Contrib Mineral Petrol (2005) 149: 113–128 DOI 10.1007/s00410-004-0641-8
O R I GI N A L P A P E R
Christian Liebske Æ Alexandre Corgne Æ Daniel J. Frost David C. Rubie Æ Bernard J. Wood
Compositional effects on element partitioning between Mg-silicate perovskite and silicate melts
Received: 30 July 2004 / Accepted: 9 November 2004 / Published online: 14 January 2005 Springer-Verlag 2005
Abstract High-pressure melting experiments were performed at 26 GPa and 2,200–2,400C on synthetic peridotite compositions with varying FeO and Al2O3 contents and on a synthetic CI chondrite analogue composition. Peridotite liquids show a crystallisation sequence of ferropericlase (Fp) followed down temperature by Mg-silicate perovskite (MgPv) + Fp, which contrasts a sequence of MgPv followed by MgPv + Fp observed in the chondritic composition. The difference in crystallisation sequence is a consequence of the different bulk Mg/Si ratios. MgPv/melt partition coefficients for major, minor and trace elements were determined by electron microprobe and secondary ion mass spectrometry. Partition coefficients of tri- and tetravalent elements increase with increasing Al concentration in MgPv. A lattice strain model indicates that Al3+ substitutes predominantly onto the Si-site in MgPv, whereas most elements substitute onto the Mgsite, which is consistent with a charge-compensating coupled substitution mechanism. MgPv/melt partition coefficients for Mg (DMg) and Si (DSi) are related to the melt Mg/Si ratio such that DSi becomes lower than DMg at low Mg/Si melt ratios. We use a crystal fractionation model, based on upper mantle refractory lithophile element ratios, to constrain the amount of MgPv and Casilicate perovskite (CaPv) that could have fractionated during a Hadean magma ocean event and could still be Editorial Responsibility: W. Schreyer C. Liebske Æ D. J. Frost Æ D. C. Rubie Bayerisches Geoinstitut, University of Bayreuth, 95440 Bayreuth, Germany A. Corgne Æ B. J. Wood Department of Earth Science, University of Bristol, Bristol, BS8 1RJ, UK C. Liebske (&) Christian Liebske, Bayerisches Geoinstitut, Universita¨t Bayreuth, 95447 Bayreuth, Germany E-mail:
[email protected] Tel.: +49-0-21-55-3878 Fax: +49-0-921-55-3769
present as a chemically distinct heterogeneity in the lower mantle today. We show that a fractionated crystal pile composed of 96% MgPv and 4% CaPv could comprise up to 13 wt% of the entire mantle.
Introduction In order to reconcile a number of geochemical and geophysical observations, it has been suggested that the Earth’s lower mantle may contain a chemically distinct region below a depth of approximately 1,600 km (Kellogg et al. 1999; van der Hilst and Karason 1999). The existence of this region was proposed to explain the common geochemical inference of an undegassed and undepleted mantle source region that may also be enriched in radiogenic heat producing elements. A possible mechanism for creating such stratification in the Earth’s interior is large-scale crystal fractionation from a deep terrestrial magma ocean, which may have formed during the accretion of the Earth by impacts of numerous smaller or even ‘‘moon forming’’ planetesimals (e.g., Tonks and Melosh 1993). A particularly attractive aspect of this idea is that fractionation of (Mg,Fe)SiO3 perovskite (MgPv) into the lower mantle might also explain the super-chondritic Mg/Si ratio of the Earth’s primitive upper mantle (Ringwood 1979). A number of studies have used experimentally determined partition coefficients to test the effects of MgPv fractionation on the geochemistry of the residual liquid that would have formed the primitive upper mantle (e.g., Kato et al. 1988a,b; Drake et al. 1993; McFarlane et al. 1994; Corgne and Wood 2002; Corgne et al. 2004; Hirose et al. 2004; Walter et al. 2004). Many refractory lithophile elements are present in the upper mantle in near chondritic proportions and significant fractionation of MgPv would have perturbed these ratios. Recent polymineralic fractionation models, that also consider Ca-silicate perovskite (CaPv) fractionation (Corgne et al. 2004; Hirose et al. 2004; Walter
114
et al. 2004), conclude that the lower mantle could contain between 5 and 13 wt% of fractionated material without significantly perturbing the geochemical signature of refractory lithophile elements in the primitive upper mantle. Approximately 60 wt% fractionation, on the other hand, would be required to explain the superchondritic Mg/Si ratio of the upper mantle. The Earth most likely accreted from a range of chondritic material (O’Neill and Palme 1998) and there is no reason to assume that a magma ocean must have been of a primitive upper mantle peridotitic composition. Most previous MgPv/melt partitioning studies, however, have focused on peridotite bulk compositions. If a deep fractionated crystal pile were 4% denser than the overlying mantle, as proposed by convection modelling (Kellogg et al. 1999), then the most obvious possibility is that it may be more Fe-rich. Furthermore, a number of studies have shown that Al content has a dramatic effect on FeO–MgO partitioning between MgPv and ferropericlase (Fp) (Wood and Rubie 1996; Frost and Langenhorst 2002) and on the oxidation state of iron in MgPv (McCammon 1997; Lauterbach et al. 2000; Frost et al. 2004). Little is known, however, about the effects of Al on MgPv/melt partitioning at conditions above the peridotite solidus. Here, we present data on element partition coefficients between MgPv and silicate melts for major, minor and trace elements in peridotitic and chondritic systems. We have specifically examined the effects of varying Al2O3 and FeO contents on MgPv/melt partition coefficients. These data are used to examine whether there are plausible magma ocean compositions that could have allowed a larger or smaller proportion of lower mantle fractionation while maintaining a residual liquid composition that is still within the bounds provided by observed upper mantle chondritic element ratios.
Experimental and analytical methods Peridotitic and chondritic (CI with total Fe as FeO) starting compositions (McDonough and Sun 1995) were synthesised from analytical grade oxides and carbonates. Reagents were ground together under alcohol and then decarbonated by heating to 1,000C. A selection of trace elements at the 100–500 ppm level were added using atomic absorption standard solutions. Powder and doped aqueous solutions were homogenised in an agate mortar and were subsequently devolatilised at 400C. Pressed pellets of this powder were reduced in a gas mixing furnace at an oxygen fugacity about two log units below the quartz–fayalite–magnetite (QFM) oxygen buffer at 1,100C for 48 h. An Al-enriched peridotite composition with a total Al2O3 concentration of 7 wt% was prepared by adding Al2O3 to the initial peridotite composition. Nominal major and minor element starting compositions are reported in Table 1.
