iron–sulphide core is postulated (Anderson et al., 1996). There is no water ice on the surface, which is young—volcanic resurfacing probably takes place everywhere within 10 Ma and there will be no trace left of the earliest formed crust of this body. The outer parts of Io are likely to have been refluxed many times through the eruption process and its accompanying volatile losses. Initial accounts (Sagan, 1979) stressed a major role for sulphur in the volcanism but the height and steepness of some of the slopes point to the abundance of strong silicate rocks rather than the weaker sulphur as principal components in this topography (Clow & Carr, 1980; Greeley et al., 1990). Young (1984) argued that the false colour images on which the identification of sulphur was based were artificially red; real colours were greyer and bluer with the average colour a slightly greenish–greyish–pale yellow. Areas shown in the false colour images as red are moderate greyish–yellow and may be greenish. BVSP (1981, section 5.8.2) left open the question of whether the lavas were predominantly sulphur or silicate. Nash et al. (1986) reviewed the silicate–sulphur controversy and Carr (1986) concluded in favour of predominantly silicate volcanism. Hapke (1989) considered that elemental S was absent on the surface and that multispectral studies were consistent with dark areas being fresh basalt and lighter areas being variably coated with SO₂, S₂O and CSO frosts. Io may be no richer in S than the Moon, whose basalts are still relatively S rich after eruption into vacuum (73).
  Topographic features include many inactive volcanoes, with five calderas >20 km in diameter per million km². These have steep walls and flat floors and the largest is 200 km in diameter. There is up to 10 km of vertical relief at the poles. Shield volcanoes 80 km across and 2·5 km high with up to 10° slopes are present (Moore et al., 1986) and require a 40 km thick lithosphere for their support. Ra Patera is a 1 km high volcano with slopes of 0·1–0·3°, high eruption rates and low-viscosity lavas. Individual lava flows are up to 250 km long, up to 15 km wide with deduced eruption rates of 60–2000 m³/s.
  The case for magmatic differentiation of Io was made by Keszthelyi & McEwen (1997), who inferred an 50 km thick crust of 2·6–2·9 g/cm³ density, which they considered would be alkali rich with much feldspar and nepheline, and they predicted a forsterite-rich mantle. Magmas were considered to have temperatures <1100°C. In contrast to this view, the highest observed eruption temperatures require mafic or even ultramafic silicate melts with lava temperatures ranging from 1430 to 1730°C and are tentatively identified as producing hypersthene-bearing lavas (McEwen et al., 1998). An eruptive event of 1990 (Davies, 1996) has been modelled as a large silicate lava flow erupting 10⁵ m³/s of lava at 1200°C. Blaney et al. (1994) modelled IR outbursts as silicate magma erupting at 3 x 10⁵ m³/s, rates comparable with those on the Earth and early Moon. Almost the entire thermal output of Io comes from recent lava flows. Consolmagno (1981) argued that the current thermal output of Io is much larger than the long-term input from tidal heating and that the volcanic activity on Io may consequently be episodic. The Galileo mission (Davies et al., 1997; Lopes-Gautier et al., 1997; McEwen et al., 1997; Schenk et al., 1997; Spencer et al., 1997a, 1997b; Stansberry et al., 1997) detected 12 previously known and 18 new hotspots with temperatures consistent with silicate volcanism.
  Volcanic plumes extend 70–280 km above the surface and spray pyroclastic material over regions 70–1000 km in diameter. The mass fraction of silicate melt in these plumes is debatable (Johnson, 1990) but the large amounts of sodium and potassium evaporated probably indicate a significant silicate component which may be more strongly represented among lower-trajectory particles.Volatilization losses from the body are extensive (Spencer & Schneider, 1996; Küppers & Jockers, 1997). Sulphur dioxide is prominent in the atmosphere and much sulphur is being lost entirely from Io. The cloud of sodium selectively lost from Io can be detected from the Earth in a region 500 times the radius of Jupiter or 10 times the diameter of the Moon in the sky. These observations suggest that extensive selective volatilization losses of Na and S would also have accompanied the eruption of the lunar lavas, which still contain more sulphur than equivalent terrestrial lavas. Wilson & Keil (1991) and Keil & Wilson (1993) went further and suggested that on asteroidal-sized bodies the violence of sulphur-driven volcanism may have accelerated basaltic material beyond the escape velocity and led to net impoverishment of the silicate fraction.
  Showman & Malhotra (1999) quoted a current rate of loss of mass from Io of 1 t/s, which translates into a current rate loss of 756 t/m² per billion years. This will be equivalent to loss of 250 m of basalt per billion years or about 1% by weight of a crustal column 25 km deep in that time, and 100 km deep during the life of the Solar System. The loss rate might have been much higher when the rocks were richer in volatiles. The potential for significant loss of sodium and more volatile elements during volcanism over small bodies is clear.
  The final phase of the Galileo mission has explored Io at high resolution (McEwen et al., 2000). Observations include active lava lakes (Pele, Loki patera); a lava curtain; active lava flows emplaced as channelized flows and by flow under insulated crust; nested caldera chains up to 200 by 50 km in extent and several kilometres deep, implying large near-surface magma chambers, and collapse of substantial shield volcano structures; and mountain chains up to 16±2 km higher than the general surface. Several plumes come from distal flows rather than eruptive sources (Keiffer et al., 2000; Lopes-Gautier et al., 2000). Gaseous sulphur has been discovered in the Pele plume where the sulphur–sulphur dioxide ratio is consistent with equilibration with relatively reduced basaltic magmas having oxygen fugacities between those of the IW and QFM buffers (Spencer et al., 2000). Despite the eruption into hard vacuum in a gravity field comparable with that of the Moon, despite the absence of evidence for high-viscosity lavas and despite exhibiting the highest rate of volcanic activity in the Solar System, the volcanic topography of Io, with its huge shields and large calderas, is strikingly dissimilar from that of the Moon.
  
• 55. Venus is very similar to the Earth in size, mass and probably in bulk composition but has a very different distribution of volatiles between rock and atmosphere and may have lost a substantial amount of water. There has been abundant volcanic and presumably intrusive igneous activity, much of it central in type. Abundant flood basalts with distinct flow fronts have not produced maria-like structures. Although no spreading axes have been recognized, up to 10 000 km of subduction suture have been identified (Schubert & Sandwell, 1995) and great lava flows analogous to terrestrial flood basalts in volume and rate are seen with deduced eruption rates up to 10⁶ m³/s and volumes up to 10⁴ km³ (Lancaster et al., 1995). Two large continent-like areas with apparent tectonic deformation are preserved but the proportional area of continental material is lower than in the case of the Earth.
  There are no postulated meteorite recoveries from Venus and no sample returns have been achieved. Unmanned probes Venera 9 and Vega 1 and 2 landed on vast plains units; Venera 8, 10, 13 and 14 on localized plains units (Weitz & Basilevsky, 1993). Volcanic features at Venera 8 and 13 are steep-sided domes, blocky lavas and ash beds, and have non-tholeiitic lava compositions. The crust of Venus is inferred from limited sampling and geomorphology to be overwhelmingly mafic with aluminous (16–18·4%) basalts some of which are highly alkaline and potassic, others tholeiitic, with rather high MgO (8·3–12·8%) and relatively low iron (7·7–9·4%) and low titanium (<1·6%) (Kargel et al., 1993). Large, possibly multi-ringed, impact basins up to 170 km in diameter, with smooth, perhaps basaltic fills, are present on Venus where they probably formed during the past 500 Ma (Alexopoulos & McKinnon, 1994). In a planet-wide review of large igneous provinces, Head & Coffin (1997) considered Venus to have exhibited a global large igneous province 300 Ma ago with possible total mantle overturn. There may be very little ancient crust preserved on Venus.
  The Earth’s surface is dominated by a two-fold subdivision. About one-fifth is thick, high-standing, broadly calc-alkaline and sedimentary continental crust, very little of which is more than 2·5 Ga old and almost none of which has survived from the period of the late heavy bombardment at >3·8 Ga. As in the case of Io and Venus, it may not be possible to directly observe the character of the earliest crust. About four-fifths of the surface is the product of igneous activity at spreading axes, is covered by ocean and is broadly basaltic with minor sedimentary cover, all of which is less than 250 Ma old and most of which is less than 120 Ma old. Superimposed on this pattern are two smaller-scale effects which are greatly influenced by plate motion. Subduction zones at plate junctions are marked by extensive igneous activity which is controlled by hydrous partial melting of upper-mantle and crustal rocks. Plume activity on various scales has resulted in the production of large basaltic igneous provinces including oceanic plateaux, continental flood basalts and great dyke swarms particularly associated with continental break-up, and long island chains within the ocean basins. Central volcanism with development of large intrusive complexes is a widespread feature.
  Nothing comparable with the large lunar impact basins has been preserved and there are no unambiguous maria-like structures (but see 126). Magma types appear to have undergone some secular evolution; the Archaean granite–greenstone belts with their komatiite–picrite volcanism have not reappeared in that form in the later record, large bodies of calcic anorthositic gabbro also appear to be restricted to the early record, and there is a distinctive massif anorthosite association emplaced widely around 1·3 Ga ago (Ashwal, 1993). The vast majority of erupted basalts closely approach low-pressure plagioclase-saturated cotectic character.
  
