Journal of Petrology Advance Access originally published online on October 1, 2004
Journal of Petrology 2005 46(1):135-167; doi:10.1093/petrology/egh066
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Journal of Petrology vol. 46 issue 1 © Oxford University Press 2004; all rights reserved
An Experimental Investigation of the Influence of Water and Oxygen Fugacity on Differentiation of MORB at 200 MPa
INSTITUT FÜR MINERALOGIE, UNIVERSITÄT HANNOVER, WELFENGARTEN 1, D-30167 HANNOVER, GERMANY
RECEIVED APRIL 15, 2002; ACCEPTED JULY 28, 2004
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONEXPERIMENTAL TECHNIQUESRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Crystallization experiments were performed at 200 MPa in the temperature range 1150–950°C at oxygen fugacities corresponding to the quartz–fayalite–magnetite (QFM) and MnO–Mn3O4 buffers to assess the role of water and fO2 on phase relations and differentiation trends in mid-ocean ridge basalt (MORB) systems. Starting from a primitive (MgO 9·8 wt %) and an evolved MORB (MgO 6·49 wt %), crystallization paths with four different water contents (0·35–4·7 wt % H2O) have been investigated. In primitive MORB, olivine is the liquidus phase followed by plagioclase + clinopyroxene. Amphibole is present only at water-saturated conditions below 1000°C, but not all fluid-saturated runs contain amphibole. Magnetite and orthopyroxene are not stable at low fO2 (QFM buffer). Residual liquids obtained at low fO2 show a tholeiitic differentiation trend. The crystallization of magnetite at high fO2 (MnO–Mn3O4 buffer) results in a decrease of melt FeO*/MgO ratio, causing a calc-alkaline differentiation trend. Because the magnetite crystallization temperature is nearly independent of the H2O content, in contrast to silicate minerals, the calc-alkaline differentiation trend is more pronounced at high water contents. Residual melts at 950°C in a primitive MORB system have compositions approaching those of oceanic plagiogranites in terms of SiO2 and K2O, but have Ca/Na ratios and FeO* contents that are too high compared with the natural rocks, implying that fractionation processes are necessary to reach typical compositions of natural oceanic plagiogranites.
KEY WORDS: differentiation; MORB; oxygen fugacity; water activity; oceanic plagiogranite
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL TECHNIQUESRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Since the late 1950s, many experimental studies have focused on the investigation of phase relations in natural basaltic systems for a better understanding of the differentiation mechanisms under crustal pressures and temperatures. Most of these experiments have been carried out at 1 atm (e.g. Osborn, 1959
; Grove & Bryan, 1983; Grove & Baker, 1984; Juster et al., 1989; Snyder et al., 1993; Thy & Lofgren, 1994; Toplis & Carroll, 1995, 1996), assuming that the small pressure difference between shallow magma chambers and the surface does not influence significantly the results and implications drawn from these studies. It was considered that the low volatile content of basaltic melts (volatiles are incorporated in melts only under pressure) has little influence on phase relations. One of the benefits of these studies was to evaluate the role of fO2 (oxygen fugacity), which is a critical parameter controlling the crystallization of Fe–Ti oxides in iron-bearing systems.
Concerning the role of volatiles, especially H2O, the broad absence of amphibole as a primary cumulate phase in tholeiitic plutonic complexes (e.g. Skaergaard, Kiglapait, Pleasant Bay, typical oceanic crust) has been taken as evidence for very low water activities during the main stage of fractionation of these melts. However, the common presence of small amounts of interstitial amphibole within typical oceanic gabbros indicates that water may play a role, at least during the late differentiation stage of mid-ocean ridge basalt (MORB) (e.g. Gillis, 1996; Dick et al., 2000; Coogan et al., 2001). Studies on water contents in basalts by Sobolev & Chaussidon (1996), Danyushevsky et al. (2000) and Kovalenko et al. (2000) showed that even for primitive MORB the H2O content varies from <0·1 up to 1·5 wt %. Assuming an initial water content for a primitive MORB liquid of 0·5 wt % at a pressure corresponding to shallow levels, the crystallization of 60 wt % of anhydrous minerals results in an increase of water activity (aH2O) up to about 0·23 [calculated after Burnham (1979) for MORB at 1100°C and 100 MPa; see below]. Thus, it can be expected that the water activity of MORB liquids increases with differentiation to values that influence phase relations in these systems (especially at high degrees of fractionation), even if the stability field of amphibole may not have been reached.
Few experimental data are available to understand the effect of aH2O on the differentiation trends of primitive MORBs at moderate pressure (e.g. 
200 MPa). Spulber & Rutherford (1983) performed experiments using a primitive MORB from the Galapagos Spreading Centre to assess the origin of plagiogranites in oceanic ridges, but aH2O was not varied systematically. Further experimental investigations have been carried out in more alkali-rich compositions. The comparison of results from melting and crystallization experiments of Holloway & Burnham (1972) on the 1921 Kilauea tholeiite at aH2O 
0·6 with those of Yoder & Tilley (1962) and Helz (1973, 1976), at water-saturated conditions, shows that water influences phase relations and that the occurrence of amphibole influences the differentiation from andesitic to rhyolitic residual melts. Sisson & Grove (1993a, 1993b) investigated phase relations in calc-alkaline systems at water-saturated conditions, and Kawamoto (1996) carried out melting experiments at aH2O 1 in similar systems. Those workers emphasized the dramatic role of water and pointed out that the absence of amphibole in these systems does not necessarily mean that dry conditions are prevailing.