Table 1 Nominal starting compositions of materials used for melting experiments
SiO2 TiO2 Al2O3 Cr2O3 FeO MnO MgO CaO Na2O NiO Total
H2026, fertile peridotite
H2033a, Al-enriched peridotite
H2029, chondrite analogue
44.9 0.2 4.4 0.4 8.0 0.1 37.7 3.5 0.4 0.2 100.0
43.7 0.2 7.0 0.4 7.8 0.1 36.7 3.4 0.3 0.2 100.0
34.0 a) 2.4 0.6 34.8 0.4 23.9 1.9 a) 2.0 100.0
S3237: Fertile peridotite + 30 wt% metallic Fe a) Added as trace element
High-pressure melting experiments were performed in a multianvil apparatus at the Bayerisches Geoinstitut using 8 mm Cr2O3 doped MgO octahedra with a straight LaCrO3 furnace. The pressure cell was compressed by tungsten carbide cubes with edge lengths of 32 mm and truncations of 3 mm. The pressure calibration is based on melting phase relations in peridotitic compositions, which indicate run pressures of 26 GPa (Trønnes and Frost 2002). The precision in these experiments is estimated to be ±1 GPa. Capsule materials were either Re metal or graphite. Re capsules were spark eroded from solid rods. All ceramic parts of the high-pressure cells were fired at 1,000C for several hours prior to assembling. The sample powder was stored in a vacuum oven at 250C. After preparation, the high-pressure cell was stored in a vacuum oven until it was used the next day. As discussed by Trønnes and Frost (2002) and Walter et al. (2004), measurements of temperatures 2,300C (i.e., temperatures required for melting our samples at 26 GPa) are often unreliable and the thermocouple itself may also cause damage to soft graphite capsules during compression. Therefore, in many of our experiments a thermocouple was excluded. Run temperatures leading to partially molten samples were estimated on the basis of reported solidus and liquidus temperatures for peridotites and chondrites by Trønnes (2000), Trønnes and Frost (2002) and Asahara et al. (2004). The uncertainty in temperature in this approach is most likely on the order of the liquidus temperature interval, which is estimated to be 150C based on the aforementioned studies. Run conditions are summarised in Table 2. Polished samples have been examined by micro-Raman spectroscopy for phase identification and electron microprobe for chemical analyses. Major and minor element concentrations of the solid phases and coexisting liquids were analysed using a Cameca SX-50 electron microprobe operating at an acceleration voltage of 15 kV. MgPv was analysed using a focused electron beam with a 10–15 nA beam current and counting times of 20 s on the peak and 10 s on the background. The
115 Table 2 Experimental run conditions and products Run
Starting material
Capsule material
Time (min)
Phase assemblage
S3237 H2026 H 2029 H 2033a
Peridotite + 30 wt% Fe0 Fertile peridotite Chondrite analogue Al-enriched peridotite
Graphite Rhenium Rhenium Rhenium
12 10 16 10
Liq, Liq, Liq, Liq,
a
Fp, MgPv Fp, MgPv MgPv, Fp Fp, MgPv, Mj
Pressure: all experiments were performed at 26 (±1) GPa Estimated temperatures: S3237, H2026, H2033a: 2,300C; H2029: 2,200C; see text for details Abbreviations: Liq: Liquid; Fp: Ferropericlase; MgPv: Mg-silicate perovskite; Mj: Majorite
a
quenched liquid, which consisted entirely of quench crystals, was analysed with a defocused beam of 10– 30 lm and beam currents between 15 and 30 nA. Synthetic and natural silicates and oxides were used as standards. Trace element concentrations were measured using a Cameca 4f secondary ion mass spectrometer at the University of Edinburgh. The primary O beam with beam current of 10 nA was accelerated by 10 kV before hitting the gold coated sample. The diameter of the ion beam on the sample surface was focused to approximately 15 lm. Secondary ions were then accelerated by 4,500 V into the mass spectrometer unit. To reduce transmission of molecular species an offset of 75 V was
applied. Trace element concentrations were calculated relative to 30Si after calibration on NIST 610 standard glass. Further analytical details can be found in Corgne et al. (2004).
Results Major and minor element concentrations of coexisting phases are summarised in Table 3. Backscattered electron images of recovered and polished run products are shown in Fig. 1. Thermal gradients, which are unavoidable in high-pressure multianvil experiments lead to the segregation of the silicate melt to the hottest
Table 3 Chemical compositions of coexisting phases measured by the electron microprobe. Errors are two standard deviations of the mean S3237 (fertile peridotite + Fe0) Melt
SiO2 TiO2 Al2O3 Cr2O3 FeO MnO MgO CaO Na2O NiO Total
H2026 (fertile peridotite)
Fp
MgPv
Fp
MgPv
n=22
2r
n=17
2r
n=38
2r
n=50
2r
n=10
2r
n=10
2r
43.81 0.23 4.47 0.29 15.48 0.14 31.11 4.52 0.08 0.03 100.15
0.58 0.02 0.13 0.03 0.53 0.02 1.04 0.39 0.01 0.02
0.19 0.03 1.54 0.35 23.45 0.12 73.86 0.08 0.05 0.05 99.71
0.02 0.005 0.01 0.01 0.12 0.01 0.09 0.004 0.003 0.01
53.96 0.30 4.17 0.18 5.61 0.06 35.13 1.15 0.01 0.01 100.60
0.13 0.005 0.03 0.00 0.07 0.004 0.12 0.04 0.002 0.004
45.93 0.27 4.54 0.46 7.44 0.14 37.08 3.40 0.05 0.23 99.54
0.17 0.01 0.04 0.01 0.14 0.01 0.27 0.13 0.004 0.01
0.18 0.01 1.73 0.62 10.10 0.10 84.94 0.04 0.04 1.04 98.80
0.01 0.005 0.02 0.01 0.06 0.01 0.52 0.003 0.003 0.02
54.41 0.31 4.25 0.37 3.33 0.05 36.04 0.78 0.01 0.04 99.60
0.61 0.02 0.08 0.02 0.16 0.03 0.46 0.07 0.01 0.03
H2029 (CI analogue) Melt
SiO2 TiO2 Al2O3 Cr2O3 FeO MnO MgO CaO Na2O NiO Total
Melt
H2033a (Al-enriched peridotite)
Fp
MgPv
Melt
Fp
MgPv
Mj
n=32
2r
n=20
2r
n=22
2r
n=23
2r
n=16
2r
n=23
2r
n=2
2r
31.41 0.01 1.59 0.61 41.40 0.48 15.79 3.20 0.69 0.81 96.00
0.34 0.01 0.04 0.01 0.46 0.02 0.28 0.08 0.02 0.03
0.29 0.01 0.57 0.86 60.60 0.45 32.25 0.09 0.95 2.99 99.07
0.01 0.002 0.01 0.01 0.30 0.01 0.30 0.004 0.01 0.02
50.60 0.01 3.39 0.45 14.67 0.21 28.93 1.06 0.14 0.20 99.68
0.47 0.01 0.05 0.02 0.49 0.02 0.19 0.04 0.01 0.04
43.51 0.25 7.04 0.40 9.01 0.18 34.10 4.07 0.10 0.24 98.90
0.22 0.01 0.04 0.01 0.05 0.01 0.10 0.02 0.004 0.01
0.26 0.02 3.02 0.59 13.14 0.13 80.86 0.06 0.06 1.15 99.29
0.01 0.01 0.04 0.02 0.11 0.02 0.37 0.01 0.01 0.02
51.57 0.38 6.43 0.34 4.28 0.07 35.21 0.87 0.01 0.05 99.20
0.16 0.01 0.08 0.01 0.11 0.02 0.14 0.06 0.00 0.02
48.32 0.03 14.43 0.37 3.34 0.10 28.31 4.45 0.07 0.05 99.47
0.01 0.02 0.37 0.02 0.22 0.03 0.30 0.07 0.01 0.04
116 Fig. 1 Backscattered electron images of run products. (a) S3237, (b) H2026, (c) H2029 and (d) H2033a. Most of the additional Fe-metal in sample S3237, which segregated completely into the molten area of the capsule, was lost during sample preparation. Small amounts of the metal phase can be seen in (a)
part of the sample and the creation of zones of solid phases. It is believed that these zones broadly reflect the crystallisation sequence with decreasing temperature down the thermal gradient (e.g., Zhang and Herzberg 1994). Experimentally, a liquidus phase in samples with no or small thermal gradients is properly identified by showing the presence of a minor amount of crystals coexisting with a large melt pool, which is chemically almost identical with the bulk composition. These conditions were not achieved in our experiments and we determine the liquidus phase by making the simplified assumption that it is the phase that is the closest to the melt pool. This approach is discussed below.
