A Comment on ‘Postcumulus Processes Recorded in Whole Rock Geochemistry: A Case Study from the Mirabela Layered Intrusion, Brazil’ by S.J. Barnes and M. Williams

Abstract

In their analysis of whole-rock data from the ultramafic section of the Mirabela Layered Intrusion, Barnes and Williams suggest that the rocks formed as a closed system of crystals + trapped liquid and that compaction played essentially no role in the formation of the rocks. Here it is shown that their stratigraphic trends are fully consistent with a system undergoing concurrent crystallization and compaction (CC&C). Specifically, phosphorus decreases upward owing to more time for compaction prior to apatite saturation away from the cooler margins. In contrast, sodium and strontium (mainly incorporated in interstitial plagioclase) increases upward owing to the initial interstitial liquid becoming more enriched in the plagioclase components over the crystallization of the ultramafic section, causing interstitial plagioclase to crystallize earlier in the compaction history up-section. Local coincident peaks in P, Na, and Sr, typically related to changes in rock type, can be ascribed to porosity variations that can develop during compaction of a layered, heterogenous mush. These and other observations lead to a conclusion contrary to that of Barnes and Williams: the Mirabela intrusion is instead an excellent case study for the consequence of CC&C in layered intrusions and potential role of non-conventional fractionation mechanism of the parent magma.

INTRODUCTION

In their recent work, Barnes and Williams (2024) used data from four drill cores through the ultramafic section of the Mirabela Layered Intrusion to test a hypothesis first suggested by Meurer & Boudreau (1996, 1998). In this earlier work, Meurer and Boudreau suggested that in any crystal mush undergoing concurrent crystallization and compaction (CC&C) the idea that the mush can be considered a closed system of trapped liquid and ‘cumulus’ crystals is not valid. The estimates of any residual trapped liquid fraction based on the concentration of nominally incompatible elements will depend on both that amount of liquid remaining at any point during compaction and when the main host minerals for any particular incompatible element begin to crystallize. In general, the earlier a host mineral for any element incompatible with the original solids crystallizes, the higher will be the estimate of a residual liquid fraction.

For example, consider an ultramafic crystal + liquid mush consisting of liquid and orthopyroxene ± olivine. The expected crystallization order of minerals crystallizing from the interstitial liquid would be plagioclase crystallizing well before apatite. In a true closed system, the initial liquid fraction can be estimated by the abundance of Na or Sr (or plagioclase mode) or P (or apatite mode) and the trapped liquid porosities based on both should be identical within error, assuming a good estimate of the original liquid concentration of these elements is known. Furthermore, because the mafic minerals will continuously re-equilibrate with the crystallizing pore liquid, the solids will become variably Fe-enriched owing to the ‘trapped liquid shift effect’ on the mg# (Barnes, 1986), the amount of the shift a function of the original proportion of interstitial liquid. Finally, although fractional crystallization of the parent magma will result in an absolute increase in these incompatible elements during crystallization of the ultramafic section, the stratigraphic variation in the Na/P and Sr/P ratios will remain constant until the magma becomes plagioclase-saturated. None of the evolved, partly solidified interstitial liquid moves from its original location or is returned to the magma chamber because there is no significant compaction or loss of a late, low-density liquid as buoyant plumes lost from the mush.

However, for an open system undergoing CC&C, then plagioclase may begin to crystallize early in the compaction process while there is still abundant interstitial liquid present. In contrast, apatite crystallizes after a longer period of liquid loss to compaction. In this case, the estimates of trapped liquid abundance would be higher based on Na, Sr, or, plagioclase mode than for P or apatite mode. Karykowski & Maier (2017) present evidence of open system behavior in the crystal mush during formation of the Lower Zone of the Rustenburg Layered Series of the Bushveld complex using the non-covarying modes of late-crystallizing minerals. Furthermore, estimates of the trapped liquid effect on the mg# shift of the mafic minerals will be lower for estimates based on the expected lower bulk rock P concentrations than for those based on Na or Sr and in both cases can underestimate the true magnitude of the shift. In addition, compaction can result in local increase in porosity as liquid ponds below low-permeability layers as described more fully below.

