The decline and renaissance over the past 50 years of the concept that assimilation of crustal rocks may play a significant role in basalt genesis is reviewed and attention is drawn to the very restricted sampling of the potentially contaminated lavas which is available for most oceanic basalts. A simplified quantitative model of the behaviour of ideally behaved trace elements during the evolution of a hypothetical oceanic island volcano is explored and qualitative consideration given to the thermal and isotopic implications of this model. Three prominent features of the model are the importance of thermal contamination, which may be the clearest indicator of extensive country rock assimilation; the robustness of incompatible trace element signals once these elements have been contributed to the magma chamber; and the apparent decoupling of output lava composition from the concurrent input parental liquid composition, such that no closed system process can even approximately reproduce the relationships.
Conceptual importance of assimilation in magma genesis
Assimilation, contamination and hybridization in the evolution of magmas were important concepts in petrogenesis before 1950. This was partly in response to the ‘space problem’ posed by large, predominantly calc-alkaline, plutons which fuelled the ‘granite controversy’, and partly in response to field and petrographic relationships observed at the margins of plutonic bodies, which provided evidence of the operation of these processes. When the focus of attention switched to basalts (Tilley, 1950), the implications of the space problem were largely ignored. The problem was less obtrusive and the uniformity and distinctive geochemical character of successive flows could be held to preclude extensive assimilation and contamination (Carmichael et al., 1974).
Over the past 50 years, the rise, fall and renaissance of appreciation of the importance of assimilation and contamination by the petrological community are chronicled by the weight given or withheld in a succession of influential texts. Initially this appreciation was largely in the context of the plutonic calc-alkaline and alkaline rocks, and was driven by the field relationships. It has only recently become extended first to continental, and now to oceanic basalts, where the field evidence may seem to be far less compelling.
For example, Bowen', (1928) text assigned the whole of chapter 10 to a consideration of the possible role of assimilation and contamination in magma genesis. The 50th anniversary update of Bowen's text (Yoder, 1979) was constrained to a similar format yet the petrogenesis section of chapter 17 makes no reference to assimilation or contamination. The companion volume (Hargraves, 1980) containing the modern developments made only passing reference to assimilation and contamination in the section on trace element geochemistry. The standard text on igneous petrogenesis for the 1950s and 1960s (Turner & Verhoogen, 1951, 1960) gave substantial treatment to the issues of assimilation and contamination, but by the time of its major revision (Carmichael et al., 1974) this treatment had shrunk to a mention covering barely more than one page of the greatly expanded section on igneous rocks. A major treatise on rocks of basaltic composition (Hess, 1968) made no mention of the topic in two volumes. It received no mention in Ringwood's, (1975) text except in the context of wall-rock reaction within the upper mantle, and only a minor mention in Yoder's (1976) book on basalt genesis. A major geochemical symposium volume (Ahrens, 1977) made only two mentions of the topic, both in relation to plutonic rocks. Mason's, (1966) text on geochemistry and the text by (Ehlers & Blatt, 1982) made no mention of the topic. A recent compendium on oceanic basalts (Floyd, 1991) made no mention of assimilation or contamination. In the context of the petrogenesis of lunar rocks, Taylor (1975) made no mention of assimilation and contamination and only very minor mention of the topic in a wider context (Taylor, 1982). Yet Rb-Sr data for some mare basalts yield ‘ancient’ whole-rock isochrons, providing a strong case for contemplating significant contamination of the trace element and isotopic signals by assimilation of KREEP (Basaltic Volcanism Study Project, 1981, pp. 961–963).
Cox et al. (1979), however, gave substantial treatment to assimilation and contamination and, among more recent texts, Middlemost (1985) made substantial mention of assimilation, but nowhere in the context of basalt petrogenesis. Hall (1987) gave the topic significantly greater attention and Wilson (1989) presented a broad treatment, extending to continental flood basalts and briefly to oceanic islands, but not to back-arc basins or mid-ocean ridges. Summarizing then, assimilation of crustal rocks was not widely entertained as an important factor in basalt petrogenesis between 1950 and 1980. However, the space problem never went away and it is once again fashionable to contemplate a significant role for assimilation, even in the evolution of basalts.