The lack of experiments in basaltic systems at moderate pressure (100–500 MPa) in the presence of volatiles is due to experimental problems. The formation of quench crystals is a serious problem when performing experiments with low-viscosity basaltic compositions, and often hampers the interpretation of results from phase equilibrium studies (e.g. Hamilton et al., 1964; Holloway & Burnham, 1972; Helz, 1973, 1976). In this study, all experiments were carried out in internally heated pressure vessels (IHPV) using a rapid-quench sample holder to avoid the formation of quench crystals during cooling of the experimental charge (Holloway et al., 1992; Roux & Lefevre, 1992; Berndt et al., 2002). A first series of experiments was undertaken to provide information on the influence of aH2O on phase relations in a primitive MORB system at 200 MPa and fO2 corresponding to the MnO–Mn3O4 buffer using both a primitive and a fractionated synthetic 10-component MORB glass (Table 1). As the fO2 of natural MORB lavas is found to vary generally between QFM – 2 and QFM + 1 (where QFM is the quartz–fayalite–magnetite buffer) (e.g. Carmichael & Ghiorso, 1986; Christie et al., 1986), we performed a second series of crystallization experiments under redox conditions corresponding to the QFM buffer. For these experiments, the sample holder was equipped with a hydrogen membrane to control and adjust fO2 (Berndt et al., 2002). All experiments were performed in the temperature range 1150–950°C and at 200 MPa, a typical pressure of MORB magma chambers under the mid-ocean ridge system at depths between 3 and 6 km (e.g. Fisk, 1984; Nicolas, 1989). For a better understanding of the role of water during differentiation of MORB, the initial water content of our experimental systems was varied systematically ranging from 0·35 wt % to 4·7 wt % H2O. The results are used to estimate the influence of H2O and fO2 on stability of Fe–Ti oxides and MORB liquid line of descent.
|
|
EXPERIMENTAL TECHNIQUES |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL TECHNIQUESRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Apparatus
All experiments were performed in a large-volume IHPV working vertically with pure Ar or Ar–H2 mixtures as pressure medium. A detailed description of the apparatus has been given by Berndt et al. (2002)
. Formation of quench crystals during cooling the experimental charge was avoided using a rapid-quench system [see experimental setup described by Berndt et al. (2002)] in which up to six capsules are fixed to a Pt-wire in the hot zone of the furnace. By fusing the Pt-wire, the capsules drop onto a cold copper block placed at the bottom of the sample holder (about 25°C). The intrinsic fO2 (pure Ar as pressure medium) of the IHPV has been measured at four temperatures (900–1200°C) using the Ni–Pd solid redox sensor technique after Taylor et al. (1992) and is equivalent to the MnO–Mn3O4 solid oxygen buffer (Fig. 1).
|
Starting material
For experiments under oxidizing conditions (MnO–Mn3O4 buffer) two synthetic starting compositions (Table 1) were used for the crystallization experiments: a primitive MORB glass (B1) and a more evolved MORB glass (B2) corresponding to a residual glass composition obtained after 50% crystallization of B1 at 1100°C (run 42; Table 1). Only the B1 composition was used for experiments under reducing conditions (QFM buffer). Starting materials were made from synthetic oxide and carbonate powders. Oxide powder mixtures were placed into a Pt crucible and fused twice (grinding between fusions) at 1600°C, 1 atm, and a log fO2 corresponding to about –0·68 (air). To avoid alkali loss from the glass, melting duration was <1·5 h. Electron microprobe analyses of chips extracted from the top, middle and bottom of the glass showed that they are homogeneous. The glass B1 lies within the compositional range of natural primitive MORBs (Table 1).
Sample preparation and experimental procedure
First attempts were made by preparing the samples with dry powder of the starting glasses (grain size <50 µm) and a given amount of water added into Au80Pd20 capsules. After loading the capsules into the sample holder, they were held for about 1 h above the liquidus temperature at 200 MPa to obtain a homogeneous distribution of H2O in the melt and to allow redox equilibrium of the melt. In a second step, temperature was lowered to the desired run temperature. Resulting products clearly showed disequilibrium features such as complex zonation of minerals. Compositional variations for Al2O3 in clinopyroxene were as high as 5–9 wt % within a single crystal. To avoid these effects, charges with dry glass plus water were brought directly to the desired run temperature. Results at relatively high temperatures (1050–1150°C) did not show disequilibrium features, but mineral sizes and distributions were heterogeneous at lower temperatures and low water activity (aH2O). This is probably due to crystallization taking place before homogeneous distribution of water is achieved in the glass powder.
Therefore, the experimental strategy consisted in using pre-hydrated glass powder as starting materials (Gaetani & Grove, 1998; Müntener et al., 2001). Large amounts of dry glass powder were placed into Au80Pd20 capsules [internal diameter (i.d.) 0·5 cm; outer diameter (o.d.) 0·54 cm at 3 cm length] and stuffed with the help of a steel piston. Distilled water was added to the charge using a microsyringe in five steps to achieve homogeneous H2O distribution within the capsule. Sealed capsules were heated at 1250°C and 200 MPa (above the liquidus) at the desired fO2 (either MnO–Mn3O4 or QFM buffer conditions) for 24 h in an IHPV equipped with the rapid-quench hydrogen-membrane device described above. Homogeneous crystal- and bubble-free glasses were obtained and the water contents were determined by Karl–Fischer titration (KFT) (Table 2; for technique and precision see Behrens (1995)]. Final bulk water contents of the glasses vary between 0·35 and 4·7 wt % and are shown in Table 3.