Phase relations Three slightly different peridotite compositions were investigated. Experiment H2026 was performed on a ‘‘normal’’ model fertile peridotite composition, H2033a was enriched in Al2O3 (7 wt%) and S3237 contained additional metallic Fe (30 wt%) and was performed in a graphite capsule. These three compositions, however, crystallised with an identical sequence of Fp on the liquidus followed down temperature by MgPv + Fp, consistent with phases relations by Trønnes and Frost (2002) extrapolated to 26 GPa (see Fig. 1 in Trønnes and Frost 2002). In addition, the Al-enriched peridotite
117 a)
b)
Temperature
sample (H2033a) also contained majoritic garnet (Majorite, Mj) close to the liquidus. The fact that the sample containing metallic Fe (S3237) shows an identical crystallisation sequence to the experiments in Re capsules implies that oxygen fugacity has no resolvable effect on the phase relations under these conditions, assuming that the elevated Fe/(Mg + Fe) ratio of S3237, caused by oxidation of some FeO, has no influence on the crystallisation sequence. For the chondritic composition (model CI, H2029), Fe concentrations in all phases were much higher compared to the peridotitic samples. Contrary to experiments performed in the peridotite system, MgPv was observed on the liquidus of the chondritic sample followed down temperature by MgPv + Fp. Melting relations for Fe-rich chondritic compositions have been reported by Agee (1990), Agee et al. (1995), Trønnes (2000) and Asahara et al. (2004). These studies all agree that majoritic garnet (Mj) is the liquidus phase below approximately 25 GPa. Although Agee (1990) and Agee et al. (1995) observed a transition to Fp on the liquidus above 25 GPa, Asahara et al. (2004) describe a direct change of the liquidus phase from Mj to MgPv between 24 and 25 GPa. In our study, MgPv is the liquidus phase at 26 GPa, in agreement with Asahara et al. (2004). This implies that the appearance of Fp on the liquidus in a chondritic composition, if it occurs at all, must be restricted to a relatively small pressure interval of 1–2 GPa. This is a remarkably small interval when compared to peridotitic compositions, where Fp is the liquidus phase over a pressure interval of approximately 8 GPa, i.e., from 23.5 GPa (Trønnes and Frost 2002) to 31 GPa (Ito et al. 2004). As Fp is the most FeO-rich phase in both compositions, it seems unlikely that decreasing the Fe/Mg ratio will expand the Fp stability field. It seems more likely that changes in the liquidus phase relations are related to differences in the Mg/Si ratios of both compositions. This is illustrated on a schematic MgO–MgSiO3 phase diagram (Fig. 2). Between 2 and 22.6 GPa, the eutectic composition, arising from the reaction Mg2SiO4 + MgSiO3 = liquid, moves in the direction of increasing MgO content with increasing pressure as summarised by Presnall et al. (1998). At 20 GPa, the eutectic composition is more MgO-rich than either the chondrite or peridotite compositions, resulting in MgSiO3 Mj being on the liquidus for both compositions (Fig. 2a). The observed phase relations in peridotites and chondrites suggest that the shift in the eutectic composition must change direction with increasing pressure and that the eutectic must move quite sharply towards MgPv between 22.6 and 25 GPa, since periclase appears on the liquidus. The situation at approximately 25 GPa is shown in Fig. 2b where Fp is on the liquidus in the peridotite system while the chondrite composition is very close to the eutectic and may have either Fp or MgPv on the liquidus, depending on the exact bulk composition. Above 25 GPa, the liquidus stability field of MgPv must expand at the expense of Fp as a result of a shift
c)
Mg/Si Fig. 2 Phase diagrams along the MgO–MgSiO3 join at (a) 20 GPa, (b) 25 GPa and (c) 33 GPa, inferred from Presnall et al. (1998) and Ito et al. (2004). Differences in phase relations between peridotites and chondrites can be rationalised by their different bulk Mg/Si ratios and changes in eutectic melt composition with pressure. Arrows indicate the direction the eutectic is moving at each pressure
in the eutectic composition towards MgO, because MgPv becomes the liquidus phase in a peridotite composition (Ito et al. 2004) at 31 GPa (Fig. 2c). Thus, we believe that most lower mantle melting studies in complex compositions are quite consistent in the framework of phase relations in the MgO–MgSiO3 system. From this, it follows that the equivalent cotectic composition in a complex system must cross any peridotitic composition twice between 23 and 33 GPa. This likely contributes to a relatively shallow pressure dependence of the liquidus temperature over this region, which would be consistent with the melting temperatures observed by Ito et al. (2004) between 20 and 30 GPa. The description of the phase relations presented here depends critically on the correct identification of the liquidus phase. Thermal gradients and possibly resulting compositional gradients in the capsules may complicate the evaluation of liquidus phase relations from experi-
118
mental run textures. Nonetheless, we emphasise that the consistency between phase relations reported in different experimental studies is remarkable and that some studies carefully evaluated crystallisation sequences using experiments performed at isobaric conditions but varying temperatures (e.g., Trønnes and Frost 2002). This consistency clearly supports the proposed melting systematics and our model to explain liquidus phase relations in chondritic and peridotitic compositions based on their different Mg/Si ratios. Element partitioning All run products contained MgPv crystals that were sufficiently large (20–40 lm diameter) for ion microprobe trace element analysis. Following the terminology of Beattie et al. (1993), the element partition coefficient Di = CMgPv /Cmelt (where Ci denotes the weight fraction i i of an element i) was calculated from measured element concentrations of MgPv and silicate melt. Element
concentrations measured by SIMS are given in Table 4. Partition coefficients are reported in Table 5 and are shown in Fig. 3. SIMS analyses are strongly matrix dependent and we are aware that very high FeO contents as observed in the melt of run H2029 (41.4 wt% FeO) may have an effect on derived partition coefficients, such that high bulk FeO concentrations may result in higher ion yields relative to a matrix having a low FeO content (R. Hinton, personal communication). We observed slightly lower partition coefficients for Al and Mg within the Fe-rich samples S3237 and H2029 from SIMS analyses relative to those derived from the electron microprobe (see Table 5), which may be related to differences in the relative ion yields. The good agreement between the other partition coefficients for major and minor elements obtained from the two techniques does not give strong evidence for a systematic influence of the Fe content on SIMS analyses. However, for consistency, partition coefficients for major and minor elements presented in figures and tables below were therefore calculated from electron microprobe analyses.