• 56. Mercury, the innermost planet, is 2·75 times more voluminous, 4·5 times more massive, and more than one and a half times denser than the Moon, and it contains a much higher proportion of iron metal. It has an ancient, heavily cratered crust with highland, lowland and intercrater plains features (BVSP, 1981, section 5.5; Kieffer et al., 1992; Beatty et al., 1999). Smooth plains units are downbowed, suggesting mascon-like features. No central volcanic constructs were seen on the 45% of the surface imaged by Mariner 10 but radar images of the unseen part suggest the presence of a 500 km diameter shield volcano topped by a 70 km caldera (Beatty et al., 1999). There is much less contrast in albedo on the surface than in the case of the Moon and mid-IR spectroscopy points to the presence of anorthositic breccias very similar to those on the Moon, and to iron- and titanium-poor (Fe + Ti <4%) basalts (Sprague et al., 1994). Kiefer & Murray (1987) preferred a volcanic origin for Mercury’s smooth plains and viewed volcanism as a global process. Rava & Hapke (1987) confirmed a crust low in Fe²⁺ and Ti and found no evidence for a second wave of global melting, identifying local outpourings of basalt only. Some evidence was reported for late pyroclastic activity richer in iron and for the presence of more Fe-rich material underlying the surface. Representation of samples of Mercury among the meteorites is considered a low-probability event and none can be positively identified (Love & Keil, 1995). Aubrites were considered the most likely group. Given that the planet is rich in iron, the very low iron oxide content of probable partial melt rocks at the surface points to a very low oxygen fugacity in the silicate mantle and possibly throughout the body. If it accreted in its present position relative to the Sun without significant contributions from comets, low volatile content, except for carbon and sulphur, and low oxygen fugacity could be envisaged as a primordial feature. If a lunar magma ocean is rejected, however, similarities between the crustal compositions on the Moon, Mercury and Mars (below) may favour a model of initial volatile accretion followed by losses during igneous processing in all three bodies.
  The surface of Mars is now well surveyed and numerous summaries are available (BVSP, 1981, section 5.6; Carr, 1981; Kieffer et al., 1992; Beatty et al., 1999). Ancient cratered highlands cover the southern hemisphere; the northern hemisphere has been extensively resurfaced with basalt and has some of the largest central volcanoes in the Solar System. The evolution of volcanism with time has been summarized graphically in BVSP (1981, fig. 5.6.1b). The Mars global surveyor laser altimeter experiment has shown that the southern hemisphere is on average 6 km higher than northern. Internal convective processes are required to thin the northern hemisphere (Smith et al., 1999). Crustal remnant magnetism is mostly confined to the ancient heavily cratered highlands, where there are broad E–W-trending linear features up to 2000 km long, on a much larger spatial scale than associated with spreading axes on Earth (Connerney et al., 1999). Any internal convective activity was greatly diminished before the growth of the great central volcanoes because there is no suggestion of development of Hawaiian–Emperor type volcano chains.
  At least a dozen samples from poorly constrained locations on Mars are available from meteorite collections (McSween & Treiman, 1998). Four extrusive or hypabyssal types include heavily shocked, relatively iron-rich ‘tholeiitic’ andesine or labradorite basalts or dolerites marked by pigeonites strongly zoned to augite. The samples are possibly enriched in cumulus pigeonite, which is foliated either by the flow or cumulus process. Oxygen fugacities are moderate to low, there are no Eu anomalies in the bulk materials and chalcophile elements are depleted. Three lherzolitic samples are conspicuously richer in the ferrous components of olivine, orthopyroxene, pigeonite and augite, and in the albite component of plagioclase than most comparable terrestrial rocks. Crystallization ages of all the above are in the range 330–180 Ma and they may all come from the same region of Mars and been ejected in a single impact event. There are three 1·3 Ga old clinopyroxenite or wehrlite samples composed of iron-rich augite, some pigeonite and olivine with minor oligoclase, orthoclase and titanomagnetite. They are thought to be pyroxene cumulates within a lava flow. The samples have been subject to hydrous and oxidizing weathering before ejection. Also 1·3 Ga old is a cumulate of chromite and iron-rich olivine with interstitial augite, pigeonite and sodic plagioclase. The final specimen is ALH84001, a relatively magnesian very ancient (4·5 Ga) orthopyroxenite cumulate showing later development of carbonates and disputed evidence of life on Mars. Fe/Mn ratios in Martian samples are distinctively low relative to rocks of the Earth and Moon and somewhat lower than in the eucrites. Despite the relatively high oxygen fugacity evident in available samples from Mars, a positive Ce anomaly is not reported (Laul et al., 1986).
  Over one-third of heavily cratered Martian highlands are ancient cratered plains units which are probably volcanic outpourings and there are some volcanic features suggestive of ash-flow tuff eruptions in Tyrrhena Patera, Hadriaca Patera, Amphitrites Patera and other ancient volcanoes within the highlands which have the morphological characteristics of pyroclastic-containing composite volcanoes (Scott & Tanaka, 1981). McSween et al. (1999) summarized the geology at the Pathfinder landing site on ancient intercrater plains in Chryse Planitia, specifically on a flood deposit. Blocks of andesite, similar to icelandite rather than terrestrial orogenic andesite, were analysed by Golombek et al. (1997) and Rieder et al. (1997). These workers viewed any inferences about the role of water and plate tectonics in the petrogenesis as premature.
  Sparsely cratered uplands plains are present in Lunae Planum, Syria Planum, Sinai Planum and Hesperia Planum. Wrinkle ridges and inconspicuous flow fronts are identified. There are no recognizable vents. Layered units are exposed in the walls of canyons and channels. Floor fractured craters occur, as on the Moon, associated with the junction between the cratered highlands and the basalt plains. They have been interpreted as volcanically modified craters with basaltic magma intrusion lifting and cracking the debris fill.
  Mars preserves three large impact basins (Hellas, Argyre, Isidis) in the older, heavily cratered highland terrain of the southern hemisphere, a lower density of such impacts than on the Moon and consistent with either a smaller number of impacts or more prolonged ancient crust formation which has obliterated some basins. The largest, Hellas, is 9 km deep, and 2300 km wide with a raised rim up to 2 km high extending 4000 km from the centre of the basin (Smith et al., 1999). Isidis has a diameter of 1900 km, Argyre 1200 km. Only the Isidis basin appears to have been substantially flooded by basalt magmas in a manner comparable with that seen on the nearside of the Moon and the Isidis floor contains low domes possibly formed along eruptive fissures. Some volcanic units are present on the floor of Hellas and Argyre; there are wrinkle ridges and possible flow fronts in both. The Hellas basin is surrounded by old sprawling volcanoes. Hadriaca Patera on the NE rim has a 60 km diameter smooth-floored caldera. Several similar volcanoes occur on the south rim, with large flow channels leading down into the floor of Hellas.
  The oldest basalt-like surfaces on Mars are the Old Ridged Plains units, which occur as small patches throughout the cratered highlands and also near Isidis. These basalts do not build central constructs and have lunar mare-like surfaces without visible flow fronts and with wrinkle ridges set on broad 10 km wide uplifts. Circular ridges occur on the margins of these plains units, possibly as a result of settling of the crust of lava or of a pyroclastic flow over the rims of preflow craters (blackbird-like features). A large caldera-like feature is mapped within this early basalt unit in Syrtis Major.
  Later igneous activity largely resurfaced the northern hemisphere with the production of vast basalt plains distinguished by the presence of many flow units 10–80 m thick, mostly 20–30 m thick, which are visible at resolutions not as good as in available imagery of the lunar maria (Plescia, 1981). A Northern Plains Unit may be a product of sub-glacial volcanism. There is some evidence of regional dyke swarms which do not extend into the older cratered highlands, and several highly conspicuous, huge central volcanic constructs with very large calderas. Even small volcanoes have large calderas and presumably large shallow-seated magma chambers (Plescia, 1994). Large (up to 1300 km across) shield volcanoes and possible stratovolcanoes, involving very fluid, possibly ultrabasic lavas or pyroclastic flows, have been identified around the huge Hellas impact basin (Peterson, 1978). Central volcanoes abound in the later volcanic history of Mars. Early patera apparently formed by rapid eruption of low-viscosity lavas which created large, low-angle shields distinguished by huge central calderas, indicative of equally large high-level magma chambers in which partial crystallization of parental magmas and imposition of low-pressure cotectic character may have affected at least the later erupted magmas. Alba Patera is 1600 km in diameter, with a 100 km diameter caldera; Apollinaris Patera is 400 km in diameter with a 70–100 km diameter caldera. Patera heights are 3 km, with slopes of <0·5°, implying lavas more fluid than those of Hawaii. Individual flows are 400 km³, small relative to the subsidence volumes of the calderas.
  Large central volcanoes reach heights of 17–20 km above the surrounding plains units, 27 km above datum. Volcano volumes are up to 100 times those of the largest examples on Earth. The aureole of Olympus Mons may be formed from early pyroclastic flows and extends to a 1000 km radius, and a sub-glacial birth has been suggested. These structures again have huge central calderas; for example, Arsia Mons is 110 km in diameter; Ascraeus Mons 40–80 km in diameter with four depressions, 4 km deep at maximum; Olympus Mons is up to 90 km in diameter with the youngest caldera 23 km in diameter; Pavonis Mons is of 45 km diameter with an older flooded caldera of 100 km diameter; even the smaller Elysium Mons, which is 14 km high and 170 km in diameter, has a 12 km diameter caldera. Wrinkle ridges occur on intermediate age floor of the Olympus caldera (BVSP, 1981, fig. 5.6.4), presumably on a consolidated lava lake. Wrinkle ridges also occur in the circular depression (old caldera) of Pavonis Mons. Individual Hawaiian volcanoes are 9 km high shields of <120 km diameter and would fit inside some of the Martian calderas. Crater counts suggest activity extending over billions of years. Arsia Mons, with a volume of 10⁶ km³, has a magma productivity as low as 0·0005 km³/year, much lower than that of Hawaii, which is 0·01–0·02 km³/year. These long-term average productivity rates imply 10⁷ years to fill the collapse space of the largest caldera, which in turn suggests an episodic supply and long intervals between eruptions, yielding a high probability of extensive low-pressure partial crystallization in magma evolution. Flows from embayments in the sides of volcanoes suggest that upward growth had stalled, implying a depth to source of 160 km, and a thick strong lithosphere to support the edifice.
  Mars is distinguished by hotspots stationary for >1 Ga leading to massive shield volcanoes of 1·5 x 10⁶ km³, equivalent to a total large igneous province on the Earth (Head & Coffin, 1997). Plume activity has been suggested as the source of the large central volcanoes, but clearly if there has been any plate motion on Mars the movement must have ceased before the bulk of the eruptions which formed the large central volcanoes. Parallels have been drawn between the Mackenzie dyke swarm and possible dyke swarms around the Pavonis centre on Mars (McKenzie & Nimmo, 1999). Conspicuously absent are any indications of senescent alkaline undersaturated pyroclastic activity such as that which marks Mauna Kea.
  
• 57. Volatiles including water were accreted to Mars in abundance and at some stage there was an atmosphere sufficiently dense to have sustained surface floods. Both the Northern Plains basaltic unit and parts of Olympus Mons may have formed in sub-glacial conditions, suggesting much greater water loss (Beatty et al., 1999). Atmospheric loss to space and to freezing out in the soils, coupled with waning of outgassing by the igneous activity, has greatly reduced the density of the atmosphere. If the ancient Martian crust proves to be as feldspathic as that of the Moon, there will be no compelling need for a global magma ocean to provide for its genesis.
  
• 58. Bodies the size of the largest asteroids clearly did not accrete with sufficient short-lived radioactive sources or sufficient accretional energy to power substantial partial melting, a conclusion reinforced by the suggestion that the much larger Callisto may be a little differentiated rock–ice mixture. Other bodies smaller than the largest asteroids have undergone thermal metamorphism, partial melting, possible core separation and igneous activity (50), which must be due to causes other than internal radioactive heating.
  
• 59. Two external heat sources are proximity to the Sun and heating by tidal deformation. The evidence from the four major satellites of Jupiter, all formed in the same region of space, points to tidal energy being a major factor in determining the differences in thermal history among the satellites and asteroids. Proximity to the Sun may have been a major factor in the rate of volatile depletion once igneous activity had commenced. The generation of four bodies with silicate portions similar in size and density to that of the Moon, but without any neighbouring planet to provide a differentiated silicate target, does nothing to support the large impactor hypothesis for creation of the Moon. Williams (2000), however, inferred from geological evidence that the Moon has never been in close proximity to the Earth, but this extrapolation may be vulnerable at 3·5–4·2 Ga.
  
• 60. There is an array of asteroids and satellites which range from undifferentiated and unmetamorphosed, through metamorphosed to partially melted, and even perhaps to largely melted in their early evolution, most of which provide no evidence for generation of a global magma ocean or flotation of plagioclase-rich crusts. The onset of partial melting should precede any more advanced melting in such a scenario. Based on a porous flow model for magma segregation within the Moon, essentially radioactive heating, and with radioactive heat sources concentrated into the magma, between 5 and 10% partial melting of the deep lunar interior has been calculated (Turcotte & Ahern, 1978). In a survey of accretional heating models, Ransford (1979) explored large body impact effects and concluded that a Moon with some melting in its outer layers but little towards the centre was predicted. Generally <10% partial melting was predicted in the outer few hundred kilometres but many assumptions had to be made to reach this conclusion, prominent among them being the duration of the main accretion phase—the faster, the hotter—and the magnitude of the tidal input from the early Earth–Moon system.
  Why should remelting of already consolidated cumulates occur in the Moon, as is required by the conventional hypothesis for mare basalt petrogenesis, in the absence of an external heat source or internal convection? Lunar basalt generation is more easily explained if it is viewed as the end stage of continuing partial melting with a geothermal gradient which first increased and later may have decreased in a Moon which had not already largely melted, fractionated and cooled.
  
• 61. Remote sensing of the lunar surface and upper crust has been provided by photogrammetry, laser altimetry, radar sounding, seismic studies, gravity measurements, reflectance studies in the IR to UV, and X-ray, -ray and neutron spectrometry. Data have been obtained from Earth-based telescopes, orbiting Apollo command modules, Mariner 10, Galileo, Clementine and Lunar Prospector. The geochemical data measure the local surface regolith composition and are calibrated against returned regolith samples from nine sampling sites. Assumptions implicit in the interpretation of remote-sensing data to support the ‘conventional’ petrogenetic model are that the regolith developed over all highland regions is representative of the immediately underlying bedrock, but that the regolith developed over the maria is not. This paper, however, views mare regoliths as just as representative of bedrock as those from the highlands (110–120).
  Large-scale variations in the lunar highland crust are evident in the vertical sense. Anorthositic regolith is present on higher ground and large impacts excavate through to less feldspathic materials (Tompkins & Pieters, 1999). Thorium contents in highland rocks are inversely correlated with elevation of the surface. The mega-regolith covering much of the lunar highlands is 1–2 km deep, and smooths out a certain amount of lateral variation but large-scale lateral variations at greater depth are evident in the differences between materials excavated and exposed in larger impacts (Pieters, 1986). This implies that the Moon grew to much its present size before the end of the igneous differentiation which created the bulk of the highlands crust. Small-scale variations are consistent with short-range heterogeneity or local late volcanism (Hubbard et al., 1978). Galileo imaging of large areas of the Moon (Greeley et al., 1993; Pieters et al., 1993b) displayed regional differences in the highlands crust, with lowish-albedo material near the South Pole–Aitken basin and relatively little high-albedo (anorthositic) material. Some of the highland light plains materials appear to be too young to be impact sheets and may be volcanic.
  A plot of Th/Ti ratio against Fe (wt %) in lunar samples distinguishes between ferroan anorthosites (Warren & Kallemeyn, 1993), mare basalts, and a variety of other highland rocks. All three quantities have been measured for 18% of the lunar surface by the -ray spectrometer experiment. Each pixel in the results represents an average of the concentration of the element or ratio of elements in 10⁴ t of material. These values, controlled against ground truth, represent good average figures and extend the composition estimates to much greater volumes and surface areas of the lunar crust than are directly represented by the returned samples [see Taylor (1975, fig. 5.15 for Al/Si), BVSP (1981, plates 2.2 and 2.3 for Fe and Ti, and plate 2.4 for Al/Si) and Spudis & Pieters (1991, plates 10.2 and 10.3 for Fe and Ti)]. These compositions could also be related less precisely to ground- and spacecraft-based optical and IR reflectance studies to extend the findings over more than 50% of the lunar surface, leading to the conclusion that the reflectance properties of the maria surfaces are those of the regoliths, not those of the hand-specimen compositions. The plot has been used to survey petrological provinces across the Moon by Davis & Spudis (1987) with results also given by Spudis & Pieters (1991, plate 10.8). The bulk of the lunar highlands have compositions intermediate between ferroan anorthosite and mare basalt in composition. The mare input in the highlands has to be interpreted as older, reworked mare component which it would be desirable to minimize because basaltic clasts are not particularly abundant. Ferroan anorthosite (FAN) clasts appear to be more abundant in examined breccias than clasts representing the Mg-suite or KREEP. The FAN component is expected to predominate in the acceptable solutions. The other immediate constraints to be observed are an acceptable level of total REE, and an absolute value of Th close to 0·91 ppm (Metzger et al., 1977), the most precise of the individual parameters measured from orbit. Solutions which incorporate the pure KREEP component as end-member admit very little KREEP component, otherwise Th rises too high. Too high a mare basalt component in the average highland mixtures conflicts with the petrological observations, unless a substantial part of the mare contribution is in the form of plutonic gabbros and norites which have escaped separation from the Mg-suite rocks. Solutions which incorporate the Th-poor Mg-rich suite end-member do not achieve the required absolute values of Sm and Th in the average compositions.
  Soon maps of the distribution of concentrations of H, Fe, Ti, K, Th, Ca, Al and some REE over the entire lunar surface will be available (Binder, 1998), rather than just for the restricted equatorial swaths provided by the Apollo 15 and 16 spacecraft orbits. Preliminary maps of the global Th distribution on the lunar farside and at the poles (Lawrence et al., 1998) confirm large areas of low and very low concentration which may have been under-represented in the original limited coverage by the Apollo 15 and 16 data, and a prominent area of higher values roughly antipodal to the Imbrium basin which was also under-represented. The new global average for Th concentration in the lunar crust is unlikely to fall significantly below the 0·76 ppm below which a substantial positive Eu anomaly in the lunar highlands might be guaranteed (62). Definitive results for Fe and Ti are awaited and there is some discrepancy between the optical and -ray results (Elphic et al., 1998; Lucey et al., 1998). The thermal and fast neutron experiment (Feldman et al., 1998a) has confirmed the association of more mafic materials with lower altitudes in the lunar highlands and in the immediate surrounds of the large mare basins; of probable large-scale lateral heterogeneity within the deeper highland crust; and of more feldspathic materials with higher altitudes in the lunar highlands, the latter particularly in an annulus surrounding the large South Pole–Aitken basin. The results also identify the basalts of the Crisium and Smythii basins as compositionally distinct from those surfacing the other maria. Water discovered as shallowly buried ice beneath regolith in the permanently shadowed regions in craters at the north and south lunar poles is less in amount than could be expected to have accumulated from cometary impacts over the past 2 Ga and has nothing to say towards the case for an initially hydrous Moon (Feldman et al., 1998b).
  The distribution of TiO₂ contents in lunar mare basalts has been found to be unimodal and strongly skewed with a mode 2·2%, median of 2·5% and mean 3·2% (Giguere et al., 2000), rather than having the strongly bimodal distribution previously apparent from Apollo surface sampling. Although advanced as consistent with the conventional model, this distribution is equally consistent with a low-pressure, gabbro-dominated fractionation trend leading from parental low-titanium, incompatible-element-poor basalts with low or negligible Eu anomalies, towards reducing quantities of high-titanium, incompatible-element-rich basalts with large negative Eu anomalies.
  A recent brief summary (Jolliff et al., 2000) has introduced a new Lunar Science Initiative ‘New Views of the Moon Enabled by Combined Remotely Sensed and Lunar Sample Data Sets’, initiated by the Curation and Analysis Planning Team for Extraterrestrial Materials and supported by NASA’s cosmochemistry program, the Lunar and Planetary Institute, Houston and the US Geological Survey, Flagstaff. Figure 1 of Jolliff et al. (2000) supports the inferences from the Al/Si ratios in Fig. 2 above—FeO values in the lunar maria, remotely sensed by Clementine, peak around 17% and tail off rapidly above 18·5%. Average hand-specimen high-titanium basalts average 19%, low-titanium basalts 20·2% FeO and those advanced as candidate primary magmas contain 19·6–22·5% FeO. To produce average mare regolith FeO concentrations of 17% FeO by addition of average highland materials with 4·7% FeO to such basalt compositions would require the presence of 17·5–31% highland material in the regoliths, which is not observed, contravenes expectations from cratering dynamics, and is inconsistent with the diminutive reciprocal scattering of mare basalt into the highlands.
  