Solutions to the space problem
Simply stated, rocks which are identified as plutonic or hypabyssal igneous products frequently occupy space apparently occupied previously by crust or earlier eruptives. Where has that country rock gone? The larger the body, the larger looms the problem, but in geochemical terms the problem is potentially just as acute whatever the size of the body. Solutions involve either displacement or replacement of the country rock or some combination of these.
Displacement of country rock may have been by uplift, by lateral compression, by lateral extension including slumping (which is admissible but not necessarily the whole answer in the cases of oceanic islands and spreading centres), by excavation in the case of large impact basins, by subsidence or by piecemeal foundering of the roof (stoping). Replacement, discussed further below, does not require either the transformation of pre-existing country rock by the now discredited mysterious ichors or emanations (see Holmes, 1965, pp. 182–185), nor the simple linear addition of country rock to parent magma to create the observed compositions (McBirney, 1979).
Assimilation as a factor in basalt petrogenesis—better assimi-late than never
A suggestion that isotopic ratios of basalts might change during magma chamber processes (O'Hara, 1973, 1975) developed into a model of the refilled, fractionated advancing magma chamber. This explicitly reintroduced the space problem into the basalt debate and overturned the logic of some of the arguments previously adduced against significant assimilation in basalt petrogenesis. It recognized the hazards of extensive assimilation of crust and earlier eruptives (O'Hara, 1977, 1980; O'Hara & Mathews, 1981) in basaltic rocks, even in the generation of oceanic crust at spreading axes. These proposals were coincident with the development of the model of simultaneous fractionation and assimilation (DePaolo, 1981) chiefly in the context of calc-alkaline volcanics.
Roof assimilation by an advancing magma chamber redistributes the added compatible components into the cumulates, possibly producing some bias in the precipitated mineral proportions relative to those expected from cotectic crystallization within a closed magma chamber. It provides for homogenization of the contaminated liquids and smooths out the effects of short time-scale variation in the nature and amount of the contaminant. It redistributes the incompatible components into the remaining liquid in the magma chamber where they may have long residence times, and into the lavas. Incompatible trace element contents and ratios in that evolving magma may depart from the constraints imposed by any closed system partial crystallization model. Isotopic ratios even in the least altered, most homogeneous erupted lavas may reflect contamination.
Superheating with respect to the low-pressure liquidus would be anticipated in volatile-poor parental magmas rising by simple isentropic decompression from upper-mantle source regions. This superheat is apparently not present—erupted lavas are generally saturated with two or more crystal species. The thermal energy required for assimilation may in part account for the missing superheat. This energy demand drives cooling and partial crystallization of the magma, whose major element chemistry and phase equilibria are thereby forced to conform to the requirements of low-pressure phase equilibria. In principle, a given mass of parental magma can yield by its own cooling, first to the low-pressure liquidus and then to the solidus by crystallization, enough heat to totally remelt (i.e. comprehensively assimilate) a greater mass of roof rock (which in this model is largely a residual liquid from partial crystallization of the parent magma). In practice, much less can be achieved because of the heat losses from the system by conduction and convection—but even very much less assimilation can nevertheless amount to significant contamination in geochemical terms. It may yet transpire that the persistent occurrence of low-pressure cotectic character in erupted basalts, when considered in relation to the relatively short times available for heat transfer out of the magma chambers, will provide the most compelling argument in favour of large-scale assimilation. This proposition is difficult to test in the terrestrial environment because of the difficulties of quantifying the major role played by hydrothermal cooling. It may be easier to test eventually in the water-poor environments of Mars and Venus if it transpires that their central volcanic structures have also persistently erupted low-pressure cotectic magmas.
In the absence of some other acceptable explanation of the textural differences, the simple fact of the juxtaposition of slowly cooled, coarse-grained plutonic rocks against faster cooled, fine-grained hypabyssal and eruptive rocks at the roof of the magma chamber may be compelling evidence of a change in the thermal environment as a result of the advance of the magma chamber roof.
How little do we know about ocean island volcanoes?
The volume of a uniform angle circular cone is proportional to the cube of its height. If the Hawaiian volcanoes are regarded as uniform cones with three-tenths of their height exposed above sea-level, then the total volume above sea-level represents less than 3% of the total erupted volume, much of it currently inaccessible to direct sampling. Even if half the height were above sea-level, this represents only 12.5% of the volume. The volume of Loihi represents much less than 1% of the speculative volcano ultimately to be built upon its site. The question of how deep one has to cut or drill into the edifice to recover earlier eruptives is addressed later.