Table 4 Concentrations and standard deviations of trace elements in MgPv and silicate melts (in ppm)
Li Be B Na Mg Al Si P K Ca Sc Ti V Cr Fe Mn Co Ni Ga Ge Rb Sr Y Zr Nb Mo Ba La Ce Nd Sm Eu Lu Hf Pb Th U
S3237 (fertile peridotite + Fe)
H2026 (fertile peridotite)
H2029 (CI analogue)
H2033a (Al-rich peridotite)
melt (4) ±
melt (3) ±
melt (5) ±
Melt (6) ±
MgPv (5) ±
532 311 322 793 164,308
24 70 68 116 3,076
19 6 5 62 180,592
10 2 4 17 980
203,377 261 146 22,567 427 1036 36 38 2807 2,707 75,747 24 1,271 1 85 2014 3 222 28 10 1 137 8 80 3 90 3 65 18 2 201 0.5 94 0.3 94 0.4 474 2 451 2 177 9 396 10 341 1 17 1 104 1 97
39 23 1,319 13 142 0 102 5,880 86 7 120 12 3 1 49 4 20 21 5 97 40 40 169 128 45 68 81 6 45 47
241,056 43 28 4,988 1,127 1,591 38 2562 4,3412 606 32 722 128 15 3 1 72 215 82 10 4 1 8 36 72 39 576 929 24 7 19
5 8 822 240 22 1 69 2,574 47 2 30 8 1 0.2 1 8 39 5 1 4 1 4 19 20 9 15 203 6 5 5
MgPv (5) ±
396 203,960 25,757 204,732 309 114 2,7696
52 102 15,803 174,769 119 19,824 252,227 39 197 31 60 1,797 7,687
1,245
107
1,335
2,409 145 1,576 19,9402 18,998 59,718 1,290 52 487
MgPv (2) ±
242 92 83 46 311 4,598 207,604 654 23,254 214,688 25 168 24 38 1,485 18,929 554 47 1,263
19 19 19 15 6,934 329
13 3 2 27 191,456 20,399 254,331 23 47 7 5 1,483 4,252 17 1,191 45 1,367
8 3 2 17 835 205
74 3,526 6,260 77,750 47 1,123 102
61 1,081 51 2
210
MgPv (3) ±
40 90 86 572 2,790 328
9 1 714 372 28
501 932 992 6595 100,859 9,763 14,6771 419 590 28,750 173 92
40 14 7 757 171,929 17,421 236,525 38 40 35 32 1,290 8,940 8 524 2 41
2,651 32,967 463 26
5 2,303 32 0.2
5,555 501,666 4,698 127
201 8,557 83 4
3,927 173,507 1,920 44
6
96
7
422
17
155
103 97 86 81
5 10 14 9
6 71 200 55
3 3 38 3
64 84 102 75
15 3 5 2
0.6 61 210 46
0.3 6 44 5
348 203 81 320
29 5 1 27
3 104 100 32
90 94 92 487 477 186 401 296 66 114 103
5 5 8 50 52 20 55 62 9 23 23
6 4 5 36 72 38 543 780 28 5 15
4 2 2 9 10 5 37 172 3 2 3
34 227 62 333 355 145 438 409 17 93 91
11 24 3 16 7 1 14 19 1 4 2
0.4 1 4 17 46 27 527 908 24 3 11
0.4 1 1 4 11 6 8 189 2 2 4
418 176 320 203 272 260 174 157 94 152 180
30 17 29 16 17 15 3 3 12 17 18
3 2 7 6 19 27 181 241 8 3 5
9 6 8 132 153 155 2 9 386 7 1
119 Table 5 Partition coefficients for experimental runs. For major and minor elements two values are given. The first numbers were derived from SIMS analyses given in Table 4, while second values S3237 Li Be B Na Mg Al Si P K Ca Sc Ti V Cr Fe Mn Co Ni Ga Ge Rb Sr Y Zr Nb Mo Ba La Ce Nd Sm Eu Lu Hf Pb Th U
0.26 0.86 0.77 1.23 0.64 0.54 0.28
(12) / 0.16 (4) (7) / 1.31 (4) (3) / 0.93 (3) / 1.23 (2) (11) (25) (6) /0.25 (2)
1.07 (10) / 1.33 (14) 0.65 (5) / 0.64 (6) 0.30 (4) / 0.36 (1) 0.38 (4) / 0.42 (7) – / 0.50 (45)
0.06 0.73 2.32 0.69
(3) (8) (59) (9)
0.07 (4) 0.042 (16) 0.06 (2) 0.074 (20) 0.15 (3) 0.20 (3) 1.36 (21) 2.64 (81) 0.43 (8) 0.05 (2) 0.15 (5)
H2026
H2029
H2033a
0.055 (33) 0.03 (3) 0.02 (2) 0.06 (6) / 0.19 0.92 (3) / 0.97 0.88 (2) / 0.93 1.18 / 1.18 (1) 0.28 (7) 0.12 (4) 0.22 (2) / 0.23 2.15 (68) 1.08 (4) / 1.15
0.08 (2) 0.015 (6) 0.007 (7) 0.12 (2) / 0.20 1.70 (5) / 1.83 1.78 (6) / 2.14 1.61 / 1.61 (2) 0.093 (9) 0.05 (1) 0.31 (2) / 0.33 3.02 (15) 0.44 (2) / –
0.04 (2) 0.02 (1) 0.014 (12) 0.08 (2) / 0.09 (3) 1.10 (2) / 1.03 (1) – /0.91 (1) 1.19 / 1.19 (1) 0.16 (3) 0.19 (6) 0.22 (4) /0.21 (1) 2.64 (56) 1.54 (21) / 1.49 (6) 1.07 (2) 0.91 (4) / 0.84 (3) 0.57 (6) / 0.47 (1) 0.48 (5) / 0.37 (9) 0.37 (3) 0.36 (3) / 0.23 (7) 0.58 (5) 0.54 (7) 0.28 (5) 0.010 (6) 0.9 (1) 2.38 (68) 1.25 (40) 0.52 (15) 0.02 (2) 0.009 (7) 0.08 (5) 0.08 (5) 0.16 (6) 0.22 (8) 1.46 (25) 2.73 (88) 1.37 (57) 0.07 (6) 0.2 (1)
(12) (1) (2)
(2) (8)
0.75 (1) / 0.81 (5) 0.42 (2) / 0.45 (2) 0.41 (3) / 0.40 (22) 0.257 (5) – / 0.19 (12) 0.46 (4)
0.71 (3) 0.35 (1) 0.41 (1) 0.35 (1) – / 0.24 0.37 (2)
0.010 (5) 0.73 (9) 2.06 (44) 0.62 (8)
0.010 (4) 0.51 (3) 1.23 (4) 0.10 (1)
0.01 (1) 0.006 (2) 0.06 (1) 0.05 (1) 0.13 (3) 0.19 (4) 1.20 (4) 2.22 (47) 1.46 (15) 0.03 (2) 0.12 (4)
0.008 (4) 0.009 (3) 0.023 (2) 0.029 (3) 0.071 (8) 0.10 (1) 1.04 (5) 1.54 (7) 0.09 (1) 0.017 (8) 0.026 (4)
Although D values among the samples investigated show some variation, the partition coefficient patterns presented in Fig. 3 demonstrate the same general trends. The majority of elements are found to be incompatible
(2) (4) (6)
(2)
/ 0.75 (4) / 0.35 (1) / 0.44 (4) (4)
in MgPv and only Si, Mg, Sc, Zr, Lu and Hf are consistently compatible in all samples where they have been analysed. Partition coefficients for transition metals (Sc to Ni) systematically decrease with increasing atomic
10
Partition Coefficient Di (CiMg-pv/Cimelt )
Fig. 3 Element partition coefficients (MgPv/melt) for all elements analysed in this study
were calculated from electron microprobe analyses given in Table 3. Numbers in parentheses are uncertainties in terms of the last units cited calculated from errors in Tables 3 and 4
1
0.1
0.01 H2033a (Al-enriched peridotite) S3237 (fertile peridotite + Fe) H2026 (fertile peridotite) H2029 (CI analogue)
0.001 Li Be B Na Mg Al Si P K Ca Sc Ti Cr Fe Mn Co Ni Ga Ge Rb Sr Y Zr Nb Ba La Ce NdSmEu Lu Hf Pb Th U
120 Fig. 4 Comparison between partitioning data from run H2026 with recent MgPv partitioning coefficients for major, minor and trace elements from Corgne et al. (2004), Hirose et al. (2004) and Walter et al. (2004)
Partition Coefficient Di (CiMg-pv/Cimelt )
10
1
0.1
this study; H2026 (fertile peridotite) Walter et al. (2004); #62 Walter et al. (2004); #249 Hirose et al. (2004); LO#113 Corgne et al. (2004); H2016a Corgne et al. (2004); H2020a Corgne et al. (2004); H2033b
0.01
0.001 Li Be B Na Mg Al Si P K Ca Sc Ti Cr Fe Mn Co Ni Ga Ge Rb Sr Y Zr Nb Ba La Ce Nd Sm Eu Lu Hf Pb Th U
number. Ni appears to be the most incompatible with D’s between 0.19 and 0.33. Partition coefficients for REE increase with atomic number but decrease with ionic radius, with La being highly incompatible (DLa = 0.01) and Lu being compatible (DLu up to 1.5). MgPv-melt partition coefficient patterns reported here follow the same general trend as reported in previous studies (Kato et al. 1988b; Drake et al. 1993; McFarlane et al. 1994; Kato et al. 1996; Taura et al. 2001; Trønnes
Fig. 5 (a) MgPv/melt partition coefficients for selected elements as function of Al per formula unit in MgSiO3 (normalised to two cations). Very high partition coefficients 6 for Zr from Kato et al. (1988a, 1996) are not shown in this diagram. (b) Partition coefficients for incompatible elements as a function of the Al content of MgPv This Study Hirose et al. 2004 McFarlane et al. 1994
Corgne et al. 2004 Taura et al. 2001 Kato et al. 1988b
Walter et al. 2004 Kato et al. 1996 Ito et al. 2004
This Study Hirose et al. 2004
wt% Al2O3 in MgPv
a)
0
1
2
3