|
The pre-hydrated glasses were finally crushed in a steel mortar and sieved to fractions of 100–200 µm. About 50–80 mg of the hydrous glass powders were sealed in Au80Pd20 capsules (i.d. 0·3 cm; o.d. 0·34 cm at 1 cm length). A set of up to four capsules (corresponding to the four bulk H2O contents of B1 and B2) were fixed to the rapid-quench device and brought directly to run temperature. Phase relations were investigated at 200 MPa, at 50°C intervals in the temperature range 1150–950°C (Table 3). The run durations for oxidizing experiments were
20 h, except for charges 62 and 61 (B1; 72 h) and 111 (B2; 96 h). The products of these longer experiments were compared with those of the 20 h runs conducted at the same pressure and temperature confirming identical results (runs 45, 42 and 104; see below and Table 3a and b). Experimental duration for runs under reducing conditions was limited by the diffusive Fe loss to the capsule material and varied with temperature from 2 to 9·2 h (Table 3c). After quenching each capsule was weighed to check for leaks and then punctured to determine if a fluid phase was present or not. A cross-section of each capsule (including the sample container) was prepared as a polished section for electron microprobe analysis of solid phases and the noble metal.
|
|
|
Loss of iron
To avoid iron loss to the sample container (e.g. Johannes & Bode, 1978
; Sisson & Grove, 1993a), Au80Pd20 capsules were used because Fe solubility is lower in Au80Pd20 than in Pt, and Au80Pd20 has a higher melting point than pure Au, which allowed us to perform experiments up to 1250°C (Kawamoto & Hirose, 1994). The glasses and capsule walls were analysed with the electron microprobe to check for iron loss or enrichment, respectively. No iron was detected in the capsule material of experiments performed under oxidizing conditions and the total Fe content of the hydrous glasses was equivalent to that of the dry starting material. Experiments performed under QFM buffer conditions showed significant iron loss as a result of Fe diffusion into the Au80Pd20 capsule material. Therefore, the Au80Pd20 capsules were pre-saturated in iron at fO2 corresponding to the QFM buffer at 200 MPa and 1200°C (conditions similar to the desired fO2 in the crystallization experiments at aH2O = 1) following the procedure described by Ford (1978). Thus, iron loss could be limited in our study. A melt composition profile of run 130 (Table 3c), performed above the liquidus temperature at 1150°C and with about 4·58 wt % H2O melt water content, showed that iron has been lost mainly from the rim of the experimental charge, which is in direct contact with the capsule material. In the middle of the capsule the initial iron concentration of the glass was preserved. The average iron content along the profile was lowered by about 4·8% relative to the starting material. Iron loss from the other experiments performed at 1150°C was estimated to be between 6 and 8% relative by mass balance calculations. At temperatures below 1100°C, Spulber & Rutherford (1983) and Sisson & Grove (1993a) emphasized that no significant amount of iron was lost from their samples to the capsule material, which is in accordance with our results. Therefore, the iron content of the glasses analysed in the middle of the capsule is believed to be realistic. This is confirmed by Fe2+–Mg olivine–melt partition coefficients, which agree with published data for this mineral.
Calculations of aH2O and fO2
It has been shown that the thermodynamic model of Burnham (1979) is not always correct for the calculation of aH2O and melt water contents, especially at high pressures (e.g. water solubility can be underestimated by 20% relative at 500 MPa; Holtz et al., 1995). At 200 MPa, calculated H2O solubilities after Burnham (1979) for synthesized MORB glasses B1 and B2 agree well with the contents determined by KFT for these glasses (Table 2). Thus, we have used the model of Burnham (1979) to calculate melt water content at aH2O = 1. In addition, for runs in which aH2O was <1, water activities have also been calculated using the model of Burnham (1979) and the water contents of the melts determined by mass balance and electron microprobe (see below).
For runs performed under oxidizing conditions the fH2 was imposed by the vessel and corresponds to the MnO–Mn3O4 solid oxygen buffer [based on the equation of Chou (1978)]. For experiments at lower oxygen fugacities a proportion of H2 was added to the pressure medium (Ar) to reach a final pH2 at run conditions resulting in an fO2 corresponding to the QFM buffer at water-saturated conditions (Table 3c). It should be noted that in water-undersaturated experiments fH2O and thus fO2 decrease with aH2O (Scaillet et al., 1995), resulting in much more reducing fO2 conditions at low aH2O (Table 3).
Melt Fe2+/Fe3+ ratios were calculated using the method of Kress & Carmichael (1991). As emphasized by Baker & Rutherford (1996) and Gaillard et al. (2001), the expression of Kress & Carmichael (1991) has not been calibrated for hydrous glasses. For H2O-saturated rhyolitic melts Baker & Rutherford (1996) and Gaillard et al. (2001) showed that water has an oxidizing effect on the melt Fe2+/Fe3+ ratio relative to the Kress & Carmichael (1991) expression, especially at fO2 < NNO + 1·5 (where NNO is the nickel–nickel oxide buffer). For the hydrous experiments performed in this study at the QFM buffer (which is below the NNO buffer conditions), the calculated melt Fe2+/Fe3+ ratios were not corrected with respect to the melt water content, because the data of Baker & Rutherford (1996) and Gaillard et al. (2001) were obtained for water-saturated conditions in very silicic melts only. Thus, the effect of dissolved H2O on the Fe2+/Fe3+ ratio in basaltic melts, especially in runs at aH2O <1, is unknown.