• 62. The geochemical ‘facts’ about lunar highland petrology have at times been in doubt (e.g. Anders, 1978) as they are now about mare basalts. This uncertainty afflicts the REE in particular. The REE are present in lunar rocks and magmas predominantly as trivalent cations and in this form behave as incompatible trace elements rejected into the liquid by all mineral phases except the phosphate minerals. At these low oxygen fugacities europium is present substantially as the divalent cation and in that form is readily accepted into the crystal structure of plagioclase (as are Sr and Ga). Consequently, there should be a preferential concentration of Eu in plagioclase and in plagioclase-enriched materials relative to the amount of europium, Eu^*, which would notionally have been present had the element behaved as expected from its position within the rare earth series as a whole, yielding a positive Eu anomaly (Eu/Eu^* >1·0) in the plagioclase itself. Progressive fractionation of plagioclase and its separation from the other crystallizing minerals might, therefore, yield plagioclase cumulates with low REE and positive Eu anomalies, REE-poor cognate mafic cumulates with complementary negative Eu anomalies, and residual liquids with increasing REE and increasingly negative Eu anomalies if the bulk extract has a positive Eu anomaly. Cotectic crystallization of plagioclase with clinopyroxene and apatite in terrestrial anorthosites does not, however, generate major Eu anomalies in bulk cumulate or residual liquid (Morse & Nolan, 1985) but both the oxygen fugacities and the alkali contents of the magmas are higher in this instance.
  The compositions of hand specimens of lunar maria basalt mostly display significant negative Eu anomalies (i.e. Eu/Eu^* <1·0, often <<1·0). Separation of liquid from residual plagioclase in the mantle during partial melting or from cumulus plagioclase in gabbros during partial crystallization processes could provide an explanation for this negative Eu anomaly in the magmas. The latter interpretation lies at the heart of the proposition (110–120) that the aluminous basalts and average regolith compositions, which are close to plagioclase saturation at the liquidus at low pressure, represent the true parent liquid compositions of maria basalt flows. Those parent magmas would be required to have undergone gabbro fractionation or some other interaction with plagioclase crystals within the lunar crust before eruption.
  However, the proposed primary magmas based on the hand-specimen compositions have olivine, pyroxene and titanium-rich minerals as liquidus phases at all likely pressures (BVSP, 1981, figs 3.4.3 and 3.4.5; Taylor et al., 1991, tables 6.2–6.5). They are not in, or close to, equilibrium with plagioclase crystals, at any pressure. The interpretation chosen by most workers is to postulate the existence of a ferro-magnesian silicate source region which has an in-built negative Eu anomaly which is inherited by the partial melts. That required source is conveniently provided by the deep-seated cumulate products of a postulated lunar magma ocean, from which all the plagioclase and its accompanying europium had been abstracted by flotation during crystallization.
  The highly feldspathic compositions of the lunar highland crust cannot be derived as liquids by partial melting of any likely lunar mantle material in the absence of water. Because the Moon was held to have been volatile depleted from its birth, either a fortuitous accretion of highly feldspathic material late in lunar history or physical separation of plagioclase crystals from other lunar materials was required. Hence the hypothesis that flotation of plagioclase crystals from a lunar magma ocean created the lunar highland crust, a hypothesis which arose from petrological need. The only compelling evidence that plagioclase flotation really occurred and the only compelling evidence that a lunar magma ocean actually existed, is to be found in the alleged positive europium anomaly in the average lunar highlands composition. Demonstration of a positive Eu anomaly in average lunar highland materials is thus essential (necessary but not sufficient—see 93) to the hypothesis of primary magma status for the maria basalt hand specimens because it would validate the requisite complementary negative Eu anomaly in the underlying mantle. Demonstration of this positive Eu anomaly in average lunar highland materials is equally essential to the magma ocean hypothesis and its complementary plagioclase flotation mechanism. The entire edifice of conventional lunar petrogenetic interpretation rests heavily on the alleged positive Eu anomaly in the average lunar highlands. This crucial matter is discussed at some length here.
  An elegant logic underlies the estimate of average lunar highland composition, and was originally laid out by Taylor et al. (1973a, 1973b), Taylor & Jake (1974) and Taylor & Bence (1975), and clearly explained by Taylor (1975, pp. 249–253). Many returned samples from highland sources have been analysed. The concentrations of incompatible trace elements in these samples display excellent linear correlations with one another in log–log concentration plots. If the concentration of any one of these elements in the average lunar highlands composition can be obtained, then all the rest can be predicted with a high degree of confidence. Thorium, whose concentration in the surface rocks had been measured in a wide equatorial swath of the lunar crust by its -ray emissions, provides the required global index to a plausible average composition. Knowing thorium in the average composition, the concentration of the sum of the REE in the average composition can be predicted. The concentration of Eu in highland rocks varies much less than the concentrations of the other REE. Anorthositic samples with low total REE have excess Eu and a positive Eu anomaly; as the concentration of total REE increases in other samples the positive Eu anomaly declines, becomes non-existent and then becomes a pronounced negative Eu anomaly in the KREEPy samples richest in total REE. Knowing the total REE in the average lunar highlands one can, therefore, predict the sign and magnitude of the Eu anomaly in that material. This led to the confident assertion:
  ‘A consequence of the REE abundance patterns is that the average highland composition has a positive Eu anomaly. The consequences of a positive Eu anomaly in the highland rocks are profound’ (Taylor, 1975, p. 251).
  Application of the above procedure had resulted in a representation of chondrite-normalized REE patterns in lunar highland samples which has been reproduced in numerous places (e.g. Taylor, 1975, fig. 5.20) and shows a pronounced positive anomaly (Eu/Eu^* 1·4) in the deduced average highland composition at a total REE concentration of 65 ppm (20 times chondritic) derived from an average Th concentration of 1·5 ppm. The actual correlation between total REE and the magnitude and sign of the Eu anomaly, which underpins the vital step in this argument, was not presented for evaluation. This omission can be rectified using the data of Taylor et al. (1973a, 1973b) and Taylor & Bence (1975) in log–log plots of Th concentration vs concentration of total REE and of Th concentration vs Eu/Eu^*. Taylor et al. (1973a) showed a small negative anomaly. Taylor et al. (1973b, fig. 2) showed no anomaly. Taylor & Jake (1974, fig. 2) showed a positive anomaly. Bence et al. (1975), Taylor (1975) and Taylor & Bence (1975) showed a substantial positive anomaly. The combination of Th, total REE and value of Eu/Eu^* in this latter average composition falls outside the envelope of the data points (Fig. 3). At the adopted average of 1·5 ppm Th a small but definite negative anomaly should have been expected as originally deduced by Taylor et al. (1973a).
  The case for a positive Eu anomaly in the lunar highlands was, therefore, weak. Two developments have promised rehabilitation. The Th values measured from orbit have been revised downwards and it has become clear that the surface of the lunar farside contains a far greater proportion of Th-poor, REE-poor anorthositic materials, in which positive Eu anomalies are more likely to be encountered, than do the lunar nearside highlands. The new situation has been reviewed by Haskin & Warren (1991), who presented a revised representation of highland REE patterns which does not include an updated estimate for the average lunar highland composition (Haskin & Warren, 1991, fig. 8.8). They also presented revised linear correlations of incompatible element concentrations against the concentration of Sm (Haskin & Warren, 1991, table 8.2) together with values for concentrations in chondrites (Haskin & Warren, 1991, table 8.1):
  

  where Eu_C, Sm_C and Gd_C represent the concentrations of these elements in chondrites. Then dividing by the product of Eu_C and Sm_C, and then replacing Eu/Eu_C, etc. by Eu_N, etc., where Eu_N is the chondrite-normalized concentration of the element,
  

  
  But Eu_N^* may also be defined as the arithmetic mean of Sm_N^* and Gd_N^*, or
  

  
  From this it follows that
  

  
  Equation (4) is readily solved for the value of Sm at the condition (Eu/Eu^*)_N = 1·0, when it is to be expected that there will be no Eu anomaly in lunar highland materials. The sum of REE and probable average Th concentration at this condition may then be estimated. This solution yields Sm = 2·649 ppm; REE 48 ppm; Th 0·763 ppm.
  This value for Sm concentration when there is no Eu anomaly lies just within the limits of validity of the Eu–Sm correlation (Sm >2·5 ppm) quoted by Haskin & Warren (1991). Regions of the lunar highland crust which have Th concentrations greater than 0·763 ppm can, on this basis, be expected to contain negative Eu anomalies. At Th 1·5 ppm the new correlations imply (Eu/Eu^*)_N = 0·65, a marked negative anomaly in contrast to the (Eu/Eu^*)_N = 1·4 presented by Taylor (1975); a Th value of 0·6 ppm translates into Sm = 2·231 ppm, slightly outside the valid limits of correlation, and a tentative value of (Eu/Eu^*)_N = 1·15, a distinct but not startling anomaly compared with (Eu/Eu^*)_N 30 reported from analysed lunar highland anorthosites.
  Some words of caution are needed, however. The Eu–Sm log–log concentration plot given by Haskin & Warren (1991, fig. 8.10d) actually shows a range of Eu/Sm in the region of ‘good’ correlation 0·2–1·1 although most data points fall closer to the correlation given; and among selected rock samples whose analyses are listed in the appendices to Taylor et al. (1991), the boundary between the few specimens showing small positive or negative Eu anomalies lies at Th between 1·03 and 1·26 ppm. Europium is approximately constant in amount in highland samples which contain less than 13·5 times chondritic values of Sm (and total REE), but Sm (and total REE) contents are closely correlated throughout the concentration range (Haskin & Warren, 1991). Highland rocks with >0·76 ppm Th, corresponding to Sm 2·65 ppm, 14 times chondritic, with REE >48 ppm are on average expected to have negative Eu anomalies, but a few alkali-rich rocks have flat REE patterns and no Eu anomaly at 100 times chondritic REE (Papike et al., 1998). Some hand-specimen samples with slightly greater than 1·0 ppm Th have small positive or no Eu anomaly.
  It is convenient for the next argument to take 0·9 ppm Th as representing the boundary between highland rocks which contain a positive Eu anomaly and those which do not, because orbital mapping of the -ray flux and inferred Th contents has been published; for example, by BVSP (1981, plate 2.1), where pink was assigned to regions with <0·9 ppm, purple to regions with 0·9–1·9 ppm, blue to regions with 1·9–2·8 ppm, green to regions with 2·8–3·8 ppm, yellow to regions with 3·8–4·7 ppm, orange to regions with 4·7–5·7 ppm and dark orange to regions with >5·7 ppm. Even discounting the dark orange, orange and some of the yellow areas which correlate with the nearside maria basalts (but also exclude some highland KREEP contributions), it is visually obvious that the pink area is greatly exceeded by the purple, blue, green and yellow areas. A very rough calculation indicates that, because of the way in which Eu and Sm vary in the rocks, the area of pink would need to be greater than the sum of purple plus twice that of blue plus three times that of green plus four times that of yellow (exclusive of maria) for there to be a likelihood of a positive Eu anomaly in the average highlands—and this neglects altogether the contribution of nearside KREEP. It is also visually obvious from Haskin & Warren (1991, fig. 8.7o and t) that a majority of analysed lunar highland soils, regolith and polymict breccias, representing a substantial mass of sample, contain >2·7 ppm Sm and >0·9 ppm Th. Spudis & Pieters (1991) did not quote results from the orbital -ray experiment directly but presented a colour-contoured map of the deduced Th contents in the lunar surface (Spudis & Pieters, 1991, plate 10.1). This is seen graphically to have values >0·9 ppm across almost the whole of the nearside highlands surveyed and to lie close to or above that value over about half of the farside highlands, but unfortunately the violet coloration applied to all values of Th <0·9 ppm embraces the boundary between highlands which are expected to have a negative Eu anomaly (Th >0·76 ppm) and those which probably have a positive anomaly. On the basis of these results it is impossible to be confident that the average lunar highland composition across the whole Moon will contain a positive Eu anomaly at all, still less one large enough to support the hypothesis of massive plagioclase flotation into the crust. Latest uncorrected counting results from Lunar Orbiter (Lawrence et al., 1998), however, indicate that the surface of much of the lunar highlands away from the Imbrium and South Pole basins is composed of rocks with low Th contents which may have positive Eu anomalies.
  Korotev & Haskin (1988), in an important paper not referred to by Haskin & Warren (1991) or Papike et al. (1998), concluded that there is no significant positive anomaly in the average lunar highland crust, that Eu/Al ratios are chondritic, and that most of the lunar Al, REE and other highly incompatible elements are now in the crust. It follows that any hypothesis proposing significant plagioclase flotation from a reduced lunar magma ocean is excluded. The requisite substantial positive Eu anomaly is absent, unless the bulk of the plagioclase now in the crust is related to later intrusive activity (Mg-suite or mare-related), not to the ferroan anorthosite suite (FAN). FAN could then be envisaged as relics of a much thinner primordial crust. Palme & Wänke (1977) independently deduced a probable positive Eu anomaly in the lunar interior based on trace element correlations and in particular on the correlation between Eu and Sr in KREEP and mare basalts.
  