and Frost 2002; Corgne et al. 2004; Hirose et al. 2004; Walter et al. 2004). As shown in Fig. 4, the partition coefficient pattern of sample H2026 (fertile peridotite) is comparable with the most recent MgPv-melt partitioning data from Corgne et al. (2004), Hirose et al. (2004) and Walter et al. (2004). Most of the variation in individual partition coefficients between these studies is related to compositional effects, as discussed below. The various compositions studied allow us to test for the effects of FeO and Al2O3 contents on MgPv/melt element partitioning. We observe no systematic influence of bulk FeO content on element partitioning even up to concentrations of 35 wt%, which is consistent with partition coefficients derived from the laser-ablation ICPMS data from Corgne et al. (2004) (samples H2016a, H2020a and H1625), spanning bulk FeO contents of
4
5
6
7
DU
D Lu
1.2 1.0 0.8
0
1
2
0.00
0.02
0.04
3
4
5
6
7
0.06
0.08
0.10
0.12
0.14
0.30 0.25 0.20 0.15 0.10 0.05 0.00 0.10 0.08
D Th
2.0 1.5
0.06 0.04
1.0
0.02
0.5 1.6
0.4
0.00 0.30 0.20 0.25 0.15 0.10 0.05
3.0
0.12
D Sm
1.2 0.8
D Ce
D Hf
D Ti
D Zr
0.6 2.5
Walter et al. 2004 Kato et al. 1996
wt% Al2O3 in MgPv
b)
1.4
Corgne et al. 2004 Taura et al. 2001
2.0 1.0
0.08 0.04 0.00
0.00
0.02
0.04
0.06
0.08
0.10
Al in MgPv (pfu)
0.12
0.14
Al in MgPv (pfu)
121
Discussion Crystal chemistry of Mg-perovskite By exploring MgPv/melt partition coefficients in the framework of the ‘‘lattice strain model’’ (Brice 1975; Blundy and Wood 1994), we can produce a simple parameterisation of the results, learn about the site preference of various elements and gain some insights into the crystal chemistry of MgPv. In this model, a partition coefficient Di for a particular element i with radius ri is expressed as 4pENA r0 2 1 3 Di ¼ D0 exp ðri r0 Þ þ ðri r0 Þ ; 3 RT 2 ð1Þ where D0 is the partition coefficient for a fictive cation with the radius r0 that enters a cation site without causing elastic strain. E is the apparent Young’s modulus of that site, NA is Avogadro’s number, R and T are the gas constant and temperature in Kelvin, respectively. This model predicts an energy penalty, for incorporation of cations, which deviate from the ideal cation radius r0 resulting in a partition coefficient that is smaller than D0. Orthorhombic MgPv has two cation sites, a large distorted site occupied by 12-fold coordinated Mg2+ with eight of the oxygen anions being closer to Mg2+ than the remaining four and a smaller site, occupied by sixfold coordinated Si4+. In the following considerations, cations on the Mg-site are assumed to be in eightfold coordination. It should be noted that ionic radii for 12fold coordination are not available for most elements. Results of lattice strain fits for MgPv/melt partition coefficients are summarised in Fig. 6, and the fit parameters are listed in Table 6. Sets of parabolas with maxima
close to the effective ionic radii of Si4+ = 0.40 A˚ and Mg2+ = 0.89 A˚ (all cation radii taken from Shannon 1976), corresponding to the two cation sites of MgPv, can be distinguished. Some uncertainty arises from the unknown effective radii of Be2+, Al3+ and Ga3+ for eightfold coordination. These radii have been estimated by extrapolating differences in ionic radii between sixand eightfold coordination of other cations of the second and third group of elements in the periodic table. Anomalously high values for the very incompatible elements Sr and La in sample S3237 (fertile peridotite with added metallic Fe) most likely indicate a small amount of melt contamination in the ion probe analyses of MgPv crystals. The level of this contamination is such that it only affects these highly incompatible elements. Other samples show no evidence of melt contamination. The partitioning behaviour of some transition metal elements is poorly described by the lattice strain model as a result of either crystal field stabilisation energies (CFSEs) (Burns 1993) or the possibility of heterovalent element substitution. The former effect may explain the deviation of Co2+ and Ni2+ from the lattice strain curves shown in Fig. 6, while both effects may influence partition coefficients for multivalent Fe, Mn and Cr. DMn is in poor agreement with the divalent curve even though Mn2+ has a 5d electron configuration and 10
Hf Zr
Ti
Si
S3237 H2026 H2029 H2033a
Ge
1
U
Nb
P
Th
0.1
MgPv/melt Partition Coefficient Di
8–15 wt%. However, DSi, DMg and DAl (derived from the electron microprobe) are all larger in the chondritic composition compared to peridotitic and DMg > DSi, as discussed further below. Partition coefficients for tri- and tetravalent elements increase with the Al2O3 concentration of MgPv. In Fig. 5, a strong correlation can be seen between absolute values of partition coefficients and the Al content (per formula unit, normalised to two cations) of MgPv. Data from the literature are also included (see Fig. 5 for reference). Ti, Lu and Zr change from incompatible to compatible with increasing Al content. It would seem that most of the variations observed in previously reported data are best explained by the effect of Al on MgPv/melt partitioning, which is supported by the crystal chemical considerations presented below. However, other parameters that may contribute to variations in experimentally determined partition coefficients are effects of disequilibrium as a result of short experimental run times (Corgne et al. 2004, Walter et al. 2004) and possibly temperature.
4+ and 5+ 0.01 0.001 10
Sc
Al
3+
Cr
1 0.1
Lu Y
Ga Fe3+ Mn3+
B
Eu Sm NdCe La
0.01 0.001 10
Mg 1
Fe Be
1+ and 2+
0.1
Ni
Ca
Co Mn
Na Sr
Ba
Li
K
0.01 0.001 20
40
60
80
100
120
140
160x10-12
Cation radius [m]
Fig. 6 Fits of the lattice strain model to experimental determined partition coefficients as a function of cation radius. Fitting was performed using Be, Mg, Ca, Sr for 2+ curves; Ga, Sc, Y, REE for 3+ and Ti, Hf, Zr, Th, and U for 4+ elements. Open triangles show partition coefficients from run H2033a that have not been included in the fits of Eq. 1. Parabolas shown for the Si-site are only estimates since the number of data points is less than the number of adjustable parameters in Eq. 1
122 Table 6 Fit parameters obtained from the lattice strain model Run 1+ H2029 2+ S3237a H2026 H2029 H2033a 3+ S3237b H2026 H2029c H2033a 4+ S3237 H2026 H2029 H2033a 3+ Si-site H3033ad
D0
E(GPa)
r0 (A˚)
0.2
60
1.21
1.5 1.1 2.5 1.3
(1) (1) (1) (1)
359 290 403 354
(35) (44) (26) (33)
0.96 0.95 0.96 0.95
2.0 1.8 2.1 2.5
(3) (1) (1) / 3.0 (3)
416 425 534 467
(50) (16) (13) / 661 (8) (43)
0.90 0.90 0.90 / 0.90 0.90
2.3 2.2 1.3 2.5
(4) (3) (1) (5)
527 (93) 804 (89) 1029 (36) 482 (107)
0.83 0.85 0.86 0.83
1,300
0.50
1.0
T = 2573 K except for H2029 (T=2473 K) DBe = 0.02 estimated b DGa= 0.5 estimated c Additional values: D0 fixed to DSc d r0 = 0.5 A˚ fixed in regression a
should be insensitive to crystal-field effects. We note that if Mn were totally in the 3+ state, there would be much better agreement for DMn with the trivalent lattice strain curve (Fig. 6). A similar observation can be made for DFe, which would be an excellent fit to the trivalent curve if all Fe in the system is assumed to be ferric. Mo¨ssbauer measurements and electron energy loss spectroscopy on MgPv produced under similar conditions, however, indicate maximum Fe3+ /S Fe ratios of 0.4–0.5 (Frost et al. 2004). As the lattice strain curves predict that DFe2þ should be greater than DFe3þ , this implies that some other influence, such as CFSE, may indeed complicate the partitioning of Fe.