Two thoughts follow from this consideration of relative volumes in conical structures. We know chasteningly little from direct observation about the geochemistry and field relationships within the central 95% of the total erupted volume in such oceanic volcanoes. Geochemical speculation is comparatively unconstrained. Geophysical data indicate the presence of very large volumes of consolidated magma chamber products, which must have formed at the expense of space previously occupied by oceanic mantle or crust, or by previously erupted lavas, within such volcanoes (Watts et al., 1985; Carress et al., 1995). Low-pressure cotectic character and other petrochemical characteristics of the erupted lavas (Jamieson, 1970) point to the extensive modification of magma compositions in high-level magma chambers or conduits. Even the early stages of erosion or shallow drilling of these structures exposes the contents of high-level magma chambers (MacDonald, 1949; McBirney & Aoki, 1968; Upton & Wadsworth, 1972; Upton, 1982; Augé et al., 1989; Rançon et al., 1989) and xenolithic suites abound in plutonic gabbroic products which have probably crystallized within the superstructure (Jackson, 1968). The active magma conduits (Ryan et al., 1981) clearly occupy space previously filled in succession by air or water, then by basalt, and finally by intrusives. Replacement of the space previously occupied by erupted lavas by the advancing magma chambers has, therefore, proceeded into the last 5%, probably into the final 1% by volume of the eruptive history and must be assumed to have influenced the whole of the volcanic edifice until proven otherwise.
Unless the establishment of the magma chambers can be demonstrated to have taken place entirely by shouldering aside the previously erupted lavas there is a space problem to be addressed (and shouldering aside makes no contribution to solving the thermal problem). Some solutions to that space problem introduce a high probability that magmatic evolution involved substantial assimilation (refluxing) of previously erupted lavas and intercalated sediments, some of which may have been altered by contact with seawater or fresh water, either by weathering or through hydrothermal activity.
How little do we know about mid-ocean ridge volcanoes?
The evidence for continuing high-level magma chamber processes, with their inherent potential for contamination and assimilation, was explicit in the ocean-wide identification of a layer of gabbro overlying the Moho (e.g. Christiansen & Salisbury, 1975) and in early petrological studies of mid-ocean ridge basalt (MORB) (Muir & Tilley, 1964). It was implicit in the recognition of low-pressure cotectic character in such rocks (O'Hara, 1968). If the roof of that gabbro chamber anywhere invades the overlying dolerites and basalts, there is a space problem to be addressed here also (O'Hara & Mathews, 1981). We know depressingly little about the field relationships of plutonic and extrusive rocks in this environment.
Modelling an Oceanic Island Volcano
Outline of the model
This study attempts to model what might happen during the establishment of a central volcano within an oceanic basin, paying due attention to the space problem. Figure 1 indicates the initial and advanced stages envisaged in the model. Recent oxygen isotope data (Eiler et al., 1996) provide some justification for the extensive assimilation and replacement of (altered) oceanic crust which is postulated here.
The generalized model involves feeding parental magma batches into a chamber under circumstances which guarantee the advance, growth and eventual decline of a magma chamber. Partial crystallization of the magma in each cycle increases throughout. Lava escape decreases throughout. Assimilation of substantial amounts of roof, proportional to the superheat provided by the parent magma inputs plus the heat released by partial crystallization, creates the space for the advancing magma chamber. In its present form this is a very simplified model, but even so a large number of assumptions, summarized below, have to be made to construct a working simulation. The ultimate definitive statement of the formulation of the general model and the explicit and implicit assumptions made in each specific model is, however, contained in the Mathematica code (Wolfram Research, 1994) used for the simulations (available on request from the author). Almost all of the assumptions are potentially contentious. It is here emphasized, however, that exploration of alternative assumptions in many cases has indicated that it is the detail, not the substance, of the results which is affected.
Dominating general factors in all the specific models examined and in the results presented are as follows:
The bulk of the mass of the lava-pile is erupted during the first half of the sequence.
The magma chamber grows in size in the first half of the sequence, with low crystallization to eruption ratios and low ratios of roof mass consumed to parent magma mass in the total input to each cycle of the magma chamber.
The magma chamber shrinks in the second half of the sequence, with increasing ratios of crystallization to eruption and increasing ratios of roof to parent magma in the gross input to each cycle of the magma chamber.