Understanding the role of both processes is fundamental to understanding how magmas cool and crystallize. The in situ, closed system model proposed by Barnes and Williams implies that formation of low-porosity mush requires a relatively sharp transition between a slowly crystallizing magma and the crystal pile. Fractionation requires the magma to circulate and equilibrate with the mush along a relatively ‘hard ground’ interface. The CC&C model envisions a more extensive crystallization front. Both would attribute orthocumulate rocks (those with abundant late-crystalizing minerals) to faster cooling. Both models also envision that adcumulate rocks (with few late-crystallizing minerals) form where cooling is slower. However, adcumulates are envisioned to form directly during crystallization from the magma at the top of the crystal pile in the closed system whereas they are the result of more efficient compaction for the CC&C model. Finally, if fractionated interstitial liquid is lost from the mush and mixes with the evolving magma, then the incompatible element concentrations of the evolving parent magma will not follow a conventional Raleigh fractionation path but will be modified more along the lines suggested by the in situ fractional crystallization model proposed by Langmuir (1989).

Barnes and Williams have a unique database for many dozens of drill cores with 1-m sampling over roughly 200 ± m of stratigraphic section to explore, but the dataset is incomplete in that the major elements SiO₂ and CaO are missing, no lithologies or modes are reported (lithology is estimated based on bulk composition), and the analyses are apparently done on strict 1-m increments without regard to lithologic boundaries. In their analysis, they did not consider what the different elements from individual samples or drill cores may have suggested, but instead both plotted the data into undifferentiated clouds of data points and used averages to suggest the incompatible element concentrations are inherited from a variable amount of liquid initially present in the rocks. This averaging masks important stratigraphic and lateral variations. In addition, their graphs of composition versus drill core depth show several overlapping compositional variables in one plot for each drill core and it is difficult to see pertinent trends. Here is presented a different interpretation of the data that suggests that CC&C was an important control on the geochemistry of the ultramafic section of the Mirabela intrusion.

An alternative description of the data

To simplify the discussion, here will be considered the example discussed above, an early crystallizing plagioclase and a later crystallizing apatite in the ultramafic section of the Mirabela. Drill core MBS-047 was used by Barnes and Williams as an example drill core from the center of the intrusion. It was likely least affected by sidewall cooling and is a good example of a compacting mush column cooling mainly from below. In Fig. 1 are shown Al, Na, Sr, P, and the bulk rock Mg# variations with stratigraphic height from this drill core.

P is taken to be a suitable proxy for apatite. To test which of the elements Al, Na, or Sr might best serve as proxies for plagioclase, it is noted that Na and Sr have a robust positive correlation with each other for all samples in this core (Fig. 2a). In contrast, Al, which can have significant portioning onto pyroxene as well as plagioclase, shows distinct trends for pyroxene-rich rocks as compared with pyroxene-poor rocks (Fig. 2b). In the following, Na and Sr will be considered as proxies for the plagioclase mode while recognizing that the small concentrations in other phases can bias results in rocks with only minor amounts of plagioclase and apatite. Sr would be the ideal element, as the D_liq/pl ≅ 2, whereas Na is modestly incompatible especially in anorthite-rich plagioclase. However, what Sr makes up in compatibility is in part counteracted by the lower analytic precision reported by Barens and Williams in Sr-poor rocks.

Examination of Fig. 1 shows that Na and Sr have opposite stratigraphic trends to P. Al shows the expected offset in the more orthopyroxene-rich rocks but otherwise parallels the extrapolated Al trend from lower in the section. If the rocks formed as a closed solid + liquid mixture, the Na, Sr, P, and Al (corrected) trends should all parallel each other. The inverse stratigraphic trends shown in Fig. 1 are generally characteristic of all the Central zone drill cores (Fig. 3). The interpretation of Barnes and Williams that there is no fractionation of the two during crystallization of the interstitial liquid is not true in detail.