among our samples from 0.08 to 1.5, we suspect some analytical bias and have excluded Pb from further considerations. Trivalent elements REE’s and Y substitute exclusively onto the Mg-site in MgPv. A comparison between the Ga and Al partition coefficients from our experiments shows that DAl is always higher than DGa although the estimated effective radius of Al3+ of 0.68 A˚ in eightfold coordination is smaller than that of Ga3+ (0.75 A˚). This indicates that Al is predominantly present on the Si-site in MgPv, whereas Ga3+ may substitute mainly on the Mg-site. This view is supported by a study of Andrault (2003), who found a high solubility of an [GaAl] component in MgPv. Calculation of ‘‘Mg-site’’ Al partition coefficients from lattice strain fits indicate that about 10% of the Al may also be present on this site in MgPv, in agreement with the estimate of Corgne et al. (2004). Sc, which likely substitutes predominantly onto the Mg-site, shows a slightly anomalous partition coefficient in sample H2029, such that DSc is higher than constrained from the lattice strain model, when fitted together with Ga and REE (see Table 6). A similar behaviour of Sc can be seen in the data from Corgne et al. (2004) in their runs H2016a and H2020a. This is potentially a result of some small amount of mixing of Al, Ga and possibly Sc between both sites in MgPv. Parabolas for the MgPv Si-site shown in Fig. 6 are only estimates. Figure 7 shows partition coefficients for trivalent elements as a function of cation radii for varying MgPv Al contents. All samples were produced under nearly identical experimental conditions. The figure shows, that 10 Sc
Divalent elements There seems to be no indication that divalent cations are influenced by variations in the Al content of MgPv. Be seems to be well assigned to the Mg-site, even given its small estimated ionic radius (0.66 A˚). Although being highly incompatible, Ba has a partition coefficient that is significantly higher than predicted from the lattice strain model, consistent with previous studies (Corgne et al. 2004; Hirose et al. 2004; Walter et al. 2004). Walter et al. (2004) suggested a substitution of Ba2+ for O2 by the formation of a Si4+ vacancy, based on the observation that Ba2+ has an ionic radius similar to O2. We have accordingly excluded Ba from lattice strain fits. Pb2+ is significantly more compatible in all samples than predicted by the lattice strain model, which may be related to its electronic structure (Blundy and Wood 1994) or possibly arises from a contribution of some Pb4+. However, given that DPb shows the largest variations
Partition coefficient Di
Lu Y
1
Ga Eu Sm Nd Ce
0.1 La
0.01 0.130 Al pfu, H2033a, this study 0.085 Al pfu, H2026, this study 0.007 Al pfu, H2033b, Corgne et al. 2004
0.001 70
80
90
100
110
120x10-12
Cation Radius [m] Fig. 7 Partition coefficients for trivalent elements and lattice strain fits as a function of cation radius. The peridotitic samples have different bulk Al concentrations but were equilibrated under nearly identical pressure and temperature conditions. The lattice strain parabola to the data of Corgne et al. (2004) has been adjusted by fixing D0 = 1.6 in order to match DSc (E = 646 GPa, r0 = 0.90 A˚)
123
the increase of partition coefficients with increasing MgPv Al content is a general feature of trivalent cations. From Fig. 7, it seems that the most obvious effect of adding Al is to increase the parameter D0 of the lattice strain model. However, since D0 is strongly related to DSc, a direct correlation between the Al content of MgPv and the parameters of the lattice strain model is complicated by uncertainties in some values of DSc and by strong coupling of D0 and E during data fitting. Partition coefficients for Ce are significantly higher than predicted from the lattice strain model, as indicated in Fig. 7. This can also be seen in previous data sets (Corgne et al. 2004; Walter et al. 2004). Melt contamination seems to be excluded as a possible explanation, because DLa should then also be raised but this is generally not the case. A good explanation for this ‘‘Ceanomaly’’ would be its presence in both trivalent and tetravalent states, as it has been observed in natural zircons (e.g., Hinton and Upton 1991). The lattice strain model predicts that D4+ Ce should be an order of magnitude larger than D3+ Ce such that a relatively small component of Ce4+ could raise the apparent partition coefficient to its observed level. Although the oxygen fugacity in our experiments is not constrained and may be quite high, we note that a Ce anomaly of similar magnitude occurs in our experiment performed in the presence of metallic Fe (S3237) and in the experiments of Walter et al. (2004) performed in diamond capsules. Such effects for Ce are not observed at low pressure, even in partitioning studies performed in air (e.g., Hill et al. 2000), which may indicate that either pressure or MgPv crystal chemical effects favour the stability of the smaller Ce4+ cation.
Tetravalent cations As seen in Fig. 5a, b, partition coefficients for tetravalent cations also increase with increasing Al content of MgPv. In accordance with their effective radii, the elements Hf4+, Zr4+, U4+ and Th4+ substitute exclusively onto the Mg-site. The assignment of Ti4+ is more difficult, however, as it may substitute onto both sites and it is also possible that its site preference changes with the Al concentration of MgPv. At high Al contents, DTi is greater than DSi and much greater than DGe, which means that if Ti were entirely on the Si-site the corresponding elastic strain parabola would require an unrealistically high value of D0 (10) or a very low value of E4+ and there would be very poor agreement with Ge. The relationship between Si and Ge in fact implies that only a relatively small amount of Ti is on the Si-site ( DSi for MgPv/melt partitioning, as observed for the chondritic sample (H2029). These observations clearly demonstrate that PUM is highly unlikely to be a derivative of large scale MgPv fractionation from a chondritic magma ocean.
2.0
This Study Corgne et al. 2004 Walter et al. 2004 Hirose et al. 2004 Trønnes 2000 Asahara et al. 2004 Ito et al. 2004 Trønnes & Frost 2002 McFarlane et al. 1994 Agee 1993 Gasparik 1996 (Fe-free)
OC/EC
MgPv partition coefficient
1.8
1.6
CaC/CI
1.4
PUM
DSi
1.2
1.0
DMg
0.8 0.6
0.8
1.0
1.2
1.4
molar Mg/Si ratio (melt)
Fig. 8 Partition coefficients for Mg and Si as a function of the Mg/ Si ratio of the residual melt derived from various starting compositions. Open symbols represent DSi and filled symbols DMg, respectively. The size of the symbols represents approximately the uncertainty of data points. DSi and DMg show a crossover at chondritic Mg/Si ratios. Fractionation of MgPv from a melt having a Mg/Si ratio > Dj. There are two problems, however, with applying this methodology to determining the extent of maximum possible perovskite fractionation. The first is that the aforementioned ‘most discerning element ratios’ have also been strongly fractionated in the mantle as a result of extraction of the continental crust and cannot, therefore, be assumed to be chondritic in PUM. The second is that since the majority of elements that are incompatible in MgPv are actually compatible in CaPv (e.g., REEs), the combined fractionation of MgPv and CaPv will partly compensate the leverage of one phase on an element concentration ratio, thereby extending the possible amount of fractionated material and creating two unknowns, i.e., the total amount of plausible perovskitic fractionation and the proportion of MgPv and CaPv in the crystallising assemblage. In Fig. 9 we have plotted two ratios, Ca/Sc and Yb/ Ca, that are, with suitable uncertainties, chondritic in PUM (Walter et al. 2004) but have complimentary sensitivities to the fractionation of each perovskite phase. Partition coefficients for CaPv are taken from experiment H2020b of Corgne et al. (2004). We use a 10% uncertainty on each element concentration consistent with estimates by Walter et al. (2004). The amount of pure MgPv fractionation is constrained by Ca/Sc to be approximately 10% but if we add CaPv to the fractionating assemblage the amount of allowable fractionation increases and can reach 100% for a 70 : 30 assemblage of MgPv and CaPv. The ratio Yb/Ca con-
125
CI chondrite normalised ratios
1.6 96% MgPv 4% CaPv 92% MgPv 8% CaPv 100% MgPv
1.4
Ca/Sc
1.2 1.0 0.8 0.6
Yb/Ca
0.4 0
5
10
15
20
25
30
wt% perovskite fractionation
Fig. 9 Chondrite normalised ratios for Ca/Sc and Yb/Ca as a function of perovskite fractionation calculated from Eq. 7. MgPv/ melt partition coefficients were taken from experiment H2026 with DYb being calculated from Eq. 1 and parameters given in Table 6. Data for CaPv are from Corgne et al. (2004). Horizontal dashed lines represent the uncertainty on both element ratios. Solid, dashed and dotted lines show the effect of different proportions of MgPv and CaPv in the fractionating assemblage