• 63. Solidification of a deep body of dry magma has been the subject of intense thought and speculation among the lunar science community because of its relevance to the solidification of the postulated lunar magma ocean (Walker et al., 1975; Longhi, 1977; Solomon & Longhi, 1977; Herbert et al., 1978; Minear & Fletcher, 1978; Minear, 1980; Morse, 1987). Minear (1980) concluded that solidification would be rapid (60 Ma) with only a thin crust developed during most of this time. Solomon & Longhi (1977), among many, concluded that remelting of the magma ocean cumulates would require the retention of trapped liquid during solidification, a conclusion with important consequences for the plagioclase saturation story (93, 98) although there is no necessity for this in major element terms because of the extensive crystalline solutions of basaltic components in the pyroxenes at the relevant pressures. Morse (1987) has argued that, rather than flotation of actual plagioclase crystals, flotation of the less dense, more feldspathic residual liquids from extraction of ferro-magnesian minerals may be the important mechanism. This allows segregation of a potentially feldspar-rich material to the surface long before, and from a much greater volume of magma, than that which would have precipitated actual plagioclase crystals. If the Morse (1987) mechanism operates, a substantial part of the cumulate pile was precipitated from magmas which were not plagioclase saturated at depth, yet formed during enrichment in potential plagioclase of the magmas from which the crust would later form. This mechanism can explain a plagioclase-rich crust which has only a limited positive Eu anomaly. It does not implant a complementary negative Eu anomaly in the cumulate mantle until plagioclase is saturated in the crystallizing magma and so does not ease the problem of generating the marked negative Eu anomalies in the mare basalts. Nor does it ease the problem of generating magmas which do not have plagioclase precipitating as their second silicate phase at low pressure.
  The topic has long been the subject of detailed field, petrological and geochemical studies of smaller, but still very deep and non-hypothetical terrestrial magma bodies. The former concept of crystals nucleating in a static magma body as it cools and settling through the magma to accumulate at the floor fails to explain many features of the petrology of the cumulates (Campbell, 1978; McBirney & Noyes, 1979), although other features appear to rehabilitate the settling of crystals following nucleation in the upper boundary layer (Irvine et al., 1998). In a deep body of dry magma the pressure gradient ensures that the liquidus temperature of a fixed basaltic composition declines from the base of the intrusion to the top, at a gradient steeper than the adiabatic gradient for a flow of convecting magma. If the magma is homogeneous at the outset (Walker et al., 1975), magma cooled at the roof and descending to the floor is likely to encounter the liquidus, become saturated and supersaturated with respect to crystals, nucleate them and then separate a residual liquid. Some form of small packet crystallization (Langmuir, 1989) seems more likely than whole body fractional crystallization. Hot magma rising from the floor of the body is moving towards conditions above its liquidus temperature and will tend to dissolve any plagioclase crystals and nuclei which it contains (Morse, 1986a, 1993).
  The viscosity of a magma increases as it cools and increases dramatically in the temperature interval just above the first appearance of plagioclase. The low alkali contents of lunar basalts ensure that, other things being equal, lunar magmas will have much lower viscosities than terrestrial basalts. Vertical segregation of phenocrysts within lunar basalt magmas will have been more rapid than in terrestrial lavas in spite of the lower gravitational acceleration. However, plagioclase, whatever its size, is likely to be almost neutrally buoyant in lunar and terrestrial basaltic magmas and may be intrinsically buoyant in many mafic magmas (Scoates, 2000) although examples of plagioclase settling in Icelandic pillow lavas are known, as at Litla Skogafell (map ref. 329894), Reykjanes peninsula, Iceland. Unless it undergoes froth-flotation by adhesion of gas bubbles, plagioclase is unlikely to float and probably will not sink in lunar magmas after their eruption. Plagioclase crystals might become relatively enriched in any part of the lava or magma, however, if ferro-magnesian phases were preferentially removed.
  The residual liquid formed during crystallization at the floor of the intrusion is likely to be less dense than the main body of liquid if plagioclase is absent from the crystallizing phases, more dense if plagioclase is among the nucleating phases (Huppert & Sparks, 1980b; Stolper & Walker, 1980). Escape and mingling of the residual liquid in the former case with the supernatant magma will promote convective motion. In the latter case, its mingling will tend to set up density stratification which opposes large-scale convective motion. If some plagioclase develops as suspended crystals and is carried upwards by convecting magma, it will tend to dissolve, reducing the local magma density, adding to the density stratification which opposes convective motion and changing the composition of the liquid towards compositions which should have plagioclase as liquidus phase on cooling. Double diffusive layering may develop (Huppert & Sparks, 1980a). The potential complications as this process continues appear too extensive to be reliably predicted, especially if volatiles are also involved (Huppert et al., 1982). Morse (1986b), however, argued that this process would not operate in the presence of crystals and in the absence of walls to an intrusion. It is necessary to turn to the abundant terrestrial evidence to determine what the outcome might be, bearing in mind the possible differences between the terrestrial and lunar situations in terms of further magma additions, stirring by impact, differences in initial magma composition, and the differences in the adiabatic and liquidus gradients when expressed per kilometre of depth in the two situations.
  The evidence from terrestrial layered gabbro complexes (e.g. the Bushveld, Kiglapait, Muskox, Skaergaard and Stillwater complexes) covers a considerable range of initial magma compositions from picritic through noritic towards anorthositic, and exhibits a wide range of major and minor magma recharge events during consolidation. It indicates that plagioclase nucleates and grows predominantly at the floor of the body, with minor solidification at the roof and walls. Plagioclase accumulation at the roof is not a significant process. Morse (1982, 1986b) noted that unzoned plagioclase of the type found in lunar ferroan anorthosites was indicative of near-equilibrium conditions and probable adcumulus growth at the floor of intrusive bodies. Lunar highland igneous clasts have been derived from many individual plutons in which plagioclase accumulated together with spinel, olivine and pyroxene.
  Complex assimilation and refluxing of plagioclase in the upper layers of a magma ocean was proposed as an alternative mechanism for arriving at an anorthositic crust from a chondritic source composition (Longhi & Boudreau, 1979). Walker (1983), recognizing the inadequate Eu anomaly in the lunar highlands, moved away from the magma ocean concept towards serial magmatism in the lunar highland crust. Longhi & Ashwal (1985) went a stage further, proposing generation of both lunar and terrestrial anorthosites by accumulations within rhythmic layered intrusions followed by tectonic mobilization of the plagioclase-rich portions.
  
• 64. The eucrite parent body and several other disrupted asteroidal bodies (87) have apparently developed deep basic magma bodies in which the fractional crystallization of magmas closer in character to lunar magmas has taken place, also apparently with bottom crystallization and cumulate formation, but without development of anorthositic materials by flotation [and see Scoates (2000)].
  
• 65. The absence of an average positive Eu anomaly (62) precludes any overall enrichment of the lunar crust by movement of plagioclase crystals unaccompanied by a KREEPy component. The enrichment of plagioclase in the highlands must otherwise be accomplished by the movement of liquids containing high normative plagioclase [see Morse (1987)]. The small negative Eu anomaly which may be present in the average lunar highlands is consistent with plagioclase or amphibole crystals being residual in the upper-mantle source of those liquids.
  
• 66. In the absence of a substantial positive Eu anomaly in the lunar highlands (62) there is no other requirement for a magma ocean to have existed even if one cannot be excluded. Current thinking inclines towards a lunar highland crust dominated by the effects of serial feldspathic magmatism whose products are marked by almost total near-surface differentiation into cumulates and KREEPy residual liquids.
  
• 67. The postulated global lunar magma ocean provided for the petrogenesis of the feldspathic lunar highlands in the assumed absence of volatiles. The four lunar-sized rocky satellites which have conserved icy crusts and copious volatiles presumably accreted without developing global magma oceans or their water and volatile contents would not have survived (51). The former existence of an accretionally generated lunar magma ocean remains speculative, therefore, as must the generation of magma oceans in the evolution of Mars, Venus or the Earth.
  Whether or not a magma ocean will form during the free accretion of a planet from near-infinite space is critically dependent upon the accretion rate and on the size of individual impacting projectiles (Wetherill, 1985; Melosh, 1989, section 12.3; Tonks & Melosh, 1993). Kaula’s (1979) model for Earth accretion over a 25 Ma time-span from moderate-sized projectiles suggests that the Moon would have been too small to generate a magma ocean by impact heating. The currently popular hypothesis of rapid reassembly of the Moon in orbit around the Earth from materials ejected by a giant impact of a Mars-like body with the proto-Earth provides yet another scenario. What is beyond question is that partial melting of the outer parts of a planet by accretional heating becomes increasingly likely as it passes through the maximum in accretion rate which precedes the decline in available material. Assuming melt movement towards the surface on a time-scale short relative to that of accretion, this would act as a modified process of zone refining with an initially increasing, later decreasing partial melt fraction and a steadily increasing proportion of melt in the outer parts of the accreting body. It is a matter of opinion whether the Moon or any other terrestrial planet passed through this stage and on into the stage of generation of a magma ocean which then cooled and differentiated as a single body of melt.
  
• 68. The effect of increased water pressure and water solubility on the phase equilibria of basic magmas is to expand the liquidus phase volume of olivine at the expense of those of plagioclase and pyroxenes, and that of calcium-rich pyroxene at the expense of those of plagioclase and calcium-poor pyroxene (37). Melts which have been produced by small mass fractions of partial melting in equilibrium with olivine, two-pyroxenes and plagioclase in a water-bearing regime will, on depressurization and water loss, precipitate anorthosites, troctolites and norites. These rock types dominate lunar highland petrology.
  
• 69. Partial melting of a planet is more probable than global melting (60). The probability and consequences of partial melting and differentiation during the final 50% of growth, rather than after the final accretion of the Moon, were explored by Smith (1981) in a process which resembles partial melting with source material recharge (REXM, O’Hara, 1993) but is complicated by cumulate growth and the presence of large pressure gradients. Methods of generating an anorthositic crust from a partially molten magma ocean have been explored by Shirley (1983) and have led on to serial magmatism models.
  The average composition of the relatively low-density lunar highland crust, 10% by volume of the Moon, contains about 0·9 ppm Th and about 12 times chondritic REE, consistent with the lunar crust representing an 0·08 mass fraction melt of the whole lunar interior assuming the bulk Moon to have been of broadly chondritic composition initially. Such high concentrations of elements which are incompatible in plagioclase are inconsistent with the highland crust representing a substantial cumulus of plagioclase from a global magma ocean, unless the Moon itself is grossly enriched in such elements. The latter solution seems to be excluded in the case of Th by the measured heat flow from the Moon, the known Th content of the crust and the measured U/Th ratios. The positive identification of a small Fe or larger FeS core segregated within the Moon suggests that the whole Moon, not just an outer 500 km thick accretionally heated layer, has been involved in a global chemical differentiation process.
  The Eu vs Sm plot given by Haskin & Warren (1991, fig. 8.10d) shows a scattered but in the main relatively constant value of Eu 0·8–2·0 in samples with Sm <2·5 and a well-populated correlation beyond. This is the classic pattern, familiar to petrologists examining correlations between an element concentrated in a possibly cumulus phase and some index of advancing differentiation. The well-populated correlation is usually interpreted as a liquid ‘line’ of descent [Bowen, 1928; Turner & Verhoogen, 1951, 1960; Carmichael et al., 1974, pp. 46–50; Cox et al., 1979, pp. 6, 22–40, 166–173 (for a clear exposition of the complex issues involved); Wilson, 1989, pp. 14–17]. In the lunar case this would use Sm concentration as the index of differentiation and would be interpreted as showing liquid descent from a parental liquid with 2·5 ppm Sm (13 x chondritic) and a small negative Eu anomaly, i.e. a liquid close in this respect to the average lunar highland composition. Plagioclase (carrying up to 2 ppm Eu and negligible Sm) and at least one other REE-poor phase (pyroxene and olivine are both reported in possibly cumulus textures with plagioclase from highland rocks) accumulate in variable amounts from those liquids to create the Sm-poor compositions. The implied crystal–liquid distribution coefficient for total Eu in plagioclase is a little lower than that for Eu²⁺ alone, as would be expected.
  Physiographic units identified as volcanic rocks before the mission to the Apollo 16 site are chemically distinguished also in the orbital X-ray data (Andre & El-Baz, 1981). The Cayley Plains materials have closer affinities to basalts than to typical terra physically, spectrally and chemically.
  The above interpretation of the Eu–Sm variation further predicts that plots of bulk Cr or Sc against Sm, or any other index of reduction in residual liquid volume, will display a related pattern. At the relevant oxygen fugacities bulk Cr²⁺ is mildly incompatible in olivine, compatible in pyroxene and markedly compatible in spinel, and overall mildly incompatible during cotectic crystallization of norite-like materials. Cr²⁺ appears to follow Fe²⁺, whereas Cr³⁺ is the species strongly concentrated into pyroxene and spinel. This is not capable of immediate testing from plots presented by Haskin & Warren (1991). A plot of the relevant data from selected individual rock specimens (Taylor et al., 1991) shows extreme scatter and is convincing of nothing.
  