Analytical techniques and determination of phase proportions
Electron microprobe analyses of glasses and minerals were performed with a Cameca Camebax and a Cameca SX 100 instrument and are listed in Table 4. Analytical conditions for minerals were 15 kV, 15 nA beam current, and counting times of 5 s for Na and K, and 10 s for all other elements. Glasses were analysed with a defocused beam of 20 µm with a 5 nA beam current and counting times of 2 s for Na and K, and 5 s for the other elements. In crystal-rich runs (glass < 30 wt %) the beam was defocused as much as possible, but loss of alkalis was observed for spot sizes <15 µm. In this case analyses were corrected by a loss factor using the B1 hydrous starting glasses as standards with known water contents.
Phase proportions (wt %) were determined by mass balance (glass analyses were normalized to 100 wt %). Melt water contents in runs with aH2O < 1 were calculated using the known melt fractions and bulk water content of the starting materials. No attempts were made to consider the water content in amphiboles, because runs within the amphibole stability field (Table 3; see Fig. 4) were water saturated.
The melt water content was also estimated for all runs with the ‘by-difference’ method (Devine et al., 1995; Koepke, 1997). Comparison of melt water contents obtained by mass balance (aH2O < 1) and from the ‘Burnham model’ (aH2O = 1) with the water contents determined with the ‘by-difference’ method show good agreement (Fig. 2).
|
, B2) and by using the ‘Burnham model’ (aH2O = 1; 
, B1; 
, B2) with the H2O contents obtained with the ‘by-difference’ method (Devine et al., 1995In experiments with low melt fractions, in which water-saturated conditions mostly prevailed, melt water contents obtained by the ‘by-difference’ method were relatively higher than those from the ‘Burnham model’ (see triangles in Fig. 2 for runs with aH2O = 1). This is probably due to too low microprobe totals resulting from analyses of very small melt pools with a beam that could not be defocused. Analyses of glasses in runs with low melt fractions were possible because small batches of crystal-free residual liquid were located along the capsule rim. This was especially the case for run 151 containing about 8 wt % melt. However, for the near-solidus experiments 47 and 46 the remaining glass pools were too small to be analysed even with a focused beam. Hence, phase proportions (vol. %) have been estimated from back-scattered electron (BSE) images using analySIS© 3.0 software (Gardien et al., 1995
; Koepke et al., 1996). The proportion in wt % (Table 3a) was recalculated assuming typical average densities for the solid phases and for the melt (melt density of 2·28 g/cm3 is estimated from glass composition of run 48 after Lange (1997), Lange & Carmichael (1990) and Ochs & Lange (1999)]. For these two experiments the melt compositions were roughly determined by mass balance calculations.
Experimental limitations and general aspects on the effect of H2O on experimental results
It should be emphasized that there are different ways to understand the effect of H2O on mineral compositions and liquid lines of descent. The effect of bulk water content can be estimated by comparing experimental charges obtained at identical temperature with different bulk water content of the charge. However, comparison of such experiments shows that the crystal amount is not the same at different bulk water contents, especially for high-temperature runs (e.g. runs 42–45 at 1100°C; Table 3a; Fig. 3a). In these high-temperature experiments, residual melts have different water contents and therefore different aH2O. In contrast, for low-temperature runs (e.g. runs 34–37 at 1000°C, Table 3a) the crystal content is always high enough to attain fluid-saturated conditions (as a result of crystallization of mostly anhydrous phases). Thus, at high temperatures, a decrease of the bulk water content results in a strong decrease of aH2O in the experimental charge. In contrast, at low temperature, only small changes in aH2O with changing bulk water content are expected. In subsequent sections, results are discussed as a function of either bulk water content or aH2O. It is emphasized that these two parameters are not identical.
|
From the discussion above, it could be expected that, at low temperatures (e.g. 950 and 1000°C), experimental conditions are always fluid saturated (water saturated), and all other parameters being equal (P, T, bulk composition, fO2), the phase assemblage and composition should be identical. The results in Fig. 4 show that this is not the case because the melt proportion, the relative proportions of minerals and, therefore, melt composition (see Figs 9–11) differ as a function of the bulk water content. An additional possible explanation is that despite the presence of a free fluid phase in these runs (Table 3) aH2O may be decreased by the presence of air in the capsules. Thus, aH2O slightly decreases with decreasing bulk water content because of the higher N2/H2O ratio in experiments with starting glasses containing low water contents. The XH2O [defined as XH2O/(XH2O + XN2)] in these charges was estimated to be 0·92 (for lowest bulk water contents). The small changes in aH2O for given bulk water contents at low temperature influence both composition and stability of experimental phases. For example, in B1 experiments at 1000°C, orthopyroxene is stable only at low bulk water contents (0·89 and 1·49 wt % H2O) and the stability of amphibole is reached only at the highest bulk water content of 4·70 wt % (Fig. 4a).
|
A further point is that a change in the water content of the melt (and thus aH2O), while holding all other parameters constant, results in a change of fO2 (see above). Oxygen fugacity is identical in all runs with aH2O = 1 at isothermal conditions. In this case the fO2 corresponds to the adjusted fO2 conditions of the IHPV. In contrast, at higher temperatures, when H2O saturation has not been reached, fO2 depends also on the prevailing aH2O (Scaillet et al., 1995
; Table 3). Hence, it is difficult to discriminate between the individual effects of aH2O and fO2, especially in the discussion on the evolution of iron-bearing phases.