The geochemical ‘fertility’ (in terms of incompatible trace elements) of what is being assimilated at the roof of the magma chamber increases progressively. In the final stages this increase may be very rapid if the magma chamber roof has advanced far enough to invade relatively late-stage lavas.
Incompatible elements in the magma chamber have relatively long residence times, which will give any concentration or isotopic signal in the incompatible trace element inputs a conspicuous longevity in the lava output.
Compatible elements in the magma chamber have relatively short residence times, which yields the opposite effect (despite the choice of a specific partial crystallization model which maximizes the retention of compatible elements in the magma).
Relationship between volcano cross-section, density, product thickness and mass
All materials involved (mantle, magma, lava and cumulate) are assumed, arbitrarily and solely in the interests of simplicity, to have identical density. The cross-section of the volcanic products is assumed uniform throughout its vertical extent except where lava spreadout has occurred at the surface. Consequently, mass of any volcanic product is proportional to its thickness, modified where necessary by the lava spreadout factor.
Initial mantle and crust
The source mantle is assumed to have unit concentration of every element relative to some normalizing standard, which is obviously not chondritic but might be taken as chondritic for all non-volatile, non-siderophile elements. All results obtained are in effect normalized to the source mantle composition. [Figures 3 and 4 below have obvious and intentional similarities to the primitive mantle normalized trace element variation diagrams or ‘spidergrams’ used in studies of basalt geochemistry by Thompson et al. (1984) and by Sun & McDonough (1989) for example.]
The assumptions made about crust composition are important if substantial roof consumption is permitted. The assimilated roof may be a major source of incompatible elements. Here a crude ‘oceanic’ crust has been modelled by assuming that its bulk composition is that of a 10% equilibrium partial melt of its source region, but that it has separated into a lower crust which is the solid in equilibrium with the residual liquid which then forms the upper crust, after 50% of the original partial melt has solidified. This is intended to model a crust in which the lower part is dominated by cumulates and the upper part by residual liquids conjugate with those cumulates. A 2% equilibrium partial melting depletion of the source mantle before formation of the ‘crust’ has been arbitrarily assumed for the models presented here, sufficient to guarantee a degree of relative depletion of highly incompatible elements throughout the whole of the oceanic crust so formed. This prior depletion has not been applied to the source of the parent liquids for the later central volcano which penetrates that crust.
Partial melting process
A wide variety of models might be assumed, but it has been shown elsewhere that many of these yield ideal trace element results not vastly different from an equilibrium batch melt formed in a simple system with uniform mass fraction of melting. This latter is assumed in many of the models investigated. As an extreme alternative, some models have assumed that the source region is progressively fractionally depleted by the melt extraction process. It has been assumed that 0.5% trapped melt is present during all partial melting processes. These assumptions evade altogether the possible complications arising from integrated partial melting when the process is markedly non-modal and involves exchange with small amounts of scarce minerals (e.g. zircon, monazite) in which selected, normally highly incompatible, elements are in fact highly compatible.
Parental magma inputs
Exactly 1000 cycles (0–999) of input, lava ejection, roof consumption and precipitation are assumed. Implicitly it is envisaged that these are evenly distributed in time through the life of the volcano. Three cases which have been modelled assume that the size of input (1) has a uniform magnitude (0.5 unit mass) throughout; or (2) decreases linearly with the number of cycles from unit mass to zero; or (3) increases from very small to a maximum value of 0.75 units in the middle of the evolution and wanes to zero at the end. Each of these assumptions ensures that the total parental magma input is 500 unit masses.
Cooling and precipitation
Cooling is assumed to increase towards the surface. A linear increase in the mass crystallized in each cycle from zero in the zeroth cycle to some maximum mass in the final (999th) cycle has been assumed. This maximum is related to the magnitude of the unit mass of input parental magma via a simple factor, constant during each simulation but user variable. The model of partial crystallization process assumed here is one of small packet crystallization with simple equilibrium partial crystallization (SPC-EPC, O'Hara & Fry, 1996) and 0.5 mass fraction crystallization of each small packet (this yields the same result for ideally behaved trace elements as small packet crystallization with integrated perfect fractional crystallization under these conditions, but could yield very different results when the process involves integrating non-modal crystallization products with the appearance of carrier-phases very rich in selected trace elements at low mass fractions of residual melt). This choice, given the periodically recharged, periodically tapped and continuously fractionated nature of the (non-steady-state) magma chamber, provides for excellent maintenance of relatively high concentrations of highly compatible elements in the residual liquids.