• 70. The Moon may have acquired a complement of volatiles during accretion (51) sufficient to support an early calc-alkaline style of volcanism similar to that which may have formed andesitic crust on Mars. Initial eruptive activity on the Moon would then have been violently explosive; later activity may have involved basaltic nuées ardentes akin to those of Ulawun, New Britain (Melson et al., 1972), although the latter appear to involve fragmented solid rather than liquid basalt. Early eruptions on the Moon might have been accompanied or followed by spectacular carbon- and sulphur-gas driven basaltic fire-fountaining similar to that displayed on Io. Even the final consolidated flows on the Moon are conspicuously vesicular.
  
• 71. It is postulated that in a peridotite system containing hydrogen, carbon and sulphur compounds and undergoing progressive heating at pressures up to 2 GPa, the onset of partial melting will be marked by initial development of a silicate melt with much dissolved water. After removal of most of the water in this form, accompanied by some carbon and sulphur compounds, further, higher-temperature melts would contain less potential volatile material, mainly in the form of dissolved carbon and sulphur. Proving this postulate would require an extensive experimental programme which paid due attention to the speciation of these three elements and their compounds in solids and melt as functions of pressure and oxygen fugacity.
  
• 72. The initial water-rich volcanism (70, 71) may have been short lived and provided only a thin feldspathic crust. The extraction of water-poor basaltic melts from aluminous peridotites in the outer 400 km of the Moon would supply melts to the surface which were oversaturated in plagioclase (93) and could deposit substantial amounts of gabbros and norites while evolving residual liquids which would be low-pressure plagioclase-saturated cotectic residual liquids with negative Eu anomalies.
  Mare volcanism took place over a time-span of at least 1200 Ma, from 4·2 Ga or earlier to 3·0 Ga, and may have been more active in the period before the major impacts around 3·9 Ga than in the subsequent period. A high proportion of the materials occult in the highland breccia and regolith compositions have been assigned to the mare basalt suite on remote-sensing and geochemical grounds but basalt fragments are scarce in recovered breccias. These observations are consistent with a substantial presence in the breccias and soils of fragmented coarse-grained gabbroic cumulates which are complementary to the older erupted basalts. A eucrite-like gabbro 61223/4 has been reported among pristine samples from Apollo 16 (Marvin & Warren, 1980; Takeda et al., 1981) but is currently grouped with the Mg-rich highland suite.
  Model bulk lunar compositions which have been suggested, all predicated on the assumption of a magma ocean and a plagioclase-flotation crust, range from 6·0 to 27·2% Al₂O₃ and from 32·4 to 12·9% MgO (Kesson & Ringwood, 1977, table 1). All of these compositions are higher in alumina, and many much lower in magnesia, than chondrites. Thirteen estimates of the bulk composition of the silicate portion of the Moon (Warren & Wasson, 1979, table 1) range from 3·7 to 26·6% Al₂O₃ and from 39·3 to 13·1% MgO, generally higher in alumina and lower in magnesia than estimates of the terrestrial upper-mantle composition. These results follow naturally from the predicate that the highlands are a feldspar-rich cumulate from a feldspar-saturated upper mantle. Lower alumina and higher magnesia values for the bulk Moon follow automatically from the postulate that the lunar highlands are a (wet) feldspathic partial melt product.
  
• 73. Wallace & Carmichael (1992) reported that MORB are the most reduced terrestrial magmas and most terrestrial basalts are saturated with sulphur gases on eruption. Sulphur, a potent volatile, is not depleted in lunar surface rocks and is twice as abundant in lunar as it is in terrestrial basalts. Sulphur-bearing gas pressures will have been high (Sato et al., 1973; Sato, 1979). Sato (1976) calculated vapour pressures in equilibrium with basalt 74275, assuming saturation with FeS and carbon, which imply high volatility of S₂ and SO₂ with pressures up to 10⁶ times ambient and Na up to 10⁴ greater than this with CO as much higher again. CO and COS pressures would dominate the volcanism. Similar results were obtained with glass 74220 (Sato, 1978). Some basalt specimens, whose textures suggest relatively slow cooling and consequently relatively deep burial within the flow units, also contain undeformed vesicles. Gas pressures in the vesicles and hence volatile activities in the magmas must have been in the range 0·1–1·0 bar, 10¹² times the ambient surface pressure. The upper part of the flow must have been a froth or ash-flow during emplacement and initial eruption may have been powered by much higher gas pressures (81).
  Losses of sulphur gases from terrestrial basalts at much higher ambient pressures are conspicuous, as evidenced following the Laki fissure eruption of 1783 and by suggested global effects in the wake of the Deccan and Siberian flood basalt events earlier in geological history. Among the last eruptive products on the lunar surface are areas covered in pyroclastic glass beads. Volatiles do not seem to have been in any shorter supply in the Moon than in Io today. Volatilization losses of sodium and sulphur gases during melt eruption into hard vacuum, similar to that observed from Io, may have modified important geochemical characteristics of the lunar surface rocks during eruption (84).
  
• 74. Roedder & Weiblen (1978) reported early sulphide droplet saturation in very low titanium (VLT) basalts. Grove (1981), seeking to explain the large chemical variations among pyroclastic green glass beads from Apollo 15, found that removal of an FeS-rich immiscible liquid was necessary. He further noted the sulphur-rich coatings on the spheres, implying a sulphur-rich volatile phase in the eruption process which powered fire-fountaining. His preferred interpretation involved production of sulphur-bearing but not sulphide-saturated melts at depth under relatively oxidizing conditions, followed by reduction on eruption accompanied by formation and some separation of an FeS-rich melt.
  
• 75. Why is sulphur not strongly depleted in lunar rocks and the basaltic achondrites when so many other potentially volatile elements are? Many of the ‘vapour-mobilized’ elements recognized by Haskin & Warren (1991, fig. 8.1, specifically Cu, Zn, As, Se, Ag, Cd, In, Te, Hg, Tl, Pb) are also strongly chalcophile. So also are a majority of the non-volatile ‘siderophile’ elements recognized by Haskin & Warren (1991, fig. 8.1, specifically Ni, Mo, Ru, Rh, Pd, Sb, Re, Os, Ir, Pt and possibly Au) which are also depleted. These elements are strongly concentrated into sulphide-bearing layers deposited during partial crystallization and magma recharge events in large magma chambers in terrestrial igneous activity.
  The possibility that separation of sulphur-rich phases has played a significant role in the geochemistry of the lunar rocks needs to be sympathetically re-evaluated. It was scarcely entertained in work summarized by Haskin & Warren (1991), where the term chalcophile is defined, but not employed, and loss in the gas phase was not considered. Sulphur, together with the major elements iron and oxygen respectively, has the capacity to act as the controlling element in the formation of two highly contrasted carrier-phases either of which might have separated from lunar silicate magmas. These are an immiscible sulphide liquid (into which many of the siderophile and some of the volatile elements depleted in the lunar rocks would partition strongly) and a sulphur–oxygen-dominated vapour (into which potentially volatile elements might be expected to partition closely in accordance with their activities in the silicate melt). When such carrier-phases are fractionally removed from the system even in small mass fractions, those trace elements which partition strongly into the carrier-phase are rapidly removed from the residual liquid [see O’Hara et al. (2001)]. The supply of such trace elements can be outlasted by that of the major components S, Fe and O, which condition the appearance of the carrier-phases. These major elements remain in the system in proportions controlled by whatever cotectic equilibria are involved in the continuing formation of the carrier-phases.
  
• 76. Righter & Drake (1996) have discussed core formation in the Moon, Mars and Vesta. Levin (1979) concluded that a small iron core perhaps with moderate amounts of FeS was required in the Moon. The improved moment of inertia data from Lunar Prospector are consistent with a lunar Fe-rich core comprising 0·5–2·2% of the lunar mass, with a radius of 220–370 km, or an FeS-rich core comprising 0·9–5·4% of the lunar mass, with a radius of 330–590 km (Konopliv et al., 1998). Seismic data fix the maximum radius at 450 km, consistent with either an Fe-rich or an FeS-rich core, but the high sulphur content of the mare basalts might point to the latter. The simple existence of such a core may require that the whole Moon has been differentiated, not just the outer regions involved in the postulated magma ocean. Mobility of an FeS-rich liquid may have required relatively high oxygen fugacities (Gaetani & Grove, 1999).
  An argument based on the geochemical behaviour of tungsten has been advanced in support of the case that the materials of the Moon have ‘seen’ a substantial metal phase separation, a metal phase not now preserved inside the Moon. It is based on the coherent ratios of W (siderophile) to La (refractory and highly magmaphile) in lunar and terrestrial samples and their 19-fold depletion factor relative to carbonaceous chondrites (Rammensee & Wänke, 1977). A more oxidized lunar interior would exacerbate the problem; a sulphide-rich core would offer no direct solution, but the mass fraction of metal with which the remaining silicates are required to have equilibrated can be reduced if the iron-rich metal phase is poor in nickel. Martian meteorites have coherent W/La ratios which are much less depleted relative to carbonaceous chondrites (Laul et al., 1986). However, diogenites, howardites and eucrites all have similar ratios to the surface rocks of the Earth and Moon. There is as yet no evidence of a substantial iron core within Vesta and the coincidence suggests that there may be further twists to come in this story.
  
• 77. Lunar highland anorthositic samples, including lunar meteorites, contain a small (10%) but persistent positive Ce anomaly (Masuda et al., 1972; Takahashi & Masuda, 1978) which decreases as the REE content increases. Statistical correlations for lunar highland rocks (Haskin & Warren, 1991) support this general observation but are consistent with there being only a very small positive anomaly in the average lunar highland composition. KREEP and mesostasis has a small negative Ce anomaly but the positive anomaly is carried in the olivines, pigeonites, ilmenite and above all the calcium-rich clinopyroxene of the rocks. Basalt hand specimens have an even more marked positive anomaly. The prevalent small positive Ce anomaly reported from many lunar highland and mare samples is consistent with an episode in their evolution in which REE were distributed between crystals and liquid under conditions more oxidizing than have prevailed in the Earth’s mantle throughout the past 3 Ga, even at destructive margins. Takahashi & Masuda (1978) attributed this anomaly to an alteration effect, possibly hydrothermal alteration or interaction with ice early in lunar history. It cannot be wholly attributed to terrestrial alteration effects because the anomaly was first recognized in the Apollo samples, which have not been exposed to terrestrial weathering.
  Canil (1999) has explored the partitioning of vanadium between silicates and silicate liquid, which is very sensitive to the oxygen fugacity during crystal–liquid partitioning. Vanadium becomes (more) incompatible as the oxygen fugacity rises, with major changes around an oxygen fugacity three orders of magnitude lower than that of the equilibrium Ni–NiO (NNO x 10^–3) where V³⁺ goes to V⁴⁺ not far from where Cr²⁺ goes to Cr³⁺. Canil argued that Archaean komatiites developed at oxygen fugacities close to that of NNO, younger komatiites and basalts at NNO x (10^–1–10^–3) and boninites and arc related picrites at NNO x (10¹–10²). Cr is expected to be largely Cr²⁺ at the oxygen fugacities characteristic of lunar basalts after their eruption, with Eu substantially as Eu²⁺ and Ti partly as Ti³⁺ (Schreiber, 1977). If there has been an earlier episode in lunar petrogenesis marked by higher oxygen fugacities, evidence for it may be found in the ratios and concentrations of this group of transition metals and the two rare earth elements but the argument will be complicated by the lack of agreement about parental magma compositions and the high compatibility in specific mineral phases of some of the species involved.
  A speculative six-step scenario for the generation of the positive Ce anomaly involves:
  
  • (i) formation of a proto-Moon about half the present mass (Cameron & Canup, 1998; Halliday & Lee, 1999) by giant impact on the Earth. This proto-Moon and proto-Earth continue to accrete average carbonaceous chondrite material containing relatively large concentrations of hydroxyl, carbon and sulphur compounds and with a relatively high inherent oxygen fugacity.
    
  • (ii) Partial melting sets in throughout and a small iron–sulphide-rich core separates in the deeper portions. Separation of the sulphide liquid would have been greatly promoted by a high oxygen fugacity. A relatively water-rich melt, potentially feldspathic and with affinities to terrestrial calc-alkaline basic magmas, is formed in the outer parts under conditions of fO₂ somewhat higher than in the current terrestrial mantle at destructive margins. A proportion of the Ce is present as Ce⁴⁺ and partitions more strongly into the liquids in this form than do the trivalent REE, giving rise to a small positive anomaly in the bulk melt. Because of the oxidizing conditions much Eu is present as Eu³⁺ and there is no significant Eu anomaly in the bulk melt.
    
  • (iii) This water-rich melt migrates to the surface and there erupts and irrupts explosively under well-stirred conditions because the Moon is still accreting fairly rapidly and its radius increasing. A high degree of reworking of earlier formed crustal materials may have attended the eruptive process. Plagioclase is a liquidus phase as a result of the diminished pressure and solubility of water in the melts. Water and other volatiles are lost in large quantities and the magmas become reduced by the losses of carbon and sulphur gases. Ce⁴⁺ is mostly reduced to Ce³⁺ and subsequently behaves like the other REE. The small positive anomaly is handed on to the rocks and minerals which crystallize, but some mineral (?clinopyroxene or phosphate) takes a slight excess of Ce, leading to development of a small negative Ce anomaly in the late residual liquids. Eu³⁺ is substantially reduced to Eu²⁺ as the oxygen fugacity falls and thereafter behaves like Sr, becoming strongly concentrated in early plagioclase and excluded from most other minerals relative to the other REE. A substantial negative Eu anomaly develops in liquids which are residual from plagioclase fractionation.
    