Achievement of equilibrium
Crystallization experiments are known to be the best method to approach equilibrium conditions (Pichavant, 1987) when compared with melting experiments in which non-equilibrated residual minerals can remain for long run durations in the surrounding melt. The following observations are taken as evidence for near-equilibrium conditions. (1) All phases are typically homogeneous and their compositions change with the experimental conditions, following systematic compositional trends. (2) The crystal distribution is homogeneous in all experiments (Fig. 3a–c). (3) Phase relations and compositions were identical in runs performed at identical conditions and for different run duration (24 h and up to 96 h), at high and at low temperatures (for B1 compare runs 62 and 61 with 45 and 42, respectively, and for B2 compare run 111 with 104; Table 3a and b). (4) Mineral–melt and mineral–mineral partitioning coefficients (for olivine, plagioclase, amphibole and Fe–Ti oxides) are in good agreement with published data from other studies (see below). (5) The formation of quench crystals was avoided (Fig. 3a–c) so that there is no need to distinguish between quench and stable phases or to perform integrative measurements of glass and quench crystals, as has been done in previous studies (e.g. Helz, 1976).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL TECHNIQUESRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Phase relations
Experimental results and phase proportions are listed in Table 3. Phase relations for compositions B1 and B2 are shown in T–H2O diagrams (Fig. 4). The influence of water on phase relations is presented in terms of both bulk water content and contours of equal aH2O in the melt. The relation between aH2O and water content of the melt is given in Table 3.
Primitive MORB system (B1) at MnO–Mn3O4 buffer conditions
Olivine is the liquidus phase for all bulk water contents (Fig. 4a). At aH2O = 1 and aH2O = 0·6 the saturation boundary is at about 1125°C and 1150°C, respectively. In runs below 1000°C olivine is no longer stable. This is in agreement with the decreasing calculated modal abundance of olivine at 1000°C compared with 1050°C (Table 3a) and with changes from euhedral to rounded olivine crystals at 1050°C and 1000°C. Plagioclase and clinopyroxene saturation curves are almost identical and the relative order of crystallization is unknown. Plagioclase and clinopyroxene start to crystallize between 1100°C and 1050°C at aH2O = 1 and at about 1150°C at aH2O = 0·2. Magnetite was found below 1050°C. Amphibole is stable under nearly water-saturated conditions only. It is present in all runs at 950°C and at 1000°C with 4·70 wt % bulk H2O content. Orthopyroxene is stable in runs with 0·89 and 1·49 wt % bulk H2O content at 1000°C and in all runs at 950°C. The ilmenite stability field is restricted to 950°C in charges with 0·89 and 1·49 wt % bulk H2O content.
Differentiated MORB system (B2) at MnO–Mn3O4 buffer conditions
In contrast to B1, olivine was not observed in experiments using the B2 starting composition. Magnetite crystallizes as the liquidus phase for all bulk water contents (Fig. 4b) and its liquidus temperature is below 1150°C for a bulk water content of 4·50 wt %. Clinopyroxene is observed between 1100°C and 1050°C at aH2O = 1 and 1150°C and 1100°C for aH2O < 0·3. The plagioclase saturation curve depends strongly on aH2O. At water-saturated conditions plagioclase is stable only below 1000°C. At lower water activities the saturation temperature of plagioclase rises to 1100°C (aH2O = 0·25). The ilmenite saturation curve depends also on aH2O and is approximately 50°C below that of plagioclase. In contrast to B1, orthopyroxene is observed in B2 at higher temperature for low bulk water contents. It crystallizes at 1050°C for an aH2O of 0·38 (run 96; Table 3b) but is not stable at 950°C and 4·5 wt % bulk H2O.
Primitive MORB system (B1) at QFM buffer conditions
Olivine is the liquidus phase for all bulk water contents and starts to crystallize between 1125°C and 1150°C at water-saturated conditions (Table 3c; Fig. 4c). At aH2O < 0·4 olivine crystallizes at temperatures higher than 1150°C. The plagioclase and clinopyroxene saturation temperature is about 50°C below that of olivine. Amphibole is stable only in experiments at 950°C in which water-saturated conditions have been reached even for low bulk water contents. Ilmenite crystallizes in runs 151 and 150 at 950°C (1·12 and 0·55 wt % bulk water content, respectively).
The solidus temperature was reached in run 150 and crystals were too small to be analysed correctly. Thus, ilmenite as well as all other phases were identified only qualitatively in this experiment by back-scattered electron images.
At a given temperature, the phase diagrams at both low and high fO2 (Fig. 4) show variations in the phase assemblages within the fluid-saturated field. This is attributed to the effect of the dilution of the fluid phase by air (see discussion above), resulting in a slight decrease of aH2O in capsules with low bulk water content (higher N2/Ar ratio).
Phase chemistry
Olivine
XFo of olivines obtained at MnO–Mn3O4 buffer conditions are generally higher (XFo between 92 and 83; Fig. 5a) than at QFM buffer conditions (XFo between 88 and 66; Fig. 5b). At constant temperature, XFo varies as a function of bulk water content (e.g. at 1100°C from XFo92 to XFo84 for a bulk water content of 4·7 and 0·89, respectively; Fig. 5a). At high fO2 XFo falls with falling temperature from 1150 to 1050°C to minimum values of XFo89 (4·70 wt % bulk H2O content) and XFo83 (0·89 wt % bulk H2O content), but rises again at temperatures <1050°C. This is a consequence of increasing magnetite crystallization leading to slightly higher MgO and lower FeO* (total iron) concentrations in the melt (Figs 5 and 9; Table 4a). At low fO2, the minimum XFo is observed in olivine crystallizing at 950°C and is identical, independent of bulk water content (XFo of 66). This is related to melt Mg-number, which is almost identical in all 950°C experiments (ranging from 38 to 35 for all bulk water contents).