The stratigraphic trends are confounded, in part, by local porosity variations. For example, the opposite stratigraphic trends of P as compared with Na and Sr would suggest that one might expect X-Y composition plots of P versus Na or P versus Sr for all Central zone drill core to show a negative correlation, but this is not always clearly observed (Fig. 4). It can be minimized by using the 3-m composite analyses of Barnes and Williams (Fig. 4c and d). The more random trends present in X-Y composition plots as compared with the stratigraphic trends is a consequence that any sample composition has both a stratigraphic component and a local porosity component, the effect of the latter of which is most apparent in the coincident peaks of high Na, Sr, and P concentrations. A comparison of MBS-047 with drill core MBS-170 from the Northern Margin region is shown in Fig. 4e and f. Although having overall higher incompatible element abundances, MBS-170 also has negative P/Na and P/Sr stratigraphic trends but when plotted with MBS-047 suggests both combined have a broad positive trend. The linear correlations on the X-Y plots are assumed by Barens and Williams to be a function of trapped liquid proportion alone (aside from a minor fractionation in the parent liquid over the course of the studied section) and is used to define a parent magma composition. While it is possible to calculate an average element ratio from these plots (e.g. P/Na), it is not clear that these ratios are the same as for the parent magma without an independent estimate of the parent magma composition (from an uncontaminated chilled margin or contemporaneous dike, for example).

As noted above, there are a number of coincident peaks that correlate across the P, Na, Sr, and other element profiles, several of which are evident in Fig. 1. A plot of all the peaks from the Central region for which P ≥ 100 ppm is shown in Fig. 5. If the mush initially forms as a closed system composed of the liquid and the bulk solid, then the bulk composition of a perfectly incompatible element should be a function of porosity alone and lie on a line connecting the origin to the estimated parent liquid. When compared with these trend lines, the coincident peaks tend to be enriched in Na and especially P, consistent with the more incompatible nature of Na relative to Sr in plagioclase-saturated liquids and the much later apatite saturation expected for P. Furthermore, while a number of peaks can have Na and P that exceed estimated parent liquid values, this is not as strongly manifested in the more compatible Sr.

The Central region drill core MBS-047 discussed above sampled the section where cooling was mainly through the base and compaction occured normal to layering . Three other drill cores shown in Fig. 2 of Barnes and Williams were from closer to the intrusion margins. Here the regular trends seen in the central section are confounded by additional cooling from side margins and the more pronounced inward-dipping attitude of the layering and are only briefly discussed here. P in these other drill cores is more variable and peaks in P may or may not correlate with peaks in Al or other elements. Where they correlate, they are likely formed by the same porosity variation mechanism described more fully below. Where the P peaks show no obvious correlation or weak coincident peaks, this can be interpreted as the migration of a very late, minor, P-rich liquid that does not significantly affect the abundance of other elements. In the slower cooling interior mush these late liquids presumably were more effectively lost from the crystal pile, perhaps as low-density plumes.

DISCUSSION

In summary, the stratigraphic trends in MBS-047 are characterized by an overall decrease in P but an increase in Na and Sr up-section, with a few peaks where all increase sharply. In terms of modal mineralogy, one would expect this to be expressed in a decrease in the amount of apatite upward while the amount of plagioclase should increase. If the rocks were formed by the closed system process suggested by Barnes and Williams, one would expect all three elements and the amounts of apatite and plagioclase to co-vary, their abundance only controlled by the original amount of trapped liquid. This is not the case. If the rocks were undergoing compaction, the decrease in P up-section in MBS-047 can be explained as due to more efficient compaction in the slower cooling upper part rather than near the base. But again, this conflicts with the upward increase in Na and Sr, and conventional interpretations imply that the residual liquid fraction increases with height as well.

A similar increase in interstitial plagioclase up-section was addressed by Boorman et al. (2004) in a textural study of the Lower and Critical zones of the Bushveld Complex. They noted that the boundary between the Lower Zone and the Lower Critical Zone of the Bushveld Complex is defined by an otherwise rather unremarkable increase in the amount of interstitial plagioclase, from nearly absent in the LZ to several percent in the overlying LCZ. They interpreted and modeled this as a consequence of CC&C.

To understand how the two opposite stratigraphic trends can develop by CC&C, Boorman et al. noted that the composition of the initial interstitial liquid present at any stratigraphic level must represent the evolving parent magma composition. During crystallization of the Bushveld Lower Zone the interstitial liquid is initially far from plagioclase saturation and the mush can compact efficiently such that very little liquid remains once it does reach plagioclase saturation. At the base of the Bushveld Lower Critical Zone, however, the initial liquid is closer to plagioclase saturation and becomes saturated sooner when porosity is still relatively high.