The mass of roof which is consumed in each cycle has been set at some fixed fraction (the assimilation factor) of the sum of the masses of the new input magma in each cycle and the mass crystallized in each cycle, these being the principal sources of the heat required for the assimilation process. It is assumed that the contaminated material is well and rapidly mixed with the rest of the liquid in the magma chamber in each cycle. More sophisticated assumptions can readily be envisaged.
Uplift, sinkage and spreadout
In all models the net effect is the transfer of 500 mass units of partial melt across the Petrological Moho. Finally, the contents of the lava-pile (minus what is eaten by the advancing magma chamber) plus the contents of the solidified magma chamber have to be equivalent to the mass derived from the source region (500 units) plus however much of the original crust has been consumed (240 units if the magma chamber advances through the whole of the crust without any displacement of pre-existing crust, which it does in many of this particular group of models). In one set of models the magnitude of what is erupted is arbitrarily set, and part of the space problem is resolved by allowing the cumulate pile and even the base of the magma chamber to sink below the original position of the Moho (the requirement that the cumulates retain a right cylindrical form can be relaxed). Part of the space problem is of course resolved by eruption. In another set of models, all the products of the process are required to end up above the original position of the Moho and the mass erupted in each cycle is controlled by the requirement that all space adjustments are made within the crust and superstructure, but without any lateral or vertical displacement of country rock or lava-pile. When mass is erupted from the magma chamber, it is assumed to spread out as a lava, so that the mass directly overlying the advancing magma chamber is less by some factor reflecting the relative areas than the actual mass erupted. This spreadout factor is assumed to be large during the early eruptions and to decline linearly as the number of cycles increases.
The mass of roof assimilated, the mass of cumulates formed and the mass of liquid erupted are assumed to be in some way related to the magnitude of the parental magma input in each cycle. Some degree of flexibility in the calculations has been retained by using a simple fudge factor to specify what fraction or multiple of the input mass is used in these relationships. Parameters are adjusted so that the roof of the magma chamber never emerges at the surface (i.e. the mass assimilated is never greater than the sum of the crust plus the mass of the overlying lava-pile); so that at no stage the magma chamber shrinks to zero and then is reborn; and so that the mass required to be erupted at the start of a given cycle is never greater than the mass of residual liquid available for eruption at that moment. In general, parameters were chosen so that the size of the magma chamber increases steadily during the early cycles, passes through a maximum and then declines. Depending upon the choice of the various parameters, the magma chamber size may decrease to zero before the end of the 1000 cycles (in which case the remaining inputs of parental liquids might be supposed to erupt directly onto the surface as ‘primary’ magmas), or it may still have a finite size at the end of the inputs (in which case the remaining liquid might be supposed to crystallize by a closed system process with little further eruption). Particular attention has been paid to the combinations of parameters which lead to a final magma chamber size which is non-zero but small relative to the maximum size attained during the evolution.
Thus far no specific treatment of isotopic contamination has been attempted. For elements with bulk distribution coefficients close to unity, such as oxygen, the bulk contamination will be controlled by the ratio of material assimilated to parental liquid input; in general, this will increase markedly in the later stages of the evolution. Highly compatible elements, which have relatively short residence times in the magma chamber, might display greater effects—only sulphur and osmium of the elements with isotopes which are commonly determined might behave in this way. Highly incompatible elements which have long residence times in the magma chamber may preserve the signatures of their source regions more successfully, i.e. the effects for oxygen may be somewhat decoupled from those for Rb, Sr, REE, etc.
In both models for which further results are presented here, the mass of magma input in each cycle declines linearly from unity to zero. Lower- and upper-crustal thicknesses were set at 120 mass units each and the lava spreadout factor was allowed to decrease linearly from four to 0.5. The assimilation factor was set at 0.33 (two-thirds of the energy potentially available for roof consumption escapes by conduction or convection) and the relative mass factor was set at 0.99. In model A the input magma composition is constant and assumed to be a 0.1 mass fraction equilibrium partial melt of an undepleted source region. In model B the input is an imperfect fractional melting product (1000 equilibrium decrements to remove a total of 0.1 mass fraction of the original source) with 0.5% trapped melt held in the residue at each step [IFM-ISFM model of O'Hara, (1993)]. Part of the space problem is resolved by allowing cumulates to subside below the original position of the Moho.