|
|
One of the most significant minor elements in olivine is calcium (Libourel, 1999
), with concentrations depending on XFo ranging between 0·19 wt % and 0·47 wt % CaO (Table 4a and c). It is generally accepted that neither oxygen fugacity, nor temperature or pressure, directly influence Ca partitioning (Roeder, 1974; Longhi et al., 1978; Watson, 1979; Jurewicz & Watson, 1988; Libourel, 1999). As demonstrated by Libourel (1999), Ca concentration in olivine depends on the olivine composition as well as on the alkali and alumina content of the melt. Generally, the partition coefficient 
(CaOOlivine/CaOMelt, in wt %) increases with decreasing XFo and increasing melt alkali and CaO content. Figure 6 shows vs XFo obtained in this study. As expected from the generally high XFo and low melt alkali content, Ca concentration in olivine is typically low in our system. Comparison with other data from experimental studies on more iron-rich, tholeiitic and other basaltic systems confirms values obtained in this study, suggesting equilibrium conditions with respect to Ca partitioning between olivine and melt.
|
Pyroxene
Clinopyroxene and orthopyroxene obtained in runs with B1 at high fO2 vary only moderately with changing experimental conditions and have an average composition of En45Fs12Wo43 (±1·3 mol %) and En82Fs1Wo3 (±0·9 mol %), respectively (Table 4a). Clinopyroxene synthesized in experiments with composition B2 is characterized by a larger compositional variation (Table 4b). At water-saturated conditions clinopyroxene has an average composition of En42Fs12Wo46 (±2·3 mol %; 4·50 wt % bulk water content), which changes continuously with decreasing bulk water content to En48Fs13Wo39 (±1·4 mol %; 0·74 wt % bulk water content). B2 orthopyroxene in equilibrium with the most En-rich clinopyroxene crystallized at 0·74 wt % bulk water content has the highest Wo content (En71Fs20Wo9 ± 2·2 mol %). All other orthopyroxenes vary only little in their mole proportions (En79Fs16Wo5 ± 2 mol %), showing slight decreases in Fs and Wo contents with increasing bulk water content. At QFM buffer conditions clinopyroxene has an average composition of En44Fs14Wo42 ± 2 mol %. The average
of clinopyroxene [defined as (XCpx–FeO/XCpx–MgO) x (XMelt–MgO/XMelt–FeO)] is 0·26 but generally increases with melt Mg-number. As reported by Hoover & Irvine (1977) and Toplis & Carroll (1995), values increase independently of fO2 with increasing melt Mg-number up to 0·26 (for melt Mg-number of about 40). At QFM buffer conditions for relatively low melt Mg-number, the obtained values are in good agreement with published data (Table 4). For melt Mg-number above 50, the deviates from the linear trend with values up to 0·41. The same observation has been made for partitioning of Mg and Fe2+ between clinopyroxene and melt under MnO–Mn3O4 buffer conditions, which confirms that there is no dependence of on fO2 (Toplis & Carroll, 1995). It is noteworthy that the bulk water content has no influence on .
|
Plagioclase
The composition of plagioclase as a function of temperature and H2O is shown in Fig. 7. The anorthite (An) content decreases continuously by
0·05 mol % per °C with falling temperature for all bulk water contents. Plagioclase synthesized at high fO2 is slightly more An-rich than that crystallized at similar T–H2O conditions and low fO2. Consequently, the highest An contents (An88, 1050°C; An79, 950°C) are obtained for B1 in runs with 4·70 wt % bulk H2O content (Fig. 7a). The most Ab-rich plagioclase for B1 (An66; Fig. 7b) is stable at 950°C with 1·12 wt % bulk water content. In agreement with previous studies (e.g. Turner & Verhoogen, 1960; Carmichael et al., 1974; Panjasawatwong et al., 1995; Martel et al., 1998; Scaillet & Evans, 1999; Berndt et al., 2001), at a given temperature the An content of plagioclase increases with increasing melt water content and therefore increasing aH2O. However, at fluid-saturated conditions, the plagioclase compositions are also systematically more Ca rich with increasing bulk water content, although the same melt water content (aH2O = 1) apparently has been reached in runs at lower temperatures. This is attributed to the presence of N2 in the capsules (as already discussed), resulting in a slight decrease of water activity with decreasing bulk water content and higher N2/H2O ratio.
|
The direct influence of H2O on plagioclase–liquid equilibrium compositions is shown in Fig. 8. For a constant Ca/Na ratio in the melt, the An content of plagioclase obtained at nearly water-saturated conditions (4·55–5·25 wt % H2O in melt) is noticeably higher when compared with An contents in plagioclases in equilibrium with melts containing about 2 wt % H2O. Consequently,
[defined as molar (Ca/Na)Plag/(Ca/Na)Liq] increases with increasing aH2O of the melt. Continuous lines in Fig. 8 indicate values given by Sisson & Grove (1993a) for plagioclase–melt pairs obtained in high-alumina basalts with 4 wt % H2O in the melt at 100 MPa (
; Sisson & Grove, 1993b) and 2 wt % H2O in the melt at 200 and 500 MPa (
; Baker & Eggler, 1987). This indicates that the effect of pressure variations between 100 and 500 MPa on is negligible. As emphasized by Sisson & Grove (1993a,) the pressure dependence of in dry systems in the range 1 atm to 20 kbar is always significantly lower than the influence of water. values obtained in this study for given melt water contents in primitive and differentiated MORB compositions agree well with the data of Sisson & Grove (1993a, 1993b) and Baker & Eggler (1987) for high-alumina basalts. This confirms that plagioclase composition mainly depends on aH2O and melt Ca/Na.
|
), and about 2 wt % H2O in melt at 200 and 500 MPa (Baker & Eggler, 1987Compared with experimental studies in dry systems (e.g. Toplis & Carroll, 1995
), the An content of plagioclase obtained in our study decreases only by about 0·05 mol % per °C [compared with 0·5 mol % per °C in the study of Toplis & Carroll (1995) for a given bulk water content with falling temperature (Fig. 7)]. As discussed above, in water-bearing systems, for a given bulk water content, increasing crystal fraction (as a result of falling temperature) results in an increase of melt water content. Thus, the effect of decreasing Ca/Na in the melt with progressive crystallization, resulting in a strong decrease of An content in dry systems, is compensated by increasing melt water content.