  • (iv) Partial melting declines in productivity and now produces partial melts which are less water rich, possibly boninitic, but still relatively rich in carbon and sulphur compounds, and still of relatively high inherent oxygen fugacity. The melts produced continue to develop significant positive Ce anomalies but are produced without Eu anomalies (which there is no means of generating both because of the oxygen fugacity and the fact that many melts are not plagioclase saturated on separation). These mare basalt precursors also erupt explosively, less violently than previously, but more violently than is the norm for modern terrestrial basalts. Surface flow emplacement was predominantly as glowing avalanche deposits giving way to violent fire-fountaining as eruptive activity and volatile contents waned. Extensive volatile loss and reduction by loss of carbon and sulphur compounds takes place, with the effects noted above on the oxidation states and geochemical behaviour of Ce and Eu. Production of these water-poor mare basalt precursors overlapped in time with continuing production of water-rich melts from other parts of the mantle.
    
  • (v) Assimilation of, and interaction with, earlier crustal materials as the outer parts of the Moon accreted eventually imparted a volatile-poor, reduced character to all the basic magma irrupting. Reduced melts formed shallow intrusions and large lava lakes in which partial crystallization led to gabbro fractionation and development of marked enrichment of TiO₂, up to the point of ilmenite and armalcolite coprecipitation at 8% TiO₂ in the melt. Other incompatible elements, including the REE with the exception of Eu, also became enriched. Sulphide saturation and possible metal precipitation led to chalcophile and siderophile element depletion. Most mare basalts have been processed through this type of geochemical filtering.
    
  • (vi) As volatiles in mantle regions were progressively depleted and temperatures of partial melting continued to rise, relatively small amounts of volatile-poor magmas developed and erupt as visible conventional flows, dykes, surface ridges of more viscous magma, domes and maars.
    
  
  The bulk lunar mantle is predicted by this model still to have an inherent oxygen fugacity higher than that typical of the modern terrestrial upper mantle. The lunar mantle will have a negative Ce anomaly but the latest partial melts may still carry small positive Ce anomalies because of the high oxygen fugacity at the source. The same broad scenario of early evolution of water-rich highly explosive calc-alkaline magmas, succeeded by voluminous volatile-rich basaltic materials giving rise to mare-like surface features, succeeded in turn by large-scale basaltic flood volcanism with abundant visible flow units and finally by localized central volcanism can be fitted to the available evidence from Mars (57). Internal conditions in the Martian mantle may also have been oxidizing and a positive Ce anomaly might be present in the older volcanics at least.
  This scenario will not be encountered in bodies too small to develop pressures of 0·1–0·2 GPa through substantial masses of material relatively close to the surface, and to sustain internal radioactive heating for a prolonged period of time. Bodies the size of Vesta, whatever their initial bulk composition, could be expected to be dominated by volatile-poor volcanism because of their low central pressure. Feldspathic calc-alkaline melts could not develop and positive Ce anomalies would not be expected. There is no Ce anomaly in the basaltic achondrites and mesosiderites thought to be the products of basaltic volcanism on small asteroidal bodies where internal pressures may have been insufficient to retain water and other volatiles and the whole body may be reduced. Likewise, there is no Ce anomaly in fresh terrestrial igneous rocks which are derived from an evolved upper mantle with locally variable but overall moderately high oxygen fugacity, nor in the igneous rocks of the Martian meteorites, also derived from an evolved mantle with a lengthy evolutionary history. The possibility that relatively strongly oxidized mantles are a feature of intermediate-sized rocky planets, large enough to accrete and retain volatiles but too small to sustain extensive subsequent mantle evolution, may be worth considering. It would predict that the surface rocks of Europa and perhaps Mercury should also display positive Ce anomalies, whereas those of Io should not.
  The alternative scenario calls for an initially ice-covered Moon with early volcanic rocks being erupted and altered in a strongly oxidizing sub-glacial environment, which would terminate any debate about water-rich volcanism in the lunar highland crust (70).
  
• 78. Processes akin to those which created the major PGE concentrations in the Bushveld, Stillwater and Sudbury complexes, or in the vents during eruption of flood basalts at Noril’sk, may account for some part of the depletion of chalcophile and siderophile elements in lunar mare basalts, although direct precipitation of molten metal alloys or metal–sulphide mixtures is also credible at these low oxygen fugacities. A higher production of such a phase may have some bearing on the marked depletion of Au in lunar basalts relative to those of the HED parent body (Taylor, 1975, pp. 166–170, 176–177). Prolific ores of the noble metals may exist close to the lunar surface where they might be detected by high-resolution remote sensing in the walls and debris of suitable craters.
  
• 79. Within the pressure range in the outer 500 km of the Moon, the potential solubility of water in silicate melts exceeds that of carbon, sulphur and their gases. Water would be preferentially eliminated into the silicate melts forming at the lowest temperatures during radiogenic heating of an initially cool body of broadly carbonaceous chondrite bulk composition. These earliest water-rich silicate melts which would have risen to form the early crust would have been biased towards highly feldspathic calc-alkaline and andesitic compositions (57). On loss of water from such melts plagioclase is a very prominent and abundant early crystallizing phenocryst as is seen in eruptives at terrestrial destructive margins. A high activity of water in the lunar interior combined with an oxygen fugacity within the range of terrestrial or lunar magmas implies a relatively high activity of hydrogen also. Whereas water loss from initially hydroxyl-bearing phases will be neutral in its effect on the oxidation state of the residual liquid, any loss of the highly mobile hydrogen while water is still present will lead to some oxidation of the melt as further water dissociates.
  
• 80. Carbon, sulphur and hydroxyl compounds in silicate liquids are soluble at high pressures and volatile at low pressures. The total effect of losses of dissolved water and water formed by reaction with escaping hydrogen on the residual silicate melt is likely to be reducing. The activity of water in any melt which had contained water at elevated pressure in the Moon would probably greatly exceed that of the volatile carbon and sulphur compounds and water would be lost rapidly from erupting magmas.
  The evolution of carbon-oxide and sulphur-oxide gases from silicate melts underpins the industrial process of iron-smelting and the reduction of oxidized metal ores. Carbon and sulphur react with oxygen in the oxidized ore, leading eventually to separation of immiscible molten metal. At low pressures, even low activities of elemental carbon and sulphur in the silicate melt imply very reduced melts whose oxygen fugacities are close to or below those of the iron–wüstite equilibrium. The effect of pressure on the equilibria is, however, profound. An increase of pressure would be expected to favour the distribution of carbon- and sulphur-oxide gases into denser assemblages of elemental carbon, carbides or sulphides with the oxygen combined with oxidized iron in the melt. High activities of carbon and sulphur can then be sustained in equilibrium with relatively oxidized silicate melts. Graphite can be used as a container for terrestrial basaltic melts in experiments at pressures as low as 5 kbar without smelting problems, and immiscible sulphide liquids are in equilibrium with relatively oxidized terrestrial basic magmas at fairly low pressures. Ferri-ilmenites indicative of very high inherent oxygen fugacities are a characteristic mineral in kimberlites carrying free carbon as diamond.
  
• 81. The lunar samples are strongly depleted, relative to the Earth and chondrite meteorites, in all of the potentially volatile elements with the exception of sulphur, which is more than twice as abundant in lunar basalts as in MORB. When did volatile depletion occur? Was it during initial accretion of the Mars-size impactor thought to have provided a substantial part of the Moon’s mass; during final assembly of the Moon in orbit around the Earth from the ejected and vaporized material; in the outer parts of the Moon during continuing impacts into a lunar magma ocean formed in the latter stages of reaccretion, if such ever existed; during irruption of magmas at the lunar surface possibly followed by formation of lava lakes; or during final eruption of the individual flows which gave rise to the samples recovered? Volatile depletion relative to the Earth might arise or be enhanced at any of these stages.
  Selective volatilization loss during eruption or selective failure to condense or be retained during accretion is a complex topic. Each element may be present in the condensed and vapour phases as a variety of chemical species. Transfer of material will attempt to equilibrate condensed and vapour phases by equalizing the activity of all possible species involving an element in the two phases. In many circumstances the results of the process may be controlled by kinetic as much as by equilibrium phenomena. Even at equilibrium there are no unique quantities to be defined such as the volatility of an element or the distribution coefficient of an element between two phases, and both properties are likely to be functions of both oxygen and sulphur fugacity. What happens in detail will depend upon the compositions of both phases, the speciation of the element within each, and the kinetics of the system, especially if one of the phases (the vapour) can readily escape from that part of the system which is recovered for examination. The relative volatilities of the pure elements are no guide to their relative volatilities from complex silicate melts.
  Little experimental work has been done to define the behaviour of basalt magmas erupting under such conditions. Dobar (1965) reported ‘frothing’ of (partially) molten terrestrial basalt and granite on exposure to a vacuum considerably higher in pressure than that ambient at the lunar surface. Mackin (1969) anticipated the first Apollo mission with a prediction that the maria were filled by nuées ardentes deposits, hence the ghost craters, and also asserted that lavas erupted on the Moon should froth. O’Hara et al. (1970a, 1970b) reported violent volatilization loss and rapid conversion of a molten terrestrial flood basalt composition into lunar-like geochemistry in a vacuum furnace. Vinogradov (1971) also predicted spraying of basalt during eruption on the Moon, and Naughton et al. (1971, 1972) reported on vaporization from heated lunar samples. The large vacuum facility at the Lunar Receiving Laboratory was decommissioned before experiments which would have settled the debate could be carried out. The available field evidence from the Moon does little to constrain basalt behaviour under such conditions, beyond requiring that flows were exceptionally fluid and had very low effective viscosities during the main period of formation of the maria (flow thicknesses estimated 10 m, vast extent, few flow fronts preserved) and were still low in the final stages of filling of Mare Imbrium when very large flow units formed with flow thicknesses of 80 m. Less fluid magmas may have erupted in very small quantities at a very late stage in the Marius Hills. Terrestrial basaltic ash-flow eruptions are scarce (Melson et al., 1972) but if lunar magmas contained even small proportions of volatiles pyroclastic eruption and ash-flow magmatism probably ensued.
  Eruptive activity took place into hard vacuum. Sulphur contents in the mare lavas, like those in the basaltic achondrites, are higher than in terrestrial basalts and would sustain a vapour pressure many times the present confining pressure at the lunar surface. Fire-fountaining certainly occurred at a late stage, producing the Apollo 15 green glasses and Apollo 17 orange and black glasses. Many returned hand specimens of basalt are vesicular, demonstrating that gases were still evolving from the basalt as flow ceased and the magma consolidated. Carbon and sulphur gases are likely to have evolved, disrupting even fairly thick layers of basalt magma. Semi-continuous lava fountaining is favoured as the main type of lunar explosive volcanism (Hörz et al., 1991) but it may have been the only type of volcanism during early mare formation. Models of flow of a gas–particle mix in the Earth and Moon point to much larger vents and much higher eruption velocities in the lunar case, other parameters being equal (Pai et al., 1978) but Housley (1978) argued that lunar basalts would fire-fountain less vigorously than those on the Earth.
  
• 82. Two quotations encapsulate the ‘conventional’ view:
  ‘Some petrologists appealed to loss of Na and K to account for the order of magnitude depletion [of lunar lavas] in sodium compared with terrestrial lavas, perhaps reluctant to believe that so fundamental a distinction could exist in basalt chemistry ... The lunar lavas were extruded at temperatures approaching 1200°C into a hard vacuum of at least 10^–9 torr [10^–12 bar]. Although such conditions might be thought to favor loss of volatile elements, the lithospheric pressure of the lava exceeds that of the vapor pressure of the elements at depths greater than 10^–3 cm. Thus loss could only occur from a thin skin ... Very thorough stirring, mixing or bubbling would be needed to lose elements from depth ... Loss of volatiles is of course inhibited by rapid freezing of the surface of the extruded lavas. The consensus is that volatile loss from the maria basalts during extrusion is trivial’ (Taylor, 1975, p. 150).
  ‘Surface loss of volatiles during extrusion or meteorite impact has not found much favour. During surface extrusion of the lavas, freezing of the surface will rapidly occur, and loss of volatiles will not occur’ (Taylor, 1975, p. 166).
  Taylor’s arguments and final assessment ignored the specific proposal that volatilization losses occurred during fire-fountaining on eruption (Biggar et al., 1971, 1972). Diffusion distances to a surface would then be small and diffusion rates through the melt would not be an important constraint. The surface area between melt and vapour would be huge, lithospheric pressure would not be an impediment to vapour development, and surface cooling of droplets within the eruption plume would be negligible during 100 s flight times (10 km height plume on the Moon) which would be required to produce significant losses of sodium at 0·1 mm diameter droplet sizes. Fire-fountaining on the scale observed over Io would ensure significant losses from larger droplet sizes.
  
• 83. Gas bubbles, vesicles, are a form of ‘phenocryst’ indicating early separation of a volatile phase from a magma. The vesicles present indicate no more than the amount of volatiles evolved after the magma became static and viscous enough to prevent their upward escape. There is no easy way to estimate how much gas has evolved and been lost entirely from the system.
  