Presentation and discussion of specific results
Figure 2 displays curves of the evolution of various aspects of the volcano as functions of the number of cycles of magma input, crystallization, assimilation and eruption for model A. Thicknesses are expressed in mass units following the simplifying assumption of uniform density as explained in the text. Each mass unit is approximately equivalent to 35 m in the oceanic situation with a volcanic superstructure rising about 3500 m above the original ocean floor. The upper figure plots the evolution of the thicknesses of the total cumulate pile and of the magma chamber itself (heavy curves), of the lid of unassimilated original crust and previously erupted lavas above the magma chamber (broken curve), the sum total of these three which represents the total thickness from the base of the cumulates to the top of the volcano, and the total maximum lava thickness generated before any reassimilation is considered (light curves). The lower figure plots the evolution of the positions of the top of the volcano, of the base of the cumulates assuming subsidence as a simple piston (light curves), and of the floor and roof of the magma chamber (heavy curves), as thicknesses relative to the original Moho (horizontal labelled axis) and original surface (broken line). Results for model B differ only in minor detail from the above. The roof of the magma chamber enters the more ‘fertile’ upper crust (i.e. the upper 50% of the oceanic crust, as defined above in the assumptions used in these models) after ∼380 cycles and the lava-pile after ∼760 cycles in both models.
Figure 3 plots the predicted concentrations of ideally behaved trace elements in the lavas as a function of the logarithm of their effective crystal–liquid partition coefficients (taking into account the effect of the assumed mass fraction of trapped melt in the partial melting process). Concentrations are displayed as a function of the number of cycles. These concentrations are shown on a linear scale of relative concentration, and are in effect normalized to the concentrations of the same elements in the source mantle. Both models show extreme enrichment of the incompatible trace elements in the later stages when the ratios of crystallization to lava escape, and of assimilation to new magma input, are both high, and incompatible element concentrations in the contaminant are also high. The spike of incompatible elements in the early cycles of model B is a product of the effects of the early stages of fractional partial melting of the source assumed in that model. No spike is present in the early stages of model A but incompatible element concentrations are uniformly higher throughout the middle stages. The reversed discrimination (relative concentrations of highly incompatible elements less than those of moderately incompatible trace elements) of the more incompatible elements, which characterizes most of the model B products, is present but barely detectable in the middle cycles of the model A products. Model B provides an alternative mechanism for persistent and large-scale production of lavas which display reversed discrimination combined with high relative concentrations of the incompatible elements additional to those mechanisms involving only partial melting (O'Hara, 1995).
Figure 4a plots what are in effect sections across Fig. 3a after selected numbers of cycles, together with data for the assumed compositions of the lower and upper crust, and the composition (constant throughout all 1000 cycles in this model) of the initial parental magma input to the magma chamber, all normalized to the composition of the source mantle. Results are here plotted on a logarithmic scale for the relative concentration. Late stage products display incompatible trace element concentrations which would, in a simple world, point to 1% to 0.1% partial melting of the true source, despite the fact that the input magma remains constant as a 10% partial melt of that source. Figure 4b is a similar plot for the dramatically evolving compositions of the fresh magma inputs in model B, which display strong reversed discrimination from cycle 100 onwards and severe incompatible element depletion from about cycle 250 onwards. The curves for input magma composition in the first and 500th cycles are reproduced in Fig. 4c, together with the assumed lower- and upper-crust compositions and the compositions predicted for the lava compositions in the 100th, 400th, 700th and 1000th cycles. Features to note are the strong reversed discrimination in the middle cycles and the remarkable robustness of the incompatible element signal throughout despite the rapid depletion of these elements in the input. These concentrations are sustained principally by the long residence times of these elements in the first half of the sequence and are boosted by the contributions from the assimilated materials in the later stages. We observe in both Fig. 3 and Fig. 4 the similarity in the late stage products (those most easily observed in ocean island volcanoes) between the predicted geochemistry of lavas derived from very different partial melting models.