strains the maximum amount of pure MgPv fractionation to be approximately 17%, however, this is reduced if CaPv is added to the crystallising assemblage. The maximum amount of fractionation allowed by both of these ratios is approximately 13% of a 96 : 4 MgPv : CaPv mixture. These ratios are therefore the most discriminating for determining the maximum allowable fraction of a lower mantle perovskitic layer as a number of other possibly useful and reliably chondritic ratios (Al/Yb, Sc/Yb, Sm/Yb, Zr/Y; see Walter et al. 2004) all remain within their PUM uncertainties for this amount and proportion of fractionation. This estimate is larger than in some previous studies (Kato et al. 1988a, 1988b), which had to employ less reliable partitioning data. Using our partitioning data we can examine the influence of varying chemical composition on the allowable size of a perovskitic layer. Assuming a crystallising assemblage of 96% MgPv and 4% CaPv, bulk partition coefficients for Fe and Ca are 0.4 and 0.5, respectively, and the concentration of both elements would, therefore, increase in the liquid as fractionation proceeds. As described in previous sections, even extreme enrichment of a melt in Fe does not have a pronounced effect on MgPv/melt partitioning and will not have major consequences on the allowable amount of perovskite fractionation. The Ca content of the melt, however, is negatively correlated with CaPv/melt partition coefficients for trivalent and divalent cations. All currently available partition coefficients for CaPv come from experiments with enriched bulk CaO contents (e.g., Corgne and Wood 2002; Corgne et al. 2004; Hirose et al. 2004), as CaPv does not appear near the liquidus of normal peridotite assemblages at conditions equivalent to the top of the lower mantle. In determining the plausible amount of perovskite fractionation above we used values for CaPv partitioning from H2020b of
Corgne et al. (2004), an experiment which contained 8 wt% CaO in the melt. Any plausible magma ocean composition likely had a lower CaO content. Using data from Corgne et al. (2004) and Hirose et al. (2004) we can make a rough parameterisation of partition coefficients vs the melt CaO content. By extrapolating the CaO content to a typical peridotite value (3.5 wt%) we find that the permissible amount of total perovskite fractionation and the CaPv content of the assemblage decreases to 10 wt% fractionation in the proportion 97% MgPv and 3% CaPv as a result of the increase in CaPv/melt partition coefficients. The Al2O3 concentration has a strong effect on MgPv/melt partition coefficients but given that the bulk Al2O3 partition coefficient is close to unity, the concentration of Al2O3 in MgPv will not vary enough during fractionation to significantly affect our conclusions on the size of a potential perovskitic layer. Any perovskitic reservoir is likely to be Fe and Ca depleted and thus probably less dense compared to the rest of the mantle, which may violate the requirements for a dynamically stable reservoir in the lower mantle (Kellogg et al. 1999). If a magma ocean extended to pressures corresponding to the present day core–mantle boundary, an initially perovskite-rich layer could have become more dense as a result of phase transformation to a post-MgPv phase (Murakami et al. 2004). This could have stabilised such a chemically distinct layer at the bottom of the mantle against subsequent mantle convection. In addition to chemical variables, it is likely that pressure and possibly temperature also influences partition coefficients. In recent studies it has been proposed that D0 can be related to the free energy of fusion of a hypothetical trace element component (Blundy et al. 1995; Wood and Blundy 1997; van Westrenen et al. 2001). As silicates usually show a positive Clapeyron slope of fusion, this may contribute to an increase in D0 with increasing pressure. In this case, however, increasing absolute values of mineral/melt partition coefficients will increase their leverage on RLE ratios, and therefore decrease the total amount of allowable perovskite fractionation. This implies that the 10–13 wt% MgPv/CaPv fractionation, that is still consistent with upper mantle RLE ratios, is a maximum boundary and that the effects of pressure in a magma ocean significantly deeper than 700 km may even lower this value. Figure 10 shows BSE normalised element abundances of a ‘‘perovskite depleted mantle’’ (PDM) and the complementary perovskitic reservoir formed by 13 wt% fractionation (96 : 4 MgPv : CaPv). The remarkable flat element abundance pattern of the PDM is a result of bulk partition coefficients close to unity and the relatively high melt fraction. The calculated element patterns for the depleted mantle with and without perovskite fractionation are almost indistinguishable, assuming the continental crust was formed by 1.5 wt% melt extraction (Hofmann 1988) and using partitioning data and mineral modes from Salters and Stracke (2004). Although there is no direct evidence for the
126
BSE normalised abundance
100
10
1
0.1 CC from 1.5 wt% melt extraction from BSE DM after 1.5 wt% melt extraction from BSE PDM from 13 wt% MgPv/CaPv fractionation (96:4) perovskitic reservoir 96:4 MgPv:CaPv DM after 1.5 wt% melt extraction from PDM
0.01
0.001 Ba
K
Th
Nb
U
La
Ce
Hf
Sr
Nd Sm
Ti
Yb
Lu
Fig. 10 BSE normalised abundance patterns for highly incompatible elements during formation of the continental crust (CC). Elements are sorted by decreasing incompatibility. BSE concentrations are taken from McDonough and Sun (1995). Filled black symbols represent a ‘‘perovskite depleted mantle’’ (PDM) and the complementary perovskitic reservoir after 13 wt% MgPv/CaPv (96 : 4) crystallisation calculated from Eq. 5 (partition coefficients for MgPv from H2026; CaPv data are from Corgne et al. 2004). Open diamonds and squares show the depleted mantle (DM) with and without previous perovskite fractionation as a residual of 1.5 wt% crustal melt extraction. The latter ‘‘low melt fraction’’ processes have been calculated using the batch melting equation
presence of a lower mantle perovskitic reservoir, the geochemistry of the upper mantle is not inconsistent with 13 wt% MgPv/CaPv fractionation. Such a reservoir could of course be smaller, could have been homogenised by subsequent mantle convection or may never have existed. Assuming 13 wt% fractionation and taking least square regressions to the Mg and Si partitioning coefficients presented in Fig. 8 together with the present day Mg/Si ratio of PUM (1.26, Walter et al. 2004), we calculate that a magma ocean (or the BSE) would have an initial Mg/Si ratio of 1.23, which is essentially peridotitic. We conclude, in agreement with previous studies (e.g., Corgne et al. 2004; Hirose et al. 2004; Walter et al. 2004), that the depletion of Si in PUM relative to Mg and CI-chondrites is a feature of the bulk silicate Earth, that was caused by either volatilisation during accretion (e.g., Ringwood 1979), incorporation of Si into the core (e.g., Allegre et al. 1995) or was simply inherited from an as yet unsampled proto Earth material.
Conclusions Differences in liquidus phase relations, particularly the Fp stability field, between peridotitic and chondritic compositions can be rationalised by changes in the eutectic melt composition with pressure, which are analogous to changes observed in the simple MgO– MgSiO3 system. MgPv/melt partition coefficients for tri- and tetravalent cations increase with increasing Al concentration in MgPv but are not significantly influenced by even ex-
treme variations in the bulk Fe content. The positive correlation between the Al content of MgPv and partition coefficients of trivalent cations can be envisaged by a charge compensating coupled substitution mechanism involving Al3+ predominantly on the Si-site and other trivalent cations on the Mg-site. In the case of tetravalent cations, the actual charge compensating mechanism cannot be identified with certainty, but it may involve a similar mechanism, by substitution of two Si4+ and one Mg2+ by two tri- and one tetravalent cation on the Siand Mg-site, respectively. An additional or complementary exchange mechanism, which involves formation of Mg-vacancies, is also possible. Partition coefficients of heterovalent elements (e.g., Fe, Mn, Ce and possibly Pb) indicate that MgPv has an affinity for highly oxidised cations. Our partitioning data have been used to place constraints on possible early mantle differentiation processes involving MgPv fractionation from a Hadean magma ocean. Partition coefficients of Mg and Si are a function of the melt Mg/Si ratio and converge such that the ratio is less efficiently fractionated by MgPv for chondritic compositions. Therefore, it appears unlikely that the present day Mg/Si ratio of the primitive upper mantle (PUM) has evolved by MgPv fractionation from a chondritic magma ocean. The chondritic Ca/Sc and Yb/Ca ratios observable in PUM limit fractionation of a mixture of MgPv and CaPv perovskite into a chemically distinct reservoir in the lower mantle to less than 13 wt% in the proportion of 96% MgPv and 4% CaPv. Likely variations in the Al2O3, CaO or FeO contents in the residual melts during perovskite crystallisation do not have a significant effect on the maximum amount of combined MgPv/CaPv fractionation. An accumulated MgPv-rich layer would likely be depleted in Fe and Ca relative to the overlying mantle and therefore probably less dense, indicating that it may not have resisted against subsequent mantle convecting, unless it was stabilised by densification from a post-(Mg)perovskite phase transformation slightly above the core–mantle boundary. Acknowledgements We would like to thank Heinz Fischer, Hubert Schulze, Georg Herrmannsdo¨rfer and Detlef Krausse for technical support. We greatly acknowledge assistance during ion microprobe analysis by R. W. Hinton and J. Craven at University of Edinburgh. C.L. is grateful for support by the German Science Foundation (DFG, project, Fr 1555/01). A.C. acknowledges receipt of a Postgraduate Scholarship from the University of Bristol. We would like to thank Reidar Trønnes and Wim van Westrenen for very helpful and constructive reviews.