Amphibole
According to the classification of Leake et al. (1997) the synthesized amphibole is pargasite. The Mg-number of amphibole (calculated with all iron as FeO*) increases with increasing bulk water content (Table 4). The number of Si atoms in the tetrahedral position of the amphibole synthesized at 1000°C is lower than that of amphibole obtained at 950°C, whereas AlIV increases with rising temperature. This is in agreement with previous experimental studies (Helz, 1981; Sisson & Grove, 1993a). At 950°C, Si and Al remain constant for all bulk water contents. Sisson & Grove (1993a) showed that the tetrahedral Al/Si ratio of amphibole is linearly related to the Al/Si ratio of the melt and can be described as 
. An average 
of 1·01 (±0·08) from coexisting liquid–amphibole pairs obtained in B1 and B2 agrees well with the value of 0·94 (±0·06) given by Sisson & Grove (1993a) for liquid–amphibole couples in compositions ranging from high-alumina basalt to high-silica rhyolite.
The occurrence of amphibole in natural systems is often used to argue whether magmas were ‘dry’ or hydrous in a broadly qualitative sense. Our results support previous experimental findings in hydrous basaltic and calc-alkaline systems (e.g. Holloway & Burnham, 1972; Helz, 1973; Spulber & Rutherford, 1983; Sisson & Grove, 1993a; Johnson et al., 1994) showing that a relatively high amount of H2O in the melt is necessary to stabilize amphibole (5 wt % in our study). In our study the stability field of amphibole is always restricted to runs under nearly water-saturated conditions (Fig. 4); however, not all experiments containing a free fluid phase contain amphibole, probably because of higher N2/H2O ratio in charges with low bulk water contents and thus slightly lower aH2O (see above). This indicates that very small variations in aH2O may influence amphibole stability. However, the absence of amphibole is clearly not evidence for ‘dry’ conditions.
Fe–Ti oxides
Owing to the small size of magnetite and ilmenite crystals, analyses of these phases are often contaminated by the surrounding glass as indicated by variable SiO2 contents (Table 4a and b). However, no attempt was made to correct the analyses.
Magnetite in B1 and B2, which is stable only at high fO2, is characterized by relatively low ulvöspinel contents [XUsp; calculated after Andersen et al. (1993)] ranging from 5 to 19 mol %, depending mainly on the bulk water content and, to a lesser extent, on temperature. XUsp systematically increases with decreasing bulk water content for all temperatures (Table 4a and b).
For a given temperature, XUsp in B1 experiments with different bulk water contents remains relatively constant (e.g. 8 and 9·5 mol % at 1050°C and 950°C, respectively, in runs with between 4·70 and 0·89 wt % bulk H2O content). In contrast to B1, magnetite from B2 shows a stronger XUsp variation as a function of temperature. At 1050°C and 950°C, the difference in XUsp between experiments with 4·50 and 0·74 wt % bulk water content is 14 and 4 mol %, respectively. MgO contents of B1 and B2 magnetite decrease with falling temperature for constant bulk water contents and with decreasing bulk water content for a given temperature. This evolution of MgO content in magnetite is related to the Mg-number of the melt. Al2O3 content also decreases with falling temperature for a constant bulk water content.
The ilmenite content [XIlm; calculated after Andersen et al. (1993)] of rhombohedral oxides synthesized in B2 decreases regularly with falling temperature (from XIlm = 43 mol % at 1050°C to XIlm = 33 mol% at 950°C for 0·74 wt % bulk water content; Table 4b) and increases systematically at constant temperature with decreasing bulk water content. The same effect of water is observed for B1 ilmenite crystallized in runs with 0·89 and 1·49 wt % bulk water content at 950°C, which have generally higher XIlm than those obtained at similar conditions in B2. Compared with magnetite, MgO, Al2O3, and MnO contents in ilmenite are generally lower. As for magnetite, MgO content falls with decreasing bulk water content and falling temperature.
The results of Bacon & Hirschmann (1988) can be used as a test for coexisting Fe–Ti oxide equilibrium using Mg/Mn partitioning between magnetite and ilmenite. Fe–Ti oxide pairs from B1 and B2 experiments (Table 4a and b) satisfy the Mg/Mn partitioning criterion of Bacon & Hirschmann (1988), having an average magnetite log(Mg/Mn) of 1·26 (± 0·13) and ilmenite log(Mg/Mn) of 1·40 (± 0·08).