Figure 5a illustrates the contrast that develops between the composition of the lava that erupts from the magma chamber and the composition of the magma currently being fed into the chamber in the case of model A; i.e. it presents the variation in the compositions of the output lavas normalized to the composition of the input liquid. Exactly the same figure would be produced by normalizing to the input liquid from any cycle because the input liquid composition in model A is always constant. Figure 5b and Fig. 5c present similar plots for the lava output in model B, normalized to the current magma input and to the initial magma input, respectively.
Figure 5a displays the influence of contamination by the relatively impoverished lower crust (as defined in the assumptions relating to the models), which reduces the incompatible elements and enhances compatible elements in the gross inputs, but the effects are small. Simultaneously, there is a relative decline in the compatible elements, reflecting the more evolved nature of the contaminant. Once the upper, more fertile, oceanic crust is being assimilated it imparts to the incompatible element signal a minor but steadily increasing relative enrichment of incompatible elements and slight reversed discrimination. Enrichment continues and is much enhanced once the lava-pile starts to be consumed. Here the magma chamber both shrinks and undergoes more partial crystallization relative to lava escape, leading to the virtual elimination of the reversed discrimination among highly incompatible elements. The relative reduction in compatible elements, which is enhanced by the increased ratio of crystallization to escape in the later cycles, is reversed for an interval when the magma chamber roof enters the base of the lava-pile. Here lavas from the earliest cycles, which are relatively compatible element enriched, form the contaminant. Normal service is soon resumed.
Figure 5b displays effects among the compatible elements clearly related to those in Fig. 5a, but is dominated by the huge relative enrichment of the incompatible elements, which decline rapidly in the current input compositions used for normalization (Fig. 4b). Reversed discrimination is not apparent because the liquids to which the results are normalized themselves display such extreme reversed discrimination. Normalizing to the initial input liquid, which shows no reversed discrimination and which is close in composition to an initial partial melt of the source material, yields the results shown in Fig. 5c. The relative concentration of incompatible elements falls very rapidly and remains low until assimilation of the incompatible-rich lavas at the base of the pile commences. Reversed discrimination is strong throughout.
If the erupted lavas in each cycle were required to be evolved directly from the actual parental magma batch then rising from the mantle, the geochemical changes from the input liquid which would be required are those displayed in Fig. 5a and b. There is no closed system crystallization process yet postulated which could yield the required high enrichment and sometimes negative discrimination of incompatible elements combined with minimal depletion of and negligible discrimination between the highly compatible elements.
Magma evolution with assimilation: the thermal, trace element and isotopic signals
Given the model for the establishment of a central oceanic island volcano through the ocean floor, conveyed through the figures, some general conclusions may be reached from the analysis of the problem carried out so far.
Most lava erupts in the early history of the volcano, when the ratio of assimilated material to fresh input magma is at its lowest. The geochemical impact of the contamination will also be minimal because of the nature of the material assimilated at this stage. The principal evidence for assimilation in this part of the eruptive sequence may be thermal, identified through major element, experimental and petrographic studies and manifested as persistent low-pressure cotectic character in the erupted basalts, which has been achieved despite the possible difficulties of extracting heat from the relatively well-insulated magma chamber by either conduction or hydrothermal circulation. Preliminary calculations suggest that a requirement for significant thermal contamination will be dependent upon a high rate of eruption during the formation of the main bulk of the superstructure—uniform growth over the possible total life-span of an island volcano may not require much thermal contamination.
The majority of readily accessible lavas in a well-established oceanic island volcano represent relatively late stages in the history of the volcano, when there may be significant geochemical and isotopic effects caused by contamination. The shrinkage of the magma chamber renders it more vulnerable to contamination by relatively small amounts of previously erupted basalt. The details of the incompatible element chemistry in this region can be expected to be very variable and critically dependent upon how far into the eruptive pile the chamber roof has advanced. Marked inter-island and inter-volcano variability is predicted in the very late stage products of extinct volcanoes.
Isotopic contamination has not as yet been specifically modelled but some general conclusions may be reached by inspection. The majority of the lava output is erupted during the first half of the sequence and will come from a chamber which is not assimilating materials of particularly distinctive isotopic or geochemical character (especially where the oceanic crust penetrated by the volcano has been formed a relatively short time previously and the crustal material being assimilated is the less hydrothermally altered, less fertile part of the section, with negligible chance of involvement of chemical sediments). The lavas come from a chamber in which the ratio of uncontaminated parent liquid input to assimilated contaminant is high. Geochemical evidence for roof assimilation will be minimized.