References Agee CB (1990) A new look at differentiation of the Earth from melting experiments on the Allende meteorite. Nature 346:834– 837 Agee CB, Li J, Shannon MC, Circone S (1995) Pressure–temperature phase diagram for the Allende meteorite. J Geophys Res 100, B9:17725–17740
127 Akber-Knutson S, Bukowinski MST (2004) The energetics of aluminium solubility into MgSiO3 perovskite at lower mantle conditions. Earth Planet Sci Lett 220:317–330 Allan NL, Du ZM, Lavrentiev MY, Blundy JD, Purton JA, van Westrenen W (2003) Atomistic simulation of mineral-melt trace-element partitioning. Phys Earth Planet Inter 139:93– 111 Allegre CJ, Poirier JP, Humler E, Hofmann AW (1995) The chemical composition of the Earth. Earth Planet Sci Lett 134:515–526 Andrault D (2003) Cation substitution in MgSiO3 perovskite. Phys Earth Planet Inter 136:67–78 Asahara Y, Kubo T, Kondo T (2004) Phase relations of a carbonaceous chondrite at lower mantle conditions. Phys Earth Planet Inter 143–144:421–432 Beattie P, Drake M, Jones J, Leeman W, Longhi J, McKay G, Nielsen R, Palme H, Shaw D, Takahashi E, Watson B (1993) Terminology for trace-element partitioning. Geochim Cosmochim Acta 57:1605–1606 Blundy JD, Wood BJ (1994) Prediction of crystal-melt partition coefficients from elastic moduli. Nature 372:452–454 Blundy JD, Falloon TJ, Wood BJ, Dalton JA (1995) Sodium partitioning between clinopyroxene and silicate melts. J Geophys Res 100 B8:15,501–15,515 Brice JC (1975) Some thermodynamic aspects of the growth of strained crystals. J Cryst Growth 28:249–253 Burns RG (1993) Mineralogical applications of crystal field theory. Cambridge University Press, Cambridge, UK, pp 551 Corgne A, Wood BJ (2002) CaSiO3 and CaTiO3 perovskite-melt partitioning of trace elements: Implications for gross mantle differentiation. Geophys Res Lett 29(19):1903, DOI: 10.1029/ 2001GL014398 Corgne A, Allan NL, Wood BJ (2003) Atomistic simulations of trace element incorporation into the large site of MgSiO3 and CaSiO3 perovskites. Phys Earth Planet Inter 139:113–127 Corgne A, Liebske C, Wood BJ, Frost DJ, Rubie DC (2004) Silicate perovskite-melt partitioning of trace elements and geochemical signature of a deep perovskitic reservoir. Geochim Cosmochim Acta (in press) Drake MJ, McFarlane EA, Gasparik T, Rubie DC (1993) Mg-perovskite/silicate melt and majorite garnet/silicate melt partition coefficients in the system CaO–MgO–SiO2 at high temperatures and pressures. J Geophys Res 98(E3):5427–5431 Frost DJ, Langenhorst F (2002) The effect of Al2O3 on Fe–Mg partitioning between magnesiowu¨stite and magnesium silicate perovskite. Earth Planet Sci Lett 199:227–241 Frost DJ, Liebske C, Langenhorst F, McCammon CA, Tronnes RG, Rubie DC (2004) Experimental evidence for the existence of Fe-rich metal in Earth’s lower mantle. Nature 428:409–412 Gasparik T (1996) Melting experiments on the enstatite–diopside join at 70–224 kbar, including the melting of diopside. Contrib Mineral Petrol 124:139–153 Hill E, Wood BJ, Blundy JD (2000) The effect of Ca-Tschermaks component on trace element partitioning between clinopyroxene and silicate melt. Lithos 53:203–215 van der Hilst RD, Karason H (1999) Compositional heterogeneity in the bottom 1000 kilometers of Earth’s mantle: toward a hybrid convection model. Science 283:1885–1888 Hinton RW, Upton BGJ (1991) The chemistry of zircon—variations within and between large crystals from syenite and alkali basalt xenoliths. Geochim Cosmochim Acta 55:3287–3302 Hirose K, Shimizu N, van Westrenen W, Fei Y (2004) Trace element partitioning in Earth’s lower mantle and implications for geochemical consequences of partial melting at the core–mantle boundary. Phys Earth Planet Inter 146:249–260 Hofmann AW (1988) Chemical differentiation of the Earth: the relationship between mantle, continental crust, and oceanic crust. Earth Planet Sci Lett 90:297–314 Ito E, Kubo A, Katsura T, Walter MJ (2004) Melting experiments of mantle materials under lower mantle conditions with implications for magma ocean differentiation. Phys Earth Planet Inter 143–144:397–406
Kato T, Ringwood AE, Irifune T (1988a) Experimental determination of element partitioning between silicate perovskites, garnets and liquids: constraints on early differentiation in the mantle. Earth Planet Sci Lett 89:123–145 Kato T, Ringwood AE, Irifune T (1988b) Constraints on element partition coefficients between MgSiO3 perovskite and liquid determined by direct measurements. Earth Planet Sci Lett 90:65–68 Kato T, Ohtani E, Ito Y, Onuma K (1996) Element partitioning between silicate perovskite and calcic ultrabasic melt. Phys Earth Planet Inter 96:201–207 Kellogg LH, Hager BH, van der Hilst RD (1999) Compositional stratification in the deep mantle. Science 283:1881–1884 Kro¨ger FA, Vink HJ (1956) Relation between the concentration of imperfections in crystalline solids. Solid State Phys 3:307–435 Lauterbach S, McCammon CA, van Aken P, Langenhorst F, Seifert F (2000) Mo¨ssbauer and ELNES spectroscopy of (Mg,Fe)(Si,Al)O3 perovskite: a highly oxidised component of the lower mantle. Contrib Mineral Petrol 138:17–26 McCammon C (1997) Perovskite as a possible sink for ferric iron in the lower mantle. Nature 387:694–696 McDonough WF, Sun S-s (1995) The composition of the Earth. Chem Geol 120:223–253 McFarlane EA, Drake MJ, Rubie DC (1994) Element partitioning between Mg-perovskite, magnesiowu¨stite, and silicate melt at conditions of the Earth’s mantle. Geochim Cosmochim Acta 58:5161–5172 Murakami M, Hirose K, Kawamura K, Sata N, Ohishi Y (2004) Post-perovskite phase transition in MgSiO3. Science 304:855– 858 O’Neill SC, Palme H (1998) Composition of the silicate Earth: implication for accretion and core formation. In: Jackson I (ed) The earth’s mantle: compositions, structure, and evolution. Cambridge University Press, Cambridge, UK, pp 3–182 Presnall DC, Weng YH, Milholland CS, Walter MJ (1998) Liquidus phase relations in the system MgO–MgSiO3 at pressures up to 25 GPa—constraints on crystallisation of molten Hadean mantle. Phys Earth Planet Inter 107:83–95 Richmond NC, Brodholt JP (1998) Calculated role of aluminum in the incorporation of ferric iron into magnesium silicate perovskite. Am Mineral 83:947–951 Ringwood AE (1979) Origin of the Earth and Moon. Springer, Berlin Heidelberg New York, p 295 Salters VJM, Stracke A (2004) Composition of the depleted mantle. Geochem Geophys Geosyst 5:Q05004, DOI:10.1029/ 2003GC000597 Shannon RD (1976) Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides. Acta Crystallogr A 32:751–767 Stebbins JF, Kroeker S, Andrault D (2001) The mechanism of solution of aluminum oxide in MgSiO3 perovskite. Geophys Res Lett 28:615–618 Taura H, Yurimoto H, Kato T, Sueno S (2001) Trace element partitioning between silicate perovskites and ultracalic melt. Phys Earth Planet Inter 124:25–32 Tonks WB, Melosh HJ (1993) Magma ocean formation due to giant impacts. J Geophys Res 98 (E3):5319–5333 Trønnes RG (2000) Melting relations and major element partitioning in an oxidized bulk Earth model composition at 15– 26 GPa. Lithos 53:233–245 Trønnes RG, Frost DJ (2002) Peridotite melting and mineral-melt partitioning of major and minor elements at 22–24.5 GPa. Earth Planet Sci Lett 197:117–131 Walter MJ, Nakamura E, Trønnes R, Frost DJ (2004) Experimental constraints on crystallization differentiation in a deep magma ocean. Geochim Cosmochim Acta 68:4267–4284 van Westrenen W, Wood BJ, Blundy JD (2001) A predictive thermodynamic model of garnet-melt trace element partitioning. Contrib Mineral Petrol 142:219–234 Wood BJ, Blundy JD (1997) A predictive model for rare earth element partitioning between clinopyroxene and anhydrous silicate melt. Contrib Mineral Petrol 129:166–181
128 Wood BJ, Rubie DC (1996) The effect of alumina on phase transformations at the 660-kilometer discontinuity from Fe–Mg partitioning experiments. Science 273:1522–1524 Yamamoto T, Yuen DA, Ebisuzaki T (2003) Substitution mechanism of Al ion in MgSiO3 perovskite under high pressure
conditions from first-principles calculations. Earth Planet Sci Lett 206:617–625 Zhang J, Herzberg C (1994) Melting experiments on anhydrous peridotite KLB-1 from 5.0 to 22.5 GPa. J Geophys Res 99 (B9):17729–17742