It is suggested that the same thing occurred in the Mirabela intrusion. The Central region drilling sampled rocks relatively far from the country rock contact. The low plagioclase abundance low in the section (as reflected in the low bulk rock Sr and Na content) and the relatively monotonic upward increase in the plagioclase components reflect the residual porosity at the time of plagioclase saturation, which occurs progressively earlier in the compaction history with height. Just below the gabbronorite in the underlying websterite plagioclase is expected to start crystallizing soon after the pyroxenes accumulated on the floor by whatever mechanism one may envision. Here, the rocks contain a maximum of ~0.3 wt.% Na which would put this minimum porosity at ~25%, assuming the interstitial liquid crystallized plagioclase in the same proportions as required to produce the ~1.2% Na seen in the overlying plagioclase-bearing gabbronorite. Near the base of the drill core, however, where Na ≅ 0.05 wt %, the same reasoning would suggest only ~4% liquid at the time of plagioclase saturation. A lower estimate would result if some Na partitioned into pyroxene.

Although P would also be expected to increase in the initial interstitial liquid as the pile continues to grow, its main host mineral apatite is still relatively far from saturation at the top of the ultramafic section. The longer time allowed for compaction away from the cooler base still results in more efficient liquid loss such that P decreases up-section. Estimates of liquid fraction at the time of apatite crystallization require an independent estimate of initial parent magma P concentration. Using the average P concentration of 375 ppm suggested by Barnes and Williams for the parent magma, the ~5–10 ppm present just below the gabbronorite would suggest only ~1–3% porosity or about an order or magnitude lower than the Na estimate of ~25%. In contrast, at the base of MBS-047 P ≅ 45 ppm, equivalent to a porosity of ~12%, or three times higher than the ~4% estimated from the Na concentration.

One might surmise that the plagioclase abundance just below the contact with the plagioclase-saturated gabbroic section would record the initial porosity, but not necessarily so. The system is compacting as plagioclase crystallizes, and hence the 25% porosity noted above only records a minimum porosity. This is also true for P and apatite crystallization.

The coincident peaks of correlated P, Na, Sr, and other element profiles is also consistent with CC&C. Compacting igneous systems can form high-porosity zones when an overlying, low-porosity layer essentially prevents underlying liquid from escaping as the matrix below continues to compact. These liquid segregations can develop as segregation sheets even in thick lava flows (Boudreau and Philpotts, 2002). Consider the peak occurring at a depth of about 175 m just below the base of the websterite unit in MBS-047 shown in Fig. 1. Meurer & Boudreau (1996) have modeled how the porosity can evolve when the compacting system crosses an ultramafic–gabbronorite boundary in which there is a marked change on the matrix–liquid density difference owing to the appearance of abundant plagioclase. In short, compaction in the gabbronorite is slower than in the ultramafic mush, which allows for a higher permeability above the boundary that allows the uppermost mush of the ultramafic section to lose liquid faster that it would have without the density change (Fig. 6). This results in a low porosity at the top of the ultramafic section (= websterite unit). The websterite acts to divide the local region between two compacting units. The Upper unit responds to the websterite as if it were a hard floor to the compacting gabbronorite (allowing for eventually very low residual porosity at the base of the gabbronorite) whereas the mush below the websterite experiences the websterite as a low-permeability lid and liquid ponds below it. If the lower compaction profile were a box of cereal, it would have the disclaimer ‘Contents may have settled…’, to assure consumers that they were not cheated out of product.

An in situ fractionation model.

Summarizing the above, the P, Na, and Sr trends represent the combination of the initial liquid becoming saturated in plagioclase progressively earlier owing to the liquid becoming more evolved with height, resulting in an increasing abundance with height. In contrast, P is only incorporated into the pile much later and records more efficient compaction away from the more rapidly cooled lower sections and producing a decreasing trend with height. Local variations record local porosity variations that can include layers that are largely liquid.

In the analysis of Barnes and Williams, the assumed lack of any fractionation in the ultramafic mush implies one can use the bulk rock trends of strongly incompatible elements to estimate their abundance in the magma that crystallized the analyzed section of the Mirabela intrusion; the ratio of the various incompatible elements in the mush is assumed the same as in the parent magma. The interpretation suggested here is that P was fractionated from Na and Sr owing to the earlier crystallization of plagioclase during CC&C and was lost, in part, from the pile. This would imply that the observed trends evolved from a slightly different parent magma but preferentially lost a P-rich liquid component.