The maximum input of ‘light’ oxygen and radiogenic strontium is likely to come from the hydrothermally altered upper crust and lower part of the volcanic pile. If gabbro is the main product of the magma chamber by this stage, both O and Sr will have relatively short residence times in the magma; if not, the strontium may behave as a more incompatible element and its signal may become separated from that of the oxygen. Any distinctive isotopic signal from assimilation of ocean-floor sediments and altered oceanic crust may be maximized in the later middle stages of the history of the volcano, in lavas which may not be well represented by readily accessible samples (but if a 10 km high edifice is built as a series of similar cones, with the given assumption regarding decrease in the mass of erupted material in each cycle, then a conical surface only 2 km deep into the edifice is composed of material erupted when ∼50% of the total lava mass has erupted, i.e. when considerably fewer than half of the cycles have been completed).
Isotopic contamination of highly incompatible elements will have long-term consequences, which in model A will be maximized in the latest stages, when the chamber may be refluxing incompatible element-rich eruptives which already carry an isotopic signal derived from ocean floor sediments and altered crust. Moreover, this refluxed material will be input at high ratios relative to the mass of parental magma input in each cycle. In model B, however, the maximum isotopic contamination of highly incompatible elements is likely to occur as altered upper-crust and sea-floor sediment is assimilated (i.e. around cycles 700–760) and this will thereafter be diluted by the assimilation of an uncontaminated isotope signal from the earliest, incompatible element rich lavas.
The author is of the firm opinion that attempts to apply conceptual and theoretical models are best left to, or carried out in conjunction with, the ‘owners’ of the natural data sets, who best understand the field relationships and the limitations of the sampling and analysis. Numerous additional points would need to be discussed, first among which is the choice of trapped melt fraction in the partial melting and partial crystallization processes. This dramatically affects the behaviour of the incompatible trace elements, because no element can have an effective bulk distribution coefficient less than the numerical value of the assumed mass fraction of trapped melt. The lower- and upper-crust compositions as well as the input parental liquids used in the above models will consequently display no positive discrimination of relative concentration among elements whose crystal–liquid distribution coefficients are significantly smaller than 0.005 (but there is growing negative discrimination combined with very low mass contributions in the inputs in model B). There will be markedly subdued discrimination among elements with true crystal–liquid distribution coefficients of less than 0.01. It is tempting to assert that the extent of discrimination between elements with very low crystal–liquid distribution coefficients sets some upper limit to the permissible mass fraction of trapped liquid in the partial melting and crystallization processes, but this is only true if ideal trace element behaviour can be assumed at all mass fractions of melt which contribute to the eventual erupted products in integrated melting regimes, or in small packet crystallization with integrated partial crystallization. Additional complications in the modelling would arise if the supply of source mantle or melt production from that mantle was episodic, as it would be if the source material was ascending as a string of isolated hot blobs rather than as a continuously flowing stream.
The results presented above show that with the partial crystallization process assumed, the lack of discrimination inherited from the sources survives little changed even into the late eruptive products. In these models provision was made for ‘real’ equilibrium crystal–liquid distribution coefficients to be used in the evaluation of the partial crystallization processes. However, the assumed partial crystallization model (small packet crystallization) and the chosen parameters have the effect of making the effective distribution coefficient almost double the crystal–liquid distribution coefficient for elements with d between 0.001 and 0.01 [O'Hara & Fry, 1996,fig. 5b and appendix, equation (7)] and the effect becomes greater as the amount of crystallization in each packet increases. Other choices of trapped melt fraction, crystallization parameter and crystallization process lead to significantly different results in the detail of highly incompatible element concentrations.
Entertaining all of these potential variations in the models, explaining and justifying the choices made, and arguing the selection of compositions for the oceanic crust and the unseen 95% of the eruptive sequence even for a single oceanic island data set is beyond the scope of this paper.
Figures 2–5 were prepared and all calculations were carried out in Mathematica (Wolfram Research, 1994). I am indebted to G. Bergantz, B. G. J. Upton, M. Wilson and another unnamed reviewer for suggestions which have significantly improved this paper. This study was largely funded by Cardiff University, and conference attendance was supported by the Royal Society and Cardiff University.