Glass composition at MnO–Mn3O4 buffer conditions
Compositions of glasses obtained in B1 and B2 experiments are given in Table 4a and b and the concentrations of SiO2 and FeO* are plotted against temperature in Figs 9 and 10. Depending on experimental conditions, glass compositions range from basaltic through dacitic to rhyolitic (run 106; classification using the total alkalis–silica (TAS) diagram after, for example, Le Bas et al. (1986)] and follow a calc-alkaline differentiation trend. SiO2, Na2O, and K2O concentrations continuously increase for B1 and B2 with falling temperature and decreasing bulk water content. However, compared with B2, SiO2 contents of B1 vary only slightly for a given temperature with changing bulk water content (Figs 9a and 10a). The maximum SiO2 concentration is about 64 wt % in B1 and 70 wt % in B2. Despite different initial Na2O contents of B1 and B2 the same maximum Na2O concentration of about 5·5 wt % to 4 wt % is reached in both systems for the lowest and highest bulk water contents, respectively (at 950°C).
|
The MgO and CaO concentrations in glasses from B1 and B2 decrease with falling temperature. In runs with high bulk water contents (4·70 and 4·50 wt % H2O in B1 and B2, respectively), the CaO remains constant (B1) or rises slightly (B2), but decreases once clinopyroxene begins to crystallize. At constant temperature, MgO and CaO contents fall with decreasing bulk water content, in contrast to SiO2, Na2O and K2O. MgO contents decrease from their initial values of 9·77 wt % (B1) and 6·49 wt % (B2) to a minimum of approximately 2·5 wt % (highest bulk water contents) and 1·2 wt % (lowest bulk water content).
The TiO2 concentrations in B1 glasses decrease with increasing bulk water contents, which can be correlated with the increasing XUsp in magnetite with increasing bulk water content (see above). For a given bulk water content TiO2 concentrations first increase and then decrease with falling temperature. The maximum value is reached at 1100°C for 0·89 and 1·49 wt % bulk water content and at 1000°C for 2·53 and 4·70 wt % bulk water content. In B2, TiO2 concentrations at low temperatures are mainly controlled by the crystallization of ilmenite. In all experiments containing ilmenite, the TiO2 concentration in the glass is below 1·3 wt %. In addition, TiO2 contents show a general decrease with falling bulk water content and temperature.
The evolution of FeO* content in B1 and B2 glasses is similar to that of TiO2. In B1, the maximum FeO* values are observed at 1100°C (10·19 wt % for 1·49 wt % bulk water content; 10·98 wt % for 0·89 wt % bulk water content; Fig. 9b). The FeO* concentration in melts from experiments with 2·05 and 4·70 wt % bulk water content show only little variation down to 1050°C (about 8·9 wt %). In B2, FeO* contents of all glasses in experiments containing ilmenite are below 5·8 wt % (Fig. 10b). In addition to the effect of ilmenite, FeO* also decreases with falling temperature (for aH2O = 1, 4·5 wt % bulk water) and with increasing water content of the melt (compare runs at 1100°C; Fig. 10b).
|
Glass composition at QFM buffer conditions
Experimental melt compositions are given in Table 4c. The evolutions of melt SiO2 and FeO* with temperature as a function of bulk water content are shown in Fig. 11. In contrast to oxidizing conditions, melts follow a tholeiitic differentiation trend (see Figs 14 and 15;Irvine & Baragar, 1971
; Miyashiro, 1974), ranging from basaltic to andesitic compositions. SiO2 content of the melt (Fig. 11a) increases with falling temperature and bulk water content up to 61 wt % (run 151). FeO* concentration generally increases with falling temperature, but shows significant variations with changing bulk water content (Fig. 11b). For water-saturated melts, FeO* content increases slightly from 1150°C to 950°C (8·24 wt % at 1150°C and 8·97 wt % at 950°C). For lower bulk water contents, FeO* first increases with falling temperature and decreases at temperatures below 1050°C. The highest FeO* is observed for about 0·5 wt % bulk water content with 10·95 wt % FeO* at 1050°C (run 143). The FeO* content in water-saturated melts at 950°C is identical in runs 153 and 152 (no ilmenite present). However, the melt FeO* content is significantly lower in run 151 containing ilmenite. With falling temperature and decreasing bulk water content, the melt MgO and CaO concentrations also decrease. In contrast, Na2O and K2O rise in concentration with falling temperature and bulk water content. A maximum TiO2 content is reached at 1000°C and decreases generally in all experiments at 950°C.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL TECHNIQUESRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Comparison of phase relations at high and low fO2
Phase relations obtained for the B1 starting composition at oxidizing and reducing oxygen fugacities can be compared to estimate the effect of fO2, as all other parameters are equal (pressure, temperature, bulk water content). For both high and low oxygen fugacities olivine is the liquidus phase crystallizing at similar temperatures between 1150°C and 1100°C, depending on bulk water content. In contrast to B1 phase relations at oxidizing conditions, orthopyroxene is not stable at low fO2. This is in agreement with previous studies (e.g. Grove & Juster, 1989
), which show that decreasing melt Mg-number (as a result of decreasing fO2) strongly confines the low-Ca pyroxene stability. Because orthopyroxene is not stable at reducing conditions (melt Mg-number of about 35 at 950°C; Table 4c), olivine is present in the low-temperature runs at 950°C, in contrast to oxidizing conditions (melt Mg-number of 61–68). Magnetite did not crystallize at low fO2 because of the lower melt Fe3+/Fe2+ ratio. Ilmenite stability is known to be less dependent on the oxidation state of the melt (e.g. Toplis & Carroll, 1995), and this mineral is observed at the same temperature and bulk water contents in both reducing and oxidizing conditions. The clinopyroxene and plagioclase saturation is also widely independent of fO2, as both phases crystallize between 1150°C and 1050°C.
Mg/Fe partitioning between olivine and melt
Partitioning of Fe and Mg between olivine and melt, expressed as 
(Roeder & Emslie, 1970) is known to show only a small dependence on melt composition, temperature and fO2, with a value of about 0·3 (± 0·02). The determined in this study is slightly higher with an average value of 0·33 (± 0·01). This higher value is in agreement with the results of Ulmer (1989), who showed a slight positive pressure dependence of , thus suggesting that equilibrium conditions between olivine and melt have been reached in our experiments. Although