The interpretation suggested here also implies that the conclusion of Barnes and Williams that the mush initially formed as a low liquid fraction ‘hard ground’ is not likely. Instead, the results suggest that the initial mush formed with, at minimum, the porosity expected from random loose packing of mineral grains. Indeed, it is suggested that the initial mush is more akin to the crystallization front observed in lava lakes and discussed by Marsh (2013) where crystallization mainly occurs over some distance at the margins of a cooling intrusion. Crystallization of the floor can be aided by the descending of partially solidified density plumes from the upper margin or density flows from the sidewalls, but the net effect is a mush column that starts with a significant liquid fraction and with a bulk composition of the initial magma.

The model proposed here also solves the problem of the poikilitic harzburgite in which olivine is in an apparent reaction relationship with orthopyroxene, as the hard ground model would require that the olivine crystals simultaneously precipitate from and react with the liquid. Barnes and Williams recognized this problem, and suggested that olivine was carried in. This is not a problem for an initially liquid-rich compacting mush in which the curvature of the olivine–orthopyroxene phase boundary can change from a cotectic to a peritectic as a result of evolving interstitial liquid composition and falling temperature. A mush can start as a cotectic assemblage of olivine + orthopyroxene + liquid only to evolve to the point where olivine + liquid → orthopyroxene. Indeed, modal information on the harzburgites can further inform on the minimum amount of liquid present before plagioclase saturation (Fig. 7).

Barnes and Williams also suggested that the thicker layers present in the central part of the intrusion as compared with the margins is inconsistent with compaction, as more efficient compaction in the center should result in thinner layers. However, the broad U-shaped nature of the layers is consistent with more compaction in the center of the intrusion depressing an originally ~horizontal layering. The layers near the margins can be thinned due to deformation (e.g. stretching) as the central region is depressed. In addition, more efficient compaction in the central region could result in a continuously present, modest depression in the central region such that partially solidified liquids descending from the roof and walls of the chamber would pond to produce a thicker section.

Finally, it is noted that the extensive drilling and chemical analyses were not performed to better understand layered intrusion processes but rather to define a region of potentially economic PGE-Ni-Cu sulfide mineralization. Barnes and Williams suggested that the presumed low initial porosity implies that the mush below the ore zone could not be the source of the ore metals remobilized by late-evolving igneous fluids, under the assumption that the ore component would only be present in an assumed minor initial pore liquid. This would not be the case if the initial porosity was much higher and sulfide precipitated early in the CC&C history only to be remobilized during later degassing to produce the very irregular, podiform Santa Rita ore zone they describe. That issue is left for a separate discussion.

CONCLUSIONS

Detailed examination of individual drill cores shows contrasting behavior of Na and Sr (and suitable corrected Al) versus P in the Central region drill cores that suggests early crystallization of plagioclase relative to apatite. Both models could be further tested if one had an independent estimate of the parent magma composition from a contemporaneous dike, for example. The Barnes and Williams suggestion that the average bulk rock Na/P, etc. ratios are the same as that of the parent liquid can lead to widely variable estimates of residual liquid, depending on which element is selected.

There are also consequences for magmatic differentiation. Conventional Raleigh fractionation assumes all minerals equilibrate with the entire mass of liquid. In the ‘hard ground’ model of Barnes and Williams, formation of a low-porosity mush and the fractionation of the magma requires that ‘cooling rates were slow and magma convection was vigorous’. This appears contradictory as vigorous convention would require rapid heat loss and hence rapid crystallization. Finally, it would seem to be mechanically difficult to approximate Raleigh fractionation by a hard ground model if all the magma was required to have circulated along a sharp solid–liquid contact in order to equilibrate with the solid, as the laminar flow expected of a more likely modest convection regime implies velocity decrease near a solid boundary and less opportunity for the floor rocks and magma to equilibrate.

The interpretation developed here is that the magma above the floor of the chamber can be affected by mixing with evolved liquid that escapes from a mush undergoing CC&C. In this regard, the mush zone and magmatic fractionation will more resemble the in situ fractional crystallization model described by Langmuir (1989), in that the magma compositional evolution is largely accomplished by the dilution of the parent magma by evolved liquids escaping the solidifying crystal mush. More complex liquid extraction models also could be considered (e.g. Boulanger & France, 2023). Indeed, the mush zone is likely mush more chemically active than is generally recognized.

Acknowledgements

This work was improved by the thoughtful and much appreciated reviews of Julia Hammer and an anonymous reviewer.

Data availability

This contribution has not utilized any new data or software.

References