• 84. Volatilization of basaltic material by stepwise heating in vacuum proceeds by very rapid loss of soda, slightly less rapid loss of potash, significant decreases in iron and magnesium oxides, and accelerating decrease in silica. Lime, alumina, magnesia and titania are enriched in the residual liquid, with magnesia losses becoming significant only when volatilization is far progressed, i.e. the residual liquid is modified towards gabbroic anorthositic compositions (Yakovlev & Basilevsky, 1994). Volatilization losses are extremely rapid and capable of converting terrestrial flood basalt compositions to lunar basalt-like compositions, dominantly by sodium loss, in the equivalent of 100 s flight time in a fire-fountain (Biggar et al., 1972; Storey, 1973; Storey et al., 1974), when the changes are from basalt towards calcic norite or calcic pigeonite basalt initially. The results of Yakovlev & Basilevsky (1994) even point to a possibility that the gabbroic anorthosite composition of the lunar highlands is a by-product of intense and long-sustained impact modification of more basaltic or gabbroic materials. Impact driven volatilization yielded granitic condensates (Yakovlev & Basilevsky, 1994) but such a component might have been lost entirely from the lunar surface during the period of heavy bombardment.
  The conclusions of Melosh (1989), O’Keefe & Ahrens (1994) and Stöffler et al. (1994) point to lunar highlands in which target melt rocks abound, some of them fairly homogeneous and some forming magma bodies large enough to undergo internal fractional crystallization. Doubt is thereby cast upon the identification of pristine lunar samples on the basis of their cumulus textures and low noble metal contents, the more so if evolution of the impact melt magmas involved separation of a sulphide melt, as it certainly did at Sudbury. Norman (1994) summarized the case against Stöffler et al.’s (1994) interpretation of Sudbury but did not refute the more general conclusions of Melosh (1989) and O’Keefe & Ahrens (1994).
  The parental magmas may have been subject to volatilization losses during their eruption as nuées ardentes or during fire-fountaining (81). Selective volatilization of sodium combined with release of iron originally present as ferric oxide, iron sulphide or carbide would, however, transpose the residual silicate melt composition away from normative plagioclase and towards normative calcium-poor pyroxene, yielding compositions which would tend to display the low-pressure crystallization sequence which is so common among mare samples. Petrologists warned early in the Apollo program of the risk of Na loss in particular (O’Hara et al., 1970a, 1970b; Brown & Peckett, 1971) because of the experimental difficulty of maintaining the sodium content of silicate glasses at liquidus temperatures during syntheses and experiments even at atmospheric pressure. Excess of alumina, released during sodium loss from albite molecule in basaltic compositions, manifests itself in the CIPW norm as extra anorthite and diminished diopside. Excess of silica from the same source manifests itself in extra hypersthene and diminished olivine (which might be tempered by an increase in FeO as a result of the reduction of any oxidized iron during sulphur loss). The most evident petrological characteristics of lunar relative to terrestrial basalts are their high anorthite content in CIPW-normative plagioclase and the high ratio of hypersthene to diopside in their CIPW norms, coupled with relatively low mg-number in the normative ferromagnesian minerals. These geochemical characteristics are precisely what should be expected if the true lunar parental magmas were more akin to terrestrial tholeiitic, or even alkali, basalts and had had their compositions severely modified by selective volatilization during eruption. The scale of the potential effects are such that this might be the end product even if the original parental magmas had been strongly undersaturated nepheline-normative basalts. Until the issue of volatilization during eruption has been resolved by experimentation it is optimistic to advance any lunar basalt as a primary magma. This effect would invalidate all oversimplified inferences about the lunar interior and its evolution derived from the chemistry or phase behaviour of the recovered materials.
  There are several specific geochemical arguments which have been adduced against selective volatilization:
  ‘Another argument for originally low concentrations of lunar vapor-mobilized elements is the low ratio of Na to Ca in plagioclase feldspars from samples representative of the bulk of the lunar highlands. If Na, which is a relatively volatile element, had been as abundant in relation to Ca on the Moon as it is on Earth, lunar plagioclase would be more sodic. It is unlikely that the Na could have evaporated from an anorthosite (with a composition essentially equivalent to pure feldspar) or from mare lavas without leaving excess silica or especially excess alumina which we do not observe’ (Haskin & Warren, 1991, p. 415).
  If one accepts the lunar magma ocean and plagioclase flotation hypotheses for lunar highland origin, then indeed it follows that the entire lunar magma ocean must itself have been Na depleted to crystallize the Na-poor plagioclase which is typical of highland rocks. One may question whether a lunar magma ocean, with its implication of sustained bombardment into a growing pool of magma, accompanied by stirring and splashing on a vast scale, could have formed without massive selective volatilization of elements such as sodium, whatever the composition of the original material accreting to form the Moon. If, as argued here, the highlands are the product of magmas formed as partial melts of an essentially solid lunar interior which then irrupted to the lunar surface, the appropriate question is whether Na and other volatiles were lost during eruption and before the crystallization of the plagioclase now seen in the anorthosites.
  The coherence and restricted range of ratios of potentially volatile trace elements (e.g. K, Rb) relative to involatile trace elements (e.g. U, Sr, La, Sm and Ba) have also been held to be strong arguments for indigenous volatile depletion without subsequent selective volatilization in the lunar rocks. Taylor (1975, fig. 4.21) quoted a ‘good correlation’ between Rb and Ba, but the figure actually shows a variation of Ba/Rb from 0·01 to 0·03. If the samples with the lowest ratio of Ba/Rb have lost no Rb, those with the highest ratio have lost no more than 67% of their Rb.
  Taylor (1975, fig. 4.23) also referred to the ‘close association’ between K and La, but that figure shows a variation of K/La 50–125 and if the samples with the highest ratio of K/La have lost no K, those with the lowest ratio have lost no more than 60% of their K.
  Taylor (1975, fig. 4.24) demonstrated a range in K/U from 8000 to 50 000 and if the samples with the highest ratio of K/U have lost no K, those with the lowest ratio have lost no more than 84% of their potassium. Haskin & Warren (1991, figs 8.10 and 8.13), with a more extensive dataset, demonstrated a variation in K/Sm ratios of 40–500 in all lunar samples, 100–300 in mare basalts. If the samples with the highest ratio of K/Sm have lost no K, those with the lowest ratio have lost no more than 67–92% of their original K. Na/Sm in the lunar samples varies from 50 to 2000. If this variation is wholly due to selective volatilization of Na it would imply sodium losses of up to 97·5% of that originally present.
  If the samples richer in the most volatile element have already lost some of the volatile element, all these maximum loss estimates must be increased. The ratios of volatile to non-volatile minor and trace elements in lunar rocks do not preclude selective volatilization, even of trace elements whose activities in, and vapour pressures over, the silicate melt would have been necessarily low relative to those of sodium, carbon and sulphur gases. Both Taylor (1975) and Haskin & Warren (1991) referred to the systematics of the Rb–Sr isotopic system, which appears to preclude the Moon’s strontium having been exposed to higher than present Rb/Sr ratios for any significant period of time, but even if there has been no selective volatilization of Rb, this imposes no direct constraint on possible losses of Na, S and O, which are petrologically much more important.
  
• 85. The howardite–eucrite–diogenite (HED) group meteorites include many regolith breccias. Petrology and petrogenesis of this group has been reviewed by BVSP (1981, section 1.2.8) and most recently by Mittlefehldt et al. (1998). These groups stressed the low-pressure olivine + pigeonite + plagioclase-saturated cotectic character of the lavas and hypabyssal samples (Stolper, 1977; Stolper et al., 1979); the presence of cumulate textures in gabbro and orthopyroxenite samples; the scarcity of vesicles in the eruptives; the development of small negative Eu anomalies in some of the lavas, and of distinct positive anomalies in the cumulate samples which are consistent with plagioclase fractionation from reduced basic magmas. The samples are siderophile depleted and relatively sulphur rich. There is no evidence for formation of anorthositic materials by plagioclase flotation or any other mechanism in the magma bodies (64). No high-titanium basalts have been reported. Vesta, a possible parent body of the HED meteorites, has a diameter of about 525 km, and a poorly constrained density of 3·7 ± 0·5 g/cm³ (Ghosh & McSween, 1998) based on an H-chondrite model. The body has a differentiated, reduced basaltic surface (eucrite) of a type which is very rare among the asteroids as a whole, although a larger number, particularly among the larger bodies, have the reflectance properties of mesosiderites. Cruikshank et al. (1991) identified three possible small source bodies for the available suite of HED meteorites other than Vesta itself. Hiroi et al. (1995) identified some small Vesta-like objects in near-Earth orbits and 20 more in Vesta-like orbits.
  
• 86. Although plagioclase-saturated low-pressure cotectic character provides a strong case for low-pressure partial crystallization in larger planets (35–41), Stolper (1977) concluded that it resulted from eucrite formation as primary magmas at a pressure not too different from atmospheric, which in turn implied an unsampled calcic plagioclase-bearing peridotite interior. Key arguments in reaching this conclusion were recognition that the mg-number of minerals in the available cumulate samples could not be directly related to those of the available lavas, and the presence of incompatible trace element features in the lavas inconsistent with perfect fractional crystallization but consistent with small and variable mass fractions of equilibrium partial melting. Subsequent developments in modelling of sophisticated crystallization processes has invalidated these trace element arguments (18–28). The main conclusion to be drawn from the cumulate eucrites is perhaps that magma chambers which could act as sites of low-pressure partial crystallization certainly existed on the parent planet. A direct link between specific lavas and specific cumulates is not required in a small sample set. There is evident uncertainty about the petrogenesis of this group of meteorites (Mittlefehldt et al., 1998). The balance of evidence suggests an equally viable alternative interpretation in terms of partial melting, perhaps of a carbonaceous chondritic source, eruption with extensive volatilization losses, formation of large magma bodies close to the surface, extensive partial crystallization and eruption of residual liquids to form a surface crust.
  
• 87. Hubble Space Telescope observations of Vesta indicate a layered structure of eucrite overlying diogenite in turn overlying peridotite which has been excavated by a 450 km (almost hemispherical) crater 8 km deep with 8–14 km high rims and a 13 km high central peak (Gaffey, 1997). Other craters are observed up to 150 km in diameter and a few kilometres deep. Miyamoto & Takeda (1994) inferred derivation of the Moore County cumulate eucrite meteorite from 8 km deep in a 10 km thick crust cooled from the solidus in <10 Ma. This depth would be consistent with crust formation from an 10% melt fraction of the whole interior of a 525 km diameter body. Gabbroic samples exist which are clearly cumulates formed within large magma bodies differentiating at low pressure. Orthopyroxenite samples are known which could represent an ultramafic layer at the base of those magma bodies. Diogenite (orthopyroxenite)-rich plutons were envisaged by Warren (1997).
  Ghosh & McSween (1998) argued for simple radiogenic heating to power the volcanism of Vesta but a magma ocean has been proposed, with development of a core between 5 and 30% of the body’s mass (Ruzicka et al., 1997). Righter & Drake (1997) went further and proposed total melting of the body, core formation followed by well-stirred equilibrium crystallization to 0·20 mass fraction of remaining melt, followed finally by gravitationally controlled fractional crystallization of the remainder.
  
• 88. HED basalts are also relatively rich in sulphur and may have separated both sulphide melts and sulphur gases during their irruption (73–75, 78).
  
• 89. Eruptions at the surfaces of small bodies without permanent atmospheres can be expected to produce optically dense fire-fountains in which particles will suffer little cooling and will fall back to form lava ponds and flows, with <1% surviving as discrete clasts (Wilson & Keil, 1997). These are the ideal circumstances for extensive selective volatilization losses during eruption (82). Such losses are essential on a large scale if the HED suite and the mesosiderites are to have evolved within small bodies from volatile-rich, somewhat oxidized parental materials such as the carbonaceous chondrites. Sodium loss in particular is essential because all the chondritic meteorites are relatively sodium rich and have actual or potential plagioclase in the oligoclase–albite range.
  
• 90. Most of the meteorites which show evidence of early igneous evolution within the asteroids are characterized by low oxygen fugacities, extremely calcic plagioclases, low sodium concentrations and high normative orthopyroxene, characteristics also shared with the lunar igneous rocks. These are predictable results of extensive volatilization losses of sulphur and sodium from basaltic compositions produced by partial melting of more chondritic peridotites.
  
• 91. If there is no positive Eu anomaly in the average lunar highland crust (62), there can be no complementary built-in negative Eu anomaly in the underlying mantle [contrast Taylor & Jake (1974, fig. 3)]. Positive Eu anomalies have been reported for some lunar metabasalts and ophitic basalts (Laul et al., 1978), and only slightly negative anomalies for some ferrobasalts when the REE concentration is about 10 times chondritic. When dealing with regolith samples, however, it is necessary to be aware of comminution and sampling effects. Crushing and mineral separation experiments on 70135,27 (Haskin & Korotev, 1977) established the extent of variability to be expected. Clinopyroxene is the major mineral carrier of REE and, together with olivine, pigeonite and ilmenite, has a positive Ce anomaly. The mesostasis, which tends to be preferentially comminuted, is 400 times richer in REE than the minerals and has a negative Ce anomaly [see also Blanchard et al. (1975)]. Haskin (1978) summarized the trace element geochemistry of the Mare Crisium fragments and concluded that the soil–rock differences mainly reflected the comminution effects, whereas inter-particle differences reflected mainly sampling problems. There may or may not be a small Eu anomaly in either sense and there is a small positive Ce anomaly in the average soil. The bottom line is that there is only a very small Eu anomaly in those basalts poorest in titanium.
  
• 92. If there is no relative deficit of Eu in the underlying lunar mantle, then there is no inbuilt negative Eu anomaly for the later mare basalts to inherit at birth. Parental melts would have contained no more than small Eu anomalies as is the case for some lunar VLT basalts which are low in overall concentrations of incompatible elements. The marked negative Eu anomalies in the other mare basalts must, therefore, have been imposed on the residual liquids by fractionation of plagioclase, probably as gabbro, after arrival of the magmas at the base of, or within, the lunar crust. An alternative would be extensive re-equilibration of the magmas with plagioclase already located in the crust, i.e. by a form of assimilative reaction. In either event, the erupted basalt compositions would be required to be close to plagioclase saturation at low pressure and cannot be regarded as primary magmas.
  Walker (1983), recognizing that the lunar mantle might not contain an adequate negative Eu anomaly to bequeath to later basalts, proposed a mechanism for an origin of the negative Eu anomaly in the mare basalts (assuming magmas of the hand-specimen compositions) involving imposition of a cryptic plagioclase crystallization signature, analogous to the cryptic clinopyroxene crystallization signal seen in many terrestrial continental flood basalts and MORB. However, although the clinopyroxene cryptic signal is readily explicable in terms of a small depressurization of essentially dry magmas between crustal magma chamber and surface eruption, the same change in the phase equilibria expressly forbids generation of a cryptic plagioclase crystallization signal in the lunar basalts by such a mechanism (93).
  If the mechanism of flotation of potentially feldspathic liquids (Morse, 1987) operates (see 63), a substantial part of the cumulate pile was precipitated from magmas which were not plagioclase saturated at depth, yet formed during the enrichment in potential plagioclase of the magmas from which the crust would later form. This mechanism can explain a plagioclase-rich crust which has only a limited positive Eu anomaly, but it does not implant a complementary negative Eu anomaly in the cumulate mantle until plagioclase is saturated in the crystallizing magma and so does not ease the problem of generating the marked negative Eu anomalies in the mare basalts. Nor does it solve the problem of generating magmas which do not have plagioclase precipitating as their second silicate phase at low pressure (see 93).
  
• 93. The liquid compositions which form during low mass fractions of partial melting of polyphase mantle mineral assemblages are necessarily close to simultaneous saturation at the pressure of their formation with all the crystal phases present at the solidus of the mantle assemblage. The phase equilibria of the liquid fraction will show a close approach to simultaneous saturation in these phases at the liquidus at that pressure [subject to note (6) and to the possible absence of a solid phase which is in reaction relationship with the liquid at the time of its formation]. The effect of pressure on the phase equilibria relevant to peridotite melting and basalt petrogenesis in dry systems has been long established (O’Hara, 1968a; Cox et al., 1979, figs 9.9–9.12). Dealing first with the remelting of cumulates (without change of pressure on the cumulate before the remelting event), liquids which are produced in equilibrium with olivine, orthopyroxene, and clinopyroxene plus plagioclase in the range 0–1·0 GPa or spinel or garnet at higher pressures up to 2·5 GPa, can be expected to display the low-pressure crystallization sequence olivine joined next by plagioclase, then by clinopyroxene and finally by calcium-poor pyroxene as temperature falls. This predicted behaviour is displayed in phase equilibria diagrams for primitive MORB formed at 8–11 kbar (BVSP, 1981, figs 3.3.24 and 3.3.35), primitive Icelandic picrite formed at 25 kbar (Maaløe & Jakobsson, 1980), primitive Hawaiian olivine tholeiite formed at 12 kbar (BVSP, 1981, fig. 3.3.25), tholeiitic CFB formed or equilibrated at 0·8 GPa (Thompson, 1972) and alkaline CFB formed or equilibrated at 1·6 GPa (BVSP, 1981, fig. 3.3.26). Real basalts, furthermore, know about these constraints. The typical phenocryst assemblages in erupted MORBs, Icelandic and Hawaiian tholeiites, the Skye Main Lava Series (Scarrow & Cox, 1995) and in non-picritic CFBs generally are olivine plus plagioclase sometimes accompanied by clinopyroxene. Pigeonite and orthopyroxene phenocrysts are rare (Jamieson, 1970b). A phase diagram for the forsterite–diopside–anorthite system (BVSP, 1981, fig. 3.3.20) indicates that substantial partial melting of an alumina-saturated olivine–pyroxene assemblage might be required at 0·7–2·0 GPa, producing melts far advanced up the olivine + pyroxene phase boundary, to generate a primary liquid which would display the low-pressure crystallization sequence of olivine followed by pyroxene. Terrestrial MORBs and CFBs are widely assumed to have been generated by 10% or greater mass fractions of partial melting. Yet this more advanced melting has either not proceeded far enough for the liquid products to precipitate olivine joined by pyroxene rather than plagioclase during low-pressure crystallization, or the melts have been modified towards more feldspathic compositions en route to the surface. The commonly observed sequence in lunar mare hand specimens is olivine joined next by calcium-poor pyroxene with or well in advance of plagioclase. Liquids with this type of low-pressure behaviour cannot be generated by small mass fractions of partial melting of originally plagioclase-saturated cumulates even at very high pressures within the Moon. Addition of some trapped melt to the equation exacerbates the problem.
  Clinopyroxene can be encountered as the second crystallizing silicate phase at low pressure in the initial partial melts of mantle assemblages saturated with garnet, when these are formed at pressures >4·5 GPa (Herzberg & O’Hara, 1998) but this is not an admissible solution for the Moon. Such liquids can also be generated at lower pressures by more advanced partial melting yielding harzburgite equilibria. The picritic eruptives of Baffin Island and Iceland (Clarke, 1970; Maaløe & Jakobsson, 1980) may have formed by partial melting of garnet-saturated lherzolite assemblages at 2·5–3·5 GPa and show the anticipated low-pressure silicate crystallization sequence of olivine joined by plagioclase before pyroxene. Other picrites which do display the low-pressure crystallization sequence olivine–clinopyroxene, joined only later by plagioclase, are voluminous among the early eruptives of continental flood basalt provinces (Karroo, Deccan, Siberian Traps), in the Solomon Islands (Stanton & Bell, 1969), at Mauna Loa and Mauna Kea, and also occur in basalt sequences related to a destructive margin in Kamchatka (Kamenetsky et al., 1995), but none of these are required to have formed with the restrictions regarding their source region which apply to the lunar mare hand specimens. All can have formed at much higher pressures and/or with much higher mass fractions of partial melting.
  The desired lunar ‘primary’ liquids might also be generated by more advanced partial melting of the cumulates at pressures of 0·5–2·5 GPa leaving residues of depleted harzburgites, but the required mass fractions of melting required for all the ‘primary’ liquids if this solution is adopted are likely to exceed those compatible with the trace element imposed requirement for small mass fractions of partial melting. Phase equilibria for postulated lunar mare primary magmas (BVSP, 1981, figs 3.4.3 and 3.4.5) exhibit temperature intervals between liquidus co-saturation with olivine and orthopyroxene and their subsequent saturation with spinel or plagioclase, of 50–100°C, which suggests that the liquids are far from the compositions of the initial primary melts of plagioclase-saturated cumulates under those conditions.
  Is the situation improved by allowing cumulates formed at one pressure to be remelted at another? If plagioclase-saturated (but plagioclase-free) magnesian harzburgites at pressures of close to 10 GPa plume upwards to lower pressures, alumina and lime dissolved in the orthopyroxene will tend to exsolve to yield actual crystals of plagioclase. The initial partial melts will remain saturated with plagioclase to higher mass fractions of partial melting in consequence. These initial partial melts will migrate in composition as the pressure falls so that they should display less plagioclase crystallization before pyroxene entry at low pressure, but most of the motion of the initial liquid compositions takes place during the last 0·2 GPa decrements of pressure (BVSP, 1981, fig. 3.3.18), i.e. within the thickness of even the thinnest parts of the lunar nearside crust, within the uppermost 5 km of the terrestrial crust and within the depth of some terrestrial shallow-seated layered gabbro complexes. There is no resolution of the problem here. Advanced partial melting into the harzburgite + liquid equilibria continues to have the advantages and disadvantages already noted but affords no resolution of the problem because the orientation of the locus of liquids in equilibrium with olivine + plagioclase + clinopyroxene is almost coplanar with the olivine + orthopyroxene control planes [e.g. O’Hara (1968a) and Fig. 4].
  If plagioclase-saturated (but plagioclase-free) ilmenite lherzolites, formed originally at pressures of close to 0·4 GPa, anchor downwards to pressures of 1·0 GPa and on to pressures of 2·5 GPa, then the initial liquids of their partial melting are no longer required to be plagioclase or spinel saturated on formation, but those alumina-saturated liquids towards which the initial liquids will tend are increasingly far displaced into the field where olivine joined by plagioclase before pyroxene will be the low-pressure crystallization sequence. A majority of Apollo 17 hand-specimen compositions show plagioclase entry virtually coincident with that of pyroxene at low pressure (O’Hara & Humphries, 1975) and clearly lie far from the high-pressure alumina-saturated cotectics, implying that substantial mass fractions of partial melting would be required to generate these compositions directly. Again, the effects of pressure changes on the phase equilibria do not offer a resolution of the basic problem. If these were true primary magmas formed in the manner postulated, they should display the crystallization sequence olivine joined by plagioclase followed after a significant interval by calcium-rich rather than calcium-poor pyroxene.
  Can anything be salvaged? The primary character of the major element composition might be maintained, and the trace element concentrations could be ascribed to assimilation of KREEPy material during transit to the surface. This relaxes the requirement for low mass fraction partial melts and so greatly eases the problem. Integrated melting (O’Hara, 1985, 1995a) can also combine trace element features suggestive of low mass fractions of partial melting with major element features suggestive of higher mass fractions of melting, but only at the expense of relatively high average mass fractions of melting. In the absence of a positive Eu anomaly in the average lunar highland crust, the requirement for plagioclase saturation in the magma ocean (and indeed the requirement for a magma ocean at all) can be abandoned and a harzburgitic mantle adopted (66, 67). The problem of the low-pressure crystallization behaviour is then solved at the expense of the primary character of the magmas. Plagioclase fractionation is required to impart the negative Eu anomaly (92) but is inadmissible unless the hand-specimen compositions represent portions of the erupted magmas which have become enriched in ferro-magnesian and oxide phenocrysts (103, 104) or have been translated in character by selective volatilization (84). Either solution spells sudden death for the primary magma hypothesis.
  
• 94. Primitive terrestrial MORB compositions are thought to have separated at moderate pressures by up to 10% average partial melting from plagioclase-undersaturated lherzolite and harzburgite residua in one popular view. In an alternative view they are thought to have separated from plagioclase-bearing cumulates at relatively low pressures. Whichever interpretation is adopted the compositions of these liquids have evolved under conditions where there was substantial ‘elbow-room’ towards compositions with high normative plagioclase content relative to pyroxene (93), allowing equilibration of relatively feldspathic partial melts with olivine and pyroxenes at moderate pressures. Just being relatively close to equilibrium with plagioclase at pressures of 0·2–1·0 GPa is sufficient to ensure that plagioclase will appear before pyroxene in the low-pressure crystallization sequence.
  
• 95. The anticipated sequence of olivine joined by plagioclase followed by clinopyroxene during low-pressure crystallization is pervasive in terrestrial MORB compositions and in many continental flood basalts.
  
• 96. None of the putative lunar primary magmas based on glass-bead or mafic hand-specimen compositions displays the olivine–plagioclase crystallization sequence at low pressure. In these compositions the crystallization sequence of liquidus olivine joined next by calcium-poor pyroxene is typical. There are in most cases extended temperature intervals at all pressures before the appearance of a third phase, which is usually Ca-rich clinopyroxene. Such liquid compositions, if they are indeed primary magmas, can only have formed from source rocks which were harzburgites or olivine pyroxenites.
  
• 97. The negative Eu anomaly cannot be inherited and must be generated by separation from plagioclase at low pressure (92). Choice of the hand-specimen or glass-bead compositions as the primary magmas leads to the impasse that plagioclase does not begin to precipitate until late in the crystallization sequence, whereas the anomaly is required to be well developed in the initial liquid composition. The hand specimens cannot represent the parental magmas.
  
• 98. If the lunar mantle has evolved from broadly chondritic material by extraction of a relatively small mass fraction of partial melt to form the lunar highlands and the mare basalts, the ultrabasic upper-mantle mineral assemblage probably still includes small amounts of an alumina-rich phase. This would be plagioclase at depths less than 200 km, spinel at depths between 200 km and 500 km, and garnet at greater depths. At the least the mineral assemblage is likely to consist of olivine and pyroxene crystalline solutions which are close to saturation with the alumina-rich phases. This still seems to be the case in the terrestrial mantle despite its having undergone more prolonged and extensive partial melting than the lunar mantle. In the ‘conventional’ lunar model, the outer parts of the mantle are required to have grown by accumulation of dense phases from an evolving magma ocean which was simultaneously crystallizing and losing plagioclase by flotation to form the early crust. Even if the extraction of plagioclase is perfect and complete in this process, and there is negligible trapped liquid, it would still be the case that the ferromagnesian assemblage would be plagioclase saturated. The first liquids to form if it were to be remelted at the pressure of its original formation would also be plagioclase saturated under those conditions.
  
• 99. The primary liquidus phase hypervolumes of different crystal species, i.e. the ranges of bulk compositions in multicomponent composition space from which a given crystal species will begin to crystallize first, are altered by changes in the pressure (O’Hara, 1968a; Herzberg & O’Hara, 1998). One of the consequences is that in any fixed basaltic or picritic bulk composition it is likely that the liquidus phase at low pressure (olivine) will be replaced as the pressure increases by one or more other phases, typically a pyroxene first and later by garnet at very high pressures. Co-saturation with olivine and a pyroxene somewhere along the liquidus as pressure increases is an almost inevitable consequence of starting with a basaltic composition and need have no great petrological significance with regard to origin of that bulk composition. This property is in no way diagnostic of primary character and a very wide range of randomly chosen compositions will display this type of materials behaviour. The data reported in table 6.5 of Taylor et al. (1991) and many of the entries in table 9 of Papike et al. (1998), for example, have little petrogenetic significance.
  The probability of primary character is greatly enhanced when it is found that the liquid displays simultaneous co-saturation at the liquidus for several silicate phases, e.g. olivine, calcium-poor pyroxene, calcium-rich pyroxene and garnet at upper-mantle pressures, as in the case of Icelandic picrite liquids. The probability of gabbro fractionation is greatly enhanced when the liquid displays simultaneous co-saturation with olivine, plagioclase and calcium-rich clinopyroxene at deep crustal pressures as in the case of Snake River flood basalt (Thompson, 1972). None of the putative lunar primary magmas displays the simultaneous co-saturation at high pressure which would be expected if compositions were those of liquids generated by either primary partial melting or subsequent deep-seated differentiation, especially if mass fractions of melting are constrained to be small by the incompatible trace element behaviour.
  
• 100. The adiabatic gradient within the probable materials composing the lunar mantle will be lower than that in the Earth’s mantle, as a result of the reduced gravitational acceleration. The gradient of the solidus temperature of the postulated cumulate pile (liquidus temperature of the postulated residual liquid/postulated initial remelt liquid) will greatly exceed this value, compounded by the additional decline in solidus temperature with height associated with the evolving compositions of liquid and cumulate. If a global magma ocean ever existed, its cumulate pile should have been subject to thermally driven convective motion from the moment of deposition. Ignoring this reservation, the pressures and temperatures of co-saturation of olivine and pyroxene in the postulated lunar mare primary magmas have been used to obtain<